Measuring Precipitation from Space: EURAINSAT and the Future (Advances in Global Change Research)

MEASURING PRECIPITATION FROM SPACE

ADVANCES IN GLOBAL CHANGE RESEARCH VOLUME 28

Editor-in-Chief Martin Beniston, University of Geneva, Switzerland

Editorial Advisory Board B. Allen-Diaz, Department ESPM-Ecosystem Sciences, University of California, Berkeley, CA, U.S.A. R.S. Bradley, Department of Geosciences, University of Massachusetts, Amherst, MA, U.S.A. W. Cramer, Department of Global Change and Natural Systems, Potsdam Institute for Climate Impact Research, Potsdam, Germany. H.F. Diaz, Climate Diagnostics Center, Oceanic and Atmospheric Research, NOAA, Boulder, CO, U.S.A. S. Erkman, Institute for Communication and Analysis of Science and Technology – ICAST, Geneva, Switzerland. R. García Herrera, Facultad de Físicas, Universidad Complutense, Madrid, Spain M. Lal, Centre for Atmospheric Sciences, Indian Institute of Technology, New Delhi, India. U. Luterbacher, The Graduate Institute of International Studies, University of Geneva, Geneva, Switzerland. I. Noble, CRC for Greenhouse Accounting and Research School of Biological Sciences, Australian National University, Canberra, Australia. L. Tessier, Institut Mediterranéen d’Ecologie et Paléoécologie, Marseille, France. F. Toth, International Institute for Applied Systems Analysis, Laxenburg, Austria. M.M. Verstraete, Institute for Environment and Sustainability, EC Joint Research Centre, Ispra (VA), Italy.

The titles published in this series are listed at the end of this volume.

MEASURING PRECIPITATION FROM SPACE EURAINSAT and the Future

edited by

V. Levizzani P. Bauer and

F. Joseph Turk

A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN-13 978-1-4020-5834-9 (HB) ISBN-13 978-1-4020-5835-6 (e-book)

Published by Springer, P.O. Box 17,3300 AA Dordrecht, The Netherlands. www.springer.com

Some of the chapters are created by an officer or employee of US Government as part of his/her official duties and are therefore protected by copyright laws as mentioned under Title 17, Section 10-5 of the United States Copyright Law Printed on acid-free paper

All Rights Reserved © 2007 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.

CONTENTS

Contributors Acknowledgments Preface

SECTION 1: CLIMATE MONITORING....………………………………....... 1

European Commission Research for Global Climate Change Studies: Towards Improved Water Observations and Forecasting Capability….….. M. Schouppe and A. Ghazi

xi xxiii xxv

1

3

2

Is Man Actively Changing the Environment?.............................................. D. Rosenfeld

7

3

The Global Precipitation Climatology Project………………………….… A. Gruber, B. Rudolf, M. M. Morrissey, T. Kurino, J. Janowiak, R. Ferraro, R. Francis, A. Chang, and R. F. Adler

25

4

Oceanic Precipitation Variability and the North Atlantic Oscillation…….. P. A. Arkin, H. M. Cullen, and P. Xie

37

5

Global Satellite Datasets: Data Availability for Scientists and Operational Users………………………………..………………….... G. A. Vicente, and the GES DAAC Hydrology Data Support Team

49

SECTION 2: CLOUD STUDIES IN SUPPORT OF SATELLITE RAINFALL MEASUREMENTS……………………………………………….

59

6

61

Cloud Top Microphysics as a Tool for Precipitation Measurements……... D. Rosenfeld

vi

Contents

7

The Retrieval of Cloud Top Properties Using VIS-IR Channels…..……... E. Cattani, S. Melani, V. Levizzani, and M. J. Costa

8

Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events Over Africa and Portugal……………………………….... M. J. Costa, E. Cattani, V. Levizzani, and A. M. Silva

79

97

9

3D Effects in Microwave Radiative Transport Inside Precipitating Clouds: Modeling and Applications……………………………...……….. 113 A. Battaglia, F. Prodi, F. Porcù, and D.- B. Shin

10

Cloud Microphysical Properties from Remote Sensing of Lightning within the Mediterranean……..……………….………….…. 127 C. Adamo, R. Solomon, C. M. Medaglia, S. Dietrich, and A. Mugnai

11

The Worth of Long-Range Lightning Observations on Overland Satellite Rainfall Estimation………………………….………………….... 135 E. N. Anagnostou and T. G. Chronis

12

Neural Network tools for Satellite Rainfall Estimation….…….……….… 149 F. J. Tapiador, C. Kidd, V. Levizzani, and F. S. Marzano

SECTION 3: RAINFALL ALGORITHMS………………………………........

163

13

Passive Microwave Precipitation Measurements at Mid- and High Latitudes…………………………………………………………………... 165 R. Bennartz

14

The Goddard Profiling Algorithm (GPROF): Description and Current Applications……….……………….…………………….…... 179 W. S. Olson, S. Yang, J. E. Stout, and M. Grecu

15

Past, Present and Future of Microwave Operational Rainfall Algorithms…...…………………………………………………………..... 189 R. R. Ferraro

16

Space-Borne Radar Algorithms.................................................................... 199 T. Iguchi

17

Rain Type Classification Algorithm…………………….……………….... 213 J. Awaka, T. Iguchi, and K. Okamoto

Contents

vii

18

Dual-Wavelength Radar Algorithm……………………………….....…… K. Nakamura and T. Iguchi

225

19

A Next-generation Microwave Rainfall Retrieval Algorithm for use by TRMM and GPM…………………………………………….... 235 C. Kummerow, H. Masunaga, and P. Bauer

SECTION 4: BLENDED TECHNIQUES………………………………....…..

253

20

The University of Birmingham Global Rainfall Algorithms….………….. C. Kidd, F. J. Tapiador, V. Sanderson, and D. Kniveton

255

21

Multivariate Probability Matching for Microwave Infrared Combined Rainfall Algorithm (MICRA)…………………………………. 269 F. S. Marzano, D. Cimini, and F. J. Turk

22

Toward Improvements in Short-time Scale Satellite-Derived Precipitation Estimates using Blended Satellite Techniques…….……….. F. J. Turk and A. V. Mehta

281

23

Global Rainfall Analyses at Monthly and 3-h Time Scales…….………… 291 G. J. Huffman, R. F. Adler, S. Curtis, D. T. Bolvin, and E. J. Nelkin

24

CPC MORPHING Technique (CMORPH)……………………………..… 307 R. J. Joyce, J. E. Janowiak, P. Xie, and P. A. Arkin

25

CMAP: The CPC Merged Analysis of Precipitation…………………….... 319 P. Xie, P. A. Arkin, and J. E. Janowiak

26

Rainfall Estimation Using a Cloud Patch Classification Map……………. K.-L. Hsu, Y. Hong, and S. Sorooshian

329

COLOUR SECTION …………………………………………………... CP 1–CP 16 SECTION 5: VALIDATING SATELLITE RAINFALL MEASUREMENTS……………………………………………………..……

343

27

Methods for Verifying Satellite Precipitation Estimates…..…………….... 345 E. E. Ebert

28

Assessment of Satellite Rain Retrieval Error Propagation in the Prediction of Land Surface Hydrologic Variables………….……… 357 E. N. Anagnostou

viii

Contents

29

EURAINSAT Algorithm Validation and Intercomparison Exercise.…….. 369 M. Kästner

30

Ground Validation for the Global Precipitation Climatology Project.......... 381 M. M. Morrissey and S. Greene

31

Validation of Rainfall Algorithms at the NOAA Climate Prediction Center………………………………………………………...... 393 J. Janowiak

32

Ground Networks: Are We Doing the Right Thing?................................... W. F. Krajewski

403

SECTION 6: MODELING PRECIPITATION PROCESSES AND DATA ASSIMILATION FOR NWP…………………………………….

419

33

Aerosol Impact on Precipitation from Convective Clouds……..………… A. Khain, D. Rosenfeld, and A. Pokrovsky

421

34

The Wisconsin Dynamic/Microphysical Model (WISCDYMM) and the use of it to Interpret Satellite-Observed Storm Dynamics….…….. 435 P. K. Wang

35

The European Centre for Medium-Range Weather Forecasts Global Rainfall Data Assimilation Experimentation…...………..…...…… 447 P. Bauer, P. Lopez, E. Moreau, F. Chevallier, A. Benedetti, and M. Bonazzola

36

Rainfall Assimilation into Limited Area Models…………...…….………. 459 A. Buzzi and S. Davolio

37

Implementing an Operational Chain: The Florence LaMMA Laboratory.................................................................................................... 471 A. Ortolani, A. Antonini, G. Giuliani, S. Melani, F. Meneguzzo, G. Messeri, A. Orlandi, and M. Pasqui

SECTION 7: APPLICATIONS TO MONITORING WEATHER EVENTS………………………………………………………………………... 483 38

Satellite Precipitation Algorithms for Extreme Precipitation Events…....... 485 R. A. Scofield and R. J. Kuligowski

Contents

ix

39

Application of a Blended MW-IR Rainfall Algorithm to the Mediterranean…………………………………………………….… 497 F. Torricella, V. Levizzani, and F. J. Turk

40

Retrieving Precipitation with GOES, Meteosat, and Terra/MSG at the Tropics and Mid-latitudes…………..………………………………. 509 C. Reudenbach, T. Nauss, and J. Bendix

41

Model and Satellite Analysis of the November 9–10, 2001 Algeria Flood……………………………………………………………… 521 C. M. Medaglia, S. Pinori, C. Adamo, S. Dietrich, S. Di Michele, F. Fierli, A. Mugnai, E. A. Smith, and G. J. Tripoli

42

Modeling Microphysical Signatures of Extreme Events in the Western Mediterranean to Provide a Basis for Diagnosing Precipitation from Space..………………………………………................ 535 G. J. Tripoli, C. M. Medaglia, G. Panegrossi, S. Dietrich, A. Mugnai, and E. A. Smith

43

Online Visualization and Analysis: A New Avenue to use Satellite Data for Weather, Climate, and Interdisciplinary Research and Applications…………………………………………………….…….. 549 Z. Liu, H. Rui, W. L. Teng, L. S. Chiu, G. Leptoukh, and G. A. Vicente

SECTION 8: THE PRESENT AND FUTURE OF SATELLITE PLATFORMS…………………………………………………………………… 559 44

The Space-Based Component of the World Weather Watch’s Global Observing System (GOS)…………………………………………. 561 D. E. Hinsman and J. F. W. Purdom

45

The Meteosat and EPS/Metop Satellite Series……………………………. 571 J. Schmetz, D. Klaes, A. Ratier, and R. Stuhlmann

46

The Evolution of the NOAA Satellite Platforms………………………….. 587 W. P. Menzel

47

Japan’s Role in the Present and Future Satellite Observation for Global Water Cycle Research…………………………………………. 601 R. Oki and Y. Furuhama

x

Contents

48

International Global Precipitation Measurement (GPM) Program and Mission: An Overview..………………………………….………….… 611 E. A. Smith, G. Asrar, Y. Furuhama, A. Ginati, A. Mugnai, K. Nakamura, R. F. Adler, M.-D. Chou, M. Desbois, J. F. Durning, J. K. Entin, F. Einaudi, R. R. Ferraro, R. Guzzi, P. R. Houser, P. H. Hwang, T. Iguchi, P. Joe, R. Kakar, J. A. Kaye, M. Kojima, C. Kummerow, K.-S. Kuo, D. P. Lettenmaier, V. Levizzani, N. Lu, A. V. Mehta, C. Morales, P. Morel, T. Nakazawa, S. P. Neeck, K. Okamoto, R. Oki, G. Raju, J. M. Shepherd, J. Simpson, B.-J. Sohn, E. F. Stocker, W.-K. Tao, J. Testud, G. J. Tripoli, E. F. Wood, S. Yang, and W. Zhang

49

Snowfall Measurements by Proposed European GPM Mission…………... 655 A. Mugnai, S. Di Michele, E. A. Smith, F. Baordo, P. Bauer, B. Bizzarri, P. Joe, C. Kidd, F. S. Marzano, A. Tassa, J. Testud, and G. J. Tripoli

50

Observing Rain by Millimetre–Submillimetre Wave Sounding from Geostationary Orbit………………………………………………….. 675 B. Bizzarri, A. J. Gasiewski, and D. H. Staelin

51

The CGMS/WMO Virtual Laboratory for Education and Training in Satellite Matters.………………………………………………………... 693 J. F. W. Purdom and D. E. Hinsman

52

The International Precipitation Working Group: A Bridge Towards Operational Applications…………..……………………………. 705 V. Levizzani and A. Gruber

List of Acronyms.………………………………………………………………. 713 List of Symbols and Functions…………………………………………….…… 721

CONTRIBUTORS

CLAUDIA ADAMO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy ROBERT F. ADLER – NASA/Goddard Space Flight Center (GSFC), Laboratory for Atmospheres, Code 613.1, Greenbelt, MD 20771, USA EMMANOUIL N. ANAGNOSTOU – Department of Civil and Environmental Engineering, University of Connecticut, 261 Glenbrook Rd., Storrs, CT 06269-2037, USA, and Marie Curie Visiting Scientist, Hellenic Center of Marine Research (HCMR), Anavissos, Greece ANDREA ANTONINI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy PHILLIP A. ARKIN – Cooperative Institute for Climate Studies (CICS) and Earth System Science Interdisciplinary Center (ESSIC), 2207 Computer and Space Science Building, University of Maryland, College Park, MD 20742-2465, USA GHASSEM R. ASRAR – Science Division, NASA Headquarters, Washington, DC 20546, USA JUN AWAKA – Department of Information Science, Hokkaido Tokai University, Minami-ku, Minami-sawa 5-1-1-1, Sapporo 005-8601, Japan FABRIZIO BAORDO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy ALESSANDRO BATTAGLIA – Meteorologisches Institut, Universität Bonn, Auf dem Hügel 20, D-53121 Bonn, Germany

xi

xii

Contributors

PETER BAUER – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom JÖRG BENDIX – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany ANGELA BENEDETTI – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom RALF BENNARTZ – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA BIZZARRO BIZZARRI – c/o Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy DAVID T. BOLVIN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA MARINE BONAZZOLA – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom ANDREA BUZZI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ELSA CATTANI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ALBERT CHANG† – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA FRÉDÉRIC CHEVALLIER – Laboratoire des Sciences du Climat et l’Environnement (LSCE), CEA-CNRS-UVSQ, Bât. 701, Orme des Merisiers, F-91191 Gif-sur-Yvette CEDEX, France LONG CHIU – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt,

†

deceased

Contributors

xiii

MD 20771, USA, and George Mason University, Fairfax, VA 22030, USA MING-DAH CHOU – Department of Atmospheric Sciences, National Taiwan University, Taipei, Taiwan THEMISTOCLIS G. CHRONIS – NASA/Marshall Space Flight Center (MSFC), Huntsville, AL 35812, USA DOMENICO CIMINI – CETEMPS, Universita’ degli Studi dell’Aquila, via Vetoio, I-67010 Coppito, L’Aquila, Italy MARIA JOÃO COSTA – Centro de Geofísica de Évora, CGE-UE, Universidade de Évora, Rua Romão Ramalho 59, PT-7000-671 Évora, Portugal HEIDI M. CULLEN – The Weather Channel, 300 Interstate North Parkway, Atlanta, GA 30339, USA SCOTT CURTIS – Department of Geography, East Carolina University, East Fifth St., Greenville, NC 27858-4353, USA SILVIO DAVOLIO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy MICHEL DESBOIS – CNRS, Laboratoire de Météorologie Dynamique (LMD), École Polytechnique, 91128 Palaiseau Cedex, France STEFANO DIETRICH – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy SABATINO DI MICHELE – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom JOHN F. DURNING – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ELIZABETH E. EBERT – Bureau of Meteorology Research Centre (BMRC), GPO Box 1289, Melbourne, Victoria 3001, Australia FRANCO EINAUDI – NASA Goddard Space Flight Center (GSFC), Earth-Sun Exploration Division, Code 610, Greenbelt, MD 20771, USA

xiv

Contributors

JARED K. ENTIN – Science Division, NASA Headquarters, Washington, DC 20546, USA RALPH R. FERRARO – Satellite Climate Studies Branch, Cooperative Research Programs (CoRP), NOAA/NESDIS/STAR, and Cooperative Institute for Climate Studies (CICS), University of Maryland, 2207 Computer and Space Sciences Building, College Park, MD 20742, USA FEDERICO FIERLI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy RICHARD FRANCIS – Met Office, FitzRoy Road, Exeter, Devon, EX1 3PB, United Kingdom YOJI FURUHAMA – Japan Aerospace Exploration Agency (JAXA), Office of Space Applications, Tsukuba Space Center, 2-1-1, Sengen, Tsukuba City, Ibaraki 305-8505, Japan ALBIN J. GASIEWSKI – NOAA Environmental Technology Laboratory (ETL), 325 Broadway R/ET1, Boulder, CO 80305-3328, USA ANVER GHAZI† – European Commission, Research Directorate General, Directorate I: Environment, Unit I2: Global Change, B-1049 Brussels, Belgium AMNON GINATI – European Space Agency (ESA), European Space Research & Technology Centre (ESTEC), Keplerlaan 1, Postbus 299, 2200 AG Noordwijk, The Netherlands GRAZIANO GIULIANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy MIRCEA GRECU – Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA J. SCOTT GREENE – College of Atmospheric and Geographic Sciences, University of Oklahoma, 100 East Boyd St., SEC Suite 710, Norman, OK 73019, USA ARNOLD GRUBER – Cooperative Institute for Climate Studies (CICS), Earth System Science Interdisciplinary Center (ESSIC), University of Maryland, 2207 Computer and Space Sciences Building, College Park, MD 20742, USA

†

deceased

Contributors

xv

RODOLFO GUZZI – Agenzia Spaziale Italiana (ASI), Viale Liegi 26, I-00198 Roma, Italy DONALD E. HINSMAN – World Meteorological Organization (WMO), WMO Space Programme, 7bis, Avenue de la Paix, Case postale No. 2300, CH-1211 Geneva 2, Switzerland YANG HONG – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA PAUL R. HOUSER – George Mason University & Center for Research on Environment and Water (CREW), 4041 Powder Mill Road, Suite 302; Calverton, MD 20705-3106, USA KUO-LIN HSU – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA GEORGE J. HUFFMAN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA PAUL H. HWANG – NASA/Goddard Space Flight Center (GSFC), Global Precipitation Measurement (GPM) Project, Code 422, Greenbelt, MD 20771, USA TOSHIO IGUCHI – National Institute of Information and Communications Technology (NiCT), Applied Research and Standards Department, 4-2-1 Nukui Kita-machi, Koganei-shi, Tokyo 184-8795, Japan JOHN E. JANOWIAK – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA PAUL JOE – Meteorological Service of Canada, 4905 Dufferin St., Downsview, Ontario, M3H 5T4, Canada ROBERT J. JOYCE – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA MARTINA KÄSTNER – Deutsches Zentrum für Luft- und Raumfahrt (DLR), Institut für Physik der Atmosphäre (IPA), Oberpfaffenhofen, D-82234 Wessling, Germany RAMESH KAKAR – Science Division, NASA Headquarters, Washington, DC 20546, USA

xvi

Contributors

JACK A. KAYE – Science Division, NASA Headquarters, Washington, DC 20546, USA ALEXANDER KHAIN – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel CHRIS KIDD – School of Geography, Earth and Environmental Sciences, The University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom DIETER KLAES – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany DOMINIC R. KNIVETON – Department of Geography, University of Sussex, Falmer, Brighton, BN1 9SJ, United Kingdom MASAHIRO KOJIMA – Japan Aerospace Exploration Agency (JAXA), Office of Space Applications, Tsukuba Space Center, 2-1-1, Sengen, Tsukuba City, Ibaraki 305-8505, Japan WITOLD F. KRAJEWSKI – C. Maxwell Stanley Hydraulics Laboratory, The University of Iowa, Iowa City, IA 52242-1585, USA ROBERT J. KULIGOWSKI – NOAA/NESDIS/Center for Satellite Applications and Research (STAR), E/RA2, World Weather Building, 5200 Auth Rd., Camp Springs, MD 20746-4304, USA CHRISTIAN KUMMEROW – Department of Atmospheric Science, Colorado State University, Ft. Collins, CO 80523, USA KWO-SEN KUO – Coelum Inc., NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA TOSHIYUKI KURINO – Japan Meteorological Agency (JMA), 1-3-4 Otemachi, Chiyoda-ku, Tokyo 100-8122, Japan GREGORY G. LEPTOUKH – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt, MD 20771, USA, and George Mason University, 4400 University Drive, Fairfax, VA 22030, USA DENNIS P. LETTENMAIER – Department of Hydrology, Wilson Ceramic Lab., University of Washington, Box 352700, Seattle, WA 98195-2700, USA

Contributors

xvii

VINCENZO LEVIZZANI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy ZHONG LIU – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive, Greenbelt, MD 20771, USA, and George Mason University, 4400 University Drive, Fairfax, VA 22030, USA PHILIPPE LOPEZ – European Centre for Medium-Range Weather Forecasts (ECMWF), Research Department, Satellite Section, Shinfield Park, Reading, Berkshire RG2 9AX, United Kingdom NAIMENG LU – National Satellite Meteorological Center (NSMC), China Meteorological Administration (CMA), Beijing 100081, China FRANK S. MARZANO – Centre of Excellence CETEMPS, University of L’Aquila, Italy, and Department of Electronic Engineering, Università “La Sapienza”, via Eudossiana 18, I-00184 Roma, Italy HIROIKO MASUNAGA – Department of Atmospheric Science, Colorado State University, Ft. Collins, CO 80523, USA CARLO M. MEDAGLIA – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy AMITA V. MEHTA – Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA SAMANTHA MELANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy FRANCESCO MENEGUZZO – Consiglio Nazionale delle Ricerche (CNR), Istituto di Biometeorologia (IBIMET), Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy W. PAUL MENZEL – NOAA/NESDIS, Center for Satellite Applications and Research, Cooperative Institute for Meteorological Satellite Studies (CIMSS) and Department of Atmospheric and Oceanic Sciences, University of WisconsinMadison, 1225 W. Dayton Street, Madison, WI 53706, USA GIANNI MESSERI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy

xviii

Contributors

CARLOS A. MORALES RODRIGUEZ – Instituto de Astronomia, Geofísica e Ciências Atmosféricas, Departamento de Ciências Atmosféricas, Rua do Matão, 1226 - Cidade Universitária, CEP: 05508-090, São Paulo, SP, Brazil EMMANUEL MOREAU – NOVIMET SA, Bât. Mermoz, 10-12 Avenue de l’Europe, 78140 Vélizy, France PIERRE MOREL – Université de Paris, Paris, France MARK M. MORRISSEY – College of Atmospheric and Geographic Sciences, University of Oklahoma, 100 East Boyd St., SEC Suite 710, Norman, OK 73019, USA ALBERTO MUGNAI – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via del Fosso del Cavaliere 100, I-00133 Roma, Italy KENJI NAKAMURA – Hydrospheric Atmospheric Research Center (HYARC), Nagoya University, Furocho, Chikusaku, Nagoya 464-8601, Japan TETSUO NAKAZAWA – Meteorological Research Institute (MRI), Japan Meteorological Agency (JMA), 1-1 Nagamine, Tsukuba-city, Ibaraki 305-0052, Japan THOMAS NAUSS – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany STEVEN P. NEECK – Science Division, NASA Headquarters, Washington, DC 20546, USA ERIC J. NELKIN – Sci. Sys. & Appl., Inc., NASA/Goddard Space Flight Center (GSFC), Code 613.1, Greenbelt, MD 20771, USA KEN’ICHI OKAMOTO – Osaka Prefecture University, Sakai, Osaka 599-8531, Japan RIKO OKI – Japan Aerospace Exploration Agency (JAXA), Earth Observation Research and application Center (EORC), 1-8-10 Harumi Chuo-ku, Tokyo 104-6023, Japan WILLIAM S. OLSON – Joint Center for Earth Systems Technology (JCET), University of Maryland Baltimore County (UMBC), Suite 320, 5523 Research Park Drive, Baltimore, MD 21228, USA

Contributors

xix

ANDREA ORLANDI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy ALBERTO ORTOLANI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy GIULIA PANEGROSSI – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA MASSIMILIANO PASQUI – Laboratorio di Meteorologia e Modellistica Ambientale (LaMMA), c/o CNR-IBIMET, Via Madonna del Piano 10 - Edificio D, I-50019 Sesto Fiorentino (FI), Italy SABRINA PINORI – SERCO-DATAMAT Consortium, c/o ESA/ESRIN, via. G. Galilei, I-00044 Frascati, Italy ALEXANDER POKROVSKY – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel FEDERICO PORCÙ – Dip. di Fisica, Università di Ferrara, Edificio C, via Saragat 1, I-44100 Ferrara, Italy FRANCO PRODI – Dip. di Fisica, Università di Ferrara, Edificio C, via Saragat 1, I-44100 Ferrara, Italy, and Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy JAMES F. W. PURDOM – Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Ft. Collins, CO 80523-1375, USA GARUDACHAR RAJU – Indian Space Research Organization (ISRO), New BEL Road, Bangalore 560 094, India ALAIN RATIER – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany CRISTOPH REUDENBACH – Fachbereich Geographie, Laboratory for Climatology and Remote Sensing (LCRS), Philipps-Universität Marburg, Deutschhausstrasse 10, D-35032 Marburg, Germany DANIEL ROSENFELD – Program of Atmospheric Sciences, Institute of Earth Sciences, The Hebrew University of Jerusalem, Jerusalem 91904, Israel

xx

Contributors

BRUNO RUDOLF – Deutscher Wetterdienst (DWD), Postfach 10 04 65, D-63004 Offenbach am Main, Germany HUALAN RUI – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive, Greenbelt, MD 20771, USA, and SSAI, Lanham, Maryland, USA VICTORIA L. SANDERSON – School of Geography, Earth and Environmental Sciences, The University of Birmingham, Edgbaston, Birmingham, B15 2TT, United Kingdom JOHANNES SCHMETZ – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany MICHEL SCHOUPPE – European Commission, Information Society and Media Directorate-General, DG INFSO - G5 - ICT for the Environment, Beaulieu 31 03/20, B-1049 Brussels, Belgium RODERICK A. SCOFIELD† – NOAA/NESDIS, World Weather Building, 5200 Auth Rd., Camp Springs, MD 20746-4304, USA J. MARSHALL SHEPHERD – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ANA MARIA GUEDES DE ALMEIDA E SILVA – Centro de Geofísica de Évora, CGE-UE, Universidade de Évora, Rua Romão Ramalho 59, PT-7000-671 Évora, Portugal JOANNE SIMPSON – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA ERIC A. SMITH – NASA/Goddard Space Flight Center (GSFC), Laboratory for Atmospheres/Code 613.1, Greenbelt, MD 20771, USA BIUNG-JU SOHN - School of Earth and Environmental Sciences, Seoul National University, NS80, Seoul, 151-747, Korea ROBERT SOLOMON – United States Department of Agriculture, Forest Service, Pacific Wildland Fire Sciences Lab., 400 N. 34th Street, Suite 201, Seattle, WA 98103, USA

†

deceased

Contributors

xxi

SOROOSH SOROOSHIAN – Center for Hydrometeorology and Remote Sensing (CHRS), Department of Civil and Environmental Engineering, University of California Irvine, E-4130 Engineering Gateway, Irvine, CA 92697-2175, USA DAVID H. STAELIN – Research Laboratory of Electronics (RLE), Department of Electrical Engineering and Computer Science, and Engineering Systems Division, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA ERIC F. STOCKER – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA JOHN E. STOUT – George Mason University, 4400 University Drive, Fairfax, VA 22030, USA ROLF STUHLMANN – European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Am Kavalleriesand 31, D-64295 Darmstadt, Germany WEI-KUO TAO – NASA/Goddard Space Flight Center (GSFC), Greenbelt, MD 20771, USA FRANCISCO J. TAPIADOR – Instituto de Ciencias Ambientales (ICAM), Universidad de Castilla-La Mancha, Av. Carlos III s/n, 45071 Toledo, Spain ALESSANDRA TASSA – SERCO-DATAMAT Consortium, c/o ESA/ESRIN, via. G. Galilei, I-00044 Frascati, Italy WILLIAM TENG – NASA/Goddard Space Flight Center (GSFC), Earth Sciences Data and Information Services Center, Distributed Active Archive (DAAC), Greenbelt, MD 20771, USA, and SSAI, Lanham, Maryland, USA JACQUES TESTUD – NOVIMET SA, Bât. Mermoz, 10-12 Avenue de l’Europe, 78140 Vélizy, France FRANCESCA TORRICELLA – Consiglio Nazionale delle Ricerche (CNR), Istituto di Scienze dell’Atmosfera e del Clima (ISAC), via Gobetti 101, I-40129 Bologna, Italy GREGORY J. TRIPOLI – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA

xxii

Contributors

F. JOSEPH TURK – Naval Research Laboratory (NRL), Marine Meteorology Division, 7 Grace Hopper Avenue, Monterey, CA 93943-5502, USA GILBERTO A. VICENTE – NOAA/NESDIS/SSD – Products Implementation Branch (PIB), WWB, Room 510, E/SP22, 5200 Auth Road, Camp Springs, MD 20746, USA PAO K. WANG – Department of Atmospheric and Oceanic Sciences, University of Wisconsin-Madison, 1225 W. Dayton Street, Madison, WI 53706, USA ERIC F. WOOD – Department of Civil and Environmental Engineering, Princeton University, Princeton, NJ 08544, USA PINPING XIE – NOAA/NWS/NCEP/Climate Prediction Center (CPC), 5200 Auth Road, Camp Springs, MD 20746, USA SONG YANG – George Mason University, and Goddard Earth Sciences and Technology Center, NASA/Goddard Space Flight Center (GSFC), Code 912.1, Greenbelt, MD 20771, USA WENJIAN ZHANG – Department of Observation & Telecommunications, China Meteorological Administration (CMA), Beijing 100081, China

ACKNOWLEDGMENTS

The idea to write this book came out at the end of the Project EURAINSAT – European satellite rainfall analysis and monitoring at the geostationary scale funded by the European Commission between 2001 and 2003. Several of us participated one way or another in the project, and a major outcome was that it contributed significantly to narrow the distance between the European research groups and the rest of the world. Many of the projects were encouraged by the launch of the first satellite mission dedicated to measure rainfall, the Tropical Rainfall Measuring Mission (TRMM). The outstanding success of TRMM has laid the foundation for a new international collaboration between scientific institutions and space agencies that are working towards another milestone, the Global Precipitation Measurement (GPM) mission. The book is the result of this renewed spirit of cooperation that will have to further increase as we are facing the challenges of the years to come. The community also felt that it was time to put together the amount of knowledge, technology and vision that is available in the field for the benefit of students, professionals and decision makers. More than 20 years have passed since the last book on this subject was printed. Many satellite missions have been launched in between while, at the same time, scientists have made substantial progresses towards transforming satellite rainfall “estimates” into real “measurements” and to produce operational rainfall products readily available for a wide field of applications ranging from climate research and numerical weather prediction to hydrology and agriculture. This book represents a significant effort and each one of the authors did not spare herself or himself in providing top class material, most of the time contributing results and ideas that were worth publishing in peer reviewed journals. The result is a book, which not only photographs the state of the art of the discipline, but also projects it into the future. Several people and organizations need to be warmly thanked and it would be difficult to do it one by one. However, we feel it is appropriate to say at least a few “thank you’s” that will enclose almost all the others.

xxiii

Acknowledgments

xxiv

First, The European Commission (EC) has to be acknowledged for its vision in financing a small group of European scientists on the occasion of the EURAINSAT project and for allowing them to reach out to the rest of the world. Dr. Anver Ghazi, former Head of EC Unit global Change, was a man of vision and unfortunately he has left us while putting together this book. It is an honour for us to have his Preface as one of the last contributions towards building the European Research Area. Two other great friends and colleagues have left us in the meantime, but we feel that they are still with us through their pages in this book: Dr. Rod Scofield and Dr. Al Chang. Thank you, friends. Each one of the authors deserves gratitude for the dedication and the patience with us during the very long time it took to finally print this volume. Our Institutions, CNR, ECMWF and NRL, were supportive in terms of allocating significant portions of our time to the project thus testifying the importance of these pages for an entire community. Two major international organizations were also very much behind us: the International Precipitation Working Group (IPWG) and its major sponsor, the World Meteorological Organization (WMO). They gave us the opportunity to work together and transform our efforts into a global strategy for the future. Finally, our families were deprived of many hours and are part of the project through their understanding and their moral and practical support. Without them this book would have never been printed.

Bologna, Reading, and Monterey, 13 October 2006 Vincenzo Levizzani, Peter Bauer and F. Joseph Turk

PREFACE

Since the last two decades, awareness by the international community of the threats hanging over the planet has increased significantly. Progress in sciences and technologies made it possible to improve our inventory of the state of the environment; evidence was given that rapid changes are occurring across the globe. At present, the complex dynamics of planet Earth and its multidimensional and interrelated processes are at the heart of current scientific investigations. In this context, it is primordial for Europe to further strengthen its capacity to understand, detect, and forecast global change. Detailed process studies and modeling relies permanently on the availability of systematic observations of atmospheric, terrestrial, and oceanic parameters including those of climate. Such a comprehensive observing capability includes large panoply of instruments on various platforms such as ground stations, ships, buoys, floats, ocean profilers, unmanned autonomous underwater and airborne vehicles, balloons, aircraft and satellites. Among these platforms, it is worth noting the unsurpassed coverage brought by Low Earth Orbiting (LEO) and Geostationary (GEO) satellites. Since about three decades, satellites are transmitting homogeneous measurements of many variables characterizing environmental processes and changes. As part of its 5th Framework program for European research and technological development (1998–2002), the European Commission has been supporting a portfolio of 26 research projects in the field of generic Earth observation technologies for the environment. The related 240 research and user organizations have cooperated and delivered innovative processing, modeling, and integration techniques. They have demonstrated elements of future monitoring systems and assessed the strengths and limitations of current space missions. Among these 26 European projects, EURAINSAT has successfully explored the real-time exploitation of data from both LEO and GEO satellites for rainfall estimation and subsequent assimilation into numerical weather prediction models. EURAINSAT has not only promoted the use of SEVIRI data from the recently declared operational METEOSAT-8 satellite,

xxv

xxvi

Preface

it has also prepared the scientific ground for the multilateral Global Precipitation Measurement (GPM) constellation to be launched in the period 2007-2008. The project, which was presented on the occasion of the Flood Media Event organized by the European Commission on 13 October 2003, builds upon several European research projects addressing the issue of precipitation such as EUROTRMM, MUSIC, and MEFFE. The project also opens new collaboration perspectives at European and global levels. It is thus a privilege to introduce this book presenting state of the art in the field of measuring precipitation from space. Anver Ghazi† Head of Unit, Global Change European Commission, Research Directorate General, Brussels, Belgium

†

Hyderabad, 7 March 1940 – Köln, 25 July 2005

Section 1 Climate Monitoring

1 EUROPEAN COMMISSION RESEARCH FOR GLOBAL CLIMATE CHANGE STUDIES: TOWARDS IMPROVED WATER OBSERVATIONS AND FORECASTING CAPABILITY Michel Schouppe and Anver Ghazi† European Commission, Research Directorate General, Brussels, Belgium

Abstract

Improved forecasting at the local, regional and global scales of the interrelated Earth system processes requires, as a prerequisite, a comprehensive global observing strategy susceptible to support the progressive establishment of a global science of integration. This paper concentrates on climate change research and observations, including the key parameters of the global water cycle. Reference is made to the related European research program in the field of “Global Change and Ecosystems” (thematic sub-priority 1.6.3. of the 6th research Framework Programme of the European Union, 2002–2006).

Keywords

Earth system, global change, water, floods, global observations, forecasting, European research, 6th Framework Programme

1

INTRODUCTION

Looking at our global environment, water is often mentioned to be the resource challenge of this century. Water must satisfy the need of all forms of life, starting with the rising water demand by a world’s population that has tripled during the 20th century. Sustainable management of water resources has to consider the increasing imbalance between the geographical and seasonal demand for, and availability of, water. It has to cope with pollution and waste of water, also where it is in abundant supply. The European Union (EU) is pursuing a common water policy since a number of years. The Water Directive of the European Parliament and of the European Council establishes †

Hyderabad, 7 March 1940 – Köln, 25 July 2005

3 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 3–6. © 2007 Springer.

4

M. Schouppe and A. Ghazi

a framework for the Community action in the field of water policy. This EU Directive calls notably for integration and coordination of river basin management across national borders. It also calls for an improved management of the hydrological extremes of floods and droughts. Water management and more generally sustainable development rely heavily on the improved understanding of the multidimensional mechanisms of the water cycle and their interactions with other climate-related processes. This paper concentrates on climate change research and observations, including the key parameters of the global water cycle.

2

WATER CYCLE AND PRECIPITATION

The global water cycle includes many components, mainly atmospheric water vapor, cloud cover, precipitation, surface and subsurface runoff, soil moisture, groundwater, oceans, snow, glaciers, and ice sheets. These components – that illustrate water in its various phases – are affected by a number of physical, chemical, biological, and human-induced processes that play a key role in the Earth’s climate. Precipitation is unanimously recognized as one of the most central variables of the global water cycle, mainly because of its direct significance for the availability of water for drinking and the agriculture, but also because of its impacts on, for instance, runoff over land, soil moisture, floods, stream flow, ocean salinity or atmospheric circulation through associated latent heating. To date, precipitation data are recorded from various remote and in situ instruments (mainly rain gauges, rain radars, and microwave, visible or infrared sensors) based on various platforms (mainly ground-based, airborne or space-borne). However, the fact that water cycle dynamics occur at a wide variety of spatial and time-scales makes them very challenging to observe, understand, and predict. Precipitation is characterized by various regimes. Moreover, determination of precipitation inputs and regimes requires ideal observation of the vertical hydrometeor structure, especially with respect to droplet size, shape, and temperature. Existing instruments and new emerging sensing technologies have each their own advantages but also limitations with respect to spatial sampling and resolution, spatial coverage, temporal coverage, cost of purchase and operation, calibration, accuracy, and consistency of the retrievals, etc. To date, the quantification of adequate precipitation products is hampered by a lack of long-term, stable and high quality observational data and a lack of integration among the various available observation data sources. The observation challenge is a prerequisite for our understanding of the water cycle and its interactions with the climate system at the local, regional, and global scales. Moreover, further research is also required in order to reduce current limitations of the climate models in their ability to simulate aspects of the water cycle such as precipitation amounts and frequency on seasonal

European Commission Research for Global Climate Change Studies

5

and longer time-scales, not only at very large grid scales, but also at lower scales such as catchment scales. At weather time-scales, new optimized nearreal-time assimilation schemes and increased lead-time of weather forecasts open the perspective for improved flood warning. Better forecasts also rely on stronger interdiscipline linkages. Meteorologists and hydrologists, for example, could further expand intelligent coupling of numerical weather forecast models with runoff models.

3

TESTS AND APPLICATIONS

Nowadays, more and more evidence is accumulated that weather and climate are characterized by chaotic aspects that could be inherited from a nonlinear behavior of the internal dynamics of the Earth system. The scientific community is step-by-step investigating the occurrence and amplitude of climate change with respect to changes in the forcings and feedbacks patterns of the Earth system. There is a consensus to categorize water vapor among the largest forcing agents and to group the processes involving water in its three phases among the most important feedback mechanisms that amplify or damp climate perturbations. This is notably true for the water vapor-cloudradiation feedback. Scientists who have studied long-term climatological data confirm increases in the frequency of extreme events – such as extreme temperatures or exceptional intensity of precipitations – decreases in sea level rise and seasonal and perennial snow and ice. In its Third Assessment Report 2001 the Intergovernmental Panel on Climate Change (IPCC) forecasts warmer climates, owing to more frequent and more intense hydrological extremes. Observed climate trends and projected scenarios raise the urgency to address several key scientific questions. To what extent is the global water cycle intensifying? What is the level of interactions between variations in the water cycle and other biogeophysical cycles? More particularly, is the observed increase in extreme events linked to climate change? Is this the result of natural variability or is it caused by human pressure on the environment, or both? One should consider the numerous interconnected factors at play. Climate change is likely to have an incremental effect. Human pressure (for instance, on the atmospheric composition) interferes with the natural flows of energy. More directly, human activities influence the hydrological cycle at various scale, notably through changes in land use, land cover including deforestation, irrigation and drainage, extended pavement, etc.

4

RESEARCH STRATEGY

Quantitative answers to the above questions require a comprehensive and integrated research strategy that builds upon ongoing disciplinary research on parts of the Earth system, but progressively evolves towards addressing interrelations between the dynamics of the Earth system as a whole. Such an

6

M. Schouppe and A. Ghazi

interdisciplinary Earth science vision is challenging and likely to take time to materialize. However, the process of establishing its foundation has already started, notably on both sides of the Atlantic. Such a science of integration is to be articulated over a comprehensive strategy that encompasses the expansion of observing capabilities, the exploitation of increasing computing capabilities, the optimization of performing assimilation schemes, the further development of numerical models in order to deliver both a deeper understanding of the Earth processes and improved forecasting and prediction capabilities.

5

EUROPEAN FRAMEWORK FOR RESEARCH

In the general frame of the 6th Framework Programme for EU research (FP6), “Global Change and Ecosystems” (sub-Priority 1.6.3) has been identified among the main thematic priorities for European research during the period 2002–2006 (http://www.cordis.lu). This “Global Change and Ecosystems” sub-priority comprises a horizontal activity focused on global observations and operational forecasting that directly contributes to the comprehensive strategy described above. The aim is to make systematic observations of primordial parameters, including those of climate, the sea, land and atmosphere, so as to improve forecasting, consolidate long-term observations for the modeling, establish common European database and help Europe to contribute in an integrated way to global observing systems such as the Global Climate Observing System (GCOS), the Global Ocean Observing System (GOOS) or the Global Terrestrial Observing System (GTOS). Focus is on Earth system-integrated observations and international cooperation. Activities to be supported could bring new research elements to the ongoing “Global Monitoring for the Environment and Security” initiative (GMES). Obviously rainfall products are essential to flood warning and mitigation systems. European research is also very active in this area which is included in the “Global Change and Ecosystems” sub-priority under “mechanisms of desertification and natural disasters.” Supported activities aim, in this case, at identifying the links to climate change, assessing and mapping risks while considering the environmental and socioeconomic consequences of floods, storms, and extreme events.

2 IS MAN ACTIVELY CHANGING THE ENVIRONMENT? Daniel Rosenfeld Hebrew University of Jerusalem, Institute of Earth Sciences, Jerusalem, Israel

1

PRECIPITATION AS A CENTERPIECE IN CLIMATE CHANGE

Water is the lifeblood of our livelihood on Earth. The habitability of Earth has been determined mainly by water availability. The deserts are sparsely populated not because they are too hot, but because they lack water. The availability of water to the hottest desert on Earth, e.g., the Nile River has resulted in one of the most densely populated regions on Earth. Temperaturedriven inhabitable areas are due to too low temperatures, and not due to excessively high temperatures. Therefore, our main concern with respect to climate variability and change, at least as far as our living conditions are concerned, has to be its manifestation with respect to water resources, which is determined to a large extent by precipitation. The role of precipitation goes far beyond merely replenishing our water resources. Although only radiative forcing has been considered until now in changing the climate, it is only one of the two major components that energize the climate system. Because the air is transparent to sunlight at most of the solar wavelengths, most of the solar radiation is delivered to the atmosphere indirectly by surface heating. However, most (~77%) of the solar radiation reaching the surface is consumed to evaporate water and the energy is transported in the atmosphere as latent heat within the water vapor. Only when and where the reverse process occurs, i.e., precipitation back to the surface, the released latent heat of condensation causes a net warming of the air, decreases its density and so creates pressure gradients that propel the global circulation of the atmosphere. In terms of energy budget, 47% of the atmospheric radiative loss is balanced by surface sensible (11%) and latent (36%) heat 7 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 7–24. © 2007 Springer.

8

D. Rosenfeld

fluxes. Therefore, the roles of radiative and latent heating in the climate system are of similar magnitudes. These facts have been well known for a long time. The need for a better documentation of the way the latent heating of the atmosphere propels the global circulation has been the main motivation for launching the Tropical Rainfall Measuring Mission (TRMM) satellite in late 1997. So, why have we not considered until now “latent heat forcing” in relation to climate change, in parallel with the “radiative heating forcing”? Latent heat forcing is defined here as the change in the atmospheric heating due to man-made induced changes in precipitation. The term sounds unfamiliar, because we rarely considered man-made impacts on the precipitation, and most that we have done so far with respect to precipitation was driven by radiative considerations, mainly along the following lines: the added greenhouse gases act to accelerate the hydrological cycle, whereas the radiative impacts of the added aerosols slow down the hydrological cycle and redistribute the precipitation. Under such circumstance the changes in latent heating became response and not forcing. We did not consider latent heat forcing because we had little appreciation of man-made induced changes in precipitation by cloud-aerosol interactions and land use changes. Recent observations, mainly from TRMM, have shown that the same microphysical changes that incur the “Cloudmediated aerosol radiative forcing” mainly in shallow clouds, can suppress precipitation from deeper clouds, and change the nature of the precipitation in the deepest tropical convective clouds. Separation of the response and forcing components is a challenge due to the many feedback processes. However, this does not detract from the importance of recognizing the major role of the latent heat forcing. The latent heat forcing is not likely to change the mean global temperature, but not less importantly, it would change the precipitation and climate patterns, storm tracks, etc., which can have profound regional and global impacts that likely occur now. The climate change terminology has to adjust, because in its present from it relates only to radiative effects, and so limits us conceptually from considering other energy forms. The latent heat forcing is a meaningful term if one can show that anthropogenic activity has a substantial impact on precipitation. In the subsequent sections we will address this possibility.

2

MECHANISMS FOR ANTHROPOGENIC IMPACTS ON PRECIPITATION

2.1 Greenhouse gases radiative effects on precipitation The increased GHG reduces thermal radiation from the surface, and increases instead the convective fluxes. Assuming a negligible change in the ratio between latent and sensible heat fluxes, more evaporation would occur and be balanced by the same amount of additional precipitation. This leads towards a

Is Man Actively Changing the Environment ?

9

warmer and wetter climate. The global average precipitation has increased during the last century by about 2.4 mm per decade (Dai et al. 1997, Fig. 10), in line with the nearly 3.0 mm per decade obtained for a 1% per year CO2 increase, as simulated in a coupled ocean–atmospheric GCM study (Russell et al. 1995). Spatially, the precipitation increases during the last century occurred over most of the middle and high-latitude land areas, but not over the low latitude continents (Dai et al. 1997).

2.2 Aerosol radiative effects Aerosols reflect and absorb solar radiation, thereby blocking surface heating and evaporation. Because precipitation must balance evaporation on a global scale, added aerosols to the atmosphere has to slow down the hydrological cycle (Ramanathan et al. 2001) and reduce the precipitation amount. Therefore, estimating the changes in the global energy budget can provide us indirectly the changes in global precipitation. It is much easier to evaluate the global energy budget than measuring global precipitation at the same confidence level. However, based on energy budget considerations alone we cannot predict the spatial and temporal distribution of precipitation, which matters practically much more than the globally averaged amount. An energy flux of 1 W m–2, when used for evaporating water from the Earth surface, evaporates 12.6 mm per year. Because, on a global average, 77% of the surface heating is consumed for evaporation, a net loss of 1 W m–2 of surface heating would result in a loss of evaporation and subsequent precipitation of about 10 mm per year. This would be the precipitation loss due to backscatter of short wave solar radiation to space by white aerosols, such as sulfates, which are mostly anthropogenic over the land areas. This leads towards a cooler and drier climate. The precipitation losses do not end here. In the case of absorbing aerosols, such as carbonaceous particles from smoke and vehicles, the lower troposphere is heated, but without mediation by the surface and the associated surface evaporation. This results in decreasing evaporation and hence precipitation, and leads towards a warmer and drier climate. Indeed, a decreasing trend of solar radiation and surface evaporation (Roderick and Farquhar 2002) has been observed in recent decades (Gilgen et al. 1998; Stanhill and Cohen 2001), while warming was recorded at the same time. This can occur only if a smaller fraction of the surface heating is consumed by evaporation.

2.3 Land use effects The decreasing evaporation fraction (known as the Bowen ratio) can occur not only due to the increase in absorbing aerosols, but also due to changes in land use. Extensive deforestation and cultivation induces decreases in evaporation and in the Bowen ratio. It also induces increases in the surface albedo, which

D. Rosenfeld

10

causes greater reflection of solar energy back to space, and further reduction in the surface heating and evaporation. The anthropogenic changes in the global energy budget of the surface are given in Table 1 (Wild et al. 2004). Table 1. Estimated changes in energy fluxes over global land surfaces for the period 1960–1990 (energy gain for the surface is signed positive). (After Wild et al. 2004.)

(a) Change in absorbed short wave radiation (b) Change in downward long wave radiation (a) + (b) = Surface radiative forcing (c) Change in upward long wave radiation (d) Change in net radiation (a) + (b) + (c) (e) Change in ground heat flux (f ) Change in ice melt

–6 to –9 W m–2 +2 to +3 W m–2 –3 to –7 W m–2 –2 W m–2 –5 to –9 Wm–2 –0.01 Wm–2 –0.2 Wm–2

(1, 2, 3) (4, 5, 6) (4) (7) (7)

1. Gilgen et al. (1998); 2. Liepert (2002); 3. Stanhill and Cohen (2001); 4. Wild et al. (2004); 5. Wild et al. (1997); 6. Garratt et al. (1999); 7. Ohmura (2004).

This amounts to changes in turbulent (latent and sensible heat) fluxes of +5 to +9 W m–2, which means loss of latent and sensible heat flux from the surface to the atmosphere by this amount. The increase in surface sensible heating and drying has the same effect on clouds and precipitation as increasing the amount of pollution aerosols, as described in the next section. This is so because it induces stronger thermals that become stronger updrafts at the cloud base, which produce greater vapour super-saturation and hence a larger fraction of the aerosols nucleate into a greater number concentration of cloud droplets (Williams et al. 2002; Williams and Stanfill 2003).

3

THE AEROSOL CONTROL ON CLOUD PRECIPITATION EFFICIENCY

Clouds precipitate when they survive sufficiently long for the water and/or ice particles to grow to sizes that are large enough to fall to Earth surface. This can happen in two ways. At temperatures above freezing, droplets grow by attracting water vapor through diffusion, until they reach an “effective radius” (re) of about 14 µm, which is related to the ratio of the total volume divided by the total surface area of all the drops in a given cloud volume. After this point, the droplets continue to grow by colliding and coalescing with other water droplets. Eventually, when the drops are bigger than about 0.2 mm in diameter, they fall through the cloud and reach the Earth’s surface as rain. This precipitation process is, however, highly sensitive to the size of the initial cloud droplets. Those with diameters less than about 25 µm are so small that they

Is Man Actively Changing the Environment ?

11

float in air and have a low probability of growing into raindrops by colliding and coalescing with other droplets. Droplets larger than about 30 µm, on the other hand, coalesce much faster (see in Fig. 1).

Figure 1. Large cloud droplets are collected by a falling raindrop, but small cloud droplets follow more closely the airflow streamlines and bypass the falling raindrop, thereby slowing down its further growth.

The other way in which a cloud can grow particles to precipitation size is through ice processes, which operate also when droplet coalescence processes are absent. Ice particles are first formed either when water droplets freeze as they ascend to heights where the temperatures decrease below 0oC, or when ice crystals are nucleated on aerosol particles called ice nuclei. The ice particles collect unfrozen drops faster than water drops of the same mass (Pinsky et al. 1998), and also evaporate more slowly. This process leads often to the growth of ice hydrometeors that fall to Earth, melting to form rain if they reach temperatures above 0oC. If the falling ice particles are large, they may not melt at all before reaching the ground; this is hail. Snowflakes are aggregates of ice crystals, but snow can also be enriched by collecting cloud drops. Therefore, snow that falls from clouds with smaller cloud drops is also suppressed (Borys et al. 2003). Pollution affects these precipitation processes because all cloud droplets – whether formed through the water or ice route – must initially form around an existing aerosol particle, known as a cloud condensation nucleus. But the number of these nuclei depends on the purity of the air. Clean air has relatively few cloud condensation nuclei per unit volume, which means that only about 100 cloud droplets are formed in every cm3 of air. Polluted air, in contrast, has about 1,000 or more cloud condensation nuclei per cm3, mainly in the form of

12

D. Rosenfeld

additional smoke and aerosol particles. Since the total amount of water in polluted and unpolluted clouds at a particular height is about the same, the water in dirty clouds is distributed over a much larger number of droplets, which then must be smaller. In other words, there are lots of very small droplets in polluted clouds, which cannot easily grow into larger drops of precipitation size during the lifetime of the cloud (Fig. 2). Pollution therefore hinders rainfall.

Figure 2. Clean clouds have fewer water droplets per unit volume than polluted clouds, although the size of the droplets in clean clouds increases quickly with height above the base of the cloud. Polluted clouds have a larger number of smaller drops, and their size increases only slowly with height. The small size of the drops in the polluted clouds slows their conversion into rainfall. (From Rosenfeld and Woodley 2001.)

Pollution can also affect the growth of droplets that form through the ice phase. The reason is that polluted clouds have lots of tiny droplets, which freeze more slowly at sub-zero temperatures than the larger drops found in clean clouds. Droplets that are smaller than 30 µm tend to remain in a supercooled liquid state until about –25oC and even down to a chilly –38oC if the cloud contains very small droplets in vigorously ascending air currents (Rosenfeld and Woodley 2000). These supercooled liquid droplets float in the air flowing around the falling ice precipitation particles, and therefore manage to avoid being captured. Cloud model simulations have shown that cleaning up the air in this case would have caused freezing at much higher temperatures and more than double the rainfall amount from this cloud (Khain et al. 2001). The ice particles, in other words, fail to collect enough water from the small

Is Man Actively Changing the Environment ?

13

cloud droplets to grow to precipitation size. Once again, pollution slows down rainfall. Cloud-physics measurements have been usually made from an aircraft equipped with suitable instruments. However, measurements characterizing the processes over large areas and short periods of time are impractical. A satellite-based approach was developed to meet the need for covering large areas (Section 2 of this book). This approach triggered a new era in cloud physics and led to appreciation of the extent by which man-made effects on precipitation occur. Not much was known before this new era about the impact of aerosols from anthropogenic aerosols on precipitation. For example, it was assumed initially that industrial and urban pollution inhibited precipitation (Gunn and Phillips 1957). However, later reports of enhanced rainfall downwind of paper mills (Eagen et al. 1974) and over major urban areas (Braham 1981) suggested that giant CCN caused enhancement of the precipitation (Johnson 1982). However, attempts to correlate the urban-enhanced rainfall to the air pollution sources failed to show any relationship (Gatz 1979). The most plausible explanation for the urban rain enhancement invokes the heat-island effect and increased friction, both of which would tend to increase the surface convergence, resulting in more cloud growth and rainfall over and downwind of the urban areas. On the other hand, the suggestion published in Nature (Cerveny and Balling 1998) that air pollution might enhance precipitation on the large scale in the northeastern USA and Canada and the speculative explanations for this effect would appear to confuse the issue. The truth is that rather little was known definitively about this subject.

4

AEROSOLS FROM URBAN AND INDUSTRIAL AIR POLLUTION

Space-borne measurements of ship tracks in marine stratocumulus provided the first evidence that effluents from ship stacks change cloud microstructure such that their water is redistributed into a larger number of smaller droplets (Coakley et al. 1987). Extrapolation of these observations to clouds that are sufficiently thick for precipitation (i.e., at least 2 km from base to top) would mean that the effluents have the potential to suppress precipitation. Application of the imaging scheme of Rosenfeld and Lensky (1998) to the Advanced Very High Resolution Radiometer (AVHRR) onboard the National Oceanic and Atmospheric Administration (NOAA) orbiting weather satellites have now revealed numerous “ship-track” like features in clouds over land, emanating from major urban and industrial pollution sources. Because the tracks originate evidently from pollution sources, they are named “pollution tracks” by Rosenfeld and Woodley (2003), who identified them as a frequently occurring global phenomenon. Rosenfeld (2000) analyzed the precipitation effects for such a case for pollution tracks in Australia. Satellite retrievals of cloud

14

D. Rosenfeld

microstructure clearly revealed plumes, embedded in extensive cloudy areas, in which the clouds had reduced particle sizes. These plumes originated from major urban areas and industrial facilities such as coal-fired power plants. The satellite retrievals in the polluted and unpolluted regions showed little cloud drop coalescence (as inferred by the methodology of Rosenfeld and Lensky, 1998) in the polluted region and strong coalescence in the pristine clouds. In addition, the Precipitation Radar (PR) onboard the Tropical Rainfall Measuring Mission (TRMM) satellite revealed that the plumes of polluted clouds were devoid of precipitation, whereas the ambient clouds had precipitation intensities exceeding 10 mm h–1. Although producing no precipitation, the clouds in the plumes were as thick as the adjacent precipitating clouds and had no shortage of water. In addition, the PR detected a “bright band” signature, which is indicative of melting snow, in the adjacent precipitating unpolluted clouds, showing further that the pollution also suppresses the processes leading to the growth of ice particles to precipitation size.

Figure 3. Air pollution decreases the drop sizes of convective clouds over the British Isles. This NOAA-AVHRR image from 18 April 1995, 1337 UT was analyzed by the scheme of Rosenfeld and Lensky (1998), showing convective rain clouds with large drops (re > 20 µm, well exceeding the 14 µm precipitation threshold) in the north-westerly flow from the Atlantic Ocean. The clouds interact with the air pollution over the populated land areas and become composed of small drops (re < 10 µm, too small for precipitating) that appear in yellow shades. Note that the sharp distinction of the clouds around the latitude of Glasgow. Northern Scotland is sparsely populated and hence the clouds remain pristine with large drops, as indicated by the red shades (see also color plate 2).

Is Man Actively Changing the Environment ?

15

Quantitative satellite assessment of rainfall downwind of large cities over the USA has shown that warm-season convective rainfall has increased relative to the rural areas (Shepherd et al. 2002; Shepherd and Burian 2003). This has been attributed mainly to the effect of the urban heat island that triggered thunderstorms over the city, although an added component of pollution aerosols effect on invigoration of the thunderstorms is also possible (Williams et al. 2002; Andreae et al. 2004). The invigoration occurs due to the suppression of early onset of precipitation. This has two effects: (1) the water is not lost to rain, but rather it is carried to the heights where it freezes and releases the latent heat of freezing, which adds buoyancy to the cloud volume; (2) the delay of the onset of rainfall also delays the onset of downdraft and the respective dissipation of the convective cloud cell. Givati and Rosenfeld (2004) have quantified the suppression of precipitation on a regional scale due to urban and industrial air pollution such as caused by the pollution tracks in Australia (Rosenfeld 2000), in clouds that are expected to be especially vulnerable (see Fig. 3 for an example of polluted clouds over the British Isles). Such are the orographic shallow precipitation clouds that form over topographical barriers downwind of major coastal areas. For example, the typical rainfall situation in California is from clouds that move inland with the pristine air from the Pacific Ocean. The air mass becomes polluted during its passage over the densely inhabited and industrial coastal areas. This polluted air ascends the hills to the east and forms shallow and short-living orographic clouds, which are responsible for most of the enhanced precipitation over the hills with respect to the upwind lowland. Hilltop measurements in the clouds have shown how the added pollution slows down the accretion of the cloud drops on ice hydrometeors, by about a factor of two (Borys et al. 2003). This should reduce the orographic enhancement factor of precipitation. Indeed, rain gauge analyses of century-long time series showed that the rainfall over the hills was decreased by up to 25% relative to the upwind lowland with major urban areas (Rosenfeld and Givati 2004), in both California and Israel (see Fig. 4). These precipitation losses occur in regions where water is in high demand and great shortage. For example, Israel already has started building seawater desalination plants, at a cost of more than 50 US cents per m3. This puts the economical price on the lost precipitation in Israel. The amount of lost water is estimated by more than 4 × 109 m3 of rainfall volume per year in the Sierra Nevada section of central California alone. At least 1/3 of this amount is loss of exploitable hydrological water.

D. Rosenfeld

16

Figure 4. Topographic cross section showing the effects of urban air pollution on precipitation as the clouds move from west to east from the coast to the Sierra Nevada Mountains and to the eastern slopes. The boxes show the amount of the annual precipitation (mm per year) in each topographic location and the numbers above them show the loss or gain of precipitation (mm per year) at each site. Maritime air (zone 1) is polluted over coastal urban areas (zones 2, 3) – no decrease in precipitation occurs. The polluted air rises over mountains downwind and forms new polluted clouds (zone 4) – decreases of 15%–20% (losses of 220 mm per year) in the ratio between the western slopes to the coastal and plain areas. The clouds reach to the high mountains (zone 5). All precipitation is snow – slight decrease of 5%–7% (loss of 65 mm per year) in the ratio between the summits to the plain areas. The clouds move to the high eastern slopes of the range (zone 6) – increase of 14% (gain of 66 mm per year) in the ratio between the eastern slopes to the plain. (Givati and Rosenfeld 2004, © American Meteorological Society.)

5

SMOKE AEROSOLS FROM BURNING VEGETATION

Suspicions that smoke from burning vegetation suppresses precipitation arose already more than 30 years ago, based on laboratory experiments (Gunn and Phillips 1957) and on observations that precipitation was reduced downwind of seasonal agricultural burning of sugar cane fields in Australia (Warner 1968). Vegetation burning emits large concentrations of small Cloud Condensation Nuclei (CCN) (Hobbs and Radke 1969; Kaufman and Fraser 1997), which modify the cloud drop size distribution (DSD) so that the same amount of water is redistributed over a larger number of smaller drops. The coalescence efficiency of cloud droplets into raindrops is greatly reduced when the radius of the largest cloud droplets is smaller than about 25 µm (Mason and Jonas 1974), which is equivalent to an effective radius (re) threshold of 14 µm (Rosenfeld and Gutman 1994). Effective radius is the cloud droplet size distribution parameter, which is observable by satellites. It has already been observed by satellite that re decreased below the precipitation threshold of re = 14 µm in clouds infected by smoke from burning vegetation in the Amazon (Kaufman and Fraser 1997) and Indonesia (Rosenfeld and Lensky 1998). Satellite observations of the Tropical-Rainfall-Measuring-Mission (Rosenfeld

Is Man Actively Changing the Environment ?

17

1999) show that warm rain processes in convective tropical clouds that ingest smoke from forest fires are practically shut off. The tops of the smokeinfected clouds (see Fig. 5 for a picture of the tops of smoky clouds in the Amazon) has to start developing ice, i.e., grow to altitudes colder than about – 10oC, for the clouds to start precipitating. In contrast, adjacent tropical clouds in the cleaner air precipitate most of their water before ever freezing. Similar TRMM observations were documented in the Amazon (Rosenfeld and Woodley 2003). The very deep clouds that developed there in the smoke, however, had stronger precipitation radar reflectivities and more lightning. Andreae et al. (2004) validated the satellite inferences by in situ measurements of CCN, cloud DSD and precipitation in smoky and smoke-free clouds over the Amazon.

Figure 5. Smoke rising within convective clouds and detrains from their tops. This picture was taken over the Amazon in clouds that were measured with cloud physics aircraft, which documented the small cloud drops and lack of precipitation forming processes by drop coalescence (Andreae et al. 2004).

It was observed by in situ (Andreae et al. 2004) and satellite (Rosenfeld and Lensky 1998) that clouds have to reach greater heights for onset of precipitation in the more polluted and smoky conditions. The lack of early precipitation allows updrafts to accelerate and transport cloud water in deep convection to the high and supercooled regions, where it can release additional latent heat of freezing, which it would not have delivered in the clean case of early rainout.

18

D. Rosenfeld

The added water is available for production of intense ice precipitation, hail and lightning, creating more violent convective storms. Confining the heating to the lower troposphere in the case of cleaner air leads to more confined tropical response and less energy propagation to higher latitudes. Higher latitude propagation is more efficient when the level of maximum heating associated with the release of latent heating in the convective updrafts is seen in the middle to upper troposphere, typical of thunderstorms. Shallow convection, or even cumulus congestus have a maximum heating level in the lower troposphere and thus affect only the regional climate. A shift to more thunderstorms has the effect of generating planetary scale upper level waves that affect the global climate. For example, a perturbation in convection in tropical South America may affect the weather in Europe and Asia in the time scale of intraseasonal oscillations (Grimm and Silva Dias 1995). Quantitative assessment of the smoke microphysical effect on a regional basis is still lacking. Indications of the magnitude of the effect can be obtained from a radar study of rainfall in Thailand, under conditions of suppressed and intense cloud drop coalescence (Rosenfeld and Woodley 2003, Fig. 6). This is relevant, because it has been already shown that cloud drop coalescence in the tropics is dominated by the aerosols, and the variability in the coalescence in the studied clouds in Thailand appeared to be related to the amount of smoke from agricultural burning. The radar estimates of rain cell properties were partitioned using in situ observations of the presence or absence of detectable raindrops on the windshield of the aircraft as it penetrated the updrafts of growing convective towers, 200–600 m below their tops at about the –8oC level (about 6.5 km amsl). Cells observed to contain detectable raindrops during these aircraft penetrations were found to have smaller first-echo depths than cells without observed raindrops when growing through the aircraft penetration level. This faster formation of raindrops is attributed to a rapid onset of coalescence in the convective cells. Convective cells exhibiting a rapid onset of coalescence (Category 2 clouds with detectable raindrops when growing through the aircraft penetration level) produced over a factor of two more rainfall than cells in which the onset of coalescence was slower (Category 1 clouds with no detectable raindrops when growing through the aircraft penetration level). This was true also for convective cloud systems over a fixed-size area covering about 2,000 km2 (Table 1 of Rosenfeld and Woodley 2003).

Is Man Actively Changing the Environment ?

19

10

7

10

6

10

5

10

4

10

3

10

2

3

Rvol [m ]

Strong Coalescence Activity W eak Coalescence Activity

3

5

7 9 11 13 15 17 Hm ax [km]

Figure 6. The mean rain volumes of convective rain cells as a function of maximum precipitation echo top height for cells growing on days with weak coalescence activity (Category 1) and on days with strong coalescence activity (Category 2). The data are plotted at the center point of 2 km intervals of maximum echo top height (Hmax). For the purpose of showing the trend on a logarithmic scale, a value of 102 m3 was used instead of the zero value of RVOL for the 3 km maximum echo top height interval for cells growing on days with weak coalescence activity (Category 1). (After Rosenfeld and Woodley 2003, © American Meteorological Society.)

Figure 6 quantifies the increase of rain production of convective cells with maximum echo top height in both coalescence categories, but the increase is greater for clouds exhibiting a rapid onset of coalescence. The data are plotted at the center point of 2 km intervals of maximum echo top height, Hmax. The difference in rainfall production is the largest for the shallowest clouds, but even the deepest convective cells, in which mixed-phase precipitation forming processes are dominant, the rain-volume from clouds with suppressed coalescence was significantly smaller than the rainfall from clouds with active coalescence. For example, deep tropical cumulonimbus cells (all cells with Hmax >10 km) that exhibit a rapid onset of coalescence produce twice as much more rain volume than cells with slower coalescence activity with a P-value of 0.002. Most important, lifecycle analysis of convective cloud systems in which the convective cells reside showed that the results for the cell scale are preserved on the scale of cloud systems. These radar estimates are likely the lower bound on the actual rainfall differences, because rainfall that occurs in clouds with suppressed coalescence has larger drops than raindrops of the same rain intensity that fall from clouds with active coalescence (Rosenfeld and Ulbrich 2003). This means that the

D. Rosenfeld

20

radar estimated rainfall from the clouds with active coalescence is relatively underestimated with respect to the rainfall from clouds with suppressed coalescence, and hence the estimated effect of coalescence on rainfall is just the lower bound.

6

REVERSING POLLUTION INDUCED SUPPRESSION OF PRECIPITATION

The coin of demonstrated sensitivity of precipitation forming processes to aerosols has two sides. We have already seen how smoke and air pollution aerosols produce large concentration of small CCN that induce small cloud droplets, which are slow to coalesce and form precipitation. In contrast, large soluble aerosols can produce large cloud droplets that are faster to coalesce and form precipitation. Enhancing precipitation by seeding clouds with large hygroscopic particles and ice nuclei is not a new idea. However, this practice did not enjoy much credibility, mainly due to the large natural variability in the conditions compared to the magnitude to the effects of advertent cloud seeding (Silverman 2001, 2003). Natural hygroscopic cloud seeding occurs on a grand scale by salt aerosols from evaporated sea spray. This was shown to be an important mechanism that restores the precipitation from clouds that form in polluted air that flows to the Indian Ocean from south Asia (Rosenfeld et al. 2002), and so cleansing the air from the particulate air pollution. This process leads to the conversion of the polluted continental air mass to a pristine maritime air mass. Therefore, it appears that the oceans are the “green lungs” of the atmosphere, to a large extent because they are salty. The effect occurs not only over oceans, but also over salty land surface areas, such as the desiccated bed of the Aral Sea (Rudich et al. 2002). Due to the ocean salt aerosols, the detrimental impacts of air pollution on precipitation are limited mainly to the land areas, most strongly in the densely populated areas, where people rely on the precipitation for their livelihood. The net effect can take the form of redistribution of precipitation from land to the ocean. More quantitative assessment of that awaits global circulation models that can take these effects into account. Desert dust has mixed effects, because it can contain large particles with varying amounts of soluble materials (Rosenfeld et al. 2001). In any case, desert dust can enhance precipitation due to its strong ice nucleating ability (Rosenfeld and Farbstein 1992; Rosenfeld et al. 2001). This provides us with hope that we will be able to affect the precipitation favorably by both controlling emission and deliberate release of aerosols that incur the desired effects. Spaceborne measurements of the natural, inadvertent and advertent effects on precipitation are an emerging technology with large but yet little exploited potential. The METEOSAT Second Generation (MSG) satellite that was commissioned in early 2004, has opened a new era in which continuous

Is Man Actively Changing the Environment ?

21

(scan every 15 minutes) microphysical satellite retrieval is possible both during day and night times (see Section 2 in this book).

7

IMPACTS ON THE GENERAL CIRCULATION OF THE ATMOSPHERE

The evidence so far shows that anthropogenic aerosols and changes of land surface properties do affect substantially the precipitation and hence the latent heat release. Having defined “latent heat forcing” as the difference between natural and anthropogenically modified latent heat release, and in view of the substantial anthropogenic effects on precipitation, the latent heat appears to be a major component in the energetics of climate change, on a par with the greenhouse gases forcing. If this happens, these changes already must be with us and may possibly explain some of the oddities of recent climatic events. These changes are additional to those induced by the increased greenhouse gases. A global observation system from satellites that measures aerosols, clouds composition and precipitation, is the best way to document quantitatively these effects. General circulation models (GCM) do not contain yet the necessary processes to fully address this issue quantitatively. The best that could be done until now is sensitivity studies, which show considerable sensitivity of the climate system to the impact of anthropogenic aerosols on the precipitation (Nober et al. 2003). Because the hydrological cycle and the anthropogenic effects on it are such a major constituents of the climate system, proper observations and simulations of them are a necessary direction in future development, if we want to have any hope in proper understanding of man-made impacts on our environment and water resources. These new insights no doubt will be featured in future debates on climate change and the impact of pollution on the environment with water resources at the top of the list. Finally, in recognition of the recent findings the 14th Council of World Meteorological Organization, Geneva, May 2003, accepted the following resolution: “Congress noted with concern the new additional evidence, also presented at the 8th WMO Scientific Conference on Weather Modification, that was pointing to an apparent substantial reduction of the rainfall efficiency of clouds by plumes of smoke caused by biomass burning (agricultural practices, forest fires, cooking and heating) and industrial processes. Congress also noted the evidence that such non-raining clouds could regain their raining ability once they moved over oceans or large bodies of water (such as the Aral Sea) because sea-salt was then mixed into the clouds and overrode the detrimental effect of the smoke particles. Therefore, Congress recommended CAS to establish an ad hoc Group on Biomass Burning and Smoke Plumes in general, charge it to prepare a summary report for information of the Members,

D. Rosenfeld

22

addressing relevant issues such as (1) the climatology of smoke and weather active aerosol (Cloud Condensation Nuclei or CCN) plumes, (2) the in situ and remote measurement of CCN and cloud droplet concentrations, (3) strategies to reduce biomass burning and hence the density of smoke plumes, and (4) the seeding procedures and evaluation methods to re-establish raining ability of clouds affected by smoke plumes, and CAS to report to Fifteenth Congress.”

8

REFERENCES

Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Borys, R. D., D. H. Lowenthal, S. A. Cohn, and W. O. J. Brown, 2003: Mountain and radar measurements of anthropogenic aerosol effects on snow growth and snowfall rate. Geophys. Res. Lett., 30, 1538, doi: 10.1029 /2002GL016855. Braham, R. R., Jr., 1981: Summary of urban effects on clouds and rain. Meteorological Monographs, Boston, 18, 141–152. Cerveny, R. S. and R. C. Balling. Jr, 1998: Weekly cycles of air pollutants, precipitation and tropical in the coastal NW Atlantic region. Nature, 394, 561–563. Coakley, J. A., R. L. Bernstein, and P. R. Durkee, 1987: Effects of ship-stack effluents on cloud reflectivity. Science, 237, 1020–1022. Dai, A., I. Y. Fung, and A. D. Delgenio, 1997: Surface observed global land precipitation variations during 1900–88. J. Climate, 10, 2943–2962. Eagen, R. C., P. V. Hobbs, and L. F. Radke, 1974: Particle emissions from a large Kraft paper mill and their effects on the microstructure of warm clouds. J. Appl. Meteor., 13, 535–552. Garratt, J. R., D. M. O’Brian, M. R. Dix, J. M. Murphy, G. L. Stephens, and M. Wild, 1999: Surface radiation fluxes in transient climate simulations. Global Planet. Change, 20, 33–55. Gatz, D. F., 1979: Investigation of pollutant source strength rainfall relationships at St. Louis. J. Appl. Meteor., 18, 1245–1251. Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of short wave irradiance at the surface estimated from Global Energy Balance Archive Data. J. Climate, 11, 2042–2061. Givati, A. and D. Rosenfeld. Quantifying precipitation suppression due to air pollution. J. Appl. Meteor., 43, 1038–1056. Grimm, A. M. and P. L. Silva Dias, 1995: Analysis of tropical extratropical interactions with influence functions of a barotropic model. J. Atmos. Sci., 52, 3538–3555. Gunn, R. and B. B. Phillips, 1957: An experimental investigation of the effect of air pollution on the initiation of rain. J. Meteor., 14, 272–280. Johnson, D. B., 1982: Role of giant and ultragiant aerosol particles in warm rain initiation. J. Atmos. Sci., 39, 448–460. Kaufman, Y. J. and R. S. Fraser, 1997: The effect of smoke particles on clouds and climate forcing. Science, 277, 1636–1638. Khain, A. P., D. Rosenfeld, and A. Pokrovsky, 2001: Simulating convective clouds with sustained supercooled liquid water down to –37.5oC using a spectral microphysics model. Geophys. Res. Lett., 28, 3887–3890. Liepert, B. G., 2002: Observed reductions of surface solar radiation at sites in the United States and worldwide from 1961 to 1990. Geophys. Res. Lett., 29, art. no. 1421. Nober, F., H.-F. Graf, and D. Rosenfeld, 2003: Sensitivity of the global circulation to the suppression of precipitation by anthropogenic aerosols. Global Planet. Change, 37, 57–80. Ohmura, A., 2004: Cryosphere during the twentieth century. Geophys. Monogr., 150, 239–257. Pinsky, M. B., A. P. Khain, D. Rosenfeld, and A. Pokrovsky, 1998: Comparison of collision velocity differences of drops and graupel particles in a very turbulent cloud. Atmos. Res., 49, 99–113.

Is Man Actively Changing the Environment ?

23

Ramanathan, V., P. J. Crutzen, J. T. Kiehl, and D. Rosenfeld, 2001: Aerosols, climate and the hydrological cycle. Science, 294, 2119–2124. Roderick, M. L. and G. D. Farquhar, 2002: The cause of decreased pan evaporation over the past 50 years. Science, 298, 1410–1411. Rosenfeld, D., 1999: TRMM Observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105–3108. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287, 1793–1796. Rosenfeld, D. and H. Farbstein, 1992: Possible influence of desert dust on seedability of clouds in Israel. J. Appl. Meteor., 31, 722–731. Rosenfeld, D. and I. M. Lensky, 1998: Spaceborne sensed insights into precipitation formation processes in continental and maritime clouds. Bull. Amer. Meteor. Soc., 79, 2457–2476. Rosenfeld, D. and C. W. Ulbrich, 2003: Cloud microphysical properties, processes, and rainfall estimation opportunities. Chapter 10 of “Radar and Atmospheric Science: A Collection of Essays in Honor of David Atlas”. R. M. Wakimoto and R. Srivastava, Eds., AMS, Meteorological Monographs, 52, 237–258. Rosenfeld, D. and W. L. Woodley, 2000: Convective clouds with sustained highly supercooled liquid water down to –37.5oC. Nature, 405, 440–442. Rosenfeld, D. and W. L. Woodley, 2001: Pollution and Clouds. Physics World, Institute of Physics Publishing LTD, Dirac House, Temple Back, Bristol BS1 6BE, UK, 33–37. Rosenfeld, D. and W. L. Woodley, 2003: Closing the 50-year circle: From cloud seeding to space and back to climate change through precipitation physics. Chapter 6 of “Cloud Systems, Hurricanes, and the Tropical Rainfall Measuring Mission (TRMM)” W.-K. Tao and R. F. Adler, Eds., 234 pp., AMS, Meteorological Monographs 51, 59–80. Rosenfeld, D., Y. Rudich, and R. Lahav, 2001: Desert dust suppressing precipitation – A possible desertification feedback loop. Proc. Natl. Acad. Sci.USA, 98, 5975–5980. Rosenfeld, D., R. Lahav, A. P. Khain, and M. Pinsky, 2002: The role of sea-spray in cleansing air pollution over ocean via cloud processes. Science, 297, 1667–1670. Rudich Y., D. Rosenfeld, and O. Khersonsky, 2002: Treating clouds with a grain of salt. Geophys. Res. Lett., 29, doi:10.1029/2002GL016055, 2002. Russell, G. L., J. R. Miller, and D. Rind, 1995: A coupled atmosphere – ocean model for transient climate change studies. Atmos. – Ocean, 33, 687–730. Silverman, B. A., 2001: A critical assessment of glaciogenic seeding of convective clouds for rainfall enhancement. Bull. Amer. Meteor. Soc., 82, 903–923. Silverman, B. A., 2003: A critical assessment of hygroscopic seeding of convective clouds for rainfall enhancement, Bull. Amer. Meteor. Soc., 84, 1219–1230. Shepherd, J. M., H. Pierce, and A. J. Negri, 2002: Rainfall modification by major urban areas: Observations from spaceborne rain radar on the TRMM satellite. J. Appl. Meteor., 41, 689–701. Shepherd, J. M. and S. J. Burian, 2003: Detection of urban-induced rainfall anomalies in a major coastal city. Earth Interactions, 4, 1–17. Stanhill, G. and S. Cohen, 2001: Global dimming: a review of the evidence for a widespread and significant reduction in global radiation with discussion of its probable causes and possible agricultural consequences. Agricultural and Forest Meteorology, 107, 255–278. Wild, M., A. Ohmura, and U. Cubasch, 1997: GCM simulated surface energy fluxes in climate change experiments. J. Climate, 10, 3093–3110. Wild, M., A. Ohmura, H. Gilgen, and D. Rosenfeld, 2004: On the consistency of trends in radiation and temperature records and implications for the global hydrological cycle. Geophys. Res. Lett., 31, L11201, doi:10.1029/2003GL019188.. Williams, E. and S. Stanfill, 2002: The physical origin of the land-ocean contrast in lightning activity. C. R. Physique, 3, 1–16.

24

D. Rosenfeld

Williams, E., D. Rosenfeld, N. Madden, J. Gerlach, N. Gears, L. Atkinson, N. Dunnemann, G. Frostrom, M. Antonio, B. Biazon, R. Camargo, H. Franca, A. Gomes, M. Lima, R. Machado, S. Manhaes, L. Nachtigall, H. Piva, W. Quintiliano, L. Machado, P. Artaxo, G. Roberts, N. Renno, R. Blakeslee, J. Bailey, D. Boccippio, A. Betts, D. Wolff, B. Roy, J. Halverson, T. Rickenbach, J. Fuentes, and E. Avelino, 2002: Contrasting convective regimes over the Amazon: Implications for cloud electrification. J. Geophys. Res., 107, D20, 8082, doi:10.1029/2001JD000380.

3 THE GLOBAL PRECIPITATION CLIMATOLOGY PROJECT Arnold Gruber1, Bruno Rudolf 2, Mark M. Morrissey3, Toshiyuki Kurino4, John Janowiak5, Ralph Ferraro1, Richard Francis6, Albert Chang7†, and Robert F. Adler7 1

National Oceanic and Atmospheric Administration, National Environmental Satellite and Data Information Service, Camp Springs, MD, USA (current affiliation: Cooperative Institute for Climate Studies, Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, USA) 2 Deutscher Wetterdienst, Offenbach, Germany 3 University of Oklahoma, School of Meteorology, Norman, OK, USA 4 Japan Meteorological Agency, Tokyo, Japan 5 National Oceanic and Atmospheric Administration , National Weather Service,Climate Prediction Center, Camp Springs, MD, USA 6 EUMETSAT, Darmstadt, Germany (current affiliation: Met Office, Exeter, United Kingdom) 7 National Aeronautics and Space Administration, Goddard Space Flight Center, Greenbelt, MD, USA

1

INTRODUCTION

The global precipitation climatology project was formed by the World Climate Research Program in 1986 (WCRP 1986) to exploit the capabilities of satellites, combined with gauges to provide the best available estimates of global precipitation. The objectives were to: • Improve understanding of seasonal to interannual and longer-term variability of the global hydrological cycle. • Determine the atmospheric heating needed for climate prediction models. • Provide an observational data set for model validation and initialization and other hydrological applications.

†

Deceased 26 May 2004

25 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 25–35. © 2007 Springer.

A. Gruber et al.

26

Figure 1. Organization structure of the GPCP.

Initially the goal was to provide 10 years (1987–1996) monthly mean fields of precipitation on a 2.5 degree × 2.5 degree latitude/longitude grid. However, as the GPCP evolved it was possible to extend the time period back to 1979 (Adler et al. 2003) and produce data sets at pentad, 2.5 degree × 2.5 degree scales (Xie et al. 2003 ) and global daily estimates of precipitation at 1 degree × 1 degree latitude/longitude scales (Huffman et al. 2001 ). Examples of these data sets will be shown later.

2

ORGANIZATION

Since first described by Arkin and Xie (1994) the organization of the GPCP has been modified to the current structure shown in Fig. 1. The geostationary satellite operators of Japan (GMS), Europe (METEOSAT), and the USA (GOES) collect geostationary infrared satellite data for the estimation of precipitation using the GPI (Arkin and Meisner 1987). The USA also provides NOAA polar orbiting data for areas where there is no geostationary coverage, typically over the Indian Ocean area. These data, covering the latitudes 40N–40S are combined at the Geostationary Satellite Precipitation Data Center located at the NOAA National Weather Service. Microwave estimates are provided by two centers. One is located at NASA/GSFC, USA, and provides oceanic estimates of rainfall based on the emission characteristics of precipitation at low frequencies (19 GHz) (Wilheit et al. 1991) and the other is obtained over land using the scattering characteristic of raining clouds (Ferraro and Marks 1995). The latter is

The Global Precipitation Climatology Project

27

located at NOAA/NESDIS in the USA. Gauge data are collected and analyzed at the Global Precipitation Climatology Center (GPCC) located at the Deutscher Wetterdienst (DWD), Germany. These estimates of rain (MW, IR, and gauges) are sent to the GPCP Merge Development Center where they are merged and global precipitation maps are prepared. The basic merging procedure including error estimates is described by Huffman et al. (1997). For the latest version of the monthly mean products rain estimates from the TOVS sounder are utilized in high latitude precipitation estimates where the infrared and microwave estimates are either not available or unreliable. The Surface Reference Data Center located at the University of Oklahoma has the responsibility for collecting and maintaining high quality reference data and for performing validation of the rain products. In the following sections a brief review of the data sets and some of the applications of the data will be provided.

3

DATA SETS

The GPCP produces three global rainfall data sets; monthly mean on 2.5 × 2.5 degree latitude/longitude grid, pentad mean also on a 2.5 × 2.5 degree grid and a daily rainfall map on 1 × 1 degree latitude/longitude grid. These data sets share much data in common but there are differences in the input data. One important factor is that there are different satellite sources of data with different lengths of time especially for the monthly and pentad means. The input data to the various products are summarized in Table 1. More details including the procedures for the product development are in the references cited in the text below, which briefly summarizes each product. The monthly mean is the second version of this data set and differs from the first version in that it is extended back in time to 1979 and has complete global coverage (Version 1 began in 1987 and suffered extensive gaps pole ward of 60 N, S latitude). The monthly mean is a blend of satellite infrared and microwave estimates of rainfall and gauges. The analysis procedure is to use stepwise bias corrections; i.e., infrared estimates are adjusted to microwave estimates (presumed less biased) and then satellite estimates adjusted to gauges. The final blending uses inverse error weighting. Version 1 is described more completely in Huffman et al. (1997) and Version 2 is described by Adler et al. (2003). The annual average rainfall from Version 2 for the period 1979–1998 is shown in Fig. 2. The global average rainfall is 2.61 mm per day. The average over land is 1.96 mm per day and the value over the oceans is 2.85 mm per day. The largest amounts of rain occur over the tropical oceans approaching 10 mm per day in the equatorial western Pacific. It is interesting to note that the precipitation in the NH mid-latitude storm tracks exhibit values that are as high as some of the tropical areas, especially over land.

