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 <1, and the proportion of males affected of 14, 4, and 1%. The differences were a function of the location of the voles in the landscape and indicated the importance of estimating a realistic exposure level. 3) The orchard scenario from the first case was rerun with varying values for the NOEL, doubling each time from 1.5625 mg/kg body weight (bw) to 50 mg/kg, producing 6 simulations of NOEL. The resulting reduction in impact followed a clear pattern, but was much less than the 5 times reduction in toxicity assumed by changing the NOEL (Figure 7.6).
Discussion Complex agent-based simulations are here to stay; they are currently finding usage in many fields, including economics, human geography, traffic control, and the biological sciences (e.g., Grimm et al. 2005; Fowler 2007; Pena et al. 2007; Worden and
65
Incorporating Realism into Ecological Risk Assessment Pesticide Sprayed in the Years Within the Two Arrows
Population Size Relative to Baseline
1.2 1 0.8
100% Exposure ‘Realistic’ Exposure
0.6 0.4 0.2 0
0
20
40 60 80 Year from Simulation Start
100
120
Figure 7.5 Simulated vole population depressions after the application of pesticide under 2 scenarios, 100% exposure or realistic application to orchards.
Population Depression (%)
16 14 12 10 8 6 4 2 0
0
10
20 30 NOEL (mg/kg bw/day)
40
50
Figure 7.6 Simulated vole population depression with decreasing toxicity (expressed as NOEL) for a “vinclozolin-like” pesticide. Halving the NOEL led to a linear decrease in population impact.
Levin 2007). There is little doubt that when properly constructed these models are capable of capturing realistically the nonequilibrium dynamic properties of systems that are a defining feature of the world around us. The resulting predictions are not always easy to reduce to simple relationships, but these patterns are produced from mechanistic processes and are repeatable and analyzable. In fact, one of the primary advantages of having realistic models is that these patterns or “secondary predictions” can be tested. When found to be accurate these patterns are strong indicators that the underlying model structure captures the essence of the system being modeled (Grimm et al. 2005). Hence, one of the major strengths of ABMs, resulting from
66
Ecological Models for Regulatory Risk Assessments of Pesticides
their rich diversity in structure and behaviors, is that they can be tested in a variety of ways and much more thoroughly than more simply structured models. Predictions of the models can be evaluated via experimentation with the model itself; hence, guesswork as to the cause of an observed effect can be eliminated. This together with realism is the key trade-off against the increased data and resource requirement of building these models. The examples presented earlier were designed to demonstrate some of the realistic and complex patterns derived from ALMaSS, and to indicate that the effect of altering seemingly simple assumptions regarding environmental structure, number of stressors, or exposure and toxicity can have large impacts. General rules of thumb may be elusive. For example, it would be attractive to conclude from the vole that increasing realism decreases our perception of risk, but as the beetle and spider example shows, this is not always the case. Space and time affect exposure probabilities, and these interact with the behavior and ecology of the animals (e.g., voles avoiding heavily managed grasslands in Example 4-2), toxicology, environmental fate, patterns of usage (e.g., crop distributions), weather, and so forth. While the list of potential factors that could be important may appear bewildering at first sight, the fact that these factors can be integrated in ABM simulations should be an encouraging sign that we may not be far from being able to make much more realistic risk assessments at population levels. What is now needed is a standardized method of implementing these models, and their testing. Testing is not as simple as testing statistical significance because we are dealing with mechanistic constructs; hence, techniques such as pattern-oriented modeling (Grimm et al. 2005) need to be employed, which are data intensive. However, if a concentrated effort could be directed at a few standard environments and species, then this process need not take many years. Developing standard test environments and species for testing would also alleviate the problem of defining the system to be tested and which factors to include, that is, create a level playing field for pesticide testing. This could be seen as a natural progression of the current proposed changes to EU Directive 91/414 resulting in the zoning of the EU (European Commission 2002a, 2002b, 2002c). Development of standard landscapes incorporating current agricultural and forestry practices for different climatic zones could be a first step at standardizing ABM simulation approaches across the EU. A list of key species exhibiting a range of life histories could be developed for each area analogous to the species currently used for toxicological testing in pesticide approval procedures. Doing this would not only provide a focus for pesticide regulatory cases, but would also bring ecotoxicology into line with other biological sciences by putting greater emphasis on the importance of complexity (e.g., Service 1999; Bell and Koithan 2006).
Models 8 Ecological Supporting Management of Wildlife Diseases Hans-Hermann Thulke and Volker Grimm Contents Introduction............................................................................................................... 67 Examples................................................................................................................... 69 Ecological Models Substituting Field Studies..................................................... 69 Background..................................................................................................... 69 Problem........................................................................................................... 69 Conceptual Model........................................................................................... 69 Ecological Model............................................................................................ 69 Results............................................................................................................. 70 Ecological Models Supporting Design of Field Application............................... 72 Background..................................................................................................... 72 Problem........................................................................................................... 72 Conceptual Model........................................................................................... 72 Ecological Model............................................................................................ 72 Results............................................................................................................. 74 Discussion................................................................................................................. 74
Ecological epidemiology of wildlife diseases is a field in which ecological models are increasingly used for decision support, including control legislation. Models are used to design field experiments and sometimes even to substitute for field experiments. The establishment of ecological models 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.
Introduction Many wildlife diseases are caused and transmitted by pathogens and can therefore become epidemics. There are several possible reasons why we might need to 67
68
Ecological Models for Regulatory Risk Assessments of Pesticides
control these epidemics. Unwanted consequences of wildlife epidemics include conservation issues regarding endangered species, dramatic consequences for agricultural food production, and direct harm to human beings. Fortunately, against many of the known wildlife disease agents, intervention methods were developed to limit or even eradicate the pathogens. Such control methods have 1 thing in common: control applications rely on widespread distribution of medication to the environment aiming at a homogeneous supply of treatment to the pathogen’s host individuals. Hence, despite all the obvious differences between pesticide application and wildlife medication, interesting parallels exist in the basic situation and the potential for ecological modeling to support management (Table 8.1). However, in contrast to regulatory pesticide risk assessment, where so far ecological models have only been used occasionally, models are frequently used to inform the management strategies of wildlife disease control, including control legislation. Here, we provide examples, which show how models assisted regulatory decisions and improved disease management in complex situations. We briefly present the example problems and discuss how the corresponding models supported management decisions and why they were accepted as a basis for legislative decisions or management plans. We emphasize the important role of confidence and how the main purpose of ecological modeling — gaining understanding about how ecological systems work — helps building up confidence among nonmodelers and decision makers. Finally, we summarize how ecological modeling can increase objectivity of decision making, as well as what it cannot provide. Table 8.1 Comparison of Characteristic Aspects of Controlling Wildlife Pathogens or Crop Pests, and of Potentials of Ecological Modeling in These Two Fields of Application Characteristic aspects Harmful agent in some population Disastrous or conflicting human interest Spreading by contact, by vectors or by wind Management by laboratory-approved control methods Search for objective support (modeling) Consequence estimation of intervention Feasibility justification (strategy, economy, public interest) Strategy optimization Cost–benefit
Wildlife pathogens
Pests
Yes Yes Yes Yes
Yes Yes Yes Yes
Seldom Yes
Yes Yes
Yes Yes
Potential Potential
Ecological Models Supporting Management of Wildlife Diseases
69
Examples Ecological Models Substituting Field Studies In some cases carefully parameterized, tested, and validated ecological models can substitute field studies and experiments. This is shown by the following example from rabies management in red fox populations. Background The rabies virus has been under human consideration for more than 2000 years due to rabies’ lethality. The devastating experience of humans dying of rabies even left traces in our language: for example, the adjective “rabid.” Therefore, in the last quarter of the 20th century global efforts resulted in the invention of an orally administrable vaccine that is distributed via baits (Baer et al. 1971). Since then millions of baits were distributed from airplanes over virtually all of Central Europe. This campaign removed the virus from the Central European fox population. Problem Presently, one of the tasks for European regulators is to prepare control guidelines for new outbreaks within an unprotected fox population (Freuling et al. 2008). A new outbreak will be locally restricted at first, but will put surrounding areas at risk due to successive spatial spread of the infection. However, due to logistic constraints immediately after the detection of the outbreak (e.g., limited amount of vaccines), resources for control measures will be limited. Thus, the challenge of localized emergency control will be the race between buildup time of a properly immunized population and the distance the disease might spread in the meantime. If the infection passes the outer border of the emergency control area (Figure 8.1), the management action will have failed. Conceptual Model Combining the existing knowledge on how effective, how fast, and by which pathways the virus might spread, regulators developed a strategy of “fencing the outbreak” within a ring of vaccinated animals (Figure 8.1). The appeal of the “living fence” is the gain in time: until the disease’s front wave hits the vaccinated ring-shaped area, sufficient protection can be established in the latter (i.e., enough animals baited and protected so that the infection chain might be cut in the ring area). This “ring strategy” thus seems logical and was favored in discussions among regulators. Ecological Model In order to test the proposed ring strategy, the underlying knowledge was transferred into a rule-based simulation model. A spatially explicit, individual-based modeling approach was needed because 1 single infected animal leaving the controlled area would matter. The temporal resolution was 1 week, as this is the period of rabies infectivity. The spatial unit corresponds to the home range of fox families represented by quadratic grid cells. The rabies transmission between fox family groups, contact transmission within fox families, and longer-distance transmission
70
Ecological Models for Regulatory Risk Assessments of Pesticides
Figure 8.1 Graphical depiction of the alternative management concept (light gray areas depict the proposed spatial configuration of treatment). In the circle/combat strategy (left panel) resources for countermeasures, for example, vaccination, are concentrated around the outbreak (large black cross). Then the infection quickly might reach (small light and dark gray crosses symbolizing 2 generations of disease transmission) and pass the outer border (encircled cross). In the ring/contain strategy, although based on the same resources, the latter “escape” case would be still in the controlled area.
via the individual dispersal of infected subadults were modeled stochastically (Thulke et al. 1999; Eisinger et al. 2005). The simulation area (i.e., 256 × 256 cells) corresponds to ~65.000 km². The detailed structure of the model, sources of parameterization, and validation and sensitivity considerations are described in Eisinger (2007). Virtual vaccination campaigns were scheduled according to the standard protocol: in the autumn and spring 20 baits are distributed per square kilometer. The treated area was always centered on the first detected rabies case (Figure 8.2). Results The simulation experiments clearly revealed that any ring-shaped approach would fail (Figure 8.3). The main drawback of the ring strategy was identified: the neglected or delayed control of the disease inside the region encircled by the ringshaped baiting area (Eisinger et al. 2005). In conclusion, the immediate control of a compact area around the detected case is mandatory for an emergency strategy. In the 2004 to 2006 emergency situation in Germany’s federal state Hessen, the ring strategy was planned to be performed as a field trial. Fortunately, the simulation study was available just in time to convince managers to use the compact, or circle, strategy. Applying the compact strategy finally gave the successful elimination of the disease from the area within 4 campaigns, as predicted by the model analyses. Thus, the ecological model substituted many unsuccessful field trials that would have had to have been undertaken if trusting the insufficient conceptual or mind model.
71
Ecological Models Supporting Management of Wildlife Diseases
Figure 8.2 (See color insert.) Screenshot of the model landscape of the rabies model, applied to compare alternative emergency control options. Pixels represent fox families that are either completely healthy (green), contain infected animals (red), contain at least 1 contagious animal (black), contain immunized animals (light blue), or are died off (white).
Risk of Breakout
100%
(10.000 Repetitions)
Ring
10%
1%
0%
Circle 0
2
4 6 Time (Campaigns)
8
10
Figure 8.3 The performance of the ring vaccination (dotted line) compared to the compact application of vaccination resources (solid line) around the disease outbreak (after Eisinger et al. 2005). Although in the very beginning the longer distance to the outer border of the control area delays some breakouts, the situation quickly changes when in the circle strategy control becomes successful. Already after 4 campaigns the circle approach had either failed or achieved eradication (saturation of the breakout risk). In the alternative scenario the disease inside the vaccinated ring is not controlled and spreads. The resulting pressure on the vaccinated ring zone then produces more breakouts before eradication is achieved.
72
Ecological Models for Regulatory Risk Assessments of Pesticides
Ecological Models Supporting Design of Field Application Here, we show how models can help optimize the design of field interventions with an example from management of the small fox tapeworm that parasitizes the red fox. Background The small fox tapeworm is a parasite of the red fox. The worm’s eggs are spread into the environment by feces of infested foxes. Subsequently, the eggs are ingested by voles in which a larval stage develops. When foxes prey on such infested voles, they ingest the larvae, which then mature into fecund worms. For humans, the concern arises if a human does not notice such an adhesive egg attached to berries or mushrooms. After ingestion, the larval stage develops in the human’s liver, where it can cause substantial damage, often with lethal consequences. Again, the method of intervention is medication of the foxes via baiting. However, the successful treatment only kills the worms in the fox; currently there is no immunity. Problem In the late 1990s a large-scale field trial was scheduled in a high-risk area of the federal state Brandenburg (Tackmann et al. 2001). Baits containing a deworming drug were regularly distributed in repeated campaigns, but the optimum frequency and duration of treatment were unclear. The ad hoc design of the control campaigns led to a drastically reduced parasite abundance in foxes, even below the statistical detection level. This seemingly indicated eradication (Tackmann et al. 2001). After the control program stopped, however, parasite abundance recovered quickly to the original situation; that is, the control turned out to be unsustainable. Conceptual Model The biannual schedule of antirabies vaccination was apparently insufficient for tapeworm control. Taking the worms’ ecology into account, it was assumed that the ideal approach would be to deworm foxes as fast as a new worm generation could mature in the gut of freshly infested foxes. This led to choosing 6-week intervals. Furthermore, after 1 year of application, leading to a drastic decline in worm abundance, control intensity was lowered to 12-week intervals to save costs. Ecological Model The model is formulated as a rule-based simulation model (Hansen 2001). It is grid based (Durrett and Levin 1994) and individual based (DeAngelis et al. 1994). It includes 4 types of entities (Figure 8.4): grid cells (voles’ home range, which could include infected fox feces), voles (relative abundance within grid cells), foxes (individuals), and worms (implicit as eggs and larvae, explicit as adults). The parasitic cycle was modeled via the predator–prey relationship between foxes and voles plus the infection of voles due to the defecation behavior of foxes (Figure 8.4). Numerical values for parameters were extracted from the literature. Simulation experiments were performed for 20-year periods. Scenarios with different treatment intervals and different total numbers of campaigns (duration of control program) were tested (Figure 8.5).
Ecological Models Supporting Management of Wildlife Diseases
73
Figure 8.4 (See color insert.) Schematic representation of the fox tapeworm model. Grid cells represent vole families; cylinders symbolize individual feces either loaded with eggs (red) or not containing eggs (black). Accordingly, if a vole family lives inside a defecated grid cell, it switches to infected for several weeks (pink vole). While the modeled individual foxes prey on the grid, they might randomly catch an infected vole from a grid cell where infestation was assigned formerly. The worms are developing during the next time steps and the cycle starts again after the next defecation (after Hansen 2001).
Prevalence After 20 Years (%)
50% 40% 30% 20% 10% 0% 0
2 4 6 8 Duration of Control Program (years)
10
Figure 8.5 Simulation results regarding the sustainability of strategies to control fox tapeworm infections in red fox populations. Shading symbolizes different intensity of application (interval between consecutive baiting campaigns: 2 weeks, 4 weeks, 6 weeks; field trial strategy [6 weeks in first year, 12 weeks in later years]). Simulations of control are performed with prolonging time period, that is, up to maximum of 10 years of repeated baiting campaigns (x-axis). Per assumed duration of control period points represent the corresponding end result found after 20 years in terms of average prevalence of tapeworm infestation in foxes (after Hansen et al. 2003).
