Ecological Models for Regulatory Risk Assessments of Pesticides: Developing a Strategy for the Future (Society of Environmental Toxicology and Chemistry)

Ecological Models for Regulatory Risk Assessments of Pesticides Developing a Strategy for the Future

Other Titles from the Society of Environmental Toxicology and Chemistry (SETAC) Veterinary Medicines in the Environment Crane, Boxall, Barrett 2008 Relevance of Ambient Water Quality Criteria for Ephemeral and Effluentdependent Watercourses of the Arid Western United States Gensemer, Meyerhof, Ramage, Curley 2008 Extrapolation Practice for Ecotoxicological Effect Characterization of Chemicals Solomon, Brock, de Zwart, Dyev, Posthumm, Richards, editors 2008 Environmental Life Cycle Costing Hunkeler, Lichtenvort, Rebitzer, editors 2008 Valuation of Ecological Resources: Integration of Ecology and Socioeconomics in Environmental Decision Making Stahl, Kapustka, Munns, Bruins, editors 2007 Genomics in Regulatory Ecotoxicology: Applications and Challenges Ankley, Miracle, Perkins, Daston, editors 2007 Population-Level Ecological Risk Assessment Barnthouse, Munns, Sorensen, editors 2007 Effects of Water Chemistry on Bioavailability and Toxicity of Waterborne Cadmium, Copper, Nickel, Lead, and Zinc on Freshwater Organisms Meyer, Clearwater, Doser, Rogaczewski, Hansen 2007 Ecosystem Responses to Mercury Contamination: Indicators of Change Harris, Krabbenhoft, Mason, Murray, Reash, Saltman, editors 2007 For information about SETAC publications, including SETAC’s international journals, Environmental Toxicology and Chemistry and Integrated Environmental Assessment and Management, contact the SETAC Administratice Office nearest you: SETAC Office 1010 North 12th Avenue Pensacola, FL 32501-3367 USA T 850 469 1500 F 850 469 9778 E [email protected]

SETAC Office Avenue de la Toison d’Or 67 B-1060 Brussells, Belguim T 32 2 772 72 81 F 32 2 770 53 86 E [email protected]

www.setac.org Environmental Quality Through Science®

Ecological Models for Regulatory Risk Assessments of Pesticides Developing a Strategy for the Future Edited by

Pernille Thorbek, Valery E. Forbes, Fred Heimbach, Udo Hommen, Hans-Hermann Thulke, Paul J. Van den Brink, Jörn Wogram, Volker Grimm

Boca Raton London New York

CRC Press is an imprint of the Taylor & Francis Group, an informa business

Information contained herein does not necessarily reflect the policy or views of the Society of Environmental Toxicology and Chemistry (SETAC). Mention of commercial or noncommercial products and services does not imply endorsement or affiliation by the author or SETAC. Published in collaboration with the Society of Environmental Toxicology and Chemistry (SETAC) 1010 North 12th Avenue, Pensacola, Florida 32501 Telephone: (850) 469-1500 ; Fax: (850) 469-9778; Email: [email protected] Web site: www.setac.org © 2010 by The Society of Environmental Toxicology and Chemistry (SETAC) SETAC Press is an imprint of the Society of Environmental Toxicology and Chemistry. No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number: 978-1-4398-0511-4 (Paperback) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging‑in‑Publication Data Ecological models for regulatory risk assessments of pesticides : developing a strategy for the future / editors, Pernille Thorbek … [et al.] p. cm. Proceedings of a workshop. Includes bibliographical references and index. ISBN 978-1-4398-0511-4 (pbk. : alk. paper) 1. Pesticides--Environmental aspects--Congresses. 2. Pesticides--Risk assessment--Congresses. 3. Ecological risk assessment--Simulation methods. I. Thorbek, Pernille. QH545.P4E27 2010 632’.95042--dc22 Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com and the SETAC Web site at www.setac.org

2009035207

SETAC Publications Books published by the Society of Environmental Toxicology and Chemistry (SETAC) provide in-depth reviews and critical appraisals on scientific subjects relevant to understanding the impacts of chemicals and technology on the environment. The books explore topics reviewed and recommended by the Publications Advisory Council and approved by the SETAC North America, Latin America, or Asia/Pacific Board of Directors; the SETAC Europe Council; or the SETAC World Council for their importance, timeliness, and contribution to multidisciplinary approaches to solving environmental problems. The diversity and breadth of subjects covered in the series reflect the wide range of disciplines encompassed by environmental toxicology, environmental chemistry, and hazard and risk assessment, and life-cycle assessment. SETAC books attempt to present the reader with authoritative coverage of the literature, as well as paradigms, methodologies, and controversies; research needs; and new developments specific to the featured topics. The books are generally peer reviewed for SETAC by acknowledged experts. SETAC publications, which include Technical Issue Papers (TIPs), workshops summaries, newsletter (SETAC Globe), and journals (Environmental Toxicology and Chemistry and Integrated Environmental Assessment and Management), are useful to environmental scientists in research, research management, chemical manufacturing and regulation, risk assessment, and education, as well as to students considering or preparing for careers in these areas. The publications provide information for keeping abreast of recent developments in familiar subject areas and for rapid introduction to principles and approaches in new subject areas. SETAC recognizes and thanks the past coordinating editors of SETAC books: A.S. Green, International Zinc Association Durham, North Carolina, USA C.G. Ingersoll, Columbia Environmental Research Center US Geological Survey, Columbia, Missouri, USA T.W. La Point, Institute of Applied Sciences University of North Texas, Denton, Texas, USA B.T. Walton, US Environmental Protection Agency Research Triangle Park, North Carolina, USA C.H. Ward, Department of Environmental Sciences and Engineering Rice University, Houston, Texas, USA

Contents Abbreviations.............................................................................................................ix List of Figures............................................................................................................xi List of Tables........................................................................................................... xiii Preface...................................................................................................................... xv Acknowledgments...................................................................................................xvii About the Editors.....................................................................................................xix Workshop Participants.......................................................................................... xxiii Chapter 1 Executive Summary of the LEMTOX Workshop: Lessons Learned and Steps to Be Taken.............................................................1 Volker Grimm, Valery E. Forbes, Fred Heimbach, Pernille Thorbek, Hans-Hermann Thulke, Paul J. Van den Brink, Jörn Wogram, and Udo Hommen Chapter 2 Introduction to the LEMTOX Workshop............................................ 11 Pernille Thorbek Chapter 3 Short Introduction to Ecological Modeling......................................... 15 Volker Grimm Chapter 4 Regulatory Challenges for the Potential Use of Ecological Models in Risk Assessments of Plant Protection Products................. 27 Jörn Wogram Chapter 5 Development and Use of Matrix Population Models for Estimation of Toxicant Effects in Ecological Risk Assessment......... 33 John D. Stark Chapter 6 MASTEP: An Individual-Based Model to Predict Recovery of Aquatic Invertebrates Following Pesticide Stress............................... 47 Paul J. Van den Brink and J.M. (Hans) Baveco Chapter 7 Incorporating Realism into Ecological Risk Assessment: An ABM Approach................................................................................... 57 Chris J. Topping, Trine Dalkvist, and Jacob Nabe-Nielsen vii

viii

Contents

Chapter 8 Ecological Models Supporting Management of Wildlife Diseases............................................................................................... 67 Hans-Hermann Thulke and Volker Grimm Chapter 9 State of the Art of Ecological Modeling for Pesticide Risk Assessment: A Critical Review........................................................... 77 Volker Grimm, Pernille Thorbek, Amelie Schmolke, and Peter Chapman Chapter 10 The Role of Ecological Modeling in Risk Assessments Seen from an Academic’s Point of View..................................................... 89 Valery E. Forbes Chapter 11 Potential Role of Population Modeling in the Regulatory Context of Pesticide Authorization.....................................................97 Franz Streissl Chapter 12 Ecological Modeling: An Industry Perspective................................ 105 Pernille Thorbek, Paul Sweeney, and Ed Pilling References.............................................................................................................. 111 Index....................................................................................................................... 121

Abbreviations ABM AFSSA ALMaSS BEFORE DAR DEBtox EEC EFSA EM EU FOCUS IBM LOC MASTEP NOEC NOEL ODD OECD PDF PPP PRAPeR SETAC TER UBA USEPA

Agent-based model Agence Française de Securité Sanitaire des Aliments (French Agency for Food Safety) Animal, Landscape, and Man Simulation System BEech FOREst Draft assessment report Dynamic Energy Budget aquatic toxicity test software Expected environmental concentration European Food Safety Authority Ecological model European Union Forum for the Co-ordination of Pesticide Fate Models and their Use Individual-based models Level of concern Metapopulation Model for Assessing Spatial and Temporal Effects of Pesticides No observable effect concentration No observed effects level Overview–design concepts–detail Organisation for Economic Co-operation and Development Probability density function Plant protection product Pesticide Risk Assessment Peer Review Unit (EFSA) Society of Environmental Toxicology and Chemistry Toxicity exposure ratio Umweltbandesamt (German Federal Environment Agency) US Environmental Protection Agency

ix

List of Figures Figure 3.1  Schematic representation of different tasks involved in developing and using models.................................................................................... 18 Figure 5.1  Deterministic population projection of a hypothetical species........ 37 Figure 5.2  Stochastic population projection for a hypothetical species; average population abundance over time (dotted line) and associated 95% confidence limits....................................................................................................... 39 Figure 5.3  Comparison of deterministic population projections for 3 fish species and Daphnia pulex....................................................................................... 41 Figure 5.4  Dose–response curves for tephritid fruit flies exposed to acephate................................................................................................................ 43 Figure 5.5  Deterministic population projections for an oriental fruit fly control population and a population exposed to the acephate EEC resulting in 83% mortality............................................................................................................ 43 Figure 5.6  Deterministic population projections for a melon fly control population and a population exposed to the acephate EEC resulting in 7% mortality....................................................................................................................44 Figure 5.7  Stochastic population projections for oriental fruit fly exposed to the EEC of acephate..............................................................................................44 Figure 6.1  Overview of the scheduling of state change for an Asellus individual in the Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides (MASTEP). ............................................................................ 50 Figure 6.2  Dynamics of population numbers in all treatment levels (a) for the treated 100 m stretch, (b) the complete 600 m stretch, and (c) 95% confidence intervals of the dynamics of numbers of the treated 100 m stretch....... 53 Figure 6.3  Visual representation of the dynamics of abundance for one of the runs of the 10 m buffer zone treatment level...................................................... 53 Figure 7.1  ALMaSS screenshot of a typical 10 × 10 km landscape used for simulations. ........................................................................................................60 Figure 7.2  (a) Variation in carrying capacity (K) among weather years within a single 500 × 500 m2 in the 10 × 10 km natural landscape. (b) Variation in population size among the 400 squares in the weather year 1995, which was used repeatedly over 200 simulation years. ........................................... 61 Figure 7.3  A section of a 10 × 10 km landscape before and after rounding of landscape features and subsequent randomization of their position. .................. 62 xi

xii

List of Figures

Figure 7.4  Beetle population numbers plotted against time for decreasingly realistic landscape structures. . ........................................................... 62 Figure 7.5  Simulated vole population depressions after the application of pesticide under 2 scenarios, 100% exposure or realistic application to orchards..... 65 Figure 7.6  Simulated vole population depression with decreasing toxicity (expressed as NOEL) for a “vinclozolin-like” pesticide. ......................................... 65 Figure 8.1  Graphical depiction of the alternative management concept........... 70 Figure 8.2  Screenshot of the model landscape of the rabies model, applied to compare alternative emergency control options. .................................... 71 Figure 8.3  The performance of the ring vaccination compared to the compact application of vaccination resources around the disease outbreak............ 71 Figure 8.4  Schematic representation of the fox tapeworm model. ................... 73 Figure 8.5  Simulation results regarding the sustainability of strategies to control fox tapeworm infections in red fox populations........................................... 73 Figure 9.1  Distribution of model types in ecological models that have been used to assess the risk of pesticides for nontarget organisms in papers published between 2000 and May 2007................................................................... 82 Figure 9.2  Toxicity endpoints and risk measures used in ecological models that have been used to assess the risk of pesticides for nontarget organisms in papers published between 2000 and May 2007. (a) Toxicity endpoints. (b) Measure used to quantify risk. ......................................................... 82 Figure 9.3  Use of calibration, verification, and validation in ecological models that have been used to assess the risk of pesticides for nontarget organisms in papers published between 2000 and May 2007.................................. 83 Figure 11.1  The number of substances for each group of organisms for which an acceptable risk was not fully demonstrated............................................ 100 Figure 11.2  The number of uses for which risk assessments of focal species (FS) were accepted or rejected based on the following refinements: proportion of diet taken from the treated field (PT), proportion of different food types (PD), or refinement of residues (R)....................................................... 101 Figure 11.3  The most sensitive group of organisms driving the aquatic risk assessment as percent of the total of 50 substances in list 2............................ 102 Figure 11.4  The number of substances for which mesocosm studies were submitted and the number of substances for which the risk was deemed acceptable or unacceptable based on the mesocosm endpoint............................... 102

List of Tables Table 5.1  Toxicity and Hazard Ratio of Acephate to 3 Fruit Fly Species........... 42 Table 5.2  Comparison of Recovery Time and Generation Time for 3 Fruit Fly Species Exposed to the Acephate EEC..............................................................44 Table 7.1  Impact Assessments of Insecticide to All Arable Fields for 4 Species...................................................................................................................... 63 Table 8.1  Comparison of Characteristic Aspects of Controlling Wildlife Pathogens or Crop Pests, and of Potentials of Ecological Modeling in These 2 Fields of Application................................................................................................. 68

xiii

Preface The protection goal of pesticide regulation is, in most cases, populations. At the population level, the effects of crop protection products on nontarget organisms depend not only on the exposure and sensitivity to the toxicant in question, but also on factors such as life history characteristics (e.g., dispersal abilities, fecundity, generation time, age or life stage-specific sensitivity), population structure, density dependence, timing of exposure, ecosystem processes such as predation and competition, and landscape structure. Ecological modeling presents an excellent tool whereby the importance and interaction of such factors can be explored and the effects on populations can be predicted. Therefore, ecological modeling has the potential to be implemented into ecological risk assessment and the regulatory process. The focus of the workshop “Ecological Models in Support of Regulatory Risk Assessments of Pesticides: Developing a Strategy for the Future,” referred to as “LEMTOX,” was on population models such as unstructured models, matrix models, and individual- or agent-based models. In the area of environmental fate, modeling is already used routinely to increase the realism, relevance, and robustness of exposure assessments. In contrast, ecological modeling has received very little attention in effects assessments. This workshop brought together 35 experts from academia, regulatory authorities, contract research organizations, and industry from Europe, the United States, and Asia to discuss the benefits of modeling in the context of registrations, to identify obstacles preventing ecological modeling being used routinely in regulatory submissions, and to agree on actions needed to overcome these obstacles. Ecological models were identified as potentially important tools for the following topics in pesticide risk assessment: • to extrapolate effects and recovery patterns observed in higher-tier tests (e.g., mesocosm studies) to species with different biological traits, to longerterm effects, or to the landscape scale; • to extrapolate effects from one exposure pattern to another; and • to explore the impacts of sublethal effects on the population level (especially for vertebrates) to analyze and predict indirect effects caused by the interactions of populations from different species. However, there are also obstacles to overcome if ecological modeling is to be efficiently used in regulatory risk assessments of pesticides. We discussed the following issues: • uncertainties in the process of model development, for example, selection of model type, model design (e.g., how much complexity is necessary), model analysis (sensitivity analysis, verification, validation), and model documentation and communication; • availability of data and background information needed for good modeling; xv

xvi

Preface

• limited knowledge of modeling; and thus • lack of confidence in the outcome of ecological models and their reliability in pesticide risk assessment. The workshop participants concluded that these obstacles for a wider use of ecological models should be solved on 3 levels:

1) by providing a guidance document, that is, good modeling practice, on the modeling process (including design, testing, application, documentation, and reporting); 2) by increasing the confidence in ecological models by confirming that they have the ability to provide more ecologically relevant risk assessments through case studies; and 3) by training the people who generate or evaluate results obtained by ecological models.

Thus, as follow-up actions of the LEMTOX workshop, it is planned to organize a workshop especially to develop guidelines for good modeling practice with the focus on pesticide registration and to develop well-documented examples (case studies) of model development, documentation, and application that can be used to offer training and a basis for further discussion for stakeholders and a wider user group in workshops and short courses.

