Fire Detection

ENVIRONMENTAL SCIENCE, ENGINEERING AND TECHNOLOGY FIRE DETECTION No part of this digital document may be reproduced, s...

Author: Roger P. Bennett

98 downloads 1708 Views 11MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

ENVIRONMENTAL SCIENCE, ENGINEERING AND TECHNOLOGY

FIRE DETECTION No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in rendering legal, medical or any other professional services.

ENVIRONMENTAL SCIENCE, ENGINEERING AND TECHNOLOGY Additional books in this series can be found on Nova’s website under the Series tab.

Additional E-books in this series can be found on Nova’s website under the E-books tab.

ENVIRONMENTAL SCIENCE, ENGINEERING AND TECHNOLOGY

FIRE DETECTION

ROGER P. BENNETT EDITOR

Nova Science Publishers, Inc. New York

Copyright © 2011 by Nova Science Publishers, Inc. All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without the written permission of the Publisher. For permission to use material from this book please contact us: Telephone 631-231-7269; Fax 631-231-8175 Web Site: http://www.novapublishers.com

NOTICE TO THE READER The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or reliance upon, this material. Any parts of this book based on government reports are so indicated and copyright is claimed for those parts to the extent applicable to compilations of such works. Independent verification should be sought for any data, advice or recommendations contained in this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to persons or property arising from any methods, products, instructions, ideas or otherwise contained in this publication. This publication is designed to provide accurate and authoritative information with regard to the subject matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other professional services. If legal or any other expert assistance is required, the services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS. Additional color graphics may be available in the e-book version of this book.

LIBRARY OF CONGRESS CATALOGING-IN-PUBLICATION DATA Fire detection / editor, Roger P. Bennett. p. cm. Includes index. ISBN: 978-1-61122-369-9 (eBook) 1. Forest fires--Detection. 2. Forest fires--Research. I. Bennett, Roger P. SD421.375.F57 2010 634.9'618--dc22 2010036040

Published by Nova Science Publishers, Inc. † New York

CONTENTS Preface Chapter 1

vii Early Detection of Forest Fires from Space Based on the RTM Method G.G. Matvienko, S.V. Afonin and V.V. Belov

Chapter 2

Fire Surveillance and Evaluation by Means of Lidar Technique Andrei B. Utkin, Alexander Lavrov and Rui Vilar

Chapter 3

An Introduction to Uncertainty in Remotely Sensed Fire Maps and Historic Fire Regime Reconstructions Brean W. Duncan

Chapter 4

Chapter 5

Chapter 6

Chapter 7

Chapter 8 Index

Aerosol and Trace Gas Retrievals from Remote Sensing Fire Products Gabriel Pereira, Nelson Jesus Ferreira, Francielle da Silva Cardozo, Fabrício Brito Silva, Elisabete Caria Moraes, Yosio Edemir Shimabukuro, Saulo Ribeiro de Freitas and Karla Maria Longo

1 41

79

103

The Role of Magnetic Measurements in Detecting Past Fire Signatures in Soils and Sediments Frank Oldfield

119

Forest and Fire Risk Dynamics in the Great Xing’an Mountains, Northeastern China: A Spatial Simulation Study Zhihua Liu, Hong S. He, Yu Chang and Jian Yang

129

Detection of The Positions and Computing the Rate of Spread of Fire Fronts Using a Radiative Flame Model and Inverse Method K. Chetehouna, O. Séro-Guillaume and D. Bernardin Large Scale Forest Fires in Alaska: Detection and Prevention Hiroshi Hayasaka

149

173 191

PREFACE Forest fires are a serious problem affecting many terrestrial ecosystems and causing substantial economic damage. Due to the increase of frequency and severity of large forest fires and wildland-urban interface fires, the World Health Organization has identified this problem as a threat to public health security in the 21st century. This book gathers and presents current research from across the globe in the study of fire detection techniques and applications. Some topics discussed, herein, include: early detection of forest fires from space using the RTM method; using the Lidar technique (light detection and ranging) for evaluation and fire surveillance; remotely sensed fire maps and historic fire regime reconstructions; aerosol and trace gas retrievals from remote sensing fire products; as well as using wireless thermal sensors to detect the positions of the fire front. Chapter 1 - This chapter considers the results of the theoretical and practical works dealing with forest fire detection from space; they are performed at Zuev Institute of Atmospheric Optics (IAO), SB RAS, since 1997. The chapter consists of three parts. Its first part addresses the results of forest fire detection on the territory of Tomsk region for period of 1998-2008 with the application of AVHRR/NOAA satellite system. The authors studied the question of early detection of small-sized fires and analyzed the dependence of fire detection results on the time of the monitoring. The detection results, obtained at IAO with the help of the AVHRR/NOAA regional algorithm, are compared with those based on the MODIS Fire Products (MOD14) global algorithm, used for MODIS/EOS system. The second part of the chapter presents the methodic foundations of the RTM approach to the multispectral monitoring of the Earth’s surface. The authors analyzed the influences of the distorting effect of the atmosphere and uncertainties of the specification of the atmospheric optical and meteorological parameters on the results of retrieval of the Earth’s surface temperature from space. The third part describes the software for implementation of the RTM approach and the results of its practical application. Chapter 2 - Lidar (light detection and ranging) is an active remote detection technique that uses a pulsed laser beam to probe the atmosphere. When the laser radiation illuminates a target, such as a smoke plume originating from a forest fire, part of the incident radiation is backscattered, the intensity of this radiation is measured as a function of time by a suitable detector, and the resulting signal is analyzed by artificial intelligence methods. If the signature of a smoke plume is identified, an alarm is emitted. Precise position of the smoke

viii

Roger P. Bennett

plume is derived from the current azimuth/elevation angles of the laser beam (provided by the scanning system) and the distance to the target (calculated from the detection time). Being an active detection technique, lidar presents better sensitivity than conventional fire surveillance methods based on visible or infrared imaging. Instead of visualizing the fire, lidar functions by interacting with smoke, without needing line-of-sight observation of the flames. The lidar signals resulting from smoke plumes can be subjected to an inversion procedure, which yields the distribution of the extinction and backscattering coefficients along the laser beam path and, finally, the concentration of smoke particles. This characteristic makes the lidar methods an invaluable tool for the investigation of fire and smoke behavior in natural conditions and for the experimental verification of atmospheric gas-dynamic models. The authors describe simple and robust algorithms of lidar-signal recognition based on the fast extraction of sufficiently pronounced peaks followed by their classification with the help of an artificial intelligence method (neural network). The investigations to be presented include tracing smoke-plume evolution, restoring the smoke concentration and representing the results as contour plots on the topographic map, estimating forest-fire alarm promptness, and smoke-plume location by azimuth scanning of the probing beam. The possibility of locating a smoke plume whose source is out of line-ofsight and detection under extremely unfavorable visibility conditions are also demonstrated. The eye hazard problem caused by laser radiation is addressed and the possibilities of providing eye safety conditions are indicated. Chapter 3 - Uncertainty exists in all mapped geographic features. Geographic fire maps are no exception. Fire maps are produced using many techniques with remote sensing being among the most widely used methodologies for both single fire event mapping and recent historic fire regime reconstructions. Acknowledgement and incorporation of spatial map uncertainty in fire maps produced by any method is rare with few exceptions. Including uncertainty within fire mapping products will represent an important step in the evolution and maturation of fire mapping science. This chapter explores the chief sources of uncertainty in mapping fires, particularly when mapping fires using remote sensing techniques. A case study is presented that utilizes confidence information to reduce uncertainty in historic fire regime maps. Chapter 4 - Annually, anthropogenic fires devastate large areas of forest and grasslands all over the world, releasing a large amount of greenhouse gases and aerosols into the atmosphere. This issue affects the environment, altering the atmospheric and surface radiation balance, besides the biogeochemical and hydrologic cycles. The main objective of this work is to use the fire radiative energy (FRE) release rate to estimate carbon monoxide (CO), particulate matter of less than 2.5 microns in diameter (PM2.5µm) and the amount of biomass consumed by fires in the South America 2002 season. For this, combustion experiments near the Laboratory of Radiometry (LARAD) of Remote Sensing Division at the National Institute for Space Research (DSR/INPE) were performed to obtain the coefficient that relates the consumption of biomass with the Fire Radiative Energy (FRE) released rate. The emission inventory estimated by the Moderate Resolution Imaging Spectroradiometer (MODIS) and Wildfire Automated Biomass Burning Algorithm (WFABBA) from Geostationary Operational Environmental Satellites (GOES) Fire Radiative Power (FRP) measurements to the South America 2002 dry season were included in Coupled Chemistry-Aerosol-Tracer Transport model coupled to the Brazilian developments on the regional Atmospheric

Preface

ix

Modeling System (CCATT-BRAMS). The model results were evaluated with South America 2002 surface data collected in the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) Smoke, Aerosols, Clouds, rainfall and Climate (SMOCC) and the Radiation, Cloud and Climate Interactions (RaCCI) experiment. Evaluation of model and ground data revealed a good conformity with SMOCC/RaCCI data in the general pattern of temporal evolution. The results showed high correlations, with values between 0.80 and 0.95 (significant at 0.01 level by student t test), in PM2.5µm and CO emissions simulated in CCATT-BRAMS model. For the period analyzed, biomass consumed by fires can exceed 5 Tg (teragrams) in South America, with a daily average of 2.2 Tg (0.8 Tg estimated by MODIS and 1.32 Tg estimated by GOES). As a result, the coefficient derived from the relationship between biomass consumption and FRE released estimated the biomass burned from July to November 2002 in the South America dry season in approximately 0.28 ± 0.07 Pg (pentagrams). Chapter 5 - Fire can transform both paramagnetic and imperfect antiferromagnetic iron minerals with low magnetic susceptibility and, in the case of paramagnets, zero magnetic remanence, into strongly magnetic minerals, often with high susceptibility values and distinctive remanence characteristics. The present chapter outlines the ways in which magnetic measurements can be used to detect fire signatures in soils and sediments. The magnetic products of burning that are most readily detected and characterized are finegrained, dominantly superparamagnetic, ferrimagnetic minerals (maghemite and nonstoichiometric magnetite). By using a combination of magnetic susceptibility and remanence measurements, these can sometimes be distinguished from ferrimagnetic minerals produced by other processes. The research applications considered here include archaeological prospecting, reconstructions of fire histories from sedimentary evidence and sediment tracing. Chapter 6 - Natural disturbance-based forest management, based primarily on the understanding of natural disturbance regimes and forest dynamics, provide sustainable forest management paradigms to maintain biodiversity and essential ecological function in managed, forested regions. So understanding how forest ecosystem and fire dynamics respond to historic and current fire regime pose great significance in designing scientifically sound management plans for Great Xing’an Mountains in the Northeast China. The authors used a spatially explicit landscape dynamics model, LANDIS, to simulate the long-term forest response and fire dynamics under historic fire regime (before 1950s) and the fire suppression (after 1950s). Specifically, the authors compare how the fuel loads and fire hazards, and forest tree species abundance response under the two scenarios. Under the fire suppression scenario, fire risk will quickly increased to a dangerous level, about 80% of the landscape will carrying a high level of fire risk at the end of the simulation; both fine fuel and coarse fuel will rise to medium-high level after a few decades’ suppression. Generally, fires tend to be more catastrophic and less frequent. Fire suppression results in less frequent, but more intense fires. Fire suppression can also decrease the proportion of coniferous forests, increase the proportion of deciduous forests and alter forest age structures. The results suggest that extensive additional forest management activities, such as prescribed burning , fuel load reduction, uneven-aged harvesting should be implemented to maintain low level of fire risk and forest type diversity. Further studies are needed to evaluate the effects of prescribed burning and fuel load reduction under the fire suppression, and to find the fine balance among fire suppression, maximal timber production, and sound ecological function.

x

Roger P. Bennett

Chapter 7 - Usually models describing forest fires at large scale are compared to measurements obtained in prescribed burnings. These types of fires are set on fields of several tenths meters sizes. They usually are instrumented, using thermocouples and/or systems for recovering the fire front position. If the distribution of vegetation is homogeneous and if the wind is constant, the fire front is a straight line for a line ignition, its position can be obtained by the peaks of temperature. If the fire front is no more a straight line, other methods must be elaborated for determining the fire front position. Due to the size of the terrain, the use of thermocouples is uneasy and devices using optical remote sensing have been designed. However, the determination of the fire front position by optical methods is not so easy because of the size of the field in prescribed burning, and of the possibility of smoke. This paper is a methodological paper and the question addressed here is: is it possible to determine the fire line position measuring the radiative flux field? By the measurement of the radiative heat flux coming from the flame, we propose an approach for the computation of the fire positions and the rate of spread of flame fronts which are necessary to validate the numerical simulations of the propagation models of forest fires. In the first step of the method, the heat fluxes are measured by a specific thermal sensor in four horizontal directions. This thermal sensor is simple and convenient for installation and use, its scale is adapted to the one used in physical models of propagation, has a low cost, is not destroyed by the flames, and its acquisition is sent to computer by wireless connection. The wireless thermal sensor consists in a steel body, a thin heat insulating layer, a glue layer, 4 copper plates and 9 type K thermocouples. The nine measured temperatures are related to the heat fluxes by means of a transfer function utilizing intrinsic physical parameters of the thermal sensor. In the second step of this method, the heat fluxes are calculated using an approximate resolution of the radiative transfer equation. Subsequently, the positions of the fire front and the flame characteristics are determined by applying an inverse method. The rate of spread is deduced by applying a least-squares regression on the positions values. A theoretical extrapolation of this approach to any shape of fire front in more complex experimental scenarios such as prescribed burnings is presented. Chapter 8 - In 2004, wildfires burned 26,700km2 in Alaska. Nine individual fires exceeded 1,000km2 in size during a summer characterized by record high temperatures and extreme drought. A substantial portion of fire growth was realized on just a few days when strong pressure gradient winds occurred. The total burn area in 2004 was the largest since record-keeping began in Alaska in 1956. Combined with an additional 19,000km2 burned in 2005, the area burned equals 10% of Alaska’s boreal forest area in just two years. Such regional fire events are believed to be climate driven. We analyzed local and regional weather factors with fire growth derived from daily MODIS “hotspot” imagery, using the 2,180km2 Boundary Fire as an example.

In: Fire Detection Editor: Roger P. Bennett

ISBN 978-1-61122-025-4 © 2011 Nova Science Publishers, Inc.

Chapter 1

EARLY DETECTION OF FOREST FIRES FROM SPACE BASED ON THE RTM METHOD G.G. Matvienko1,2 , S.V. Afonin1,2 and V.V. Belov1,2 1

Zuev Institute of Atmospheric Optics of the SB RAS, Tomsk, Russia 2 Tomsk State University, Tomsk, Russia

ABSTRACT This chapter considers the results of the theoretical and practical works dealing with forest fire detection from space; they are performed at Zuev Institute of Atmospheric Optics (IAO), SB RAS, since 1997. The chapter consists of three parts. Its first part addresses the results of forest fire detection on the territory of Tomsk region for the period of 1998-2008 with the application of AVHRR/NOAA satellite system. We studied the question of early detection of small-sized fires and analyzed the dependence of fire detection results on the time of the monitoring. The detection results, obtained at IAO with the help of the AVHRR/NOAA regional algorithm, are compared with those based on the MODIS Fire Products (MOD14) global algorithm, used for MODIS/EOS system. The second part of the chapter presents the methodic foundations of the RTM approach to the multispectral monitoring of the Earth’s surface. We analyzed the influences of the distorting effect of the atmosphere and uncertainties of the specification of the atmospheric optical and meteorological parameters on the results of retrieval of the Earth’s surface temperature from space. The third part describes the software for implementation of the RTM approach and the results of its practical application.

INTRODUCTION In remote sensing of the ground surface from space, the actual problem of real-time detection of fires in forests and industrial objects is solved. It is obviously important to detect a seat of fire in early stages (when its area is less than 5–10 ha) when its extinguishing does

2

G.G. Matvienko, S.V. Afonin and V.V. Belov

not require great efforts. In this case, reliable algorithms automatically detecting small-sized high-temperature objects (HTO) with area less than 0.1% of the pixel size are required. An analysis of the algorithms for the detection of fire seats from space published in the literature has allowed us to conclude the following. In most fire detection algorithms used in practice, the decision rule P{x} > dP is used, where dP is the threshold value of the function P{x}, and its parameters {x} are satellite measurements of albedos and brightness temperatures (or of their functions). The threshold dP values are fixed or can be determined based on statistical analysis of {x} in the vicinity of a potential fire seat. However, the actual parameters of the atmosphere during satellite measurements are in fact, disregarded in algorithms used in practice. Efficient solution of fast control over the environment state from space is possible only on the base is of application of high-accuracy algorithms of thematic processing and atmospheric correction of space measurements. This is particularly important in temperature sensing of the environment under complex observational conditions including the detection of low-intensity fires at early stage of their development. Regretfully, the presently available methods do not solve this problem. The importance of correction of satellite IR measurements for the distorting effect of the atmosphere with the use of information on the atmospheric state (meteorological and aerosol parameters of the atmosphere) and on the geometry of observations during satellite measurements is obvious for obtaining maximum accuracy. The main objectives of this chapter are: 1. To outline a procedure of real-time monitoring of forest fires at the IAO SB RAS with the use of the AVHRR data. 2. To analyse the results of satellite monitoring of boreal forest fires in the Tomsk Region in 1998–2008; to investigate the dependence of the results of fire monitoring on the time of day; to assess the possibility of early detection of fires from satellites. 3. To develop the multispectral RTM approach for monitoring of the Earth’s surface from space and for detecting the small-sized forest fires; to test the method in practice.

PART 1. MONITORING OF BOREAL FOREST FIRES IN THE TOMSK REGION OF WESTERN SIBERIA Real-time detection and monitoring of forest fires in vast and not easily accessible territories of Siberia and the Far East are urgent problems for Russia. In 1998-2008, according to the data of the Tomsk Forest Protection Services, 3205 forest fires with a total area of more than 213 000 ha burned in boreal forests in the Tomsk Region of Western Siberia (55°N to 60°N, 75°E to 90°E, see the map in figure 1). In the last decade, satellite monitoring of forest fires (SMFF) based, as a rule, on the AVHRR/NOAA data has been widely used in Russia. Since 1996, the National Forest Fire Centre of Russia has had an Internet site (http://nffc.infospace.ru) in which the satellite sensor data on forest fires for the most part of the territory of Russia are daily updated in firehazardous seasons.

Early Detection of Forest Fires from Space Based on the RTM Method

3

Along with the National Forest Fire Centre of Russia, there are some independent regional centers of satellite monitoring of forest fires (in Krasnoyarsk, Irkutsk, Novosibirsk, Tomsk, and Yakutsk). In our opinion, they significantly increase the efficiency of fire detection because of the knowledge of specific conditions of monitoring in local territories and implementation of satellite sensor data processing algorithms adapted to these conditions. It is also obvious that regional centres of satellite monitoring can maintain continuous contact with Forest Protection Services. Since 1998, the Institute of Atmospheric Optics (IAO) of the SB RAS has carried out real-time satellite monitoring of forest fires in the territory of Tomsk Region. The Institute has all the required components to solve this problem, including:

Figure1. Map of forest fires detected by Forest Protection Services in the Tomsk Region in 1998–2008 (the largest rivers are mapped).

• • • •

ScanEx station (http://scanex.ss.msu.ru) for the acquisition of digital satellite sensor data, knowledge of the theory of image transfer through the atmosphere and methods of correction of satellite measurements for the distorting effect of the atmosphere, long-term practical experience in satellite sensor data interpretation, and standard and original software packages for satellite sensor data processing (see, for example, [1-5]).

To develop a system of real-time satellite monitoring of forest fires, we first analyze the experience accumulated in Russia and abroad. The majority of works published in the literature contain data on fire monitoring in middle and southern latitudes (Africa, Brazil,

4


Spain, the USA, Canada and Northern Europe). To a lesser degree, such investigations were carried out in the Russian boreal forest zone. In addition, only a few papers addressed the efficiency of early detection of forest fires (EDFF) from satellites. The term “efficiency of early detection of forest fires” can be treated in two ways. First, it can be considered as the probability of detecting a small fire with an area of about 1 ha and smaller. This probability, in fact, represents the accuracy of satellite detection of fires in the initial stage of fire burning. Second, this term can be defined as the difference ΔT = T1 – T2 between the time of the first satellite fire detection (T1) and that of its detection by a Forest Protection Services (T2). Then the value of ΔT characterizes an advance (ΔT > 0) or a delay (ΔT < 0) of satellite systems of fire monitoring compared to conventional (ground-based or airborne) systems of fire detection. This characteristic allows the role of satellite sensor data in routine operation of Forest Protection Services to be determined in fire-hazardous seasons (even as a rough approximation).

1.1. IMAGE PROCESSING Satellite monitoring of boreal forest fires in the Tomsk Region is carried out every year in the fire-hazardous season from April (May) through September. The procedure of monitoring includes three main stages (described in detail in [38]). 1) Acquisition of digital data from NOAA-12 to NOAA-19 satellites. Pre-processing of satellite sensor data, including calculations of geographic position based on the wellknown SGP4 program and data from the Two-Line Elements Set available via Internet source (http://celestrak.com/NORAD/elements/noaa.txt). Refinement of geographic position in semi-automatic operation using reference points and contour hydrographic lines by a special program that allows the geographic position error to be reduced down to 1–2 km. 2) AVHRR data processing by two algorithms for automatic detection of hotspots on the underlying surface. Automatic rejection of sun glints in satellite images. The results of automatic satellite image processing and the quality of false alarms rejection are controlled by an operator. To improve the reliability of detecting small fires, a special computer program for automatic comparison of previous satellite images is used. This program identifies the presence of the fire being detected in these images. 3) Generating a map of the Tomsk Region based on the results of satellite monitoring of fires (fragment of the map is exemplified in figure 2). The map is then delivered to the Forest Protection Services with a text file describing some characteristics of the hotspots detected.

Earrly Detection of o Forest Firess from Space Based B on the RTM R Method

5

Figure 2. Fragment F of thee map with the results r of NOA AA–12 satellite monitoring m of forest fo fires in thee Tomsk Regionn at 18:26 LT on o July 5, 2000. The fires deteccted are markedd by small squaares and numbeered.

wo types of Fire Detecction Algorithhms. Accordinng to the availlable literaturee data [[6], tw saatellite algoritthms – Fixed Threshold Techniques T annd Spatial Anaalysis Techniques – are m mainly used to solve the probblem of autom matic detectionn of forest firees. Having tested various thhreshold algorrithms [[5] in the initial staage of fire moonitoring, we chose an algoorithm best addapted to the Siberian condditions. It was the algorithm m based on [[112] and develooped at the C Centre of Spacce-Borne Monnitoring at the Institute of Solar-Terrestri S al Physics (IS STP) of the SB B RAS [13]. b tem mperature in thhe AVHRR This algoriithm uses fourr fixed thresholds for the brightness baand 3 (T3) deppending on thee albedo in thee AVHRR bannd 1 (A1): 1) 2) 3) 4)

T3 > 29 90 K for A1 < 1 (and T4 > 2665 K), T3 > 30 06 K for A1 < 4, T3 > 31 16 K for A1 < 10, T3 > 32 20 K for A1 < 25 (and T4 > 265K), 2

giiven that the rule r T3 – [T4 + 3·(T4 – T5)] > 4 is fulfilledd, where T4 and a T5 are the brightness teemperatures inn the AVHRR bands 4 and 5, 5 respectivelyy. When testing the threshoold algorithmss for boreal foorests in the terrritory of Tom msk Region w the use off the actual sattellite sensor data with d for 1998 [5], the best results r were obtained for thhe ISTP algoriithm. At the saame time, the results of testiing indicated a need for impproving the

6


satellite algorithms for detecting forest fires in the case of monitoring under unfavorable conditions of broken cloudiness, low-intensity seats of fires, and sun glints from the water surface and clouds. For this purpose, the original Spatial Analysis Algorithm [14] was developed at the IAO SB RAS in 1999. The algorithm allowed the contribution of solar radiation in the 3rd AVHRR channel to be considered and the reliability of fire detection under unfavorable conditions of measurements to be increased. It was tested for the satellite sensor data in 1998-2000 and showed high efficiency, versatility, and reliability of its operation. Thus, since 1999, we have used two algorithms for fire detection from satellites: the basic IAO algorithm and the backup ISTP algorithm. We believe that this approach to monitoring with the use of two different algorithms increases the reliability of the results of fire monitoring where fires are detected by both algorithms. In addition, while the basic algorithm offers advantages over the backup one, our experience shows that in some cases, the latter provides additional data on seats of fires, thereby improving the results of implementation of the former. Problem of sun glints. It is well known that rivers, lakes, wetland, and cloudiness produce sun glints in satellite images when the geometry of observations is unfavorable, and these glints may cause false alarms in the monitoring of forest fires. In the Tomsk Region, there are hundreds of rivers (the largest rivers are mapped in Figure 1), hundreds of lakes, and large wetland areas, and the average amount of clouds over this territory in the fire-hazardous season (May–September) is about 50%. Therefore, the problem of sun glints is very serious. This fact is supported by the data recorded on June 19, 2000 and tabulated below (chart 1). Chart 1. Number of hotspots in satellite images recorded on June 19, 2000 NOAA satellites

Image LT

15 14 12 15

02:10 17:03 17:50 20:01

Sun elevation, deg 33.9 44.9 39.4 22.6

Number of hotspots T3>305 K 3407 8533 11765 681

T3>310 K 263 333 4387 17

T3>315 K 2 7 2206 2

T3>320 K 0 4 996 0

Notes: • LT is the local time of recording of satellite images (LT=GMT+8), • T3 is the radiant temperature in the AVHRR band 3.

On that day, the number of hotspots increased by several thousand within a very short period (between 17:03 and 17:50) due to sun glints. The frequency of glint recurrence in satellite images of this region is 5–10% of the number of all images. These situations are most typical of postmeridian images (more than 70% of all images). In the literature [6,7, 10], the following criteria for rejecting sun glints are described: (1) the angle between the hotspot-to-sensor vector and the specular reflection vector is small (~10°) and (2) albedos measured in visible channels exceed the preset hotspot threshold (~0.3). An analysis of the map in Figure 1 suggests that most forest fires burned in the immediate proximity of rivers or other water reservoirs. Another important circumstance follows from the statistics of our measurements: more than 75% of false alarms caused by sun


7

glints in satellite images are from 1 to 3 pixels in size. Consequently, for boreal forests in the Tomsk Region, we must not only solve the problem of sun glint rejections but also distinguish between neighboring small-size sun glints and fires in satellite images. Our experience demonstrates that in our case, the use of the above-enumerated criteria alone does not solve these problems with sufficient accuracy. Therefore, we use a more complex procedure of automatic rejection of sun glints including: 1. specialized spatial analysis of satellite measurements, 2. threshold values and albedos A1 and A2 as well as their relationships determined for the sensor–hotspot–sun geometry of the Tomsk Region, 3. comparison with the previous image in which sun glints are absent, 4. hydrographic data of an electronic map of boreal forests in the Tomsk Region. Practical application of this procedure demonstrated its high (close to 100%) efficiency.

1.2. RESULTS The results of satellite monitoring of boreal forests in the Tomsk Region performed at the IAO SB RAS in 1998–2008 were analyzed using specially developed software (described in detail in [39]). Fire location data from the Tomsk Forest Protection Services were used as reference data. To compare the satellite sensor data with the reference data, the following characteristics of fires were used: • • •

geographic coordinates, date and time of fire detection (TDET), localization (TLOC), and extinguishing (TEXT), fire areas at the time of their detection (SDET) and extinguishing (SEXT).

To estimate the efficiency of early detection of forest fires from satellites, the fire duration was defined as [T0, TEXT], where T0 = TDET – ΔT (ΔT depends on SDET). These characteristics are illustrated by Figures 3 that show the following statistical data for 1998– 2008: (а) histograms of fire distributions over the areas SEXT and histograms of fire distributions over the duration [TDET, TEXT].

16

Number of fires, %

Number of fires, %

32

8 4 2 1 0,5

0,1

1

10

100

Area, ha

Figure 3. Continued on next page.

1000

100 90 80 70 60 50 40 30 0,1

1

10 100 Area, ha

1000


16

Number of fires, %

Number of fires, %

8

8 4 2 1 0,5

0,1

1

10

100 90 80 70 60 50 40 30 20 10

Duration, day

0,1

1 10 Duration, day

Figure 3. Size distribution of forest fires and duration distribution of forest fires with the corresponding integral functions.

It follows that small-size fires with areas less than 1 ha account, on average, for about 53% of all fires, and fires with short duration ( 50 ha. The minimum area of forest fires detectable from satellites is about 0.1–0.2 ha, and the probability of detection of these fires is about 10%.

PFDET, %

(a) Efficiency of SMFF 100 90 80 70 60 50 40 30 20 10 0 0,1

1

10

100

1000

Area SEXT, ha

(b) Efficiency of EDFF

PFED, %

100 90 80 70 60 50 40 30 20 10 0 0,1

1

10

100

1000

Area SDET, ha Figure 4. Efficiency of SMFF (a) and EDFF (b).

The experience accumulated by Forest Protection Services demonstrates that a nearground fire with an area less than 5 ha can be extinguished with high probability. Turning to Figure 4b, we can notice fairly high (>15%) efficiency of early detection of fires already at SDET ≈ 1 ha, increasing to 35–45% for SDET ≈ 5 ha. Thus, in our case the satellite sensor data can be used efficiently for early detection of fires, when their extinguishing is not very expensive. Imaging Time-of-Day Effect. In this section, we examine the information content of satellite images for monitoring forest fires as a function of the time of day. To this end, all of the SMFF data was divided by the types of satellites (NOAA-12, -14, -15) and orbits (antemeridian and postmeridian). The urgency of this problem is motivated by the fact that


11

some monitoring procedures used in Russia are based only on postmeridian images or even on a single daytime image. This approach is caused by the evident circumstance that postmeridian images are most informative and close in time to breaking-out fires and spreading of ones already burning . Chart 3. Results of satellite monitoring of forest fires as functions of the time of day Orbits

Antemeridian (a.m.)

Satellites

NOAA12

NOAA-14

Postmeridian (p.m.) NOAA-15 NOAA-12

NOAA-14

NOAA- Total 15

1998 LT

08:14– 10:02

04:56–06:50 –

18:28–20:12

15:08–16:47

–

NIMG

91

93

–

86

83

–

PFDET,%

24.7

34.0

–

76.3

80.4

–

94.8%

PFED,%

9.3

23.3

–

34.9

72.1

–

85.7%

LT

07:52– 09:37

05:39–07:30 –

18:04–19:49

15:44–17:36

20:01– 21:40

NIMG

150

148

–

153

152

153

PFDET,%

22.2

23.7

–

74.7

67.0

57.7

96.4%

PFED,%

19.8

14.0

–

54.7

52.3

48.8

91.9%

LT

07:31– 09:45

06:21–09:00

09:51– 12:01

17:44–19:45

16:20–18:20

19:59– 21:54

NIMG

121

132

81

126

141

83

1999

2000

PFDET,%

13.0

15.6

26.0

66.2

75.3

46.8

94.8%

PFED,%

5.9

0.0

32.4

55.9

55.9

35.3

85.3%

Notes: LT is the local time of recording of satellite images (LT=GMT+8), NIMG is the number of satellite images, the total SMFF results (PFDET) and the efficiency of early detection (PFED) are given in per cent of the number of all fires throughout the year (see Table 2).

The data in Chart 3 demonstrates the SMFF results as functions of the time of day. Antemeridian images have relatively low efficiency, whereas postmeridian images are characterized by far higher (by a factor of 2–3) efficiency. However, from Chart 3, it is clear that the use of only one (even the most informative) image reduces the efficiency of satellite fire monitoring by 20–25%, and the efficiency of early detection decreases 1.5–2 times. The last column in Chart 3 gives the total efficiency of satellite monitoring with the use of only postmeridian images. It follows from this data that about 95% of the SMFF results and more than 85% of the early detection results have been detected using only the postmeridian images. Thus, even though the information content of postmeridian images is high, some useful satellite sensor data is lost due to the rejection of antemeridian images, and the efficiency of early detection of fires decreases noticeably. Consequently, only a full SMFF

12


procedure including all satellite images provides a maximum SMFF efficiency and real-time monitoring of the fire dynamics and state of cloudiness in the site of fire. Comparison with MODIS Fire Product (MOD14). Another important element of estimating the efficiency of the technology of real-time forest fire detection from space developed at the IAO SB RAS is its comparison with other algorithms widely used in other centers of satellite monitoring. The MODIS Fire Product [10] belongs to these global algorithms. The algorithm uses brightness temperatures derived from the MODIS 4-and 11-μm channels, denoted by T4 and T11, respectively. The MODIS instrument has two 4-μm channels numbered 21 and 22, both of which are used by the detection algorithm. Channel 21 saturates at nearly 500 K; channel 22 saturates at 331 K. Since the low-saturation channel (22) is less noisy and has a smaller quantization error, T4 is derived from this channel whenever possible. However, when channel 22 saturates or has missing data, it is replaced with the highsaturation channel to derive T4. T11 is computed from the 11-μm channel (channel 31), which saturates at approximately 400 K for the Terra MODIS and 340 K for the Aqua MODIS. The 12-μm channel (channel 32) is used for cloud masking; brightness temperatures for this channel are denoted by T12. After the cloud- and water-covered pixels are identified, the potential fires are determined with the use of three conditions: T4 > 310 K, 2) ΔT=T4 – T11> 10 K, 3) ρ0.86 < 0.3, where T4 and T11 are brightness temperatures in channels 21/22 and 31 of the EOS/MODIS sensor, and ρ0.86 is the reflectance in channel 2 of this sensor. Then for background pixels adjacent to the potential fires, the following statistical characteristics are determined: mean values (T4*, T11*, ΔT*) and mean absolute deviations (μ4, μ11, μΔT) for T4, T11, and ΔT, respectively. Further, the pixels flagged as potential fires are examined through the series of tests: Test 1. T4 > 360 K (320 K for nighttime pixels). Test 2. ΔT > ΔT* + C1μΔT. Test 3. ΔT > ΔT* + C2. Test 4. T4 > T4* + C3μ4. Test 5. T11 > T11* + μ11 – C4. (C1 = 3.5, C2 = 6.0, C3 = 3.0, С4 = 4.0). After testing, some pixel is classified as a fire, providing the following conditions are fulfilled: a) test 1 or (test 2 + test 3 + test 4 + test 5) for daytime pixels, b) test 1 or (test 2 + test 3 + test 4) for nighttime pixels. To perform our comparative analysis, we took 1610 MOD14-type granules of the firedangerous season in 2003. These files contained results of detecting high-temperature anomalies, including geographical coordinates of “hot” pixels, satellite measurements at these


13

points, and statistical characteristics in their vicinities. The official data of Tomsk Fire Protection Service on forest fires in the territory of the Tomsk Region in June–September, 2003 was used as test data. The efficiency of the technology of early forest fire detection from space developed at the Institute (at least, on the regional level) is illustrated by Chart 4 presented below. It gives the data on the number of forest fires detected in the territory of the Tomsk Region obtained by processing of the AVHRR (IAO SB RAS) and MODIS (MOD14) satellite data. Chart 4. Results of comparison of the efficiency of satellite fire detection from the AVHRR (IAO) and MODIS (MOD14) data in the Tomsk Region in 2003 The number of early detected fires is given in the parentheses June

July

August

September

Total

AVHRR / IAO

16 (7)

60 (22)

82 (37)

28 (11)

186 (77)

MOD14

7 (4)

28 (11)

53 (16)

10 (6)

98 (37)

MOD14/ Т

6 (3)

20 (6)

43 (13)

9 (6)

78 (28)

MOD14/ A

6 (4)

21 (7)

40 (8)

7 (4)

74 (23)

32

(b) Efficiency of EDFF

(a) Efficiency of SMFF NOAA (IAO) MOD14

28

20

Number of fires

24 15

20 16

10

12 8

5

4 0

0 0,1

1

10

100

Area, ha

1000

0,1

1

10

100

1000

Area, ha

Figure 5. Results of comparative analysis of the efficiency of forest fire detection using the MOD14 and IAO algorithms

Here, we have used the following designations: MOD14 indicates that we used the MODIS data of both satellites (Terra and Aqua), MOD14/T indicates that we used only the data of the Terra satellite, and MOD14/A indicates that we used only the data of the Aqua satellite. The data given in Chart 4 is supplemented by its graphic illustration (Figure 5), where the estimates of the SMFF efficiency are given for both methods depending on the site area. The data presented in the Chart above and in the figure gives us grounds to conclude that the efficiency of the technology of early detection of forest fires from space developed at the IAO

14


SB RAS is at least on the regional level, twice higher than that of the MODIS Fire Product algorithm used in the routine operation mode on the global scale.

PART 2. SOLUTION OF PROBLEMS OF THE TEMPERATURE MONITORING OF THE EARTH’S SURFACE FROM SPACE ON THE BASIS OF THE RTM METHOD 2.1. Formulation of the Problem of Fire Detection From Space We now formulate basic relationships of the algorithm of reconstructing the brightness characteristics of a small-sized fire in the “surface + atmosphere + fire” system. Let a hightemperature object (fire) characterized by area SF and temperature TF (TF > 600 K) be located on the surface of area SFOV (SF dB will be independent of the optical-geometrical conditions of observations. Thus, the use of the RTM approach in practice for detection of small-sized fires requires the fulfillment of the following key conditions: • • • •

adequate model of IR radiative transfer through the atmosphere, real-time information: meteorological and optical parameters (of required volume and accuracy) of the atmosphere, geometry of the observations, “fast” software for atmospheric correction of the satellite IR images, retrieval of the land surface temperature (LST) in the channels λ = 4 μm and λ = 11 μm.

Next, it is necessary to consider preliminarily the following questions: 1. To obtain statistical data on the variability of the values Pλ and IBG for the NOAA/AVHRR infrared channels. 2. To estimate the relative contribution of ISRF, IATM, IRFL, and ISCT to background radiance IBG. 3. To demonstrate the dependence of ISCT on the geometry of observations (θ, φ, Z). To attain the first three objectives, we simulated numerically the values Pλ, ISRF, IATM, IRFL, and ISCT with the use of the LOWTRAN-7 computer code. In these computations, we used the actual meteorological parameters of the atmosphere and the geometry of satellite observations (θ, φ, Z) for Tomsk (56°30′ N, 85°00′ E) in May–September 1998-2000. More than 1300 situations were considered. The data was obtained from two satellites (NOAA–12 and NOAA–14) for two orbit types (antemeridian and postmeridian). The underlying surface was assumed Lambertian. The surface emissivities in the AVHRR IR channels 3 and 4 were ε3.7S = 0.96 and ε11S = 0.98. The near-surface air temperature was set equal to the underlying surface temperature TS. The data presented in Chart 5 were recorded in channel 3 (λ = 3.75 μm) of the NOAA/AVHRR. Based on their analysis, we can conclude the following: a) Total contribution of transmitted surface radiance ISRF and atmospheric radiance IATM to IBG, which linearly decrease with increasing aerosol optical thickness (АОТ) τλaer, dominates. b) Contribution of IRFL is about 10% and decreases linearly with the АОТ. c) Contribution of ISCT increases linearly with the АОТ. d) Even a significant growth of the AOT changes IBG at most by 2-3%.

16


Chart 5. Atmospheric transmittance and background radiance (λ = 3.75 μm) averaged over the period of observations (Tomsk, May–September 1999) Visibility (Vis)

IBG

Pλ

τλaer

mol (no aerosol) 40 km, rur 40 km, urb 20 km, rur 10 km, rur 5 km, rur 2 km, rur 2 km, urb mol vs 2 km, rur mol vs 2 km, urb

0.46435 0.46510 0.46356 0.46592 0.46691 0.46854 0.47356 0.45236 +1.98 % - 2.58 %

0.74648 0.73267 0.72971 0.71791 0.68934 0.64510 0.53298 0.49339 - 28.60% - 33.90%

0 0.01867 0.02272 0.03902 0.07963 0.14596 0.33688 0.41407

Relative contribution to IBG (%) ISRF+IATM IRFL ISCT 89.77 10.22 0.02 88.58 9.95 1.47 88.96 9.83 1.21 87.35 9.68 2.97 85.07 9.19 5.74 81.33 8.45 10.21 71.06 6.67 22.27 77.14 5.25 17.60

Symbols rur and urb denote rural and urban boundary-layer aerosol models; values 2 –1 of background radiance are expressed in mW/(m ⋅sr⋅cm )

The temporal variability of the atmospheric transmittance and the background radiance in channel 3 (λ = 3.75 μm) and channel 4 (λ = 10.8 μm) of the NOAA/AVHRR can be estimated from the simulated data. We can conclude the following: •

The range of variations and the standard deviation of the atmospheric transmittance P3.75 are three times less than those P10.8. This can be easily explained by a stronger dependence of P10.8 on the atmospheric temperature and humidity that possess high spatiotemporal variability. • On the other hand, the relative variability of background radiance IBG in the channel with λ = 3.75 μm is three times less than in the channel with λ = 10.3 μm. This is due to the fact that the relative temperature variations of Planck’s function Bλ(T) are by a factor of 2.8 greater for λ = 3.75 μm compared to λ = 10.8 μm. From the viewpoint of complexity of computational algorithms, the volume of the required a priori information and difficulties in assigning actual a priori information with required accuracy, the calculation of IRFL and ISCT is the most labor-consuming problem. It is based on knowledge of: 1) Geometrical conditions of observations (the field-of-view angle θ, solar zenith angle Z, and relative azimuth angle of measurements φ). 2) Meteorological parameters of the atmosphere. 3) Optical characteristics of the atmospheric aerosol. In this regard, we consider some results of numerical simulation of the ISCT value.


17

Figure 6. Dependence of the scattered radiance TSCT on geometry of satellite observations; rural boundary-layer aerosol, V = 5 km.

To study the peculiarities of the behavior of the scattered solar radiance in the channel with λ = 3.75 μm, the ISCT value was simulated numerically using the LOWTRAN-7 computer codes for a cloudless atmosphere and the following satellite observation conditions:

a) Meteorological parameters for the atmospheric model for the midlatitude in summer b) Rural and urban boundary-layer aerosol models (visibility range V = 50 - 2 km) c) Geometry of observations: θ = 0–55°, HS = 90°, Z = 75–0° (solar elevation angle), and φ = 0–180°. As we can see in Figure 6, the results of modeling demonstrate a complex dependence of the ISCT value on the satellite observation conditions. For convenience, the ISCT value in the figure is expressed as an increment TSCT to the brightness temperature Tλ. First, it should be noted that the azimuth dependence of TSCT value becomes more pronounced with increasing scanning angle and decreasing V value (increase of the atmospheric turbidity). For solar elevation angle HS of the order of 10° and azimuth angles φ < 50°, an abnomalous local maximum of the TSCT value is observed. The amplitude of this maximum depends on θ and on the optical characteristics of the boundary-layer aerosol. An analysis of the NOAA satellite data on the relationship between the geometrical parameters HS and φ for the Tomsk Region demonstrated the abnormal growth of TSCT. This must be taken into account when correcting satellite measurements for the distorting effect of the atmosphere.