74
Ecological Models for Regulatory Risk Assessments of Pesticides
Results The simulations revealed that the sustainability of a control program mainly depends on the duration of application of a particular control scheme (Figure 8.5). Thus, a minimum number of years is needed to eradicate the disease. Obviously, the higher the frequency of baiting, the faster eradication can be achieved, but a higher frequency is also more expensive. A detailed cost–benefit analysis of different control schemes is included in Hansen et al. (2001, 2003). The strategy applied in the field trial, which was based on the conceptual model described earlier, never was sustainable, no matter how long it was run.
Discussion Why was it beneficial to use ecological models in the 2 examples? Obtaining the same insights by performing field experiments would have been impossible. Such experiments would be far too expensive, take far too much time, would be hard to repeat, and would often be unfeasible for ethical reasons. Thus, the alternatives were either to base decisions on conceptual or mind models, or to use ecological models. Mind models are often not formulated explicitly, and simple reasoning is usually not sufficient for checking whether these models are appropriate and consistent. The advantage of ecological models is that all assumptions on how the ecological system works have to be made explicit; once translated to a set of equations or a computer program, the rigorousness of mathematics or computer logics is utilized to explore the consequences of the model assumptions in a rigorous way (Chapter 3). Why should managers take into account management recommendations that are based on results obtained from a simplifying representation of reality, that is, an ecological model? The representation could, although being explicit and using computer logics, be inappropriate, leading to the wrong conclusions. The key is to explain to the managers that models do nothing more than explore the logical consequences of the assumptions of their already existing conceptual models. In addition, modelers will of course have to do their “homework”: verification and validation (Chapter 3; Chapter 1), and sensitivity, uncertainty, and robustness analyses, which show how robust the model-based conclusions are (Rykiel 1996). For the rabies model, over a period of more than 10 years different model versions were used to address a suite of management questions. Confidence in the ecological models emerged from the demonstration of validity against a suite of well-known patterns observed in the real system: large-scale spatial “waves” of rabies prevalence, disease foci ahead of the wave front, and dynamics of prevalence at local and regional scales (Jeltsch et al. 1997; Tischendorf et al. 1998; Thulke et al. 1999). Particularly, the ecological model used to represent the concepts of efficient emergency control was shown to be “structurally realistic” (Chapter 3) because it made correct secondary predictions about long-term data of hunted fox (Eisinger and Thulke 2008). By structural realism we mean that key aspects of rabies epidemiology, fox ecology, and their interaction where captured in the model. This demonstration of validity encouraged managers to trust the results obtained from systematic simulations.
Ecological Models Supporting Management of Wildlife Diseases
75
Ecological modeling can thus support management decisions by making assumptions about how systems work explicitly. These assumptions reflect our current understanding and knowledge of the system. By using mathematics and computers, we explore whether these assumptions are complete and consistent, and whether they adequately predict observed properties and behaviors of the real system. Consequently, the assistance to be expected from ecological models for management decisions is not the complete or unbiased representation of the natural world, but rather a test bed for exploring the consequences of our current knowledge. Ecological models represent the expert’s current knowledge, but they do not necessarily replicate everything the experts conclude from their knowledge. With both examples we showed that improved understanding is one of the key benefits of using ecological models. Management decisions should not be based on mere numbers predicted by models, as these numbers often are quite uncertain due to uncertainty of parameters and model structure. More important for the understanding of what drives risk and how this risk can be managed is to see how the model outputs (numbers) change when we alter model assumptions and parameters. For example, once the 2 alternative emergency control strategies of rabies were implemented, it was possible to explain verbally why the ring strategy cannot be appropriate, without referring to specific numbers or parameter values. Likewise, once the model for the fox tapeworm existed, it became evident that too short control campaigns, or campaigns with too low treatment frequency, will fail because the parasite population cannot be eliminated from the controlled area. What can we learn from all this about the role ecological modeling can play in pesticide risk assessment? In both fields, management of wildlife diseases and pesticides is similar but, of course, not identical (Table 8.1). In both cases modeling population dynamics is required, that is, for nontarget species and for the pathogen’s host, respectively. However, nothing similar to the tiered approach in pesticide risk assessment exists in epidemiology. Pesticide risk assessment is thus much more “numbers driven,” including safety factors whose ecological significance is largely unknown. The lesson to be learned from ecological epidemiology is that yes, ecological models can lead to improved management decisions, even though these models usually do not make very precise quantitative predictions. Ecological models assist in making “mind models” explicit, and this sometimes provides the only pathway to explore their consequences. If field tests are limited, then decisions based on ecological models reflect understanding of how the system in question is organized and works, and are consequently more likely to be successful. Of course, quantitative risk assessment is still necessary, even when ecological models are used, but in addition to specific, absolute quantifications, a main area of ecological models could be relative risk assessments: Which application scheme is more efficient in terms of interest, for example, control efficiency, effects on nontarget organisms, and costs?
of the Art of 9 State Ecological Modeling for Pesticide Risk Assessment A Critical Review Volker Grimm, Pernille Thorbek, Amelie Schmolke, and Peter Chapman Contents Introduction............................................................................................................... 78 Meaning of Testing, Verification, Calibration, and Validation.................................. 79 Overview of Types of Population Models.................................................................80 Methods.....................................................................................................................80 Results....................................................................................................................... 81 Discussion................................................................................................................. 83 Appendix................................................................................................................... 85
While many ecological models addressing pesticide risk assessment can be found in the academic literature, they are rarely applied in pesticide registrations. In order to determine issues with existing models as published in literature, we assess the state of the art of ecological models addressing pesticide risk assessments. We focus on how toxic effects were modeled, how risk was quantified, and on the verification and validation of the models. From an extensive literature scan, 41 models were selected for review. Three main classes of models are used for pesticide risk assessment: difference and differential equations (14 out of 41 models), matrix models (17), and individual-based models (IBMs; 8). Additionally, 2 grid-based simulations were included. In the majority of models, toxicity was implemented as increased mortality, and risk was quantified by the effects on population dynamics. In some publications, both measures remained unclear. Apart from a subset of the IBMs, no validation was performed on the models. Another problem could be identified in the often ambiguous description of the models, and a lack of justification of the choice of model type. From this review of models as published in scientific literature, we conclude that most models do not address key issues of pesticide risk assessment sufficiently. We provide recommendations for the future use of ecological models in pesticide risk assessment. 77
78
Ecological Models for Regulatory Risk Assessments of Pesticides
Introduction Numerous ecological models addressing pesticide risk assessment have been published in the academic literature (e.g., Pastorok et al. 2001, 2003; Sibly et al. 2005). On the other hand, for pesticide registration it seems that ecological models so far have only been used occasionally. What is the reason for this discrepancy? One obvious reason is that the purposes of scientific publications and regulatory dossiers are quite different. In basic science, models are assessed with regard to their originality and the general insights they provide. In ecological applications including pesticide registrations, decision making requires that all protocols and tools used are transparent, standardized to some degree, and reliable enough to inform real-world decision making. With ecological models, however, there still is much skepticism about their value for decision making (e.g., Beissinger and Westphal 1998; Tannenbaum 2007). Some see modeling as a black box approach, others feel that models can be manipulated to give the results one wants, and there is also a perception that there might be very wide limits of uncertainty around the predictions given by models (Glaser and Bridges 2007). Are these feelings based on prejudice and ignorance, or do they reflect important issues that so far prevented ecological models from being routinely used for decision making in pesticide registration? To answer this question, we here take stock and assess the state of the art of the use of ecological models to support risk assessments of pesticides. Our working hypothesis is that so far the academic literature does not appropriately address key issues relevant for using models for decision support. For a model to be trusted, it must meet certain minimum criteria:
1) to prove a model’s worth, it must be validated and its ability to give useful answers needs to be ascertained; 2) the assumptions underlying the model need to be clear, and to achieve this, justification is needed for choice of model type in particular and for model design in general; 3) the model needs to be clearly and unambiguously described, for example, how are toxic effects modeled; and 4) the outputs from the models need to be relevant for the risk assessments in question.
We therefore scanned ecological models that have been used for risk assessment in publications for these 4 criteria. The aim of our stock taking was to prove that these key issues need to be addressed explicitly and in a systematic, standardized way. There needs to be a clearly defined set of criteria by which models can be assessed so that all stakeholders involved in pesticide registration gradually build up confidence in ecological models as a tool for decision making. In this way, we could eventually start utilizing the potential of ecological models to overcome the inherent limitations of risk assessments that focus on effects at the level of individuals (Chapter 2). Ecological models have already been used to great effect in various areas and have successfully informed decision making. Examples include models for silviculture
State of the Art of Ecological Modeling for Pesticide Risk Assessment
79
(Pretzsch et al. 2002), for the control of wildlife diseases (Eisinger and Thulke 2008), and for developing cost-effective conservation schemes (Drechsler et al. 2007; Ulbrich et al. 2008). In many of those cases, however, it took more than a decade to convince decision makers, and this was achieved by continually proving the worth of the models in practice (H. Pretzsch and H. Thulke, personal communication). In order to foster the use of ecological models for decision making in general, and for pesticide registration in particular, in a more strategic and effective way, we have to formulate general guidelines for model development, testing, communication, and application (Grimm et al. 2006). These guidelines would define a good modeling practice for applied ecological modeling. The purpose of the guidelines would not be to prescribe every single modeling step in detail, which is neither possible nor desirable, but to guarantee transparency: What, exactly, is the model? Why is it designed in a certain way? How can it contribute to decision making? And how are uncertainties dealt with? Our review is critical in the sense that we wanted to identify important impediments for a wider use of ecological modeling for pesticide registrations. We are not criticizing any single study per se; we acknowledge that most of the models we reviewed were not developed for directly contributing to pesticide registration. In the following, we first briefly define testing, verification, calibration, and validation because many different, and sometimes even contradictory, definitions exist, and because we need these terms for our review and for comparing model types. Then we give an overview of the main types of population models that can be used for risk assessment, and present our review.
Meaning of Testing, Verification, Calibration, and Validation By testing we refer to the process of making sure that a model is technically correct; that is, it was correctly implemented on computers so that the results are not artifacts, and it was correctly analyzed using appropriate and rigorous methods. With verification we refer to the process of making sure that the model “mimics the real world well enough for its stated purpose (Giere, 1991)” (Rykiel 1996, p. 230). This is usually achieved by comparing model output to data. With validation we refer to the process of making sure that we can place confidence “in inferences about the real system that are based on model results (Curry et al., 1989)” (Rykiel 1996, p. 230; note that Rykiel and many others combine verification and validation under one term, either “validation” like Rykiel or “verification”). Calibration means to tune parameter values of an existing model so that the match of model output and observations is optimized. Another word for calibration is “fitting.” Verification and calibration are inseparably bound to each other and can lead to quite impressive matches of model output and data. However, it is possible to fit a model that is structurally unrealistic, that is, reproduces the right patterns for the wrong reasons (Levin 1992). If so, we would have little confidence in the model’s predictions. Grimm and Railsback (2005; Grimm et al. 2005), therefore, propose
80
Ecological Models for Regulatory Risk Assessments of Pesticides
employing independent or secondary predictions for validation, that is, predictions of features of patterns in the real system that were detected during model analysis, but not used or preferably not even known during model development and calibration. For example, the beech forest model BEFORE was verified by comparison to spatiotemporal patterns of so-called developmental stages (Rademacher et al. 2004). The model was then validated by detecting patterns in the age–structure of the model’s canopy trees and comparing these patterns to real data that were not considered before by the developers of BEFORE.
Overview of Types of Population Models In theoretical population ecology, there is a broad spectrum of model types. Although in principle this spectrum is a continuum, some major types can be identified. Different classifications exist (Grimm and Railsback 2005, Chapter 10), but in the context of pesticide risk assessment, it is sufficient to distinguish 3 main classes: differential and difference equations, matrix models, and individual- or agent-based simulation models (IBMs or ABMs). These model types are described in more detail in Chapters 3, 5, 6, and 7. All three of these main model types have their specific strengths and weaknesses. No single type of model serves all possible purposes of a modeling equally well, but every model type has its main field of application. Differential and difference equations are good for exploring general ideas and principles, but they are often hard to validate. Matrix models are designed for projections, not for predictions; that is, they extrapolate the consequences of current environmental conditions into the future. Consequently, matrix models are verified, not validated. An important advantage of matrix models is that they can, based on quite limited and easily obtained data, compare the impact a certain exposure will have on species with different life cycles and reproductive patterns (e.g., Stark et al. 2004). IBMs and ABMs, finally, are very flexible and can include a wide range of factors thought to be important for population dynamics. They are the most demanding model type in terms of data needed for parameterization. They can, if designed for this purpose, be used for predictions and thus be validated, but their development and testing are time consuming and can be limited by a lack of sufficient data. Often, they are neither fully communicated nor analyzed to the point that their main results are understood.
Methods We searched for publications of the years 2000 to May 2007 that included ecological population models for pesticide risk assessment. First, the databases BIOSIS, CAB Abstracts, and PASCAL were searched for papers, including a combination of the following keywords in title, abstract, or keywords: (risk assessment) AND (simulation? OR model$) AND (population? OR ecolog$) AND (crop protection product OR pesticide OR insecticide OR fungicide OR herbicide).
State of the Art of Ecological Modeling for Pesticide Risk Assessment
81
Of the publications fulfilling these conditions, the titles and abstracts were scanned to identify relevant publications. These were then reviewed and the following characteristics were assessed: • Purpose • Model type (type and justification for choice of model type)* • Scales (resolution and extent of the representation of time and, if applicable, space) • State variables (variables characterizing the model’s entities) • Processes (i.e., what processes are represented in the model?) • Scheduling (applying to simulation models only: Which entity is doing what at what time and in what sequence? When are state variables updated?) • How is toxicity modeled?* • Other factors or stressors included (e.g., temperature or predation) • Any tests against independent data (if yes, details)?* • Parameterization sources, that is, how much guesswork? • How is risk quantified? (method and endpoint, i.e., aggregated output variable)* • Paper’s main conclusion • Notes Here, we quantified the set of publications regarding the 4 characteristics marked by an asterisk. A more detailed analysis of, for example, model structure, purpose, and conclusions, will be presented elsewhere (Schmolke et al. in press).
Results The search of the BIOSIS, CAB Abstracts, and PASCAL databases yielded 444 publications. After scanning titles and abstracts, 73 publications were identified as potentially relevant for our review. We cross-checked these publications with our own databases, and found that our filters did not detect all relevant publications, for example, Stark et al. (2004). We included 19 additional publications from our own databases, ending up with 92 publications. After the detailed evaluation of these publications, 39 publications were selected for our quantitative assessment of the state of the art. One of these publications, Bartell et al. (2003), contained 3 different models, adding up to a number of 41 models evaluated in our review. Overall, we believe that our sample should be representative, although we certainly were not able to detect all publications of the years 2000 to May 2007 that are relevant for our review. Of the 41 models, 14 (34%) were differential equations, 17 (41%) matrix models, 8 (20%) IBMs, and the 2 remaining were grid-based simulation models (Figure 9.1; in their simplest form referred to as “cellular automata”; Wissel 2000). Note that five of the IBMs were developed within the ALMaSS framework (Chapter 7). Some models used more than 1 measure of toxicity. We found 31 models (75%) where toxicity was represented by increased mortality, which, however, usually was representing only 1 exposure concentration matched by 1 mortality figure (Figure 9.2a). In 7 cases sublethal effects were considered, 2 cases included specific
82
Ecological Models for Regulatory Risk Assessments of Pesticides
Matrix models 41%
Other 5%
Differential equations 34% IBMs 20%
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.
toxicity measures (dose–response curves, DEBtox), in 6 cases other measures were used, or it remained unclear how toxicity was represented. Different quantifications of risk were also possible within the same model. For 17 models (41%) the effect of the pesticide at the population level was quantified via effects on population dynamics, that is, total population size, mostly by presenting Population dynamic N(t)
Increased mortality
Population growth rate
Sublethal effects
Recovery Specific measures
Risk of a certain effect
Unclear or none 0
5
10
15
20
Unclear 25
Number of Models (a)
30
35
0
2
4
6
8
10
12
14
16
18
Number of Models (b)
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. Note that some publications made use of more than 1 toxicity endpoint and more than 1 measure of risk.