Acknowledgments This book presents the proceedings of a Society of Environmental Toxicology and Chemistry (SETAC) European workshop that took place in Leipzig, Germany, September 9 to 12, 2007. The 35 scientists involved in this workshop represented 11 countries. We thank all of the participants for contributing actively to the debate. In addition, we thank all who commented on the report, and especially Lorraine Maltby, Anne Alix, Bin-le Lin, Patrice Carpentier, Dieter Schaefer, Peter Dohmen, and anonymous reviewers for constructive criticism. The workshop was made possible by the generous support of the following institutions: • Syngenta

• Helmholtz Centre for Environmental Research — UFZ

• Bayer CropScience

• RifCon

xvii

About the Editors Pernille Thorbek is an ecological modeler at Syngenta, United Kingdom, where she develops ecological models supporting highertier risk assessments. She is also involved in several research projects that aim to develop ecological models for risk assessments of pesticides, and she is currently industry supervisor for 6 PhD students and postdocs in the area. She is on the steering committee of the SETAC Europe Advisory Group on Mechanistic Effect Models for Ecological Risk Assessment (MEMoRisk) and has coorganized several sessions on ecological modeling at SETAC conferences. She has a PhD in ecology from University of Aarhus, Denmark, and an MSc in ecology from University of Copenhagen, Denmark. Valery E. Forbes has her PhD in coastal oceanography from the State University of New York at Stony Brook. She is currently head of the Department of Environmental, Social and Spatial Change and professor of aquatic ecology and ecotoxicology at Roskilde University, Denmark. She is director of an international center of excellence, Centre for Integrated Population Ecology (CIPE), is on the editorial board of several international journals, and provides scientific advice to the private and public sectors. Specific research topics include population ecology, fate and effects of toxic chemicals in sediments, and ecological risk assessment. She has published about 100 internationally peer-reviewed articles and 2 books on these topics.

xix

xx

About the Editors

Fred Heimbach works as a consultant scientist at RifCon GmbH in Leichlingen, Germany. He obtained his MSc degree and PhD in conducting research on marine insects at the Institute of Zoology, Physiological Ecology at the University of Cologne. From 1979 until 2007 he worked at Bayer CropScience in Monheim, Germany, on the side effects of pesticides on nontarget organisms. In addition to his work, he gave lectures on ecotoxicology at the University of Cologne. Dr. Heimbach has researched the development of single-species toxicity tests for both terrestrial and aquatic organisms and has worked with microcosms and mesocosms in the development of multispecies tests for these organisms. As an active member of European and international working groups, he participated in the development of suitable test methods and risk assessment of pesticides and other chemicals for their potential side effects on nontarget organisms. He has served for several years on SETAC Europe and the SETAC World Council, and he has been an active member of the organizing committees of several European workshops on specific aspects of the ecotoxicology of pesticides. Udo Hommen is senior scientist at the Fraunhofer Institute of Molecular Biology and Applied Ecology in Schmallenberg, Germany. He obtained his MSc degree and PhD at the RWTH Aachen University working on the use of freshwater model ecosystems and ecosystem models for chemical risk assessment. At the Fraunhofer IME he is responsible for aquatic micro- and mesocosm studies. His general research interest is higher-tier risk assessments of plant protection products and other chemicals, and he is also working on the statistical evaluation of laboratory and field tests, as well as monitoring studies, probabilistic risk assessment, and ecological modeling. His teaching activities include lectures at RWTH Aachen University, short courses at SETAC conferences, and (until 2002) summer schools on ecological modeling. Hommen has actively participated at several international SETAC workshops; he is council member of the SETAC Europe German Language Branch, steering committee member of the SETAC Advisory Group in Aquatic Macrophyte Ecotoxicology (AMEG), and chair of the SETAC Europe Advisory Group on Mechanistic Effect Models for Ecological Risk Assessment (MEMoRisk).

About the Editors

xxi

Hans-Hermann Thulke studied mathematics and biology at the University of Leipzig. In 1996 he obtained his PhD working on probability theory. He is senior scientist at the Helmholtz Centre for Environmental Research — UFZ and is supervising the project group Ecological Epidemiology at the Department of Ecological Modelling. In close collaboration with national and international authorities or research bodies, the group develops management-oriented ecological models particularly tailored to support strategic decision making in disease control and emergency planning. His teaching assignments covered university lecturing (University of Cape Town, Leipzig, and Halle); international graduate courses in modeling, statistics, or risk assessment (ESF, Peer-Network, UFZ — winter school); and supervision of PhD, diploma, and undergraduate students. He is a member of WDA and SVEPM. He is involved in EFSA activities as a scientific expert and has fulfilled consultancies for public and private bodies regarding modeling, risk assessment, and strategy planning. Paul J. Van den Brink is a professor of chemical stress ecology and works at the research institute Alterra and the Aquatic Ecology and Water Quality Management Group of Wageningen University, both belonging to the Wageningen University and Research Centre. He is involved in supervising and executing international projects on the scientific underpinning of higher-tier risk assessment procedures for contaminants. Recent research topics are the development of effect models (e.g., food web, metapopulation, and expert base models), trait-based ecological risk assessment (TERA), the validation of risk assessment procedures (e.g., uniform principles and species sensitivity distribution concept), and human and ecological risk assessment of pesticide use in developing countries in the tropics. Since 1994, Van den Brink has published more than 85 peer-reviewed papers, for 2 of which he won an international prize. In 2006 he won the LRI-SETAC Innovative Science Award. He also organized and took part in many international workshops and courses. He is presently a member of the SENSE research school (www.sense.nl), associate fellow of the Canadian River Institute, president of SETAC Europe, and an editor of the journal Environmental Toxicology and Chemistry.

xxii

About the Editors

Jörn Wogram is a biologist with special interest in aquatic ecotoxicology and environmental risk assessment. In 2003, he joined Germany’s Federal Environmental Agency (UBA), where he is now responsible for the coordination of the environmental risk assessment of plant protection products. Wogram studied biology with an emphasis on ecology at the University of Braunschweig, where he also obtained his PhD in 2001. During his academic career, he led or contributed to several scientific projects in the fields of limnology, effect and exposure monitoring of pesticides in streams, development of aquatic micro- and mesocosm test systems, and the environmental behavior and ecotoxicology of veterinary pharmaceuticals. Wogram has co-organized several scientific workshops and congresses. He has a teaching position at the University of Landau and is a council member of SETAC Europe. Volker Grimm is an expert in ecological modeling and theory. He has been involved in the development of more than 20 ecological models of mainly populations, including insects, mammals, birds, plants, and trees. He has coauthored a book including user-friendly software for metapopulation viability analysis. His main research interest is the method of ecological modeling. He promotes the general modeling strategy of “pattern-oriented modeling.” Topics addressed in his methodological work are optimizing model complexity, model verification and validation, and model communication. In 2006, he published with 28 coauthors the first general protocol for describing individual- and agentbased models (the “ODD protocol”). Since 2006, he has been involved in establishing ecological modeling for pesticide risk assessment. He co-organized the first international workshop focusing on ecological models for pesticide risk assessment (LEMTOX), which took place at the UFZ in 2007. He is on the editorial board of 6 international journals and a member of the Scientific Council of the Center for Ecological Research, Poland. He is teaching at the University of Potsdam and giving international courses in ecological modeling and individual-based modeling.

Workshop Participants* Hans Baveco Alterra, Wageningen University and Research Centre The Netherlands

Udo Hommen Fraunhofer Institute for Molecular Biology and Applied Ecology Germany

Eric Bruns Bayer CropScience Germany

Matthias Liess Department System Ecotoxicology, Helmholtz Centre for Environmental Research — UFZ Germany

Patrice Carpentier Unité Ecotoxicologie–Environnement, DiVE — Direction du Végétal et de l’Environnement AFSSA — French Food Safety Agency France Peter Dohmen BASF Germany Virginie Ducrot Equipe d’écotoxicologie et qualité des milieux aquatiques, INRA — Centre de Rennes France

Bin-Le Lin Research Center for Chemical Risk Management, National Institute of Advanced Industrial Science and Technology Japan Steffen Matezki Federal Environment Agency (UBA) Germany Vibeke Møller Environmental Protection Agency Denmark

Valery E. Forbes Department of Environmental, Social and Spatial Change Roskilde University Denmark

Jacob Nab-Nielsen Section for Climate Effects and System Modelling, National Environmental Research Institute Denmark

Nika Galic Alterra, Wageningen University and Research Centre The Netherlands

Ed Odenkirchen Office of Pesticide Programs, US Environmental Protection Agency United States

Volker Grimm Department of Ecological Modelling, Helmholtz Centre for Environmental Research — UFZ Germany

Rob Pastorok Integral Consulting, Inc. United States

*

Affiliations were current at the time of the workshop.

xxiii

xxiv

Workshop Participants

Ed Pilling Syngenta United Kingdom

Paul Sweeney Syngenta United Kingdom

Thomas Preuss Institute for Environmental Research, RWTH Aachen University Germany

Hans-Hermann Thulke Department of Ecological Modelling, Helmholtz Centre for Environmental Research — UFZ Germany

Melissa Reed Pesticides Safety Directorate United Kingdom Thorsten Schad Bayer CropScience Germany Dieter Schaefer Bayer CropScience Germany Amilie Schmolke Department of Ecological Modelling, Helmholtz Centre for Environmental Research — UFZ Germany Alois Staněk State Phytosanitary Administration Czech Republic John D. Stark Department of Entomology, Washington State University United States Franz Streissl Pesticide Unit (PRAPeR), EFSA Italy

Chris J. Topping National Environmental Research Institute Denmark Magnus Wang RifCon Germany Jörn Wogram Federal Environment Agency (UBA) Germany Csaba Szentes Central Agricultural Office, Directorate of Plant Protection, Soil Conservation and Agri-environment Hungary Paul J. Van den Brink Aquatic Ecology and Water Quality Management Group, Wageningen University and Research Centre The Netherlands Peter Van Vliet Board for the Authorisation of Plant Protection Products and Biocides, Ctgb The Netherlands

xxv

Workshop Participants

Workshop participants:

1) Franz Streissl 2) Ursula Schmitz 3) Jörn Wogram 4) Chris J. Topping 5) Peter Van Vliet 6) Csaba Szentes 7) Alois Staněk 8) Nika Galic 9) Hans Baveco 10) Melissa Reed 11) Thierry Caquet 12) Hans-Hermann Thulke 13) Rob Pastorok 14) Thorsten Schad 15) Vibeke Møller 16) Thomas Preuß 17) Eric Bruns 18) Udo Hommen 19) Amelie Schmolke 20) Virginie Ducrot

21) John D. Stark 22) Dieter Schaefer 23) Paul J. Van den Brink 24) Peter Dohmen 25) Ed Pilling 26) Valery E. Forbes 27) Magnus Wang 28) Patrice Carpentier 29) Paul Sweeney 30) Volker Grimm Not shown: Matthias Liess Bin-Le Lin Jacob Nab-Nielsen Ed Odenkirchen Fred Heimbach Pernille Thorbek Annette Schmidt Steffen Matezki (Photo credit: A. Künzelmann, UFZ)

Summary of 1 Executive the LEMTOX Workshop Lessons Learned and Steps to Be Taken Volker Grimm, Valery E. Forbes, Fred Heimbach, Pernille Thorbek, Hans-Hermann Thulke, Paul J. Van den Brink, Jörn Wogram, and Udo Hommen Contents Why LEMTOX?..........................................................................................................2 Views from Academia, Industry, and Regulators........................................................3 V. Grimm: Introduction to Ecological Modeling...............................................3 J. Wogram: Is There a Place for EMs in the Risk Assessment under EU Directive 91/414 EEC?..........................................................................3 H.-H. Thulke and V. Grimm: EMs for Risk Assessment of Wildlife Diseases.................................................................................................4 V. Grimm et al.: Current State of EMs in Pesticide Risk Assessment...............4 V. Forbes: Academia’s View on EMs.................................................................4 F. Streissl: EFSA’s View on EMs.......................................................................4 P. Thorbek et al.: Industry’s View on EMs........................................................4 Three Example Models...............................................................................................5 Results from Group Discussions.................................................................................5 What Are the Benefits of Using Ecological Modeling as a Pesticide Risk Assessment Tool?..........................................................................5 What Are the Barriers for Ecological Modeling Being Used More Frequently in Pesticide Risk Assessment?.............................................7 What Will It Take for Ecological Modeling to Become More Widely Used in Pesticide Risk Assessment?......................................................8 Discussion...................................................................................................................9

The workshop titled “Ecological Models in Support of Regulatory Risk Assessments of Pesticides: Developing a Strategy for the Future,” referred to as “LEMTOX,” had the aim to bring together experts from business, academia, and regulatory authorities 1

2



Ecological Models for Regulatory Risk Assessments of Pesticides

1) to discuss the benefits of ecological models (EMs) for pesticide risk assessment, 2) to identify issues that so far have prevented a wider use of EMs in the regulatory context, and 3) to agree on the next steps to be taken to establish EMs as a more widely used tool for pesticide risk assessment.

During the workshop, in a series of keynote lectures, the perspectives of industry, academia, and regulatory authorities were summarized, and 3 example models were presented. In breakout group and plenary discussions, the 3 topics of the workshop were discussed. There was general consensus that EMs have a high potential, or might even be the only way, to achieve risk assessments that are more ecologically relevant. A major challenge in establishing EMs as a common tool is the lack of a guidance document defining good modeling practice. The next steps to be taken in order to establish EMs in pesticide registration are to build a core group of stakeholders to organize a follow-up workshop where a draft guidance document on good modeling practice is produced, and to perform carefully selected case studies that follow the guidance document and that clearly demonstrate the added value of risk assessments that are supported by ecological models.

Why LEMTOX? The aim of the LEMTOX workshop was to discuss the potential role of ecological modeling for pesticide risk assessment and registration. The focus of the workshop was on population models. Participants came from highly diverse backgrounds, that is, academia, regulatory authorities, contract research organizations, and industry, and from different countries, that is, Austria, the Czech Republic, Denmark, France, Germany, Hungary, Japan, the Netherlands, the United States, and the United Kingdom. Either they were ecological modelers, who had used EMs to support risk assessments or who had evaluated risk assessments with EMs, or they were interested in how EMs could be used in regulatory risk assessments of pesticides. Thus, the issue at hand was not whether models are a useful tool for decision making (e.g., Tannenbaum 2007), nor was it our intention to defend EMs against extreme skepticism (e.g., Beissinger and Westphal 1998; Glaser and Bridges 2007). Rather, the overall approach toward EMs in the regulatory context was pragmatic optimism: The need for and potential of EMs to improve risk assessment for certain important questions such as population-level risk assessments was acknowledged, but concerted actions will be required by all stakeholders involved to establish EMs as a routinely used tool for pesticide risk assessment. During the workshop, 8 keynote presentations were given by representatives from academia, industry, and regulatory authorities to summarize the corresponding perspectives. The keynotes were aimed at giving introductions to ecological modeling and how it may fit into the current regulatory framework, to show how it is used to improve ecological management in other areas, and to give insights into the perspectives of the different stakeholder groups. Three further keynote presentations

Executive Summary of the LEMTOX Workshop

3

presented examples of different types of EMs that can, and have been, used for pesticide risk assessment and summarized the current state of the art. In addition to the presentations, there were 3 breakout group sessions where the following topics were discussed: • What are the benefits of using ecological modeling as a pesticide risk assessment tool? • What are the obstacles that have so far prevented a wider use of EMs for pesticide registration? • What are the steps to be taken to put the potential of EMs for decision support into practice and establish them as a standard tool? The breakout sessions were followed by plenary sessions. In the following paragraphs, we will summarize the keynote presentations and the results of the breakout group sessions and plenary discussions (see also Forbes et al. 2009). These summaries are based on the workshop summary given at the end of the workshop by the rapporteur (Udo Hommen).

Views from Academia, Industry, and Regulators V. Grimm: Introduction to Ecological Modeling Models are defined as purposeful representations. They are thus always much simpler than their real-world counterpart, and a clear purpose (problem, research question) is needed to develop and assess a model. The diversity of existing model types partly reflects different kinds of questions that can be asked about the same system, for example, a population that is affected by pesticides. Nevertheless, many decisions made by modelers on the choice of model type and structure seem to be ad hoc. For EMs to be used for decision support, we must go beyond ad hoc choices and find clear criteria for deciding when to use what type of model so that we obtain similar solutions to similar problems (Chapter 3).

J. Wogram: Is There a Place for EMs in the Risk Assessment under EU Directive 91/414 EEC? The answer to this question is yes! For example, EMs are useful for extrapolating recovery processes and identifying the ecological relevance of effects observed in standard lab tests, in particular for birds and mammals. However, care must be taken not to model only species that are used in generation of standard ecotox endpoints because these are often chosen for their ease of culture rather than for the representativeness of their life histories. It is important to include vulnerable species in order to get a representative picture of the effects of pesticides on nontarget organisms (Chapter 4).

4

Ecological Models for Regulatory Risk Assessments of Pesticides

H.-H. Thulke and V. Grimm: EMs for Risk Assessment of Wildlife Diseases Ecological epidemiology of wildlife diseases is a field in which EMs are increasingly used for decision support. Models are used to design field experiments and sometimes even to substitute for field experiments. The establishment of EMs in this field took about 10 years and was based on increasing confidence of the stakeholders involved. The models in use reflect expert knowledge on small-scale processes and extrapolate them to larger scales. Validation and mechanistic understanding were key issues for the acceptance of models aimed at decision support (Chapter 8).

V. Grimm et al.: Current State of EMs in Pesticide Risk Assessment What is the state of the art of EMs for pesticide risk assessment? In a review of 39 publications, Grimm et al. (Chapter 9) scanned the models for certain characteristics. They found a high diversity of model types, with a dominance of simple models that often were neither verified nor validated. Toxicity was often included in a simplistic way, not fully making use of the potential of EMs. Moreover, risk was often not quantified in a way that would allow using model output in regulatory risk assessments.

V. Forbes: Academia’s View on EMs A central issue for pesticide risk assessment is extrapolation from individual- to population-level effects and from small temporal and spatial scales to larger ones. Empirical methods to tackle these issues are limited. Models are thus the only way to explore the full range of ecological complexities that may be of relevance for ecological risk assessment. However, EMs are not a silver bullet. Transparency is key, and certain challenges exist, for example, translating model output to useful risk measures. To make full use of models and get them established for risk assessment, we need case studies that clearly demonstrate the added value of this approach (Chapter 10).