18


In this part, we study how the quality of the a priori meteorological information (AMI) affects the accuracy of reconstructing the object radiance from satellite data in the spectral range 3.5–4 μm. The atmospheric transmittance and the characteristics of upwelling radiation in the NOAA/AVHRR infrared channels were calculated for the atmospheric conditions in Tomsk with allowance for the actual geometry of satellite observations and meteorological parameters of the atmosphere specified in accordance with the IAO SB RAS data for the period of May–September 1998–2000. Satellite measurements were simulated with the use of meteorological data closest in time to satellite observations. A high-temperature object with a temperature of 800–1200 K and area of 10–1000 m2 was simulated within the radiometer field of view. Different types of AMI and sources of information about the background surface temperature TS were used for atmospheric correction of the simulated “satellite measurements.” As a result, a correlation was found between the AMI characteristics and the accuracy of radiance reconstruction for a high-temperature object. For an object with an area smaller than 100-200 m2, the results demonstrate the marked effect of the AMI quality on the results of reconstruction. Depending on the AMI type, the RMS value of the reconstructed radiance may vary several times, but remains at least half as small as that without atmospheric correction.

2.2. RETRIEVAL OF THE LAND SURFACE TEMPERATURE In the past 25 years, of concern has been the active development of satellite methods of LST retrieval [15-[21]], referred to as “split-window methods” (SW) . As a part of this approach, the IR measurements in two spectral channels of a “split” atmospheric transparency window of 10–13 μm are used and the well-known method of differential absorption is implemented to account for the distorting water vapor effect. Application of the SW algorithm [22] is based on linear relations between LST and satellite measurements in two spectral channels close to 11 and 12 μm. The relations can also include values of the underlying surface emissivity ελ for these channels. Parameters of these relations are calculated from model data or data of combined analysis of LST satellite and ground measurements. As an example, we can consider the standard MODIS algorithm for LST remote measurement [20]:

T0 = C + α(T11 + T12)/2 + β(T11 – T12)/2, α = A1 + A2(1 – ε)/ε + A3(Δε/ε2), β = B1 + B2(1 – ε)/ε + B3(Δε/ε2), ε = (ε11 + ε12)/2, Δε = (ε11 – ε12)/2, where coefficients Ai and Bi (i = 1, 3) depend on the satellite zenith angle and the integral moisture content of the atmosphere. In practice, these algorithms are very simple and efficient for global LST monitoring. However, their users cannot disregard a number of serious practical limitations:


19

1) The LST retrieval error (δTS) strongly depends on errors of δTλ measurements. For instance, it is reported [[20]] that δTS ≈ 6.19δTλ. For NOAA/AVHRR instrument, δTλ ≈ 0.12 K, i.e., δTS ≈ 0.7 K; while for EOS/MODIS system, δTλ ≈ 0.05 – 0.07 K, i.e., δTS ≈ 0.3 – 0.4 K. 2) The surface emissivities ε11 and ε12 as well as their difference Δε should be well known. It is underlined in [16,17] that at δTS ≈ 0.5 K the relative error δε of ε specification should be no more than about 0.5 – 1%, and for Δε it should be no worse than 0.25 – 0.5%. 3) The coefficients of the algorithms are determined only for a given range of “standard” situations in the clear-sky atmosphere. 4) The algorithms take into account the thermal absorption by the water vapor; at the same time, the distortions caused by aerosol and cirrus clouds, are ignored. Thus, the standard LST retrieval algorithms used in practice, do not provide for confident and universal solution of the problem of atmospheric correction of IR measurements, especially under complex (non-standard) observational conditions. Another more correct approach is in the use of thermal radiative transfer models. The RTM method accounts for the distorting characteristics of the atmosphere with the use of widely known computer programs of the type of LOWTRAN, MODTRAN, ATCOR, etc., on the basis of the a priori optical-meteorological information on the atmospheric state at the moment of satellite observations. As examples of such approach, the atmospheric correction of the MSU-SK, NOAA/AVHRR, Landsat, ASTER [1,3 ,23 -26] radiometer data can be cited. Undoubtedly, this approach offers the universality and explicit accounting for all distorting factors in solution of the problem of LST retrieval from space, though its practical implementation requires invoking a large amount of real-time a priori information of the required quality and high-speed calculations. The intensive development of computation methods and modern technologies of parallel computer programming [27] eliminates labor consumptions of enormous computations. Moreover, a combined approach was suggested [28]: the fast SW method for standard situations and the RTM method for situations beyond the standard limits (in the presence of aerosol and semitransparent or cirrus clouds). The software package is also described in the same work, allowing the user, by means of accessible facilities (IMAPP and MODTRAN) and on the basis of EOS/MODIS satellite information, to employ the RTM method for integrated temperature monitoring of the Earth’s surface, including LST retrieval and monitoring of high-temperature objects (HTO), i.e., fires and industrial thermal sources.

2.3. DISTORTIONS OF THERMAL RADIATION BY MOLECULAR ATMOSPHERE It is well known that the main factors of the molecular distortion of thermal radiation in EOS/MODIS channels include: the selective absorption by spectral lines of atmospheric gases and continuum absorption by line wings of H2O and N2. Though estimates of the influence of these factors on characteristics of upward fluxes of thermal radiation are available in the

20


literature, the permanent development of thermal radiative transfer models necessitates some improvement of these estimates for a tighter relevance to the problems of the land surface temperature retrieval. We used for these purposes the well-known software package LBLRTM_v11.3 (11/2007) [29], built upon the spectral line database HITRAN-2004 [30] (including all changes made before January 1, 2007) and molecular continuum models MT_CKD_2.1 [31].

2.3.1. Selective Absorption by Spectral Lines of Atmospheric Gases An analysis of the HITRAN-2004 data on the total intensity of the molecular spectral lines and integrated gas content WGAS allows one to separate out from the total list consisting of 39 molecules the optically active molecules (in the considered EOS/MODIS spectral channels), which determine the required accuracy of LST retrieval by the RTM method: H2O, CO2, O3, N2O, and CH4. Figure 7 presents the absorption functions of thermal radiation calculated with the use of LBLRTM_v11.3 in the considered EOS/MODIS channels. The influence of the selective absorption by each molecule (and their sum) on the accuracy of the RTM method can be estimated quantitatively by calculating the change of the radiation (brightness) temperature measured in satellite channels provided the chosen molecule is not taken into account in the line-by-line (LBL) calculations.

Figure 7. Continued on next page.


21

Figure 7. Absorption of thermal radiation in EOS/MODIS spectral channels 20, 21/22, 31, and 32. Midlatitude summer. Shade of grey shows continuum; peaks are for lines + continuum.

Thus, it is necessary to determine the difference δTλ(mol) = Tλ(∑) – Tλ(∑–mol), where Tλ(∑) and Tλ(∑–mol) are calculated brightness temperatures, for which either all absorbing components (∑) are taken into account or the chosen molecule (∑–mol) is not taken into account. Chart 5. Optical depth τ of atmospheric gases and distortion of brightness temperature δTλ (mol), K. Midlatitude summer Spectral channel 20 δTλ τ LBLRTM_v11.3 data H 2O 0.1267 0.935 CO2 0.0013 0.022 O3 N 2O 0.0175 0.319 CH4 0.0102 0.145 Other molecules All molecules 0.1572 1.420 All (tropics) 0.1949 1.923 MODTRAN_v3.5 data H 2O 0.1288 0.926 All molecules 0.1584 1.416 All (tropics) 0.1969 1.917 Molecules

21 τ

δTλ

31 τ

δTλ

32 τ

δTλ

0.0035 0.0017

0.020 0.027

0.0859 0.0027

0.685 0.034

0.0827 0.0049 0.0003

0.662 0.070 0.015

0.0174 0.0045

0.260 0.067

0.0273 0.0298

0.376 0.436

0.0915 0.1209

0.857 1.151

0.0007 0.0889 0.1155

0.015 0.765 0.994

0.0040 0.0315 0.0342

0.022 0.429 0.496

0.0756 0.0876 0.1126

0.615 0.969 1.231

0.0911 0.0979 0.1239

0.772 0.896 1.119

Chart 5 presents the δTλ, estimates allowing us to draw certain conclusions: 1) First of all, obviously, the influence of the selective absorption by atmospheric gases in all EOS/MODIS channels exceeds the 0.25 K level and hence, should be accounted for within the RTM method. 2) In channels 20 and 21, the distorting effect of the selective absorption is determined by lines of H2O, N2O, and CH4 molecules.

22

G.G. Matvienko, S.V. Afonin and V.V. Belov 3) In channels 31 and 32, it is sufficient to take into account only the contribution of H2O lines, with much less accounting for the CO2 line contribution.

Thus, in the framework of the RTM method, the problem of fast specification of the confident a priori information concerns only the temperature and humidity profiles. One more important condition of the successful use of the RTM method in practice is good accuracy and high speed of calculation of the selective absorption coefficients in processing of large-volume satellite information. Obviously, the direct use of the LBL methods in the framework of the RTM method is impossible in view of their laboriousness; therefore, it is advisable to use simplified radiative transfer models, tested in practice and accessible to wide user community, such as the commonly known MODTRAN program. Presently, the program MODTRAN_v4.x [33] is a commercially available product; however, its predecessor MODTRAN_v3.x [32] and its codes are accessible (that is important) to users. Chart 5 presents δTλ(mol) calculations with the use of the program MODTRAN_v3.5, based on the parameters of spectral lines from HITRAN-96 [34] and models of molecular continuum CKD_v2.1_rev.3.3 [35]. The comparison of δTλ(mol) values, obtained by MODTRAN_v3.5 and LBLRTM_v11.3, suggests that this data disagrees by less than 0.15 K, reasonably meeting the practical accuracy requirements to the satellite-borne LST retrieval.

2.1.2.Continuum Absorption by Spectral Line Wings of Atmospheric Gases According to the MT_CKD_v2.1 model, in addition to the selective absorption of the thermal radiation by lines inside the spectral channels, a marked influence is exerted by the continuum absorption by line wings of intense H2O, CO2, O3, and N2 bands lying outside these spectral channels (see figure 7). The 3.5–4 μm atmospheric transparency window is characterized by a weak H2O continuum, while at wavenumbers ν < 2600 cm–1, there is a stronger N2 continuum. In the 10–13 μm transparency window strong H2O continuum is dominated, with a simultaneous presence of weak CO2 continuum. The H2O continuum is presented by two components corresponding to self-broadening of lines (H2O– H2O) and air-caused line broadening (H2O–AIR). To quantitatively estimate the influence of each continuum component, we calculated δTλ(cont) by analogy with δTλ(mol): δTλ(cont) = Tλ(∑) – Tλ(∑–cont), where Tλ(∑) and Tλ(∑– cont) are the calculated brightness temperatures, for which all absorbing components (∑) are taken into account or the chosen continuum component (∑–cont) is not accounted for. Chart 6 presents the results of δTλ(cont) calculations, whose analysis allows us to make the following conclusions. 1) The effect of H2O and CO2 continuums on the brightness temperature in channels 20 and 21 is less than 0.05 K. The effect of N2 continuum in channel 20 has the same order of magnitude; however, it markedly increases in channel 21, exceeding a level of 1 K. 2) The component H2O–H2O (≈ 1–2 K) dominates in channels 31 and 32 at a much less (than 0.2 K) influence of the component H2O–AIR. The effect of CO2 continuum is almost insignificant (less than 0.01 K).


23

3) Comparing the δTλ(cont) values, obtained via LBLRTM_v11.3 and MODTRAN_v3.5 programs, we see that they disagree in channels 20, 31, and 32 by less than 0.1 K, and they increase up to 0.2 K due to N2 continuum only in channel 21. Chart 6. Optical depth τ of molecular continuum and distortion of brightness temperature δTλ (cont), K. Midlatitude summer Spectral channel 20 δTλ τ LBLRTM_v11.3 data H2O–H2O 0.0019 0.008 H2O–AIR 0.0026 0.014 CO2 0.0003 0.003 LBLRTM_v11.3 data N2 0.0058 0.067 All 0.0106 0.093 All (tropics) 0.0131 0.119 MODTRAN_v3.5 data H2O–H2O 0.0044 0.020 H2O–AIR 0.0034 0.018 N2 0.0074 0.093 All 0.0152 0.131 All (tropics) 0.0199 0.173 Components

21 τ

δTλ

31 τ

δTλ

32 τ

δTλ

0.0029 0.0001 0.0034

0.013 0.000 0.041

0.2959 0.0112 0.0001

1.400 0.072 0.002

0.3956 0.0278 0.0003

1.825 0.174 0.006

0.1052 0.1115 0.1135

1.251 1.309 1.454

0.3072 0.5525

1.483 3.064

0.4237 0.7558

2.032 4.060

0.0061 0.0002 0.1174 0.1236 0.1278

0.027 0.001 1.445 1.475 1.647

0.3188 0.0008

1.491 0.005

0.4390 0.0049

1.986 0.030

0.3196 0.5849

1.497 3.170

0.4439 0.8099

2.022 4.144

2.1.3. Influence of Errors in Setting Profiles of Meteorological Parameters To date, the current databases of spectral line parameters, molecular continuum models, and thermal radiative transfer models, overall, ensure high accuracy of accounting for the distorting influence of the atmosphere, when using the a priori valid information on the key meteorological parameters of the atmosphere X(z), where z is the height. Since the vertical profiles of X(z) contain measurement (retrieval) errors δX(z), it seems reasonable to estimate the effect of these errors on the accuracy of the RTM method. The estimates were made as follows: 1) for the chosen profile of atmospheric meteorological parameters (e.g., meteorological model for the midlatitude summer), we calculated the brightness temperature Tλ(0); 2) some changes δX(z) were introduced in a given profile and the value of Tλ(δX) was calculated for the distorted profile; and 3) the difference δTλ(δX) = Tλ(0) – Tλ(δX) was calculated, being the measure of the influence of the inaccuracy in setting the meteorological parameters on the brightness temperature.

24

G.G. Matvienko, S.V. Afonin and V.V. Belov Chart 7. Change of brightness temperature caused by variations of the profiles of meteorological parameters: the air temperature δTAIR, the humidity δWH2O, and the minor atmospheric gas content δWGAS. LBLRTM_v11.3 data Parameter Midlatitude summer δTAIR = +2 K δWH2O = +20% δWGAS = +40% Tropics δTAIR = +2 K δWH2O = +20% δWGAS = +40%

Spectral channel 20 21

31

32

+0.206 −0.153 −0.168

+0.150 −0.010 −0.151

+0.632 −0.659 −0.068

+0.786 −0.820 −0.043

+0.241 −0.218 −0.186

+0.147 −0.020 −0.169

+0.968 −1.199 −0.075

+1.170 −1.418 −0.043

Chart 7 presents the results of calculations of δTλ(δX) for the air temperature and moisture content, as well as the content of other atmospheric gases. For the air temperature and moisture content, we have chosen δTAIR = +2 K and δWH2O = +20% at all atmospheric levels. They can be considered as characteristic retrieval errors of atmospheric meteorological parameters according to EOS/MODIS data [36]. For profiles of other atmospheric gases, we have chosen δWGAS = +40% as a certain limiting value. Thus, the data of Chart 7, overall, reflects the maximal effect of meteorological parameter profile errors on Tλ. Accounting for the limiting character of these estimates, we can make the following conclusions. 1) In channels 20 and 21, the influence of variations of profiles of all meteorological parameters in absolute value is less than 0.25 K, which, in principle, permits one in practice to optimize the volume of calculations of the distorting atmospheric parameters. 2) In channels 31 and 32, the value of δTλ for a given δWGAS does not exceed 0.1 K; therefore, setting of a priori information on the atmospheric content of minor gas constituents in these channels does not require high accuracy. At the same time, the effect of uncertain setting of the temperature and air humidity profiles is significant (δTλ > 0.5 K) for the correct treatment of the molecular distortion of the thermal radiation in the framework of the RTM method. Note that the absolute value of δTλ is determined by the degree of the thermal radiation absorption in the channel; therefore, |δTλ| is less in channel 31 than in channel 32. This circumstance can be used to compensate for the effect of imperfect setting of meteorological parameters in the RTM method. 3) Note that identical signs of δТAIR and δWH2O correspond to differently signed δTλ values. That is, in the presence of the positive correlation between δТAIR and δWH2O, this circumstance may lead to mutual error compensation in setting meteorological parameters, which are key ones to the atmospheric correction. Thus, the atmospheric correction of remote IR measurements of LST becomes possible on the basis of the meteorological information with relatively low accuracy characteristics.


25

The latter two conclusions should be complemented with some important comments. First, the analysis of the satellite methods of the temperature and humidity profiles retrieval [36] allows us to suppose with a high degree of confidence that errors in the retrieval of the temperature and humidity have just a positive correlation. Second, the difference between δTλ(δX) values in channels 31 and 32 allows us to propose the compensation for the δX effect through the application of the RTM method by the split-window principle, that is, the LST is to be determined as follows: TS = TS,31 – ΔTS; ΔTS = CERR(TS,32 – TS,31), where TS,31 and TS,32 are LST values retrieved in 31 and 32 channels and CERR ≈ 2.0 is the coefficient obtained from simulation calculations. This will ensure the RTM method resistance to uncertainties in setting the a priori meteorological information.

PART 3. APPLICATIONS OF THE RTM METHOD 3.1. Structure of the Program Complex Note first that two types of the basic software for thematic processing of information from EOS/MODIS system are known to the present time. The first type includes DRL licensing programs (Direct Readout Lab, GSFC/NASA), where basic algorithms are grouped in PGE (Product Generation Executive), which include the program texts and necessary data for their assembling, setting, and exploiting. In Russia, this software is successively used for many years at the Center of Space Monitoring of the Altay State University. The second type is presented by the well-known program package IMAPP (International MODIS/ AIRS Processing Package). The design, maintenance, and distribution of the Package are conducted under GNU General Public License at the Space Science and Engineering Center (SSEC), being a division of the University of Wisconsin–Madison (ftp:// ftp.ssec.wisc.edu / pub / IMAPP/MODIS/). To solve our problem, we chose the IMAPP v. 2.0 package and adapted it to the operation medium Windows. General scheme (Figure 8) of the designed software for thematic processing of the EOS/MODIS data includes three stages. 1) At the initial stage (levels 0 and 1), the satellite file EOS/MODIS is unpacked with the help of IMAPP program from PDS format to a set of HDF-EOS formats; the geographic assignment of data and calibration of the space measurements are performed. 2) At the second stage (level 2), the a priori information from MODIS on the parameters of the atmospheric state is prepared for processing. This information includes: – – – –

cloud mask (MOD35), optical characteristics of aerosol (MOD04), total column precipitable water vapor (MOD05), characteristics of clouds (MOD06),

26

G.G. Matvienko, S.V. Afonin and V.V. Belov –

atmospheric profiles of the geopotential, temperature, humidity of air, and ozone content (MOD07).

Figure 8. Schematic view of the atmospheric correction of remote measurements of the LST with the use of the EOS/MODIS satellite system.

3) Based on the a priori information, formed at stage 2, with the use of the program block “Atmospheric correction”, characteristics of distorting effects of the atmosphere are calculated and the space measurements of LST and its spectral reflectance are corrected. At present, this is made by the well-known program MODTRAN. To set values of the emissivity ελ in the IR channels at λ = 3.96, 11, and 12 μm of EOS/MODIS device, data of MODIS UCSB Emissivity Library (http: //www.icess.ucsb.edu /modis/EMIS/html/em.html) or the database Global Infrared Land Surface Emissivity Database (http://cimss.ssec.wisc.edu/ iremis/) is used.

3.2. AN EXAMPLE OF THE RTM METHOD APPLICATION To preliminary test the program complex performance, the temperature of an area of 60 × 60 km of the Luginetsk oil-gas condensate field (58.15° N, 78.89° E) was sensed under


27

different atmospheric conditions. Figure 9 shows space patterns of the area with spatial resolution of 250 m, resulted from the composition of three spectral channels of the visible range of EOS/MODIS space system for June 2 and 5, 2004. Torches of the oil-gas condensate field are located in the image center. A) Cloudlesss atmosphere (06/02/2004; 12:38 LT)

B) Smoke, cloudiness (06/05/2004; 13:15 LT)

Figure 9. MODIS space images of the Luginetsk oil-gas condensate field; geographical projection – Albers Conical Equal Area.

In image A, where the atmosphere is free of clouds and the aerosol has the background content, two basic types of surface are clearly seen: areas covered with vegetation (dark) and open areas (light). Analysis of a series of the cloud-free images allows us to conclude that the spatial distribution of the underlying surface temperature follows the image outlines and is sufficiently stable and that the temperature of light areas is by 2–3 K higher than that of dark ones. A different situation is seen in image B: the smoke of the forest fires, as well as dense and semi-transparent cloudiness noticeably distort spatial outlines of the underlying surface. Space images in Figure 9 are supplemented by data from Figure 10, which are constructed by MODIS photographs in Cartesian coordinates. For case A, open areas are outlined; the characteristic values of albedo measurements at λ = 466 nm (ρ466) and the brightness temperature at λ = 11 μm (T11) are marked. Bright

28


flames (λ = 466 nm) and the sand quarry (to the right) are well seen in the image center. The differences of brightness temperatures in the channels λ=11 and 12 μm characterize the scale of spatial dissimilarity of the atmosphere distorting properties in the IR spectral range. λ = 466 nm (ρ466)

contour: ρ466 = 0,1 λ = 11 μm (T11)

contours: T11 = 294.5 and 295.5 K Figure 10. Continued on next page.


29

contour: ρ466 = 0.1

contours: T11 = 288.5, 295, and 299 K Figure 10. MODIS space data: albedo values in the channel λ = 466 nm, brightness temperatures in the channel λ = 11 μm.

Analysis of the values T11 – T12 for case A allows a conclusion that distorting properties of the atmosphere in this case can be considered quasihomogeneous (one range of values 0.5– 1.3 K, r.m.s. = 0.11 K). Vice versa, in the case B, the values of T11 – T12 fall in the range 0.4– 7.1 K and r.m.s. = 0.71 K due to the smoke and cloudiness. Naturally, the results of operation of the basic SW algorithm [20] are of interest (a standard product is called MOD11_L2) in the both cases under consideration. This data,

30


obtained from the site Land Processes Distributed Active Archive Center (LP DAAC, http://edcdaac.usgs.gov/datapool/ datapool.asp), is presented in figure 11. In case of the cloudless atmosphere, the standard algorithm MOD11 retrieves the LST everywhere, excluding only a few pixels, because the cloud mask due to bright pixels (erroneously, in our opinion) fixes the presence of partial cloudiness. In this case, LST spatial structure is almost similar to the spatial structure of brightness temperatures. In case B, spatial structures of ρ466, T11, and LST are noticeably distorted by the smoke and cloudiness. The number of white pixels, where MOD11 data is absent, significantly increased. A) LST (MOD11_L2)

contours: LST = 297.5 and 298.5 K B) LST (MOD11_L2)

contour: LST = 302 K Figure 11. Continued on next page.

B') LST (correction)


31

contours: LST = 301 and 302 K B) λ = 11 μm (T11)

contour: T11 = 298.5 K Figure 11. MODIS space data: albedo values in the channel λ=0.466 μm, brightness temperatures in the channe l λ = 11 μm.

The retained T11 structures are outlined in the image. Note that the results of LST retrieval in this case, aside from omissions, contain explicitly underestimated LST values, falling into the limits of cloud outlines. Now we have to fill white omissions in LST in order to retrieve the temperature structure similarly to the case of the cloudless atmosphere through the atmospheric correction of data at the areas, where the thermal radiation passes mostly through the smoke and aerosol. The designed program complex was used for this purpose. As the a priori information, we used the IMAPP-retrieved data from MODIS measurements. Using the MODTRAN

32


program in [28], the distorting characteristics of the atmosphere were calculated for the channel λ = 11 μm (the bottom in Figure 10), measurements of the brightness temperatures T11 were corrected, and new spatial LST distribution B′ was found. After the correction, a part of the temperature structure of the area, centered at the point (Y = 25, X = 52), was retrieved. Note an important fact that the difference between the retrieved LST values for B and B′ in the vicinity of this point does not exceed 1 K. In addition, the surface temperature near torches (Y = X = 30) was also retrieved. Consider one more result of the program complex application, i.e., the retrieval of the thermal brightness temperature of flames TF from measurements of brightness temperatures T4 in the channel λ = 3.96 μm, accounting for (based on IMAPP data) the optical-meteorological state of the atmosphere [37]: B(Tλ) = IF + IBG,

BF = S(θ)ελFB(TF)Pλ , IBG = ISRF + IATM + IRFL + ISCT, BF = [B(Tλ) − IBG]/Pλ,

where B(Tλ) is the Planck function; Tλ is the brightness temperature of the thermal radiation; IF is the intensity of flame emission attenuated by the atmosphere; IBG is the background radiation intensity; ISRF is the contribution of the surface thermal radiation attenuated by the atmosphere; IATM is the contribution of thermal radiation of the atmosphere; IRFL is the contribution of the incident flows of thermal and solar radiation, reflected from the surface; ISCT is scattered (thermal and solar) radiance along the path; Pλ = exp{–τλ} is the atmospheric transmittance; τλ is the optical thickness of the atmosphere; S(θ) is the ratio of the flame area to pixel size. Chart 8 illustrates the results of problem solution provided the flame diameter is 19 m. Chart 8 Case A B

LST, K 300.0 302.0*

Pλ, τλ 0.822 (0.196) 0.344 (1.067)

T11, K 297.4 293.8

T4, K 334.0 321.0

TF, K 1252 1261

* LST = T11,cor – the temperature is retrieved after atmospheric correction of measurements in the channel λ = 11 μm with the help of the designed program complex (see Figure 11, case B′).

Thus, retrieving results for flame temperature are very close despite different atmospheric conditions during space observations. It should be noted that excluding this operation (atmospheric correction) results in significantly different values for cases A and B: 1192 К and 956 К, i.e., the overestimation for case B exceeds 300 K.


33

3.3. APPLICATION OF THE RTM METHOD TO DETECTION OF HIGHTEMPERATURE OBJECTS The RTM method was tested using data of 97 files (granules) of the telemetric information from EOS/MODIS (Terra satellite, daytime images) for June 2006, pertaining to the West Siberian territory. As test objects for observations, we have chosen 13 flames from combustion of accompanying gas in oil-gas fields of Tomsk and southern Tyumen Regions. The choice of flames was determined by their stability and availability of their geographic coordinates, necessary for the torch identification. Thus, the flames were a set of varyingintensity thermal objects, allowing the effective elaboration of the methods of satellite-based HTO detection under different conditions of satellite observations. For elaboration of satellite methods, we used two variants of the standard algorithm MOD14_v5.0.1 [10], as well as the RTM methods with the use of our methodical innovations and software [28]. To increase the sensitivity of this algorithm in detection of HTOs with a relatively low intensity of thermal emission, we have modified the algorithm MOD14 with: a) considerably lowering the thresholds (5): T4 > 302 K (versus former 310 K) and ΔT > 3.5 K (versus 10 K); and b) changing the coefficients C1…C4 (6): C1 = 2.5, C2 = 5.0, and C3 = 2.0.

3.3.1. Description of the Algorithm Based on the RTM Method Stage 1. Based on the EOS/MODIS satellite telemetry, the IMAPP program is used to determine the a priori optical-meteorological information on the atmosphere state for regions of detecting high-temperature sources. The a priori information includes the following data: – – –

a spatial resolution of 1 km: the cloud mask (MOD35), the integral atmospheric moisture content (MOD05); a spatial resolution of 5 km: vertical profiles of the geopotential, the air temperature, humidity, ozone content (MOD07), and cloud characteristics (MOD06); a spatial resolution of 10 km: aerosol optical characteristics (MOD04).

Emittances of the pixels ελ are determined by the standard method based on maps of surface types and charts of the correspondence of ελ to these Earth’s surface types. Stage 2. The cases of water pixels, as well as pixels, covered with thick clouds, are rejected with the use of MOD35, MOD06, and MOD05 data. Stage 3. For channels 21/22 (henceforth, channel 21), 31, and 32, the a priori information, obtained earlier, is used to calculate the characteristics of the thermal radiation distortion by the modified version of program MODTRAN_v3.5. Then, based on the solution of thermal radiative transfer equation, TS,21, TS,31, and TS,32 are calculated, i.e., LST values, retrieved in channels 21, 31, and 32. To obtain correct temperature and humidity profiles in the absence of LST values, the condition of approximate equality TS,21 ≈ TS,31 ≈ TS,32 is to be satisfied.

34


Stage 4. If TS,31 ≠ TS,32, then one of the reasons for this are errors in profiles of the meteorological parameters. In this case, the simplest compensation of these errors is conducted through calculating corrections of the form ΔTS = CERR(TS,32 – TS,31) and of a new value of TS,31 = TS,31 – ΔTS. Stage 5. In the case of influence of cirrus and semitransparent clouds, the retrieved LST values are corrected: TS,21 = TS,21 + ΔT21,CLD, and TS,31 = TS,31 + ΔT31,CLD, where the “cloud” corrections are determined via Look-Up-Table of the influence of cloud characteristics on LST retrieval results and the mutual analysis of MOD35, MOD06, and MOD05 data. Stage 6. The HTO detection is performed with the use of two conditions: TS,21 > 302 K and ΔT = TS,21 – TS,31 > 3.5 K.

3.3.2. Detection Results Chart 9 presents the results of detection of test objects (flames) with the use of two (original and author-modified) algorithms MOD14, as well as the RTM method for the temperature monitoring of the Earth surface, proposed by us. Chart 9 gives results of flame detection, summed over all flames (N∑), the number of detections of each flame, and average temperature for each flame (T21,av). When testing the algorithm, a total of 38 128 pixels in the flame neighborhood were processed. Note that the condition TS,21 ≈ TS,31 ≈ TS,32 in the absence of clouds and HTOs does hold, signifying a good quality of the atmospheric correction of satellite LST measurements. For instance, for the sample, consisting of 30985 pixels, corresponding to conditions of the clear-sky atmosphere, average retrieved LSTs were: TS,21 = 298.4 K, TS,31 = 298.4 K, and TS,32 = 298.7 K. That is, the uncertainty in accounting for the molecular absorption in the EOS/MODIS channels 21, 22, 31, and 32 was, on the average, less than 0.5 K. The number of flame detections N∑ with the use of the MOD14 v5.0.1 algorithm was 60, with identification of 6 test objects from 13. For the algorithm MOD14_v5.0.1, modified by us, (MOD14*), N∑ = 83 at identified 10 test objects. With the use of the RTM method, N∑ reached 122, and all 13 test objects were observed at a varying frequency. Chart 9. Results of detection of 13 test objects (flames) from space with the help of three satellite methods Method

N∑

Torches F1 F2

MOD14

60

4

MOD14* 83

6

RTM T21,СР

122 13

F3

F4

F5

F6

F7

F8

F9

F10 X1

X2

X3

−

−

−

−

−

−

−

1

1

14

14

26

2

−

1

−

1

1

−

6

4

18

18

26

4

3

4

2

8

1

1

8

9

21

21

27

309 304 306 306 305 305 308 303 307 306 314 320 329

Thus, the RTM method is on the average by a factor of two more efficient than the standard algorithm MOD14_v5.0.1. In the modified algorithm version MOD14*, the detection


35

thresholds of potential fires coincide with thresholds in the RTM method. However, in this case again, the RTM method is far (almost by a factor of 1.5) more efficient than MOD14*. Speaking about comparative estimates of the efficiency of these three algorithms, it is very important to note the following. Among test objects we can distinguish three bright flames (X1…X3, see Chart 9), located in the south of the Tyumen Region, for which the detection frequency is markedly higher than for other flames at a less dependence on the choice of the method. Considering that the RTM method shows its main advantages in the detection of relatively low-intensity thermal sources, it is advisable to obtain comparative estimates of application of the methods to such sources, namely, ten flames (F1…F10, see Chart 9), located at the Tomsk Region. In this case, N∑ ratio for three considered algorithms is already 6:21:53 for MOD14, MOD14*, and RTM, respectively, therefore; advantages of the RTM method markedly increase. Let us compare the efficiency of the application of the RTM method and the algorithm used at IAO SB RAS [38,39], namely, the algorithm of the forest fire detection from data of the satellite system NOAA POES, when detecting low-intensity flames. In this case, the N∑ ratio will be 36:53 for the IAO algorithm and the RTM method, respectively. In the detection of high-temperature objects, the RTM method has considerable advantages over standard approaches, especially for the problem of detection of low-intensity sources under complex optical-meteorological observation conditions. Thus, among the considered satellite methods, the RTM method, proposed by us, is the most efficient. Then the IAO algorithm follows and two variants of the algorithm MOD14 conclude the list.

CONCLUSION The main results and data of satellite monitoring of boreal forest fires in the territory of the Tomsk Region are the following. 1. The effectiveness of the application of the AVHRR/NOAA satellite system for satellite monitoring of forest fires in this territory is in the range 19–48% (about 38% on average) and depends on the seasonal characteristics of fires themselves (size and time of burning) and state of cloudiness. 2. The probability of early fire detection from satellites is in the range 13–27% (about 18% on average). 3. The minimal sizes of the forest fires fixed in the regime of automatic AVHRR data interpretation is about 0.1–0.2 ha, and is detected with a probability of about 10%. For the limiting area of effective fire extinguishing (about 5 ha), the detection probability increases to 35–45%, which suggests the possibility of efficient application of satellite systems for early detection of fires in their early stages. 4. The efficiency of the regional algorithm developed at the IAO SB RAS is much greater than that of the MODIS Fire Product algorithm. 5. The maximum efficiency of satellite monitoring of forest fires is provided only by the complete SMFF scheme, including satellite image recording irrespective of the time of day.

36

G.G. Matvienko, S.V. Afonin and V.V. Belov 6. The RTM method based on real-time satellite meteorological data on the atmospheric state at the moment of satellite observation allows the distorting effect of the molecular atmosphere to be considered with errors less than 0.5 K. Application of the RTM method based on the split-window principle makes this solution stable in the sense of errors of assigning a priori meteorological information. 7. The RTM method has significant advantages over the standard approaches in detecting high-temperature sources, especially low-intensive fire sites under unfavorable meteorological conditions of observations. 8. The software prototype for real-time correction of satellite IR MODIS measurements for the distorting effect of the atmosphere presented in this work allows the capabilities of the existing methods of temperature sensing of the underlying surface to be extended due to the allowance for the distorting effect of the aerosol and semitransparent cloudiness.

ACKNOWLEDGMENTS In conclusion, we would like to express our gratitude to the managers and staff management in Forest Protection Services for a fruitful and long-term cooperation. We are grateful to former deputy director, Prof. V.V. Koshelev (Insitute of Solar-Terrestrial Physics SB RAS), deputy director, Prof. E.A. Loupian (Space Research Institute RAS), and Dr. A.I. Sukhinin (Sukachev Institute of Forest SB RAS) for useful advice and discussions. We are thankful to our colleagues from Laboratory of Optical Signal Propagation (IAO SB RAS) Yu.V. Gridnev, D.V. Solomatov, M.V. Engel, and N.V. Kabanova for a great contribution to the detection of fires. We are also thankful to V.P. Protasova for help in preparation of this chapter.

REFERENCES [1]

[2] [3]

[4] [5] [6]

Zuev,V. E., Selivanov, A.S., Fomin, V.V., Panfilov, A.S., Romanov, A.V., Afonin, S.V., and Khamarin, V. I., Measuring the ocean surface temperature with the MSU-SK sensor from the Kosmos-1689 satellite. Opt. Atm. 1: 76-80 (1988). Zuev, V. E., Belov, V. V., and Veretennikov, V. V. Systems with Applications in Scattering Media. Publishing House of the SB RAS, Tomsk. 1997. Afonin, S.V., Panfilov, A.S., Romanov, A.V., Selivanov, A.S., Fomin, V.V., and Khamarin, V.I., Satellite experiments on testing radiometric accuracy of IR-channels of the MSU-SK sensor placed onboard Resurs1 Satellite No.2 during LKI. Proc. SRA “Planeta” No. 42: 33-37. 1993. Belov, V.V., Afonin, S.V., and Makushkina, I.Yu., Image transfer through the atmosphere. Atmos. Oceanic Optics.10: pp.449-462 (1997). Belov, V.V., Afonin, S.V., Gridnev, Yu.V., and Protasov, K.T., Thematic processing and atmospheric correction of satellite images. Atmos. Ocean Opt.12: 991-997. 1999. Kaufman, Y.J., and Justice, C., MODIS ATBD: Fire Products (Version 2.2 Nov. 10. EOS ID#2741.P. 77. 1998.

Early Detection of Forest Fires from Space Based on the RTM Method [7] [8]

[9]

[10] [11]

[12] [13]

[14] [15] [16] [17] [18]

[19]

[20]

[21]

37

Boles, S.H. and Verbyla ,D.L., Comparison of three AVHRR-based fire detection algorithms for interior Alaska. Remote Sens.Environ. 72: 1-16. 2000. Li Z. Q., Nadon S., Cihlar J. Satellite-based detection of Canadian boreal forest fires: development and application of the algorithm Int. J. Remote Sens. 21: 3057– 3069. 2000. Li Z., Kaufman, Y. J., Ichoku C., Fraser R., Trischenko A., Giglio L., Jin J., Yu X. A review of AVHRR-based active fire detection algorithms: principles, limitations, and recommendations. In F. Ahern, J. Goldammer, C. O. Justice (Eds.), Global and Regional Wildfire Monitoring from Space: Planning a Coordinated International Effort (pp. 199-225). The Hague, The Netherlands: SPB Academic Publishing. 2001. Giglio L., Descloitres J., Justice C., Kaufmann Y. An enhanced contextual fire detection algorithm for MODIS . Remote Sens. Environ: 87: 273-282. 2003. Loupian E.A., Mazurov A.A., Flitman E.V., Ershov D.V., Korovin G.N., Novik V.P., Abushenko N.A., Altyntsev D.A., Koshelev V.V., Tashchilin S.A., Tatarnikov A.V., Csiszar I., Sukhinin A.I., Ponomarev E.I., Afonin S.V., Belov V.V., Matvienko G.G. and Loboda T., Satellite monitoring of forest fires in Russia at federal and regional levels. Mitigation and Adaptation Strategies for Global Change: 11. No 1: 113-145. 2006. Franca J. R. de A., Brustet J.-M., Fontan J., Multispectral remote sensing of biomass burning in West Africa . J. Atmos. Chem. 22: 81–110. 1995. Minko, N.P., Abushenko, N.A., and Koshelev, V.V., Forest fire detection in East Siberia forests using AVHRR/NOAA data. In: Proceedings of International Simposium on Hyperspectral Remote Sensing and Application (Robert O. Green, Qingxi Tong, Editors), Proceedings of SPIE, 3502: 192-200.1998. Gridnev, Yu.V., Detection of small fires from NOAA/AVHRR data. Atmos. Oceanic Optics.15: 659-662. 2002. Becker, F., Li, Z.L. Towards a local split window method over land surface. Int. J. Remote Sens. 11, No. 3: 369–393. 1990. Ottle, C., Vidal-Madjar, D. Estimation of land surface temperature with NOAA 9 data. Remote Sens. Environ. 40: No. 1: 27–41. 1992. Li, Z.L., Becker, F., Feasibility of land surface temperature and emissivity determination from AVHRR data. Remote Sens. Environ.. 43: No. 1: 67–85. 1993. Sobrino, J.A., Li, Z.L., Stoll, M.P., Becker, F. Improvements in the split window technique for land surface temperature determination. IEEE Trans. Geosci. Remote Sen.: 32, No. 2: 243–253. 1994 Wan, Z., Dozier, J. A generalized split-window algorithm for retrieving land surface temperature measurement from space. IEEE Trans. Geosci. Remote Sens: 34, No. 4: 892–905. 1996. Wan, Z. MODIS Land-Surface Temperature Algorithm Theoretical Background Document (LST ATBD), version 3.3 Inst. for Comput. Earth Syst. Sci., Univ. of Calif., Santa Barbara. 1999. http://modis.gsfc.nasa.gov/data/atbd/atbd_mod11.pdf Mao, K., Qin, Z., Shi, J., Gong, P. A practical split-window algorithm for retrieving land surface temperature from MODIS data. Int. J. Remote Sens: 26, No. 15: 3181– 3204. 2005.

38


[22] Сoll C, Caselles V. A split-window algorithm for land surface temperature from advanced very high resolution radiometer data: Validation and algorithm comparison. J. of Geophys. Res., 102 (D14): 16, 697-16,713. (1997). [23] Afonin, S.V., “Design and application of the atmospheric radiation model for detection of the ocean temperature from space sensing data,” Cand. Phys.-Math. Sci. Dissert., Tomsk, 192 p. 1987. [24] Belov, V.V., Afonin, S.V., From Physical Foundations, Theory, and Simulation to Thematic Processing of Satellite Images. Publishing House of IAO SB RAS, Tomsk, Ru. 266 p. 2005. [25] Thome, K., Palluconi, F., Takashima, T., Masuda, K., Atmospheric correction of ASTER. IEEE Trans. Geosci. Remote Sens: 36. No. 4: 1199–1211. 1998. [26] Sobrino, J.A., Jiménez-Muñoz, J.C., Paolini, L. Land surface temperature retrieval from LANDSAT TM 5. Remote Sens. Environ: 90, No. 4: 434–440. 2004. [27] Wang, P., Karen, Y.L., Cwik, T., Green, R. MODTRAN on supercomputers and parallel computers. Parallel Computing: 28, No. 1: 53–64. 2002. [28] Afonin, S.V., Solomatov, D.V., Solution of problems of atmospheric correction of satellite IR measurements accounting for optical-meteorological state of the atmosphere. Atmos. Oceanic Opt: 21, No. 2: 125–131. 2008. [29] Clough, S.A., Shephard, M.W., Mlawer, E.J., Delamere, J.S., Iacono, M.J., CadyPereira, K., Boukabara, S., Brown, P.D. Atmospheric radiative transfer modeling: a summary of the AER codes, Short Communication. J. Quant, Spectrosc. Radiat. Transfer: 91, No. 2: 233–244. 2005. [30] Rothman, L.S., Jacquemart, D., Barbe, A., Benner, D.C., Birk, M., Brown, L.R., Carleer, M.R., Chackerian, C., Jr., Chance, K., Dana, V., Devi, V.M., Flaud, J.-M., Gamache, R.R., Goldman, A., Hartmann, J.-M., Jucks, K.W., Maki, A.G., Mandin, J.Y., Massie, S.T., Orphal, J., Perrin, A., Rinsland, C.P., Smith, M.A.H., Tennyson, J., Tolchenov, R.N., Toth, R.A., Auwera Vander, J., Varanasi, P., Wagner, G., The HITRAN 2004 Molecular Spectroscopic Database. J. Quant. Spectrosc. Radiat. Transfer: 96, No. 2: 139–204. 2005. [31] Mlawer, M.J., Tobin, D.C., Clough, S.A. A Revised Perspective on the Water Vapor Continuum: The MT_CKD Model. J. Quant, Spectrosc. Radiat. Transfer. 2004. (in press). [32] Kneizys, F.X., Abreu, L.W., Anderson, G.P., Chetwynd, J.H., Shettle, E.P., Berk, A., Bernstein, L.S., Robertson, D.C., Acharya, P., Rothman, L.S., Selby, J.E.A., Gallery, W.O., Clough, S.A. The MODTRAN 2/3 Report and LOWTRAN 7 Model, Phillips Laboratory, Hanscom AFB contract F19628-91-C-0132 with Ontar Corp. 1996. [33] Berk, A., Anderson, G., Acharya, P., Hoke, M., Chetwynd, J., Bernstein, L., Shettle, E., Matthew, M., Adler-Golden, S. MODTRAN4 Version 3 Revision 1 User’s Manual, Air Force Res. Lab., Hanscom Air Force Base, Mass. 2003. [34] Rothman, L.S., Rinsland, C.P., Goldman, A., Massie, S.T., Edwards, D.P., Flaud, J.-M., Perrin, A., Camy-Peyret, C., Dana, V., Mandin, J.Y., Schröder, J., McCann, A., Gamache, R.R., Wattson, R.B., Yoshino, K., Chance, K.V., Jucks, K.W., Brown, L.R., Nemtchinov, V., Varanasi, P., The HITRAN molecular spectroscopic database and HAWKS (HITRAN atmospheric workstation): 1996 edition. J. Quant Spectrosc. Radiat. Transfer: 60, No. 5: 665–710. 1998.