State of the Art of Ecological Modeling for Pesticide Risk Assessment
83
No Verification or validation 73% Validation 7% Calibration or comparison 19%
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.
population size with and without pesticide application (Figure 9.2b). For 9 models (22%), which were all matrix models, effects on the population growth rate λ were determined. Recovery was explicitly quantified in 4 models and the risk (i.e., probability) of a certain effect on population size or λ was considered in 2 models. In 7 models, the quantification of risk remained unclear. Out of the 30 models with no reference to verification or validation (Figure 9.3), 26 were differential equations or matrix models. For 8 models, some kind of verification or calibration was presented, that is, some evidence that the model was able to reproduce observations, for example, population dynamics or structure. Only for 3 models validations via independent predictions were presented. All three of these models were IBMs developed within the ALMaSS framework (Chapter 7).
Discussion Our aim was to obtain a quantitative overview of ecological population models used for questions regarding risk assessment of pesticides that were published in the academic literature. Matrix models were the most dominating model type, closely followed by differential or difference equations (Figure 9.1). Those model types can be combined due to their very simple structure, representing 75% of the models reviewed here. Simple models have a high potential to generate understanding and projections. However, they are usually not designed to be verified or validated, although in principle this is possible to some degree. For example, if a model predicts that a certain level of sublethal effects will cause no reduction in overall population size, then that could be tested against independent data. Nevertheless, in 73% (30) of the models
84
Ecological Models for Regulatory Risk Assessments of Pesticides
reviewed here, verification and validation were not considered at all (Figure 9.3). We find this problematic in terms of creating confidence in the reliability of ecological modeling in a regulatory context. We found that 20% of the models were IBMs, but that likely is an overestimation. Five of the 8 IBMs were developed within the ALMaSS framework, and thus rather reflect the work of 1 very active group than the general usage of IBMs for pesticide risk assessment. Regarding the quantification of risk or, more generally, the effects of pesticides, we found only a few cases where measures were used that directly quantified risk. In most cases, possible effects on population dynamics were presented. This is helpful for the demonstration but not for the quantification of effects and, in particular, risk. Likewise, the quantification of toxicity was often only specified as a mortality rate, and in some cases it was not explained how this mortality rate was calculated. In most instances, only increased mortality was considered, for example, mortality following constant exposure, although one of the potentials of population models is to consider sublethal effects, effects of different exposure patterns, and time-dependent effects. Model descriptions were not always clear and unambiguous; for example, for some models it was not explained how toxicity was represented, and in many it would not be possible to repeat the work as the description of, for example, central algorithms, was lacking and/or parameter values were not disclosed. Moreover, for quite a few papers, not enough thought has gone into how the model outputs can be used to quantify the risk. This is perhaps expected, as these are academic publications, but it is paramount for use in regulatory risk assessments. For a model to be easily assessed, its assumptions should be clearly communicated. This is true for many steps along model development, and starts already with the choice of model type, as different types of models have different underlying assumptions. In general we found that for most models the choice of model type had not been justified. Thus, quite a diverse range of models is used for answering essentially very similar questions. While this diversity can be productive from an academic point of view, it is not for regulatory risk assessments. Because so far there are no agreed, established criteria for knowing when a model is to be trusted, the diversity of model types used for similar questions would increase the workload of the regulatory authorities, who would have assess to a very wide range of models and model types. To support decision making in pesticide registration, similar questions should lead to similar solutions, that is, to use similar model types; or alternatively, it should be demonstrated that different model types lead to similar results. To summarize, our stock taking of published population models for pesticide risk assessment confirmed our working hypothesis that so far the academic literature does not appropriately address key issues relevant for using models for decision support. This is not surprising because many, if not most, models published in the literature were developed for addressing basic scientific questions rather than directly supporting decision making. However, so far the academic literature was the main source of information for stakeholders from industry and regulatory authorities who wanted to learn about the potential use of ecological models for pesticide registration. Thus, the current hesitation and lack of confidence regarding ecological models to a large degree reflect the state of the art of the literature.
State of the Art of Ecological Modeling for Pesticide Risk Assessment
85
We conclude with recommendations for modelers and regulators. Modelers should describe their models thoroughly and, preferably, using a common format (e.g., the ODD [Overview–Design concepts–Detail] protocol for individual-based models; Grimm et al. 2006). They should explain why they used a certain model type for answering a certain question. They should quantify risk in a way that is relevant for regulators and represent toxicity in a way that utilizes the potential of ecological models to go beyond simple assumptions in lower-tier risk assessments. And last but not least, the issues of testing, calibration, verification, and validation should be addressed explicitly. Regulators would benefit from training in the basic rationale of ecological modeling (Chapter 3), which would enable them to understand the pros and cons of the different model types. The regulators should check whether modelers followed the recommendations presented in the previous paragraph. They should require transparency at all levels, but they also should state as clearly as possible what they expect from the models and modelers, for example, what kind of data (e.g., field or laboratory) they would require for accepting the model for decision making. Ideally, these recommendations to modelers and regulators should be combined and detailed in general guidelines for modeling for regulatory ecological risk assessment of pesticides. These guidelines, which would define a “good modeling practice” in the regulatory context, should be formulated in collaboration with all stakeholders involved (Chapter 1). One last but important issue is terminology: we found it very difficult to find the relevant papers by searches in the databases. In particular, the word “model” is used in so many different ways and contexts that it was hard to find the type of dynamic, ecological population models we were interested in. We therefore propose to use a unified terminology by making sure that the keyword “ecological model” appears in title, abstract, or keyword list, no matter what specific type of population model has been used.
Appendix Listed are the references to the 39 publications evaluated in our review. Adams BM, Banks HT, Banks JE, Stark JD. 2005. Population dynamics models in plant– insect herbivore–pesticide interactions. Math Biosci 196:39–64. Bartell SM, Campbell KR, Lovelock CM, Nair SK, Shaw JL. 2000. Characterizing aquatic ecological risks from pesticides using a diquat dibromide case study III. Ecological process models. Environ Toxicol Chem 19:1441–1453. Brown AR, Riddle AM, Winfield IJ, Fletcher JM, James JB. 2005. Predicting the effects of endocrine disrupting chemicals on healthy and disease impacted populations of perch (Perca fluvialis). Ecol Model 189:377–395. Byers JA, Castle SJ. 2005. Areawide models comparing synchronous versus asynchronous treatments for control of dispersing insect pests. J Econ Entomol 98:1763–1773. Chandler GT, Cary TL, Bejarano AC, Pender J, Ferry JL. 2004. Population consequences of fipronil and degradates to copepods at field concentrations: an integration of life cycle testing with Leslie matrix population modeling. Environ Sci Technol 38:6407–6414.
86
Ecological Models for Regulatory Risk Assessments of Pesticides
Choi WI, Moorhead DL, Neher DA, Il Ryoo M. 2006. A modeling study of soil temperature and moisture effects on population dynamics of Paronychiurus kimi (Collembola: Onychiuridae). Biol Fertil Soils 43:69–75. Davidson RS, Armstrong DP. 2002. Estimating impacts of poison operations on non-target species using mark-recapture analysis and simulation modeling: an example with saddlebacks. Biol Conserv 105:375–381. Demyanov V, Wood SN, Kedwards TJ. 2006. Improving ecological impact assessment by statistical data synthesis using process-based models. Appl Statist 55:41–62. Forbes VE, Calow P, Sibly RM. 2001. Are current species extrapolation models a good basis for ecological risk assessment? Environ Toxicol Chem 20:442–447. Gosselke U, Triltsch H, Rossberg D, Freier B. 2001. GETLAUS01: The latest version of a model for simulating aphid population dynamics in dependence on antagonists in wheat. Ecol Model 145:143–157. Gurney WSC. 2006. Modelling the demographic effects of endocrine disruptors. Environ Health Perspect 114:122–126. Gutierrez AP, Ponsard S. 2006. Physiologically based demographics of Bt cotton-pest interactions. I. Pink bollworm resistance, refuge and risk. Ecol Model 191:346–359. Jepsen JU, Topping CJ, Odderskaer P, Andersen PN. 2005. Evaluating consequences of landuse strategies on wildlife populations using multiple-species predictive scenarios. Agric Ecosyst Environ 105:581–594. Kimmerer WJ, Cowan JHJ, Miller LW, Rose KA. 2001. Analysis of an estuarine striped bass population: effects of environmental conditions during early life. Estuaries 24:557–575. Kriticos DJ, Brown JR, Maywald GF, Radford ID, Nicholas DM, Sutherst RW, Adkins SW. 2003. SPAnDX: a process-based population dynamics model to explore management and climate change impacts on an invasive alien plant, Acacia nilotica. Ecol Model 163:187–208. Kriticos DJ, Dowling PM, King WM, Sheppard AW, Jones RE. 2002. 4SPaM: a process-based and weather-driven population model for managing weeds in temperate pastures. In: 13th Australian Weeds Conference: weeds “threats now and forever?” Perth, Western Australia, September 8 to 13, 2002. Victoria Park (Australia): Plant Protection Society of Western Australia, p. 244–247. Lopes C, Pery ARR, Chaumot A, Charles S. 2005. Ecotoxicology and population dynamics: Using DEBtox models in a Leslie modeling approach. Ecol Model 188:30–40. Mamedov A, Udalov S. 2002. A computer tool to develop individual-based models for simulation of population interactions. Ecol Model 147:53–68. Miller DH, Ankley GT. 2004. Modeling impacts on populations: fathead minnow (Pimephales promelas) exposure to the endocrine disruptor 17β-trenbolone as a case study. Ecotoxicol Environ Safety 59:1–9. Murdoch AJ, Watson SJ, Park JR. 2003. Modelling the soil seed bank as an aid to crop management in integrated arable farming systems. In: The BCPC International Congress: crop science and technology (Proceedings of an international congress held at SECC, Glasgow, Scotland, November 10 to 12, 2003). Vols. 1 and 2. Alton (UK): British Crop Protection Council, p. 521–526. Naito W, Miyamoto K, Nakanishi J, Masunaga S, Bartell SM. 2003. Evaluation of an ecosystem model in ecological risk assessment of chemicals. Chemosphere 53:363–375. Niquil N, Kerleguer G, Leguerrier D, Richard P, Legrand H, Dupuy C, Pascal PY, Bacher C. 2006. How would the loss of production due to an herbicide have repercussions in the food web of an intertidal mudflat? Sensitivity analysis of an inverse model for Brouage mudflat, Marennes-Oleron Bay, France. Cahiers de Biologie Marine 47:63–71.
State of the Art of Ecological Modeling for Pesticide Risk Assessment
87
Pertoldi C, Topping C. 2006. Impact assessment predicted by means of genetic agent-based modeling. Crit Rev Toxicol 34:487–498. Raimondo S, McKenney CLJ. 2005. Projected population-level effects of thiobencarb exposure on the mysid, Americamysis bahia, and extinction probability in a concentrationdecay exposure system. Environ Toxicol Chem 24:564–572. Raimondo S, McKenney CLJ. 2006. From organisms to populations: modeling aquatic toxicity data across two levels of biological organization. Environ Toxicol Chem 25:589–596. Spromberg JA, Meador JP. 2005. Relating results of chronic toxicity responses to population-level effects: modeling effect on wild chinook salmon populations. Integr Environ Assess Manage 1:9–21. Spromberg JA, Meador JP. 2006. Relating chronic toxicity responses to population-level effects: a comparison of population-level parameters for three salmon species as a function of low-level toxicity. Ecol Model 199:240–252. Stark JD, Bamfo S. 2002. Population-level outcomes of differential susceptibility among life stages of the aphid parasitoid Diaeretiella rapae to pesticides. In: Proceedings of the 1st International Symposium on Biological Control of Arthropods, Honolulu, Hawaii, January 14–18, 2002. Forest Service (WA): US Department of Agriculture, p. 314–317. Stark JD, Banks JE. 2004. Estimating susceptibility of biological control agents to pesticides: influence of life history strategies and population structure. Biol Control 29:392–398. Stark JD, Banks JE, Vargas R. 2004. How risky is risk assessment: the role that life history strategies play in susceptibility of species to stress. Proc Natl Acad Sci USA 101:732–736. Tang S, Xiao Y, Chen L, Cheke RA. 2005. Integrated pest management models and their dynamical behaviour. Bull Math Biol 67:115–135. Thorbek P, Topping CJ. 2005. The influence of landscape diversity and heterogeneity on spatial dynamics of agrobiont linyphiid spiders: an individual-based model. Biocontrol 50:1–33. Topping CJ, Hansen TS, Jensen TS, Jepsen JU, Nikolajsen F, Odderskaer P. 2003. ALMaSS, an agent-based model for animals in temperate European landscapes. Ecol Model 167:65–82. Topping CJ, Odderskaer P. 2004. Modeling the influence of temporal and spatial factors on the assessment of impacts of pesticides on skylarks. Environ Toxicol Chem 23:509–520. Traas TP, Janse JH, Van Den Brink PJ, Brock TCM, Aldenberg T. 2004. A freshwater food web model for the combined effects of nutrients and insecticide stress and subsequent recovery. Environ Toxicol Chem 23:521–529. Vale GA, Grant IF. 2002. Modelled impact of insecticide-contaminated dung on the abundance and distribution of dung fauna. Bull Entomol Res 92:251–263. Wang G, Edge WD, Wolff JO. 2001. Demographic uncertainty in ecological risk assessments. Ecol Model 136:95–102. Watanabe K, Yoshimura C, Omura T. 2005. Stochastic model for recovery prediction of macroinvertebrates following a pulse-disturbance in river. Ecol Model 189:396–412. Wennergren U, Stark J. 2000. Modeling long-term effects of pesticides on populations: beyond just counting dead animals. Ecol Appl 10:295–302.
Role of Ecological 10 The Modeling in Risk Assessments Seen from an Academic’s Point of View Valery E. Forbes Contents Introduction...............................................................................................................90 Advantages of Using Ecological Models in Risk Assessment..................................90 Disadvantages of Using Ecological Models in Risk Assessment............................. 93 Model Requirements for Regulatory Purposes, Especially the Registration of Pesticides...................................................................................................................94 Challenges for the Future.......................................................................................... 95 Conclusion................................................................................................................96
In this chapter, I explore how ecological modeling can contribute to pesticide risk assessment. I argue that, despite some challenges that need to be addressed, increased application of ecological models has the potential to greatly improve our estimates of risk and can do so in a cost-effective manner. Ecological models cannot and should not be expected to entirely replace laboratory and field studies, and the costs associated with developing and applying models need to be balanced against the benefits that they can provide for risk assessment. As a general rule, a model should be no more complex than necessary to address the questions of interest. Thus, different types of ecological models may be best suited to different types of risk assessment questions. An important challenge is to demonstrate, clearly and convincingly, that the use of ecological models leads to substantially better risk assessments than those provided by current procedures. Overcoming this challenge will require development of a carefully designed research program, increased training in model use and interpretation, and active involvement of all stakeholders.