F. Streissl: EFSA’s View on EMs The analysis of environmental risk assessments of 50 pesticides revealed that risks to birds and mammals often were not fully addressed. EMs could be useful for filling this gap, but should also be useful for aquatic organisms and nontarget arthropods, particularly in relation to questions of recovery. A major obstacle to the full use of EMs is the lack of clear protection goals in terms of population-level effects. Validation is key to the acceptance of models in the regulatory context, and guidance is needed for the assessment of EMs and their outputs (Chapter 11).

P. Thorbek et al.: Industry’s View on EMs Ecological modeling is superbly suited for combining such factors as exposure, toxicity, species ecology, and landscape characteristics. Typical questions industry would like to be answered by EMs are: Is there potential for recovery? What will

Executive Summary of the LEMTOX Workshop

5

the recovery time be? What impacts do sublethal effects have on the population level? How do landscape characteristics affect persistence of populations? So far, however, the lack of guidelines for model construction, testing, and assessments creates uncertainty about how models will be assessed by regulatory authorities (Chapter 12).

Three Example Models Matrix models have a long history in ecology (Chapter 5). They are easy to use and understand and provide a means to project the population-level consequences of effects at the individual level. They cannot represent spatial effects. Including stochasticity and density dependence is possible but makes their use more complicated. The model MASTEP (Chapter 6) is an example of a “simple” individual-based model that includes life cycle, spatial effects, and density dependence but ignores variability within cohorts and includes no adaptive decision making by individuals. The model is designed to predict recovery of aquatic invertebrates following pesticide stress. It is also designed to directly add a biological component to the FOCUS surface water scenarios. MASTEP’s focus is on insight and visualizing model output, so that confidence in the model can build up gradually. ALMaSS (Chapter 7) is a complex, spatially explicit individual- or agent-based model that includes a high-resolution realistic representation of a specific landscape in Denmark, the behavior and farming practice of farmers, and detailed behaviorbased models of certain species, for example, skylarks or bank voles. The model has a high potential to explore a wide range of scenarios of pesticide exposure and risk assessment. For example, it was demonstrated in a study on a hypothetical pesticide that for risk assessment, exposure and animal behavior often are more important than toxicology. Developing high-resolution models like ALMaSS requires a lot of resources, and documenting the model and its analyses is a challenge. The pros and cons of matrix and individual-based models, as well as of differential equation models, are discussed in Chapters 3 and 9.

Results from Group Discussions What Are the Benefits of Using Ecological Modeling Pesticide Risk Assessment Tool?

as a

In general, EMs bring more ecology into ecological risk assessment. They are an excellent tool for exploring the importance of ecological complexities. They allow integrating exposure, effects, and ecology, and can lead to more realistic and thus better risk assessments. With regard to data, the benefits of EMs can be grouped according to 3 steps in risk assessment:

6







Ecological Models for Regulatory Risk Assessments of Pesticides

1) Benefits of EMs before or instead of creating more data. EMs can guide the design of studies (laboratory and field experiments, as well as monitoring studies) and identify the most sensitive species by evaluating which organism or life cycle characteristics increase vulnerability to pesticides. Models can help identify both potentially important (exposure) scenarios that should be tested and which additional data will reduce the uncertainty of the risk assessment the most. Once a well-validated EM exists, it can also be used to identify the right scale, frequency, and duration of postregistration monitoring. In general, an important role of EMs for ecological applications (e.g., population viability analysis) is to provide a comprehensive stock taking of existing data and identify important gaps in these data, which then have to be collected in the field and laboratory. EMs allow comparison and ranking of compounds with different toxicity, application, and fate profiles (e.g., reproductive effects vs. acute mortality). And, last but not least, EMs may save animal testing. 2) Benefits of EMs in analyzing data. With EMs, we can extract more information from data; for example, with matrix models we can extract population growth rate from life table data. We also can use EMs to support the interpretation of complex empirical results, for example, provide a quantitative basis for interpreting lab and mesocosm data (e.g., Lin et al. 2005). For example, Beaudouin et al. (2008) use an individual-based model (IBM) of mosquito fish to virtually increase the number of replicates of their cosm experiments. With EMs, we can provide scientifically based safety factors and thereby increase confidence in these factors. Sensitivity analyses of EMs reveal the processes and parameters that affect risk most strongly. More importantly, however, EMs can aid in ensuring that the protection goal is achieved, for example, where individual organism effects do not result in population-level effects. Moreover, cost–benefit analyses can be performed that could provide metrics for common quantifications of effects and economic impacts. 3) Benefits of EMs in using data. EMs can be used to predict effects at the population level, in particular extrapolation: from survival and sublethal effects (e.g., reproduction, growth, or behavior) to population-level endpoints (e.g., persistence, spatial structure, age distributions); from direct effects on a population to indirect effects on community or ecosystem level (e.g., effects in food chains); from temporally discrete effects to long-term consequences (e.g., multigeneration effects and impact of adaptive mechanisms). Extrapolation is in principle also possible for effects between life cycle traits, for example, from “fast” test species to “slow” ones in the field. Probably the most important extrapolation is to predict the effects between different environmental conditions, in particular regarding climate and habitat type and different exposure scenarios, including risk management and mitigation strategies. Another important issue is to extrapolate from smaller to larger scales, for example, recovery from mesocosm to ecosystem or landscape level.

Executive Summary of the LEMTOX Workshop

7

What Are the Barriers for Ecological Modeling Being Used More Frequently in Pesticide Risk Assessment? There are quite a few challenges if EMs are to be established as a widespread tool for pesticide risk assessment, but the overall consensus was that these indeed are challenges, not insurmountable obstacles. Challenges require work and concerted actions, but there was agreement that the potential to obtain better risk assessments warrants the effort required to overcome the challenges.

1) Challenges before starting the modeling process (problem formulation). For EMs to be successful and rigorous, it is necessary to formulate clear questions that are directly linked to risk management issues, in particular protection goals and assessment endpoints. The appropriate set of species of concern has to be defined, including the question to what degree generic species, that is, hypothetical species that represent species with certain types of ecology and/or life histories, could or should be considered. Appropriate spatial and temporal scales need to be defined, which must be compatible with the scales of regulatory decision making. And, most importantly, the degree of generality, precision, and realism (sensu Levins 1966) desirable in model-based risk assessments needs to be made explicit before the modeling process starts. 2) Challenges for developing EMs. What model type should be chosen for what type of question, and what is the right level of complexity within each of these model types? So far, choice of model type seems to be more or less ad hoc and often represents the preference of the model developer for a certain type of model rather than the very question to be answered with the model (Chapter 9). For pesticide risk assessment and regulations, the choice of model type must be made more explicit, including the selection of state variables and driving processes. The availability of ecological parameters must be addressed in a more systematic way, including reference ranges for different habitats, geographic regions, landscape structures, and multiple stressors. There is also lack of guidance for model analysis (sensitivity analysis, verification, validation; Chapter 9) and model documentation. The lack of guidance is one of the main obstacles to the acceptance of EMs into the regulatory process. Appropriate guidance documents should provide clear checklists and assessment criteria that can be used by both modelers and regulators. Model documentation should include why a specific model type and complexity has been selected; documentation should also include information about model inputs, outputs, assumptions, analysis, interpretation, limitations, and uncertainties. It should be demonstrated that a model is sensitive enough to show adverse effects (positive control). It was also suggested during the workshop that standardized submodels could facilitate the development and use of EMs for regulatory decision making. 3) Challenges for using models in risk assessment. The most important challenge for using EMs in risk assessment is overcoming the general lack of confidence in EMs. Examples and case studies are needed that convincingly

8

Ecological Models for Regulatory Risk Assessments of Pesticides

demonstrate the benefits of EMs, that is, that ecological modeling will lead to better risk management decisions (e.g., knowledge of false positive and false negative error rates). Ecological modeling so far is not included in the curricula of most universities, which is one of the reasons for the widespread lack of knowledge and understanding of EMs: How does it work? Which are the main approaches? What are the potentials and limitations? Due to poor documentation, EMs can lack transparency, leading to the black box syndrome. Besides the constraints in knowledge and understanding, decision makers may simply not have the time to assess every single submitted model from scratch. Better documentation, guidance documents, standardized models, and software would make evaluation of submitted models more efficient and reproducible. In general, EMs for pesticide risk assessments need to explicitly address the issues of regulations, for example, safety factors.

What Will It Take for Ecological Modeling to Become More Widely Used in Pesticide Risk Assessment? The main purpose of the LEMTOX workshop was to agree on actions, that is, the next steps to be taken to make EMs more suitable and useful for pesticide registrations and risk assessment. These steps should tackle the following issues:



1) How to get EMs more widely used. The state of the art of ecological modeling for pesticide risk assessment needs to be improved. The most important means for improvement is to provide guidance for good modeling practice (model development, analysis, documentation, and evaluation). Confidence in model results can then be increased by applying good modeling practices, which possibly could include peer review of EMs for risk assessment. Next, we need to provide case studies that include clear problem formulations, cover different but common problems solved with models of different types, demonstrate the added value of EMs for risk assessment, and include verification and validation. In analogy to the tiered risk assessment approach, a tiered modeling approach could also be developed, where relatively simple models are developed first. And, training of all stakeholders involved is necessary, ranging from longer courses in different modeling techniques to shorter courses introducing certain specific models and evaluation techniques. 2) How can we manage that? As with any new approach to be established, a core stakeholder group has to initiate concerted actions. The most important outcome of the LEMTOX workshop is that such a group representing the different stakeholders emerged from it. This group will try to get funding for the necessary activities: prepare a guidance document for good modeling practice; discuss the need for standardized and/or tiered models; prepare case studies, preferably new ones that follow the good modeling practice guidance and demonstrate that the resulting models are better suited for risk assessment than are existing models; and offer training for stakeholders and wider user groups, that is, introduce good modeling practice via

Executive Summary of the LEMTOX Workshop

9

demonstration of successful case studies at the European scale (workshops, short training courses, Internet repository).

Discussion The most important result of the LEMTOX workshop was the establishment of a core group of stakeholders from industry, academia, and regulatory authorities that will initiate steps to be taken to enable wider use of EMs in risk assessments, in particular for pesticide registration. In this respect, the workshop was successful. This book is the first European volume addressing, in a systematic way, the issue of EMs for regulatory pesticide risk assessment. It summarizes the perspectives of the stakeholders, introduces ecological modeling and different model types, summarizes the benefits and challenges of ecological modeling, and outlines the next steps to be taken. In the final discussion by the workshop participants, it was agreed that in order to get EMs established as part of the regulatory process, concerted actions are needed: Industry needs to know what regulators expect and require; regulators need to better understand the methods, potentials, and limitations of ecological modeling and how they must formulate protection goals so that models can be designed to address these goals; and scientists from the academic world need to better acknowledge the specific requirements of models that are supposed to support decision making. EMs are being used for decision support in other fields, including wildlife epidemiology (Chapter 8), forestry, fishery management, and species conservation. However, in all these fields, it usually takes 10 to 20 years to get the use of EMs established and accepted. In the field of pesticide risk assessment, there is the chance to proceed faster. This can be achieved by learning from the use of EMs in other fields (e.g., Chapter 8), by producing a guidance document for good modeling practice now, and by starting carefully selected and planned case studies that follow good modeling practice and clearly demonstrate the added value of risk assessments based on EMs. In contrast to most other fields where EMs are used to support management decisions, for pesticide risk assessment a very active organization exists that has the declared aim to coordinate the actions of business, academia, and government: the Society of Environmental Toxicology and Chemistry (SETAC). There is thus a good chance that EMs can help produce improved and ecologically relevant risk assessments within the next 5 years. Workshop participants agreed on the following next steps to be taken: Organize a follow-up workshop where a draft guidance document for good modeling practice is produced; write grant proposals to, for example, the European Commission, for joint projects where good modeling practice is applied in a set of case studies that cover high-priority nontarget organisms, scales, and effects; and start bilateral collaborations between industry and academia, where good modeling practice is used to develop models for real pesticide registrations.* *

By February 2009, a SETAC Europe Advisory Group was established (Mechanistic Effect Models for Ecological Risk Assessment of Chemicals [MEMoRISK]) and a proposal for a European joint project was successfully submitted to the European Commission (Marie Curie Initial Training Network CREAM: “Mechanistic Effect Models for Ecological Risk Assessment of Chemicals”).

to the 2 Introduction LEMTOX Workshop Pernille Thorbek In order to obtain registration of plant protection products, extensive risk assessments are carried out to demonstrate that the products’ use will cause no unacceptable effects on nontarget organisms. Currently, ecological risk assessments of plant protection products make extensive use of toxicity exposure ratios (TERs) gained from single-species tests, which are normally performed under laboratory conditions. Such TERs are good measures of the risk posed to individuals (European Commission 1997, 2002a, 2002b), and combined with safety factors (assessment factors), they are highly conservative and therefore suitable for lower-tier risk assessment. However, for most species, the protection goal is not the individual but the population or community (European Commission 2002a, 2002b; Pastorok et al. 2002, 2003; Sibly et al. 2005). Model ecosystem tests aim at measuring effects on populations and communities. Thus, in higher-tier risk assessments, the TERs are often calculated using results from semifield (e.g., aquatic mesocosm studies) or field studies (Campbell et al. 1990). These types of studies do indirectly take some of the factors important for population-level effects into account. However, such studies are time consuming and expensive, so it is not possible to carry out large-scale experiments for all possible scenarios. Furthermore, species composition and population characteristics might differ from those in natural ecosystems. For instance, recovery cannot always be observed in these systems because of the isolated nature of the semifield experiments and the short time span of most studies. Because recovery is an item of increasing importance, there is a need for a tool that can extrapolate recovery patterns from semifield experiments to the ecosystem level. Such a tool would also enable the prediction of recovery patterns of species not always present in test systems, in particular those with a limited number of life cycles per year. Such a tool can also test the ecological significance of laboratory-based findings, for example, population-level impact of sublethal effects such as endocrine disruption. The effects of crop protection products on populations of nontarget organisms depend not only on the exposure and toxicity, but also on factors such as life history characteristics (e.g., dispersal abilities, generation time, fecundity), population structure, density dependence, timing of exposure, landscape structure, community structure, and occurrence of other stressors. Ecological modeling presents an excellent tool whereby the importance and interaction of such factors can be explored and the effects on populations can be predicted. Thus, in cases where the protection goal is populations or communities, ecological modeling has the potential to provide additional ecologically relevant endpoints that can support risk assessments 11

12

Ecological Models for Regulatory Risk Assessments of Pesticides

that would otherwise take only exposure and measured ecotoxicological endpoints into account. In principle, therefore, ecological modeling should be able to play an important role in risk assessments of pesticides. Indeed, ecological modeling has been used to demonstrate how the effects of toxicants interact with life history (Stark and Banken 1999; Stark et al. 2004), landscape structure (Halley et al. 1996; Topping et al. 2003, 2005; Thorbek and Topping 2005), difference in sensitivity among life stages (Meng et al. 2006), timing of application (Thompson et al. 2005), density dependence (Forbes et al. 2001, 2003), endocrine disrupters (Brown et al. 2005), and multiple stressors (Koh et al. 1997). Models are useful for extrapolating from lab or semifield studies to field situations (Naito et al. 2003; Lopes et al. 2005; Van den Brink et al. 2006a) or to predict recovery time (Barnthouse 2004; Van den Brink et al. 2006a). Ecological modeling is also being used successfully to inform ecosystem management in other areas, for example, epidemiological models of livestock diseases (Eisinger and Thulke 2008; Chapter 8) and conservation of endangered species (Burgman et al. 1993; Frank et al. 2002). In principle, therefore, ecological modeling should be able to play an important role in ecological risk assessments of pesticides. However, although there already exists a wide range of models that are suitable for ecological risk assessments (reviewed in Pastorok et al. 2001), the use of ecological models to support pesticide registration has been limited. This is in marked contrast to environmental fate assessments, in which modeling is used routinely to increase the realism, relevance, and robustness of the risk assessments (FOCUS 2001). Environmental fate and effects will of course have to be dealt with in different ways; the level of exposure may be mitigated, but intrinsic toxicity endpoints cannot be mitigated. There also may be differences in the model complexity for environmental fate models and population models. Nonetheless, in the area of environmental fate, it is generally accepted that models are useful for estimating exposure, whereas in the area of ecological modeling, there is no consensus on whether models can improve ecological risk assessments. The aim of the SETAC Europe workshop “Ecological Models in Support of Regulatory Risk Assessments of Pesticides: Developing a Strategy for the Future,” referred to as “LEMTOX,” was thus to explore the reasons for this discrepancy between the potential of ecological models to support decision making and its limited use for pesticide registrations. The focus of the LEMTOX workshop was ecological models such as unstructured population models, stage-structured matrix models, or individual- or agent-based models (IBM, ABM). Although community models, food web models, and pharmacokinetic and toxicodynamic models may also be relevant for ecological risk assessments of pesticides, they were not covered by this workshop. During the LEMTOX workshop, the role of ecological modeling in support of regulatory risk assessments was explored via keynote lectures, breakout sessions, and plenary discussions. Among the participants from academia, industry, and regulatory authorities in Europe, Asia, and the United States, the purpose of the workshop was to discuss and where possible answer the following questions: • What are the benefits of using ecological modeling as a pesticide risk assessment tool?

Introduction to the LEMTOX Workshop

• What barriers for ecological modeling are being used more frequently in pesticide risk assessments? • What actions are needed for ecological modeling to become more widely used in pesticide risk assessments?