39

[35] Clough, S.A., Kneizys, F.X., Davies, R.W. Line shape and the water vapor continuum. Atmos. Res: 23, No. 3–4: 229–241. 1989. [36] Seemann,S.W., Li,J., Menzel,W.P., Gumley,L.E. Operational retrieval of atmospheric temperature, moisture, and ozone from MODIS infrared radiances. J. Appl. Meteorol: 42, No. 8: 1072–1091. 2003. [37] Afonin, S.V, To the problem of atmospheric correction of satellite data in space monitoring of small-sized forest fire sources. Atmos. Oceanic Opt: 18, No. 4: 299–301. 2005. [38] Afonin, S.V., Belov, V.V. and Gridnev, Yu.V., System of the space-based monitoring of forest fires on the territory of Tomsk Region. Part 1. Organization of the space-based monitoring system. Atmos. Oceanic Opt.: 13, No. 11: 921–929. 2000. [39] Afonin, S.V., Belov, V.V., System of the space-based monitoring of forest fires on the territory of Tomsk Region. Part 2. Estimation of efficiency of space monitoring. Atmos. Oceanic Op: 14, No. 8: 634–638. 2001.



Chapter 2

FIRE SURVEILLANCE AND EVALUATION BY MEANS OF LIDAR TECHNIQUE Andrei B. UtkinA1, Alexander LavrovA and Rui VilarB2 A INOV - Inesc Inovação, Rua Alves Redol 9, Lisbon 1000-029, Portugal B Departamento de Engenharia de Materiais, Instituto Superior Técnico, Technical University of Lisbon, Av. Rovisco Pais 1, 1049-001, Lisbon, Portugal

ABSTRACT Lidar (light detection and ranging) is an active remote detection technique that uses a pulsed laser beam to probe the atmosphere. When the laser radiation illuminates a target, such as a smoke plume originating from a forest fire, part of the incident radiation is backscattered, the intensity of this radiation is measured as a function of time by a suitable detector, and the resulting signal is analyzed by artificial intelligence methods. If the signature of a smoke plume is identified, an alarm is emitted. Precise position of the smoke plume is derived from the current azimuth/elevation angles of the laser beam (provided by the scanning system) and the distance to the target (calculated from the detection time). Being an active detection technique, lidar presents better sensitivity than conventional fire surveillance methods based on visible or infrared imaging. Instead of visualizing the fire, lidar functions by interacting with smoke, without needing line-ofsight observation of the flames. The lidar signals resulting from smoke plumes can be subjected to an inversion procedure, which yields the distribution of the extinction and backscattering coefficients along the laser beam path and, finally, the concentration of smoke particles. This characteristic makes the lidar methods an invaluable tool for the investigation of fire and smoke behavior in natural conditions and for the experimental verification of atmospheric gas-dynamic models. The authors describe simple and robust algorithms of lidar-signal recognition based on the fast extraction of sufficiently pronounced peaks followed by their classification with the help of an artificial intelligence method (neural network). 1 Telephone: +351 213 100 426; fax: +351 213 100 401; email: [email protected]. 2 Telephone +351 218 418 120; fax +351 218 418 121; email: [email protected].

42

Andrei B. Utkin, Alexander Lavrov and Rui Vilar The investigations to be presented include tracing smoke-plume evolution, restoring the smoke concentration and representing the results as contour plots on the topographic map, estimating forest-fire alarm promptness, and smoke-plume location by azimuth scanning of the probing beam. The possibility of locating a smoke plume whose source is out of line-of-sight and detection under extremely unfavorable visibility conditions are also demonstrated. The eye hazard problem caused by laser radiation is addressed and the possibilities of providing eye safety conditions are indicated.

1. INTRODUCTION Lidar (light detection and ranging) technology is a promising tool for forest-fire monitoring because, due to its very high sensitivity and spatial resolution, it enables efficient detection and location of the small smoke plumes that originate from forest fires in their early stages of development over a considerable range (tens kilometers). The analysis of gaseous emission produced by power plants, factories, and forest fires was amongst the first applications of lidar [1,2]. Since those early experiments, interest in lidar has been steadily increasing, and lidar methods, along with sophisticated algorithms for lidar signal processing and analysis [3-6], are now widely used for atmosphere research and monitoring [7-9]. On the contrary, the use of lidar as a tool for fire detection did not receive enough attention. The reported works are mostly concerned with large-scale spatial and temporal phenomena, such as on-ground and airborne analysis of smoke clouds resulting from large forest fires [10-13], weapon firing exercises [14], burning oil [15], measuring the density of smoke generated by large forest fires in the atmosphere and stratosphere [16,17], and investigating the correlation between the smoke and ozone concentrations [18]. Potentials of forest-fire detection using lidar were the object of a preliminary theoretical study carried out by Andreucci and Arbolino [19, 20]. The more conclusive predictions made by Vilar and Lavrov [21, 22] soon received experimental confirmation: work by Utkin et al. [23] testified that small fires with a burning rate of about 0.03 kg of wood per second could be promptly detected from a distance of 6.5 km. The fundamentals of the forest-fire detection by lidar methods are presented in the next section. The implementations of lidar fire surveillance stations are discussed in Section 3. The applications of lidar sensors for fire surveillance, including the early detection of forest fires, fire surveillance in industrial environments, and the study of the internal structure of smoke plume are considered in Section 4. Section 5 is devoted to the artificial intelligence algorithms used for automated fire detection. Comparison of the lidar technique with other fire detection methods is made in Section 6 and the prospects of lidar-assisted fire detection are summarized in Section 7.

Fire Surveillance and Evaluation by Means of Lidar Technique

43

2. PRINCIPLES OF SMOKE DETECTION BY LIDAR 2.1. Fundamentals of the Lidar Technique A lidar instrument (Figure 1) consists of a radiation emitter (pulsed laser and beamforming optics); a receiver that collects and measures the backscattered-radiation intensity and usually consists of a light gathering optical train, photodetector and preamplifier; and a computer-based control and data acquisition unit. The laser source produces short and intense radiation pulses, which propagate through the atmosphere. When the emitted laser beam hits a target, part of this radiation is backscattered and collected by the receiver, where it is converted into an electric signal. This electric signal is amplified and directed to the computerized data-acquisition unit, to be recorded, in digital form, as a function of time passed since the laser-pulse emission. Lidars for automated surveillance are supplemented with a signal processing system, capable of classifying the target signatures and issuing, if appropriate, an alarm signal containing information about the target that caused the alarm situation, including its location.

Figure 1. Lidar equipment and detection principles.

The distance between the lidar and the target R may be calculated from the time delay t between the laser-pulse emission and the reception of the backscattered signal by the equation

R = ct / 2 ,

(1)

where c is the velocity of light. The raw lidar signal S is the receiver-unit output voltage recorded during a period of time immediately after the laser-pulse emission ( t = 0 ). This time is related to the distance to the target via equation (1), and the conversion from t to R is a simple rescaling. As a result, the raw lidar signal (peaks due to retroreflection from different targets and the noise background, as illustrated in Figure 2) is often represented as a dependence of the electric signal measured in volts S on the distance R rather than on the time t

44

Andrei B. Utkin, Alexander Lavrov and Rui Vilar

S (t ) ⇒ S (R ) = GIubph (R )RL + S0 ,

(2)

where G is the total electronic gain, I ubph (R ) the unbiased photodetector current, RL the load resistance and S 0 the background component representing all types of low-frequency noise, which can be assumed to be constant during the relatively short measurement time (about 67 μs for a range of 10 km, according to Eq. (2)).

Figure 2. Lidar signal contaminated by noise.

The unbiased photodetector current is proportional to the retroreflected radiation power

Pr

I ubph (R ) = ξ ph Pr (R ) ,

(3)

where ξ ph is the photodetector responsivity. The retroreflected radiation power Pr is given by the lidar equation [7]:

Pr (R ) = E l

R cβ (R ) Arec τ trτ rec exp⎛⎜ − 2 ∫ α (R') dR ' ⎞⎟ , 2 0 ⎝ ⎠ 2 R

(4)

where E l is the output laser pulse energy, β the backscattering coefficient of the medium, Arec the entrance area of the light gathering optics, τ tr and τ rec the transmitter and receiver

efficiencies, and α the extinction coefficient. For the cylindrical light gathering optical systems


Arec =

2 π d rec ,

45 (5)

4

where d rec is the diameter of the entrance pupil. At the early stage of fire, the characteristic spread of the smoke plume in the laser-beam propagation direction is of the order of a few tens of meters. To reveal the specific internal structures that allow differentiating the smoke-plume signatures from other lidar returns, the data-acquisition unit must measure the photodetector output with a sampling interval δR ~ 1.5 m. This eventually yields the discrete-time lidar signal in the form S (R (t )) = C 0

β (R ) R2

R c exp ⎛⎜ −2 ∫ α (R ') dR ' ⎞⎟ + S 0 , C 0 = Gξ ph El Arecτ trτ rec = const , 0 ⎝ ⎠ 2

(6)

rec digitized at the points ti = 2Ri / c, Ri = i δR, i = 0,1,K , imax , imax = Rmax / δR ,

Si = S (Ri ) = S (R(ti )) ,

(7)

rec rec represents the maximum range of signal recording. Note that Rmax is exclusively where Rmax det defined by the signal recording time and differs from the lidar detection range Rmax , which

depends on the overall lidar sensitivity, cross-section and reflectivity of the target, and the noise level.

2.2. Receiver and Clutter Noise As illustrated in Figure 2, in a lidar signal the smoke-plume signatures are observed against a background contaminated by electronic and atmospheric noise. The background component of the signal can be estimated by averaging the signal recorded beyond the range of the instrument,

S0 ≈

i2 1 rec det < Ri1 < Ri2 ≤ Rmax , ∑ S i , Rmax i2 − i1 + 1 i =i1

(8)

where no signal attributable to retroreflection is expected [24]. Let us assume that the backscattered radiation intensity is measured by an avalanche photodetector (APD). Then the total noise power Pn associated with the lidar signal comprises the following components: thermal and amplification noise, Pther and Pamp , shot noise Pshot , noise due to background radiation Pbgnd , and noise associated with the APD dark current Pdark .

46

Andrei B. Utkin, Alexander Lavrov and Rui Vilar According to Yariv [25], Youmans et al. [26], and Overbeck et al. [27], the noise powers

Pther and Pamp can be estimated as Pther = 4k BTB, Pamp = 4k BTa B ,

(9)

where k B is the Boltzmann constant, T the absolute detector temperature and B the electronic bandwidth, which depends on the laser-pulse duration tl

B = 1 /(2tl ) .

(10)

The value of Ta characterizes the effective noise temperature, with is related to the temperature and the noise figure of the amplifier N fig via the equation

Ta = (N fig − 1) T .

(11)

The shot noise results from the finite number of electrons that carry energy associated with the signal S . Its mean power can be estimated as [26, 27].

Pshot = 2ePr FexM 2ξ APDBRL ,

(12)

where e is the charge of an electron, Fex is the excess-noise factor, and M and ξ APD are, correspondingly, the APD gain and responsivity. The power associated with the background-radiation noise can be assessed by

Pbgnd = 2eIbgnd Fex M 2ξ APD BRL ,

(13)

where

I bgnd = Arec Brecτ rec

π Γ2 4

Lλ ,

(14)

is the background-radiation power [26, 27]. Here Brec stands for the receiver optical bandwidth, Γ is the full angle of field of view of the receiver and Lλ the background radiance of the sky at the operating wavelength λ . The noise component due to the avalanche-photodiode dark current is characterized by the power [25, 28].

Pdark = 2eidark Fex M 2 BRL ,

(15)


47

where idark is the dark current. The power associated with the electric lidar signal is given by the equation [8, 25, 27].

Psig = (Mξ APD Pr ) RL , 2

(16)

and the ratio of the signal power to that of noise (signal-to-noise ratio, SNR ) is

SNR =

Psig Pther + Pamp + Pshot + Pbgnd + Pdark

.

(17)

Substituting Eqs. (6) and (9)-(16) into Eq. (17) and solving the resulting equation with respect to El yields an estimation of the minimum laser pulse energy required for single-shot target detection with some predefined value of SNR

El (SNR ) =

2 SNR eFex B

cξ APD

Arec τ trτ rec β exp (− 2α R ) R2

⎛ ⎛ 2 k B (T + Ta ) ⎞ ⎞⎟ 2 ⎜⎜ ⎟ × ⎜1 + 1 + + + ξ I i APD bgnd dark ⎟ . ⎜ SNR eFex B ⎝ eM 2 Fex RL ⎠ ⎟⎠ ⎝

(18)

Another type of photodetector widely used in lidar equipment are photomultipliers (PMT). This case is analyzed by Measures [7] who provides the final equations for the signalto-noise ratio and E l in the form

SNR =

ξ PMT Pr , 2eFPMT B(ξ PMT Pr + ξ PMT I bgnd + idark )

El (SNR ) =

cξ PMT

2 SNR 2 eFPMT B Arec τ trτ rec β exp(− 2αR ) R2

(19)

(20)

⎛ 2 (ξ PMT I bgnd + idark ) ⎞⎟⎟, × ⎜⎜1 + 1 + 2 SNR eFPMT B ⎝ ⎠ where ξ PMT and FPMT are respectively the PMT responsivity and the noise factor. Although accumulating lidar returns corresponding to several laser pulses emitted along the same optical path requires more time than the single-shot operation, the stochastic noise can be greatly reduced. A rough estimation of this noise reduction based on the normal-

48


distribution model [7] leads to a dependence of the signal-to-noise ratio with the square root of the number of lidar returns accumulated, nacc :

SNR ≈ nacc SNR (1) .

(21)

Correspondingly, the required laser-pulse energy lowers form

(

)

(

)

El( nacc ) (SNR ) ≈ El SNR (1) = El SNR / nacc .

El defined by Eq. (18) to (22)

2.3. Smoke Plume Structure To apply the equations derived in the previous Subsection to smoke plume detection, one needs to know the distribution of two basic parameters – the extinction ( α ) and the backscattering ( β ) coefficients – within the plume. This requires modeling of the smokeplume structure. One of the simplest models, based on the "top-hat" Morton’s approximation [29], was proposed by Lavrov and Vilar [21]. In their model the influence of wind was neglected and the smoke plume was assumed to have axial symmetry with respect to the z axis drawn in the vertical direction. The characteristic radii of the velocity, temperature, and particleconcentration distributions in the plume are assumed to be equal. The velocity in the vertical direction u , the temperature T , and the plume density ρ are averaged over this common radius R p = R p (z ) , so that in the cylindrical coordinates r, ϕ , z , oriented in such a manner that z represents the distance to the ground, one has R p ( z ) ⎛ u (r , z ) ⎞ ⎛ u (z ) ⎞ ⎜ T ( z )⎟ = 2 ⎜ ⎟ ⎜⎜ ⎟⎟ R 2 ( z ) ∫ ⎜ T (r , z ) ⎟ r dr . p 0 ⎝ ρ (r , z )⎠ ⎝ ρ ( z )⎠

(23)

Further, it is assumed that the molar mass of the gas in the hot plume is equal to that of air and that the local pressure equilibrium is achieved, thus within the framework of the idealgas equation

ρ T = ρ airTair ,

(24)

where the index air denotes the ambient air parameters. With the introduction of the normalized temperature

θ (z ) =

T − Tair ρ air − ρ g= g, Tair ρ

(25)


49

where g is the gravity acceleration, the plume parameters for a steady-state combustion process are defined by the system of ordinary differential equations

d ( ρ R 2pu ) = 2 ρ a Eu R p , dz

d ( ρ R p2 u 2 ) = ρ θ R 2p , dz

d ( ρ θ R 2pu ) = 0 , dz

(26)

where E is the entrainment coefficient. Following the suggestion of Morton et al. [29, 30], E can be found from the quasi-empirical relation

E = 0.12

ρ . ρa

(27)

For given values of the fire site radius and burning rate, which identify the initial values of velocity and temperature (see Lavrov and Vilar [21]), the system of equations (26) can be solved using an explicit finite-difference scheme. The solution yields the plume density ρ and radius R p as functions of the distance to the ground z , which allows calculating the desired extinction and backscattering coefficients. A more complicated model that takes into account the influence of crosswind was proposed by Utkin et al. [31]. The fire source is assumed to be a rectangular. The hot plume mixes with ambient air, grows up due to the buoyancy forces and is deflected by the wind. Initial velocity and temperature values were calculated using the thermodynamic considerations discussed in detail by Andreucci and Arbolino [19]. The flow of the smokeplume gases is described via the three-dimensional system of Reynolds-averaged NavierStokes equations [32-34] with appropriate boundary conditions, guaranteeing the uniqueness of the solution. This complete set of equations was solved with the help of the SIMPLE algorithm [32, 35] using suitable software [36]. The model adequately describes all essential factors affecting the dispersion of a hot smoke plume in the presence of wind — the turbulent mixing, action of buoyancy, and deflection — and may be used for the semi-qualitative assessment of the evolution of smoke plumes, simulation of training patterns for artificial intelligence systems of smoke recognition, and prediction of the lidar sensitivity and range with respect to smoke.

2.4. Interaction of Smoke Plume with Laser Radiation The interaction of the laser beam with the smoke plume depends not only on the plume density distribution predicted by the model, but as well on the shape and dimensions of the smoke particles. Although it is not the case, most feasible models of light scattering by smoke are based upon the assumption that the particles have a spherical shape [37], which enables one: (i) to neglect the orientation of the particle with respect to the light propagation direction, (ii) to characterize the variety of particle shapes and sizes only by a one-dimensional particlesize distribution, and (iii) to estimate the backscattering-to-concentration ratio c β n on the basis of Mie theory. This results in the integral formula [7, 19, 38].

50

Andrei B. Utkin, Alexander Lavrov and Rui Vilar ∞

∫ π rpθb (rp , λ , n*) N p (rp )drp , cβ n = = 0 ∞ n ∫0 N p (rp )drp β

2

where rp is the particle radius, [39],

(28)

θb the backscattering efficiency given by the Mie series

λ the radiation wavelength, n* = n1 − n2 i the complex refraction index, which

( )

depends on the burning material, and N p rp the particle-size distribution. The terms of the Mie series are calculated on the basis of derivatives of Legendre polynomials and spherical Bessel functions. The convergence of the series and, correspondingly, the number of terms to be calculated for achieving a predefined accuracy depend on the parameter 2π rp / λ [40].

( )

The distribution N p rp

can be assessed from results of Kent [41] while the complex

refraction index can be estimated as n* = 1.53 − 0.03i [42] and n* = 1.95 − 0.66 i [43] for wood and oil smoke respectively. On the basis of experimental [44] and computational data [45], the value of the backscattering-to-extinction ratio can be assessed as

cβα =

β ≅ 0.033 sr–1. α

(29)

This dependency is a particular case of a more complicated relationship

β = const × α κ , κ = const

(30)

widely used for the characterization of lidar targets within the framework of so-called Klett's algorithm for signal inversion [6, 46-50]. For a smoke plume, both α and β are proportional to the particle concentration n , from which follows β / α = const and κ = 1 . The Klett's algorithm involves the representation of the lidar signal in the logarithmic range-adjusted form

⎛ R 2 S (R ) − S0 ⎞ ~ ⎟⎟ . S ( R) = ln⎜⎜ 2 ⎝ R0 S (R0 ) − S0 ⎠

(31)

where R0 is the reference distance for which the extinction coefficient α is known or can be estimated by some value α 0 :

α (R0 ) = α 0 .

(32)

Substituting Eq. (6) into equation (31) and differentiating with respect to the variable R , ~ one can verify that S (R ) satisfies the ordinary differential equation


51

~ d S 1 dβ = − 2α . dR β dR

(33)

For the case of relation (30) this equation reduces to the Bernoulli (homogeneous Riccati) equation, which has the stable analytical solution found by Klett [50].

(

)

~ exp κ −1S (R ) α ( R) = , R < R0 . R0 ~ α 0−1 + 2κ −1 ∫ exp κ −1S (R′) dR′ R

(

(34)

)

~

On the boundary of the computation domain R = R0 , the definition (31) yields S (R0 ) = 0 , which meets condition (32). Finally, after retrieving α (R ) from the experimental data S (R ) , one can estimate the particle concentration profile along the direction of the probing laser beam

n( R) =

cβα α ( R) . cβn

(35)

As will be shown in Subsection 4.1, by combining several lidar scans corresponding to different laser-beam azimuth and elevation angles, the three-dimensional structure of the smoke plume can be revealed, providing reliable information about the evolution of the plume during the fire and its dissemination in the atmosphere. For fire-surveillance applications it is of utmost importance that detection is automated, as early as possible, and with maximum possible range. This implies automatic recognition of the smoke-plume signatures and rejection of false alarms, which frequently result from objects causing radiation backscattering, such as hills, trees, aerial cables, and so on. Inhomogeneities of the medium caused by atmospheric phenomena may also lead to spurious peaks in the lidar signal. According to Eqs. (2)-(6), the shape of the peaks correspondent to smoke plumes in the lidar signal depends in a complicated way on the distribution of the extinction and backscattering coefficients within the smoke plume volume. Both parameters are closely connected with the distribution of soot particles. Experimental investigations [23, 51-53] showed that both the soot-particle density distribution and corresponding radiation retroreflection peaks vary greatly, following the inherently stochastic process of smoke-plume evolution in natural conditions [54, 55]. Although important for the prediction of the range of the lidar instrument, gas-dynamics smoke-plume models cannot provide a solid basis for the extraction of characteristic features from the smoke plume signatures, especially for smalland medium-size fires developing in undulated terrain and in the presence of wind. This fact is illustrated by Fig. 3, a photograph of an experimental fire taken during Gestosa experiments [56, 57], where ascending and descending smoke plumes are observed simultaneously. Due to this lack of reliable parametric models, automated fire surveillance is mainly based on the statistical analysis of the experimental signals [53] or on artificial-intelligence algorithms such as neural networks [58].

52


Figure 3. Wildland fire developing in terrain with complicated relief in the presence of wind. A downward-propagating smoke flow is observed along with conventional buoyancy-dominated plume.

2.5. Eye Safety One aspect of active fire sensing that requires consideration is the potential danger to the eye caused by the probing laser beam. In general, the lasers commonly used in lidar instrumentation, such as Q-switched Nd:YAG lasers, produce in the near field an intense pulse of radiation with 1064 or 532 nm wavelength and an energy density several orders of magnitude higher than the maximum permissible single-pulse exposure of the human eye [38]. In traditional lidar applications, such as the ozone-layer investigation or the study of aerosols in the upper atmosphere, the laser beam is directed vertically or nearly vertically to the sky, so the risk of an accidental eye injury is minimal. This is not the case in wildfire surveillance applications, where the laser beam travels along predominantly horizontal paths, at low heights above the ground. The problem of eye safety was analyzed from a theoretical standpoint by Utkin et al. [52]. The eye hazard factor for the distance R , EHF (R ) , is defined as the ratio of the laser beam energy density within an illuminated spot of effective diameter d R to the maximum permissible single-pulse eye exposure for looking directly into the laser beam MPEλ : EHF (R ) =

4 El 4 El 4 El , = ≈ 2 π d MPEλ π (d 0 + 2 R tan (γ / 2)) MPEλ π (d 0 + γ R )2 MPEλ 2 R

where d 0 is the waist of the laser beam and γ is the laser-beam divergence.

(36)


53

The laser beam is eye-safe for distances such that EHF (R ) ≤ 1 , and according to Eq. (36), a lidar presents no danger outside the eye-hazard range

Rehr =

⎞ 1 ⎛⎜ 4 El ⎟. d − 0 ⎟ γ ⎜⎝ π MPEλ ⎠

(37)

In the simplest lidar configurations no beam formation optics is used, so the raw laser beam is emitted directly to the atmosphere. In this case

d 0
256 . acc

SNR 5 assures excellent sensitivity. For SNR ~ 3 the probability of automatic detection drops down to approximately 50%, which can be considered as a good result, bearing in mind that suspicious areas can be rescanned to verify the alarm as a part of the automated surveillance procedure. SNR = 3 corresponds to the instrument range of 315 m, which allows to cover an area of 10 ha. Table 4. Parameters of the experimental campfires Campfire

Fuel mass, kg

Burning rate kg/s

Fire diameter, m

a b

210 540

0.06 0.15

0.5 0.75

Distance to the lidar, m 100 250

Table 5. Illustrative experimental data Camp-fire

APD gain

a

30

b

100

Distance to the smoke plume, m 85.5 94.5 100.5 96 97.5 87 259.5 279 287.25 288 292.5 268.5 270 270 279.75 291.75

SNR 5.1 3.5 3.8 4.7 6.3 4.1 4.4 4.0 3.5 4.8 3.6 5.4 4.0 7.7 3.5 5.2

nacc 2 2 2 256 256 1024 1 1 1 1 1 256 256 256 256 256


69

Figure 18. Typical lidar signal, containing a smoke-plume signature in the vicinity of R = 264 m. The burning rate is about 0.15 kg/s, nacc = 512.

5. AUTOMATED DETECTION Detectors for automated fire surveillance must be supplemented with a signal recognition system, performing the classification of target signatures and issuing, if needed, an alarm signal containing information about the target that caused the alarm situation, including its location. In principle, a lidar detector locates the smoke plume with a precision of a few meters, thus allowing for a very accurate mapping of the fire. The angular target position (the azimuth ϕ and elevation ϑ , see Figure 19) is given by the laser beam direction, but the calculation of the distance to the smoke plume Rsp is based on the analysis of the return signal.

Figure 19. The main spatial parameters of smoke-plume detection.

Neural network (NN) architectures and algorithms suited for lidar data extraction have been discussed in the literature since the 1990s [95]. The authors developed a simple and fast

70


preprocessing method for facilitating the recognition of low intensity retroreflection peaks from distributed targets. Following the same principles as the radial-basis function algorithms [96, 97], the recognition efficiency of a perceptron-based NN is enhanced by a special binarization procedure that uses a 2D grid in the signal-distance plane for the waveform representation and a point-to-node proximity criterion for assigning one or zero to the grid nodes. Each node is treated as a separate input component, increasing the network input dimension, number of adjustable weights and, according to Cover's theorem [97], improving pattern separability. When the vertical grid spacing equals the lidar sampling distance so that all the signal points are located on the vertical grid lines, the above algorithm reduces to a simple point binarization of the signal with resolution corresponding to the horizontal grid spacing. The threshold binarization procedure, corresponding to a point binarization in which a one is assigned to each grid point situated below any point already assigned to one, is even easier for hardware implementation (a batch of threshold detectors with linearly increasing thresholds) and results in less sparse and more compact binarized samples: the bottom line always contains ones and can be discarded. The application in question is characterized by the following difficulties: rec / δR ~6.7×103, is The length of the discrete-time sequence to be processed, imax = Rmax much larger than in other lidar applications, and fire may occur anywhere within the surveillance range, so no narrower region of interest can be selected a priori. As a result, conventional NN algorithms [95, 98] cannot be directly applied because they would require excessive computation time and resources. The peaks corresponding to smoke-plume signatures are narrow. As seen from Eq. (6), for a starting fire the characteristic spread of a smoke-plume peak ΔRss , within which the backscattering coefficient β is sufficiently large to produce the signal above the noise level, is restricted by the diameter of the plume: ΔRss ≤ ΔRsp ≈ 10 m. Well-developed fires result in much wider plumes, but denser smoke increases the laser-beam extinction up to α ~ 0.2 m–1 [99], and in these circumstances the smoke-plume signal decreases down to the noise level at a distance of the order of α–1, resulting in ΔRss ~ 5 m. Measured as a number of points in the digitized signal, N ss = ΔRss / δR , the peak width varies from 5 to 10, being always much less than that for the cases described by Bhattacharya et al. [95] and Mitra et al. [98] (hundreds of points). The small peak width and the wide variety of peak shapes preclude the application of statistics-based algorithms for noise reduction and signal compression [98]. The distance to the target Rsp must be determined simultaneously with the forest fire signature resulting in an additional output, codifying Rsp in the case of positive detection. Due to the fact that a constant background can be represented as a sum of uniformly distributed peaks, the problem of peak recognition is not linearly separable a priory and cannot be solved without introduction of preprocessing and/or non-linearity. According to general indications given by Anderson and Rosenfeld [100], automatic smoke detection requires the development of a specialized NN algorithm, incorporating all a priori information about the targets in order to simplify the overall structure and facilitate the recognition. Depending on its nature, the knowledge of the input signal can be represented as a transformation, selection rule and/or invariant and then built into the system via specific design or preprocessing procedures [97]. The analysis of the lidar signal provides the


71

following a priori information: (a) The smoke-plume signatures manifest themselves in the raw lidar signal as peaks whose characteristic width ΔRss (several meters) is much less than the typical distance to the smoke plume Rsp (from hundred meters to several kilometers). (b) The position of the smoke-signature maximum corresponds to the desired distance to the smoke plume. (c) The local noise level may be estimated as the root-mean-square of the signal just before and after the peak and a segment of lidar signal of length ~ 3ΔRss , containing the smoke-signature maximum in its center, is supposed to provide information on both the smoke-plume signature shape and local noise. (d) The ratio of the peak amplitude to the mean local noise, called peak-to-noise ratio (PNR), represents an important scaleindependent characteristic of the peak magnitude, closely linked with the probability of the peak to be a target signature rather than a clutter signal. For this reason, it is worthwhile to treat PNR as an invariant characteristic feature to be extracted and independently presented for recognition. Utkin et al. [58] showed that within the range 10ΔRss ≤ R ≤ Rmax the shape of the smokeplume signature is invariant with respect to the distance. Obviously, the effect of noise increases with distance, so the pattern-recognition problem in question can be treated as distance-independent in the sense that the recognition conditions for a tenuous smoke plume are equivalent to those for a denser plume observed at a greater distance provided that the signal-to-noise ratio is the same.

Figure 20. Stages of the smoke-signature recognition procedure.

72


The knowledge and invariances are built into the system via a preprocessing procedure. The raw lidar signal, consisting of several thousand points, is viewed by the preprocessing software through a window of several tens of points ( ~ 3ΔRss / δR ) that moves along the signal curve in order to define a region of interest. The window motion stops whenever a local maximum coincides with the window center and the corresponding peak-to-noise ratio is calculated. If this peak-to-noise ratio is lower than a threshold value PNRthr (typically, from 3 to 5), the peak is considered to be too small to correspond to a smoke plume and the observation window continues its motion along the curve. Otherwise the pattern within the region of interest is sent to the recognition unit. The corresponding peak-to-noise ratio is also introduced to the NN, but via a special input (Figure 20). The patterns sent to the NN are classified using the supervised learning procedure implemented through the least-squares filtering [97]. For a given training set, it yields a unique deterministic solution for the desired interconnection weights as a product of the pseudoinverse of the matrix composed from the binarized training samples and the vector of the corresponding classification tags (here, 1 for the smoke-signature peaks and –1 for the spurious peaks). Following Bishop's recommendations [96], the instability arisen from the sparse nature of the binary-sample matrices and incomplete ranks is overcome by stabilized pseudoinversion on the basis of singular-value decomposition [101]. The binary input has two additional entries: one for the constant activation bias (+1) and the other for the PNR value that passes to the perceptron without binarization. The alarm signal is accompanied by the current position of the moving window center Rw ≈ Rsp that corresponds to the maximum of the retroreflected radiation and thus provides the desired distance to detected smoke plume. Figure 21 illustrates the comparison of this threshold-binarization algorithm with three more complicated artificial-intelligence methods developed for smoke-signature recognition in lidar signals by Fernandes et al. [102]. The threshold-binarization algorithm demonstrates a better efficiency in the area of false alarm probability greater than 0.65%, resulting in 100% detection of the smoke-plume signatures in the validation set with a false alarm rate as low as 0.84%.

Figure 21. ROC curves corresponding to the developed threshold-binarization algorithm and the three committee machines described by Fernandes et al. [102].


73

At the same time, the proposed algorithm yields nearly one order of magnitude faster training and the developed supervised learning procedure is not connected with the choice of the best classifier, so it can be strictly formalized and performed by untrained users. As well, the learning procedure is fast and of predictable duration: it does not involve repetitive/iterative routines like the training epochs in the case of gradient-descent methods and the global minimum of the classification error for a given training set is readily achieved by a sequence of matrix operations of guaranteed stability. For short lidar-to-target distances, R ≤ 10ΔRss , the shape factor of the smoke-plume signature depends on Rsp . However, it was concluded that the variation in shape does not affect the recognition system because due to very high PNR value the alarm is activated even if the input from the binary-sample nodes does not match perfectly the smoke-plume signature information stored in the NN. This neural-network algorithm is extremely flexible and it was successfully used for automated signal processing in a variety of lidar applications. As compared to alternative methods of fire detection automation based on radiometry and video or infrared imaging, the present active technique, due to its potentially higher sensitivity, offers quicker response to the alarm situation: automation of the 1D lidar signal processing is an easier task than fire or smoke-plume recognition in the 2D images provided by video/infrared cameras.

6. LIDAR AND OTHER FIRE DETECTION METHODS One of the most important fire surveillance methods is satellite-born detection using infrared cameras. This method is predominantly important for surveillance of large and homogeneous regions [103]. Conventional thermal and smoke detectors are widely used nowadays, but they typically take charge of a limited area. That is why the majority of local-observation techniques for outdoor fire surveillance use different remotely controlled cameras. Detectors based on video imaging extract the smoke plume features from the background of vegetation and sky. When black and white (grayscale) video cameras are used, this extraction is based on light-intensity differences. To avoid generating false alarms from natural illumination, the detection systems based on color cameras use decision rules derived from the color composition, such as the one developed by Chen et al. [104]. Systems based on infrared cameras with high resolution can detect fires with the radius from about 1 m up to fires spanning several kilometers [105]. Apart from lidar, the active fire detection methods include radar, sodar and radio-acoustic sounding technologies [106]. Although all these methods demonstrated fire detection capability, they cannot be used isolated. The satellite-based imaging covers large areas, but the spatial resolution is poor (about 1 km for the best systems) and the alarm promptness is low (15-30 min delay) [103]. Ground-based cameras provide quicker response, but require very complicated automated detection algorithms, inevitably performing with high false-alarm probability as the image features for smoke are seldom adaptive to lighting conditions, smoke density, or background scene. In addition, infrared cameras are ineffective in daytime in summer, while video cameras cannot function with low natural light intensity. The active detection methods provide much better sensitivity and lower false alarm rate, but the alarms produced entirely

74


on this basis are not supported by images of the fire site, which significantly reduces the credibility of the alarm. Moreover, active methods often result in complicated equipment with special maintenance requirements. As discussed in the literature [103, 105], significant improvement of fire surveillance can be achieved by combining information from various sensors based on different detection methods. The neural-network classification algorithm described in the previous section has excellent characteristics for this multi-technique approach. The NN output can be easily fused with data provided by other sensors by including the lidar network into the global neuralnetwork decision algorithm. Alternatively, the lidar signal classification output can be treated as an independent measure of the fire probability and included as a separate term in the fuzzy decision function governing the eventual alarm generation, just as in the case of the heterogeneous imaging system described by Arrue et al. [105].

7. CONCLUSION Lidar is a powerful method of fire detection and study, which offers different solutions for a wide range of detection conditions and applications, resulting in reliable and costefficient fire surveillance. As compared to alternative automated fire detection methods, which are mostly based on video image processing and analysis, this method offers higher sensitivity and quicker response to the alarm situation. In addition, automatic fire signature detection in lidar systems is easier than in IR/VIS images. Of special interest is the integration of lidar sensor(s) into a multi-technique fire surveillance system. Such integration would reduce single system disadvantages in the area of fire detection and extend the sensitivity of the combined system, improve its applicability at low illumination condition, and significantly reduce the false alarm rate. The use of modern lasers and light gathering optics can significantly increase the range of the lidar instrument, making the technology much more robust and competitive in today’s market of fire surveillance in forestry and industry.

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9]

Hamilton, P. M. Philos. Trans. R. Soc. Lond. A 1969, 265, 153-172. Hinkley, E. D. Laser monitoring of the atmosphere; Springer: Berlin, 1976 Zuev, V. E.; Naats I.E. Inverse Problems of Lidar Sensing of the Atmosphere; Springer: Berlin, 1983. Bissonnette, L. R. Appl. Opt. 1986, 25, 2122-2125. Weinman, J. A. Appl. Opt. 1988, 27, 3994-4001. Wei, H.; Koga, R.; Iokibe, K.; Wada, O.; Toyota, Y. J. Opt. Soc. Amer. A 2001, 18, 392-398. Measures, M. R. Laser Remote Sensing; Wiley: New York, NY, 1984. Jelalian, A. V. Laser radar systems; Artech House: Boston, MA, 1992. Bösenberg, J.; Brassington, D.; Simon, P. C. Instrument development for atmospheric research and monitoring; Springer: Berlin, 1997.


75

[10] Uthe, E. E., Morley, B. M., Nielsen, N. B. Appl. Opt. 1982, 21, 460-463. [11] Banta, R. M., Oliver, L. D., Holloway, E. T., Kropeli, R. A., Bartram, B. M., Cupp, R. E. J. Appl. Meteorol. 1992, 31, 1328-1349. [12] Targ, R., Steakley, B. C., Hawley, J. G., Ames, L. L., Forney, P., Swanson, D., Stone, R., Otto, R. G., Zarifis, V., Brockman, P., Calloway, R. S., Klein, S. H., Robinson, A. Appl. Opt. 1996, 35, 7117-7127. [13] Pershin, S., Hao, W. M., Susott, R. A., Babbitt, R. E., Riebau, A. Proc. SPIE 1999, 3757, 60-66. [14] Uthe, E. E. Appl. Opt. 1981, 20, 1503-1510. [15] Eberhard, W. L. Appl. Opt. 1983, 22, 2282-2285. [16] Muller, D., Wagner, F., Althausen, D., Wandinger, U., Ansmann, A. Geogr. Res. Lett. 2000, 27, 1403-1406. [17] Fromm, M., Alfred, J., Hoppel, K., Hornstein, J., Bevilacqua, R., Shettle, E., Servranckx, R., Li, Z. Q., Stocks, B. Geogr. Res. Lett. 2000, 27, 1407-1410. [18] Longo, K. M., Thompson, A. M., Yamasoe, M. A. J. Geophys. Res. 1999, 104, 1211312119. [19] Andreucci, F., Arbolino, M. V. Nuovo Cimento 1993, 16C, 35-50. [20] Andreucci, F., Arbolino, M. V. Nuovo Cimento 1993, 16C, 51-65. [21] Lavrov, A., Vilar, R. Proc. SPIE 1999, 3868, 473-477. [22] Vilar, R., Lavrov, A. Appl. Phys. B 2000, 71, 225-228. [23] Utkin, A. B., Lavrov, A. V., Costa, L., Simões, F., Vilar, R. Appl. Phys. B 2002, 74, 7783. [24] Durieux, E., Fiorani, L., 1997. Data Processing. In Instrument Development for Atmospheric research and Monitoring; Bösenberg J., Rrassington D., Simon P.C.; Eds.; Springer: Berlin, pp 87-94. [25] Yariv, A., Optical Electronics, Saunders College Publishing: Philadelphia, PA, 1991. [26] Youmans, D. G., Garner, R., Peterson, K. R. Proc. SPIE 1994, 2271, 13-28. [27] Overbeck, J. A., Salisbury, M. S., Mark, M. B., Watson, E. A. Appl. Opt. 1995, 34, 7724-7730. [28] Saleh, B. E. A., Teich, M. C. Fundamentals of Photonics; John Wiley and Sons: New york, NY, 1991. [29] Morton, B. R. In: Tenth International Symposium on Combustion; The Combustion Institute: Pittsburgh, PA, 1965, p 973. [30] Morton, B. R., Taylor, G., Turner, J. S. Proc. R. Soc. London, Ser. A 1956, 234, 1-23. [31] Utkin, A. B., Lavrov, A., Vilar, R. Proc. SPIE 2006, 6359, 635915-1 - 635915-11. [32] Patankar, S. V. Numerical heat transfer and fluid flow; Hemisphere Publishing Corporation: Washington, DC, 1980. [33] Novozhilov, V. Prog. Energy Combust. Sci. 2001, 27, 611-666. [34] Mokhtarzadeh-Dehghan, M. R., Konig, C. S., Robins, A. G. Atmos. Environ. 2006, 40, 3909–3923. [35] Patankar, S. V., Spalding, D. B. Int. J. Heat Mass Transfer 1972, 15, 1787-1806. [36] CFD for design engineers, CHAM - UK Headquaters: London, 2005, http://www.cham.co.uk [37] Chylek, P., Ramaswary, V., Cheng, R., Pinnick, R. G. Appl. Opt. 1981, 20, 2980-2984. [38] Carnuth, W., Trickl, T. Proc. SPIE 1992, 1714, 192-198.

76


[39] Hulst, van de H. C. Light Scattering by Small Particles; John Wiley and Sons: New York, NY, 1957. [40] Wiscombe, W. J. Appl. Opt. 1980, 19, 1505-1509. [41] Kent, G. S. Appl. Opt. 1978, 17, 3763. [42] Benech, B., Dinh, P.V., Ezcurra, A., Lesne, J. L. Atmos. Environ. 1988, 22, 1071-1084. [43] Patterson, E. M., McMahom, C. K. Atmos. Environ. 1984, 18, 2541-2551. [44] Althausen, D., Muller, D., Ansmann, A., Wandinger, U., Hube, H., Clauder, E.,Zorner, S. J. Atmos. Oceanic Technol. 2000, 17, 1469-1482. [45] Liu Z., Voelger P., Sugimoto N. Appl. Opt. 2000, 39, 3120-3137. [46] Klett, J. D. Appl. Opt. 1983, 22, 514-515. [47] Klett, J. D. Appl. Opt. 1985, 24, 1638-1643. [48] Fernald, F. G. Appl. Opt. 1984, 23, 652-653. [49] Matsumoto, M., Takeuchi, N. Appl. Opt. 1994, 33, 6451-6456. [50] Klett, J. D. Appl. Opt. 1981, 20, 211-220. [51] Utkin, A. B., Fernandes, A., Simões, F., Vilar, R., Lavrov, A. In Forest fire research and Wildland Fire Safety: Proceedings of the IV International Conference on Forest Fire Research, Luso, Portugal; Viegas, D. X.; Ed.; Millpress: Rotterdam, 2002, p 58. [52] Utkin, A. B., Fernandes, A., Simões, F., Lavrov, A., Vilar, R. Int. J. Wildland Fire 2003, 12, 159-166. [53] Utkin, A.B., Fernandes, A.M., Lavrov, A.V., Vilar, R.M. Int. J. Wildland Fire 2004, 13, 401-412. [54] North, G.R., Cahalan, R.F. J. Atmos. Sci. 1981, 38, 504-513. [55] Arnold, L. Random Dynamical Systems; Springer: New York, NY, 1998. [56] Viegas, X. Int. Forest Fire News 2000, 23, 102-105. [57] Viegas, D. X., Cruz, M. G., Ribeiro, L. M., Silva, A. J., Ollero A., Arrue B., Dios R., Gómez-Rodríguez F., Merino L., Miranda A. I., Santos P. In Forest fire research and Wildland Fire Safety: Proceedings of the IV International Conference on Forest Fire Research, Luso, Portugal; Viegas, D. X.; Ed.; Millpress: Rotterdam, 2002, p. 121. [58] Utkin, A. B., Lavrov, A. V., Vilar, R. M. Opt. Spectrosc. 2009, 106, 926-936. [59] LaserAce 1000; Laseroptronix: Valentuna, Sweden, 2003. [60] Riaño, D., Meier, E., Allgöwer, B., Chuvieco, E. In Forest fire research and Wildland Fire Safety: Proceedings of the IV International Conference on Forest Fire Research, Luso, Portugal; Viegas, D. X.; Ed.; Millpress: Rotterdam, 2002, p. 122. [61] Industrial fiber lidar; INO - Institut National d’Optique: Sainte-Foy, Canada, 2003. [62] Falcon lidar sensor system; Toposys - Topographische Systemdaten GmbH: Ravensburg, Germany, 2003. [63] Saito, Y.; Saito, R.; Kawahara, T. D.; Nomura, A.; Takeda, S. Forest Ecology and Management 2000, 128, 129-137. [64] Takeuchi, A., Saito, Y., Kawahara, T. D., Nomura, A., Suzuki, T. Proc. SPIE 2001, 4153, 22-29. [65] Foy, B. R., McVey, B. D., Petrin, R. R., Tiee, J. J., Wilson, C. W. Appl. Opt. 2001, 40, 4344-4352. [66] American National Standard for the Safe Use of Lasers: ANSI Z-136.1; Laser Institute of America: Orlando, FL, 1993. [67] Vaidyanathan, M., Killinger, D. K. Appl. Opt. 1994, 33, 7747-7753. [68] Huffaker, R. M., Reveley, P. A. Pure Appl. Opt. 1998, 7, 863-873.