A. Gruber et al.

28 Table 1. Summary of data sources for GPCP products.

GPCC Gauge Data GHCN +CAMS Gauge Data SSM/I Data Geostationary IR MSU TOVS Polar IR * OPI GTS gauge data

Monthly Mean V2 1986–present

Pentad

Daily

07/1987–11/1987, 01/1988–present 1986–present 1979–present

07/1987–11/1987, 01/1988–present 1997–present

1979–1986 07/1987–11/1987, 01/1988–present 1986–present 07/1987–11/1987, 01/1988–present 1986–present 01/1979–06/1987, 12/1987

1997–present 1997–present 1979–present

* Used to fill gaps in geostationary IR OPI – Outgoing longwave radiation precipitation Index GHCN – Global Historical Climate Network CAMS – Climate Assessment Monitoring System (Climate Prediction Center, NOAA) SSM/I – Special Sensor Microwave Imager MSU – Microwave Sounding Unit TOVS – TIROS Operational Vertical Sounder GTS – Global telecommunication System GPCC – Global precipitation Climatology Centre.

Figure 2. GPCP annual mean precipitation 1979–1998 based on Version 2.

The Global Precipitation Climatology Project

29

The Pentad data set also is a blend of satellite estimates and gauges and extends from 1979 and as the Version 2 is continuing. The analysis procedure is somewhat different than used in the monthly mean product. Satellite rain estimates are combined by maximum likelihood estimates, and then bias is removed by solving a Poisson equation with gauges as boundary conditions. The pentad data sum to the GPCP Version 2 monthly means. The pentad data development is described in Xie et al. (2003). The daily global precipitation product is based only satellite estimates. The methodology of using both microwave and infrared is somewhat different than in Version 2 and is fully described by Huffman et al. (2001). The main reason for satellite only is that daily gauges are not uniform in their definition of a day so that it is not practical to obtain a global analysis from gauges that will be homogeneous. However, the influence of the gauges is felt since the daily estimates are constrained to sum to the monthly totals from Version 2. The advantages of the higher temporal resolution of the daily and the pentad data sets is that important variability in precipitation such as the 30–60 day Madden–Julian Oscillations can be detected and monitored. And the higher spatial resolution of the daily data are more useful for hydrological and water management applications. An example of what is gained from the higher spatial temporal resolution is shown in Fig. 3 which shows time longitude sections for the latitude band 5N-5S for the period January 1997–October

Figure 3. Time longitude sections at 5N-5S, January 1997–October 1998; monthly, pentad and daily data (see also color plate 1).

A. Gruber et al.

30

1998, which encompasses the 97/98 El Niño. Monthly, pentad and daily sections are shown. Note that the El Niño is well depicted in all the sections but the higher resolution data can depict Madden–Julian Oscillations and daily disturbances.

4

VALIDATION

An important aspect of the work done by the GPCP is validation of the data. Figure 4 shows some the validation of the monthly mean GPCP rain estimates against data from a mesoscale gauge precipitation network over Oklahoma, for a specific grid box. These gauges were not used in the gauge analysis that was merged with the satellite estimates of rain. The upper part of the figure shows a time series from January 1994 through 1998. The red curve is the multisatellite (MW and IR) estimates of rain, the green curve is the satellite and gauge merged estimates and the blue line is the reference gauge data.

Figure 4. Validation of GPCP monthly mean precipitation. Top –time series, bottom – scatter plot. See text for explanations.

The Global Precipitation Climatology Project

31

This comparison shows the importance of the gauges in the blending procedure in determining the correct magnitude of the rainfall. The lower part of the figure is a scatter plot between of the GPCP rain against the reference data. There is essentially no bias and the correlation between the two data sets is 0.93. More information about validation can be obtained from the Surface Reference Data web page: http://srdc.evac.ou.edu/.

5

APPLICATIONS

As indicated in the introduction some of the goals were to produce a data set that can be used to understand the seasonal and longer variability of precipitation and use these data for model validation. Some examples of the application of these data will be shown in this section. Seasonal variability is shown in the time longitude sections of precipitation anomalies of Fig. 5 for the tropical zone 5N-5S for the period July 1987 through 1998. This is a particularly good area to look at seasonal variability since it is influenced by the quasi-periodic El Niño. The right hand side of the figure shows anomalies of rain which is contrasted to anomalies in sea surface temperature shown on the left-hand side. The points to notice are the good agreement between the anomalies especially in depicting El Niño and La Niña conditions. Noteworthy are the dominant El Niño of 1997/98 and the weaker and extended El Niño conditions of 1991–1995.

Figure 5. Time latitude sections of sea surface temperature anomalies (left) and monthly mean precipitation anomalies (right).

32

A. Gruber et al.

With regard to looking for trends in the data one has to be cautious since the record is relatively short about 23 years and composed of estimates from different satellites and times and different algorithms. When we look at the time series of globally averaged precipitation, shown in Fig. 6, we see a weak negative trend; however, in view of the variability of the time series and the use of various satellite inputs over the period this weak negative trend is not believable. However, when we look at trends on a regional basis we can find areas where the trend is more significant (Fig. 7). For example the area in the western pacific shows a region of large negative and positive trends. These trends are consistent with analysis of atoll data performed by Morrissey and Graham (1996). Another application of the data is validation of model outputs. In the example below we show difference maps between GPCP and NCEP/NCAR Reanalysis data (Janowiak et al. 1998). Figure 8 shows that over the ocean heavy rain areas (ITCZ, SPCZ, storm tracks) the reanalysis underestimates rain and in the tropical dry zones it overestimates rain. There are also overestimates of rain over tropical land areas of NE South America extending into the Caribbean and SE. Comparisons such as this can provide invaluable information for diagnosing deficiencies in the model physics and parameterization schemes.

Figure 6. Time series of globally averaged monthly mean precipitation. Red line is 12-month running mean, blue line is 5-month running mean. Trend is mm per day per month.

The Global Precipitation Climatology Project

33

Figure 7. Regional trends – mm per day per month. Note values are multiplied by 100.

Figure 8. Mean annual difference between NCEP/NCAR reanalysis data and the GPCP product.

6

CONCLUDING REMARKS AND FUTURE OUTLOOK

The GPCP has been successful in developing and producing rainfall data sets that are now useful and will continue to do so. We have exceeded our original mandate which was to produce a 10-year monthly mean data set on a 2.5 degree × 2.5 degree grid. In fact we are producing monthly mean and pentad

A. Gruber et al.

34

data set from 1979 onward on a 2.5 degree grid and a daily data from 1997 onward on a 1 degree × 1 degree grid. Nevertheless there are still challenges for the GPCP. Among the them are: • The determination of absolute values of rain over the open ocean. This is difficult to achieve because there is generally no reference data available. However, the use of the TRMM precipitation radar data should prove to be of value for this problem. • The determination of solid precipitation rates is still largely unknown, although the water equivalent of solid precipitation is included over land through the gauges. Recently some work has been done that suggests that high frequency (150, 183 GHz) passive microwave data available from AMSU may help in identifying solid precipitation. • Accurate estimates of precipitation in regions of complex terrain. This is a challenge for both in situ and remote-sensing measurements.

Acknowledgments: The NOAA Office of Global Programs is to be acknowledged for their support. The statements contained within the manuscript are not the opinions of the funding agencies or the US government, but reflect the authors’ opinions.

7

REFERENCES

Adler, R. F., G. J. Huffman, A. Chang, R. Ferraro, P.-P. Xie, J. Janowiak, B. Rudolf, U. Schneider, S. Curtis, D. Bolvin, A. Gruber, J. Susskind, and P. A. Arkin, 2003: The Version 2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–present). J. Hydrometeor., 4, 1147–1167. Arkin, P. A. and B. N. Meisner, 1987: the relationship between large scale convective rainfall and cold cloud over the western hemisphere during 1982–1984. Mon. Wea. Rev., 115, 41–74. Arkin, P. A. and P.-P. Xie, 1994: The Global Precipitation Climatology Project: First Algorithm Intercomparison Project. Bull. Amer. Meteor. Soc., 75, 401–419. Ferraro, R. and G. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground based radar measurements. J. Atmos. Oceanic Technol., 12, 755–770. Huffman, G. J., R. F. Adler, P. A. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The global precipitation climatology project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78, No.1, 5–20. Huffman, G. J., R. F. Adler, M. M. Morrissey, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multi-satellite observations. J. Hydrometeor., 2, 36–50. Janowiak, J. E., A. Gruber, C. R. Kondragunta, R. E. Livezey, and G. J. Huffman, 1998: A comparison of the NCEP/NCAR reanalysis precipitation and the GPCP rain gauge-satellite combined data set with observational error considerations. J. Climatol., 11, 2960–2979. Morrissey, M. and N. E. Graham, 1996: Recent trends in rain gauge precipitation measurements from the Tropical Pacific: evidence for an enhanced hydrological cycle. Bull. Amer. Meteor. Soc., 77, 1206–1219.

The Global Precipitation Climatology Project

35

WCRP, 1986: Report of the workshop on global large scale precipitation data sets for the World Climate Research Programme. WCP-111, WMO/TD-No. 94, 45 pp. Wilheit, T. J., A. T. C. Chang, and L. S. Chiu, 1991: Retrieval of monthly rainfall indices from microwave radiometric measurements using probability distribution functions. J. Atmos. Oceanic Technol., 8, 118–136. Xie, P.-P., J. E. Janowiak, P. A. Arkin, R. F. Adler, A. Gruber, R. Ferraro, G. J. Huffman, and S. Curtis, 2003: GPCP pentad precipitation analyses: An experimental data set based on gauge observations and satellite estimates. J. Climate, 16, 2197–2214.

4 OCEANIC PRECIPITATION VARIABILITY AND THE NORTH ATLANTIC OSCILLATION Phillip A. Arkin1, Heidi M. Cullen2, and Pingping Xie3 1

ESSIC, University of Maryland, College Park, MD, USA The Weather Channel, Atlanta, GA, USA 3 Climate Prediction Center, NOAA/NWS/NCEP, Camp Springs, MD, USA 2

1

BACKGROUND

The North Atlantic Oscillation (NAO) was originally discovered by Sir Gilbert Walker (Walker and Bliss 1932). Walker found that the strengths of the semipermanent low and high pressure systems in the North Atlantic Ocean were inversely correlated during Northern Hemisphere winter. Periods when both systems were stronger than average, resulting in an intensified pressure gradient and stronger than average westerly winds, alternate with periods when both are weaker than average and the oceanic westerlies are relatively weak. Changes from strong to weak gradient conditions are characterized by a “seesaw” of atmospheric mass from south to north in the North Atlantic Ocean. The NAO is the dominant coherent mode of climate variability in Boreal winter in the region (van Loon and Rogers 1978), with a strong signature throughout the troposphere. Variability associated with the NAO has been shown to control the location and intensity of the storm track in the North Atlantic and Western Europe (Hurrell 1995), and exerts a strong influence on temperature and precipitation anomalies in eastern North America, Europe and the Mediterranean region (Cullen et al. 2002). Indices representing the variability of the NAO have been computed from a variety of atmospheric data. The original and simplest such index is based on the normalized difference between surface pressure at stations that represent the variations of the Azores High and the Icelandic Low (van Loon and Rogers 1978). More recently, in an attempt to capture more effectively

37 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 37–47. © 2007 Springer.

P. A. Arkin et al.

38

the large-scale variability of the full atmospheric phenomenon, indices based on empirical orthogonal function analysis of surface pressure or midtropospheric geopotential height have been developed and published (Barnston and Livezey 1987). All indices of the NAO are formulated so that high values represent periods when the pressure gradient in the North Atlantic, and hence the westerlies, are stronger than average. Spectral analysis of NAO indices indicates that, while variability is found on all time scales longer than a few days, significant peaks are found at periods of 2–3 and 7–10 years, along with an increasing trend over the last 50 years (Hurrell 1995). Precipitation anomalies associated with the NAO have been described by Walsh and Portis (1999) and Cullen et al. (2002). However, precipitation variability over the Atlantic Ocean associated with the NAO and the associated atmospheric circulation features have not been well described, principally due to the lack of reliable oceanic precipitation data sets. The recent availability of global precipitation data sets may have reduced this impediment, and the remainder of this paper is an attempt to use these new data sets to provide a thorough description of the climate variability associated with the NAO. We begin with a description of the data used in the study, followed by a presentation of the regional manifestations of the NAO in precipitation, storminess and circulation during both the winter and summer seasons. In section 4 we investigate the tropical variations in precipitation associated with the NAO, and we finish in section 5 with conclusions and a discussion of further work required.

2

DATA

We utilize three primary data sets in this study. Precipitation will be obtained from the CPC Merged Analysis of Precipitation (CMAP; Xie and Arkin 1996, 1997), while analyses of the large-scale atmospheric circulation are derived from the NCAR/NCEP reanalysis (Kalnay et al. 1996). These will be supplemented by an analysis of the frequency of cyclonic storminess derived from surface pressure data. CMAP provides monthly maps of precipitation averaged over 2.5° × 2.5° areas for the period January 1979–December 2002 (regularly updated) based on a variety of estimates derived from satellite observations and rain-gauge data. The CMAP algorithm uses a linear combination of all available satellite estimates, with weights over land derived from a comparison to a gauge-only analysis. Over ocean, the weights are determined from retrospective analysis comparing the different satellite-derived estimates to atoll rain-gauge observations. These weights are extrapolated into higher latitudes in a seasonally and latitudinally varying pattern based on the characteristics of each product. In regions where no usable satellite-based estimate is available,

Oceanic Precipitation Variability and the North Atlantic Oscillation

39

a version of CMAP using precipitation forecasts from the NCEP/NCAR Reanalysis (Kalnay et al. 1996) is created. The model-derived precipitation amounts are treated as another estimate, with weights derived in a manner similar to the others. This permits the creation of two versions of CMAP: one that is spatially complete and includes information that is partially model-dependent and another that is purely observation-based but with gaps in coverage. In both cases, the intermediate product is combined with a gauge-based analysis to minimize bias. The combination is done in such a manner as to preserve the gradients in the satellite combination while using the absolute value of the gauge analysis where sufficient gauges are available. In this study we will use the version of CMAP that includes model-derived information as our primary source, because we expect that the model forecasts are sufficiently skillful in the high latitudes of the North Atlantic Ocean to be useful. The influence of the model forecasts on our results will be determined by comparing them to a parallel analysis using the observation-only CMAP. Circulation data used in this study will be derived from the NCEP/NCAR reanalysis (Kalnay et al. 1996). This reanalysis utilized a modern data assimilation/forecast system together with as complete as possible an observational database. Its results have been used in more than 1,000 published studies (Kalnay, personal communication, 2002). We have examined winds and geopotential heights at 1,000 and 500 hpa, and will use the 500 hpa geopotential height variability in this paper to describe the NAO signal in atmospheric circulation. Since it has been shown that the NAO is the principal control on the winter storm tracks in the region (Hurrell 1995), we will utilize an additional data set based on an automated tracking of cyclonic circulation centers in surface pressure fields. This analysis was produced by Chandler and Jonas (personal communication, Center for Climate Systems Research at Columbia University and NASA/GISS) and consists of gridded maps of the number of cyclonic centers passing through each region during each month from January 1962 to December 1998. In this study we use the results for the 20year period from January 1979 to December 1998.

3

REGIONAL MANIFESTATIONS

The manifestations of the NAO have been most extensively studied in the region of the North Atlantic Ocean and the adjacent portions of North America and Europe, and in the Northern Hemisphere winter season. In the following sections, we will investigate the precipitation, storminess and circulation anomalies that characterize the NAO during the Boreal winter (December–March) and Boreal summer (June–September) seasons.

40

P. A. Arkin et al.

3.1 Boreal winter In the long-term winter (DJFM) mean, the fields of precipitation, storm frequency and height (Fig. 1) exhibit a consistent relationship over the North Atlantic, with collocated axes of maximum storminess and precipitation associated with the strongest gradients in geopotential height. A bifurcation appears in the eastern Atlantic, with the strongest axis continuing over northern Europe and a secondary maximum extending over the Mediterranean. The highest values of precipitation are located to the southwest of the most frequent storminess. Monthly means were subtracted to obtain anomalies, and correlations calculated against the NAO index published on the Climate Prediction Center website. The patterns obtained (Fig. 2) are consistent with previous work but exhibit significant new details over the ocean. Precipitation exhibits an alternating pattern of correlation with the NAO, with positive values over the highest latitudes, negative between 30°–40° N, positive between the equator and 20° N, and negative further south. These correlations are strongest over the ocean, but extend into the continents.

Figure 1. Mean December–March CMAP precipitation (mm per day. top left), 500 hpa heights (dm, top right) and storm frequency (storms per month in each 2º × 2º area, below left). Contours are 0.2, 0.5, 1, 2, 4, 6, 8, 12 and 16 for precipitation, the same with an additional one at 0.1 for storminess, and at 10 dm interval for height, with the southernmost contour at 580 dm.

Oceanic Precipitation Variability and the North Atlantic Oscillation

41

Figure 2. Correlation of CMAP precipitation (top left), 500 hpa heights (top right) and storm frequency (below left) with NAO index during December–March. Contours for precipitation and storminess are at ± 0.15, 0.3, 0.4, 0.5 and 0.6; for height at ± 0.25, 0.4, 0.5, 0.6, 0.7 and 0.8.

The 500 hpa height correlations are less complex, with strong negative correlations in the North Atlantic, positive correlations to the south, and an additional center of negative correlation further south. This pattern is to be expected, considering that this NAO index is based on empirical orthogonal function analysis of 700 hpa height anomalies. The frequency of storms exhibits a correlation pattern similar to that of precipitation in the North Atlantic, but with essentially no sign of the area of positive correlations south of 25°N. Composites of precipitation, storminess frequency, and height computed based on quartiles of the NAO index using monthly anomalies confirm the implications of the correlation plots: high index periods are accompanied by increased precipitation and storminess in the higher latitudes, with an anomalous height gradient somewhat further south than the maxima in precipitation and storm frequency (see Fig. 3). Low index periods exhibit decreases in precipitation and storminess in the northern latitudes, with increases between 30°–40° N extending from the Atlantic eastward as far as the Middle East. The westerly anomalous wind maxima at 1,000 and 500 hpa are located south of the greatest increases in precipitation and storminess.

42

P. A. Arkin et al.

Figure 3. Composite CMAP precipitation (mm per day, left) and storm frequency (storms per month, right) anomalies based on quartiles of NAO index during December–March. Plots are difference (high minus low) between high and low index months. Contours are at ± 0.2, 0.5, 1, 2, 4 and 8.

While the consistency of these results with those of earlier studies is encouraging, we attempted to explicitly address their robustness. Since the NAO exhibits relatively little serial correlation on the monthly time-scale (one month lag correlation of 0.16), correlations with an absolute value of 0.3 or greater are highly significantly different from zero. As one can see from Fig. 2, the areas of significance in geopotential height are large; while the values of the precipitation correlations are slightly lower, they are still significant over substantial areas and the pattern is similar. The negative correlation between precipitation and NAO near the equator in the Atlantic is significant as well. Assessing the robustness of composite patterns is more challenging; explicit statistical tests are not well suited as many of the underlying assumptions are violated. Here we have used Monte Carlo testing to determine the relationship between composites chosen based upon an NAO

Figure 4. Composite CMAP precipitation anomalies during the highest (left) and lowest (right) quartiles of December – March months based upon the NAO index normalized by the standard deviation of 1000 randomly selected composites. Contours are at ±1, 1.5, 2 and 3 standard deviations.

Oceanic Precipitation Variability and the North Atlantic Oscillation

43

index and those chosen based on random time series. A set of 1,000 composites of CMAP precipitation for the months December–March based on random series was created, using the same methodology as described above. The standard deviation of the sets of high and low composites was calculated and used to normalize the composites based upon the NAO index. The results are shown in Fig. 4, and indicate that at least the northern triad of composite anomalies is reliably distinct from noise. Further, these patterns are very similar, although reversed, during periods of high and low NAO index.

3.2 Boreal summer The long-term summer mean circulation, precipitation, and storm frequency patterns are displaced northward and weakened relative to those of winter. The subtropical features in geopotential height are stronger relative to the higher latitude features. The oceanic precipitation maximum in the middle latitudes shows evidence of strong tropical influence, while the maximum in storm frequency is displaced well northward and is quite weak over Europe.

Figure 5. Correlation of CMAP precipitation (top left), 500 hpa heights (top right) and storm frequency (below left) with NAO index during June – September. Contours as in Fig. 2 except for height, which here has the same contours as precipitation and storminess.

44

P. A. Arkin et al.

Correlations between the NAO index and precipitation, storm frequency and 500 hpa geopotential height (Fig. 5) show that, while the NAO continues to influence the circulation during summer, its manifestations are markedly weaker than in winter. The areas of correlation with absolute values >0.3 are smaller, and the multicomponent pattern observed during winter is not present. In geopotential height, particularly at 1,000 hpa (not shown), the correlation maxima are more fragmented than during winter, while in precipitation and storm frequency they are more restricted to northern latitudes. Precipitation correlations are strongest over Northern Europe, while storminess correlations extend more uniformly from eastern North America across the Atlantic and into Europe. One feature, the area of positive correlation between precipitation and the NAO in the western subtropical Atlantic, raises the question of whether some connection exists between the NAO and tropical storm activity. Some aspects of the changes in geopotential height correlation between winter and summer are notable. At 500 hpa, both the middle and the southern band of correlations exhibit two clear centers during the summer (Fig. 5), while during winter (Fig. 2) only the southernmost band exhibits dual centers. At 1,000 hpa (not shown), the singular dipole seen in winter is replaced in summer with dual centers in the southern band of positive correlations. Note that this implies that no simple index based on one pair of stations can capture the full variability associated with the NAO. In addition, the centers during the summer are displaced eastward relative to the winter. Composites of precipitation, storm frequency and geopotential height based on the NAO index for the months from June to September (not shown) largely confirm the results suggested by the correlations shown in Fig. 5. The storminess and precipitation features are displaced well northward relative to the winter, and the southern features found in precipitation during winter are not seen in summer. The composite anomalies in precipitation and geopotential height are stronger over the European continent during summer, rather than being strongest over the ocean as observed during winter. The storm frequency composite anomaly is more uniform over land and ocean than the other composites. The positive correlation between NAO index and precipitation in the western subtropical Atlantic is borne out by the composite precipitation anomaly, but is not supported by the other composites. The Monte Carlo calculations of precipitation composites give results supportive of these findings: the composite precipitation anomalies over the European continent and in northern latitudes are robust, with substantial areas exceeding 1.5 standard deviations. These features are approximately symmetric between low and high index periods. In the western tropical Atlantic, interestingly, an area of positive anomaly in precipitation that exceeds 1.5 standard deviations is found during high index months without any indication of a corresponding negative anomaly during low index months.

Oceanic Precipitation Variability and the North Atlantic Oscillation

4

45

GLOBAL ASSOCIATIONS

Hoerling et al. (2001) have shown that the increasing trend in the NAO index during the period from 1950 to 2000 is associated in atmospheric models forced by observed sea surface temperature variability with a trend in tropical precipitation over the Indian and Pacific oceans. Here we examine our shorter period of record to determine whether periods of high and low NAO index exhibit a systematic relationship to tropical precipitation variability. The correlations between the NAO and CMAP precipitation over the globe exceed 0.3 in absolute value in few areas outside the tropical and Northern Hemisphere Atlantic Ocean (not shown). The area of negative correlations hinted at in the southernmost latitudes in Fig. 2 (December– March) extends to about 20° S centered on the east coast of South America. The location and shape hints at a relationship to the South Atlantic Convergence Zone, and areas of negative correlation 14 µm as for “warm rain” clouds. In mature or stratiform clouds with slow vertical air motions there is sufficient time for the cloud drops to freeze and become ice crystals that aggregate into snow flakes. The super-cooled water at the upper and hence colder portions of such clouds is typically completely converted into ice crystals and precipitation particles, so that the clouds are said to be glaciated. Ice crystals that form by heterogeneous freezing in a super-cooled cloud grow on expense of the cloud drops, so that in such glaciated cloud each ice crystal contains a water amount that was previously distributed in many drops. Therefore, the ice crystals are much larger than the cloud droplets from which they were formed, and hence possess much larger re than that of the

64

D. Rosenfeld

source water cloud. These ice crystals aggregate with time into snow flakes. When they fall through super-cooled clouds the snow flakes continue growing by accreting the cloud droplets. Distinction must be made here between this situation and ice clouds that form by homogeneous freezing of cloud droplets near or above the –38°C isotherm. Such homogeneously frozen drops retain their mass and become similarly small and numerous ice crystals, that have no efficient mechanism to aggregate into precipitation particles at such cold temperatures and small sizes. This situation can be detected by the existence of ice clouds with small re at temperatures 14 µm (Rosenfeld and Gutman 1994; Gerber 1996). This can be used quantitatively for improving the accuracy of rainfall measurements from space, as demonstrated by Lensky and Rosenfeld (1997) for the NOAA/AVHRR. This principle was applied to an operational rainfall product (Ba and Gruber 2001).

4

THE DEPENDENCE OF RAINFALL REMOTE SENSING ON HYDROMETEOR SIZE DISTRIBUTIONS

The previous sections provided a physical basis for indirect measurements of precipitation based on the retrieved cloud top composition and temperature, using the visible and IR wavebands. Direct measurements of precipitation use the microwave frequencies that interact directly with the precipitation size particles (diameter >0.1 mm). Direct measurements are divided into active and passive. Active microwave measurements require a radar instrument that transmits pulses of radiation and receives the back scattered echoes. The echo intensity is converted into precipitation intensity according to the radar equation. The backscatter occurs mainly in the Rayleigh regime, where the intensity of the scattered radiation is proportional to the 6th power of the particle size. This highly nonlinear relation causes a serious problem of nonuniqueness between the echo and precipitation intensities. Small concentrations of large hydrometeors can produce the same reflectivity factor (Z, [mm6 m–3]) as much larger concentrations of smaller hydrometeors that form much greater equivalent rain intensity (R, [mm h–1]). This nonuniqueness in the Z–R relations has been historically the weakest point in radar rainfall measurements from both surface and space-borne platforms. Rainfall measurements from space with passive microwave rely on both thermal emission and back scatter. The thermal emission does not depend so strongly on the particle size, but because of that it can’t distinguish between cloud water and precipitation. The thermal emission can be used mainly above the cold background of the oceans, which appear cold due to the small microwave emissivity of flat water surfaces. Most of the signal from deep convection comes from the backscatter of the upwelling thermal radiation back downward. This signal strongly depends on the particle size, as in the case for the radar. The larger the hydrometeors the more energy is backscattered to the surface and the lower the satellite measured brightness temperature becomes. It can be also viewed as larger particles backscatter more strongly the 3 K background of the outer space, but it is inaccurate in a

D. Rosenfeld

68

strict physical sense. Here again the same rain intensity that is associated with larger hydrometeors would be interpreted as a stronger passive microwave signal and heavier rain. Ice hydrometeors are colder and have smaller emissivity than raindrops, and hence they would create lower brightness temperature for the same precipitation intensity, and more so when they reside higher in the cloud and at lower temperatures. Therefore, the correct interpretation into rainfall of both active and passive microwave measurements depends strongly on the relations between the hydrometeor size distributions, types and the rain intensity. It is essential to obtain information about the hydrometeor sizes for achieving reasonable accuracy of the precipitation intensities. This can be achieved by space borne radar measurements with multiple wavelength radar, as planned for the Global Precipitation Measurement (GPM; Smith et al. 2007) mission. The already available space borne radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite has only a single wavelength, and hence requires external independent information about the Z–R relations. Passive microwave measurements are conducted in several wavelengths simultaneously, but have very limited capability to resolve the particle size, especially in deep convective clouds.

5

CLOUD MICROSTRUCTURE AND Z–R RELATIONSHIPS

In the previous section we have seen that independent external information on hydrometeors type and sizes is essential for improving the accuracy of both radar and passive microwave precipitation measurements. Such information can be obtained from the inferred cloud microstructure and precipitation forming processes as obtained from the T–re relations. We will review here the precipitation evolution in microphysically maritime and continental clouds. Microphysically maritime clouds are composed of low concentration of large cloud drops that coalesce readily into warm rain. In contrast, microphysically continental clouds are composed of small drops that form precipitation mainly by ice processes.

5.1 Microphysically maritime clouds: Evolution of warm rain In a hypothetical rising cloud column with active coalescence, the initial dominant process would be widening of the cloud drop size distribution into large concentrations of drizzle drops; the drizzle continues to coalesce with other drizzle and cloud drops into raindrops, which will continue to grow asymptotically to the equilibrium raindrop-size distribution (RDSD), with the median volume drop diameter D0e = 1.76 mm (Hu and Srivastava 1995). Therefore, during the growth phase of the precipitation particles the rain rate R increases with D0, median volume drop diameter, and this would increase

Cloud Top Microphysics as a Tool for Precipitation Measurements

69

D0 for a given R. Ideally, for rainfall with drops that fall from cloud top while growing, R would increase with the fall distance from the cloud top, mainly by growth of the falling drops due to accretion and coalescence, and to a lesser extent by addition of new small raindrops, until the raindrops grow sufficiently large for breakup to become significant. Shallow orographic clouds can present conditions such as some distance below the tops of convective clouds. Therefore, similar evolution of R can be observed on a mountain slope, such as documented by Fujiwara (1965). Different values of R near cloud top or in shallow orographic clouds can come mainly from changing NT, the total concentration of raindrops, because the drop size is bounded by the limited vertical fall distance along which they can grow. This would cause orographic precipitation to have small drops and for R to depend mainly on NT, and more so with shallower clouds and stronger orographic ascent, because the stronger rising component supplies more water for the production of many small raindrops not too far below cloud top, which are manifested as a larger R.

5.2 Microphysically continental clouds: Evolution of cold rain Microphysically “continental” clouds are characterized by narrow cloud drop size distributions, and therefore by having little drop coalescence and warm rain. Most raindrops originate from melting of ice hydrometeors that are typically graupel or hail in the convective elements, and snowflakes in the mature or stratiform clouds. Graupel and hail particles grow without breakup while falling through the super-cooled portion of the cloud, and continue to grow by accretion in the warm part of the cloud, where they melt. Large melting hailstones shed the excess melt-water in the form of a RDSD about which little is known. The shedding stops when the melting particles approach the size of the largest stable raindrops, which are later subject to further breakup due to collisions with other raindrops. In fact, new raindrop formation is limited only to the breakup of preexisting larger precipitation particles. Therefore, we should expect that in such clouds there would be, for a given R, a relative dearth of small drops and excess of large drops compared to microphysically “maritime” clouds with active cloud drop coalescence. Deep continental convective clouds would therefore initiate the precipitation by forming large drops that with maturing approach DSDe from above. This is in contrast with the approach from below for maturing maritime RDSD. Recent satellite studies (Rosenfeld and Lensky 1998) have shown that microphysically maritime clouds are associated typically with a “rainout” zone, i.e., the fast conversion of cloud water to precipitation cause the convective elements to lose water to precipitation while growing. This leaves less water carried upward to the super-cooled zone, so that weaker ice precipitation can develop aloft. Williams et al. (2002) have recognized this

D. Rosenfeld

70

as a potential cause to the much greater occurrence of lightning in continental compared to maritime clouds. Williams et al. (2002) noted that frequent lightning occurred also in very clean air during high atmospheric instability, probably because the strong updraft leaves little time to the formation of warm rain, and carries the large raindrops that manage to form up to the super-cooled levels of the clouds, where they freeze and participate in the cloud electrification processes (Atlas and Williams 2003). This difference between continental and maritime clouds means that mostly warm rain would fall even from the very deep maritime convection, which reaches well above the freezing level, whereas precipitation from continental clouds would originate mainly in ice processes. Therefore, the expected difference in RDSD between microphysically maritime and continental clouds is expected to exist also for the deepest convective clouds that extend well into the sub-freezing temperatures.

1000 Florida Cont Florida Mar LBA Cont LBA M ar India Cont India Mar Kw aj M ar

-3

N [m m m ]

100

10

1

0.1

0.01 0

1

2

3

4

5

6

D [m m ] Figure 2. Disdrometer measured RDSDs of continental (solid lines) and maritime (dashed lines) rainfall, as microphysically classified by VIRS overpass. The RDSD is averaged for the rainfall during ± 18 h of the overpass time, and the concentrations are scaled to 1 mm h-1. The disdrometers are in Florida (Teflun B), Amazon (LBA), India (Madras) and Kwajalein. (From Rosenfeld and Ulbrich 2003, courtesy of Amer. Meteor. Soc.)

Cloud Top Microphysics as a Tool for Precipitation Measurements

71

5.3 Quantifying the role of cloud top re on RDSD The ultimate test for the role of cloud microstructure is comparing the RDSD of clouds at the same location, but at different times, when they possess maritime or continental microstructure. Rosenfeld and Ulbrich (2003) did exactly that. They used the VIRS (Visible and Infrared sensor) onboard TRMM (Tropical Rainfall Measuring Mission) satellite to retrieve the microstructure of rain clouds over disdrometer sites. The clouds were classified into continental, intermediate and maritime, using the methodology of Rosenfeld and Lensky (1998). The RDSDs from the continental and maritime classes during the overpass time + 18 h were lumped together and plotted in Fig. 2. Indeed, the continental and maritime RDSDs are well separated in Fig. 2, with the continental clouds producing greater concentrations of large drops and smaller concentrations of small drops. A comparison between the directly measured disdrometer rainfall and the calculated accumulation by applying the TRMM Z-R relations (Iguchi et al. 2000) to the disdrometer measured Z resulted in a relative overestimate by more than a factor of two of the rainfall from the microphysically continental clouds compared to the maritime clouds. The evidence shows that it is mainly the cloud microstructure that is responsible for the large systematic difference in the RDSD and Z–R relations between maritime and continental clouds. There are several possible causes for these differences, all working in the same direction:

5.4 Extent of coalescence The cloud drop coalescence in highly maritime clouds is so fast that rainfall is developed low in the growing convective elements and precipitates while the clouds are still growing. The large concentrations of raindrops that form low in the cloud typically fall before they have the time to grow and reach equilibrium RDSD, thereby creating the rainout zone (Rosenfeld and Lensky 1998) less than 2 km above cloud base height. Therefore, D0 remains much smaller than D0e, as was shown in Fig. 6b of Rosenfeld and Ulbrich (2003). In microphysically continental clouds with suppressed coalescence the cloud has to grow into large depth before start precipitating, by either warm or cold processes. The raindrops that fall through the lower part of the cloud grow by accretion of small cloud drops, so that they tend to breakup much less than drops that grow mainly by collisions with other raindrops, as is the case for maritime clouds. This process allows D0 to exceed D0e in the growing stages of the precipitation, and later approach it from above when the raindrop collisions become more frequent with the intensification of the rainfall.

D. Rosenfeld

72

5.5 Warm versus cold precipitation processes The rainout of the maritime clouds (Rosenfeld and Lensky 1988) depletes the cloud water before reaching the super-cooled levels (Zipser and LeMone 1980; Black and Hallett 1986), so that mixed phase precipitation would be much less developed in the maritime clouds compared to the continental. This is manifested in the smaller reflectivity aloft in the maritime clouds (Zipser and Lutz 1994), which is a manifestation of the smaller hydrometeors that form there (Zipser 1994). In contrast, the suppressed coalescence in continental clouds leaves most of the cloud water available for growth of ice hydrometeors aloft, typically in the form of graupel and hail. These ice hydrometeors can grow indefinitely without breakup, until they fall into the warm part of the cloud and melt. The melted hydrometeors continue to grow by accretion of cloud droplets, until they exceed the size of spontaneous breakup or collide with other raindrops. Therefore, convective rainfall that originates as ice hydrometeors would have D0 > D0e, and would approach D0e from above with maturing of the RDSD.

5.6 Strength of the updrafts Updrafts are typically stronger in more continental clouds, and therefore contribute to more microphysically continental clouds and less warm rain processes, as discussed already above. In addition, stronger updrafts allow drops with greater minimal size to fall through them. In addition, stronger updrafts leave less time for forming of warm rain and rainout, and advect more cloud water to the super-cooled zone. Therefore, due to the reasons already discussed in (a) and (b), the stronger updrafts are likely to lead to precipitation with greater D0 and smaller R for the same Z.

5.7 Evaporation More continental environments have typically higher cloud base and lower relative humidity at the sub-cloud layer. Evaporation depletes preferentially the smaller raindrops and works to increase D0.

6

RELATIONS BETWEEN PRECIPITATION MEASUREMENT BIASES AND CLOUD MICROSTRUCTURE

Now we can return and try explaining the large discrepancies between satellite measurements and rain-gauge estimates that were found over central Africa, while in the Amazon regions the rain gauges coincide closely with satellite estimates (McCollum et al. 2000). This is consistent with the in situ microphysical observations showing that clouds in the Amazon are microphysically maritime, similar to equatorial pacific clouds (Stith et al. 2002), except for periods when they are polluted by smoke from forest fires (Andreae

Cloud Top Microphysics as a Tool for Precipitation Measurements

73

et al. 2004). McCollum et al. (2002) have also shown that remote sensing of rainfall measurements by both passive microwave (SSM/I) and surface radar measurements have relative overestimates when moving from the east coast of the USA to central USA by 25–30% (see Fig. 3). McCollum et al. (2002) suggested that this bias is caused by the greater continentality of the rain clouds in central USA. When they used multispectral algorithm that takes into account cloud top microstructure the systematic bias somewhat decreased.

Figure 3. Spatial distribution of area-averaged multiplicative bias for the SSM/I with respect to the estimates of the bias-adjusted hourly digital precipitation radar rainfall estimates on a national grid. (From McCollum et al. 2002, courtesy of Amer. Meteor. Soc.)

Additional indication of the precipitation forming processes is the lightning activity. Clouds are electrified when graupel collides with ice crystals in a super-cooled water cloud. Therefore lightning is a manifestation of intense ice precipitation forming processes. Tropical maritime clouds have between one and two orders of magnitude less lightning for the same amount of rainfall of continental clouds (Petersen and Rutledge 1998). Satellite rainfall estimates in the Amazon regions and central Africa are comparable in magnitude, while there is much more lightning activity over central Africa with much less rain gauge measured rainfall. We postulate that rainfall regime over the Amazon is less microphysically continental than that over central Africa, and hence having smaller hydrometeors and larger extent of cold anvils for the same rainfall amounts. This suggestion is further supported by findings of Petersen and Rutledge (2001). Greater continentality is characterized by larger amounts of cloud water carried up to the upper portions of the cloud, where it freezes and forms large ice hydrometeors, and the released latent heat of freezing invigorates the updrafts and loft the large ice particles to great heights (Andreae et al. 2004). The large ice particles aloft produce smaller passive microwave brightness temperatures that are interpreted as greater rain intensities, by a

74

D. Rosenfeld

factor of 2–3 compared to the maritime clouds. The large raindrops that form when these ice particles melt equally cause radar overestimates of rain intensities by a similar factor of 2–3 compared to the maritime clouds. As shown earlier in this section and elsewhere (Rosenfeld and Lensky 1998; Andreae et al. 2004), the continentality of the clouds can be quantified independently of the radar by satellite retrieved T–re relationships.

Figure 4. MSG image from 20 May 2003 1342 UTC, over central Africa at a 1200 × 1200 km2 rectangle between 1–12N and 15–26E. The area shows the transition between the relatively microphysically maritime clouds over the forested area (dark surface) and microphysically continental clouds over the dry lands of the Sahel to the north (bright surface). The T–re relations of the continental clouds (1) show much smaller re for a given T compared to the maritime clouds (2). The median re of the maritime clouds (the yellow line) saturates near T = –20°C, indicating glaciation at that temperature. The small median re at area 1 even above the –40°C isotherm indicates homogeneous glaciation of the cloud water and hence low precipitation efficiency. The color scheme is red for the visible, green for 3.9 µm reflectance component, and blue for temperature. For full description and interpretation of the color table is given in Rosenfeld and Lensky (1998). The T–re lines represent percentiles of re for a given T in 10% steps for each line, between 5–95%. The median is between the yellow and green lines. (see also color plate 3)

Cloud Top Microphysics as a Tool for Precipitation Measurements

75

Figure 5. The measurement bias of the TRMM precipitation radar (PR, middle panel) and TRMM passive microwave (TMI, lower panel) in relation to the continentality of the rain clouds, as given by the mean 30 dBZ echo top height in precipitation features with TMI signal of ice scattering. Note the large overestimate where large ice hydrometeors exist high in the clouds. (Presented by S. Nesbitt at the TRMM Hawaii Scientific Conference, Honolulu, HI, 22–26 July 2002.)

7

CONCLUSIONS

Differences in clouds microstructure can explain systematic biases of up to a factor of 3 in passive MW and radar direct rainfall measurements. The cloud microstructure can be obtained by T–re relations that are obtained from the operational NOAA orbital satellites. The MSG, which was commissioned in early 2004, is the first of a new generation of geostationary satellites that have sufficient resolution for providing useful T–re relations of convective clouds (e.g., Fig. 4). Combining the cloud’s microphysical continentality from T–re analyses such as shown in Fig. 4 with the radar and passive microwave measurements has the potential of eliminating much of the measurement biases shown in Fig. 5. Night time capabilities for microphysical measurements are also emerging (Lensky and Rosenfeld 2003a). Indirect rainfall measurements can be also substantially improved using the information about cloud top composition. Only the first steps have been done so far in this direction during daylight (Lensky and Rosenfeld 1997; Ba and Gruber 2001) and night (Lensky and Rosenfeld 2003b). The implication for future missions is that rainfall measuring satellite should include both microwave and VIS/IR sensors, and the rain estimation should use this added information, without which systematic bias errors greater than a factor of 2 are difficult to avoid.

D. Rosenfeld

76

8

REFERENCES

Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Arking, A. and J. D. Childs, 1985: Retrieval of cloud cover parameters from multispectral satellite images. J. Climate Appl. Meteor., 24, 322–333. Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the western hemisphere during 1982–84. Mon. Wea. Rev., 115, 51–74. Atlas, D. and T. Bell, 1992: The relation of radar to cloud area-time integrals and implications for rain measurements from space. Mon. Wea. Rev., 120, 1997–2008. Atlas, D. and C. R. Williams, 2003: An anatomy of a continental tropical convective storm. J. Atmos. Sci., 60, 3–15. Ba, M. and A. Gruber, 2001: GOES Multispectral Rainfall Algorithm (GMSRA). J. Appl. Meteor., 40, 1500–1514. Black, R. A. and J. Hallett, 1986: Observations of the distribution of ice in hurricanes. J. Atmos. Sci., 43, 802–822. Chang, F. L. and Z. Li, 2003: Retrieving vertical profiles of water-cloud droplet effective radius: Algorithm modification and preliminary application. J. Geophys. Res., 108 (D24), 4763, doi:10.1029/2003JD003906. Fujiwara, M., 1965: Raindrop-size distribution from individual storms. J. Atmos. Sci., 22, 585–591. Gerber, H., 1996: Microphysics of marine stratocumulus clouds with two drizzle modes. J. Atmos. Sci., 53, 1649–1662. Hu, Z. and R. Srivastava, 1995: Evolution of the raindrop-size distribution by coalescence, breakup and evaporation: Theory and observations. J. Atmos. Sci., 52, 1761–1783. Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM Precipitation Radar. J. Appl. Meteor.,: 39, 2038–2052. Ebert, E. E., M. J. Manton, P. A. Arkin, R. J. Allam, G. E. Holpin, and A. Gruber, 1996: Results from the GPCP algorithm intercomparison program. Bull. Amer. Meteor. Soc., 77, 2875–2887. Karlsson, K.-G., 1997: An introduction to remote sensing in meteorology. SMHI, Norrköping, 315 pp. Kidder, S. Q. and T. H. Vonder Haar, 1995: Satellite meteorology. An Introduction. Academic Press, San Diego, 466 pp. Lensky, I. M. and D. Rosenfeld, 1997: Estimation of precipitation area and rain intensity based on the microphisical properties retrieved from NOAA AVHRR data. J. Appl. Meteor., 36, 234–242. Lensky, I. M. and D. Rosenfeld, 2003: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds at night-time. J. Appl. Meteor., 42, 1227–1233. Lensky, I. M. and D. Rosenfeld, 2003: A night rain delineation algorithm for infrared satellite data. J. Appl. Meteor., 42, 1218–1226. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. McCollum, J. A., A. Gruber, and M. Ba, 2000: Comparison of monthly mean satellite estimates of precipitation with gauges over Africa. J. Appl. Meteor., 39, 666–679. McCollum, J. A., W. F. Krajewski, R. R. Ferraro, and M. B. Ba, 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 1065–1080. Nakajima, T. and M. D. King, 1990: Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements. Part I: Theory. J. Atmos. Sci., 47, 1878–1893.

Cloud Top Microphysics as a Tool for Precipitation Measurements

77

Petersen, W. A. and S. A. Rutledge, 1998: On the relationship between cloud-to-ground lightning and convective rainfall. J. Geophys. Res., 103, 14025–14040. Petersen, W. A. and S. A. Rutledge, 2001: Regional variability in tropical convection: observations from TRMM. J. Climate, 14, 3566–3586. Petty, G. W., 1995: The status of satellite-based rainfall estimation over land. Remote Sens. Environ., 51, 125–137. Rosenfeld, D., 1999: TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105–3108. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287, 1793–1796. Rosenfeld, D. and A. Gagin, 1989: Factors governing the total rainfall yield of continental convective clouds. J. Appl. Meteor., 28, 1015–1030. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the tops of potential rain clouds by multispectral analysis of AVHRR data. Atmos. Res., 34, 259–283. Rosenfeld, D. and I. M. Lensky, 1998: Space borne based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc., 74, 2457–2476. Rosenfeld, D. and C. W. Ulbrich, 2003: Cloud microphysical properties, processes, and rainfall estimation opportunities. Chapter 10 of “Radar and Atmospheric Science: A Collection of Essays in Honor of David Atlas”. Edited by Roger M. Wakimoto and Ramesh Srivastava. Meteorological Monographs, 52, 237–258, AMS. Rosenfeld, D. and W. L. Woodley, 2003: Closing the 50-year circle: From cloud seeding to space and back to climate change through precipitation physics. Chapter 6 of “Cloud Systems, Hurricanes, and the Tropical Rainfall Measuring Mission (TRMM)” edited by Drs. Wei-Kuo Tao and Robert Adler, 234pp., pp. 59–80, Meteorological Monographs, 51, AMS. Rosenfeld, D., E. Cattani, S. Melani, and V. Levizzani, 2004. Considerations on daylight operation of 1.6 µm vs 3.7 µm channel on NOAA and METOP Satellites. Bull. Amer. Meteor. Soc., 85, 873–880. Smith, E. A., G. Asrar, Y. Furuhama, A. Ginati, C. Kummerow, V. Levizzani, A. Mugnai, K. Nakamura, R. F. Adler, V. Casse, M. Cleave, M. Debois, J. Durning, J. Entin, P. Houser, T. Iguchi, R. Kakar, J. Kaye, M. Kojima, D. Lettenmaier, M. Luther, A. Mehta, P. Morel, T. Nakazawa, S. Neeck, K. Okamoto, R. Oki, G. Raju, M. Shepherd, E. Stocker, J. Testud, and E. Wood, 2007: International Global Precipitation Measurement (GPM) program and mission: an overview. In: Measuring precipitation from space. EURAINSAT and the future. V. Levizzani, P. Bauer and J. F. Turk, eds., Springer, 611–654. Stith, J. L., J. E. Dye, A. Bansemer, and A. J. Heymsfield, 2002: Microphysical observations of tropical clouds. J. Appl. Meteor., 41, 97–117. Williams, E., D. Rosenfeld, N. Madden, J. Gerlach, N. Gears, L. Atkinson, N. Dunnemann, G. Frostrom, M. Antonio, B. Biazon, R. Camargo, H. Franca, A. Gomes, M. Lima, R. Machado, S. Manhaes, L. Nachtigall, H. Piva, W. Quintiliano, L. Machado, P. Artaxo, G. Roberts, N. Renno, R. Blakeslee, J. Bailey, D. Boccippio, A. Betts, D. Wolff, B. Roy, J. Halverson, T. Rickenbach, J. Fuentes, and E. Avelino, 2002: Contrasting convective regimes over the Amazon: Implications for cloud electrification. J. Geophys. Res., 107, D20, 8082, doi:10.1029/2001JD000380. Zipser, E. J., 1994: Deep cumulonimbus cloud systems in the tropics with and without lightning. Mon. Wea. Rev., 122, 1837–1851. Zipser, E. J. and M. A. LeMone, 1980: Cumulonimbus vertical velocity events in GATE. Part II: Synthesis and model core structure. J. Atmos. Sci., 37, 2458–2469. Zipser, E. J. and K. Lutz, 1994: The vertical profile of radar reflectivity of convective cells: a strong indicator of storm intensity and lightning probability. Mon Wea. Rev., 122, 1751–1759.

7 THE RETRIEVAL OF CLOUD TOP PROPERTIES USING VIS-IR CHANNELS Elsa Cattani1, Samantha Melani2, Vincenzo Levizzani1, and Maria João Costa3 1

Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy Institute of BioMeteorology, IBIMET-CNR, La.M.M.A. (Laboratory for Meteorology and Environmental Modelling), Florence, Italy 3 Department of Physics and Evora Geophysics Centre, University of Evora, Evora, Portugal 2

1

INTRODUCTION

The remote sensing of the cloud optical and microphysical properties from solar reflection and emission measurements is exceedingly important for an improved understanding of the Earth’s climate system, since clouds are a strong modulator of the shortwave and longwave components of the Earth’s radiation budget (King et al. 1992). Cloud microphysical characterization using multispectral techniques in the visible (VIS), near-infrared (NIR), and infrared (IR) range is an efficient proxy for the detection of precipitating systems. Recent studies have demonstrated that cloud properties can be effectively exploited to identify the genesis and evolution of the cloud mass. Rosenfeld and Lensky (1998) examined the evolution of the effective radius (Re) of convective cloud particles versus cloud top temperature (Tc) to infer information about the efficiency of the precipitation forming processes. They associated five different microphysical stages in the temporal evolution of a cloud with peculiar trends of Tc as a function of Re: diffusional droplet growth, coalescent droplet growth, rainout zone, mixed-phase precipitation, and glaciation. This kind of analysis provides a potential tool to improve satellite rainfall estimation techniques (Rosenfeld and Gutman 1994; Rosenfeld 2007) based on microwave data, especially in case of rain from clouds without large ice particles, whose detection is more difficult over land due to the high surface emissivity. 79 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 79–95. © 2007 Springer.

E. Cattani et al.

80

Sensors on board polar and geostationary platforms have recently entered a new era. The new generation of satellite sensors is characterized by a greater number of spectral channels in the VIS, NIR and IR, a better spatial resolution, and an increased data availability. Noteworthy among these satellites is the MODerate resolution Imaging Spectroradiometer (MODIS) on board the Earth Observing System (EOS) satellites (Barnes et al. 1998) equipped with 36 wavebands whose spatial resolution ranges from 250 m to 1 km. On the geostationary side it is worth mentioning the Spinnig Enhanced Visible and InfraRed Imager (SEVIRI) (Schmetz et al. 2002), the main instrument on board the new European geostationary platform Meteosat Second Generation (MSG) with 12 spectral channels and a spatial resolution of 4.8 km at the sub-satellite point, due to an oversampling factor of 1.6 and a 3 km sampling distance, for all channels except the High Resolution Visible (HRV) channel (1.67 km spatial resolution, with a sampling distance of 1 km at the nadir and an oversampling factor of 1.67). These improved observational capabilities are ideal to enhance our understanding of cloud microphysics through cloud top retrievals of hydrometeor radiative properties. It is unfeasible to discriminate among the wealth of cloud physical properties using data out of a single narrow portion of the electromagnetic spectrum. Moreover, an appropriate spatial resolution is needed to limit the occurrence of partially cloud covered pixels. Finally, the SEVIRI 15 min repeat cycle opens unprecedented scenarios to analyze the temporal evolution of cloud systems. In this work radiative transfer simulations are documented for reviewing the physical concepts behind the retrieval of cloud parameters from satellite sensor data. Radiance data in the VIS, NIR, and IR channels were simulated in the presence of water and ice clouds to estimate the sensitivity of the spectral radiances to the effective radius, cloud optical thickness, cloud top temperature/height, and thermodynamic phase. Uncertainties associated with satellite measurements were considered, and the influence of the solar and viewing geometry, the surface radiative properties, and the atmospheric water vapor amount on the cloud parameter retrieval were investigated.

2

THE RADIATIVE TRANSFER MODEL AND ITS SET UP

Simulated radiances were produced via the Signal Simulator for Cloud Retrieval (SSCR) radiative transfer model (RTM) (Nakajima and Tanaka 1986, 1988; Stamnes et al. 1988). SSCR is a 1D, plane parallel model conceived to compute radiances, transmittances, and plane and spherical albedoes in presence of water and ice clouds at VIS, NIR and IR wavelengths. The RTM can simulate radiance data as measured by a satellite sensor using the response functions of the various channels. The radiative transfer calculations are based on a combined discrete-ordinate/matrix-operator method, with the

The Retrieval of Cloud Top Properties Using VIS-IR Channels

81

delta-M approach for the representation of the phase function and the corrections for the single scattering radiation. The atmosphere is vertically divided in several homogenous sublayers and limited at the ground either by a Lambertian surface, characterized by a user-specified nonspectral albedo, or by an ocean surface, defined by a built-in albedo model that takes into account wind speed. The atmospheric models, which include vertical profiles of temperature, pressure and 28 gaseous species, are deduced from the LOWTRAN7 radiative transfer model (Kneizys et al. 1988). Only the water vapor profile can be modified by the user. The gas absorption is computed by means of a three term k-distribution. SSCR represents clouds as vertically homogeneous layers defined by the following input parameters: thermodynamic phase, Re, size distribution type (choice between three different types of functions, i.e., power law, log-normal, and modified gamma), optical thickness at 0.5 µm (τ), and top and bottom heights. The vertical superimposition of different cloud layers is not allowed and Mie theory is applied to cloud particle scattering considering only spherical ice particles. SSCR was used to analyze the sensitivity of the radiance data measured at various satellite sensor channels to the cloud optical and microphysical properties and to explore the possibility of using these spectral data for cloud property retrieval. The simulated data refer to satellite sensor channels centered at 0.6, 1.6, 3.7, 11 and 12 µm. The response functions of MODIS have been used to carry out the computations, but analogous results can be obtained using the response functions of other sensors characterized by similar spectral characteristics as SEVIRI’s. The simulations were performed with the following model setup: • •

• • • •

mid-latitude summer atmospheric profile; variations in the illumination and viewing conditions were simulated using 3 different solar and satellite zenith angle values (θ0 and θ, respectively), 5°, 30° and 70°, and one relative azimuth angle value (Φ), 80°; the surface contribution to the total signal was accounted for different surface types, i.e., sea, vegetation, snow, and desert soil; the cloud optical thickness at 0.5 µm ranged from 1 to 200, to evaluate the radiative behavior of a wide set of cloudy scenarios; a log-normal size distribution was assumed for cloud particles, with Re in the range [1, 40] µm for water clouds and up to 200 µm for ice clouds; the cloud bottom and top heights were fixed at 0.5 and 1.5 km, respectively, for water clouds, and 8.5 and 10 km for ice clouds.

Finally, the results of the sensitivity study were analyzed taking into account the radiometric performances of the MODIS and SEVIRI sensors.

E. Cattani et al.

82

3

SENSITIVITY ANALYSIS

The last few years have witnessed the development of numerous algorithms that exploit multispectral satellite measurements for a global monitoring of cloud microphysical and optical properties, such as particle size and optical thickness, and cloud macrophysical properties, i.e., the cloud height and thermodynamic phase. Most of these methods rely on the synergetic use of various spectral channels and on the combination of the information content of the different channels. Substantial efforts were devoted to retrieve the optical thickness and the effective particle size because they are crucial elements for the accurate determination of cloud radiative bulk properties (single-scattering albedo, phase function, asymmetry parameter, etc.) (King et al. 1992; Yang and Baum 2003), that very often are parameterized in the RTMs as functions of R e (e.g., Key and Schweiger 1998). Among the most widespread techniques for the retrieval of τ and Re, the VIS/NIR bispectral technique dwells on the different dependencies of the reflected solar radiation in the VIS (0.6, 0.8 µm) and NIR (1.6, 2.1 and 3.7 µm) channels on cloud optical thickness and particle size. In the VIS channels the scattering of incident radiation by cloud particles is conservative and thus the single-scattering albedo is about 1 and does not depend on cloud particle size. At NIR wavelengths the single-scattering albedo is less than unity and is affected by cloud particle size. Nakajima and King (1990) and Nakajima and Nakajima (1995) applied the technique to data from the Advanced Very High Resolution Radiometer (AVHRR). Other examples can be found in Arking and Childs (1985) and Han et al. (1994). NIR channels (in particular those centered at 1.6 and 3.7 µm) are often used in conjunction with the thermal IR (11 and 12 µm) channels for the thermodynamic phase detection. An accurate determination of the cloud particle phase is a fundamental prerequisite for the retrieval of Re and τ. Cloud phase is considered an a priori information needed by many Re and τ retrieval methods since the optical properties of liquid and solid particles are distinct. A wrong cloud phase attribution may thus result in a wrong Re and τ retrieval. Several methods for cloud phase determination were developed such as that of King et al. (1992) who use the ratio between the 1.6 and 0.6 µm reflectances. Clouds of different phase but with similar particle size and optical thickness are characterized by similar VIS reflectances and distinct NIR reflectances, due different absorption effectiveness of water with respect to ice (see Fig. 1). For this reason water clouds are expected to exhibit larger values of the reflectance ratio than ice clouds. The brightness temperature differences (BTD) between 11 and 12 µm have been analyzed together with the BTDs between 3.7 and 11 µm and 3.7 µm reflectance, as shown by Key and Intrieri (2000) who proposed an algorithm for the day and

The Retrieval of Cloud Top Properties Using VIS-IR Channels

83

Figure 1. Imaginary part of the refractive index for water and ice with superimposed relevant VIS/NIR/IR channels.

nighttime determination of cloud phase. Much earlier Inoue (1987) used the split window technique for cloud type classification. The analysis of twodimensional histograms of BTDs as a function of the brightness temperature at 11 µm with a selection of appropriate thresholds allowed to identify various cloud types, as cirrus, dense cirrus, cumulonimbus and cumulus. Cloud top height or cloud top temperature are macrophysical cloud characteristics that are very important for the cloud phase detection and cloud type classification. The simplest method to infer them is to use IR radiance data or the equivalent brightness temperature (at 11 µm), eventually corrected for thermal radiation emitted by the surface, and then to derive the cloud top as the height where the brightness temperature matches the temperature profile (Nakajima and Nakajima 1995). A more complex method is the CO2 slicing, that exploits radiance data around the CO2 absorption band at 15 µm. This method proved to be especially effective for detecting thin cirrus clouds that are often missed by simple IR and VIS approaches (Zhang and Menzel 2002).

3.1 VIS channel In this section the results of the sensitivity analysis of 0.6 µm simulated reflectances are presented. The dependence of the reflectances on cloud optical thickness and the influence of surface reflection on the relation between the reflectances and τ will be discussed.

E. Cattani et al.

84

(a)

(b)

Figure 2. VIS reflectances as a function of cloud optical thickness at 0.5 µm for water (a) and ice (b) clouds. Reflectances were computed for solar (θ0) and satellite (θ) zenith angles of 30°, a relative azimuth angle (Φ) of 80°, and a sea surface characterized by albedo ωs = 0.05. The different curves refer to various Re values, varying between 3 and 39 µm (water) and 3 and 200 µm (ice).

In Fig. 2 VIS reflectances are plotted as a function of cloud optical thickness at 0.5 µm for water (Fig. 2a) and ice (Fig. 2b) clouds. Reflectances were computed for θ0 = θ = 30° and Φ = 80°. The simulations were carried out for a sea surface characterized by albedo ωs = 0.05 in order to avoid interference from surface reflection. In each graph the different curves refer to various Re values, ranging from 3 to 39 µm for water clouds and from 3 to 200 µm for ice clouds. VIS reflectances show a marked dependence on the optical thickness, but a very scarce sensitivity to Re. A slight increase of the reflectance with decreasing Re is evident from Fig. 2, but this makes it only possible to distinguish small cloud particles with Re < 6 µm from considerably larger particles with Re > 40 µm. This implies that it is not possible to exploit this dependence for a retrieval of the effective radius. No capabilities to distinguish cloud phase can be attributed to the channel: from an analysis of Fig. 2, water and ice clouds with the same Re and τ values exhibit quite similar reflectances, due to the fact that at VIS wavelengths the scattering is conservative. Considering the radiometric performances of MODIS and SEVIRI and the sensitivity of the VIS reflectances to τ, an estimation of the errors on the retrieved τ values was done. For MODIS it was possible to obtain an estimate of the radiometric error (standard deviation) out of the values of the Uncertainty Index disseminated with Level 1B data sets. The percent uncertainties derived from the index in case of VIS channels are directly applicable to the reflectance data. Using several Uncertainty Indexes for a number of MODIS granules an average percent uncertainty value of ± 2.5% was found. For SEVIRI the official EUMETSAT short-term radiometric error reported by Schmetz et al. (2002), 0.27 at 5.3 W m–2·sr–1·µm–1, was used. From the propagation of errors, neglecting the error associated to the

The Retrieval of Cloud Top Properties Using VIS-IR Channels

85

solar irradiance, and assuming a constant percent error on radiance, the percent error on the reflectance was determined to be equal at ±5%. In Fig. 3 each τ value used in the simulations is displayed with its uncertainty interval (shaded areas) whose extremes were computed as the τ values that correspond to ρ ± ∆ρ, where ρ is a reflectance value at the fixed geometry, Re and τ, and ∆ρ is the error associated to the reflectance. Figure 3 refers to a water cloud with an effective radius of 12 µm and the solar and satellite geometry is the same as in Fig. 2. The τ uncertainty intervals are computed for ∆ρ/ρ = 2.5%·(light gray shaded area) and 5%·(dark gray shaded area). Note that from Fig. 3 the retrieval of cloud optical thickness is not reliable for τ values greater than ≈30–40. For τ = 30 a percent error of about ±16% is found, in case of ∆ρ/ρ = 5%, whereas for ∆ρ/ρ = 2.5% and τ = 40 the percent error is of ±10%. After that the decreasing sensitivity of the reflectance data and the increasing ∆ρ damage the τ retrieval capability. Surface reflection can substantially influence the top of the atmosphere signal and hence the relation between reflectance and τ. In Fig. 4 the VIS reflectance vs τ curves for the same cloudy scenario of Fig. 2 refer to different surface albedo (ωs) values that are taken from the ωs data set of the RTM Streamer (Key and Schweiger 1998). The curve with ωs = 0 is also plotted as a reference.

Figure 3. Uncertainties in the retrieved τ depending on measurement (reflectance) errors for a water cloud with Re = 12 µm. The solar and satellite geometries are the same as in Fig. 2. The light gray shaded area represent the τ uncertainty for a percent error on VIS reflectance of ±2.5%, whereas the dark gray shaded area is relative a percent error on VIS reflectance of ±5%.

E. Cattani et al.

86

reflectance at 0.6 µm

1.0 0.8 0.6

Re = 12 µm water cloud

0.4

ωs = 0 ωs = 0.05 sea ωs = 0.14 veg. ωs = 0.19 sand ωs = 0.83 snow

0.2 0.0 1

10

100

cloud optical thickness at 0.5 µm Figure 4. Simulated reflectances at 0.6 µm as a function of τ and surface albedo ωs under the same conditions as in Fig. 2. The effective particle radius is fixed at Re = 12 µm.

Surface reflection largely contributes to the satellite signal in particular for optical thickness lower than 10–20, having as a consequence an overestimation of the cloud optical thickness. Moreover, highly reflecting surfaces, like snow in Fig. 4, can completely cancel the dependence of the VIS reflectance on τ.

3.2 NIR channels Data in the NIR channels, in particular at 3.7 µm, are widely used for the retrieval of the cloud effective radius. Radiances at these wavelengths depend almost exclusively on Re, especially in the case of thick clouds. The thermodynamic phase is another cloud parameter that can modulate the NIR signal, due to the different absorption properties of water and ice at these wavelengths (see Fig. 1). However, several phenomena may intervene by modifying the relation between the radiance values and Re or the cloud phase, i.e., the surface reflection, the water vapor absorption and, for the waveband at 3.7 µm, the thermal emission. Radiative transfer simulations at 1.6 and 3.7 µm were carried out in order to exemplify the phenomena previously summarized. The behavior of the reflectances as a function of Re for water clouds is shown in Fig. 5a and b for the 1.6 and 3.7 µm channels, respectively. Similarly in Fig. 6a and b are plotted the reflectances at 1.6 and 3.7 µm for ice clouds.

The Retrieval of Cloud Top Properties Using VIS-IR Channels

87

(a)

(b) Figure 5. Reflectances at 1.6 (a) and 3.7 µm (b) for water clouds as a function of the cloud effective radius. The different curves refer to various VIS optical thickness values.

The various curves refer to different cloud optical thickness values at 0.5 µm. Reflectances shown in Fig. 5 and 6 represent only the cloud signal without any other contributions, so they were computed in dry atmospheric conditions and for a surface with ωs = 0. Moreover, the thermal emitted component at 3.7 µm was omitted. The reflectances at both NIR channels are very sensitive to the cloud particle effective radius: in particular the reflectances decrease with increasing Re, due to the increase of the absorption by cloud particles, which is more efficient at 3.7 µm than at 1.6 µm for both ice and water clouds. A slight dependence of the NIR reflectances on τ can be noted, which is

E. Cattani et al.

88

a)

b) Figure 6. Same as in Fig.5 but for ice clouds.

drastically reduced in case of Re > 3–4 µm and thicker clouds (τ > 20 at 1.6 µm and τ > 10 at 3.7 µm). These conditions are necessary for the simultaneous retrieval of τ and Re, exploiting the combined use of NIR and VIS reflectances. Only in this case the sensitivity of the VIS and NIR reflectances to τ and Re is nearly orthogonal. This means that τ and Re can be determined nearly independently and thus measurement errors in one channel have little impact on the cloud property determined primarily by the other channel (Nakajima and King 1990). Additionally, for Re < 3–4 µm and thin clouds multiple solutions of τ and Re are possible, as stated by Nakajima and King (1990) and Nakajima and Nakajima (1985). By comparing Fig. 5a with Fig. 6a and Fig. 5b with Fig. 6b, the sensitivity of the NIR reflectances to the cloud thermodynamic phase can be

The Retrieval of Cloud Top Properties Using VIS-IR Channels

89

understood, being due to the different absorption properties of water and ice clouds in both spectral channels (see also Fig. 1). As in the previous section for τ, the uncertainty in the retrieved Re depending on the reflectance errors was evaluated for both 1.6 µm and 3.7 µm channels. The percent errors on reflectance at 1.6 µm for MODIS and SEVIRI were determined in the same way as for the VIS channel: from the Uncertainty Index values in percent error of about ±5% was found for MODIS, whereas a percent error of ±10% was computed from the short-term radiometric error for SEVIRI. The determination of the percent error on reflectance for the 3.7 µm channel is more complex because it is very difficult to weigh the error due to the extraction of the thermal emitted component from the total signal. This error has to be combined with the one from the total signal to obtain the uncertainty in the solar reflected component. Therefore, for this channel the uncertainty in the retrieved Re was evaluated for increasing values of the percent error on the reflectance, from 5 to 30%, in order to fix a maximum limit in the reflectance error beyond which the Re retrieval is surely unreliable. Figure 7 shows the uncertainty in the retrieved

a)

b)

c)

d)

Figure 7. Uncertainty in the retrieved Re as a function of the reflectance errors for water and ice clouds at 1.6 µm (a and b) and 3.7 µm (c and d). The different gray shaded areas are relative to different reflectance error values.

90

E. Cattani et al.

effective radius as a function of the percent error on reflectances at 1.6 µm (Fig. 7a and b) and 3.7 µm (Fig. 7c and d) for water and ice clouds. The uncertainty intervals associated to each Re value are represented by shaded areas (different gray shades are relative to various percent errors on the reflectances) and were computed using the NIR reflectances displayed previously in Figs. 5 and 6 for τ = 200, in order to eliminate the dependence of the reflectance data on cloud optical thickness. From Fig. 7 it comes out that the effective radius retrieval is not reliable in case of Re < 4 µm for both spectral channels and cloud phases. These Re values are characterized by percent errors of up to 50% and uncertainty intervals that overlap, showing the very scarce sensitivity of the reflectance to such small Re values. For water clouds at 1.6 µm (Fig. 7a) the Re retrieval proves reliable in case of a percent error on reflectances of 5%. In this case the errors on the effective radii are of about 10%, especially for Re > 10 µm. While increasing the error on the reflectance data up to 10% a deterioration in the capability to retrieve Re arises with Re errors >20% thus preventing to appreciate the Re variations. At 3.7 µm (Fig. 7c) the errors on the retrieved Re are 70 µm the uncertainty intervals associated to each Re value overlap because of the weak dependence of the reflectance on the effective radius (see Fig. 6a), and the retrieval loses its reliability. The Re retrieval is completely compromised by a reflectance error of 10%. The Re retrieval for ice clouds at 3.7 µm (Fig. 6d) is effective if ∆ρ/ρ ≤ 15%. With increasing ∆ρ the uncertainties on the retrieved Re values are such that it is not possible to distinguish different Re values. Moreover, the retrieval is not efficient for cloud particles with R e > 45–50 µm, due to the very low reflectance values and the almost complete absence of sensitivity of the reflectance on Re (see Fig. 6b). NIR radiance data can be affected by different spurious signals that can modify the sensitivity to the cloud parameters. Figure 8 shows the effects of water vapor absorption, surface reflection and thermal emission on the cloud signal at 3.7 µm and for water clouds with τ = 5. A high reflecting surface (ωs = 0.370 at 3.7 µm) increases the total signal, particularly in case of small Re values where the cloud absorption is weaker. The water vapor above cloud top is responsible for the absorption of radiation coming out of the cloud top. This phenomenon is more effective the lower the cloud top because of the greater amount of water vapor, and the greater θ and θ0. 3.7 µm radiances are more affected by the water vapor absorption than the 1.6 µm ones. The thermal emission, which is present only at 3.7 µm, produces an additive signal that is scarcely dependent on Re, as can be noted from Fig. 8 by comparing the curves with empty triangles and squares. For this reason several retrieval methods subtract this component from the total signal and use only the solar reflected component (Nakajima and Nakajima 1995; King et al. 1997).

3.3 IR channels Brightness temperatures (BT) at 11 and 12 µm were simulated for ice and water clouds in order to determine their sensitivity to the principal cloud parameters, Re, τ and the thermodynamic phase, and to evaluate the influence, which the water vapor can exert. In Fig. 9 the BTDs between 11 and 12 µm are plotted as a function of the brightness temperatures at 11 µm (BT11) for a water cloud. Two values of the effective radius, 5 µm (circles) and 15 µm (triangles), and 18 τ values ranging from 0 to 200 have been considered. Cloud top and bottom heights were fixed at 1.5 and 0.5 km, respectively. The behavior of the BTD and BT11 was evaluated (1) in absence of water vapor below and above the cloud layer (solid line), (2) with water vapor below the cloud layer (dashed line), and (3) with water vapor below and above the cloud (dotted line). For these simulations the standard water vapor

92

E. Cattani et al.

Figure 9. Brightness temperature differences as a function of the brightness temperature at 11 µm for a water cloud. The different symbols refer to R evalues used in the simulations, 5 µm (circles) and 15 µm (triangles). The simulations were carried out for dry atmospheric conditions (solid line), in presence of water vapor below the cloud layer (dashed line) and with water vapor above and below the cloud (dotted line). The different τ values used are specified in the figure.

amounts relative to the Midlatitude Summer profile were used. The surface was considered as a black body with a temperature of 294.20 K. Figure 10 is the same as Fig. 9 but for an ice cloud with top and bottom heights at 10 and 8.5 km, respectively. The brightness temperatures simulated in dry atmospheric conditions for both water and ice clouds are very sensitive to cloud optical thickness variations: BT11 decreases with increasing τ, ranging from the value of surface temperature for τ = 0 to the cloud top temperature for high τ values, as the cloud absorbs increasing amount of surface emitted radiation. For τ > 20 the cloud is sufficiently thick to be considered a black body and thus BT11 does not vary anymore. Less marked is the sensitivity of BT11 to Re: the greater the effective radius values the lower the BT11, due to the greater absorption of ice or water cloud particles. As for the BTD their dependence on τ can not be used in a retrieval of cloud optical thickness, due to the fact that the same BTD value can correspond to very different τ values, whereas the high sensitivity to Re could be exploited for the retrieval. Note that cloudy scenarios with 1 < BTD < 5 K can be equally associated to water or ice clouds (Giraud et al. 2001) as can be seen from Figs. 9 and 10. Therefore

The Retrieval of Cloud Top Properties Using VIS-IR Channels

93

Figure 10. Same as in Fig. 9 but for an ice cloud.

it could be very difficult to give a correct interpretation of BTD signatures without phase information. The introduction of water vapor below cloud bottom affects BT11 and BTD only in case of thin cloud layers (τ < 2). Water vapor behaves as an absorber of the surface emitted radiation and hence BT11 decreases with respect to the values assumed in dry conditions, whereas BTD increases. In general this phenomenon can be neglected for water clouds as the majority of them have optical thickness greater than 2. Only water clouds are affected by the presence of water vapor above cloud top because the water vapor profiles have their maximum at low altitudes. The effect can be noted at every τ value since in this case the water vapor absorbs the radiation coming out of the cloud top.

4

CONCLUSIONS

Radiative transfer simulations were used to evaluate the sensitivity of radiance data at VIS, NIR and IR satellite sensor channels to the principal cloud parameters, and to explore the possibility of exploiting it in retrieval procedures. Reflectance data at 0.6 µm proved to be the most suitable for the retrieval of the cloud optical thickness. The retrieval is feasible for τ values up to 30–40, with relative errors not greater than 10% in case of errors on the reflectance ∆ρ/ρ ≤ 5%. The surface reflection and the slight dependence of the reflectance on Re have to be taken into account for a correct τ retrieval.

E. Cattani et al.

94

The Re dependence of the VIS reflectance can introduce significant errors if not accounted for during the τ retrieval. For this reason VIS channels are often used in conjunction with NIR channels in iterative retrieval procedures, where the Re estimates from NIR wavebands are exploited as first guess for the τ retrieval and vice versa, until convergence is achieved. The Re retrieval can be carried out using radiances at 1.6 or 3.7 µm, according to the different τ values, cloud phase, surface reflection and water vapor amount. 3.7 µm radiances can accurately provide Re for much thinner clouds (τ > 10) than 1.6 µm radiances (τ > 20). Actually these τ values together with Re ≥ 3–4 µm are requested to have reflectances that don’t depend on the cloud optical thickness and to prevent τ and Re multiple solutions. In case of water clouds the Re retrieval is more reliable at 3.7 µm, with errors on the retrieved Re lower than 10% for reflectance errors ≤ 15%, than at 1.6 µm, where Re errors of about 10% can be obtained only for reflectance errors not greater than 5% and Re > 10 µm. The performances of the 1.6 µm channel improve in case of ice clouds, with Re error values ≤ 10% for ∆ρ/ρ = 5%, whereas those of the 3.7 µm reflectance data are quite similar to the previous ones obtained for water clouds. Moreover the 1.6 µm reflectances are sensitive to much greater Re values than the reflectances at 3.7 µm are (up to ~70 µm for 1.6 µm and Re = 45–50 µm for 3.7 µm). The surface reflection can bring about an underestimation of the retrieved Re value. Generally greater τ values are needed at 1.6 µm with respect to 3.7 µm to neglect this phenomenon. This is due to the fact that water and ice cloud particles absorb much more radiation at 3.7 µm than at 1.6 µm, and numerous surface types are characterized by greater albedo values at 1.6 µm than at 3.7 µm. Also the presence of water vapor above the cloud top has to be taken into account, as it can significantly modify the top of the atmosphere signal, especially at 3.7 µm, for low clouds and zenith views. BTD vs BT11 diagrams proved to be useful for the determination of τ, the cloud top temperature and Re. By examining the diagrams relative to large enough areas, containing clear and cloudy satellite sensor pixel with variable τ values, it is possible to extract some information about these cloud parameter from the interpretation of the typical ‘arch’ signature of the BTD vs BT11 curves. Also at these wavelengths the water vapor absorption represent a spurious effect that has to be taken into account, particularly in case of low clouds.

5

REFERENCES

Arking, A. and J. D. Childs, 1985: Retrieval of cloud cover parameters from multispectral satellite images. J. Climate Appl. Meteor., 24, 322–333. Barnes, W. L., T. S. Pagano, and V. V. Salomonson, 1998: Prelaunch characteristics of the MODerate Resolution Imaging Spectroradiometer (MODIS) on EOS-AM1. IEEE Trans. Geosci. Remote Sens., 36, 1088–1100.

The Retrieval of Cloud Top Properties Using VIS-IR Channels

95

Giraud,V., O. Thouron, J. Riedi, and P. Goloub, 2001: Analysis of direct comparison of cloud top temperature nad infrared split window signature against independent retrievals of cloud thermodynamic phase. Gephys. Res. Lett., 28, 983–986. Han, Q., W. B. Rossow, and A. A. Lacis, 1994: Near-global survey of effective droplet radii in liquid water clouds using ISCCP data. J. Climate, 7, 465–497. Inoue, T., 1987: A cloud classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 3991–4000. Key, J. R. and A. J. Schweiger, 1998: Tools for atmospheric radiative transfer: Streamer and Fluxnet. Computers and Geosciences, 24, 443–451. Key, J. R. and J. M. Intrieri, 2000: Cloud particle determination with the AVHRR. J. Appl. Meteor., 39, 1797–1804. King, M. D., Y. J. Kaufman, W. P. Menzel, and D. Tanré, 1992: Remote sensing of cloud, aerosol, and water vapor properties form the MODerate Resolution Imaging Spectrometer (MODIS). IEEE Trans. Geosci. Remote Sens., 30, 2–27. King, M. D., S.-C. Tsay, S. E. Platnick, M. Wang, and K.-N. Liou, 1997: Cloud retrieval Algorithms for MODIS: optical thickness, effective radius, and thermodynamic phase. ATBD-MOD-05. Kneizys, F. X., E. P. Shettle, L. W. Abreu, J. H. Chetwynd, G. P. Anderson, W. O. Gallery, J. E. A. Selby, and S. A. Clough, 1988: Users Guide to LOWTRAN 7. AFGL-TR-88-0177. Nakajima, T. and M. D. King, 1990: Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements. Part I: Theory. J. Atmos. Sci., 47, 1878–1893. Nakajima, T. Y. and T. Nakajima, 1995: Wide-area determination of cloud microphysical properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J. Atmos. Sci., 52, 4043–4059. Nakajima, T. and M. Tanaka, 1986: Matrix formulation for the transfer of solar radiation in a plane-parallel scattering atmosphere. J. Quant. Spectrosc. Radiat. Transfer, 35, 13–21. Nakajima, T. and M. Tanaka, 1988: Algorithms for radiative intensity calculations in moderately thick atmospheres using a truncation approximation. J. Quant. Spectrosc. Radiat. Transfer, 40, 51–69. Rosenfeld, D., 2007: Cloud top microphysics as a tool for precipitation measurements. In: Measuring precipitation from space – EURAINSAT and the future. V. Levizzani, P. Bauer, and F. J. Turk, eds, Springer, 61–78. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the top of potential rain clouds by multispectral analysis of AVHRR data. Atmos. Res., 34, 259–283. Rosenfeld, D. and I. M. Lensky, 1998: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc., 79, 2457–2476. Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977–992. Stamnes, K., S.-C. Tsay, W. Wiscombe, and K. Jayaweera, 1988: Numerically stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media. Appl. Opt., 27, 2502–2509. Yang, P. and B. A. Baum, 2003: Satellite remote sensing/Cloud properties. Vol. 5, Encyclopedia of Atmospheric Sciences, ed. J. R. Holton, J .A. Curry and J. A. Pyle, Academic Press, 1956–1965. Zhang, H. and W. P. Menzel, 2002: Improvement in thin cirrus retrievals using an emissivityadjusted CO2 slicing algorithm. J. Geophys. Res., 107, 4327–4340.

8 CLOUD MICROPHYSICAL PROPERTIES RETRIEVAL DURING INTENSE BIOMASS BURNING EVENTS OVER AFRICA AND PORTUGAL Maria João Costa1, Elsa Cattani2, Vincenzo Levizzani2, and Ana Maria Silva1 1

Department of Physics and Evora Geophysics Centre, University of Evora, Evora, Portugal Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy

2

1

INTRODUCTION

Clouds are a major driving force behind climate mechanisms. They strongly modulate the energy balance of the Earth through absorption and scattering of solar radiation and absorption and emission of terrestrial radiation, and on the other hand, clouds and precipitation are the regulating factors of the hydrologic cycle. Although the importance of clouds is widely recognised, their impact is associated with great uncertainties due to the complexity and space-time variation of the cloud phenomena, therefore the global monitoring of their optical and microphysical properties retrieved from multispectral satellite sensor data becomes a main task/necessity. A great number of studies were conducted on the possible modification of cloud properties through the interaction with atmospheric aerosol particles, as this may lead to important changes of the Earth’s climate. On the one hand, aerosol particles acting as cloud condensation nuclei may be responsible for direct modifications of the cloud properties (Albrecht 1989; Bréon et al. 2002; Kawamoto and Nakajima 2003) – first indirect effect. It consists of a decrease of droplet size due to an increase in droplet concentration (assuming a constant liquid water content), when a cloud is polluted with anthropogenic aerosol particles serving as additional cloud condensation nuclei (Twomey 1974). On the other hand, due to the decrease of cloud particle size, aerosol particles may indirectly interfere with cloud lifetime and precipitation efficiency, producing the second indirect effect, 97 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 97–111. © 2007 Springer.

98

M. J. Costa et al.

which is characterized by the diminution of the efficiency of precipitation forming processes, tending to increase the liquid water content, cloud lifetime and thickness (Rosenfeld 1999, 2000). Both effects may interfere with the planetary albedo. The biomass burning aerosol is considered one of the main responsible of this kind of cloud properties modification. In addition, this type of aerosols contains organic compounds and black carbon, the latter being a strong absorber of solar radiation that can greatly impact cloud formation and evaporation (Ackerman et al. 2000). Many regions of the Earth are characterized by the emission of biomass burning aerosol connected to the agricultural practises and forest fires, but the quantification of the effects of the aerosol-cloud interaction is far from being completely achieved. Satellite measurements provide indispensable data for effective global observations of clouds properties. However, satellite retrievals are in general less accurate than in situ observations, which are an essential tool for comparisons with satellite-derived parameters, constituting the quality control of global satellite products. The development of the present methodology is motivated by the existence of a new generation of geostationary (GEO) satellite measurements such as those from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) flying on Meteosat Second Generation (MSG) (Schmetz et al. 2002). This innovative sensor opens new perspectives with respect to past and present GEO systems since it provides the necessary additional spectral measurements, supplied until now exclusively by low Earth orbit (LEO) satellite sensors. The doubled sampling frequency and improved spatial resolution prompts for global monitoring of cloud properties and eventually advances in cloud-aerosol interaction studies, facilitating the task of comparing the derived cloud properties with in situ measurements as well. Data from MSG-1 SEVIRI was not yet distributed on a regular basis at the moment the present study was carried out and therefore measurements from the MODerate resolution Imaging Spectroradiometer (MODIS) (Barnes et al. 1998) onboard Terra and Aqua LEO satellites were used, which present comparable spectral channels in the visible (VIS), thermal infrared (IR), and near IR (NIR). The methodology for the characterization of the cloud properties is based on satellite multi-spectral measurements used in combination with radiative transfer calculations to retrieve the cloud optical thickness (COT), particle effective radius (ER) and cloud top temperature (CTT). The retrieval procedure is applied to strong aerosol events from intense biomass burning aerosol transports that occurred in Southern Africa and Portugal in summer 2000 and 2003, respectively, in order to investigate possible alterations of the cloud properties. Comparisons between the retrieved parameters and MODIS cloud products have been carried out. Moreover, for the case study over Southern Africa comparisons with in situ measurements of the Southern African Fire-Atmosphere Research Initiative 2000 (SAFARI 2000)

Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events

99

(http://mercury.ornl.gov/safari2k/) field campaign are presented, in the attempt to improve the understanding of the interactions between clouds and aerosol.