13

Introduction to 3 Short Ecological Modeling Volker Grimm Contents Introduction............................................................................................................... 15 What Is a Model?...................................................................................................... 16 Why Modeling?........................................................................................................ 17 How Modeling Works............................................................................................... 17 Types of Population Models..................................................................................... 19 Differential and Difference Equations............................................................. 19 Matrix Models.................................................................................................20 Individual-Based Models................................................................................. 22 Beyond Ad Hoc Design............................................................................................24 Discussion.................................................................................................................25

Ecological modeling has the potential to become a standard tool in higher-tier risk assessments of pesticides. However, most stakeholders involved are not familiar with the rationale and methods of ecological modeling, and as a consequence, they do not know how to assess models and when models are to be trusted. Therefore, I here give a short introduction to ecological modeling. After presenting the 3 main types of population models relevant for pesticide risk assessment, I explain why choice of model type and structure should not be ad hoc, but instead be determined by a detailed specification of the model’s purpose. Finally, I recommend that modelers offer short courses where nonmodelers from industry and regulatory authorities are introduced to ecological modeling in more detail, and where modelers can learn more about the specific requirements of the regulatory process.

Introduction Modeling plays a key role in ecology that together with observation and experiments improves understanding and management of ecological systems. The use of models aims to overcome the limitations of experiments that are a result of the complexity, extent, and slow dynamics of real systems or for ethical reasons. Ecological models of all types are used not only to study fundamental ecological processes but also for practical applications. Examples include models supporting forest management

15

16

Ecological Models for Regulatory Risk Assessments of Pesticides

(Porte and Bartelink 2002; Huth and Tietjen 2007), fishery (Pauly et al. 2000), and biological conservation (Lindenmayer et al. 1995). A developing and important field for ecological modeling is risk assessment of pesticides, in particular in the regulatory context. A main problem, however, is that like in all fields of ecological applications, the managers, decision makers, and stakeholders involved in the registration process are not usually familiar with the scope and methods of ecological modeling. Industry and regulatory authorities currently acknowledge the potential of ecological modeling, but it remains unclear how models are to be assessed and, in particular, under what conditions they could be trusted strongly enough to base regulatory decisions on their outcomes. Therefore, I will give a brief introduction to ecological modeling and explain what a model is, why models are developed, and how the process of modeling works. I will briefly describe the main types of population models that can be useful for risk assessments of pesticides, and how model choice and design can be made less ad hoc in the future. More in-depth introductions to general principles of ecological modeling are given by Starfield et al. (1990), Grimm and Railsback (2005), and Haefner (2005). The role of ecological modeling for ecotoxicological risk assessment is discussed in Pastorok et al. (2001, 2003), Bartell et al. (2003), and Sibly et al. (2005).

What Is a Model? Intuitively, it is clear that a model is some kind of simplified representation of a real thing or system. But the terms “simplification” or “representation” are insufficient to characterize models. The defining feature of a model is that it serves a purpose: “A model is a purposeful representation” (Starfield et al. 1990). The purpose of a model determines how and to what extent we simplify the real system in a model. Starfield et al. (1990) describe the purpose as a kind of “customs officer” who decides whether a certain feature of the real system “passes” to be included in our model. For example, without knowing the purpose of a forest model we would not know how to represent a tree. Should crown architecture and the root system be included, or maybe even the leaves? Once we have clearly formulated the model’s purpose, we can more easily make experimental decisions on model design: for forest management, the trunk and its size or volume are most important, but crown structure and even leaves can probably be represented in a more aggregated way, for example, by considering only the crown’s diameter and the leaves’ total area. If we ask questions other than those relevant for forest management, for example, about species composition or how the forest will respond to changing climate, we might need to represent crowns and leaves in a more detailed way. Why do we want to simplify? The answer is that we have no choice anyway: we cannot know, for example, the detailed crown architecture of all trees in a forest. And even if we would know, we probably wouldn’t care. Including too much detail in a model would make the model unnecessarily complex. The model should be as simple as possible so that we can more easily understand what the model does. The type of model that I am going to discuss in the following is dynamic mechanistic models that describe how and why populations or other ecological systems change in the course of time. This is in contrast to descriptive, or statistical, models

Short Introduction to Ecological Modeling

17

that describe the correlation between certain variables of interest without referring to causal relationships or mechanisms.

Why Modeling? The purpose of a model can be answering a scientific question or solving a certain management problem. Modeling is thus problem solving (Grimm and Railsback 2005). In most cases, the foremost problem to be solved is to identify the key factors that determine the internal organization and dynamics of a system. In this way we develop a better understanding of a system, which allows better solving of management problems. If, for example, we have developed a good understanding of how grazing, rainfall, and recovery of perennial grasses interact in a semiarid system, we can make adaptive farming decisions regarding stocking rates and the spatiotemporal usage of the different parts of a farm (Müller et al. 2007). Or, when we understand how a nontarget population recovers from a pesticide application, we can optimize exposure timing and patterns in order to reduce impact and facilitate recovery. The key word regarding ecological modeling is thus “understanding.” Dynamic models that try to represent real processes but do not produce understanding are black boxes that are not to be trusted for decision support. Once we have sufficiently tested and analyzed our models, we can take advantage of the fact that they are dynamic: we can extend the models’ time horizon into the future and thus project or predict. Projection means that we assume that current conditions extend into the future so that we can project the current situation to 20 or 50 years into the future; for instance, many matrix models make projections of population dynamics without taking factors such as density dependence or time-varying survival into account. Prediction includes future changes of environmental conditions and the population’s response. This can be achieved by mechanistic models, where demographic rates are not imposed but emerge from mechanistic submodels describing individuals and their behaviors. If the submodels were parameterized for a wide range of environmental conditions, the model could be used to predict how the population would respond even to changes in the environment that have not been observed before. For example, when we understand how feeding behavior, the distribution of food, and the dynamics of abiotic factors determine the winter mortality of seabirds like oystercatchers, we can use the model to predict, with striking precision, how the loss of one-third of the birds’ feeding area due to construction will affect winter mortality (GossCustard et al. 2006). But how can we identify the key factors of a system that should be studied in order to gain better understanding, and when can we trust a model enough to base decisions on its projections and predictions? To answer these questions, it is important to know and understand the different tasks of ecological modeling.

How Modeling Works The basic idea of modeling is that we formulate simplifying assumptions about what constitutes a system and how it works. Then we use mathematics and computers to

18

Ecological Models for Regulatory Risk Assessments of Pesticides

Communicate the model

Formulate the question

Assemble hypotheses Analyze the model

Implement the model

Chose model structure

Figure 3.1  Schematic representation of different tasks involved in developing and using models. Models are developed by iteratively performing these tasks, which constitute the “modeling cycle” (modified after Grimm and Railsback 2005).

explore the consequences of our assumptions (Wissel 1992). In virtually all cases, our first assumptions will turn out to be incomplete or simply wrong, so that the model does not appropriately represent the real system. We thus have to revise our assumptions and test the model again. Usually, modeling means to iterate through this cycle of formulating assumptions and exploring the consequences many times, until we decide, according to some criteria, that the model is good enough to serve its purpose. The modeling cycle consists of several tasks (Figure 3.1). Often, we iterate only through parts of the cycle before completing the entire cycle. These tasks are (Grimm and Railsback 2005):





1) Formulate the question. We need to start with a very clear research or management question. Often, formulating a clear and productive question is by itself a major task. 2) Assemble hypotheses for essential processes and structures. We make experimental, or preliminary, assumptions about what the key elements and processes are. These assumptions reflect our current conceptual model of the system that we will often represent graphically. Our conceptual model is based on empirical knowledge, existing theory, and modeling heuristics. 3) Choose scales, state variables, processes, and parameters. We produce a written formulation of the model. Producing and updating this formulation is essential for the entire modeling process, including final publication or delivery to clients (Grimm et al. 2006). With mathematical models, we use equations to formulate the model; with simulation models, we use a mixture of verbal descriptions, pseudocode, model rules, and the equations that are implemented in the computer programs.

Short Introduction to Ecological Modeling





19

4) Implement the model. Here we use the logic of mathematics and computer programs to translate our verbal model description into an “animated” object (Lotka 1925). The implemented model has its own, independent dynamics, driven by the internal logic of the model. Our assumptions may be wrong or incomplete, but the implementation itself is — in an ideal world — always right. 5) Analyze, test, and revise the model. This task, analyzing a model and learning from it, should be the most time consuming and demanding one. We have to make sure that the model is implemented correctly, observe model output, compare it to data, and test how changes in the model assumptions affect the model’s behavior. Finally, we can also try to deduce new predictions for validation: Does the model predict phenomena or patterns that we did not know and use in some way for model development and calibration?

Types of Population Models For pesticide risk assessment, 3 major types of population models can be distinguished (Bartell et al. 2003): difference or differential equations, matrix models, and individual- or agent-based models. Within each type, further distinctions can be made, for example, regarding the inclusion of stochasticity or spatial effects. However, these distinctions are less fundamental than the choice of the model type itself.

Differential and Difference Equations Ordinary differential equations or, when time is described as proceeding in discrete time steps, difference equations are well known from physics, where they are used with great success. A differential equation is defined by including not only a timedependent function but also 1 or more of the function’s derivatives. For example, to determine a function describing the growth of a population, we can set up an equation that describes how N(t), the size of a population at a certain time t, changes in a small time interval, dt. The most simple assumption we can make is that the change of the population’s size, dN(t)/dt, is proportional to N(t) itself:



dN (t ) = r N (t ) dt

This is an ordinary differential equation because it contains a function and its first derivative; “ordinary” means that the only independent variable is time, t. The parameter r is the population’s per capita growth rate and includes the difference of the per capita birth and death rates. The purpose of the equation is to determine the “solution,” N(t), that is, the population dynamics. In this case, the solution is simple to find because we know that only the exponential function is equal to its derivative:

N (t ) = N 0 er t

20

Ecological Models for Regulatory Risk Assessments of Pesticides

where N0 is the population size at time t = 0. Thus, if we assume that the per capita growth, r, is constant, the population dynamics equals exponential growth. Setting up differential equations is relatively easy, but solving them can be more demanding. They are often solved numerically, that is, N(t) is determined by computer simulations. Or, not N(t) itself is considered, but only the stability of the equilibrium solution, that is, dN(t)/dt = 0. This approach has several advantages: the powerful and flexible language of mathematics is used, the model is easy to communicate in this language, and for keeping the equations tractable, the modeler is forced to represent the real system in an extremely simplified way. This makes the model easier to understand and more likely that only key factors are included (but not necessarily all key factors). The main disadvantage is that many aspects of real populations, which can be important in many situations, are ignored: spatial relationships, local interactions, differences among individuals, adaptive behavior, and environmental stochasticity. By ignoring all these factors it is also difficult to test the models. Some of these limitations can be overcome by using sets of differential equations, one for each variable of concern. The equations then include terms describing interactions among the variables, for example, among different species (e.g., Tang et al. 2005) or life history stages. A further limitation is that demographic parameters (birth and death rates) are imposed (Railsback 2001; Grimm and Railsback 2005): they are based on data from certain field studies that were performed under certain environmental conditions. It is not possible to extrapolate these models to new environmental conditions that are outside the range observed in the field studies. In the context of pesticide registrations, single differential equations may be of limited use (e.g., limited to describing only specific, detailed aspects of population dynamics or submodels; see also later) because of their limited potential for verification and validation. Sets of differential equations, however, can be useful. In addition, the potential of differential equations to produce insights into general mechanisms, that is, understanding, can also be important in the regulatory context. For instance, a differential equation can represent a simplified version of a more complex matrix or individual-based simulation model. This simplified version would be easier to understand and could also guide analysis of the corresponding more complex models (e.g., Pagel et al. 2008).

Matrix Models Matrix models are sets of mostly linear difference equations. Each equation describes the dynamics of 1 class of individuals. Matrix models are based on the fundamental observation that demographic rates, that is, fecundity and mortality, are not constant throughout an organism’s life cycle but depend on age, developmental stage, or size. Ecological interactions, natural disturbances, or pesticide applications usually will affect different classes of individuals in a different way, which can have important implications for population dynamics and risk. In the following, I will only consider age-structured models, but the rationale of the other types of matrix models is the same. For an example of this approach applied to pesticide risk assessment, see Stark (Chapter 5).

21

Short Introduction to Ecological Modeling

In an age-structured matrix model, the state of a population at a certain time t is described by a vector consisting of m elements, each of the elements describing the number of individuals in one of m age classes; that is, n1 is the number of individuals in age class 1, and so forth:  n1     n   2  N (t ) =  ..     ..    nm  N (t ) The new state of the population after 1 time step, which often equals 1 year for population dynamics, is calculated in the following way:



 n1   F1     n  S  2   2  0  ..  =     ..   0    nm   0 N (t +1)

F2 0 S3 0 0

.. 0 0 .. 0

.. 0 0 0 Sm

Fm   n1     0   n2     0 × ..     0   ..     0  nm 



(3.1)

N (t )

where F and S are the age-specific fecundities and survival rates. The symbol × denotes the multiplication of a vector with a matrix: n1 at time t + 1 is calculated as the sum of all offspring produced in year t. To calculate this, the fecundity F of each age class is multiplied with the number of individuals in that age class:

n1,t +1 = F1n1,t + F2 n2,t +  + Fm nm ,t

The number ni at time t + 1 of all other age classes is calculated by multiplying the number of individuals in the preceding age class, ni–1, by the survival rate of those individuals into the next age class, Si, for example: n2.t +1 = S2 n1 . The other age classes of the population vector cannot contribute to n2. Thus, the transition or Leslie matrix in Equation 3.1 has a typical structure: age-dependent fecundities in the first row and age-dependent survival rates in the subdiagonal (Caswell 2001; Chapter 5). It can be shown that the population’s long-term growth rate only depends on the transition matrix, not on the initial state of the population. If all elements of the matrix are constant, the growth rate can be calculated as the first eigenvalue of the matrix (i.e., the constant that is obtained by solving Equation 3.1). If restricted to constant demographic rates, matrix models do not require any modeling at all, because

22

Ecological Models for Regulatory Risk Assessments of Pesticides

the model structure is already given by Equation 3.1. Rather, the demographic parameters have to be determined in the field or laboratory. The elements of the matrix can be made dependent on population density or the environment, but then there is no longer a simple solution via eigenvalues. Instead, Equation 3.1 has to be iterated on computers and resembles simple individual-based models (IBMs; see next section). The advantages of matrix models include their clear structure, which makes communication simple; data requirements are clear; population growth rates can easily be calculated; it is easy to link observed field or lab individual-level data to population-level effects; and the sensitivity of different age classes to changes in their demographic parameters can easily be checked (Caswell 2001). The main disadvantage lies in the assumption of constant demographic rates, which is unrealistic (Chapter 5). The elements of the matrix represent average values observed under certain conditions. The model projects the consequences of these conditions into the future. For example, we can ask: What if every year 4% of the second and third age class are lost due to pesticide application? Does the population still have a positive growth rate? Simple matrix models are not validated because their purpose is projection, not prediction. Even verification usually is not addressed; rather, demographic rates are extracted from data and plugged into the transition matrix to project mainly population growth. If the purpose of a model-based risk assessment is projection rather than prediction, matrix models are an important tool. As with differential equations, understanding of the effects of pesticides can be obtained more easily from matrix models than from simulation models. Therefore, it can be useful to simplify an existing individual-based simulation model to a matrix model. This can be achieved by measuring average demographic rates in the simulation model, and using them as input for the matrix model (Topping et al. 2005; Pagel et al. 2008). However, in case a model has to be predictive, matrix models will usually not be appropriate because they are difficult to test (validate).

Individual-Based Models Individual-based models describe the life cycle of individual (discrete) organisms. The organisms can differ and display autonomous behavior (DeAngelis and Mooij 2005; Grimm and Railsback 2005). The entities of an IBM — individuals, habitat units, and the abiotic environment — are characterized by sets of state variables, for example, sex, age, body mass, location (individuals); vegetation cover, soil moisture, food level (habitat units); or temperature, rainfall, and disturbance rate (environment). In contrast to the previous 2 model types, IBMs have to be implemented as computer programs. In the past, this made IBMs hard to communicate and, as a result, understand. Recently, however, a common protocol for describing IBMs was proposed (Grimm et al. 2006; see Van den Brink et al. 2007 for an example application), the ODD protocol (Overview–Design concepts–Detail). ODD provides a common structure for IBM descriptions, but also helps us to think about IBMs in a structured way. For example, developers of IBMs have to make the following decisions, which correspond to the 7 elements of ODD (see also the tasks of the “modeling cycle”):

Short Introduction to Ecological Modeling





23

1) Purpose: What is the purpose of the model? 2) State variables and scales: What are the entities and their state variables? What spatial and temporal resolution and extent does the model have? 3) Process overview and scheduling: What processes are represented in the model and how are they scheduled? What entity is doing what at which point in time and in what sequence? 4) Design concepts: What are the concepts underlying the model’s design? For example, how does the model take into account, if at all, emergence, fitness, adaptation, stochasticity, and observation? For details, see Grimm and Railsback (2005). 5) Initialization: What is the state of the model’s entities at the beginning of the simulations? 6) Input: Is there any external input driving environmental variables, for example, time series of annual average rainfall that is taken from external data files? 7) Submodels: In detail, how are the model’s processes implemented? For example, how is an individual’s growth linked to temperature and food availability?