77

[69] Hellström, J., Karlsson, G., Pasiskevicius, V., Laurell, F. Opt. Lett. 2001, 26, 352-354. [70] High-energy, eyesafe lidar for long-range, high-resolution aerosol detection. Final report on NASA Small Business Innovative Research (SBIR) program, Contract NAS120476; Q-Peak Inc.: Bedford, MA, 1998. [71] Harrell-Klein, S., Wilcox, W. E., Killinger, D. K., Rines, G. A., Schwarz, R.A. Proc. SPIE 1995, 2366, 354-357. [72] Hutchinson J. A., Trussell C. W., Allik T. H., Hamlin S. J., McCarthy J. C., Bowers M. S., Jack M. Proc. SPIE 1999, 3707, 222-233. [73] Patterson, E. M., Roberts D. W., Gimmestad G. G. Appl. Opt. 1989, 28, 4978-4981. [74] Karning H., Ruger J. F., Weispfenning M. Proc. SPIE 1998, 3436, 433-440. [75] Ready, J. F. Industrial Applications of Lasers; Ch. 2. Properties of Laser Light; 2nd ed.; Academic Press: San Diego, CA, 1997. [76] Kent, G. S., Hansen, G. M. Appl. Opt. 1999, 38, 6383-6387. [77] Moorgawa A., Michaelis M. M., Diab R. D., Anderson J., Kuppen M., Prause A. R., McKenzie E., Porteneuve J., Leveau J., Bencherif H., Morel B., Cunningham P. F. Proc. SPIE 2000, 4065, 324-337. [78] NASA patter. Appl. Opt. 1989, 28, 4948. [79] Spinhirne, J. D. Micro pulse laser radar; United States Patent n. US5241315 of August 31, 1993. [80] Harms, J., Lahmann, W., Weitkamp, C. Appl. Opt. 1978, 17, 1131-1135. [81] Brown de Colstoun, F., Chambaret, J. P., Chambaret, Y., Saige, V. A. G., Moscovici, J.-C. M. Station for detecting and locating through laser beams an object or a substance likely to diffuse back at least one part of the incident laser ray and system for sensing a substance such as smoke in particular of a fire such as a forest fire; United States Patent n. US4893026 of January 9, 1990. [82] Duck, T. J., Sipler, D. P., Salah, J. E., Meriwether, J. W. (2001) In Advances in Laser Remote Sensing; Dabas A., Loth C., Pelon J.; Eds.; Ecole Polytechnique: Palaiseau Cedex, France, 2001; pp. 73-76. [83] Eloranta, E. W., Ponsardin, P. In Optical Remote Sensing; Sawchuk, A., Ed.; OSA Trends in Optics and Photonics; Optical Society of America: Washington, DC, 2001; Vol. 52, paper OMC4. [84] Streicher, J., Werner, C. Proc. SPIE 1999, 3707, 590-598. [85] Vilar R. M. C. S., Simões F. S., Costa J. L. V., Utkin A. B., Lavrov A. Lidar system controlled by computer for smoke identification applied, in particular, to early stage forest fire detection. International Patent WO 03/073128 A1, of September 04, 2003. [86] Killinger, D. K., Menyuk, N. Science 1987, 235, 37-45. [87] Lavrov, A., Utkin, A. B., Vilar, R., Fernandes, A. Appl. Phys. B 2003, 76, 87-95. [88] Cordoba, A., Vilar, R., Lavrov, A., Utkin, A. B., Fernandes, A. Opt. Laser Technol. 2004, 36, 393-400. [89] Lavrov, A., Utkin, A. B., Vilar, R., Fernandes A. Int. J. Therm. Sci. 2006, 45, 848-859. [90] Lavrov, A., Utkin, A. B., Vilar, R. Opt. Spectrosc. 2010, 109, 1211-1217. [91] Fernandes, A. M., Utkin, A. B., Lavrov, A. V., Vilar, R. M. Fire Saf. J. 2006, 41, 144154. [92] Davidson, G. A., Slawson, P. R. Atmos. Environ. 1982, 16, 223-229. [93] Heikes, K. E., Ransohoff, L. M., Small, R. D. Atmos. Environ. 1990, 24A, 297-307. [94] Netterville, D. D. J. Atmos. Environ. 1990, 24A, 1061-1081.

78


[95] Bhattacharya, D., Pillai, S. R., Antoniou, A. IEEE Trans. Geosci. Remote Sens. 1997, 35, 699-707. [96] Bishop, C. M. Neural Networks for Pattern Recognition; Claredon Press: Oxford, United Kingdom, 1995. [97] Haykin, S., Neural Networks; 2nd edition; Prentice Hall: London, United Kingdom, 1999. [98] Mitra, V., Wang, C. J., Banerjee, S. Proc. of Application of Neural Networks in Engineering 2003, 13, 695-700. [99] Kozlov, V.S., Panchenko, M.V. Combustion, Explosion, and Shock Waves 1996, 32, 577-586. [100] Anderson, J. A., Rosenfeld, E. Neurocomputing: Foundations of Research; MIT Press: Cambridge, MA, 1988. [101] Press, W. H., Flannerly, B. P., Teukolsky, S. A., Vetterling W. T. Numerical Recipes: The Art of Scientific Computing; Cambridge Univ. Press.: Cambridge, United Kingdom, 1986. [102] Fernandes, A. M., Utkin, A. B., Lavrov, A. V., Vilar, R. M. Pattern Recognition 2004, 37, 2039-2047. [103] Diagne, M., Drame, M., Ferrão, C., Marchetti P. G., Pinto, S., Rivolta, G. IEEE Geosci. Remote Sens. Lett. 2010, 7, 93-97. [104] Chen, T., Wu, P., Chiou, Y. In Proceeding of IEEE International Conference on Image Processing (ICIP 2004), Singapore; IEEE: Piscataway, NJ, 2004, pp 1707–1710. [105] Arrue, B. C., Ollero, A., Martinez de Dios, J. R. IEEE Intell. Syst. 2000, 3, 64-73. [106] Yasar, Y. G., Ince, T. Sensors 2009, 9, 1485-1498.



Chapter 3

AN INTRODUCTION TO UNCERTAINTY IN REMOTELY SENSED FIRE MAPS AND HISTORIC FIRE REGIME RECONSTRUCTIONS Brean W. Duncan Innovative Health Applications Mail Code: IHA-300 Kennedy Space Center, Florida 32899, USA

ABSTRACT Uncertainty exists in all mapped geographic features. Geographic fire maps are no exception. Fire maps are produced using many techniques with remote sensing being among the most widely used methodologies for both single fire event mapping and recent historic fire regime reconstructions. Acknowledgement and incorporation of spatial map uncertainty in fire maps produced by any method is rare with few exceptions. Including uncertainty within fire mapping products will represent an important step in the evolution and maturation of fire mapping science. This chapter explores the chief sources of uncertainty in mapping fires, particularly when mapping fires using remote sensing techniques. A case study is presented that utilizes confidence information to reduce uncertainty in historic fire regime maps.

INTRODUCTION Fire is a primal force that has helped shape the distribution of terrestrial resources on earth (Bond and van Wilgen 1996). The spatial distribution of its influence is critical to many processes and ultimately to life on earth (Bond et al. 2005). Humans have altered fire regimes globally, altering fire frequencies, fire intensities, fire seasonality, fuel continuity, and carbon dynamics (Leach and Givnish 1996; Cochrane 2003; Duncan and Schmalzer 2004; Roy and Boschetti 2009). Mapping historic and contemporary fire distribution is critical for understanding global ecosystem dynamics and the survival of many fire-dependent

80

Brean W. Duncan

organisms. Uncertainty is inherent in mapped geographic features (Zhang and Goodchild 2002). Reducing uncertainty is important and has drawn considerable scientific attention (Mowrer and Congalton 2000; Foody and Atkinson 2002; Zhang and Goodchild 2002). Geographic fire features are no exception with their unique sources of uncertainty, but have drawn less scrutiny in the scientific literature. Fire maps are produced using many techniques with remote sensing being among the most widely used methodologies for both single fire event mapping and recent historic fire regime reconstructions (Minnich 1983; Salvador et al. 2000; Fisher et al. 2006; Duncan et al. 2009). Direct acknowledgement and incorporation of map uncertainty in fire maps produced by any method (dendrochronology, sediment varve analysis, remote sensing, etc.) is rare with few exceptions (Jordan et al. 2005). Integrating uncertainty within fire mapping products will represent an important step in the evolution and maturation of fire mapping science. It is more than maturing a science however, since land and resource managers often rely on geographic fire products to make decisions and allocate resources. For these reasons, reducing uncertainty and maximizing accuracy and confidence of mapped fire features deserves more attention. This chapter highlights the most important sources of uncertainty in mapping fires. It makes no attempt to discuss general forms of uncertainty that exist in all maps because this information can be found in previously available sources (Mowrer and Congalton 2000; Hunsaker et al. 2001; Foody and Atkinson 2002; Shi et al. 2002; Zhang and Goodchild 2002). In particular it will focus on the uncertainty associated with mapping fires using remote sensing techniques. Uncertainty in both single fire event mapping and historic fire regime reconstruction will be discussed. This chapter is subdivided into six major sections. The first section defines and describes the difference between important accuracy assessment terms setting the stage for the exploration of uncertainty in fire mapping. The second section covers uncertainty issues related to mapping fire using optical remote sensing. The third section focuses on sources of heterogeneity in and around fire scars and how it may contribute to the level of uncertainty of mapped geographic fire features. The fourth section outlines hard and soft image classifications and their advantages/disadvantages in regard to geographic fire feature uncertainty. The fifth section deals with accuracy assessment and the limitations of accuracy for controlling uncertainty in fire maps. The sixth and final section presents a case study which maps a historic fire regime using time-series satellite imagery. Because this study was historic in nature, formal accuracy assessment could not be performed, so a confidence attribute was assigned to each mapped fire feature based on ancillary information. This type of analysis includes information on uncertainties inherent to the data, allowing the freedom to include or exclude fire features based on the features’ mapped confidence.

ERROR, UNCERTAINTY AND CONFIDENCE It is helpful to discuss the difference between error and uncertainty at a theoretical level and to define a few terms useful for the basis of this chapter. Accuracy assessment is an integral part of the mapping sciences. Traditionally, specifying error has been the chief means of performing accuracy assessments, particularly in the remote sensing field. There are many definitions for error, but for this chapter, it can be defined as the difference between observed

An Introduction to Uncertainty in Remotely Sensed Fire…

81

and true values. It becomes evident that specifying error is dependent on defining truth. This is problematic: defining truth has proven difficult because class definitions are vague and often two experts can’t agree on a true class definition. This is the reason that the mapping sciences are moving to an uncertainty framework. Certainty can be defined as the quality or state of being certain (free from doubt) especially on the basis of evidence. The important subtlety here is that certainty is based on evidence and not absolute truth. Uncertainty can be defined as not being certain or not known beyond doubt. Using uncertainty moves us away from absolute truth to a more relative measure of truth by way of evidence. At a fundamental level, mapping project objectives need to be clearly defined and then empirical evidence can be gathered in support of those specific project objectives to evaluate uncertainty of geographic output. Simply put, uncertainty is the degree to which a given geographic output leaves its users uncertain about the true nature of the world in claims to represent. If objectives are clearly defined and empirical evidence can be gathered that are consistent with those objectives, it is theoretically possible to specify uncertainty. Confidence is another measure based on evidence that is increasingly being used in the literature to supplement measures of uncertainty. Confidence is simply defined as the quality or state of being certain. Pivotal in an uncertainty framework is the clear definition of mapping objectives making it possible to measure the quality or certainty that those mapping objectives have been satisfied. The evidence for the basis of specifying confidence can be based on the position of boundaries or can be vegetation composition inside a mapped patch or could be both combined. For this chapter it will be assumed that this evidence framework is utilized for the basis of all uncertainty measures including references for measuring uncertainty.

OPTICAL REMOTE SENSING AND UNCERTAINTY Remote sensing is an efficient means for mapping geographic features. Fire scientists are commonly using remote sensing methods as a means to map spatial fire information. Remote sensing is not without error. Error potentially exists in many stages of the remote sensing process. Error exists in the data itself, in the use of the data for particular application, and in the burn area classification process, all which can contribute to uncertainty of mapped fire information. There are many relevant sources for complete information on error and remote sensing in environmental science applications (Friedl et al. 2001; Foody and Atkinson 2002; Shao and Wu 2008). Here there is coverage of the most likely sources for error and uncertainty in the remote sensing process for mapping fire features. Image resolution is an important variable dictating many aspects of mapped feature quality. There are four image resolutions: spatial, spectral, radiometric, and temporal (Jensen 2005). Spatial resolution refers to the grid cell size of the digital imagery. Images are comprised of raster grid cells and the size of these cells dictates the image resolution. Each cell has a brightness value assigned to it by the sensor based on how much electromagnetic radiation it senses/collects within that cell. The brightness value represents the reflectance of all ground features within each image cell for the corresponding geographic location. There are tradeoffs with different image resolutions. An image with coarse spatial resolution (large grid cells) will have small file size and will require relatively little processing effort. Fine-

82

Brean W. Duncan

resolution images provide more detail but require more storage and processing power to classify. Spectral resolution refers to the number of electromagnetic regions that the sensor can capture data in. The electromagnetic spectrum is divided up into bands representing these different regions of the electromagnetic spectrum. The higher the number of bands the finer the spectral resolution. It is generally more advantageous to use fine spectral resolution data. This is because subtle differences in burn characteristics may be best separated in specific regions of the electromagnetic spectrum making it relatively easy to separate and classify these differences. The drawback for this advantage is that the large numbers of image bands are difficult to store and process. Radiometric resolution defines how coarsely or finely the electromagnetic signal strength is recorded by the image sensor within each spectral band. It is simply the sensitivity of the sensor to discern different levels of signal strength. This information is typically communicated by referring to the number of bits. Early sensors like Landsat 1 had a precision of 6-bits (density value range of 0 to 63) and later sensors like QuickBird and IKONOS record in 11-bit precision (density value range of 0 to 2,047). This is important because the detail that fine radiometric resolution provides within each band helps distinguish subtle ground features from one another, such as variable rates of fuel consumption in a fire scar. Temporal resolution simply refers to how frequently in time imagery is collected for a location. Overpass frequency of the satellite carrying the instrument varies with each satellite platform. The Landsat system has a revisitation frequency of 16 days while some sensors have pointable off-nadir imaging capabilities, making it possible to image an area of interest more frequently than using only on nadir overpasses. Imaging in an off-nadir oblique manner creates other issues that can add to classification uncertainty such as bidirectional reflectance inconsistencies between images (Jensen 2005). Geostationary satellites, such as the GOES series, used for weather observations, offer the most frequent image collection options but are of little use for terrestrial mapping applications. Environmental factors such as cloud cover can also limit the frequency and availability of high-quality images. In regions with fast-growing vegetation, high temporal resolution is required to map quality fire-scar boundaries without obstruction from fast vegetation recovery. Uncertainty can be introduced if the resolution of the imagery is not optimum for each application. If the spatial resolution is too coarse, then the fire boundaries will not be detailed enough, while low spectral resolution can hinder the separation of burned from unburned vegetation. The greater the number of bands the more separable each feature should become, as most geographic features have unique signatures for each band. Bands can be used individually or combined to enhance features for classification (Jensen 2005). Classification routines may be the most important determinant of output feature quality. This is why the literature is dominated by detailing different types of classification routines. There are two general divisions, supervised and unsupervised. A supervised classification requires an analyst to make decisions at all stages of the classification. An unsupervised classification also needs an analyst to set up the classification but a machine clustering algorithm actually performs the classification portion of the process. The analyst needs to select the clustering parameters (e.g., thresholds such as the number of clusters, number of iterations, convergence, etc.). Because the unsupervised classification requires less human involvement and thus is slightly less subjective. Notice that it is only slightly less subjective, since some level of subjectivity is inevitable, since each cluster of pixels grouped by the machine classifier still needs to be labeled by a human analyst. We will further


83

subdivide classification techniques into hard and soft classifications after a discussion of fire feature heterogeneity.

HETEROGENEITY AND FIRE MAPPING Heterogeneity within natural features of the earth's surface is a major source of uncertainty in the mapping sciences (Zhang and Goodchild 2002). Fire scar features are no exception with heterogeneity existing because of the degree of fuel consumption (fire intensity), type and amount of exposed soil/bedrock, amount and degree of terrain-induced shading, and presence of different fuel/vegetation types, to name only a few factors (White et al. 1996; Rogan and Franklin 2001; Duncan et al. 2009; Fernandez-Manso et al. 2009). I will use the degree of fuel consumption as an example of burn-scar heterogeneity for the purposes of discussion. Heterogeneity within fuel consumption is a result of variability in fire weather (temperature, humidity, wind speed and direction ), fuel type and loading, fuel moisture content, soil type and moisture content, topography, and other variables. Fire severity varies geographically within each fire feature. Certain areas will burn extremely hot, consuming all available fuels, with other areas not burning at all. Completeness of fuel consumption may be uniform but more likely will vary with enclaves of unburned fuels varying in size. Boundaries between consumed and un-consumed fuels can be abrupt or gradual over some distance. As with most geographic features, heterogeneity (in regard to degree of fuel consumption) may increase as one moves out of the center of a fire feature toward its outside boundaries. Mapping crisp fire boundaries is therefore a very challenging proposition. The heterogeneity of fire scar boundaries is complicated further in many regions where vegetation growth rates are rapid. Not only was there variable consumption of fuels by the fire, but as time since burn increases, the re-growth of vegetation begins to obscure fire-scar boundaries further (Duncan et al. 2009). Vegetation growth rates vary regionally but may also vary by community type within regions and seasonally with community types. Herbaceous types, for example, may recover very rapidly from a growing-season fire but much slower from a dormant-season fire in the same location. Soil reflectance can also influence the amount of heterogeneity within burn scars. After vegetative and ground-litter fuels have been burned away, the reflective characteristics of different soils become important. Organic soils found in hydric and mesic sites are dark, while sandy soils found in xeric sites will have a much brighter appearance and image signature. Image classification routines need to deal with variability to produce high-quality fire maps and reduce uncertainty. Certain classification systems have the ability to incorporate heterogeneity better than others, ultimately reducing uncertainty. The next section describes some classification systems capable of incorporating heterogeneity for fire mapping.

UNCERTAINTY AND HARD/SOFT CLASSIFICATION FOR FIRE MAPPING The tendency to lump or split is deeply rooted in human nature. Categorical thematic maps are desired because they are clean and easy to understand, and represent complex

84

Brean W. Duncan

variables very simply, so simply in fact that we either ignore or are willing to accept their generalization of complex features. Creating categorical maps necessitates the use of crisp boundaries but crisp boundaries rarely exist in nature. There are some instances where boundaries are very abrupt and relatively clearly defined: lakes and geopolitical or urban boundaries are good examples. Most boundaries in nature, however, are mixtures of features transitioning from dominance on one side to the other across a boundary zone. Traditional hard-classification techniques have difficulty accommodating the heterogeneity of natural boundaries (Zhang and Goodchild 2002). Soft classification methods exist and can accommodate heterogeneity by allowing multiple class membership (Gopal and Woodcock 1994; Zhan et al. 2002). Many forms of soft classification exist. In the simplest form of soft classification each pixel would have more than one label. There are more advanced forms such as fuzzy classification, neural networks, spectral mixture analysis, and others (Jensen 2005). Fuzzy classification techniques are widely used (Gopal and Woodcock 1994; Jensen 2005; Legleiter and Goodchild 2005; Lowry et al. 2008) and serve as a good example to discuss concepts related to soft classification. In hard classifications it is assumed that there is internal homogeneity and the study area can be divided into non-overlapping units (Zhang and Goodchild 2002). In soft classification it is expected that areas are non-homogeneous, that classes can have membership in multiple land-cover elements, and no matter how a land area is partitioned, it is certain that areas will not be uniform. As is commonly the case when mapping fire scars, the fire boundary is gradual with regard to fuel consumption. At the center of the fire scar there is complete consumption of fuels which gradually transitions to unconsumed fuels around the extreme border of the fire scar. To aid the classification process, some classes of fuel consumption should be derived. Coarse or detailed classifications can be used depending on the resources and needs of the study. For demonstration purposes, I will use a coarse classification with four classes; no fuel consumption (0%), little fuel consumption (1-50%), moderate fuel consumption (51-99%), and complete fuel consumption (100%). Field work is required to find suitable training areas to represent these classes of fuel consumption. The classification can be performed and each pixel in the classification would have partial membership (Table 1). These fuzzy membership values reveal that Pixel A was almost completely unburned, Pixel B had fuels that were mostly completely consumed, Pixel C was a true mixed pixel with representation of all fuel consumption classes, and Pixel D was almost completely consumed. Table 1. Example of partial membership among fuel consumption classes for select image pixels. The membership values for each pixel must sum to one Fuel Consumption Class None (0%) Little (1-50%) Moderate (51-99%) Complete (100%)

Pixel A 0.8 0.2 0.0 0.0

Pixel B 0.0 0.0 0.4 0.6

Pixel C 0.1 0.2 0.4 0.3

Pixel D 0.0 0.0 0.1 0.9

Soft classification techniques attempt to separate (un-mix) land-cover elements within individual pixels by assigning proportions to each element. Because the proportion of each


85

dominant land-cover element (known as endmembers) can be determined, it is possible to construct continuous map distributions. Continuous distributions can solve many of the problems that have been discussed, because they have the ability to most closely match cognitive and scientific distributions; that is, if it is possible to view or measure a graduated boundary or distribution of some sort, in this case it might be degree of fuel consumption, then that same distribution can theoretically be mapped. There is less abstraction of reality than with hard-classification techniques. Examples of pixel-based continuous fire distributions exist in the literature (Wimberly and Reilly 2007; Miller et al. 2009). Mixed-pixel soft classification techniques are still relatively rare in the fire literature but are beginning to be applied to fire mapping applications (Cochrane and Souza 1998; Rogan and Franklin 2001; Roldan-Zamarron et al. 2006; Robichaud et al. 2007; Smith et al. 2007). Hybrid approaches combining mixed-pixel techniques with other classifiers for mapping fire severity are being developed and used (Fernandez-Manso et al. 2009). A large part of reducing map uncertainty is an understanding how well these mapping techniques represent actual observed fire-severity patterns (Brewer et al. 2005; Hudak et al. 2007; Allen and Sorbel 2008; Kasischke et al. 2008). The soft-classification approach lends itself not only to a more detailed classification but also to more analysis options. For instance, each value or value range can be selected and mapped separately to look for correlations between fuel consumption and other environmental variables. More effort is required to achieve these results, but the outcome is more representative of reality than would be the case if mapping used a hard classification.

ACCURACY ASSESSMENT A distinction can be made between traditional remote sensing vs. cartographic accuracy assessments (Goodchild 1994). Here the focus will be on the theoretical aspects of assessing remote sensing accuracies. Traditionally an accuracy assessment is conducted to complete a mapping project (Congalton and Green 2009). There are typically at least four stages in any remote sensing mapping effort, classification, labeling, verification, and accuracy assessment. After the classification has been completed each classified area must be labeled. The labeling is done by the analyst and then these labels must be verified. Verification of labels can be performed in many ways but the verification process requires that there is some form of confirmation that the classes generated by the classification routine are optimum and are being labeled properly. At this stage, adjustments can be made to the classification scheme and classification routine to optimize the thematic results. After all adjustments have been made and the classification is complete, the last step is typically an accuracy assessment. To perform the accuracy assessment, reference data need to be collected. Reference data are simply information that represents evidence of land cover at a given location. This information is best obtained by ground site visits to locations within the study site but can also be gathered from another independent source that is of significantly better quality and more detailed than the map being produced (Zhang and Goodchild 2002). Common sources for reference data other than field visits are fine-resolution aerial imagery, as an example. The mapped class and actual ground class are compared to determine accuracy. This is typically done by randomizing the location where comparisons are made and then creating a

86

Brean W. Duncan

contingency table summarizing the information (Card 1982; Congalton 1991). Specific information on the accuracy of each land-cover class and the overall accuracy is produced by this contingency table. Other measures such as Kappa can also be derived from the contingency table information. Kappa is a single measure designed to indicate how the map agreement differs from what would be expected by chance (Congalton and Green 2009). The contingency table is typically used for assessing the accuracy of categorical maps because it tracks marginal totals for each category and is not just based on one single accuracy value. This allows the tracking of user’s and producer’s accuracies, which assign an accuracy value from a map user’s and map producer’s perspectives respectively (Story and Congalton 1986). There are many useful examples of accuracy assessment of geographic fire products in the literature (Brewer et al. 2005; Shao and Duncan 2007; Henry 2008; Fernandez-Manso et al. 2009; Roy and Boschetti 2009). Accuracy assessment is important and through this process much can be learned about what inaccuracies may exist within a map. But performing an accuracy assessment by itself will not limit uncertainty. Limiting uncertainty can best be achieved by acknowledging sources of uncertainty and devising a systematic plan to limit their influence in mapped output. A comprehensive strategy for limiting uncertainty must be incorporated into the mapping methodology. When this is done properly, we should then expect accuracy values to be high. Accuracy values can be high for a map that contains large amounts of uncertainty, because the accuracy assessment does not always test in the areas of greatest uncertainty. Most accuracy assessments focus on thematic accuracy, largely ignoring boundary error. Because random points often fall or are selected to be in the center of patches, away from boundaries, we often miss out on information indicating the accuracy of boundaries between categories. Classification error matrices (contingency tables) are not natural tools for analyzing spatial variation in accuracy, or whether points near boundaries have different accuracies in comparison with internal locations (Steele et al. 1998). Other limitations exist for historic mapping applications when field visits and other high-quality spatial evidence are not available. The following case study provides one example of how uncertainty in historic fire maps may be dealt with when field visits cannot be performed.

CASE STUDY Labeling Mapped Confidence In A Historic Fire Regime Reconstruction Historic fire regime reconstructions are especially challenging. All of the difficulties and uncertainties found while mapping single fire events are present plus there are additional considerations. The main additional difficulty is that the fire evidence is often obscured with the passage of time since fire. Field visits to verify and conduct accuracy assessments are not possible. For this reason, historic fire reconstructions may suffer from larger mapping uncertainties that single fire-event maps. Accuracy assessment of single fire-event maps are not uncommon. This is especially true when new image-classification routines are developed for mapping fire scars. To determine the effectiveness of new classification techniques, an accuracy assessment is typically conducted. Because conducting a classic accuracy assessment is not an option for historic reconstructions, other techniques need to be employed. Assessing uncertainty in land-cover


87

classification has a rich history (Goodchild 1994; Steele et al. 1998; Friedl et al. 2001). Global and sub-pixel uncertainty have been richly explored (Zhang and Foody 1998; Foody and Atkinson 2002; Zhan et al. 2002; Goncalves et al. 2009). One of the interesting methods uses confidence information to describe inherent uncertainty in classified geographic features. Confidence information can be assigned to any mapped geographic feature (area or individual pixel) through the use of agreement and disagreement rules (Liu et al. 2004). For agreement and disagreement to be assessed, there must be reference information available. Assuming that historic ancillary reference data are available, confidence information for both location and interpretation can be assigned (Wickham et al. 2004). Fire science can use this concept for improving the amount of information available from all geographic fire maps, particularly historical fire-regime mapping efforts. Although confidence information can take many forms (Joria and Jorgenson 1996; Zhu et al. 2000; Sarmento et al. 2009) for helping to label uncertainty, what follows is a case study which represents a unique application of labeling mapped confidence for identifying uncertainty in a historical fire-regime reconstruction on the federal properties on and surrounding John F. Kennedy Space Center, Florida.

Background The federal government purchased land on a barrier island complex along the central east coast of Florida beginning in 1949 for space-launch operations. This land is known as the John F. Kennedy Space Center (KSC), Merritt Island National Wildlife Refuge (MINWR), Canaveral National Seashore (CNS), and Cape Canaveral Air Force Station (CCAFS) (Figure 1). We will take the first letter from each and refer to the collective properties as KMCC. After the land was purchased fire suppression was implemented. In 1981 the landscape was divided up into fire-management units (FMU) and a managed fire regime was implemented by the MINWR to reduce dangerous fuel loads and maintain habitat for native fire-dependent species, many of which inhabit these properties. The initial goals for this work were to first map as many of the fires that occurred since 1981, creating a complete managed fire-regime history, and secondly to be able to label uncertainty contained within each fire feature. This confidence labeling would give the data user the ability to include or exclude fire features based on their mapped confidence values. This case study will focus on the second goal. For further details on the work and the first goal see Duncan et al. (2009) for details.

Methods Classification of Burn Areas A time series of multispectral satellite imagery was used to map fire scars. The image data consists of multiple bands collected in the visible and infrared spectral wavelengths that are used for classification and discriminative purposes.

888

Breaan W. Duncann

Fiigure 1. The geoographic locatioons of Kennedyy Space Center, Merritt Island National N Wildliife Refuge, Canaveral Nationnal Seashore, annd Cape Canaveeral Air Force Station. S

Two images a year y were useed to maximize the number of fire scars mapped, m due to t the rapid veegetation growth rates folllowing disturrbance on KM MCC (Schmaalzer and Hinnkle 1992; Scchmalzer 2003 3). A total of 40 satellite scenes s were used u dating from fr 1984 to 2005, 39 weere Landsat Thematic Map pper (TM) im mages and 1 was w a SPOT image needeed to fill a gap g in TM avvailability. Thhe TM sensorss collect data at a 30 meter sppatial resolutioon, were first launched l in 19982, and havee been providing continuouus data since that t date makking it ideal foor mapping thhis managed fire f regime. The T first high--quality imagees of our site were availablle in 1984. U Using the first image from 1984, it was possible p to map m some of thhe fires that occurred o in 19983. Because there was onnly one SPOT T image, a connventional unsupervised claassification (JJensen 2005) was w employedd on the originnal bands and used u the MINW WR fire records to select thhe best classifi fied image. Thhe image proceessing techniqque that was used u to classify fy fire scars inn each individdual Landsat TM scene waas more rigorrous and folloowed Shao annd Duncan (22007). This so ource should be consultedd for details on o the techniqque, includingg accuracy asssessment infoormation. Thee results of claassifying this image time series s and describing the m managed fire reegime can be found f in Dunccan et al. (2009). A brief desscription of the classificatioon routine folloows. Each sateellite scene was rectified too State Plane NAD83 N Meters to be comppatible and cliipped to the geographic g bouundaries of thhe federal propperties. A nonpparametric sepparation indexx (SI) was usedd to select the best bands foor classifying burned areas.. The ideal baands have burrned and unbuurned areas seeparated by thheir spectral siignature, makiing them uniqque and easy too classify, hennce the separaation index. O TM band (TM4 – nearr infrared), andd three transfformed bands (Normalized Difference One V Vegetation Ind dex, Principaal Componentt 4, and Tassseled Cap 2) 2 (Jensen 2005) were coollectively ussed for diffeerentiating buurned from unburned arreas. The unnsupervised cllassification algorithm a ISO ODATA was employed e because it is a consistent and repeatable cllassification method m suitable for use on an a image time series. The nuumber of specctral classes


89

was 20, the number of iterations was 20, and the convergence threshold was 0.99 for all the classifications with different band combinations. The 20 spectral classes were then manually recoded into two information classes, burned and unburned, to form the classified fire-scar maps. Following each classification, the fire-scar maps were masked with a GIS data layer of the burned fire management units (FMUs). This masking process, called post-classification cleaning, took advantage of the MINWR fire records and masked out any unburned FMUs. This step removed commission errors outside burned FMUs and helped produce a highquality fire-scar map.

GIS Database After the fire-scar maps were visually inspected and identified problems were rectified, the final thematic maps were converted from ERDAS Imagine (Leica Geosystems 2008) into ArcGIS GRID format (ESRI 2008), and then to a vector format. Attribute information such as burn date, FMU, type of burn (prescribed vs. natural), and age (time since last burn), were added to the fire-scar maps. Because MINWR maintained a database containing both natural and prescribed fires on KMCC since 1977, it was possible to assess and label fire boundary confidence by comparing visual evidence of burn scars on the satellite images and the classified burn scars with the MINWR fire records. If there was agreement between all forms of evidence, the burn scar was labeled with a high confidence value, and if not, the burn scar was labeled with a lower value of confidence. The confidence value (CV) ranged from 1 to 4 and was also added to the fire scar maps (Figure 2). A CV of 1 indicates low confidence in fire scar boundaries with a value of 3 or 4 indicating high confidence in mapped fire boundaries. This is a similar application of classifying landcover confidence (Liu et al. 2004), but modified for application to mapping fire scars. Results are presented with confidencelevel information, allowing the selection of mapped features based on the confidence with which the fire scars were mapped. The confidence values are important because they provide a means for documenting mapped feature quality despite the inability to conduct a formal accuracy analysis due to the historic nature of this study. The time difference between each fire date and the date of the first image acquired after that fire (used to map that fire scar) was recorded in months and called the delta burn date. This was done for each recorded fire using the MINWR fire database and combined with the confidence item information. The goal was to learn how quickly the rapidly growing vegetation in this region obscures fire scars, indicating how many images are required per year to map high-quality (high confidence) fire-scar boundaries. Insight into this question could be gained by exploring how the mapped confidence decreased with increased time since burn (delta burn date). In addition, each fire scar was categorized into one of three dominant land-cover types (wetland, flatwoods, or scrub) to determine how re-growth rates of each land-cover type influences the ability to map high-quality fire-scar boundaries.

900

Breaan W. Duncann

R Results Seasonality/area/size managed m fire regime r elemen nts A total of 54,175 5 ha wass mapped as burned betweenn 1983 and 20005. Of that tootal, 48,601 haa were mapped as burned with w a CV > 1. Only 10 perccent of the mappped burn areaa had a CV = 1. The amoun nt of area burnned peaked inn 2003 for all confidence c vaalues and peakked in 1997 foor CV > 1, wiith reduced am mounts of burnned area in 19999 and 1990,, respectively (Figure 3). A Area burned peeaked in the month m of Noveember with the lowest amouunt in October (total can bee found by taaking the aveerage multiplied by numbeer of years = 21) (Figure 4). 4 Annual vaariability in monthly m area burned was generally low w, with variaability being greatest in N November, the month with the highest avverage and tootal burn area.. Area burnedd reached a m maximum in thhe winter seasson and a minnimum in the spring for all CVs and a minimum m in thhe summer foor CV > 1 (Fiigure 5). Annnual variabilityy in season burned b is veryy low with unniformly smalll standard erroor bars.

Fiigure 2. The pro ocess of determ mining and labeling fire boundaary confidence values v (CV). Thhe diagram onn the left symbo olizes the fire reecords kept by Merritt M Island National N Wildliffe Refuge (MIN NWR). The diiagram in the middle m representts fire scars mappped from satelllite imagery. Thhe diagram on the t right shhows how the in nformation is ussed to label mappped confidence values.


91

Figure 3. Mapped burn scar area by year for Kennedy Space Center, Merritt Island National Wildlife Refuge, Canaveral National Seashore, and Cape Canaveral Air Force Station. Areas were summarized by confidence values 1 and 2 through 4.

Figure 4. Annual average burn area by month for Kennedy Space Center, Merritt Island National Wildlife Refuge, Canaveral National Seashore, and Cape Canaveral Air Force Station. Winter was comprised of Dec., Jan., and Feb.; Spring was Mar., Apr., May; Summer was Jun., Jul., Aug.; and Fall was Sep., Oct., Nov. Areas were summarized by confidence values 1 and 2 through 4 for the period of 1984-2004. Error bars represent standard error for confidence values 1 through 4. 1000 900 800

Area (ha)

700 600 Confidence 1

500

Confidence 2-4

400 300 200 100 0 Winter

Spring

Summer

Fall

Season

Figure 5. Annual average burn area by season for Kennedy Space Center, Merritt Island National Wildlife Refuge, Canaveral National Seashore, and Cape Canaveral Air Force Station. Winter was comprised of Dec., Jan., and Feb.; Spring was Mar., Apr., May; Summer was Jun., Jul., Aug.; and Fall was Sep., Oct., Nov. Areas were summarized by confidence values 1 and 2 through 4 for the period of 1984-2004. Error bars represent standard error for confidence values 1 through 4.

92

Brean W. Duncan

Areal extents for single fires had a mean of 198 ha, a median of 112 ha, a minimum of 0.73 ha, and a maximum of 1,324 ha for all CVs. For CVs > 1, the mean was 209 ha, the median was 126 ha, the minimum was 1.26 ha, and the maximum was the same at 1,324 ha.

Frequency/Return-Interval Managed Fire Regime Element Fire frequency peaked in 1997 for all confidence values (Figure 6). The mean fire frequency was 12 fires per year (274 total fires/23 years), the minimum was four fires per year, and the maximum was 24 fires per year for all confidence values. For CVs > 1, the mean fire frequency was 10 per year (233 total fires/23 years), the minimum fire frequency was one per year, and the maximum was 19 per year. The fire cycle is defined as the amount of time needed to burn an area equal to the study site, in this case, 27,500 ha (area of open water excluded). The fire cycle (fire rotation) at KSC/MINWR, excluding CCAFS and CNS, was 12 years for all CVs and 13 years for CVs > 1. Because the fire cycle is measured in years, the initial year (1983) of study was excluded from the calculation, because the available satellite imagery did not allow mapping of all fires for that year. The calculation started with 1984 and each annual burn total was added until the flammable area of the study site (27,500 ha) was reached, and the number of years added became the fire cycle. The fire cycle was very similar to the return interval of 11.5 years for all confidence values and 13 years for CVs > 1. The return interval is calculated by dividing the upland flammable area (27,500 ha) by the average area burned each year (2,393 ha). The same calculation was followed for CVs > 1 (2,120 ha).

Figure 6. Mapped burn frequency by year for Kennedy Space Center, Merritt Island National Wildlife Refuge, Canaveral National Seashore, and Cape Canaveral Air Force Station. Frequencies were summarized by confidence values 1 and 2 through 4.

An Introduuction to Uncerrtainty in Rem motely Sensed Fire…

93

Spatial Patteern Manageed Fire Regiime Elemen nt The landsccape mosaic maps m cover 21,,528 ha for alll CVs and 20,6659 ha for CV Vs > 1. The laandscape age mosaic map with w confidennce values > 1 represent arreas that mostt assuredly bu urned (Figuree 7). Includinng all confideence values inncreases the burn b area butt it mainly frragments the age a blocks andd makes the agge distributionn on the landsccape younger (Figure 8). The histogram insets in thesee figures show w the age distriibutions for thheir respective maps. The y old age class c for both maps, m but the magnitude peeak in age distribution occuurred in the 3 year w the greatesst for the map including all confidence was c vallues. Fire frequeency was maniifested in mucch finer scale patterns p than the t age mosaiic map, and thhere is a singlle fire frequenncy hot spot in the same location l on both maps (Figgure 9 and Fiigure 10). Th here are subtle difference between b thesee maps, but including i all confidence vaalues creates one o major diffference, the addition a of a frequency f classs making a tootal of nine (F Figure 10) verrsus eight (Figure 9).

M Mapped Con nfidence and d Fire Boun ndary Degraadation

Fiigure 7. Landsccape age mosaicc map and assocciated area (inseet) for Kennedyy Space Center, Merritt Issland National Wildlife W Refugee, Canaveral Naational Seashoree, and Cape Cannaveral Air Forcce Station. A is the time since last burn, initialized Age i from m 2006, the yearr the mapping was w complete. Areas A shown arre for all confid dence values.

944

Breaan W. Duncann

Fiigure 8. Landsccape age mosaicc map and assocciated area (inseet) for Kennedyy Space Center, Merritt Issland National Wildlife W Refugee, Canaveral Naational Seashoree, and Cape Cannaveral Air Forcce Station. A is the time since last burn, initialized Age i from m 2006, the yearr the mapping was w complete. Areas A shown arre for confidencce values greateer than one.

The delta burn b date valuues were the sm mallest for weetlands and larrgest for scrubb landcover tyypes (Table 2)). This trend was w the same for f the delta burn b dates withh CVs of 3 annd 4. There w were 24 fire scars labeled with w a confiddence greater than 3 and a delta burn date d period grreater than six months (tthese were thhe largest delta burn datees and had the t highest coonfidence valu ues). All exceept one of theese fires weree growing seaason fires indiicating that grrowing season n fire scars may m have a lonnger residencyy time on the landscape maaking them

An Introduuction to Uncerrtainty in Rem motely Sensed Fire…

95

eaasier to map using u remote sensing. s The growing g seasoon varies for each species, b grrowing seasonn for dominantts in this systeem is from April through earrly October.

Fiigure 9. Frequenncy Confidencee 1-4 Fire frequuency map and associated a area (inset) for Kennnedy Space Center, Merritt Issland National Wildlife W Refugee, Canaveral Naational Seashore, and Cape Caanaveral Air Foorce Station. Frrequencies derivved by overlyinng all mapped fiires (all confideence values) andd summing thhe number of tim mes each area burned. b

966

Breaan W. Duncann

Fiigure 10. Fire frrequency map and a associated area a (inset) for Kennedy K Spacee Center, Merrittt Island N National Wildlifee Refuge, Canaaveral National Seashore, S and Cape C Canaverall Air Force Statiion. Frrequencies derivved by overlyinng all mapped fires fi with a conffidence value grreater than one and suumming the num mber of times each area burnedd.