89
90
Ecological Models for Regulatory Risk Assessments of Pesticides
Introduction The overall aim of regulatory risk assessment of pesticides is to quantify the likelihood that such chemicals may cause undesirable impacts on ecosystems and human health. This information ideally allows risks to be managed while allowing society to benefit from products that facilitate cost-effective plant production. For good reasons, legislative procedures for registering and managing pesticides are more stringent than those for other industrial chemicals. Pesticides are designed to kill or repel at very low exposure concentrations, and they are intentionally applied to ecosystems for this purpose. Yet despite the greater stringency of regulatory risk assessment procedures for pesticides, compared to other chemicals, there is very little ecology involved in the assessment process. Assessment of ecological effects is primarily based on the results of single-species laboratory tests, in which links between measured test results and ecological protection targets are made by applying safety factors to test results, or by fitting the test results to a statistical distribution — both of which approaches make fairly profound (and largely implicit) assumptions about how ecological complexities are likely to influence risk in real ecosystems. Indeed, extrapolation from the handful of test results that usually form the basis of effects assessments to likely impacts on intact ecosystems and their parts is one of the greatest challenges — and least scientifically robust parts — of current chemical legislative procedures. In the remainder of this chapter I shall explore how ecological modeling can contribute to chemical risk assessment. I shall argue that, despite some challenges that need to be addressed, increased application of ecological models has the potential to greatly improve our estimates of risk, and can do so in a cost-effective manner. My focus will be on population models of various complexities and on how such models can improve regulatory risk assessments of pesticides. Many types of population models have been developed; these may be deterministic or stochastic (Maltby et al. 2001), and they may range from general models based on very simple stage- or agebased structure that ignore ecological complexities (such as density dependence and migration; Calow et al. 1997) to advanced individual-based models that incorporate a wealth of species- and habitat-specific detail (Topping et al. 2003). It is beyond the scope of the present chapter to review the range of population models available for use in pesticide risk assessment (but see Chapter 9; Pastorok et al. 2001). Rather, my aim is to discuss the advantages and limitations of population models in general.
Advantages of Using Ecological Models in Risk Assessment Ecological risk assessment differs importantly from human health risk assessment in that the latter attempts to extrapolate effects of chemicals on a handful of model species to 1 target species (humans), whereas the former attempts to extrapolate effects on a handful of model species to all species in potentially exposed ecosystems. Even if it were possible to test all species in all ecosystems (which it is not), there would be both moral and economic reasons for not wishing to do so. Ecological models are the only practical way to cover the range of situations for which we wish to estimate risk.
The Role of Ecological Modeling in Risk Assessments
91
This is admittedly a strong statement, but it is based on the acknowledgment of the diversity, complexity, extent, and slow dynamics of most ecosystems. Modeling has become an indispensable tool for ecology and is employed for addressing all kinds of theoretical and applied problems. There is no reason to believe that pesticide risk assessment should be an exception. Although the influence of relevant biotic and abiotic variables on the fate and effects of chemicals can, to a certain extent, be explored through controlled experiments and observations of natural systems, the combinations of factors that can be tested in practice are very limited. Mesocosm and field studies are often expensive to perform, can be difficult to replicate sufficiently, and are frequently complicated to interpret. Because they typically represent one unique scenario (species composition and density, temperature, light, nutrient level, and timing of pesticide application in relation to the environmental conditions), questions are often raised about the generality and robustness of the results. Likewise, unexpected or uncontrollable events may occur (e.g., it may have been an unusually rainy, sunny, hot, or cool season), the influence of which on the estimate of risk can be difficult to assess. Models, once they have been tested, verified, and validated to a sufficient degree (Chapter 9), have the important advantage of allowing simulation of all possible relevant scenarios and can be a highly cost-effective tool for exploring how changes in selected environmental variables are likely to influence risk (see, e.g., Chapter 7, and the discussion of IBMs in Chapter 3). This can be useful information in itself, but can also be used to optimize the design of experimental work. As ecologists, we may be concerned that current legislative risk assessment procedures ignore many of the ecological complexities that may influence risks of chemicals to populations in the field. An effective way to address these concerns is to use models to explore the influence of key biological (e.g., life history, population density) and environmental (e.g., habitat structure and complexity, climate change) variables on target populations and the risks of chemicals to them. In other words, we can use model simulations to ask “if–then” questions. If a pesticide is applied before, rather than after, the main reproductive period of nontarget species X, would recovery be faster or slower? How much would risk of pesticide Y to species X depend on seasonal rainfall pattern? On temperature? On patchiness of species X’s habitat with respect to the sprayed area? Whereas we may recognize, and be concerned by, the lack of ecology in current regulatory risk assessment procedures, it is likely that many of the ecological complexities of natural systems have little or no influence on the risks of chemicals to ecosystems and the populations in them. Or it may even be the case that incorporation of more ecological realism reduces risk (compared to current regulatory assessments) and adds an extra margin of safety to current simplified risk assessments. Would this not be useful to know? And would it not give us greater confidence in the “ecologically naïve” approaches in current use? Alternatively, if certain ecological complexities are found to exacerbate risk, and thus lead to underprotective risk assessments if ignored, we could improve present approaches to take these into account. Ecological models can offer a powerful approach for addressing such issues and are the only way that ecological complexities (to the extent necessary) can be incorporated into risk assessment protocols in a quantitative and consistent manner.
92
Ecological Models for Regulatory Risk Assessments of Pesticides
A very useful application of ecological models is for obtaining insight into the mechanisms behind observed or predicted ecological impacts. For example, population dynamics models are often used in combination with elasticity and decomposition analyses (Caswell 2000; De Kroon et al. 2001). Elasticity analyses can be used to determine the relative sensitivity of population dynamics to changes in different life history traits. If different chemicals target different life-history stages or variables, elasticity analysis can help to predict which of the chemicals is likely to have the greatest (or least) impact on population dynamics. Decomposition analysis is used retrospectively to quantify the contribution of different life-history traits to observed impacts (e.g., of chemicals) on population dynamics. The contribution is determined by a combination of the demographic importance of a trait (its elasticity) and its sensitivity to chemical exposure. Some traits may be very sensitive to chemical exposure, but effects on them may have less of an impact on population dynamics than expected if they have a low elasticity. Another important advantage of employing ecological models in risk assessment is that they provide outputs that are closer to the protection goals of the assessment (i.e., populations, communities, and ecosystems) than other approaches. Application of even very simple population models provides a better measure of response to chemicals than the individual-level endpoints (survival, reproduction, development) most widely in use for risk assessments today, not the least because such models integrate effects on all of the key life-history traits that determine population-level attributes (Forbes and Calow 1999). Although it should be desirable to incorporate as much ecology as necessary in risk assessments, increasing efforts to reduce animal testing in combination with the increasing availability of sensitive biochemical and molecular assays is leading to a trend toward less, rather than more, ecology in risk assessment. Such a trend is likely to add, rather than reduce, the uncertainty in assessment of effects because the data on which the risk assessments are based are further removed from what one is trying to protect (Forbes et al. 2006). For example, there is increasing interest in using microarrays and other molecular biological methods to assess risk. Microarrays are being developed that measure the upregulation and downregulation of selected genes in vitro. But there are profound uncertainties involved in linking the expression (or lack thereof) of individual genes in model species exposed to a chemical in a Petri dish to adverse impacts on populations of other (untested) species in nature. The uncertainties associated with these alternative test methods can lead to both false positives (overestimates of risk) and false negatives (underestimates of risk), and it is therefore difficult to see how the application of such methods is likely to improve risk assessment. If we want to do a better job of assessing the risks of pesticides and other human impacts on ecological systems, our focus has to be on developing models that are grounded in ecological theory that can directly and explicitly incorporate the ecological entities that the assessment aims to protect.
The Role of Ecological Modeling in Risk Assessments
93
Disadvantages of Using Ecological Models in Risk Assessment Clearly ecological models are not a “silver bullet” that will solve all of the challenges faced in assessing the risks of toxic chemicals to ecological systems. Ecological models, whether simple or complex, cannot (and should not) be expected to entirely replace laboratory and field studies and so may seem to make the registration process more demanding, particularly when they are first implemented into regulatory protocols. However the costs associated with including ecological models in risk assessments need to be balanced against the increasing confidence that they can provide in the decision-making process. Ecological models, like all models, are simplifications of reality and are subject to the constraints and assumptions imposed by the modeler. A disadvantage, particularly for more complex models, is that their assumptions and constraints can be difficult to check and their outputs problematic to validate. Alternatively, it can be argued that in simple models much of the complexity is hidden or at least not made explicit. Although complex models might be more difficult to test, they have a much higher potential for validation than structurally too simple models (Chapter 3). However, the more complex a model, the more difficult it usually is to communicate. Although, as mentioned, the outputs of ecological models are closer to ecological protection targets, nevertheless the results may not provide a direct estimate of risk. For example, in much of our own work, we have used the simple 2-stage matrix model of Calow and Sibly (1990) in combination with laboratory-based life table response experiments (Levin et al. 1996; Hansen et al. 1999). The output of the model is a measure of population growth rate (λ), which, as Caswell (2001) emphasizes, should not be used to predict some concrete future population size, but which rather provides important information about the current state of the population. We have compared differences in λ between control and chemically exposed populations to determine the exposure concentration at which there is a statistically significant impact on population dynamics. But how should such results be extrapolated to provide an estimate of risk for field populations? Can we assume that the same impairment in λ will result in the same risk (e.g., of extinction) to different species? No, because this will depend on life history. Can we assume that the same impairment in λ will be equally as risky in a growing as in a shrinking population, or in a highly variable as in a rather stable population? No, probably not. Is λ the appropriate population-level endpoint on which to focus, or are other endpoints (e.g., equilibrium population size) more appropriate in certain situations? This is an area in which more research is needed. So although the outputs of ecological models may get us closer to ecological protection targets than other effects’ endpoints, translating model outputs into appropriate measures of risk is not necessarily straightforward. A somewhat related issue has to do with the skills needed for ecological modeling. Even simple ecological models require some proficiency in mathematics and/or computer programming. Although most ecologists probably have at least some familiarity with matrix algebra and can compute a simple life table, the mathematics involved in matrix population models can quickly become daunting for most of us (present company included!). Although it is possible to produce simple individual-based models
94
Ecological Models for Regulatory Risk Assessments of Pesticides
without much programming expertise, using freeware such as NetLogo (Wilensky 1999), more complicated IBMs, such as ALMaSS (Topping et al. 2003), are highly demanding and require a sophisticated degree of programming skill to develop, apply, and interpret. Such skills are not widespread among ecologists or regulators.
Model Requirements for Regulatory Purposes, Especially the Registration of Pesticides In order for ecological models to become a standard part of regulatory risk assessment protocols, a number of requirements must be met. First, both the assumptions and the constraints of the models must be understood and articulated (i.e., there must be transparency). In addition, the outputs must be interpretable in a form that can be used for decision making. In practice, risk assessment of pesticides is kept separate from risk management, and thus risk assessment focuses on quantifying likely impacts of pesticides, whereas management involves decisions about what impacts are acceptable or not. For the results of ecological models to provide useful information for risk assessment, it is necessary that value judgments can be made about them (see also Chapter 9). Many ecological attributes and processes may be modeled, but they are not all equally important for maintaining the long-term sustainability of ecosystems. A serious weakness in current pesticide risk assessment protocols is that the outputs of the exposure and effects assessments are not directly comparable. In particular, the type of exposure concentration provided by the exposure assessment is not that which gives the best correlation to ecotoxicological effects (Boesten et al. 2007). Time-varying exposure concentrations are the rule rather than the exception for most pesticides under realistic field conditions. When expressing an effect concentration for mesocosm studies, for example, (in which exposure frequently varies with time), it is often unclear whether the effect concentration should be expressed in terms of the initial exposure concentration, the peak concentration, a time-weighted average concentration, the concentration measured at the time the effects were detected, or some other measure. Which exposure measure is chosen can have an important influence on the outcome of the risk assessment (Brock et al. forthcoming). For ecological models to contribute usefully to assessing risk, they should be flexible enough to handle variable exposure scenarios, and the estimates of exposure and ecological effect must be expressed in units that are directly comparable. The models should also be flexible enough that they can be easily adapted to different species and ecosystem types. As a general rule, a model should be no more complex than necessary to address the questions of interest. This is likely to mean that different types of ecological models are more or less appropriate for different tiers in regulatory risk assessment protocols (Pastorok et al. 2001). It also is likely to require that the questions addressed by the different tiers need to be articulated more explicitly in ecological terms. Sometimes the questions asked by risk assessments are general and ecologically vague. For example, Directive 91/414/EEC (European Commission 2002a,
The Role of Ecological Modeling in Risk Assessments
95
2002b, 2002c) requires that plant protection products be assessed in terms of their potential to pose “unacceptable” risks to nontarget aquatic and terrestrial species in general. For this type of question, a few species and exposure scenarios are chosen to represent aquatic and terrestrial species in general, and the major challenge is to extrapolate from the specific conditions of the tests performed to ecological systems in general. It is unlikely in such cases that the use of complex and highly species-specific models applied to the tested species will provide a better estimate of overall risk to nontarget species in general. In other cases, the questions addressed in risk assessments may be highly site and/or species specific. For example, in retrospective risk assessments, the aim is often to identify causes of observed impacts on defined populations occupying an actual habitat. For these types of risk assessment it is feasible and often necessary to employ models that incorporate much more of the relevant ecological and biological details because these can be clearly defined and their incorporation can reduce the uncertainty in the risk assessment. It may also be important to decide whether the model output will be used to identify exposure conditions that are protective (e.g., if the model output gives a result less than X, there are unlikely to be impacts on field populations at exposures equal to or less than the one modeled) or whether the results are intended to be used in a more predictive manner (e.g., the model might be used to predict the time necessary for full population recovery following a defined pesticide application scenario). Whether the model result should be predictive, or merely protective, is likely to place different constraints on the model’s complexity. Regardless of what types of models are to be used, careful attention has to be paid to adequately testing, verifying, and validating them (see Chapter 9).
Challenges for the Future An important hurdle that needs to be overcome before ecological modeling can become a credible tool in risk assessment relates to the skepticism and suspicion of models in general. This can perhaps be facilitated by starting to incorporate very simple models that require a minimum of mathematical or programming skills, that have a long history of ecological theory to support them, and whose constraints and assumptions are clearly understood. However, although simple models are most useful for addressing general questions, they are usually not designed for making specific testable predictions of real systems. Ultimately, the best way to overcome potential model skepticism is to demonstrate, clearly and convincingly, that the use of ecological models leads to substantially better risk assessments than those provided by current procedures. This could involve a series of case studies in which both existing risk assessment predictions and risk assessments incorporating relevant modeling results are compared with carefully designed field studies. A way forward may be to agree on a suite of relevant risk assessment questions, where standard methods fail, and then to carefully select the most appropriate model type for each question, while making the limitations of the selected model very explicit. As noted, it is essential to have an understanding of and consensus on how much complexity is needed in ecological models to provide the answers necessary for risk
96
Ecological Models for Regulatory Risk Assessments of Pesticides
assessment. This is an issue in need of urgent attention. In addition, there is a need for agreement on the roles of models in the risk assessment process (At what tiers should they be employed? Will they replace or add to current steps in the process? What happens when model results are not consistent with other lines of evidence?). These are nontrivial questions that need to be thoroughly debated by all stakeholders. On a more practical side, there is an urgent need to improve training and education in ecological modeling at all levels. Another practical approach to encouraging model use would be the development of common software with a user-friendly interface, although it is known from other fields that nonmodelers using such software often are not aware of how to properly use the model and interpret its results. Finally, having overcome the scientific and technical issues involved in incorporating ecological models in risk assessment, there will remain the hurdle of getting them formally incorporated into the regulatory procedures, which are notoriously resistant to change. But this hurdle can be overcome. Given that models have become a critical and generally accepted feature of the exposure assessment of pesticides (i.e., FOCUS surface water scenarios), there are grounds to be optimistic.
Conclusion Ecological models offer a potentially powerful tool for risk assessment. They can more cost-effectively capture the range of possible scenarios for which estimates of risk are more desirable than field experiments or mesocosms. Exploring the importance of ecological complexities on risk using models can both add confidence to existing approaches (in which such complexities are ignored) and help to focus experimental work on those ecological variables that really matter to risk assessment. If we want to do a better job of assessing risks to ecological systems, we should be developing models for risk assessment that are grounded in both ecological theory and empirical patterns and that can unambiguously incorporate those ecological entities that we wish to protect.