In social sciences, similar models have long been used, namely, agent-based models (ABMs; Gilbert and Troitzsch 2005), which have a stronger focus on an organism’s or agent’s behavior and decision making than IBMs (see Topping et al. 2009, Chapter 7, for an example ABM, and Van den Brink and Baveco, Chapter 6, for an example IBM). IBMs often include imposed demographic rates, whereas in ABMs demographic rates often emerge from the individual’s behavioral decisions. In the last 10 years, however, the differences between IBMs and ABMs are fading away in ecology, so that they should be treated as one big class of models that can be referred to either way (Bousquet and LePage 2004; Grimm 2008). The main advantage of IBMs is that they are the most flexible approach. All kinds of factors that are possibly important for the purpose of the model can easily be included in IBMs: individual variability, development, local interactions, and adaptive behavior. Thus, potential key factors are not sacrificed for mathematical convenience. IBMs can also be surprisingly predictive, if they were designed for this purpose. Examples include the trout models of Railsback and coworkers (Railsback and Harvey 2002), the shorebird models of Goss-Custard, Stillman, and co-workers (Goss-Custard et al. 2006; Stillman and Simmons 2006), and applications of the ALMaSS framework for ABMs (see Chapter 7, and the literature cited there). The reason for this predictive capacity of IBMs is that demographic rates are not imposed but emerge from the organisms’ behavioral responses to changes in their environment. If the submodels, for example, regarding feeding and metabolism, are parameterized and tested for a range of environmental conditions that is wider than observed in real populations, the IBM can be used to predict how the population responds to environmental conditions that have not been observed in reality so far. The disadvantages of IBMs are closely linked to their complexity: developing and testing IBMs can be very time consuming and require vast amounts of data and empirical knowledge. With the trout, shorebird, and ALMaSS models, it took years until they could be used and validated for the first time (subsequent applications

24

Ecological Models for Regulatory Risk Assessments of Pesticides

require less time, though). Setting up such full-fledged IBMs should thus be compared to setting up a virtual laboratory, which is also time consuming and expensive, but once it is there, all kinds of simulation experiments (Peck 2004) can be performed. The majority of IBMs, however, are much simpler. Regarding potentials and limitations, they are somewhere in the middle between full-fledged ABMs and matrix models (e.g., Van den Brink et al. 2007; Wang and Grimm 2007; Chapter 6). General problems with IBMs are that often they are not fully understood, they are poorly analyzed, and their uncertainty regarding parameter values and model structure remains unexplored. Developers of IBMs often seem to be stuck in the “trap of realism”: in trying to make their model realistic, which is laudable in principle, they avoid analyzing unrealistic scenarios (Grimm 1999), for example, constant or homogeneous environments or nonadaptive behavior. Such scenarios are, however, decisive for understanding the significance of, for example, environmental heterogeneity or adaptive behavior. IBMs are very likely to play an important role for pesticide risk assessment because of the high potential of being realistic, flexible, testable, and predictive. However, the issues with required resources, communication, testing, analysis, and clearly defining their purposes in risk assessment have to be resolved. In order to economize the resources of both model developers and the regulatory authorities who will have to assess them, it might be worth considering agreeing submodels or parameter values for focal species.

Beyond Ad Hoc Design Having the 3 main model types in mind, it is important to realize that the earlier description of the modeling cycle is idealized in one important way: ideally, tasks 2 and 3 — the choice of model type and structure — should be driven only by the model’s purpose and our current understanding of the system. In practice, however, this is not always the case. The choice of the model type is often to a large degree also influenced by the personal background of the modelers, that is, their education, experience, specific skills, and personal preferences. In science, this is not necessarily problematic because the modeler will reformulate the original purpose of the model in such a way that the preferred model type can be used. For ecosystem management, however, we might end up with completely different model types that cannot easily be compared and, in the worst case, might not lead to the same management recommendations. For ecological applications like pesticide risk assessment, we have to find ways to avoid ad hoc choices of model types and ad hoc design of the models themselves. As for the choice of model type, it is important that managers understand the strength and weaknesses of different model types so that they formulate the purpose of the model as precisely and detailed as possible. Part of the specification of the model’s purpose is also a detailed specification of the acceptance criteria of the model (Bart 1995): What kind of evidence is required to show that a model has been verified and validated? As for the design of the models themselves, this relates mainly to IBMs because matrix and mathematical models have much fewer degrees of freedom in their structure. With IBMs, it has been observed that model design often is ad hoc, rather than

Short Introduction to Ecological Modeling

25

being based on established submodels or designs (Grimm 1999; Railsback 2001). Grimm and Railsback (2005) formulated a framework, consisting of design concepts, pattern-oriented modeling, and a theory–development cycle that could make model design more systematic. In any case, a main acceptance criterion for ecological models in risk assessment is a clear and systematic documentation of their structure, design, and output (Grimm et al. 2006).

Discussion If ecological modeling is to become a standard tool for risk assessments of pesticides, all stakeholders involved need to understand what models are and how they are developed. It is important to realize that models by definition ignore many, if not most, aspects of real systems in order to identify key factors of the system’s internal organization. Models are based on assumptions that are implemented, tested, and revised. Model development is iterative and ends when certain acceptance criteria are fulfilled. To assess or evaluate a model, we need to know its purpose. In risk assessment, verification and validation of models are more important than, for example, in theoretical ecology, where often models are accepted that even cannot be tested at all. Verification means to prove that the model is an appropriate representation of the real system in question. This is usually proven by comparing model output to observed patterns and data. But still, the model may reproduce the right patterns for the wrong reasons, for example, by searching for the “right” parameter values long enough (calibration). Validation refers to model predictions regarding aspects of the real system that were not used for model construction or calibration; ideally, they were even unknown while the model was developed. If these independent or secondary predictions are correct, we have achieved the strongest evidence we can have that the model is “structurally realistic” (Wiegand et al. 2003; Grimm and Railsback 2005). For example, the beech forest model BEFORE (Rademacher et al. 2004) contained information about the age structure of canopy trees, but this information was never used, or even observed during model development, calibration, and testing. But then, it turned out that neighboring canopy trees typically differ in age by 60 years. This secondary prediction could be verified by rescanning the literature about old-growth beech forests (Rademacher et al. 2001). This short introduction is, of course, not sufficient to make stakeholders in pesticide risk assessment familiar enough with ecological modeling, but it aims to provide a first overview. To establish ecological modeling in risk assessment, some more time has to be invested by both nonmodelers and modelers. I have been involved in teaching ecological modeling in classes lasting for 1 or 2 weeks, after which the participants had a very good idea of what modeling is and how it works. Modelers interested in risk assessment thus should offer such courses to people from industry and regulatory authorities who consider using ecological modeling and models. In this way, confidence in ecological modeling will increase, leading to first applications of models for real risk assessments, so that eventually enough momentum will build up to get modeling established in this context. In this process, standard models will be developed for certain species, and guidelines for modeling in the regulatory

26

Ecological Models for Regulatory Risk Assessments of Pesticides

context will be formulated (good modeling practice). My personal and, I think, not too optimistic prediction is that all this will happen within the next 5 years.

Challenges 4 Regulatory for the Potential Use of Ecological Models in Risk Assessments of Plant Protection Products Jörn Wogram Contents Regulatory Framework: Do Models Fit In?.............................................................. 27 Regulatory Challenges..............................................................................................28 Past Experiences with Population Models in Regulatory Risk Assessment and Lessons Learned.............................................................................................. 30

The current regulatory framework is considered compatible with a future implementation of ecological and population models into the environmental risk assessment of plant protection products. The need for interpretation and extrapolation tools is most evident in the context of studies of aquatic model ecosystems, field studies of terrestrial arthropods, and studies of birds and mammals. Ecological sensitivities vary among species; that is, a species’ ecology determines how sensitive populations are to exposure of plant protection products. Some ecological models submitted in the past were rejected mainly because they were not considered to be representative of the variability of ecological sensitivities found among species. In order to ensure a better acceptance of ecological models in future risk assessments, starting with a proper definition of the regulatory question should be considered an integral part of good modeling practice.

Regulatory Framework: Do Models Fit In? According to the principles of decision making defined in EU Directive 91/414 EEC, Annex VI C 2.5.2.2 (European Commission 2002a), a plant protection product (PPP) failing the acceptability criteria in a standard risk assessment will not be authorized “unless it is clearly established through an appropriate risk assessment that 27

28

Ecological Models for Regulatory Risk Assessments of Pesticides

under field conditions no unacceptable impact on the viability of exposed species […] occurs.” This text passage is the basis of the so-called tiered approach in regulatory risk assessments and enables the submitter to apply nonstandard methods that are tailored to the specific regulatory question of concern. Characteristically, the directive does not define or restrict in any way the variety of methods that can be used. Actually, the regulatory authorities will evaluate any submitted information in a higher-tier risk assessment as long as it is deemed relevant for the regulatory question and complies with the current state of the scientific and technical knowledge. In principle, this freedom to choose a scientific approach when planning and performing a higher-tier risk assessment provides an excellent basis for any scientific innovation in the regulatory context — including the development and implementation of ecological modeling. The knowledge of the key characteristics of potentially exposed ecosystems and coenoses is generally better for risk assessments of PPPs than for other classes of chemicals, for example, industrial chemicals or human pharmaceuticals. The reason for this is that in order to gain authorization of a PPP, the conditions of use need to be defined precisely; this includes giving information on the time frame, location (type of crop), and amount of a pesticide to be applied. Consequently, it is possible to focus the risk assessment on specific situations and ecosystems that constitute a worst-case situation in terms of exposure (e.g., edge-of-field water bodies, field margins, bird species with a high preference for the crop in question). These definitions of worst-case scenarios enable risk assessors to define the key factors driving the risk caused by exposure and the resulting effects; this knowledge could obviously be used in a model-based prediction of exposure and effects. In fact, the use of models in exposure assessment is well established and worked out in detail. For instance, in the context of the EU active substances program, the FOCUSsurface water model package is routinely used to predict a realistic worst-case exposure in edge-of-field water bodies (FOCUS 2001). In contrast to this, models predicting ecological effects have so far hardly been used for decision making in regulatory risk assessment.

Regulatory Challenges As has been shown by Streissl (Chapter 11), this cannot be explained by a lack of regulatory problems, for the risk of pesticides to nontarget organisms was often not fully addressed in the context of the active substances program. In fact, the limits of a risk assessment relying on ecotoxicological testing alone are reached when the expected specific characteristics of a predicted exposure scenario or a specific population dynamic cannot be simulated by experiments such as simulated ecosystem studies. This may be the case if the voltinism (i.e., the number of broods or generations in a year) of the test species differs from vulnerable species expected in the field, or if the exposure profile is thought to be different under field conditions. Consequently, in higher-tier risk assessments a key challenge is to discuss the available data in the light of their representativeness for the real world (Liess et al.

Regulatory Challenges for the Potential Use of Ecological Models

29

2005). The following regulatory questions, which usually remain unsolved in national application procedures of PPP in Germany, reflect this challenge quite well:

1) Extrapolation of mesocosm results to natural ecosystems. Generation time is one of the most important determinants of population recovery potential (Barnthouse 2004). Consequentially, the OECD Guidance Document on Simulated Lentic Freshwater Field Tests states that “extrapolations have to be made when transferring the results of a microcosm/mesocosm study to … ecosystem components that have not been tested (e.g., recovery of semior univoltine species)” (OECD 2006). The relevance of this demand for extrapolations has recently been shown exemplarily in an ecological characterization of streams, ditches, and ponds in agricultural landscapes in some European countries (Brock et al., forthcoming). The characterization indicated that univoltine life cycles are most common among aquatic invertebrates, and even semivoltine species are quite frequent. Even though it has been shown that the toxicological sensitivity of organisms does not depend on the length of their life cycle, it was concluded that species-specific characteristics such as the number of generations per year and dispersal abilities should not be ignored when considering recovery in risk assessments (Brock et al. in press). Typical mesocosm findings that would require extrapolations are • Recovery of populations was demonstrated, but voltinism of the taxa present in the study was not deemed representative for vulnerable species present in the field. • Recovery was demonstrated but the exposure design was not deemed representative for intended use (e.g., single application tested but multiple applications applied; or early application tested but late application applied). • Recovery was not demonstrated within the time frame of the study (especially when univoltine species are tested, e.g., Amphipods). Up to now, no generally accepted methods are available to enable such extrapolations. 2) Extrapolation of arthropod in-field studies to off-field situations. According to the Guidance Document on Terrestrial Ecotoxicology (European Commission 2002b) for higher-tier risk assessments of terrestrial invertebrates, field studies are the “ultima ratio.” Generally, such studies are performed on test plots located in the crop in question. Due to the overlap of test plot and intended application area, field tests are generally considered to be suitably representative for the ecology of relevant in-field communities. In contrast to that, the use of in-field tests in risk assessments for offcrop habitats (European Commission 2002b) is restricted by the relatively unstable character of cropped habitats (ploughing, sum of pesticide application). Such conditions promote communities with a high proportion of r-strategists, that is, species with high reproduction rates and short life cycles (Begon et al. 1990) as well as species with good dispersal ability (e.g., hover flies). Off-field habitats like meadows or hedges constitute more

30



Ecological Models for Regulatory Risk Assessments of Pesticides

stable conditions and contain less mobile species and species with longer life cycles (Roß-Nickoll et al. 2004). Consequently, in order to use in-field studies for risk assessments of off-crop habitats extrapolations are needed. These extrapolations are similar to those described earlier for extrapolations from mesocosms to ecosystems. 3) Evaluation of the ecological relevance of effects observed in toxicological tests of birds and mammals. In contrast to the exposure assessments for aquatic ecosystems and for terrestrial plants and arthropods, the exposure assessments for birds and mammals include some ecological considerations: According to the Guidance Document on Risk Assessment for Birds and Mammals (European Commission 2002c) exposure potential depends on the behavior of the species and is taken into account by including food intake rate, habitat preference, and food type preferences. However, the current guidance document does not provide methods to assess the population level relevance of risk measured at the level of individual, for example, partial loss of fertility. For further details, see Streissl (Chapter 11).

Past Experiences with Population Models in Regulatory Risk Assessment and Lessons Learned The aforementioned examples illustrate the need for tools to bridge the gap between test systems and natural ecosystems in risk assessments, and suggest that population models may be useful. In the Guidance Document of the Scientific Panel on Plant Protection Products and Their Residues for the Risk Assessment for Birds and Mammals (EFSA 2007), the potential use of models in future risk assessment is discussed. In the past 10 years, several models have been submitted as part of dossier submissions with risk assessments of aquatic and terrestrial invertebrates. Although there have been some submissions of invertebrate population models in the context of national application procedures of PPPs in Germany, I do not have knowledge of any acceptance based on an ecological model in regulatory decision making. The aim of the models submitted to the Federal Environment Agency (UBA, regulatory authority in Germany) was to predict population recovery from effects measured in single species test or in mesocosm studies, and the models were primarily individual based (IBM; Grimm and Railsback 2005). In all cases, the model organism was the same aquatic or terrestrial invertebrate species that had turned out to be most toxicologically sensitive to the insecticide in question. The species modeled were the ladybird Coccinella septempunctata, the water flea Daphnia magna, the amphipod Gammarus pulex, and the phantom midge Chaoborus crystallinus (confidential studies, not published). The C. crystallinus model, based on an IBM by Strauss et al. (unpublished manuscript), was able to simulate the population dynamics in isolated and connected test systems with sufficient accuracy, and its predictions in terms of the population recovery were considered plausible and reliable. Why have none of those models been accepted by the UBA? The answer is surprisingly simple: in all cases the life cycle traits of the modeled species were not considered to reflect a realistic worst case for natural ecosystems in agricultural

Regulatory Challenges for the Potential Use of Ecological Models

31

landscapes. In fact, my colleagues and I had the impression that the submission of those models was due to a misunderstanding of the surrogate species principle in risk assessment. Surrogate species are meant to represent the protection aim, that is, the populations of species in potentially exposed ecosystems, in terms of their toxicological (physiological) sensitivity. Their ecological traits, like generation time or dispersal ability, however, might be totally different from that of vulnerable species in the field. Notably, this applies to typical laboratory test species like Cladocerans, which have been chosen because of their short life cycle and ease of culture. Consequently, a population model simulating the recovery of a toxicologically sensitive species might fail to result in a protective regulatory decision if the ecological sensitivity of the species is rather low (e.g., if the reproductive or dispersal potential of the modeled species is higher than for the majority of species that should be protected). In France, some models have also been submitted as part of regulatory risk assessments (Patrice Carpentier, AFSSA, and Anne Alix, AFSSA, personal communication). One model was considered as species, scenario, and ecotoxicological inputs were deemed relevant. One model was not considered because it did not make use of the relevant ecotoxicological endpoint. The lessons learned from that are just as simple as the main reason for the rejection of models in the past: it should be considered an integral part of good modeling practice to start with a proper definition of the regulatory question. As regards the combination of ecological and toxicological vulnerability, the model species should reflect a realistic worst case.* In combination with a proper validation of models, this will probably lead to a better rate of acceptance of models submitted in the context of environmental risk assessment — although not in either case to a more pleasant result with regard to the level of the predicted risk.