97

Table 2. Delta burn date statistics for fire scars mapped between 1983 and 2005 on KMCC, Florida. Delta burn date is the time difference in months between a fire and the next image in time series (after that fire) used to map the burn scar. Confidence values of a) 1 through 4 and b) 3 through 4. (a) Cover type Wetlands Flatwoods Scrub

(b) Cover type Wetlands Flatwoods Scrub

Mean 3.0 3.7 4.0

Mean 2.3 3.9 4.0

Median 2.0 3.0 3.0

Median 2.0 4.0 3.0

Minimum 1 1 1

Minimum 1 1 1

Maximum 11 10 10

Maximum 6 10 10

Discussion Uncertainty, Variability, and Fire Regime Elements There is significant information contained in the results of this study describing a managed fire regime. These results can be compared and contrasted to what is known about natural fire regimes to aid land managers in their effort to mimic the results of natural fire regimes (Duncan et al. 2009). For purposes here however, it is most important to focus on the influence of the mapped confidence and how this may alter the results of a fire regime mapping effort. The results from this study reveal that there are differences in information content based on the users’ acceptance of mapped confidence levels. There was a difference of 10% in total area burned between all CV levels and CV >1. The mapped CV also determined both the season and year the amount of burned area reached a peak. Average individual fire size was also influenced by the selection of CV levels with a difference of about 5% between all CVs and CVs > 1. Annual fire frequency also varied by 17%, the fire cycle was 8% different, and the return interval was 12% different between all CVs and CVs > 1. The landscape age and frequency maps are very similar with a difference in mapped area of 4% because most areas have recently been mapped with a CV > 1. The mapped confidence information helped estimate the range of variation. The range of variation for uncertainty in mapped fire features was between 4% and 17% depending on the fire regime element. The confidence value in this case encompassed many types of uncertainty. Positional uncertainty was included (Was the fire scar mapped in the same location as the fire records indicated or was a portion of the fire in another management unit where it was not recorded?). Boundary sharpness/distinctness was also included (providing information about how difficult it was to map boundaries that clearly indicated fires presence in accordance with the fire records). Boundary sharpness has two components, degree of fuel consumption and vegetation re-growth rates that were included together and not split apart in the CV method used here. Generally speaking, the closer in time the mapping takes place to

98

Brean W. Duncan

the date of the fire the better, minimizing the influence that vegetation re-growth will have on the uncertainty of mapping fire-scar boundaries.

Fire Boundary Degradation As discussed earlier, temporal resolution is an important consideration when mapping fire scars. Uncertainty about the number of images required in time series to assure mapping firescar boundaries with high quality can be addressed using the confidence information. Because of the rapid vegetation growth rates in this region, it was not know how many images would be needed annually to guarantee that every fire that occurred in our study could be mapped. Experience indicated that one a year would not be suitable to map accurate boundaries so two were acquired (one spring and one fall) each year. Because the time lag between images was not always exactly six months (some lags were longer), it allowed us to explore the limits of the classification technique to delineate high-confidence fire-scar boundaries after time intervals exceeding six months following fire. The confidence values helped provide guidance on mapping quality (high-confidence) fire-scar boundaries and their degradation with time since burn. The outer limits of detectability were tested; for example, using the first image in the series, it was attempted to map fires as far back into 1983 as possible but the ability to detect any fire scars occurring eleven months prior to the date of image acquisition was lost. Getting the optimum number of images in series is important so that an ideal balance can be created between reducing imagery costs, minimizing classification effort, and maximizing the quality of fire-regime reconstruction. If the primary objective was to map fire scars in marshes, than we would need a larger number of images each year, likely a minimum of three. It was concluded that for general mapping of fire scars, two images a year spaced about six months apart, acquired in the spring and fall, is reasonable in relation to the tradeoffs discussed above. The number of images may be dependent on the time of year also; for example, to map marsh fire scars it might be necessary to have images every two months during the growing season but further apart during other times of the year. More study may be required to truly optimize the number of images in this system or any other. Using the confidence information, it was determined that the most persistent fire scars were left by growing-season fires. These fires had the largest delta burn dates and this may signify that growing-season fires take a longer time to reestablish vegetative cover following disturbance. This makes sense as the large flush of leaves occurs at the beginning of the growing season (generally late March) prior to most of these fires, and then the plants are dormant in fall/winter.

UNCERTAINTY AND FUTURE OF FIRE MAPPING The incorporation of uncertainty in remote-sensing classification and land-cover mapping is now common. Through time, new techniques have evolved to deal with quantifying uncertainty in these fields (Couturier et al. 2009; Goncalves et al. 2009). The same process needs to take place in fire mapping; however, many of the techniques used in these other disciplines can be adapted, speeding the transition. The beginning of this process may be


99

more general and qualitative then specific and quantitative. For example, the case study presented here groups many uncertainties and uses a very coarse labeling system to convey mapped confidence. While this leaves much to be desired, it is a start in the right direction. Other techniques for visualizing areas that may be subject to greater uncertainty than surrounding areas may be implemented (Buttenfield 2001). The implementation of rigorous soft-classification techniques will greatly improve the incorporation of uncertainty with the ultimate goal of possibly mapping continuous surfaces showing all levels of variation. To begin this process, fire scientists must first acknowledge uncertainty in maps and in the mapping process. Measuring uncertainty in geographic fire products is required for a complete understanding of how to reduce uncertainty in fire mapping projects. Applying measuring processes like epsilon bands for determining uncertainty in mapped fire boundaries will provide much insight. It is this continuous process of recognition, measurement, and improvement in mapping technique that will bring fire mapping to a mature level ensuring data users that all important uncertainty sources have been considered.

ACKNOWLEGMENTS The work for this chapter was completed under NASA contracts NAS10-12180 and NNK08OQ01C. I thank Denise Thaller and Burton Summerfield at NASA for their support. I would also like to thank the Merritt Island National Wildlife Refuge personnel particularly Frederic Adrian for support and help with the case study presented here. I would like to thank Mike Goodchild for his inspiration of the many ideas embedded in this work and for his direct involvement in contributing to this chapter. I would also like to thank Guofan Shao for his contribution to the case study and his editorial help. I would also like to thank Paul Schmalzer and Carlton Hall for their continuous help and support.

REFERENCES Allen, J. L. and Sorbel, B. (2008). Assessing the differenced normalized burn ratio's ability to map burn severity in the boreal forest and tundra ecosystems of alaska's national parks. International Journal of Wildland Fire 17(4): 463-475. Bond, W. J. and van Wilgen, B. W. (1996). Fire and plants. London, Chapman and Hall. Bond, W. J., Woodward, F. I. and Midgley, G. F. (2005). The global distribution of ecosystems in a world without fire. New Phytologist 165(2): 525-537. Brewer, C. K., Winne, J. C., Redmond, R. L., Opitz, D. W. and Mangrich, M. V. (2005). Classifying and mapping wildfire severity: A comparison of methods. Photogrammetric Engineering and Remote Sensing 71(11): 1311-1320. Buttenfield, B. P. (2001). Mapping ecological uncertainty. Spatial uncertainty in ecology: Implications for remote sensing and gis applications. C. T. Hunsaker, M. F. Goodchild, M. A. Friedl and T. J. Case. New York, Springer: 402. Card, D. H. (1982). Using known map category marginal frequencies to improve estimates of thematic map accuracy. Photogrammetric Engineering and Remote Sensing 48(3): 431439.

100

Brean W. Duncan

Cochrane, M. A. (2003). Fire science for rainforests. Nature 421(6926): 913-919. Cochrane, M. A. and Souza, C. M. (1998). Linear mixture model classification of burned forests in the eastern amazon. International Journal of Remote Sensing 19(17): 34333440. Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sended data. Remote Sensing of Environment 37(1): 35-46. Congalton, R. G. and Green, K. (2009). Assessing the accuracy of remotely sensed data: Principles and practices. Boca Raton, CRC Press, Taylor and Francis Group. Couturier, S., Mas, J. F., Cuevas, G., Benitez, J., Vega-Guzman, A. and Coria-Tapia, V. (2009). An accuracy index with positional and thematic fuzzy bounds for land-use/landcover maps. Photogrammetric Engineering and Remote Sensing 75(7): 789-805. Duncan, B. W. and Schmalzer, P. A. (2004). Anthropogenic influences on potential fire spread in a pyrogenic ecosystem of florida, USA. Landscape Ecology 19(2): 153-165. Duncan, B. W., Shao, G. F. and Adrian, F. W. (2009). Delineating a managed fire regime and exploring its relationship to the natural fire regime in east central florida, USA: A remote sensing and gis approach. Forest Ecology and Management 258(2): 132-145. Fernandez-Manso, O., Quintano, C. and Fernandez-Manso, A. (2009). Combining spectral mixture analysis and object-based classification for fire severity mapping. Investigacion Agraria-Sistemas Y Recursos Forestales 18(3): 296-313. Fisher, R., Bobanuba, W. E., Rawambaku, A., Hill, G. J. E. and Russell-Smith, J. (2006). Remote sensing of fire regimes in semi-arid nusa tenggara timur, eastern indonesia: Current patterns, future prospects. International Journal of Wildland Fire 15(3): 307-317. Foody, G. M. and Atkinson, P. M., Eds. (2002). Uncertainty in remote sensing and gis. West Sussex, John Wiley and Sons. Friedl, M. A., McGwire, K. C. and McIver, D. K. (2001). An overview of uncertainty in optical remotely sensed data for ecological applications. Spatial uncertainty in ecology: Implications for remote sensing and gis applications. C. T. Hunsaker, M. F. Goodchild, M. A. Friedl and T. J. Case. New York, Springer: 258-283. Goncalves, L. M. S., Fonte, C. C., Julio, E. and Caetano, M. (2009). A method to incorporate uncertainty in the classification of remote sensing images. International Journal of Remote Sensing 30(20): 5489-5503. Goodchild, M. F. (1994). Integrating gis and remote-sensing for vegetation and analysis and modeling - methodological issues. Journal of Vegetation Science 5(5): 615-626. Gopal, S. and Woodcock, C. (1994). Theory and methods for accuracy assessment of thematic maps using fuzzy-sets. Photogrammetric Engineering and Remote Sensing 60(2): 181-188. Henry, M. C. (2008). Comparison of single- and multi-date landsat data for mapping wildfire scars in ocala national forest, florida. Photogrammetric Engineering and Remote Sensing 74(7): 881-891. Hudak, A. T., Morgan, P., Bobbitt, M. J., Smith, A. M. S., Lewis, S. A., Lentile, L. B., Robichaud, P. R., Clark, J. T. and McKinley, R. A. (2007). The relationship of multispectral satellite imagery to immediate fire effects. Fire Ecology 3(1): 64-90. Hunsaker, C. T., Goodchild, M. F., Friedl, M. A. and Case, T. J., Eds. (2001). Spatial uncertainty in ecology: Implications for remote sensing and gis applications. New York, Springer-Verlag.


101

Jensen, J. R. (2005). Introductory digital image processing: A remote sensing perspective. New Jersey, Prentice Hall. Jordan, G. J., Fortin, M. J. and Lertzman, K. P. (2005). Assessing spatial uncertainty associated with forest fire boundary delineation. Landscape Ecology 20(6): 719-731. Joria, P. E. and Jorgenson, J. C. (1996). Comparison of three methods for mapping tundra with landsat digital data. Photogrammetric Engineering and Remote Sensing 62(2): 163169. Kasischke, E. S., Turetsky, M. R., Ottmar, R. D., French, N. H. F., Hoy, E. E. and Kane, E. S. (2008). Evaluation of the composite burn index for assessing fire severity in alaskan black spruce forests. International Journal of Wildland Fire 17(4): 515-526. Leach, M. K. and Givnish, T. J. (1996). Ecological determinants of species loss in remnant prairies. Science 273(5281): 1555-1558. Legleiter, C. J. and Goodchild, M. F. (2005). Alternative representations of in-stream habitat: Classification using remote sensing, hydraulic modeling, and fuzzy logic. International Journal of Geographical Information Science 19(1): 29-50. Liu, W., Gopal, S. and Woodcock, C. E. (2004). Uncertainty and confidence in land cover classification using a hybrid classifier approach. Photogrammetric Engineering and Remote Sensing 70(8): 963-971. Lowry, J. H., Ramsey, R. D., Stoner, L. L., Kirby, J. and Schulz, K. (2008). An ecological framework for evaluating map errors using fuzzy sets. Photogrammetric Engineering and Remote Sensing 74(12): 1509-1519. Miller, J. D., Knapp, E. E., Key, C. H., Skinner, C. N., Isbell, C. J., Creasy, R. M. and Sherlock, J. W. (2009). Calibration and validation of the relative differenced normalized burn ratio (rdnbr) to three measures of fire severity in the sierra nevada and klamath mountains, california, USA. Remote Sensing of Environment 113(3): 645-656. Minnich, R. A. (1983). Fire mosaics in southern-california and northern baja california. Science 219(4590): 1287-1294. Mowrer, H. T. and Congalton, R. G., Eds. (2000). Quantifying spatial uncertainty in natural resources: Theory and applications for gis and remote sensing. Chelsea, Ann Arbor Press. Robichaud, P. R., Lewis, S. A., Laes, D. Y. M., Hudak, A. T., Kokaly, R. F. and Zamudio, J. A. (2007). Postfire soil burn severity mapping with hyperspectral image unmixing. Remote Sensing of Environment 108(4): 467-480. Rogan, J. and Franklin, J. (2001). Mapping burn severity in southern california using spectral mixture analysis. International Geoscience and Remote Sensing Symposium, Sydney, Australia. Roldan-Zamarron, A., Merino-De-Miguel, S., Gonzalez-Alonso, F., Garcia-Gigorro, S. and Cuevas, J. M. (2006). Minas de riotinto (south spain) forest fire: Burned area assessment and fire severity mapping using landsat 5-tm, envisat-meris, and terra-modis postfire images. Journal of Geophysical Research-Biogeosciences 111(G4). Roy, D. P. and Boschetti, L. (2009). Southern africa validation of the modis, l3jrc, and globcarbon burned-area products. Ieee Transactions on Geoscience and Remote Sensing 47(4): 1032-1044. Salvador, R., Valeriano, J., Pons, X. and Diaz-Delgado, R. (2000). A semi-automatic methodology to detect fire scars in shrubs and evergreen forests with landsat mss time series. International Journal of Remote Sensing 21(4): 655-671.

102

Brean W. Duncan

Sarmento, P., Carrao, H., Caetano, M. and Stehman, S. V. (2009). Incorporating reference classification uncertainty into the analysis of land cover accuracy. International Journal of Remote Sensing 30(20): 5309-5321. Schmalzer, P. A. (2003). Growth and recovery of oak-saw palmetto scrub through ten years after fire. Natural Areas Journal 23(1): 5-13. Schmalzer, P. A. and Hinkle, C. R. (1992). Species composition and structure of oak-saw palmetto scrub vegetation. Castanea 57: 220-251. Shao, G. and Duncan, B. W. (2007). Effects of band combinations and gis masking on firescar mapping at local scales in east-central florida, USA. Canadian Journal of Remote Sensing 33(4): 250-259. Shao, G. F. and Wu, J. G. (2008). On the accuracy of landscape pattern analysis using remote sensing data. Landscape Ecology 23(5): 505-511. Shi, W., Fisher, P. F. and Goodchild, M. F., Eds. (2002). Spatial data quality. London, Taylor and Francis. Smith, A. M. S., Lentile, L. B., Hudak, A. T. and Morgan, P. (2007). Evaluation of linear spectral unmixing and dnbr for predicting postfire recovery in a north american ponderosa pine forest. International Journal of Remote Sensing 28: 5159-5166. Steele, B. M., Winne, J. C. and Redmond, R. L. (1998). Estimation and mapping of misclassification probabilities for thematic land cover maps. Remote Sensing of Environment 66(2): 192-202. Story, M. and Congalton, R. G. (1986). Accuracy assessment - a users perspective. Photogrammetric Engineering and Remote Sensing 52(3): 397-399. White, J. D., Ryan, K. C., Key, C. C. and Running, S. W. (1996). Remote sensing of forest fire severity and vegetation recovery. International Journal of Wildland Fire 6(3): 125136. Wimberly, M. C. and Reilly, M. J. (2007). Assessment of fire severity and species diversity in the southern appalachians using landsat tm and etm plus imagery. Remote Sensing of Environment 108(2): 189-197. Zhan, Q., Molenaar, M. and Lucieer, A. (2002). Pixel unmixing at the sub-pixel scale based on land cover class probabilities: Application to urban areas. Uncertainty in remote sensing and gis. G. M. Foody and P. M. Atkinson. West Sussex, Wiley: 307. Zhang, J. and Foody, G. M. (1998). A fuzzy classification of sub-urban land cover from remotely sensed imagery. International Journal of Remote Sensing 19(14): 2721-2738. Zhang, J. and Goodchild, M. F. (2002). Uncertainty in geographical information. London, Taylor and Francis. Zhu, Z. L., Yang, L. M., Stehman, S. V. and Czaplewski, R. L. (2000). Accuracy assessment for the us geological survey regional land-cover mapping program: New york and new jersey region. Photogrammetric Engineering and Remote Sensing 66(12): 1425-1435.



Chapter 4

AEROSOL AND TRACE GAS RETRIEVALS FROM REMOTE SENSING FIRE PRODUCTS Gabriel Pereira11, Nelson Jesus Ferreira2, Francielle da Silva Cardozo1, Fabrício Brito Silva1, Elisabete Caria Moraes1 ,Yosio Edemir Shimabukuro1, Saulo Ribeiro de Freitas2and Karla Maria Longo3 1

Remote Sensing Division (DSR), National Institute for Space Research (INPE). 2 Center for Weather Forecasting and Climate Studies (CPTEC), National Institute for Space Research (INPE) 3 Center for Space and Atmospheric Sciences (CEA), National Institute for Space Research (INPE)

ABSTRACT Annually, anthropogenic fires devastate large areas of forest and grasslands over the world, releasing a large amount of greenhouse gases and aerosols into the atmosphere. This issue affects the environment, altering the atmospheric and surface radiation balance, besides the biogeochemical and hydrologic cycles. The main objective of this work is to use the fire radiative energy (FRE) release rate to estimate carbon monoxide (CO), particulate matter of less than 2.5 microns in diameter (PM2.5µm) and the amount of biomass consumed by fires in the South America 2002 season. For this, combustion experiments near the Laboratory of Radiometry (LARAD) of Remote Sensing Division at the National Institute for Space Research (DSR/INPE) were performed to obtain the coefficient that relates the consumption of biomass with the Fire Radiative Energy (FRE) released rate. The emission inventory estimated by Moderate Resolution Imaging Spectroradiometer (MODIS) and Wildfire Automated Biomass Burning Algorithm (WFABBA) from Geostationary Operational Environmental Satellites (GOES) Fire Radiative Power (FRP) measurements to the South America 2002 dry season were included in Coupled Chemistry-Aerosol-Tracer Transport model coupled to the Brazilian 1 Corresponding Author. Phone: 55-12-3945-6668, E-mail address: [email protected].

104

Gabriel Pereira, Nelson J. Ferreira, Francielle da Silva Cardozo et al. developments on the regional Atmospheric Modeling System (CCATT-BRAMS). The model results were evaluated with South America 2002 surface data collected in the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) Smoke, Aerosols, Clouds, rainfall and Climate (SMOCC) and Radiation, Cloud and Climate Interactions (RaCCI) experiment. Evaluation of model and ground data revealed a good conformity with SMOCC/RaCCI data in the general pattern of temporal evolution. The results showed high correlations, with values between 0.80 and 0.95 (significant at 0.01 level by student t test), in PM2.5µm and CO emissions simulated in CCATT-BRAMS model. For the period analyzed, biomass consumed by fires can exceed 5 Tg (teragrams) in South America, with a daily average of 2.2 Tg (0.8 Tg estimated by MODIS and 1.32 Tg estimated by GOES). As a result, the coefficient derived from the relationship between biomass consumption and FRE released estimated the biomass burned from July to November 2002 in the South America dry season in approximately 0.28 ± 0.07 Pg (pentagrams).

Keywords: Fire radiative power, Smoke emission coefficient, biomass burning consumption, Amazon Rainforest.

1. INTRODUCTION Every year, wildfires consume immeasurable areas of grassland and tropical forests, releasing a large and unknown amount of aerosols and trace gases into the atmosphere (Crutzen and Andreae, 1990). In South America, temporal and spatial variability in land-use and land-cover due to agricultural land clearing, grassland management and deforestation of the Amazon tropical rainforest leads to variations in anthropogenic biomass burning (Kaufman et al., 1990, 1992; Ward et al., 1992; Werf et al., 2006). Although the impacts of biomass burning on land surface albedo, atmospheric chemistry, atmospheric and surface net radiation, hydrological and biogeochemical cycles and regional climate change have been largely studied (Andreae, 1991; Barbosa et al., 1998; Bremer and Ham, 1999; Rosenfeld, 1999; Andreae and Merlet, 2001; Freitas et al., 2007, Pereira et al., 2009), burning biomass emissions require an accurately estimation to provide a high-quality assessment of the environmental and climate effects of aerosols and trace gases. Traditional biomass emission methods of estimating aerosols and trace gases generally utilize emission factors associated with fuel load characteristics and burn efficiency (Andreae and Merlet, 2001). Furthermore, while emission factors for different species are accurately known, burn efficiency depends on fuel load moisture content (Chuvieco et al., 2004) and burned area which is usually accessible a long time after the fire is over (Roy et al., 2002; Silva et al., 2005). Recently, new methods have been developed for deriving the burned biomass and fire emissions from environmental satellite Fire Radiative Power (FRP) (Wooster, 2002; Wooster et al., 2003; Ichoku and Kaufman, 2005). The FRP can be defined as the part of the chemical energy emitted as radiation in the biomass burning process. The temporal integration of FRP gives the FRE. In theory, radiative intensity released by fires is linearly correlated with the burned biomass and might be independent of vegetation type (Wooster et al., 2005; Freeborn et al., 2008). Also, satellite measurements of FRE released rate could be associated with aerosol optical depth (AOD) to provide regional smoke emission coefficients (Ichoku and Kaufman,

Aerosol and Trace Gas Retrievals from Remote Sensing Fire Products

105

2005). These methods allow near real-time estimation of the concentration of aerosols and trace gases emitted into the atmosphere using chemistry transport models (Chatfield et al., 2002; Horowitz et al., 2003; Freitas et al., 2006). However, FRP-based biomass burned consumption (Wooster et al., 2005; Freeborn et al., 2008) and smoke emission coefficients (Ichoku and Kaufman, 2005; Jordan et al., 2008, Pereira et al., 2009) must be evaluated with ground data to verify the agreement between satellite estimation and ground data. In this study, inventories of carbon monoxide (CO) and particulate matter with diameter smaller than 2.5 micrometers (PM2.5µm) are obtained through FRP smoke emissions and burned biomass coefficients applied in Wildfire Automated Biomass Burning Algorithm (WFABBA) on the Geostationary Operational Environmental Satellite (GOES) and Moderate Resolution Imaging Spectroradiometer (MODIS) on the Terra and Aqua Earth Observing System (EOS) fire products. Thus, to evaluate both methods, the inventory emissions are assimilated into the regional atmospheric transport model Coupled Chemistry-Aerosol-Tracer Transport model coupled to the Brazilian developments on the regional Atmospheric Modeling System (CCATT-BRAMS) adapted for the 2002 South America dry season. Finally, model results of CO and PM2.5µm are compared with ground and atmospheric data collected in the Large Scale Biosphere–Atmosphere in Amazonia (LBA) SMOCC/RaCCI during the 2002 dry-to-wet transition season (Andreae et al., 2004), accuracy of the model results is determined by using linear regression model and standard error of estimate.

2. DATA AND METHODOLOGY 2.1. Study Area South America has a distinct and important biodiversity, spatially distributed in many natural ecosystems such as the Amazon Tropical Rain Forest. However, this complex system is constantly exposed to deforestation, agricultural expansion and anthropogenic burning. Figure 1 shows the study area separated into three large regions with SMOCC/RaCCI flights site location (white circles), according to homogeneous ecosystems characteristics: Amazon Tropical Rain Forest and Brazilian Cerrado in (a); Brazilian Caatinga biome (in northeastern South America) in (b); and Atlantic Forest (tropical and subtropical moist forest, tropical dry forest, tropical savannas, mangrove forests) and grassland biomes in (c).

2.2. Model Description The CATT, an Eulerian transport model fully coupled to BRAMS, is a numerical model that simulates biomass burning emissions, deposition and transport at grid and sub-grid scales. In this model, the trace gas and aerosol emissions, deposition and transport estimation are obtained simultaneously with the evolution of atmospheric condition, using the dynamic and physical parameterizations of the atmospheric model integration. The mass continuity equation for CO and PM2.5µm in the form of a tendency equation is expressed as:

Gaabriel Pereira, Nelson J. Ferrreira, Franciellle da Silva Caardozo et al.

1006 ∂s ∂t

⎛ ∂s ⎞ ⎟ + ∂t ⎠advv ⎝123 3

=⎜

I

⎛ ∂s ⎞ + ⎜ ⎟ ⎝ ∂t ⎠ PBLdiff 1424 3 II

⎛ ∂s ⎞ + ⎜ ⎟ ⎝ ∂t ⎠deepconv 14243 III

⎛ ∂s ⎞ + ⎜ ⎟ ⎝ ∂t ⎠shallow conv 1 14243 IV

⎛ ∂s ⎞ + WPM + R ⎜ ⎟ { + Qpr ⎝ ∂t ⎠chemCO 123 VII { VIII 1424 3 VI 2.5μm

V

(1) where is the grid box meaan tracer mixinng ratio, term w m (I) is the 3-dd transport (addvection by m mean wind), teerm (II) repressents the sub-ggrid scale difffusion in the planetary p bounndary layer (P PBL), terms (III) ( and (IV)) are the subb-grid transpoort by deep and a shallow convection, c reespectively. Teerm (V) is appplied to CO which w is treatedd as a passive tracer with a lifetime of 300 days (Seinfe feld and Pandiis, 1998), term m (VI) is the wet w removal applied a to PM M2.5µm, term (V VII) refers to the dry depoosition appliedd to gases annd aerosol parrticles and, finnally, term (V VIII) is the source s term that includes the plume rise r mechanissm associatedd with the veegetation firess (Freitas et al., 2007; Longoo et al., 2007).

Fiigure 1 – Study area divided innto three large ecosystems e withh white circles that t represent thhe SM MOCC/RaCCI flight sites locaation. (a) Amazon Tropical Raiin Forest and Brazilian B Cerraddo; (b) Brazilian Caatingga biome; (c) Atlantic A Forest and a grassland biiomes.

The modell simulations for 15 July––15 Novemberr 2002 are peerformed withh a 50 km hoorizontal resollution grid andd represent thee South Amerrica dry seasonn. The verticall resolution vaaried telescoppically with hiigher resolutioon at surface (150 m) withh a ratio of 1..07 up to a m maximum vertiical resolutionn of 850 m, with w the top model m altitude of 20 km in 38 vertical leevels. CO and d meteorologiccal fields are initialized i usinng horizontallly homogeneoous profiles asssociated with h a backgrounnd situation. Also, A model iss run during a period of 155 days with soources for the spin-up. Lateeral boundary condition is defined d as havving constant inflow and raadiative variaable outflow. CO emissionns from urbaan-industrial vehicular v activities and


107

biogenic sources are not included in this study, because in the areas where the model discussion was focused biomass burning evaluation is the most important emission source. The total time integration model was 124 days, and for these days the initial atmosphere and boundary conditions were acquired from the European Center for Medium range Weather Forecasting (ECMWF) and from Center for Weather Forecasting and Climate Studies (CPTEC/INPE) models.

2.3 Fire Products and FRP-Based Coefficients The fraction of chemical energy emitted from burning biomass as radiation can be defined as FRP, and the temporal integration of FRP gives the Fire Radiative Energy (FRE). Initial studies with FRP were performed with the MODIS Airborne Simulator (MAS) in the SCAR-C and SCAR-B (Smoke, Cloud and Radiation - California / Brazil) experiments (Kaufman et al., 1998). FRP is available from the MODIS fire products, also called MOD14 (Terra) and MYD14 (Aqua), which utilize a contextual algorithm applied to brightness temperatures in the 4 µm and 11 µm infrared radiation channels (Justice et al., 2002; Giglio, 2005). The FRP is calculated for each fire detection with the method proposed by Kaufman et al. (1996, 1998a, 1998b). Moreover, while MODIS gives information of active fires 4 to 5 times per day for the same area, the WFABBA fire product, based on GOES observations, is available for South America with a higher observation frequency,, approximately 48 times per day, and with a nominal spatial resolution of 4 x 4 km at NADIR. The WFABBA algorithm for the detection of fire pixels, as well as MODIS, uses two bands, located on the 3.9 µm and 10.7 µm channels (Prins et al., 1998). Whereas MODIS fire products use a semi-empirical formula to estimate FRP, WFABBA/GOES products are not included operationally in this estimate. Thus, the Stefan-Boltzmann law (Equation 2) is used to calculate this physical property for GOES data. 4 FRP = Afire .σ .Tfire

(2)

where A fire represents the fire fractional area (m²), σ is the Stefan-Boltzmann constant (5.67x10-8 W.m-2.K-4), T fire is the fire temperature and FRP is the Fire Radiate Power in Megawatts (MW or MJ.s-1). To assess GOES’s capability to retrieve regional-scale FRP, Pereira et al. (2009) evaluated MODIS and GOES products, showing a correlation greater than 96% (significant at 0.05 level by student t test) between monthly FRP measurements derived from both sensors. Also, due to GOES instrument characteristics approximately 6% of 2000-2008 fires detected by WFABBA/GOES presented an early saturation in mid-infrared channels. For saturated pixels, we can use several methods which involve a variety of errors in this process, consequently, the technique described in Equation (3) was chosen due to the lack of background information in WFABBA/GOES fire products.

108

Gabriel Pereira, Nelson J. Ferreira, Francielle da Silva Cardozo et al.

GOES FRP =

A fire a

.σ

4.03

∫

B (λ , T ) d λ − Lb

(3)

3.76

where a is a constant fit based on the GOES MIR spectral channel (Wooster et al., 2005), B(λ,T) is the Planck’s Radiation Law, λ is wavelength (µm), T is temperature (K) and Lb is the spectral radiance emitted by background (110 MW).

2.4 Source Emission Parameterization and Evaluation With Observed Data Table 1 shows the FRE-based smoke emission coefficients for MODIS (Ichoku and Kaufman, 2005) and GOES (Pereira et al., 2009) fire products, the emission factors acquired from Wiedinmyer et al. (2006) used to estimate PM2.5µm and CO emissions and the biomass consumption coefficient (BCC) derived in sixteen small-scale combustion experiments performed in the Laboratory of Radiometry (LARAD) of Remote Sensing Division (DSR) at the National Institute for Space Research (INPE). To calculate the South America 2002 emissions inventory, the three FRE-based smoke emissions coefficients for generic biomes that provide the smoke flux are converted in PM2.5µm and CO emission values by a relationship between biomes particulate matter with a diameter less than 10 µm (PM10µm) and with respectively PM2.5µm and CO emission factors. Also, the same emission factors are used to convert the time-integrated biomass burned in aerosols and trace gases emissions. Table 1 – FRE-based smoke emission coefficients (kg.MJ-1) for MODIS and GOES for Amazon Tropical Rain Forest and Brazilian Cerrado (a), Brazilian Caatinga biome (b), Atlantic Forest and grassland biomes (c); Average emission factors (EF) for PM10µm, PM2.5µm and CO (g.kg-1); PM2.5µm and CO ratio assigned in inventory estimation; Biomass Consumption Coefficient (BCC) in kg.MJ-1

Ecosystems (a) (b) (c)

MODIS Ce 0.063 0.048 0.061

GOES Ce 0.03 0.006 0.02

PM10µm EF 12.5 6.9 6.9

PM2.5µm EF 9.9 5.6 5.6

CO EF 117.0 84.0 84.0

PM2.5µm Ratio 0.79 0.81 0.81

CO Ratio 9.4 12.2 12.2

BCC 0.95 0.95 0.95

The CCATT-BRAMS model simulations of CO and PM2.5µm are compared with ground data from LBA SMOCC/RaCCI campaign, collected near 62.37o W and 10.75o S in the Amazon basin. The PM2.5µm and CO near-surface measurements were made at the Ouro Preto do Oeste pasture site from September to November 2002. The PM2.5µm particle mass concentration was measured with a TEOM (Tapered Element Oscillation Mass Balance) with a 30-min temporal resolution from 10 September to 4 November 2002. Therefore, an intercomparison of the PM2.5µm and CO model results at 12:00 UTC with the daily average centered at 12:00 UTC was done to evaluate these simulations. Moreover, the accuracy of the model results was determined by using linear regression model, standard error of estimate and the student t test.


109

Furthermore, assessments between simulated CO profiles in the planetary boundary layer (PBL) and lower troposphere with SMOCC/RaCCI campaign airborne measurements were performed. The airborne CO profiles measurements were acquired using an Aero-Laser (AL5002) instrument operating at 1Hz (Guyon et al., 2005; Freitas et al., 2007) onboard the INPE Bandeirante aircraft during September and October of 2002. The sixteen flights took place over the deforestation arc with a high incidence of vegetation fires situated in Rondônia, Acre, Amazonas and Mato Grosso States (Figure 1). The typical maximum altitude reached by the SMOCC/RaCCI aircraft was 5km and the measurement accuracy is approximately ±5%.

3. RESULTS AND DISCUSSIONS 3.1. CCATT-BRAMS Surface Simulation The time series and normalized index among the CCATT-BRAMS model simulation by smoke emission and biomass burning coefficients and ground data collected in the LBA SMOCC/RaCCI campaigns are shown in Figure 2. During September to November, the Amazon basin is largely dominated by the South Atlantic Subtropical High (SASH), switching the convection patterns to the northwest of South America, decreasing the precipitation rate and increasing the fire activity (Freitas et al., 2007). The time series of PM2.5µm (Figure 2a and Figure 2c) and CO (Figure 2b and Figure 2d) show the biomass burning emissions from the model results at 12:00 UTC with the daily average of ground measurements centered at 12:00 UTC, revealing a good conformity between temporal evolution of the model simulation and the observed data. The normalized index, shown at the top of each simulation, represents the difference between the modeled and observed data, showing positive values wherever emissions values collected in LBA SMOCC/RaCCI are greater than emissions modeled by CCATT-BRAMS FRP method. During the CCATT-BRAMS simulation, three distinct regimes of rainfall could be observed in the Amazon basin. The period from September to the beginning of October shows characteristics of dry season with low precipitation rates and a high number of detected fires not only in experiment site locations in Rondônia but over the whole South America region. These characteristics are evidently noticed on the surface PM2.5µm and CO measurements. During this period, PM2.5µm and CO values can reach approximately 160 µg.m-3 and 2400 ppb, respectively. It is important to note that the time series show a strong variability, showing that the fires are close to site location during this period. In the second period, from early to the end of October, the Amazon basin showed an increasing precipitation rate, causing a reduction in the occurrence of burning biomass at the site, but a few hot spots are still observed. From the end of October to the beginning of November, the wet season reduces the number of fires in the whole Amazon Basin. This reduction is evidenced by surface PM2.5µm and CO values. Moreover, model time series results have shown a good agreement with observed data.

110


Fiigure 2 – Time series with nearr surface PM2.55µm (µg.m-3) and a CO (ppb) estimated from the t smoke em missions coefficcients (light greey) with the obsserved data colleected in LBA SMOCC/RaCCI S I (dark grey) inn (a) and (b); Tim me series with near n surface PM M2.5µm (µg.m--3) and CO (ppbb) estimated by burning biiomass consump ption coefficiennt (light grey) with w observed daata in (c) and (dd).

To analyzee CCATT-BR RAMS model uncertainties, the bootstrapp technique was w used to caalculate the coorrelation andd slope for thee four simulattions (Efron, 1982). In this method, a poopulation of 1.0x104 reconsstructs the origginal curve andd provides thee parameters too create the coonfidence inteerval for the model m estimatee. The linear regression with standard errror bars to siimulate smokee emission annd biomass consumption deerived PM2.5µm ncentrations m and CO con annd LBA SMO OCC/RaCCI ground data is shown in Figure 3. Allso, at the sidde of each reegression grap ph, the confiddence interval for slope (abbove) and corrrelation (below) derived frrom bootstrap technique is shown. s Through thhe bootstrap teechnique, the highest correllation frequenccies for PM2.55µm and CO arre found betw ween 0.80 to 0.95 (significannt at 0.01 leveel by student t test) for bothh methods. Fuurthermore, thhe slope analyysis reveals a considerable c siimilarity betw ween the aerosool emission esstimates. In bo oth methods, an a underestim mation of PM2.55µm emission is i evident, witth CCATTB BRAMS modeled values 20% % lower than observed data. The undereestimation occcurs mainly

Aerrosol and Tracce Gas Retrievvals from Rem mote Sensing Fire F Products

111

onn days with large fires. The T PM2.5µm estimation by b smoke em mission method, initially prroposed by Icchoku and Kaaufman (2005) for MODIS S fire productss and adaptedd to GOES (P Pereira et al., 2009), show ws overestim mated values on days withh low burninng biomass obbservations su uch as in the November N weet transition and a underestim mated emissioons in large fires. Also, the PM2.5µm simuulated with thhe burned biom mass consumpption coefficieent initially prroposed by Wooster W et al. a (2005) annd performedd in the Labboratory of Radiometry R (L LARAD/INPE E) showed the same tendenccy to underestiimate high em missions and ovverestimate loow fire emissio ons. As shown in Figures 3cc and 3d, CO emissions were w underestim mated by bothh methods. Fuurthermore, th he smoke emission method estimated higgh fire emissioons more accuurately than thhe burning bioomass consum mption coefficient method, with slopes frrom 1.0-1.1 and a 1.2-1.4, reespectively. Also, A this resuult shows a good g agreemeent between ground g and model m data. H However, when n the intensityy of the burning biomass iss too high, thee values modeeled by the FR RE-derived coefficients c arre underestim mated, probablly due to the influence off smoke in saatellite FRP measurements. m

Fiigure 3 . Linearr regression withh standard errorr bars between PM P 2.5µm (µg.m-33) and CO (ppb)) estimated frrom the smoke emissions e coeffficients (a) and (c) and by biom mass consumptioon coefficient (b) and (d), reespectively.

112


Fiigure 4 – Averaage South Amerrica CO (ppb) biomass b burningg emissions at thhree different altitudes m modeled in CCA ATT-BRAMS beetween 15 July 2002 and 15 November 2002. (a) - (c) shows daily avverage CO inveentories estimateed by smoke em mission coefficients for surfacee, 1600 m and 7000 7 m with avverage wind streeam, respectiveely. (d) - (f) reprresent average values v of CO thhrough biomass burning coonsumption metthod.

3.2. CO Spattial Distribu ution and Prrofiles In South America, A num merous weatheer systems coould change the t transport of burning biiomass emissiions, modifyinng the chemical air compoosition, the raadiation budgeets and the cllouds microphhysical propeerties (Freitass et al., 20077). Figure 4 shows CO (ppb) ( daily avveraged values for the Soutth America biomass burning season in thhree vertical atmospheric leevels, obtained d through smooke emission (a ( - c) and biom mass consumpption (d - f) methods. m As shhown in Figurre 4a and 4d, there t are conssiderable diffeerences in surfface CO valuees modeled by y FRP-based methods. m Average vaalues of CO simulated s withh the smoke emission e methhod (Figure 4a) 4 seem to esstimate large fires more accurately in thhe Southwest of o South Ameerica (below 20º 2 S) than thhe burned biom mass method (Figure 4d) ass shown in sloope values derived from thee bootstrap teechnique. Thee highest inccidence of firres occurs inn the region known as thhe “arc of deeforestation”, where the areea average suurface values of o CO could be b greater thaan 600 ppb, w maximum with m daily values between b 2000 and 3500 ppbb. In the Amazon basin, fiire flame temp mperature can exceed 1600 Kelvin (K), and a usually raanges between n 830 – 1440 K (Riggan et al., 2004). With W the high FRP F intensity released in biiomass burninng activity, thee plume emitteed into the atm mosphere can reach r more thaan 6,000 m

Aerrosol and Tracce Gas Retrievvals from Rem mote Sensing Fire F Products

113

abbove ground level. Figures 4b; 4 4e; 4c andd 4f show the CO concentraation at 1700 m and 7000 m above grounnd level for thhe smoke em mission and biomass burnedd consumptionn methods, reespectively. These T figures show the trransport of biomass b burniing emissionss from the A Amazon basin to subtropicall regions, in thhe Amazon baasin transport to t the eastern side of the coontinent is caaused by thee Andes Mouuntains and thhe South Atllantic Subtroppical High (S SASH), and too the African continent c (70000 m) by high levels jets. Average ob bserved CO values v (ppb) with w standard deviation aree shown in Figure 5 (in bllack and light grey, respectiively) followeed by data modeled throughh FRP-based methods m (in dootted dark greey). CO profilees assessmentt indicates a good g agreemennt between moodeled data annd observed SMOCC/RaCC S CI airborne meeasurements. As shown in i Figure 5a, the t FRP-basedd method initiaally proposed by Ichoku andd Kaufman (22005) presented a better fit fi between CO O surface proofiles, frequenntly inside thhe standard deeviation rangee. In initial verrtical levels thhe modeled daata is quite cloose to the observed data. Fuurthermore, between b 1500 m and 30000 m above ground levvel the CO profile is unnderestimated d due to diffficulties in estimating ann appropriatee spatial andd temporal diistribution of the t emission source s energy necessary forr the plume risse mechanism coupled in C CCATT-BRAM MS (Freitas ett al., 2007). Inn upper verticcal levels CO values presennted a good aggreement. Mo oreover, the buurned biomasss consumptionn coefficient generated g low w values for C profiles inn initial and middle CO m verticaal levels, duee to underestiimated burninng biomass em missions at thee surface.

Fiigure 5 – CO prrofiles from SM MOCC/RaCCI data d (black line)) with standard deviation d (lightt grey) and m modeled data at the t same verticaal levels using (a) ( the smoke emission methodd and (b) the bioomass coonsumption coeefficient.

3.3. 2002 Sou uth Americaa Biomass Burned B The total of o above-grounnd biomass in the Amazon basin b and in other o ecosystem ms exhibits siignificant variiations dependding on the methodology addopted. Factorrs such as the amount of caarbon in vegeetation and the carbon sequuestration in burned b area are a difficult too calculate. Saaatchi et al. (22006) estimateed the total biomass in the Amazon A basinn as 86 Pg (peentagrams), appproximately 300 to 400 tons t per hectaare (ha). Furtthermore, the total amountt of above-3 grround biomasss for the Am mazon basin was w estimated as 4 to 15 kg.C.m k by Olson O et al. (22002). Many studies s such as a Fearnside and a Barbosa (11998), Houghton (1999), Houghton H et all. (2001), Houughton and Hackler H (2006)), Fernandes et e al. (2007) examined e the amount of

114


biiomass consumed by fires in South Am merica and othher continentss, however a consent in biiomass estimaation is not so evident. Using FRP P-based consuumption coeffiicient derivedd from the LA ARAD/INPE experiment, e thhe burned biom mass in Southh America 20002 was estimaated for 15 Juuly - 15 Noveember 2002 tim me period. Figure F 6 show ws the Southh America daily d biomass burned estim mation (in teeragrams – Tgg) through thee biomass connsumption metthod applied to t WFABBA//GOES and M MODIS/EOS FRP F measurem ments. In Souuth America, the highest consumption c o biomass of occcurs in Augu ust, Septemberr and October,, when the daiily burned biom mass could exxceed 5 Tg, w average off 2.2 Tg (0.888 Tg estimatedd by MODIS and 1.32 Tg estimated with e by GOES G FRP daata). Howeverr, the presencee of clouds andd the lack of satellite s imagees generate low w values in thhe burned biom mass graph evven in intense fire periods. These featurees are related to t different w weather system ms acting in Soouth America and a due to cloouds occurrencce. Furthermorre, the FRP-bbased method adapted from m Wooster et al. (2005) esstimated in 0..28 ± 0.07 Pgg the amount of o biomass buurned in the South S Americaa 2002 fire seeason. This vaalue is similarr to those founnd in the literaature, such as Houghton et al. a (1999) andd Tian et al. -1 (11998), 0.2 Pg C.year C and 0.3 0 Pg C.year-11, respectivelyy.