Role of 11 Potential Population Modeling in the Regulatory Context of Pesticide Authorization Franz Streissl Contents Introduction...............................................................................................................97 Terrestrial Vertebrates............................................................................................... 98 Nontarget Arthropods................................................................................................99 Aquatic Organisms....................................................................................................99 Regulatory Acceptance and Outlook...................................................................... 101
An analysis of the available environmental risk assessments of 50 pesticides revealed that of all environmental compartments, it was the risk to terrestrial vertebrates (birds and mammals) that most often failed to be fully addressed. Population modeling is 1 option that has the potential to further refine the risk assessment. Population modeling may also be useful for other groups of organisms (e.g., aquatic organisms or nontarget arthropods). The main advantage of population modeling in a regulatory context is that the outcome of the risk assessment could be compared directly with clearly defined protection goals. However, the lack of protection goals in terms of population-level effects hampers the use of population modeling in a regulatory context. To enhance the acceptability of risk assessments based on population modeling, it would be beneficial for models to be validated and to agree on appropriate models for use in risk assessment of pesticides. Furthermore, guidance would be needed on how to assess the model calculations.
Introduction The Pesticide Risk Assessment Peer Review Unit (PRAPeR) of the European Food Safety Authority (EFSA) is responsible for the peer review of initial risk assessments of active substances. For each substance, a member state carries out this initial risk assessment, which is presented in a draft assessment report (DAR) that is then peer reviewed by experts across the European Union (EU) in a process 97
98
Ecological Models for Regulatory Risk Assessments of Pesticides
coordinated by EFSA. At the end of the peer review, a scientific report is produced by EFSA concluding on the risk assessment for that substance. The final environmental risk assessments of 50 substances in stage 2 of the work program of the pesticides review of the European Commission under Directive 91/414/EEC were analyzed in order to identify areas where population modeling may be applied as a risk refinement step. The 50 substances consisted of 24 insecticides including some acaricides and nematicides, 15 herbicides, and 11 fungicides.
Terrestrial Vertebrates Terrestrial vertebrates (birds and mammals) were the group of organisms for which an acceptable risk most often failed to be fully demonstrated, with 25 substances (50%) in list 2, followed by nontarget arthropods, with 14 (28%) substances, and aquatic organisms, with 12 (24%) substances (Figure 11.1). For birds, the regulatory trigger values of Annex VI of Council Directive 91/414/ EEC (European Commission 1997) were not met in a first-tier risk assessment for 37 substances out of 50. For most of those substances a refined risk assessment was submitted; there were only a few exceptions where applicants did not provide any further risk refinement, or they submitted it too late for it to be taken into consideration in the peer review. The usual options for risk refinement are refinements of selected focal species, for example, proportion of diet taken from the treated area (PT), proportion of different food types taken in the treated area (PD), and measured residues in food items (R). Refined risk assessments for 51 uses (several representative uses proposed for most substances) were peer reviewed. Of these assessments, 25 were accepted and 30
Number of Substances
25 20 15 10
Sewage treatment
Soil microorganisms
Terrestrial plants
Bees
Soil macroorganisms
Earthworms
Aquatic organisms
Non target arthropods
0
Terrestrial vertebrates
5
Groups of Organisms
Figure 11.1 The number of substances for each group of organisms for which an acceptable risk was not fully demonstrated.
99
Role of Population Modeling in Regulatory Context 25
Number of Uses
20 15
4 10
10 5 0
Not accepted Accepted
16 7
10 3 FS
16
4
PT PD Risk Refinement Options
R
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).
26 were rejected by the experts from regulatory authorities. Most risk refinements were rejected because the proposed proportion of diet taken from the treated area (PT) was not sufficiently justified; that is, the PT was not supported by data from radio-tracking or observation studies (Figure 11.2). With regard to the risk to birds, no safe use was demonstrated for 20 of the 50 substances, although considerable efforts were made to refine the exposure via contaminated food items (R). No risk refinements based on population modeling were submitted.
Nontarget Arthropods The risk to nontarget arthropods was not sufficiently addressed for 14 substances. If a potential high risk is indicated in the standard glass plate test with the 2 indicator species Aphidius rhopalosiphi and Typhlodromus pyri, a further species will normally be tested. If the potential for recolonization is not demonstrated in extended laboratory studies, field studies are required. Field trials investigating impacts on infield populations of nontarget arthropods are the highest-tier data currently accepted for risk refinement. However, in the future an option would be to explore using population modeling before initiating field studies. Thus, modeling could be a preliminary step and be used to guide the field study so it can be targeted better or, if sufficient confidence on the modeling approach can be ensured, to replace a field study.
Aquatic Organisms In the aquatic risk assessments, 16 out of 50 substances met the regulatory trigger values in a first-tier risk assessment. For 34 substances the trigger was not met without consideration of no-spray buffer zones and/or further risk refinement. Aquatic invertebrates (daphnids) were identified as the most sensitive group of aquatic organisms, and they were driving the risk assessment in about 50% of the
100
Ecological Models for Regulatory Risk Assessments of Pesticides 9% 28%
13%
Fish Invertebrates Algae Macrophytes
50%
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.
cases, followed by fish with 28% (Figure 11.3). The high proportion of aquatic invertebrates in the most sensitive group was caused by the fact that about half of the substances in stage 2 were insecticides. The risk to aquatic invertebrates and algae is often refined by mesocosm data. In the majority of cases, mesocosm data were sufficient to demonstrate a safe use (see Figure 11.4). On the basis of the endpoints from 17 mesocosm studies, the risk was considered acceptable for 13 substances (including risk mitigation such as no-spray buffer zones). Whereas the potential adverse effects on rapidly reproducing species (planktonic and benthic invertebrates and algae) are usually investigated during a mesocosm study, the long-term effects on slowly reproducing species with low potential of recovery within a short period of time (e.g., univoltine insect species) are usually not studied, and hence the risk is potentially not covered. Some of those species are of 18 Number of Mesocosms
16 14 12 10 8 6 4 2 0 Mesocosm studies submitted
Risk acceptable
Risk not acceptable
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.
Role of Population Modeling in Regulatory Context
101
high ecological and conservational value (e.g., autochthonous crayfish species, dragonflies). Population modeling may be used to assess the risk for such invertebrate species that are not covered by mesocosm experiments and where the basic data set indicates a high susceptibility. In 28% of the cases, fish were the most sensitive group of aquatic organisms. Potential adverse effects on fish are usually not investigated in mesocosm studies since their presence can cause difficulties in assessing effects on the invertebrate community (because the predation pressure is too high). The effect of endocrine disruptors on fish is of increasing concern, for example, triazole fungicides in the peer review of pesticides. Test guidelines are available for the investigation of effects on fish reproduction, but long-term effects on fish populations are not investigated further. Currently, there are 2 options for risk refinements that are regularly applied. The first option is to refine exposure with FOCUS step 4 calculations, where the predicted environmental concentration is calculated for specific geoclimatic conditions and where risk mitigation measures such as no-spray buffer zones are included in the calculations. The second option is to test a larger number of species to reduce the uncertainty around differences in sensitivity. Modeling of fish populations would provide a further option for refinement of long-term effects on fish populations.
Regulatory Acceptance and Outlook Overall it is concluded that population modeling is of potential use, particularly to refine the risk to terrestrial vertebrates where the risk is very often not sufficiently addressed by refinements focusing only on exposure. Experiments with vertebrates should be reduced to a minimum because of animal welfare considerations. Hence, further testing to reduce uncertainty related to differences in species sensitivity should be avoided as a standard risk refinement option. Field studies investigating effects on population level are cost-intensive, and population modeling may provide a more cost-effective alternative. Population modeling has some advantages in a regulatory context: • The outcome of the risk assessment is expressed in real-world effects instead of a toxicity exposure ratio (TER) number, for which it is unclear how protective it is in terms of effects in the field. • The results of population modeling could be compared directly with clearly defined protection goals (e.g., expressed as a certain percentage of reduction in abundance or increase of risk of extinction of exposed populations). • It would be possible to set different levels of protection for different species (e.g., for some species an initial effect of 30% may be acceptable because they have a strong potential for recovery, whereas for other species an initial effect of 10% would be unacceptable). • The outcome of population modeling could be used directly in risk management, that is, by guiding efficient mitigation strategies. However, population modeling also has disadvantages. The evaluation of a risk assessment based on population modeling would be more complex than a traditional
102
Ecological Models for Regulatory Risk Assessments of Pesticides
TER-based deterministic approach. Furthermore, it is unclear what would be the thresholds for acceptable and nonacceptable effects on populations (lack of protection goals). Thus, the lack of clearly defined protection goals in terms of effects on the population level is a major problem in using population modeling in the regulatory context. Decision-making criteria for authorization of plant protection products included in Annex VI of Directive 91/414/EEC (European Commission 1997) are given as TERs or as hazard quotients, for example, “no authorization should be granted if the acute and short-term TER for birds and other terrestrial vertebrates is less than 10 on the basis of LD50 or the long-term TER is less than 5.” The directive also states: “unless it is clearly established through an appropriate risk assessment that under field conditions no unacceptable impact occurs after use of the plant protection product according to the proposed conditions of use.” This leaves room for other means of addressing the risk under field conditions, but it does not specify what unacceptable impacts would be. The guidance document on terrestrial ecotoxicology (European Commission 2002b) states that “poisoned birds and mammals probably constitute the highest social concern, while reproductive effects, although less evident, represent a higher ecological hazard,” and “there is a common understanding that the ecological risk assessment aims not at individuals but at the protection of populations. In general the continuance of populations of non-target organisms should be ensured. Structural and functional endpoints should be regarded of equal importance.” Although the Guidance Document on Birds and Mammals (European Commission 2002c) discusses individual-based effects versus population effects, there is no definition of acceptable and unacceptable effects on populations. In the guidance document on birds and mammals, it is mentioned that “extrapolation tools aiming at bridging the gap between individuum level and population level by means of population modeling techniques are not yet satisfactory (Kendall and Lacher 1994),” but that “population effects should be considered at least qualitatively.” However, progress has been made in science, and it may be possible to predict effects with more accuracy if the latest science is applied. The guidance document for aquatic organisms (European Commission 2002a) points out that “protection of species is a relevant assessment endpoint but difficult to evaluate and therefore not appropriate as a measurement endpoint. Due to the complexity of the matter, particularly when biodiversity issues are included, there are no agreed proposals on these points either in the scientific or in the regulatory community. In general the sustainability of populations of non-target organisms should be ensured.” The guidance document refers to unacceptable effects listed by Brock and Ratte (2002) in the CLASSIC workshop document (Community Level Aquatic System Studies Interpretation Criteria), for example, decrease in biodiversity (overall species richness and densities, population densities of ecological key species, population densities of indicator species), impact on ecosystem functioning and functionality (water quality parameters, harvestable resources, e.g., fish), and decrease in perceived aesthetic value or appearance of the water body (disappearance of species with a popular appeal, e.g., dragonflies, water lilies, visual mortality of fish, frogs, water fowl, and other vertebrates).
Role of Population Modeling in Regulatory Context
103
Directive 91/414/EEC (European Commission 1997, 2002a, 2002b, 2002c) leaves open how an appropriate risk assessment should demonstrate that no unacceptable impact on nontarget organisms occurs under field conditions. Furthermore, it is not specified what would be unacceptable impacts; the respective guidance documents mention protection goals in a general way. For instance, the protection goals for long-term effects on exposed populations are not formulated in a precise way, and no clear decision criteria are given for risk assessors or risk managers. The usefulness of population modeling can therefore be questioned in the context of regulatory pesticide registration. Without clearly defined protection goals it is not clear for risk assessors what the threshold for acceptable effects would be. Judgment would be left to the individual risk assessor, who would have to make decisions on acceptable effects. Without agreed and clearly defined protection goals (in terms of population effects), there is also a high potential for arbitrary decisions. Apart from the lack of clear protection goals, there are also other hurdles for using population models in a regulatory context. Various models do exist, and it is not clear which would be the most appropriate for pesticide risk assessment. It would be beneficial if the models could be validated with regard to their accuracy, or conservatism, to predict effects in the field. Many regulatory authorities are understaffed, so risk assessors have to work under time pressure. This situation is not conducive to the introduction of new risk assessment concepts, which are potentially more time consuming. Therefore, it would be beneficial to agree on a few models, which are suggested to be used in the risk assessment. Simple models would be preferred in order to enhance acceptability of population modeling. Ideally, it should be possible for the risk assessor to understand the model and to use it without being a specialist in population modeling, and guidance should be made available on how to assess the model calculations (e.g., parameterization, quality criteria for input data).
Modeling 12 Ecological An Industry Perspective Pernille Thorbek, Paul Sweeney, and Ed Pilling Contents Introduction............................................................................................................. 105 Which Type of Questions Can Modeling Be Used For?......................................... 106 Suitability of Existing Model.................................................................................. 107 What Needs to Be Covered by Guidance on Good Modeling Practice?................ 108 Conclusions............................................................................................................. 109
Ecological modeling is a tool that can make better use of existing data to provide scientifically sound and ecologically relevant risk assessments. Here, we give an industry perspective on what type of questions ecological modeling can help answer and what are the main obstacles for using ecological modeling in regulatory submissions.
Introduction Plant protection products (PPPs) are manufactured and applied to crops with the purpose of controlling pests and diseases in order to increase yields, thus producing high-quality yet affordable foodstuffs. PPPs are inherently toxic, at least to the pest in question; therefore, ecological risk assessments are carried out to ascertain that no unacceptable harm is done to nontarget organisms, that is, the organisms that are not pests. In order to carry out those risk assessments, a wide range of studies are conducted to describe exposure (environmental fate) and effects (ecotoxicology), which are subsequently combined to calculate toxicity exposure ratios (TERs; or risk quotients in America) used in the ecological risk assessments. The protection goal of most risk assessments is the populations, not the individuals: • “In general, the sustainability of populations of non-target organisms should be ensured” (Guidance Document on Aquatic Ecotoxicology; European Commission 2002a). • “There is a common understanding that the ecological risk assessment aims not at individuals but at the protection of populations” (Guidance Document on Terrestrial Ecotoxicology; European Commission 2002b).
105
106
Ecological Models for Regulatory Risk Assessments of Pesticides
On the population level, the effects of PPPs on nontarget organisms depend not only on exposure and toxicity but also on life history characteristics, phenology, timing of application, presence of refuges in time and space, and landscape structure. In the lower tiers the toxic effects of exposure are mainly estimated by lab studies, thus mainly taking the individual-level effects into account, which is a very conservative approach, and therefore suited for lower-tier risk assessments. In the higher tiers, the toxicity may be estimated by semifield experiments, field studies, or even large-scale monitoring programs. These experiments indirectly take into account some of the other factors affecting risk at a population scale. The complexity of higher-tier studies is increasing, and these studies are very time consuming and expensive, and some also require animal testing, so it is not feasible to carry out large-scale experiments for all scenarios. Industry is under increasing regulatory pressure. Mesocosm studies and field studies are being increasingly challenged by the regulatory authorities because they may be carried out under conditions that they find do not cover the scenario for the risk assessment in question (e.g., discussed at the SETAC workshop Aquatic Mesocosms in Pesticide Registration in Europe: Recent Experiences [AMPERE]). Ecological modeling can be used to extrapolate the findings from studies to cover a broader range of conditions, thus providing assurance that relevant conditions are taken account of.