*

The vulnerability depending on the combination of several different biological traits; it may be necessary to run models with different species in parallel.

and Use of 5 Development Matrix Population Models for Estimation of Toxicant Effects in Ecological Risk Assessment John D. Stark Contents Introduction...............................................................................................................34 Risk Quotient Method of Risk Assessment and Its Limitations...............................34 Can Population Models Improve Our Ability to Estimate Risks of Chemical Exposure to Populations?................................................................................ 35 Why Use Matrix Models over Other Models to Determine the Effects of Toxicants on Populations?............................................................................... 36 Matrix Models........................................................................................................... 36 Deterministic Matrix Models.................................................................................... 38 Stochastic Matrix Models......................................................................................... 38 Advantages of Matrix Models over Other Models................................................... 39 Disadvantages of Matrix Models Compared to Other Models................................. 39 Rat–Elephant Phenomenon or Why the Use of Surrogate Species May Be a Bad Idea...........................................................................................................40 Example of Population Modeling with 3 Insect Species Exposed to a Pesticide..... 42 Conclusions............................................................................................................... 43

Population models based on the Leslie matrix have a long history of use by ecologists and are one of the modeling approaches that have potential for use in estimating effects of toxicants on populations of organisms. Matrix models are easy to construct and interpret and, as such, can be quite valuable for estimating impacts of chemicals on populations of species we seek to protect. Because matrix models can take into account lethal and sublethal effects and differences in life history parameters among species, they can be effective in providing guidance for management of threatened and endangered species. 33

34

Ecological Models for Regulatory Risk Assessments of Pesticides

Introduction Traditionally, estimation of the ecological impacts of chemical toxicants has been done by comparing individual-level endpoints of toxicity to an expected, or range of expected, environmental concentrations (Suter 1993; Klaine et al. 1996; Solomon et al. 1996; Giesy et al. 1999). However, the ability of this approach to protect threatened and endangered species has been questioned (Stark et al. 2004), and the use of population modeling has been suggested as a more feasible means to ascertain the fate of populations under stress (Nacci et al. 2005). Modeling is a process of building simple, abstract representations, usually based on mathematical equations and/or computer logics, of complex systems to gain knowledge of how a system, such as a population or community, works. Models can be used to predict the way systems might behave in the future, and can be used to guide the decision-making process where populations need to be managed (Akçakaya et al. 2008). A shift from studying effects of toxicants on individuals to populations is occurring with increasing emphasis on population-level assessments by regulatory agencies such as the US Environmental Protection Agency (USEPA) (Applied Biomathematics and Woodlot Alternatives 2003). However, there are relatively few examples of population models being used in ecological risk assessment of pesticides. The reason that models have not been widely adopted for this purpose is that there is no consensus on which models to use, what data are necessary to develop for use in models, what time frame to use for data collection, which organisms to evaluate, or even which life stages of an organism to evaluate. In this chapter, an overview of the development and use of matrix population models for estimation of toxicant effects is presented. Matrix models are one of the modeling approaches that can be used to determine whether populations of organisms will remain the same, increase, or decline, and thus have great potential for use in ecological risk assessment.

Risk Quotient Method of Risk Assessment and Its Limitations Before we get into the workings of matrix models, it is important to understand how risk or hazard assessment of chemicals, for example, pesticides, in some countries is conducted. The simplest way to evaluate whether a chemical may be hazardous to an ecosystem is through the use of risk quotients. Risk quotients involve comparing susceptibility of a species to potential exposure. Risk quotients are developed by dividing an expected environmental concentration (EEC) by a toxicity endpoint such as the acute LC50 to determine whether a particular toxicant may cause damage to an organism or community. Expected environmental concentrations can be determined in a number of ways. For example, the USEPA uses a modeling approach, while other methods involve actual measurements of chemical concentrations in specific ecosystems (USEPA 2007a). In Europe the approach is essentially the same, but the quotient is reversed. Thus, toxicity exposure ratios (TERs) are calculated

Development and Use of Matrix Population Models

35

by dividing the toxicity endpoint by the EEC (EU Directive 91/414 EEC, European Commission 2002a). An example of the development of a risk quotient is illustrated by considering that the EEC for a pesticide in surface water is 10 µg/L and the LC50 for trout is 5 µg/L. To develop a risk quotient, you simply divide the EEC by the LC50, which in this case results in a hazard ratio of 2 (Europe makes use of TER, where toxicity is divided by exposure; the TER for the above example is 0.5). Quotients greater than 1 indicate that the pesticide poses a hazard to the species in question. There is no probability associated with this type of assessment, and the endpoint of effect is a point estimate (LC50), where in actuality the LC50 is a number that falls within a range of numbers, the 95% confidence limits, and has an associated slope. Risk quotients can also be developed with chronic toxicity data such as the no observable effect concentration (NOEC) for reproduction. However, the concept of the NOEC has been criticized (Laskowski 1995). The USEPA uses levels of concern (LOCs) for risk quotients for pesticides. LOCs are compared to the quotient to determine whether a particular chemical may pose a hazard to certain species. LOCs were developed to take into account uncertainty in the risk assessment process. Additionally, LOCs differ depending upon the type of pesticide being evaluated, the type of toxicity data available (acute LC50 versus chronic NOEC), and whether the species are terrestrial, aquatic, or threatened and endangered (USEPA 2007b).

Can Population Models Improve Our Ability to Estimate Risks of Chemical Exposure to Populations? An important question is: Can population modeling tell us anything different from the quotient method with individual endpoints of toxicity? Before we try to answer this question, another question must be asked: Does the risk quotient method work for the protection of populations? It seems that if a population suffers only acute mortality, then the quotient method should work. However, exposure to toxicants can result not only in mortality but also in multiple sublethal effects. Additionally, effects on populations can different greatly from effects in individuals (Stark 2005). A comparison of risk quotients for several chemicals and species to population-level effects showed that the quotients using acute mortality and an EEC work well for some species– chemical combinations but not for others (Stark unpublished). Furthermore, the same level of mortality in 2 species may result in very different outcomes due to differences in life history strategies (Stark et al. 2004), and thus even a simple measure of mortality among species may not provide enough information to protect a population. The answer to the question — Can population modeling tell us anything different from the quotient method with individual endpoints of toxicity? — is yes. The reason that modeling tells us more about population viability is that modeling can be used to determine the probability that a population will become extinct, whether it will recover or remain the same. Clearly this cannot be accomplished with the quotient method.

36

Ecological Models for Regulatory Risk Assessments of Pesticides

Why Use Matrix Models over Other Models to Determine the Effects of Toxicants on Populations? There are many types of models that can be used to estimate the effects of toxicants such as pesticides on populations (Grimm et al., Chapter 9). However, matrix models are easy to understand and use, and they have a long history of use by conservation biologists and applied ecologists (Caswell 2001; Morris and Doak 2002).

Matrix Models Populations of most organisms consist of different ages and stages. Therefore, individuals within a population are not the same with respect to their contribution to population growth at a specific time because each individual will have a particular probability of dying and reproducing, and thus have a different influence on population dynamics. The most commonly used age- or stage-based population models are based on the Leslie matrix (Leslie 1945). With the Leslie matrix, the population contains n ages or stages and is described by a column vector, as shown next. I present a hypothetical insect in this example, which has an egg, a nymph, and an adult stage. Furthermore, all individuals in this example are females. Vector (at time t) Eggs Nymphs Adults

10 5 1

This initial vector is a stage-structured population and shows that the population at time t consists of 3 stages — eggs, nymphs, and adults — and the total number of individuals in the population is 16. For each stage, 2 demographic parameters or vital rates are needed for the model. The first is the survival rate, which indicates the probability that an individual in stage class x will be alive 1 time step into the future. This is also known as the transition rate because an individual can theoretically remain in a stage or age class and survive. The next vital rate is fecundity, and it represents the number of offspring produced per individual in a stage of age class. Survival and fecundity can be organized into a matrix, M, which is graphically represented as f0 p0 0 0 0 0

f1 0 p1 0 0 0

f2 0 0 p2 0 0

… … … … … 0

fx–1 0 0 0 0 px–2

fx 0 0 0 0 0

where fx and px are fecundity and survival rates for class or age x, respectively.

37

Development and Use of Matrix Population Models

In the matrix presented next, the survival rates show that 50% mortality occurs from the egg to the larval stage (0.5), reproduction only occurs in the adult stage, and 0 eggs were laid. To calculate the abundance of the population at the next time step (t + 1), the vector is multiplied against the matrix as follows: Vector (time t)

Matrix (M)

Vector (time t + 1)

10

0

0

10

5 × 1 Abundance (t)

1 0

0 0.5

0 0

10(0) =

10(1) 10(0)

5(0) +

5(0) 5(0.5)

1(10) +

1(0) 1(0)

10 =

16

10 2.5 Abundance (t + 1) 22.5

As this example illustrates, the population starts with 16 individuals and after 1 time step (1 week) consists of 22.5 individuals. Obviously, you can’t have 0.5 individual, and this number will be rounded off as the multiplications continue. Therefore, the Leslie matrix model projects population abundance and population structure into the future based on starting conditions. In Figure 5.1 we see results of the matrix model after 52 multiplications. For a matrix model to be valid, enough matrix multiplications must be done for the stable age distribution to be reached. This occurs when the growth rate (λ) stabilizes. Some modeling programs, such as RAMAS GIS (Applied Biomathematics, Setauket, New York), precalculate the stable age distribution and the starting vector is in the stable age distribution. Therefore, the data necessary to develop a simple Leslie matrix model are a measure of survival and fecundity. These data can be developed in the field through sampling of wild populations or in the laboratory with life table experiments.

Number of Individuals

2e+5

2e+5

1e+5

5e+4

0

0

20

40

60

Time (weeks)

Figure 5.1  Deterministic population projection of a hypothetical species.

38

Ecological Models for Regulatory Risk Assessments of Pesticides

Deterministic Matrix Models The simple matrix model depicted earlier is an example of a deterministic matrix model. Deterministic matrix models have no measure of randomness (stochasticity), assume constant demographic parameters, and ignore density dependence. Moreover, estimated growth rate and stable stage or age structure refer to an exponentially increasing (or decreasing) population. Deterministic models can be used to compare stressed (e.g., toxicant exposed) and unstressed populations. Stressed populations can be compared to unstressed populations (controls) by comparing the time it takes the control population to reach a predetermined number of individuals to the time it takes the stressed population to reach the same number of individuals (Wennergren and Stark 2000; Stark et al. 2004, 2007). This recovery time interval can then be compared to generation time to look for levels of stress that cause a delay of 1 or more generation time intervals. Stark et al. (2007) stipulate that if a population is delayed for >1 generation time interval, this population will be impaired. This is particularly true for species that are univoltine or have long life cycles. Thus, even if a population is not headed toward extinction (negative growth rate) it still may be impaired. If a population of a species is an important food source for small fish, for example, long recovery times would result in lower food supplies and therefore may have a negative impact on aquatic food webs.

Stochastic Matrix Models Stochasticity is randomness or unpredictability, and thus a stochastic model is one in which randomness is incorporated within the model. There are several types of stochasticity that can be incorporated into matrix population models (Akçakaya 2005). Environmental stochasticity reflects the fact that environmental conditions change and are not constant through time (Meng et al. 2006). Fluctuations in environmental conditions can have a direct effect on survival and fecundity of individuals within a population. If sample sizes are small, demographic stochasticity can be important (Akçakaya 2005). Catastrophes are extreme adverse events, such as a drought, while bonanzas are particularly good years in terms of resources (Morris and Doak 2002). Both catastrophes and bonanzas are infrequent events but should be considered in population models when appropriate. Sampling error stochasticity has to do with errors in sampling or variability among individuals who sample populations. All of these factors can lead to errors in population models. Some modeling programs such as RAMAS give the user the ability to assign levels and types of stochasticity within their models. The advantage of stochastic matrix models over deterministic models is that stochastic models can give you probabilities of extinction, risk of decline, and probability of recovery. Furthermore, they are more realistic than deterministic models because factors such as carrying capacities, competition, and immigration can also be incorporated. In Figure 5.2, the same modeling data presented in Figure 5.1 are modeled using demographic stochasticity with scramble competition (see Akçakaya 2005 for definition) and a carrying capacity. In this example, we see the population abundance over

39

Development and Use of Matrix Population Models

Number of Individuals

2e+5 2e+5

1e+5

5e+4

0

0

20

40 Time (weeks)

60

Figure 5.2  Stochastic population projection for a hypothetical species; average population abundance over time (dotted line) and associated 95% confidence limits (solid and dashed lines).

time and an associated 95% confidence limit. This means that the population lies between the 2 outer bounds 95% of the time.

Advantages of Matrix Models over Other Models Simple constant–parameter matrix models are well suited to link laboratory test data to population level effects (Kuhn et al. 2000, 2001, 2002; Lin 2005) for the evaluation of field data and are easy to understand and explain (Caswell 2001; Morris and Doak 2002). Additionally, they have proven to be useful for the protection of threatened and endangered species (Crouse et al. 1987; Doak et al. 1994; Fujiwara and Caswell 2001; Hole et al. 2002; Kendall et al. 2003; Norris and McCulloch 2003; Wilson 2003), biological invasions (Thomson 2005), restoration strategies (Endels et al. 2005), toxicology (Wennergren and Stark 2000; Stark and Banks 2001; Stark et al. 2004; Van Kirk and Hill 2007), and amphibian decline (Vonesh and De La Cruz 2002). Thus, matrix models have a rich history in applied ecology and are a widely used approach for conservation of threatened and endangered species. Matrix models have also been useful for assessing how changes in specific vital rates affect population outcomes through elasticity and sensitivity analysis (Hansen et al. 1999; Caswell 2001).

Disadvantages of Matrix Models Compared to Other Models One of the major disadvantages of matrix models is that they are based on present conditions (constant parameters) and project what might occur in the future based on these conditions (Banks et al. 2008). However, demographic and environmental parameters change through time. For example, reproductive and mortality rates often

40

Ecological Models for Regulatory Risk Assessments of Pesticides

vary temporally, especially with exposure to toxicants that degrade in the environment over time. It is possible to modify matrix models with time-varying parameters (Caswell 2001; Gotelli and Ellison 2006; Meng et al. 2006), but these modifications are mathematically complex and prone to the proliferation of parameter estimation errors (Wood 1994). Some software programs, for example, RAMAS GIS (Applied Biomathematics, Setauket, New York), have the ability to vary both environmental and vital rates over time. For example, pesticide degradation and the ensuing reduction of effects on organisms can be specified in RAMAS GIS. Another disadvantage of matrix models is that the data necessary to develop a model are more difficult to obtain than acute mortality data or single measures of chronic effects on reproduction. Although certainly not limited to matrix models, the data used to construct models are often developed in the laboratory where the organisms have been cultured under optimal conditions. Hence, fecundity is often overestimated and mortality is underestimated. This is not an issue when comparing different life histories when all species have been raised under the same conditions, but can result in erroneous conclusions if the model is used for a risk assessment of a specific species in the wild. Thus, field-collected data are essential when trying to protect endangered species.

Rat–Elephant Phenomenon or Why the Use of Surrogate Species May Be a Bad Idea Another reason to use modeling for the determination of effects of toxicants on threatened and endangered species has to do with the fact that surrogate species are often used in ecological risk assessment to make predictions about the fate of endangered species (http://www.epa.gov/oppfead1/endanger/consultation/ecoriskoverview.pdf; Stark 2006). Surrogate species are often used because numbers of threatened and endangered species are low, and therefore toxicological data cannot be developed. Furthermore, it is illegal to take endangered species from their habitat or harm them. The issue of how accurately surrogate or indicator species can be used to predict effects on threatened and endangered species continues to be contentious in toxicology and risk assessment circles (Stark 2006). Stark et al. (2004) showed that different species exposed to the same levels of stress (mortality, reductions in the number of viable offspring, or a combination of both of these factors) do not react the same over time periods where reproduction can occur. Stark (2006) calls this “the rat–elephant phenomenon.” This phenomenon is best illustrated by imagining that we have 2 animal populations (rats and elephants) consisting of 100 individuals each, and we kill 50% of each population. The question then is which population will recover the fastest? The answer is obvious: the rat population will recover much faster than the elephant population. The rat population recovers faster than the elephant population because it has a higher intrinsic rate of increase, higher reproductive rate, and shorter generation time than the elephant. The conclusion of the study by Stark et al. (2004) was that you cannot compare species with respect to toxicity endpoints such as the LC50 or NOECs for reproduction over long time periods unless they have extremely similar life history traits, and therefore the use of

41

Development and Use of Matrix Population Models

Number of Individuals

100000 80000 60000 40000 Atlantic Herring Yellowtail Flounder Fathead Minnow D. pulex

20000 0

0

10

20

30

Time (years)

Figure 5.3  Comparison of deterministic population projections for 3 fish species and Daphnia pulex.

surrogate species may have little value for risk assessment unless life history traits among the species used as surrogates and the species being considered for protection are very similar. An example of this phenomenon is seen when comparing Daphnia to fish (Figure 5.3). Daphnids have extremely short life spans, generation times, and very high population growth rates, which can be measured as growth rate (λ)/day, and produce many broods of offspring during their lifetime of approximately 2 months under ideal laboratory conditions. Many fish species have 1 brood of offspring per year, and their growth rates are measured as λ/year. In this example, the starting population of each species is 100 individuals. Populations are allowed to grow to 100 000 individuals. The daphnid species reaches 100 000 individuals within 3 weeks, while the yellow tail flounder, Atlantic herring, and fathead minnow reach the same number of individuals in 13, 16, and 20 years, respectively. Another factor that may confound the use of surrogate species for determination of effects on another species is differential susceptibility. Differential susceptibility to a specific chemical can be extremely large even among closely related species. For example, Stark and Vargas (unpublished manuscript) found large differences in susceptibility to fipronil applied to soil to stop adult emergence of 3 species of tephritid fruit flies: the Mediterranean fruit fly, Ceratitis capitata (Wiedemann); melon fly, Bactrocera cucurbitae (Coquillett); and oriental fruit fly, Bactrocera dorsalis (Hendel). The melon fly was found to be 435 times more susceptible than the Mediterranean fly at LC50. Differential susceptibility has to do with differences in uptake, detoxification, and elimination of toxicants, which vary even among closely related species. The combination of differences in life history parameters and differential susceptibility to population viability has not been thoroughly explored to date.