2

METHODOLOGY

The first step of the methodology consists of the cloud detection over the area selected for the study, as well as the particle phase determination (liquid water or ice), assuming that clouds at one time are made of either liquid water or ice particles, hence no mixed phase clouds are considered in the study. The pixel classification procedure relies on a bi-spectral technique that uses MODIS measurements in the VIS and IR channels, centred at 0.65 and 11 µm respectively. The satellite measurements are initially classified in terms of the underlying surface (land or water) using a land-sea mask. Subsequently, the histograms of the VIS radiance measurements and IR brightness temperature values are analysed to determine threshold values that define the limits between clear sky, water clouds and ice clouds. Such threshold classification is done at the pixel level and when the pixel is cloudy, four possible cases are distinguished: water clouds over the ocean, ice clouds over the ocean, water clouds over land, and ice clouds over land (see Table 2). The VIS, NIR (centred at 3.75 µm) and IR radiance measurements corresponding to the pixels classified in the four categories are used to retrieve COT, ER and CTT using the algorithm proposed by Nakajima and Nakajima (1995) and Kawamoto et al. (2001). The four categories are treated separately because relevant differences in cloud and surface characterization must be taken into account. The algorithm relies on the comparison between the modelled cloud radiances in the three spectral bands and the corresponding satellite radiance measurements, deprived of the undesirable components, such as the solar radiation reflected by the surface and the thermal radiation emitted from the cloud layer and the surface, in order to obtain only the cloud signal. These corrections are based on the use of LookUp Tables (LUTs) calculated using the radiative transfer code (RTC) RSTAR (Nakajima and Tanaka 1986, 1988). The LUTs contain the radiative quantities necessary for the cloud properties retrieval, namely the cloud reflected radiances and spherical albedo in the VIS and NIR, the transmission in the VIS, NIR and IR and the reflection and atmospheric emitted radiation in the NIR and IR spectral bands. The RTC calculations are done taking into account the MODIS spectral response functions for each of the three spectral channels. The LUTs are built for a grid of selected values of the COT, ER, CTT, equivalent water vapour above the cloud (ewvu), equivalent water vapour of the cloud layer (ewvc), solar zenith (θ0), satellite zenith (θ ) and relative azimuth (Φ) angles, as shown in Table 1. The cloud is characterised by a lognormal size distribution and mean values of surface temperature and

M. J. Costa et al.

100

reflectance, as well as standard atmospheric vertical profiles (McClatchey et al. 1971), are provided according to the actual conditions (see Table 3). Besides the COT, ER and CTT, the cloud top height (CTH), and pressure (CTP) are also retrieved, from the CTT values by linear interpolation of the selected atmospheric vertical profile. The code was modified to yield also the cloud type as output. In its original version, the cloud type or the cloud geometrical thickness had to be initially defined. In this modified version, the cloud type is established from the COT and CTP values, using the ISCCP cloud classification (Rossow and Schiffer 1999). Table 1. Grid of values used to build the LookUp Tables.

Grid parameter COT ER (µm) CTT (K) ewvu (g m–2) ewvc (g m–2) θ0 (º) θ (º) φ (º)

Grid values Cloud phase Liquid water

Ice

1, 2, 4, 6, 9, 14, 20, 30, 50, 70

0.1, 0.5, 1, 2, 4, 8, 16, 32, 48, 64 5, 10, 20, 40, 60, 80, 100, 120, 140, 160, 2, 4, 6, 9, 12, 15, 20, 25, 30, 35, 40 180 250, 260, 270, 280 220, 230, 240, 250, 260, 270 50, 5000, 10000, 20000, 30000, 40000, 50000 50, 5000, 10000, 20000, 30000, 40000, 50000 0, 5, 10, 20, 30, 35, 40, 45, 50, 55, 60, 65, 70 0, 5, 10, 20, 30, 35, 40, 45, 50, 55, 60 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180

Comparisons between the retrieval results and time – space collocated MODIS official cloud product data sets (King et al. 1998) of COT, ER and CTT were carried out. The idea was to validate the present retrievals against a state of the art retrieval algorithm, in particular as regards to the setup of the algorithm and the selection and classification of cloudy pixel. In situ measurements taken during the intensive SAFARI 2000 campaign in Southern Africa (Swap et al. 2003), conducted from 13 August to 25 September 2000, were used as a further source of comparison data for the African case study. The datasets used are the cloud effective radius measurements made by the UK Met Office C-130 aircraft (Keil and Haywood 2003), and the cloud and aerosol layers bottom and top heights as a part of the cloud and aerosol measurements from the ER-2 Cloud Physics Lidar (CPL) (McGill et al. 2003).

3

CASE STUDIES

The methodology described in the previous section is applied to two selected episodes where marine stratocumulus and convective clouds are observed in the presence of intense biomass burning events, one in the southern hemisphere and the other in the northern hemisphere.

Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events

101

Figure 1. Absorbing aerosol particles detected by the Earth Probe TOMS. The grayish spots locate the absorbing aerosol. The white boxes delimit the location of the MODIS granules used for the cloud top properties retrieval. Images available online at http://toms.gsfc. nasa.gov/ aerosols/aerosols.html.

The first case study concerns the intense fire season that occurred in Southern Africa during summer 2000 with a peak in late August and early September. The region is subject to one of the most frequent occurrences of biomass burning in the world. The heaviest burning was in western Zambia, southern Angola, northern Namibia, and northern Botswana. Yet, the smoke from these fires may be transported over substantial distances during several days or even weeks being detected some thousands of kilometres away from the fire sources. The present analysis focused on the African coasts between Southern Angola and Northern Namibia. This oceanic area is normally characterized by the presence of semipermanent stratiform clouds, so it provides a good opportunity to increase the understanding of cloud microphysics and cloud-aerosol interactions. The aerosol index maps from the Total Ozone Mapping Spectrometer (TOMS) onboard the Earth Probe (Torres et al. 1998) presented in Fig. 1 for the 5, 11 and 13 September 2000 reveal the presence of absorbing aerosol particles as smoke. The greyish spots over the dominating background colour delimit the areas where the absorbing particles are present. The white boxes delimit the approximate location of the MODIS-Terra granules used in each of the days. 5 September 2000 is representative of background conditions since the aerosol event can be barely noted on the upper right part of the white box, which corresponds to a cloud-free area. On 11 September aerosol and background conditions co-exist since the aerosol event does not concern the lower part of the granule, but only the upper part, where clouds are also detected. A vast aerosol mass extends over the ocean in a cloudy area on 13 September. The second case study focuses on a smoke transport event originated from the numerous uncontrolled fires burning across continental Portugal during August 2003, which turned out to be the worst fire season that Portugal faced in the last 23 years if not ever. The fires continued to spread during several days taking advantage of the hot, windy and dry conditions all over

M. J. Costa et al.

102

Figure 2. Same as Fig. 1 but for the Portuguese fires.

the country. It is estimated that an area of about 5.6% of the entire Portuguese forest area burnt until 20 August 2003. Nevertheless, the fires continued at least until mid September, contributing to aggravate the scenario. The TOMS images presented in Fig. 2 illustrate the regions with absorbing aerosol particles, coming from the fires burning across Portugal and Spain. Clouds are detected on both days in the areas affected by the absorbing aerosol plumes. The white boxes indicate once again the geographical location of the MODIS-Terra (4 August) and MODIS-Aqua (5 August) granules used for the retrieval.

4

RESULTS OF THE RETRIEVAL PROCEDURE

The threshold determined from the bi-spectral analysis in terms of the brightness temperature (TB) and of the VIS radiance (IVIS), are summarised in Table 2. In Table 3, the vertical profiles, as well as the surface reflectance and temperature used, are indicated. One of the apparent effects out of the cloud-aerosol interaction is the reduction of the cloud particle size due to the increase of cloud condensation nuclei that induce a redistribution of the cloud water content. The analysis was therefore concentrated on the retrieved ER values of water clouds to detect any cloud modification due to the aerosol effect. Figure 3 shows the frequency histograms of the ER (for the water clouds over the ocean) corresponding to the three days analysed. The classification of the MODIS pixels for 5 September 2000 allowed for distinguishing water clouds located mainly over the ocean, but not in the area where the absorbing aerosols are detected (Fig. 1a). Therefore, we are probably in this case dealing with clouds in the absence of any absorbing aerosol from the biomass fires. The frequency histogram (Fig. 3a) presents a peak of the ER around 11 µm with a low spread of the values. In the 11 September case, the water clouds detected by the classification cover most of the oceanic area of the granule, however only the upper part of the granule corresponds to the coexistence of the absorbing aerosols and water clouds, as shown in Fig. 1b. The histogram of the 11 September shows a bimodal distribution with a peak of the ER towards 14 µm and a smaller peak around 6 µm. This second peak can be connected to the presence of

Cloud Microphysical Properties Retrieval During Intense Biomass Burning Events 05/09/2000 1015UTC

11/09/2000 0940UTC

(a)

103

3/09/2000 0925UTC

(b)

(c)

Figure 3. Histograms of the ER obtained for the African biomass burning event. They refer to water clouds detected over ocean. Table 2. Threshold values obtained from the bi-spectral cloud classification.

African fires Iberian Peninsula fires

Ice clouds

Water clouds Over ocean Over land

Case study

260K1.5 [km] means shallow rain “certain”. Over land, however, the judgment is always shallow rain “possible” no matter how low is the height Hstorm because Hfreeze may not be trustworthy over land. When the region of shallow rain is isolated from the other non-shallow rain certain areas, this shallow rain is called as shallow isolated. Shallow nonisolated is defined as the shallow rain which is not shallow isolated.

2.3 H-method The H-method also classifies rain into three categories: stratiform, convective, and other, but with the definitions of these being different from those by the V-method. The H-method is based on the University of Washington convective/stratiform separation method (Steiner et al. 1995), which examines the horizontal pattern of Zm at a given height; this original horizontal pattern method is applicable to the data with a 2 km horizontal resolution. Since the horizontal resolution of the TRMM PR is about 4.3 km, the original horizontal pattern method is not readily applicable to the TRMM PR data. Besides, the TRMM PR observes precipitation not only over low flat

Rain Type Classification Algorithm

219

areas but also over high mountain areas; in the latter case, examining the horizontal pattern of Zm at a given height would become impossible sometimes. To enable the horizontal pattern method applicable to the TRMM PR data, the following modifications are made: 1. Instead of examining a horizontal pattern of Zm at a given height, a horizontal pattern of Zmmax(in R) is examined; here Zmmax(in R) is the maximum of Zm in the rain region, R, which is defined as the clutter free region below Hfreeze (minus 1 km margin). The quantity Zmmax(in R) is obtained for each angle bin. 2. Parameters are changed so that they are suitable to the TRMM data with a 4.3 km horizontal resolution. Choice of parameters was made before the launch of the TRMM satellite using a ground radar data in such a way that the result with a simulated 4.3 km horizontal resolution data produces almost the same result as that with a 2 km horizontal resolution data. 3. Third category of other type is also introduced to handle the noise case. In the H-method, detection of convective rain is made first. If Zmmax(in R) exceeds 40 dBZ, or Zmmax(in R) stands out against the background area, this pixel is regarded as a convective center. (Here, we use the term “pixel” for identifying the location of Zmmax(in R) in a horizontal plane.) Rain type for a convective center is convective, and rain type for the (four) pixels nearest to the convective center is also convective. If rain type is not convective and if the rain echo is certain to exist, rain type is stratiform. Rain type by the H-method is other if the radar echoes below Hfreeze (minus 1 km margin) at a given angle bin are very weak so that the echo there may possibly be noise. This means that the other type by the H-method consists of (i) cloud only case and (ii) noise only case.

2.4 Unification of rain types Since the algorithm 2A23 uses two independent methods for classifying rain types, it would not be friendly to the users if 2A23 outputs the rain types by the two methods separately. To make the result user-friendly, 2A23 outputs the unified rain type. In version 6 of 2A23, the unified rain type is expressed by a 3-digit number (while in the previous versions, the unified rain type is expressed by a 2-digit number): the first digit indicates the main category of the type (1: stratiform, 2: convective, 3: other), and the remaining second and third digits distinguish the sub-category. Let the rain type for the ith angle bin be rainType[i]. Then the main category of the unified rain type is obtained by rainType[i]/100. The main category of unified rain type would satisfy the need of most users.

220

J. Awaka et al.

It should be mentioned that the unification of rain types is made in such a way that the rain type by the V-method and that by the H-method can be reconstructed from the unified rain type by using a suitable look-up table. One of the most surprising findings obtained immediately after the launch of the TRMM satellite is on ubiquitous shallow isolated rain. Shallow isolated rain is observed in a dot-like manner, which implies that shallow isolated rain may not be stratiform which is characterized by a rather continuous wide rain area. But the strength of the radar echo from shallow isolated rain is usually weak. Hence, regarding the shallow isolated rain as convective may contradict with our usual notion that the convective rain should be strong. In version 5, shallow isolated rain is classified as stratiform, convective, or other depending on the strength of radar echo, and the most of shallow isolated rain is classified as stratiform because of weak radar echo.

Figure 3. Angle bin (i.e., antenna scan angle) dependence of the counts of three main rain types (left panel), and that of detected BB (right panel).

In version 6 of 2A23, all the shallow isolated rain is classified as convective because shallow isolated rain has convective characteristics (Schumacher and Houze 2003), though the radar echo of shallow isolated rain is usually weak. Note that shallow rain consists of shallow isolated and shallow non-isolated rain (see Section 2.2.2). Most of the shallow non-isolated rain is classified as stratiform because its radar echo is usually weak.

3 SOME STATISTICAL RESULTS Figure 3 shows angle bin dependence of the count of the following: stratiform, convective, and other types on the left panel, and detected BB on the right panel. The angle bin number 25 in the abscissa corresponds to the nadir direction, and the angle bin numbers 1 and 49 to the scan edges, where the

Rain Type Classification Algorithm

221

antenna scan angle, i.e., the off nadir angle, is about ±17°. The figure is based on one month statistics of data in February 1998; the total data for both over land and ocean are used. The left panel of Fig. 3 indicates that among the three main categories of rain type, the most populous is stratiform, and the least is other. The count of “other” type in version 6 of 2A23 becomes very small when compared with that in version 5 (not shown), because of the changes in the parameters for characterizing “other”. In version 6, “other” means ice cloud only or possibly just noise. Figure 3 indicates that the count of stratiform and that of convective depend on the angle bin, but the count of other is almost independent of the angle bin. The count of detected BB shows a large dependence on the angle bin. In what follows, the reasons for these angle bin dependencies are examined by turns. First on the stratiform: in Fig. 3, the count of stratiform is rather constant near nadir, but the count decreases near scan edges. This occurs because the stratiform precipitation includes a large number of shallow non-isolated; when the storm top is low, the rain echo may be masked by a smeared surface clutter near scan edges. (The surface echo is a clutter to the observation of precipitation.) Since the TRMM PR has a rather small range resolution, ∆R, which is 250 m, and a large size of footprint of about 4.3 km (see Table 1), a sharp surface echo at nadir smears to a large extent near a scan edge as illustrated in Fig. 4.

Figure 4. Shapes of surface echo, at nadir (left) and near scan edge (right).

In Fig. 3, the count of the detected BB shows a sharp decrease near scan edges. This occurs because the shape of BB smears near scan edges due to a mechanism similar to the one illustrated in Fig. 4, and because it becomes very difficult to detect the smeared BB. Figure 5 shows the angle bin dependence of the counts of stratiform under the following conditions: Hstorm > 2, 3, 4, and 5 [km] (thin lines). The figure also shows the angle bin dependence of the unconditional count of stratiform (thick line), which is identical to that of stratiform in Fig. 3. The thin curves in Fig. 5 have almost the same shape with parallel shifts in the vertical

222

J. Awaka et al.

Figure 5. Unconditional stratiform count (thick), and the stratiform counts under the condition that the height of the storm top is greater than 2 km, 3 km, 4 km, and 5 km (thin).

direction. The shape of the thin curve for Hstorm > 2 [km] near scan edges is almost the same as that of the thick curve near scan edges, which indicates that the smearing of the surface echo extends up to about 2 km at scan edges. A closer look at the thin curves in the figure indicates that the count of stratiform slightly decreases near scan edges, which implies that a sensitivity of the TRMM PR may be slightly lower near scan edges. Let us move on to a discussion on the count of convective, which exhibits a dependency on the angle bin as shown in Fig. 3. If the majority of convective precipitation has a tall storm top and a strong precipitation rate, we would expect that the count of convective may be almost independent of the angle bin, because the high storm top is free from the effect of surface clutter, and the strong precipitation rate produce a large radar echo which is not affected by a sensitivity of the TRMM PR. In version 6 of 2A23, however, all the shallow isolated is classified as convective (see Section 2.2.3). Figure 6 separately shows the angle bin dependency of the count of shallow isolated (thick line) and that dependency of the count of convective which excludes the shallow isolated (thin lines). The count of shallow isolated shows angle bin dependence because of the masking effect by the surface clutter. The count of convective which excludes shallow isolated is almost independent of the angle bin, which is a characteristics of tall and strong precipitation. Finally, the count of other type in Fig. 3 is almost constant over the angle bin. This is what we expect because the other type in version 6 of 2A23 consists of ice cloud and noise as described in the early part of this section. The ice cloud may be observed uniformly over the entire angle bins because

Rain Type Classification Algorithm

223

Figure 6. Count of shallow isolated (SI) shows a large dependence on angle bin, i.e., on antenna scan angle (thick line). When the SI count is subtracted from, the remaining convective count becomes almost constant over the angle bin (thin line).

of its high storm top, which is too high to be masked by the surface clutter, and the noise may also appear uniformly over the entire angle bins because of its randomness; thus making the count of other type almost constant against the angle bin. It should be noted in Fig. 3 that at the very edges of angle bins, i.e., at the angle bin number being 1 and 49, the count of stratiform increases a little and the count of convective decreases a little. This phenomenon occurs because the applicability of the H-method is not guaranteed at the very edges of angle bins. (The stand-out condition for the convective precipitation by the H-method requires an averaged Zm in the background area, which cannot be well defined at the very edges of angle bins because Zm is available in about one half of the background area only; the other half of the background area is in the outside of the PR swath and the data there is missing).

4 CONCLUDING REMARKS Though the rain type classification algorithm 2A23 works fine, there are several things that we should be aware of. Among other things, it should be noted that all the shallow isolated is classified as convective in version 6 of 2A23; the strength of most shallow isolated is weak, which is not the characteristics of the ordinary convective precipitation. Detection of BB, which plays a key role in the rain type classification by the V-method, should be improved. A substantial improvement in the detection of BB is anticipated in the future Global Precipitation Measurement (GPM) project, because for GPM a dual frequency radar is planned to be used (Kobayashi and Iguchi 2003); at least the discrimination of a true BB peak from a false BB peak, arising due to a large attenuation effect in the strong convective precipitation, would become relatively easy.

224

J. Awaka et al.

Acknowledgments: Development of 2A23 has been sponsored by Japan Aerospace Exploration Agency, JAXA (former National Space Development Agency, NASDA).

5 REFERENCES Amitai, E., 1999: Relationships between Radar properties at high elevations and surface rain rate: Potential use for spaceborne rainfall Measurements. J. Appl. Meteor., 38, 321–333. Awaka, J., T. Iguchi, and K. Okamoto, 1998: Early results on rain type classification by the Tropical Rainfall Measuring Mission (TRMM) precipitation radar. Proc. 8th URSI Commission Final Open Symp., Aveiro, Portugal, 143–146. Battan, L. J., 1973: Radar Observation of the Atmosphere, University of Chicago Press, Chicago, IL. Fabry, F. and I. Zawadzki, 1995: Long-term radar observations of the melting layer of precipitation and their interpretation. J. Atmos. Sci., 52, 838–851. Klaassen, W., 1988: Radar observations and simulation of the Melting layer of precipitation. J. Atmos. Sci., 45, 3741–3753. Kobayashi, S. and T. Iguchi, 2003: Variable pulse repetition frequency for the Global Precipitation Measurement Project (GPM). IEEE Trans. Geosci. Remote Sens., 41, 1714–1718. Kozu, T., T. Kawanishi, H. Kuroiwa, M. Kojima, K. Oikawa, H. Kumagai, K. Okamoto, M. Okumura, H. Nakatsuka, and K. Nishikawa, 2001: Development of precipitation radar onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. IEEE Trans. Geosci. Remote Sens., 39, 102–116. Kummerow, C., J. Simpson, O. Thiele, W. Barnes, A. T. C. Chang, E. Stocker, R. F. Adler, A. Hou, R. Kakar, F. Wentz, P. Ashcroft, T. Kozu, Y. Hong, K. Okamoto, T. Iguchi, H. Kuroiwa, E. Im, Z. Haddad, G. Huffman, B. Ferrier, W. S. Olson, E. Zipser, E. A. Smith, T. T. Wilheit, G. North, T. Krishnamurti, and K. Nakamura, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39, 1965–1982. Meneghini, R. and T. Kozu, 1990: Spaceborne Weather Radar, Artech House, Boston·London. Schumacher, C. and R. A. Houze, Jr., 2003: The TRMM Precipitation Radar’s view of shallow, isolated rain. J. Appl. Meteor., 42, 1519–1524. Steiner, M., R. A. Houze, Jr., and S. Yuter, 1995: Climatological characterization of threedimensional storm structure from operational radar and rain gauge data. J. Appl. Meteor., 34, 1978–2007.

18 DUAL-WAVELENGTH RADAR ALGORITHM Kenji Nakamura1 and Toshio Iguchi2 1

Hydrospheric Atmospheric Research Center, Nagoya University, Japan National Institute of Information and Communications Technology, Japan

2

1 INTRODUCTION The dual-wavelength precipitation radar (DPR) is one of the key instruments for the GPM core satellite. Precipitation observation using the DPR along with nearly simultaneous observation by a microwave radiometer is essential for accomplishing the GPM mission. The 3-hourly global rain mapping will be achieved by the constellation satellites with microwave radiometers, and the microwave radiometer rain estimate will be improved or tuned by the information provided by the DPR. TRMM achievements have clearly shown that the comparison of the results from radar with microwave radiometer data is very effective to improve the rain retrievals. To achieve the GPM’s global precipitation mapping, the radar onboard the core satellite should have: (a) high sensitivity to detect weak rain and snow, (b) capability to discriminate solid precipitation from liquid one, and (c) better accuracy of rain retrieval than the TRMM PR. To meet the above requirements, the DPR has been designed, and the rain retrieval algorithms are under development. Here, the basic types of DPR algorithms and the generic rain profile retrieval will be described. It should be emphasized that the DPR algorithms are far from matured, and still open to many new ideas. In addition, groundbased or airborne radar experiments are required to validate each algorithm.

2 BASIC SPECIFICATION OF THE DPR For the study of the DPR algorithms, the performance of the DPR should first be specified. The specifications include the sensitivity, the swath, the range resolution, etc. At least the same performance of the TRMM

225 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINAST and the Future, 225–234. © 2007 Springer.

226

K. Nakamura and T. Iguchi

precipitation radar (PR) was required for the lower frequency radar (14-GHz Ku-band radar), which makes the DPR observation a natural extension of the TRMM PR observation. For the higher frequency radar, a 35-GHz channel was selected. Radio waves in this band suffer from much more rain attenuation than the Ku-band radio waves and the rain echoes will show a very different measured profiles from those in the Ku band. The direct combination of the measured profiles at the two bands makes possible the accurate precipitation profile retrievals. One of the big decisions was the nominal range resolution of the Ka-band radar (the higher frequency part of the DPR) to be 500 m, though the TRMM PR and the 14-GHz radar of the DPR has 250-m range resolution. The main reason for it is to attain the sensitivity of nearly 12 dBZ. For the global precipitation observation, significant sensitivity improvement is required for snow observation and for detection of very light rain or highly rainattenuated rain signatures. Sacrificing the range resolution by a factor of two will improve the minimum detectable signal by 6 dB. Though very shallow rain/snow may be missed by the degraded range resolution, such rain/snow is hard to be detected even with 250-m range resolution due to range smearing and ground clutter. The swath of the Ka-band radar was another issue and the current solution is to scan the radar beams across the swath of about a half of the TRMM PR swath, that is, about 100 km. The minimum requirement for the DPR swath was that the largest footprint (a size of about 40 km) of the microwave radiometer onboard the same core satellite should be covered. Another requirement was that the swath should be wide enough to cover the core area of a rain system. TRMM PR data analysis shows that (in tropical area of 120–150E and 10–20N, August 1998) 95% of the core area of precipitation with more than 35 dBZ has a linear dimension of less than 100 km. The current major parameters of the DPR are: Frequency: 13.60 and 35.55 GHz (dual-wavelength). Range resolution: 250 m (14 GHz), 250/500 m (35 GHz). Swath: about 220 km (14 GHz), about 100 km (35 GHz). Horizontal resolution: about 5 km at nadir. Sensitivity: nearly 12 dBZ for 35 GHz. For the DPR algorithms, the beam matching at the two channels is crucial. The DPR is carefully designed to realize sufficient beam matching. This issue is related to not only the sensor design but also calibration techniques for the beam patterns.

3 TYPES OF THE DPR ALGORITHMS The DPR algorithms may fall into several types according to the combination of the sensor data depending on the swath where the algorithms are applied. The algorithm for the narrowest swath may be for the swath of the

Dual-Wavelength Radar Algorithm

227

Ka-band radar, where both Ka- and Ku-band radar data will be available. The second one may be for the Ku-band radar swath, and the third may be for the microwave radiometer swath.

3.1

For the Ka/Ku narrow swath

The first one is the combination of the Ka-band and Ku-band radar data. This combination can be applied in the narrowest swath of the DPR, since both data must be available. This is the primary algorithm of the DPR. Here several algorithms are conceivable. The first one is a Ku-band single wavelength algorithm which is the same as the TRMM PR algorithm (Iguchi et al. 2000) utilizing the Hitschfeld-Bordan solution with iterations and the so-called surface reference technique (SRT) (Meneghini et al. 2000). The second one may be the Ka-band single wavelength method. Though the Ka-band radar suffers from strong attenuation, it may well be used for weak rain cases. In addition, the SRT should work well because of strong attenuation. Thus, an algorithm like the TRMM PR algorithm may be applicable to the Ka-band radar data. The third is the Ka/Ku combined algorithm. This algorithm is a new one and is expected much to improve the precipitation profiling. The fourth may be the combined algorithm using Ka/Ku radar data and microwave radiometer data. The attenuation caused by components other than precipitation particles becomes crucial in the Ka-band. Gaseous and cloud attenuation may be estimated from the microwave radiometer data even though the pixel sizes are different. Algorithms for a dual-frequency radar have been investigated using airborne radars, and are still being developed (Meneghini et al. 1992, 1997; Kozu et al. 1991). The algorithms may fall into several categories. One is a simple dual-wavelength radar algorithm that utilizes the difference of the rain attenuation between two radar reflectivity profiles. The algorithm in this category takes advantage of the fact that the rain rate is less sensitive to the variation of the raindrop size distribution (DSD) than the radar reflectivity. This algorithm is applicable even to two profiles over a limited range interval to retrieve rain rate. One crucial disadvantage is that this technique suffers from the error when the scattering deviates from Rayleigh scattering (in other words, Mie scattering effect). The second is more sophisticated using a two-parameter DSD and it retrieves two parameters at each range bin using two radar reflectivities at Ka and Ku-band radio waves (Mardiana et al. 2004) (see next section). This technique uses a coupled pair of the first order differential equations, and needs initial values. The SRT may be used for the determination of the initial values. Another interesting extension from the TRMM PR is the application of the SRT to the DPR. The use of SRT in the DPR algorithm is expected to improve the accuracy of rain attenuation correction over the ocean, and the

K. Nakamura and T. Iguchi

228

combination of surface echoes at two channels may make the SRT applicable over land. As another application, the SRT has its own beam filling problem (Nakamura 1991), which in turn means that the SRT for the DPR may be used to correct the beam filling errors.

3.2

For the Ku/Ka wide swath

The Ku-band radar is currently designed to have a more than two times wider swath than the Ka-band radar. Inside the Ka-band radar swath, precise precipitation profile measurement can be done. The extension of the precise profile and/or the raindrop size distribution to the wider swath is the core of the algorithm. This part, however, has not yet been developed.

3.3

For the radiometer swath

The combination of PR and radiometer data has been applied in a TRMM algorithm. The basic idea is to use the PR fine resolution data to correct the beam filling bias in the microwave radiometer rain retrieval. One crucial issue in the passive microwave rain retrieval is the height of the liquid precipitation layer (Masunaga et al. 2002; Ikai and Nakamura 2003). The error in the estimate of the liquid precipitation layer directly affects the accuracy of rain retrieval in the so-called emission mode where the column integrated absorption is primarily used. The precipitation profile retrieved from the DPR has a potential to improve the instantaneous rain retrieval using the microwave radiometer. However, these algorithms which use the DPR data to improve directly the microwave radiometer rain retrieval is yet to be well developed.

4 FORMULATION OF THE DPR COMBINED ALGORITHM In this section, we describe a formulation of the DPR profiling algorithm. We derive a coupled pair of differential equations for two parameters in the gamma distribution model for the DSD in this example. However, the formulation can be easily generalized to other types of model. In particular, almost exactly the same formulation can be used if one of the two parameters is proportional to the scale factor (N0 in this example) because we do not use any special property of the gamma distribution when we define Ib(D0) and It(D0) below. In a gamma distribution model, the number density N(D,r) of drops whose diameter is between D and D+dD at range r is given by

N ( D, r )dD = N 0 (r ) D μ exp[−( μ + 3.67) D / D0 (r )]dD

(1)

Dual-Wavelength Radar Algorithm

229

where D0 is the mass-weighted median diameter of the DSD, and μ the shape parameter. The measured radar reflectivity factor Zm(r) is given as the true effective radar reflectivity factor Ze(r) multiplied by the attenuation factor: r

Z m (r ) = Z e (r ) exp[−0.2 ln(10) ∫ k ( s )ds ]

(2)

0

Here, k is the specific attenuation expressed in dB km–1 and is defined in terms of the DSD parameters and the extinction cross section σ t ( D, λ ) of a drop with diameter D for electromagnetic waves with wavelength λ

k (r ) = ck N 0 (r ) I t ( D0 (r ), μ , λ ) = ck N 0 (r ) ∫ σ t ( D, λ ) D μ exp[−( μ + 3.67) D / D0 (r )]dD

(3)

ck = log10 e if It is in m2, N in m-3, and r in m. However, if It is measured in mm2 and r is in km, ck = 10 −3 log10 e . Similarly, the effective radar reflectivity factor Ze(r) is defined as

Z e (r ) = N 0 (r ) I b ( D0 (r ), μ , λ )

(4)

where

I b ( D0 (r ), μ , λ ) = cZ (λ ) ∫ σ b ( D, λ ) D μ exp[−( μ + 3.67) D / D0 (r )]dD

(5)

and 2

cZ (λ ) = λ4 /(π 5 K )

(6)

We assume that the shape factor μ does not change with range. In what follows, we are not going to indicate the dependence of functions Ib and It on μ and λ explicitly. Suppose we have measurements of Zm(r) at two frequencies, f1 and f2. If we use suffixes 1 and 2 for corresponding variables, equation (1) becomes r

Z m1 (r ) = N 0 (r ) I b1 ( D0 (r )) exp[−a ∫ N 0 ( s ) I t1D0 ( s )ds ] 0

(7)

K. Nakamura and T. Iguchi

230 r

Z m 2 (r ) = N 0 (r ) I b 2 ( D0 (r )) exp[−a ∫ N 0 ( s ) I t 2 D0 ( s )ds ]

(8)

0

Here a = 0.2ck ln(10) . Taking the logarithms of these equations and differentiating them with respect to r, we obtain

d d ln N 0 d ln( I b1 ( D0 )) dD0 (r ) ln[ Z m1 (r )] = + − dr dr dD0 dr

(9)

a exp(ln N 0 ) I t1 ( D0 (r )) and

d d ln N 0 d ln( I b 2 ( D0 )) dD0 (r ) ln[Z m 2 (r )] = + − dr dr dD0 dr

(10)

a exp(ln N 0 ) I t 2 ( D0 (r )) Rearrangement of these equations gives

dD0 (r ) 1 = dr b1 − b2 ⎡ d ln(Z m1 ) d ln(Z m 2 ) ⎤ ×⎢ − + aN 0{I t1 ( D0 ) − I t 2 ( D0 )}⎥ dr dr ⎣ ⎦

(11)

and

1 dN 0 (r ) 1 = N 0 dr b1 − b2 d ln( Z m 2 ) ⎡ d ln( Z m1 ) ⎤ × ⎢b2 − b1 + aN 0{b2 I t1 ( D0 ) − b1 I t 2 ( D0 )}⎥ dr dr ⎣ ⎦

(12)

where

d ln( I b1 ( D0 )) dD0 d b2 = ln( I b 2 ( D0 )) dD0

b1 =

(13) (14)

Dual-Wavelength Radar Algorithm

231

These are the equations which D0(r) and N0(r) must satisfy. These equations are numerically solvable and give a set of solutions. Since only the derivatives of ln Z m1 and ln Z m 2 appear in equations (4) and (5), constant offsets in ln Z m1 and ln Z m 2 are irrelevant in possible solutions. Among the possible solutions we need to choose the one that most closely reconstructs the measured profiles, Zm1 and Zm2. This is the case when the radar is well calibrated and when there is no uncertainty of attenuation other than the attenuation by rain itself. If we need to solve the equations in the interval that lies in the middle of the entire observation range with unknown attenuation to the boundary of the interval, we need to know the attenuation to a range that is within the interval. Solving the equations backward for a given pair of the path-integrated attenuations to the surface is such a case. When b1=b2 the right-hand sides of equations (4) and (5) become singular. This situation happens at the point where the quantity Ib1/Ib2 takes its maximum and also at the point D0=0.0. The latter case corresponds to the Rayleigh scattering case in which the particle dimensions are much smaller than the wavelength. In this case, since Ib1=Ib2, the combination of equations (2) and (3) gives

ln Z m1 (r ) − ln Z m 2 (r ) = A1 (r ) − A2 (r )

(15)

where rs

A1 (rs ) = −a ∫ N 0 ( s ) I t1 ( D0 ( s ))ds

(16)

0

rs

A2 (rs ) = −a ∫ N 0 ( s ) I t 2 ( D0 ( s ))ds

(17)

0

This is the case when the difference between the radar reflectivity factors measured at two different frequencies can be attributed solely to the attenuation difference. Note that the non-Rayleigh characteristics appear firstly in It and then Ib as the diameter D0 increases. Therefore, this formulation may be applicable to the combination of some relatively long wavelengths. Note that in this case we cannot separate the effect of N0 from D0 on the attenuation difference. In other words, we can estimate only a single parameter in the DSD model. Therefore, we need to assume some functional relationship between N0 and D0. Nevertheless, in the Ka-band, the rainfall rate R is nearly proportional to the attenuation and its dependence on the DSD parameters is rather small. Therefore, without retrieving N0 and D0

232

K. Nakamura and T. Iguchi

separately, we can still infer the rain distribution (this corresponds to the simple combined algorithm).

5 REMARKS ON THE DPR COMBINED ALGORITHM The objective of the DPR combined algorithm is to retrieve precipitation profiles. The radars are assumed to detect only precipitating particles, since the expected sensitivities are 17 dBZ and 12 dBZ for the Ku and Ka-band radar, respectively. This sensitivity is too poor to detect non-precipitation particles such as cloud particles. Though the precipitation profile consists of liquid particles (rain), solid particles (ice, graupel, etc.) and a mixture of them, in the previous section we restricted our problem only to liquid particles, since the profiling of solid and/or mixed phase particle using the DPR has only a few investigations (Liao et al. 1997). For the liquid particle profiling, the target is to retrieve the raindrop size distribution. Since only two profiles will be obtained from the DPR measurement, a two parameter DSD may be appropriate. We have not only the two profiles but also the surface signatures. As well developed as the SRT in TRMM PR algorithm, the surface signatures can be used to correct the rain attenuation and/or calibrate the radar constants. The surface signatures can be used as additional constraints in the rain estimation or be used as an initial value to solve the equations for profiling. A method similar to the SRT is the use of the mirror image of rain echo over ocean. Since the ocean surface is a good reflector of microwaves, the rain echo over ocean has the mirror images. The ratio of the intensity of the mirror image to the direct echo includes the effect of path attenuation and surface conditions (Meneghini and Atlas 1986; Li and Nakamura 2002). However, a simple use of mirror images in the TRMM PR data has not contributed to the improvement of rain retrieval yet. The mirror image observed by the DPR may help understanding the characteristics of the mirror image and for the exploration of its use. The accuracy of the estimation of the rain profile depends on the range over which the finite differences are applied, since the radar signature fluctuates due to incoherent scattering. For the conventional Z-method, the fluctuation is not a major error source, but for the DPR profiling technique, it becomes a major error source, since the fluctuation is not negligible after taking finite differences. The practical DRP algorithm should combine the low range resolution profiling algorithm and a high resolution singlewavelength profiling algorithm. The Ka-band radar data are not available in strong rain due to severe attenuation. In this case, only a single wavelength algorithm can be applied but may be with surface signatures at two wavelengths. An algorithm that realizes a smooth transition from a dual-frequency mode to a singlefrequency mode should be developed.

Dual-Wavelength Radar Algorithm

233

Another important issue is the beam filling and beam mismatching effect. Due to the finite differences, any error may be emphasized and potentially become a significant error source.

6 CONCLUSIONS The DPR algorithms consist of several categories. The algorithms in each category still need further development. Among them, the Ka/Ku-band combined algorithm is very important and also very interesting, and is one of the keys for the success of the GPM. The basic idea of the combined algorithm is to determine two parameters of the raindrop size distribution at each range bin using the two profiles of measured radar reflectivity from the DPR. The two parameter retrieval seems feasible at least for the liquid precipitation region. After the development of the rain retrieval for the TRMM precipitation radar, radar rain retrieval showed a big progress. Before the TRMM PR, radar rain estimates are thought to be rather qualitative instead of quantitative. The rain retrieval using the ground-based radars are operationally calibrated by available rain gauge networks in Japan. There is a common phrase “distribution is by radar and accuracy is by rain gauges.” The TRMM PR changed this situation at least for the space-borne radar. Before the TRMM launch, attenuated radio waves at 14 GHz adopted in the PR were thought to make the rain retrieval complicated. After the launch, however, the attenuating frequency was found to have a potential to deduce DSD variations by incorporating the SRT. The combined algorithms in the DPR will fully utilize the attenuating characteristics of the radar radio waves. The DSD variation is a big issue even though the combination of the SRT gives us some hints. One of the biggest expectations to the DPR is the global mapping of the DSD variation. There are, however, other precipitations as solid precipitation and mixed phase precipitations. In addition, gaseous and cloud attenuation should also be taken into consideration. Much more studies with simulations and field tests are essential.

7 REFERENCES Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM precipitation radar. J. Appl. Meteor., 39, 2038–2052. Ikai, J. and K. Nakamura, 2003: Comparison of rain rates over the ocean derived from TRMM microwave imager and precipitation radar. J. Atmos. Oceanic Technol., 20, 1709– 1726. Kozu, T., K. Nakamura, R. Meneghini, and W. C. Boncyk, 1991: Dual-parameter radar rainfall measurement form space: A test results from an aircraft experiment. IEEE Trans. Geosci. Remote Sens., 29, 690–703.

234

K. Nakamura and T. Iguchi

Li, J. and K. Nakamura, 2002: Characteristics of the mirror image of precipitation observed by the TRMM precipitation radar. J. Atmos. Oceanic Technol., 19, 145–158. Liao, L., R. Meneghini, T. Iguchi, and A. Detwiler, 1997: Estimation of snow parameters from dual-wavelength airborne radar. Prepr. 28th Conf. Radar Meteor., Am. Meteor. Soc., 510–511. Mardiana, R., T. Iguchi, and N. Takahashi, 2004: A dual-frequency rain profiling method without the use of surface reference technique. IEEE Trans. Geosci. Remote Sens., 42, 2214–2225. Masunaga, H., T. Iguchi, R. Oki, and M. Kachi, 2002: Comparison of rainfall products derived from TRMM microwave imager and precipitation radar. J. Appl. Meteor., 41, 849–862. Meneghini, R. and D. Atlas, 1986: Simultaneous ocean cross-section and rainfall measurements from space a nadir-looking radar. J. Atmos. Oceanic Technol., 3, 400–413. Meneghini, R., T. Iguchi, T. Kozu, L. Liao, K. Okamoto, J. A. Jones, and J. Kwiatkowski, 2000: Use of the surface reference technique for path attenuation estimates from TRMM precipitation radar. J. Appl. Meteor., 39, 2053–2070. Meneghini, R., T. Kozu, H. Kumagai, and W. C. Boncyk, 1992: A study of rain estimation methods from space using dual-wavelength radar measurements at near nadir incidence over ocean. J. Atmos. Oceanic Technol., 9, 364–382. Meneghini, R., H. Kumagai, J. R. Wang, T. Iguchi, and T. Kozu, 1997: Microphysical retrievals over stratiform rain using measurement from an airborne dual-wavelength radar-radiometer. IEEE Trans. Geosci. Remote Sens., 35, 487–506. Nakamura, K., 1991: Biases of rain retrieval algorithms for spaceborne radar caused by nonuniformity of rain. J. Atmos. Oceanic Technol., 8, 363–373.

19 A NEXT-GENERATION MICROWAVE RAINFALL RETRIEVAL ALGORITHM FOR USE BY TRMM AND GPM Christian Kummerow1, Hirohiko Masunaga1, and Peter Bauer2 1

Department of Atmospheric Science, Colorado State University, Fort Collins, CO, USA European Centre for Medium-Range Weather Forecasts, Reading, UK

2

1 INTRODUCTION Passive microwave rainfall algorithms have evolved steadily from those designed for the early Electronically Scanning Microwave Radiometer (ESMR), through the Scanning Multichannel Microwave Radiometer (SMMR) on Nimbus-7, and the Special Sensor Microwave Imager (SSM/I) instruments flying on the Defense Meteorological Satellite Program (DMSP). A number of algorithms fitting roughly three classes have emerged. These are (a) the “emission type” algorithms (e.g., Wilheit et al. 1991; Berg and Chase 1992; Chang et al. 1999) that use low-frequency channels to detect the increased radiances due to rain over radiometrically cold oceans; (b) the “scattering” algorithms (Spencer et al. 1983; Grody 1991; Ferraro and Marks 1995) that correlate rainfall to radiance depressions caused by ice scattering present in many precipitating clouds; and (c) the “multichannel inversion” type algorithms (Olson 1989; Mugnai et al. 1993; Kummerow and Giglio 1994; Smith et al. 1994; Petty 1994; Bauer et al. 2001; Kummerow et al. 2001) that seek to invert the entire radiance vector simultaneously. Among these, the Wilheit et al. (1991) and Kummerow et al. (2001) algorithms are used operationally for the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) as well as the Advanced Microwave Scanning Radiometer (AMSR-E) flying on Aqua, while the Wilheit et al. (1991) and Ferraro and Marks (1995) algorithms are used with SSM/I in the Global Precipitation Climatology Project (GPCP) over ocean and land, respectively. The Bauer et al. (2001) algorithm is used at ECMWF 235 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 235–252. © 2007 Springer.

236

C. Kummerow et al.

for rain assimilation experiments. In each case, algorithms have been optimized for the corresponding satellite sensor. Algorithm intercomparison efforts, initially aimed at identifying the “best” algorithms have not been able to make much headway, as each algorithm appears to have strengths and weaknesses related to specific applications, while none appears to be universally better than the others. The next advance in global precipitation monitoring, the Global Precipitation Measurement (GPM) mission, is providing new impetus towards a common algorithm framework. The GPM concept consists of a core satellite, with a dual-frequency precipitation radar (DPR), and a multichannel microwave imager (GMI). This component is similar in concept to the TRMM design but with improved radar capabilities and an orbit that will cover between 65–70° of latitude. In addition, the GPM concept uses a constellation of operational and dedicated radiometers to produce global, three hourly rainfall products required by many applications. The fact that radiometers for the GPM constellation are not fully specified and will evolve throughout the mission based on contributions from a number of different space agencies immediately imposes a number of highlevel requirements upon any algorithm designed for these sensors. Of utmost importance is the need for a transparent, parametric algorithm that insures uniform rainfall products across all sensors. The requirement for transparency is clear. A mission of GPM’s scope should not rely on a single black box operated by any one individual. Instead, it requires an open architecture that will allow the international community to participate in the algorithm development, its refinement, and its error characterization. The requirement for a parametric algorithm is also self-evident. Since GPM is being designed as an ongoing cooperative concept among many agencies, algorithms cannot be designed for specific radiometers with defined frequencies, viewing geometry, spatial resolutions or noise characteristics. The algorithm should be applicable to any sensor. Such a requirement leads naturally to a generalized framework that avoids the need for specific frequencies for their application. Finally, the algorithm should be robust in such a way that differences between sensors can be confidently interpreted as physical differences between observed scenes rather than artefacts of the algorithm. Together with the above requirements, algorithms designed for the future should also be able to fully characterize uncertainties at any space and time scale being considered by the users. This ranges from instantaneous estimates needed for many hydrologic and weather forecasting applications to large space and time averages required for climate model verification and climate trend monitoring. While such a requirement is also perhaps selfevident, such a complete error characterization does not currently exist and is undoubtedly the greatest challenge facing the community.

Microwave Rainfall Retrieval Algorithm

237

2 THE ALGORITHM Rainfall retrieval algorithms are not fully constrained. Instead, a priori information must be supplied to help constrain the estimated 3-dimensional (3D) properties of precipitating clouds. The requirement that the GPM algorithm be adaptable to any satellite sensor and that it produces realistic uncertainty estimates for global application reduces the large set of previous algorithms to those that involve physical forward/inverse modeling where the statistical properties of the a priori information and the models can be formulated in a consistent way. In a physical framework, the optimum estimate of a state vector (precipitation profile), x, must be obtained using an observation vector (brightness temperatures), y, plus additional a priori information. Due to errors in modeling and observation (error covariance R), the relation between state and observation is usually described by probability density functions (pdf’s)This can be formalized with Bayes’ theorem (e.g., Rodgers 2000):

P(x | y ) =

P ( y | x) P ( x) P(y )

(1)

P(x|y) is the posteriori probability of x when y is observed. P(y|x) is the probability of making observation y when x is present, while P(x) and P(y) are the a priori probabilities of x and y, respectively. The latter may come from global statistics of state and observations. The determination of P(y|x) requires a model that translates between state and observation space. This model may also be used to compute P(y) if P(x) is assumed to fully describe the a priori distribution of x. Examples of the application of the above principle are the ‘Bayesian’ rainfall retrieval schemes that found rather wide distribution in recent years (Evans et al. 1995; Kummerow et al. 1996; Olson et al. 1996; Haddad et al. 1997; Marzano et al. 1999; Bauer et al. 2001, Kummerow et al. 2001; Viltard et al. 2004). One particular problem associated with these rainfall retrievals is that the model that connects states and observations, i.e., y = F(x) + ε (where ε is the modeling error), is generally nonlinear. This immediately implies that the inversion of this relation is state dependent, and the inversion must be formulated differently depending on whether (a) a first guess of the actual state, xb, and its error covariance, B, is known and Gaussian with respect to the true state or (b) only a pdf of state x is known from which the pdf of y can be calculated. If (a) applies, Eq. (1) can be transformed to:

1 ⎧ 1 ⎫ T T P (x | y ) = exp⎨− [y − F(x)] R −1 [y − F(x)] − [x − x b ] B −1 [x − x b ]⎬ (2) 2 ⎩ 2 ⎭

C. Kummerow et al.

238

The probability of P(x|y) is maximized when the first derivative of Eq. (2) vanishes. This can be solved numerically by iterative procedures, and represents the “variational” retrievals. If Eq. (2) applies, it is more appropriate to seek the expected value of x. From practical considerations, the expected value is often expressed as (Olson et al. 1996):

E ( x) =

∑x i

i

{

}

exp − 0.5[y − F (x i )] R −1 [y − F(x i )]

{

T

}

∑ exp − 0.5[y − F(x i )] R −1 [y − F(x i )] T

(3)

i

even though the formulation in Eq. (3) makes the assumption that P(x) and P(y) are well known. This will be called “Bayesian” method even though both approaches are based on Bayes’ theorem. Both solutions employ a forward model that often consists of combined cloud resolving and radiative transfer models, the latter involving clear-sky atmospheric and surface models. In the variational framework, these models have to be directly inverted because the difference between model-calculated and observed brightness temperatures must be translated into increments to the initial physical state. Here, adjoint models have recently been developed for rainfall retrieval purposes (Moreau et al. 2003). The first-guess state and its error characteristics, however, are difficult to obtain for precipitation and this method is only possible in a well-constrained large-scale model (Moreau et al. 2003, 2004). The main advantage of such a method is its global applicability and its flexibility with respect to any input data, while its disadvantage is the requirement of a well-defined first guess and the computational cost. In the Bayesian method, the biggest challenge is the definition of the a priori database, P(x), because it is not well known for precipitating clouds. Historically, Bayesian schemes used precipitation profiles derived from a set of existing cloud-resolving model (CRM) simulations to construct the a priori database of potential precipitation structures that might be seen by a radiometer. The CRMs provide a physically consistent set of full 3D hydrometeor and latent heating profiles. They also provide a simple method to use understood physical processes to constrain an inversion. The biggest disadvantage of the Bayesian algorithm is its lack of general applicability because only a few CRM simulations are available (and useful) to construct a valid P(x). A first-guess constraint may be possible to help constrain P(x) in the future. Inherent to both methods is the impact of the dynamical and microphysical formulations in the forward model that often dominate the uncertainties of the radiative transfer modeling.

Microwave Rainfall Retrieval Algorithm

239

3 THE PARAMETRIC ALGORITHM FRAMEWORK Most likely, future algorithms will be composed of elements present in both approaches, i.e., pdf-type estimators with static databases and variational elements where reliable first-guess information is available. The Bayesian, as well as variational techniques have the added advantage that they are intrinsically parametric and uncertainties in the a posteriori rainfall can be computed in a straightforward fashion. Nonetheless, the approaches are not without pitfalls. Variational approaches are contingent upon a good first guess, which is often difficult to make in the case of precipitation systems. Current Bayesian schemes, in addition to relying on incomplete CRM simulations, use procedures such as rainfall screening, freezing-level estimates and convective/stratiform classification in order to improve the retrieval performance. Incomplete databases require procedures to solve the problem when none of the simulated profiles are close to the observed brightness temperature vector. More subtle, but perhaps more important to both schemes, are the errors in the a priori database itself. Errors in the CRM simulations will cause radiometers with different channel combinations to retrieve different rainfall amounts. This potential aliasing is simple to illustrate with a hypothetical CRM that consistently produces too much ice in the simulation—thus creating simulations with large Tb depressions at high frequency for even modest rainfall rates. A sensor with only low frequency channels (e.g., 10–37 GHz) may not be susceptible to this problem and will retrieve approximately the correct rainfall (all else being correct). A sensor with only high frequency channels (e.g., 85 and 150 GHz), on the other hand, will match large Tb depressions to relatively modest rain cases with large Tb depressions found in the CRM simulation. This will cause a consistent underestimation and a different result than that obtained from the first sensor. Conceptually, a parametric algorithm must therefore address two distinct issues. It must avoid any channel specific procedures in the algorithm, and it must create an a priori database that is consistent with all the brightness temperatures that may be observed by individual radiometers. Avoiding channel specific procedures will be seen to be a relatively straightforward task in both the Bayesian and variational frameworks. Building a representative a priori database with a verifiable error model is a far more challenging task. The error model, in particular, is very difficult to construct because of the role of CRMs in the forward computations. While they are useful in the sense that they add physical constraints to the clouds that may be retrieved, they are extremely difficult to verify quantitatively since they are constructed to simulate physically consistent scenes rather than the details

240

C. Kummerow et al.

of any one observed cloud realization. A more complete and verifiable a priori database appears thus to be crucial for any algorithm in the GPM era. Following is a description of a prototype parametric algorithm using both variational as well as Bayesian methods. While the various aspects of the algorithm are being developed separately and for different sensors, they are presented here as parts of the same conceptual algorithm for illustration purposes. In this conceptual algorithm, the a priori database is constructed from a combination of TRMM precipitation radar (PR), TMI and CRM information when the TMI footprint contains rainfall as determined from the PR. When the radiometer footprint does not contain rainfall, a variational technique is used with the radiometer data to obtain the clear air parameters. Together, these two components lead to a consistent 3D distribution of geophysical parameters that are fully representative of the observed scenes and fully consistent with the observation vector. Benefits of this more representative database are discussed in Viltard et al. (2004). The rainfall retrieval itself follows the Bayesian formalisms cited earlier.

3.1 The non-raining simulations Over oceans, passive microwave radiances depend upon column-integrated water vapor, cloud liquid water, sea surface temperature, and surface wind speed. These geophysical parameters can be retrieved simultaneously from the TMI itself for which the TRMM PR shows no rainfall. Techniques such as those described in the literature (e.g., Wentz 1997) do exactly this. Unfortunately, that technique has some shortcomings with respect to the current objectives. It only works over ocean, and it seeks consistency only among the channels used in the particular inversion. An alternative algorithm for clear-sky applications makes use of the previously introduced variational approach. Due to the many free parameters, in particular over land surfaces, the physical framework has to be kept very simple and only a few bulk parameters may be retrieved. As an example, we chose a set of four or six free parameters for ocean and land, respectively. Over ocean, these are surface skin temperature, near-surface wind speed, water vapor path, and cloud liquid water path, while over land these are surface skin temperature, effective water coverage, vegetation coverage, surface roughness, water vapor path, and cloud liquid water path. The surface skin temperature also determines the effective atmospheric temperature by assuming a constant lapse rate. The effective atmospheric temperature has to be understood as the temperature of the lower atmosphere where most of the water vapor is present, while the effective water coverage over land summarizes the true coverage with open water and soil moisture. The effect of soil moisture and open water on land surface emissivity is very similar. Atmospheric absorption was calculated according to Liebe et al. (1992), sea-surface emissivity with the model of Ellison et al.

Microwave Rainfall Retrieval Algorithm

241

(2003) and land surface emissivity with the model of Bauer and Grody (1995). The atmosphere consists of seven layers with constant depths.

Figure 1. Example of variational retrieval of surface skin temperature (a), effective water coverage (b), effective vegetation coverage (c) and water vapor path (d) using SSM/I data on November 1, 2003. Scales at the bottom refer to units [K], [], [], [kg m–2], respectively.

Figure 2. TB-departures (observation minus simulation) using first-guess (a, b) and after variational retrieval (c, d) at 19.35 GHz (a, c) and 85.5 GHz(b, d) and horizontal polarization. Scales at the bottom refer to units [K].

First-order climatological values were assumed for the above parameters to initialize the minimization of some SSM/I overpasses over South America and the Southern Caribbean on November 1, 2003. Figure 1 shows the resulting retrievals for the surface skin temperature, effective water coverage, vegetation coverage and water vapor path, respectively. The fields represent reasonable distributions showing the Amazon River basin in both Fig. 1b and 1c as well as the orography-dependent surface temperature distribution. The water vapor fields over ocean reach very low values in the presence of clouds, which indicate a possible aliasing effect between water vapor and liquid water absorption. Nonetheless, these results indicate that a variational retrieval is feasible for clear-sky applications providing background fields

242

C. Kummerow et al.

for the hydrometeor retrievals in the presence of rain. More realistic first-guess values as well as error covariances may be obtained from climatological fields produced by global model analyses. Potential refinements in the physical models used in the inversion may also lead to further improvements. Figure 2 illustrates the brightness temperature departures before and after the retrieval at 19.35 and 85.5 GHz. The departures are quite large in areas with strong water vapor gradients and in the presence of clouds over sea with values above 20 K. This indicates that the first-guess values chosen for this example are only appropriate for illustration rather than operational application; however, the minimization performs well and reduces the departures to within instrument error limits. Clouds and light precipitation may be present in those areas where large departures remain after the retrieval (in particular at 85.5 GHz). The problem with precipitation is artificial, as the real database would be constructed from TMI data for which rain/no rain information is available from the PR.

3.2 The raining scene Figure 3 illustrates the overall flow of the algorithm described here to derive precipitation profiles consistent with both radar and radiometer measurements. The non-raining parameter retrievals, indicated by blue-colored items in Fig. 4, were introduced in the previous section. In this section, the rain-profiling scheme using PR, TMI, and CRM information is outlined. The PR identifies pixels with radar echoes significantly above the noise threshold as “rain certain.” The weakest detectable signal by PR corresponds roughly to 0.5 mm h–1 in rain rate. The GPM 35-GHz radar is currently planned to have a threshold of approximately 12 dBZ, which corresponds to roughly 0.2 mm h–1. If PR detects a rain signal, the rain profile that best fits the PR reflectivity profile is selected from a set of precomputed CRM simulations. The reflectivity of the cloud-model profiles was obtained by computing single particle backscattering and extinction properties based upon Mie theory and assuming a gamma drop size distribution (DSD) with a given median volume diameter (D0) and µ = 3. As an example, the initially assumed DSD model may be taken, which was constructed to be consistent with the Z-R relations assumed in the TRMM PR operational rain-profiling algorithm (2A25) developed by Iguchi et al. (2000). The particle size distributions of other hydrometeor species are the same as adopted by the CRM except for melting particles. Here, the microwave properties of melting hydrometeors are simplified in such a way that the particle size distributions are linearly transformed under an averaged dielectric function from ice to liquid within a half-kilometer layer below the freezing height. This simplified treatment of melting particles could be

Microwave Rainfall Retrieval Algorithm

243

replaced with a more elaborate microphysical model in the future (Bauer 2001; Olson et al. 2001; Battaglia et al. 2003). The CRMs used in the study were the Goddard Cumulus Ensemble Model (GCE) and the University of Wisconsin Non-hydrostatic Modeling System (UW-NMS), which are the same simulations used in the a priori database used in the Goddard Profiling Algorithm (Kummerow et al. 2001).

Figure 3. Algorithm flowchart. Blue colored items are related to the non-raining (NR) parameter retrieval, yellow to the PR profile matching, and red to the comparison of matched profiles in the Tb space.

The best fit in the PR reflectivity matching is defined as the one having the least root-mean-squared difference between observed (PR-1C21 attenuationuncorrected reflectivity) and computed reflectivity. When the observed PathIntegrated Attenuation (PIA) from the PR is sufficient to provide a robust signal, there is additional DSD information available from the radar. In this case, the best-fit solution with respect to both the reflectivity profile as well as the PIA is sought by searching simulated profiles with several different DSD assumptions. While computationally different from the PR algorithm developed by Iguchi et al. (2000), philosophically this step matches the PR procedure by

244

C. Kummerow et al.

adjusting the retrieved DSD to match both the reflectivity and the PIA when it is deemed robust. Figure 4 shows a snapshot of surface rain rate given by the matched CRM profile and 2A25 surface rain, exhibiting good qualitative agreement. A direct pixel comparison of the observed and reconstructed surface rainfall for this scene shows a bias of 1.5% with a correlation of 0.96. The light grey portion of Fig. 3, which includes an iterative step if PIA is robust, shows the flow of the above procedure. At this point in the retrieval, the raining and non-raining scenes must be merged. As discussed in the previous section, the non-raining retrievals are applied to all TMI footprints in which PR observed no rain. The rain retrieval, however, is applied to PR pixels that generally have higher spatial resolution than the TMI footprints. This difference in resolution can lead to small areas within partially raining TMI footprints for which the clear air retrieval could not be performed but for which PR observes no rain. These areas must be filled by an interpolation scheme before the final step in the raining retrieval can be completed. This interpolation scheme is also used to prescribe the surface conditions under the raining pixels.

Figure 4. Top: Surface rain rate given by CRM profiles that best fit the measured PR profiles. Bottom: 2A25 surface rain rate for the same scene as the top panel.

Figure 5 illustrates this procedure, originally developed by Shin and Kummerow (2003), for the cloud liquid water field over ocean. All clear air fields are treated in a similar manner. Figure 5a represents the TMI retrieval for column water vapor (CWV) in this example. Figure 5b shows the CWV field associated with the raining retrievals, and Fig. 5c the final CWV field in which TMI CWV has been mapped to PR pixel locations and missing values have been interpolated. The slightly lower CWV values in precipitation (relative to the non-raining surroundings) might be an artifact

Microwave Rainfall Retrieval Algorithm

245

of the algorithm still under development. Inspection of the raining profiles, however, indicate they are nearly all saturated. As such, the lower CWV values might be real if one takes into account the lower temperatures associated with evaporation cooling associated with light precipitation. The interpolation over land will introduce greater uncertainties as the effective water coverage, vegetation cover, and surface roughness can vary rapidly. Precipitation profiles obtained from the PR matching technique are assigned to the satellite swath. The geophysical parameters unobservable by PR such as SST, surface wind speed, water vapor, and cloud water are provided from the TMI retrieval in non-raining scenes and interpolated to the raining field of view (FOV). Radiative transfer calculations using the Eddington approximation (Kummerow 1993) are then performed along slant paths that intersect a few neighboring PR pixels to properly take into account the TMI incidence angle of 52.8°. The computed brightness temperatures are convolved with the Gaussian antenna pattern using the 3dB-beam width of each TMI channel. Figure 6 shows the retrieved liquid and ice water contents, along with the computed brightness temperatures along the scan center of the rain feature shown in Fig. 4. The observed and computed brightness temperatures generally exhibit good but not perfect agreement. Figure 5. Simulation steps for a non-raining scene over ocean. (a) Shows the columnar water vapor retrieved from the TMI data for pixels in which PR detected no rain. (b) Column water vapor in raining pixels as determined through the selected cloud resolving model profile. (c) The merged and interpolated final column water vapor field.

The dark grey portion of Fig. 3 summarizes the final procedure. If the computed Tb’s at the lower frequency channels are lower than the observed ones, the assumed drop sizes can be decreased in order to increase the liquid water content determined from PR. As can be seen from the diagram, however, this can only be done for those pixels for which PR is not able to determine its own DSD through the PIA estimate. Variational

246

C. Kummerow et al.

methods that seek to adjust the DSD to simultaneously fit PR and TMI observations are also possible. Such solutions may eventually prove superior to the current approach. They are, however, less transparent. In the current formulation, the final iterative procedure will adjust DSD, but only if the DSD is the one assumed by PR and not when it can be directly observed by the sensor. In addition to any Tb disagreements in the emission channels, Fig. 6 shows occasional discrepancies in the scattering channel (85 GHz), which can be attributed to an uncertainty in the microphysical treatment of ice hydrometeors in CRM. This discrepancy is minimized by interactively updating the ice density in the CRM model. Precipitation profiles consistent with both radar and radiometer are thus obtained by repeating the entire procedure with updated DSD and ice density models.

3.3 The a priori databases Construction of the a priori databases is straightforward once the 3D raining and non-raining parameters derived from TRMM TMI and PR swath overlap data have been determined. Compared to previous efforts that relied solely on the CRMs to provide cloud structures, the current methodology insures that the a priori database is more fully grounded in observations, which would improve the databases’ representativeness of actual rainfall spectra. Figure 6. Top two panels: Vertical cross section at the scan center of precipitation water and ice given by CRM profiles that best fit the measured PR profiles. Bottom four panels: Observed TMI brightness temperatures (solid lines) and computed brightness temperatures (dashed lines) at 10 GHz, 19 GHz, 37 GHz, and 85 GHz (vertical polarization).

Through the database construction process, furthermore, it is possible to relate the derived rainfall profiles to the environmental geophysical parameters controlling rainfall formation such as surface and upper-level

Microwave Rainfall Retrieval Algorithm

247

humidity, wind velocity field, freezing height, and aerosols using other satellite retrievals and/or objective analysis data archives. Since a set of observed brightness temperatures are not always sufficient to single out a proper rainfall profile, those quantities could be used to separate and index the databases to better constrain the retrieval. The resultant a priori databases would not only improve the algorithm performance but also provide climatological insights on the physical processes governing rainfall properties. A priori databases can be constructed for any sensor once the sensor characteristics are defined. One important exception is that the current procedure only refers to microwave window channels. Sounding channels in the 60- and 118-GHz oxygen absorption bands as well as the 183-GHz water vapor absorption channels depend upon details of the temperature or humidity profiles that are not observed directly by TRMM. These are not well represented by the above procedure and would not be well represented by the a priori database.

3.4 The retrieval Once the a priori databases of hydrometeor profiles and clear scenes, as well as their corresponding Tb’s are constructed for each sensor, a Bayesian retrieval methodology can be used to select those profiles that are consistent with the observations. Synthetic retrievals using procedures similar to those described here for a number of radiometer designs are presented in Shin and Kummerow (2003). Synthetic, in this case, meant that satellite brightness temperatures were simulated from the 3D geophysical parameter derived for the a priori database. Results from that work shows very small biases between satellites ( 0 and IR > 0: the merged value is based on an error variance-weighted averaging. If MW = 0 and IR = 0: the merged value is zero. If MW > 0 and IR = 0 or MW = 0 and IR > 0: the rain/no-rain decision is made according to the sensor with the highest rain detection Heidke skill score (HSS) (Heidke 1926).

The error variance and HSS statistics for MW and IR rain retrievals were evaluated using as reference matched hourly WSR-88D rainfall fields. Table 1 shows those statistics. Note a deviation in the rainfall error statistics between the SSM/I and TMI MW sensors. The MW error statistics were augmented by a representativeness error term used to account for the large time lags between a MW overpass and a certain hourly rain estimate. For this purpose we used hourly radar rainfall fields to determine the error variance and rain detection HSS score as function of time lag. The error statistics are shown in Fig. 1.

Figure 1. Representativeness error variance (left) and HSS score (right) presented as function of time lag. Table 1. Error variance and HSS of MW and IR rainfall estimates.

Variance HSS

MW (TMI) 0.65 0.41

MW (SSM/I) 0.8 0.37

IR 1.65 0.1

5 SIMULATION EXPERIMENT AND RESULTS As mentioned in the Introduction, this study aims at studying the effect of rainfall retrieval error on the simulation of land surface parameters assuming

Assessment of Satellite Rain Retrieval Error Propagation

363

that our knowledge of the physical processes and other input parameters is accurate. The CLM, which provides a physically based framework for land surface processes, is used to model the 1D grid-averaged processes at two distinct, in terms of vegetation cover, sites. Three simulation scenarios are considered where CLM is forced with the three different rainfall data sets (radar, IR, and merged MW/IR). During the spin-up period of the model (January to April) the grid-averaged radar rainfall was commonly used as input to the model. The other meteorological forcing parameters such as radiation budget, air temperature, wind speed, and relative humidity were based on grid-averaged MESONET measurements, which were considered to be associated with insignificant error. Consequently, the only source that would differentiate the predicted land surface variables among the three simulation scenarios is the differing precipitation input. This difference is defined in relative terms as following:

εx =

Vx − Vref Vref

(1)

V is used to symbolize the different hydrologic variables (including precipitation). Subscript “x” refers to the variables predicted from satellite IR or merged rain estimates and subscript “ref” refers to variables predicted on the basis of grid-averaged rain gauge-calibrated radar rainfall. The land surface variables evaluated in this study are: latent heat flux (LE), sensible heat flux (SH), surface runoff (Roff), soil moisture content at 21 cm depth ( θS ), and soil temperature at 21 cm depth ( θ T ). In Table 2 we summarize the relative error (ε) statistics (mean and standard deviation [STD]) of the two satellite retrievals in terms of precipitation and CLM-predicted variables. The statistics are presented for three spatial scales (0.25, 0.5, and 1 degree) and the two vegetation regimes (Hveg and Lveg). A first general observation is that the merged rain retrieval is associated with less bias and error variance in all predicted variables compared to the IR rain retrieval. In terms of bias (or mean error) we note an almost 250% increase between low (Lveg) and high (Hveg) vegetation sites for the IR retrieval, while the corresponding increase in the case of merged rain estimates is moderate (~80%). An interesting observation for the IR retrieval is that the rain estimation bias magnifies significantly in runoff prediction (~3 times) in the case of Lveg, while the corresponding runoff-to-rainfall bias ratio for the merged rain retrieval is almost one. In the case of Hveg site the runoff bias

E. N. Anagnostou

364

Table 2. Error statistics (bias and STD) of rainfall and CLM-predicted variables derived from the simulation experiment.

Lveg

P

Mean error (bias) 0.25° 0.5° 60 (26) 60 (26)

LE SH Roff

34 (19) –26 (–15) 115 (37)

31 (18) –25 (–14) 115 (35)

32 (18) –27 (–16) 116 (30)

θS

7 (0.5)

7 (0.2)

8 (0.2)

θT

–0.4(–0.2)

–0.4(–0.2)

θS

–0.5(– 0.3) –12 (–15) –11 (–6) 13 (7) –7 (–24) –5 (–4)

0.25° 124 (113) 76 (69) 35 (30) 187 (144) 113 (117) 40 (34)

–12 (–15) –12 (–7) 15 (9) –7 (–25) –5 (–4)

–12 (–15) –14 (–8) 17 (10) –6 (–27) –5 (–3)

93 (88) 49 (44) 29 (26) 93 (79) 71 (52)

88 (81) 49 (44) 29 (26) 91 (71) 63 (43)

76 (67) 47 (42) 27 (24) 92 (59) 60 (41)

θT

0.02 (0.0)

0.02 (0.0)

0.05 (0.0)

15 (13)

15 (13)

13 (10)

1.0° 60 (26)

(21 cm)

Hveg

(21 cm) P LE SH Roff

STD 0.5° 129 (113) 74 (67) 34 (30) 213 (148) 108 (110) 39 (32)

1.0° 130 (109) 68 (60) 32 (27) 232 (146) 118 (120) 36 (29)

(21 cm) (21 cm)

seems to reduce (smoothing effect) by about 30%. In terms of soil moisture and other variables, we note a significant reduction of bias for both satellite retrievals and vegetation sites. In terms of the relative STD error statistic our observations are the following. We demonstrate a moderate reduction of STD from Lveg to Hveg, and notable spatial scale dependence for STD in the Hveg regime. The STD of rainfall, runoff, and soil moisture error for the merged (IR) rain estimates reduces by an average of 23% (12%) going from 0.25 to 1-degree grid resolution. We show magnification in the error STD going from rainfall to runoff variable, and a smoothing effect for all other variables. The rate of STD magnification for the IR retrieval exhibits a spatial scale dependence (ranging from 60% to 80%) in the case of Lveg, while the corresponding rate of magnification for the merged rain estimate is consistent across the scales at ~30%. In Hveg site the IR retrieval error STD magnification in runoff is shown to vary from about 5% (at 0.25 degree) to 25% (at 1 degree), while for the merged rain estimates we note a consistent smoothing effect (~10% error reduction). In all other variables the STD error propagation is associated with smoothing, which varies across the different variables. In soil moisture, for example, the STD reduces by about 12% (8%) in the merged (IR) retrieval case. The error STD smoothing is significant for the energy-related variables and fluxes

Assessment of Satellite Rain Retrieval Error Propagation

365

(θT, LE, and LH), indicating that errors in precipitation alone would have minimal effect in those variables. We finally used our experimental data to investigate the spatial structure of error in the various hydrological and land surface variables. This is presented in Fig. 2 in terms of spatial correlation of error (ε) as function of distance. We present two curves in each panel, one representing the IR and the second the merged rain retrievals. Left and right panels represent Lveg and Hveg sites, respectively. We make the following observations. First, error in rainfall and runoff decorrelate faster than in the other variables. Second, the IR retrieval error is associated always with lower lag-correlation than the corresponding IR retrieval error. This explains why spatial averaging resulted in a more significant STD reduction in the merged case compared to the IR retrieval case.

Figure 2. Spatial lag-correlation of error. Left and right panels correspond to Lveg and Hveg sites, respectively.

E. N. Anagnostou

366

6 CLOSING REMARKS This paper presented a framework for investigating hydrologically relevant satellite rain retrievals. The strategy adopted is, firstly, to compile the hydrometeorological variables needed by an LSM including three different sources of precipitation data, i.e., gauge-calibrated radar rainfall and two satellite retrieval (an IR and a merged MW/IR rain retrieval), assuming that the model parameters are well representative of the land surface processes. The CLM was selected to simulate the land surface processes at three scales, 0.25, 0.5, and 1 degree, and two distinct vegetation conditions. Simulated results from the two satellite rain retrieval-forcing parameters were compared to corresponding simulations derived from the radar rainfall input considered as reference. The primary conclusions of the present study are as follows: •

• • •

Rainfall error propagates nonlinearly in hydrologic simulation uncertainty-precipitation error structure and land surface conditions affect this error propagation (we noted this specifically in the case of runoff, and to a lesser extent in soil moisture). Error variance reduces (nonlinear smoothing effect) in most of the hydrological variables and fluxes, but runoff. We noted lower error statistics when using the merged MW/IR rain input compared to the IR-only input. Overall, spatial averaging reduces the error in most of the land surface variables, but this reduction depends on the rain retrieval error characteristics and land surface conditions.

An important observation from this study is that the effect of vegetation and structural characteristics of rain retrieval error are factors affecting the error propagation in land surface variables. This study investigated only a limited number of land surface conditions and satellite retrieval schemes and, thus, should only be viewed in a qualitative sense. Apparently, using different rain retrievals and hydrometeorological regimes could lead to different error characteristics. We would like to view this experimental error propagation framework as the basis for achieving a more comprehensive hydrologic assessment of satellite retrievals, and/or for developing merging techniques that would optimize hydrological prediction error statistics.

7 REFERENCES Adler, R. F., G. J. Huffman, D. T. Bolvin, S. Curtis, and E. J. Nelkin, 2000: Tropical rainfall distributions determined using TRMM combined with other satellite and rain gauge information. J. Appl. Meteor., 39, 2007–2023.

Assessment of Satellite Rain Retrieval Error Propagation

367

Anagnostou, E. N., A. J. Negri, and R. F. Adler, 1999: A satellite infrared technique for diurnal rainfall variability studies. J. Geophys. Res., 104 (D24), 31477–31488. Beven, K. J. and M. J. Kirkby, 1979: A physically based variable contributing area model of basin hydrology. Hydrol. Sci. Bull., 24, 43–69. Bonan, G. B., 1996: A land surface model (LSM version 1.0) for ecological, hydrological, and atmospheric studies: Technical description and user’s guide. NCAR Technical Note, NCAR/TN-417+STR, Boulder, CO. Bonan, G. B., S. Levis, L. Kergoat, and K. W. Oleson, 2002: Landscapes as patches of plant functional types: an integrating concept for climate and ecosystem models. Global change Biochem Cycles, 16 (2), Art No. 1021. Bonan, G. B., K. W. Olsen, M. Vertenstein, S. Levis, X. B. Zeng, Y. J. Dai, R. E. Dickinson, and Z. L. Yang, 2002: The land surface climatology of the community land model coupled to the NCAR community climate model. J. Climate, 15, 3123–3149. Brock, F. V., K. C. Crawford, R. L. Elliot, G. W. Cuperus, S. J. Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma Mesonet: a technical overview. J. Atmos. Oceanic Technol., 12, No. 1, 5–19. Dai, Y. J. and Q. C. Zeng, 1997: A land surface model (IAP94) for climate studies 1: formulation and validation in off-line experiments. Adv. Atmos. Sci., 14, 433–460. DeFries, R. S., M. C. Hansen, and J. R. G. Townshend, 2000a: Continuous fields of vegetation characteristics: a linear mixture model applied to multi-year 8 km AVHRR data. Int. J. Remote Sens., 21, 1389–1414. DeFries, R. S., M. C. Hansen, J. R. G. Townshend, A. C. Janetos, and T. R. Loveland, 2000b: A new global 1-km dataset of percentage tree cover derived from remote sensing. Global Change Biol., 6, 247–254. Dickson, R. E., A. Henderson-Sellers, P. J. Kennedy, and M. F. Wilson, 1986: BiosphereAtmosphere Transfer Scheme (BATS) for the NCAR community climate model. Technical Note; NCAR/TN-275+STR, NCAR, Boulder, CO. Fulton, R. A., J. P. Breidenbach, D. J. Seo, D. A. Miller, and T. O’Bannon, 1998: The WSR88D rainfall algorithm. Wea. Forecasting, 13 (2), 377–395. Global Soil data Task, 2000: Global soil data products CD-ROM (IGBP-DIS). International Geosphere-biosphere Programme – data and Information Available Services. Available at: http://www.daac.ornl.gov Grecu, M. and E. N. Anagnostou, 2001: Overland precipitation estimation from TRMM passive microwave observations. J. Appl. Meteor., 40 (8), 1367–1380. Hossain, F., E. N. Anagnostou, and T. Dinku, 2004: Sensitivity analyses of satellite rainfall retrieval and sampling error on flood prediction uncertainty. IEEE Trans. Geosci. Remote Sens., 42, 130–139. Hsu, K., X. Gao, S. Sorooshian, and H. V. Gupta, 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36 (9), 1176–1190. Huffman, G. J., R. F. Adler, E. F. Stocker, D. T. Bolvin, and E. J. Nelkin, 2003: Analysis of TRMM 3-hourly multi-satellite precipitation estimates computed in both real and postreal time. Proc. 12th Conf. Sat. Meteor. and Oceanog., 9–13 Feb. 2003, Long Beach, CA. Huffman, G. J, R. F. Adler, M. M. Morrisey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one degree daily resolution from multisatellite observations. J. Hydrometeor., 2, 36–60. Janowiak, J. E., R. J. Joyce, and Y. Yarosh, 2000: A real-time global half-hourly pixel resolution infrared dataset and its applications. Bull. Amer. Meteor. Soc., 82, 205–217. Koster, R. and M. Suarez, 1996: Energy and water balance calculations in the MOSAIC LSM. NASA Tech Memo 104606, Vol. 9. Liang, X., E. Wood, and D. Lettenmaier, 1996: Surface and soil moisture parameterization of the VIC-2L model: Evaluation and modifications. Global Planet. Change, 13, 195–206.

368

E. N. Anagnostou

Loveland, T. R., B. C. Reed, J. F. Brown, D. O. Ohlen, Z. Zhu, L. Yang, and J. W. Merchant, 2000: Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int. J. Remote Sens., 2, 1303–1330. Mitchell, K. and Co-authors, 1999: GCIP Land Data Assimilation System (LDAS) project now underway. GEWEX News, 9 (4), 3–6. Mitchell, K. and Co-authors, 2000: Recent GCIP-sponsored advancements in coupled landsurface modeling and data assimilation in the NCEP Eta mesoscale model. Prepr. 15th AMS Conf. on Hydrology, Long Beach, CA, Paper P1.22. Nijssen, B. and D. P. Lettenmaier, 2004: Effect of precipitation sampling error on simulated hydrological fluxes and states: Anticipating the Global Precipitation Measurement Satellites. J. Geophys. Res., 109, D02103, doi: 10.1029/2003JD003497. Seo, D. J., J. Breidenbach, and R. Fulton, 2000: Real-time adjustment of range-dependent biases in WSR-88D rainfall estimates due to nonuniform vertical profile of reflectivity. J Hydrometeor., 1 (3): 222–240. Seo, D. J., J. P. Breidenbach, and E. R. Johnson, 1999: Real-time estimation of mean field bias in radar rainfall data. J. Hydrology, 223 (3–4): 131–147. Todd, M. C., C. Kidd, D. Kniveton, and T. J. Bellerby, 2000: A combined satellite infrared and passive microwave technique for estimation of small scale rainfall. J. Atmos. Oceanic Technol., 18, 742–754.

29 EURAINSAT ALGORITHM VALIDATION AND INTERCOMPARISON EXERCISE Martina Kästner DLR, Oberpfaffenhofen, Germany

Abstract

This study compares four satellite rain estimations based on microwave (MW), infrared (IR) or combined MW–IR techniques and contrasts them with the mesoscale Bologna local area model (BOLAM) rain analysis or the network-based gauge data from the Global Precipitation Climatology Centre (GPCC) for a period from 08 to 13 November 2001 over the western Mediterranean Sea during a severe weather event, which resulted in a disastrous flood in Algeria. The PR-adjusted TMI Estimation of Rainfall (PATER) and frequency difference algorithm (FDA) are applied to MW TRMM data, the neural rain estimator (NRE) uses geostationary IR Meteosat data and the combined NRL Turk algorithm uses both MW data from low orbiting satellites and IR data from a geostationary orbit. The unique gridded data provide an effective basis to compare instantaneous space measurements with different algorithms. Validation results indicate that there is generally a better performance for heavy rain than for weak rain. Both MW algorithms, PATER, and FDA, perform rather similarly whereas PATER is applicable exclusively over the ocean and shows some rain detection problems due to thick aerosol loads originating from the desert. The BOLAM model performs rather well in this case study; although only a small location error of a heavy rain area was analyzed. The IR-based techniques have the advantage of a high temporal repetition rate but both algorithms, NRL and NRE, have problems with identifying the correct rainy areas compared to MW results. Overall, the results suggest combining both advantages, the wellknown rain physics of the MW channels with the high temporal resolution of IR algorithms, to retrieve precipitation from satellite data.

Keywords

Satellite rain retrieval, microwave, infrared, validation, Algerian flood 2002, severe weather event

369 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 369–380. © 2007 Springer.

370

M. Kästner

1 INTRODUCTION Various climate models predict a decrease of precipitation in the future over many parts of the subtropics, particularly in the winter (Bolle 2003). Therefore, it is essential to have not only climatological data from land, but also from over the seas. Remote areas not covered by conventional observation networks can now be continuously monitored by low orbiting and geostationary satellites. The first satellite rain retrievals in both the infrared (IR) and the microwave (MW) spectra date back to the 1970s, some recent works are Turk et al. (2000), Bauer (2000, 2001), Bauer et al. (2001), Levizzani et al. (2001), Grose et al. (2002), Oh et al. (2002), Tapiador (2002), and Kidd et al. (2003) among others. A cross-comparison of TRMM and GPCP rainfall data sets are described in Adler et al. (2002). The evaluation of passive MW precipitation algorithms which are directly linked to the 3D structure of the precipitating system use measurements from different sensors on different satellites, like SSM/I on DMSP, TMI/PR on TRMM, or AMSU on NOAA. Passive MW techniques perform much better over the oceans than over land. The MW techniques are directly related to the hydrometeors through scattering and emission, but the low earth orbits and less frequent coverage hinders tracking of developing severe storms. While the daily course of precipitation is not easily obtained from TRMM data, IR-based techniques from geostationary satellites have been widely used due to the high revisit period. However, they have an inherent weakness regarding the physical relation between cloud top temperatures and underlying rain rate (RR). Further, the rain characteristics vary with different climate regimes, hence, any developed method has to be validated against appropriate in situ measurements taken over the region of interest. An intercomparison of PMW and/or IR-based algorithms with BOLAM model data and GPCC gauge measurements at different spatial resolutions is performed for the Algerian flood in early November 2001. Although the validity of the results obtained is restricted to the case studies some general information could be extracted. Further details can be found in Kästner (2003) and Kästner et al. (2006).

2 DATA The validation of various rain algorithms is performed for a severe weather event between 08 and 12 November 2001 on the Algerian coast and the Balearic Islands. The synoptic situation was characterized by strong surface winds and heavy rainfall. An intense upper-level trough pushed far to the south of Europe where a cutoff low developed. The METEOSAT-7 IR image (Fig. 1) shows the clouds with heavy rainfall on the Algerian coast. The rainfall started on late 9 November and ended the next day at about

Eurainsat Algorithm Validation and Intercomparison Exercise

371

noon on 10 November, when 150 L m–² within 6 h were observed. Together with the cutoff low process heavy thunderstorms developed in a cyclogenesis over the Balearic Islands the next day. The precipitation was reported to be greater than 400 mm over 2 days, with a maximum of 68 L m–² in 6 h (Thomas et al. 2003). Two processes intensified the convective development: (1) the cold maritime arctic air that crossed over the still 18°C warm Mediterranean Sea where it picked up moisture, destabilized, and met initially maritime subtropic air and (2) the strong surface winds blowing against the high mountains along the African coast (>2,300 m) caused intense orographic rainfall which led to the flooding disaster in Algiers with more than 750 deaths.

Figure 1. 10 November 2001; left: TRMM TMI brightness temperatures for 19h (top), 19v (mid ) and 85v (bottom) GHz channels (v = vertical, h = horizontal polarization), orbit time 0025 UTC; right: METEOSAT-7 IR, 1200 UTC. A = Algiers.

Input data for PMW algorithms (PATER, FDA) are the TMI brightness temperatures (TB) of nine channels (10.7v,h, 19.4v,h, 21.3v, 37.0v,h, 85.5v,h GHz) with varying resolutions from 70 to 6 km. Figure 1 shows differences of emission over water and over land for both polarizations and for different TMI channels. Over water, the rainy areas appear to be warmer than their surroundings, while over land they appear to be colder due to the high MW emission of land. The applied rain retrievals and rain data are briefly described in the following: The BOLAM is a hydrostatic, primitive equation, grid point model in σ -coordinates, using horizontal wind components, potential temperature, specific humidity, and surface pressure as basic dependent variables. The initial and boundary conditions are obtained from the ECMWF 6-hourly analyses. The frequency difference algorithm (FDA) of Kidd et al. (2003) uses the 19v and 19h GHz channels and relates it to the rain rate (RR), as described

372

M. Kästner

in Ebert et al. (1996). It is a PMW satellite rain algorithm that is operationally applied to TRMM as well as SSM/I orbit data. Independent 1-degree daily (1DD) data from the Global Precipitation Climate Centre (GPCC) base on recordings of the very dense rain-gauge network in Europe, the SYNOP reports are automatically checked and corrected for systematic measuring errors (Rudolf et al. 1996). The neural rain estimator (NRE) is an operational rapid update IR-based algorithm to diagnose half-hourly near-surface rainfall. It is an empirical technique suited for acquisitions from geostationary satellites. It uses some relevant features of the cloud top evolution and structure and information from an NWP model. The basic strategy for the Naval Research Laboratory (NRL)-blended technique (Turk et al. 2000) draws upon the probability matching methods developed in the radar meteorology field for specific “tuned” Z–R relations. Time- and space-coincident IR and MW pixels are collected from different satellites and used to produce dynamically updated TB–RR lookup tables. The over-ocean satellite rainfall algorithm PATER is a physical algorithm that uses only two empirical orthogonal functions instead of the nine TBs from the TMI channels. The retrieval database is generated from several 3D cloud model simulations including the melting layer (Bauer 2001). The algorithm (Bauer et al. 2001) has a stand-alone PMW component based on TRMM TMI (1B11) data and an optionally carefully colocated calibration with PR (2A25) data (~5 km) downscaled to the lower TMI spatial resolution for the 10 GHz (~50 km). Currently, it is foreseen for operational implementation in the ECMWF assimilation scheme.