Which Type of Questions Can Modeling Be Used For? Ecological modeling presents an excellent tool to estimate risk assessments at the population scale, as it can combine a range of relevant factors into a single risk assessment. Thus, ecological modeling can combine information on use pattern, environmental fate, ecotoxicology, life history, and landscape structure to assess the risk of a PPP to a focal species in a way that is both ecologically relevant and scientifically sound. Furthermore, ecological modeling can help identify which PPPs and exposure scenarios are of most concern, thereby targeting the effort to where it’s most needed. Ecological modeling is especially relevant for answering the following types of questions and issues: • Potential for recovery: If there is an initial population-level effect of a PPP, is there potential for recovery? If yes, can recovery be expected within an acceptable time frame? This may be especially relevant for PPPs with high toxicity but short persistence in the environment, or PPPs that target a narrow taxonomic range, so only a few species are of concern. • Extrapolations: This could, for instance, be relevant if there are data from a mesocosm in 1 region but the regulatory authorities have concerns about whether the results also cover conditions in another region. There could be similar issues where it is relevant to extrapolate from species to species or use pattern to use pattern. • Spatial variations in exposure: This could, for instance, be relevant if some species failed to recover in a mesocosms study, whereas in the real world migration would enable recovery. In such situations ecological modeling may be used to extrapolate from mesocosm scale to landscape scale (e.g.,
Ecological Modeling
107
Van den Brink et al. 2006a). This is especially relevant for habitats with high connectivity or for species with good dispersal abilities. • Pulsed exposure: Many studies are carried out at maintained concentrations; however, in reality, exposure varies over time and the persistence of the PPP in the environment is often short. Ecological modeling may be used to predict the effect of a range of different pulsed exposure scenarios (e.g., Ashauer et al. 2006, 2007), thus making better use of existing data. • Long-term effects of sublethal effects: Often the sublethal effects of a PPP are described well on the level of individuals, whereas the likely impact on population level is less well known. In some cases, the effects of density dependence, for example, will make the effects much less pronounced on the population level than on the individual level. In other cases, the sublethal effects may have pronounced effects on population density and structure. Ecological models can be used to put the effects observed on the individual level into the more ecologically relevant population level. • Guiding mitigation strategies and stewardship activities: Industry spends substantial effort on developing stewardship schemes and mitigation strategies. Ecological modeling could be used to guide the development of stewardship activities aiming to conserve species or biodiversity. The guidance opens up for the use of ecological models; however, there is no guidance on how the models should be designed, parameterized, and validated (e.g., EFSA 2007).
Suitability of Existing Model The type of models that would be useful to industry would typically be ones that are able to answer the aforementioned questions. We have over time looked at a large number of published models to see whether they could help answering such questions in a regulatory context. What follows is not a critique of published models, as these were mainly developed with other questions than regulatory risk assessments in mind. It is merely a view on how useful they are for use in regulatory risk assessments. We noticed the following points: • Many of the models simulate the population dynamics of the species often used in ecotoxicological laboratory studies. Often these species are highly relevant for risk assessments, but they are also characterized by life cycles that make them suitable for laboratory culture. Therefore, regulators are sometimes more concerned about other species that are perceived to have life histories that make them more vulnerable or which are thought to be more important for ecosystem functioning. • A large proportion of the models focus on the effects of constant exposure, whereas in reality exposure is highly variable in time and space. • Sometimes the model endpoints (outputs) are difficult to use in risk assessments; for example, often population growth rate λ is reported but what is needed in the risk assessment is time to recovery.
108
Ecological Models for Regulatory Risk Assessments of Pesticides
• Some models are overly simple, assuming constant environment (e.g., optimal conditions with plenty of food and no predation at constant temperature), whereas other models are overly complex for the problem at hand (e.g., predict detailed population dynamics over several years, whereas the information needed is whether recovery can take place within a few weeks or not). In general, the best models are designed to answer specific questions (Grimm and Railsback 2005; Chapter 9), and it seems that models are needed that are designed specifically to address the questions posed in regulatory risk assessments.
What Needs to Be Covered by Guidance on Good Modeling Practice? Ecological modeling can increase the ecological relevance of risk assessments in a scientifically sound and reliable way. However, there are 2 main obstacles for using ecological modeling to support risk assessments. The first is that it is not clearly defined when an initial impact is acceptable, or what duration of effects is acceptable; however, that was not within the scope of this workshop. The other obstacle is that there is no guidance on how to use ecological models in regulatory submissions. It is therefore not clear for industry what would be acceptable to the regulatory authorities; for those in the regulatory authorities without a background in ecological modeling, it may also not be clear what constitutes a scientifically sound approach to modeling. Guidance for the use of ecological modeling in regulatory risk assessments should be flexible enough for the models to target specific questions relevant for different PPPs, but at the same time it should give sufficient direction to ensure that the models are scientifically sound and sufficiently transparent to ascertain thorough assessment of the methods used. A certain level of standardization is desirable both to ensure consistency in risk assessments and to ease the effort involved in assessing the models, as well as enabling industry to develop models that meet the regulatory expectations of authorities. To ensure this, guidance should give direction on the following points: • How to choose species of concern: For instance, should it be the species that is most sensitive to the PPP or a species with a life cycle that is thought to make its population especially sensitive? • How to choose model type: For example, when is a simple matrix model enough, and when should an individual-based model be used? • Model design: Which factors should be included? How much complexity is enough to assess the risk? That is, when is it enough to include the effects of PPP, and when should effects of other stressors also be included? • How to describe the models: There should be agreement on standardized, comprehensive, and transparent ways of describing the models. A good starting point could be the ODD protocol (Grimm et al. 2006).
Ecological Modeling
109
• How to report models’ outputs and how to apply them to different types of questions relevant for risk assessments. • How to parameterize and calibrate models: Which type of data is acceptable for which type of submodels or functions? • How to perform and report sensitivity analysis. • How to validate models: Agreement is needed on what level of validation is needed. Should the overall outputs be validated, or is it more important to validate individual processes. Is it enough to demonstrate that the model is conservative, or should it reproduce real population dynamics accurately? A starting point could be to develop models for focal species and agree on standard scenarios. We believe that it is best to start simple and show the benefits of modeling in risk assessments. In this way the methods and outputs can be made transparent, and that will build confidence in ecological modeling as a risk assessment tool.
Conclusions Ecological modeling is a tool that enables better use of existing data to provide scientifically and ecologically relevant risk assessments. However, the lack of guidance on good modeling practice makes it difficult for industry to develop models that meet the expectations of the regulatory authorities and for regulatory authorities to assess the submitted models. Therefore, stakeholders need to get together and start developing guidance.
References Akçakaya HR. 2005. RAMAS GIS: Linking spatial data with population viability analysis. Version 5.0. Setauket (NY): Applied Biomathematics. Akçakaya HR, Stark JD, Bridges TS. 2008. Demographic toxicity: methods in ecological risk assessment. Oxford (UK): Oxford University Press. Applied Biomathematics, Inc. and Woodlot Alternatives, Inc. 2003. A stochastic population model incorporating PCB effects for wood frogs (Rana sylvatica) breeding in vernal pools associated with the Housatonic River, Pittsfield to Lenoxdale, Massachusetts. Prepared for US Environmental Protection Agency. Available from: http://www.epa. gov/region1/ge/thesite/restofriver/reports/final_era/SupportingInformation%20and%20 Studies%20for%20the%20HousatonicRiverProject/Frog%20Modeling.pdf. Ashauer R, Boxall ABA, Brown CD. 2007. Modeling combined effects of pulsed exposure to carbaryl and chlorpyrifos on Gammarus pulex. Environ Sci Technol 41:5535–5541. Ashauer R, Boxall ABA, Brown CD. 2006. Predicting effects on aquatic organisms from fluctuating or pulsed exposure to pesticides. Environ Toxicol Chem 25:1899–1912. Baer GM, Abelseth MK, Debbie JG. 1971. Oral vaccination of foxes against rabies. Am J Epidemiol 93:487–492. Banks JE, Dick LK, Banks HT, Stark JD. 2008. Time-varying vital rates in ecotoxicology: selective pesticides and aphid population dynamics. Ecol Model 210:155–160. Barnthouse LW. 2004. Quantifying population recovery rates for ecological risk assessment. Environ Toxicol Chem 23:500–508. Bart J. 1995. Acceptance criteria for using individual-based models to make management decisions. Ecol Appl 5:411–420. Bartell SM, Pastorok RA, Akçakaya HR, Regan H, Ferson S, Mackay C. 2003. Realism and relevance of ecological models used in chemical risk assessment. Hum Ecol Risk Assess 9:907–938. Baveco JM, Smeulders AMW. 1994. Objects for simulation: smalltalk and ecology. Simulation 62:42–56. Beaudouin R, Monod G, Ginot V. 2008. Selecting parameters for calibration via sensitivity analysis: an individual-based model of mosquitofish population dynamics. Ecol Model 218:29–48. Begon M, Harper JL, Townsend CR. 1990. Ecology: individuals, populations and communities. Oxford: Blackwell Scientific Publications. Beissinger SR, Westphal MI. 1998. On the use of demographic models of population viability in endangered species management. J Wildlife Manage 62:821–841. Bell IR, Koithan M. 2006. Models for the study of whole systems. Integrative Cancer Ther 5:293–307. Bilde T, Topping CJ. 2004. Life-history traits interact with landscape composition to influence population dynamics of a terrestrial arthropod: a simulation study. EcoScience 11:64–73. Boesten JJTI, Köpp H, Adriaanse PI, Brock TCM, Forbes VE. 2007. Conceptual model for improving the link between exposure and effects in the aquatic risk assessment of pesticides. Ecotoxicol Environ Safety 66:291–308. Bousquet F, LePage C. 2004. Multi-agent simulations and ecosystem management: a review. Ecol Model 176:313–332. Brock TCM, Alix A, Brown CD, Capri E, Gottesbüren BFF, Heimbach F, Lythgo CM, Schulz R, Streloke M. Linking aquatic exposure and effects: Risk assessment of pesticides. Society of Toxicology and Chemistry (SETAC) Eurpoe Workshop on Linking Aquatic 111
112
References
Exposure and Effects in the Risk Assessment of Plant Protection Products; 2007 Mar 14–16, Sep 19–21; Bari (IT) and Wageningen (NL): Pensacola and Boca Raton (FL): SETAC and CRC Press. Forthcoming. Brock T, Arts G, Belgers D, Van Rhenen-Kersten C. 2009. Ecological characterization of drainage ditches in the Netherlands to evaluate pesticide stress. In: Brock TCM, Alix A, Brown CD, Capri E, Gottesbüren PFF, Heimbach F, Lythgo CM, Schulz R, Streloke M. (Eds.) Linking aquatic exposure and effects: Risk assessment of pesticides. Pensacola and Boca Raton (FL): SETAC and CRC Press. Brock TCM, Ratte H-T. 2002. Ecological risk assessment of pesticides. In: Giddings JM, Brock TCM, Heger W, Heimbach F, Maund SJ, Norman SM, Ratte HT, Schäfers C, Streloke M, editors. Community-level aquatic system studies. Interpretation criteria (Proceedings from the CLASSIC workshop held at the Fraunhofer Institute Schmallenberg, Germany, 1999). Pensacola (FL): SETAC. p 33–39. Brown AR, Riddle AM, Winfield IJ, Fletcher JM, James JB. 2005. Predicting the effects of endocrine disrupting chemicals on healthy and disease impacted populations of perch (Perca fluviatilis). Ecol Model 189:377–395. Burgman MA, Ferson S, Akçakaya HR. 1993. Risk assessment in conservation biology. London (UK): Chapman & Hall. Calow P, Sibly RM. 1990. A physiological basis of population processes: ecotoxicological implications. Funct Ecol 4:283–288. Calow P, Sibly RM, Forbes VE. 1997. Risk assessment on the basis of simplified life-history scenarios. Environ Toxicol Chem 16:1983–1989. Campbell PJ, Arnold DJS, Brock TCM, Grandy NJ, Heger W, Heimbach F, Maund SJ, Streloke M, editors. 1990. Guidance document on higher-tier aquatic risk assessment for pesticides (HARAP). Brussels: SETAC-Europe Publications. Caswell H. 2001. Matrix population models: Construction, analysis, and interpretation. 2nd ed. Sunderland (MA): Sinauer. Crouse DT, Crowder LB, Caswell H. 1987. A stage-based population model for loggerhead sea turtles and implications for conservation. Ecology 68:1412–1423. Curry GL, Deuermeyer BL, Feldman RM. 1989. Discrete simulation: fundamentals and microcomputer support. Oakland (CA): Holden-Day. Dalkvist T, Topping CJ, Forbes VE. 2009. Population-level impacts of pesticide-induced chronic effects on individuals depend more on ecology than toxicology. Ecotoxicology and Environmental Safety doi:10.1016/j.ecoenv.2008.10.002. Available from: http:// dx.doi.org/10/1016/j.ecoenv.2008.10.002. DeAngelis DL, Mooij WM. 2005. Individual-based modeling of ecological and evolutionary processes. Annu Rev Ecol Evol Syst 36:147–168. DeAngelis DL, Rose KA, Huston MA. 1994. Individual-oriented approaches to modeling ecological populations and communities. In: Levin SA, editor. Frontiers in mathematical biology. Lecture notes in biomathematics. Vol. 100. Berlin: Springer, p 390–410. De Kroon H, Van Groenendael J, Ehrlén J. 2000. Elasticities: a review of methods and model limitations. Ecology 81:607–618. Doak D, Kareiva P, Kleptetka B. 1994. Modeling population viability for the desert tortoise in the western Mojave Desert. Ecol Appl 4:446–460. Drechsler M, Johst K, Ohl C, Wätzold F. 2007. Designing cost-effective payments for conservation measures to generate spatiotemporal habitat heterogeneity. Conserv Biol 21:1475–1486. Durrett R, Levin S. 1994. The importance of being discrete (and spatial). Theor Popul Biol 46:363–394.
References
113
EFSA. 2007. Guidance document of the scientific panel on plant protection products and their residues for the risk assessment for birds and mammals under Council Directive 91/414/EEC. Draft for public consultation, November 2007. Available from: http:// www.efsa.eu.int/EFSA/DocumentSet/EFSA_GD_RA_birds_and_mammals_REV_ CONSULTATION,0.pdf. Eisinger D. 2007. A new management model for developing small scale contingency plans of rabies [PhD dissertation]. University of Potsdam (Germany). Eisinger D, Thulke H-H. 2008. Spatial pattern formation facilitates eradication of infectious diseases. J Appl Ecol 45:415–423. Eisinger D, Thulke H-H, Müller T, Selhorst T. 2005. Emergency vaccination of rabies under limited resources — combating or containing? BMC Infect Dis 5(10) (online publication). Endels P, Jacquemyn H, Brys R, Hermy M. 2005. Rapid response to habitat restoration by the perennial Primula veris as revealed by demographic monitoring. Plant Ecol 176:143–156. Englung G, Hambäck PA. 2004. Scale-dependence of movement rates in stream invertebrates. Oikos 105:31–40. European Commission. 1997. Council Directive 97/57/EC establishing Annex VI to Directive 91/414/EEC. Concerning the placing of plant protection products on the market. Off J Eur Commun L265:87–109. European Commission. 2002a. Guidance document on aquatic ecotoxicology in the context of the Directive 91/414/EEC. SANCO/3268/2001 rev. 4 final. Brussels. Available from: http://ec.europa.eu/food/plant/protection/evaluation/guidance/wrkdoc10_en.pdf. European Commission. 2002b. Guidance document on terrestrial ecotoxicology under Council Directive 91/414/EEC. SANCO/10329/2002 rev. 2 final. Brussels. Available from: http:// ec.europa.eu/food/plant/protection/evaluation/guidance/wrkdoc09_en.pdf. European Commission. 2002c. Guidance document on risk assessment for birds and mammals under Council Directive 91/414/EEC. SANCO/4145/2000 final. Brussels. Available from: http://ec.europa.eu/food/plant/protection/evaluation/guidance/wrkdoc19_en.pdf. FOCUS. 2001. FOCUS surface water scenarios in the EU evaluation process under 91/414/ EEC. EC Document Reference SANCO/4802/2001-rev. 1. Report of the FOCUS Working Group on Surface Water Scenarios. Brussels. Forbes VE, Calow P. 1999. Is the per capita rate of increase a good measure of population-level effects in ecotoxicology? Environ Toxicol Chem 18:1544–1556. Forbes VE, Hommen U, Thorbek P, Heimbach F, Van den Brink P, Wogram J, Thulke H-H, Grimm V. 2009. Ecological models in support of regulatory risk assessments of pesticides: developing a strategy for the future. Integr Environ Assess Manage 5:167–172. Forbes VE, Palmqvist A, Bach L. 2006. The use and misuse of biomarkers in ecotoxicology. Environ Toxicol Chem 25:272–280. Forbes VE, Sibly RM, Calow P. 2001. Toxicant impacts on density-limited populations: a critical review of theory, practice, and results. Ecol Appl 11:1249–1257. Forbes VE, Sibly RM, Linke-Gamenick I. 2003. Joint effects of population density and toxicant exposure on population dynamics of Capitella sp. I. Ecol Appl 13:1094–1103. Fowler CS. 2007. Taking geographical economics out of equilibrium: implications for theory and policy. J Econ Geogr 7: 265–284. Frank K, Lorek H, Koester F, Sonnenschein M, Wissel C, Grimm V. 2202. META-X: software for metapopulation viability analysis. Berlin: Springer. Freuling C, Selhorst T, Müller T. 2008. The financial challenge of keeping a large region rabiesfree — the EU example. In: Dodet B, Fooks AR, Müller T, Tordo N, in collaboration with the Scientific and Technical Department of the OIE, editors, Towards the elimination of rabies in Eurasia: developments in biologicals. Vol. 131. Basel (SZ): Karger. Fujiwara C, Caswell H. 2001. Demography of the endangered North Atlantic right whale. Nature 414:537–541.