4. CONCLUSION O NS A biom mass burning presents p a spaatial and tempporal variabilitty, directly In South America, asssociated with h land-use annd land-cover management.. The highestt incidences of o fires are loocated in the arc a of deforesttation, locatedd in the Amazoon forest bordder. Moreover,, the use of FR RP products is very promising in estimatting the trace gases g and aeroosols emitted by b biomass bu urning in nearr-real time. The FRP-bbased smoke emission e and burned biomaass consumptiion coefficiennts, derived thhrough the rellationship betw ween the smooke emission rate r (kg.s-1), burned b biomasss (kg) and FR RE released rate r (MW), shows s a goodd conformity with w the grouund data from m the LBA SM MOCC/RaCC CI campaigns and a has the pootential to be a new methodd for estimatinng trace gas annd aerosol em missions. How wever, the FRP P-based smokke emission cooefficients proomise great im mprovement, in i accuracy and a new coeff fficients for sm mall areas couuld be created in South A America, reduccing the variabbility in regionnal coefficients which coverr different biom mes.

Fiigure 6 – Daily biomass burnedd by fires in thee 2002 South Am merica dry seasson estimated byy biomass coonsumption coeefficient and GO OES and MODIIS FRP measureements.


115

REFERENCES Andreae, M.O. (1991). Biomass burning: Its history, use and distribution and its impact on the environmental quality and global climate. In J.S. Levine (Ed.), Global biomass burning: Atmospheric, climatic and biospheric implications (pp. 2-21). Cambridge: MIT Press Andreae, M.O., and Merlet, P. (2001). Emission of trace gases and aerosols from biomass burning. Global Biogeochem. Cycles, 15, 955-966 Andreae, M.O., Rosenfeld, D., Artaxo, P., Costa, A., Frank, G., Longo, K. M., Silva Dias, M.A.F. (2004). Smoking rain clouds over the Amazon. Science, 303, 1342-1345 Barbosa, P.M., Pereira, J.M.C., Grégoire, J.M. (1998). Compositing criteria for burned area assessment using multitemporal low resolution satellite data. Remote Sensing of Environment, 65, 38-49 Bremer, D.J., and J.M. Ham (1999). Effect of spring burning on the surface energy balance in a tallgrass prairie. Agric. Forest Meteorol., 97, 43-54 Chatfield, R., Guo, Z., Sachse, G., Blake, D., and Blake, N. (2002). The subtropical global plume in the Pacific Exploratory Mission-Tropics A (PEM-Tropics A), PEM-Tropics B, and the Global Atmospheric Sampling Program (GASP): How tropical emissions affect the remote Pacific. J. Geophys. Res., 107, 4278, doi:10.1029/2001JD000497 Chuvieco, E., Cocero, D., Aguado, I., Palacios, A., Prado, E. (2004). Improving burning efficiency estimates through satellite assessment of fuel moisture content. J. Geophys. Res., 109, D14S07, DOI:10.1029/2003JD003467 Crutzen, P.J., and Andreae, M.O. (1990). Biomass burning in the tropics: impact on atmospheric chemical and biochemical cycles. Science, 250, 1669-1678 Efron, B. (1982). The jackknife, the bootstrap, and other resampling plans. Society of Industrial and Applied Mathematics CBMS-NSF Monographs, 38 Fearnside, P.M., and Barbosa, R. I. (1998). Soil carbon changes from conversion of forest to pasture in Brazilian Amazonia. For. Ecol. Mgmt, 108, 147-166 Fernandes, S. D., Trautmann, N.M., Streets, D.G., Roden, C.A., Bond, T.C. (2007). Global biofuel use, 1850?2000. Global Biogeochemical Cycles, 21,GB2019 Freeborn, P.H., Wooster, M.J., Hao, W.M., Ryan, C.A., Nordgren, B.L. Baker, S.P., Ichoku, C. (2008). Relationships between energy release, fuel mass loss, and trace gas and aerosol emissions during laboratory biomass fires. J. Geophys. Res., 113, D01102, 10.1029/2007JD008489 Freitas, S.R., Longo, K.M., Chatfield, R., Latham, D., Silva Dias, M.A.F., Andreae, M.O., Prins, E., Santos, J. C., Gielow R., Carvalho Jr., J.A. (2006). Including the sub-grid scale plume rise of vegetation fires in low resolution atmospheric transport models. Atmos. Chem. Phys. Discuss., 6, 11,521-11,559 Freitas, S.R., Longo, K., Dias, M., Chatfield, R., Dias, P., Artaxo, P., Andreae, M., Grell, G., Rodrigues, L., Fazenda, A., Panetta J. (2007). The Coupled Aerosol and Tracer Transport model to the Brazilian developments on the Regional Atmospheric Modeling System (CATT-BRAMS). Part 1: Model description and evaluation. Atmos. Chem. Phys. Discuss., 7, 8525-8569 Giglio, L. (2005). MODIS Collection 4 Active Fire Product User’s Guide. Available on the Internet: http://maps.geog.umd.edu/products/ MODIS_Fire_Users_Guide_2.1.pdf .

116


Guyon, P., Frank, G., Welling, M., Chand, D.; Artaxo, P., Nishioka, G., Rizzo, L. V., Lloyd, J., Kolle, O., Silva Dias, M. A. F., Gatti, L. V., Cordova, A. M., Andreae. M. O. (2005). Airborne measurements of trace gases and aerosol particles emissions from biomass burning in Amazonia. Atmospheric Chemistry and Physics Discussions, 5, 2791-2831 Horowitz, L., Walters, S., Mauzerall, D., Emmons, L., Rasch, P., Granier, C., Tie, X., Lamarque, J.-F., Schultz, M., and Brasseur, G. (2003). A global simulation of tropospheric ozone and related tracers: Description and evaluation of MOZART, version 2. J. Geophys. Res., 108, 4784, doi:10.1029/2002JD002853 Houghton, R. A. (1999). The annual net flux of carbon to the atmosphere from changes in land use 1850-1990. Tellus, 51, 298-313 Houghton, R.A., Skole, D.L., Nobre, C.A., Hackler, J.L., Lawrence, K.T., Chomentowski, W.H. (1999). Annual fluxes of carbon from deforestation and regrowth in the Brazilian Amazon. Nature, 403, 301-304 Houghton, R.A., Lawrence, K. T., Hackler, J. L., Brown, S. (2001). The spatial distribution of forest biomass in the Brazilian Amazon: a comparison of estimates. Global Change Biol., 7, 731-746 Houghton, R. A., and Hackler, J. L. (2006). Emissions of carbon from land use change in subSaharan Africa. Journal of Geophysical Research, 111, G02003, doi:10.1029/ 2005JG000076 Ichoku, C., Kaufman, Y.J. (2005). A method to derive smoke emission rates from MODIS fire radiative energy measurements. IEEE Transactions on Geoscience and Remote Sensing, 43, 2636-2649 Jordan, N.S., Ichoku, C., Hoff, R.M. (2008). Estimating smoke emissions over the US Southern Great Plains using MODIS fire radiative power and aerosol observations. Atmospheric Environment, 42, 2007-2022 Justice, C.O., Giglio, B., Korontzi, S., Owens, J., Morisette, J.T., Roy, D.P., Descloitres, J., Alleaume, S., Petitcolin, F., Kaufman, Y. (2002). The MODIS fire products. Remote Sensing of Environment, 83, 244-262 Kaufman, Y.J., Tucker, C.J., Fung, I. (1990). Remote sensing of biomass burning in the tropics. J. Geophys. Res., 95, 9927-9939 Kaufman, Y.J., Setzer, A.W., Ward, D., Tanre, D., Holben, B.N., Menzel, P., Pereira, M.C., Rasmussen, R. (1992). Biomass Burning Airborne and Spaceborne Experiment in the Amazonas (BASE-A). J. Geophys. Res., 97, 14,581-14,599 Kaufman, Y. J., Remer, L.A., Ward, D.E., Kleidman, R., Flynn, L., Shelton, G., Ottmar, R.D., Li, R.-R., Fraser, R.S., McDougal, D. (1996). Relationship between remotely sensed fire intensity and rate of emission of smoke: SCAR-C Experiment. In J.S. Levine (Ed.), Global biomass burning: Atmospheric, climatic and biospheric implications (pp. 685696). Cambridge: MIT Press Kaufman, Y.J., Justice, C.O., Flynn, L., Kendall, J.D., Prins, E.M., Giglio, L., Ward, D.E., Menzel, W.P., Setzer, A.W. (1998). Potential global fire monitoring from EOS-MODIS. Journal of Geophysical Research, 103, 32,215-32,238 Longo, K., Freitas, S.R., Setzer, A., Prins, E., Artaxo, P., Andreae, M. (2007). The Coupled Aerosol and Tracer Transport model to the Brazilian developments on the Regional Atmospheric Modeling System (CATT-BRAMS). Part 2: Model sensitivity to the biomass burning inventories. Atmos. Chem. Phys. Discuss., 7, 8571-8595


117

Olson, J.S., Watts, J.A., Allison, L.J. (2002). Major World Ecosystem Complexes Ranked by Carbon in Live Vegetation: A Database (Revised November 2000). Available on internet: http://cdiac.esd.ornl.gov/ndps/ndp017.html Pereira, G., Freitas, S. R., Moraes, E. C., Ferreira, N. J., Shimabukuro, Y. E., Rao, V. B., Longo, K.M. (2009). Estimating trace gas and aerosol emissions over South America: Relationship between fire radiative energy released and aerosol optical depth observations. Atmospheric Environment, 43, 6388-6397 Prins, E.M., Felts, J.M., Menzel, W.P., Ward, D.E. (1998). An overview of GOES-8 diurnal fire and smoke results for SCAR-B and 1995 fire season in South America. J. Geophys. Res., 103, 31,821–31,835 Riggan, P., Tissell, R., Lockwood, R., Brass, J., Pereira, J., Miranda, H., Campos, T., Higgins, R. (2004). Remote measurement of energy and carbon flux from wildfires in Brazil. Ecol. Appl., 14, 855-872 Rosenfeld, D. (1999). TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall. Geophys. Res. Lett., 26, 3105-3108 Roy, D.P., Lewis, P.E., Justice, C.O. (2002). Burned area mapping using multi-temporal moderate spatial resolution data a bi-directional reflectance model-based expectation approach. Remote Sensing of Environment, 83, 263-286 Saatchi, S.S., Houghton, R.A., Alvará, R.C.S., Soares, J.V., Yu, Y. (2006). Distribution of Aboveground Live Biomass in the Amazon Basin. Available on the Internet: http://wwwradar.jpl.nasa.gov/carbon/ab/fbc.htm Seinfeld, J. and Pandis, S. (1998). Atmospheric Chemistry and Physics. John Wiley and Sons Inc., New York. Silva, J.M.N., As, A.C.L., Pereira, J.M.C. (2005). Comparison of burned area estimates derived from SPOT-VEGETATION and Landsat ETM+ data in Africa: Influence of spatial pattern and vegetation type. Remote Sensing of Environment, 96, 188-201 Tian, H., Melillo, J.M., Kicklighter, D.W., Mcguire, A.D., Helfrich, J.V.K., Moore, B., Vörösmarty, C.J. (1998). Effect of interannual climate variability on carbon storage in Amazonian ecosystems. Nature, 396, 664-667 Ward, D.E., Susott, R.A., Kauffman, J.B., Babbitt, R.E., Cummings, D.L., Dias, B., Holben, B.N., Kaufman, Y.J., Rasmussen, R.A., Setzer, A.W. (1992). Smoke and Fire Characteristics for Cerrado and deforestation Burns in Brazil: BASE-B Experiment. Journal of Geophysical Research, 97, 14,601-14,619 Werf, G.R., Randerson, J.T., Giglio, L., Collatz, G.J., Kasibhatla, P.S., Arellano Jr., A.F. (2006). Interannual variability in global biomass burning emissions from 1997 to 2004. Atmospheric Chemistry and Physics, 6, 3423-3441 Wiedinmyer, C., Quayle, B., Geron, C., Belote, A., McKenzie, D., Zhang, X., O’Neill, S., Wynne, K.K. (2006). Estimating emissions from fires in North America for air quality modeling. Atmospheric Environment, 40, 3419–3432 Wooster, M.J. (2002). Small-scale experimental testing of fire radiative energy for quantifying mass combusted in natural vegetation fires. Geophys. Res. Lett., 29, doi:10.1029/2002GL015487 Wooster, M.J., Zhukov, B., Oertel, D. (2003). Fire radiative energy for quantitative study of biomass burning: Derivation from the BIRD experimental satellite and comparison to MODIS fire products. Remote Sens. Environ., 86, 83-107

118


Wooster, M.J., Roberts, G., Perry, G.L.W., Kaufman, Y.J. (2005). Retrieval of biomass combustion rates and totals from fire radiative power observations: FRP derivation and calibration relationships between biomass consumption and fire radiative energy release. J. Geophys. Res., 110, D24311, doi:10.1029/2005JD006318



Chapter 5

THE ROLE OF MAGNETIC MEASUREMENTS IN DETECTING PAST FIRE SIGNATURES IN SOILS AND SEDIMENTS Frank Oldfield Department of Geography, School of Environmental Science, University of Liverpool, Liverpool, United Kingdom

ABSTRACT Fire can transform both paramagnetic and imperfect antiferromagnetic iron minerals with low magnetic susceptibility and, in the case of paramagnets, zero magnetic remanence, into strongly magnetic minerals, often with high susceptibility values and distinctive remanence characteristics. The present chapter outlines the ways in which magnetic measurements can be used to detect fire signatures in soils and sediments. The magnetic products of burning that are most readily detected and characterized are fine grained, dominantly superparamagnetic, ferrimagnetic minerals (maghemite and nonstoichiometric magnetite). By using a combination of magnetic susceptibility and remanence measurements, these can sometimes be distinguished from ferrimagnetic minerals produced by other processes. The research applications considered here include archaeological prospecting, reconstructions of fire histories from sedimentary evidence and sediment tracing.

INTRODUCTION The first papers to document systematically the role of fire in creating and transforming magnetic minerals were those of Le Borgne (1955;1960). He showed that the magnetic susceptibility of soils heated in air increased dramatically, and hypothesized that weakly magnetic iron minerals were converted to magnetite during heating under the reducing conditions created by organic matter combustion, then partially oxidised to maghemite on cooling in air. Subsequent studies have included papers dealing with the stoichiometry and magnetic grain size of the ferrimagnetic minerals formed by a natural fire (Longworth et al. 1979), demonstrations of the incorporation and survival of fire-enhanced minerals in the

120

Frank Oldfield

allochthonous component of lake sediments during and after the vegetation and soils of lake catchments have been burnt (Rummery 1981; Rummery et al. 1979; Gedye et al. 2000), the use of magnetic susceptibility to detect archaeological sites affected by fire (Oldfield et al. 1985; Thompson and Oldfield,1986; Crowther, 2003), the discrimination of the fine-grained ferrimagnetic minerals in fire-enhanced soils from those arising from pedogenesis in the absence of fire (Oldfield and Crowther, 2007), the use of artificially fire-enhanced river bedload in tracing experiments (Oldfield et al. 1981; Arkell et al. 1983) and similarly treated beach sand in coastal sediment tracing (van der Post et al. 1994), and the use of magnetic measurements in surface process studies where fire has been a major agent (Blake et al. 2006).

FIRE-INDUCED MAGNETIC MINERALS AND DIAGNOSTIC SIGNATURES Field observations and laboratory experiments (e.g. Rummery, 1981, Oldfield et al. 1981) confirm that the magnetic minerals produced by fire are varied, depending on the iron content of the material fired, the rate of heating and cooling, the peak temperature reached and the degree to which atmospheric conditions during heating and cooling are reducing or oxidizing. At one extreme, slow heating and cooling in an oxidizing atmosphere throughout is likely to give rise to mainly imperfect anti-ferromagnetic haematite (αFe2O3), unless the material heated is self-reducing. At the other extreme, rapid heating and cooling with reducing conditions maintained throughout, leads to magnetite (Fe3O4) formation. Where partial oxidation occurs during cooling, the likely product will be maghemite (γFe2O3). Both magnetite and maghemite have cubic spinel structures and are ferrimagnetic with susceptibility values per unit mass several orders of magnitude higher than paramagnetic or imperfect anti-ferromagnetic minerals. As Longworth et al. (1979) show from Mossbauer experiments, the products of a wildfire in the field included non-stoichiometric magnetite approximating to Fe2.9O4. Assuming a reducing atmosphere at least during heating, followed by rapid cooling, enhancement from burning can begin at temperatures as low as 100oC, though significant increases in susceptibility are slight below ~ 200oC and tend to peak above 500oC. Given the diversity of possible magnetic mineral products from burning, the wide range of variables involved and the general lack of carefully controlled experimental data documenting all their effects, it is, at first sight, rather surprising that there is any prospect for detecting fire signatures from magnetic measurements. In reality, detection depends on the extent to which the magnetic minerals in burnt materials include ferrimagnets, on the high magnetic susceptibility values of the latter and on the fine-grained nature of those minerals produced rapidly by burning. In the case of soils, the combination of combustible organic matter, with its reducing potential, at or near the surface, and iron-rich mineral soil immediately below, provides a suitable context for magnetic enhancement by burning. Even the low iron content in most peats can, if sufficiently concentrated by combustion of all the organic matter, produce an extremely magnetic ash residue. It follows from the above, that in some contexts, as for example many archaeological sites, magnetic susceptibility measurements alone may suffice to identify locations and patterns of burning. In other situations, it is necessary to look for the distinctive magnetic

The Role of Magnetic Measurements in Detecting Past Fire Signatures…

121

properties of the fine-grained ferrimagnets generated by burning. Note that grain size here refers to magnetic grain size and not particle size. Although the two are inevitably not completely independent, their relationship is far from straightforward (Oldfield et al. 2009) and the distinction between the two is essential in this account. Magnetic susceptibility gives no more than a rough indication of the concentration of ferrimagnets and it tells us nothing about magnetic grain size. Within the sub-micron size range, ferrimagnets show contrasting behaviours depending on whether or not the magnetic grain size lies above or below the threshold, around 20-25nm in diameter, below which thermal randomization at room temperature overcomes retention of a magnetic remanence once a sample is removed from the magnetizing field. The finest grains, below this threshold, are terms superparamagnetic (SP). Above this threshold, responses to magnetic fields in the laboratory depend on whether one or more magnetic domains exist within each grain. The finest grains above the threshold are termed Stable Single Domain (SD). Detecting and characterizing fire-induced magnetic minerals depends on the predominantly SP nature of the ferrimagnets produced. Such grains have the highest susceptibilities per unit mass (Maher, 1988) but do not retain any magnetic remanence. In practice, magnetic mineral assemblages in natural materials span a range of sizes and those that are predominantly SP, will include many that lie close to the upper size threshold. Such grains exhibit what is termed magnetic viscosity, which is to say, their magnetic properties are both time and frequency dependent. They acquire and lose magnetic remanence over time and this property can be quantified by measuring the decay of magnetic remanence once a sample has been removed from the magnetic field (Higgitt et al. 1991). Frequency dependence can be characterized by measuring magnetic susceptibility at different frequencies. Manipulating the magnetic experience of a sample in the laboratory thus provides a basis for detecting those ferrimagnetic assemblages dominated by grains at and below the SD/SP threshold. Four items of equipment form the ideal tool kit for detecting and characterizing fireinduced magnetic minerals: (i) a Bartingon MS2 dual frequency susceptibility meter that measures at 4.7kH and 0.47kH, (ii) a sensitive magnetometer for measuring remanent magnetization once a sample has been magnetized and removed from the field, and equipment for generating a range of (iii) DC and (iv) AF magnetic fields. One of the most important distinctions depends on growing and measuring Anhysteretic Remanent Magnetization (ARM). This involves placing the sample in a smoothly increasing and decreasing alternating field, usually peaking at 100mT, upon which is superimposed a weak DC biasing field usually between 0.4 and 1mT. ARM is acquired in the biasing field and is normally expressed as the susceptibility of ARM (χARM). This involves normalizing the ARM to the biasing field and it allows comparisons to be made between results obtained in laboratories using different DC fields. Used in conjunction, these items of equipment allow measurement of the following magnetic properties in addition to χARM: -

Low frequency magnetic susceptibility (χlf) measured at 0.47 kHz and high frequency magnetic susceptibility (χhf), at 4.7 kHz. The difference between the two measurements gives the frequency dependent susceptibility expressed either per unit mass (χfd) or as a percentage of χlf (χfd%).

122

Frank Oldfield -

Saturation Isothermal Remanent Magnetisation (SIRM) using a field of 1 T generated by a Pulse Magnetiser and reverse field demagnetization of the SIRM at fields of, for example, −20 mT, −40 mT, −100 mT and −300 mT.

Several prior publications give a fuller account of these magnetic properties and their significance (Walden et al. 1999). The properties that help to distinguish fire-induced magnetic minerals include the following: 1) High χlf values in conjunction with χfd% values generally >8%. 2) Low χARM/χlf, (< 5) and χARM/χfd (< 50) quotients in conjunction with high χφd% values. 3) Rapid reverse field demagnetization of SIRM, especially in the lower fields of 20mT and -40mT, relative to the samples with which the burnt material is being compared. The first of these properties confirms that the dominant grain size of the ferrimagnetic assemblage lies close to or below the SD/SP threshold and certainly not above the SD size range of ~ 25 – 100nm. This is an essential criterion for interpreting the other two. Provided this criterion is met, the low χARM/χlf, and χARM/χfd quotients point to a grain size assemblage dominantly in the SP rather then SD range (Oldfield, 1994; 2007). As is noted below, this is an essential distinction in lake sediment studies. The third criterion indicates the probability that the ferrimagnetic assemblage is rich in viscous grains. This can be confirmed by plotting the relatively steep and linear loss of IRM against the log. of time. Figure 2, based on Oldfield and Crowther (2007) illustrates the use of the quotients noted in (2) above to identify fire-enhanced soil. Figure 4, based on Gedye et al (2000) also shows the way in which the reverse field behaviour noted in (3) helps to identify horizons strongly influenced by catchment fires in a lake sediment record.

APPLICATIONS IN ENVIRONMENTAL RESEARCH So clear and strong are the indications of fire in the palaeoecological record in many parts of the world, especially from the charcoal record (Carcaillet et al. 2002) that there can be little doubt that it is a significant contributor to the magnetic signatures in the sedimentary record in many lakes. In the early days of research in environmental magnetism the role of fire received a good deal of attention (Oldfield, 1991;1992). Figure 1 shows the results of an early study designed to test the survival of fire generated magnetic minerals in lake sediments adjoining areas of dated forest fires (Rummery et al. 1979). Extensive fires in the Landes region of SW France culminated in 1949. The burnt area included parts of the catchments of two of the shallow coastal lagoons, the Etang de Biscarosse and the Etang de Sanguinet. Magnetic measurements on cores taken from each lake confirmed the survival of peaks in magnetic concentration linked to the main fire event. Subsequent to that early study, and only after the near ubiquity of biogenic magnetite in the form of the magnetosomes within the cells of magnetotactic bacteria was confirmed by elegant Transmission Electron Microscope (TEM) images (Petersen et al. 1986; Snowball,


123

1994) was the importance of fire in the sedimentary record set into a more realistic context (Oldfield 1994; 2007). Whereas, as noted above, the magnetite/maghemite grains formed by fire are predominantly superparamagnetic (SP) in size or very close to the boundary between SD and SP grains, ie. around or below 20 - 25nm in diameter, those formed by magnetotactic bacteria are predominantly of stable single domain (SD) size. The behaviour of ferrimagnetic grain assemlages on either side of this size transition thus provides the basis for distinguishing between assemblages of grains predominantly SD or SP in size (Oldfield,1994) and it is this distinction that Oldfield and Crowther (2007) used not only to differentiate fire-induced grains from magnetosomes, but also to identify a much more subtle difference in mean size between fire-induced and non-fire induced secondary, pedogenic ferrimagnetic minerals in soils (Figure 2). In view of all the above studies, it is perhaps timely to attempt some realistic assessment of the potential role of magnetic measurements in the reconstruction of past fire incidence and in others types of environmental research. Magnetic susceptibility surveys are routinely used in archaeological prospecting (see e.g. Clark, 2001) and sites of former burning give rise to some of the strongest signatures recorded. Burning, however, is not the only process leading to variations in magnetic susceptibility at archaeological sites as susceptibility variations may also record features such as ditch fills, post holes and former cultivation. An early successful attempt using magnetic measurements to detect the effects of in situ burning is illustrated in Figure 3.

Figure 1. Detecting magnetic signatures from historically documented fires (Rummery et al. 1979; Thompson and Oldfield, 1986). The location map on the left shows the position of two of the lakes in the Landes region of S W France in relation to extensive areas of pine forest burned in the 1940s. In the plots to the right, SIRM is used as an indicator of magnetic concentrations in sediment profiles from two of the lakes. In both cases, peak values correspond with, or lie immediately above peaks in charcoal frequency and in pine pollen breakage. In the case of the profile from the Etang de Biscarosse, dating by 137Cs confirms that the peak in magnetic concentrations is close in age to the date of the most severe fire, in1949.

Burnt timbers were found in the embankment of Maiden Castle, an Iron Age fortified site on the Mid-Cheshire Ridge. In order to establish whether the timbers were burnt before or

124

Frank Oldfield

after emplacement, magnetic measurements were carried out around the burnt timbers and comparisons made with unburnt material remote from the timbers. From these measurements it was clear that the wood was burnt in situ, possibly as part of a failed attempt at vitrification (Oldfield et al. 1985; Thompson and Oldfield, 1986). This study predated the development the full range of equipment outlined in the previous section. For the identification of sites where fire was used or occurred accidentally, the combination of approaches outlined by Crowther (2003) and Oldfield and Crowther (2007) is likely to provide secure and reliable evidence in most contexts, provided the temperature reached was sufficiently high (ideally at least 400°C) and the material affected was sufficiently rich in non-ferrimagnetic iron minerals prior to burning.

Figure 2. Bilogarithmic plot (see text) illustrating the identification of burnt samples on the basis of ferrimagnetic grain size. The burnt archaeological soil envelope spans the range of values for the most strongly burnt samples from a range of archaeological sites using the criteria described in Crowther (2003). The Cotswold soils were taken from old mature deciduous woodland. The samples falling within the unburnt envelope were then burned experimentally in the laboratory using the procedure described in Crowther (2003) and the subsequent measurements all plotted within the Cotswold woodland (burnt) envelope. The other envelopes of values are explained more fully in Oldfield (2007).


125

Figure 3. Magnetic measurements from the area surrounding charred timbers emplaced in the rampart of Maiden Castle, Cheshire (Oldfield et al. 1985; Thompson and Oldfield, 1986). The lower diagram plots the increased magnetic concentrations (χ and SIRM) towards the top of the section in the vicinity of the charred timber. The upper graph plots quotients that are independent of magnetic concentration and confirms clear qualitative differences between unburnt sands remote from the charred timbers and samples in close proximity. The envelope of values for samples close to the timbers overlaps that for samples of unburnt sand subsequently burnt under controlled conditions in the laboratory.

The detection procedure outlined in Crowther (2003) involves burning samples under controlled conditions to establish the extent to which any previous burning has already enhanced the susceptibility values. In cases where further enhancement by controlled burning in the laboratory is minimal, the likelihood of previous burning is greatest and vice versa. Where the burnt material was highly magnetic before any possibility of being affected by fire, for example in the case of basalts, experiments designed to reveal magnetic grain size differences may be useful. Where burnt soil was already strongly magnetically enhanced prior to any putative burning, discrimination will always be more difficult since the distinctions

126

Frank Oldfield

made in the Oldfield and Crowther (2007) study are subtle and can also be influenced by particle (as distinct from magnetic grain) size (Oldfield et al. 2009). The use of artificially enhanced gravel or sand in tracing experiments (Arkell et al. 1983; von Post et al. 1994) has met with only limited success and other approaches that are more sensitive, quantitative and deal more effectively with burial are more practical. Equally, attempts to use magnetic measurements as passive tracers of surface processes have mainly focused on contexts where recurrent fire has been an integral part of the ecosystem and distinguishing the impact of any given sequence of recent events, such as a major forest fire, is rendered complicated by the persistence of fire-enhanced magnetic minerals from previous events (Blake et al. 2006). Using magnetic measurements as part of a strategy to reconstruct fire history from the sedimentary record using the full range of measurements outlined above is best illustrated by Gedye et al. (2000; Figure 4) in which magnetic measurements are set alongside evidence from both pollen and charcoal analysis. In this instance, there is both remarkable consistency between the magnetic evidence for fire incidence, and indications in either the pollen or charcoal record, or both. There is also the intriguing possibility that the magnetic record may add addditional insights since the minerals probably reflect within catchment surface processes and hence the geomorphic response to fire. In this case study, it is possible to identify the magnetic signatures linked to fire incidence against the changing magnetic properties that reflect variations in magnetosome dominance and a long term shift in erosional regime. This is certainly not always the case. Not only will unrelated variations in allochthonous magnetic input to the sediments complicate detection, magnetic measurements may also reflect selective dissolution of ferrimagnetic minerals through sub-oxic diagenesis as well as authigenic growth of the strongly magnetic iron sulphide, greigite (Fe3S4).

Figure 4. Selected magnetic properties, charcoal influx and pollen and spore relative frequencies from the late Holocene part of a core from the Lago di Origlio, S. Switzerland (Gedye et al. 2000). The dotted lines (A to G) mark depths at which the charcoal record points to fire occurrence. Correspondence with peaks in χlf, χfd% and χfd/χARM indicate the likely presence of fine grained, fire derived magnetic minerals from the catchment. At these depths, the consistently high χARM/SIRM quotient indicative of dominance by bacterial magnetosomes is temporarily reduced in response to the finer mean grain size of the catchment derived ferrimagnetic minerals (see text). Horizon A shows a strong magnetic signature, but only a small peak in charcoal influx. There is, however, a strong response from heather (Calluna) pollen and fern spores, confirming that the fire event had a major impact on catchment vegetation.


127

At the very least, magnetic measurements should be used alongside charcoal and pollen counts before there is any possibility that the evidence they yield might enhance interpretations.

CONCLUSIONS Fire often gives rise to magnetic mineral assemblages that can be distinguished from those generated by other processes. The distinctive characteristics of fire induced ferrimagnets are a function of their fine magnetic grain size. Under favourable circumstances, the resulting characteristic ‘signature’ can be detected using a combination of dual frequency magnetic susceptibility, isothermal and anhysteretic remanence measurements. The magnetic approach to fire detection is most successful in archaeological studies where a combination of experimental burning and careful measurement has confirmed its applicability. Detection of fire signatures in sedimentary records is more problematical, but where it is possible, it may enhance the insights derived from other techniques. The use of artificially enhanced magnetic materials in tracing experiments has met with only limited success.

REFERENCES Arkell, B., Leeks, G. Newson, M. and Oldfield, F. (1983). Trapping and tracing some recent observations of supply and transport of coarsesediments from upland Wales. Special Publications of the International Association of Sedimentologists 6, 107–19. Blake, W.H., Wallbrink, P., Doerr, S.H., Shakesby, R.A. and Humphreys, G.S. (2006). Magnetic enhancement in fire-affected soil and its potential for sediment-source ascription. Earth Surface Processes and Landforms 31, 249-269. Carcaillet C., Almquist H., Asnong H., Bradshaw R.H.W., Carrión J.S., Gaillard M.-J., Gajewski K., Haas J.N., Haberle S.G., Hadorn P., Müller S.D., Richard P.J.H., Richoz I., Rösch M., Sánchez Goñi M.F., von Stedingk H., Stevenson A.C., Talon B., Tardy C., Tinner W., Tryterud E., Wick L., Willis K.J. (2002) Holocene biomass burning and global dynamics of the carbon cycle. Chemosphere 49, 845 – 863. Clark, A. (2001). Seeing Beneath the Soil. Routledge, New York. Crowther, J. (2003). Potential magnetic susceptibility and fractional conversion studies of archaeological soils and sediments. Archaeometry 45, 685-701. Gedye, S.J., Jones, R.T., Tinner, W., Ammann, B. and Oldfield, F.2000: The use of mineral magnetism in the reconstruction of fire history: a case study from Lago di Origlio, Swiss Alps. Palaeogeography Palaeoclimatology Palaeoecology 164, 101–10. Higgitt, S.R., Oldfield, F. and Appleby, P.G. (1991). The record of land use change and soil erosion in the late Holocene sediments of the Petit Lac d'Annecy, eastern France The Holocene 1, 14-28. Le Borgne, E. (1955). Abnormal magnetic susceptibility of the topsoil. Annales de Geophysique 11, 399-419, Le Borgne , E.(1960). The influence of fire on the magnetic properties of soil, schist and granite. Annales de Geophysique 16, 159-196.

128

Frank Oldfield

Longworth, G., Becker, L.W., Thompson, R., Oldfield, F., Dearing, J.A. and Rummery, T.A. 1979: Mossbauer-effect and magnetic studies of secondary iron-oxides in soils. Journal of Soil Science 30, 93–110. Maher, B., (1988). Magnetic properties of some sub-micron synthetic magnetites. Geophysical Journal 94, 83–96. Oldfield, F. (1991). Sediment magnetism – soil-erosion, bushfires, or bacteria. Geology 19, 1155–56. Oldfield, F. (1992). The source of fine-grained magnetite in sediments. The Holocene 2, 180– 82. Oldfield, F. (1994). Toward the discrimination of fine-grained ferrimagnets by magnetic measurements in lake and near-shore marine sediments. Journal of Geophysical Research-Solid Earth 99, 9045–50. Oldfield, F. (2007). Sources of fine-grained magnetic minerals in sediments: a problem revisited. The Holocene 17, 1265-1271. Oldfield, F. and Crowther, J. (2007). Establishing fire incidence in temperate soils using magnetic measurements. Palaeogeography, Palaeoclimatology, Palaeoecology 249, 362– 369. Oldfield, F., Thompson, R. and Dickson, D.P.E. (1981). Artificial magnetic enhancement of stream bedload – a hydrological application of superparamagnetism. Physics of the Earth and Planetary Interiors 26, 107–124. Oldfield, F., Krawiecki, A., Maher, B.A., Taylor, J.J. and Twigger, S. (1985). The role of mineral magnetic measurements in archaeology. Paleoenvironmental Investigations: Research Designs, Methods, and Data Analysis, 29–44. Oldfield, F. Hao, Q. Bloemendal, J. Gibbs-Eggar, Z.Patil, S. and Guo, Z. (2009). Links between particle size and magnetic grain size: general observations and some implications for Chinese loess studies. Sedimentology doi: 10.1111/j. 13653091.2009.01071.x.1365-3091.2 Petersen, N, von Dobeneck, T. and Vali, H. (1986). Fossil bacterial magnetite in deep-sea sediments from the south Atlantic Ocean. Nature 320, 611-615. Rummery, T.A. (1981). The effects of fire on soil and sediments. Unpublished PhD Thesis. University of Liverpool. Rummery, T.A., Bloemendal, J., Dearing, J., Oldfield, F. and Thompson, R. (1979). Persistence of fire-induced magnetic oxides in soils and lake sediments. Annales de Geophysique 35, 103–107. Snowball, I.F. (1994). Bacterial magnetite and the magnetic properties of sediments in a Swedish lake. Earth and Planetary Science Letters, 126, 129-142. Thompson, R. and Oldfield, F. 1986: Environmental magnetism.Allen and Unwin, 227 pp. van der Post, K., Oldfield, F. and Voulgaris, G. (1994) Magnetic tracing of beach sand. Coastal Dynamics Proceedings, Barcelona, pp. 323-333. Walden, J., Oldfield, F. and Smith, J. editors (1999). Environmental magnetism: a practical guide. QRA technical guide No. 6. 243pp.



Chapter 6

FOREST AND FIRE RISK DYNAMICS IN THE GREAT XING’AN MOUNTAINS, NORTHEASTERN CHINA: A SPATIAL SIMULATION STUDY Zhihua Liu*1,3, Hong S. He1,2, Yu Chang1 and Jian Yang1 1

Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang, China; School of Natural Resources, University of Missouri, , Columbia, MO, USA; 3 Graduate School of Chinese Academy of Sciences, Beijing, China

2

ABSTRACT Natural disturbance-based forest management, based primarily on the understanding of natural disturbance regimes and forest dynamics, provide sustainable forest management paradigms to maintain biodiversity and essential ecological function in managed forested regions. So understanding how forest ecosystem and fire dynamics respond to historic and current fire regime pose great significance in designing scientifically sound management plans for Great Xing’an Mountains in the Northeast China. We used a spatially explicit landscape dynamics model, LANDIS, to simulate the long-term forest response and fire dynamics under historic fire regime (before 1950s) and the fire suppression (after 1950s). Specifically, we compare how the fuel loads and fire hazards, and forest tree species abundance response under the two scenarios. Under the fire suppression scenario, fire risk will quickly increased to a dangerous level, about 80% of the landscape will carrying a high level of fire risk at the end of the simulation; both fine fuel and coarse fuel will rise to medium-high level after a few decades’ suppression. Generally, fires tend to be more catastrophic and less frequent. Fire suppression results in less frequent, but more intense fires. Fire suppression can also decrease the proportion of coniferous forests, increase the proportion of deciduous forests and alter forest age structures. The results suggest that extensive additional forest management activities, such as prescribe burned, fuel load reduction, uneven-aged harvesting should be implemented to maintain low level of fire risk and forest type diversity. Further studies are needed to evaluate the effects of prescribed burning and fuel

* Corresponding author, Email: [email protected]

130

Zhihua Liu, Hong S. He, Yu Chang, et al load reduction under the fire suppression, and to find the fine balance among fire suppression, maximal timber production, and sound ecological function.

Keywords: Fire suppression; Historic fire regime; LANDIS; spatially explicit landscape simulation model; forest landscape; Northeastern China

1 INTRODUCTION Forest fire is one of the key natural disturbances in shaping the dynamics of forest ecosystems, affecting species composition (Noble and Gitay 1996; Gardner et al. 1999) and age structure (Heinselman 1973; Van Wagner 1978). However, many factors, especially fire suppression have dramatically changed natural fire regimes worldwide (Baker 1992; Barrett 1994; Finney 1999; Chang 2007), such as by lengthening mean fire return intervals (Guyette and Larsen, 2000; Lesieur et al. 2002; Shang et al.2004), increasing fuel loads and fire risk (Bury 2004; He et al 2004), and leading to major changes in vegetation succession trajectories and forest landscape dynamics (Ryan 2002, Chang et al 2007). All of these effects have the potential to cause negative ecological consequences, which were not well understood so far. Thus, understanding the effects of long-term fire suppression on fire regimes and forest ecosystems has become increasingly important in designing scientifically sound forest management plans (Dombeck et al., 2004). The boreal forest of Great Xing’an Mountains in the northeastern china provides the most timber of any forest areas in China; simultaneously it encompasses rather unique ecological and environmental system in the region (Zhou, 1991; Xu, 1998). In Great Xing’an Mountains, effective fire suppression has been carried out since early 1950s，and will probably continue in the future. After about more than 50 years of fire suppression and high intensity harvesting, natural fires have been largely suppressed and fire regimes have been significantly changed in this area (Xu 1998), and the catastrophic fires are more frequently than historically (Chang et al 2007). The fire return interval has extended from 120 to 150 years before the 1950s (Xu 1998) to about 500 years presently (Chang et al 2007). The success of fire suppression, coupled with a warmer, drier climate due to global warming (Xu, 1998), has led to flammable fuel buildup and resulted in fires of greater intensities and extents than those that occurred historically in the region. In 1987, a catastrophic fire occurred in the Great Xing’an Mountains of Northeastern China and burned a total area of 1.3×106 ha. Among the other man-made factors, fire suppression contributed most to this disastrous fire (Wang et al. 2007). After the 1987 catastrophic fire，proposal of reintroduce the historic fire into forest management has been put forward by many forest managers and policy makers. While the long-term effects of fire suppression on forest landscapes and fire regime has not been well understood and how the forest landscape response to historic fire regime and fire suppression were rarely study in this area. Over the past decade, there has been an increasing interest in forest management approaches based on natural disturbance dynamics on the landscape scales (Bergeron and Harvey 1997; DeLong 2007; Bergeron and Drapeau. 2007). The basic conception of natural disturbance-based management favoring the development of stand and landscape

Forest and Fire Risk Dynamics …

131

compositions and structures similar to those of natural origin should maintain essential ecological functions. But first and foremost, natural disturbance regimes must be well understood (Attiwill, 1994). Forest fire is one of the most important natural disturbance agents in shaping the dynamics of forest ecosystems of the Great Xing’an Mountains in the Northeast China (Zhou, 1991; Xu, 1998). So, fully understanding of forest response and fire dynamics to current fire suppression can provide valuable information needed for determining ecologically sound forest management planning. Therefore, our goal is to examine the forest response and fire regime resulting from the effective fire suppression. To achieve this goal, we used a spatially explicit forest landscape model, LANDIS (He and Mladenoff 1999a), to explore forest landscape responses to the constructed simulation scenarios in the Great Xing’an Mountains. As long-term empirical studies of fire effects over large landscapes are virtually impossible to conduct, and ecological models have proved to be useful tools for studying fire disturbance at landscape scales (e.g., Baker, 1992; He and Mladenoff, 1999a; Hargrove et al., 2000). We construct two simulation scenarios: results from re-introduction of the historic fire regime before 1950s (NFS) and effective fire suppression regime after 1950s (FS). We construct these two simulation scenarios for two main reasons, 1) comparison between these two scenarios (NFS and FS) could help us understand the potential problems caused by long-term effective fire suppression; 2) we can have a well understanding of historic forest ecosystem and natural fire regime in the Great Xing’an Mountains. More specifically, we modelled the long-term potential effects of two different fire simulation scenarios on tree species abundance and age structure, fuel loads, and fire risk. For both scenarios (NFS and FS), we compared predicted fuel loads, vegetation changes, fire frequency, area burned and fire risk under these two fire regimes.

2. METHOD 2.1. Study Area Our study area, Huzhong Forest Bureau (Fig. 1), encompassing approximately 937 244 ha on the Great Xing’an Mountain, is in the north-western area of Heillongjiang Province in north-eastern China (52◦25’00”N 122◦39’30”E to 51◦14’40”N 124◦21’00”E). The area falls within the cool temperate zone (Zhou et al. 1991) affected by the Siberian cold air mass. It possesses a terrestrial monsoon climate with a long and severe winter. Annual average precipitation is ~500 mm, more than 60% of which occurs between June and August. The annual average temperature is 4.7℃ with an average of −28.9◦C for February, the coldest month in the year. The average temperature for July, the hottest month in the year, is 17.1℃, with a highest recorded temperature of 35.3℃. The vegetation of this area belongs to cool temperate coniferous forests, which are the southern extension of eastern Siberian boreal forests (Zhou et al. 1991). The forest area accounts for 86.98% of the study area. The canopy species composition is relatively simple, including larch (Larix gmelini), pine-s (Pinus sylvestris var. mongolica), spruce (Picea koraiensis), birch (Betula platyphylla), two species of aspen (Populus davidiana, Populus suaveolens), willow (Chosenia arbutifolia), and an important shrub species pine-p (Pinus

132

Zhihua Liu, Hong S. He, Yu Chang, et al

pumila). With the exception of some portions of wetland near the river, larch is widely distributed and accounts for 65% of the study area. Birch and pine are mixed with larch in most areas owing to fire disturbance and forest harvesting, with pine-s having a small area of distribution (1.8%). Aspen and willow are confined to terraces along the rivers where water is plentiful. Spruce, being highly shade tolerant, occurs mostly in valleys and high elevation areas, pine-p occurs mostly in >800m elevation (Xu 1998). Analyses of the dynamics of every species in this area are unfeasible. Therefore, in the present study, we analyzed four representative species: larch, pine, spruce, and birch. We selected these species for their characteristics: larch is a late successional, climax species; birch is an early successional, pioneer species; pine-s is common species with high economic value, and pine-p has unique ecological and economic values (zhao et al 1997)

Figure 1. The geographic location of the study area and different land types, among which water and non-forest land types are not simulated in the model.