3 METHOD OF ANALYSIS In a pre-study the RR of the PATER over-ocean algorithm were merged from three TRMM overflights daily and then downscaled to a 1° × 1° grid for comparison with the independent RR gauge data from the 1DD GPCC data. For the main study a common area, period, grid, and format were essential for combining data sets with different temporal or spatial resolutions sometimes describing different physical observables. All subsequent tasks like calibration, sampling, or error analysis needed a common grid that allowed an equivalent evaluation. For the joint effort of validation and intercomparison of several rain algorithms applied in the scope of the EURAINSAT project, continuous and categorical statistics were used according to Ebert et al. (1996, 1998). The Algerian severe weather event was used as a common case study for an intercomparison of PMW, IR, combined MW/IR rainfall algorithms and the BOLAM model. The common area extended from 15 W to 20 E and from 30 to 60 N; with a common period from 09 to 11 November 2001; and with RR

Eurainsat Algorithm Validation and Intercomparison Exercise

373

data resampled into a 0.25° × 0.25° lat–long grid (~28 km). TRMM and IRE data did not both consistently cover the complete common area, therefore, the subsequent comparison was conducted only over the available common area. The temporal coincidence was optimal for the different PMW algorithms, because PATER and FDA have the same TRMM database, otherwise the temporal window was better than ±15 min for comparisons with IR (NRL, NRE) and in most cases better than ±90 min for comparisons with the independent model data, which have a 3-hourly temporal resolution. Comparisons were made for single orbits as well as for merged data within 3-h periods.

4 RESULTS 4.1 1° intercomparison – PATER retrieval with GPCC Although the GPCC data set was over land and the PWM PATER data set was over water; coastal boxes contained in both data sets, could be compared (see Fig. 2).

Figure 2. Rain rate (RR) analysis in mm day–1 on 10 November 2001; left: 1DD GPCC (Global Precipitation Climate Centre) RR distribution; right: PATER retrieval, RR from merged orbits (0025 and 0210 UTC) downscaled to 1DD.

In order to get an adequate number of observations (n = 146) all the coastal boxes of the southwestern Mediterranean Sea over 5 days (different symbols in Fig. 3) during a severe rain and flood event were taken into account. The result of this validation study is a rather high correlation coefficient of 0.71 between rain gauges and PATER estimates. A closer examination shows three outcomes. (1) Most of the data points are within the red ellipse indicating a good agreement. (2) A few data points are within the green ellipse showing higher rain-gauge values than PWM RRs . This is explained

M. Kästner

374

by the different temporal data structure of instantaneous orbit data versus accumulated gauge data. (3) Biased (too high) PATER RRs for low rain gauge measurements (blue ellipse). This third outcome was expected, because the sensitivity of the PATER algorithm is given by Bauer (2001) with lwc = 0.07 g m–3 equivalent to RR = 24 mm day–1. In fact the false alarm rate indicates an even higher sensitivity to about 17 mm day–1 in this case.

Figure 3. Rain rate comparison of the 1-degree daily (1DD) GPCC gauge data vs. PMW PATER rain rate estimations in mm per day for the period 09–13 November 2001 (Algerian case).

4.2 0.25° intercomparison – PATER and other EURAINSAT retrievals A validation study of the satellite rainfall estimations has been performed for the Algerian case using independent intercomparison of different rainfall retrievals, including pure PMW, pure IR, combined MW/IR techniques, and model results using BOLAM before nudging with satellite rainfall data. Both PMW algorithms, PATER, and FDA, rely on the same TMI orbit data and so it was expected that their comparison would result in a rather similar rainfall region and intensity. Both algorithms are assessed to be of equal quality in this heavy rainfall event, considering that the PATER algorithm is restricted to ocean surfaces and to rain events above 1 mm h–1. The most successful

Eurainsat Algorithm Validation and Intercomparison Exercise

375

comparison of pure PMW with other techniques was in this case the BOLAM model followed by the NRL blended MW/IR technique both of which performed better than the pure IR technique. The results of the intercomparison of different rain retrievals are given for visual inspection for a single date (10 November 2001, 0300 UTC) (Fig. 4 and 5). A complete analysis of the categorical statistics of all possible combinations of the above-described algorithms within the period 08 to 13 November 2001 is given in Table 1.

Figure 4. 10 November 2001, TMI orbit 22741 + 22742, 0025 + 02:10 UTC; left: PATER algorithm – MW; right: FDA algorithm – PMW.

Figure 5. 10 November 2001, 0300 UTC; upper left: BOLAM – model; upper right: NRL Turk algorithm – combined IR–MW; bottom: NRE (neural rain estimator) – IR. (see also color plate 14)

M. Kästner

376

Figure 4 shows three areas of heavy rainfall, one west of the Canary Islands, the biggest one is around Algiers, and one southeast of Sardinia. The three rain areas coincide very well with results of both MW algorithms. Even the rain intensities are similar except for the Sardinia area. FDA works over land and ocean, PATER exclusively over ocean and for rainfall rates above 1 mm h–1. The low RRs erroneously detected over the sea south of Sicily (Fig. 4 left) are attributed to strong desert aerosol also detected as aerosol fallout in Rome the next day. The rainy speckles over the Atlantic are due to cumulus convective showers within the cold air. Compared to the PMW techniques, the BOLAM model rainfall (before nudging) shows wide agreement of the strong rain bands (Fig. 5 upper left). The rainfall intensities were rather similar; only the position of the rainfall had an error for the Sardinia area. On the other hand, the shower pattern over the Atlantic is not well matched and there are too many areas with light rain. The two IR-based algorithms, NRL and NRE (Fig. 5), give heavy rain areas over the Mediterranean Sea, but not at the correct position. The Algerian coast, where the maximum precipitation fell, is hardly classified as a heavy rain area. The combined MW/IR NRL algorithm shows a much better performance than the IR algorithm NRE alone. Overall it seems to be worthwhile to combine the high temporal resolution of IR with the better rainfall identification performance of MW techniques for monitoring purposes. The NRL algorithm or the now available TRMM 3B42RT products belong to this category. The low MW pixel resolution makes a 0.25° lat–long grid appropriate, but better spatial resolution is desired by the users. Table 1. Categorical statistics for different satellite rain retrievals.

P vs. B P vs. NRE P vs. NRL B vs. NRE B vs. NRL NREvs.NRL F vs. P F vs. B F vs. NRL F vs. NRE Total

Compared pairs

Hit

Miss

False alarm

Correct negative % 62.2 56.1 67.2 39.2 65.9 52.6 79.7 66.3 75.9 56.3

Heidke skill score Best= 1 0.67 0.63 0.70 0.61 0.76 0.73 0.82 0.78 0.81 0.69

N = 100% 8.346 2.595 14.619 3.774 13.880 6.320 3.353 6.128 7.142 931

% 4.9 7.2 3.3 22.3 9.9 20.6 2.5 11.2 5.0 12.5

% 21.1 16.6 14.3 10.0 13.3 15.6 8.9 18.3 15.2 23.3

% 11.9 20.0 15.2 28.6 10.9 11.1 8.9 4.2 3.9 7.8

67.088

9.9

15.7

12.3

Bias Best = 1 0.65 1.14 1.05 1.58 0.89 0.88 1.01 0.52 0.44 0.57

62.1

0.72

0.87

Eurainsat Algorithm Validation and Intercomparison Exercise

377

The event occurrences of the contingency table for comparisons within the time period 09 to 11 November 2001 consider the above-mentioned temporal windows (Table 1). The number of compared pairs (N) differs for each algorithm pair; thus, the occurrences are given in percent of each N for better comparison between the cases, the maxima (hits) or minima (false alarms) are italicized. The following abbreviations are used: B = BOLAM, F = FDA, P = PATER. The accuracy of all intercomparisons is measured in terms of Heidke skill score (HSS), and ranges from 0.61 (BOLAM vs. NRE), to respectable 0.78 (BOLAM vs. FDA), to optimal 0.82 (PATER vs. FDA, Fig. 6). No bias between the data sets is indicated when the bias value is unity, thus, PATER and NRL have only a small bias, while again PATER performs best when compared with FDA. Note that even in this heavy rain event only 10% of the gridded pixels are hits (rain/rain), whereas the majority (62%) is correct negatives (no-rain/no-rain), thus, the correct negatives dominate the statistics.

Figure 6. Comparison of PATER with FDA. The accuracy via HSS*1,000 (thick) depends on the rain/no-rain threshold RR-min.

378

M. Kästner

5 DISCUSSION A heavy rainfall situation was selected for a validation analysis of different satellite rainfall algorithms representing MW, IR, or combined MW/IR retrievals and the BOLAM model. The rainfall data were gridded on 1° and 0.25° in a common area over Western Europe and categorical statistics are used. Special interest was laid on the PATER algorithm that uses TRMM active and passive MW data. The PATER retrieval is crucially dependent on the forward 3D cloud model calculations. There is a need for better and more comprehensive cloud models that cover the whole spectrum of natural clouds, particularly clouds with moderate and light rain and snow. The correlation coefficient r is often used in comparison tasks. It has already been shown that r is dependent on the choice of the grid size (Turk et al. 2002). In this study the maximum r is reached for both PMW techniques, namely r is 0.88 for PATER vs. FDA (orbit 22,741, 0.25° grid). The accuracy is measured in terms of the HSS, which is a combination of hits, false alarms, misses, and correct negatives. HSS is truly dependent on the threshold RR-min, discriminating rain from no-rain pixels. Figure 6 results from varying the minimum detectable RR (RR-min) for the rain algorithm pair PATER vs. FDA. The event occurrences for each category are given on the y-axis. For RR-min = 1 mm h–1, false alarms (red) and misses (blue) are of the same order, which means that in this case both the algorithms are unbiased. The impact of RR-min on the accuracy is evident; the strong increase of HSS from 0.3 to 0.8 for RR-min greater than 0.7 mm h–1 indicates that higher RRs are more accurately detected than low ones. This implies that not only the chosen grid size but also the problem on where the rain/no-rain threshold is set is inherently associated with the accuracy problem. As the statistics are dominated by the correct negatives and not by the hits, maybe the use of entity-based methods, like contiguous rain area (CRA) verification give further insight into algorithm performances (Ebert 2000). In this case study the PMW techniques performed better than IR techniques. Overall, the results suggest combining both advantages, the wellknown rain physics of the MW channels with the high temporal resolution of IR algorithms, to retrieve precipitation from satellite data. Acknowledgment: This research is funded in by the EURAINSAT project, a shared-cost project (contract EVG1-2000-00030), co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development subprogramme 5th Framework Programme).

Eurainsat Algorithm Validation and Intercomparison Exercise

379

6 REFERENCES Adler, R. F., G. Huffman, and D. Bolvin, 2002: TRMM and GPCP initial cross-comparison. GEWEX News, 12, 5–6. Bauer, P., 2001: Over-ocean rainfall retrieval from multisensor data of the tropical rainfall measuring mission. Part I: design and evaluation of inversion databases. J. Atmos. Oceanic Technol., 18, 1315–1330. Bauer, P., D. Burose, and J. Schulz, 2000: Rain detection over land surfaces using passive microwave satellite data. ECMWF Tech. Memo., No. 330, Reading, England. Bauer, P., P. Amayenc, C. D. Kummerow, and E. A. Smith, 2001: Over-ocean rainfall retrieval from multisensor data of the Tropical Rainfall Measuring Mission. Part II: algorithm implementation. J. Atmos. Oceanic Technol., 18, 1838–1855. Bolle, H.-J. (ed.), 2003: Mediterranean Climate – Variability and Trends. Springer, Berlin, 372 pp. Buzzi, A., M. D’Isidoro, and S. Davolio, 2003: A case study of an orographic cyclone formation south of the Alps during the MAP-SOP. Quart. J. Roy. Meteor. Soc., 129, 1795–1818. Ebert, E. E., 1996: Results of the 3rd Algorithm Intercomparisons Project (AIP-3) of the Global Precipitation Climatology Project (GPCP). Revision I. Bureau of Meteorology Research Centre, Melbourne, Australia, 199 pp. Ebert, E. E. and M. J. Manton, 1998: Performance of satellite rainfall estimation algorithms during TOGA COARE. J. Atmos. Sci., 55, 1537–1557. Ebert, E. E. and J. L. McBride 2000: Verification of precipitation in weather systems: determination of systematic errors. J. Hydrol., 239, 179–202. Grose, A., E. A. Smith, H.-S. Chung, M. L. Ou, B. J. Sohn, and F. J. Turk, 2002: Possibilities and limitations for QPF using nowcasting methods with infrared geosynchronous satellite imagery. J. Appl. Meteor., 41, 763–785. Kästner, M., 2003: Inter-comparison of precipitation estimations using TRMM microwave data and independent data. In: Proc. 3rd GPM Workshop – Consolidating the Concept, Noordwijk, The Netherlands, 24–26 June 2003, AP-5. http://www.estec.esa.nl/conferences/ 03C06/ and in: Proc. CD-ROM 5th EGS Plinius Conference on Mediterranean Storms, Ajaccio, France, 1–3 Oct 2003. Kästner, M., F. Torricella, and S. Davolio, 2006: Intercomparison of satellite-based and model-based rainfall analyses. Meteor. Appl., 13, 213–223. Kidd, C., D. Kniveton, M. Todd, and T. Bellerby, 2003: Satellite rainfall estimation using a combined passive microwave and infrared algorithm. J. Hydrometeor., 4, 1088–1104. Levizzani V., J. Schmetz, H.J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. Oh, H. J., B. J. Sohn, E. A. Smith, F. J. Turk, A. S. Seo, and H. S. Chung, 2002: Validating infrared-based rainfall retrieval algorithms with 1-minute spatially dense raingauge measurements over the Korean peninsula. Meteor. Atmos. Physics, 81, 273–287. Rudolf, B., H. Hauschild, M. Reiß, and U. Schneider, 1992: Beiträge zum Weltzentrum für Niederschlagsklimatologie – Contributions to the Global Precipitation Climatology Centre. Meteorologische Zeitschrift, 1, 7–84. Tapiador, F., 2002: A new algorithm to generate global rainfall rates from satellite infrared imagery. Revista de Teledeteccion, 18, 57–61. Thomas, W., F. Baier, T. Erbertseder, and M. Kästner, 2003: Analysis of the Algerian severe weather event in November 2001 and its impact on ozone and nitrogen dioxide distributions. Tellus B, 55B, 993–1006.

380

M. Kästner

Turk, F. J., J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Combining SSM/I, TRMM and infrared geostationary satellite data in a near real-time fashion for rapid precipitation updates: advantages and limitations. Proc. The 2000 EUMETSAT Meteorological Satellite Data Users’ Meeting, Bologna, Italy, 29 May – 2 June 2000; EUM P 29, 452–459. Turk, F. J., E. E. Ebert, H. J. Oh, B. J. Sohn, V. Levizzani, E. A. Smith, and R. R. Ferraro, 2002: Validation of an operational global precipitation analysis at short time scales. In: Proc. 1st Workshop Intern. Precip. Working Group (IPWG), Madrid, 225–248.

30 GROUND VALIDATION FOR THE GLOBAL PRECIPITATION CLIMATOLOGY PROJECT Mark M. Morrissey and Scott Greene University of Oklahoma, Norman, OK, USA

Abstract

This paper discusses the role of the Surface Reference Data Center (SRDC) and the activities associated with data collection, error characterization, and validation associated with the Global Precipitation Climatology Project (GPCP). Housed in the Environmental Verification and Analysis Center (EVAC) at the University of Oklahoma, the EVAC/SRDC has built upon work from past NOAA-supported projects to become a unique location for scientists to obtain scarce rain gauge data and to conduct research into verification activities. These data are continually analyzed to produce error-assessed rainfall products. Scientists need only to access the EVAC/SRDC web site (http://www.evac.ou.edu/ srdc) to obtain critical global rain gauge data sets. Many of these data sets are impossible to obtain elsewhere. In this paper we will discuss the data collection, analysis, and validation methodology activities of the SRDC.

1 INTRODUCTION During the initiation of the Global Precipitation Climatology Project (GPCP) in the early 1980s it was recognized that confidence could only be placed in a satellite-derived global precipitation analyses if an active verification program was set up concurrently with the GPCP. Thus, the idea of the Surface Reference Data Center (SRDC) was formed and initially located at the US National Climatic Data Center (NCDC) in Ashville, North Carolina. The primary mission of the SRDC at that time was to collect and analyze rain gauge data from special high density and quality networks located around the world. In addition, the data from these networks were to be interpolated in an optimal fashion to produce rainfall estimates of sufficient quality to provide useful comparisons with GPCP estimates at the same time

381 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 381–392. © 2007 Springer.

382

M. M. Morrissey and S. Greene

and space scales. The interpolated rain gauge estimates were placed on the SRDC web site to be downloaded for verification purposes by GPCP scientists. During the mid-1990s the SRDC was transferred to the Environmental Verification and Analysis Center (EVAC) at the University of Oklahoma. EVAC specializes in the development of creative verification methodology, which is primarily stochastic in nature. At this time, the SRDC took on an additional responsibility of performing verification analysis of GPCP products. EVAC specializes in locating and providing hard-to-find rain gauge data sets to the research community. For example, the Comprehensive Pacific Rainfall Database (PACRAIN, Morrissey et al. 1995a), which is part of the SRDC data base, is the most extensive Pacific island rain gauge data base in the world. Data have been collected from hundreds of Pacific island stations, with some records going back as far as the 1800s. It is currently available to scientists via the internet at www.evac.ou.edu/pacrain. Data analysis and quality control of these data are essential and considered an operational task of the SRDC. This is especially important for the Pacific region since data collection there is a variable and sometimes inconsistent process. The main web site for access to the SRDC/EVAC data, analyses, research results, and project descriptions is located at www.evac.ou.edu.

2 DATA COLLECTION AND AVAILABILITY The mission of the SRDC is to collect historical and near-real time rain gauge data only from regions of the world where satellite precipitation verification is especially important, such as in the tropical oceanic regions of the world and where special, high quality, dense rain gauge networks can be found. In addition, these data sets must be independent from the GPCP algorithms. In other words, if the GPCP products are calibrated they should not be calibrated using SRDC data sets. This would, of course, invalidate any verification efforts for such an algorithm. In regions such as the tropical Pacific, dense rain gauge networks are not in existence. However, the SRDC has developed a specialized stochastic method (i.e., the NCR method, see below; Morrissey 1991; Morrissey and Greene 1993) that allows very useful comparisons of the statistical properties of rain gauge-collected data with satellite rainfall estimates. This method was used in the Precipitation Intercomparison Project (PIP-3, Adler et al. 2003) to assess the uncertainties associated with different satellite rainfall estimates over the tropical Pacific region. In regions of the world where high time and space resolution rain gauge networks exist (e.g., the Oklahoma Mesonetwork, Brock et al. 1995), the SRDC maintains a database of interpolated areal estimates of gauge data at scales compatible with the GPCP products. For such networks, the SRDC

Ground Validation for The Global Precipitation Climatology Project

383

has developed stochastic methods to assess the sample error variance and the signal to noise ratio associated these networks (Morrissey et al. 1995b). The mathematical development of these statistics is available on the SRDC web site.

2.1 Pacific rain gauge data Due to important role of the tropical Pacific as the primary driver of the earth’s circulation through the tremendous release of latent heat and its variations, the amount and changes in rainfall in this region must be measured accurately for global climate models to simulate and predict changes in the earth’s weather and climate. These measurements cannot come from the existing rain gauge networks in the Pacific due to the lack of sufficient gauge density. They must come from satellite estimates. However, for verification purposes the existing rain gauge networks are invaluable. A review of the 3rd Algorithm Intercomparison Project (Ebert 1996; Ebert and Manton 1998) indicated that during a comparison of 57 different satellite rainfall algorithms using two shipboard couple ocean atmosphere response experiment (COARE; Webster and Lukas 1992) radars, the radars significantly underestimated rainfall by 30% compared with the average satellite estimate over the same region and time period. A comparison of islandbased rain gauge statistics (Morrissey et al. 1994) showed a strong consistency in the expected rain rate given that it is raining among the four widely separated tipping bucket gauges in the western Pacific. When the expected rain rate given rain is multiplied by the fraction of time raining (within a month) a reasonable estimate of monthly rainfall is obtained. A comparison of monthly rainfall with both the averaged satellite estimates and the COARE radar estimates strongly suggested that it was the radar estimates that contained the largest bias. This was mostly likely the result of calibrating the radar estimates to largely untested optical rain gauges on moored buoys in the western Pacific (Ciesielski 1998). Considering the size of the Pacific basin and importance of rainfall as a tracer of latent heat released, an error of this magnitude most likely results in a similarly sized error in model predictions. Thus, the importance of Pacific rain gauge data cannot be underestimated. The SRDC/EVAC Pacific rain gauge database, PACRAIN, consists of daily and monthly rain gauge data from 643 stations throughout the tropical Pacific, with some records going back into the 1800s. Much of these data have been collected through arrangements with local Pacific meteorological services, as well as New Zealand’s National Institute of Water and Atmospheric Research (NIWA), Meteo-France and US NCDC. The database currently contains over 1.3 million daily observations from 653 sites, extending from 1971 to the present. There are more than 40,000 monthly observations from 201 sites, extending from 1874 through 1970. Daily data

384

M. M. Morrissey and S. Greene

are organized into files by observation site, and monthly data are organized into files by month. There is also an interactive query form, which allows the user to select data based on criteria like date and location. A key component of the SRDC and the Pacrain database is the Schools of the Pacific Rainfall Climate Experiment (SPaRCE; www.evac.ou.edu/ sparce/). The SPaRCE program, funded by NOAA, has over 200 schools, technical centers, and other local organizations across the Pacific interested in taking part in the global climate research effort. Each organization is equipped with a direct read rain gauge, a GPS, a camera, and detailed instructions on setup and maintenance of a professional weather observation site. Several sites are equipped with instruments in addition to rain gauges, such as thermistors and hygrometers. Each local participant group takes daily rainfall measurements, which are quality controlled and then provided to the research community through inclusion in the PACRAIN database. The SPaRCE program also supplies participants with education materials (e.g., books, video tapes) and workshops in an effort to increase awareness of the necessity of enhancing the quality and quantity of Pacific environmental data. It is extremely important to the participants that their efforts make a direct and vital contribution to the global effort of understanding climate change and the potential effects of such a change on their specific locales. The SPaRCE program is an internationally recognized program, and just received an excellent review from the United Nations Economic and Social Commission for Asia and the Pacific (ESCAP) (www.unescap.org/drpad/ vc/conference/ex_pi_167_spr.htm).

2.2 Specialized rain gauge networks In addition to rain gauge data from the Pacific region, the SRDC also archives rainfall estimates from specialized networks located in a variety of countries and the USA. For example, the Oklahoma Mesonetwork (Brock et al. 1995; Morrissey and Greene 1998) has over 100 well-located sites within Oklahoma. The gauge data are collected at 5-min intervals which allows comparisons of analyzed satellite estimates within almost zero time differences. Some of the specialized gauge networks in the SRDC database are located in Darwin, Australia, Florida, Texas, Kenya, Brazil, and a very highdensity gauge network located in South Korea. In addition to these networks, EVAC has installed a very high-density gauge network (i.e., the Piconet) which consists of 15 pairs of gauges, uniformly spaced within 1.0 km area at the Will Rogers World airport in Oklahoma City, Oklahoma. The primary purpose of installing this network was to study the fine scale of rainfall events in an effort to better quantize areal rainfall for larger, less dense gauge networks. A complete description of this network can be found at www.evac.ou.edu/piconet/powerpoint_ams/ piconet_presentation_ files/frame.htm.

Ground Validation for The Global Precipitation Climatology Project

385

3 VALIDATION METHODOLOGIES A main thrust of the SRDC is to provide research in validation methodology as well as provide routine validation of GPCP products. The sections to follow provide an example of some of the statistical validation procedures developed as part of the ground validation efforts of the GPCP. Following that, a few examples of routine validation and simple comparative analysis of rain gauge and GPCP product estimates are shown.

3.1 The spatial sampling error variance A measure of the relative accuracy of a time-space average of areal rainfall taken from a network of rain gauges having any distribution in space can be obtained using a method developed by Morrissey et al. (1995b) based upon a method developed by Parker (1984) for time series. The method provides a baseline error statistic from which to decide whether to utilize a more sophisticated spatial averaging method such as block Kriging (Journel and Huijbregts 1989; Delfiner and Delhomme 1975; Bastin et al. 1984; Bras and Rodriguez-Iturbe 1985; Lebel et al. 1987; Davis 1986). Given a network of gauges, one can place a large square grid containing many sub-grid squares over the network where each sub-grid square contains one or zero gauges. Using this grid setup, the following relation for the error variance was obtained (more details are in Morrissey et al. 1995b):

σ e2 = σ p2 (

m m 2δ ( j ) ρ ( d i , j ) 1 1 m−1 m 2 ρ (d i , j ) + +∑ ∑ − ∑i =1 ∑ j =1 2 m n i =1 j =i +1 m mn

+ ∑i =1

m −1

∑

m j =i +1

2δ (i )δ ( j ) ρ (d i , j )

(1)

n2

where Φ 2p is the variance of the point values (about the long-term mean rainfall), ∆(di,j) is the distance correlation between values located at sub-grid square boxes i (i = 1, 2, … east) and j ( j = 1, 2, south), and ∗(i) is equal to 1 if small box i, j contains a value and zero if it does not. The total number of sub-grid boxes is n and the total number of these boxes that contain a rain gauge is m. The numbering scheme associated with the grid system has no bearing on the resulting standard error equation so long as an appropriate transformation between the index-lagged correction and the distance-lagged correlation is made. Using the method described above, representative locations having low standard error can be selected to perform validation exercises.

386

M. M. Morrissey and S. Greene

3.2 The signal to noise ratio A statistic with which to assess the usefulness of areal rainfall from a gauge network is a combination of two statistics, i.e., the spatial sampling error variance (see above), and the time-space rainfall variance (i.e., Φ 2T). The ratio of the time-space rainfall variance to the spatial sampling error variance (i.e., Φ 2T/Φ 2e) provides an excellent estimate of the signal to noise ratio. The development of the time-space rainfall variance is described by RodriguezIturbe and Mejia (1974) and Morrissey (1991). A detailed description of this method is provided on the SRDC web site.

3.3 The noncontiguous rain gauge method When all that is available within an important region where satellite verification must be conducted is a sparsely distributed network of rain gauges, such as that in the tropic Pacific region, another stochastic method is available for useful assessment of the uncertainty in satellite rainfall estimates. The non-contiguous rain gauge method (i.e., NCR, Morrissey 1991; Morrissey and Greene 1993) uses the second-order statistics of rainfall obtained from widely scattered rain gauges to assess the natural variance of time-space rainfall at the desired scales associated with given values of satellite estimates. This allows one to put error limits on the satellite estimates. The primary assumption required for this method to be valid is that the satellite algorithm uncertainty is constant within the test region. The rain gauge values are also assumed to be homogeneous and stationary. Thus, it is necessary to test for inhomogeneities before this method is applied.

4 EXAMPLE OF VALIDATION WORK: GPCP’S 1.0 DEGREE DAILY AND 2.5 DEGREE MONTHLY RAINFALL PRODUCTS TESTED OVER OKLAHOMA AND THE PACIFIC Much work has been completed comparing the GPCP’s satellite rainfall products over two very different climate regimes, the middle USA (i.e., Oklahoma) and over the tropical Pacific. A summary of a comparison GPCP’s 1.0 degree daily (1DD) products (Huffman et al. 2001) and the GPCP monthly product version 2 (i.e., V2; Huffman et al. 1995, 1997) with gauge data from the Okalahoma Mesonetwork is given below. Boxes at scales comparative with the satellite estimates were selected using the error characterization method described above. The GPCP produces two 1DD products, one which includes information from collocated rain gauge data (i.e., the PSG product) and one that does include this information (i.e., the PMS product). Both products were tested using the Mesonetwork data. It should be noted that the Mesonetwork data are not included in the

Ground Validation for The Global Precipitation Climatology Project

387

development of the PSG product and are thus produce rainfall estimates independent of the PSG product.

4.1 Oklahoma 1.0 degree daily A description of the GPCP 1DD rainfall product is given by Huffman et al. (2001). As an example of the approach the SRDC takes in analyzing the uncertainties associated with satellite estimates, we compared both 1DD product’s values from 1997 to August 2002 with simple areal-averaged rain gauge data from the Oklahoma Mesonetwork shown below. Three 1.0 degree boxes located in northwest, center, and southeast Oklahoma were selected for the comparison (Fig. 1). These boxes were selected due to the relatively low standard error and high signal to noise ratio (see description above).

Figure 1. Three 1 degree boxes contain mesonetwork rain gauges selected for their high signal to noise ratio. The results from each box are shown below using scatter diagrams (the box location is noted in the figure title).

A striking result is that all three boxes indicate very similar correlation and slope values (Fig. 2). A bias appears to be associated with the PSG product with that product overestimating high rainfall. Moreover, the bias which can be observed in each box also appears to be similar in magnitude. This is only an example of the work found at the SRDC web site (additional details and updated results can be found at www.evac.ou.edu/srdc). It should be noted that the non-normality and multi-colinearity inherent in rainfall data strongly suggests that standard linear regression analysis is inappropriate for this type of comparison. Thus, these analyses should only be considered as a first or “quick” look at the comparison between the satellite and rain gauge data.

M. M. Morrissey and S. Greene

388

Figure 2. A comparison between the boxaveraged daily rainfall data from the three Mesonetwork boxes and the GPCP 1 degree daily product.

4.2 Oklahoma 2.5 degree monthly Using the same Mesonet rain gauge data, two 2.5 degree boxes can be used to compare with the GPCP monthly product (Fig. 3). The GPCP V2 product data were compared with box-averaged gauge data from 1994 to August 2002. The results in scatter diagram form are shown in Fig. 4.

Figure 3. The two 2.5 degree monthly boxes selected for the comparison with the monthly GPCP product.

Ground Validation for The Global Precipitation Climatology Project

389

Figure 4. A comparison between the satellite and Mesonetwork box averages for the GPCP version 2 monthly product.

The monthly product over Oklahoma appears to be relatively unbiased and relatively accurate compared to the 1DD product. However, at large satellite value the Mesonetwork box averages tend to underestimate rainfall compared to the satellite value. Too much cannot be read into the accuracy of the product from this preliminary comparison due to de the reasons mentioned above. However, these “quick looks” do suggest areas of further, more in-depth research. Time series and error and bias comparison are also available on the SRDC web site.

4.3 PACRAIN 2.5 degree monthly Three boxes were selected within the Pacific region (Fig. 5) which contained a sufficiently low standard error and significantly high signal to noise ratio to allow a quick look at the accuracy of the monthly GPCP V2 product. It should be noted that gauge data are not assimilated into the monthly GPCP product. The scatter plots of the comparison for the three boxes are shown in Fig. 6. The record of comparison was from 1979 to August 2002. An initial analysis of the scatter diagrams suggests a statistically unbiased relationship with the possible exception of box 1,725. A quick look into the gauge data records for that box indicates that the records are quite sporadic and that the comparison may have been contaminated with poor gauge data and low spatial temporal and sampling. Assuming accurate gauge measurements, another look at the results seem to suggest perhaps a small underestimate by the GPCP monthly product of approximately 500 mm per month. While these comparisons are not statistically rigorous, they do provide the producers of the GPCP satellite algorithms with a quick look at the comparisons and allow them to pursue more stringent research methods if necessary.

390

M. M. Morrissey and S. Greene

Figure 5. Tropical western Pacific region. Outlined boxes are those having the highest signal to noise ratio for the monthly data. The crosses indicate rain gauge sites.

Figure 6. The scatter plots associated with each of the three Pacific 2.5 degree monthly boxes.

Ground Validation for The Global Precipitation Climatology Project

391

5 SUMMARY The SRDC’s role in the GPCP is to provide researchers with hard-to-find quality rain gauge data sets and to provide initial comparisons between the GPCP products and analyzed and interpolated rain gauge data sets. In addition, the SRDC researchers actively produce statistical methods which researchers can use to assess the usefulness of a set of rain gauge data and to perform statistical comparison of satellite estimates with sparsely distributed rain gauge data in region like the tropical Pacific.

6 REFERENCES Adler, R. F., C. Kidd, M. Goodman, A. Ritchie, R. Schudalla, G. Petty, M. Morrissey, and S. Greene, 1996: PIP-3 Intercomparison Results. Precip. Intercomparison Proj (PIP-3) Workshop, 18–20 November 1996, College Park, MD. Bastin, G., B. Lorent, C. Duqué, and M. Gevers, 1984 Optimal estimation of the average areal rainfall and optimal selection of rain gauge locations. Water Resour. Res., 20, 463–470. Bras, R. L. and I. Rodriguez-Iturbe, 1985: Random Functions and Hydrology, Dover, NY, 559 pp. Brock, F. V., K. C. Crawford, R. L. Elliott, G. W. Cuperus, S. J. Stadler, H. L. Johnson, and M. D. Eilts, 1995: The Oklahoma Mesonet: a technical overview. J. Atmos. Oceanic Technol., 12 (1), 5–19. Ciesielski, P. E. and R. H. Johnson, 1999: Precipitation estimates for the COARE intensive observing period. Proc. Conf. on the TOGA Coupled Ocean-Atmosphere Response Experiment (COARE98), Boulder CO, 7–14 July 1998, pp. 193–194, WCRP-107, Geneva, Switzerland. Davis, J. C., 1986: Statistics and Data Analysis in Geology. John Wiley, New York, 646 pp. Delfiner, P. and J. P. Delhomme, 1975: Optimum interpolation by kriging. In: Display and Analysis of Spatial Data, edited by J. C. Davis and M. J. McCullagh, John Wiley, New York, pp. 96–114. Ebert, E. E., 1996: Results of the 3rd algorithm intercomparison project (AIP-3) of the Global Precipitation Climatology Project (GPCP). BMRC Report No. 55, BMRC, GPO Box 1289K, Melbourne, Vic., Australia 3001, 199 pp. Ebert, E. E. and M. J. Manton, 1998: performance of satellite rainfall estimation algorithms during TOGA-COARE. J. Atmos. Sci., 55, 1537–1557. Huffman, G. J., R. F. Adler, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, and J. Susskind, 2001: Global precipitation at one-degree daily resolution from multiobservations. J. Hydrometeor., 2, 36–50. Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider, 1997: The Global Precipitation Climatology satellite Project (GPCP) combined precipitation dataset. Bull. Amer. Meteor. Soc., 78, 5–20. Huffman, G. J., R. F. Adler, B Rudolf, U. Schneider, and P. R. Keehn, 1995: Global Precipitation estimates based on a technique for combining satellite-based estimates, raingauge analysis and NWP model precipitation information. J. Climate, 8, 1284–1295. Janowiak, J. E. and P. A. Arkin, 1991: Rainfall variations in the tropics during 1986–1989. J. Geophys. Res., 96, 3359–3373. Janowiak, J. E., P. A. Arkin, P. Xie, M. L. Morrissey, and D. R.Legates, 1995: An examination of the east Pacific ITCZ rainfall distribution. J. Climate, 8, 2810–2838. Journel, A. G. and C. J. Huijbregts, 1989: Mining Geostatistics. Academic Press, San Diego, CA, 600 pp.

392

M. M. Morrissey and S. Greene

Lebel, T., G. Bastin, C. Obled, and J. D. Creutin, 1987: On the accuracy of areal rainfall estimation: a case study. Water Resour. Res., 23, 2123–2134. Morrissey, M. L, 1991: Using sparse rain gages to test satellite-based rainfall algorithms. J. Geophys. Res., 96, 18561–18571. Morrissey, M. L. and J. S. Greene, 1993: Comparison of two satellite-based rainfall algorithms using Pacific atoll rain gage data. J. Appl. Meteor., 32, 411–425. Morrissey, M. L. and Y. Wang, 1995: Verifying satellite microwave rainfall estimates over the open ocean. J. Appl. Meteor., 34, 794–804. Morrissey, M. L., M. A. Shafer, S. E. Postawko, and B. Gibson, 1995a: Pacific rain gauge data. Water Resour. Res., 31, 2111–2113. Morrissey, M. L., J. A. Maliekal, J. S. Greene, and J. Wang, 1995b: The uncertainty of simple spatial averages using rain gauge networks. Water Resour. Res., 31, 2011–2017. Parker, D. E., 1984: The statistical effects of incomplete sampling of coherent data series. J. Climatol., 4, 445–449. Postawko, S. E., M. L. Morrissey, and B. Gibson, 1994: Schools of the Pacific Rainfall Climate Experiment: combining research and education. Bull. Amer. Meteor. Soc., 75, 2296–2311. Rodriguez-Iturbe, I. and J. M. Mejia, 1974: The design of rainfall networks in time and space. Water Resour. Res., 10, 713–728. Webster, P. J. and R. Lukas, 1992: TOGA COARE: The Coupled Ocean-Atmosphere Response Experiment. Bull. Amer. Meteor. Soc., 73, 1377–1416.

31 VALIDATION OF RAINFALL ALGORITHMS AT THE NOAA CLIMATE PREDICTION CENTER John Janowiak NOAA Climate Prediction Center, Camp Springs, MD, USA

1 INTRODUCTION The validation of precipitation estimates from satellite is an essential activity for numerous reasons, among them being to assess the skill of the estimation algorithms, to provide users with the accuracy of them, and to provide feedback to algorithm developers that may potentially help improve their methodology. The primary motivation for the validation activity that will be discussed in this paper is directed toward the latter, i.e., to provide useful feedback toward the improvement of satellite estimates of precipitation. Validation activities can be conducted at various time and space scales depending on the availability of reference data sets that are suitable to be used as “truth”. Ideally, these reference data sets have known error characteristics that can be incorporated into the validation process. In reality, however, this information is not widely available and is limited to relatively small-scale regions with high rain gauge density. In this paper, the focus is on a continental-scale validation effort over the USA. The system that has been implemented for the USA has been modeled after the excellent work of Dr. E. Ebert of the Australian Bureau of Meteorology Research Center (BMRC) who implemented a validation system over Australia for the International Precipitation Working Group (IPWG) and which is discussed in a separate paper of this section.

2 VALIDATION DATA Two sources of precipitation validation data are available over the USA. One is composed of rain gauge data while the other is radar. Rain gauges provide 393 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 393–401. © United States Government 2007.

J. Janowiak

394

the only totally direct measurement of precipitation at the surface and provide measurements that are continuous in time. However, rain gauges provide point measurements that are not spatially complete and formidable gaps in coverage exist in certain areas of the country (Fig. 1). Conversely, radar provides much better spatial coverage than gauges and the spatial coverage characteristics are quite similar to the satellite estimates, although radar does not provide a direct estimate of precipitation. By comparing the satellite estimates to both the rain gauge and radar data, a more complete validation exercise can be conducted than if only one source were used.

Figure 1. Typical distribution of rain gauge data.

2.1 Rain gauge analyses The main validation data that are used for this effort are objective analyses of over 7,000 rain gauge observations over the continental USA that provide rainfall totals over the 24-h period from 1,200 UTC to 1,200 UTC. These observations are objectively analyzed to a 0.25° × 0.25° latitude/longitude grid using a modified Cressman (1959) technique. While it is widely known that this analysis procedure tends to spread out light values of rainfall and also to deamplify heavy values, this rainfall validation data set provides the best available nationwide precipitation information on a daily basis in the USA at this point in time. A typical distribution of the rain gauge locations that are input to the analysis scheme is shown in Fig. 1. As part of the analysis process,

Validation of Rainfall Algorithms at the NOAA Climate Prediction Center

395

various types of quality control are performed, including climatological checks, “buddy” checks (i.e., comparisons with neighboring values), and a check against IR satellite data to ensure that clouds are present when rainfall has been reported. Further information is available from Higgins et al. (2000).

2.2 Radar All of the satellite-generated rainfall estimates are validated against the NEXRAD “Stage II” radar product that is produced by NOAA/NWS in addition to the rain gauge analyses. The Stage II radar estimates are not integrated with rain gauge observations, although a large-scale bulk correction based on rain gauge data is used to reduce bias in these estimates (R. Kuligowski, personal communication, Camp Springs 2005). Although the radar estimates are not considered as accurate as data from rain gauges, the spatial density that radar offers is more closely matched to the satellite estimates. The radar data are interpolated to a 0.25° × 0.25° latitude/longitude grid that matches the rain gauge analysis grid.

3 SATELLITE AND NUMERICAL MODEL ESTIMATES At the time of this writing, 14 different satellite estimates are validated along with short-term forecasts of precipitation from two numerical forecast models. The precipitation estimates are interpolated to match the validation grid, if necessary, using a Bessel interpolation scheme. There is a variety of information that is used by the various satellite precipitation estimation techniques. Some use only passive microwave information, some use only infrared data, and others use both. A list of the precipitation estimates that are presently validated is presented in Table 1. The validation system is designed with flexibility so that results from new algorithms can be incorporated rather easily.

4 VALIDATION STATISTICS AND RESULTS The validation over the USA is conducted on a continental scale, i.e., comparisons are made for the USA as a whole (excluding Alaska and Hawaii). Validation statistics that use the rain gauge analyses as the reference standard also include the northern half of Mexico. Only those locations where nonmissing estimates are available for all of the satellite techniques are used when computing statistics to ensure that the same domain is validated for each of the estimates. This is important over the USA particularly in winter because some techniques cannot provide precipitation estimates over snow-covered surfaces. Radar estimates are used both as a validating tool, and as a “contestant” in which the radar estimates are compared to the rain gauge analyses. The

J. Janowiak

396

statistics discussed in the following section are very elementary tools that are described in most statistics texts. The definitions below are condensed from Wilks (1995). Table 1. List of products that are presently being validated over the USA (TRMM Tropical Rainfall Measuring Mission (Simpson et al. 1988)).

Product name IR, passive microwave-blended methods: NRLGEO 3B42RT CMORPH PERSIANN Passive microwave only MWCOMB 3B40RT NRBLD AMSU IR only GPI IRRAIN HYDROE 3B41RT IR and visible data GMSRA GMSRAD NWP model forecasts GFS NOGAPS

Source, definition Turk et al. (2003) NASA/TRMM: merge of products “3B40RT” and “3B41RT” Joyce et al. (2004) Sorooshian et al. (2000) NOAA/CPC: merge of SSM/I, TMI, and AMSU-B precipitation estimates NASA/TRMM: merge of SSM/I and TMI-derived precipitation estimates Naval Research Lab: merge of SSM/I, TMI, and AMSU-B precipitation estimates Ferraro et al. (2000) Arkin and Meisner (1987) NOAA/CPC NOAA/NESDIS – also uses model data NASA/TRMM: IR estimates calibrated by passive microwave Ba and Gruber (2001) NOAA/NESDIS: also uses radar data NOAA/NWS: Global Forecast System model (formerly “MRF”) US Naval Research Laboratory global forecast model

4.1 Statistics for assessing errors in magnitude Mean rain rate. This is simply the average of the daily mean rainfall rate over the entire validation domain. The difference between the mean of the remotely sensed estimates and observations is the mean bias. Mean absolute error. This statistic is the country-wide average of the absolute difference (i.e., negative differences are changed to positive) between the estimates and observations. Absolute errors retain the differences in magnitude that would otherwise be reduced because positive and negative differences would cancel each other to some degree.

Validation of Rainfall Algorithms at the NOAA Climate Prediction Center

397

Root mean square error. Similar to mean absolute error, except that the differences are squared before summing, and the square root of the average is derived. Maximum rain rate. This is simply the largest value recorded each day regardless of location. This statistic is useful to define the range of values of the satellite estimates compared to observations when collected over a sufficient length of time.

4.2 Statistics for assessing errors in the spatial distribution of precipitation All of the statistics discussed in this section, with the exception of spatial correlation, are computed from a 2 × 2 contingency table that summarizes “hits” and “misses” for two categories: (1) daily rainfall amounts less than 1 mm and (2) daily rainfall of 1 mm or more. These statistics are used to assess the rainfall detection aspects of the estimates and not the magnitudes. Spatial correlation. This is the correlation between an estimated field and observed field that is conducted on daily data which yields information about the degree of agreement in the spatial variability of the two fields. Probability of detection (POD). The ratio between the number of occurrences where satellite estimates of rainfall > 1 mm day–1 were correctly observed and the total number of observations of rainfall exceeding that threshold. Values range from the worst possible score of “0” (precipitation never correctly detected by satellite) to the best possible score “1” (precipitation always correctly detected by satellite). False alarm ratio (FAR). The ratio between the number of satellite estimates of rainfall >1 mm day–1 that were detected incorrectly and the total number of satellite estimates that exceeded that threshold (whether correct or not). Values range from a perfect score of “0” (precipitation never observed when not detected by satellite) to the worst possible score of “1” (precipitation never observed when detected by satellite). POD and FAR should always be considered together, because good or even perfect values in either case individually are easily obtained. For example, if an algorithm has a wet bias such that it always detects rain everywhere all the time, the POD will be perfect (“1”). However, the FAR will also be close to the worst score (“1”). Bias ratio. The ratio between the number of satellite estimates > 1 mm day–1 and the number of observed amounts that exceeded that threshold. A value of “1” is perfect. Skill score. A score that attempts to assess the skill of the estimates with respect to random chance. The scores usually range from “0” (no skill over chance) to “1” (perfect skill).

398

J. Janowiak

Figure 2. Time series of statistics of a comparison with validating rain gauge analyses over the USA during June–August, 2003. Thick line is radar, dotted line is GFS model forecasts, thin solid line is the satellite estimate with the best statistic for each day.

4.3 Validation results A comparison of several estimates of precipitation with the validating rain gauge analyses over the USA are now discussed. Results are presented for both the warm season, when precipitation is primarily convective in nature, and for one cool season when stratiform precipitation dominates. The warm season results, which are displayed in Fig. 2, show time series of spatial correlation and Heidke skill score for radar (thick solid line), the NWS/NCEP global forecast model (GFS) 12–36-h precipitation forecast (dotted line), and the best score by any of the satellite estimates for a given day (thin solid line). For almost every day during June–August 2003, the radar performs best and the model forecasts worst compared to the rain gauge analyses. Note that the satellite estimates are very close to the radar values in both statistics over the entire 92-day period. In contrast, the GFS model predictions perform much better during the cool season (Fig. 3) and the performance measures are much closer among the radar, satellite, and model. In fact, the model forecasts often outperform the radar and satellite estimates both in terms of spatial correlation and skill during the cool season.

Validation of Rainfall Algorithms at the NOAA Climate Prediction Center

399

Figure 3. Same as Fig. 2, except for September–November.

Over the course of analyzing the validation results during the US summer season, a consistent bias in the satellite estimates was “rediscovered”, namely persistent overestimates in the western states. This observation is depicted in Fig. 4, which depicts areas of eastern Montana with rainfall amounts in excess of 40 mm day–1 from the satellite estimates but amounts of less than 5 mm day–1 from the rain gauge data. Note that radar also overestimates considerably although the amounts are somewhat lighter than the satellite estimates. To ensure that the gauge analysis was not in error, the gauge results were verified by contacting the Glasgow, Montana NWS forecast office, who in turn verified the precipitation measurements of cooperative observers in the region. This “rediscovery” is consistent with the earlier studies of Scofield (1987), Rosenfeld and Mintz (1988), and more recently McCollum et al. (2001) who found that significant evaporation occurs in semiarid regions between the cloud base and surface. In fact, Rosenfeld and Mintz (1988) estimate conservatively that 30% of the rainfall evaporates in the first 1.6 km below the cloud base in semiarid regions at rainfall intensities as high as 80 mm h–1. One way to account for this overestimation is to use relative humidity data to modulate the rainfall estimates. Scofield (1987) adopted this approach by using the mean humidity from the surface to 500 hPa from numerical forecast model analyses.

400

J. Janowiak

Figure 4. Rainfall over the 24-h period 1200 UTC 12 August 2003–1200 UTC 13 August 2003.

5 SUMMARY The example above is just one way that information from continental-scale validation efforts as described in this paper can provide helpful feedback to algorithm developers who can then modify and improve their estimation techniques. A similar continental-scale validation effort is underway over Australia and in the planning stages over Europe and Brazil. Certainly, several such efforts over different climatological regions have the potential to provide substantial useful feedback that will help the precipitation estimation algorithm community.

Validation of Rainfall Algorithms at the NOAA Climate Prediction Center

401

There are obviously other aspects of validation that can be conducted than have been presented here. In addition to such large-scale examinations, comparatively small-scale validation efforts over regions with dense groundbased validation data and known error characteristics would complement this validation effort in a positive fashion. Efforts such as these are underway in the USA under the direction of the IPWG and the Global Precipitation Climatology Project (GPCP), both of which are sponsored by the World Meteorological Organization.

6 REFERENCES Arkin, P. A. and B. N. Meisner, 1987: The relationship between large-scale convective rainfall and cold cloud over the western hemisphere during 1982–1984. Mon. Wea. Rev., 115, 51–74. Ba, M. B. and A. Gruber, 2001: GOES multispectral rainfall algorithm (GMSRA). Bull. Amer. Meteor. Soc., 40, 1500–1514. Cressman, G. F., 1959: An operational objective analysis system. Mon. Wea. Rev., 87, 367– 374. Ferraro, R. R., F. Weng, N. C. Grody, and L. Zhao, 2000: Precipitation characteristics over land from the NOAA-15 AMSU sensor. Geophys. Res. Lett., 27, 269–2672. Higgins, R. W., W. Shi, E. Yarosh, and R. Joyce, 2000: Improved United States precipitation quality control system and analysis. NCEP/Climate Prediction Center ATLAS No. 7, 40 pp., Camp Springs, MD 20746 USA. Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie, 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydromet., 5, 487–503. McCollum, J. R., W. F. Krajewski, R. R. Ferraro, and M. B. Ba, 2002: Evaluation of biases of satellite rainfall estimation algorithms over the continental United States. J. Appl. Meteor., 41, 1065–1080. Rosenfeld, D. and Y. Mintz, 1988: Evaporation of rain falling from convective clouds as derived from radar measurements. J. Appl. Meteor., 27, 209–215. Scofield, R. A., 1987: The NESDIS operational convective precipitation estimation technique. Mon. Wea. Rev., 115, 1773–1792. Simpson, J. R., R. F. Adler, and G. R. North, 1988: A proposed Tropical Rainfall Measuring Mission (TRMM) satellite. Bull. Amer. Meteor. Soc., 69, 278–295. Sorooshian, S, K. Hsu, X. Gao, H. V. Gupta, B. Imam, and D. Braithwaite, 2000: Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteor. Soc., 81, 2035–2046. Turk, F. J., E. E. Ebert, B.-J. Sohn, H.-J. Oh, V. Levizzani, E.A. Smith, and R. Ferraro, 2003: Validation of a global operational blended-satellite precipitation analysis at short time scales. 12th AMS Conf. on Sat. Meteor. and Ocean, CD-ROM, 13–17 February, Long Beach, CA. Wilks, D. S., 1995: Statistical methods in the atmospheric sciences. Academic Press, San Diego, CA, 467 pp.

32 GROUND NETWORKS: ARE WE DOING THE RIGHT THING? Witold F. Krajewski IIHR-Hydroscience & Engineering, University of Iowa, Iowa City, IA, USA

Abstract

The author discusses principal issues in designing ground-based networks of rainfall observing instruments for validation of space-based methods of rainfall estimation. He considers tipping bucket rain gauges, disdrometers, vertically pointing and X-band polarimetric scanning radars. The design issues emphasize the need for studies of small-scale rainfall variability, characterization of the measurement error of the instruments used, and observations of spatial and temporal variability of rainfall characteristics at scales smaller than those of the space-based methods.

1 INTRODUCTION As this entire volume testifies, there is a lot of excitement in the science community about using satellites for observing rainfall from space. There are many scientific applications where such observations offer new insights into the workings of the coupled atmosphere, land, and ocean system. For many of those applications, assessment of our ability to quantify rainfall using the signal measured by the satellite-based sensors is not critical. For other applications, such as agriculture, prediction of precipitation-induced natural hazards, and operation of water resources systems, considerable level of quantitative performance is necessary (e.g., Nijssen and Lettenmaier 2004.) The term validation is often used as a synonym of our attempts to evaluate algorithm or model performance. However, despite its popular use few authors bother to define the precise meaning of the term in their studies. To some, validation is the process of quantitative evaluation of the outcomes of algorithms and models and their subsequent improvements. Others view it as a statistical characterization of the outcome’s uncertainty (Krajewski and Smith 2002; Gebremichael et al. 2003). In this article, we adopt the latter view as 403 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 403–417. © 2007 Springer.

404

W. F. Krajewski

availability of such statistical description of outcome’s uncertainty provides a reference against which improvements can be measured. As the focus of this volume is on precipitation, we refer to the outcomes as precipitation products or maps. It is thus clear that mere comparison of products of two or more different algorithms cannot be considered as validation in the above sense. Still, such exercises are valuable and we will term them assessment or evaluation. These two terms are broad enough to include many different studies where the precise meaning of the nature of the investigation is not critical.

2 OBJECTIVES AND SCOPE Based on the above discussion, we define the objectives of this paper. We discuss a variety of issues related to the use of ground-based sensors as they promise to provide information relevant to validation of space-based rainfall products. We emphasize the estimation of rainfall rather than snowfall but much of the discussion is relevant to both. We set forth as the goal for validation to be the determination of the space-time joint distribution of the product errors. Determination of probability distribution function (pdf) of errors for a fixed space and time scale, the pdf conditional on rainfall intensity or accumulation for a fixed space and time scale, the spatial correlation function for fixed space and time scale and the temporal correlation function for fixed space and time scale are all example of special cases of the general problem and thus somewhat simpler. This by no means implies that these problems are simple to solve as we are still far away from solving any of them. In essence, this paper presents our view of a research agenda necessary for a practical solution of the validation problem for satellite-based rainfall estimation much as Krajewski and Smith (2002) did for radar-rainfall estimation. Before we proceed, let us define what we mean by error. Two fundamental quantities of interest are the difference and the ratio between the true and estimated quantity. The choice between them is not obvious and depends mainly on whether the mechanism causing error is additive or multiplicative. This is often unknown and therefore we recommend analysis of both, whenever possible. This approach has always provided interesting insights. Mathematical convenience is often a useful guiding aspect, which is fine, as long as it does not lead to unverifiable assumptions. As a general principle, based on the central limit theorem, uncertainties of products at large scale tend to be Gaussian and thus amenable to additive error models. How large is large is not well explored in the case of rainfall and thus, to some degree, is a subjective matter. The error of satellite rainfall products can be conceptualized as having three basic components: sampling, estimation, and instrumental effects. These components are not independent. Instrumental effects are typically small, which does not mean that they are negligible. The estimation error

Ground Networks: Are We Doing the Right Thing?

405

includes the conversion of the measured signal into the estimation quantity and thus contains algorithm error, the propagated effect of the instrumental error, the spatial sampling (known as field of view) error and the georeferencing error. If the product is rainfall accumulation over some time scale (e.g., daily), it also contains the temporal sampling error. Thus, the sampling error does not affect other errors if the product is rainfall intensity. Many studies have been conducted on the sampling error (e.g., Bell et al. 1990; North et al. 1993; Oki and Sumi 1994; Bell and Kundu 1996, 2000) of time-integrated products. Most assumed perfect (error-free) rainfall intensity fields. Relatively few studies address the total error (Krajewski et al. 2000; Gebremichael et al. 2003) or error decomposition (e.g., Short and North 1990; North and Polyak 1996; McCollum and Krajewski 1998). Since it is the total error that is of interest to users of products, we focus on a strategy for obtaining its statistical description using ground-based networks of sensors. In our discussion we emphasize the instrumental and estimation error as opposed to the sampling error on the basis that the latter has been studied much more extensively. Establishing ground-based observing systems for quantitative assessment of satellite-based estimates of rainfall is fundamentally important and challenging (e.g., North and Nakamoto 1989). According to our definition of validation, ground-based estimates of rainfall should serve as a reference for error studies of space-based products only if their own estimation error can be characterized, or if it can be shown that they are much more (order of magnitude) accurate than the space-based products. There are several types of sensors that need to be considered: single- and multiparameter scanning radar, vertically pointing radars (i.e., profilers) with their ability to provide detailed vertical view of precipitating atmosphere, disdrometers, and networks of rain gauges. Clearly, scanning radar offers probably the most promise due to its ability to provide spatially continuous and temporally frequent observations of rainfall. However, the situation with validation of its products is about the same as we discussed for the satellite case: we do not know the error structure of radar-rainfall products. Since Krajewski and Smith (2002) and Krajewski and Ciach (2003) provide recent relevant discussion, we limit our scope to other sensors and issues. The exception is networking of small, inexpensive radars which we discuss in Section 5.

3 RAIN GAUGE NETWORKS We begin with rain gauge networks and the characteristics they should have to serve as useful reference for satellite-based products. First, we address the issue of random error of tipping bucket rain gauge. To date we know of two rigorous studies on this subject. Habib et al. (2001) performed a data-based simulation and estimated the error distributions as function of rainfall

406

W. F. Krajewski

magnitude, time integration scale, and bucket size. Recently, Ciach (2003) conducted an experimental study and synthesized the results for two different methods of tip data processing. His results are in good agreement with the earlier study by Habib et al. (2001). A very interesting result is the scaling behavior of the errors. These results are also confirmed by analysis from the double gauge data at the Iowa City Municipal Airport. The main conclusion from these studies is that tipping bucket rain gauges, when well maintained and deployed as a pair, provide accurate observation of rainfall accumulations at scales from 10 min up. The standard errors decrease with increasing rain amount and time integration scale. Deploying the tipping bucket rain gauges in pairs, advocated by Ciach and Krajewski (1999) and Steiner et al. (1999), has many advantages. The most important one is data quality control. Since rainfall displays significant variability in space and time, analysis of “reasonableness” of single gauge record often passes cases of gauge malfunctioning. For example, a piece of dirt partially blocking inlet of the funnel will change estimated rain intensity without making it look suspicious. The main assumptions of the concept of using two gauges at a single location are: (1) it is highly unlikely that two gauges would fail in exactly the same way; and (2) that rainfall variability at the scale of gauge separation distance (~1 m) is negligible. Thus, when the gauges function well the data show good agreement with each other. A disagreement is a sign of at least one of them malfunctioning. The site should be checked by a qualified technician. Note, that if malfunction can be attributed to one of the gauges, the data is not lost at the given site as the second gauge worked properly. An added benefit of having a pair of gauges is a reduction of the random error (Ciach 2003). The second issue is that of designing networks of gauges for validation studies. Operational networks are of little help as their average separation distance is typically larger than the scale of interest to validation. The density of operational rain gauge networks varies from country to country, but it is safe to assume that it is rarely higher than 0.002 km–1. As satellite maps of rainfall can have a resolution on the order of 25 km2 , such density is clearly not adequate. Consider two objectives for design of validation networks: (1) characterizing statistical behavior of rainfall in space and time; and (2) estimating rainfall over an area as accurately as possible. It turns out that these two objectives lead to quite different network configurations. To illustrate this, let us pose a question relevant to direct validation (Krajewski and Smith 2002): “How many gauges are required in a 2 × 2 km2 area (equivalent to a typical radar-rainfall product) to obtain areal estimate with high accuracy (say, better than 5%)?” If the accuracy is specified in terms of mean square error, answering this question requires knowledge of the spatial covariance function of the relevant rainfall regime. To estimate the shape of covariance (or correlation) function of rainfall intensity or accumulation over short time

Ground Networks: Are We Doing the Right Thing?

407

scales the network has to sample several separation distances. In particular, accuracy of gauge-based estimated rainfall over an area comparable is size (linear) with the decorrelation distance of rainfall is sensitive to the shape of the covariance function near the zero separation distance (see Habib and Krajewski (2002) for an illustration). Thus, the network has to include gauges that are close to each other as well as far from each other. Formal statement of the design problem is not straightforward especially when little is known about the shape of the function the network is supposed to identify. Prior experience is limited but it could be helpful. Habib and Krajewski (2002) estimated correlation functions for Florida summer rainfall. Krajewski et al. (2003) estimated it for five different locations around the world and reviewed early experiments (in the 1960s and 1970s) motivated by the needs of the communication community (Crane 1990). However, these studies were based on short-lived experiments and the analyses are subject to significant sample size and other limitations. In particular, short-duration experiments and thus small samples prevent conditional analyses. Identifying the effects of rainfall intensity and amount, season, rainfall type, and linking these to synoptic situation and other observables is important for the validation problem but impossible with small samples. For example, when designing an experimental network at the Iowa City Municipal Airport in 1998, we included inter-gauge distance as short as 10, 100, 200, and 500 m. We have learned since that there is little variability at the scale of 200–500 m. Based on our experience thus far, distances as short as 250 m should be included in networks designed for studying the shape of covariance function in the tropics and 500 m in midlatitudes. Other statistics may require different separation distance considerations. Now let us turn our attention to the second objective, i.e., estimating rainfall over an area. Here, the network design problem has been well studied in the past as hydrologists need mean areal rainfall as input to their rainfall-runoff models (e.g., Bras and Rodriguez-Iturbe 1985). In general, the gauges should be organized on a uniform grid covering the area of interest. Such design maximizes rain cell detection. Estimation of network sampling errors associated with simple averaging of rain gauge values can be accomplished numerically using the methodology proposed by Morrissey et al. (1995). Thus, questions on network density required for achieving certain levels of accuracy, or questions of network expansion to improve accuracy can be easily studied if the rainfall spatial covariance function is known. Based on earlier studies of such function (e.g., Krajewski et al. 2003), we attempt to answer the question of the number of gauges needed to estimate rainfall over a 2 × 2 km2 pixel with accuracy better than 5%. We consider two cases: exponential decay with the correlation distance of 5 and 15 km. In the first case we need about 20 gauges uniformly covering the pixel to reach the 5% error level. In the second case only 5–8 gauges will achieve the same objective (Moore et al. 2000). The pixel size is more appropriate to the

408

W. F. Krajewski

validation of radar-rainfall problem. For the scale of a satellite rainfall product pixel (about 5 × 5 km2), 5% error level can be achieved with about 35 gauges for highly variable rainfall and with only about 10–12 for the less variable regime typical of midlatitudes. We present these cases to show that solving the validation problem is both feasible and not terribly expensive (i.e., compared to the cost of building and maintaining a satellite). Rain gauge networks of the size we mentioned above can be easily deployed and operated. As a matter of fact, the smaller the area the easier and cheaper it is to maintain a network. For example, clusters of the 2 × 2 km2 size can be placed at the airport where many other meteorological instruments are often placed. Cell phone-based data communication provides a simple way of near real-time operation and dual-gauge design provides means for demand-only maintenance visits (Kruger and Kanukurthy 2004). Since optimal network configurations may be difficult to archive in practice due to constraints imposed by the existing infrastructure or its lack, it is useful to study the sensitivity of a particular design using geographic information system (GIS) technology. Krajewski and Goska (2004) developed an ArcView GIS utility that calculates mean square error associated with a certain network configuration over an arbitrarily shaped area. Another rain gauge network design objective results from considerations of ground networks that include scanning radar. Although the subject of radar-rainfall estimation is outside the scope of the current paper, it is well known that radar-rainfall uncertainty depends on radar range. Thus, if we are to benefit from radar-estimated rainfall for the validation of satellite rainfall products, we need to organize observational networks that can address the radar-range effects (Krajewski and Ciach 2003). A simple solution is placing dense clusters along a radar beam. The question is “How many?” and “How should they be spaced?” The answer depends on the operating range of a given radar. There are two major effects that the spacing should be able to capture. The first is the detection of the so-called bright band (reference). In some rainfall regimes the effect may easily take place at a radar range shorter than 100 km. The second effect is overshooting precipitating clouds. The range of this effect is longer than the bright band effect and depending on the rain regime may take place anywhere between 100 and 200 km. Studies by Smith et al. (1996) demonstrate the shape of the range-dependent bias in early NEXRAD system estimates of rainfall in the USA while the theoretical considerations made by Krajewski and Vignal (2004) give a more general means of estimating both bias and error variance. Their model could be used as guidance in designing the cluster locations to capture the range effect shape. Another solution would be to place small cluster of about four sites every 10 km along the radar range.

Ground Networks: Are We Doing the Right Thing?

409

4 DISDROMETER NETWORKS Rain gauge networks provide surface-based information on the main variable of interest, i.e., rainfall on the ground as this is the variable that affects hydrologic response and thus is of concern to operators of water resources systems. However, much more general information about rainfall processes is included in the measurements of surface drop size distribution (DSD). From DSD data, not only point rainfall can be calculated, but also many other quantities relevant to remote sensing of rainfall and its hydrologic applications. For example, one can calculate variables such as radar reflectivity, optical extinction, and kinetic energy. Thus, why not simply replace the rain gauges in the experimental (i.e., validation) networks with disdrometers? The main answer is the cost of the instruments. The cost of a disdrometer may be an order or even two orders of magnitude higher than the cost of a tipping bucket rain gauge. This situation, however, is quickly changing. There are several new models of optical disdrometers on the market that are priced competitively with some high-end rain gauges. There are also efforts within the research community to further drive this cost down by exploring new design ideas. Another important consideration in wide-scale deployment of disdrometers is their power consumption and operation. Virtually all current design requires power supply and housing for a computer that control data acquisition system. Clearly, for the devices to be ready for remote operation both issues need to be addressed. The devices need to be low-power so that they can be operated from a solar-panel rechargeable battery and have to have a computer processor and data acquisition system embedded in their design. Technologies for addressing both issues exist and will not increase the overall cost of the devices significantly. Since disdrometers measure DSD indirectly and the cumulative experience with their operation is much less than in the case of rain gauges, they require thorough testing. Several intercomparison experiments point to sensitivity of the obtained results to the instrument type (Sheppard and Joe 1994; Campos and Zawadzki 2000; Williams et al. 2000; Tokay et al. 2001; Miriovsky et al. 2004.) These experiments used different types of instruments collocated or in a close proximity of each other. To distinguish the sampling error effects associated with a particular instrument from the crossinstrument differences it is necessary to compare several collocated disdrometers of the same kind. Ciach (2003) conducted and documented such an experiment using tipping bucket rain gauges but we do not know of a similar experiment using disdrometers. Another issue is the limited sampling area of the present-day disdrometers. The sampling area of some 50 cm2, typical for optical devices causes significant sampling errors (Smith 1993; Jameson and Kostinsky 2002) Mini radar-type devices such as the POSS (Sheppard 1990) suffer less from this

410

W. F. Krajewski

problem but require making assumptions on the velocity versus drop size relationship necessary for estimating the DSD. This is an important issue as some studies indicate often significant departures of the measured velocity from the often-used experimentally determined terminal velocity relationships (e.g., Gunn and Kinzer 1949; Beard 1976) It is also likely that the drop velocity near the ground surface is different from the terminal velocity aloft. This could be caused by turbulence, vertical air motions associated with updrafts and downdrafts, and other effects. Therefore, we recommend using instruments that have the ability to measure drop velocity independently of their size. Clearly, the POSS provides measurement of the drop velocity spectrum (through Doppler effect), but not independently of the size measurement. Other DSD measurement issues include the effect of the wind and the measurement of the drop shape. The two-dimensional (2D) video disdrometer (e.g., Kruger and Krajewski 2002) is capable of providing information on drop shape. This is important for the interpretation of polarimetric radar observations of rainfall (Bringi and Chandrasekar 2001). The wind effect is associated with air flow distortion by the instrument itself. For example, Nešpor et al. (2000) demonstrated that the large size and the shape of the 2D video disdrometer made by Johanneum Research in Austria, under certain conditions may cause significant distortion of the observed DSD. Other optical instruments with smaller structures are less susceptible to this effect, but may suffer from strong directional dependence on wind directions. With the current-day computational fluid dynamics technologies undertaking relevant studies is relatively straightforward although significant practical and theoretical issues remain (Habib and Krajewski 2001). Once the instrumental effects are well understood, we should undertake efforts to improve our knowledge of the spatial variability of variables relevant to remote-sensing rainfall. The most prominent variable is radar reflectivity. Its variability at the scale of radar-rainfall pixel directly affects quantitative interpretation of the estimated rainfall maps and products. The issues of observational network design, i.e., the number and distribution in space of the disdrometers are similar to those we discussed in the section above. As radar reflectivity is a higher-order moment of the DSD than rainfall rate, and there is no unique relationship between the two, it is likely that its characteristic scale (e.g., correlation distance) is significantly different from that of rainfall. Thus, we may need to organize experiments that will be able to capture a wide range of distances so that we can model the shape of the covariance function and other measures of association adequately. Only when we understand the error characteristics of the instruments we use will our interpretation of the results be meaningful. Thus, we need to continue supporting development of new, less-expensive disdrometers, conducting

Ground Networks: Are We Doing the Right Thing?

411

their intercomparisons, and designing and carrying out small-scale studies of spatial and temporal variability of various raindrop characteristics. These studies should be conducted in a continuous deployment mode as opposed to the more traditional campaign style typical for atmospheric experiments. This is necessary to enable conditional analysis of the collected data. Regarding the conditioning, it seems that the best strategy is to use the variables that can be observed by larger-scale remote sensors. For example, based on a vertically pointing radar we may be able to classify rainfall type into, say, convective and stratiform on a minute by minute basis, but if a nearby scanning radar is not able to resolve such scale conditional (on rain type) analysis is not very helpful in studies of radar performance. Another very appealing argument is that such conditioning would allow bringing data from many sites into the analysis provided they are conducted using the same or similar instruments. This is clearly the case using satellite-based observations. However, selection of appropriate variables is not an obvious task. It requires comprehensive analyses of the existing data via physically based modeling and/or data mining approaches.

5 OTHER GROUND-BASED OBSERVING SYSTEMS 5.1 Vertically pointing radars Vertically pointing radars provide crucial information for space-borne remote sensing. They are capable of observing vertical profiles of precipitating clouds and identify features affecting remote-sensing-based estimates. These features include thickness and height of the melting ice at cloud-based (i.e., bright band problem), precipitation phase, convective cores, updrafts and downdrafts, etc. They are also capable of providing estimates of the vertical profile of DSD. These estimates are more reliable if the profiler operates at multiple frequencies so that air and raindrop motion can be distinguished. However, for such radars the sampling volumes are not well matched. For high variability conditions such as convective cores this is a cause of concern as volume mismatch results in increased uncertainty of the measurements. Profiler-based studies of precipitation systems and the related instrumental and estimation issues have been well documented in a number of publications, for example, Wakasugi et al. (1986), Gage et al. (1999, 2000, 2002), Williams (2002), Williams et al. (2000), and Kollias et al. (2002). We propose to use these proven technologies to explore the spatial variability of the vertical profile of precipitating clouds. The scale of interest is the subscale of the resolution of space-based microwave and infrared sensors, i.e., about 15–25 km2. Although such scale domains can be explored by scanning radars, the main advantage of the profilers is their ability to

412

W. F. Krajewski

measure the Doppler velocity of the raindrops and thus provide estimates of the DSDs. They also identify the location and the thickness of the bright band much more accurately. Although the time resolutions of the VPRs is also very high, essentially providing continuous observations, our understanding of the space-time relation for rainfall processes is still limited (e.g., Fabry 1996.) Thus, having several, say five, such instruments evenly distributed over a single pixel, would provide additional insight into this relationship and the special variability of the processes affecting space-based sensors. This is particularly important for rainfall estimation over land where varied emissivity of the land surface introduces difficulties into interpretation of the satellite data.

5.2 Networks of small polarimetric radars Recently, another attractive technology is emerging that may offer many advantages to address the problem of validation of space-based remote sensing of precipitation: special purpose networks of inexpensive radars. Several groups have demonstrated advantages of using X-band polarimetric radars for rainfall estimation (Matrosov et al. 2002; Anagnostou et al. 2004). Because X-band radars are widely used for navigation, many manufacturers compete in this market. This drives down the cost of waveguides, microwave sources, and test equipment. Relatively small antennas can give high azimuthal resolution at X-band, compared to C-band and S-band radars. For example, to achieve 1.5° resolution requires an antenna with a diameter of 2 m, which translates to a tremendous cost advantage compared to C- and S-band. It is easy to install this size antenna on a small building or to mount it on a trailer. The polarimetric measurements at X-band also offer certain advantages, such as increased sensitivity to rainfall, as compared to longer wavelengths (e.g., Matrosov et al. 1999, 2002; Zrnic and Ryzhkov 1999). X-band (3 cm) waves are subject to more attenuation than the longer Cband or S-band waves in heavy rainfall. This is an issue if one needs a long operating range, but our focus is on short range and high resolution. If the network radars’ use is limited to 20 km, and there are multiple radars looking at the same area from different directions, the resultant multiradar estimates of rainfall will not suffer much from attenuation. The radars are polarimetric, and some polarimetric observations, such as KDP, are insensitive to partial attenuation. The physical concept behind polarization diversity is that, under aerodynamic forces, falling hydrometeors take oblate shapes, which depend on their size, and as a result, impact differently the propagation and backscattering of incoming horizontal (H) and vertical (V) electromagnetic waves. The most common polarimetric radar measurements are: (1) the reflectivity factors at H and V polarization (ZH, ZV); (2) the differential

Ground Networks: Are We Doing the Right Thing?

413

reflectivity factor (ZDR); and (3) the propagation differential phase (ΦDP). Over a certain radial distance ∆r, one can calculate the specific differential phase shift (KDP). All of these parameters, in various combinations, have been shown to contribute to improve rainfall estimation and enable retrieval of DSD (Bringi and Chandrasekar 2001; Anagnostou et al. 2004). Development and operation of a network of radars offers numerous advantages. For example, consider a network of four radars overlooking a regular dense network of rain gauges. Its operation leads to: Improved accuracy of rainfall algorithms. Sound algorithms can only be developed from large samples of data, since rainfall is highly variable in both space and time, and is highly intermittent (long periods of no rain between short duration events). This is contrary to past practice where a great deal of research and conclusions were based on case studies. As we emphasized in other sections of this paper, evidence is mounting that proper evaluation of remote sensing of rainfall can be established only from large sets of data. This need for large samples is increasing, as there is a trend to develop different algorithms for different atmospheric situations. To perform such conditioning for a sufficiently large sample, the data set from which the sample is drawn should be as large as possible. Increased reliability. It rains only about 5% of the time, so if the radar happens to be down we lose data. With four radars, it is unlikely that we will miss any rainfall events. We will be able to reduce the measurement error variance, and thus the uncertainty of the estimates of rainfall. Reduced development and operating costs. As network radars share spare parts and technical support, a network of four X-band radars may cost as little as $1 million. Repeatability. Credibility of the system, and therefore the confidence of users of the data will be greatly increased if the individual radars in the system demonstrate consistent performance when considered individually. On the other hand, viewing the same storm from different aspect angles will mitigate the adverse effects of signal attenuation and data noise. Still, much research remains to be done to fully realize the above benefits. These include technological advancements of radar hardware, software to operate the radar as a true network and not simply a collection of four individual radars, and, of course, rainfall estimation algorithms.

6 CONCLUSIONS AND RECOMMENDATIONS In this article we advocate an experimental framework for use of groundbased observation of rainfall for the quantitative evaluation of space-based methods of precipitation estimation and monitoring. We favor this approach over an alternative, i.e., the error propagation modeling on the basis of several arguments. The most important ones are: (1) the experimental approach

414

W. F. Krajewski

leads to total error estimates of the variable that is of ultimate interest, i.e., rainfall on the ground; (2) many of the parameters (characteristics) of the atmospheric and instrumental processes needed for the error propagation approach are not easy to observe or are not observable at all; (3) statistical distributions of the input variables are not known. We also caution against using approximate (engineering) methods of uncertainty propagation computations, i.e., the methods based on Taylor series expansion of the nonlinear functions governing the rainfall system and its observations. These methods give good results only when input variable errors are small (small coefficient of variation) and have symmetric distributions. We recommend a hierarchical approach to the validation problem. The hierarchy of instruments should cover the space-time scale gap that exists between the point observations by direct sensors such as rain gauges and observations from satellite platforms. The connecting device is weather radar. Used at different frequencies and configurations it brings down the resolution to the order of hundreds of meters in space and minutes in time. Such scales make the link with the point scale feasible via simple experimental setups. This in turn, permits error characterization of the radarbased estimates. With such knowledge we will be in a good position to attack the space-based technologies. Our hierarchical approach to the problem of validation of spacebased remote sensing of rainfall allows, of course, multisensor approaches (Krajewski 1987; Tustison et al. 2003). However, to optimally combine data from different sensors – motivated by the fact that they often have complementary characteristics in terms of sensing and relevance to rainfall – requires knowing error characteristics of the individual sensors. The combination could be done in a multisensor, multiscaling fashion or via a data assimilation schemes using physically based models. In either case error characterization is crucial. We hope that our discussion clearly illustrates the need for investing in basic research on rainfall via ground-based observational systems. The space-borne remote-sensing validation problem and the basic understanding of rainfall processes are strongly coupled. Without comprehensive understanding of the process, we will not be able to take advantage of the possibilities offered by remote sensing and without remote sensing we will not fully understand rainfall and its impact at global, regional, and local scale water cycle and other processes it affects. Therefore, what is the answer to the question we posed in the title? We do not know about many networks that were designed using the principles we discussed above. If we are serious about quantifying the uncertainty of space-based rainfall products we have to follow a systematic approach to designing, deploying, and using information from ground-based observational networks.

Ground Networks: Are We Doing the Right Thing?

415

Acknowledgments: Many of the ideas expressed in this article are result of collaborations and stimulating discussions over the years with Grzegorz Ciach, Anton Kruger, Mark Morrissey, Emad Habib, Paul Kucera, Brian Nelson, Mekonnen Gebremichael, Chris Kummerow, George Huffman, Bob Adler, Christopher Williams, Ken Gage, and James A. Smith, as well as with the participants of a recent workshop on small-scale variability of rainfall hosted by Jeff Austin in Auckland, New Zealand. These included Isztar Zawadzki, Frederic Fabry, Gyu Won Lee, Ian Cluckie, Chris Collier, Charles Lin, and Alan Seed, among others.

7 REFERENCES Anagnostou, E. N., M. N. Anagnostou, A. Kruger, W. F. Krajewski, and B. Miriovsky, 2004: High-resolution rainfall estimation from X-band polarimetric radar measurements. J. Hydrometeor., 43(1), 106–118. Beard, K. V., 1976: Terminal velocity and shape of cloud and precipitation drops aloft. J. Atmos. Sci., 33, 851–864. Bell, T. L., A. Abdullah, R. L. Martin, and G. R. North, 1990: Sampling errors for satellitederived tropical rainfall: Monte Carlo study using a space-time stochastic model. J. Geophys. Res., 95(D3), 2195–2205. Bell, T. L. and P. K. Kundu, 1996: A study of the sampling error in satellite rainfall estimates using optimal averaging of data and a stochastic model. J. Climate, 9, 1251–1268. Bell, T. L. and P. K. Kundu, 2000: Dependence of satellite sampling error on monthly averaged rain rates: Comparison of simple models and recent studies. J. Climate, 13, 449–462. Bras, R. L. and I. Rodriguez-Iturbe, 1993: Random functions and hydrology. Dover, 559 pp. Bringi, V. N. and V. Chandrasekar, 2001: Polarimetric Doppler weather radar: Principles and applications, Cambridge University Press, Cambridge, 636 pp. Campos, E. and I. Zawadzki, 2000: Instrumental uncertainties in Z–R relations. J. Appl. Meteor., 39, 1088–1102. Ciach, G. J., 2003: Local random errors in tipping-bucket rain gauge measurements. J. Atmos. Oceanic Technol., 20(5), 752–759. Ciach, J. G. and W. F. Krajewski, 1999: On the estimation of radar rainfall error variance. Adv. Water Resour., 22(6), 585–595. Crane, R. K., 1990: Space-time structure of rain rate fields, J. Geophys. Res., 95, 2001–2020. Fabry, F., 1996: On the determination of scale ranges for precipitation fields. J. Geophys. Res. 101, 12819–12826. Gage, K. S., C. R. Williams, W. L. Ecklund, and P. E. Johnston, 1999: Use of two profilers during MCTEX for unambiguous identification of Bragg scattering and Rayleigh scattering. J. Atmos. Sci., 56, 3679–3691. Gage, K. S., C. R. Williams, P. E. Johnston, W. L. Ecklund, R. Cifelli, A. Tokay, and D. A. Carter, 2000: Doppler radar profilers as a calibration tools for scanning radars. J. Appl. Meteor., 39, 2209–2222. Gage, K. S., C. R. Williams, W. L. Clark, P. E. Johnston, and D. A. Carter, 2002: Profiler contributions to Tropical Rainfall Measuring Mission (TRMM) ground validation field campaigns. J. Atmos. Oceanic Technol., 19, 843–863.

416

W. F. Krajewski

Gebremichael, M., W. F. Krajewski, M. Morrissey, D. Langerud, G. Huffman, and R. Adler, 2003: Error uncertainty analysis of GPCP monthly rainfall products: A data based simulation study. J. Appl. Meteor., 42(12), 1837–1848. Gunn, R. and G. Kinzer, 1949: The terminal velocity of fall for water droplets in stagnant air. J. Meteor., 6, 243–248. Habib, E., W. F. Krajewski, and A. Kruger, 2001: Sampling errors of fine resolution tippingbucket rain gauge measurements. J. Hydrol. Eng. 6(2), 159–166. Habib, E. and W. F. Krajewski, 2001: An example of computational approach used for aerodynamic design of a rain disdrometer. J. Hydraul. Res., 39(4), 425–428. Habib, E. and W. F. Krajewski, 2002: Uncertainty analysis of the TRMM Ground Validation radar-rainfall products: Application to the TEFLUN-B field campaign. J. Appl. Meteor., 41(5), 558–572. Jameson, A. R. and A. B. Kostinski, 2002: Spurious power-law relations among rainfall and radar parameters. Quart. J. Roy. Met. Soc., 128(7), Part B No. 584, 2045–2058. Kollias, P., B. A. Albrecht, and F. Marks Jr., 2002: Why Mie? Accurate observations of vertical air velocities and raindrops using a cloud radar. Bull. Amer. Meteor. Soc., 83, 1471–1483. Krajewski, W. F., 1987: Radar-rainfall data quality control by the influence function method. Water Resour. Res., 23(5), 837–844. Krajewski, W. F. and G. J. Ciach, 2003: Towards probabilistic quantitative WSR-88D algorithms: Preliminary studies and problem formulation. NOAA / N WS Report for Contract DG133W-02-CN-0089. Krajewski, W. F. and R. Goska, 2004: A GIS utility for observational network design for area-average estimation, to be submitted to Computers and Geosciences. Krajewski, W. F. and J. A. Smith, 2002: Radar hydrology: Rainfall estimation. Adv. Water Resour., 25, 1387–1394. Krajewski, W. F. and B. Vignal, 2004: Parameterization of the range dependent error in radar rainfall estimates based on a vertical profile of reflectivity analysis. to be submitted to J. Hydrometeorology. Krajewski, W. F., G. J. Ciach, and E. Habib, 2003: An analysis of small-scale rainfall variability in different climatological regimes. Hydrol. Sci. J., 48(2), 151–162. Krajewski, W. F., G. J. Ciach, J. R. McCollum, and C. Bacotiu, 2000: Initial validation of the Global Precipitation Climatology Project over the United States. J. Appl. Meteor., 39(7), 1071–1086. Kruger, A. and W. F. Krajewski, 2002: Two-dimensional video disdrometer: A description. J. Atmos. Oceanic Technol., 19, 602–617. Kruger, A. and K. Kanukurthy, 2004: A cell phone-based data logger and network for monitoring environmental variables. IEEE Trans. Instr. Meas. (near submission). Matrosov, S. Y., K. A. Clark, B. E. Martner, and A. Tokay, 2002: X-band polarimetric radar measurements of rainfall. J. Appl. Meteor., 41, 941–952. Matrosov, S. Y., R. A. Kropfli, R. F. Reinking, and B. E. Martner, 1999: Prospects for measuring rainfall using propagation differential phase in X- and Ka-radar bands. J. Appl. Meteor., 38, 766–776. McCollum, J. R. and W. F. Krajewski, 1998: Investigations of error sources of the Global Precipitation Climatology Project emission algorithm. J. Geophys. Res., 103(D22), 28711–28719. Miriovsky, B. J., A. A. Bradley, W. N. Eichinger, W. F. Krajewski, A. Kruger, B. R. Nelson, J.-D. Creutin, J.-M. Lapettite, G. W. Lee, I. Zawadzki, and F. L. Ogden, 2004: An experimental study of small-scale variability of reflectivity. J. Appl. Meteor. 5(1), 110–128. Moore, R. J., D. A. Jones, D. R. Cox, and V. S. Isham, 2000: design of the HYREX raingauge network. Hydrol. Earth System Sci., 4, 523–530.