114
References
Giere RN. 1991. Understanding scientific reasoning. New York: Harcourt Brace Jovanovich College Publishers. Giesy JP, Solomon KR, Coats JR, Dixon KR, Giddings JM, Kenaga EE. 1999. Chlorpyrifos: ecological risk assessment in North American aquatic environments. Rev Environ Contam Toxicol 160:1–129. Gilbert N, Troitzsch KG. 2005. Simulation for the social scientist. Buckingham (UK): Open University Press. Glaser D, Bridges TS. 2007. Separating the wheat from the chaff: the effective use of mathematical models as decision tools. Integr Environ Assess Manage 3:442–449. Goss-Custard JD, Burton NHK, Clark NA, Ferns PN, McGrorty S, Reading CJ, Rehfisch MM, Stillman RA, Townend I, West AD, Worrall DH. 2006. Test of a behavior-based individual-based model: response of shorebird mortality to habitat loss. Ecol Appl 16:2215–2222. Gotelli NJ, Ellison AM. 2006. Forecasting extinction risk with nonstationary matrix models. Ecol Appl 16:51–61. Grimm V. 1994. Mathematical models and understanding in ecology. Ecol Model 75:641–651. Grimm V. 1999. Ten years of individual-based modelling in ecology: what have we learned and what could we learn in the future? Ecol Model 115:129–148. Grimm V. 2008. Individual-based models. In: Jørgensen SE, Fath BD, editors, Ecological models. Vol. 3 of Encyclopedia of ecology, 5 vols. Oxford (UK): Elsevier, p 1959–1968. Grimm V, Berger U, Bastiansen F, Eliassen S, Ginot V, Giske J, Goss-Custard J, Grand T, Heinz SK, Huse G, Huth A, Jepsen JU, Jørgensen C, Mooij WM, Müller B, Pe’er G, Piou C, Railsback SF, Robbins AM, Robbins MM, Rossmanith E, Rüger N, Strand E, Souissi S, Stillman RA, Vabø R, Visser U, DeAngelis DL. 2006. A standard protocol for describing individual-based and agent-based models. Ecol Model 198:115–126. Grimm V, Railsback SF. 2005. Individual-based modeling and ecology. Princeton (NJ): Princeton University Press. Grimm V, Revilla E, Berger U, Jeltsch F, Mooij WM, Railsback SF, Thulke H-H, Weiner J, Wiegand T, DeAngelis DL. 2005. Pattern-oriented modeling of agent-based complex systems: lessons from ecology. Science 310:987–991. Haefner JW. 2005. Modeling biological systems: principles and applications. New York: Springer. Halley JM, Thomas CFG, Jepson PC. 1996. A model for the spatial dynamics of linyphiid spiders in farmland. J Appl Ecol 33:471–492. Hansen F. 2001. Modellierung der raum-zeitlichen Dynamik des Fuchsbandwurms (Echinococcus multilocularis) [PhD dissertation]. University of Potsdam, Germany. Hansen F, Forbes VE, Forbes TL. 1999. Using elasticity analysis of demographic models to link toxicant effects on individuals to the population level: an example. Funct Ecol 13:157–162. Hansen F, Tackmann K, Jeltsch F, Staubach C, Thulke H-H. 2001. If space changes all — the small-scale epidemiology of the fox tapeworm. In: Menzies FD, Reid SWJ, editors, Proceedings SVEPM. Noordwijkerhout (NL). p 73–85. Hansen F, Tackmann K, Jeltsch F, Thulke H-H. 2003. Köderauslageintervalle und Dauer der Bekämpfung des Kleinen Fuchsbandwurms. Berliner und Münchener Tieräztliche Wochenschrift 116:299–305. Hole DG, Whittingham MJ, Bradbury RB, Anderson GQA, Lee PLM, Wilson JD, Krebs JR. 2002. Widespread local house-sparrow extinctions — agricultural intensification is blamed for the plummeting populations of these birds. Nature 418:931–932. Huth A, Tietjen B. 2007. Management strategies for tropical rain forests: results of ecological models and requirements for ecological-economic modelling. Ecol Econ 62:207–215. Jeltsch F, Müller MS, Grimm V, Wissel C, Brandl R. 1997. Pattern formation triggered by rare events: lessons from the spread of rabies. Proc R Soc Lond Ser B 264:495–503.
References
115
Kendall RJ, Lacher TE. 1994. Wildlife toxicology and population modeling. SETAC special publication. Boca Raton (FL): Lewis Publishers. Kendall WL, Hines JE, Nichols JE. 2003. Adjusting multistate capture-recapture models for misclassification bias: manatee breeding proportions. Ecology 84:1058–1066. Klaine SJ, Cobb GP, Dickerson RL, Dixon KR, Kendall RJ, Smith EE, Solomon KR. 1996. An ecological risk assessment for the use of the biocide, dibromonitrilopropionamide (DBNPA), in industrial cooling systems. Environ Toxicol Chem 15:21–30. Koh HL, Hallam TG, Lee HL. 1997. Combined effects of environmental and chemical stressors on a model Daphnia population. Ecol Model 103:19–32. Kuhn A, Munns Jr WR, Champlin D, McKinney R, Tagliabue M, Serbst J, Gleason T. 2001. Evaluation of the efficacy of extrapolation population modeling to predict the dynamics of Americamysis bahia populations in the laboratory. Environ Toxicol Chem 20:213–221. Kuhn A, Munns Jr WR, Poucher S, Champlin D, Lussier S. 2000. Prediction of populationlevel response from mysid toxicity test data using population modeling techniques. Environ Toxicol Chem 19:2364–2371. Kuhn A, Munns Jr WR, Serbst J, Edwards P, Cantwell MG, Gleason T, Pelletier MC, Berry W. 2002. Evaluating the ecological significance of laboratory response data to predict population-level effects for the estuarine amphipod Ampelisca abdita. Environ Toxicol Chem 21:865–874. Laskowski R. 1995. Some good reasons to ban the use of NOEC, LOEC, and related concepts in ecotoxicology. Oikos 73:140–144. Leslie PH. 1945. On the use of matrices in certain population mathematics. Biometrika 33:183–212. Levin L, Caswell H, Bridges T, DiBacco C, Cabrera D, Plaia G. 1996. Demographic responses of estuarine polychaetes to pollutants: life-table response experiments. Ecol Appl 6:1295–1313. Levin SA. 1992. The problem of pattern and scale in ecology. Ecology 73:1943–1967. Levins R. 1966. Strategy of model building in population biology. Am Sci 54:421–431. Liess M, Brown C, Dohmen P, Duquesne S, Hart A, Heimbach F, Kreuger J, Lagadic L, Maund S, Reinert W, Streloke M, Tarazona JV. 2005. Effects of pesticides in the field. Brussels: SETAC. Lin BL, Tokai A, Nakanishi J. 2005. Approaches for establishing predicted-no-effect concentration for population-level ecological risk assessment in the context of chemical substances management. Environ Sci Technol 39:4833–4840 Lindenmayer DB, Burgman MA, Akçakaya HR, Lacy RC, Possingham HP. 1995. A review of the generic computer programs ALEX, RAMAS/space and VORTEX for modeling the viability of wildlife metapopulations. Ecol Model 82:161–174. Lopes C, Pery ARR, Chaumot A, Charles S. 2005. Ecotoxicology and population dynamics: using DEBtox models in a Leslie modeling approach. Ecol Model 188:30–40. Lotka AJ. 1925. Elements of physical biology. Baltimore (MD): Williams and Wilkins. Maltby L, Kedwards TJ, Forbes VE, Grasman K, Kammenga JE, Munns Jr WR, Ringwood AH, Weis JS, Wood SN. 2001. Linking individual-level responses and population-level consequences. In: Baird JD, Burton GA Jr, editors. Ecological variability: separating natural from anthropogenic causes of ecosystem impairment. Pensacola (FL): SETAC, p 27–82. Meng YB, Lin B-L, Tominaga M, Nakanishi J. 2006. Simulation of the population-level effects of 4-nonylphenol on wild Japanese medaka (Oryzias latipes). Ecol Model 197:350–360. Morris WF, Doak DF. 2002. Quantitative conservation biology: theory and practice of population viability analysis. Sunderland (MA): Sinauer Associates.
116
References
Müller B, Linstädter A, Frank K, Bollig M, Wissel C. 2007. Learning from local ecological knowledge: modeling the pastoral-nomadic range management of the Himba, Namibia. Ecol Appl 17:1857–1875. Nacci D, Pelletier M, Lake J, Bennett R, Nichols J, Haebler R, Grear J, Kuhn A, Copeland J, Nicholson M, Walters S, Munns W. 2005. An approach to predict risks to wildlife populations from mercury and other stressors. Ecotoxicology 14:283–293. Naito W, Miyamoto K, Nakanishi J, Masunaga S, Bartell SM. 2003. Evaluation of an ecosystem model in ecological risk assessment of chemicals. Chemosphere 53:363–375. Norris K, McCulloch N. 2003. Demographic models and the management of endangered species: a case study of the critically endangered Seychelles magpie robin. J Appl Ecol 40:890–899. OECD. 2006. Guidance document on simulated freshwater lentic field tests (outdoor microcosms and mesocosms). OECD Environment Health and Safety Publications, Series on Testing and Assessment, No. 53. Environment Directorate, OECD, Paris. Available from: http://www.olis.oecd.org/olis/2006doc.nsf/LinkTo/NT00000E82/$FILE/JT03208281. PDF. Pagel J, Fritzsch K, Biedermann R, Schröder B. 2008. Annual plants under cyclic disturbance regimes — better understanding through model aggregation. Ecol Appl 18:2000–2015. Pastorok RA, Akçakaya HR, Regan H, Ferson S, Bartell SM. 2003. Role of ecological modeling in risk assessment. Hum Ecol Risk Assess 9:939–972. Pastorok RA, Bartell SM, Ferson S, Ginzburg LR. 2001. Ecological modeling in risk assessment: chemical effects on populations, ecosystems and landscapes. Boca Raton (FL): CRC Press. Pauly D, Christensen V, Walters C. 2000. Ecopath, ecosim, and ecospace as tools for evaluating ecosystem impact of fisheries. Ices J Mar Sci 57:697–706. Peck SL. 2004. Simulation as experiment: a philosophical reassessment for biological modelling. Trends Ecol Evol 19:530–534. Peeters ETHM, Camu JM, Beijer JAJ, Scheffer M, Gardeniers JPP. 2002. Response of the waterlouse Asellus aquaticus to multiple stressors: effects of current velocity and mineral substratum. J Aquatic Ecosyst Stress Recov 9:193–203. Pena J, Hinchey MG, Resinas M, Sterritt R, Rash JL. 2007. Designing and managing evolving systems using a MAS product line approach. Sci Comput Program 66:71–86. Porte A, Bartelink HH. 2002. Modelling mixed forest growth: a review of models for forest management. Ecol Model 150:141–188. Pretzsch H, Biber P, Dursky J. 2002. The single tree-based stand simulator SILVA: construction, application, and evaluation. Forest Ecol Manage 162:3–21. Rademacher C, Neuert C, Grundmann V, Wissel C, Grimm V. 2001. Was charakterisiert Buchenurwälder? Untersuchungen der Altersstruktur des Kronendachs und der räumlichen Verteilung der Baumriesen mit Hilfe des Simulationsmodells BEFORE. Forstwissenschaftliches Centralblatt 120:288–302. Rademacher C, Neuert C, Grundmann V, Wissel C, Grimm V. 2004. Reconstructing spatiotemporal dynamics of Central European natural beech forests: the rule-based forest model BEFORE. Forest Ecol Manage 194:349–368. Railsback SF. 2001. Concepts from complex adaptive systems as a framework for individualbased modelling. Ecol Model 139:47–62. Railsback SF, Harvey BC. 2002. Analysis of habitat-selection rules using an individual-based model. Ecology 83:1817–1830. Roß-Nickoll M, Lennartz G, Fürste A, Mause R, Ottermanns R, Schäfer S, Smolis M, Theißen B, Toschki A, Ratte HT. 2004. Die Arthropodenfauna von Nichtzielflächen und die Konsequenzen für die Bewertung der Auswirkungen von Pflanzenschutzmitteln auf den terrestrischen Bereich des Naturhaushaltes. UBA Text 10/04. Berlin.
References
117
Rykiel EJ. 1996. Testing ecological models: the meaning of validation. Ecol Model 90:229–244. Service RF. 1999. Exploring the systems of life. Science 284:80–83. Sibly RF, Akçakaya HR, Topping CJ, O’Connor RJ. 2005. Population-level assessment of risks of pesticides to birds and mammals in the UK. Ecotoxicology 14:863–876. Sibly RM, Nabe-Nielsen J, Forchhammer MC, Forbes VE, Topping CJ. 2009. Spatial and temporal heterogeneity in population dynamics. BMC Ecology 9:18 doi10.1186/14726785-9-18. Available from: http://www.biomedcentral.com/1472-6785/9/18. Solomon KR, Baker DB, Richards RP, Dixon DR, Klaine SJ, La Point TW, Kendall RJ, Weisskopf CP, Giddings JM, Giesy JP, Hall Jr LW, Williams WM. 1996. Ecological risk assessment of atrazine in North American surface waters. Environ Toxicol Chem 15:31–74. Starfield AM, Smith KA, Bleloch AL. 1990. How to model it. New York: McGraw-Hill. Stark JD. 2005. How closely do acute lethal concentration estimates predict effects of toxicants on populations? Integr Environ Assess Manage 1:109–113. Stark JD. 2006. Toxicity endpoints used in risk assessment: What do they really mean? SETAC Globe 7:29–30. Stark JD, Banken JAO. 1999. Importance of population structure at the time of toxicant exposure. Ecotoxicol Environ Safety 42:282–287. Stark JD, Banks JE. 2001. “Selective pesticides”: are they less hazardous to the environment? Bioscience 51:980–982. Stark JD, Banks JE, Acheampong S. 2004. Estimating susceptibility of biological control agents to pesticides: Influence of life history strategies and population structure. Biol Control 29:392–398. Stark JD, Banks JE, Vargas R. 2004. How risky is risk assessment: the role that life history strategies play in susceptibility of species to stress. Proc Natl Acad Sci USA 101:732–736. Stark JD, Sherman M. 1989. Toxicity, penetration, and metabolism of acephate in 3 fruit fly species (Diptera: Tephritidae). J Econ Entomol 82:1564–1571. Stark JD, Vargas R, Banks JE. 2007. Incorporating ecologically relevant measures of pesticide effect for estimating the compatibility of pesticides and biocontrol agents. J Econ Entomol 100:1027–1032. Stillman RA, Simmons VL. 2006. Predicting the functional response of a farmland bird. Funct Ecol 20:723–730. Suter GWII. 1993. Ecological risk assessment. Boca Raton (FL): Lewis Publishers. 538 p. Tackmann K, Löschner U, Mix H, Staubach C, Thulke H-H, Ziller M, Conraths FJ. 2001. A field study to control Echinococcus multilocularis-infections of the Red Fox (Vulpes vulpes) in an endemic focus. Epidemiol Infect 127:577–587. Tang S, Ziao Y, Chen L, Cheke RA. 2005. Integrated pest management models and their dynamical behaviour. Bulletin of Mathematical Biology 67:115–135. Tannenbaum LV. 2007. And so we model: the ineffective use of mathematical models in ecological risk assessments. Integr Environ Assess Manage 3:473–475. Thompson HM, Wilkins S, Battersby AH, Waite RJ, Wilkinson D. 2005. The effects of four insect growth-regulating (igr) insecticides on honeybee (Apis mellifera L.) colony development, queen rearing and drone sperm production. Ecotoxicology 14:757–769. Thomson DM. 2005. Matrix models as a tool for understanding invasive plant and native plant interactions. Conserv Biol 19:917–928. Thorbek P, Bilde T. 2004. Reduced numbers of generalist arthropod predators after crop management. J Appl Ecol 41:526–538. Thorbek P, Topping CJ. 2005. The influence of landscape diversity and heterogeneity on spatial dynamics of agrobiont linyphiid spiders: an individual-based model. BioControl 50:1–33.