2.2. Description of LANDIS LANDIS is a spatially explicit forest landscape model in response to natural and anthropogenic disturbances, succession and management, and has been described extensively elsewhere (He et al., 1999; He and Mladenoff, 1999a; Mladenoff and He, 1999; Gustafson et al., 2000, Mladenoff 2004) and therefore, only a general description of the model is provided here. LANDIS is a raster-based model that simulates ecosystem dynamics including forest succession, seed dispersal, species establishment, fuel accumulation and decomposition, fire and windthrow disturbance, timber harvest, and fuel treatment with a 10-year time step. LANDIS is designed to simulate ecological dynamics and forest management at large extents (103 to 106 ha) over long time spans (101 to 103 years). LANDIS stratifies a heterogeneous landscape into relatively homogeneous land types, which are generated from GIS layers of climate, soil, or terrain attributes (slope, aspect, and


133

landscape position). A single land type contains a somewhat uniform suite of ecological conditions, resulting in similar species establishment patterns and fire disturbance characteristics, including ignition frequency, fire cycle (mean fire return interval), and fuel decomposition rates (He and Mladenoff, 1999). LANDIS simulates fire levels of fire intensity from surface fires to crown fires. Fire intensity is determined by the amount of fuel on a site. LANDIS fuel module (Shang and He, 2003; He et al., 2004) are designed to track fuel dynamics, to estimate potential fire risk, and to evaluate the effects of various fuel treatments. In LANDIS, Fuel are simulated for three distinct types: fine fuel, coarse fuel and live fuel. Fine fuel typically corresponds to 1 and 10-h lag (Brown and Davis, 1973) and is the primary determinant of fire ignitions (Andrews and Chase, 1989). Coarse fuels, also called coarse woody debris (CWD) correspond to 100- and 1000-h fuels and are primarily responsible for determining fire intensities (Burgan, 1987). Live fuels, also called canopy fuels, are live trees that may provided vertical continuity between strata and allow fire to be transmitted from the surface into the crowns. Fuel loads as well as wildfire intensity are modeled as five categorical classes from very low (class 1) to very high (class 5). Fuel load can be modified by Land type, fire, wind, harvest, and biological disturbance.

2.3. Parameterization of LANDIS Parameterization of LANDIS 4.0 for the Huzhong Forest Bureau involved several aspects: species’ vital attributes, a forest composition map that contains species presence/absence and age classes information at each cell, a land type map, species establishment probabilities for each land type and fire disturbance regimes for each land type. The available materials for parameterization include: two Landsat TM scenes taken in 1990, the fire records from 1990 to 2000 to derive fire suppression regime, a digital elevation model (DEM), and a forest stand map and a stand attribute database compiled from the forest inventory taken in 1990 in the Huzhong area. The stand attribute database provides the relative percentage occurrence of each canopy species, the average age of dominant canopy species, timber volume, and crown density, among other factors

2.3.1. Species Attributes and Forest Composition Map A total of eight tree species were incorporated into LANDIS, and species’ vital attributes (Table 1) were estimated based on existing studies (Ai et al., 1985; Hu et al., 1991; Xu, 1998; He et al., 2002a; Xu et al. 2004) in the region and consultation with local experts. Forest composition map was derived from a forest stand map in 1990, the following stand attribute database and two scenes of Landsat TM imagery taken in 1990. The stand attribute database provides information including the relative percentage of canopy species, and the average age of dominant canopy species, timber production, and crown density. It also contains subdominant and accompanying tree species, but with no age information available. To reduce computational loads during model simulations, the forest composition map was resampled at 90m×90m resolution, which yielded 1480 rows×1274 columns. Each cell contains the presence/absence and age cohorts of all eight tree species.

134


Table 1. Species’ vital attributes derived for Huzhong Forestry Administrative Bureau ED, effective seeding distance (m); FT, fire tolerance class (1 to 5, with 1 for the least tolerant and 5 for the most tolerant); LONG, longevity (years); MD, maximum seeding distance (m); MTR, age of maturity (years); MVP, minimum age of vegetative reproduction (years); ST, shade tolerance class (1 to 5, with 1 for the most shade intolerant and 5 for the most shade tolerant); VP, vegetative reproduction probability Species name LONGMTR ST Larch (Larix gmelini) 300 20 3 Pine-s (Pinussylvestris var. 210 40 1 mongolica) 300 30 4 Spruce (Picea koraiensis) 150 15 1 birch (Betula platyphylla) 180 30 1 Aspen-d (Populus davidiana) Aspen-s (Populus suaveolens) 150 25 1 250 30 2 willow (Chosenia arbutifolia) pine-p (Pinus pumila) 250 30 3

FT 4

ED 100

MD 200

VP 0

MVP 0

2

50

200

0

0

2 3 3 4 2 1

50 200 -1 -1 -1 50

150 2000 -1 -1 -1 100

0 0.8 1 1 0.9 0

0 40 40 40 30 0

We assumed all present tree species randomly distributed in a stand, but in different proportion based on forest inventory data. For the dominant species, the algorithm assumes that the species will occur on every cell in a stand. For the non-dominant species, each cell in a stand has a probability of being assigned with that species, which is determined by the relative occurrence of the species within the stand (0 ~ 1). For example, if a stand has a relative occurrence P1 for non-dominant species 1, then the (P1/P)*100 (P stand for occurrence of the dominant species) percent of the cell in a stand will be assigned stochastically with that species. Similar algorithms will be used to assign other species if there were other species present. If there are no species present in the stand, then all the cells in the stand will be blank. There will be at least one species on each cell if the stand isn’t blank. Species information map created by this method should be reasonable because the previous study indicated that all species in a stand will be occurred in a 30m×30m forest inventory sampling(Zhou, 1991; Xu, 1998), if extrapolated to 90m×90m resolution to be simulated in the present study, the result should be acceptable The species assigned to each cell is also assigned to an age cohort. If it is a dominant species on the cell in a stand, it will be directly assigned with the age cohort of the dominant species recorded in the stand attribute table. Because there was no age information available for non-dominant species in the stand, we assigned non-dominant species to age cohorts determined by the area weighted average age of this species in the corresponding compartment (calculated from the stands where the species of interest is the dominant species). If there are no stands in the corresponding compartment where the species of interest is the dominant species, the non-dominant species is assigned to age cohort determined by the area weighted average age of this species in the whole study area. This was based on the assumption that the area weighted average age of species at the compartment level or the landscape level (whole study area) reflects the age of this species at the stand level. Age information for Pinus pumila was missing in the forest inventory, so we assigned P. pumila an age of 100 years based on tree ring investigations conducted by us in July 2001 and July 2002.


135

2.3.2. Ecological Land Types Map Land type stratifies the heterogeneous landscape into relatively homogeneous units (land type units). Within each land type, similar environments for species establishment are assumed (Mladenoff and He, 1999). In this study, we derived six land types based primarily on terrain attributes: southern slope, northern slope, ridge top, terrace, residential land (nonforest land), water body (Fig.2). All land types were interpreted from the 2001 Enhanced Thematic Mapper imagery and DEM. Non-active land types (not simulated in LANDIS) (water and residential areas) account for 0.76% of the total area, whereas terrace, southern slope, northern slope, and ridge top account for 4.78, 37.25, 42.53, and 14.68% of the study area, respectively. Terrace is a linear feature and in most cases has a width of 2500m along the Huma River.

Figure 2. Simulated landscape coverage area of each species by age class under the two scenarios. Each 0.81 ha pixel could contain at least one species. (A) larch under historic fire regime scenario; (B) larch under fire suppression scenario; (C) birch under historic fire regime scenario; (D) birch under fire suppression scenario; (E) pine-p historic fire regime scenario; (F) pine-p fire suppression scenario; (G) pine-s under historic fire regime scenario; (H) pine-s under fire suppression scenario.

136


The establishment coefficients for all species for each landtype are listed in Table 2. These were determined based on the relative suitability of each species to establish itself within each land type, which was empirically derived from available literature (Li et al. 1987; Zhao et al. 1997; Xu 1998; Liu et al. 1999; He et al.2000a; Hu et al.2004) Table 2. species establishment coefficients for each land type in the Huzhong area. These coefficients were empirically derived from available literature (Li et al. 1987; Zhao et al. 1997; Xu 1998; Liu et al. 1999). The residential land and water body were not used for effective area during our simulations and the parameters were set to 0. landtype

Species establishment coefficients Larch Pine-s Spruce birch Aspen-d Aspen-s southern slope 0.4 0.2 0.03 0.3 0.2 0 northern slope 0.4 0.1 0.05 0.2 0.2 0 ridge top 0.3 0.08 0 0.05 0 0 terrace 0.01 0 0 0.05 0.05 0.07 residential land 0 0 0 0 0 0 water body 0 0 00 0 0 0

willow 0 0 0 0.2 0 0

pine-p 0 0 0.1 0 0 0

2.4. Fire Regime Forest harvesting ceased in 1999 when project of Natural Forest Preservation was implemented, and windthrow occurs rarely in this area; only fire disturbance was simulated in the present study. The fire suppression regime for our simulations was parameterized based on a database of 10-year fire dating from 1990 to 2000 (Hu et al., 2004). Current fire return intervals were estimated by calculating the reciprocal of the annual proportion of forest land burned within each Landtype in the Bureau. Only those fires greater than the resolution of the simulation recorded in the database (0.81 ha) were used to parameterize the fire regime (Table.3); While under historic fire regime, the MRI for these land types are roughly between 120 and 150 years (Xu 1998),

2.5. Simulation Scenarios We selected two scenarios: the first scenario approximated a historic fire regime before effective fire suppression (NFS). The second scenario approximated an effective fire suppression regime (FS) since 1950s. We began the simulation with realistically parameterized forest composition and landtype maps with species/age that represent the initial status of the landscape in 1990. From this starting point, the entire study area was simulated for 300 years. Each scenario was replicated ten times, with different random seed numbers. we used the same initial species composition and age cohorts for both simulation scenarios for two reasons, one is we can derive the historic species composition, age structure and fire dynamics of Great Xing’an forests landscape under the historic fire regime, the other is that we could readily compare the potential effects for these two scenarios over time. To ensure both scenarios are correctly simulated, we used historic fire data (Xu 1998) and the fire


137

records from 1990 to 2000 to calibrate the model iteratively until both scenarios matched the input data. To improve the clarity of figures and tables and to aid interpretation of the results, age class data were summarized into the following forest age classes: sapling (0–40 years), midage (41–80 years), near-mature (81–100 years), mature (101–140 years), and old-growth (>140 years) for conifers, and sapling (0–30 years), mid-age (31–50 years), near-mature (51– 60 years), mature (61–80 years), and old-growth (>80 years) for broadleaf trees (Xu 1998; Shifley et al. 2000). Results from the simulations were summarized as percentage cover of the study area. We used SPSS (version 13.0; SPSS, Chicago, IL, USA) to evaluate the simulated results. Specifically for MRI (mean fire return interval), a Wilcoxon signed ranks test was used to examined whether a significant difference exists between the derived and the simulated results, for MFS( mean fire size) and FPD (fire frequency per decade), a one sample t-test was used to evaluate whether a significant difference exists between the derived and the simulated results. Table.3. Statistical test for mean fire return intervals (MRI) to evaluate whether there is significant difference between inferred and simulated MRI for the fire suppression (FE) and historic fire regme (NFE) scenarios at 95% confidence landtype

MRI under FS scenario MRI under NFS scenario inferred simulated inferred simulated Terrace 1500 1631 500 518 South slope 600A 573 160 B 158 A North slope 500 498 150 B 147 377 140 B 166 Ridge top 400A Wilcoxon signed ranks test P**=0.715 P**=0.730 Values were calculated based on real fire records, with locations and burned area, from 1990 to 2000. Values were derived based on published data (Xu 1998). **stand for 95% confidence A B

3. RESULTS 3.1. Verification of the Simulated Fire Regimes The results showed that there was no significant difference (95%) between the MRI derived from the fire records and the MRI simulated on various land types in both the FE (P =0.715) and NFE (P =0.730) scenarios (Table 3). There was no significant difference (95%) between the MFS under FE (P =0.26) and NFE (P =0.37) scenarios. There was also no significant difference (95%) between the FPD under FE (P =0.57) and NFE (P =0.43) scenarios. Therefore, the model correctly simulated key fire characteristics (e.g. MFS, fire frequency, MRI) of the two fire regimes. This ensured the validity of subsequent analysis involving fire and the interaction between fire and vegetation.

138


3.2. Tree Species Composition and Age Structure Simulation results show distinct differences in species coverage under the two scenarios (Fig.2), Percent coverage of larch was slight lower under the fire suppression regime scenario (Fig. 2B) than historic fire regime scenario (Fig. 2A) before simulation year 160, but become the almost same percent coverage after that, before year 120, larch increased in the number of sites they occupied under both scenario, but more rapidly under the historic fire regimes scenario (Fig. 2A and B). At the end of the simulation, the percent coverage of the larch is slightly higher than that at the beginning. Percent coverage of birch decreased rapidly before year 130 and increased rapidly after that under both scenarios (Fig. 2C and D), percent coverage of birch is higher under fire suppression scenario at the end of the simulation. Tree in both the pine-s(Fig. 2G and H)and pine-p (Fig. 2E and F) decreased in the number of sites they occupied under both scenarios, but more rapidly under the fire suppression regime, at the end of the simulation, Percent coverage of pine-s and pine-p become slightly higher than zero under both scenarios. Tree size class structure also differed between the two scenarios. For example, though all age classes of larch are slightly higher under the historic fire regime scenario, revealing the same trend of the percent coverage of the species, but in different proportion in different age class, generally between 0.2%for mid-age to 1% for old-growth. For birch, age class for nearmature and old growth are much higher under fire suppression scenario, reaching to 8.3% at simulation year 110, smuch higher proportion also existed in other age classes under fire suppression scenario. All size classes of pine-sand pine-p decreased in percent coverage under two scenarios, but more rapidly under the fire suppression regime.

3.3. Fuel Load and Fire Risk Results showed differences in fine fuel loading between the two simulation scenarios (Fig. 3).Frequent fires under the historic fire regime scenario reduced the fine fuel loads to a relative low level along the 300-year simulation period. At the year 100, the percent of medium fine fuel load is about 10% lower than that under fire suppression regime. By the end of the 300-year simulation, about 30% of the study area had a low level of fine fuel accumulation, 66% had medium fine fuel loads, and another 4% had extremely high fine fuel loads. Under the fire suppression scenario, fine fuels were at relative higher levels. At simulation year 300, on about 20% of landscape the fine fuel loads were low or medium-low, more than 60% of landscape had medium fine fuel loads, and about 10% of landscape had extremely high fine fuel loads.


139

Historic Fire Regime

Year 0

Year 50

Year 100

Year 150

Year 200

Year 250

Year 200

Year 250

Year 300

Fire Suppression Regime

Year 0

Water Nonforest Medium

Year 50

Year 100

Low Medium-low Medium-high

Year 150

Year 300

High

Figure 4.Comparison of fine fuel loads between the two scenarios. Fine fuel loads were classified into five categories: low (class1), medium-low (class2), medium (class3), medium-high (class4) and high (class5).

The coarse fuel loads differed distinctly between these two scenarios in some respects (Fig. 4). Under the historic fire regime, coarse fuel load were at lower level during the whole simulation period. About 30–40% of the landscape carries a medium-low or low coarse fuel load during the most of the simulation period except for simulation year 200, only 15% of the landscape carrying medium-low coarse fuel load in simulation year 200. The coarse fuel load accumulated gradually from 20% of the whole landscape to 50% at the ending of the simulation. Only < 10% of the landscape carries a high coarse fuel load at year 300. However, under the fire suppression scenario, Only 0 δϕ(y) = −ϕ(y) ∀y ∈ Λr (xi ) cal mes ⎪⎨ M x M x − ≥ ( )) ( ) 0 ( r i r i ) ⎪⎪if ϕ(y) = 0 δϕ(y) = 0 ∀y ∈ Λr (xi ) ⎪⎩

(45)

Relations (45) mean that if a point (a cell for the numerical calculation) is ignited it must be switched off and remained switched off if it was.

(M rcal (xi )) − M rmes (xi )) < 0

⎧if ϕ(y) > 0 δϕ(y) = 0 ∀y ∈ Λr (xi ) ⎪ ⎪ ⎨ ⎪ if ϕ(y) = 0 δϕ(y) = ϕ(y) ∀y ∈ Λr (xi ) ⎪ ⎪ ⎩

(46)

Relations (46) mean that if a point is ignited it remains so, and if it was switched off it is switched on. For the second type of variations, we would have to determine the fire front as a curve in place of fire zone. A possible way of doing that is to introduce the characteristic function of the burning zone:

⎧⎪0 if y ∉ S f ( ) y χ = ⎪⎨ ⎪⎪1 if y ∈ S f ⎩

(47)

Let us remind that the gradient of the characteristic function is proportional to the unit normal vector to the boundary: ∇χ = −nδ∂S f (48) Then relation (40) can be written: Ns

δ2J = −2∑ (M rcal (xi , S f ) − M rmes (xi )) ∫ ϕ(y)G (xi − y)τ ⋅ ∇χds y i =1

Sf

(49)

If the transformation (33) is equal to the identity, the global transformation is only geometrical, i.e. is a virtual displacement. In this case the following relation holds:

Detection of The Positions and Computing the Rate of Spread of Fire…

∂χ + ∇χ ⋅ τ = 0 ∂ε

169

(50)

Then relation (49) leads: Ns

δ2J = 2∑ (M rcal (xi , S f ) − M rmes (xi )) ∫ ϕ(y)G (xi − y) i =1

Ωf

∂χ ds y ∂ε

(51)

Therefore the minimizing algorithm is very similar to the preceding one setting:

ϕ(y)

∂χ = δϕ(y) ∂ε

(52)

This short mathematical presentation of the extension of the identification from a line fire front to any shape of the fire shows that the method is powerful. We post pone to a subsequent work the exposition of the numerical implementation of the method.

7. CONCLUSION Some attempts are made at reconstructing a fire front by image processing video recorded fire. This optical technique, giving essentially the geometrical variables of flame, is difficult because the forest fires are often accompanied by thick smoke and the videos become really noisy in these situations. The work presented here is an attempt to propose an alternative to image processing in vegetation fire metrology at different scales (laboratory, fire tunnel and field). From the measurement of the heat flux received by a specific wireless thermal sensor in four directions of space we are able to determine some characteristics of the flame, the positions and the rate of spread of a line fire front in laboratory experiments. The aim of this chapter is not to discuss the validity of the flame model in propagation models of forest fires so we have considered a model of isothermal flame derived from the Radiative Transfer Equation (R.T.E.). This model, using the inverse method, provides realistic parameters for fire front positions, flame height, thicknesses of the burning zone, extinction or absorption coefficient and temperature of flames. Concerning the fire front positions and the rate of spread, this new physical method gives comparable results with those obtained by an image processing method described in [30] and adapted from the one used by Pastor et al. [4]. The values of the different characteristics of flame (thicknesses, height, temperature and extinction coefficient) can be compared favourably to experiment and literature. From these results, the device used here is demonstrated to be a robust instrument for the reconstruction of the line fire front at laboratory scale. However, the present model of flame seems to be a good candidate to modelling the radiative transfer in the forest fires propagation models. Let us notice that, as the radiative heat flux is one of the leading processes involved in propagation, it is important for propagation models relying on energy balance to have the best flame model possible. The system presented here could help validate a flame model.

170

K. Chetehouna, O. Sero-Guillaume and D. Bernardin

REFERENCES [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8] [9]

[10]

[11]

[12]

[13]

[14]

Iverson, L.R., Yaussy, D.A., Rebbeck, J., Hutchinson, T.F., Long, R.P. and Prasad, A.M., A comparison of thermocouples and temperature paints to monitor spatial and temporal characteristics of landscape-scale prescribed fires, International Journal of Wildland Fire 13(3), pp. 311-322, 2004. Zhou, X., Weise, D. and Mahalingamx, S., Eperimental measurements and numerical modeling of marginal burning in live chaparral fuel beds, Proceedings of the Combustion Institute 30, pp. 2287-2294, 2005. Santoni, P.A., Simeoni, A., Rossi, J.L., Bosseur, F., Morandini, F., Silvani, X., Balbi, J.H., Cancellieri and D., Rossi, L., Instrumentation of wildland fire: Characterisation of a fire spreading through a Mediterranean shrub, Fire Safety Journal 41, pp. 171-184, 2006. Pastor, E., Àgueda, A., Andrade-Cetto, J., Muñoz, M., Pérez, Y. and Planas, E., Computing the rate of spread of linear flame fronts by thermal image processing. Fire Safety Journal, 41, pp. 569-579, 2006. Dupuy, J.L., Vachet, P. and Maréchal, J., Fuel bed temperature measurements in laboratory fires spreading over a slope. V International Conference on Forest Fire Research, D.X. Viegas (Ed), 2006. Mendes-Lopes, J.M.C. and Ventura, J.M.P., Flame characteristics in fires propagating in beds of pinus halepensis needles. V International Conference on Forest Fires Research, D.X. Viegas (Ed), 2006. Martinez-de Dios, J.R., Arrue, B.C., Ollero, A., L. Merino and Gomez-Rodrıguez F., Computer vision techniques for forest fire perception. Image and Vision Computing, 26, pp. 550–562, 2008. Hartley, R. and Zisserman, A., Multiple view geometry in computer vision, 2nd ed. Cambridge: Cambridge University Press, 2003. Chetehouna, K., Séro-Guillaume, O., Sochet, I. and Degiovanni, A., On the experimental determination of flame front positions and of propagation parameters for a fire. International Journal of Thermal Sciences, 47(9), pp. 1148-1157, 2008. Silvani, X. and Morandini, An intrusive sensor system for experimental investigations of free wildland fires. International Symposium on Identities and Mediterranean Area. pp. 110-114, July 2006. Chetehouna, K., Séro-Guillaume and Degiovanni, A., Identification of the upward gas flow velocity and of the geometric characteristics of a flame with a specific thermal sensor. International Journal of Thermal Sciences, 44(10), pp. 966-972, 2005. Huang, C.H. and Huang, C.Y., An inverse problem in estimating simultaneously the effective thermal conductivity and volumetric heat capacity of biological tissue. Applied Mathematical Modelling, 31(9), pp. 1785-1797, 2007. Shi, J. and Wang, J., Inverse problem of estimating space and time dependent hot surface heat flux in transient transpiration cooling process. International Journal of Thermal Sciences (2008), doi:10.1016/j.ijthermalsci.2008.11.019. Mandel, J., Bennethum, L.S., Beezley, J.D., Coen, J.L., Douglas, C.C., Kim, M. and Vodacek, A., A wildland. Mathematics and Computers in Simulation, 79(3), pp. 584606, 2008.

Detection of The Positions and Computing the Rate of Spread of Fire…

171

[15] Séro-Guillaume, O., Ramezani, S., Margerit, J. and Calogine, D., On large scale forest fires propagation. International Journal of Thermal Sciences, 47(6), pp.680-694, 2008. [16] Ferragut, L., Asensio, M.I. and Monedero, S., A numerical method for solving convection-reaction-diffusion multivalued equations in fire spread modelling. Advances in Engineering Software, 38(6), pp. 366-371, 2007. [17] Asensio, M. I. and Ferragut, L., On a wildland fire model with radiation. International Journal for Numerical Methods in Engineering, 54(1), pp. 137-157, 2002. [18] Sacadura, J.F., Radiative heat transfer in fire safety science. Journal of Quantitative Spectroscopy and Radiative Transfer, 93(1-3), pp. 5-24, 2005. [19] Balbi, J.H., Rossi, J.L., Marcelli, T. and Santoni, P.A., A 3D physical real-time model of surface fires across fuel beds. Combustion Science and Technology, 179 (12), pp. 2511-2537, 2007. [20] Dupuy, J.L., Testing Two Radiative Physical Models for Fire Spread Through Porous Forest Fuel Beds. Combustion Science and Technology, 155(1), pp. 149-180, 2000. [21] Siegel, R. and Howell, J., Thermal radiation heat transfer, fourth ed., Taylor and Francis, New York, 2002. [22] Monod, B., Collin, A., Parent, G. and Boulet, P., Infrared radiative properties of vegetation. Fire Safety Journal, 44(1), pp. 88-95, 2009. [23] Huang, Ch., Özisik, M.N., Inverse problem of determining unknown wall flux in laminar flow through a parallel plate duct. Numerical Heat Transfer, Part A: Applications, 21(1), pp. 55-70, 1992. [24] Li, H.Y. and Yan, W.M., Inverse convection problem for determining wall heat flux in annular duct flow. Journal of Heat Transfer, 122 (3), pp. 460-464, 2000. [25] Lee, K.H., Baek, S.W., and Kim, K.W., Inverse radiation analysis using repulsive particle swarm optimization algorithm. International Journal of Heat and Mass Transfer, 51(11-12), pp. 2772-2783, 2008. [26] Bergmann, M., Séro-Guillaume, O., and Ramezani, S., Note on the determination of the ignition point in forest fires propagation using a control algorithm. Communications in Numerical Methods in Engineering, 24(11), pp. 879-896, 2008. [27] Beck, J.V., Blackwell, B. and Clair, C.R.St.Jr. Inverse Heat Conduction, Ill-Posed Problems, John Wiley and Sons Inc., New York, 1985. [28] Michael, L.R. and Torczon, V., Pattern Search Algorithms for Bound Constrained Minimization. SIAM Journal on Optimization, 9(4), pp. 1082-1099, 1999. [29] Audet, C. and Dennis Jr., J.E., Analysis of Generalized Pattern Searches. SIAM Journal on Optimization, 13(3), pp. 889-903, 2003. [30] Chetehouna, K., Zarguili, I., Séro-Guillaume, O., Giroud F., and Picard, C., On the two ways for the computing of the fire front positions and the rate of spread. Modelling, Monitoring and Management of Forest Fires, WIT Transactions on Ecology and the Environment, 119, pp. 3-12, 2008. [31] Margerit, J. and Séro-Guillaume, O., Modelling forest fires. Part II: reduction to twodimensional models and simulation of propagation. International Journal of Heat and Mass Transfer, 45(8), pp. 1723-1737, 2002. [32] Séro-Guillaume, O., Zouaoui, D. Bernardin, D. and Brancher, J.P, The shape of magnetic liquid drop. Journal of Fluid Mechanics, 241, pp. 215-232, 1992.



Chapter 8

LARGE SCALE FOREST FIRES IN ALASKA: DETECTION AND PREVENTION *

Hiroshi Hayasaka Graduate School of Engineering, Hokkaido University, Kita-ku, N13, W8, Sapporo, 060-8628, Japan

ABSTRACT In 2004, wildfires burned 26,700km2 in Alaska. Nine individual fires exceeded 1,000km2 in size during a summer characterized by record high temperatures and extreme drought. A substantial portion of fire growth was realized on just a few days when strong pressure gradient winds occurred. Total burn area in 2004 was the largest since recordkeeping began in Alaska in 1956. Combined with an additional 19,000km2 burned in 2005, the area burned equals 10% of Alaska’s boreal forest area in just two years. Such regional fire events are believed to be climate driven. We analyzed local and regional weather factors with fire growth derived from daily MODIS “hotspot” imagery, using the 2,180km2 Boundary Fire as an example.

Keywords: Lightning ; Drought ; Hot Spot ; Boundary Fire ; Fire Weather ; MODIS

INTRODUCTION The boreal forest (Taiga) occupies one-third of the world’s forest area. During summer, the risk of fire is high in this region due to relatively low rainfall (average < 300mm) (Kasischke E.S., 2000). Recent trends toward warmer, drier summers associated with global climate change are expected to increase the number and size of boreal forest fires (Campbell *

A version of this chapter was also published in Forest Fires: Detection, Suppression and Prevention, edited by Eduards Gomez and Kristina Alvarez, published by Nova Science Publishers, Inc. It was submitted for appropriate modifications in an effort to encourage wider dissemination of research.

174

Hiroshi Hayasaka

I. D. and Flannigan, M. D., 2000). Recent regional fire events appear to support these predictions. In 2002, many large-scale forest fires occurred near Yakutsk, the capital of the Sakha Republic in Siberia, burning a total area estimated at over 23,000km2 -- the largest reported in Sakha since 1955 and about ten times greater than mean burnt area for protected forest, or about 2,400km2. In 2003, forests burned near the Baikal Lake in Siberia were especially severe, and total burnt area in Russia (Siberia) estimated in excess of 234,000km2. In 2004, many large-scale forest fires occurred in Alaska due to record-breaking lightning strikes aggravated by severe drought conditions and the occurrence of foehn wind. The total burnt area in 2004 was about 26,000km2, the largest on historical record since 1956. (NASA MODIS image with hotspots and fire clouds taken from Alaska Air line are shown in Figure 1 and Figure 2 respectively.) In 2005, many large-scale forest fires recurred in Alaska. Many fires became very active in the middle of August due to drought and foehn winds. The total burnt area in 2005 was about 26,000km2, the third largest area since 1956. Active fire occurrence of two consecutive years has seldom occurred. Trends in and features of large forest-fire occurrence and weather hold the key to understanding how climate change may affect boreal ecosystems, and fire in the boreal forest is now being actively researched. Kasischke et al. (2002) studied the patterns of large fires over the last 50 years in Alaska by analyzing the large-fire database (LFDB) and other information. The influence of El Niño weather events was investigated by Hess et al. (2001), who concluded that many of the largest fire years occurred during or just after a warm ENSO episode. Duffy et al. (2005) cleared impacts of tele-connection of large-scale atmosphericocean variability on Alaskan fire season severity. Stevens and Dallison (2005) urged that recent climate change in the boreal forest could substantially impact on the number and size of wildfires in Alaska and should be more thoroughly studied.

Figure 1. NASA satellite image in 2004.

Large Scale Forest Fires in Alaska: Detection and Prevention

175

Figure 2. Large Fire Clouds Observed in Alaska

1. OUTLINE OF ALASKAN FOREST FIRES IN LAST HALF-CENTURY 1.1. Brief Overview on Alaska and Vegetation Alaska, the largest of the 50 United States of America, is located at the northwesternmost corner of the North American Continent, mainly located between North Latitude 58 o to 71o and West Longitude 141 o to 166 o. Forest in Alaska is so-called boreal forest and mainly exists in interior Alaska surrounded by Brooks and Alaska Ranges. Fairbanks located at North Latitude 64.8 o and West Longitude 147.9 o is the center of interior Alaska. Alaska’s forest, which cover about 460,000 km2, consist mostly of black spruce, white spruce, aspen, birch, sphagnum moss, and lichens. Forest fires in Alaska sometimes spread due to so-called crown fires, and lightning is mainly responsible for large burnt areas.

1.2. Forest Fire History Map and Data A map of Alaska’s wilderness fire history from 1942 to 2007 made by the Alaska Fire Service (AFS) is shown in Figure 3. Alaskan Forest fire history data from 1956 to 2005 provided by the University of Alaska Fairbanks and the Alaska Fire Service was analyzed to distinguish anthropogenic fires from natural or lightning-caused fires (Figure 4). In this article, fire data before 1956 was not used here due to relatively low reliability of data. The bar graph in Figure 4 indicates burnt area and the line graph shows the number of fires. Smaller bars and lower lines indicate lightning-caused forest fires. There is a apparent difference between the two lines but the difference between the two bars is very small,

176

Hiroshi Hayasaka

indicating that forest fires in Alaska are mainly caused by lightning and account for much of the burnt area and for about 40% of the number of fires.

1.3. Recent Fire Activity In 2004, many fires occurred despite markedly high precipitation in May for most of Alaska. Rainfall was observed in mid-May through mid-June due to strong convection, but a long drought of about one month in June-July and strong foehn winds increased the activity of fires ignited by lightning, making the total burnt area in 2004 the largest since 1956. Combined with an additional 19,000km2 burned in 2005, the area accounted for 10% of Alaska's boreal forests in just two years. In 2005, precipitation in August was only 6.1mm -- the lowest since 1956. Many fires become very active due to severe drought in August despite high precipitation from May to July. Strong foehn winds during a severe drought in August increased fires, making the total burnt area in 2005 the third largest since 1956.

2. FOREST FIRES AND WEATHER TRENDS IN ALASKA FIRE AND WEATHER IN LAST HALF-CENTURY 2.1. General Trends Alaskan forest fire and weather trends are summarized in Table 1 using weather data from the Alaska Climate Research Center, University of Alaska Fairbanks. The average number of fires and area burned each year for three periods -- 1956-2005, 1956-1989, and 1990-2005 -- are summarized to show the increase in fire activity in Alaska between 1990 and 2005. The average number of fires and burnt area from 1990 to 2005 are clearly greater than in the other two periods. Average burnt area from 1990 to 2005 is 2.3 times greater than that of between 1956 and 1989. To determine the cause of this recent increase in fire activity in Alaska, precipitation and average maximum temperature were analyzed and summarized in Table 1. The decrease in precipitation in spring (March to May) from 1990 to 2005 ranged from 0.2 and 0.9mm compared to that between 1956 and 1989. The recent increase in maximum temperatures in spring (March to May) ranged between 1.1 to 2.8oC compared to those from 1956 to 1989. A marked temperature rise of 2.8oC occurred in April. In addition to this spring weather change, trends of lower precipitation and higher temperature are found in June, when precipitation was -0.7mm lower and 0.4oC warmer than normal. This drier trend in June may slightly increase lightning activity and fire-ignition probability by lightning. Weather changes in April and June thus indicate drier springs, increased lightning activity, and a slightly high ignition probability by lightning – all of which may be important causes of recent increased fire occurrence in Alaska.


1

Figure.3 Map of Alaska Fire History (1942-2009) *

Figure.4. Alaska Fire History (1956-2007).

1

Original map “akfirehist07.jpg” from http://agdc.usgs.gov/data/blm/fire/index.html.

177

178

Hiroshi Hayasaka Table 1. Trends in Forest Fires and Weather in Alaska

2.2. High Fire Years and Weather The top ten forest fires in Alaska are listed in Table 2. Six of these occurred after 1990. In Alaska, lightning causes most forest fires in June and July, peaking around July 1 and corresponding to the first lightning occurrence peak. Occurrence trends in forest fires and lightning in Alaska are shown in Fig.5 (Hayasaka, et. al., 2003). Most large fires (the 10 largest fires in that 14-year period, indicated by numbers 1 to 10 in Fig.5) in Alaska started from June. This is because forest fires could last several weeks due mainly to low precipitation and could become active flaming fire under favorable weather conditions such as low humidity, high temperature, and proper wind velocity. Table 2. Top Ten Forest Fires and Weather in Alaska

To determine the cause of the top ten fires from 1956 in Alaska, precipitation and the average maximum temperature are summarized in Table 3. Departures from averages for both ΔP10 (precipitation) and ΔT10 (temperature) are showed in Table 3 for easier understanding.

Average departures ΔP10 and ΔT 10 for the top ten fires and average absolute values for precipitation and temperature are listed in the bottom two columns in Table 3. Precipitation and temperature in summer (June to August) may be very important to large-scale fires. Large negative values of less than -20mm are easily found in columns for


179

the three months of June through August in Table 3. These large negative values for precipitation are found mainly in columns in the top four fire years. Values exceeding 3oC are also easily found in temperature columns in Table 3, but these temperatures are found mainly in columns in the top five fire years. These abnormal conditions for precipitation and temperature may be strongly related to fire activities. The three months of June through August show very low precipitation and high temperature in the first and second fire years, i.e., 2004 and 1957. Three months of low precipitation and high temperature also found in 1997. Two months of such conditions were found in 1977. Four other fire years except for 1990 and 2002 had one month of low precipitation and high temperature. Summer weather changes in precipitation and temperature also indicate drier summers.

ΔP10 indicate drier summers, especially in June and August, but average departures of ΔT 10 clearly show markedly high temperatures of 1 or 2 oC from March

Average departures of

to August or spring and summer. This temperature trend may be one of the most important causes of the recent increased in fire occurrence in Alaska. Table 3. Low Forest-Fire Years and Weather in Alaska

Figure.5. Fire and Lightning Occurrence Trends.

180

Hiroshi Hayasaka

2.3. Low Forest-Fire Years and Weather Low forest-fire years in Alaska are listed in Table 4. Four of five occurred in the 1960s. Table 4 shows surprisingly low burnt areas compared to the average of 2,690km2 in Table 3. The lowed burnt area is found in 1964 totaling only 14km2 or about 1/192 of the average area. To clarify the reason for low burnt area, departures from average, ΔP5 and ΔT5 , and

average departures, ΔP5 and ΔT 5 , are shown in Table 4 as in Table 3. For large fires, low precipitation and high temperature in summer were essential weather condition, while high precipitation and low temperature were key weather condition for low fires occurrence. No markedly high precipitation departure is found in Table 4, however, but high precipitation such as 80, 70, and 48mm are found in Table 3. This indicates that fire activity does not depend on precipitation much in Alaska. Precipitation exceeding 100mm is needed to suppress fires completely, but no such large precipitation is expected because annual precipitation in Alaska is about 250mm and originally as low as in the desert. Low temperature departures of less than -1oC were easily found in Table 4. Markedly low temperatures from March to May were found in the low forest-fire years of 1964 and 1961. In 1965, a low temperature trend starting in April lasted until August. In 1978 and 1963, markedly low temperatures and high precipitations occurred in June – the month most important for fire activity in Alaska. Based on the above discussion of the correlation between fire and weather conditions lead us to conclude that forest fires in Alaska are mainly controlled by temperature rather than precipitation. Table 4. Low Forest-Fire Years and Weather in Alaska

3. RECENT DETAIL FIRE ANALYSIS USING HOTSPOTS BY SATELLITES AND WEATHER DATA IN SITU 3.1. Fire Monitoring by Satellites and Fire Occurrence Tendency in 2004 MODIS (Moderate Resolution Imaging Spectrometer) satellite imagery captured by Terra and Aqua has proven useful to detect wildfires in remote locations. The number of daily “hotspots” detected by NASA using MODIS images was plotted in Fig.6. A “hotspot” (HS for short hereafter) is detected by the infrared radiation sensor with a spatial resolution of about 1.1 km2. A HS does not always mean fire, as they can be also produced by hot industrial facilities such as furnaces, boilers and so on, bare, dry and hot land areas, or


181

sunlight reflection off lakes. Still, the MODIS HS is an extremely useful tool to understand fire behavior in a large area such as Alaska or other remote areas. The cooperation of GINA (Geographic Information Network of Alaska) in downloading, preparing, and providing MODIS HS to fire management agencies in Alaska was very effective especially in detecting new fire starts or significant increases in fire activity. Histograms of the number of daily HS detected by MODIS show graphically the progression of fire activity through the summer in 2004 (Fig. 6). The first HS representing a fire was found at day number (DN for short hereafter) 162 (10/June). Spring of 2004 was rainy in interior Alaska, setting record rainfall for the month of May (49.8mm). This ended in early June and record heat followed, making June 2004 (average 19.3 oC) the warmest in Fairbanks since 1948 (Alaska Climate Research Center, 2006). In fact, June 2004 was the warmest on record for Nome, Fairbanks, Anchorage, Valdez, Juneau and King Salmon. With the strong high pressure ridge and heat of June came record-setting thunderstorms. Lightning strikes in 2004 totaled 147,642 strikes-- almost five times average (Shulski et al., 2005). The highest number of lightning strikes--about 9,000--ever recorded in one day occurred on July 15 (DN=197). Over 8,000 lightning strikes were also observed on June 14 (DN=166), August 18 (DN=231) and August 20 (DN=233). Fire activity, as reflected by MODIS HS began to show up soon after. Three HS peaks exceeding 2500 are found in Fig.6. These fire peaks occurred on June 29 (DN=181), July 13 (DN=195) and August 21 (DN=234). Three days of strong northeast winds at the end of June caused fires to rapidly spread. The Boundary Fire, just north of Fairbanks, grew from about 263 km2 to about 1,052 km2 during that time. This Boundary Fire was used in the next section as an example of modeling growth with MODIS HS data.

Figure.6. Fire Occurrence Tendency Using Hotspot

182

Hiroshi Hayasaka

3.2 Smoke from Forest Fires and Fire Expansion Observed by Satellite A Terra satellite image for the most fire active day on June 29 (DN=181) is shown in Fig.1. From this image, the concentration of fires in interior Alaska and western Canada is easily apparent. Fires have long smoke trails extending almost due west, indicating the strong pressure gradient winds driving rapid fire spread. The amount of smoke produced was so large it resembles a cloud shield on the image. The smoke (haze) from fire impact to Alaskan cities and towns was extreme. Air quality in interior and northeast Alaska was considered unhealthy or hazardous for 52 days during 2004. Fairbanks subdivisions were exposed to high levels of carbon monoxide (> 10 ppm) and smoke particulates (PM2.5) in excess of 1,000 mcg/m3, over eight times the previously recorded high from wildfire. Even indoors, borough air quality specialists observed “hazardous” levels of smoke particulates at >300 mcg/m3 over a 24-hour period (Shulski et al., 2005). Low visibility grounded air tankers and helicopters, closed airports, restricted grocery and medical services to many Alaskan towns, and all but shut down such tourism industries as flight-seeing, fly-in fishing, and remote lodges. Because fire detection aircraft were periodically grounded by low visibility, fire suppression agencies relied heavily on MODIS HS imagery to track fire expansion. Two fire maps near Fairbanks are shown in Fig.7 to display fire location and size clearly. Left-hand and right-hand maps in Fig.7 are for June 28 (DN=180) and August 21 (DN=234) respectively. Numbers from 1 to 24 were used to identify individual fires. The “Boundary Fire” (No. 9 in Fig.7) ultimately became one of the largest fires of Alaska in 2004.

Figure 7. Forest Fire Maps: Fire Location and Size.

Figure.8. Precipitation and Drought.


183

Besides the extent of fires, drought was an important factor which contributed to the unusually large amount of smoke produced by the 2004 fires. Precipitation measured at the Fairbanks Airport is plotted in Fig.8, where the red line of cumulative rainfall illustrates drought periods. A 39-day drying period began in mid-June through July, with a second dry spell in August. During these warm, dry periods deep layers of feather moss and organic duff on the forest floor became exceedingly dry (Jandt et al. 2005). Since fuel moisture is the most important factor controlling depth of burn, fires burned deep in 2004. By late July many fires were burning all the way down to mineral soil, increasing biomass available for combustion by roughly 50%.

3.3. The Boundary Fire And Weather In Situ The Boundary Fire, located about 32 km Northeast of Fairbanks, was ignited by lightning June 13 (DN=165) although MODIS first detected the fire on June 18 (DN=170). Drought and strong northeast winds caused the fire to expand rapidly, ultimately reaching 2,176 km2. The Boundary Fire was the second largest fire in 2004 and accounted for 8.4% of area burned statewide. Peaks in HS activity appear on June 29 and July 17 (Fig.9). Each peak in activity lasts about 3-5 days. The Boundary Fire did not show a third HS peak in August when other large fires expanded (Fig.6). This may be attributed to fire suppression actions that were taken on the fire as it threatened Fairbanks subdivisions. Weather data, including air temperature, relative humidity (RH), wind speed and direction, and solar radiation, was measured hourly from June 1 (DN=153) to July 31 (DN=213) at the top of Caribou Peak (773m) at the west end of the Boundary Fire (Fig.10).