Ground Networks: Are We Doing the Right Thing?

417

Morrissey, M. L., J. A. Maliekal, J. S. Greene, and J. Wang, 1995: The uncertainty in simple spatial averages using rain-gauge networks. Water Resour. Res., 31(8), 2011–2017. Nešpor, V., W. F. Krajewski, and A. Kruger, 2000: Wind-induced error of rain drop size distribution measurement using a two-dimensional video disdrometer. J. Atmos. Oceanic Technol., 17, 1483–1492. Nijssen, B. and D. P. Lettenmaier, 2004: Effect of precipitation sampling error on simulated hydrological fluxes and states: Anticipating the global precipitation measurement satellites, J. Geophys. Res., 109(D2), D02103 10.1029/2003JD003497. North, G. R., S. S. P. Shen, and R. Upson, 1993: Sampling errors in rainfall estimates by multiple satellites. J. Appl. Meteor., 32(2), 399–410. North, G. R. and S. Nakamoto, 1989: Formalism for comparing rain estimation designs. J. Atmos. Oceanic Technol., 6(6), 985–992. North, G. R. and I. Polyak, 1996: Spatial correlation of beam-filling error in microwave rainrate retrievals. J. Atmos. Oceanic Technol., 13(5), 1101–1106. Oki, R. and A. Sumi, 1994: Sampling simulation of TRMM rainfall estimation using radar AMeDAS composites. J. Appl. Meteor. 33(12), 1597–1608. Sheppard, B. E., 1990: Measurement of raindrop size distributions using a small Doppler radar. J. Atmos. Oceanic Technol., 7, 225–268. Sheppard, B. E. and P. I. Joe, 1994: Comparison of raindrop size distribution measurements by a Joss-Waldvogel disdrometer, a PMS 2DG spectrometer, and a POSS Doppler radar. J. Atmos. Oceanic Technol., 11, 874–887. Short, D. A. and G. R. North, 1990: The beam filling error in Nimbus-5 ESMR observations of GATE rainfall. J. Geophys. Res. 95, 2187–2193. Smith, J. A., D. J. Seo, M. L. Baeck, and M. D. Hudlow, 1996: An intercomparison study of NEXRAD precipitation estimates. Water Resour. Res., 32, 2035–2045. Smith, P., 1993: A Study of sampling-variability effects in raindrop size observations. J. Appl. Meteor., 32, 1259–1269. Steiner, M., J. A. Smith, S. J. Burges, C. V. Alonso, and R. W. Darden, 1999: Effect of bias adjustment and rain gage data quality control on radar rainfall estimates. Water Resour. Res., 35, 2487–2503. Tokay, A., A. Kruger, and W. F. Krajewski, 2001: Comparison of drop size distribution measurements by impact and optical disdrometers. J. Appl. Meteor., 40(11), 2083–2097. Tustison, B., D. Harris, and E. Foufoula-Georgiou, 2003: Scale-recursive estimation for multi-sensor QPF verification: A preliminary assessment. J. Geophys. Res., 108(D8), 8377–8390. Wakasugi, K., A. Mizutani, M. Matsuo S. Fukao, and S. Kato, 1986: A direct method for deriving drop-size distributions and vertical air velocities from VHF Doppler radar spectra. J. Atmos. Oceanic. Technol., 3, 623–629. Williams, C. R., 2002: Simultaneous ambient air motion and raindrop size distributions retrieved from UHF vertical incident profiler observations. Radio Sci., 37, 101029/ 2000RS002603. Williams, C. R., A. Kruger, K. S. Gage, A. Tokay, R. Cifelli, W. F. Krajewski, and C. Kummerow, 2000: Comparison of simultaneous rain drop size distributions estimated from two surface disdrometers and a UHF profiler, Geophys. Res. Lett., 27(12), 1763–1766. Zrnic, D. S. and A. V. Ryzhkov, 1999: Polarimetry for weather surveillance radars. Bull Amer. Meteor. Soc., 80(3), 389–406.

Section 6 Modeling Precipitation Processes and Data Assimilation for NWP

33 AEROSOL IMPACT ON PRECIPITATION FROM CONVECTIVE CLOUDS Alexander Khain, Daniel Rosenfeld, and Alexander Pokrovsky The Hebrew University of Jerusalem, Jerusalem, Israel

Abstract

Mechanisms through which atmospheric aerosols affect cloud microphysics, dynamics, and precipitation are investigated using a spectral microphysics cloud model. Significant effect of aerosols on cloud updrafts and cloud top height is found. Maritime aerosol leads to earlier formation of raindrops that fall down through cloud updrafts. This is one of the reasons of comparatively low vertical velocity in maritime convective clouds. An increase in the small cloud condensational nuclei (CCN) concentration leads to formation of a great number of small droplets with low collision rate. The direct consequence of this is a time delay in raindrop formation. This delay prevents the decrease in vertical velocity and increases the duration of the diffusion droplet growth stage, increasing latent heat release by condensation and freezing. As a result, vertical velocities in clouds developing in smoky (continental-type aerosol) air turn out to be larger and clouds attain higher levels. The decrease in precipitation efficiency of clouds arising in smoky air can be attributed to the higher loss of precipitating mass due to higher sublimation and evaporation. In case of very strong atmosphere instability and low air humidity (very continental thermodynamic conditions, forest fires, etc.), a great number of small droplets reach the upper troposphere and freeze with the formation of ice crystals that do not contribute to precipitation. Under more stable conditions, the delay in raindrop formation leads to the fact that the raindrops fall down from higher levels, as compared to those in case of clouds developing in clean air. In case of comparatively low air humidity and a certain wind shear, these raindrops fall trough a dry deep layer. As a result, precipitation from single cumulus clouds decreases significantly. Under certain conditions wide deep clouds developing in continental aerosol conditions produce stronger downdrafts and stronger convergence in the boundary layer. As a result, secondary clouds arise that can form a squall line. Under similar thermodynamic conditions, clouds developing under maritime aerosols do not produce strong downdrafts, and do not lead to the squall line formation. Formation of secondary clouds and squall lines increases precipitation over the area considered.Thus, the “aerosol effect” on precipitation can be understood only in combination with the “dynamical effect” of aerosols. This fact should be taken into account in schemes of parameterization of aerosol effects on precipitation.

421 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 421–434. © 2007 Springer.

422

A. Khain et al.

1 INTRODUCTION Analysis of the Tropical Rainfall Measuring Mission (TRMM) satellite data demonstrated that smoke from burning vegetation can practically shut off warm rain formation in tropical clouds (Rosenfeld 1999). Rosenfeld and Woodley (1999) observed that in polluted areas over Thailand and Indonesia smoky clouds do not precipitate at all because of the narrow spectra of small droplets. At the same time, similar clouds begin precipitating in clear air in 20–25 min after their formation. A decrease in precipitation in urban areas was reported (Rosenfeld 2000). Observations in the Amazon region (Kaufman and Nakajima 1993; Andreae et al. 2004) show significant effects of biomass burning on droplet spectra and precipitation formation. Some of these effects were numerically simulated by Khain et al. (1999, 2001) using the two-dimensional (2D) spectral microphysics Hebrew University Cloud Model (HUCM). Detailed investigation of raindrop formation in ascending cloud parcels (Segal et al. 2004) showed that the spectrum of cloud condensational nuclei (CCN) can be divided into three main ranges: CCN with radii rCCN < 0.01 µm are not usually activated and do not influence the cloud microphysical structure; CCN of the intermediate size with 0.01 µm < rCCN < ~1 µm are, as a rule, activated and give rise to droplet formation. An increase in the concentration of CCN of this size leads to an increase in the droplet concentration and slows the diffusional growth of droplets. Droplet spectra become narrower and the height of the collision triggering level increases. This leads to a delay in raindrop formation (Andreae et al. 2004; Khain et al. 1999). The last, CCNs with rCCN > ~1 µm give rise to formation of largest droplets, which foster raindrop formation at lower levels (e.g., Yin et al. 2000; Rosenfeld et al. 2002; Segal et al. 2004). A delay or acceleration in raindrop formation does not automatically lead to a decrease or increase in the accumulated rain. In this study we investigate physical mechanisms, through which aerosols influence cloud microphysics, dynamics, and accumulated rain.

2 NUMERICAL MODEL The model microphysics (see in more detail, Khain and Sednev 1996, Khain et al. 2000) is based on solving an equation system for eight size (number) distributions for water drops, ice crystals (columnar, plate-like, and dendrites), snowflakes (aggregates), graupel, and hail/frozen drops, and CCN. Each size distribution is represented by 33 mass doubling categories (bins), so mass mk in the category k is determined as mk = 2 mk–1, where k = 2,…,33. The minimum mass in the hydrometeor mass grids (except aerosols) corresponds to that of a 2 µm-radius droplet. The mass grids used for hydrometeors of all types are similar. This simplifies the calculation of interaction between

Aerosol Impact on Precipitation from Convective Clouds

423

hydrometeors of different bulk densities. The model microphysics is specifically designed to take into account the effect of atmospheric aerosols on cloud development and precipitation formation and effects of clouds on CCN concentration in the atmosphere. Nucleation (CCN activation) of droplets is based on the utilization of a separate size distribution function for CCN. In the current model version, the initial size distribution of CCN is calculated using a dependence of concentration N of activated CCN on supersaturation with respect to water Sw as described by Khain et al. (2000). In particular, the empirical dependence (Pruppacher and Klett 1997) can be written in the form of N = N0Skw, where Sw (in %), N0, and k are measured constants. Using supersaturation Sw calculated in the course of model integration, the value of critical size of dry CCN rcrit is determined at each model time step. Aerosol particles with radii rN > rN,crit are activated and transformed into droplets. Corresponding bins of the CCN size distributions become empty. In case there are no aerosol particles with rN > rN,crit in the CCN spectra in a particular grid point, no new droplet nucleation takes place at this point. The size of fresh nucleated droplets is calculated as follows. In case the radii of CCN rN < 0.03 µm, the equilibrium assumption (according to the Köhler equation), is used to calculate the radius of a nucleated droplet corresponding to rN (see Khain et al. 2000 for more detail). In case rN > 0.03 µm, the radius of water droplet formed on these CCN is simply equal to five times the radius of the dry aerosol particle (Kogan 1991; Khain et al. 1999; Yin et al. 2000). Since large CCN do not reach their equilibrium size at cloud base, this approach prevents nucleation of unrealistically large droplets and inhibit too fast raindrop formation. Nucleation of ice crystals is described proceeding from the formula presented by Meyers et al. (1992) relating the number concentration of deposition and condensation-freezing ice nuclei (IN), Nd, to supersaturation with respect to ice, Sice = Nd exp (ad + bdSice), where Nd = 10–3 m–3, ad = –0.639, bd = 12.96. Nucleation is prevented for temperatures warmer than –5°C. The number of newly activated ice crystals at each time step in a certain grid point, dNd, is calculated as follows:

⎧b N dS if dSice > 0 dN d = ⎨ d d ice 0 if dSice ≤ 0 ⎩

(1)

where dSice is calculated using a semi-Lagrangian approach (more details in Khain et al. 2000). The type of ice crystals nucleated depends on temperature. According to Takahashi et al. (1991) temperature-dependent nucleation proceeds as follows: plate-like crystals form at –8°C > Tc ≥ –14°C

424

A. Khain et al.

and –18°C > Tc ≥ –22.4°C, columnar crystals arise at –4°C > Tc ≥ –8°C, Tc < –22.4°C, and dendrites (branch-type crystals) form at –14°C > Tc ≥ –18°C. Secondary ice generation is described by a Hallett and Mossop (1974) mechanism, according to which at T = –5°C each 250 collisions of droplets having radii exceeding 24 µm with graupel particles leads to the formation of one ice splinter. According to measurements, this process is assumed to occur within –3 to –8°C temperature range. We suppose that the density of splinters is the same as that of pure ice (0.9 g cm–3) and, hence, the splinters are assigned to plate-type ice crystals. The rate of drop freezing is described following the observations of immersion nuclei by Vali (1975, 1994) and homogeneous freezing by Pruppacher (1995). The rate of freezing is calculated using a semi-Lagrangian approach allowing one to calculate changes in supersaturation and temperature in moving cloud parcels reaching model grid points (Khain et al. 2000). At each time step supersaturations with respect to water and ice were calculated by solving an equation system of corresponding differential equations (Khain and Sednev 1996). Besides droplet and ice nucleation, these values of supersaturation are used for calculation of diffusion growth/evaporation of water droplets and deposition/sublimation of ice particles. We take into account the shape of ice crystals to calculate diffusion growth of different ice crystals. An efficient and precise method of solving the stochastic kinetic equation for droplet collisions (Bott 1998) was extended to a system of stochastic kinetic equations that are used to calculate water–water, water–ice, and ice– ice collisions. The model uses height-dependent drop–drop and drop–graupel collision kernels calculated using a hydrodynamic method valid within a wide range of drop and graupel sizes (Khain et al. 2001; Pinsky et al. 2001). Ice–ice collision rates are assumed to be temperature dependent. An increase in the water–water and water–ice collision kernels by turbulent/inertia mechanism was taken into account following Pinsky et al. (2000). As a result of riming, ice crystals, and snowflakes can convert to graupel or to hail depending on temperature. Collisions between ice crystals lead to snow (aggregates) formation. Khain et al (2000) describe in detail the procedure for the conversion of hydrometeor types as a result of different kind of collisions. Recently, a description of collisional breakup has been implemented in the HUCM microphysics (Seifert et al. 2005). The changes of the drops size distribution due to breakup are represented by the well-known stochastic breakup equation (Pruppacher and Klett 1997). The coalescence efficiency and the fragment size distributions are parameterized following Low and List (1982) with some corrections for small raindrops using parameterizations given by Beard and Ochs (1995). The breakup is conducted for drops exceeding 100 µm in diameter.

Aerosol Impact on Precipitation from Convective Clouds

425

3 RESULTS: AEROSOL EFFECTS ON PRECIPITATION 3.1 Unstable continental conditions Simulations of cloud development under unstable conditions in Texas during summertime (Rosenfeld and Woodley 2000) were performed for a continental-type CCN (C-case), as well as for microphysically maritime CCN (M-case). In the simulations conducted, the initial size distribution of CCN was approximated as NCCN = ASwk. In C-case, A was set 1,260 cm–3 and k = 0.308. In addition, in C-case the maximum size of dry CCN was assumed equal to 0.6 µm. In M-case, coefficient A was set equal to 100 cm–3, while coefficient k = 0.462. It was also assumed, that there were no small CCN in the CCN spectrum that could be activated at supersaturation values exceeding 1.1% in the “maritime” case. This assumption is based on measurements by Hudson (1984, 1993) and Hudson and Frisbie (1991) indicating no increase in the CCN concentration in extreme maritime cases for S > 0.6%, which suggests a lack of small CCN. Thus, we assumed that under not very extreme conditions, maritime air does not contain small aerosols to be activated at S > 1.1%. This limitation allows us to keep droplet concentration to be typical of maritime clouds rain rates as the functions of time and x-coordinate are shown in Fig. 1. One can see a significant decrease in precipitation amount in clouds that developed in smoky (continental CCN) air. The difference in precipitation can be attributed to the following. When CCN concentration is low (M-case), raindrops and large graupel form at comparatively low levels, fall down and reach the surface without a significant evaporation. In C-case, high concentration of small droplets (up to 1,000 cm–3) arises by nucleation (Khain et al. 2001). These droplets have low collision efficiency, as well as low freezing rate. As a result, they reach the level of homogeneous freezing (~9.5 km) and give rise to formation of ice crystals with concentration of several hundred per cm–3. These crystals as well as small graupel and snowflakes formed at higher levels spread over a large area and sublimate, and do not contribute to the precipitation (they sublimate in a cloud anvil). Droplets falling from higher levels also experience significant evaporation. As a result, the increase in CCN concentration decreases the precipitation efficiency of clouds that develop in strong unstable low humidity air. Note that mass contents of water drops (with maximum of about 3 cm–3), as well as ice particles (total ice content maximum was also about 3 cm–3) were quite significant. Thus, precipitation efficiency (defined as the ratio of precipitation amount to the amount of hydrometeors formed in clouds by condensation and deposition) is quite low in these continental clouds (in C-case).

A. Khain et al.

426

3.2 Single maritime clouds Aerosol effects on clouds and precipitation under maritime (as well as under comparatively wet and not extremely unstable conditions) are more complicated and are highly affected by vertical wind shear and air humidity. The investigation of aerosol effects is performed using the same initial CCN distributions as were used under the continental thermodynamic conditions. However, in these simulations the GATE (day 261) temperature profile (Ferrier and Houze 1989) was used. Since the maximum CCN radius in the model is 2 µm, the maximum radius of nucleated droplets in M-case did not exceed ~10 µm. The role of giant CCN was not investigated in the study.

Maritime CCN

Continental CCN

Precipitation inhibition

Figure 1. Rain rates as the functions of time and x-coordinate under unstable continental conditions.

The development of single clouds was triggered by 5 min of heating within the zone of the 0.5 km width in the surface layer. By varying the intensity of the heating, single clouds of different top height were simulated. The cloud evolution was simulated up to the cloud dissipation. Figure 2 shows the dependence of accumulated rain on the cloud top height (determined by the 10 dBZ level) of convective clouds in three sets of simulations with the GATE (day 261) temperature profile. In the first set no wind shear (NO WS) was assumed. In the second set (with a weak wind shear, WS) the wind speed increases from 0.4 m s–1 at z = 0 km to 5 m s–1 at z = 9 km. Above z = 9 km the wind speed was assumed equal to 5 m s–1. In both cases air humidity was quite high, with ~90% in the lower 2 km layer and about 50– 60% in the middle troposphere. The third set of simulations was similar to the second one except the humidity was decreased by 10% in the lower 2 km layer (RH = 80%) and by about 20–25% in the middle troposphere. Each set

Aerosol Impact on Precipitation from Convective Clouds

427

of the experiments consisted of experiments with maritime CCN (M-case) and continental CCNs (C-case).

Figure 2. Dependences of accumulated rain on cloud depth under different conditions in case of maritime (GATE, 261 day) temperature profile.

Figure 3. Vertical profiles of convective heating, cooling, and net heating calculated under continental and maritime CCN. The clouds chosen produced nearly similar precipitation amounts.

One can see that (a) single clouds that developed in M-cases produce larger amounts of accumulated rain. To produce the same rain amount, clouds that developed in C-case, should be higher; (b) wind shear, and especially, the air humidity significantly change the accumulated rain amount, as well as the

428

A. Khain et al.

relationship between the rain amounts in C- and M-cases. At the same time, sensitivity of the accumulated rain to the CCN concentration under comparatively stable maritime temperature conditions is smaller than that under unstable continental conditions; (c) while in all sets of simulations the accumulated rain monotonically increases with the cloud top height in Mcases, in C-cases with a nonzero wind shear, precipitation stops growing, beginning with a certain cloud top height. The physical explanation of the lower precipitation efficiency in single clouds developed in microphysically continental CCN conditions can be derived from the analysis of vertical profiles of convective heating/cooling. Figure 3 shows such profiles plotted for two clouds that developed in simulations with no wind shear. Firstly, the cloud developing in the C-case reaches the top height of 9 km. Secondly, the cloud that develops in M-case, reaches the maximum top height of 6.5 km. According to Fig. 2, these clouds produce comparable accumulated rain amounts. These profiles were obtained by averaging of heating/cooling in the horizontal direction over the computational area (64 km) and over time of simulation (4 h). Dashed and dasheddotted lines corresponding to positive values of latent heat release reflect contribution of condensation, freezing and deposition. Along similar lines, but corresponding to negative values of latent heat release, reflect cooling caused by the droplet evaporation, ice melting and ice sublimation. Thick and thin solid lines show profiles of net convective heating for continental and maritime CCN, respectively. The squares (integrals) formed by these solid lines reflect the total atmospheric heating due to phase transitions and are a measure of the precipitation amount. The increase in the CCN concentration leads to a strong increase in heating both due to the increase in diffusion growth and due to more intense droplet freezing. At the same time, the cooling in the C-case is larger as compared to the M-case because of the larger droplet evaporation and ice sublimation. The higher droplets evaporation and ice sublimation in the C-case can be attributed to the following. In the C-case raindrops and ice particles are smaller than in the M-case and ascend to higher levels. Besides, they have smaller sedimentation velocity. As a result, the time duration of their sedimentation is longer. The detrainment of liquid and ice particles at upper levels leads to the fact that ice particles and drops falling from higher levels tend to sediment through a comparatively dry air. At the same time, in the M-case raindrops form at lower levels, and fall down through the vertically narrow layer of cloudy wet air (or in the close vicinity of the cloud). As a result, the evaporation (and cooling) in the maritime aerosol case is lower as compared to the continental CCN case. The net heating in the C-case is extended to higher levels. The minimum in the net heating at ~4 km in the C-case is related to the cooling caused by the melting of ice (mainly graupel). No such minimum is seen in the profile of net convective heating in the M-case simulation indicating a smaller contribution of melted rain in the last case.

Aerosol Impact on Precipitation from Convective Clouds

429

Figure 4. Vertical profiles of convective heating for clouds developed in continental and maritime aerosols in case of a comparatively weak wind shear.

The higher loss in the precipitating mass by the drop evaporation and ice sublimation is the main cause of the lower precipitation efficiency of clouds developed in the continental CCN case. Another example of latent heat release profiles for clouds developed in case of upper level wind shear in the C- and M-case is shown in Fig. 4. These clouds reached the same top heights (about 10 km), but have quite different accumulated rain (see Fig. 2). Stronger wind at upper levels moves water droplets and ice downwind fostering their fall through dry air and their evaporation and sublimation. Since water droplets and ice particles ascend to higher levels in C-cases, wind shear leads to higher loss in precipitation for C-cases. We believe that this is the reason, why in the presence of wind shear the precipitation amount in C- cases does not increase with cloud top height (beginning with a certain cloud depths): a significant fraction of hydrometeors ascending above certain level evaporates and does not contribute to precipitation. This can explain the results of some simulations with high wind shears at upper levels (not shown), when the growth of cloud top height in C-cases was accompanied by a decrease in precipitation amount. The decrease in precipitation with a decrease in humidity (Fig. 2) is explained by the increase in the loss in precipitating mass by evaporation and sublimation. This loss is higher in clouds that develop under continental aerosol conditions. A comparatively low sensitivity of precipitation in M-cases to both wind shear and humidity is related to the

430

A. Khain et al.

formation of raindrops at low levels and to their fall within wet air in, or in the close surrounding of the cloud.

3.3 Dynamical aerosol effects It is well known that vertical velocities in maritime convective clouds are significantly smaller than in continental clouds. While in continental clouds vertical updrafts easily exceed 20–30 m s–1, only 5% of deep maritime clouds have maximum updraft velocity exceeding 10–15 m s–1 (Zipser and LeMone 1983; Emanuel 1994). It is widely accepted that this difference is caused by higher instability of the continental atmosphere. This is usually attributed to a potentially higher surface temperature (which, say, in summertime Texas conditions can exceed 36°C (Rosenfeld and Woodley 2000), while the sea surface temperature (SST) hardly exceeds 31°C. Note, however, that the land surface temperature often does not exceed the SST, while the existence of a higher vertical humidity gradient (a higher gradient of virtual temperature) and lower cloud base level should foster formation of strong updrafts in maritime clouds too. Nevertheless, under the same surface temperatures, vertical velocities in maritime clouds still remain lower than those in continental cumulus clouds. As it was shown in Section 3.2, one of the factors, decreasing cloud updrafts in maritime clouds is the low CCN concentration that leads to early formation of raindrops and their fall through the cloud updrafts. The dynamical effect was observed even in experiments with the Texas unstable conditions (not shown): both Figure 5. Maximal updrafts and downmaximum values of convective updrafts in clouds developed using the GATE drafts as well as downdrafts are profiles under continental and maritime aerohigher in clouds developed in smoky sol conditions. The development of secondair (continental aerosols). ary clouds in experiments cloud with a Since the instability of the marisuccesssive increase in the intensity of the first. time atmosphere is comparatively

Aerosol Impact on Precipitation from Convective Clouds

431

low, deep maritime convective clouds are often forced by surface level convergence caused by dissipating clouds (Ferrier and Houze 1989; Emanuel 1994). As a result, formation of secondary clouds turns out to be dependent on the properties of primary clouds. In Section 3.2, primary clouds were forced by an initial heating within a comparatively narrow zone of the subcloud layer. Besides, the wind shear was weak in these simulations. As a result, these clouds did not produce secondary clouds either in the M- and Ccases. As the width of the initial heating triggering the convection was successively increased from 2 to 4 km, the primary clouds became wider and the maximum vertical velocity increased. Besides, the wind shear was also increased to that measured at 261 day of the GATE (~ 7 m s–1 per 5 km in the middle troposphere). Figure 5 shows that in the C-case experiments the strengthening of the primary cloud leads to formation of a secondary cloud (seen by the formation of new vertical velocity maximum). At a certain stage, this secondary cloud gives rise to squall-line formation. Note that secondary clouds formed in corresponding M-case simulations were much weaker and did not develop into a squall line. This effect can be attributed to the following. Figures 3 and 4 show that convective heating by condensation and freezing signifycantly larger in C-cases than in M-cases. Because of wind shear transporting cloud hydrometeors downwind, cooling due to evaporation and sublimation takes place at some distance from the cloud updraft. This spatial shift between heating and cooling is accompanied by a decrease in the loading in the updraft zone (in contrast to M-cases, where raindrops can fall through cloud updraft). These factors lead to formation of dynamically induced vorticity that increases both vertical updrafts and downdrafts in clouds arising within microphysically continental aerosol. As a result, the maxima of updraft and downdraft vertical velocity are larger in the C-case clouds than in M-case clouds (Fig. 6).

Figure 6. Precipitation rate in M-case and C-case in simulations when GATE (261 day) sounding in case of squall-line formation.

432

A. Khain et al.

Figure 6 shows precipitation rate in both the M-case and C-case in simulations when the GATE (261 day) sounding was used, and the first cloud was triggered by a surface layer heating within the area of the 4 km width. Profiles of convective heating suggest (not shown) a significant precipitation rate, as the heating significantly exceeds the cooling. We attribute the increase in the precipitation efficiency to an increase in the relative humidity within the area of the squall line and to a significant increase in cloud cover (squall line is accompanied by significant melted rain falling from stratiform clouds behind the line).

4 CONCLUSION A 2D spectral microphysics cloud model has been used for the investigation of aerosol effects on cloud dynamics, microphysics, and precipitation. It is shown that (a) increase in concentration of CCN with radii (0.01 µm < rCCN < 1 µm) drastically decreases precipitation from clouds developed under unstable dry continental conditions; (b) an increase in the CCN concentration decreases precipitation from individual maritime clouds (and, as follows from supplemental simulations, under intermediate continental conditions) as well. This decrease is, however, smaller than in case of unstable continental conditions and highly depends on wind shear and air humidity. The reduction of the precipitation efficiency with the increase in the CCN concentration is caused by the increase in the loss of precipitation during sedimentation through a deep layer of dry air. An increase in the CCN concentration increases the intensity of convection and fosters the formation of squall lines. In this case precipitation rate significantly increases, indicating nonlinear effects in the cloud–aerosol interaction. The cloud lifetime as well as the area covered by clouds also increases with the increase in the concentration of CCNs of intermediate size. These aerosol effects should affect the radiation balance of the atmosphere. It is widely accepted that an increase in the concentration of aerosols in the atmosphere leads to atmospheric (and climatic) cooling. It was shown here that the increase in the aerosol concentration leads to an increase in cloud cover at higher levels in the atmosphere that leads to atmospheric heating. Thus, dynamical effects of aerosols on cumulus convection make net aerosol effects on climate not so obvious. This topic requires further investigation. The increase in cloud intensity with the CCN concentration, as well as the cloud top height should foster the lightning formation, and, possibly, other dangerous meteorological phenomena. It is shown that aerosols influence precipitation efficiency. Clouds of similar cloud top heights will precipitate differently under different aerosol

Aerosol Impact on Precipitation from Convective Clouds

433

conditions. Thus, the aerosol effects on precipitation should be taken into account in different rain retrieval algorithms. Acknowledgments: The study was performed under support of the Israel Ministry of Science (German–Israel collaboration in Water Resources, grant WT 0403), by the Israel Water Company (Shaham) as well as by EU project EURAINSAT.

5 REFERENCES Andreae, M. O., D. Rosenfeld, P. Artaxo, A. A. Costa, G. P. Frank, K. M. Longo, and M. A. F. Silva-Dias, 2004: Smoking rain clouds over the Amazon. Science, 303, 1337–1342. Beard, K. V. and H. T. Ochs, 1995: Collisions between small precipitation drops. Part II: Formulas for coalescence, temporary coalescence, and satellites. J. Atmos. Sci., 52, 3977– 3996. Bott, A., 1998: A flux method for the numerical solution of the stochastic collection equation. J. Atmos. Sci., 55, 2284–2293. Emanuel, K. A., 1994: Atmospheric convection. Oxford University Press, Oxford, 580 pp. Ferrier, B. S. and R. A. Houze, 1989: One-dimensional time dependent modeling of GATE cumulonimbus convection. J. Atmos. Sci., 46, 330–352. Hallett, J. and S. C. Mossop, 1974: Production of secondary ice crystals during the riming process. Nature, 249, 26–28. Hudson, J. G., 1993: Cloud condensational nuclei near marine cumulus. J. Geophys. Res., 98, 2693–2702. Hudon, J. G. and P. R. Frisbie, 1991: Cloud condensation nuclei near marine stratus. J. Geophys. Res., 96, 20795–20808. Hudson, J. G. and Y. Xie, 1999: Vertical distribution of cloud condensation nuclei spectra over the summertime northeast Pacific and Atlantic Ocean. J. Geophys. Res., 104, 30219– 30229. Kaufman, Y. J. and T. Nakajima, 1993: Effect of Amazon smoke on cloud microphysics and albedo-analysis from satellite imagery. J. Appl. Meteor., 32, 729–744. Khain, A. P. and I. Sednev, 1996: Simulation of precipitation formation in the Eastern Mediterranean coastal zone using a spectral microphysics cloud ensemble model. Atmos. Res., 43, 77–110. Khain, A., A. Pokrovsky, and I. Sednev, 1999: Some effects of cloud-aerosol interaction on cloud microphysics structure and precipitation formation: Numerical experiments with a spectral microphysics cloud ensemble model. Atmos. Res., 52, 195–220. Khain, A. P., M. Ovtchinnikov, M. Pinsky, A. Pokrovsky, and H. Krugliak, 2000: Notes on the state-of-the-art numerical modeling of cloud microphysics. Atmos. Res., 55, 159–224. Khain, A. P., M. B. Pinsky, M. Shapiro, and A. Pokrovsky, 2001: Graupel-drop collision efficiencies. J. Atmos. Sci., 58, 2571–2595. Khain, A. P., D. Rosenfeld, and A. Pokrovsky, 2001: Simulation of deep convective clouds with sustained supercooled liquid water down to –37.5°C using a spectral microphysics model. Geophys. Res. Lett., 28 (20), 3887–3890. Kogan, Y. L., 1991: The simulation of a convective cloud in a 3-D model with explicit microphysics. Part I: Model description and sensitivity experiments. J. Atmos. Sci., 48, 1160–1189. Low, T. B. and R. List, 1982: Collision, coalescence and breakup of raindrops, Part II: Parameterization of fragment size distributions. J. Atmos. Sci. 39, 1607–1618.

434

A. Khain et al.

Meyers, M. P. and W. R. Cotton, 1992: Evaluation of the potential for wintertime quantitative precipitation forecasting over mountainous terrain with an explicit cloud model. Part I: Two-dimensional sensitivity experiments. J. Appl. Meteor., 31, 26–50. Pinsky, M., A. P. Khain, and M. Shapiro, 2000: Stochastic effects on cloud droplet hydrodynamic interaction in a turbulent flow. Atmos. Res., 53, 131–169. Pinsky, M., A. P. Khain, and M. Shapiro 2001: Collision efficiency of drops in a wide range of Reynolds numbers: Effects of pressure on spectrum evolution. J. Atmos. Sci., 58, 742– 764. Pruppacher, H. R., 1995: A new look at homogeneous ice nucleation in supercooled water drops. J. Atmos Sci., 52, 1924–1933. Pruppacher, H. R. and J. D. Klett, 1997: Microphysics of clouds and precipitation, 2nd edn. Kluwer Academic, Dordrecht, 914 pp. Rosenfeld, D., 1999: TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 20, 3105. Rosenfeld, D., 2000: Suppression of rain and snow by urban and industrial air pollution. Science, 287 (5459), 1793–1796. Rosenfeld, D. and W. Woodley, 2000: Deep convective clouds with sustained supercooled liquid water down to –37.5°C. Nature, 405, 440–442. Rosenfeld D., R. Lahav, A. Khain, and M. Pinsky, 2002: The role of sea spray in cleaning air pollution over ocean via cloud processes. Science, 297, 1667–1670. Segal, Y., A. Khain, and M. Pinsky, 2004: Effects of atmospheric aerosol on precipitation in cumulus clouds as seen from 2000-bin cloud parcel microphysical model: sensitivity study with cloud seeding applications. Quart. J. Roy. Meteor. Soc., 130B (587), 561–582. Seifert, A., A. Khain, and U. Blahak, 2005: Possible effects of collisional breakup on mixedphase deep convection simulated by a spectral (bin) cloud model. J. Atmos. Sci., 62, 1917–1931. Takahashi, T., T. Endoh, and G. Wakahama, 1991: Vapor diffusional growth of free-falling snow crystals between –3 and –23°C. J. Meteor. Soc. Japan, 69, 15–30. Yin, Y., Z. Levin, T. Reisin, and S. Tzivion, 2000: The effects of giant cloud condensational nuclei on the development of precipitation in convective clouds: A numerical study. Atmos. Res., 53, 91–116. Vali, G., 1975: Remarks on the mechanism of atmospheric ice nucleation. Proc. 8th Int. Conf. on Nucleation, Leningrad, 23–29 Sept., I.I. Gaivoronsky, ed., Gidrometeoizdat, 265–269. Vali, G., 1994: Freezing rate due to heterogeneous nucleation. J. Atmos. Sci., 51, 1843–1856. Zipser, E. J. and M. A. LeMone, 1980: Cumulonimbus vertical velocity events in GATE. Part 2: Synthesis and model core structure. J. Atmos. Sci., 37, 2458–2469.

34 THE WISCONSIN DYNAMIC/MICROPHYSICAL MODEL (WISCDYMM) AND THE USE OF IT TO INTERPRET SATELLITE-OBSERVED STORM DYNAMICS Pao K. Wang Department of Atmospheric and Oceanic Science, University of Wisconsin–Madison, Madison, WI, USA

1 BRIEF DESCRIPTION OF THE CLOUD MODEL WISCDYMM AS CURRENTLY CONFIGURED This paper will briefly report on the cloud model that has been used in our group for various researches but especially those related with the physics and dynamics atop thunderstorms. The successful applications of this model to investigate several satellite-observed dynamical processes atop thunderstorms will also be summarized. The model is the Wisconsin dynamic and microphysical model (WISCDYMM), developed in the author’s research group. The earliest form was described in Straka (1989) and subsequently modified by others (Johnson et al. 1993, 1994; Lin and Wang 1997; Wang 2003). Its properties are described in the following sections.

1.1 Grid configuration and time step WISCDYMM uses a uniform staggered grid in all directions, placing the wind components on the normal grid cell faces and the remaining variables at the grid cell centers (Arakawa-C grid). Typically, these cells are assigned equal dimensions in both horizontal directions while it is usually smaller in the vertical. The time step size, assumed uniform, is dictated by quasicompressible computational stability requirements. The computational domain and resolution can be changed according to the needs of specific purposes. A typical setup of the grid for studying severe storm dynamics is given in the following: the grid mesh is 1.0 km horizontally and 0.5 km 435 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 435–446. © 2007 Springer.

436

P. K. Wang

vertically, with corresponding dimensions of 55 and 20 km for the model domain, while the time step is 3 s. In some cases, the resolution remains the same but the horizontal domain is expanded to 120 × 100 km and the top boundary is set at 30 km. For the purpose of studying the water vapor transport across the tropopause (Wang 2003), the vertical resolution was set at 200 m. The model also has been run with horizontal resolution of 500 m and good results were obtained.

1.2 Initialization The initial fields in WISCDYMM simulations consist of two components: (a) a horizontally homogeneous base state closely adapted from a prestorm rawinsounding, with no condensate and specifying the surface pressure from the sounding data and (b) an impulse to initiate the modeled storm. The rawinsounding data, which have irregular vertical spacing, are linearly interpolated (without smoothing) to the appropriate model grid levels as potential temperature, either water vapor mixing ratio or relative humidity, and the horizontal wind components relative to the earth. Base-state pressure values above ground are derived from the surface pressure by assuming a hydrostatically balanced environment. So far, the initial impulse has been an ellipsoidal warm bubble in the lower central part of the model domain, with the same relative humidities as in the base state.

1.3 Model physics WISCDYMM predicts the three wind components, turbulent kinetic energy, potential temperature, pressure deviation and mixing ratios for water vapor, cloud water, cloud ice, rain, graupel/hail, and snow. The model adapts the quasicompressible, nonhydrostatic primitive equation system of Anderson et al. (1985), rearranging the mass continuity equation to predict the pressure deviation much as in the fully compressible 3D cloud model of Klemp and Wilhelmson (1978), but allows time steps approximately three times larger by assigning acoustic waves a reduced pseudo-sound speed roughly twice the maximum anticipated wind speed. As in Klemp and Wilhelmson, subgrid transports are parameterized via 1.5-order “K-theory” to predict turbulent kinetic energy, from which a time- and space-dependent eddy coefficient is diagnosed for momentum and set 35% larger for the heat and moisture predictands (Straka 1989). As elaborated by Straka (1989), a version of WISCDYMM called hail parameterization model (HPM) features a bulk microphysics parameterization that entails water vapor and five hydrometeor types: cloud water, cloud ice, rain, graupel/hail, and snow, with 37 individual transfer rates (source/sink terms). Adapted largely from Lin et al. (1983) and Cotton et al. (1982, 1986), this package treats all hydrometeors as spheres except for cloud ice, which is treated as small hexagonal plates. Cloud water and cloud ice are assumed

The Wisconsin Dynamic/Microphysical Model

437

monodisperse, with zero fall speed relative to the air. All three precipitation classes have inverse-exponential size distributions, with temperature-dependent intercepts for snow and graupel/hail, while the intraspectral variation of particle fall speed versus diameter for each is assumed to satisfy a power law. If necessary, WISCDYMM can also be run in the HCM mode in which the evolution of hailstones can be tracked by studying the growth of hail sizes in a number (e.g., 25) of size bins. The HCM has been tested successfully in Straka (1989). WISCDYMM is also programmed to activate one or more of the following three iterative microphysical adjustments (Straka 1989) where and if needed: •

• •

A saturation adjustment is performed to: (a) condense cloud water (or depose cloud ice) to eliminate supersaturation, releasing latent heat of evaporation (or sublimation), or (b) evaporate cloud water (or sublimate cloud ice) in subsaturated air until either saturation is reached or the cloud water (or cloud ice) is exhausted, absorbing latent heat instead. Cloud water is adjusted first and cloud ice second, incrementing the water vapor and temperature to suit. In-cloud saturation mixing ratios are weighted between their values with respect to water and ice in proportion to the relative amounts of cloud water and cloud ice respectively. Where no cloud is present, saturation is taken with respect to water or ice if the temperature is above or below 0°C, respectively. More than three iterations are rarely needed. Prior to the saturation adjustment, any cloud ice at temperatures above 0°C is melted, and any cloud water at temperatures below –40°C is frozen, respectively absorbing or releasing latent heat of fusion. After partial update of the moisture fields by advection and turbulent mixing, the decrement in each hydrometeor field due to the net sink (the sum of the individual microphysical sink terms) is compared to its available supply, defined as its partially updated mixing ratio plus the increment due to its net source (the sum of the individual microphysical source terms). If the net sink of a hydrometeor class exceeds its available supply, it is prorated downward along with each of its components so as to not exceed 25% of the available supply. The procedure is iterative because prorating down the sinks of one class also reduces the corresponding source terms for one or more other classes, but more than two iterations are rarely needed.

1.4 Boundary conditions The lateral boundary conditions are similar to those in Klemp and Wilhelmson (1978). Reflection of outward-propagating gravity waves is suppressed by “radiation” conditions which advect each horizontal normal wind component out with a velocity equal to itself plus a prescribed constant

438

P. K. Wang

outward gravity wave propagation of speed c*, except assigning zero advection in inflow of speed greater than c*. The other predictands at lateral boundary outflow locations are advected by upstream differencing. Both the upper and lower boundaries are rigid lids. Variables at the top are held undisturbed, while reflection of upward-propagating gravity waves off the lid is suppressed by imposing an upper-level Rayleigh damping layer that abuts it. The lower boundary has four options: free-slip with no surface heat flux; semi-slip with no surface energy budget; no-slip with heat flux and no surface energy budget; or no-slip with heat flux, insolation, and surface energy budget.

1.5 Interior numerics The current version of WISCDYMM uses forward-in-time differencing and sixth-order flux-conservative Crowley spatial differencing (Tremback et al. 1987). To suppress nonlinear instability, a fourth-order numerical diffusion operator with a constant coefficient, as in Klemp and Wilhelmson (1978), is added in the discretized predictive equations at each time step.

2 SIMULATION OF THE 2 AUGUST 1981 CCOPE SUPERCELL THUNDERSTORM Up to the present, WISCDYMM has been used primarily for simulating deep convective events. It has successfully simulated many cases of severe thunderstorms occurring in the USA and other parts of the world. To illustrate the capability of this model, the simulation results of the 2 August 1981 supercell storm that occurred in Montana in the Midwest of USA will be briefly presented. In later sections, the simulation results of this storm will be used as examples for understanding some thunderstorm dynamical processes as observed by meteorological satellites.

2.1 A brief description of the 2 August 1981 CCOPE supercell The storm chosen for the simulation for illustrating the plume-formation mechanism is a supercell that passed through the center of the Cooperative Convective Precipitation Experiment (CCOPE) (Knight 1982) observational network in southeastern Montana on 2 August 1981. The storm and its environment were intensively observed for more than 5 h by a combination of seven Doppler radars, seven research aircraft, six rawinsonde stations and 123 surface recording stations as it moved east–southeastward across the CCOPE network. Miller et al. (1988) and Wade (1982) provided many of the observations in this section, especially those on the history of the storm. This case was chosen because it is a typical deep convective storm in the US High Plains and it provides much detailed observational data for comparison with

The Wisconsin Dynamic/Microphysical Model

439

model results with regard to dynamics and cloud physics, and the author’s group has obtained successful simulations of it previously (Johnson et al. 1993, 1994).

2.1.1.

Environmental conditions

The initial conditions for the simulation are based on a 1,746 (mountain daylight time (MDT) sounding (Fig. 1) taken at Knowlton, Montana, approximately 90 km ahead of the storm. This sounding provided the most representative temperature and moisture profile available, with a massive convective available potential energy (CAPE) 3,312 J kg–1 distributed over a comparatively shallow layer from the level of free convection LFC = 685 hPa to the equilibrium level EL = 195 hPa. The subcloud layer (below 730 hPa) was nearly dry-adiabatic and well mixed, with a potential temperature close to 311.5 K, and also relatively moist because a surface low in north central Wyoming advected water vapor mixing ratios of 12–13 g kg–1 into the region on easterly winds. Above the subcloud region, a strong capping dry layer existed at approximately 710 hPa, caused by warmer and drier air that had unexpectedly moved into the region after 1,300 MDT. Wade (1982) gives some possible causes of this warming. The dry layer was significant in that it allowed the low-level air mass to continue warming for the remainder of the afternoon and become even more potentially unstable. From the dry layer to 450 mb, the environmental lapse rate was nearly dry adiabatic. The calculated indices from the Knowlton sounding (totals index = 60, lifted index = –9.4, and a K index = 38) indicated that the air mass over eastern Montana on 2 August was very unstable, and hence very favorable for the development of deep convection. Large vertical wind shear between lower and midlevels was also conductive to severe weather development. The 1,746 MDT Knowlton hodograph (not shown) indicated strong subcloud flow, veering nearly 70° from the surface layer to cloud base at 1.6 km AGL. The magnitude of the mean shear over the lowest 6 km was 0.008 s–1. There was little directional shear above the cloud base, but vertical speed shears between the cloud base and 9 km were 0.006 s–1 (Miller et al. 1988). Taking into account the vertical wind shear and buoyancy effects, the Bulk Richardson Number for the prestorm environment was 25, in the expected range for supercell storms. Some previous studies have pointed out that clockwise curvature of the wind shear vector over the lowest 2 km of the hodograph also favored development of the right-moving supercell.

2.1.2.

Examples of simulated microphysical and dynamical fields

To illustrate the model performance, Figs. 2 and 3 show the simulated hydrometeor mixing ratio and vertical velocity fields in the central east–west

P. K. Wang

440

Figure 1. The 1,746 MDT Knowlton, Montana sounding on 2 August 1981. The solid curve is for temperature and dashed curve for dew point. The portion of dew point curve above 300 hPa, which was missing in the original sounding, is constructed using an average August 1999 HALOE water vapor profile over 40–60N.

vertical cross section of the storm at t = 40 min. In general, the model results agree with the observed behavior of the supercell very well.

2.1.3.

Simulation of anvil top cirrus plumes

As indicated before, WISCDYMM has been used successfully to study cloud dynamical and physical processes atop deep convective storms. One recent example is the identification of the physical mechanism that produces the cirrus plumes above the anvils of some severe thunderstorms. Such plumes have been observed from satellite visible and infrared (IR) images and some details of them have been studied by a number of investigators (Setvak and Doswell 1991; Levizzani and Setvák 1996). Figure 4 shows an example of such plumes. Since the anvils of these severe storms were already at the tropopause level, the plumes must have been higher up in the stratosphere. In one case, Levizzani and Setvák (1996) estimated that the

The Wisconsin Dynamic/Microphysical Model

441

plume was located at ~15 km which was about 3 km above the local tropopause at the time. Since water vapor in the stratosphere has significant implications on the global climate process because of its strong absorption of IR radiation, it is important to understand where the water vapor source is and how much is injected into the stratosphere.

Figure 2. Simulated hydrometeor mixing ratios and storm-relative vector wind projection field for the CCOPE storm of 2 August 1981, in x–z (east–west) vertical cross sections through the maximum updraft as of 90 min. Solid – snow, dotted – graupel and hail, dashed – cloud droplets, dash dot – rain, dash dot dot – cloud ice.

CCOPE Supercell 2 Aug 1981 y = 27 km t = 40 min 0

0 0.5 0

18

1

0

-0.2 -1 5

01

-0.25

0 0. 5

7

-0.25

10

5 -0.2

-1

-0.5 3 -0.5 0.5 0-3 -1 7 10 -0.25 20 1 30

0

-1

1

.5 -0

-0.5

0

5

1 -1 5

3

1 0 0.5 -1

-0.5

-0 5 .2

25

0

-0.5

7

0.5 -3

20

0

15

-0 .2 5

0.5

10

-0.5

z (km)

0. 5

-3

30

7

0

10

-0 .5

-5

-0.5

0

5

-0.2 05.5 1

0

.2 5

-0.25

-0

0

20

.25

-3

-0

4

-1

-3

-1

-0.5

6

2

5

1

-0.5

0 0.5

8

0

0

-0.5 -1

-5 3 40

-0.2 5

10

5

12

-1

0

0. 5 1

-0.2

0

0.5

0

14

-0-0 .2.5 5 1

25 -0.

16

30

35

40

45

50

55

x (km)

Figure 3. The vertical velocity (w) field in the central east–west cross section of the simulated storm at t = 40 min. Solid (dashed) contours represent positive (negative) w (in m s–1).

442

P. K. Wang

Figure 4. GOES-8 composite image (cannel 1 + channel 3 + channel 4) on 6 May 2002 0015 UTC showing plumes on top of thunderstorms (white arrows) in Texas and Oklahoma border region (courtesy of NOAA).

Figure 5. Snapshots of modeled RHi (relative humidity with respect to ice) profiles at t = 24, 32, 40, 80, 96, and 112 min in the central east–west cross section (y = 27 km), showing the plume feature above the anvil. Only the portion near the cloud top is shown. The vertical axis range is 10–20 km and horizontal axis range 20–55 km (from Wang 2003).

The Wisconsin Dynamic/Microphysical Model

443

In order to identify the physical mechanism responsible for the plume formation and the source of its water vapor, we studied the simulation results of the CCOPE supercell to see whether or not the simulated storm produces the same plume phenomenon. The answer turned out to be positive. Figures 5 and 6 show the simulated plume phenomenon in the central cross-sectional view and 3D cloud top view, respectively (Wang 2003). The model-produced plumes exhibit nearly the same major characteristics of the observed ones, hence it is highly likely that the observed plumes must have been produced in a manner similar to that occurred in the simulated storm. Careful analysis of the model results indicate that the moisture forming the plumes come from the storm below, and the mechanism that eject moisture from the troposphere into the stratosphere is the breaking of cloud top gravity waves (Wang 2003). This demonstrates that the results generated by the cloud model are realistic and can be used as a substitute (when appropriate) for studying physical processes in thunderstorms whereas in situ or remote observations are either difficult or can provide only limited temporal and spatial coverage.

Figure 6. Snapshots of 3D renderings for the 30% RHi contour surface at t = 24, 32, 40, 80, 96, and 112 min, showing the plume features above the anvil. Data below 10 km are windowed out. (From Wang 2003.)

P. K. Wang

444

2.1.4.

Simulation of Fujita’s jumping cirrus

Fujita (1982) described the observation of the jumping cirrus phenomenon above a thundercloud from an aircraft as follows: One of the most striking features seen repeatedly above the anvil top is the formation of cirrus cloud which jumps upward from behind the overshooting dome as it collapses violently into the anvil cloud.

Figure 7. Snapshots of the RHi profiles in the central east–west cross section of the simulated storm from t = 1,320 –2,640 s. The RHi scale is similar to that in Fig. 5.

The Wisconsin Dynamic/Microphysical Model

445

In a later paper he made a more elaborated report on this phenomenon (Fujita 1989). Again due to the limited temporal and spatial coverage that can be provided by such observations, it is generally difficult to obtain detailed data for conclusively interpreting this phenomenon. At the time there were questions about whether it is possible to have cirrus clouds jumping upstream (i.e., against the wind). But by carefully studying the cloud top dynamics using the same WISCDYMM-simulated results of the CCOPE supercell, it can be seen that the gravity wave breaking related to the anvil top plume phenomenon is also responsible for the jumping cirrus phenomenon. The cirrus is not really jumping against the wind but it moves upstream only relative to the storm. If we keep in mind that the simulated storm is moving at ~30 m s–1 during its development, then the cirrus is still moving downwind relative to the surface. Figure 7 shows a series of the rendered cloud top humidity profiles that illustrate this process. Acknowledgments: This work is partially supported by NSF Grants ATM0234744 and ATM-0244505 to the University of Wisconsin-Madison.

3 REFERENCES Anderson, J. R., K. K. Droegemeier, and R. B. Wilhelmson, 1985: Simulation of the thunderstorm subcloud environment. Prepr. 14th Conf. Severe Local Storms. Indianapolis, IN., Amer. Meteor. Soc., 147–150. Cotton, W. R., G. J. Tripoli, R.M., and E. A. Mulvihill, 1986: Numerical simulation of the effects of varying ice crystal nucleation rates and aggregation processes on orographic snowfall. J. Climate Appl. Meteorol., 25, 1658–1680. Cotton, W. R., M. A. Stephens, T. Nehrkorn, and G. J. Tripoli, 1982: The Colorado state University three-dimensional cloud model – 1982. Part II: An ice phase parameterization. J. Rech. Atmos., 16, 295–320. Fujita, T. T., 1982: Principle of stereographic height computations and their application to stratospheric cirrus over severe thunderstorms, J. Meteor. Soc. Japan., 60, 355–368. Fujita, T. T., 1989: The Teton-Yellowstone tornado of 21 July 1987. Mon. Wea. Rev., 117, 1913–1940. Johnson, D. E., P. K. Wang, and J. M. Straka, 1994: A study of microphysical processes in the 2 August 1981 CCOPE supercell storm. Atmos. Res. 33, 93–123. Klemp, J. B. and R. B. Wilhelmson, 1978: The simulation of three-dimensional convective storm dynamics. J. Atmos. Sci., 35, 1070–1096. Knight, C. A. (ed.), 1982: The Cooperative convective precipitation experiment (CCOPE), 18 May–7 August 1981. Bull. Amer. Meteor. Soc., 63, 386–398. Levizzani, V. and M. Setvák, 1996: Multispectral, high resolution satellite observations of plumes on top of convective storms. J. Atmos. Sci., 53, 361–369. Lin, H.-M. and P. K. Wang, 1997: A numerical study of microphysical processes in the 21 June 1991 Northern Taiwan mesoscale precipitation system. Terres. Atmos. Oceanic Sci., 8, 385–404.

446

P. K. Wang

Lin, Y. L., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22, 1065–1092. Miller, L. J., D. Tuttle, and C. A. Knight, 1988: Airflow and hail growth in a severe northern High Palins supercell. J. Atmos. Sci., 45, 736–762. Setvak, M., R. M. Rabin, C. A. Doswell III, and V. Levizzani, 2002: Satellite observations of convective storm tops in the 1.6, 3.7 and 3.9 µm spectral bands. Atmos. Res., 67–68, 607–627. Straka, J. M., 1989: Hail growth in a highly glaciated central High Plains multi-cellular hailstorm. Ph.D. Diss., Dept. Meteorology, University of Wisconsin, Madison, WI, 413 pp. Tremback, C. J. P., W. R. Cotton, and R. A. Pielke, 1987: The forward-in-time upstream advection scheme: Extension to higher order. Mon. Wea. Rev., 115, 540–555. Wade, C. G., 1982: A preliminary study of an intense thunderstorm which move across the CCOPE research network in southeastern Montana. Proc. 9th Conf. on Weather Forecasting and Analysis. Seattle, WA, Amer. Meteor. Soc., 388–395. Wang, P. K., 2003: Moisture pumes above thunderstorm anvils and their contributions to cross tropopause transport of water vapor in midlatitudes. J. Geophys. Res., 108 (D6), 4194, doi: 10.1029/2003JD002581. Wang, P. K. et al., 2001: A cloud model interpretation of the enhanced V and other signatures atop severe thunderstorms. Prepr. 11th Conference Satellite Meteorology and Oceanography, 15–18 October 2001, American Meteorological Society, Madison, WI, 402–403.

35 THE EUROPEAN CENTRE FOR MEDIUMRANGE WEATHER FORECASTS GLOBAL RAINFALL DATA ASSIMILATION EXPERIMENTATION Peter Bauer1, Philippe Lopez1, Emmanuel Moreau2, Frédéric Chevallier3, Angela Benedetti1, and Marine Bonazzola1 1

European Centre for Medium-Range Weather Forecasts, Reading, UK NOVIMET, Velizy, France 3 Laboratoire des Sciences du Climat et l'Environnement, Paris, France 2

1 INTRODUCTION The quality of today’s numerical weather prediction (NWP) systems is driven by the quality of data that is used to determine the present state of the atmosphere and the quality of the representation of physical processes in the model. The data usage is optimized if the analysis, that is the methodology for deriving the most realistic state of the atmosphere–land–ocean system at a given time, is capable of capturing the four-dimensional (4D) development of this highly variable system in accordance with observations that are distributed in space and time. Principally, this represents an optimization problem because a state is sought that agrees best with a priori information from a short-range forecast that is based on a previous analysis and observations. The observations may be in geophysical terms (e.g., temperature or humidity) or electromagnetic terms (e.g., radiance or reflectivity). As in all optimization problems, the errors associated with each component provide a weight for each component in the analysis. At ECMWF, 4D data assimilation was first implemented in 1997 (e.g., Rabier et al. 2000) and run very successfully for global analyses and forecasts since then. A fundamental principle of the assimilation at ECMWF is its “incremental formulation”. In essence, this means that the short-range forecast is accurate enough to assume a linear dependence of model physics on state variable increments in the vicinity of the forecasted (first-guess) 447 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 447–457. © 2007 Springer.

448

P. Bauer et al.

state. The increments are produced from the difference between model and observation related states (Courtier et al. 1994). Figure 1 summarizes the logical flow. A high-resolution (~45 km in the horizontal and on 60 model levels) calculation of the model trajectory over the (12-h) assimilation window is carried out that is initialized with a short-range forecast from the previous analysis. The model state is compared to observations and interpolated to a lower resolution. The first minimization loop (inner loop) is based on the lowresolution trajectory model fields and the departures between observations and the high-resolution trajectory model fields. The resulting increments serve for updating the high-resolution fields. At the first iteration, these originate from the first guess. After, these are replaced by the higher-iteration updates (outer loop).

Figure 1. Incremental 4D-Var data assimilation algorithm (for details see text). x = model control variable, S = interpolation operator, d = departures, H = observation operator, y = observations, J = cost function, δ = increment, ∇ = gradient, indices ‘0’, ‘b’, ‘i’, ‘a’ denote control variable before assimilation, background, during update, and from analysis, respectively. Block arrows indicate temporal integration (Tremolet 2004).

Data from several observation types is assimilated. The bulk of the observations is obtained from satellites and their impact has recently proven to exceed the impact of conventional observations even in the northern hemisphere (Simmons and Hollingsworth 2002). Secondly, the difference in

The European Centre for Medium-Range weather Forecasts

449

forecast skill for the northern and southern hemispheres has been reduced dramatically indicating the beneficial impact of satellite observations in otherwise data-sparse areas. While the forecast of the dry atmospheric state (wind, temperature, surface pressure) has reached an unprecedented accuracy, the quality of both humidity analysis and forecast are still unsatisfactory due to the insufficient accuracy of the parameterization of moist physical processes and due to insufficient observations. The parameterization schemes have to cover the response of the atmospheric moisture fields to large-scale dynamics, cloud and precipitation formation, and fluxes at the surface–atmosphere boundary. Beljaars (2003) assesses the ECMWF model performance with respect to the atmospheric moisture representation. His main conclusions are that too much precipitation is generated over the entire dynamic range of rain rates but that the model has a dry bias outside areas with precipitation. The precipitationbias and its reflection in total column water vapor (TCWV) was confirmed by Marécal et al. (2001, 2002) when comparing near-surface rain estimates from satellite data and model fields, as well as the atmospheric moisture required for producing the respective rainfall intensities. The clear-sky dry bias has been noted at least since the assimilation of satellite data that is almost entirely sensitive to the TCWV (Gérard and Saunders 1999) and its manifestation in the so-called tropical precipitation spin-down that is the overproduction of rain at the beginning of the forecast originating from the moistening of the atmosphere in the analysis. This, of course, feeds back into the large-scale dynamics through the release of latent heat. Apart from these problems, the global analyses are biased towards clearsky observations because almost without exception cloud and rain contaminated data are rejected. This is because: • Large discrepancies between model forecast and observations can be expected due to oversimplified parameterizations. • The observation operator, i.e., the model that translates between model state variables and observables may be nonlinear due to the strongly nonlinear response of cloud and precipitation schemes to moisture increments (Marécal and Mahfouf 2003), as well as due to the nonlinear relationship between radiances and water vapor/condensates. • Even if these observations were assimilated, the effect could dissipate over the subsequent forecasts. The first two aspects may be in conflict with the general linearity assumption in the incremental 4D-Var formulation and can lead to convergence failures in the minimization. The last issue may lead to a negligible impact on model forecasts and therefore may not solve the fundamental shortcomings in the models’ description of the water cycle. Therefore, the assimilation of observations related to clouds and precipitation represents a major challenge for NWP and requires the combination of

450

P. Bauer et al.

better parameterizations, a numerically efficient and stable observation operators, and quality-controlled observations. The experience with operational rainfall assimilation produced mixed results due to its little impact on model dynamics in existing data assimilation systems (e.g., since 2001 in NCEPGDAS, since 2002 in JMA-MSM). However, the ECMWF 4D-Var system provides an ideal environment for evaluating the optimum configuration for rainfall data assimilation given the current quality of moist physical parameterizations at the highest available horizontal and vertical resolution of the ECMWF model.

2 1D-VAR Since 1998, large efforts have been made to prepare the assimilation of precipitation information at ECMWF. A major part of the studies focused on the above issues for which Marécal and Mahfouf (2000, 2002) have laid the foundation for a methodology that became operational at ECMWF in 2005. In this methodology, a one-dimensional variational retrieval (1DVar) algorithm is applied with near-surface rain rates or passive microwave brightness temperatures (TBs) as observables and TCWV as the retrieval variables. The main reason for subsetting the 4D-Var analysis with a 1D-Var retrieval of TCWV is the nonlinear relationship between the control vector and the precipitation affected observations that may violate the incremental 4D-Var setup. Secondly, the lower model resolution and the simplified physics schemes in the inner loop (~120 km) may deteriorate the convergence if precipitation observations would be assimilated directly in 4D-Var. These limitations may be overcome in the near future with a better model resolution and more consistent physics included in the inner loop. The observation operator that is used for the 1D-Var algorithm uses new linearized cloud (Tompkins and Janisková 2004) and convection schemes (Lopez and Moreau 2004), as well as a radiative transfer model (Bauer 2002; Moreau et al. 2003a). The latter has recently been integrated into the operational RTTOV package (Saunders et al. 1999) that is available for a large NWP community. One of the issues in the 1D-Var retrieval is whether to use near-surface rainfall estimates or TBs as observables. Both options were analyzed by Moreau et al. (2003b). Figure 2 shows an example of TCWV increments that were produced by the 1D-Var retrieval using rainrates (1D-RR) or TBs (1DTB) from Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) observations over tropical cyclone “Mitag” on 3 March 2002. The different coverage originates from different screening methods. In this example, 1D-TB tends to produce larger increments due to less saturation at larger rain intensities if all lower TMI channels between 10 and 37 GHz are used (Moreau et al. 2003a, b).

The European Centre for Medium-Range weather Forecasts

10 5 3 2 1 0.5 −0.5 −1 −2 −3 −5 −10 −20 121E 123E 125E 127E 129E 131E 133E 135E 137E 139E

30

21

RbR0

40

50 60 −0.3 0.0 0.3 Mean T increments (K)

30 Model levels

21

RbR0

20

19N 18N 17N 16N 15N 14N 13N 12N 11N 10N 9N 8N 7N 6N

10 5 3 2 1 0.5 −0.5 −1 −2 −3 −5 −10 −20 121E 123E 125E 127E 129E 131E 133E 135E 137E 139E

21

21

RbR0

30

30

40

40

50

50

60 −1.5 0.0 1.5 -1 Mean Qv increments (g kg )

60 −0.3 0.0 0.3 Mean T increments (K)

Model levels

20 19N 18N 17N 16N 15N 14N 13N 12N 11N 10N 9N 8N 7N 6N

451

RbR0

40

50 60 −1.5 0.0 1.5 -1 Mean Qv increments (g kg )

Figure 2. TCWV increments (kg m–2) from 1D-Var retrievals using rain rates (top left panel) and brightness temperatures (top right panel). Bottom panels show averaged profiles of temperature and specific humidity increments, respectively. Cyclone “Mitag” on 3 March 2002, 1200 UTC.

Figure 2 also shows mean profiles of temperature and humidity increments produced by the 1D-Var. If temperature increments are transferred to humidity by assuming saturation, their values are 5–10 times smaller than those of humidity itself. This suggests the dominance of the sensitivity of cloud and convection schemes to moisture changes and that it is reasonable to only assimilate moisture increments in 4D-Var (through TCWV) following the 1D-Var retrieval. The profiles also show that increments are larger with 1D-TB. For 1D-RR, the profile shapes are rather homogeneous because they are only produced by the moist-physics parameterizations and the shape of the background error standard deviation profiles. For 1D-TB, also the sensitivity of the TBs to changes in hydrometeor profiles (TB-Jacobians) plays a role. These peaks at different altitudes per channel and hydrometeor type (cloud water, rain, snow). The maximum of increments near model level 50 (850–900 hPa) is a result of the domination of the background error standard deviation profile. When investigating the 1D-Var performance more thoroughly, Moreau et al. (2003b) found no clear indication that 1D-RR is outperformed by 1DTB. However, 1D-RR requires retrieval algorithms that invert TBs to rain rates and that provide a retrieval error estimate that is compatible with the

452

P. Bauer et al.

model rainfall error estimate (Bauer et al. 2002). There is a rather large range of products for the estimation of near-surface rainfall retrieval accuracy from comparisons to local reference measurements; however, efforts to estimate global retrieval accuracy have not been successful in identifying a single retrieval approach as being superior to others (e.g., Ebert and Manton 1998). The choice of TBs as observables circumvents this problem, but only partially, because a proper estimation of modeling errors from the observation operator (cloud/convection + radiative transfer) is required for the 1D-Var algorithm as well. Practically, the error estimation can be solved from the requirement that the departures between observed and modeled TBs are consistent with the errors associated with the model variables (expressed in TB-space) and those of the observations. Therefore, the error estimation aims at a balance between realistic observation errors, model forecast errors (again in TB-space), and the difference between observations and model simulations in the analysis. The biggest advantage of using TBs, however, is the larger flexibility with regard to which channel and which sensor is employed compared to algorithms that are usually sensor-specific. The results of Moreau et al. (2003b) also indicate that 1D-TB does not require the presence of rain or clouds in the background because TBs are sensitive to clear-sky TCWV as well. The usage of 1D-RR may increase/decrease rain when present in the background but cannot create rain where the background is rain-free because there, the derivative of model state with respect to rain is not defined. This is illustrated in Fig. 3 by comparing the model background, 1D-RR and 1D-TB, as well as TRMM precipitation radar (PR) data over tropical cyclone “Ami” on 14 January 2003, 18 UTC. All states are expressed in equivalent PR reflectivity. The PR observations show that the maximum rain intensities in cyclone “Ami’s” rain-band are displaced with respect to the model background. Both 1D-Var retrievals succeed to correct the background but 1D-RR fails to produce sufficient rainfall near the maximum of the rain intensities at 26S/172W due to missing rainfall in the background.

3 1D-VAR + 4D-VAR Figure 4 shows examples of precipitation forecasts based on a control experiment (operational ECMWF model) and 4D-Var analyses using 1D-RR and 1D-TB TCWV retrievals, respectively. The rain patterns and rain intensities are significantly modified by assimilating rain affected observations. For both 1D-RR and 1D-TB, the cyclone intensity is increased and this intensification is maintained over the selected forecast period. The fact that the rain assimilation can displace precipitating systems ensures that there remains a continuous impact in the forecast with respect to cloud system location. However, when regionally averaged precipitation intensities are analyzed, no significant change of the hydrological budget was observed (not shown here).

The European Centre for Medium-Range weather Forecasts

453

Figure 3. Radar reflectivities from TRMM PR (upper left), model background (upper right), 1D-RR and 1D-TB analyses. Tropical cyclone “Ami” on 14 January 2003, 1800 UTC.

Figure 4. Forecasts of accumulated precipitation (mm) from control, 1D-TB and 1D-RR assimilation experiments (from left to right) vs. rain-rate retrieval (lower right panel). Cyclone “Zoe” on 26 December 2002, 1200 UTC. Color scale: 0.1, 0.3, 1, 2, 3, 5, 10, 20, 50, 100 mm.

454

P. Bauer et al.

Figure 5 shows the global distribution of TCWV analysis increments from a 1D-Var + 4D-Var experiment using TMI rain-rate observations. Figure 5a contains the humidity increments of all observations while Fig. 5b isolates the increments that originate from the assimilation of precipitation information. The nature of the latter is featured in the finer structures and the change in increment sign on small scales. This is the result of the displacement of rain systems, i.e., the drying of the model background where there was no or little rain in the observation and vice versa. The magnitude of increments of all and rain observations is comparable. The 4D-Var analysis spreads humidity increments beyond the limits of data availability (40N– 40S) and therefore the assimilation of TMI data has a global impact. The direct impact of moisture increments on the precipitation forecast is illustrated in Fig. 5c that contains the accumulated precipitation between days 1 and 2 of the forecast period. Here, the displacement of precipitating systems is even more obvious than from the increments in Fig. 5b.

4 CONCLUDING REMARKS In the way the assimilation of precipitation information is carried out at ECMWF, an observation operator is required that translates between model control variables, e.g., temperature and humidity, and observables, e.g., rain rates, brightness temperatures, or reflectivities. This operator contains physical parameterizations for cloud and rain generation, as well as a radiative transfer model. The constraint on the model analysis that is produced by these observations depends on the balance between model background and observation error statistics. If profile information is assimilated, the vertical distribution of analysis increments is crucial as well because it will affect the response of the model physics to the assimilated data in the forecast. For example, identical moisture increments added near the bottom or the middle of the troposphere will have a different effect on cloud and rain generation during the forecast. All the above elements contain rather large uncertainties and the success of rainfall assimilation is strongly determined by the accuracy of the observation operator and the protection of the analysis from those observations that can lead to inconsistencies in the minimization. This safety requirement is even more difficult to fulfill because many other observations are assimilated in the vicinity of precipitation whose effect may interact with that of observations inside precipitation. Most of the recent work carried out at ECMWF dealt with the implementation of a computationally efficient yet accurate observation operator and with the investigation of the 4D-Var analysis performance once rain information is assimilated.

The European Centre for Medium-Range weather Forecasts

455

Figure 5. Analysis increments of TCWV (in kg m–2) on 7 April 2003, 0000 UTC from all observations (a) and rain observations (b). (c) 48–24 h precipitation forecast (in 10–3 mm) initialized on 1 April 2003, 1200 UTC with rain observations (see also color plate 16).

456

P. Bauer et al.

Due to the remaining uncertainties in this process, above all the comparably simple physical parameterization schemes applied in global models, some longer-term assimilation studies have been started to optimize the usage of rainfall information in the ECMWF assimilation system prior to its operational implementation. Since the model physics reacts sensitively to a change of moisture state, future options will include the combined use of microwave observations that are sensitive to liquid precipitation in lower layers and frozen precipitation and cloud water/ice in the upper layers. First studies on rainfall radar data assimilation at ECMWF have indicated that the vertical distribution of moisture increments can have a significant effect on the latent heating profile and therefore the relation between moisture and energy budget. In summary, the assimilation of rainfall and cloud information offers a large potential for future applications and its success can be considered as a benchmark towards improving moist physical parameterizations in global modes and data assimilation systems, as well as the prediction of severe weather systems.

5 REFERENCES Bauer, P., 2002: Microwave radiative transfer modeling in clouds and precipitation. Part I: Model description. NWP SAF Report No. 5. Available from The Met Office, Exeter, UK, pp. 27. Bauer, P., J.-F. Mahfouf, S. di Michele, F. S. Marzano, and W. S. Olson, 2002: Errors in TMI rainfall estimates over ocean for variational data assimilation. Quart. J. Roy. Meteor. Soc., 128, 2129–2144. Beljaars, A., 2003: Some aspects of modelling of the hydrological cycle in the ECMWF model. Proceedings of ECMWF/GEWEX Workshop on Humidity Analysis, 8–11 July 2002, ECMWF, Reading, UK, 191–202. Courtier, P., J.-N. Thépaut, and A. Hollingsworth: A strategy for operational implementation of 4D-Var using an incremental approach. Quart. J. Roy. Meteor. Soc., 120, 1367–1388. Ebert, E. and M. Manton, 1998: Performance of satellite rainfall estimation algorithms during TOGA-COARE. J. Atmos. Sci., 55, 1537–1557. Gérard, E. and R. Saunders, 1999: Four-dimensional variational assimilation of Special Sensor Microwave/Imager total column water vapour in the ECMWF model. Quart. J. Roy. Meteor. Soc., 125, 1453–1468. Lopez, P. and E. Moreau, 2004: A convection scheme for data assimilation: Description and initial tests. Technical report. ECMWF Technical Memorandum No. 411. Available from ECMWF Shinfield Park, Reading, UK, pp. 29. Mahfouf, J.-F., V. Marécal, and P. Bauer, 2003: The assimilation of SSM/I and TMI rainfall rates in the ECMWF 4D-Var system. Quart. J. Roy. Meteor. Soc., 128, 2737–2758. Marécal, V. and J.-F. Mahfouf, 2000: Variational retrieval of temperature and humidity profiles from TRMM precipitation data. Mon. Wea. Rev., 128, 3853–3866. Marécal, V. and J.-F. Mahfouf, 2002: Four-dimensional variational assimilation of total column water vapour in rainy areas. Mon. Wea. Rev., 130, 43–58.

The European Centre for Medium-Range weather Forecasts

457

Marécal, V. and J.-F. Mahfouf, 2003: Experiments on 4D-Var assimilation of rainfall data using an incremental formulation. Quart. J. Roy. Meteor. Soc., 129, 3137–3160. Marécal, V., J.-F. Mahfouf, and P. Bauer, 2002: Comparison of TMI rainfall estimates and their impact on 4D-Var assimilation. Quart. J. Roy. Meteor. Soc., 128, 2737–2758. Marécal, V., E. Gérard, J.-F. Mahfouf, and P. Bauer, 2001: The comparative impact of the assimilation of SSM/I and TMI brightness temperatures in the ECMWF 4D-Var system. Quart. J. Roy. Meteor. Soc., 127, 1123–1142. Moreau, E., P. Bauer, and F. Chevallier, 2003a: Variational retrieval of rain profiles from spaceborne passive microwave radiance observations. J. Geophys. Res., 108, ACL 11–1—11–18. Moreau, E., P. Lopez, P. Bauer, A. M. Tompkins, M. Janiskovà, and F. Chevallier, 2003b: Variational retrieval of temperature and humidity profiles using rain-rates versus microwave brightness temperatures. Quart. J. Roy. Meteor. Soc., 130, 827–852. Rabier, F., H. Jarvinen, E. Klinker, J.-F. Mahfouf, and A. Simmons, 2000: The ECMWF operational implementation of four-dimensional variational assimilation. Part I: Experimental results with simplified physics. Quart. J. Roy. Meteor. Soc., 126, 1143– 1170. Saunders, R. W., M. Matricardi, and P. Brunel, 1999: An improved fast radiative transfer model for assimilation of satellite radiance observations. Quart. J. Roy. Meteor. Soc., 125, 1407–1426. Simmons, A. and A. Hollingsworth, 2000: Some aspects of the improvement in skill of numerical weather prediction. Quart. J. Roy. Meteor. Soc., 128, 647–677. Tompkins, A. and M. Janiskovà, 2004: A cloud scheme for data assimilation: Description and initial tests. Quart. J. Roy. Meteor. Soc., 130, 2495–2517. Tremolet, Y., 2004: Diagnostics of linear and incremental approximations in 4D-Var. ECMWF Technical Memorandum No. 399. Available from ECMWF Shinfield Park, Reading, UK, pp. 16.

36 RAINFALL ASSIMILATION INTO LIMITED AREA MODELS Andrea Buzzi and Silvio Davolio ISAC-CNR, Bologna, Italy

Abstract

A physical assimilation technique based on humidity nudging has been developed for application to satellite-derived rainfall fields, in the framework of the European project “EURAINSAT”. The aim of the forcing procedure is to improve the short-range precipitation forecasts with particular attention to specific meteorological phenomena, such as heavy orographic precipitation and small-scale “hurricane-like” cyclones in the Mediterranean area. The nudging scheme forces the model humidity profile in order to get model precipitation closer to the observed precipitation. The forcing is a function of the difference between the rain rates, observed and forecasted, and of precipitation type, convective, or stratiform. In addition, a modelling tool to reproduce the idealised development of midlatitude baroclinic unstable modes, including humidity in the atomsphere and a full water cycle, has been developed with the purpose of investigating the effects and capabilities of assimilation of precipitation in an idealised frame. More realistic experiments have been also performed by implementing a lagged forecast procedure, in order to evaluate, with an observing system simulation experiment (OSSE)-type strategy, the scheme’s performance in terms of improvements of short-range precipitation forecasts and impact on the dynamics of the meteorological evolution. Finally, satellite rain estimates, based on combined microwave (MW) and infrared (IR) techniques, have been assimilated into the limited area meteorological model trying to improve the short-range precipitation forecasts.

Keywords

Rainfall data assimilation, nudging, cyclogenesis, satellite estimates

1 INTRODUCTION Accurate quantitative forecasting of precipitation, especially during severe weather episodes, is one of the most challenging tasks of meteorological modelling. Data assimilation techniques are devoted to attain this aim. In 459 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 459–470. © 2007 Springer.

460

A. Buzzi and S. Davolio

particular, the frequent assimilation of variables directly related to the formation of precipitation and the water cycle may contribute to a better definition of model latent heating, vertical velocity, and moisture and consequently may lead to an improvement of the short-range precipitation forecasts (Manobianco et al. 1994). Recently, the problem of assimilating precipitation data from different sources (satellite, radar, rain gauge, etc.) into limited area meteorological models has received increasing attention, not only in the tropical area, but also in midlatitudes. Since precipitation is not a prognostic model variable, but an end result of complex dynamical and microphysical processes, it cannot be directly assimilated into numerical weather prediction (NWP) models. However, rain observations can be used to correct humidity and temperature profiles, and consequently latent heat release, in order to obtain simulated precipitation closer to the reality. Even if somehow empirical, the nudging is a quite simple, physically based method. In spite of its simplicity, it proved to be suitable for rainfall assimilation for synoptic-scale and mesoscale numerical forecasting, also in an operational framework (Falkovich et al. 2000; Macpherson 2001). The aim of this study is to investigate the effects and capabilities of the assimilation of precipitation both in an idealised and realistic frame. Therefore, the nudging technique has been implemented in a periodic channel version of the meteorological model BOLAM (Buzzi et al. 2003), which reproduces the idealised development of a midlatitude baroclinic unstable mode at finite amplitude. In particular, the application of rainfall assimilation to a concepttual model of cyclogenesis is useful to explain real assimilation results and persistence properties and provides a theoretical basis to the interpretation of the impact of assimilation on the dynamics of midlatitude-growing cyclones. Then the nudging scheme has been applied to the limited area version of the same model, used to analyse a couple of severe weather episodes.

2 THE NUDGING SCHEME The precipitation assimilation scheme has been developed on the basis of a procedure proposed by Falkovich et al. (2000) that modifies the specific humidity profile of the model according to the difference between observed and forecast rain rate. Moisture changes lead to changes of temperature and other dynamical variables through the model precipitation scheme (explicit and convective parameterisation). The procedure starts with a comparison between the forecast (Rm) and target (Rt) total precipitation accumulated over a suitable period of time. After rain-rate comparison, moisture profiles are nudged at grid points where the two values differ, according to the following equation:

Rainfall Assimilation Into Limited Area Models

461

∂q ( k ) = −ν S ,C ( k )τ −1 q ( k ) − ε S ,C q * ( k ) ∂t

{

}

(1)

where k is the model σ-level, q(k) is the specific humidity profile prior to the nudging, q*(k) is the saturation humidity profile (obtained from the model), τ is a relaxation time, εs,c is an over/undersaturation coefficient and νs,c(k) is a vertical modulation profile, whose value may vary in the interval [0–1]. A mean constant rain rate is assumed for the target precipitation within the accumulation interval. Once the model precipitation is available, that is at every convective adjustment time step (about 20 min), rain rates are computed (for both observed and forecast precipitation) using the rainfall accumulated up to the current time step, when the comparison takes place. Therefore, the scheme does not instantaneously adjust the rain rate at each time step, but rather compares and adjusts the rain accumulated until the current time step, seeking to recover the observed precipitation at the end of the accumulation interval. Convective precipitation and stratiform precipitation are handled differently. Model-generated precipitation is used in order to discriminate the precipitation type at a specific grid point. Different vertical modulation profiles νs,c(k) and different coefficients εs,c are used in the two cases in order to introduce/remove humidity only where it is needed. For stratiform precipitation νs(k) is defined in such a way that the humidity is changed only in the middle–lower troposphere, where it is assumed that most of the large-scale condensation takes place. Conversely, in case of pure convective precipitation νc(k) is such that humidity is changed mainly in the boundary layer, which is assumed to represent the source of humidity for the convective adjustment. In case of stratiform precipitation, if the model underestimates the rainfall with respect to the observed value (Rm < Rt), then q(k) is forced towards a slightly supersaturated profile ε S+ q*(k). If the model overestimates the precipitation, then q(k) is decreased gradually towards a subsaturated value ε S− q*(k). In case of convective precipitation, if Rm < Rt, then q(k) is forced

gradually towards a slightly undersaturated profile ε C+ q*(k). If Rm > Rt, then q(k) is decreased gradually towards a low relative humidity value ε C− q*(k). In case of coexistence of both types of precipitation, the sum of the modulating profiles does not exceed unity. At locations where rainfall is observed but not forecasted, both types of precipitation are provisionally considered, unless all the surrounding grid points are experiencing only one type of precipitation. If this is the case, only the appropriate modification, for large scale or convective rainfall, is applied.

462

A. Buzzi and S. Davolio

As for the computed convective (and all physical) tendencies, the nudging adjustment is distributed over all the time steps in the interval between two times at which rain rates are compared.

3 APPLICATION TO A CONCEPTUAL MODEL The basic conceptual models of mid- and high-latitude cyclogenesis were proposed by Petterssen and Smebye (1971), with a twofold classification (types A and B). Type A is considered to be closer to the classical theoretical Eady and Charney models of normal mode growth on a westerly baroclinic flow, while type B has a more local character, being related to the interaction between an upper level potential vorticity (PV) anomaly with a low-level thermal (or PV) anomaly. This basic classification has been supported also in recent observational studies (e.g., Deveson et al. 2002). In the last decade, however, the fundamental dynamical role played by humid processes in the development of cyclones has been emphasised in theoretical, numerical, and observational studies. The release of latent heat in general increases the growth rate, or even destabilises otherwise stable modes, at least for relatively uniform distributions and small amplitude perturbations of the mean westerly flow. In cases of finite amplitude and/or localised disturbances, however, the role of humidity is more complex and can be partly interpreted in terms of generation of low and midtropospheric-positive PV anomalies, interacting with the upper level PV anomalies (see, e.g., Fantini 2004). In this respect, a new type (C) of cyclogenesis has been identified in the literature, in which the latent heat exchange plays a major role and makes the cyclone characteristics very different from the classical conceptual models of Petterssen and Smebye (Plant et al. 2003). In view of the importance of the diabatic humid effects on cyclones and of the role played by precipitation assimilation techniques in altering the latent heat release in a growing cyclone, the effects of such assimilation on the growth properties (growth rate, phase speed) and structure of simple unstable modes have been investigated. In the present idealised framework, the impact of the application of the assimilation in terms of persistence of the modifications introduced can be evaluated more easily and more generally than in individual case studies. The simulations have been performed using a channel version of the BOLAM model. BOLAM is a primitive equation, hydrostatic, limited area meteorological model, whose description can be found in Buzzi et al. (2003). The precipitation assimilation system has been adapted to build up a conceptual model of baroclinic growth modified by the assimilation. The model geometry represents a midlatitude channel, comprised between 18 and 68 degrees N and extending in longitude with a period of 44 degrees. The adopted horizontal grid resolution is of about 50 km. An idealised initial state representing an unstable baroclinic zonal jet is defined, with a

Rainfall Assimilation Into Limited Area Models

463

meridional mean temperature difference of 28 K (applied also to the sea surface temperature), a maximum wind speed of about 30 m s–1 above the simulated tropopause, and a relative humidity decreasing from about 90% near the north wall to about 60% near the south wall (relative humidity is not constant to prevent the development of convective instability near the southern boundary). A barotropic wind perturbation in geostrophic equilibrium is prescribed in the initial condition, sinusoidal in longitude, and vanishing near the lateral walls. A 120-h simulation is considered as reference. Starting from the same initial condition, additional integrations have been performed, including a 24-h period of assimilation, starting after 36 h of unconstrained forecast. Therefore, the control and nudging simulations are identical during the first 36 h. Then, the rainfall assimilation modifies the nudging runs. Three data assimilation experiments have been performed. Two experiments are considered first in order to assess the effect of precipitation amount, one in which the assimilated precipitation (target) is null, the other in which the amount is doubled with respect to the reference experiment, but keeping the same spatial distribution. The target data are accumulated over 3-hourly intervals. Another experiment has been conducted to explore the response of the growing mode to the introduction of longitudinal phase changes by the precipitation assimilation procedure. The target precipitation fields are constituted by the 3-hourly accumulated fields obtained in the reference experiments, but shifted in longitude. The purpose is to investigate the possibility of altering (reducing) phase errors in model fields by means of the assimilation of precipitation. The phase shift is prescribed of the order of a few hundreds of kilometres, corresponding to a few degrees, in longitude. After a few days of integration, an unstable baroclinic wave develops into a large amplitude cyclone and anticyclone couplet, with fully developed PV anomalies, frontal structures, and precipitation areas. The mean sea level pressure field after 84 h is depicted in Fig. 1 (left panel). This is considered as a reference case with respect to cases in which rainfall assimilation is applied. The rainfall assimilation has a strong impact on precipitation fields for the experiments with modified target intensity. At the end of the nudging period, the rainfall has been completely suppressed when the target precipitation is null, while has been remarkably increased in the case of doubled target precipitation. Moreover, it appears immediately that the strength of the system has been altered substantially (Fig. 1). For example, doubling the precipitation target has the effect of deepening the cyclone by about 6 hPa, while the efficient suppression of precipitation, and hence of the associated latent heat release, during the assimilation period has increased the surface pressure in the cyclone minimum by about 10 hPa. The two experiments mentioned above indicate that the assimilation has a strong and long-lasting effect on this kind of development, by altering the growth

464

A. Buzzi and S. Davolio

rate of the system for a significant period of time, either increasing or decreasing the important contribution of latent heat release on the cyclone life cycle.