42

Ecological Models for Regulatory Risk Assessments of Pesticides

Modeling can be used to get a better idea of what might happen to a population after exposure to a toxicant, but if surrogate species have to be used, then differences in life histories and differential susceptibility should be taken into account if possible.

Example of Population Modeling with 3 Insect Species Exposed to a Pesticide Stark and Sherman (1989) developed toxicity data for the 3 tephritid fruit fly species mentioned earlier after exposure to the organophosphorous insecticide, acephate. They found that the oriental fruit fly was the most susceptible species, followed by the melon fly and Mediterranean fly (Table 5.1). Although these are pest species, for this example I will consider them to be beneficial species and invent an EEC of 1 µg/g. Quotient ratios based on the LD50 and the above-mentioned EEC indicate that acephate poses a hazard only to the oriental fruit fly (Table 5.1). By plotting the dose–response curves for these species after exposure to acephate and comparing the EEC to these curves, we see that the EEC predicted mortalities are 83, 7, and 1% for the oriental fruit fly, melon fly, and Mediterranean fly, respectively (Figure 5.4). The growth rates (λ) for these species were determined to be 2.7484, 1.6897, and 1.5992/week for the oriental, melon, and Mediterranean fly, respectively (modified from Vargas et al. 1984). Matrix models were developed for each species (controls) and with the predicted mortalities resulting from exposure to acephate (Figures 5.5 and  5.6). The recovery time interval was the endpoint of interest here. The Mediterranean fly was unaffected by 1% mortality, and therefore no graph is presented. However, the melon fly population had a recovery period of 2 weeks (Figure 5.6), while the oriental fruit fly had a recovery period of 7 weeks (Figure 5.5). Comparing recovery time interval to generation time, we find that exposure to the EEC of acephate resulted in a delay of >1 generation time interval only for the oriental fruit fly (Table  5.2). Therefore, results of this exercise show that these 3 closely related species exhibited differences in life history traits and susceptibility to acephate, which resulted in very different outcomes at the population level. The quotient method correctly indicated that acephate posed a hazard only to the oriental fruit fly. However, sublethal effects were not considered in this model and the range of effects (7-week recovery) versus no delay in the Mediterranean fly could not be predicted by the quotient method. Table 5.1 Toxicity and Hazard Ratio of Acephate to 3 Fruit Fly Species Species Mediterranean fly Melon fly Oriental fruit fly

LD50 µg/g

EEC µg/g

6.72 1.46 0.84

1 1 1

Hazard Ratio EEC/LD50 0.149 0.680 1.19

43

% Mortality

Development and Use of Matrix Population Models 99 98 95 90 80 70

83% mortality

50 30 7% mortality 20 10 5 2 < 1% mortality 1 0.5 0.2 0.1 0.1 0.5

Oriental Melon Med

EEC = 1 ppb 0.8 1 1.4 2 Dose (µg/g)

3

4 5 6

8 10

15 20

Figure 5.4  Dose–response curves for tephritid fruit flies exposed to acephate.

Number of Individuals/ha

2.5e+9

Effect of 83% mortality on OFF

2.0e+9 1.5e+9

7 week recovery time interval

1.0e+9 Control 83% mortality

5.0e+8 0.0 0

10

20 30 40 Time (weeks)

50

60

Figure 5.5  Deterministic population projections for an oriental fruit fly control population and a population exposed to the acephate EEC resulting in 83% mortality.

The same models can be run stochastically (Figure 5.7). For the oriental fruit fly we see that although the population does not go to extinction, 83% mortality results in a population that remains at a much lower level than the control.

Conclusions Matrix models have a long history of use by conservation biologists and applied ecologists. They are easy to use and understand, and the data necessary for these models can be developed in the laboratory and/or the field. There are disadvantages to the use of matrix models; the most important of these is that they are based on initial conditions and project what might occur based on these conditions. Thus, matrix models are projection, not prediction, models. Because matrix models can

44

Ecological Models for Regulatory Risk Assessments of Pesticides

Table 5.2 Comparison of Recovery Time and Generation Time for 3 Fruit Fly Species Exposed to the Acephate EEC Species

Generation Time (Days)

Recovery Time (Days)

12.30 41.60 38.44

0 14 98

Mediterranean fly Melon fly Oriental fruit fly

Number of Individuals/ha

2.5e+9

Recovery Time/ Generation Time 0 0.34 2.55

2 week recovery time interval

2.0e+9 1.5e+9 Control 7% mortality

1.0e+9 5.0e+8 0.0 0

10

20 30 40 Time (weeks)

50

60

Figure 5.6  Deterministic population projections for a melon fly control population and a population exposed to the acephate EEC resulting in 7% mortality.

Number of Individuals/ha

10000

Control

8000 6000 83% mortality

4000 2000 0

Effect of 83% mortality on OFF

0

10

20 30 40 Time (weeks)

50

60

Figure 5.7  Stochastic population projections for oriental fruit fly exposed to the EEC of acephate.

Development and Use of Matrix Population Models

45

take into account lethal and sublethal effects and differences in life history parameters among species, they can be effective in providing guidance for management of threatened and endangered species. Because species have different life history traits and susceptibilities to toxicants, surrogate species employed for prediction of effects on threatened and endangered species should be used cautiously. Species with similar life histories should make better surrogates for endangered species, but the assumption here is that susceptibilities to the toxicant in question are also similar. Population modeling of the effects of toxicants on species should improve our ability to protect species.

6 MASTEP An Individual-Based Model to Predict Recovery of Aquatic Invertebrates Following Pesticide Stress Paul J. Van den Brink and J.M. (Hans) Baveco

Contents Introduction............................................................................................................... 48 Materials and Methods.............................................................................................. 49 Purpose............................................................................................................ 49 State Variables and Scales................................................................................ 49 Process Overview and Scheduling................................................................... 49 Design Concepts.............................................................................................. 50 Initialization..................................................................................................... 51 Input ............................................................................................................. 51 Submodels........................................................................................................ 51 Life Cycle............................................................................................ 51 Reproduction........................................................................................ 51 Mortality.............................................................................................. 51 Density Dependence............................................................................ 51 Dispersal and Movement..................................................................... 51 Pesticide Mortality............................................................................... 52 Scenarios.......................................................................................................... 52 Landscape............................................................................................ 52 Exposure.............................................................................................. 52 Results....................................................................................................................... 52 Discussion................................................................................................................. 54 Assumptions of the Model............................................................................... 54 Uncertainty in the Parameters.......................................................................... 54 Outlook............................................................................................................ 54

47

48

Ecological Models for Regulatory Risk Assessments of Pesticides

For the ecological risk assessment of pesticides, recovery of affected aquatic populations is an important aspect. Due to spatial and temporal constraints, recovery cannot be studied experimentally for all species. For instance, species that lack a resistant life stage and also lack aerial dispersal cannot recover after becoming extinct in microcosms or mesocosms. It is therefore proposed to estimate recovery using ecological modeling. In this chapter we present an individual-based population model (Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides [MASTEP]). MASTEP describes the effects on, and recovery of, populations of the water louse Asellus aquaticus following exposure to a fast-acting, nonpersistent insecticide caused by spray drift for pond, ditch, and stream scenarios. The model used the spatial and temporal distribution of the exposure in different treatment conditions as an input parameter. A dose–response relation derived from a hypothetical mesocosm study was used to link the exposure with the effects. The modeled landscape was represented as a lattice of 1 × 1 m cells. The model included processes of mortality of A. aquaticus, life history, random walk between cells, density-dependent population regulation, and in the case of the stream scenario, medium-distance drift of A. aquaticus due to flow. All parameter estimates were based on the results of a thorough review of published information on the ecology of A. aquaticus and expert judgment.

Introduction One of the major issues in the environmental risk assessment of pesticides in Europe is the estimation of recovery of affected (aquatic) organisms after pesticide-induced stress (European Commission 1997). For organisms with resistant life stages or an aerial life stage and multiple life cycles per year, for example, Daphnia and Chaoborus, recovery can be estimated experimentally using outdoor microcosms or mesocosms when the study duration is long enough and a source of colonizers is available (Van den Brink et al. 1996). For nonflying organisms with no insensitive life stages, like Gammarus and Asellus, however, the isolated nature of the mesocosms and the limited duration of the experiment prevent the study of recovery of these species. In this chapter, we, therefore, describe a spatially explicit model for water louse Asellus aquaticus populations and their recovery after pesticide stress. The model can be used to estimate combined recovery through autogenic (recovery from inside, e.g., through reproduction or insensitive life stages) and allogenic (recolonization from outside of the system) processes after a spray drift event involving an insecticide in a stream in Northwest Europe. Asellus aquaticus is a widely distributed freshwater crustacean common in both standing water (ponds, lakes) and flowing water (streams, rivers). Population dynamics vary according to temperature: typically, there is 1 breeding peak in summer in Northern Europe, 2 peaks (1 in spring and 1 in autumn) in Northwest and Central Europe, and either year-round reproduction or winter breeding in Southern Europe. Since this chapter focuses on Northwest Europe, only life cycle characteristics representative of this region were used.

MASTEP

49

The water louse A. aquaticus was used as an example for invertebrates because it is relatively sensitive to insecticides and has a presumed low capacity for allogenic and autogenic recovery. Its low recovery potential is caused by its lack of ability to recolonize via terrestrial life stages and because it was believed to have a relatively low population growth rate. The decision to use A. aquaticus meant that there was no need to model multiple nonconnected watercourses, because exchange of individuals between these watercourses would not occur directly without interference of other agents like man and waterfowl. We therefore only modeled connected watercourses, though the model concept easily allows for the inclusion of nonconnected watercourses in the future. The model is described in full by Van den Brink et al. (2007).

Materials and Methods This section describes the MASTEP model and its application for Asellus aquaticus. We follow the standard protocol for describing individual-based models as proposed by Grimm et al. (2006). We chose an individual-based approach because the individual level is easily linked to the population level, that is, the level we are interested in from a risk assessment point of view, and because it allows us to use available data at both the individual and the population level. It is a natural approach because it describes the very entities comprising a population and their behavior. Furthermore, as exposure varies in time and space individuals will not receive the same exposure — something that individual-based models (IBMs) are well suited to model. MASTEP was developed in VisualWorks Smalltalk (smalltalk.cincom.com) using the EcoTalk modeling framework (Baveco and Smeulders 1994).

Purpose The purpose of the model is to quantify population effects and recovery after pesticide exposure.

State Variables and Scales The model included 2 types of entities: female individuals and quadratic grid cells comprising the habitat. The individuals were characterized by the state variables: identity number, generation number, location (coordinates of a grid cell), and an array of experienced local densities (density history). The time unit was a day and simulations usually lasted for 1 year (365 days). A grid cell’s size represented 1 × 1 m, and the habitat contained a number of grid cells depending on the spatial scenario (e.g., an array of 600 cells for a ditch scenario).

Process Overview and Scheduling State changes were scheduled as discrete events (see Figure 6.1). The events of reproduction and death due to aging were scheduled at the “birth” of each organism (i.e., when it first appeared in the simulation). If the individual was still alive at the time of the reproduction event, it would reproduce. At the time of the mortality event the

50

Ecological Models for Regulatory Risk Assessments of Pesticides “Birth”

Local density Density-dep. mortality

Move to new cell

Update density history

Residence time PDF

1 Day time step Reproduce Death

Reproductive age PDF Life span PDF

Figure 6.1  Overview of the scheduling of state change for an Asellus individual in the Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides (MASTEP). In boxes the different events in its life history are shown, and in italics the origin of the time delay after which the event takes place. Arrows without text point to events that take place immediately (time delay of 0). The main loops are the ones occurring with a 1-day time delay checking for density-dependent mortality, and the movement loop. Pesticide application was scheduled as a separate event.

individual would be removed from the simulation. Movement was also scheduled as a sequence of discrete events in continuous time. The timing of movement events was determined by the residence time probability density function (PDF). The timing of reproduction and mortality due to aging were determined by the age at reproduction PDF and life span PDF, respectively. The check on local (within-cell) density and the effectuation of density-dependent mortality was scheduled with a fixed delay of 1 day, equivalent to what would happen in a time-step-based model with a 1-day time step.

Design Concepts The model did not include adaptive behavior or individual decision making, so it was similar to matrix models, that is, based on demographic rates and further empirical parameters. The representation of the processes’ reproduction, mortality, and movement or dispersal included stochasticity. As Figure  6.1 shows, the timing of most events was stochastic. In addition, some vital rates were represented as probabilities, for example, density-dependent mortality and the number of offspring. Stochasticity was included in order to incorporate individual variability in a natural way, and to avoid artifacts due to unrealistic synchronization (e.g., all offspring appearing at the same day). The observation variables were density of individuals, either in the 100 m sprayed part of the scenario or the whole modeled water body (600 m). The number of

MASTEP

51

individuals from the different generations were summed. The 95% confidence intervals of the results were obtained from at least 5 replicate runs.

Initialization Initial population size amounted to 1000 individuals, randomly distributed over the 600 cells.

Input The model did not include any driving environmental variable; that is, the environment was assumed to be constant.

Submodels For details, see Van den Brink et al. (2007). Life Cycle The model focused on a single annual cycle, comprising several generations. The first generation 1) consisted of individuals born in the previous year. These individuals reproduced around day 120 (day 1 being January 1), causing the first population peak. The next generation of individuals 2) reproduced 70 days later (around day 190), leading to the second population peak. Reproduction Clutch size was set to depend on age at reproduction and mean local density encountered by the individual. Mean local density was calculated as the mean of all the within-grid cell densities encountered by the individual. The number of offspring could never exceed twice the default clutch size. Mortality The model set the life span of each individual at birth in a probabilistic way. Density Dependence The density-dependent mortality rate was assumed to be linearly related to actual local density. Density-dependent reproduction was incorporated by decreasing the number of offspring with average experienced density for each individual. Dispersal and Movement Individual movement by walking was modeled as a jump from 1 cell to a randomly selected neighboring cell at a time set by the (probabilistic) residence time. The probability density function was obtained from a simulation of a random walk process with parameters derived from experimental work (Englund and Hambäck 2004). The model incorporated passive movement downstream by implying that 1% of the movement to other cells was long-distance movement (drift) in a downstream direction. Drift distance was incorporated as an exponential distribution, with an assumed average of 10 m.

52

Ecological Models for Regulatory Risk Assessments of Pesticides

Pesticide Mortality Survival at a given initial (peak) concentration in the water was defined by a dose– response curve using a logistic model, with mortality occurring immediately after exposure. The parameters of this curve are obtained from the results of a hypothetical mesocosm experiment. The numbers of A. aquaticus collected 1 week after application of the chemical were regressed on the peak concentrations of the chemical occurring immediately after application.

Scenarios Landscape The structure of the stream scenario was 600 × 1 m2 cells. To obtain more realistic boundary conditions the first cell was connected to the last one (periodic boundary conditions): the individuals that migrated out of the system downstream entered it on the upstream side. The pesticide could be transported downstream, but no farther than 600 m. Periodic boundary conditions simulated a simultaneous treatment 600 m upstream of the system (and 1200 m, 1800 m, etc.). Exposure Since this chapter focuses on the effect side of the model, no justification will be provided for the peak exposures that are used to calculate the pesticide-induced mortality. The 4 profiles used, however, were representative of normal agricultural use of a fast-acting, fast-dissipating insecticide, using a 10, 12.5, 15, and 17.5 m buffer zone.

Results The untreated population density showed the expected trend of a spring peak and a (higher) summer peak (Figure 6.2a and b). Figure 6.2c shows the results with the 95% confidence intervals for the control and the 17.5 and 10 m buffer zone treatment levels. All buffer zone treatment levels were chosen to result in insecticide predicted environmental concentrations (PECs) at or above the EC50, and thus leading to a clear decrease in densities. Spatially, the application of the insecticide led to a drop in densities in the first 100 m stretch (Figure  6.2a). There were many cells from which all individuals disappeared, although a rapid recovery after treatment was observed; that is, all treatment conditions returned to control levels within 50 days (Figure 6.2a). This shows that the long-distance movement of Asellus was a very important factor determining its population recovery. After 75 days, the effect became apparent again, as a result of the periodic boundary conditions. When the numbers in the entire 600 m stretch are taken into account, differences persisted longer (Figure 6.2b). This means that effects were “exported” to untreated cells due to a reduced influx either by walking or by drift from the treated cells. This is clearly visualized in Figure 6.3. In the figure time runs from top to bottom and the stream from left to right. The left part of the stream is sprayed with the insecticide on day 130 and causes a complete

53

MASTEP Stream scenario (100 m)

1000 100 10

10000

Number (ind)

Number (ind)

10000

75 100 125 150 175 200 225 250 275 300 325

Stream scenario (600 m)

1000

100

75 100 125 150 175 200 225 250 275 300 325

Time (d)

Time (d) (b)

(a)

Number (ind)

10000

Stream scenario (100 m)

17.5 m buffer zone

1000

15 m 12.5 m

100 10

Control

10 m 75 100 125 150 175 200 225 250 275 300 325

Time (d) (c)

Figure 6.2  (See color insert.) Dynamics of population numbers in all treatment levels (a) for the treated 100 m stretch, (b) the complete 600 m stretch, and (c) the 95% confidence intervals of the dynamics of numbers of the treated 100 m stretch. The pesticide was applied on day 130.

Time (d)

0

0

Distance (m)

365 600

Figure 6.3  (See color insert.) Visual representation of the dynamics of abundance for one of the runs of the 10 m buffer zone treatment level. The x-axis represents the 600 m stretch, while the y-axis represents the temporal dimension (each day adding a row). The colors represent population density, with black for low and blue for high densities. The results of the complete 600 m stretch are shown; the first 100 m stretch was treated with an insecticide on Julian day 130.