3.4 Detailed Fire Expansion Analysis Using Hotspot Data Hotspot data released from NASA is spatially and temporally explicit, qualities which make it useful for plotting fire expansion. Hotspot distribution superimposed on a map of the Boundary Fire is shown in Fig.10. There were 208 hotspots reported on June 28 as shown by the square symbols. Each square is displayed with the same dimension as one pixel of the infrared radiation sensor on MODIS, namely 1.1x1.1 km. As illustrated in Fig.10, there is substantial overlapping of adjacent squares, creating problems for spatial analysis of burn area with raw HS. Transforming HS data to a composite figure using a CAD (Computer-aided Design) software routine (Vector Works, A&A Co., Ltd, see Fig.11) allows an approximation of burned area based on the indicated area of the composite figure. The accuracy of burned area obtained in this fashion is limited by the heat-detecting capability of infrared sensor in MODIS. However, present estimates from hotspot data closely follow daily burn acreage totals reported by the Alaska Fire Service fire (see Fig.12). Their final acreage estimate is about 20% greater than that developed from MODIS HS data. This difference may arise from undetectable small fires or from the fact that agencies base report burn area on burn perimeter, including any unburned islands. The unburned islands within a large fire perimeter can easily exceed 20% of the total area within the perimeter on an individual.

184

Hiroshi Hayasaka

Figure 9. Fire Occurrence Tendency of the Boundary Fire Using Hotspot Data.

Figure 10. Map of the Boundary Fire and Hotspot Distribution.

Figure 11. CAD Software Routine.


185

Figure 12. Comparison of Burnt Area Values.

3.5 Fire Weather Parameters and Fire Behavior According to MODIS Continuous records of temperature and RH are shown in Fig.13. The two red rectangles show active fire days when number of hotspots exceeded 100. Blue vertical lines show the first and last days hotspots were detected (see Fig.9). It can be seen that fires were more active when minimum RH was < 50% and maximum daily temperature was >20 oC (Fig.13). Five days after the fire was ignited by lightning, June 18 (DN=170), temperatures warmed dramatically, as high as 28ºC, while RH dropped to near 20% (Fig.13). The fire grew steadily but slowly until June 27 (DN=179), when very low RH was recorded for several consecutive days. At the same time, winds shifted to the northeast with the passage of an arctic cold front (Fig.14). Fire activity on the Boundary Fire increased dramatically and this behavior was reflected by more than 100 HS/day (Fig.9).

Figure 13. Relative Humidity and Air Temperature.

186

Hiroshi Hayasaka

The wind event or foehn wind from June 27 to July 2 (DN=179-184) was widespread and contributed to crown fire behavior and rapid fire expansion (Fig.14, Fig.1). Interestingly, dense smoke caused a substantial decrease in solar radiation during this period. Solar radiation decreased from 750 W/m2 on June 26 (DN=178) to 300 W/m2 on June 30 (DN=182). When winds shifted to the west toward the end of the week, they were still relatively strong (6m/s) but because of associated higher RH and cooler temperatures— attributed in part to shading from smoke--fires became relatively inactive, and fire behavior was mainly creeping and smoldering. The dampening effect of smoke shielding on fire behavior was noted all over the interior—the first time this negative feedback phenomenon has been documented on a regional scale in Alaska.

Figure 14. Wind Velocity and Direction.

4. FOR FUTURE FIRE PREVENTION STRATEGY Fire management agencies in Alaska in cooperation with GINA (Geographic Information Network of Alaska) have been using MODIS hotspot data for three years as a means of detecting new fire starts or significant increases in fire activity. Analyzing forest fire data with weather parameters using MODIS proved useful to track daily fire activity on the Boundary fire. In other words, MODIS hotspot data can be used quantitatively to monitor daily fire growth. This method may be useful to fire management agencies, like the Alaska


187

Fire Service, particularly when weather or budget limit use of aircraft for fire detection and mapping. Agency daily acreage reports are dependent on their ability to fly and map fires. It may be days or weeks after a fire occurs until the acreage is mapped and attributed to the cumulative burn acreage. Scientific investigations that require higher temporal resolution, such as investigating daily fire behavior and daily weather or fire danger indices, should consider the use of present method for deriving daily burn area update from MODIS hotspots. Observed lightning, fire weather and fire behavior (as captured by MODIS) were able to be directly correlated each and every day of the burning period on the Boundary Fire. The direct documentation of the negative feedback due to solar radiation reduction of massive smoke production on fire behavior at a regional scale was firstly described here. About one third of total burned area in 2004 came in August (Fig.3), which is normally the rainiest month in interior Alaska. Due to drought conditions, small fires expanded and burned deeper than usual during late summer. Autumn fires also occurred in the Taiga forest near Yakutsk (Far East Siberia) in September 2002 (Hayasaka, H., 2004). Changes in global circulation and longer, warmer growing seasons in boreal forest regions are expected to result in more late summer fire occurrence. Two of the three largest fire seasons ever recorded in Alaska have occurred in two consecutive years. This occurrence is unprecedented and indicates more resources will have to be allocated to fire management and control if this trend continues. Also management agencies need to continue to seek more efficient and costeffective ways to monitor more fires simultaneously. Remote sensing tools are one method that is being actively investigated in Alaska. Lastly, new fire-management strategies may need to be explored under climate change to protect crucial wildlife habitats, subsistence resources, and human inhabitants in the boreal forest from incendiary impacts. The author is now seeking strategies through comparative studies of forest fires and weather in Alaska in the Northern America, California in the Western America, Mongolia in the middle Eurasia Continent, Sakha in the Far Eastern Siberia, Kalimantan in Indonesia, and Victoria in Australia..

ACKNOWLEDGMENTS Many data used in this paper were obtained from various agencies and universities in the United States. Fire history data was provided from University of Alaska Fairbanks (UAF) and Alaska Fire Service (AFS). Weather data was from UAF. Satellite image and hotspot data was courtesy of MODIS Rapid Response Project at NASA/GSFC. Fire maps were provided by the USFS Remote Sensing Applications Center and BLM Alaska Fire Service. The author would like to express his appreciation for their assistance and cooperation. This research is partly supported by Research Revolution 2002 (RR2002), MEXT (Ministry of Education, Culture, Sports, Science and Technology) in Japan and IARC/JAXA Arctic Research, JAXA (Japan Aerospace Exploration Agency) in Japan.

188

Hiroshi Hayasaka

REFERENCES Alaska Climate Research Center, (2006): http://climate.gi.alaska.edu/Climate/ Location/ Interior/Fairbanks.html Barber, C. V. & Schweithelm, J. (2000). Trial by fire. Forest fire and forestry policy in Indonesia’s era of crisis and reform. World Resources Institute. Forest frontiers initiative in collaboration with WWF-Indonesia and Telapan Indonesia Foundation. pp.76. Boehm, H.-D.V., Siegert, F., Rieley, J. O, Page, S.E., Jauhiainen, J., Vasanser, H. & Jaya, A., (2001). Fire impacts and carbon release on tropical peatlands in Central Kalimantan, Indonesia. In Proceedings of the 22nd Asian Conference on Remote Sensing, 5-9 November 2001 Singapore. Bock, C. (1881). The headhunters of Borneo: A narrative of travel up the Mahakam and down the Barito, also journeyings in Sumatra. Sampson Low, Marston, Searle and Rivington, London, UK. Reprinted Oxford University Press, Singapore, 1985. Campbell I. D. & Flannigan, M. D. (2000). Long-Term Perspectives on Fire-ClimateVegetation Relationships in the North American Boreal Forest. Ecological Studies 138, 151-172, Springer. Chandrasekharan, C. (1998). The mission on forest fire prevention and management to Indonesia and Malaysia (Serawak). In: Nugroho, A., Miyakawa, H., Matsui, Y., Sato, Y., Chandrasekharan, C., Efransjah., Kuswanda, M (Eds). Tropical Forest fire, prevention, control, rehabilitation and trans-boundary issues. Bappenas-Indonesia, JICA and ITTO. pp 204-282. Dennis, R. (1999). A review of fire projects in Indonesia (1982-1998), Center for International Forestry Research (CIFOR), ISBN 979-8764-30-7. Duffy, P. A., Walsh, J. E., Graham, J. M., Mann, D. H. & Rupp, T. S. (2005). Impacts on large-scale atmospheric-ocean variability on Alaskan fire season severity. Ecological Applications, 15-4, pp. 1317-1330. Goldammer, J. G. & Siebert, B. (1990). The impact of droughts and forest fires on tropical lowland rain forest of East Kalimantan. In : J.G.Goldammer (ed.) Fire in the tropical biota. Ecosystem process and global challenges, 11-31. Ecological Studies Vol. 84. Springer-Verlag, Berlin, Germany. Harjdowigeno, S. (1996). Pengembangan lahan gambut untuk pertanian, suatu peluang dan tangtangan, Orasi Ilmiah IPB. Bogor. Kanapathy, K. (1976). Fertilizer requirement on peat soils. Malaysian Agricultural Journal 50:292-307. Hayasaka, H. (2003). Recent Tendencies of Forest Fire and Lightning in Alaska, Bulletin of Japan Association for Fire Science and Engineering, 53-1, 17-22 (in Japanese). Hayasaka, H. (2004). Large Forest Fires Occurred in Yakutsk Area and Lightning, 2004 Proceedings of Japan Association for Fire Science and Engineering, 266-268 (in Japanese). Hess, J. C., Scott, C. A., Hufford, G. L. & Fleming, M. D. (2001). El Niño and its impact on fire weather conditions in Alaska. Int. J. of Wildland Fire, 10, 1-13. Jandt, R. R., Allen, J. A. & E. A. Horschel. (2005). Forest floor moisture content and fire danger indices in Alaska. BLM Technical Report 54:30 pp.


189

Kasischke, E. S. (2000). Boreal Ecosystems in the Global Carbon Cycle, Ecological Studies 138, 19-30, Springer. Kasischke, E. S., Williams, D. & Barry, D (2002). Analysis of the patterns of large fires in the boreal forest region of Alaska. Int. J. of Wildland Fire, 11, 131-144. Latham, D. & Williams, E. (2001), Lightning and Forest Fires, Forest Fires, Academic Press, pp.375. Maltby, E. & Immirzi. C. P. (1993). Carbon dynamic in peatland and other wetland soil. Regional and global perspectives. Chemosphere 27(6): 999-1023. Muraleedharan, T. R., Radojevic. M., Waugh.A. & Caruana. A. (2000). Emission from the combustion of peat: an experimental study. Atmospheric Environment, 34: 3033-3055. Nugroho. K., Gianinazzi. G. & Wijaya Adhi. I. P. G., (1997). Soil hydraulic properties of Indonesia peat. In: Rieley, J. O and Page, S. E (Eds.) Biodiversity and Sustainability of Tropical Peatlands. Samara Publishing. Cardigan, UK, pp. 147-155. Page. S. E., Siegert, F., Rieley. J. O., Boehm. H.-D. V., Jaya. A. & Limin. S. H. (2002). The amount of carbon released from peat forest fire in Indonesia during 1997. Nature 420: 61-65. Shulski, M., Wendler, G., Alden, S. & Larkin, N. (2005). Alaska’s Exceptional 2004 Fire Season, Sixth Symposium on Fire and Forest Meteorology (CD-ROM), Canmore, Canada, 2005. Siegert, F., Ruecker, G., Hinrichs, A. & Hoffmann, A. A. (2001a). Increased damages from fires in logged forests during droughts caused by El Niño. Nature, 414: 437-440. Siegert, F., Boehm, H. D. .V., Rieley, J. O., Page, S. E., Jauhiainen, J., Vasander, H. & Jaya A. (2001b). Peat fire in Central Kalimantan Indonesia: Fire impact and carbon release. In: Rieley, J. O and Page, S. E (Eds.) with Setiadi, B. Proceeding of Jakarta Peatland Symposium on Peatland for people, natural resource, functions and sustainable managements. BPPT and Indonesian Peat association. pp 142-154. Stevens, E. & Dallison, D, (2005). An extraordinary summer in the interior of Alaska, P3.24, Report on the panel at University of Alaska Fairbanks. Wein, R. W. (1983). Fire behavior and ecological effects in organic terrain. In: Wein, R. W., MacLean, D. A. (ed.) The role of fire in Northern Circumpolar Ecosystem. John Wiley and Sons. New York. pp 81-95.

INDEX 2 21st century, vii, 150

A absorption, 18, 19, 20, 21, 22, 24, 34, 61, 156, 169 abstraction, 85 accounting, 19, 22, 23, 32, 34, 38 accuracy, 2, 4, 7, 15, 16, 18, 20, 22, 23, 24, 36, 57, 80, 85, 86, 88, 89, 99, 100, 102, 105, 108, 109, 114, 162, 183 ADC, 56 advantages, 6, 35, 36, 80 aerosols, viii, 52, 103, 104, 105, 108, 114, 115 Africa, 3, 37, 116, 117 age, ix, 67, 89, 93, 94, 97, 123, 129, 130, 131, 133, 134, 135, 136, 137, 138, 141, 142, 144 agencies, 181, 182, 183, 186, 187 Air Force, 38, 87, 88, 91, 92, 93, 94, 95, 96 air quality, 117, 182 air quality model, 117 air temperature, 15, 24, 33, 183 Alaska, v, x, 37, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 186, 187, 188, 189 algorithm, vii, 1, 5, 6, 12, 14, 18, 29, 30, 33, 34, 35, 37, 38, 49, 50, 57, 59, 70, 72, 73, 74, 82, 88, 107, 134, 162, 166, 168, 169, 171 Amazon, 104, 105, 106, 108, 109, 112, 113, 114, 115, 116, 117 Amazon Tropical Rain Forest, 105, 106, 108 ambient air, 48, 49, 56, 58, 156 amplitude, 17, 71 Anhysteretic Remanent Magnetization, 121 anthropogenic fire, viii, 103, 175 archaeological sites, 120, 123, 124 ARM, 121, 122, 126 artificial intelligence, vii, viii, 41, 42, 49 assessment, 49, 80, 85, 86, 88, 100, 101, 102, 104, 113, 115, 123

assimilation, 170 ATCOR, 19 atmosphere, vii, viii, 1, 2, 3, 14, 15, 16, 17, 18, 19, 23, 26, 27, 28, 29, 30, 31, 32, 33, 34, 36, 38, 41, 42, 43, 51, 52, 53, 56, 58, 61, 74, 103, 104, 105, 107, 112, 116, 120 Atmospheric Modeling System, ix, 104, 105, 115, 116 automated fire surveillance, 51, 69 automatic scanning, 56, 63 automation, 73 AVHRR/NOAA, vii, 1, 2, 35, 37

B background information, 107 background radiation, 32, 45, 56 backscattering, viii, 41, 44, 48, 49, 50, 51, 70 bacteria, 122, 128 bandwidth, 46 base, 2, 70, 105, 108, 112, 113, 114, 143, 145, 150, 183 biodiversity, ix, 105, 129, 141 biofuel, 115 biomass, viii, 37, 55, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 127, 145, 183 biotic, 142, 143 boilers, 180 Boltzmann constant, 46, 107 boreal forest, x, 2, 4, 5, 7, 9, 35, 37, 99, 130, 131, 146, 147, 173, 174, 175, 176, 187, 189 boundary conditions, 107 bounds, 100 Brazil, 3, 107, 117 burn, x, 8, 81, 82, 83, 89, 90, 91, 92, 93, 94, 97, 98, 99, 101, 104, 173, 183, 187

C cables, 51

192

Index

CAD, 183, 184 calibration, 25, 118, 154 campaigns, 109, 114 carbon, viii, 79, 103, 105, 113, 115, 116, 117, 127, 182, 188, 189 carbon monoxide, viii, 103, 105, 182 case study, viii, 79, 80, 86, 87, 99, 126, 127, 147 catchments, 120, 122 category a, 86 CCATT-BRAMS, ix, 104, 105, 108, 109, 110, 112, 113 cell size, 81 challenges, 188 chemical, 104, 107, 112, 115 Chicago, 137 China, v, ix, 129, 130, 131, 144, 145, 146, 147, 148 circulation, 187 class, 81, 84, 85, 93, 102, 133, 134, 135, 137, 138, 148 classes, 84, 85, 88, 133, 137, 138, 142 classification, viii, 41, 69, 72, 73, 74, 81, 82, 83, 84, 85, 86, 87, 88, 98, 100, 101, 102, 147 cleaning, 89 climate, x, 104, 115, 117, 130, 131, 132, 142, 144, 146, 173, 174, 187, 188 climate change, 104, 173, 174, 187 clustering, 82 clusters, 82 CNS, 87, 92 CO2, 20, 21, 22, 23 collaboration, 188 colonization, 141 color, iv, 73 combustion, viii, 33, 49, 67, 103, 108, 118, 119, 120, 154, 156, 161, 183, 189 commercial, 151 communication, 56, 57 community, 22, 83, 156 comparative analysis, 12, 13 compensation, 24, 25, 34 competition, 143 complexity, 16, 53, 54 composition, 27, 56, 73, 81, 102, 112, 130, 131, 133, 136, 141, 142, 146 compression, 70 computation, x, 19, 51, 70, 149, 168 computer, x, 4, 15, 17, 19, 43, 56, 57, 77, 150, 170 computing, 150, 171 conception, 130 conductivity, 67, 152, 170 configuration, 66, 147 configurations, 53, 159 conformity, ix, 104, 109, 114

conservation, 145, 151 constant rate, 156 constituents, 24 construction, 56, 57 consumption, viii, 57, 82, 83, 84, 85, 97, 103, 104, 105, 108, 110, 111, 112, 113, 114, 118 contingency, 86 continuous data, 88 contour, viii, 4, 28, 29, 30, 31, 42, 59 convergence, 50, 82, 89 cooling, 58, 119, 120, 151, 170 cooling process, 170 cooperation, 36, 181, 186, 187 copper, x, 150, 152, 154 correlation, 18, 24, 25, 42, 107, 110, 180 correlations, ix, 85, 104 cost, x, 53, 54, 55, 57, 67, 74, 150, 151, 187 cotton, 150 Coupled Chemistry-Aerosol-Tracer Transport, viii, 103, 105 creep, 186 crown, 133, 175, 186 crown fires, 133, 175 crowns, 9, 133 cultivation, 123 cycles, viii, 103, 104, 115

D damages, iv, 189 danger, 52, 53, 187, 188 data processing, 3, 4 database, 20, 26, 38, 89, 133, 136, 174 decay, 121 decomposition, 72, 132, 133, 165 deforestation, 104, 105, 109, 112, 114, 116, 117 degradation, 68, 98 Delta, 97 demonstrations, 119 Department of Agriculture, 145, 146 deposition, 105, 106 depth, 21, 23, 104, 117, 154, 183 derivatives, 50, 162 detectable, 10 detection, iv, vii, viii, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 33, 34, 35, 36, 37, 38, 41, 42, 43, 45, 47, 48, 51, 53, 54, 55, 57, 61, 64, 65, 66, 67, 68, 69, 70, 72, 73, 74, 77, 107, 120, 125, 126, 127, 182, 187 detection system, 61, 73 detection techniques, vii deviation, 16, 61, 113 diaphragm, 56 differential equations, 49

Index diffusion, 106, 162, 165, 171 diode laser, 55, 57, 67 disadvantages, 74, 80 discrimination, 120, 125, 128 diseases, 141, 147 dispersion, 49, 60 displacement, 166, 168 distinctness, 97 distortion, 19, 21, 23, 24, 33 distribution, viii, x, 8, 25, 27, 32, 41, 48, 49, 50, 51, 60, 68, 79, 85, 93, 99, 113, 115, 116, 132, 148, 149, 161, 165, 183 disturbances, 130, 132, 143, 145, 148 divergence, 52, 53, 55, 57, 64 diversity, ix, 102, 120, 129, 144 DOI, 115 dominance, 84, 126 drought, x, 173, 174, 176, 183, 187 drying, 165, 183

E ecology, 99, 100 economic damage, vii, 150 economic values, 132 ecosystem, ix, 79, 100, 126, 129, 131, 132, 143, 145 editors, 128 Efficiency, 10 El Niño, 174, 188, 189 elaboration, 33 election, 70 electrical conductivity, 67 electromagnetic, 81 electron, 46 electrons, 46 emission, viii, 32, 33, 42, 43, 56, 68, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113, 114, 116 emitters, 66 empirical studies, 131, 142 energy, 44, 46, 47, 48, 52, 53, 54, 55, 56, 57, 64, 77, 104, 107, 113, 115, 116, 117, 118, 156, 165, 169 energy density, 52 engineering, 161 environment, viii, 2, 67, 103, 148 environmental quality, 115 environmental variables, 85 equality, 33, 165 equilibrium, 48 equipment, 43, 47, 53, 74, 121, 124 erosion, 127, 128 Eurasia, 187 Europe, 4 European Union, 150 evaporation, 165

193

evidence, ix, 81, 85, 86, 89, 117, 119, 124, 126, 127 evolution, viii, ix, 42, 49, 51, 58, 61, 62, 79, 80, 104, 105, 109 exclusion, 141 exploration, 80 exposure, 52 extinction, viii, 41, 44, 48, 49, 50, 51, 70, 156, 163, 169 extraction, viii, 41, 51, 69, 73

F factories, 42 Fairbanks, 175, 176, 181, 182, 183, 187, 188, 189 false alarms, 4, 6, 51, 73 federal government, 87 ferrimagnets, 120, 121, 127, 128 ferromagnetic, 120 fiber, 76 Field Of View, 14 field tests, 65 fire detection, vii, 1, 2, 3, 4, 6, 7, 9, 13, 35, 37, 42, 57, 61, 64, 67, 73, 74, 107, 127, 182, 187 fire detection algorithms, 2, 37 fire event, viii, x, 79, 80, 86, 122, 126, 173, 174 fire hazard, ix, 129 fire mapping, viii, 79, 80, 83, 85, 98, 99 fire radiative energy, viii, 103, 117, 118 Fire Radiative Energy, viii, 103, 107 fire radiative energy (FRE), viii, 103 Fire Radiative Power, viii, 103, 104 fire suppression, ix, 87, 129, 130, 131, 133, 135, 136, 137, 138, 139, 141, 142, 143, 144, 145, 147, 148, 182, 183 fishing, 182 flame, x, 32, 34, 112, 149, 150, 151, 155, 156, 157, 158, 159, 162, 163, 166, 169, 170 flight, 106, 182 flights, 105, 109 fluid, 75 fluorescence, 53 force, 58, 79 forest ecosystem, ix, 129, 130, 131, 141, 145 forest fire, vii, x, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 27, 35, 37, 39, 41, 42, 57, 61, 64, 65, 66, 70, 77, 101, 102, 117, 122, 126, 146, 147, 148, 149, 150, 156, 162, 165, 169, 170, 171, 173, 174, 175, 176, 178, 180, 186, 187, 188, 189 forest management, ix, 129, 130, 132, 141, 142, 143, 144, 145 Forest Protection Services, 2, 3, 4, 7, 9, 10, 36 formation, 53, 55, 120 formula, 49, 107 foundations, vii, 1

194

Index

FOV, 14 fragments, 93 France, 77, 122, 123, 127, 149, 150, 154 FRE, viii, 103, 104, 107, 108, 111, 114 freedom, 80 frequencies, 79, 99, 110, 121, 126 FRP, viii, 103, 104, 105, 107, 109, 111, 112, 113, 114, 118 fuel consumption, 82, 83, 84, 85, 97 fuel loads, ix, 87, 129, 130, 131, 138, 139, 140, 142 fuel management, 142 function estimation, 162 fuzzy membership, 84 fuzzy sets, 101

G geometrical parameters, 17 geometry, 2, 6, 7, 15, 17, 18, 150, 156, 159, 170 Geostationary Operational Environmental Satellites, viii, 103 Geostationary satellites, 82 Germany, 76, 188 germination, 147 global climate change, 173 global scale, 14 global warming, 130 glue, x, 150, 152 GOES, viii, 82, 103, 105, 107, 108, 111, 114, 117 grain size, 119, 121, 122, 124, 125, 126, 127, 128 granules, 12, 33 graph, 110, 114, 125, 175 grassland biomes, 105, 106, 108 grasslands, viii, 103 gravity, 49 Great Xing’an Mountains, v, ix, 129, 130, 131, 141, 143, 146, 148 Greece, 150 greenhouse, viii, 103 greenhouse gases, viii, 103 growth, x, 15, 17, 83, 88, 89, 97, 98, 126, 137, 138, 141, 147, 148, 173, 181, 186 growth rate, 83, 88, 89, 97, 98 guidance, 98

H habitat, 87, 101, 141, 148 habitats, 141, 187 hardwood forest, 147 harvesting, ix, 129, 130, 132, 136, 143, 144, 146 hazards, ix, 129 haze, 182 health, vii, 150

heat capacity, 165, 170 heat transfer, 75, 154, 171 height, 23, 58, 59, 60, 61, 150, 156, 159, 163, 169 heterogeneity, 80, 83, 84, 148 histogram, 7, 9, 93 history, 87, 115, 126, 127, 142, 143, 175, 187 Holocene, 126, 127, 128 homogeneity, 84, 141 hot spots, 109 hotspots, 4, 6, 8, 174, 180, 183, 185, 187 House, 36, 38, 74 human, 52, 82, 83, 144, 187 human nature, 83 humidity, 16, 22, 24, 25, 26, 33, 83, 165, 178, 183 hybrid, 101 hypothesis, 143, 156

I IAO, vii, 1, 2, 3, 6, 7, 12, 13, 18, 35, 36, 38 ideal, 48, 88, 98, 121 identification, 33, 34, 77, 124, 151, 163, 169 identity, 166, 168 illumination, 73, 74 image, 3, 4, 7, 8, 11, 27, 28, 31, 35, 73, 74, 80, 81, 83, 84, 86, 87, 88, 89, 97, 98, 101, 150, 163, 164, 169, 170, 174, 182, 187 imagery, x, 80, 81, 85, 87, 90, 92, 98, 100, 102, 133, 135, 173, 180, 182 images, 4, 6, 7, 9, 10, 11, 15, 27, 33, 36, 73, 74, 82, 88, 89, 98, 100, 101, 114, 122, 150, 154, 180 imaging systems, 53 impacts, 104, 174, 187, 188 incidence, 109, 112, 123, 126, 128 independent variable, 166 Indonesia, 187, 188, 189 industrial environments, 42, 57 industries, 182 industry, 74 inequality, 165 Institute of Atmospheric Optics, vii, 1, 3 integration, 74, 104, 105, 107, 157 integrity, 143 intelligence, vii, viii, 41, 42, 51, 57, 72 interface, vii, 150 inversion, viii, 41, 50 ionization, 67 IR spectra, 28, 108 iron, ix, 119, 120, 124, 126, 128 islands, 183 issues, 80, 82, 100, 144, 188 Italy, 150

Index

J Japan, 173, 187, 188 John F. Kennedy Space Center, 87 Jordan, 80, 101, 105, 116

K Kennedy, John F., 87 kill, 142

L labeling, 60, 85, 87, 90, 99 Laboratory of Radiometry, viii, 103, 108, 111 lakes, 6, 84, 122, 123, 181 laminar, 161, 171 Landsat Thematic Mapper, 88 landscape, ix, 87, 93, 94, 97, 102, 129, 130, 131, 132, 134, 135, 136, 138, 139, 141, 142, 143, 144, 145, 146, 147, 148, 170 landscapes, 130, 131, 145, 146, 147 laptop, 151, 152 LARAD, viii, 103, 108, 111, 114 Large Scale Biosphere-Atmosphere Experiment in Amazonia, ix, 104 laser radiation, vii, viii, 41, 42, 61 lasers, 52, 53, 55, 57, 67, 74 laws, 151 LBA, ix, 104, 105, 108, 109, 110, 114 lead, 24, 51, 60, 142, 180 learning, 72, 73 lidar, viii, 41, 42, 43, 44, 45, 47, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 64, 66, 67, 68, 69, 70, 72, 73, 74, 76, 77 lidar detection, 45, 64 lidar signal, viii, 41, 42, 43, 45, 47, 50, 51, 57, 58, 62, 68, 69, 70, 72, 73, 74 Lidar technique, vii lifetime, 106 light, vii, 27, 41, 42, 43, 44, 49, 55, 56, 57, 73, 74, 110, 113 light detection and ranging, vii, 41, 42 light scattering, 49 Limitations, 143 line graph, 175 localization, 7 longevity, 134, 141, 144 low temperatures, 180 LOWTRAN, 15, 17, 19, 38 lying, 22, 142

195

M magnetic field, 121 magnetic materials, 127 magnetic properties, 121, 122, 126, 127, 128 magnetism, 122, 127, 128 magnetization, 121 magnitude, 22, 52, 54, 68, 71, 73, 93, 120, 155 majority, 3, 73 management, ix, 36, 87, 89, 97, 104, 114, 129, 130, 132, 141, 142, 143, 144, 145, 146, 147, 148, 181, 186, 187, 188 mangrove forests, 105 mapping, viii, 53, 69, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 92, 93, 94, 97, 98, 99, 100, 101, 102, 117, 187 marsh, 98 MAS, 107 masking, 9, 12, 89, 102 mass, 48, 57, 68, 105, 108, 115, 117, 120, 121, 131, 165 mass loss, 115 materials, 120, 121, 127, 133 matrix, 72, 73, 153 matter, iv, viii, 84, 103, 105, 108, 119, 120 measurement, x, 18, 23, 37, 44, 99, 109, 117, 121, 127, 149, 150, 169 measurements, viii, ix, x, 2, 3, 6, 7, 12, 16, 17, 18, 19, 24, 25, 26, 27, 31, 32, 34, 36, 38, 59, 62, 103, 104, 107, 108, 109, 111, 113, 114, 116, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 149, 150, 151, 161, 162, 165, 166, 170 media, 53 median, 92 medical, 182 Mediterranean, 170 membership, 84 memory, 56 Merritt Island National Wildlife Refuge, 87, 88, 90, 91, 92, 93, 94, 95, 96, 99 meter, 54, 55, 64, 88, 121 methodology, 86, 101, 113, 151 Mie theory, 49 Ministry of Education, 187 MINWR, 87, 88, 89, 90, 92 mission, 188 missions, 112 Missouri, 129, 142, 145, 146, 147, 148 mixing, 58, 106 mixture analysis, 84, 100, 101 modeling, 17, 38, 48, 100, 101, 144, 145, 146, 170, 181 modelling, 154, 156, 162, 165, 169, 171

196

Index

models, viii, x, 16, 17, 19, 20, 22, 23, 41, 48, 49, 51, 57, 66, 105, 107, 115, 131, 143, 144, 147, 149, 151, 154, 156, 165, 169, 171 Moderate Resolution Imaging Spectrometer, 180 Moderate Resolution Imaging Spectroradiometer, viii, 103, 105 modifications, 149, 173 MODIS, vii, viii, x, 1, 12, 13, 14, 18, 19, 20, 21, 24, 25, 26, 27, 29, 31, 33, 34, 35, 36, 37, 39, 103, 105, 107, 108, 111, 114, 115, 116, 117, 173, 174, 180, 181, 182, 183, 185, 186, 187 MODTRAN, 19, 21, 22, 23, 26, 31, 33, 38 modules, 144 moisture, 18, 24, 33, 39, 83, 104, 115, 183, 188 moisture content, 18, 24, 33, 83, 104, 115, 188 molecules, 20, 21 Mongolia, 187 monitoring, vii, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 18, 19, 34, 35, 37, 39, 42, 74, 116 mortality, 141, 144, 147 mosaic, 93, 94 moving window, 72 MRI, 136, 137

N National Forest Fire Centre of Russia, 2, 3 national parks, 99 natural disturbance, ix, 129, 130, 143, 145 natural resources, 101 Nd, 52, 53, 54, 64 negative feedback, 186, 187 neglect, 49 neodymium, 54 Netherlands, 37 neural network, viii, 41, 51, 84 neural networks, 51, 84 nodes, 70, 73 noise, 43, 44, 45, 46, 47, 61, 68, 70, 71, 72 North America, 117, 175, 188 numerical analysis, 58

O obstruction, 82 oil, 26, 27, 33, 42, 50, 57 old age, 93 one sample t-test, 137 open spaces, 141, 143 operations, 73, 87 optical parameters, 15 optical systems, 44 ordinary differential equations, 49 organic matter, 119, 120

oscillations, 155 oxidation, 120 ozone, 26, 33, 39, 42, 52, 116

P Pacific, 115 paints, 170 parallel, 19, 38, 161, 171 parameter estimation, 162 particle mass, 108 pasture, 108, 115 peat, 188, 189 PGE, 25 Philadelphia, 75 photodetector, 43, 44, 45, 47, 56 photographs, 27 pinus, 170 pioneer species, 132, 141 plants, 98, 99 platform, 82 PM, 103 polarization, 55 policy, 130, 145, 188 policy makers, 130 pollen, 123, 126, 127 population, 110 porosity, 165 Portugal, 41, 76, 150 positive correlation, 24, 25 power plants, 42 precipitation, 109, 131, 176, 178, 179, 180 predators, 141 preparation, iv, 36 prescribed burning, ix, x, 129, 142, 144, 149, 150, 151, 165 pressure gradient, x, 173, 182 prevention, 188 principles, 37, 43, 67, 70, 144 probability, 4, 8, 9, 10, 35, 68, 71, 72, 73, 74, 122, 134, 139, 143, 144, 176 probe, vii, 41 programming, 19 project, 81, 85, 136 propagation, x, 45, 49, 54, 60, 61, 149, 162, 165, 169, 170, 171 proposition, 83 protection, 148 prototype, 36 public health, vii, 150 pyrolysis, 165

Index

Q qualitative differences, 125 quantization, 12

R RaCCI, ix, 104, 105, 106, 108, 109, 110, 113, 114 racing, ix, 119, 120 radar, 73, 74, 77, 117 radiation, vii, viii, 6, 18, 19, 20, 21, 22, 24, 31, 32, 33, 38, 41, 42, 43, 44, 45, 46, 50, 51, 52, 53, 54, 55, 56, 61, 64, 67, 72, 81, 103, 104, 107, 112, 156, 162, 171, 180, 183, 186, 187 Radiation, ix, 19, 49, 104, 107, 108 Radiation, Cloud and Climate Interactions, ix, 104 radio, 73 radiometer, 18, 19, 38 radius, 48, 49, 50, 73 rain forest, 188 rainfall, ix, 104, 109, 117, 173, 181, 183 real time, 114, 151, 152 reality, 85, 120 reception, 43 recognition, viii, 41, 49, 51, 69, 70, 71, 72, 73, 99 recommendations, iv, 37, 72 reconstruction, 18, 80, 87, 98, 123, 127, 169 recovery, 82, 102 recurrence, 6 reflectivity, 45 reform, 188 refraction index, 50 regression, x, 105, 108, 110, 111, 150, 156, 164 regression model, 105, 108 regrowth, 116 rehabilitation, 188 rejection, 4, 7, 11, 51 relevance, 20 reliability, 4, 6, 54, 175 relief, 52, 57, 60 remote sensing, vii, viii, x, 1, 37, 79, 80, 81, 85, 95, 99, 100, 101, 102, 149 replacement, 141, 146 reproduction, 134 requirements, 22, 61, 74 RES, 149 reserves, 141 resistance, 25, 44, 154 resolution, x, 27, 33, 38, 42, 61, 70, 73, 77, 81, 85, 88, 98, 106, 107, 108, 115, 117, 133, 134, 136, 143, 147, 150, 180, 187 resources, 70, 79, 84, 101, 141, 187

197

response, ix, 73, 74, 126, 129, 130, 131, 132, 141, 144, 148 restoration, 148 rights, iv risk, ix, 52, 57, 67, 129, 130, 131, 133, 139, 140, 141, 142, 144, 146, 147, 173 room temperature, 121 root, 48, 71 root-mean-square, 71 routines, 73, 82, 83, 86 RTM method, vii, 19, 20, 21, 22, 23, 24, 25, 33, 34, 35, 36 rules, 73, 87, 148 Russia, 1, 2, 3, 11, 25, 37, 174

S safety, viii, 42, 52, 53, 54, 55, 171 satellite monitoring, 2, 3, 4, 5, 7, 8, 9, 11, 12, 35 satellites, 2, 4, 6, 7, 8, 9, 10, 13, 15, 35, 82 saturation, 12, 107 Saturation Isothermal Remanent Magnetisation, 122 scatter, 63 scattering, 49, 156 science, viii, 79, 80, 81, 87, 100, 171 scientific understanding, 145 sea level, 60 seasonality, 79 security, vii, 60, 150 sediment, ix, 80, 119, 120, 122, 123, 127 sedimentary records, 127 sediments, ix, 119, 120, 122, 126, 127, 128 seed, 132, 136, 141, 144, 146 seeding, 134 seedlings, 144 senescence, 141 senses, 81 sensing, vii, viii, x, 1, 2, 36, 37, 38, 52, 67, 77, 79, 80, 81, 85, 95, 98, 99, 100, 101, 102, 116, 149, 187 sensitivity, viii, 33, 41, 42, 45, 49, 53, 55, 61, 65, 67, 68, 73, 74, 82, 116 sensors, vii, 42, 57, 67, 74, 82, 88, 107, 150, 166 services, iv, 182 shade, 132, 134 shape, x, 39, 49, 51, 58, 68, 71, 73, 79, 150, 151, 169, 171 showing, 99, 107, 109, 155 shrubs, 101, 143 Siberia, 2, 37, 174, 187 signals, viii, 41, 51, 57, 58, 62, 63, 64, 68, 72, 152, 155 signal-to-noise ratio, 47, 48, 53, 54, 56, 62, 68, 71 signs, 24

198

Index

simulation, ix, 16, 25, 49, 109, 116, 129, 130, 131, 136, 138, 139, 141, 142, 143, 146, 147, 148, 171 simulations, x, 106, 108, 110, 133, 136, 137, 149 Singapore, 78, 188 SIRM, 122, 123, 125, 126 SMOCC, ix, 104, 105, 106, 108, 109, 110, 113, 114 smoke plume, vii, viii, 41, 42, 45, 48, 49, 50, 51, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 71, 72, 73 software, vii, 1, 3, 7, 15, 19, 20, 25, 33, 36, 49, 56, 57, 72, 183 soil erosion, 127 soil type, 83 solution, 2, 19, 32, 33, 36, 49, 51, 67, 72 South America, viii, 103, 104, 105, 106, 107, 108, 109, 112, 113, 114, 117 Space Science and Engineering Center, 25 Spain, 4, 150 Spatial Analysis Algorithm, 6 specialists, 182 species, ix, 87, 95, 101, 102, 104, 129, 130, 131, 132, 133, 134, 135, 136, 138, 141, 142, 143, 144, 146 specifications, 151 spin, 106 spore, 126 Spring, 91, 181 SSEC, 25 stability, 33, 54, 73 standard deviation, 16, 61, 113 standard error, 90, 91, 105, 108, 110, 111 state, 2, 12, 19, 25, 32, 33, 35, 36, 38, 49, 55, 57, 81, 142 statistics, 6, 70, 97 steel, x, 150, 152, 154 stochastic model, 143 stoichiometry, 119 storage, 57, 82, 117 stress, 144 structure, 30, 31, 32, 42, 48, 51, 58, 59, 60, 61, 65, 70, 102, 130, 131, 136, 138, 141, 145, 147 subjectivity, 82 sub-Saharan Africa, 116 subsistence, 187 succession, 130, 132, 142, 143, 146, 147 superparamagnetic, ix, 119, 121, 123 suppression, ix, 87, 129, 130, 131, 133, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 147, 148, 182, 183 surface energy, 115 surveillance, vii, viii, 41, 42, 43, 51, 52, 55, 56, 57, 61, 63, 64, 67, 68, 69, 70, 73, 74 survey, 102

survival, 79, 119, 122, 143, 147 susceptibility, ix, 119, 120, 121, 123, 125, 127 Sweden, 76 Switzerland, 126 symmetry, 48

T tags, 72 target, vii, 41, 43, 45, 47, 57, 69, 70, 71 techniques, vii, viii, 73, 79, 80, 83, 84, 85, 86, 98, 127, 142, 150, 170 technologies, 19, 73 technology, 12, 13, 42, 74, 150 TEM, 122 terraces, 132 terrestrial ecosystems, vii, 150 territory, vii, 1, 2, 3, 5, 6, 9, 13, 33, 35, 39 test data, 13 testing, 5, 12, 34, 36, 117 thermal resistance, 154 thinning, 142, 147 threats, 150 timber production, ix, 130, 133, 142, 144 time series, 87, 88, 97, 98, 101, 109 tissue, 170 Tomsk Region, 2, 3, 4, 5, 6, 7, 9, 13, 17, 35, 39 tourism, 182 Toyota, 74 tracks, 86 training, 49, 72, 73, 84 transformation, 70, 168 transformations, 166 transmission, 56, 152 transparency, 18, 22, 56 transpiration, 170 transport, 105, 106, 112, 113, 115, 127 treatment, 24, 132, 142 tropical dry forest, 105 tropical forests, 104 tropical savannas, 105 tundra, 99, 101 turbulent mixing, 49

U UK, 75, 147, 188, 189 unburned fuels, 83 uniform, 63, 83, 84, 133 unique features, 61 United, 77, 78, 119, 147, 175, 187 United Kingdom, 78, 119 United States, 77, 147, 175, 187 universality, 19

Index universities, 187 urban, 16, 17, 84, 102, 106 urban area, 102 urban areas, 102 USA, 4, 79, 100, 101, 102, 129, 137, 147 USDA, 145, 146

V validation, 72, 101 valleys, 132 vapor, 18, 19, 25, 39 variables, 83, 84, 120, 161, 162, 163, 166, 169 variations, 16, 24, 104, 113, 123, 126, 146, 166, 167, 168 vector, 6, 72, 89, 158, 159, 162, 168 vegetation, x, 27, 53, 66, 73, 81, 82, 83, 88, 89, 97, 98, 100, 102, 104, 106, 109, 113, 115, 117, 120, 126, 130, 131, 137, 142, 145, 147, 149, 156, 165, 169, 171 vegetative cover, 98 vegetative reproduction, 134 velocity, 43, 48, 49, 60, 170, 178 versatility, 6 video, 67, 73, 74, 150, 154, 169 videos, 169 viscosity, 121 vision, 150, 170

199

W Wales, 127 Washington, 75, 77 water, 6, 12, 18, 19, 25, 33, 39, 92, 132, 135, 136, 151, 165 water vapor, 18, 19, 25, 39 wavelengths, 54, 56, 87 WD, 148 West Africa, 37 Western Australia, 145 Western Siberia, 2 wetlands, 94 WFABBA, viii, 103, 105, 107, 114 wilderness, 175 wildfire, 52, 99, 100, 120, 133, 143, 147, 148, 182 Wildfire Automated Biomass Burning Algorithm, viii, 103, 105 wildland, vii, 150, 170, 171 wildland-urban interface, vii, 150 wildlife, 141, 187 wood, 42, 50, 124, 165 woodland, 124 workstation, 38 World Health Organization, vii, 150 worldwide, 130

Y yield, 127

Fire Detection

BS 5839 - 2002 Fire Detection

Fire Detection and Alarms BS 5839

Detection Unlimited

Quickest detection

Fire Fire

Multiuser Detection

Detection Unlimited

Fault Detection

Detection Unlimited

Detection Unlimited

Malware Detection

Detection Unlimited

Quickest Detection

Fire

Fire

Fire

Fire

Fire

Fire

Fire

Fire With Fire

Fire to Fire

Fire

Fire

Fire ()

Fire to Fire

Fire to Fire

Fire

Fire

Fire Detection

BS 5839 - 2002 Fire Detection

Fire Detection and Alarms BS 5839

Detection Unlimited

Quickest detection

Fire Fire

Multiuser Detection

Detection Unlimited

Fault Detection

Detection Unlimited

Detection Unlimited

Malware Detection

Detection Unlimited

Quickest Detection

Fire

Fire

Fire

Fire

Fire

Fire

Fire

Fire With Fire

Fire to Fire

Fire

Fire

Fire ()

Fire to Fire

Fire to Fire

Fire

Fire

Recommend Documents