Figure 1. Mean sea level pressure evolution at day 3.5 (84 h) of an unstable model in the zonal channel. Left panel: reference experiment. Centre panel: assimilation to double rainfall rate than in the reference. Right panel: assimilation to null precipitation.

Figure 2. 3-h accumulated precipitation at the end of the assimilation period, after 60 h of integration. Left: reference experiment. Centre: target precipitation (as the reference run, but shifted in longitude). Right: assimilation run. Contour interval is 2 mm.

If, instead of altering the amount of target precipitation, the same pattern is maintained but shifted in longitude, the forcing during the assimilation phase seems again able to modify the forecast precipitation field (Fig. 2), even if the rainfall intensity is slightly weaker. A comparison between the tempe-

Rainfall Assimilation Into Limited Area Models

465

rature fields at 850 hPa, of the reference and nudged case, in the region of the warm intrusion associated with the precipitation area, at 72 h (i.e., 12 h after the end of the assimilation period) emphasises the persistence of the modified temperature structure and the intensification of the warm front. This reflects the local alteration of the dynamical structure of the mode introduced by the nudging. However, at variance with the case in which precipitation amount is altered substantially, in this case the changes due to the assimilation are reabsorbed in the subsequent further growth of the system, without introducing an appreciable durable change in the spatial phase of the global structure. This can be expected in cases in which the growing mode has a coherent structure, while the perturbation introduced by the transient nudging is likely to project on decaying components, or on components growing more slowly than the dominant mode. In the case in which the amount of precipitation, and hence of latent heat release, is changed substantially, the entire growing structure is altered.

Figure 3. Evolution of the equitable threat score for thresholds of 2 mm/6 h (left panel) and 10 mm/6h (right panel), for the forecast F (dashed) and the nudging run N (solid), computed for 6-h accumulated precipitation. Number in brackets indicates number of observations (control run) exceeding the threshold value. The vertical dashed line indicates the end of the nudging period.

4 APPLICATION TO A NUMERICAL METEOROLOGICAL MODEL The nudging procedure has been tuned and extensively tested in an idealised framework but using realistic meteorological fields (OSSE-type strategy), by implementing a lagged forecast scheme. Two severe weather events, both characterised by heavy rainfall, have been selected for this purpose. The first occurred in the region south of the Alps in September 1999 and was

466

A. Buzzi and S. Davolio

extensively observed during the Mesoscale Alpine Programme (MAP) fieldphase. It was associated with the passage of a frontal system and with heavy orographic rainfall. The second, which caused a major flood in the city of Algiers in November 2001, was characterised by the development of a very intense Mediterranean cyclone. Only the experiment results concerning the second event are presented here. This case study represents a suitable opportunity for evaluating the assimilation procedure, in terms not only of improvements in precipitation forecasts, but also of impact in short-range forecasting of a cyclogenesis event. Since the latent heat release and surface heat fluxes played a crucial role in the development of the Mediterranean cyclone, it is expected that a nudging procedure that modifies the humidity profiles had an appreciable impact on cyclogenesis. The nudging procedure has been applied to the BOLAM model and the lagged forecast scheme has been implemented, performing three simulations as follows. The first simulation (C) consisted of a 36-h run, initialised 10 November, 1200 UTC. The second (F) was initialised 12 h before, at 0000 UTC of the same day and lasted 48 h. C represents the reference state and provides the precipitation target data, while F is considered the “real” forecast to be improved (compared to C). Finally, a third 48-h simulation (N) was performed, starting from the same initial condition as F, but applying the nudging procedure for 12 h, from 10 November, 1200 UTC. Two-hourly model rainfall data extracted from C are used for the assimilation, with the aim of forcing the forecast towards C. The assimilation considerably improves the precipitation forecasts at the end of the nudging phase. The improvements of the rainfall assimilation extend well beyond the forcing period, as shown by the evolution of the equitable threat score (ETS) in Fig. 3. For both low and high threshold values of rainfall, the benefit of the nudging seems to last at least 18 h during the unconstrained forecast, following the forcing. Cross sections in correspondence to the rain bands show that mesoscale vertical motion has been generated by the nudging procedure and improvements in the potential temperature and relative humidity vertical distribution have been achieved. The evolution of the Mediterranean cyclone as described by the basic forecast (F) remarkably differs from the control (C): in place of a sharp low (985 hPa at 1200 UTC, 11 November) surrounded by an area of more levelled pressure, centred south-east of the Balearic Islands, F produces a weaker large-scale pressure system without an intense core, whose centre appears to be displaced northward. After the rainfall assimilation and for almost the entire following unconstrained forecast period, the evolution of the low is improved both in intensity, timing and trajectory of the pressure minimum (Fig. 4).

Rainfall Assimilation Into Limited Area Models

467

Figure 4. Surface low centre trajectory (left panel) and central MSLP (hPa) (right panel) for the control run (left: right line from the bottom; right: upper line from the left), forecast (left: left line from the bottom; right: lower line from the left)) and nudging run (left: middle line from the bottom; right: middle line from the left) computed at 2-h intervals, from 1200 UTC, 10 November to 0000 UTC, 12 November 2001.

Figure 5. 12-h accumulated precipitation at 1200 UTC, 10 November 2001 as estimated from satellite (left panel) and for the reference forecast (R, right panel) (see also color plate 15).

The nudging scheme has been also applied to the Algerian flood event in a more realistic framework, attempting to assimilate satellite precipitation estimates (combined IR-MW, Kidd et al. 2003). Rainfall estimates, available every 30 min, have been interpolated to the limited area model grid, accumulated over 2-hourly intervals and then used as target for the assimilation. In order to avoid using data possibly affected by large errors, the assimilation was performed only over the sea. Two 48-h simulations have been initialised at 0000 UTC, 10 November 2001. The first represents the reference forecast (R), the second (N) is forced by the nudging scheme during the first 12-h period, using 2-hourly satellite data.

468

A. Buzzi and S. Davolio

The 12-h accumulated precipitation field, as forecasted by the model, presents differences with respect to the satellite precipitation estimate, especially around Sardinia (Fig. 5). In particular, the model misses the rainfall area west of the island, while produces an intense rain band over the Tyrrhenian Sea. The latter rain band is present also in the observations, but weaker, displaced eastward and affecting a smaller area. The assimilation seems to reduce the forecast error in terms of both intensity and location of the rainfall patterns (Fig. 6). The precipitation nucleus west of Sardinia is correctly generated in the form of a rain band, whose intensity is however lower than the target. Over the Tyrrhenian Sea the excess of rainfall is reduced and the patterns is more similar to the observations. Similar experiments have been performed using the same rainfall data but accumulated over different intervals. Assimilating hourly data gives results that are almost the same as for 2-hourly data. Only a slight improvement is observed concerning the rainfall pattern over the Tyrrhenian Sea, but not elsewhere. A clear degradation of the assimilation impact emerges when rainfall estimates accumulated over longer interval (6 h) are used. In this case, both the reduction and the increase of the predicted rainfall becomes less evident, although still present, and the forecast is not efficiently modified by the assimilation scheme, confirming that target data are too smooth in time and space for this type of application.

Figure 6. 12-h accumulated precipitation at 1200 UTC, 10 November 2001, for the assimilation run (N) (see also color plate 15).

5 CONCLUSIONS An original technique of assimilation of precipitation into numerical meteorological models, separating stratiform and convective precipitation, has been successfully applied to the meteorological model BOLAM. The capability of the rainfall assimilation in altering the intensity of an idealised model of baroclinic cyclone development in midlatitudes has been shown

Rainfall Assimilation Into Limited Area Models

469

using a periodic channel version of the model. The experiments described above support the assumption that the water cycle in a baroclinic atmosphere is particularly effective in determining the growth and the structure of the modes. As a consequence, the assimilation of precipitation can induce a strong modification of the structure and properties of the unstable modes, affecting in a more or less permanent way the system trajectory in phase space, as a function of the projection of the perturbations introduced on the growing mode spectrum. In this sense, the results based on the conceptual numerical model can be used to interpret the results obtained in the realistic framework. The assimilation procedure has been applied to the real case studies of heavy precipitation, using both OSSE technique to generate model data and rainfall satellite estimates. In these cases, the impact of the assimilation on short-range precipitation forecasts and on cyclone growth appears to be quite important. However, as also found in the conceptual model application, the technique is less effective in introducing spatial phase correction into the meteorological evolution, so that in this case the benefit of the rainfall assimilation has a rather short duration in time. The modification of the humidity profiles, and consequently of the latent heat release through the model precipitation scheme, produces improvements in the short-range precipitation forecasts. The positive impact of the assimilation is confined, however, to a limited period of time of the unconstrained forecast, following the assimilation phase. The impact is retained for about 12–24 h, depending on the particular meteorological situation. However, it is important to point out that better rainfall forecasts are associated to a better reproduction of the vertical motion, latent heating, and vertical profile in the rainy areas. Acknowledgement: This research was funded by the EURAINSAT project, a shared-cost project (contract EVG1-2000-00030), co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development Subprogramme, 5th Framework Programme).

6 REFERENCES Buzzi, A., M. D’Isidoro, and S. Davolio, 2003: A case study of an orographic cyclone formation south of the Alps during the MAP-SOP. Quart. J. Roy. Meteor. Soc., 129, 1795–1818. Deveson, A. C. L., K. A. Browning, and T. D. Hewson, 2002: A classification of FASTEX cyclones using a height-attributable quasi-geostrophic vertical-motion diagnostics. Quart. J. Roy. Meteor. Soc., 128, 93–118. Falkovich, A., E. Kalnay, S. Lord, and M. M. Mathur, 2000: A new method of observed rainfall assimilation in forecast model. J. Appl. Meteor., 39, 1282–1298.

470

A. Buzzi and S. Davolio

Fantini, M., 2004: Baroclinic instability of a zero-PVE jet: enhanced effects of moisture on the lifecycle of baroclinic cyclones. J. Atmos. Sci., 61, 1296–1307. Kidd, C., D. Kniveton, M. Todd, and T. Bellerby, 2003: Satellite rainfall estimation using a combined passive microwave and infrared algorithm. J. Hydrometeorology, 4, 1088– 1104. Macpherson, B., 2001: Operational experience with assimilation of rainfall data in the Met Office mesoscale model. Meteor. Atmos. Phys., 76, 3–8. Manobianco, J., S. Koch, V. M. Karyampudi, and A. J. Negri, 1994: The impact of assimilating satellite-derived precipitation rates on numerical simulations of the ERICA IOP 4 cyclone. Mon. Wea. Rev., 122, 341–365. Plant, R. S., G. C. Craig, and S. L. Gray, 2003: On a threefold classification of extratropical cyclogenesis. Quart. J. Roy. Meteor. Soc., 129, 2989–3012.

37 IMPLEMENTING AN OPERATIONAL CHAIN: THE FLORENCE LaMMA LABORATORY Alberto Ortolani1,3, Andrea Antonini2,3, Graziano Giuliani2,3, Samantha Melani2,3, Francesco Meneguzzo1,3, Gianni Messeri1,3, Andrea Orlandi1,2,3, and Massimiliano Pasqui1,3 1

Institute of BioMeteorology, National Council of Research (IBIMET-CNR), Florence, Italy Foundation for Applied Meteorology (FMA), Florence, Italy 3 Laboratory for Meteorology and Environmental Modelling (LaMMA), Florence, Italy 2

1 INTRODUCTION The Laboratory for Meteorology and Environmental Modelling (LaMMA, http://www.lamma.rete.toscana.it) was set up in 1997 due to an initiative of Regione Toscana (Tuscany Region Administration), which entrusted the Foundation for Applied Meteorology (FMA) to manage the laboratory in cooperation with the National Council for Research (CNR) and some Tuscany companies belonging to Finmeccanica Group, working in the field of space technology design and exploitation. From July 2002, LaMMA is managed by the Institute of BioMeteorology (IBIMET) of CNR. The main tasks of LaMMA are research, technological transfer, and service provision in support of the regional and national operational organizations, in the fields of meteorology and environmental monitoring. The research skills span from atmospheric and ocean modelling, air quality, and remote sensing of environmental parameters to geographic information systems for environmental management. The main research results are thus continuously integrated in the operational services of LaMMA, that acts as a regional meteorological service in Tuscany, with daily public forecasts for different TV and radio programs, as well as forecasts and nowcasts for supporting (even 24 h) Civil Protection decisions during severe weather events (storms, flash floods, snow, etc.). The Regional Atmospheric Modelling System (RAMS) (Pielke et al. 1992; MRC/*Aster 2000) is the atmospheric model used

471 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 471–482. © 2007 Springer.

472

A. Ortolani et al.

operationally at LaMMA from 1999 (Pasqui et al. 2000; Meneguzzo et al. 2001; Meneguzzo et al. 2004; Pasqui et al. 2002; Soderman et al. 2003). Hydrogeological disasters are among the sources of main concern (in terms of expense and health risk) in Tuscany, but generally in Italy and in several European and world countries. If their link to rain is widely apparent, precipitation itself (together with the hydrological mechanisms) remains largely a matter of study. The way rainfall happens, space-time distribution and intensity predictability, as well as methods for reliable homogeneous rainfall measurements are not completely understood and much efforts are needed for a satisfactory comprehension and prediction of these phenomena. In the framework of EURAINSAT a large part of these issues were addressed. The approach followed by LaMMA in the project activities was to focus on research with high and soon operational applicability, trying to maximize the potential benefits that the availability of quasi real-time rainfall fields (homogeneously produced for the whole Euro-Mediterranean area) could bring to the nowcasting and forecasting activities. For this purpose: • •

•

An automatic chain for real-time satellite rainfall estimation has been set up and a validation phase over the Tuscany area has been started and is still ongoing. In order to improve quantitative precipitation forecasts (QPFs) of RAMS, a method for diabatic assimilation of the convective part of rainfall fields in the first hours of the forecast simulation has been developed and tested, both with simulated and observed data. A method for soil moisture initialization in RAMS by means of antecedent observed precipitation has been designed and implemented with the aim of improving the description of the initial state of a simulation run.

All these methods have been designed in order to be integrable in a unique nowcasting/forecasting system, and in a way compatible with the operational constrains of the routine LaMMA activities. The following sections detail the work done and the main results.

2 THE OPERATIONAL CHAIN FOR RAINFALL FIELD ESTIMATION Precipitation estimates from satellites are a relevant component in environmental monitoring, from flash flood forecasting and landslips to assimilation in numerical weather prediction (NWP) models (Levizzani et al. 2002). In this frame, the LaMMA laboratory for either regional monitoring or hazardous prevention purposes has implemented an operational chain that produces real-time instantaneous half-hourly rainfall maps.

Implementing an Operational Chain

473

The procedure is based on a blended technique (Turk et al. 2000a, b) that dynamically correlates brightness temperatures as measured by geostationary sensors and instantaneous rain rates, as computed by microwave (MW) passive radiometer data (Ferraro and Marks 1995; Ferrraro 1997), by means of a statistical correlation (Crosson et al. 1996). The quantification of precipitation levels as implemented in the operational chain involves an automatic process that constantly colocates, in space and time, newly arriving MW and infrared (IR) geostationary data. In this frame, two automatic, independent processes for METEOSAT and SSM/I data have been developed. The procedures are scheduled to start running independently every 30 min, as shown in Fig. 1.

Figure 1. Operational chain flow chart.

The overall system has been implemented by means of a Workstation equipped with UNIX operating system and TeraScan Software (SeaSpace Corporation, Poway, CA) that allows the processing and visualization of satellite data. The Workstation performs the ingestion of both IR METEOSAT and MW SSM/I channels data. A PDUS acquires METEOSAT real-time high resolution (HR) data (every half hour). The data, supplied in digital format, also contain geolocation, rectification, and calibration parameters, as well as the temperatures for the various channels. The MW data, supplied in temperature data record (TDR) format, are downloaded by ftp connection from the SAA archive (http://www.saa.noaa.gov).

474

A. Ortolani et al.

The operational service provided by the laboratory consists of instanttaneous rain field estimation and temporal evolution (i.e., the animation of the last 6 h) of the precipitation events for the Euro-Mediterranean area, both updated every 30 min. The results can be viewed online through the LaMMA web site at http://www.lamma.rete.toscana.it/previ/ita/rainmeteosat.html. Imagery is geolocated trough a regular lat/long projection. Figure 2 shows an example for 15 November 2002, at 03:00 UTC, where an intense rainfall event occurred over the Ligurian Gulf.

Figure 2. Example of instantaneous rainfall map for 15 November 2002.

A validation phase with ground-based rain gauges measurements has been undertaken over the Tuscany area. The preliminary results show that the algorithm correctly (in terms of space-time phase) associates convective cloudiness to rainfall, but misinterprets non-raining stratiform cloudiness as light rain. Further developments will also include on the web site instantaneous rainfall maps and cumulated rainfall images over the last 1, 3, 6, 12, and 24 h for other geographical areas. Finally, the upgrade of the operational chain for the ingestion of the new MSG satellite data and other MW sources is under development.

Implementing an Operational Chain

475

3 DIABATIC PRECIPITATION ASSIMILATION IN RAMS The assimilation of high-resolution satellite rainfall estimates may be really effective in order to improve QPF. In particular, it can be exploited to minimize the spin-up problem of atmospheric models, which is related to the difficulty in providing appropriate divergence and moisture fields in the initial conditions. Such a difficulty causes inconsistencies in the representation of the latent heat release evolution, mainly in the first hours of model simulations. As a consequence the timing and location of precipitation events are often inaccurate, especially when the model spin-up phase clashes with the initial stages of precipitating phenomena. Several works have been devoted to the investigation of this problem (Carr and Baldwin 1991; Davidson and Puri 1992) and some other (Pereira Fo et al. 1999; Falkovitch et al. 2000) have pointed out how observed rainfall assimilation can potentially address this problem. In the last 15 years, various approaches for precipitation assimilation have been tried for global and large-scale models, as well as for mesoscale models. For a short review of publications on this topics see, for example, Orlandi et al. (2004). In mesoscale and limited area models the most promising results have been obtained by incrementing temperature and moisture throughout a pre-forecast period (latent heat nudging [LHN]), in order to approach precipitation observations (Jones and Macpherson 1997). It has been proven (Manobianco et al. 1994; Orlandi et al. 2004) that NWP models can retain useful information from precipitation data well beyond the assimilation period (sometimes up to 30 h). Sensitivity studies demonstrated that the specification of the space-time location of the rainfall event is more relevant than the information on precipitation intensity (Manobianco et al. 1994). Many implementations of the LHN are based on the inversion of the convective parameterization scheme. Several cumulus parameterization schemes have been developed, for large-scale and for mesoscale models, with various levels of complexity. Generally they are based on two main features: (1) resolvable-scale quantities are used to establish constraints on the amount of convection, and (2) a cloud model is used to estimate the vertical structure of the convective mass flux that satisfies the constraints. The outputs of the cloud model are re-ingested in the resolved dynamics of the model, as a feedback of the effects of parameterized cumulus convection. This feedback can be exploited for assimilating observed rainfall patterns. In particular the cloud model allows to redistribute along the column the latent heating and moisture derived from observed rainfall. The assimilation technique described here (Meneguzzo et al. 2002; Orlandi et al. 2004) is based on the inversion of the Kuo scheme (Kuo 1974;

A. Ortolani et al.

476

Molinari 1985; Molinari and Corsetti 1985) which is implemented in RAMS. The large-scale constraint of this scheme is the moisture convergence I at the base of the convective air column, which is computed from the modelresolved variables. It is split into two parts, by introducing the Kuo phenomenological parameter b, which embodies the microphysics of precipitation production. The fraction (1–b) of I is condensed and precipitated down. The convective precipitation rate is thus computed such as:

Pconv = (1 − b) I

(1)

The remaining fraction b of I is stored in the cloud, and acts to increase the moisture of the convective column. The feedback of cumulus convection is computed in terms of convective tendencies along the column for potential temperature and water vapour-mixing ratio:

L(1 − b) I ⎛ ∂θ ⎞ ⎜ ⎟ = Π ⎝ ∂t ⎠conv

Q1 ( z ) zct

∫ Q ( z ')dz ' 1

zg

,

⎛ ∂r ⎞ ⎜ ⎟ = bI ⎝ ∂t ⎠conv

Q2 ( z ) zct

(2)

∫ Q ( z ')dz ' 2

zg

where L is the latent heat of condensation for water, Π is the total Exner function, and b is computed, in RAMS, as a function of environmental wind shear (Fritsch and Chappell 1980). The vertical profiles of heating Q1(z) and moistening Q2(z) are computed through a cloud model (Molinari 1985; Tremback 1990), starting from the resolved variables. The convective tendencies are then inserted in the model equations, which are forward integrated in time at each time step. The scheme is activated with a fixed time cadence (default in RAMS is 20 min), to re-compute the convective rainfall and tendencies, only on those grid points whose air column results to be convectively unstable and with a sufficient moisture convergence. The inversion of the Kuo scheme is straightforward. The moisture convergence I is computed from the observed rainfall, by inverting Eq. (1). Convective tendencies are then computed by Eq. (2) and then inserted in the model equations. This realizes a feedback mechanism which allows to modify the atmospheric conditions consistently with the observed rainfall.

Implementing an Operational Chain

Cumulated rainfall [mm]

Cumulated rainfall [mm]

477

(

(

a)

Number of hours from 0:0 a.m. 1999/09/20

b)

Number of hours from 0:0 a.m. 1999/09/20

Figure 3. Comparison between rain gauges observations and results of assimilation and control runs. Averaged data over two hydrologic basins in Tuscany region: (a) Serchio and (b) middle Arno Valley.

The inverted Kuo scheme is activated, when there is a rainfall map available, on convectively unstable grid points with enough rainfall. The “observed convective rainfall”, Pconv, is evaluated (convective/stratiform partition) somehow roughly by subtracting the model-computed resolved rainfall from the observed rainfall. In the operational implementation the inverted scheme is activated during the first hours of model simulation (typically six), assimilating a satellite rainfall map each half an hour. Observed rainfall is weighted with respect to the model computed direct Kuo rainfall, by a gradually growing nudging function, so as to “gently” push the model towards the observed conditions. The assimilation is performed only in coarse grids, where convection is parameterized, but its effects are transmitted to high-resolution grids, where convection is explicit, through the nesting mechanism (Warner and Hsu 2000; Orlandi et al. 2001). Test and tuning experiments have been performed by assimilating synthetic convective rainfall produced by the model itself. These experiments allowed to tune some of the free parameters of the procedure, and demonstrated the permanence of positive effects of the assimilation up to 30 h. Comparisons with rain gauges measurements, when assimilating real satellite rainfall, confirm such relevant improvements. The pluviometric plot in Fig. 3 shows how the initial phase of the rainfall event is better described by the assimilating model run.

478

A. Ortolani et al.

4 INITIALIZATION OF RAMS WITH SOIL MOISTURE FROM ANTECEDENT PRECIPITATION FIELDS The soil-state initialization plays a major role in numerical weather modelling on a wide range of spatial temporal scales affecting forecast skills especially on surface atmospheric fields (Chen and Avissar 1994; Avissar and Schmidt 1998; Golaz et al. 2001; Pielke 2001; Meneguzzo et al. 2002). Furthermore, NWP models which include detailed schemes for the description of soil– vegetation–atmosphere interaction exhibit strong sensitivity to soil-state initialization. As a consequence a reasonable description of the initial state is crucial to improve forecasts reliability. Dealing with case studies, the general approach is to establish a reasonable choice of the initial soil state based on available data-sets, retrieved from satellite, weather stations or specific soil state bulletins. Another strategy is to derive the initial soil state from general circulation models, but such fields are generally not directly related to real observed precipitation fields and they have a coarse spatial resolution. The soil-state initialization is very important for the correct forecasting of a wide range of weather phenomena, but the availability of such information as first guess field is by no means trivial. The fifth version of RAMS provides a method to produce initial soil state computed from simulated atmospheric and observed precipitation fields. In other words it is possible to run the Land Ecosystem Atmosphere Feedback (LEAF) model (Walko et al. 2000) prescribing both the atmosphere state and the occurred rainfall, the latter being possibly the results of observations. The state of the atmosphere is provided by a previous atmospheric RAMS simulation. Clearly the real atmospheric state which produces the observed (and ingested) precipitation could exhibit important discrepancies with respect to the simulated atmospheric state (used as a forcing), leading to significant differences in the water exchange among soil, vegetation, and air. On the other hand, such soil first-guess field brings some benefits, firstly a more realistic evaluation of the water amount with respect to what is just simulated. Furthermore, it provides a better description of heterogeneity due to the hydrological model acting within LEAF, as well as a longer, so more accurate, reconstruction of the water cycle forcing with respect to a simple (even if somehow measured) initial estimation of the soil state. The soil initialization scheme is based on a modified version of RAMS named RAMS antecedent precipitation index (RAPI) (Pasqui et al. 2004). The RAPI model needs two different types of input: 1.

The precipitation field, as a distributed map of rainfall (from satellite estimates, radar, etc.) over the area of interest for the selected time period.

Implementing an Operational Chain

2.

479

The RAMS atmospheric fields, computed in a separate RAMS run on the same time period.

Figure 4. RAPI assimilation scheme.

Using such information, the RAPI model computes energy balance, as RAMS actually does, between atmosphere, prescribed from a previous RAMS run, and the provided rainfall fields. Using observed precipitation has the advantage of improving the computation of the water budget, at the soil level, both for heterogeneity and reliability. Several benefits of using RAPI are worth to be highlighted, both from a physical and an operational point of view. The observed precipitation, once projected on the target area, has the same topography and resolution as the model grids, so a basins budget could be more reliable than soil moisture interpolation coming from a coarser grid simulation (e.g., a GCM field). A pilot version of this assimilation techniques (Fig. 4) was set up at LaMMA using a three-nested grid RAMS configuration at 32, 8, 2 km horizontal resolution and 36 vertical levels, with a resolution ranging from 50 to 1100 m and 11 ground level down to –1.5 m with a stretched resolution. This 24-h daily simulation provides the atmospheric forcing with initial and boundary conditions every 6 h from NCEP/NCAR analysis fields. The hourly precipitation data-set, based on the satellite estimation algorithm described above, is collected over Europe and North Africa for the same 24-h period. Such data are stored on the RAMS standard lat/long format (same as SST, vegetation cover, soil textural classes) and represent the observed precipitation forcing data-set. Every day these two data-sets are ingested in RAPI to compute the initial soil state for the following day RAMS simulation. Note that simulations performed on DESMO, the LaMMA operational Linux Cluster, show the

480

A. Ortolani et al.

RAPI running about 30 times faster then RAMS. Thus, this approach does not affect critically the computational demand for the entire forecast system. Preliminary tests, performed on the surface temperature field, reveal promising benefits of this approach. In particular, a reduction of the spin-up errors during the first 6 h is found, in addition to a general improvement of the precision on the forecasted minimum temperatures (Pasqui 2004).

5 CONCLUSION The memory of well-known, sometimes very recent, floods and landslides in Tuscany and in other parts of Europe is still alive in the mind of the damaged people. Unforeseen critical events as well as great false alarms survive only in the memory of people in charge of technical responsibilities on the subject, but they are not matter of less concern due to their larger number. The work performed in the EURAINSAT context has demonstrated that also recent research results and technological means can be exploited in an operational context, and that they are able to improve real-time monitoring and prediction of critical precipitation events, in terms of precision and time ahead, opening new possibilities for bad-effect mitigation. We have addressed the problem of monitoring rainfall events by means of real-time precipitation field estimation based on merged operational satellite observations, and we have used such available rainfall fields to cope with the problem of the initial state reconstruction for prognostic models in order to improve forecast capabilities of precipitation as well as other atmospheric quantities. For this purpose the model RAMS was modified both to allow forcing convective profiles to be compliant to rainfall observations for an initial simulation period, and to compute a starting soil moisture field according to previous precipitation. In other words we have designed an integrated nowcasting/forecasting system for rainfall-driven events. The validation work is still in progress, and for the moment it is performed only on single pieces of the whole rainfall nowcasting/forecasting system. System tests and integration are going on, and all relevant results will be continuously published on LaMMA web pages. In any case the tests performed up to now have demonstrated that we are on the right way, but at the same time that a large room for improvements still exists, especially on the side of precipitation estimation, and consequently on the derived applications. New satellite missions that just started or are near to be launched could be definitely part of the solutions for the open problems.

6 REFERENCES Avissar, R. and T. Schmidt, 1998: An evaluation of the scale at which ground–surface heat flux patchiness affects the convective boundary layer using large-eddy simulations. J. Atmos Sci., 55, 2666–2689.

Implementing an Operational Chain

481

Carr, F. H. and M. Baldwin, 1991: Incorporation of observed precipitation estimates during initialisation of synoptic and mesoscale storms. 1st Int. Symp. on Winter Storms, New Orleans, LA, Amer. Meteor. Soc., 71–75. Chen, F. and R. Avissar, 1994: Impact of land–surface moisture variability on local shallow convective cumulus and precipitation in large-scale models. J. Appl. Meteor., 33, 1382– 1401. Crosson, W. L., C. E. Duchon, R. Raghavan, and S. J. Goodman, 1996: Assessment of rainfall estimates using a standard Z–R relationship and the probability matching method applied to composite radar data in central Florida. J. Appl. Meteor., 35, 1203–1219. Davidson, N. E. and K. Puri, 1992: Tropical prediction using dynamical nudging, satellite defined convective heat sources, and a cyclone bogus. Mon. Wea. Rev., 120, 2501–2522. Falkovich, A., E. Kalnay, S. Lord, and M. B. Mathur, 2000: A new method of observed rainfall assimilation in forecast models. J. Appl. Meteor., 39, 1282–1298. Ferraro, R. R., 1997: Special sensor microwave imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 (D14), 16715–16735. Ferraro, R. R. and G. F. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12, 755– 770. Fritsch, J. M. and C. F. Chappell, 1980: Numerical prediction of convectively driven mesoscale pressure systems. Part I: Convective parameterization. J. Atmos. Sci., 37, 1722–1733. Golaz, J.-C., H. Jiang, and W. R. Cotton, 2001: A large-eddy simulation study of cumulus clouds over land and sensitivity to soil moisture. Atmos. Res., 59–60, 373–392. Kuo, H. L., 1974: Further studies of the influence of cumulus convection on large-scale flow. J. Atmos. Sci., 31, 1232–1240. Jones, C. D. and B. Macpherson, 1997: A latent heat nudging scheme for the assimilation of precipitation data into an operational mesoscale model. Meteorol. Appl., 4, 269–277. Levizzani, V., R. Amorati, and F. Meneguzzo, 2002: A review of satellite-based rainfall estimation methods. European Commission Project MUSIC Report (EVK1-CT-200000058), 66 pp. Manobianco, J., S. Koch, V. M. Karyampudi, and A. J. 1994: The impact of satellite-derived precipitation rates on numerical simulations of the ERICA IOP 4 cyclone. Mon. Wea. Rev., 124, 341–365. Meneguzzo, F., G. Menduni, G. Maracchi, G. Zipoli, B. Gozzini, D. Grifoni, G. Messeri, M. Pasqui, M. Rossi, and C. J. Tremback, 2001: Explicit forecasting of precipitation: sensitivity of model RAMS to surface features, microphysics, convection, resolution. In: Proc. 3rd Plinius Conf. on Mediterranean Storms. R. Deidda, A. Mugnai, F. Siccardi, eds., GNDCI Publ. N.2560, ISBN 88-8080-031-0, 79–84. Meneguzzo, F., V. Levizzani, A. Orlandi, A. Ortolani, M. Pasqui, F. Torricella, and B. Gozzini, 2002: Resolution and data assimilation issues in the operational numerical forecast of basin-scale rain storms. Proc. 4th EGS Plinius Conf. on Mediterranean Storms, October 2002 (a), Mallorca. Meneguzzo, F., M. Pasqui, G. Menduni, G. Messeri, B. Gozzini, D. Grifoni, M. Rossi, and G. Maracchi, 2004: Sensitivity of meteorological high-resolution numerical simulations of the biggest floods occurred over the arno river basin, Italy, in the 20th century. J. Hydrol., 288, 37–56. Molinari, J., 1985: A general form of Kuo’s cumulus parameterization. Mon. Wea. Rev., 113, 1411–1416. Molinari, J. and T. Corsetti 1985: Incorporation of cloud-scale and mesoscale down-drafts into a cumulus parameterisation: Results of one- and three-dimensional integrations. Mon. Wea. Rev., 113, 485–501.

482

A. Ortolani et al.

Orlandi, A., F. Meneguzzo, G. Messeri, A. Ortolani, M. Pasqui, M. Rossi, and A. Terzo 2001: Satellite rainfall assimilation to improve the quantitative precipitation forecasting, Proc. EUMETSAT Conference, Antalia, 453–460. Orlandi, A., A. Ortolani, F. Meneguzzo, V. Levizzani, F. Torricella, and F. J. Turk, 2004: Rainfall assimilation in RAMS by means of the Kuo convective parameterisation inversion: method and preliminary results. J. Hydrol., 288, 20–35. Pasqui, M. et al., 2000: Performances of the operational RAMS in a Mediterranean region as regards to quantitative precipitation forecasts. Sensitivity of precipitation and wind forecasts to the representation of the land cover. Proc. 4th RAMS Users Workshop, Cook College, Rutgers University, 22–24 May 2000, New Jersey. Pasqui, M. et al., 2002: Historical severe floods prediction with model RAMS over central Italy. Proc. 5th RAMS Users Workshop, Santorini, Greece. Pasqui, M., C. J. Tremback, F. Meneguzzo, G. Giuliani, and B. Gozzini, 2004: A soil moisture initialization method, based on antecedent precipitation approach, for regional atmospheric modeling system: a sensitivity study on precipitation and temperature. 18th Conf. on Hydrology, AMS, Seattle. Pereira Fo, A. J., K. C. Crawford, and D. J. Stensrud, 1999: Mesoscale precipitation fields. Part II: Hydrometeorologic modelling, J. Appl. Meteor., 38, 102–125. Pielke, R. A. and Coauthors, 1992: A comprehensive meteorological modelling system – RAMS. Meteor. Atmos. Phys., 49, 69–91. Pielke Sr., R. A., 2001: Influence of the spatial distribution of vegetation and soils on the prediction of cumulus convective rainfall. Rev. Geophys., 39, 151–177. Soderman, D., F. Meneguzzo, B. Gozzini, D. Grifoni, G. Messeri, M. Rossi, S. Montagnani, M. Pasqui, A. Orlandi, A. Ortolani, E. Todini, G. Menduni, and V. Levizzani, 2003: Very high resolution precipitation forecasting on low cost high performance computer systems in support of hydrological modeling. Proc. 17th Conf. on Hydrology, AMS, Long Beach, 9–13 February, CD-ROM, ISBN 1-878220-63-2. Tremback, C. J., 1990: Numerical simulations of mesoscale convective complex: model development and numerical results. Ph.D. dissertation, Dept. of Atmospheric Science, paper 465, Colorado state university, Forth Collins, CO. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000a: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, eds., VSP Int. Sci. Publisher, Utrecht, The Netherlands, 353–363. Turk, F. J., J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000b: Combining SSM/I, TRMM and infrared geostationary satellite data in a near-realtime fashion for rapid precipitation updates: advantages and limitations. Proc. 2000 EUMETSAT Meteorological Satellite Data Users’ Conf., Bologna, 29 May–2 June, 452– 459. Walko, R. L., L. E. Band, J. Baron, T. G. F. Kittel, R. Lammers, T. J. Lee, D. Ojima, R. A. Pielke, C. Taylor, C. Tague, C. J. Tremback, and P. L. Vidale, 2000: Coupled atmosphere-biophysics-hydrology models for environmental modeling. J. Appl. Meteor., 39, 931–944. Warner, T. T. and H.-M. Hsu, 2000: Nested-model simulation of moist convection: the impact of coarse-grid parameterised convection on fine-grid resolved convection. Mon. Wea. Rev., 128, 2211–2231.

Section 7 Applications to Monitoring Weather Events

38 SATELLITE PRECIPITATION ALGORITHMS FOR EXTREME PRECIPITATION EVENTS Roderick A. Scofield† and Robert J. Kuligowski National Environmental Satellite, Data, and Information Service Camp Springs, MD, USA

1 INTRODUCTION Floods and flash floods take a heavy toll each year in terms of both lives and property. According to the 2001 Disaster Report by the International Red Cross and Red Crescent Societies, floods accounted for over two-thirds of the 211 million people affected worldwide on average by natural disasters each year during the 1990s, and also for roughly 15% of the nearly 666,000 deaths from natural disasters during this period. Many of these events are triggered by extreme precipitation, often in conjunction with other factors. However, timely and reliable information on past, current, and future precipitation can be very difficult to obtain, especially in those portions of the world where resources are not available to build and support a comprehensive precipitation observing network. The three primary sources of precipitation information are rain gauges, radar, and satellite. Rain gauges have the clear advantage of directly measuring precipitation rather than to deriving it from a remotely sensed quantity. However, even relatively dense rain gauge networks are unable to depict the intensity and spatial extent of heavy precipitation (Smith et al. 1994, 1996). Furthermore, the rain gauge networks used in these studies are much more dense than those in most parts of the world. And although rain gauges are relatively low-cost on a unit basis, the expense required to install a rain gauge network suitable for extreme precipitation events is beyond the available resources for most nations. Radar offers widespread spatial coverage at high spatial and temporal resolution. However, there are difficulties both in obtaining accurate measures †

Louisville, 3 December 1942 – Pomfret, 25 February 2006

485 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 485–495. © United States Government 2007.

486

R. A. Scofield and R. J. Kuligowski

of reflectivity and in converting these reflectivity measurements into an accurate representation of the precipitation field at ground level. The former is the result of various phenomena including anomalous propagation, beam block, and beam overshoot, and is especially problematic in regions of high topographic relief where the range of useful radar data can be severely limited (e.g., Westrick et al. 1999; Young et al. 1999). Corrections have been developed and operationally implemented in an effort to mitigate these effects, such as hybrid scan strategies (Fulton et al. 1998) and vertical profile corrections (Seo et al. 2000) to improve the quality of the reflectivity data, and multiple reflectivity-rain rate (Z-R) relationships and correction of radar precipitation estimates using rain gauges (Fulton et al. 1998; Seo and Breidenbach 2002) to improve the resulting rain rate estimates. However, even beyond these issues lies the high monetary cost of building and maintaining a suitable network of radars. Precipitation estimates from sensors on geostationary satellite platforms offer an excellent complement to existing and planned rain gauge and radar networks. These estimates are available at high spatial resolution (3–5 km) and high temporal resolution (up to 15 min) for the whole globe between 60º N and 60º S latitude, and can compensate for the coverage shortcomings of rain gauge and radar data due to terrain or cost. However, the relationship between satellite-sensed radiances and rainfall rates at ground level is less robust than that between radar reflectivities and rainfall rates; consequently, these estimates should be viewed as a complement to other available sources of precipitation data. This paper describes real-time satellite precipitation algorithms for extreme precipitation events that are being developed at the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite, Data, and Information Service (NESDIS). During the last 25 years, operational satellite precipitation estimates (SPE) based on data from Geostationary Operational Environmental Satellites (GOES) have evolved from a combination of manual effort and computer algorithms to a succession of fully automated algorithms. Efforts are now underway to further improve SPE by incorporating information from other sources such as microwave radiometers, and to enhance the impact of SPE by incorporating it into hydrologic and numerical weather prediction models and by producing 1–3 h SPE-based nowcasts. Section 2 describes this progression and the associated algorithms in more detail. A case study illustrating the performance of these algorithms is given in Section 3, followed by a discussion of future directions in Section 4.

Satellite Precipitation Algorithms for Extreme Precipitation Events

487

2 REAL-TIME GOES-BASED SPE AT NESDIS 2.1 Interactive Flash Flood Analyzer (IFFA) Research into using satellite-based IR measurements to estimate precipitation began in the late 1960s (Lethbridge 1967), but it was not until the advent of operational geostationary satellites in the mid-1970s and the development of techniques for operational use that satellite-based estimates of precipitation because suitable for real-time detection of extreme precipitation events. The first of these operational techniques was the Interactive Flash Flood Analyzer (IFFA; Scofield and Oliver 1977; Scofield 1987), which has been used operationally at the NOAA/NESDIS Satellite Services Division (SSD) Satellite Analysis Branch (SAB) since the 1970s (Borneman 1988). The IFFA was originally designed for intense Mesoscale Convective Systems (MCSs) and later extended to other types of extreme precipitation events. The basis of the IFFA is the presumed relationship between cloud-top brightness temperature and rainfall rate – colder cloud tops imply stronger convective updrafts and hence higher rainfall rates than warmer cloud tops. This relationship holds well for the raining cores in convective systems, but is less valid for stratiform areas of precipitation (such as those that trail mature MCSs) and is not valid at all for cirrus anvils (which have cold cloud tops but do not produce precipitation). SAB forecasters apply the IFFA technique by manually identifying the convectively active portions of mesoscale systems, which is done using not only individual images but also comparison of consecutive images to diagnose growth or decay of convective clouds. Once this is done, rainfall rates based on cloud-top temperature are determined, and adjustments are made to account for overshooting cloud tops, cloud mergers, available moisture, low-level inflow, and the speed of the convective system. Since the satellite cloud-top temperatures do not always reflect the extremely strong updrafts and heavy rainfall that can occur before the clouds reach maturity, a rain burst factor is used to make appropriate adjustments early in the life cycle of an MCS. Finally, since certain thermodynamic profiles will support strong updrafts but have a convective equilibrium level that precludes extremely low cloud-top temperatures, the convective equilibrium level temperature is used to enhance precipitation in such instances. For additional details, the reader is referred to Scofield and Oliver (1977) and Scofield (1987). In addition to estimating past and current precipitation, SPE information is also extrapolated into the future to produce 3-h precipitation nowcasts or outlooks (Spayd and Scofield 1984) that take into account the growth, decay, movement, and propagation of individual convective systems (Shi and

488

R. A. Scofield and R. J. Kuligowski

Scofield 1987; Juying and Scofield 1989; Corfidi 2003). SAB SPE and outlooks are sent out to NOAA National Weather Service (NWS) forecasters via the Advanced Weather Interactive Processing System (AWIPS); graphics of the estimates are also available on the SSD home page (http://www.ssd. noaa.gov/PS/PCPN/index.html).

2.2 Auto-Estimator (AE) and Hydro-Estimator (HE) A significant limitation of the IFFA approach is its highly interactive nature. The required labor limits the coverage of estimates to relatively small regions – especially problematic if multiple significant precipitation events are occurring simultaneously. To improve both spatial/temporal coverage and timeliness, NESDIS developed an automated SPE algorithm for highintensity rainfall called the Auto-Estimator (AE). The original AE, developed by Vicente et al. (1998), computes rain rates from 10.7-µm brightness temperatures (hereafter referred to as T10.7) based on a relationship derived from more than 6,000 collocated radar and satellite pixels. Areas of non-raining cold cloud are identified using the spatial gradients of T10.7 and on changes in T10.7 from the previous image. Amounts are then adjusted using a multiplicative moisture adjustment consisting of precipitable water (PW) in inches multiplied by relative humidity (RH) as a decimal value, both from numerical weather model data. During 1998 and 1999, a number of enhancements were made, including the use of 15-min WSR-88D reflectivity data to screen out non-raining cold cloud, an adaptation of the IFFA equilibrium-level temperature adjustment (using numerical model data), and adjustments for parallax and orographic enhancement of precipitation. Many of these enhancements are described in Scofield (2001), Vicente et al. (2002), and Scofield and Kuligowski (2003). However, the AE is highly dependent on radar data to correctly identify non-raining cold cloud pixels because the schemes for identifying them in the AE often incorrectly classify cirrus as raining cloud, resulting in significant overestimation of the spatial extent of heavy precipitation. Since one of the advertised strengths of satellite QPE is its usefulness in regions where radar and/or rain gauge coverage are unavailable, another version of the AE called the Hydro-Estimator (HE) was developed to address this and other issues via three significant new features: • Raining pixels are defined as those with T10.7 below the average value for cloudy pixels in the region surrounding the pixel of interest. This approach substantially reduced the exaggeration of rain area compared to the AE, which in turn eliminated the need for radar as a rain/no rain discriminator. • The rain rate curve is adjusted according to the difference between the pixel T10.7 and the average T10.7 of the nearby cloudy pixels, with the

Satellite Precipitation Algorithms for Extreme Precipitation Events

489

highest rain rates assigned to pixels that are coldest relative to their surroundings. • The components of the PW*RH adjustment have been separated, with the PW used to adjust the rain rate curve according to moisture availability and the RH used to derive an amount to be “evaporated” from the rain rate. These adjustments have improved the handling of stratiform events with embedded convection, and also of wintertime precipitation, which is typically associated with low PW values. The HE has been a very useful source of information for SAB forecasters, allowing them to monitor a greater number of heavy rainfall systems and to disseminate SPEs to NWS Forecast Offices in a more timely fashion than they could when relying solely on the IFFA. However, SAB forecasters have cautioned that the HE does have limitations that still make production of IFFA estimates necessary on many occasions. These include a tendency to overestimate the area and magnitude of heavy precipitation for cold-topped (below –58ºC) systems and to underestimate the heavy rain that can fall from warm-topped (above –58ºC) systems. Also, in regions of strong wind shear, there can be differences in the location of the cloud tops and the resulting rainfall. Finally, the aforementioned convective rain burst factor has not yet been implemented in the HE, resulting in underestimates of rain rates during the early stages of storm development. In spite of these limitations, the HE has been considered sufficiently robust to replace the AE as SAB’s operational automated algorithm and to be disseminated to NWS field offices via AWIPS. As of this writing 1-h estimates of precipitation for the continental USA and surrounding regions are updated hourly. In addition, the HE is run worldwide according to the availability of IR imagery (every 15 min over the continental USA) to produce real-time instantaneous rainfall rates and accumulations over 1, 3, 6, and 24 h.

2.3 GOES Multi-Spectral Rainfall Algorithm (GMSRA) Although much effort into SPE has focused on using a single channel (usually around 10.7 µm), the utility of other channels for precipitation applications has also been investigated by numerous authors. Some of this work has been implemented by Ba and Gruber (2001) into a real-time algorithm called the GOES Multi-Spectral Rainfall Algorithm (GMSRA) that is run every 15 min over the continental USA. The GMSRA uses the five GOES Imager channels as follows: • A threshold visible albedo value below 0.4 screens out thin cirrus (Rosenfeld and Gutman 1994). • Negative values of (T10.7–T6.9) distinguish overshooting tops from anvil cirrus (Tjemkes et al. 1993).

490

R. A. Scofield and R. J. Kuligowski

• During the daytime 3.9-µm reflectance derived from 3.9-, 10.7-, and 12.0-µm radiance data during the daylight hours is related to cloud particle size, and clouds with large particles (effective radius exceeding 15 µm) are considered to be raining even for relatively warm cloud-top brightness temperatures (Rosenfeld and Gutman 1994; Rosenfeld and Lensky 1998). • During the nighttime, this is replaced with a screen that identifies as nonraining those pixels where T6.9–T10.7 and T10.7–T12.0 both exceed threshold values. • Increases in T10.7 from the previous image identify inactive (non-raining) clouds. In addition, values of T10.7–T12.0 exceeding 1 K (Inoue 1987) indicate thin cirrus, but this function was turned off when the GOES-12 satellite (which replaced the 12.0-µm channel with a 13.3-µm CO2 absorption band) was activated. For those clouds classified as precipitating, both the probability of precipitation and the conditional rain rate are computed from T10.7 using different calibrations for different regions. An adjustment for sub-cloud evaporation similar to that used in the AE (PW*RH) is also made. Ba and Gruber (2001) contains additional details the algorithm. The original rain rate calibration for the GMSRA was based on data from a 17-day calibration period; however, a version of the GMSRA has recently been implemented that uses rain gauge-corrected radar to produce updated rain rate curves on a near-real-time basis (Ba et al. 2003). This version of the algorithm was implemented for real-time testing, in parallel with the original, beginning in early November 2003.

3 CASE STUDY AND INTERCOMPARISON Regular validation of SPE at NESDIS began in the spring of 2001 (Kuligowski et al. 2001) and is now automated and posted to the Web (http://orbit-net.nesdis.noaa.gov/arad/ht/ff/validation/validation.html). A case study is presented here in which heavy rains fell in the Tennessee River Valley during 5–7 May 2003, with accumulations exceeding 125 mm covering a significant area (Fig. 1). These rains triggered flash floods and river floods that lasted for over a week, producing US$17 million damage and 3 deaths, and displacing approximately 2,000 people. Flooding in many locations was the worst in 20–40 years. A comparison of 48-h precipitation totals from the AE, HE, and (fixedcalibration) GMSRA to the Stage IV radar/rain gauge product (Fulton et al. 1998) over this region is presented in Fig. 1. Note that, in this particular case, the AE and GMSRA significantly overestimated the spatial coverage of the heaviest precipitation – the result of the aforementioned tendency of

Satellite Precipitation Algorithms for Extreme Precipitation Events

491

satellite-based precipitation algorithms to incorrectly identify cirrus as precipitating cloud. The HE captures best the magnitude and spatial extent of the heaviest precipitation, though the HE maxima are located farther to the east than the Stage IV maxima. Table 1 contains a statistical summary of the performance of these algorithms, both against the Stage IV fields (for hourly amounts) and against rain gauges (for daily amounts) as a reliability check of the Stage IV data. As indicated by both Table 1 and Fig. 1, both the AE and the GMSRA significantly overestimated the amount of precipitation for this particular case, while the HE bias was closer to unity. (Note that the bias values are different because the hourly statistics are computed against the gridded Stage IV data while the daily statistics are computed against available rain gauges.) The AE and HE exhibit similar correlations with the validation data sets, while the GMSRA is less accurate for this case. However, the previously described improvements recently implemented in the GMSRA should lead to improvements in performance.

Figure 1. Comparison total accumulated precipitation for three satellite QPE algorithms and the Stage IV data set for the 48 h ending 1200 UTC 7 May 2003. A contour at 125 mm is provided for reference.

R. A. Scofield and R. J. Kuligowski

492

It is also clear from both the figure and the table that significant improvements are still needed in the quality of SPE. However, for regions without radar data or a dense rain gauge network, these data would prove to be quite valuable in alerting forecasters to the potential for significant flooding from heavy rainfall. Table 1. Performance statistics for satellite precipitation algorithms versus Stage IV (hourly) and versus rain gauges (daily) for the 48 h ending 1200 UTC 7 May 2003: bias ratio (“Bias”) and Pearson correlation coefficient (“Corr.”). Algorithm A-E H-E GMSRA

Hourly Bias 2.72 1.17 2.70

Corr. 0.46 0.42 0.32

Daily Bias 1.84 0.97 1.54

Corr. 0.51 0.52 0.45

4 SUMMARY AND OUTLOOK This paper has presented an overview of satellite QPE algorithms for extreme events that are produced routinely at NESDIS using IR and/or visible data from geostationary-based instrument platforms. As demonstrated by the case study in Section 3, SPE is a useful companion to radar and rain gauges; however, much work remains to be done to improve the accuracy of SPE and produce a product that is suitable for direct incorporation into multisensor precipitation analyses, hydrologic models, and numerical models without the need for manual corrections. Future research and applications in operational SPE will most likely focus on the following six areas: Improving the calibration of SPE by obtaining the best possible validation data and accounting for the scale differences between the validation data and the satellite estimates (e.g., comparing point rain gauge measurements with spatially averaged SPE). Incorporating an increasing volume of data from geostationary satellites into SPE. Not only will spatial resolution continue to improve, but the next generation of geostationary imagers will have additional channels, some of which have already been demonstrated to be useful for retrieving cloud characteristics pertinent to precipitation (e.g., Ackerman et al. 1990; Baum et al. 2000). The availability of hyperspectral data will further enhance this capability (e.g., Chung et al. 2000). Improved understanding of the physical relationship between precipitation and the signals observed in the visible and IR wavelengths will be needed to make optimal use of these new data. Blending of data from geostationary sensors with that from polar-orbiting microwave sensors. Microwave-based estimates of precipitation are considered

Satellite Precipitation Algorithms for Extreme Precipitation Events

493

to be more robust than IR-based estimates because microwave radiances are affected by the vertical profile of water and ice throughout the cloud rather than by only the cloud-top characteristics (see Section 3 for details). But since microwave data are not presently available from geostationary platforms (though work in this area continues to move forward) and thus are available relatively infrequently over a given loation, blending with IR data affords the best oppurtunity to take advantage of the strengths of both data sets. These issues were addressed in detail in Section 4, but an additional algorithm of interest for operational use is the Self-Calibrating Multivariate Precipitation Retrieval (SCaMPR) algorithm (Kuligowski 2003), which operates with latency comparable to IR/visible-only SPE. Blending of SPE with information from other sensors to produce an optimal precipitation data set. Numerous blending methods have already been developed for longer time scales (see Section 4), and methodologies for shorter time scales are also under development, including QPE-SUMS (Gourley et al. 2002) and the Multisensor Precipitation Estimate (MPE) which is planned for operational application by the NWS (Kondragunta and Seo 2004). The use of SPE in prediction applications. Improvements in SPE and useful expressions of its error characteristics will lead to increased use for initializing numerical weather models, especially over oceans and sparely populated regions where other sources of precipitation information are unavailable. SPE will also play an increasing role in hydrologic forecasting for regions where other sources of precipitation data are likewise difficult to obtain or inadequate. Furthermore, given the relatively poor performance of numerical weather prediction models at short lead times (e.g., Doswell 1986), direct use of extrapolated SPE as a nowcasting tool is being investigated, and experimental 1–3 h nowcasts for the continental USA are available in real time on the NOAA/NESDIS Flash Flood Home Page (http://orbit35i.nesdis.noaa.gov/arad/ht/ff). Finally, to realize the maximum benefit from these techniques, SPE will become more and more global in their focus. Many of the experimental algorithms described in this book are already being applied globally, including the HE described in this section. The operational implementation of SPE for real-time use by weather services outside the USA is also proceeding in Mexico (Fortune and Teran 2004) and in Central America (Alfaro 2003).

5 REFERENCES Ackerman, S. A., W. L. Smith, J. D. Spinhirne, and H. E. Revercomb, 1990: The 27–28 October 1986 FIRE IFO cirrus case study: Spectral properties of cirrus clouds in the 8–12 m window. Mon. Wea. Rev., 118, 2377–2388.

494

R. A. Scofield and R. J. Kuligowski

Alfaro, R., 2003: Validation of GOES precipitation estimates over Central America. Colorado State University, Cooperative Institute for Research in the Atmosphere Technical Report ISSN No. 0737-5352-58, 24 pp. Ba, M. B. and A. Gruber, 2001: GOES multi-spectral rainfall algorithm (GMSRA). J. Appl. Meteor., 40, 1500–1514. Ba, M. B., A. Gruber, and M.-J. Jeong, 2003: Frequency distribution of rain rate with cloud top brightness temperature and near-real time calibration of GOES Multispectral Rainfall Algorithm. Prepr. 12th Conf. Satellite Meteor. and Ocean., Long Beach, CA, Amer. Meteor. Soc., CD-ROM, P4.17. Baum, B. A., P. F. Soulen, K. I. Strabala, M. D. King, S. A. Ackerman, W. P. Menzel, and P. Yang, 2000: Remote sensing of cloud properties using MODIS airborne simulator imagery during SUCCESS 2. Cloud thermodynamic phase. J. Geophys. Res., 105(D9), 11781–11792. Borneman, R., 1988: Satellite rainfall estimating program of the NOAA/NESDIS Satellite Analysis Branch. Natl. Wea. Dig., 13, 7–15. Chung, S., S. Ackerman, P. F. Van Delst, and W. P. Menzel, 2000: Model calculations and interferometer measurements of ice-cloud characteristics. J. Appl. Meteor., 39, 634–644. Corfidi, S. F., 2003: Cold pools and MCS propagation: forecasting the motion of downwind developing MCSs. Wea. Forecasting, 18, 997–1017. Doswell, C. A., 1986: Short-range forecasting. Mesoscale Meteorology and Forecasting, P. S. Ray, Ed., Amer. Meteor. Soc., 689–719. Fortune, M. A. and A. R. Teran, 2004: Satellite-based rainfall estimates as a driver of river forecasts in Mexico. Prepr. 18th Conference on Hydrology, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, P3.6. Fulton, R. A., J. P. Breidenbach, D. J. Seo, and D. A. Miller, 1998: The WSR-88D rainfall algorithm. Wea. Forecasting, 13, 377–395. Gourley, J. J., R. A. Maddox, K. W. Howard, and D. W. Burgess, 2002: An exploratory multisensor technique for quantitative estimation of rainfall. J. Hydrometeor., 3, 166–180. Inoue, T., 1987: A cloud type classification with NOAA 7 split-window measurements. J. Geophys. Res., 92, 3991–4000. Juying, X. and R. A. Scofield, 1989: Satellite-derived rainfall estimates and propagation characteristics associated with mesoscale convective systems (MCS). NOAA Tech Memo. NESDIS 25, 49 pp. Kondragunta, C. and D.-J. Seo, 2004: Toward integration of satellite precipitation estimates into the multi-sensor precipitation estimation algorithm. Prepr. 18th Conference on Hydrology, Long Beach, CA, Amer. Meteor. Soc., CD-ROM, J1.7. Kuligowski, R. J., 2002: A self-calibrating GOES rainfall algorithm for short-term rainfall estimates. J. Hydrometeor., 3, 112–130. Kuligowski, R. J., S. Qiu, R. A. Scofield, and A. Gruber, 2001: The NESDIS QPE verification program. Prepr. 11th Conf. on Satellite Meteor. and Ocean., Amer. Meteor. Soc., Madison, WI, 383–384. Lethbridge, M., 1967: Precipitation probability and satellite radiation data. Mon. Wea. Rev., 95, 487–490. Rosenfeld, D. and G. Gutman, 1994: Retrieving microphysical properties near the tops of potential rain clouds by multi spectral analysis of AVHRR data. Atmos.Res., 34, 259–283. Rosenfeld, D. and I. Lensky, 1998: Satellite-based insights into precipitation formation processes in continental and maritime convective clouds. Bull. Amer. Meteor. Soc, 79, 2457–2476. Scofield, R. A., 1987: The NESDIS operational convective precipitation estimation technique. Mon. Wea. Rev. 115, 1773–1792. Scofield, R. A., 2001: Comments on “A quantitative assessment of the NESDIS AutoEstimator”. Wea. Forecasting, 16, 277–278.

Satellite Precipitation Algorithms for Extreme Precipitation Events

495

Scofield, R. A. and R. J. Kuligowski, 2003: Status and outlook of operational satellite precipitation algorithms for extreme precipitation events. Wea. Forecasting, 18, 1037– 1951. Scofield, R. A. and V. J. Oliver, 1977: A scheme for estimating convective rainfall from satellite imagery. NOAA Tech. Memo. NESS 86, 47 pp. Scofield, R. A., M. DeMaria, and R. Alfaro, 2001: Space-based rainfall capabilities in hurricanes offshore and inland. Prepr. Symposium on Precipitation Extremes: Prediction, Impacts, and Response, Albuquerque, NM, Amer. Meteor. Soc., 297–301. Scofield, R. A., R. J. Kuligowski, and C. Davenport, 2004: The use of the Hydro-Nowcaster for mesoscale convective systems and the Tropical Rainfall Nowcaster (TRaN) for landfalling tropical systems. Prepr. Symposium on Planning, Nowcasting, and Forecasting in the Urban Zone, Seattle, WA, Amer. Meteor. Soc., CD-ROM, 1.4. Seo, D. J. and J. P. Breidenbach, 2002: Real-time correction of spatially nonuniform bias in radar rainfall data using rain gauge measurements. J. Hydrometeor., 3, 93–111. Seo, D. J., J. P. Breidenbach, R. Fulton, and D. Miller, 2000: Real-time adjustment of rangedependent biases in WSR-88D rainfall estimates due to nonuniform vertical profile of reflectivity. J. Hydrometeor., 1, 222–240. Shi, J. and R. A. Scofield, 1987: Satellite observed mesoscale convective system (MCS) propagation characteristics and a 3–12 hour heavy precipitation forecast index. NOAA Tech. Memo., NESDIS 20, U.S. Dept. of Commerce, Washington, D.C., 43 pp. Smith, J. A., A. A. Bradley, and M. L. Baeck, 1994: The space-time structure of extreme storm rainfall in the southern Plains. J. Appl. Meteor., 33, 1402–1417. Smith, J. A., D.-J. Seo, M. L. Baeck, and M. D. Hudlow, 1996: An intercomparison study of NEXRAD precipitation estimates. Water Resour. Res., 32, 2035–2045. Spayd, L. E., Jr. and R. A. Scofield, 1984: An experimental satellite-derived heavy convective rainfall short range forecasting technique. Prepr. 10th Conf. Weather Forecasting and Analysis, Clearwater Beach, FL, Amer. Meteor. Soc., 400–408. Tjemkes, S. A., L. van de Berg, and J. Schmetz, 1997: Warm water vapor pixels over high clouds as observed to Meteosat. Beitr. Phys. Atmos., 70, 15–21. Vicente, G. A., R. A. Scofield, and W. P. Menzel, 1998: The operational GOES infrared rainfall estimation technique. Bull. Amer. Meteor. Soc., 79, 1883–1898. Vicente, G. A., J. C. Davenport, and R. A. Scofield, 2002: The role of orographic and parallax corrections on real time high resolution satellite rainfall rate distribution. Int. J. Remote Sens., 23, 221–230. Westrick, K. J., C. F. Mass, and B. A. Colle, 1999: The limitations of the WSR-88D radar network for quantitative precipitation measurement over the western United States. Bull. Amer. Meteor. Soc., 80, 2289–2298. Young, C. B., B. R. Nelson, A. A. Bradley, J. A. Smith, C. D. Peters-Lidard, A. Kruger, and M. L. Baeck, 1999: An evaluation of NEXRAD precipitation estimates in complex terrain. J. Geophys. Res., 104, 19691–19703.

39 APPLICATION OF A BLENDED MW-IR RAINFALL ALGORITHM TO THE MEDITERRANEAN Francesca Torricella1,Vincenzo Levizzani1, and F. Joseph Turk2 1

Institute of Atmospheric Sciences and Climate, ISAC-CNR, Bologna, Italy Naval Research Laboratory, NRL, Monterey, CA, USA

2

1 INTRODUCTION The network of meteorological and environmental satellites is the only practicable means of monitoring and gauging rainfall on a global scale. For this reason an adequate estimation of the accuracy of global operational rain products is crucial. The first step for estimating error characteristics is perhaps a kind of local analysis, validation campaign and limited comparisons with reference data sets, in general taken from ground based instruments and networks, deemed to represent “truth” data sets. Historically, ground based radar and rain gauges supplied the reference data for such comparisons. With the advent of space based radars such as the precipitation radar of the Tropical Rainfall Measurement Mission (TRMM) reference data sets are also available from space. Moreover, the comparison with products derived from concurrent sensors (e.g., TRMM mixed radar-microwave products) is is another tool to check the overall performance of newly developed algorithms. A parallel approach consists in developing an error model of the retrieval process and assessing the uncertainties of its parameters. The key here is that the product being validated be derived on a physical basis with empirically verifiable assumptions. To date, the largest efforts of the scientific community have been aimed to the assessment of the accuracy of mean (monthly, weekly, daily) or cumulated rain products over suitable study periods and using standard evaluation statistics (Adler et al. 2001). Nevertheless, the application of satellite derived analysis to the characterization of severe rain events, meteorological applications, and flood management requires that the global achievements of the validation exercise 497 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 497–507. © 2007 Springer.

498

F. Torricella et al.

be verified on a local scale. It is necessary to understand whether useful and reliable estimates of precipitation fields can be obtained in particular climatic and geographic conditions. The goal is to analyze the performances of the method in producing instantaneous rainfall maps. It is conceivable that global algorithms need local optimization if snapshots of the precipitation field are required for the devised applications, because the physics of the rain processes strongly depends on the immediate environment. A global blended infrared-passive microwave (IR-PMW) technique producing rain rate fields at the time/space resolution and coverage of geostationary (GEO) observations is applied to rain events over Mediterranean countries. Two cases are examined: a series of intense rainstorms that affected the Emilia-Romagna area (Northern Italy) in early August 2002, and the November 2001 Algeria flood. In spite of their unusual intensity, they were selected among a set of analogous cases for they are representative of several common characteristics that a rain algorithm should be capable of coping with when functioning in such particular environment.

2 THE HYBRID PMW-IR RAINFALL ESTIMATION METHOD The idea of using data from GEO satellites to produce rain rate (RR) maps over large areas of the globe was largely exploited using visible (VIS) and IR stand-alone observations or combining them with information from different sensors, especially PMW instruments on polar platforms. At present, only GEO measurements have the spatial resolution (a few km2), repetition time (15–30 min), and spatial coverage suitable to properly follow the rapid variations of precipitation fields. Moreover, the long history and the robust technology of GEO instruments prompts for the reanalysis of historical events and guarantees a timely and reliable release of calibrated data. VIS and IR measurements give only indirect information on the precipitation field being limited to the uppermost cloud layer near the top. Their uncertainties are thus relevant per se since the precipitating hydrometeors do not interact directly with the photons collected aloft by space borne instruments at these wavelengths. Several methods have been developed that “calibrate” for example IR brightness temperature (TB) data by using the more physically based rain estimates derived from PMW instruments. This kind of blended techniques is intrinsically constantly evolving due to the ever-expanding suite of PMW sensors and multispectral GEO imagers. Moreover they should mitigate the sampling error deriving from noncontinuous precipitation sampling due to the orbital characteristics of the satellite and the spatio-temporal structure of precipitation associated with diurnal, synoptic, seasonal, and interannual variability cycles.

Application of a Blended MW-IR Rainfall Algorithm

499

The blended technique adopted hereafter (Turk et al. 2000b) has been recently validated using rain-gauge data and analyses by the Korean Meteorological Agency and the Australian Bureau of Meteorology. Looking at mean or cumulated rainfall amounts, the correlation fairly increases and bias and root mean square error decreases as either the integration/averaging period is increased (from a minimum of 1 h up to 30 days) or the grid size for spatial averaging is coarsened (from 0.1 to 3°). The original operational set up of the software (global, automatic, real time, using a suite of PMW and IR observations) was adapted to the task of analyzing test case studies. In the Turk’s method, hereafter referred to as Naval Research Laboratory technique (NRLT), rain rates derived from PMW measurements are used to create global, geolocated RR-TB relationships that are renewed as soon as new collocated data are available from both GEO and PMW instruments. The PMW RR data can be derived, in principle, from any source, provided they are geolocated rain intensities in mm h–1, and that the files containing them report the useful information on orbit (date, start time, sensor, satellite, etc.). For the present work the adopted PMW estimates are mainly derived from Special Sensor Microwave/Imager (SSM/I) data. From the brightness temperatures measured in seven polarized channels from 19.2 to 85.5 GHz, rain rates are derived by means of the NOAA-NESDIS operational algorithm (Ferraro and Marks 1995; Ferraro 1997). The NESDIS algorithm derives rainrates at the A-scan resolution of the SSM/I (~25 km) by means of nonlinear relationships involving the instrument channels (vertical and horizontal polarization) that have been calibrated using large sets of ground reference data collected by radar networks in different countries. The physical basis of such relationships are the scattering of MW radiation due to large ice particles above the freezing level occurring in precipitating clouds, and the emission from liquid water. This latter phenomenon can be sensed only above oceanic surfaces, due to high and largely unknown emissivity of land surfaces in the MW spectral range. Relying on PMW measurements only (no need of large input database of physical properties) and on simple but well founded relationships, this algorithm is very robust and lends itself to global applications. In the NRLT, to the end of calibrating IR measurements, the globe (or the study area) is subdivided in equally spaced LATLON boxes (2.5° × 2.5°). For each box, space and time coincident IR and PMW measurements are reduced to the worse spatial resolution and then collected. The colocation process allows for time and space offsets (15 min and 10 km, respectively). To form a meaningful statistical ensemble the method can look at older PMW orbit-IR slot intersections, until a certain box coverage is reached (say 75%) and a minimum number of coincident observations is gathered for a 3° × 3° boxes region. By means of this set of RR and corresponding TB, the RR-TB relationships are derived by applying a probability matching method (Calheiros and Zawadzki 1987).

F. Torricella et al.

500

a)

b)

c)

d)

e)

f)

g)

h)

i)

j)

k)

l)

m)

n)

o)

p)

Figure 1. 6 August, 2002. Rain intensity maps in mm h -1 for the Emilia-Romagna storm case study. All times are UTC. a) Radar map at 0012; b) NRLT for the slot starting at 0000; c) Radar map at 0042; d) NRLT for the slot starting at 0030; e) NRLT for the slot starting at 0630; f) PMW NESDIS algorithm for the SSM/I orbit 08D (F13) starting at 0627; g) NRLT for the slot starting at 0700; h) Radar map at 0642; i) NRLT for the slot starting at 0800; j) PMW NESDIS algorithm for the SSM/I orbit 10D (F14) starting at 0828; k) NRLT for the slot starting at 0800; l) Radar map at 0842; m) NRLT for the slot starting at 0930; n) PMW NESDIS algorithm for the SSM/I orbit 12D (F15) starting at 0941; o) NRLT for the slot starting at 1000; p) Radar map at 0942. (see also color plate 17)

501

Application of a Blended MW-IR Rainfall Algorithm

b)

d)

f)

h)

j)

Figure 2. 10 November, 2001. Rain intensity maps in mm h -1 for the Algeria case study. Time is UTC. a) SSM/I orbit 23A (F14) starting at 1919; b) SSM/I orbit 24A (F15) starting at 1958; c) NRLT for the slot starting at 0000; d) NRLT for the slot starting at 0200; e) PR for the TRMM orbit 22741 (area overpass around 0032); f) PR for TRMM orbit 22742 (area overpass around 0210); g) NRLT for the slot starting at 0000 (calibrated with PR data); h) NRLT for the slot starting at 0200 (calibrated with PR); i) 2A12 TMI rainrates for the TRMM orbit 22741; j) 2A12 TMI rainrates for the TRMM orbit 22742.

502

F. Torricella et al.

3 AUGUST 2002 INTENSE RAINSTORMS OVER EMILIA-ROMAGNA The Po Valley in Northern Italy is surrounded on three sides by high mountains and is therefore characterized by relatively high humidity and light winds at lower atmospheric levels, rather favorable conditions for the formation of line storms and Mesoscale Convective Systems. In August 2002 a number of intense storms hit the Italian peninsula, and attained a relevant intensity over a large area from Tuscany through Emilia-Romagna all the way up to northeastern Italy. Hailfall damages were widely registered in the belt between France, Switzerland, southern Germany, Austria, Hungary, and the Caucasus region. Early in the month instability conditions were fostered by a cyclonic area insisting over France, with cold air fluxes over north-central Italy, with associated strong winds, heavy rainfall, and hailstorms. The monthly rainfall accumulations reached unusually high values, with a maximum positive anomaly of 100 mm in Ferrara. Rainstorms started on the 4th and, after a few hours of pause, they intensified again in the afternoon of the 5th one after the other with similar characteristics and hitting a limited geographical area. The storm cells originated west of the Alps and moved rapidly eastward crossing the entire Po Valley. The electric activity remained quite impressive during the entire duration. The first useful SSM/I overflight is the orbit at 1601 UTC on day 5. Later on, the zone was imaged 6 times by the SSM/I sensors in about 24 h, thanks to the availability of data from three satellites (F13, F14, and F15). The starting time of the orbits on day 5 were 1601, 1802 and 1916 UTC, and there was no overpass during the night. The first overpass on day 6 corresponds to the orbit starting at 0627 UTC and the area was then covered by the orbits 0828, 0941 and 1547 UTC. After 1600 UTC the storm system left the region and moving eastward toward Slovenia. The rain maps for the overpass on day 6 are shown in Fig. 1. On the overall, the PMW algorithm detects the storm cells and gauges the high precipitation intensities up to 35 mm h–1, the maximum allowed rain intensity in the NESDIS algorithm (panels f, j, n). Note that all the SSM/I orbits cover the western part of the area (the gray shad delimits the area where the method/instrument gives results). The first row in the figure collects the results from NRLT for two slots during the night (panels b and d). For comparison nearly simultaneous radar maps are shown (panels a and c). Radar data are taken from C-band dual polarisation Doppler weather radar of the Servizio Meteorologico Regionale (SMR) in S. Pietro Capofiume, (44.654° N, 11.624° E, 11 m a.s.l.) in the southeastern sector of the Po valley. For these two slots, the last PMW overpass that calibrated the relationships for the blended rain intensities was the one on day 5 starting at 1916

Application of a Blended MW-IR Rainfall Algorithm

503

UTC, i.e., it was about 3 h old. This can be perceived by observing the precipitation cell over the Italy-Slovenia border. The NRLT correctly followed the movement of this cell, but the precipitation field became unrealistically extended, somewhat uniform, and the peak intensity was too low. Nonetheless, the NRLT located the new cell in the west, with peak intensity, location and shape in good agreement with radar measurements. This rather good performance is confirmed by looking at the NRLT map (panel e) just before the next PMW calibration (panel f) in early morning of day 6. The location of the rain cells is still preserved more than 11 h after the last calibration, and the rain field matches closely the radar data (panel h). The agreement obviously improves in picture g, due to the intervening PMW calibration (panel f). The capability of NRLT in extending the rain field information outside the PMW spatial coverage is testified in the next row of Fig. 1. The easternmost cell detected by radar in panel l (over Istria) is revealed also by NRLT even if no direct PMW measurements was available for this area since many hours. The last row in Fig. 1 (panels m, n, o, and p) confirms the NRLT performances with the exception of the feature in the upper-right corner of panel o that appears to be structureless.

4 THE NOVEMBER 2001 ALGERIAN FLOOD In early November 2001, a widespread frontal system and upper air trough from northeast Scandinavia to southwest Spain led to an extreme precipitation event in Algiers, causing severe flooding and huge mudslides. More than 120 mm of rain fell in 12 h during the night between 9 and 10 November and more than 130 mm during the next 6 h on the mountains behind Algiers. The unusually large rain rates were fed by the cold maritime arctic air that picked up moisture crossing the warm Mediterranean waters and met maritime subtropical air. An intense orographic enhancement was caused by strong surface winds oriented towards the high mountains of the African coast (>2300 m a.s.l.). The sudden onset of precipitation, the orographic complexity of the terrain, the vicinity to the coast are all elements that can introduce large errors and bias in PMW rainfall retrieval algorithms. Due to the short duration of the event (about 20 h), the area was imaged only a few times by PMW instruments. In such unfavorable condition, rapid update techniques are a powerful instrument to follow the evolution of the otherwise poorly observed rain field. The scarcity of validation data suggested a strategy involving not only SSM/I data but also data from the instruments onboard the Tropical Rainfall Measuring Mission (TRMM). The SSM/I derived RR fields were used for the calibration of the statistical relationships within the NRLT while the TRMM Microwave Imager (TMI) 2A12 operational rain product was retained as a source of comparison/validation data. Moreover,

504

F. Torricella et al.

due to the unsatisfactory results of the NESDIS SSM/I rain algorithm, the NRLT was re-applied using measurements of the TRMM Precipitation Radar (PR) (2A25 operational product) as a calibration source for the blended product. The rain maps for the central hours of the night between 9 and 10 November are collectively shown in Fig. 2 and document the peak intensity of the event. Note the following features for interpreting the rain maps: (a) the gray shaded area represents the swath coverage of the instrument or product; (b) the rain intensity in mm h–1 are represented according to the color scale included in the figure, from 0 to 35 mm h–1; (c) the first column (panels a, c, e, g, and i) refers to 0030 UTC and the second (panels b, d, f, h, and j) to 0100 UTC; (d) the town of Algiers is marked by a red circle. The inspection of PMW rain maps (panels a and b) reveals that coastal precipitation is missed altogether. Starting 50 km offshore and coming ashore the NESDIS algorithm uses the land module, i.e., precipitation in coastal environment is treated as it were over land. This is the most difficult zone to treat because of the discontinuity in atmospheric conditions and of the mixed sea-land signal collected within the field of view. In order to try to explain the complete lack of precipitation signal along the coast, the NESDIS algorithm was modified to eliminate the ad hoc treatment of coastal environment: overseas pixels were processed by means of the “sea” algorithm and overland with the “land” algorithm, no matter how far from the coast. The result somewhat unexpected of this exercise is that the “sea” part of the PMW algorithm works fairly well and detects convective cells along the coast, although at reduced rain intensity. Nonetheless, spurious rain signatures appear all along the coast even in cloud free conditions. On the contrary, no relevant rain signatures appeared inshore. It is thus evident that the “land” part of the algorithm is completely unfit for the particular rain type and/or surface characteristics. By analyzing the terrain classification applied prior to the rain computation it appears that some small (precipitation?) area is misclassified as snow. The largest part of precipitation over the Mediterranean sea is derived by means of the scattering algorithm. The heaviest precipitation is detected over the Mediterranean off the coast of North Africa, and the values never exceeded 13 mm h–1. Because of the lack of SSM/I overpasses after 1958 UTC (the successive orbit is at 0504 UTC the day after) the relationships derived from the two orbits are used to derive the rainfall maps during the night. Heavy rain started falling after this overpass, so that the algorithm faced very critical data input conditions. The analysis of the NRLT results for the same time period (b and d) do not bring about substantial improvements with respect to PMW’s. Almost no precipitation is detected over the African continent, excluding light rain over the Algeria-Tunisia border. One of the most prominent merit of the hybrid method is that it can eliminate discontinuities and mitigate the problem of

Application of a Blended MW-IR Rainfall Algorithm

505

directly deriving precipitation over the coast from PMW, but in this case even this type of technique is of no avail in detecting precipitation over Algiers. The geolocated statistical RR-TB relationships for the nine boxes covering the area show that the heaviest precipitation appear to be associated with the Algiers box and the one immediately to the North, with a maximum intensity of about 9 mm h–1. The adjacent boxes show similar relationships, but with a reduced rain intensity. The zero-rain threshold is always < –40°C (–47°C for the box containing Algiers), and the range of temperatures associated to precipitation is no more than 10°C if one considers intensities >1 mm h–1. These relationships are obviously derived from rainy pixels over the sea. Since the NRLT is based in the end on the IR TB, METEOSAT IR images for the whole night were analyzed, considering a 1°.×.1° box centered over Algiers. The TBs remained constantly higher than the zero rain threshold, and this explains why no precipitation is detected. In practice the behavior of the NRLT in the particular case reveals that the characteristics of the clouds and precipitation fields were definitely different over land (warm rain) with respect to the convective cells embedded in the storm system, which developed over the sea very close to the coastline. This is confirmed by the PR observations (panels e and f) that show very intense precipitation cells off the Morocco-Algeria coast that appear to be delimitated by the coastline. Very likely the orography played a major role in the event, especially the relief south of Algiers, giving rise to a precipitating system that, although embedded in a larger field, was neither detected by PMW (due to the low scattering) nor by the NRLT (due to the high TB). The NRLT was further calibrated using PR rain measurements. The results (panels g and h), however, look quite disappointing. The rain pattern is not much different with respect to panels c and d, but the rain intensity values rose to much higher values, according to the radar measurements. Indeed the PR data, due to very narrow swath of the instrument, drastically change only the relationships of the westernmost boxes, altering the rain intensity associated with each IR temperature, but not the overall shape of the relationship. Unfortunately, due to the low number of overpasses and the characteristics of the TRMM orbit, the PR did not take measurements over the area of the disaster. For comparison with the NRLT results TMI rainfall maps are shown in panels i and j. The maps reveal that most of the precipitation over Algiers is missed, even if the TMI rain product is derived by means of a completely different PMW algorithm, the Goddard Profiling Algorithm (GPROF; Kummerow et al. 2001; McCollum and Ferraro 2003). In general, quite low precipitation is detected over the land. Rain rates do not differ very much between the two methods, but are in general lower for NRLT. In the NRLT maps the rain field appears shifted to the North, preventing a meaningful

506

F. Torricella et al.

numerical comparison between the two, but the main features of the field seem fairly preserved, even 4 h after the calibration.

5 CONCLUSIONS Tests of a global blended PMW-IR satellite rainfall technique were conducted for two rainfall events in the Mediterranean area: (1) rainstorms over the Po Valley in Northern Italy and (2) a coastal heavy, flood-producing storm over Algiers. The technique constantly evolves in time due to the everexpanding suite of PMW sensors and multispectral GEO imagers. However crude the blending mechanism might be, as for instance in the use of a single thermal IR channel, the technique performed fairly well when the underlying MW algorithm supplies reliable input rain maps for its calibration. Results show that, on one hand, the technique suffers from the usual shortcomings of the adopted PMW algorithm, as for instance the unsatisfactory treatment of coastal environments and the possible misclassification of pixels in the screening procedure that precedes the rain retrieval. In certain environments these deficiencies prevent rainfall detection in the first place and therefore a complete representation of precipitation fields, as demonstrated by the analysis of the Algeria event. Moreover, the technique reveals intrinsic shortcomings connected mainly to the presence over the same area of quite different precipitation types, as is the case of very cold convective nuclei embedded in stratiform fields. The technique establishes geolocated rainrate/brightness temperature relationships that are the more correct the more homogeneous the characteristics of the precipitation field for the analysis area. This is obviously driven by the global design of the technique. On the other hand, if the previous unfavorable conditions do not arise, the technique is able to fairly reconstruct the precipitation field outside the space/ time domain covered by PMW observations, even in case of sparse and uneven PMW overpasses. The improvement of PMW rain estimations and the use of multispectral channel analysis will add more precision to the presently available blended products. It is conceivable that the advent of the ongoing international missions aimed to dramatically reduce the gaps between successive MW observations does not undermine the usefulness of blended techniques that remain the preferable method for producing reliable instantaneous rainfall maps, if they are required to be global, frequent and continuous in space.

6 REFERENCES Adler, R. F., C. Kidd, G. Petty, M. Morrissey, and H. M. Goodman, 2001: Intercomparison of global precipitation products: The third precipitation intercomparison project (PIP-3). Bull. Amer. Meteor. Soc., 82, 1377–1396.

Application of a Blended MW-IR Rainfall Algorithm

507

Calheiros, R. V. and I. Zawadzki, 1987: Reflectivity rain-rate relationship for radar hydrology and Brazil. J. Climate Appl. Meteor., 26, 118–132. Ferraro, R. R., 1997: Special sensor microwave imager derived global rainfall estimates for climatological applications. J. Geophys. Res., 102 (D14), 16715–16735. Ferraro, R. R. and G. F. Marks, 1995: The development of SSM/I rain-rate retrieval algorithms using ground-based radar measurements. J. Atmos. Oceanic Technol., 12, 755–770. Kummerow, C., Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling Algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000a: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and geostationary satellite data. Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia, eds., VSP, Utrecht, The Netherlands, pp. 353–363. Turk, F. J., G. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000b: Analysis and assimilation of rainfall from blended SSMI, TRMM and geostationary satellite data. Proc. 10th AMS Conf. Sat. Meteor. Ocean., 9–14 January, Long Beach, CA., 66–69.

40 RETRIEVING PRECIPITATION WITH GOES, METEOSAT AND TERRA/MSG AT THE TROPICS AND MID-LATITUDES Christoph Reudenbach, Thomas Nauss, and Jörg Bendix Faculty of Geography, University of Marburg, Germany

1 INTRODUCTION Water affects all economic, cultural, social, and ecological aspects of daily life all over the world. Hence, investigations of sustainable water management strategies and risk assessment are main topics in today’s hydrological research. Therefore, reliable knowledge about the spatio-temporal dynamics of rainfall as a key input parameter in complex and high resolution (1 × 1 km², 1 h) decision support systems on the regional water cycle like DANUBIA for the upper Danube catchment (refer to Mauser 2003; Ludwig et al. 2003) is indispensable. At present, only adapted mesoscale weather models (Schipper 2004), locally restricted weather radar networks, and optical sensors onboard of geostationary satellites (e.g., MSG) can provide this information in the required spatial and temporal resolution however, with varying accuracy. While model results are necessary for scenario simulations under a changing climate, satellite retrievals are the only way to provide global coverage of rainfall data for climatological purposes, nowcasting and model validation. Because of the high temporal (10–30 min), spatial (3–10 km at nadir) and recently increased spectral (GOES-12, MSG) resolution of geostationary satellite sensors, this provides the opportunity to retrieve short term convective processes with lifecycles even less than 2 h. Moreover, the long-lasting GOES and Meteosat missions makes available long-term validation data sets for numerical models as well as possibilities for monitoring climatic extremes like El Niño (Bendix 1997, 2000). The current paper describes the Advective-Convective Technique (ACT) rainfall retrieval algorithm which is an improved version of the Enhanced 509 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 509–519. © 2007 Springer.