54

Ecological Models for Regulatory Risk Assessments of Pesticides

die-off (black denotes absence of Asellus, blue high abundance values). Although only one-sixth of the stream is sprayed with the insecticide, Asellus died in more than half of the 600 m stretch of the stream because of contamination by water movement (Figure 6.3). Recovery was fast in the affected parts of the stream due to the movement by drift. Because of that, the “empty patch” traveled through the stream over time. Eventually, full recovery within a year was not obtained in the 3 highest treatment levels, because of the periodic boundary conditions.

Discussion Assumptions of the Model Models are by definition a simplification of reality. The model presented here was detailed in terms of population age structure and spatial structure (and movement), but many other factors were omitted or simplified. In the present study we implemented a simple link between the fate of a chemical and the effects on Asellus individuals in the model.

Uncertainty in the Parameters Some of the parameters of the model, such as mortality, age of breeding, and number of young, have been accurately reported in the literature. The situation was completely different for parameters of movement and density-dependent processes. We know of only 1 experiment studying Asellus movement, and have derived the movement parameters for the model from this experiment, which was performed in an artificial environment without food and shelter. To assess the sensitivity of the model to the invertebrate drift parameters, we conducted simulations with and without invertebrate drift (results without drift not shown). These showed that the outcome differed for the treated stretch of the stream, where densities in the absence of invertebrate drift failed to return to untreated levels, but not much for the entire population in the 600 m stretch. We were unsure what the real drift values were, although 1% did not seem unrealistically high (Peeters et al. 2002). The parameters we used for density-dependent regulation could not be underpinned with data from the literature; the main role of the density dependence in the model was to keep populations at a desired density level without affecting the population’s potential for recovery from very low densities too much.

Outlook The model presented in this chapter shows that theories on, for instance, density dependence, life cycles, and movement patterns developed in the field of ecology can be applied in the risk assessment of chemicals. It therefore also offers an example of stress ecology, that is, ecology into which a stress element is integrated. Risk assessment of pesticides is currently merely based on determining the sensitivity of organisms, while the results of this model show that life cycle characteristics might be

MASTEP

55

equally important for determining the spatial and temporal magnitude of the effects. This raises the question of what actual level of protection is achieved by the use of single-species tests in the first tier of the risk assessment, which is almost devoid of ecology (Van den Brink 2006). This first tier may still provide a sufficient level of protection because of the use of safety factors. In this chapter we used the well-studied water louse A. aquaticus as an example. Although some life cycle characteristics of this species, like age and number of offspring, are known from detailed studies, others, like density dependence and walking behavior, are not. We therefore need more research concentrating on the life cycle and movement patterns of invertebrates. If flying insects are included as well, nonconnected water bodies should also be included, so the model becomes a metapopulation model in the classical sense. Adding more life cycles and more complex landscape features will make MASTEP a tool that allows the results of microcosm and mesocosm experiments to be extrapolated to the landscape level. This would allow better regulatory decisions to be made on acceptability of effects, as a more realistic description of recovery is obtained than that provided by the microcosm and mesocosm experiments alone.

Realism 7 Incorporating into Ecological Risk Assessment: An ABM Approach Chris J. Topping, Trine Dalkvist, and Jacob Nabe-Nielsen Contents Introduction............................................................................................................... 57 ALMaSS................................................................................................................... 58 Purpose............................................................................................................ 59 State Variables and Scales................................................................................ 59 Process Overview and Scheduling................................................................... 59 Examples of ALMaSS Applications.........................................................................60 Example 1: Measuring Carrying Capacity for Bembidion.............................. 61 Example 2: Impact of Altering Landscape Structure....................................... 61 Example 3: Assumptions Regarding Other Mortalities in a Risk Assessment.................................................................................... 63 Example 4: Modeling Chronic Effects of an Endocrine Disrupter in Voles....64 Discussion.................................................................................................................64

We present ALMaSS, an agent-based simulation model (ABM) system that has been used to evaluate impacts of pesticides in a range of terrestrial applications. Four examples are presented highlighting different aspects of using ABMs with ecotoxicological problems and indicate the direction in which ABMs might play a role in regulatory risk assessment in the future.

Introduction The aim of increasing realism in our risk assessment model is naturally to increase the accuracy and predictability of our estimate of impact or risk. While this is a very 57

58

Ecological Models for Regulatory Risk Assessments of Pesticides

laudable goal, we might ask ourselves why this has not been more widely attempted to date. Perhaps the answer lies in 3 separate spheres: industry, the regulators, and scientists. While to an extent all 3 groups can see the need for greater realism and therefore accuracy in assessments, no 1 group seems able to act without the others, and consensus on how greater realism should be obtained is difficult to achieve. The rather slow embracing of greater realism in ecological modeling is to some extent explained by the short history of the science. This relatively young science has strived for recognition, competing with sciences such as physics, where simple laws and models are capable of defining many processes well. Ecology has thus strived to emulate physics and reduce its rather complex systems of study to simple mechanisms and general principles. This has led to the development of ecology by using many broad-brush simplifications, such as the application of the logistic equation to describe population growth (Begon et al. 1990). While the logistic equation embodies the general principles of density-dependent growth, it is at best a general descriptor for the growth of populations in the real world where spatiotemporal factors are usually important. It could therefore be argued that ecology and population dynamics, in particular, have throughout their short history suffered from a “physics-envy” syndrome whereby the aim of ecology was to simplify the study system to the extent that it could be captured in simple equations (Grimm 1994; Weiner 1995, 1999). The reason for this behavior is twofold: 1) because a simple system is easy to understand and explain, and 2) because the complexities of ecology were often not tractable to the methods of study available at the time. Today, with the rapid technological progress in computing, it is possible to encapsulate complex systems in computer models, and hence simplification for the mere sake of mathematical tractability is no longer necessary. These computer systems allow us to integrate mechanisms and patterns, allowing the emergence of model behaviors comparable in complexity to the real world, but generated under controlled conditions, thus allowing subsequent simulation experiments (Peck 2004). The aim of this chapter is to present examples of incorporating complexity into population dynamic models of animal systems, which could be used for risk assessment purposes, and to give an idea of what is currently being done and what might be possible. The model system used is ALMaSS (Topping et al. 2003). ALMaSS is short for Animal, Landscape, and Man Simulation System, which indicates the 3 major components of the system.

ALMaSS This model description is based upon the ODD protocol (Overview–Design concepts–Detail; Grimm et al. 2006). For a model of the dimensions of ALMaSS (ca. 70 000 lines of code), and due to space constraints, this is necessarily a gross simplification. Here we will only briefly present an overview of ALMaSS structure (for a more comprehensive account, see Topping et al. 2003).

Incorporating Realism into Ecological Risk Assessment

59

Purpose The objective of the system is to integrate a wide range of factors related to spatiotemporal patterns of habitat and food availability and interactions between these and man’s management of the landscape and the animal’s ecology. One of the main areas to which the model has been applied is incorporating spatiotemporal factors into population-level risk assessment of pesticides (Topping and Odderskær 2004; Topping et al. 2005). The system uses ABMs of specific species with the aim of incorporating the current state of knowledge regarding their ecology and behavior as it pertains to simulation of their spatial and temporal dynamics. These models range from relatively simple invertebrate models (e.g., Bembidon lampros) to mammal models involving complex behavioral components (e.g., Microtus agrestis and Capreolus capreolus).

State Variables and Scales The landscape model uses a set of state variables to describe the local conditions within habitat fragments, including details of vegetation type and structure and a record of recent human activities, for example, spraying of a crop. In addition, there are variables describing the landscape topography, farms, and stage of management for each farm and field combination. There is also a weather data set that determines the temperature, rainfall, and wind run for each day during the simulation. The landscape model is updated once each simulation day with all the human activities that were carried out for each habitat fragment, and with all vegetation growth and weather changes. The topographic landscape of ALMaSS has a resolution of 1 m2 and typically an extent of 10 km2 (e.g., Figure 7.1). Animal models have a set of variables describing their current activity, location and age, and relevant physiological parameters (e.g., weight, energy level). For animals that have complex social interactions, for example, partridges (Perdix perdix), each agent will have information about its family group (covey). The state variables are naturally very different for each species, which serves to create unique species models with only the basic model structures in common.

Process Overview and Scheduling ALMaSS is rich in processes. These include models of vegetation growth for each vegetation type, whereby all vegetation grows according to the temperature and day length, altering its biomass and height on a daily basis. Models of human activity, for example, over 50 different models for management of farm crops, activities such as cutting of roadside verges, and simulation of traffic loads on roads, are incorporated. All landscape processes are modeled on a daily time step, and while it is possible to set any time step for animal models, a daily time step is also typically used for these. Animal models include processes such as energy-based growth (vertebrates) or temperate-based growth (arthropods), movement, responses to external triggers (e.g.,

60

Ecological Models for Regulatory Risk Assessments of Pesticides

Arable fields Building Coniferous forest Coniferous forest Mixed forest Parkland/Recreational grass Permanent pasture Scrub Unmangaged grassland Water Young forest plantation

Figure 7.1  (See color insert.) ALMaSS screenshot of a typical 10 × 10 km landscape used for simulations. In this case the red dots indicate the overwintering positions of simulated beetles. Note that the map resolution is much finer than is displayed on screen.

disturbance from human activity, predation), territorial behavior, and reproductive behavior (e.g., see Topping and Odderskær 2004). Scheduling of the model’s processes is a complex matter for the animal models because this must achieve 2 goals. First, it must avoid causing bias due to problems of concurrency (see Topping et al. 1999), and second, it must ensure a sensible sequence of events between interacting agents. ALMaSS does this using a hierarchical state transition system separated into 3 stages (these stages are considered serially in the computer but represent parallel time slices of simulation time for each individual). The sequence of individual actions within each stage can be ordered randomly or sorted, for example, by location, depending on what is logically desirable. When calling the 3 stages individual agents exhibit specific behavior (e.g., dispersal), and depending on what state they are in at the end of that behavior, they may make a conditional transition to another state. This allows complex sets of behaviors to be integrated within a single day without breaking a logical sequence of events, for example, foraging adults bringing food to their chicks resulting in subsequent chick growth.

Examples of ALMaSS Applications The following is intended as a set of examples only, and space limits the details of the specific simulations that can be presented here.

61

Incorporating Realism into Ecological Risk Assessment

Example 1: Measuring Carrying Capacity for Bembidion (Sibly et al. 2009) Using the Bembidion model (Bilde and Topping 2004), measurements of simulated beetle numbers were taken on a 10 × 10 km landscape using a 500 × 500 m grid. Carrying capacity (K) was calculated for each grid square separately and weather year based on the linear relationships between the intrinsic population growth rates r and lnN. Here K was found as the population size at which r was equal to 0. There was considerable variation in K among both weather years and grid squares (Figure 7.2). This variation indicates that at the scale at which we would be considering the impact of using an agrochemical, the effect we need to measure must be considered against a background of complex local spatiotemporal population dynamics. These effects could not be captured with simple population growth models at this scale.

Example 2: Impact of Altering Landscape Structure (Nabe-Nielsen et al. in preparation) In this example, we were interested in the effect of altering the structure of the landscape but not its composition on the beetle population dynamics. Using a 10 × 10 km landscape (Figure 7.1), the shape of the landscape elements was initially rounded while keeping size and position of the center constant (simplified shape). Subsequently, landscape elements were randomly repositioned by swapping with other landscape elements of the same size (randomized positions, Figure 7.3). The result was 3 landscapes, 1 with the realistic landscape structure, and the others having varying degrees of landscape simplification. In each landscape the beetle population was monitored over a series of years using a repeated cycle of 10 weather years (Figure 7.4). In the simulation results 3 2

pgr

1 0 –1

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

20 15 10 5

–2 4 6 8 2 In (N) Adult Females on 1 June (a)

5

10 (b)

15

20

Figure 7.2  (See color insert.) (a) Variation in carrying capacity (K) among weather years within a single 500 × 500 m2 in the 10 × 10 km natural landscape. K is defined as the population size, where population growth rate (pgr) is zero for each weather year regression. (b) Variation in population size among the 400 squares in the weather year 1995, which was used repeatedly over 200 simulation years. Contours link regions with the same density. Green indicates high population density and white indicates zero population size.

62

Ecological Models for Regulatory Risk Assessments of Pesticides

Figure 7.3  (See color insert.) A section of a 10 × 10 km landscape before and after rounding of landscape features and subsequent randomization of their position. Buildings, roads, and water stay in the same place, but all vegetated habitats are potentially moved. See Figure 7.1 for key to landscape features.

beetle numbers were clearly related to the structure of the landscape even though the area covered with the different element types was not altered. Further analysis showed that the causal mechanism was the local migration behavior of the beetle interacting with habitat complementation. This indicated that oversimplification, or in this case even moderate simplification, of landscape structure can lead to bias even though all processes and parameter values were constant between runs.

Population Size (In)

6e+05

Realistic

5e+05

Simplified shape

4e+05

Randomised positions

3e+05 2e+05 1e+05 0e+00 1990 1992 1994 1996 1998 1990 1992 1994 1996 1998 1990 1992 1994 1996 1998 1990

Figure 7.4  (See color insert.) Beetle population numbers plotted against time for decreasingly realistic landscape structures. The x-axis indicates the weather year, which was cycled using a loop of 10 years from the 1990s.

63

Incorporating Realism into Ecological Risk Assessment

Example 3: Assumptions Regarding Other Mortalities in a Risk Assessment (Topping et al. in preparation) This example concerned the assessment of impact of a fictitious foliar insecticide on beetle and spider populations. The effect of variation in life-history traits of the nontarget animals was incorporated in the assessment by including 2 variations of the Bembidon lampros (beetle) model (Bilde and Topping 2004) and 2 spider species, Erigone atra and Oedothorax fuscus (Thorbek and Topping 2005). The spiders used had similar habitat requirements but differed in their breeding behavior and dispersal, whereas the 2 variations of the beetle model only differed in the daily dispersal rate (10 or 20 m). The impact was measured as the size of the population reduction compared to a baseline scenario without the treatment in question. Three assessments were made for each of the species. The first assessed the impact of soil cultivation and harvest mortalities in the absence of pesticide. The second assessment assumed that the population was only exposed to a single human-mediated mortality, that is, the pesticide under testing (90% chance of mortality with direct exposure, no effect of exposure to residual pesticide). The third included a more realistic assessment of the levels of mortality, which included the mortalities of both the pesticide treatment and other farming operations (values taken from Thorbek and Bilde 2004). The results indicated that incorporating the extra realism of other stressors in the system dramatically changed the resulting impact assessment, by up to factor 10 (Table 7.1). In addition, the relative change in impact depended upon the species under study, and in the case of the beetle its rate of movement. The explanation for the results was related to the assumptions built into the model relating to density dependence. Since density in ABMs was a local factor, the impact of density dependence was affected by the local history of events. In this case, if agricultural mortalities resulted in a reduction in population size before pesticide application, then those killed by the pesticide would have a large impact Table 7.1 Impact Assessments of Insecticide to All Arable Fields for 4 Species

Species Bembidion 10 m Bembidion 20 m Erigone atra Oedothorax fuscus

Agricultural Mortality Depression

Pesticide Depression (No Agricultural Mortality)

Pesticide Depression (with Agricultural Mortality)

0.91 0.68 0.80 0.90

0.22 0.18 0.02 0.09

0.75 0.20 0.22 0.37

Change in Impact When Including Agricultural Mortality 241% 11% 1000% 311%

Note: Columns 2 to 4 indicate the change in population size relative to the baseline (0.9 indicates 90% population size reduction). Column 5 indicates the difference in measured impact when considering the pesticide in isolation, or against the background of agricultural mortality, and indicates the scale of potential error.

64

Ecological Models for Regulatory Risk Assessments of Pesticides

on the ability of the population to grow. In the alternate case, most of the individuals the pesticide kills could be regarded as part of the doomed surplus due to the large population size.

Example 4: Modeling Chronic Effects of an Endocrine Disrupter in Voles Simulated vole populations in a 10 × 10 km landscape were exposed to a fictitious pesticide based on vinclozolin, an endocrine disrupter with epigenetic effects resulting in fertility depressions being passed undiluted down the male line. These simulations are part of a larger study (Dalkvist et al. 2009), in which the impacts of varying a range of ecological and toxicological properties were compared. Three illustrative examples are presented here: 1) a comparison of impacts between worst-case and more realistic scenarios, 2) altering the crop on which the pesticide was sprayed, and 3) altering the threshold above which the pesticide caused a toxic response, in this case using the NOEL. In all 3 cases the rate of pesticide and the area to which it was applied were the same, and unless otherwise specified, all toxicological properties were identical. The impact of the pesticide was to render 50% of male offspring of pregnant females exposed above the NOEL sterile. The rest of the male offspring would have a reduced mating success, leading to successful mating in only half of the attempts and further transmission of a gene down the male line, resulting in 50% sterility and 50% with reduced mating success.





1) A baseline comparison between an unsprayed landscape and pesticidetreated orchard covering 5% of the total landscape, and a situation where we assume all voles were exposed regardless of their location. Figure 7.5 shows the pattern of depression relative to the baseline population for pesticide applications from 31 to 60 years. 2) Application of the same pesticide at the same time, proportion of the landscape and dose but only to orchards, as in the first case, oil seed rape, or intensively managed pasture. The treatment resulted in population size depressions of 8, 1, and