510

C. Reudenbach et al.

Convective-Stratiform-Technique (ECST) (refer to Reudenbach et al. 2001). It is designed for optical sensors of the GOES and Meteosat missions and recently updated by including cloud microphysics to benefit from the increased spectral resolution of the new generation of geostationary satellite sensors (e.g., MSG-SEVIRI). Three examples may present the different stages of the ACT development together with its possible range of application. The first example shows results from mostly convective heavy rain observations during the super El Niño event 1997/98. The second focuses on the combination of a convective and advective retrieval scheme which is necessary for a proper retrieval of mid-latitude precipitation processes over the upper Danube catchment area. The third study points on the severe European summer flooding in 2002 in order to demonstrate the benefits of including cloud microphysics in precipitation retrievals.

2 THE ADVECTIVE-CONVECTIVE TECHNIQUE ACT Operational satellite-based rainfall retrievals predominantly focus on tropical/subtropical regions but only case studies have been performed for the mid-latitudes (Levizzani et al. 2001; Levizzani 2003). These studies have proven that straightforward convective schemes which normally identify potential precipitating clouds by means of their infrared brightness temperature (TBIR) usually perform well in the tropics but cannot simply be applied to the complex situation of mid-latitude frontal precipitation. Hence, a new modular retrieval scheme, the ACT that is also applicable to advective precipitation in the mid-latitudes has been developed. It consist of three modules which deal with precipitation retrieval from convective core areas, from advective cloud regions and an enhanced classification scheme for precipitating clouds by using cloud microphysical properties. Because the first two modules only require brightness-temperatures from the infrared (TBIR) and water vapour (TBWV) channels, they can be used to investigate existing long time series of geostationary data (e.g., Meteosat). However, the increased spectral resolution of the latest generation of geostationary satellites (especially spectral bands at 0.6, 1.6 and 3.9 µm) is necessary for the third module. Figure 1 presents the principal outline of the ACT scheme. The ACT convective module is based on the Enhanced Convective Stratiform Technique (ECST; Reudenbach et al. 2001; Reudenbach 2003) that uses positive TBWV-TBIR differences (DWI) in order to discriminate between deep convective, optically thick clouds (DWI > 0) and non-raining cirrus (DWI < 0, refer to Tjemkes et al. (1997)). Pixels with positive DWI are then subdivided by analysing the frequency distribution of brightness temperatures (TBIR). Areas with TBIR < 1st quartile of the frequency distribution represent overshooting tops of convective cores, those who suit the 1st quartile reveal raining systems at tropopause level and pixels with TBIR < 3rd quartile identify potentially raining cloud systems of high vertical

Retrieving Precipitation with Goes, Meteosat and Terra/MSG

511

extension. As a result, isolated convective cores can be distinguished from directly adjacent stratiform raining areas.

Figure 1. Principal overview of the ACT convective, advective, and microphysics scheme.

The second module detects rainfall areas in warmer frontal systems (e.g., warm frontal clouds). The method is based on an iterative k-means clustering algorithm (Bradley and Fayyad 1998) that is applied to TBIR, TBWV and 3 ×3 infrared standard deviations (StdvIR). It integrates the classified cloud process patterns from the convective module as core raining areas. The resulting clusters represent potentially raining cloud types (advective-stratiform precipitation) if the maximum TBIR and StdvIR of the cluster are below respective higher than specified thresholds. Each cluster is reallocated into single cloud entities for which the compactness is calculated. The resulting cloud systems are classified as raining or non-raining (e.g., Cirrus, non-raining Nimbostratus) by means of a discriminant function based on the cluster centroid temperature, the compactness of the cloud entity and the number of embedded, isolated convective cores. After the classification of raining clouds, a specific rain rate is assigned to each pixel. The rain rates are derived from idealised 3D cloud model runs with the mesoscale Advanced Regional Prediction System (ARPS; Xue et al. 2003). For that, ten years of radiosonde data over central Europe were analysed with regard to specific convective indices (i.e., total-totals) using a 1D cloud model (Zock et al. 1995; Bendix and Bendix 1998). This analysis yielded a set of representative rain-bringing profiles which were used to initialise the ARPS model. The spatio-temporal assignment of rain rate in dependence on the brightness temperature of the raining pixel is performed by aggregating the simulated cloud top temperatures (ARPS) and rainfall

512

C. Reudenbach et al.

rates with respect to the viewing geometry and scan cycle of the used sensor. The model-based relations between cloud-top temperature and rain rate leads to specific transfer functions that are used for the final processing of the classified images. The ACT algorithm can be applied to almost every optical satellite system as long as it provides at least one water vapour and one infrared channel. However, a new cloud microphysics module could be implemented due to the increased spectral resolution of recent geostationary satellite sensors. This is important for the improvement of the ACT, because potential raining cloud systems require both a minimum optical thickness and a minimum effective cloud droplet radius (Lensky et al. 2003). Both parameters are simultaneously retrieved using an improved version of Nakajimas GTR code (Nakajima and Nakajima 1995; Kawamoto et al. 2001; Bendix 2002) that is adapted to Terra-MODIS and MSG-SEVIRI spectral bands at about 0.6 and 3.9 and 11 µm. It will be shown that knowledge about optical depth and effective radius allow an even more accurate separation between non-raining and raining advective clouds within the advective module of the ACT. To demonstrate the feasibility of the ACT, three examples present the chronological development of the different modules. The first example shows results from the convective scheme over Ecuador using GOES-8 data. The second study presents one year of rainfall retrieval over the upper Danube catchment area in Germany by means of both the convective and the advective module based on Meteosat-7 data. The third example reveals the potential enhancement of the ACT for the European summer flooding event of 2002 where cloud microphysics could be additionally considered using Terra-MODIS data in order to simulate the forthcoming potential of Meteosat-8 SEVIRI (MSG).

3 PRECIPITATION DYNAMICS DURING EL NIÑO 1997/98 IN ECUADOR – APPLYING THE ACT CONVECTIVE MODULE El Niño events cause heavy precipitation and significant economical losses in the normally dry costal areas of southern Ecuador and northern Peru while La Niña has similar impacts at the eastern-Andean slopes. In order to retrieve information about the spatio-temporal rainfall distribution and formation the convective module of the ACT was applied to half-hourly band 3 (6.47–7.02 µm) and band 4 (10.2–11.2 µm) GOES-8 data (refer to Bendix et al. 2003) during the centennial super event of 1997/98. The study area that is subdivided in the coastal plains, the Andean highlands including the eastern and western Cordillera and the inter-Andean basins passing into the Amazon region is shown in Fig. 2a. Normally, this region is characterised by two rainy seasons (March/April, October/November) except for

Retrieving Precipitation with Goes, Meteosat and Terra/MSG

513

the eastern Andean slopes between 1,000 m and 3,200 m with one rainy season in July and the more arid regions of southern Ecuador/northern Peru with only one weak precipitation peak in March (Bendix and Lauer 1992). Figure 2b shows total precipitation for an 11-day period during El Niño with maximum values exactly in the arid regions mentioned above and only slight anomalous effects within the inter-Andean basins. Precipitation is decreasing with increasing southern latitude.

Figure 2. Study region (a), total precipitation map (b) and diurnal course (c) for an 11-day period during El Niño 1997/98 retrieved with the ACT convective module over Ecuador/Peru.

The diurnal course (Fig. 2c) as a partial result of mesoscale thermal systems (land-sea breeze phenomenon) shows a maximum over land between 1300 and 0100 local time (LT) whereas during night and early morning precipitation dominates over coastal waters. From 0100 to 0700 LT, a clear maximum can be detected over coastal waters as a result of a welldeveloped land-breeze phenomenon which is slightly shifted westwards between 0700 and 1300 LT. During the afternoon, maximum rainfall is observed in the Amazon region and the coastal plains of Ecuador and northern Peru with a small coast-parallel line of reduced precipitation which indicates the divergence area from the back-flowing branch of the sea-breeze system. The great importance of the sea-breeze system on the spatial structure of El Niño can also be observed between 1900 and 0100 LT where

514

C. Reudenbach et al.

only the coastal plains show significantly enhanced precipitation (Bendix et al. 2003).

4 OPERATIONAL RAINFALL RETRIEVAL FOR THE UPPER DANUBE REGION – APPLYING THE ACT CONVECTIVE AND ADVECTIVE MODULES Within the framework of the German programme on global change in the hydrological cycle (GLOWA) the aim of GLOWA-Danube is to investigate new strategies for sustainable water management in the upper Danube catchment as a representative mountain-foreland region in the mid-latitudes (Ludwig et al. 2003). This is done by implementing a coupled webdistributed model system – DANUBIA – which integrates the different models of 16 socio-economic and natural science research groups by stateof-the-art Java network technology (Barth et al. 2003). The task of the authors is to provide reliable information about the spatio-temporal distribution of rainfall as a key input parameter for DANUBIA reference scenarios between 1995 and 2003. Therefore an AtmoSat object was integrated into the Java framework of DANUBIA using the convective and advective module of the ACT to retrieve precipitation from infrared and water vapour Meteosat-VISSR imagery. To ensure comparable time-lines to the old Meteosat system, the microphysics module is not applied to MSGSEVIRI data which is already used for recent retrievals. Two other objects which provide rainfall information as well are integrated in the DANUBIA framework. AtmoStations uses a distance dependent interpolation technique in order to extrapolate rain-gauge data of 250 stations from the German, Austrian and Swiss meteorological services to the 1 km grid of DANUBIA (Mauser 2003b). AtmoMM5 combines rainfall information which is derived from the mesoscale model MM5 (MM5 = Pennsylvania State University/National Center of Atmospheric Research Fifth Generation Mesoscale Model, Grell et al., 1995) and a subsequent downscaling technique that distributes the MM5 40 km grid result to the 1 km² DANUBIA resolution (Früh et al. 2004). Figure 3 shows a comparison of the amount of rainfall for February and July 1999 retrieved from the AtmoSat, AtmoMM5, and AtmoStations modules of DANUBIA. While for February all three models reveal similar patterns with maximum rainfall at the slopes of the northern Alps with only slight variations of ±5% about the common monthly mean rainfall of 127 mm, only AtmoSat and AtmoStations show intense precipitation in the south-eastern part in July which is mainly due to uncertainties in the MM5 convective parameterization scheme. This is also the cause for the almost complete absence of local induced thunderstorms in AtmoMM5 and AtmoStations precipitation pattern which are concentrated in the alpine foreland and

Retrieving Precipitation with Goes, Meteosat and Terra/MSG

515

eastern Bavaria. On the other hand, these convective systems can be clearly identified in the AtmoSat results. This example can only adumbrate the well performance of the satellite retrieval but an intercomparison of the 5-year period between the three rainfall products (not presented in the current paper) showed that the most realistic view on rainfall dynamics in DANUBIA is given by AtmoSat.

Figure 3. Monthly sum of rainfall derived from the AtmoMM5, AtmoSat and AtmoStations model of DANUBIA for the upper Danube catchment for February and July 1999.

Figure 4 presents the annual variation in monthly mean precipitation. Regarding winter, spring and autumn which are usually dominated by advective-stratiform rainfall events, all three techniques provide similar values. However, for mainly convective induced precipitation during summer, there is a clear deviation between AtmoStation/AtmoMM5 on the one and AtmoSat on the other hand. For AtmoMM5, this is due to the uncertainties in the MM5 cumulus parameterization scheme. For AtmoStations there exists the tendency that the spatio-temporal interpolation which is based on a sparse network of rain-gauge stations expands the area of convective precipitation patterns and therefore the mean amount of rainfall.

516

C. Reudenbach et al.

On the other hand, the half-hourly updated satellite data in AtmoSat enables the identification of borders of small convective systems more precisely which implies generally smaller rainfall areas compared to AtmoStations.

Figure 4. Annual course of monthly mean precipitation derived by the AtmoMM5, AtmoSat and AtmoStations model of DANUBIA for the upper Danube catchment in 1999.

5 PRECIPITATION RETRIEVAL FOR THE SEVERE 2002 ELBE FLOOD – INCLUDING THE ACT CLOUD MICROPHYSICS MODULE The severe Elbe flood event in 2002 caused an overall economic loss of about €18.5 billion (Munich Re 2003). The meteorological reason of this hazardous situation were three consecutive events of heavy rainfall. The first period from 1 to 4 August was dominated by heavy convective precipitation induced by regional destabilization due to the high-pressure system Elke over central Europe. From 5 August, Elke declined and the low-pressure systems Hanne and Ilse became decisive for the weather until on 10 August, a cyclone over the Gulf of Genoa started its way on a Vb track towards Poland. Figure 5a shows the complex situation on 5 August with Hanne centred over the north-western Netherlands causing extensive stratiform cloud areas along the occlusion in the north/north-eastern part and shallow convection over the alps indicating it’s a dissolving cold front. Between these frontal regions intensive convection due to a high pressure ridge from France to

Retrieving Precipitation with Goes, Meteosat and Terra/MSG

517

central Germany can be clearly identified in the Terra-MODIS infrared image.

Figure 5. Terra-MODIS data during the Elbe flood for 5 August 2002, 11:05 UTC, showing (a) infrared data overlayed by the synoptic situation, (b) the results of the ACT convective and advective module merged with data from the C-band radar network of the German weather service and (c) the same merge but with the additionally activated ACT microphysics module.

Figure 5b presents an overlay of radar network data (C-band) of the German Weather Service and the retrieval results using only the ACT convective and advective module. Both, the shallow convection along the altering cold front over the Alps/south-western Germany and the convective systems over France and the German mountain foreland are identified correctly as nonraining and rainfall is assigned to the cloud clusters along the border of the high pressure ridge. Nevertheless, only tropopause near areas in the northern stratiform band where identified as raining. This is due to the extreme heterogeneous structure of the occlusion with 3 × 3 infrared standard deviations greater than 1.5 K and TBIR higher than the actual derived threshold of 232.6 K. The potential improvement of the increased spectral resolution of e.g., Meteosat-8 can be seen in Fig. 5c. Here, the ACT microphysics module was

518

C. Reudenbach et al.

activated and applied to Terra-MODIS bands at 0.6, 3.7 and 11 µm together with the convective and advective module. The identified raining area now covers almost the entire northern cloud band and the isolated systems in the centre are also well detected. Hence, the ACT results based on all three modules fits the radar image significantly better than the two-module mode of the ACT as presented in Figure 5b.

6 CONCLUSION The presented method reveals the wide range of applications for rainfall retrievals based on geostationary satellites with high temporal and at least medium spatial resolution. Thereby the modular concept of the ACT ensures interoperability of many existing and forthcoming satellite sensors. While a combination of the convective and advective module retrieves reliable results for the Tropics and Mid-latitudes at least if aggregated over 3–4 h, the microphysics module applicable to, e.g., Meteosat-8 can significantly improve the ACT especially on a single slot basis and for extreme complex synoptic situations with dominant advective dynamics. The results of the ACT three-module mode shows that the derived rainfall structure agrees well with the corresponding radar information. Hence, the presented algorithm can provide global rainfall data in high temporal resolution especially on the basis of second generation geostationary satellites. The global coverage is a clear advantage in comparison to the locally restricted radar networks. Acknowledgements: Parts of the research described in this paper is funded by the German Federal Ministry of Education and Research as part of the German programme on global change in the hydrological cycle (GLOWADANUBE, Grant No. 07 GWK 04). The authors would like to thank EUMETSAT for providing 5 years of Meteosat data, T. Nakajima for supplying the GTR sources, W. Mauser for the AtmoStations results, A. Pfeiffer and H. Schipper for the AtmoMM5 data and the LCRS diploma student H. Scholz for his contribution to the development of the advective retrieval scheme.

7 REFERENCES Barth, M., R. Hennicker, A. Kraus, and M. Ludwig, 2003: An Integrated Simulation System for Global Change Research in the Upper Danube Basin. 1st World Congress on Information Technology in Environmental Engineering, ITTE. Bendix, J., 1997: Adjustment of the Convective-Stratiform Technique (CST) to estimate 1991/93 El Niño rainfall distribution in Ecuador and Peru by means of Meteosat-3 IR data. Int. J. Remote Sens., 18, 1387–1394. Bendix, J., 2000: Precipitation dynamics in Ecuador and Northern Peru during the 1991/92 El Niño – a remote sensing perspective. Int. J. Remote Sens., 21, 533–548.

Retrieving Precipitation with Goes, Meteosat and Terra/MSG

519

Bendix, J., 2002: A satellite-based climatology of fog and low-level stratus in Germany and adjacent areas. Atmos. Res., 64, 3–18. Bendix, J. and A. Bendix, 1998: Climatological Aspects of the 1991/92 El Niño in Ecuador. Bulletin de L’Institut Francaise d’Etudes Andines, 27, 655–666. Bendix, J. and W. Lauer, 1992: Die Niederschlagsjahreszeiten in Ecuador und ihre klimadynamische Interpretation. Erdkunde, 46, 118–134. Bendix, J., S. Gämmerler, C. Reudenbach, and A. Bendix, 2003: A case study on rainfall dynamics during El Niño/La Niña 1997/99 in Ecuador and surrounding areas as inferred from GOES-8 and TRMM-PR observations. Erdkunde, 57, 81–93. Bradley, P. S. and U. M. Fayyad, 1998: Refining Initial Points for K-Means Clustering. In: Shavlik, J. (Edt.): Proc. of the15th International Conf. on Machine Learning; 91–99. Früh, B., J. W. Schipper, A. Pfeiffer, V. Wirth, and J. Egger, 2005: Using mesoscale climate simulations as a predictor for highly resolved precipitation for the us in hydrological models. Quart. J. Roy. Meteor. Soc., submitted. Grell, G., J. Dudhia, and D. Stauffer, 1995: A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR/TN 398+STR. Boulder, Colorado, USA: NCAR. Kawamoto, K., T. Nakajima, and T. Y. Nakajima, 2001: A Global Determination of Cloud Microphysics with AVHRR Remote Sensing. J. Climate, 14, 2054–2068. Levizzani, V., 2003: Satellite rainfall estimations: new perspectives for meteorology and climate from the EURAINSAT project. Annal. Geophysics, 46, 363–372. Levizzani, V., J. Schmetz, H. J. Lutz, J. Kerkmann, P. P. Alberoni, and M. Cervino, 2001: Precipitation estimations from geostationary orbit and prospects for METEOSAT Second Generation. Meteorol. Appl., 8, 23–41. Ludwig, R., W. Mauser, S. Niemeyer, A. Colgan, R. Stolz, H. Escher-Vetter, M. Kuhn, M. Reichstein, J. Tenhunen, A. Kraus, M. Ludwig, M. Barth, and R. Hennicker, 2003: Webbased modelling of energy, water and matter fluxes to support decision making in mesoscale catchments – the integrative perspective of GLOWA-Danube. Phys. Chem. Earth, 28, 621–634. Mauser, W., 2003a: GLOWA-Danube: Integrative hydrologische Modellentwicklung zur Entscheidungsunterstützung beim Einzugsgebietsmanagement. Petermanns Geograp. Mitt., 147, 68–75. Mauser, W., 2003b: DANUBIA Software-Documentation. GLOWA-Danube Papers Technical Release No. 3. Munich Re, 2003: Topics 2002. Munich Re, Munich. Nakajima, T. Y. and T. Nakajima, 1995: Wide-area determination of cloud microphysical properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J. Atmos. Sci., 52, 4043–4059. Reudenbach, C., 2003: Convective summer precipitation in Central Europe (in German). Bonner Geogr. Abh., 109, 152 pp. Sankt Augustin. Reudenbach, C., G. Heinemann, E. Heuel, J. Bendix, and M. Winiger, 2001: Investigation of summertime convective rainfall in Western Europe based on a synergy of remote sensing data and numerical models. Meteor. Atmos. Phys., 76, 23–41. Schipper, J. W., 2005: Sensitivity of MM5 precipitation to various configurations. Mon. Wea. Rev., submitted. Tjemkes, S. A., L. van de Berg, and J. Schmetz, 1997: Warm water vapour pixels over high clouds as observed by METEOSAT. Contr. Atmos. Phys., 70, 15–21. Xue, M., D.-H. Wang, J.-D. Gao, K. Brewster, and K. K. Droegemeier 2003: The Advanced Regional Prediction System (ARPS), storm-scale numerical weather prediction and data assimilation. Meteor. Atmos. Phys., 82, 139–170. Zock, A., G. Menz, and M. Winger 1995: Regionalisation of rainfall models in Eastern Africa using Meteosat Real-Time-Window data. Proceedings of the International Geoscience Remote Sensing Symposium (IGARSS’95), Florence, Italy (New York: I.E.E.E.); 250–252.

41 MODEL AND SATELLITE ANALYSIS OF THE NOVEMBER 9–10, 2001 ALGERIA FLOOD Carlo M. Medaglia1, Sabrina Pinori1,2, Claudia Adamo1, Stefano Dietrich1, Sabatino Di Michele1, Federico Fierli1, Alberto Mugnai1, Eric A. Smith3, and Gregory J. Tripoli2 1

Istituto di Scienze dell’Atmosfera e del Clima, Consiglio Nazionale delle Ricerche, Roma, Italy 2 Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, WI, USA 3 NASA/Goddard Space Flight Center, Greenbelt, MD, USA

Abstract

In this paper, we present the results of a numerical simulation study of the development of a devastating cyclonic storm that struck the Algerian coast on November 9–10, 2001, with over 200 mm of rainfall. The highresolution numerical simulation of the storm suggests that the cyclone was spawned by when a filament of potential vorticity shed from an intense tropopause fold balled up into an intense local maximum, just north of Algiers. The storm developed west of a frontal occlusion, similar to the classical formation of a polar low. Similarities include: (a) the preexistence of a major trough, (b) the occlusion of the frontal cyclone, (c) the isolation of the warm core low west of the frontal fracture, and d) the growth of the warm core vortex. Strong slantwise neutral uplift over the occluded front was instrumental in producing the heavy rains that affected the city. The numerical simulation is also used to perform more accurate rainfall rate estimates based on available TMI measurements.

1 INTRODUCTION In the Mediterranean Basin, cyclones grow mainly because of baroclinic instability (Holton 1992), which requires horizontal temperature gradients and vertical wind shear. Moreover, the interaction of a large-scale baroclinic wave with an orographic obstacle like the Alps, by virtue of the conservation of potential vorticity (PV), is the cause of a smaller-scale, orographically 521 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 521–534. © 2007 Springer.

522

C. M. Medaglia et al.

induced baroclinic lee cyclone (BLC), that is generated to the lee of the Alps in response to a larger-scale cyclone over the Atlantic. Mediterranean BLC’s have been studied in the last few years from observational, theoretical, and modeling perspectives (e.g., Romero et al. 1997, 2000; Doswell III et al. 1998; Homar et al. 1999; Campins et al. 2000). The interaction between orography and synoptic fluxes, even if applicable to a large fraction of Mediterranean cyclones, does not completely explain all the different types of events that occur over the Mediterranean Basin (Tudurì and Ramis 1997). In fact, at almost the same time as the Mediterranean BLC theory was being formalized, a different kind of subsynoptic cyclone, with similarities to tropical cyclones and polar lows, was observed. These systems are typically about 200 km in diameter and appear as small commashaped or near-circular cloud patterns, often with a clear eye at the center, and in general they are also connected to high PV anomalies that initiate the cyclonic disturbance. Some cases of such structures above the Mediterranean Sea were reported by Tsidulko and Alpert (2001). These strongly convective cyclones cannot be explained by dry baroclinic instability alone: latent heat release (Businger and Reed 1989) and sea interaction play a major role in their development. More recently, Rossa et al. (2000) have studied the evolution of a PV column formed by the concurrence of a PV anomaly in the upper troposphere with a cyclonic circulation on the ground. It appears that the PV positive anomaly maintains the cyclone, by feeding it from the higher layer – as already proposed by Hoskins et al. (1985), who suggested that a high PV anomaly induces a circulation cell of ascending motion eastward of the ridge and descending westward. In this paper, we use model simulations in conjunction with satellite observations to perform a detailed analysis of the processes leading to the devastating cyclonic storm that struck the Algerian coast on November 9–10, 2001 producing more than 200 mm of rainfall and winds of 33 m s–1 – see also Tripoli et al. (2004). In particular, this study focuses on the upper levels precursors and on the sea–air interaction to better understand the exact timing and location of upper-level PV coupling and associated low-level cyclogenesis.

2 SYNOPTIC DESCRIPTION An initial baroclinic instability associated with a large-scale tropopause fold over Western Europe was at the origin of the heavy rains that affected northwestern Algeria on November 9–10, 2001 and then the Balearic Islands. The meteorological situation was characterized by an infiltration of stratospheric air over central Europe, the Iberian peninsula and then the Gibraltar Gulf. As shown in Fig. 1, a high-pressure area centered west of Ireland and a lowpressure one centered on the Alpine region dominated the surface condition on November 9. A cold front was also present in southwestern Europe.

Model and Satellite Analysis of the Algeria Flood

523

At the same time, the 500 hPa situation showed a very different situation with an area of low pressure extending from the Scandinavian area to Gibraltar, where there were strong winds with velocity reaching 150 Kts (at 300 hPa).

Figure 1. Meteorological situation on November 9, 2001: UK Met Office surface analysis (left) and NCEP 500 hPa geopotential height (right).

On November 10, the upper level situation was the same, characterized by an occlusion of the low pressure area over Spain and North Africa. Nevertheless, the surface situation was different: a lowering of the pressure field occurred over the Algerian coasts within the lower levels of the atmosphere (850 and 700 hPa), creating a depression that evolved from southwest Algeria towards the north. Then this depression developed into a cyclone, which moved towards the Balearic Islands while growing slightly, and maintaining a perturbed flow on northwestern Algeria. The frontal situation occurring during the first hours of November 9 was suddenly modified by the injection of cold and dry air coming from higher latitudes, making the surface temperature in Southern Spain and Algeria drop 10 K at midday. This is demonstrated by Fig. 2 that shows the presence of a polar filament carrying a deep layer of stratospheric-originated cold air down into the troposphere – here, the polar filament is evidenced by the (dark) low water vapor (WV) values over Central Europe, Spain and Northwestern Africa in the METEOSAT image, as well as by the corresponding low values of columnar ozone as measured by the Total Ozone Mapping Spectrometer (TOMS) onboard the Advanced Earth Observing Satellite (ADEOS) – see Kramer (2002). Due to the presence of this cold air mass, the surface pressure rapidly dropped down, creating a vortex over the Algerian coast and allowing the formation of a cyclone during the day of November 10. On both days, extraordinary rainfalls of 120–140 mm in 12 h were reported, which led to the flooding disaster in Algeria. Then, during the night of November 10–11, this cyclone moved from the Algerian coast towards the Balearic Islands.

524

C. M. Medaglia et al.

Noteworthy, the low-level circulation developed without a correspondent low-level cold temperature field until late in the period, when a pool of cold air sat over the low-pressure surface – which is in marked contrast with developments in which surface baroclinicity is crucial (Hoskins et al. 1985).

Figure 2. Satellite observations for November 10, 2001: METEOSAT WV channel at 1200 UTC (left) and TOMS total ozone representative for the whole day (right).

3 SIMULATION OF THE EVENT We have performed a cloud resolving simulation of the event using the University of Wisconsin–Nonhydrostatic Modeling System (UW-NMS) developed by Tripoli (1992). The UW-NMS model is based on the nonBoussinesq quasi compressible dynamical equations, and employs a twoway multiply nested Arakawa “C” grid system – see Tripoli (1992), Tripoli and Cotton (1982, 1986), and Flatau et al. (1989) for a detailed description of the numerical integration scheme, of the transport equations, and of the microphysics and dynamic module of the model. To simulate this event, three nested grids were used: the first (120 × 90 grid points) with 37.5 km resolution over a large region spanning much of Europe and North Africa; the second (240 × 180 grid points) with 9.4 km resolution, covering the western Mediterranean Basin; and the finest one (160 × 96 grid points) with 2.4 km resolution, covering the region just around Algiers. The high-resolution grid was used only during the first 60 h to capture the storm around the city of Algiers. The model was initiated at 1200 UTC of November 8 from the NCEP AVN analysis and integrated for a period of 72 h. Nevertheless, different types of initialization were performed to evaluate the sensitivity of the analysis to different initial and boundary conditions – for instance, to evaluate the relative roles of the orography and sea surface latent heat flux versus the action of the upper-level PV centers.

Model and Satellite Analysis of the Algeria Flood

525

Figure 3. Top: Accumulated rainfall prediction over the 72 h simulation period; Bottom: accumulated rainfall measurements (mm) on the Algerian coast from November 9, 0600 UTC, till November 11, 1800 UTC (from the Algerian Meteorological Office).

The model simulation captured the flooding precipitation and reported winds of the event well, however, the heavy precipitation maximum over the city of Algiers was simulated to be slightly less than observations (see Fig. 3). This was not surprising, since convective storms not anchored to topography were the primary mechanisms for the heaviest precipitation. The simulated wind was more precise, as the simulated peak gust of 33 m s–1 near Algiers exactly matched reported peak gusts. The Algerian cyclogenesis process and its predictability had much in common with the “Storm of the Century” development on March 12, 1993 off the Texas coast reported by Bosart et al. (1996). Employing “PV thinking” (Hoskins et al. 1985) to examine the synoptic forcing, one finds that an exceptionally strong south–westward digging trough created an

526

C. M. Medaglia et al.

uncommonly deep tropopause fold in the days prior to the storm, forcing stratospheric PV values down to 650 hPa pressure (see Fig. 4).

Figure 4. Simulation of PV at 500 hPa (below 8 km) for November 10 1800 UTC.

Figure 5. View of simulated low level flow vectors on grid 2 (9.4 km resolution) at 10 m MSL for 1000 UTC, November 10, 2001. Topography is color shaded in θe (from 290 K in blue to 305 K in red), while mean sea level pressure isobars at 2 hPa intervals are also shown. Channeling around topography and transport of warm θe from Wind Induced Surface Heat Exchange (WISHE) are highlighted.

Model and Satellite Analysis of the Algeria Flood

527

The evolution and coupling of the surface air stream to the PV filament aloft resulted in the surface storm cyclogenesis. Figure 5 depicts the surface frontal evolution as the system moved south off of the coasts of Spain and France on the day prior to the storm genesis. Particularly interesting was the extended influence of the upstream mountain ranges in channeling the cold air movement. As the cold surges reached the southern Mediterranean shores, an easterly low-level barrier jet formed the next day north of Tunisia and Algeria. The low-level jet flowed westward into the developing upward ageostrophic circulation aloft forming in advance of the approaching upper level PV anomaly from the southwest. The strong shear across the contrasting air masses along the barrier jet resulted in a vortex sheet that would be the nucleus of future development, when convection produced strong local convergence maximum along the line. During the 24 h before the cyclogenesis, the wind induced surface heat exchange (WISHE) along the barrier jets increased the boundary layer equivalent potential temperature (θe) to critical convection levels. For coupling between the warm surface θe and PV aloft to occur, deep moist convection had to develop under the PV filament aloft. As shown in Fig. 6, simulated deep convection was initiated off shore within enhanced surface convergence formed beneath the left exit region of the upper level jet streak that was approaching from the southwest. The coupling of the surface to the forced warm θe filament aloft is strikingly evident.

Figure 6. View from the south of the simulated Algiers cyclone at 1000 UTC, November 9, 2001 on grid 2. Topography is color shaded according to the equivalent potential temperature (θe). Isobars of mean sea-level pressure are reported (white) at 1 hPa intervals.

A sensitivity experiment that was run without the Atlas Mountains showed that similar cyclogenesis did not occur without the presence of the mountains, suggesting the critical role of orography in inducing the mesoscale cyclogenesis process (see Fig. 7). Examination of the barrier jet

528

C. M. Medaglia et al.

depicted at this time on the fine grid suggests that cooler θe air sweeping of the African continent from the south, created the effect of a warm occlusion along the barrier jet, and even rotated the barrier confluence zone northward. Notably, the moist airstreams forced into the Atlas Mountains southwest of the upper level ageostrophic forcing shown in Fig. 5 did not result in deep convective storms, suggesting that there was a critical interaction between the ageostrophic forcing aloft and the orographically enhanced circulation below to create the deep convective plumes.

Figure 7. NMS 12 h accumulated precipitation for November 10, 1100 UTC: results from the runs with and without the Atlas Mountains (left and right panels), respectively. The isobars of MSPL at 4 hPa intervals are shown as while lines.

Immediately following the eruption of deep convection, coupling of the surface flow with the PV filament above took place. The surface vorticity began to strengthen dramatically, accelerating to levels in excess of 500 10–5 s–1. This appeared to be in response to the convergence of angular momentum into the localized regions of convective lifting. In effect, there was a “balling up” of the aforementioned vortex sheet (Jascourt 1997). A surface low formed by 0300 UTC and deepened to 989 hPa by 0830 UTC. Northerly near surface winds exceeding 37 m s–1 were occurring over the water on the western flank of the vortex. Convection concentrated the mesoα-scale ageostrophic convergence into a convective band 55 km across. The meso-β-scale vortex remained stationary along the shoreline for over 9 h, perhaps in response to conservation of potential vorticity in the downsloping flow on the storm’s eastern flank. This drove winds of 33 m s–1 and heavy rains at Algiers. The no-mountain sensitivity experiment seems to confirm the role of the orography in this scenario. During the intense stationary phase of the storm on the afternoon and evening of November 10, the warm θe anomaly at the storm’s surface inflow increased by 2 K due to strong WISHE forcing. The storm developed a 2 K weak warm core

Model and Satellite Analysis of the Algeria Flood

529

anomaly aloft, only about 60 km in diameter. This appeared as a shallow thermal perturbation within an otherwise much larger cold core structure. To the west, the strong north-easterly low-level air stream, that originated as the the barrier convergence zone and then spiraled cyclonically into the mesoscale vortex, ultimately decoupled from the vortex as the vortex became stationary, then drifted westward driving strong winds into the Atlas Mountains and depositing copious amounts of precipitation west of Algiers. The meso-β-scale vortex accelerated its movement southward by 2000 UTC on the 10th, apparently as the ageostrophic forcing aloft moved northeastward and decoupled. Immediately, surface pressures rose rapidly and the storm dissipated completely by 0000 UTC, November 11. In the meantime, a meso-α-scale cyclone began to strengthen to the north and east (see Fig. 8).

Figure 8. Surface equivalent potential temperature θe on November 11 0900 UTC.

Two additional sensitivity tests featuring (a) no latent heating and (b) no land surface over Africa, lead to the conclusion that the Atlas mountains and

530

C. M. Medaglia et al.

the effects of deep moist convection spawned by flow regimes created from the Atlas mountains were responsible for the meso-β-scale storm that affected Algiers (see Fig. 9). These studies demonstrate that the large meso-α-scale cyclone would have taken place anyway late on November 10 in direct response to the upper level PV forcing, even without the convective induced coupling, and would have affected the Mediterranean region to the north, but would have affected Algiers less severely. Indeed, this was already taking place in the simulation before the local orographically generated flow systems shortcircuited that scenario and plugged the upper level storm into the surface WISHE forcing via deep moist convection.

Figure 9. NMS 12 h accumulated precipitation for the “no Africa land “run for November 10, at 1100 UTC (left) and at 2400 UTC (right). The isobars of MSPL at 4 hPa intervals are shown as white lines.

4 PRECIPITATION MEASUREMENTS FROM TRMM The lack of ground radar coverage over Northern Africa and the scarcity of rain gauges make the precipitation measurements derived from satellite sensors especially important for this event. In another paper in this book, Torricella et al. (2007) analyze the results obtained using the combined microwave/infrared (MW/IR) method developed by Turk (see Turk et al. 2000). Here, we use measurements taken by the Precipitation Radar (PR) embarked on the Tropical Rainfall Measuring Mission (TRMM) satellite to test the corresponding rainfall rate estimates based on observations taken by the TRMM Microwave Imager (TMI) (see Kummerow et al. 1998 for a description of the TRMM sensor package).

Model and Satellite Analysis of the Algeria Flood

531

Figure 10. Rainfall rates measured by the TRMM PR on November 10, 2001 at 0205 UTC (left) and corresponding rainfall rates estimated by the GPROF algorithm from TMI observations for the same overpass (right). Values in mm h–1.

Figure 11. Rainfall rates (mm h–1) estimated by the BAMPR algorithm from TMI observations for the same overpass of Fig. 10. The right panel is a zoom for the northern Algeria-Morocco border region.

Results are shown in Fig. 10 for the Goddard Profiling Algorithm (GPROF) (Kummerow et al. 2001), which is the standard TMI rainfall algorithm, and in Fig. 11 for the Bayesian Algorithm for Microwave-based Precipitation Retrieval (BAMPR) (Mugnai et al. 2001; Di Michele et al. 2003, 2005). It is evident that the GPROF estimates are quite unsatisfactory – especially over land. While this may be partially due to an inadequate screening procedure, we notice that the GPROF cloud-radiation database is not specifically tailored for the Mediterranean area and it is based on events having different microphysical characteristics than the specific Algerian event. On the other hand, the BAMPR estimates match pretty well the PR measurements, both over land and sea. Apart from specific differences between the two algorithms, we believe that the main reason why the BAMPR algorithm works better is that it uses a cloud-radiation database

532

C. M. Medaglia et al.

which has been specifically built for the Mediterranean Basin within the EURAINSAT project (Tassa et al. 2003a, b). This database is composed by cloud and precipitation microphysical profiles, that have been generated by means of the UW-NMS model for several heavy precipitation events over the Mediterranean area, including the present Algerian storm. As a matter of fact, about 40% of the retrieved profiles belong to the Algerian simulation, which demonstrates the importance of having cloud-radiation databases that include the class of events under observation. At the same time, these results provide an indirect validation of the model microphysics.

5 CONCLUSIONS This study demonstrates that exact timing and location of upper level PV coupling and associated surface cyclogenesis, as described by Bosart et al. (1996), can be significantly altered by locally generated and unbalanced flow systems and driven by convection forcing. Fortunately, as is often the case for Mediterranean storms, these local flow systems seem to be well behaved and highly predictable given a model’s ability to finely resolve surface topographical and coastline features and skillfully simulate their interaction with the broader scale flow. Acknowledgments: This study has been funded by the Italian National Group for Prevention from Hydro-Geological Disasters (GNDCI), by the Italian Space Agency (ASI), and within the frame of EURAINSAT – a shared-cost project (contract EVG1-2000-00030) co-funded by the Research DG of the European Commission (5th Framework Program).

6 REFERENCES Bosart, L. F., G. J. Hakim, K. R. Tyle, M. A. Bedrick, W. E. Bracken, M. J. Dickinson, and D. M. Schultz, 1996: Large-scale antecedent conditions associated with the 12–14 March 1993 cyclone (Superstorm ’93) over Eastern North America. Mon. Wea. Rev., 124, 1865–1891. Businger, S. and R. J. Reed, 1989: Cyclogenesis in cold air masses. Wea. Forecasting, 4, 133–156. Campins, J., A. Genovès, A. Jansà, J. Guijarro, and C. Ramis, 2000: A catalogue and a classification of surface cyclones for the western Mediterranean. Int. J. Clim., 20, 969–984. Di Michele, S., F. S. Marzano, A. Mugnai, A. Tassa, and J. P. V. Poiares Baptista, 2003: Physically-based statistical integration of TRMM microwave measurements for precipitation profiling. Radio Sci., 38, 8072. Di Michele, S., A. Tassa, A. Mugnai, F. S. Marzano, P. Bauer, and J. P. V. Poiares Baptista, 2005: Bayesian algorithm for microwave-based precipitation retrieval: description and application to TMI measurements over ocean. IEEE Trans. Geosci. Remote Sens., 43, 778–791. Doswell III, C., C. Ramis, R. Romero, and S. Alonso, 1998: A diagnostic study of three heavy precipitation episodes in the western Mediterranean. Wea. Forecasting, 13, 102–124.

Model and Satellite Analysis of the Algeria Flood

533

Flatau, P., G. J. Tripoli, J. Berline, and W. Cotton, 1989: The CSU RAMS Cloud Microphysics Module: General Theory and Code Documentation. Technical Report 451, Colorado State University, 88 pp. Holton, J. E., 1992: An Introduction to Dynamic Meteorology. Academic Press, 319 pp. Homar, V., C. Ramis, R. Romero, S. Alonso, J. Garcìa-Moya, and M. Alarcòn, 1999: A case of convection development over the western Mediterranean sea: A study through numerical simulations. Meteor. Atmos. Phys., 71, 169–188. Hoskins, B. J., M. E. McIntyre, and A. W. Robertson, 1985: On the use and significance of isentropic potential vorticity maps. Quart. J. Roy. Meteor. Soc., 111, 877–946. Jascourt, S., 1997: Convective organizing and upscale development processes explored through idealized numerical experiments. PhD Thesis, University of Wisconsin–Madison, Madison, WI 53706, 267 pp. Kramer, H. J., 2002: Observation of the Earth and its Environment: Survey of Missions and Sensors. Springer-Verlag, 1510 pp. Kummerow, C. D., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 808–816. Kummerow, C. D., D. B. Shin, Y. Hong, W. S. Olson, S. Yang, R. F. Adler, J. McCollum, R. Ferraro, G. Petty, and T. T. Wilheit, 2001: The evolution of the Goddard Profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820. Mugnai, A., S. Di Michele, F. S. Marzano, and A. Tassa, 2001: Cloud-model-based Bayesian techniques for precipitation profile retrieval from TRMM microwave sensors. Proc. ECMWF/EuroTRMM Workshop on Assimilation of Clouds and Precipitation, ECMWF, Reading, UK, 323–345. Romero, R., C. Ramis, and S. Alonso, 1997: Numerical simulation of an extreme rainfall event in Catalonia: Role of orography and evaporation from the sea. Quart. J. Roy. Meteor. Soc., 123, 537–559. Romero, R., C. A. Doswell III, and C. Ramis, 2000: Mesoscale numerical study of two cases of long-lived quasistationary convective systems over eastern Spain. Mon. Wea. Rev., 128, 3731–3751. Rossa, A. M., H. Wernli, and H. C. Davies, 2000: Growth and decay of an extra-tropical cyclone’s PV-tower. Meteor. Atmos. Phys., 73, 139–156. Tassa, A., S. Dietrich, S. Di Michele, S. Pinori, and A. Mugnai, 2003a: The EURAINSAT Cloud Radiative Dataset. EURAINSAT Technical Report Series, 1, 32 pp. Tassa, A., S. Di Michele, A. Mugnai, F. S. Marzano, and J. P. V. Poiares Baptista, 2003b: Cloud-model-based Bayesian technique for precipitation profile retrieval from TRMM Microwave Imager. Radio Sci., 38, 8074. Torricella, F., V. Levizzani, and F. J. Turk, 2007: Application of blended MW-IR rainfall algorithm to the Mediterranean. Measuring Precipitation from Space: EURAINSAT and the Future, V. Levizzani, P. Bauer, and F.J. Turk, eds., Springer, 497–508. Tripoli, G. J. and W. R. Cotton, 1982: The Colorado State University three dimensional cloud/mesoscale model. Part I: General theoretical framework and sensitivity experiments. J. Rech. Atmos., 16, 185–200. Tripoli, G. J. and W. R. Cotton, 1986: An intense, quasi-steady thunderstorm over mountainous terrain. Part IV: Three-dimensional numerical simulation. J. Atmos. Sci., 43, 894–912. Tripoli, G. J., 1992: A nonhydrostatic mesoscale model designed to simulate scale interaction. Mon. Wea. Rev., 120, 1342–1359. Tripoli, G. J., S. Pinori, S. Dietrich, C. M. Medaglia, G. Panegrossi, A. Mugnai, and E. A. Smith, 2005: The 9–10 November 2001 Algerian flood: A numerical study. Bull. Amer. Meteor. Soc., 86, 1229–1235.

534

C. M. Medaglia et al.

Tsidulko, M. and P. Alpert, 2001: Synergism of upper-level potential vorticity and mountains in genoa lee cyclogenesis: A numerical study. Meteor. Atmos. Phys., 78, 261–285. Tudurì, E. and C. Ramis, 1997: The environments of significant convective events in the western Mediterranean. Wea. Forecasting, 12, 294–306. Turk, F. J., G. D. Rohaly, J. Hawkins, E. A. Smith, F. S. Marzano, A. Mugnai, and V. Levizzani, 2000: Meteorological applications of precipitation estimation from combined SSM/I, TRMM and infrared geostationary satellite data. In: Microwave Radiometry and Remote Sensing of the Earth’s Surface and Atmosphere, P. Pampaloni and S. Paloscia eds., VSP, Utrecht, The Netherlands, pp. 353–363.

42 MODELING MICROPHYSICAL SIGNATURES OF EXTREME EVENTS IN THE WESTERN MEDITERRANEAN TO PROVIDE A BASIS FOR DIAGNOSING PRECIPITATION FROM SPACE Gregory J. Tripoli1, Carlo M. Medaglia2, Giulia Panegrossi1, Stefano Dietrich2, Alberto Mugnai2, and Eric A. Smith3 1

Department of Atmospheric and Oceanic Sciences, University of Wisconsin, Madison, WI, USA 2 Institute of Atmospheric Sciences and Climate, CNR, Rome, Italy 3 NASA/Goddard Space Flight Center, Greenbelt, MD, USA

1 INTRODUCTION Precipitation occurring over the Mediterranean basin is typically unusually difficult to measure due to variability resulting from the irregular terrain. Radar based precipitation measurements also are compromised by the terrain, which contaminates reflectivity with excessive ground cover and blocks the radar beam at low elevation angles. That same terrain also is instrumental in providing for the localization of rain producing storms that results in long term and flash flooding situations, especially in the fall season. These limitations are presumably overcome by precipitation measurements taken remotely from space based observing platforms. From space, instruments directly measure upwelling electromagnetic radiation. Radiation in the microwave bands are actively scattered, absorbed, and reflected by liquid and ice condensate present in the air. Observations of variability in the amount of microwave radiation reaching space are therefore tied to the structure of the liquid and ice present in the atmosphere from which the radiation originates. It is known that surface precipitation rate is also related to the vertical structure of the liquid and ice present. The question arises: “How unique are these two relationships?” and “Can we

535 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 535–547. © 2007 Springer.

536

G. J. Tripoli et al.

infer surface precipitation rate from the measurement of the intensity of radiation from a limited number of microwave radiance measurements?” Typically space-borne microwave radiometers (like ones on board of TRMM and DSMP satellites) measure intensity of 4–6 carefully considered microwave frequencies (Spencer et al. 1989; Kummerow et al. 1998). This measurement would appear to have an insufficient number of degrees of freedom to distinguish the many complex vertical condensate structures that might evolve. However, we also understand that not all precipitation structures that relate to the observed radiance profile are actually likely to occur. The hypothesis of precipitation retrieval hinges on this observation and so that cloud structure and attendant precipitation can be inferred given knowledge of what cloud structures and attendant surface precipitation rate are most likely to lead to a given set of microwave radiance observations (Wilheit 1986).

2 USING CLOUD RESOLVING MODELS TO INTERPRET SATELLITE OBSERVATIONS The cloud structures that are most likely associated with a given radiance observation depend on: 1. Geographical location. 2. Season. 3. Basic dynamic mechanism of the storm being measured, i.e., vertical convective, slantwise convective or stratiform. 4. The location of the measurement within the storm, i.e. (anvil, convective core, warm front, cold front). We can determine those relationships through numerical cloud resolved models (CRMs) of “typical” precipitating storms (see Fig. 1). CRMs can simulate a cloud or systems of clouds through explicit simulation of the flow dynamics and the attendant microphysics and precipitation evolution. The result is a physically consistent prediction of the dynamics, microphysics and thermodynamics of the precipitating system. We can then apply a Passive Radiation Model (PRM) to the simulated atmospheric structure of the cloud to simulate the upwelling radiation that a satellite would observe. The model simulations of dynamics, microphysics and radiance are verified against conventional and special observations through Cloud Radiation Verification Studies (CRVSs). CRVSs can also be used to improve the model physics through these comparisons (Panegrossi 2004). The CRM/PRM combination is a Cloud Radiation Simulation (CRS) and results in a gridded data set containing the atmospheric dynamic, thermodynamic and microphysical structure, the surface precipitation rate, and the attendant upwelling microwave radiation. A database composed of a wide variety of CRS results is called a Cloud Radiation Data Base (CRDB).

Microphysical Signatures of Extreme Events in the Western Mediterranean

537

A CRDB then is used as a statistical basis from which one can retrieve precipitation from a space-based radiance observation. The analysis performed in the PRM is diagramed in Fig. 2. After choosing an “appropriate” model simulated vertical profile containing model simulated state parameters and microphysics, the upwelling microwave radiation must be calculated. This is accomplished through a PRM that calculates model inferred brightness temperature according to the particular microwave frequencies and path geometry viewed by the satellite.

Figure 1. Schematic describing the relationship between Cloud Radiation Simulation (CRS), space-based microwave observations, and precipitation measurement.

Figure 2. Cloud Radiation Database Generation Flow Diagram.

538

G. J. Tripoli et al.

Using the CRDB as a statistical reference from which to base retrieval has several important sources of error. First, there are errors in the model simulation itself including errors from the model’s inability to simulate a weather system dynamics or the microphysics of precipitation formation. Precipitation simulation is especially important since there are large sensitivities of the simulated upwelling radiation to the predicted hydrometeor size, habit and phase(s). Studies comparing simulated microphysics to actual cloud observations have shown mixed results. Overall the performance of bulk microphysics models in properly simulating microphysics observed in situ can be characterized as weak. However, there may be more realism in the simulation of the overall relationship between microphysics and precipitation rate, even if the microphysics simulated for a particular storm is poor. Fortunately, based on experiences comparing radiance implied by the simulated microphysics to satellite observations, the simulated microphysics occurring in Mediterranean storms appears to be less likely to produce unrealistic satellite signatures than other storms around the world that we have attempted. This is likely because of microphysical similarity between Mediterranean storms that allows a “tuned” bulk parameterization to perform well over a wide range of cloud applications and the predominance of liquid rain in most heaving precipitation situations.

Figure 3. Illustration depicting the challenges of “representativeness” for a mesoscale convective system containing a mixture of slantwise updrafts and downdrafts, vertical convective plumes and suspended anvil ice. The upwelling ration reaching the space borne sensor is highly dependent on the angle of view and the exit point of radiation ray.

Microphysical Signatures of Extreme Events in the Western Mediterranean

539

A second major source of error is in the PRM. Perhaps because of the precipitation is dominated by rain in most cases, the radiative transfer is straightforward and has minimal error. One would expect the light snow situations to be the most challenging, but they are relatively rare except in the very high terrain. A major source of error encountered when attempting to base the relationship between precipitation rate and radiance on entries from a CRDB is the representativeness of the entries employed (see Fig. 3). Representativeness is affected by the angle of the observation and whether the radiative path goes through a horizontally uniform region of the cloud or perhaps passes through differing dynamical regions of the clouds. How can one find a representative vertical atmospheric/radiation profile among the seemingly infinite possibilities of cloud samplings possible even from just a few model runs? This challenge becomes even greater as the resolution of the radiance observation increases and smaller parts of a storm are sampled by the satellite at a given time. The Mediterranean again provides a more forgiving environment for finding representative entries because the precipitation mechanisms tend to conform to topographical forcing and be less variable spatially and over different cases. This will be discussed more below. Finding an appropriate CRDB entry to reference for a precipitation retrieval calculation has been a continuing challenge since no two storms are exactly alike, a CRS is far from perfect, and only a limited number of entries are feasible either to create or store in the CRDB. Hence there is not only the question of which simulated radiance profiles of the CRDB are most appropriate, but also the question of whether any of the profiles are appropriate.

Figure 4. Mediterranean basin depicting mountainous terrain surrounding the sea.

540

G. J. Tripoli et al.

3 APPLICATION TO THE MEDITERRANEAN BASIN In view of these challenges, we propose the hypothesis that CRSs performed for the Mediterranean basin tend to be exceptionally well suited as a statistical basis for retrieval compared to many locations on the globe. This is because of the unique character of precipitating systems in that region that result in an increased robustness in structure. Over the last decade we have performed a number of CRSs of Fall season heavily precipitating storms in the Mediterranean and in each case, the results support our hypothesis.

3.1 Characteristic of Mediterranean precipitating storm Below we summarize what we have found about the uniqueness of the Mediterranean basin for the simulation and prediction of precipitation: 1. Local origins of atmospheric vapor: The Mediterranean basin (see Fig. 4) is characterized by a warm body of water ringed by severe topography on all sides. With the hot dry desert to the south and east, a cool Atlantic Ocean to the west, and a large land mass to the north, much of the moisture falling as precipitation in the basin evaporates from the Mediterranean Sea itself. 2. Since the sea surface temperature is well observed in the Mediterranean, atmospheric models can reasonably estimate surface water vapor and thermal fluxes from the sea surface. 3. Orographic and land use influences on convection initiation: The prediction of convection initiation is perhaps the greatest problem facing quantitative precipitation forecasting (QPF). Where the first convective towers occur is difficult to predict and in most places in the world, depend on initial placement of local density currents and boundaries of various types perhaps left over from previous storms, fronts land use variations, and topography. In the Mediterranean region, topographical forcing together with the land/sea effect dominates initiation (see Fig. 5) and because their influence is represented well in a high resolution model, convective initiation in the Mediterranean tends to be unusually predictable. 4. Orographic influences on stable precipitation: Stable rain is created by forced lifting by non-buoyant rising currents. These rising currents are formed primarily by baroclinic structures featuring flow-up slanted isentropic surfaces (surfaces of constant entropy or potential temperature) or flow up a topographical surface. Stable baroclinic rains are reasonably well predicted, but may have significant error in precise placement on the mesoscale simply due to a small relative error in the placement of the synoptic system. Precipitation tied to topographic lifting, however, is much less sensitive to the error in the large scale system prediction

Microphysical Signatures of Extreme Events in the Western Mediterranean

541

because the topography is well represented in a model at all times and the local slope geometries further localize the most intense lifting.

Figure 5. Schematic depicting the roles of topographic initiation, the EML, and WISHE in focusing and triggering Mediterranean convection.

5. Elevated mixed layer (EML): Elevated mixed layers are well known to provide for the capping of conditionally unstable boundary layers from the formation of unforced deep moist convection, allowing instabilities to build to high levels. Central North America is a good example of where EMLs formed over the Rockies move eastward over the High Plains and allow for extreme instabilities to build as Gulf air flows northward under the EML. The Mediterranean region also can feature an EML (see Fig. 5), especially in the fall season. At that time of the year, the Sahara is still very hot and features a deep dry Desert Boundary Layer (DBL), extending upward to 2–4 km above the surface. As a middle latitude upper level trough approaches the Mediterranean region from the northwest, the ageostrophic Sawyer-Elliasen circulation (Carlson 1991) will result in a surge of warm surface air from the south to the north. The strong thermal gradient between the cool air to the north and the hot desert air to the south will provide additional energy for and so enhance this circulation. As the warm surge moves offshore over the relatively cool Mediterranean Sea, a cool, moist internal Marine Boundary Layer (MBL) is developed adjacent to the surface. Continued Wind Induced Surface Heat Exchange (WISHE) (Yano and Emanuel 1991) occurs as the air flow northward, building high Convective Available Potential

542

G. J. Tripoli et al.

(CAPE) that is capped by the strong EML. Generally, under these circumstances, no significant convection can be triggered until the flow is pushed up a significant topographical barrier in the northern Mediterranean. This all works to further focus the areas of precipitation into predictable locations associated with the topography and land sea boundary. 6. Low Froude number flow and topographic channeling: The Froude number is defined to be the ratio of the inertial accelerations producing vertical lifting over a barrier to static stability induced decelerations to that lifting. It is expressed as: Fr = u/hN, where u is the velocity, N is the Brunt Vaisala frequency and h is the barrier height. The strong capping of the EML increases N within the southerly surge and so decreases Fr. This causes southerly low level flow to resist being drawn over topographical barriers and causes the flow to be diverted around the barriers instead. As a result, flow which is approximately southerly, tends to follow channels between topographical barriers such as the Atlas mountains, Sicily, Sardinia, Corsica, the Italian Peninsula, Crete, Cyprus, and so on (see Fig. 6). The flow also tends to be diverted around significant mountain ranges on the north shore, such as the Mediterranean Alps near Nice, France, because of the strong barrier effect. As a result, the eventual inevitable lifting tends to be focused into highly predictable locations such as Genoa, Italy, that may not contain the highest topography, but where lifting is strongest. There convection is triggered or else stable orographic rains are forced.

Figure 6. Channeling of flow focusing on Genoa Region. Flow is focused into Genoa by a combination of topographical channeling and the barrier effect of the Alps.

Microphysical Signatures of Extreme Events in the Western Mediterranean

543

7. Downslope wind storms: Mountain waves and downslope wind storms resulting from mountain waves can occur on the lee side of mountain ranges under certain atmospheric conditions depending primarily on the vertical shear and stability profile (Tudurì and Ramis 1997). These wind events can be simulated with a competent mesoscale model with a high degree of skill. Moreover, these events produce modulating effects on the initiation and maintenance of precipitating storms and even on the magnitude of surface fluxes. Our simulations of a strong precipitating storm in Friuli, Italy (Tripoli et al. 2000) was one case where the western extent of the heavy rains and the width of the Adriatic channel jet were modulated by the formation of downslope winds off the Apennine mountains (see Fig. 7). Because of their predictability, downslope wind storms also lend a degree of predictability to precipitating systems.

Figure 7. Depiction of downslope winds due to Apennines mountains affecting the channeled flow in the Adriatic Sea.

8. Land–sea breezes: Thermal circulations related to land/sea differences produce concentrated regions of convergence called sea or land breeze fronts that are preferred locations of convection initiation. These features are relatively well posed and accurately predicted in a mesoscale simulation (see Fig. 8). They play a large role in initiating convection that helps lend another degree of predictability to precipitation events.

544

G. J. Tripoli et al.

9. Obstacle flow effects: The mountain islands and mountain ranges surrounding the Mediterranean also create obstacle flow effects that behave differently as the Froude number varies. These obstacle flows produce predictable curved flows on the downstream side of islands such as Corsica that can further concentrate and enhance the surface CAPE being moved around and may act to locally initiate convection in certain regions. The interaction of these flows with each other or with other land breeze flows can produce strong predictable vortex sheets that can lead to the growth of waterspout producing deep moist convection (Golden 1974). Simulations by the authors of waterspouts occurring on 29 October were performed leading to the CAPE and vorticity distribution given in Fig. 8. Regions of intersection between the obstacle flow and the land breeze effect led to a strong vortex sheet simultaneous with strong convergence leading to the predictable initiation of waterspout producing storms.

Figure 8. CRM simulation of complex surface flows over the Tyrrhenian Sea occurring at night resulting from a combination of obstacle flow around Corsica, a land breeze of the Italian peninsula and downslope flow off of the mountains in Corsica. The CAPE field is shown to be formed by WISHE upstream of Corsica and then advected around to the south and lifted in the zone where the land breeze off of Italy meets the obstacle flow. This particular situation produced a vortex sheet along the converging flows in the Tyrrhenian Sea that led to the formation of cumulus congestus and eventually waterspouts.

10. Lee cyclogenesis: The movement of synoptic scale baroclinic waves into the Mediterranean basin is characterized by vorticity laden flow moving across major topography barriers. The principles of vorticity conservation (Maddox et al. 1979) lead to lee cyclogenesis, often over

Microphysical Signatures of Extreme Events in the Western Mediterranean

545

the Mediterranean basin. The effect of the mountains is to increase the predictability of cyclogenesis location by helping focus the location of vorticity buildup by attaching it to known topographical features. The precipitation associated with the cyclogenesis has enhanced predictability, because of the reasons listed above. 11. Potential vorticity–surface coupling: Cyclogenesis over the Mediterranean is typically associated with the injection of stratospheric potential vorticity into the troposphere as a short-wave disturbance moves around the northwest side of a major trough (Tsidulko and Alpert 2001). Coupling of the upper level feature with a surface potential vorticity maximum can lead to surface cyclogenesis. Surface potential vorticity maximum can be inferred through warm near surface anomalies found in either the potential temperature or in the equivalent potential temperature in the case where convection is occurring. Originally this coupling mechanism was explained for Gulf-of-Mexico cyclogenesis by Bosart (1996) and later shown for Mediterranean cyclogenesis by Tripoli et al. (2004) (see Fig. 9). The interaction with the well represented sea surface temperature field, the large-scale dynamics of an approaching trough and the interaction with the local orography help focus cyclogenesis and again aid in enabling atmospheric models to represent the effect deterministically.

Figure 9. Numerical simulation of cumulus arising from Mediterranean causing a coupling with an upper level tropopause fold and rapid surface cyclogenesis near Algeria. The lowered 306 θe surface is a result of a tropopause fold and is associated with a similar shaped potential vorticity anomaly. The cumulus near the surface are in the form of θe plumes also containing 306 θe. Surface isobars drawn every 2 hPa are also drawn. View is from the west-southwest showing the Atlas mountains of north Africa near the center of the figure.

546

G. J. Tripoli et al.

4 CONCLUSION AND FUTURE WORKS Overall, precipitation occurring in the Mediterranean basin has been shown to feature a high degree of predictability because of these characteristics. As a result, CRSs tend to capture the precipitation mechanisms applicable to particular situation better than in many other places of the world where there are far more degrees of freedom to storm formation and so more uncertainty to the relationship between observed radiance and precipitation rate. A CRDB consisting of a wide cross section of Mediterranean storm types, can be expected to perform exceptionally well in this environment. Nevertheless, the performance of precipitation retrieval could be improved if the storm type occurring could be more precisely defined. The overall good predictability of Mediterranean events also means that a realtime CRM model prediction of precipitation in the basin will likely predict most precipitation events albeit with some degree of error. If a real-time relatively short-time prediction is employed to form the primary CRDB entry for a particular real-time retrieval, one might expect the probability of the database being applicable to the case to be better than an attempt to find a similar case in a general “historic” database. Hence the deterministic character of many Mediterranean precipitation events will likely facilitate the use of real-time Cloud Radiation Prediction (CRP) models as the optimal methodology for creating a statistical database for retrieval. To carry this step further, an interactive use of satellite observations and the model assimilation process would likely further improve the representation of precipitation implied from space-borne observations. Increasing computing power is beginning to allow for the possibility of real-time ensemble simulation and Ensemble-Kalman Filter (Evensen 1994) (EnKf) data assimilation schemes. This technique will use the combined use of satellite and all other observations including other satellite, radar, radiosonde, ACARS, and others to determine the error covariance matrix of each ensemble member acting as a background field for a multivariate analysis. The effect will be to use the data to selectively nudge the straying ensemble members toward solutions consistent with all of the observations and preserve the solutions that are consistent with observations. The analysis then takes the form of an ensemble of analyses, the spread of which is related to the uncertainty in the analysis. The unique characteristics of the Mediterranean basin will likely produce a relatively narrow spread in a Mediterranean precipitation analysis compared to other regions of the world where the solution is not so highly forced by the local geography. In conclusion, the Mediterranean basin is an excellent laboratory for developing technologies to retrieve precipitation from space-borne observing platforms. It unique characteristics make storms more predictable and better able to be represented in CRMs. As a result, there is a greater ability to relate

Microphysical Signatures of Extreme Events in the Western Mediterranean

547

the space-borne observations to precipitation with the help of the models, than in other regions of the globe. Acknowledgments: The authors would like to thank Sabatino DiMichele of CNR (now at ECMWF) and Will Lewis of the UW for their contributions to this study. This work was supported under NASA grant PMM-0069-0153 and within the framework of EURAINSAT a shared-cost project (contract EVG1-2000-00030) co-funded by the Research DG of the European Commission within the RTD activities of a generic nature of the Environment and Sustainable Development sub-program (5th Framework Program). This study has been partially funded by the Italian Space Agency through “LAMPOS” project and within the framework of Community Initiative INTERREG IIIB CADSES - RISK AWARE project.

5

REFERENCES

Bosart, L. F., G. J. Hakim, K. R. Tyle, M. A. Bedrick, W. E. Bracken, M. J. Dickinson, and D. M. Schultz, 1996: Large-scale antecedent conditions associated with the 12–14 March 1993 cyclone (“Superstorm’93”) over Eastern North America. Mon. Wea. Rev., 124, 1865–1891. Carlson, T. N., 1991: Mid-Latitude Weather Systems. HarperCollins Academic, New York, 1991. Evensen, G., 1994: Sequential data assimilation with a nonlinear quasigeostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99 (C5), 10 143– 10 162. Golden, J. H., 1974: Scale-interaction implication for the waterspout life cycle II. J. Appl. Meteor., 13, 693–709. Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 808–816. Maddox, R. A., C. F. Chappell, and L. R. Hoxit, 1974: Synoptic and meso-α scale aspects of flash flood events. Bull. Amer. Meteor. Soc., 60, 115–123. Panegrossi, G., 2004: Learning from passive microwave measurements to improve microphysics parametrization in explicit cloud resolving models. PhD Thesis, University of Wisconsin, Madison, 267pp. Spencer, R. W., M. H. Goodman, and R. E. Hood, 1989: Precipitation retrieval over land and ocean with SSM/I. Part I: Identification and characteristics of the scattering signal. J. Atmos. Oceanic Technol., 6, 254–273. Tripoli, G. J., G. Panegrossi, A. Mugnai, S. Dietrich, and E. A. Smith, 2002: A numerical study of the Friuli, 1998 and Genoa, 1992 floods. Proceedings of the 2nd Plinius Conference on Mediterranean Storms, Siena, 127–140. Tsidulko, M. and P. Alpert, 2001: Synergism of upper-level potential vorticity and mountains in Genoa lee cyclogenesis: A numerical study. Meteor. Atmos. Phys., 78, 261–285. Tudurì, E. and C. Ramis, 1997: The environments of significant convective events in the western Mediterranean. Wea. Forecasting, 12, 294–306. Wilheit, T. T., 1986: Some comments on passive microwave measurements of rain. Bull. Amer. Meteor. Soc., 67, 1226–1232. Yano, J. and K. Emanuel, 1991: An improved model of the equatorial troposphere and its coupling with the stratosphere. J. Atmos. Sci., 48, 377–389.

43 ONLINE VISUALIZATION AND ANALYSIS: A NEW AVENUE TO USE SATELLITE DATA FOR WEATHER, CLIMATE, AND INTERDISCIPLINARY RESEARCH AND APPLICATIONS Zhong Liu1, Hualan Rui2, William L. Teng2, Long S. Chiu1, Gregory Leptoukh1 , and Gilberto A. Vicente1 GSFC Earth Sciences Data and Information Services Center, Distributed Active Archive Center, NASA/Goddard Space Flight Center, Greenbelt, MD, USA 1 CEOSR, George Mason University, Fairfax, VA, USA 2 SSAI, Lanham, MD, USA

Abstract

This article describes a new avenue to use satellite data for weather, climate and interdisciplinary research and applications: the TRMM Online Visualization and Analysis System (TOVAS). The system description, application examples, as well as future plans are given.

Keywords

TRMM, precipitation, rainfall, remote sensing, visualization, rainfall analysis, flood, drought, crop monitoring, crop forecast, satellite measurement, rainfall climatology, rainfall image, ASCII rainfall data, rainfall, rain rate

1

INTRODUCTION

Precipitation data have been widely used in weather, climate, and interdisciplinary research and applications. For example, a real-time or nearreal-time precipitation product can be used to monitor heavy rain events (Rui et al. 2003). Historical precipitation data and time series can be used to study historical events, seasonal-to-interannual variations, El Niño/Southern Oscillation (ENSO) events, etc. Precipitation products can also be used in interdisciplinary research and applications, such as, tropical infectious diseases (e.g., Anyamba et al. 2000; Zhou et al. 2002; Masuoka et al. 1998; 549 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 549–558. © United States Government 2007.

550

Z. Liu et al.

Liu et al. 2002a,b), drought and flood monitoring (Liu et al. 2002c), and crop yield estimates (e.g., the United Nations World Food Program), etc. However, in many tropical regions and parts of the mid-latitudes, precipitation estimates still remain a major challenge due to sparse rain gauges. To better develop research and applications for these regions, it is necessary to have rainfall data with adequate spatial and temporal resolutions. The Tropical Rainfall Measuring Mission (TRMM), as an important part of the NASA Earth Sciences Enterprise (ESE), is a joint US-Japanese satellite mission to monitor tropical and subtropical (40º S–40º N) precipitation and to estimate its associated latent heating. The TRMM satellite provides the first detailed and comprehensive data set on the four-dimensional distribution of rainfall and latent heating over vastly undersampled tropical and subtropical oceans and continents. The TRMM satellite was launched on November 27, 1997. Data from the TRMM satellite are archived and distributed by the NASA Goddard Space Flight Center Earth Sciences Distributed Active Archive Center (GES DAAC). TRMM data products and services can be found at the GES DAAC website (http://daac.gsfc.nasa.gov). Detailed information about TRMM can be found at the TRMM official website (http://trmm.gsfc.nasa.gov). Despite the relatively short history, TRMM rainfall products have been widely used in many areas. One of the goals of NASA ESE is to maximize the use of ESE data products. With over 6 years’ collection of TRMM data products, it is a challenging task to make them available to users at all levels. At present, there are only few Internet websites that offer data sets that have global coverage. Most of these sites provide data for downloading only and few provide browse images that are often either hard to read or often cannot be customized. To obtain precipitation information, such as, a time series, one often needs to go through these steps: (1) order the data product; (2) obtain the processing software; (3) install the software if they have the right equipment and operating system; (4) learn to use the software. Users can easily have problems in any one of these steps. In many cases, users will be very likely to find out that the product they order does not fit in their research and application requirements, therefore their time and resources have been wasted. In short, it is important, for all levels of users, to have a simple and easy-to-use system that allows everyone to access precipitation data products. To meet this requirement, the Hydrology Data Support Team (HDST) at the NASA GES DAAC initiated a project, the TRMM Online Visualization and Analysis System (TOVAS). TOVAS is an Internet-based system. It is a simple but powerful tool that enables users to concentrate on doing science. No other software or libraries are required. To display results, users simply select a product, an area, a parameter, a color option, a plot type, a time range, and an output type. With a web browser and a few mouse clicks, any user can easily obtain precipitation information from around the world. This paper will describe the system and its main functions, the data

Online Visualization and Analysis

551

products included in TOVAS, application examples, conclusions and future plans.

2

SYSTEM DESCRIPTION AND DATA PRODUCTS

TOVAS consists of three main components: input, processing and output. Figure 1 is a schematic of the system. The input component collects user’s selections from the web interface. Scripts are composed in the processing component to generate graphic or data output files. The last component sends the output files to the user’s browser. There are two web interfaces, Java and non-Java. The Java interface allows users to easily select an area of interest with a mouse. The non-Java interface allows users to use the system when Java and Javascript are disabled. The Grid Analysis and Display System (GrADS), developed by the Center for Ocean-Land-Atmosphere Studies (COLA), is used for the analysis. Standard analysis scripts are used and users can easily regenerate analysis results offline. Because of its simple design, TOVAS can be easily configured for new applications.

Figure 1. A schematic of TOVAS.

Main functions and features of TOVAS are: • Area plot – averaged or accumulated over any available data period for any rectangular area. • Time plot – time series averaged over any rectangular area. • Hovmöller plots – image view of any longitude-time and latitude-time cross sections. • ASCII output – for all plot types. • Image animation – for area plot. • Color options – for more customized images.

552

Z. Liu et al.

The TRMM satellite flies at an altitude of 402.5 km. The TRMM satellite carries three rain-measuring instruments. NASA Goddard Space Flight Center provided the TRMM Microwave Imager (TMI), the Visible Infrared Scanner (VIRS), and the observatory, and operates the TRMM satellite via the Tracking and Data Relay Satellite System (TDRSS). The Japan Aerospace Exploration Agency (JAXA) provided the Precipitation Radar (PR), the first space-borne precipitation radar, and launched the TRMM observatory. TRMM standard products at three levels are available at the GES DAAC. Level 1 products are the VIRS calibrated radiances, the TMI brightness temperatures, and the PR return power and reflectivity measurements. Level 2 products are derived geophysical parameters at the same resolution and location as those of the Level 1 source data. Level 3 products are the time-averaged parameters mapped onto a uniform spacetime grid. An evaluation of the sensor, algorithm performance and first major TRMM results appear in the Special Issue on the Tropical Rainfall Measuring Mission (TRMM), the combined publication of the Journal of Climate and Journal of Applied Meteorology (2000). The monthly distribution statistics collected at GES DAAC shows that most users prefer Level-3 products. At present, we only include Level-3 products in TOVAS (see Fig. 2). These Level-3 and other rainfall products are, 3B42RT (0.25 degree and 3-hourly), 3B42 (1 degree and daily), 3B43 (1 degree and monthly), the historical monthly rainfall (0.5 degree and monthly) provided by Cort J. Willmott and Kenji Matsuura from Center for Climatic Research Department of Geography University of Delaware (Willmott and Matsuura 1995), and the monthly precipitation product (1 degree and monthly) provided by the Global Precipitation Climatology Centre. All of the selected products provide global precipitation at different temporal and spatial scales. Since the launch of TOVAS, it has been used in a wide variety of earth science applications, such as weather events, climate, and interdisciplinary studies, agricultural crop monitoring, rainfall algorithm study, and data product comparison. Recent examples of the applications, collected by the GES HDST, are: • Study on coastal urban heat island effect on rainfall. • Additional rainfall information to supplement ground stations in Sri Lanka. • Phenology study in Africa and North America. • Crop yield estimates and flood watch in Africa and Asia. • Rainfall information for a development project in Afghanistan. • Fire monitoring activities in Africa. • Data for hydrological modeling in Africa. • Range prediction of American butterflies.

Online Visualization and Analysis

• • • •

553

Intercomparison with other products in North America. Monitoring rain events in Balkan. Investigation of the 1997–98 El Niño/La Niña event. Investigation of insect activities in the USA.

Figure 2. An example of the TOVAS interfaces for the TRMM Level-3 monthly product, 3B43.

Figure 3. Time series of the average rainfall in the affected region.

Z. Liu et al.

554

3

EXAMPLES OF WEATHER, CLIMATE AND INTERDISCINPLINARY APPLICATIONS

3.1 Morning heavy rain events The remnants of Tropical Cyclone Japhet, which moved into Mozambique on March 2, 2003, brought heavy rains to the region during the first week of March. Parts of Mozambique, Malawi and Zimbabwe experienced strong gusty winds and locally torrential rainfall, which produced areas of flooding. In Mozambique, over 50,000 people were affected, leaving thousands cut off from desperately needed food supplies according to the United Nations World Food Program (WFP). Figure 3 shows the time series of the 3-hourly precipitation over the affected region. The heaviest rainfall event can be easily identified. Figure 4 shows the spatial distribution of the heaviest rainfall event.

Figure 4. Rainfall map at 1800 Z March 6 2003. The 3-hourly near-real-time 3B42RT can be used to monitor heavy rain events, especially those over oceans where radar and gauge data are scarce.

3.2 Climate research and applications 3.2.1. Monitoring rainfall in the mid-Atlantic region With Hovmöller plot options, it is easy to obtain seasonal-to-interannual precipitation information around the world. TOVAS allows users to plot Hovmöller maps with a fixed latitude/longitude, or an averaging area with fixed latitudes/longitudes. Users could also adjust the color bar options to

Online Visualization and Analysis

555

make customized plots. Users could also obtain data in ASCII format for additional analyses and applications. With TOVAS, one could monitor and study both present and historical rainfall conditions around the world. Here is an example of the application in the Mid-Atlantic region of the USA. Figure 5 is a Hovmöller diagram for the 3B43 monthly rainfall product during the period between January 1998 and September 2003. The plot allows an easy comparison of seasonal-tointerannual variations. A period of drier months (marked by a white arrow) in 2002 can be easily identified from the plot. It was a severe drought reported in the region and the state emergencies were declared to save water. From the plot, a period of excessive rainfall following the drought (marked by a red arrow) can be identified as well.

Figure 5. Monthly 3B43 rainfall Hovmöller diagram for the Mid-Atlantic region of the USA.

Figure 6. Hovmöller diagram of rainfall along the Equator in Pacific Ocean.

3.2.2. Monitoring rainfall along the Equator Precipitation is a very important physical parameter in studies of El Niño/Southern Oscillation (ENSO) (Curtis and Adler 2000). With TOVAS,

556

Z. Liu et al.

it is easy to explore the whole data archive. Figure 6 is a Hovmöller diagram for the rainfall between January 1998 and November 2003. A large variation of rainfall is found during this time period.

3.3 Interdisciplinary applications TOVAS allows the HDST to provide many customized services to a wide variety of users. For example, recently the HDST has been providing precipitation information to the WFP for crop yield estimates and flood/drought assessment in Africa and Asia. Figures 7 and 8 are examples of many customized products that the HDST has been providing to WFP. Figure 7 is the rainfall accumulation estimate for January 1–10, 2003. Figure 8 is the maize yield projection based on the rainfall and an empirical equation.

Figure 7. January 1–10 (decade 1) 2003 rainfall accumulation estimate in mm (see also color plate 18).

Figure 8. Maize yield projection (% yield potential) based on estimated rainfall, and an empirical equation. White spaces denote zero yield potential (see also color plate 18).

Online Visualization and Analysis

4

557

CONCLUSIONS AND FUTURE PLANS

TOVAS provides a convenient way to access TRMM and other precipitation data products. TOVAS is a simple but powerful tool that enables users to concentrate on doing science, thus providing a new avenue for weather, climate and interdisciplinary research and applications. For example, before submitting a science proposal, one usually needs to do a lot of investigation work. The capability of generating countless analytical graphics and data allows users to fully explore the data products. Users from the climate community will find it particularly useful in obtaining information on seasonal-tointerannual variation. K-12 users can easily use TOVAS for their classroom projects because it does not require additional installations of software and libraries and the interface can be learnt in a few minutes. In short, TOVAS targets users at all levels. Once one learns how to use it, he or she will have nearly all the controls of the data products. TOVAS has demonstrated one of many great potentials of information technology. The merge of computation and communication technologies allows timely providing data and information and enabling users to concentrate on doing science, which will have a very profound impact on how we do science in the future. TOVAS is just a beginning of many similar ongoing prototypes in Earth science applications at the GES DAAC. A recently released MODIS Online Visualization and Analysis System (http://lake.nascom.nasa.gov/movas), developed jointly by the Aerocenter and GES DAAC at NASA Goddard Space Flight Center, is a good example to demonstrate how the TOVAS model can be applied to other disciplinary data products. Future plans for TOVAS will be concentrating on the following areas. The first to be added is intercomparison of precipitation products. The uncertainty issue in rainfall measurements is a well-known issue in research and application communities. Timely and easy access of this information will have a great impact on both research and applications. For example, errors in rainfall measurements can easily propagate to other precipitation derived products, such as, a crop yield estimates. At present, most users have to rely on referred publications where the works of investigators often either focus on global or a specific area. The conclusions often cannot be applied to the areas of their interests. With an online intercomparison system, users will be able to identify the uncertainty by intercomparing different products. In the future, related environmental variables, such as, terrain, measurements from satellites, will be added to the system to help identifying sources of errors. Several climatological precipitation products (baselines) will also be included in the system, which is very crucial for many monitoring activities. For the 3-hourly near-real-time 3B42RT product, the focus will be on enhancing and improving forecasting capabilities, such as, rainfall tendency, movement information of rain clusters. Users will be able to use the system

Z. Liu et al.

558

to monitor rain events. For the monthly products, new features, such as, anomaly and normalized (by the climatological mean) anomaly, will be added. The anomaly information will be very useful in monitoring drought events. New data products and systems will be added for interdisciplinary users. For example, a 10-day rainfall product is particularly useful for many agriculture users. Other remote-sensing products, such as, NASA QuikSCAT sea surface wind, TRMM Microwave Imager (TMI) sea surface temperature, etc. will be integrated into the system for air–sea interaction studies and applications.

5

REFERENCES

Anyamba, A., K. J. Linthicum, and R. L. Mahoney, 2000: Application of NDVI time series data to monitor Rift Valley fever outbreak patterns. Proceedings of the 25th Annual Climate Diagnostics and Prediction Workshop, Palisades, New York. Curtis, S. and R. F. Adler, 2000: ENSO indexes based on patterns of satellite-derived precipitation. J. Climate, 13, 2786–2793. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002a: TRMM rainfall products and tools for tropical infectious disease studies. 15th Conference on Biometerology/Aerobiology and 16th Congress of Biometeorology, Kansas City, MO. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002b: TRMM rainfall for human health and environment applications. International Tropical Rainfall Measuring Mission (TRMM) Science Conference, Honolulu, Hawaii. Liu, Z., L. Chiu, W. Teng, H. Rui, and G. Serafino, 2002c: TRMM rainfall data for ecosystem studies and applications in arid and semiarid regions. AGU Spring Meeting, Washington, DC. Liu, Z., L. Chiu, W. Teng, and G. Serafino, 2002: A simple online analysis tool for visualization of TRMM and other precipitation data sets. Science Data Processing Workshop 2002, Greenbelt, MD. Masuoka, P., J. Gonzalez, S. Gordon, N. Achee, P. Pachas, R. Andre, and L. Laughlin, 2002: Remote sensing and GIS studies of Bartonellosis in Peru. Poster Presentation at High Spatial Resolution Commercial Imagery Workshop, Reston, VA. Rui, H., W. Teng, Z. Liu, and L. Chiu, 2003: TRMM scientific data at the GES DAAC and their applications to monitoring tropical cyclones. IUGG2003 MC02/07P/D-002, Abstracts Week B page B.406 Special Issue on the Tropical Rainfall Measuring Mission (TRMM), combined publication of the December 2000 Journal of Climate and Part 1 of the December 2000 J. Appl. Meteor., AMS, Boston, MA. Willmott, C. J., and K. Matsuura, 1995: Smart interpolation of annually averaged air temperature in the United States. J. Appl. Meteor., 34, 2577–2586. Zhou, J., Lau, W. K., P. Masuoka, R. G. Andre, J. Chamberlin, P. Lawyer, and L. W. Laughlin, 2002. El Niño and the spread of Bartonellosis epidemics in Peru. EOS Trans., American Geophysical Union, 83(14), 157–161.

Section 8 The Present and Future of Satellite Platforms

44 THE SPACE-BASED COMPONENT OF THE WORLD WEATHER WATCH’S GLOBAL OBSERVING SYSTEM (GOS) Donald E. Hinsman1 and James F. W. Purdom2 1

World Meteorological Organization, Geneva, Switzerland2 Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, CO, USA Abstract

Satellite data are an exceptionally important component of the World Weather Watch’s (WWW) Global Observing System (GOS). Indeed, it would be difficult to find any WMO programme that does not use or depend on satellite data. During these first decades of the 21st century WMO members will continue to exploit operational meteorological satellite systems, while expanding utilization to include experimental satellites. During the next decades existing capabilities will be refined and improved, while new applications and advanced technology will migrate from the experimental realm into full operational use. This paper addresses the current space-based component of the GOS and takes a brief look to the future.

Keywords

WMO, climate, weather, satellite

1

INTRODUCTION

The space-based subsystem of the World Weather Watch’s (WWW) Global Observing System (GOS) is now (2004) comprised of three types of satellites: operational meteorological polar-orbiting and geostationary, and environmental research and development (R&D) satellites. With regard to meteorological satellites, both polar-orbiting and geostationary, they continue to prove invaluable to WMO National Hydrological and Meteorological Services (NMHS) through the provision of a multitude of services including imagery, soundings, data collection, and data distribution. In particular, the present operational meteorological satellites include the following geostationary and polar-orbiting satellites: GOES-10, GOES-12, NOAA-15, NOAA-16, and NOAA-17 operated by the USA; GMS-5 operated by Japan; GOMS N-1, METEOR 2-20, METEOR 2-21, METEOR 3-5, and METEOR 561 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 561–570. © 2007 Springer.

562

D. E. Hinsman and J .F. W. Purdom

3M N1 operated by the Russian Federation; Meteosat-5, Meteosat-6, Meteosat-7, and Meteosat-8 (formerly MSG-1) operated by EUMETSAT; and FY-2B, FY-1C, FY-1D operated by China. Additional satellites in orbit include GOES-8, GOES-9, and GOES-11 operated by the USA. It should be noted that most space agencies contributing operational polar-orbiting and geostationary satellites have in place contingency plans for satellite systems that guarantee the continued daily flow of satellite data, products and services WMO members have come to depend on. In this regard, Japan and the USA initiated a back-up operation of GMS-5 with GOES-9 on 22 May 2003. With regard to R&D satellites, NASA’s Aqua, Terra, NPP, TRMM, QuikSCAT, and GPM missions, ESA’s ENVISAT, ERS-1, and ERS-2 missions, NASDA’s ADEOS II and GCOM series, Rosaviakosmos’s research instruments on board ROSHYDROMET’s operational METEOR 3M N1 satellite, as well as on its future Ocean series and CNES’s JASON-1 and SPOT-5, either are, or will be after launch, part of the R&D constellation. The ability of geostationary satellites to provide a continuous view of weather systems make them invaluable in following the motion, development, and decay of such phenomena. Even such short-term events as severe thunderstorms, with a lifetime of only a few hours, can be successfully recognized in their early stages and appropriate warnings of the time and area of their maximum impact can be expeditiously provided to the general public. For this reason, its warning capability has been the primary justification for the geostationary spacecraft. Since 71% of the Earth’s surface is water and even the land areas have many regions which are sparsely inhabited, the polar-orbiting satellite system provides the data needed to compensate the deficiencies in conventional observing networks. Flying in a near-polar orbit, the spacecraft is able to acquire data from all parts of the globe in the course of a series of successive revolutions. For these reasons the polar-orbiting satellites are principally used to obtain: (a) daily global cloud cover; and (b) accurate quantitative measurements of surface temperature and of the vertical variation of temperature and water vapour in the atmosphere. There is a distinct advantage in receiving global data acquired by a single set of observing sensors. Satellite data are totally different in character from in situ data and have to be used in ways that reflect their characteristics. For example, numerical weather prediction (NWP) models now use satellite-measured radiances directly, instead of inverting satellite radiances into atmospheric temperatures. This direct insertion of radiances into models had a profound positive impact on NWP forecast accuracy. Some observations, such as vegetation indices, have no direct surface-based counterpart but have been found to be of great value. Often, many years of research are needed in order to use new forms of satellite data to best advantage. Indeed, new uses for cloud imagery are still being developed after four decades of routine use, taking advantage of the

Space-Based Component of the World Weather Watch’s GOS

563

improving spectral resolution of new generations of satellites and the vastly improved capabilities of computer systems. The thrust of the current generation of environmental satellites is aimed primarily at characterizing the kinematics and dynamics of the atmospheric circulation. The ability to achieve such objectives was demonstrated during the global weather experiment in 1979. This capability is now part of the global operations of the WWW. The existing network of environmental satellites, forming part of the GOS of the WWW, produces real-time weather information on a regular basis. This is acquired several times a day through direct broadcast from the meteorological satellites by more than 1,450 stations located in over 170 National Meteorological and Hydrological Services (NMHSs). Figure 1 shows the nominal configuration for the spacebased subsystem of the WWW’s GOS.

Figure 1. Nominal space-based component of the global observing system.

Note: Information on the characteristics, capabilities and uses of the current system of meteorological satellites is contained in the CGMS Directory of Meteorological Satellite Applications. Additional up-to-date information can be found via the WMO Space Programme Homepage: http://www.wmo.ch/hinsman/satsun.html.

564

D. E. Hinsman and J .F. W. Purdom

1.1 Current and future polar platforms Operational meteorological polar-orbiting satellites provide global coverage twice a day. Their orbital altitude at approximately 850 km makes it technically feasible to make high spatial resolution measurements of the atmosphere/surface. To provide a reasonable temporal sampling for many applications, the WMO requirement is for at least two satellites in the AM and two satellites in the PM orbits, thereby providing 3-hourly coverage. A backup capability for the polar orbit exists by reactivating “retired” platforms and this has been demonstrated recently. Since 1979, coverage with two polar-orbiting satellites has been achieved most of the time. They carry a multispectral imager (usually with 1 km resolution visible, nearinfrared (NIR), and infrared (IR) window bands for observing cloud cover and weather systems, deriving sea surface temperature, detecting urban heat islands and fires, and estimating vegetation indices), a multispectral IR sounder (usually with roughly 20 broad spectral bands of 20 km resolution for deriving global temperature and moisture soundings), and a multispectral microwave (MW) sounder (most recently with 20 MW channels of 50 km resolution for deriving temperature soundings even in nonprecipitating cloud-covered regions). Current operational meteorological polar orbiters include the NOAA series from the USA and the METEOR, RESURS, and OKEAN series from the Russian Federation and the FY-1 series from China. They provide image data that can be received locally. The NOAA satellites also enable generation of atmospheric sounding products that are disseminated to NWP centres on WMO’s Global Telecommunications System (GTS). In the future, the NOAA AM satellite will be replaced by the METOP satellites provided by EUMETSAT and the NOAA PM satellite will transition to the NPOESS series. The Russian Federation METEOR series will evolve into the METEOR 3M series and the Chinese FY-1 series will be replaced by the FY-3 series. R&D missions continue to make many contributions in the area of polarorbiting remote-sensing measurements. To maximize the impact of those data and the associated expenditures in resources (manpower and financial) by operational users, space agencies are committing to (a) open and timely access to the data in standard formats, (b) preparation of the community for new data usage, and (c) data continuity. NASA’s Earth Observing System includes multiple platforms. Terra has been in an AM orbit since December 1999 and is providing global data on the state of the atmosphere, land, and oceans, as well as their interactions with solar and earth radiation. Aqua followed in a PM orbit in May 2002 and will provide climate-related data with respect to clouds, precipitation, atmospheric temperature/moisture content, terrestrial snow, sea ice, and SST. Both provide X-band direct broadcast of their high resolution MODIS data which are being received by a

Space-Based Component of the World Weather Watch’s GOS

565

number of WMO Members. In addition, AIRS data from Aqua is being tested in several GDPSs across the globe. The NASA/CNES Topex/ Poseidon and Jason-1 satellites provide a wealth of observations on the status of the ocean topography, and this is used in both short-term storm analysis and climate studies. In 2004, Aura will provide a suite of chemistry measurements focusing on atmospheric trace gases in the upper troposphere and lower stratosphere. In addition, the Earth Observer series has started providing hyperspectral VIS/NIR data. ESA launched the ENVISAT platform in March 2002. It is designed to provide measurements of the atmosphere, ocean, land, and ice over a 5-year period. ENVISAT has an innovative payload that will ensure the continuity of the data measurements of the ESA Earth Remote-Sensing (ERS) satellites, as well as facilitating the development of operational applications. Thereafter, several Earth Explorer missions are planned to study the gravity field (2005), atmospheric dynamics (2007), polar ice (2004), and soil moisture and salinity (2006). NASA and JAXA launched in 1997 a joint mission, the Tropical Rainfall Measuring Mission (TRMM). TRMM is designed to monitor and study tropical rainfall and the associated release of energy that helps to power the global atmospheric circulation shaping both weather and climate around the globe. It also provides information on soil moisture. By combining TRMM precipitation data with ocean vector winds data from QuikSCAT (launched by the USA in 1999), researchers have demonstrated the ability to significantly improve hurricane track and landfall prediction. JAXA intends to launch the Advanced Land Observing Satellite (ALOS) in 2004 that will utilize advanced land-observing technology. Later this decade, the Global Change Observation Mission (GCOM) will be aimed at observing parameters over the long term (as long as 15 years), and to understand the mechanism of the global environmental change. GCOM-A1 will observe ozone and greenhouse gases and GCOM-B1 will monitor energy and the general circulation from a sunsynchronous orbit. The People’s Republic of China is providing the newest series of polar-orbiting satellites, the FY-1 series. The FY-1 series has greatly enhanced imaging capabilities from polar orbit with its 10 channel radiometer that includes the same five channels as found on NOAA’s AVHRR and five new channels.

1.2 Current and future geostationary platforms The geosynchronous orbit is over 40 times higher than a polar orbit, which makes measurements technically more difficult from geostationary platforms. The advantage of the geostationary orbit is that it allows frequent measurements over the same region necessary for now-casting applications and synoptic meteorology. Weather, and weather-related phenomena extend across a broad range of scales. In meteorology the link between the synoptic

566

D. E. Hinsman and J .F. W. Purdom

scale and the mesoscale is many times a key factor in controlling the intensity of local weather. The only observing tool capable of monitoring weather across those scales (and those scales interactions) is the geostationary satellite. Nowcasting is in many aspects mesoscale in nature, and it is here that satellite data can provide great benefit. While observations from polarorbiting satellites often detect mesoscale phenomena at the needed spatial resolutions, they lack the temporal resolution required for many now-casting applications. For example, sounder data from polar-orbiting satellites, particularly HIRS data, provide high spatial resolution data that can be used to derive information on atmospheric instability over land in cloud free areas generally once in every 6 h. Those data are valuable for assessing the broadscale features such as axes of deeper moisture and instability that may support convection, however, their temporal resolution is not optimal for convective nowcasting. Complementing the polar-orbiting sounder data, the USA’ GOES geostationary sounder provides high spatial resolution hourby-hour information of the atmosphere’s ability to support (and inhibit) deep convection. It is important to recognize that prior to geostationary satellites, the mesoscale was a “data sparse” region; meteorologists were forced to make inferences about mesoscale phenomena from macroscale observations. Today’s geostationary satellites provide multispectral high-resolution imagery at frequent intervals. Those data reveal meso-meteorological features that are infrequently detected by fixed observing sites. The clouds and cloud patterns in a satellite image provide a visualization of mesoscale meteorological processes. When cloud imagery (and products derived from sounding data such as lifted index) is viewed in animation, the movement, orientation, and development of important mesoscale features can be observed. Furthermore, such animation provides observations of convective behaviour at temporal and spatial resolutions approaching the scale of the mechanisms responsible for triggering deep and intense convective storms. From geostationary altitude, a fixed full disk view of the Earth is viewed from one satellite, thus, at least six equally spaced satellites around the equator are needed to provide global coverage; polar regions are either very poorly observed, or not observed at all because of the large zenith viewing angles. Currently, there is global coverage from geostationary orbit (more than six operational satellites for image data and products (e.g., cloud motion winds) and two satellites are also providing a sounding capability. Reactivating “retired” platforms provides backup and there have been several examples of this type of activity. Additionally, operational satellite agencies have developed the “help your neighbour” concept whereby adjacent agencies seek to provide continuity of data and services through regional contingency planning. The geoimagers typically have 1 km resolution visible and 5 km IR window bands for observing cloud cover and weather systems in motion and estimating atmospheric motion vectors. The

Space-Based Component of the World Weather Watch’s GOS

567

geosounders to date have 18 broad spectral bands of 10 km resolution for deriving atmospheric temperature and moisture trends in time. Some of the satellites provide a real-time transmission capability to allow immediate access to the imagery for real-time applications. Products are disseminated on the WWW’s GTS by the satellite operators for near realtime applications. At present there are satellites at 0° longitude and 63°E (Meteosat 8 and 5 operated EUMETSAT), 76°E (GOMS N1 operated by the Russian Federation), 105°E (FY2B operated by the People’s Republic of China), 140°E (GMS-5 operated by Japan and backed up by GOES-9), and 135°W and 75°W (GOES 10 and 12 operated by the USA. All geostationary satellite instruments are evolving to more spectral coverage and faster imaging. In 2004, EUMETSAT introduced METEOSAT Second Generation (MSG) to the operational suite of geostationary satellites, making high-resolution 12-channel imagery of the earth’s disc available every 15 min with the advanced imager SEVIRI. Japan will launch JAMI in 2004. China will launch another in the FY-2 series of imagers in 2004. The USA will evolve to an Advanced Baseline Imager and Sounder in 2012. MSG also carries the geostationary earth radiation budget (GERB) instrument, the first full spectrum earth radiation budget measuring device to fly in geostationary orbit.

2

ENHANCED MONITORING OF SELECTED EARTH SYSTEM COMPONENTS

New observing capabilities demonstrated in a research mode during the next decade will become part of an operational observing system of the future. New capabilities relevant to WMO member needs to include the areas addressed below.

2.1 Atmospheric sounding Continuous observation of tropospheric temperature/moisture profiles, wind pattern, and moisture inflow in the far field around weather systems, where the cloud cover is broken, are currently being demonstrated in polar orbit with NASA’s AIRS instrument. Operational polar-orbiting systems that will follow AIRS with high spectral resolution interferometers are IASA on METOP and CrIS on NPOESS. Currently, the only geostationary programme planning for a high spectral resolution sounding instrument is the USA with the introduction of its GOES-R series. It is anticipated that the very high spectral resolution data from GOES-R’s sounder will demonstrate a unique ability to peer continuously through many layers of the atmosphere from geosynchronous orbit with the precision and accuracy of atmospheric

568

D. E. Hinsman and J .F. W. Purdom

sounding capabilities being developed for low earth orbit systems (i.e., 1 K accuracy and 1 km vertical resolution). The instrument is intended to map the three-dimensional (3D) distribution of water vapour at different altitudes in the atmosphere and determine atmospheric temperature profiles for use in nowcasting.

2.2 Atmospheric winds Global wind field measurements will be directly applicable to NWP, and extremely valuable for scientific diagnosis of large-scale atmospheric transport, weather systems, and boundary layer dynamics. A space-based Doppler lidar system is being developed by ESA to deliver observations of tropospheric and stratospheric wind data; progress in space-borne laser technology will continue in order to make this active sounding technique available to operational uses.

2.3 Global precipitation Quantitative measurement of the time and space distribution of global precipitation is the next highest climate research priority beyond atmospheric temperature and moisture, and an essential requirement to understand the coupling among atmospheric climate, terrestrial ecosystems, and water resources. Satellite remote sensing is the only means to acquire global rainfall data, considering the paucity of surface observations over the ocean and sparsely populated land areas. Measurement of global precipitation would likely be based on frequent observations from a constellation of passive MW sensors with detailed vertical atmospheric distribution of rain data provided by a common rain radar satellite for refinement and validation of retrieval algorithms for all satellites in the constellation. An early demonstration of this concept is being conducted using the TRMM, Aqua, and ADEOS II research satellites in tandem with operational meteorological satellites. This extends TRMM-like precipitation measurements to extratropical parts of the world for the first time, and demonstrates the concept of 3-h global precipitation products with utility to a broad range of WMO Members’ objectives.

2.4 Soil water content At present, near-surface soil water content is the only primary hydrologic variable that is not measured at large spatial scale. Scientific evidence shows that near-surface soil water content is the most significant indicator of the state of the terrestrial hydrologic system, and is the governing parameter for partitioning rainwater among evaporation, infiltration, and run-off. Large antennae will be needed to meet these requirements at low MW frequencies; these remain a significant technological challenge.

Space-Based Component of the World Weather Watch’s GOS

569

2.5 Cold climate products Passive and active (radar) MW remote-sensing methods are being considered by Europe, the USA, and Japan to determine the most effective means to acquire information globally on snow extent and water equivalent, soil freezing, and thawing that strongly affect the hydrologic regime of river basins at high latitudes.

2.6 Vertical profiles of clouds and aerosols Atmospheric aerosol content is subject to substantial variation in amount and type, as concentrations are driven by natural and human activities, including agricultural and industrial practices. In the first half of this decade, the first attempt at global observation of the 3D structure of clouds and aerosol distributions will be undertaken. These involve active remote-sensing systems (i.e., lasers/lidars and MW radars) rather than the passive remotesensing systems such as radiometers that are common today. Due to the long-term nature of climate change research, such systems are likely candidates to become part of the operational climate-observing system in the future.

3

CONCLUSIONS

During the next several years the polar-orbiting component of the operational environmental satellite system will evolve to include at least six satellites in polar sun-synchronous orbits. Those satellites will carry highresolution multispectral imagers, IR interferometer sounders, advanced MW imagers and sounders, improved ozone monitoring instruments, and radio occultation (GPS) sensors. As in the early days of satellite meteorology, polar-orbiting satellites are again providing valuable information on which we will build the geostationary satellite systems of the future. In the geostationary arena, plans are for similar coverage as today with satellites operated by China, India, EUMETSAT, the Russian Federation, and the USA. The major step forward in the geostationary arena is that many operators are planning for with satellites with greatly improved spectral and temporal coverage for imaging, and the USA is planning to move to hyperspectral IR sounding. Data from research satellite instruments such as NASA’s Moderate Resolution Imaging Spectro-radiometer and AIRS, NASDA’s IMG, NASA’s Earth Observer-1 with the hyperspectral Hyperion instrument, and ESA’s Medium Resolution Imaging Spectrometer are clearly showing that innovative hyperspectral observing is the future for satellite imaging and sounding. Planning is underway for an important follow-on mission based on the success of TRMM. That mission, known as the Global Precipitation Measurement (GPM) mission, will allow for the derivation of

570

D. E. Hinsman and J .F. W. Purdom

global precipitation at least eight times a day (more frequent in polar regions due to multiple area coverage from the sun-synchronous satellite observations). Satellite data provides a user with powerful information that can be used to aid in a variety of WMO programmes. For some applications, optimal resolutions may not be attainable from any one satellite but may be approached using data from a series of satellites. The future of space-based remote sensing and its applicability to WMO member needs looks bright. The move towards an improved space-based component of the GOS, building on current capabilities should provide a firm foundation of observations upon which skill in many areas will advance. Acknowledgements: The space-based component of the GOS is a worldwide cooperative effort involving both research and operational satellite operators. It is through their dedication and hard work that this cooperative effort is being brought to fruition.

45 THE METEOSAT AND EPS/METOP SATELLITE SERIES Johannes Schmetz, Dieter Klaes, Alain Ratier, and Rolf Stuhlmann EUMETSAT, Darmstadt, Germany Abstract

The paper provides short overviews of the European meteorological satellite programmes: (1) the first generation of European geostationary Meteosat satellites, (2) Meteosat Second Generation (MSG), and (3) the future EUMETSAT Polar System (EPS). The features of the MSG satellites are presented along with examples of novel observations of cloud and atmospheric instability. Four MSG-type satellites will serve the community for the next decade and a half. Finally, the future EPS/Metop satellites jointly developed with the European Space Agency (ESA) are introduced. The first Metop satellite is scheduled for launch in 2005. EPS/Metop provides advanced observations of temperature and humidity profiles, wind, ozone, and trace gases.

Keywords

Meteosat, MSG, SEVIRI, GERB, EUMETSAT

1

INTRODUCTION

Until 1960, our knowledge of the present weather situation around the world had been almost entirely provided by ground-based observational systems. But this changed dramatically with the launch of the first US meteorological satellite from Cape Canaveral, Florida, on 1 April 1960. This experimental satellite, TIROS (Television and Infrared Observation Satellite) provided for the first time regularly pictures of the Earth’s weather systems over large areas. Europe started to contribute to the global space-borne observing system with the launch of Meteosat-1 on 23 November 1977; this satellite was the first in geostationary orbit that carried a water vapour (WV) channel in the 6.3 µm band. Today we look back to more than 40 years of meteorological satellites that have proven to be the best way to observe the weather on a large scale. Typically, operational meteorology utilizes two types of satellites to provide the required information: 571 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 571–586. © 2007 Springer.

J. Schmetz et al.

572

Low Earth-Orbiting (LEO) satellites fly at relatively low altitudes of around 800 km above the Earth, mostly in polar (sun-synchronous) orbits, and can provide information with high spatial resolution. The whole surface of the Earth can be observed twice a day. More than one polar satellite with different equatorial crossing times is required in order to attain more frequent observations. Geostationary satellites (GEO) flying in the equatorial plane at an altitude of about 36,000 km above the Earth have the same revolution time as the Earth itself and therefore always view the same area. They can perform frequent imaging, which, in animated mode, depicts the everchanging atmospheric processes. The disadvantage of the high altitude is that it limits spatial resolution and precludes the use of active instruments like radars. With the experience of seven geostationary Meteosat satellites of the first generation, the successful start of operations of Meteosat-8 (a new satellite series) and the EUMETSAT Polar System (EPS), with its Metop satellites, close to the first launch, Europe contributes significantly to the space-based global observing system, through the combined efforts of EUMETSAT, the European Space Agency (ESA), Centre National d’Etudes Spatiales, and other partners.

2

THE FIRST-GENERATION METEOSAT SATELLITES

In 1972 the Meteosat programme, originally a proposal from the French space and meteorological authorities, became “Europeanised” through legal arrangements between the European Space Research Organisation (ESRO), later to become the ESA, and the Centre National d’Etudes Spatiales (CNES). Only 5 years later the efforts culminated in the launch of the first satellite on 23 November 1977. The first image from Meteosat-1 was successfully acquired on 9 December 1977 (see Fig. 1) and the satellite continued to operate nominally until 24 November 1979, when an on-board electronic component failure resulted in the loss of all missions except that supporting the data collection systems (DCS). The launch of the second preoperational satellite, Meteosat-2, which had already been approved in 1977, took place on 19 June 1981. The undoubted success of the preoperational programme promoted the development of a follow-on operational programme. The Meteosat Operational Programme (MOP) began in 1983 and was executed by ESA, on behalf of EUMETSAT from 1986 onwards. As the construction of three new satellites took some years and Meteosat2 had a design lifetime of 3 years only, a decision was taken to refurbish the prototype satellite, Meteosat-P2, to a flight standard for launch on the first test flight of the Ariane-4 launcher. This satellite would then be capable of

The Meteosat and EPS/Metop Satellite Series

573

filling any potential gap between Meteosat-2 and the first of the new MOP satellites. Meteosat-3 (renamed from Meteosat-P2 after launch) was eventually launched in June 1988 and took over the operational missions during the following August. It was fortuitous that Meteosat-2 had been able to fulfil its prime functions of image taking and dissemination over 7 years after launch. After serving as an in-orbit spare for a further 3 years Meteosat2 was finally re-orbited, to an altitude over 300 km above geostationary orbit, in early December 1991. With the successful launch of Meteosat-4 on 6 March 1989 the MOP was underway. Then Meteosat-3 supported for nearly 4 years two important missions in support of the National Oceanic and Atmospheric Administration (NOAA). From 1 August 1991 through 27 January 1993 Meteosat-3 provided operational images of the Atlantic basin from a position at 50°W, under the called the “Atlantic Data Coverage (ADC)” mission (de Waard et al. 1992). The satellite then was moved to 75°W and provided an operational imaging service from 21 February 1993 through 1 May 1995, called the “Extended Atlantic Data Coverage (XADC)”. The MOP programme comprised two more satellites, both of which are, at this point in time, still in operational use. Meteosat-5 operates over the Indian Ocean at 63°E since July 1998 when operation began as a support to the Indian Ocean Experiment (INDOEX). A year later this support to INDOEX continued as a routine Indian Ocean Data Coverage (IODC) mission. Meteosat-6, the backup to Meteosat-7, provides so-called “rapid scan” imagery with a repeat cycle of 10 min observing Europe. Meteosat-7, launched in September 1997, is the last first-generation Meteosat. It operates in the nominal European orbit at 0° longitude.

2.1 Earth imaging with the first generation of Meteosat The main instrument on-board the satellite is a multispectral radiometer, which provides image data in three spectral bands (e.g., Mason and Schmetz 1992): • • •

0.5–0.9 µm visible band, 5.7–7.1 µm infrared (IR) WV absorption band, 10.5–12.5 µm thermal IR window band.

Images of the full Earth disk are taken in the three spectral bands every half hour. The radiometer scans the Earth from east to west by virtue of the spinning motion (100 rpm) of the satellite whilst the south to north scanning is achieved by stepping the radiometer through a small angle (1.25 × 10–4 rad) at each rotation of the satellite. There are two visible detectors (VIS1 and VIS2), that are placed in the focal plane of the primary telescope such that they observe adjacent lines of the Earth. Thus, by combining each

J. Schmetz et al.

574

individual visible image, consisting of 2,500 lines each of 5,000 pixels, it is possible to obtain a visible image of 5,000 lines of 5,000 pixels with a sampling distance of 2.5 km at the subsatellite point (SSP). The SSP sampling distance of the IR and WV images is 5 km, each image consisting of 2,500 lines of 2,500 pixels. On the preoperational satellites, i.e., Meteosat-1 to Meteosat-3, the idea to include a WV channel occurred at a late stage in the development of the radiometer and was made possible by using the electronic chain of one of the visible channels. This meant that with the first three satellites two modes of operation were possible: Either mode 1 including VIS2, WV, and IR or mode 2 with VIS1, VIS2, and IR. In practice during daylight hours the two modes were used alternately for successive slots whilst at night mode 1 was used exclusively. The VIS and WV images had a digitization of only 6 bits. With the start of Meteosat-4 (first of MOP series) full resolution VIS and WV images could be operated simultaneously and all image data were digitized with 8 bits. Noteworthy is also an improvement to the noise of the WV channel for Meteosat-4 to Meteosat-7, which was a significant step towards the successful operational derivation of atmospheric motion vectors (AMVs) in clear-sky areas from WV images (Laurent 1993). The clear-sky WV winds complemented the AMVs derived from the tracking of cloud features (Schmetz et al. 1993). Both became important in the assimilation of data for numerical weather prediction models. More recent advances in the AMV products addressed the quality control (Holmlund 1998).

3

METEOSAT SECOND GENERATION

As the first Meteosat series (i.e., Meteosat-1 through Meteosat-7), MSG (now Meteosat-8) and its successors are spin-stabilized, however, capable of greatly enhanced Earth observations. The satellite’s 12-channel imager, known as the spinning-enhanced visible and infrared imager (SEVIRI), observes the full disk of the Earth with an unprecedented repeat cycle of 15 min in 12 spectral wavelength regions or channels. The MSG programme covers a series of four identical satellites, expected to provide observations and services over at least 15 years. Each satellite has an expected lifetime of 7 years. As with the first Meteosat system, the new generation, starting with Meteosat-8, is planned as a dual-satellite service, where one additional satellite is available in orbit. As MSG is a new series of satellites (see Fig. 1), the period for commissioning of Meteosat-8 was longer than the typical 6-month period

The Meteosat and EPS/Metop Satellite Series

575

Figure 1. The first MSG satellite was launched in August 2002; it was renamed to Meteosat-8 with the start of operational services in January 2004.

and the operational service started in January 2004. A second MSG satellite, named Meteosat-9, was launched on December 2005.

3.1 Earth imaging with the second generation of Meteosat The primary mission of the second-generation Meteosat satellites is the continuous observation of the Earth’s full disk with a multispectral imager. The repeat cycle of 15 min for full-disk imaging provides multispectral observations of rapidly changing phenomena such as deep convection. It also provides for better retrieval of wind fields, which are obtained from the tracking of clouds, WV, and ozone features. The imaging is performed by utilizing the combination of satellite spin and scan mirror rotation, a process known as stepping. The images are taken from south to north and east to west. The eight thermal IR and three solar channels have a sampling distance of 3 km at nadir and scan the full disk of the Earth. The high-resolution visible channel provides images with one kilometre sampling at nadir. Data rate limitations confine the high-resolution visible images to half the Earth in an east–west direction; however, the exact coverage of the Earth is programmable.

J. Schmetz et al.

576

Table 1. Spectral channel characteristics of SEVIRI in terms of central, minimum, and maximum wavelength of the channels and the main application areas of each channel. Channel No.

Spectral band (µm)

Characteristics of spectral band (µm)

λcen

λmin

λmax

1

VIS0.6

0.635

0.56

0.71

2

VIS0.8

0.81

0.74

0.88

3

NIR1.6

1.64

1.50

1.78

4

IR3.9

3.90

3.48

4.36

5

WV6.2

6.25

5.35

7.15

6

WV7.3

7.35

6.85

7.85

7

IR8.7

8.70

8.30

9.1

8

IR9.7

9.66

9.38

9.94

9

IR10.8

10.80

9.80

11.80

10

IR12.0

12.00

11.00

13.00

11

IR13.4

13.40

12.40

14.40

12

HRV

Broadband (about 0.4–1.1 µm)

Main observational application

Surface, clouds, wind fields Surface, clouds, wind fields Surface, cloud phase Surface, clouds, wind fields Water vapour, high level clouds, atmospheric instability Water vapour, atmospheric instability Surface, clouds, atmospheric instability Ozone Surface, clouds, wind fields, atmospheric instability Surface, clouds, atmospheric instability Cirrus cloud height, atmospheric instability Surface, height of thin clouds, atmospheric instability

SEVIRI has eight spectral channels in the thermal IR, three channels in the solar spectrum, and a broadband high-resolution visible channel. The accompanying table provides more details of the characteristics of these channels, and indicates how each channel is used: for observations of clouds and surface temperatures, WV or ozone. Figure 2a and b show the location of the SEVIRI bands on top of a solar and typical thermal energy spectrum, respectively. Figure 3a and b give examples of the weighting functions of the thermal channels for a tropical standard atmosphere and nadir view (3a) and a subarctic winter atmosphere (3b) and a satellite viewing angle of 60°. The MSG level 1.5 data have a 10-bit digitization and provide the basis for all further processing and for the derivation of meteorological products. Concerning radiometric performance Meteosat-8 exceeds the specification by far; Fig. 4 shows a comparison of the specified radiometric noise in terms

The Meteosat and EPS/Metop Satellite Series

577

of NEDT at a reference temperature for the thermal IR channels with in-orbit performance measurements after Aminou et al. (2003). It should be noted that the specified NEDT refers to end of life, while the in-orbit measurements refer to beginning of the satellite’s lifetime. SEVIRI SOLAR CHANNELS

Standard Mid-Latitude Summer Nadir

2000 300

1500 0.60 LEAF REFLECTANCE

1000 0.40

SOIL REFLECTANCE

500

0.20

EBBT [K]

2

0.80

IRRADIANCE (W/ m )

REFLECTANCE / TRANSMITTANCE

1.00

250

TOA IRRADIANCE

0.00

0 0.5

1.0 WAVELENGTH (mm)

(a)

1.5

200

5

10

15

WAVELENGTH (mm)

(b)

Figure 2. Left: (a) MSG SEVIRI spectral response functions for the solar channels plotted with the spectral reflectance of vegetation, bare soil and the spectral irradiance at the top of the atmosphere. Right: (b) thermal terrestrial spectrum in terms of equivalent blackbody brightness temperature (EBBT) for a mid-latitude summer atmosphere and nadir view and MSG SEVIRI spectral response functions for the thermal infrared (IR) channels.

Figure 3. Left: (a) weighting functions for the MSG SEVIRI thermal channels, i.e., channels 4–11, for a satellite nadir view. A tropical summer standard atmosphere has been assumed for the simulation with a radiative transfer model. Right: (b) same as Fig. 3a except for a subarctic winter atmosphere and a viewing angle of 60°.

As an additional scientific payload, MSG carries a geostationary Earth radiation budget (GERB) instrument that observes the broadband thermal IR and solar radiances exiting the Earth’s atmosphere. The GERB instrument makes accurate measurements of the shortwave and longwave components

J. Schmetz et al.

578

of the radiation budget at the top of the atmosphere. GERB data are of interest to climatological studies and also for comparison with and validation of weather forecast models. As the first radiation budget experiment from geostationary orbit, the GERB instrument has great potential to shed new light on climatic processes related to clouds and WV. In particular, simultaneous observations with SEVIRI and GERB will reveal unknown physical elements of the process of deep convection (e.g., in the tropics) and its influence on the radiation budget. MSG Noise Specifications and In-Flight Measurements 2.00 1.80 1.60

NEDT (K)

1.40 1.20 1.00 0.80 0.60 0.40 0.20 0.00 3.9

6.2

7.3

8.7

9.6

10.8

12

13.4

Channe l Ce ntral Wav e le ngth (µm)

Specification

In-Flight Measurement

Figure 4. Noise equivalent temperatures of cold spectral channels of MSG (i.e., Meteosat-8). The reference temperatures for the NEDT are: 300 K at 3.9 µm; 250 K at 6.2 µm; 250 K at 7.3 µm; 300 K at 8.7 µm; 255 K at 9.7µm; 300 K at 10.5 µm; 300 K at 12 µm; 270 K at 13.4 µm. (From Aminou et al. 2003.) Note that specified NEDT refers to end of life, measured values to beginning of life.

3.2 Products from Meteosat Second Generation Continuity for meteorological products from the first-generation Meteosats is provided through core products centrally derived at EUMETSAT. Those meteorological products include atmospheric motion vectors, cloud analysis, and atmospheric humidity. In addition, there are novel products such as atmospheric instability and total ozone over the entire MSG field of view (FOV) (for details see Schmetz et al. 2002). Over and above the central processing at EUMETSAT, there are products from a geographically distribution network of services, Satellite Application Facilities (SAF), hosted by National Meteorological Services and other institutions. The idea behind the network of SAFs is that more products from MSG (and also from the future EPS) can be derived effectively capitalizing on the scientific expertise across the EUMETSAT member states. In January

The Meteosat and EPS/Metop Satellite Series

579

2004, seven SAFs exist addressing seven themes: (1) ocean and sea ice hosted by Météo-France, (2) numerical weather prediction hosted by the MetOffice, UK, (3) nowcasting hosted by the Institute for Meteorology, Spain; (4) GRAS Meteorology hosted by the Danish Meteorological Institute, (5) land surface analysis hosted by the Portuguese Meteorological Institute, (6) climate monitoring hosted by the German Met Service (DWD), (7) ozone monitoring hosted by the Finnish Meteorological Institute, and (8) support to operational hydrology and water management hosted by the Italian Meteorological Service (UGM). Each SAF provides operational services to end-users, including real-time and/or off-line product services, distribution of user software packages, and data management. The application of the second generation of Meteosat satellites ranges from short-term forecasting to numerical weather prediction and climatological studies. The most important products for numerical weather prediction are the spectral radiances themselves and the wind fields derived from tracking the displacement of clouds and WV features in successive satellite images. Both winds and radiances are assimilated into the numerical models that compute the change of the atmosphere in the future and provide the basis for weather forecasts.

Figure 5. B/W version the RGB full disk-image from Meteosat-8 observed on 5 June 2003. Channel 3 (at 1.6 µm); channel 2 (at 0.81 µm), and channel 1 (at 0.635 µm) were used. Ice clouds appear bright.

Figure 5 shows an RGB image from Meteosat-8 derived centrally at EUMETSAT, where channels 3, 2, and 1 have been assigned to the colours red, green, and blue, respectively. As ice clouds strongly absorb at 1.6 µm (channel 3) those ice clouds appear blue in Fig. 5. Another useful product is the global instability index, specifically the lifted index (König 2002).

J. Schmetz et al.

580

4

THE EUMETSAT POLAR SYSTEM/METOP

4.1 Overview The EPS is the European contribution to the joint European/US operational polar satellite system. The European part covers the mid-morning (AM) orbit, whereas the USA continues to cover the afternoon (PM) orbit. In the framework of the EPS programme (Bühler et al. 2001; Klaes et al. 2001, 2007) a space segment with associated launch services, a full ground segment are developed, expected to provide 14 years of operations. The space segment comprises three Metop satellites, which are developed in cooperation by the EUMETSAT and the ESA, the French Centre National d’Etudes Spatiales (CNES). The Metop satellite will also carry instruments provided by the National Oceanic and Atmospheric Administration (NOAA). The first Metop satellite has been launched on 19 October 2006. The first Metop spacecraft is being developed under the ESA Metop-1 Programme. This includes the development of some payload components as the Global Ozone Monitoring Experiment (GOME-2), Advanced Scatterometer (ASCAT) and the GPS Radio Occultation Sounder (GRAS). Furthermore an Advanced Very High Resolution Radiometer (AVHRR) and the Advanced TIROS Operational Sounder (ATOVS) package, comprised of High Resolution Infrared Radiation Sounder (HIRS-4), Advanced Microwave Sounding Unit (AMSU-A), and Microwave Humidity Sounder (MHS) are components of the Metop payload. MHS, which is a EUMETSAT development, replaces the AMSU-B instrument in the ATOVS suite, while NOAA provides the ATOVS and AVHRR instruments. The IASI instrument, developed by CNES, provides advanced sounding capabilities in the IR. Figure 6 provides a view of the Metop satellite and its payload.

Figure 6. Metop spacecraft of the EUMETSAT Polar System with instruments.

The Meteosat and EPS/Metop Satellite Series

581

4.2 ATOVS: Continuity Continuity is an important aspect to operational services. Today the information on temperature and humidity soundings and surface information (including clouds) for numerical weather prediction and other applications, is provided by the ATOVS suite supported by the AVHRR imager for both morning and afternoon missions (currently on the NOAA-16 (PM) and NOAA-17 (AM) satellites). The instrument suite is common to the two components of the Initial Joint Polar System (IJPS), i.e., Metop-1 and Metop-2 (AM satellites) and NOAA-N and NOAA-N’ (PM satellites) satellites.

4.3 IASI: New technology The Infrared Atmospheric Sounding Interferometer (IASI) introduces new technology into EPS. The purpose of IASI to measure temperature, WV and trace gases at a global scale. The measurement principle is Michelson interferometry providing 8,461 spectral channels, aligned in three bands between 3.62 µm (2,760 cm–1) and 15.5 µm (645 cm–1). The unapodised resolution is between 0.3 and 0.4 cm–1, with a spectral sampling of 0.25 cm–1. Included into the instrument is an integrated imaging system (IIS), consisting of a radiometer measuring between 10 and 12 µm with high spatial resolution. The FOV covers 64 × 64 pixels and provides information in the focal plane of the sounder, allowing to co-register with AVHRR, enabling an accurate navigation and also a detailed analysis of cloud properties inside the IASI sounder pixels. Figure 7 shows a typical simulated IASI spectrum for a clear and cloudy situation after Rizzi (1998).

Figure 7. IASI spectrum simulated for a clear and cloudy atmosphere. (From Rizzi 1999.) Gray = clear sky, black = cloudy sky.

J. Schmetz et al.

582

l Figure 8. Retrieval errors from simulated IASI retrievals for temperature (left) and specific humidity (right). (From Schlüssel and Goldberg 2000.) Figures are plots of RMS errors in temperature and humidity profiles, all angles, noise-averaged over IASI IFOV-quadruples, sea surface, no clouds.

The IASI data will be used in synergy with the microwave sounding instruments, to which the scan is synchronized. Products will include, besides level 1 spectra, vertical profiles of temperature, humidity, and ozone at global scale. Figure 8 depicts the simulated retrieval error for temperature and humidity after Schlüssel and Goldberg (2002) indicating that an accuracy for temperature of 1 K per 1 km layer can be achieved throughout most of the troposphere and lower stratosphere.

4.4 ASCAT and GOME-2: Operational use of research missions There are two instruments within EPS, which are in heritage of missions on the ESA Earth Remote Sensing (ERS) Satellites: The ASCAT and the GOME-2. 4.4.1 Advanced Scatterometer ASCAT is a real-aperture, polarised C-band radar in heritage from the AMI/Wind mode mission instrument on ERS-1 and ERS-2. The improved design aims at providing ocean surface winds at 50 km over a 25 × 25 km2 grid, along and across two 550 km wide swaths on both sides of the nadir track. A high-resolution wind product will be generated at 12.5 × 12.5 km2 grid, providing wind vectors at the sea surface at 25 km resolution.

The Meteosat and EPS/Metop Satellite Series

583

Further capabilities of ASCAT are the measurement of sea ice boundaries and sea ice type. Emerging applications are expected from land surface observations. 4.4.2 Global Ozone Monitoring Experiment The GOME-2 provides the possibility to monitor the ozone total column and profiles and the components related to ozone chemistry in the Earth’s atmosphere. The instrument on Metop profits from experience gained over many years of observation and data analysis with the GOME instrument on ERS-2. Improvements include: • • • • •

Increased spatial sampling of 40 × 40 km2 for total column products, Increased Earth coverage due to increased swath width (1,920 km). Improved polarization measurements (12 bands). Enhanced on-board calibration through added white light source. Increased spectral sampling.

GOME-2 measures the backscattered UV–Vis radiation in four band between 240 and 790 nm, at a spectral resolution between 0.25 and 0.5 nm. Expected accuracy of ozone total column and profiles is better than 5% and 15% above 30 hPa and better than 50% below 30, hPa respectively. The objective is 3% for columns and 10% for profiles at all levels. Additional intended products are vertical columns of BrO, OClO, NO2, and SO2, expected to be retrieved with accuracy better than 20%.

4.5 GPS Radio Occultation Sounder The GRAS makes use of the signals of the Global Positioning Satellite System (GPS) and introduces this technology into operational use for the first time. It follows experience obtained with the GPS/MET and CHAMP experiments. The basic product is the bending angle profile, which provides the basic parameter for the derivation of temperature profiles from atmospheric refractivity. Making the GPS occultation sounding operational in real time requires to introduce a complex subsystem into EPS. In order to correct the clock errors from the different involved satellites and to provide the required precise orbits of the satellites involved in the measurement, in particular the Metop satellite, the installation of a global GRAS Support Network (GSN) was required. This GSN needs to be operated in real time includes 24 global stations.

J. Schmetz et al.

584

4.6 EPS ground segment data and products The EPS ground segment comprises a central component and distributed elements, the former at EUMETSAT in Darmstadt, the latter at the SAF, hosted by EUMETSAT member states (see Section 3.2). The central facility provides global level 1 products with a latency of 2 h 15 min, and selected level 2 products (IASI and ATOVS retrieved profiles of temperature and humidity) with a latency of 3 h. The SAF provide level 2 and higher products, based on EPS products or as multi-mission products using other satellite data, including MSG data. The access to the products is assured in real-time through the Advanced High Resolution Picture Transmission (AHRPT) and the Low Resolution Picture Transmission (LRPT) service, where users have the possibility to receive (local) EPS data in real time, when the satellite is in view of their reception stations. HRPT comprises all mission data, whereas LRPT contains the full set of ATOVS sounding HIRS/4, AMSU-A, and MHS) data and JPEG compressed AVHRR data (three selected channels). LRPT replaces the analogue APT service. The near real-time access to global level 1 and level 2 products extracted at EUMETSAT is also provided to EUMETSAT member states and ECMWF. A subset of the centrally processed products will be made available to WMO users for the dissemination over the Global Telecommunication Service (GTS/RMDCN). SAF products are planned to be distributed via GTS/RMDCN as well. The EUMETSAT Unified Meteorological Archive and Retrieval Facility (U-MARF) provides off-line data access within 7 h from measurement. All centrally generated products will be archived in U-MARF, this includes the raw data. Most of the products generated in SAFs are archived locally SAFs, however catalogue information is available in U-MARF too.

5

SUMMARY AND OUTLOOK

The European geostationary Meteosat satellites have been a great success and are, nowadays, considered as indispensable for meteorological services. Seven Meteosat’s of the first generation provided operational services since 1977, imaging the full disk of the Earth every 30 min in three spectral bands. The last three of the generation, i.e., Meteosat-5, Meteosat-6, and Meteosat7, are still in operational use. With Meteosat-8 a new series of geostationary satellites has started. After the launch end of August 2002 an extended commissioning period ended with the start of operations in January 2004. Meteosat-8 and its successors, continue the successful 25-year long mission of the first-generation Meteosat satellites. The established services from the first-generation satellites will

The Meteosat and EPS/Metop Satellite Series

585

continue with a seamless operational transfer. In addition, Meteosat-8 and its successors offer a wealth of new observational capabilities: SEVIRI, the operational imager, has twelve spectral channels and observes the Earth (full disk) every 15 min. The multispectral imaging and the high temporal repeat cycle will benefit weather forecasting and will improve severe weather warning. Significant indirect benefits will come through the assimilation of products in numerical weather models by improved forecasts. Four Meteosat satellites of the second generation are foreseen to cover operational services until about 2015, when Meteosat third generation is expected to take over. The new EPS, developed jointly with the ESA and CNES, and launched in 2006 establishes the European contribution to the global polar meteorological space observing capabilities. The Metop satellites of EPS fly together with the last TIROS satellites and then with the future National Polar Orbiting Environmental Satellite System (NPOESS) and NPOESS Preparatory Programme (NPP) satellites, where Metop will be in the morning orbit around 0930 UTC. NPOESS satellites will have instruments equivalent to the ones on Metop, i.e., Cross Track Interferometer Sounder (CrIS) equivalent to IASI and ATMS equivalent to AMSU-A plus MHS. EPS provides on the one hand continuity to current systems, through continuation of the proven ATOVS instrument suite and the AVHRR imager, on the other hand it includes novel capabilities, i.e.: • IASI provides high spectral resolution sounding and radiances which will improve NWP and ensures the availability of such a service in the mid-morning orbit. • Instruments with heritage from ESA Earth observation missions (ASCAT and GOME) are utilized operationally and will provide continuous observations over a period of at least 14 years. • With GRAS the radio occultation principle is introduced for the first time into an operational system and will demonstrate the capability of such system to provide high quality soundings in near real time. • The mission duration of 14 years will assure long-term and consistent observations that provide a sustained basis for improved utilization in NWP. Furthermore, it provides the basis for climate-monitoring which could be enhanced through a regular reprocessing of data and products considering scientific advance.

6

REFERENCES

Aminou, D. M. A., H. J. Luhmann, C. Hanson, P. Pili, B. Jacquet, S. Bianchi, P. Coste, F. Pasternak, and F. Faure, 2003: Meteosat Second Generation: A comparison of on-ground and on-flight imaging and radiometric performances of SEVIRI on MSG-1. Proc. 2003 EUMETSAT Meteorological Satellite Conference, Weimar, Germany, 29 September–3 October 2003, 236–243.

586

J. Schmetz et al.

Bühler, Y., G. Mason, J. Perez, D. Klaes, and T. Brefort, 2001: The EUMETSAT Polar System: Mission, system and programmatics. 52nd International Astronautical Congress, October 1–5, 2001, Toulouse, France. de Waard, J., W. P. Menzel, and J. Schmetz, 1992: Atlantic data coverage by Meteosat-3. Bull. Amer. Meteor. Soc., 73, 977–983. Holmlund, K., 1998: The utilization of statistical properties of satellite-derived atmospheric motion vectors to derive quality indicators. Wea. Forecasting, 13, 1093–1104. König, M., 2002: Atmospheric instability parameters derived from MSG SEVIRI observations. Technical Memorandum, 9, EUMETSAT Programme Development Department, pp. 27. Klaes, D., J. Schmetz, M. Cohen, J. Figa, J.-P. Luntama, R. Munro, P. Schlüssel, and A. Ratier, 2001: The EUMETSAT Polar System within the Initial Joint Polar System: Mission objectives, expected capabilities and products. 1st Post MSG User Consultation Meeting, Darmstadt, 13–15 November 2001, 38 pp. Klaes, D., M. Cohen, Y. Bühler, P. Schlüssel, R. Munro, J.-P. Luntama, A. von Engeln, E.O. Clérigh, H. Bonekamp, J. Ackermann, and J. Schmetz, 2007: An introduction to the EUMETSAT Polar System (EPS). Bull. Amer. Meteor. Soc., in Press. Laurent, H., 1993: Wind extraction from Meteosat water vapour channel image data. J. Appl. Meteor., 32, 1124–1133. Mason, B. and J. Schmetz, 1992: Meteorological satellites. Int. J. Remote Sens., 13, 1153– 1172. Rizzi, R., 1998: Simulation of IASI radiances in presence of clouds. Final Report to EUMETSAT, Contract Number EUM/CO/96/390/DD, 34 pp. Schlüssel, P. and M. Goldberg, 2002: Retrieval of atmospheric temperature and water vapour from IASI measurements in partly cloudy situations. Adv. Space Res., 29, 1703–2706. Schmetz, J., K. Holmlund, J. Hoffman, B. Strauss, B. Mason, V. Gaertner, A. Koch, and L. van de Berg, 1993: Operational cloud motion winds from METEOSAT infrared images. J. Appl. Meteor., 32, 1206–1225. Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977– 992.

46 THE EVOLUTION OF THE NOAA SATELLITE PLATFORMS W. Paul Menzel NOAA-NESDIS, Madison, WI, USA

Abstract

The history and upcoming changes to the constellation of satellites operated by the National Environmental Satellite Data and Information Service (NESDIS) of the National Oceanic and Atmospheric Administration (NOAA) are described.

Keywords

NOAA, NESDIS, NPOESS, satellite, polar, geostationary

1

THE GLOBAL OBSERVING SYSTEM

In the last 40 years, a remote-sensing capability on polar and geostationary platforms has been established that has proven useful in monitoring and predicting severe weather such as tornadic outbreaks, tropical cyclones, and flash floods in the short term as well as climate trends indicated by sea surface temperatures, biomass burning, and cloud cover in the longer term. This has become possible first with the visible and infrared window imagery of the 1970s and has been augmented with the temperature and moisture sounding capability of the 1980s. Satellite imagery, especially the time continuous observations from geostationary instruments, dramatically enhanced our ability to understand atmospheric cloud motions and to predict severe thunderstorms. These data were almost immediately incorporated into operational procedures. Sounder data are filling important data voids at synoptic scales. Applications include temperature and moisture analyses for weather prediction, analysis of atmospheric stability, estimation of tropical cyclone intensity and position, and global analyses of clouds. Polar orbiting

587 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 587–600. © United States Government 2007.

588

W. P. Menzel

microwave measurements help to alleviate the influence of clouds for all weather soundings. Geostationary depiction of temporal and spatial changes in atmospheric moisture and stability are improving severe storm warnings. Atmospheric flow fields (wind field composites from cloud and water vapour drift) are helping to improve hurricane trajectory forecasting. Applications of these data also extend to the climate programmes; archives from the last twenty years offer important information about the effects of aerosols and greenhouse gases and possible trends in global temperature. In the next decade more improvements will be realized. The anticipated accuracies, resolutions, and cycle times for some of the meteorological parameters derived from satellite systems planned for this decade are summarized in Table 1; these include temperature and humidity profiles (from infrared and microwave radiometers) and upper winds (from tracking the movement of cloud and water vapour features). Infrared radiometers provide the highest quality profiles but only in clear sky conditions. Microwave radiometers provide data under cloudy conditions but have lower vertical resolution. Observations from both are currently widely available over the oceans, but over land varying surface emissivity is currently limiting soundings to the upper atmosphere. Upper wind observations are provided on a global basis but only where suitable tracers are available and usually only at one level in the vertical. Improvements in resolution and capability are expected over the next 5–10 years; better utilization of sounding data over land is also near at hand. As the space based remote-sensing system of the future develops and evolves, four critical areas (all dealing with resolution) will need to be addressed in order to achieve the desired growth in knowledge and advanced applications. They are: (1) spatial resolution – what picture element size is required to identify the feature of interest and to capture its spatial variability; (2) spectral coverage and resolution – what part of the continuous electromagnetic spectrum at each spatial element should be measured, and with what spectral resolution, to analyse an atmospheric or surface parameter; (3) temporal resolution – how often does the feature of interest need to be observed; and (4) radiometric accuracy – what signal to noise is required and how accurate does an observation need to be? Each of these resolution areas should be addressed in the context of the evolving space based observing system wherein the satellite(s) exist, or will exist. Higher temporal resolution is becoming possible with detector array technology; higher spatial resolution may come with active cooling of infrared detectors so that smaller signals can be measured with adequate signal to noise. Higher spectral resolution is being demonstrated through the use of interferometers and grating spectrometers. Advanced microwave radiometers measuring moisture as well as temperature profiles are being introduced in polar orbit; a geostationary complement is being investigated. Ocean colour observations with multispectral narrow band visible measurements are being

The Evolution of the NOAA Satellite Platforms

589

studied. Active sensors are being planned to supplement passive sensors with measurements of ocean height and atmospheric motions. The challenge of the future is to further the progress realized in the past decades so that environmental remote sensing of the oceans, atmosphere, and earth increases our understanding the processes affecting our lives and future generations. Table 1. Anticipated accuracies, resolutions, and cycle times for some of the meteorological parameters derived from satellite systems planned for this decade. Note that (1) the horizontal and vertical interval refers to the sampling distance between consecutive measurements, (2) cycle is the time interval between measurements and assumes two satellites in polar orbit, (3) delay is the time delay between the observation and receipt of data by the end user, (4) satellite winds are single level data with vertical sampling typically ~1 km, and (5) for ATOVS, 1–4 profiles are provided per second per satellite (per 100 km2). Element and Instrument

Humidity Sfc – 300hPa NOAA & METOP (ATOVS)6 METOP (IASI) & NPOESS (CrIS) Humidity below 50hPa METOP (GRAS) & NPOESS (GPSOS) Temperature Sfc – 10hPa NOAA & METOP (ATOVS) 6 METOP (IASI) & NPOESS (CrIS) Temperature 500hPa – 10hPa METOP(GRAS) & NPOESS (GPSOS) Wind Sfc – 200hPa MSG (SEVIRI) & GOES (Imager)

2

Horiz Res1 (km)

Vert Res2 (km)

Cycle 3

Delay4

Accuracy

(h)

(h)

(rms)

50 15

3 1

6 12

2 2

15% 10%

500

~1

12

2

10%

50 15

3 1

6 12

2 2

1.5 K 1K

500

~1

12

2

1K

50/100

One level5

1

2

2–5 m s–1

MEETING REMOTE-SENSING REQUIREMENTS IN THE NEXT TWO DECADES

Monitoring of the Earth’s environment has become an international endeavor. No one country has the observational systems necessary to provide the data it needs for its environmental monitoring and prediction operations. In the last decade and more so in the next decade, satellite remote-sensing contributions to the Global Observing System are being made by an increasing number of international partners. The collaboration and coordination among the international satellite community continues to increase. The demands for environmental data are enormous, ranging across all components of the Earth system – atmosphere, oceans, and land. The data

590

W. P. Menzel

requirements cover a broad range of spatial and time scales – 100s of metres and minutes to global, seasonal and inter-annual to decadal and centennial time frames. A number of research satellites are starting to provide advanced observations of the Earth and its atmosphere; these instruments will provide data on atmospheric, ocean, and land surface conditions with accuracy and spatial resolutions never before achieved. It is not uncommon for today’s research satellites to achieve lifetimes of several years, and many nonoperational space agencies are becoming increasingly aware of the importance of utilization of their satellite’s data streams. It is not unrealistic to foresee a time when there will be more and more of these special data made available for operational uses (blurring the distinction between research and operational platforms). Data from satellites are making contributions not only in the weather forecasting arena, but also in the fields of climate and ocean research. Satellite remote sensing is beginning to establish the level of continuity and calibration in their worldwide observations so that it will be possible to understand the physical, chemical, and biological processes responsible for changes in the Earth system on all relevant spatial and time scales. Coastal and ocean satellite remote-sensing services are expanding. A new generation of improved resolution, coastal and ocean remote-sensing capable satellites is rapidly emerging. For the first time ever, one can envision an operational coastal and ocean remote-sensing system that will assist marine fisheries management, coastal planning, and environmental quality stewardship. In the past, the satellite systems evolved mainly from a series of technology demonstrations. In the current planning, improvements in satellite remote sensing are being driven by user requirements for improved measurements and products. While these will rely on new technology demonstrations, the push is coming from user requirements more than opportunities to realize new technological capabilities. Finally, it is important to note that satellites are but one component of an Integrated Global Observing System (IGOS). In situ observations from a variety of instruments on balloons, planes, ships, buoys, and land surfaces are the other part of an IGOS. The challenge is to identify the best mix of satellite and in situ observations that will meet environmental monitoring and prediction requirements. This is a daunting task that must build upon the current systems and anticipate the future systems.

3

CURRENT AND FUTURE POLAR PLATFORMS

Polar orbiters allow a global coverage to be obtained from each satellite twice a day. To provide a reasonable temporal sampling for many applications at least two satellites are required, thereby providing 6-hourly coverage. A backup capability exists by reactivating “retired” platforms and this has been demonstrated recently. Since 1979, coverage with two polar orbiting

The Evolution of the NOAA Satellite Platforms

591

satellites has been achieved most of the time. The orbital altitude of 850 km makes it technically feasible to make high spatial resolution measurements of the atmosphere/surface. Current operational polar orbiters include the NOAA series from the USA and the METEOR, RESURS, and OKEAN series from Russia and the FY-1 series from China. They provide image data that can be received locally. The NOAA satellites also enable generation of atmospheric sounding products that are disseminated to NWP centres on the GTS. In the future, the NOAA AM satellite will be replaced by the METOP satellites provided by EUMETSAT and the NOAA PM satellite will transition to the NPOESS series. Instruments that have been or will soon be a part of the polar orbiting series of satellites include: visible and infrared radiometers, atmospheric temperature and humidity sounding, microwave all-weather radiometers, ozone monitoring, scatterometers, radiation budget, and positioning sensors. Table 2. VIIRS channel number, wavelength (µm), and primary application. Bands I are sensed with a spatial resolution of 400 m and bands M at 800 m. The signal to noise ratio in the reflective bands ranges from 25 to 1,000; the noise equivalent temperature difference in the emissive bands ranges from 0.03 to 0.4 K (larger values for the I bands). Ch number M1 M2 M3 M4 I1 M5 M6 M7 I2 M8 M9 M10 I3 M11 M12 I4 M13 M14 M15 I5 M16

Wavelength (µm) Reflective Bands 0.412 0.450 0.488 0.555 0.630 0.672 0.751 0.865 0.865 1.24 1.378 1.61 1.61 2.26 Emissive Bands 3.7 3.74 4.05 8.55 10.8 11.55 12.0

primary application ocean colour/ aerosol ocean colour/ aerosol ocean colour / aerosol ocean colour / aerosol imagery ocean colour / aerosol atm corr atm corr NDVI cld particle size cirrus snow fraction snow map clouds SST imagery / clouds SST / fires cld top properties SST cloud imagery SST

592

W. P. Menzel

3.1 Visible and infrared radiometers The Advanced Very High Resolution Radiometer (AVHRR), flown in October 1978 on TIROS N, measures radiation in five visible and IR windows at 1 km resolution. This will transition to a more capable visible and infrared imager called the Visible Infrared Imaging Radiometer Suite (VIIRS, see Table 2), when the NOAA satellites become the NPOESS series, starting with a demonstration program in 2006, called the NPOESS Preparatory Project (NPP). VIIRS will be better calibrated than the AVHRR, have higher spatial resolution (400 m vs. 1 km at nadir), and have additional spectral capability including channels that can be utilized to determine ocean colour. Parameters that may be derived from the VIIRS for use in operational as well as climate monitoring include sea surface temperature, aerosols, snow cover, cloud cover, surface albedo, vegetation index, sea ice, and ocean colour.

3.2 Atmospheric temperature and humidity sounding An important development was the remote sounding of vertical temperature and humidity profiles in the atmosphere on a worldwide basis with the TIROS Operational Vertical Sounder (TOVS). TOVS has evolved to an advanced version in 1998 and consists of the High resolution Infrared Radiation Sounder (HIRS) and the Advanced Microwave Sounding Unit (AMSU). These IR and microwave sounders can produce soundings in clear and cloudy (non-precipitating) skies every 50 km. NOAA will be transitioning to more capable sounders in the NPOESS series, starting with a demonstration program in 2006 on NPP. HIRS will be replaced by the Cross Track Infrared Sounder (CrIS), a Michelson interferometer that is designed to enable retrievals of atmospheric temperature profiles at 1 degree accuracy for 1 km layers in the troposphere, and moisture profiles accurate to 15% for 2 km layers. This is accomplished by the CRIS working together with the Advanced Technology Microwave Sounder (ATMS), being designed to be the next generation cross track microwave sounder. Comparable sounding capability will be realized on the METOP series by the Infrared Atmospheric Sounding Interferometer (IASI) in conjunction with the advanced microwave temperature sounding units (AMSU-A) and microwave humidity sounders (MHS/HSB). CrIS/ATMS will fly on afternoon (1330 ascending) and IASI/AMSU/MHS will fly in the morning (0930 descending) orbit.

3.3 Microwave all-weather radiometers A complementary series of DMSP satellites in polar orbit fly a scanning microwave radiometer called the Special Sensor Microwave Imager (SSM/I), flown since June 1987. These provide night–day, all-weather imaging of the

The Evolution of the NOAA Satellite Platforms

593

land and ocean surface because of the ability of microwave radiation to penetrate clouds. NOAA has used the DMSP SSM/I data extensively. A conical scanning version of the microwave sounder will be flown on NPOESS. The Conical Microwave Imager Sounder (CMIS) will combine the microwave imaging capabilities of Japan’s Advanced Microwave Scanning Radiometer (AMSR) on EOS PM-1, and the atmospheric sounding capabilities of the Special Sensor Microwave Imager/Sounder (SSMI/S) on the current DMSP satellites. Polarization for selected imaging channels (vertical, horizontal, and +/– 45 degrees) will be utilized to derive ocean surface wind vectors similar to what has previously been achieved with active scatterometers. Although demonstrated on airborne platforms, space based validation of the passive microwave technique for wind vector derivation is being tested in the Windsat Coriolis mission in 2003. CMIS data can be utilized to derive a variety of parameters for operations and research including all weather sea surface temperature, surface wetness, precipitation, cloud liquid water, cloud base height, snow water equivalent, surface winds, atmospheric vertical moisture profile, and atmospheric vertical temperature profile.

3.4 Monitoring ozone Another important sounding approach used the ultraviolet portion of the electromagnetic spectrum to sound atmospheric ozone. The Solar Backscatter Ultraviolet (SBUV), which provides information on ozone amounts for atmospheric 7–10 km layers, was incorporated into the operational series of NOAA polar satellites (POES) beginning with NOAA-9 in 1984. The Total Ozone Mapping Spectrometer (TOMS) flew on Nimbus-7 and provided critical image data that first identified the Antarctic ozone hole, but it has not been made operational. The Nimbus-7 TOMS lasted into the 1990s and was replaced subsequently by TOMS sensors flying on a Russian Meteor spacecraft, the Japanese ADEOS, and a NASA Earth Probe. The TOMS equivalent capability will be continued with the flight of the Dutch provided Ozone Mapping Instrument (OMI) on NASA’s Chemistry mission in 2004 and subsequently the Ozone Mapping and Profiler Suite (OMPS) on NPOESS, being developed for flight on afternoon (1330 ascending) NPOESS platforms. It consists of a nadir scanning ozone mapper similar in functionality to TOMS and a limb scanning radiometer that will be able to provide ozone profiles with vertical resolution of 3 km. Depending upon its ultimate design, the OMPS may be able to provide some of the same capability as limb scanning sensors on NASA’s UARS and EOS Chem. However in the near term, there is concern about a possible gap in TOMS type data coverage.

594

W. P. Menzel

3.5 Scatterometers The first active radar scatterometer to determine wind speed and direction over the ocean surface was flown on Seasat in 1978. Great progress in this area was possible in the 1990s with the SCAT data on ERS-1 and ERS-2; the SCAT on ERS-2 is still providing quasi-operational data. A NASA scatterometer termed NSCATT flew on Japan’s ADEOS-1 from August 1996 to June 1997; scientists were able to show a significant positive impact in predicting marine forecasting, operational global numerical weather prediction, and climate forecasting. A follow-on mission, Quickscat, launched in 1999, carries the NSCATT successor instrument, Seawinds. Another Seawinds sensor flew on ADEOS-2 in 2002. No additional US scatterometer missions are planned before NPOESS, which plans to use a passive microwave approach to determining the ocean vector wind field. This passive microwave technique is being tested as part of the Windsat Coriolis mission in 2003. Europe’s METOP series of satellites, scheduled to begin flying in 2005 include an Advanced Scatterometer (ASCAT) sensor, but ASCAT alone may not be able to provide the required geographic coverage and frequency of observation needed for operations and research. The Japanese are planning to carry a Seawinds follow-on provided by NASA on GCOM early next decade.

3.6 Radiation budget The Earth’s radiation budget and atmospheric radiation from the top of the atmosphere to the surface will be measured by the Clouds and Earth Radiant Energy System (CERES) on its afternoon (1330 local time ascending orbit) NPOESS platforms. The predecessor Earth Radiation Budget (ERB) sensors flew on Nimbus in 1978, as well as on a free flyer and on NOAA-9 and NOAA-10 in the mid-1980s. The first CERES is currently flying on the Tropical Rainfall Measuring Mission (TRMM) which was launched in November 1997. Two CERES scanners (one each working in the biaxial and cross track mode) are in orbit with EOS Terra since December 1999 and EOS Aqua since May 2002.

3.7 Altimetry Altimeters flew on the European ERS-1 and ERS-2 satellites in the 1990s and provided a major quasi-operational contribution. NOAA is planning to manifest a dual frequency microwave radar altimeter for its morning (0530 descending) NPOESS platforms. The type of altimeter, realized with JASON-1 in 2001, measures the ocean topography which provides information on the ocean current velocity, the sea level response to global warming/cooling and hydrological balance, the marine geophysical processes (such as crustal deformation), and the global sea state.

The Evolution of the NOAA Satellite Platforms

595

3.8 Positioning sensors Geometric determinations of location depend on inferences about the atmospheric temperature and moisture concentrations; they provide valuable complementary information to tropospheric infrared and microwave sounders about the tropopause and stratosphere. Ray bending and changes in the phase and amplitude of the transmitted signals allowing inference of the upper atmosphere temperature profile to the order of 1 K or better between altitudes of 8–30 km in layers (with footprints ranging between 1 × 30 km2 and 1 × 200 km2 extent) with near global coverage. The coverage would be expected to be evenly spread over the globe, excepting polar regions. The system measures upper atmospheric virtual temperature profiles so data from the lower atmosphere would require alternate data to separate vapour pressure and temperature traces. The Global Positioning System Occultation Sensor (GPSOS) will measure the refraction of radiowave signals from the GPS constellation and Russia’s Global Navigation Satellite System (GLONASS). This uses occultation between the constellation of GPS satellite transmitters and receivers on LEO satellites. The GPSOS will be used operationally for spacecraft navigation, characterizing the ionosphere, and experimentally to determine tropospheric temperature and humidity. A similar system, GPSMET, flew in 1995. NOAA is planning to manifest a GPSOS on all NPOESS platforms. A promising research GPS system is COSMIC (Constellation Observing System for Meteorology, Ionosphere and Climate). The National Space Program Office (NSPO) in China, the University Corporation for Atmospheric Research (UCAR), the Jet Propulsion Laboratory (JPL), the Naval Research Laboratory (NRL), the University of Texas at Austin, the University of Arizona, Florida State University and other partners in the university community are developing COSMIC, a project for weather and climate research, climate monitoring, space weather, and geodetic science. COSMIC plans to launch eight LEO satellites later this decade, each COSMIC satellite will retrieve about 500 daily profiles of key ionospheric and atmospheric properties from the tracked GPS radio-signals as they are occulted behind the Earth limb. The constellation will provide frequent global snapshots of the atmosphere and ionosphere with about 4000 daily soundings (see: http://www.cosmic.ucar.edu/).

4

CURRENT AND FUTURE GEOSTATIONARY PLATFORMS

The geosynchronous orbit is over 40 times higher than a polar orbit, which makes measurements technically more difficult from geostationary platforms. The advantage of the geostationary orbit is that it allows frequent measurements over the same region necessary for now-casting applications

W. P. Menzel

596

and synoptic meteorology. A disadvantage is that a fixed full disk view of the Earth is viewed from one satellite. Thus, five equally spaced satellites around the equator are needed to provide global coverage; polar regions are reviewed poorly at large zenith angles. Currently, there is global coverage from geostationary orbit (>5 operational satellites for image data and products (e.g., cloud motion winds) and two satellites are providing a sounding capability as well. Reactivating “retired” platforms provides backup and there have been several examples of this. Some of the satellites provide a real-time reception capability to allow immediate access to the imagery for real-time applications. Products are disseminated on the GTS by the satellite operators for near real-time applications. Instruments that have been or will soon be a part of the geostationary series of operational satellites include: Table 3. Advanced Baseline Imager Spectral Bands and Objectives. Spatial resolution is 2 km infrared and 0.5 km visible, full disk coverage will require 5 min. Future GOES Imager (ABI) Band 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Wavelength Range (µm)

Central Wavelength (µm)

Sample Objective(s)

0.45-0.49 0.59-0.69 0.84-0.88 1.365-1.395 1.58-1.64 2.235 - 2.285 3.80-4.00 5.77-6.6 6.75-7.15 7.24-7.44 8.3-8.7 9.42-9.8 10.1-10.6 10.8-11.6 11.8-12.8 13.0-13.6

0.47 0.64 0.86 1.38 1.61 2.26 3.90 6.19 6.95 7.34 8.5 9.61 10.35 11.2 12.3 13.3

Daytime aerosol-over-land, Color imagery Daytime clouds fog, insolation, winds Daytime vegetation & aerosol-over-water, winds Daytime cirrus cloud Daytime cloud water, snow Day land/cloud properties, particle size, vegetation Sfc. & cloud/fog at night, fire High-level atmospheric water vapor, winds, rainfall Mid-level atmospheric water vapor, winds, rainfall Lower-level water vapor, winds & SO 2 Total water for stability, cloud phase, dust, SO 2 Total ozone, turbulence, winds Sfc. & cloud Total water for SST, clouds, rainfall Total water & ash, SST Air temp & cloud heights and amounts

4.1 Visible and infrared radiometers The Visible and Infrared Spin Scan Radiometer (VISSR), flown since 1974, has been the mainstay of geostationary imaging on GOES, Meteosat, and GMS. VISSR enabled 5–7 km images of the full earth disk every 30 min in two or three visible and infrared windows and one water vapour sensitive band. The USA changed to a staring Imager with 5 channels of visible and infrared measurements at 5 km resolution with full disk coverage in 30 min in 1993. More changes are underway. The USA will evolve to an Advanced Baseline Imager (ABI, see Table 3) that makes full disk images in 16 spectral bands in 5 min at 2 km infrared and 0.5 km visible resolution. ABI offers improved performance over current GOES in all dimensions (routine

The Evolution of the NOAA Satellite Platforms

597

full Earth disk imaging while enabling mesoscale sub 1 min interval imaging, better navigation, more noise free signals, and additional spectral bands for improved moisture feature detection).

4.2 Infrared sounding With the three axis stable platform on GOES-8, NOAA was able to introduce geostationary infrared sounders. Measuring the infrared radiation in 18 spectral bands, these sounders provide temperature and moisture sounding over North America and nearby oceans every hour every 30 km (in clear skies). A variety of products and applications are described in the literature. NOAA plans to evolve to the Hyperspectral Environmental Sounder (HES) in 2012, using an interferometer, focal plane detector arrays, and on board data processing to cover 3.7–15.4 µm with 2,000 plus channels measuring radiation from 10 km resolution; contiguous coverage of 6,000 × 5,000 km2 will be accomplished in less than 60 min. NASA is planning to demonstrate the technology necessary for HES, with the Geostationary Imaging Fourier Transform Spectrometer (GIFTS) in 2008. GIFTS will improve observation of all three basic atmospheric state variables (temperature, moisture, and wind velocity) with much higher spatial, vertical, and temporal resolutions. Water vapour, cloud, and trace gas features will be used as tracers of atmospheric transport. GIFTS observations will improve measurement of the atmospheric water cycle processes and the transport of greenhouse and pollutant gases. GIFTS and HES represent a significant advance in geostationary sounding capabilities and brings temporal and horizontal and vertical sounding resolutions into balance for the first time ever.

5

THOUGHTS ON THE FUTURE GLOBAL OBSERVING SATELLITE SYSTEM

At present, soundings of temperature and humidity are primarily the domain of the polar-orbiting meteorological satellite constellation. The reason for this is historical. Sounding instruments were first developed for the polarorbiting satellites since they flew closer to the Earth and provided a more complete global coverage. However, the present user requirement is for hourly soundings that cannot be satisfied with the present constellation of polar-orbiting satellites. Additionally, there is already proven technology for soundings from geostationary orbit. Finally, there are firm plans to continue at least some of the geostationary satellites with a sounding capability. Thus, soundings of temperature and humidity should be provided from both constellations of satellites. There are several user requirements for wind vector over the ocean surface. The technology has been proven for well over a decade. There are

598

W. P. Menzel

plans to fly operational scatterometers on operational polar-orbiting satellites. Thus, the operational polar-orbiting satellites should have the capability to provide surface wind vectors over the ocean. To assure continuity, as well as to provide sufficient overlap between geostationary satellites, there is a need for at least six geostationary satellites. The concept of requiring a satellite to have the capability to make several different types of concurrent observations, e.g., soundings, imagery and scatterometry from the same polar-orbiting satellite, should be reviewed. It is possible that a series of smaller single purpose satellites would be more cost effective. Although not yet implemented on operational satellites, the concept has been successfully demonstrated on experimental and single satellite missions. Experimental observation satellites pose unique challenges for open and timely access to experimental data in standard formats, preparation of the community for new data usage, and data continuity. It is expected that a set of guidelines will be developed and agreed upon by the satellite operators. Assuming that such assurances can be obtained, a new constellation of satellites could be added to the space-based GOS. Certainly, user requirements exist in abundance for parameters not provided by the meteorological satellites including, but not limited to aerosol, cloud ice, cloud water, ozone and other trace gas profiles, land cover, land surface topography, ocean wave period and direction, ocean topography, ocean colour, significant wave height, snow water equivalent, soil moisture, and vegetation type. There is now a convergence of needs since the R&D satellite operators have also shown a keen interest for operational evaluations of their new data. The existence of experimental satellite missions capable of measuring these as yet unsatisfied requirements provides ample proof of the availability of technology and plans – although not necessarily for a continuous series of satellites. Thus, the space-based GOS could add a constellation of experimental satellites covering several different mission areas such as oceanographic, atmospheric chemistry, high-resolution land use and hydrological. Such a constellation would probably require a variety of mission oriented polarorbiting experimental satellites. Coordination of equator crossing times and geostationary positions, frequency allocations for communications, standard data formats, and open data policy remain challenges for the global community when planning the future GOS. Thus it is envisioned that the space based component of the GOS in 2015 will involve • quantitative measurements for input to 4-D continuous global data assimilation systems, • synergistic multi-satellite / multi-sensor systems,

The Evolution of the NOAA Satellite Platforms

• • • • •

599

atmospheric state variables (T,q,V) with the required accuracy and frequency, multi-level cloud and H2O imagery, small-satellite GPS occultation density profiles, higher spatial resolution multispectral (hyperspectral) land and ocean surface observations, air quality (CO and O3) monitoring.

User requirements in several applications areas (including numerical weather prediction) indicate the need for a four-polar and six-geostationary satellite system. Microwave, altimetry, scatterometry, radio occultation, and lidar systems remain unique to the low earth orbiting satellites. In the evolved GOS, (a) sounding will be accomplished with combined radiometric (infrared and microwave) and geometric (radio-occultation) systems, (b) passive and active remote sensing are combined to offer the best measurement of water vapour at resolutions commensurate with its variability in nature (c) altimetry will be pursued with a two-orbit system fully operational with real time capability wide swath (non-scanning) altimeters to enhance mesoscale capabilities, (d) atmospheric wind profiles will be accomplished with Doppler lidar systems, (e) ocean surface wind vectors now achieved with active techniques will be derived from passive measurements, (f) SST will evolve from a combined LEO and GEO systems of measurements, (g) expanded ocean colour capabilities will include increased horizontal resolution for coastal areas, and (h) SAR will belong to a multi-satellite system with a “wave mode” and sea ice/wave monitoring service. Expansion of the space based component of the GOS will be an international collaboration. There will be efforts to facilitate contributions of single instruments to larger platforms or flying in formation; the latter will mitigate the need for launching the full platform upon the loss of one critical instrument. Replacement strategies of the current or near future GOS satellites by the next generation satellites will proceed with a phased implementation approach. The role of small satellites in the GOS will be expanded. Coordination of international contributions to the polar orbiting observing system to achieve optimal spacing for a balance of spectral, spatial, temporal, and radiometric coverage will be a goal. Operational continuation of research capabilities with proven utility to the GOS will occur as much as possible without interruption of the data flow. There will be a commitment for adequate resources to sustain research developments necessary for improved utilization of these measurements. As much as possible, preparation for utilization of a given new measurement will begin prior to launch with distribution of simulated data sets that test processing systems; this will improve the percentage of the instrument post

600

W. P. Menzel

launch lifetime that is used operationally (the current 6 months to 2 years of post launch familiarization will be reduced). International algorithm development will assure best talent participation and enhance uniformity in derived products.

47 JAPAN’S ROLE IN THE PRESENT AND FUTURE SATELLITE OBSERVATION FOR GLOBAL WATER CYCLE RESEARCH Riko Oki and Yoji Furuhama Japan Aerospace Exploration Agency (JAXA), Japan

1

INTRODUCTION

1.1 Background It is said that the year-by-year increase of the effect of global warming has recently been felt as reality. Global observations from satellites have increased their importance, and have long been recognized indispensable in Japan as well. As a worldwide activity, the implementation plan of the World Summit on Sustainable Development (WSDD) that was held in 2002 in Johannesburg refers to the promotion of global water cycle research and observation including satellite remote sensing. Under such circumstances, the Council for Science and Technology Policy (CSTP), which is the organization to make decisions on governmental policy on science and technology in the Cabinet Office in Japan, has summarized in 2002 a report on the long-term vision of the future development and application of space activities. The report says that Japan should give priority to Earth environment monitoring along with national security, information and telecommunications, and positioning in the future space development and applications. In the area of Earth environment monitoring and observation, the report emphasizes that continuous observations of carbon dioxide and global water cycle should be realized with Earth observing satellites to deepen our understanding of earth science. The Space Activities Commission (SAC) also summarized in September 2003 a long-term plan of space development in which observation of global warming and water cycle was selected as a weighty program.

601 V. Levizzani et al. (eds.), Measuring Precipitation from Space: EURAINSAT and the Future, 601–609. © 2007 Springer.

602

R. Oki and Y. Furuhama

JAXA has developed satellites and space-borne sensors for environment monitoring and measurement. For example, the Advanced Earth Observing Satellite (ADEOS) was launched in 1996, and ADEOS-II in 2002. The Precipitation Radar (PR) on NASA’s Tropical Rainfall Measuring Mission (TRMM) satellite and AMSR-E (the Advanced Microwave Scanning Radiometer for the Earth Observing System) on NASA’s Aqua satellite have been providing invaluable data for atmospheric science and other research areas. Related to the above mentioned trend of Japan’s policy, study for satellite development (so-called phase B study) has started for GOSAT (Greenhouse Gas Observation Satellite) and GPM (Global Precipitation Measurement) missions recently in 2003 and 2004, respectively.

1.2 Scientific and social needs for observation of global water cycle by satellites One of the major concerns of the present day is the global change of our environment like global warming. To assess the effect of changes in each environmental factor on the global environment, we need to know its present status, relationship to other factors, and the mechanism of how it affects the environment. Among the effects of global warming, changes in global water cycle and in the distribution of water resources as a consequence of the former should be emphasized because of their close connection with our life. In a sense, monitoring global water cycle is more important than monitoring temperature environment. In fact, changes in global water cycle are not caused only by global warming. Among many environmental factors related to the global water cycle, precipitation is one of the most important components, because it affects everyone’s life and work. Too much precipitation causes floods, and too less of it causes droughts. Agricultural production depends on precipitation. It is also a true global variable that determines the general circulation through latent heating and reflects climate changes. It is a key component of air–sea interaction and eco-hydrometeorological modeling. Japan is located in the east edge of Asia where the Asian monsoon affects directly her weather and climate. Monitoring the present status and predicting the variation of global water cycle and the Asian monsoon have been an important challenge to us. Even though precipitation is such an important component of our environment, it is one of the least known physics components of cloud–weather–climate prediction models. Because of its large variability in space and time, its distribution over the globe is not very accurately known. We can never obtain accurate global rain distribution data by rain gauges and ground-based rain radars alone. Knowledge of the spatial and temporal distribution of global precipitation is a key to improve our understanding of weather and climate systems.

Japan’s Role in the Present and Future Satellite Observation

603

Japan in collaboration with the USA launched the TRMM satellite in 1997 to investigate tropical rainfall. TRMM was a pilot project to measure rain from space and the world’s first space-borne precipitation radar technology was demonstrated. TRMM’s concept, that is, precipitation observation by radar in combination with a microwave radiometer, will be succeeded by the Global Precipitation Measurement (GPM). It is also important to observe 3-dimensional (3D) cloud structure globally by active sensors in the next step to improve cloud-weather-climate models.

2

TRMM

The TRMM satellite was launched in November 1997 and has collected more than a 6-year record of tropical and subtropical rain. Great strides in rain observation technology by satellite sensors were made in the TRMM mission. The results can be summarized in two points. Firstly, it carries the world’s first space-borne precipitation radar (PR developed by Japan) that enabled measurement of the detailed vertical structure of rain systems uniformly over the globe. Secondly, by combining the PR data with data simultaneously obtained by the visible and infrared scanner (VIRS) and the TRMM microwave imager (TMI), the uncertainties in rain estimates that originated in measurement principles and sensor performances themselves have been greatly reduced. Many scientific results were obtained by analyzing PR data in Japan, and major results can be summarized as follows: • • • • • • • •

Substantial increase in quantitative measurements of rain distribution in tropical and subtropical areas. Accurate rain measurements over ocean and land in nearly equal quality (PR) for a long time period. Impact of PR observation to TMI algorithms, resulting in more physically consistent and quantitative estimates of rain rates. Revealing diurnal, annual, and long-term variations of precipitation. Observation of 3D rain structure. Improvement of short-term weather forecasting by 4D data assimilation with TMI data. Estimation of sea surface temperature from TMI data. Estimation of soil moisture from the characteristics of surface cross sections measured by PR.

The success of the PR measurements that is proven by the achievements exemplified above and the effectiveness of simultaneous measurements by radar and microwave radiometer will be inherited by GPM, the next generation mission, and help to realize a further developed observation system.

R. Oki and Y. Furuhama

604

3

AQUA/AMSR-E AND ADEOS-II/AMSR

The Advanced Microwave Scanning Radiometer for the Earth Observing System (AMSR-E) was developed and provided to NASA’s EOS Aqua satellite by the National Space Development Agency of Japan (NASDA) as one of the indispensable instruments for Aqua’s mission (Kawanishi et al. 2003; Shibata et al. 2003). AMSR-E is a modified version of AMSR that was launched in December 2002 aboard the Advanced Earth Observing Satellite-II (ADEOS-II). Both of them are dual-polarized total-power passive microwave radiometers that observe water-related geophysical parameters to clarify the mechanism of global water and energy circulation. The frequency bands include 6.925, 10.65, 18.7, 23.8, 36.5, and 89.0 GHz. AMSR also has 50.3 and 52.8 GHz bands. The hardware improvements over the existing space-borne microwave radiometers for Earth imaging include the largest main reflector of its kind and the addition of 6.925-GHz channels that has been desired for a long time. These improvements provide finer spatial resolution and the capability to retrieve sea surface temperature and soil moisture information on a global basis. In addition to the variables such as water vapor, precipitation, and sea surface wind speed, that have been proved to be measured with a radiometer, AMSR and AMSR-E can retrieve novel geophysical parameters, including sea surface temperature (SST) and soil moisture, by using new frequency channels. Near-real-time products will be used to investigate satellite data assimilation into weather forecasting models and to contribute to improve forecasting accuracy. Very high spatial resolution of AMSR and AMSR-E improves precipitation retrievals, because it lessen the effect of nonlinear response (i.e., beam filling problem) of the brightness temperatures to the total rain amount within the footprint. Many applications over land as well as sea ice investigation also benefit from this improved spatial resolution. Although TMI realized the higher spatial resolution thanks to the lower orbit altitude of the TRMM, AMSR-E improves the resolution by using a large antenna and extends its capability to global measurements. Unfortunately, the observation with AMSR abruptly terminated in less than one year after launch because of the unexpected accident of ADEOS-II. The possibility of deploying an AMSR-like instrument in space is being discussed in order to restart a ADEOS-II follow-on mission as early as possible, because it is believed necessary to collect climate data at least for 15 years continuously to study the variation of Earth environment.

4

GPM

Based on the success of TRMM, a follow-on mission was proposed both in Japan and the USA. It is the Global Precipitation Measurement (GPM) that is planned to start in approximately 2008. Feasibility studies, related science,

Japan’s Role in the Present and Future Satellite Observation

605

and mission planning have been carried out in both countries. Details of GPM are described in the article by E. Smith et al. in this book and Furuhama et al. (2002). Only the significance of the mission from Japanese point of view is reaffirmed in this article. GPM will expand the TRMM mission in two ways. Firstly, GPM will make accurate measurement of precipitation that was established with the TRMM PR. Secondly, by international collaboration it will achieve frequent measurement of precipitation that was not realized by the TRMM satellite alone. In these two points, the GPM mission is not a single satellite program, but a program that utilizes the information from multiple of satellites in international collaboration. The entire GPM system consists of the core satellite and a fleet of constellation satellites each of which carries a microwave radiometer for precipitation measurement. The core satellite will carry the dual-frequency precipitation radar (DPR) which is an improved version of the TRMM PR and the GPM microwave imager (GMI). The constellation satellites will join GPM by providing microwave radiometer data in international partnership. The frequent and accurate global precipitation measurement will be realized with this whole system. GPM is expected to give answers to many questions. For example, how are the rainfall and rainfall structure responding to changes in the Earth’s temperature and other climate variables? How directly is the surface hydrology coupled to the rainfall and evaporation? We need to observe, understand, and model the Earth’s system to learn how it is changing, and consequences for life on Earth. To do so, we need to establish the existence of trends in the rate of global water cycle. Acceleration would lead to faster evaporation, increased global average precipitation, and a general increase in extremes, particularly droughts and floods. In addition, GPM may impact many other surrounding research areas. They included studies in cloud system and radiation, ocean–land atmosphere interactions, freshwater forcing on ocean processes, ocean salinity modeling, development of hydrometeorology and carbon assimilation models, soil moisture and its impact on flood/drought prediction, and water vapor transport, to mention a few.

4.1 DPR measurement Precipitation measurement with higher accuracy than TRMM is expected with the DPR because the DPR should be capable of differentiating solid particles from liquid particles and providing some information of the drop size distribution (DSD). The DPR consists of Ku-band (13.6 GHz) and Kaband (35.5 GHz) precipitation radars. The minimum detectable rainfall rate will be improved to 0.2 mm h–1 by the Ka-band radar. Table 1 shows the main characteristics of the DPR. Accurate rainfall estimates will be provided by a dual-frequency algorithm using the matched beam data observed simultaneously by the two radars. Figure 1 shows the concept of dual-frequency

R. Oki and Y. Furuhama

606

measurement of precipitation. Because the two radars have different dynamic ranges, the Ka-band radar will be used to measure snow and light rain whereas the Ku-band radar will measure light to heavy rain. In the overlapped dynamic range, DSD information and accurate rainfall estimates will be provided by a dual-frequency algorithm. The reader should refer to Nakamura and Iguchi (2007), and Adhikari and Nakamura (2003) for the details of the DPR and the rain profiling algorithms. Table 1. Main characteristics of DPR. Frequency Swath width Horizontal resolution Tx pulse width Range resolution Observation range Tx peak power Minimum detectable rainfall rate * Measurement accuracy Beam-mating accuracy Data rate Mass Power consumption Size

KuPR 13.597/13.603GHz 245 km 5 km (at nadir) 1.6 µs (×2) 250 m 18 km to –5km ASL >1000 W 0.5 mm h–1 (18 dBZ) within ±1 dB