118
References
Thulke H-H, Grimm V, Müller MS, Staubach C, Tischendorf L, Wissel C, Jeltsch F. 1999. From pattern to practice: a scaling-down strategy for spatially explicit modelling illustrated by the spread and control of rabies. Ecol Model 117:179–202. Tischendorf L, Thulke H-H, Staubach C, Müller MS, Jeltsch F, Goretzki J, Selhorst T, Müller T, Schlüter H, Wissel C. 1998. Chance and risk of controlling rabies in large-scale and long-term immunized fox populations. Proc R Soc Lond B 265:839–846. Topping CJ, Hansen TS, Jensen TS, Jepsen JU, Nikolajsen F, Odderskaer P. 2003. ALMaSS, an agent-based model for animals in temperate Eurpean lanscapes. Ecological Modelling 167:65–82. Topping CJ, Odderskaer P. 2004. Modeling the influence of temporal and spatial factors on the assessment of impacts of pesticides on Skylarks. Environ Toxicol Chem 23(2):509–520. Topping CJ, Rehder MJ, Mayoh BH. 1999. VIOLA: a new visual programming language designed for the rapid development of interacting agent systems. Acta Biotheor 47:129−140. Topping CJ, Sibly RM, Akçakaya HR, Smith GC, Crocker DR. 2005. Risk assessment of UK skylark populations using life-history and individual-based landscape models. Ecotoxicology 14:925–936. Ulbrich K, Drechsler M, Wätzold F, Johst K, Settele J. 2008. A software tool for designing cost-effective compensation payments for conservation measures. Environ Model Software 23:122–123. [USEPA] US Environmental Protection Agency. 2007a. Data requirements for pesticide registration. Available from: http://www.epa.gov/opp00001/regulating/data_requirements. htm. [USEPA] US Envirionmental Protection Agency. 2007b. Appendix F: the risk quotient method and levels of concern. Available from: http://www.epa.gov/espp//litstatus/effects/ appendix_f_rq_method_and_locs.pdf. Van den Brink PJ. 2007. Response to recent criticism on aquatic (semi-) field studies experiments: opportunities for new developments in ecological risk assessment of pesticides. Letter to the editor. Integr Environ Assess Manage 2:202–203. Van den Brink PJ, Brown CD, Dubus IG. 2006b. Using the expert model PERPEST to translate measured and predicted pesticide exposure data into ecological risks. Ecol Model 191:106–117. Van den Brink PJ, Van Wijngaarden RPA, Lucassen WGH, Brock TCM, Leeuwangh P. 1996. Effects of the insecticide Dursban® 4E (active ingredient chlorpyrifos) in outdoor experimental ditches. II. Invertebrate community responses and recovery. Environ Toxicol Chem 15:1143–1153. Van den Brink PJ, Verboom J, Baveco JM, Heimbach F. 2007. An individual-based approach to model spatial population dynamics of invertebrates in aquatic ecosystems after pesticide contamination. Environ Toxicol Chem 26:2226–2236. Van Kirk RW, Hill SL. 2007. Demographic model predicts trout population response to selenium based on individual-level toxicity. Ecol Model 206:407–420. Vargas RI, Miyashita D, Nishida T. 1984. Life history and demographic parameters of three laboratory-reared tephritids (Dipera: Tephritidae). Ann Entomol Soc Am 77:651–656. Vonesh JR, De la Cruz O. 2002. Complex life cycles and density dependence: assessing the contribution of egg mortality to amphibian declines. Oecologia 133:325–333. Wang M, Grimm V. 2007. Home range dynamics and population regulation: an individualbased model of the common shrew Sorex araneus. Ecol Model 205:397–409. Weiner J. 1995. On the practice of ecology. J Ecol 83:153–158. Weiner J. 1999. On self-criticism in ecology. Oikos 85:373–374. Wennergren U, Stark JD. 2000. Modeling long-term effects of pesticides on populations: beyond just counting dead animals. Ecological Applications 10:295–302.
References
119
Wiegand J, Jeltsch F, Hanski I, Grimm V. 2003. Using pattern-oriented modeling for revealing hidden information: a key for reconciling ecological theory and application. Oikos 100:209–222. Wilensky U. 1999. NetLogo. Evanston (IL): Center for Connected Learning and Computer-Based Modeling, Northwestern University. Available from: http://ccl.northwestern.edu/netlogo/. Wilson PH. 2003. Using population projection matrices to evaluate recovery strategies for Snake River spring and summer chinook salmon. Conserv Biol 17:782–794. Wissel C. 1992. Aims and limits of ecological modelling exemplified by island theory. Ecol Model 63:1–12. Wissel C. 2000. Grid-based models as tools for ecological research. In: Dieckmann U, Law R, Metz JAJ, editors, The geometry of ecological interactions: simplifying spatial complexity. Cambridge: Cambridge University Press, p 94–115. Wood SN. 1994. Obtaining birth and mortality patterns from structured population trajectories. Ecol Monogr 64:23–44. Worden L, Levin SA. 2007. Evolutionary escape from the prisoner’s dilemma. J Theor Biol 245:411–422.
Index a ABMs, see Agent-based models Academic’s view, see Ecological modeling, academic’s point of view Acephate, 42 Agent-based models (ABMs), 12, 23, 57–66 ALMaSS, 58–60 process overview and scheduling, 59 purpose, 59 state variables and scales, 59 ALMaSS application examples, 60–64 assumptions regarding other mortalities in risk assessment, 63–64 impact of altering landscape structure, 61–62 measuring carrying capacity for Bembidion, 61 modeling chronic effects of endocrine disrupter in voles, 64 beetle population numbers, 62 discussion, 64–66 EU Directive 91/414, proposed changes, 66 impact assessments of insecticide to arable fields, 63 introduction, 57–58 landscape model, 59 major strength of, 65 ODD protocol, 58 pattern-oriented modeling, 66 physics-envy syndrome, 58 scheduling of processes, 60 secondary predictions, 65 vinclozolin, 64, 65 ALMaSS, 58–60 framework, IBMs development within, 83 model, 5 process overview and scheduling, 59 programming skill needed, 94 purpose, 59 state variables and scales, 59 ALMaSS application examples, 60–64 assumptions regarding other mortalities in risk assessment, 63–64 impact of altering landscape structure, 61–62 measuring carrying capacity for Bembidion, 61 modeling chronic effects of endocrine disrupter in voles, 64 Alternative management concept, 70 Antirabies vaccination, 72
Aphidius rhopalosiphi, 99 Aquatic invertebrates, see MASTEOP Aquatic organisms FOCUS step 4 calculations, 101 guidance document for, 102 mesocosm data, 100 rapidly reproducing species, 100 Asellus aquaticus, 48, see also MASTEOP
b Bactrocera cucurbitae, 41 dorsalis, 41 Beech forest model, 80 Bembidon lampros, 59, 63 BIOSIS database, 80, 81 Birds, effects observed in toxicological tests of, 30
c CAB Abstracts database, 80, 81 Capreolus capreolus, 59 Cellular automata, 81 Ceratitis capitata, 41 Chaoborus, 30, 48 Coccinella septempunctata, 30 Crop protection products, effects of on nontarget organisms, 11
d Daphnia, 30, 48 DAR, see Draft assessment report Databases, ecological population models, 80 DEBtox, 82 Difference equations, 19–20, 83 advantages, 20 demographic parameters, 20 disadvantages, 20 ordinary differential equation, 19 pesticide registrations, 20 population dynamics, 19 Draft assessment report (DAR), 97
e Ecological modeling guidance flexibility, 108 main purpose of, 68 realism in, 58 role of in pesticide risk assessment, 12 skills needed for, 93
121
122 training in basic rationale of, 85 Ecological modeling, academic’s point of view, 89–96 advantages of using ecological models, 90–92 conclusion, 96 decomposition analysis, 92 disadvantages of using ecological models, 93–94 ecological protection targets, 93 efforts to reduce animal testing, 92 future challenges, 95–96 identification of exposure conditions, 95 if–then questions, 91 influence of biological variables on target populations, 91 introduction, 90 microarrays, 92 model requirements for regulatory purposes, 94–95 NetLogo, 94 observed impacts on defined populations, 95 pesticide risk assessment protocols, weakness in, 94 population model development, 90 registration of pesticides, 94–95 research program development, 89 robustness of results, 91 software development, 96 unexpected events, 91 Ecological modeling, benefits of using as pesticide risk assessment tool, 5–6 data analysis, 6 ecology, 5 extrapolation, 6 life cycle characteristics, 6 prediction of effects, 6 Ecological modeling, industry perspective, 105–109 conclusions, 109 guidance on good modeling practice, 108–109 introduction, 105–106 lower-tier risk assessments, 106 plant protection products effects of on nontarget organisms, 106 extrapolations, 106 potential for recovery, 106 pulsed exposure, 107 purpose, 105 sublethal effects of, 107 toxicity, 105 questions that modeling can be used for, 106–107 suitability of existing model, 107–108 Ecological modeling, introduction, 15–26
Index beyond ad hoc design, 24–25 definition of model, 16–17 difference equations, 19–20 advantages, 20 demographic parameters, 20 disadvantages, 20 ordinary differential equation, 19 pesticide registrations, 20 population dynamics, 19 discussion, 25–26 forest model, 16 how modeling works, 17–19 individual-based models, 22–24 advantage, 23 agent-based models, 23 design concepts, 23 disadvantages, 23 general problems, 24 ODD protocol, 22 submodels, 23 matrix models advantages, 22 age-structured models, 20, 21 disadvantages, 22 linear difference equations, 20 population density, 22 population growth rate, 21 modeling cycle tasks, 18 model validation, 25 population dynamics, 17 population model types, 19–24 differential and difference equations, 19–20 individual-based models, 22–24 matrix models, 20–22 projection, 17 reason for modeling, 17 theory–development cycle, 25 Ecological modeling, state of the art, 77–87 ALMaSS framework, 83 appendix, 85–87 beech forest model, 80 BIOSIS database, 80, 81 CAB Abstracts database, 80, 81 cellular automata, 81 DEBtox, 82 developmental stages, 80 difference equations, 83 discussion, 83–85 good modeling practice, 85 introduction, 78–79 measures of toxicity, 81 methods, 80–81 model assumptions, 84 model diversity, 84 PASCAL database, 80, 81
123
Index
population model types, 80 quantification of risk, 84 results, 81–83 silviculture, models for, 78 testing, verification, calibration, and validation, 79–80 toxicity endpoints, 82 Ecological models (EMs), 2 academia’s view on, 4 challenges for developing, 7 costs associated with development, 89 current state of in pesticide risk assessment, 4 EFSA’s view on, 4 flexibility, 94 hurdle, 96 industry’s view on, 4–5 lack of confidence regarding, 84 management decisions supported by, 75 mechanism behind ecological impacts, 92 peer review of, 8 practical applications, 15 purpose, 16 risk assessment of wildlife diseases, 4 structurally realistic, 74 use of for decision support in other fields, 9 EEC, see Expected environmental concentration EFSA, see European Food Safety Authority EMs, see Ecological models Endocrine disrupter, 64 Erigone atra, 63 EU, see European Union EU active substances program, 28 EU Directive 91/414 EEC EMs in risk assessment under, 3 nontarget organisms, 103 pesticides review, 98 plant protection products, 27, 95, 102 proposed changes, 66 toxicity exposure ratio calculation, 34–35 European Food Safety Authority (EFSA), 97 European Union (EU), 97 Expected environmental concentration (EEC), 34
f “Fencing the outbreak” strategy, 69 FOCUS model, 5 Forest model, 16 Freeware, 94 Fruit fly species, toxicity and hazard ratio of acephate and, 42, 44
g Gammarus, 30, 48 Good modeling practice, 85, 108–109
h Human health risk assessment, ecological risk assessment vs., 90
i IBMs, see Individual-based models Individual-based models (IBMs), 12, 22–24, 49, 77 advantage, 23 agent-based models, 23 ALMaSS framework, 83 design concepts, 23 disadvantages, 23 general problems, 24 ODD protocol, 22 programming skill needed, 94 role of for pesticide risk assessment, 24 submodels, 23 Industry perspective, see Ecological modeling, industry perspective
l LEMTOX workshop, executive summary, 1–9 aim of LEMTOX, 2–3 barriers for ecological modeling used more frequently in pesticide risk assessment, 7–8 black box syndrome, 8 EM development challenges, 7 guidance documents, 7 problem formulation, 7 risk assessment model challenges, 7–8 benefits of using ecological modeling as pesticide risk assessment tool, 5–6 data analysis, 6 ecology, 5 extrapolation, 6 life cycle characteristics, 6 prediction of effects, 6 discussion, 9 diversity of model types, 3 example models, 5 ALMaSS, 5 FOCUS, 5 MASTEP, 5 guidance documents, 7 important outcome of workshop, 8, 9 keynote presentations, 2 optimistic approach of, 2 participant backgrounds, 2 results from group discussions, 5–9 views from academia, industry, and regulators, 3–5 academia’s view on ecological models, 4 EFSA’s view on ecological models, 4
124 EU directive 91/414 EEC, 3 industry’s view on ecological models, 4 introduction to ecological modeling, 3 pesticide risk assessment, 4 wildlife diseases, 4 wider use of ecological modeling in pesticide risk assessment, 8–9 LEMTOX workshop, introduction, 11–13 agent-based models, 12 crop protection products, effects of on nontarget organisms, 11 higher-tier risk assessments, 11 individual-based models, 12 laboratory-based findings, 11 pesticide registration, 12 population-level effects, 11 toxicity exposure ratios, 11 Leslie matrix, 36 Levels of concern (LOCs), 35 LOCs, see Levels of concern
m Mammals, effects observed in toxicological tests of, 30 MASTEOP (Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides), 47–55 Asellus movement, 54 discussion, 54–55 model assumptions, 54 outlook, 54–55 parameter uncertainty, 54 introduction, 48–49 materials and methods, 49–52 density dependence, 51 design concepts, 50–51 dispersal and movement, 51 exposure scenario, 52 initialization, 51 input, 51 landscape scenario, 52 life cycle, 51 mortality, 51 pesticide mortality, 52 process overview and scheduling, 49–50 purpose, 49 reproduction, 51 scenarios, 52 state variables and scales, 49 submodels, 51 model, 5 results, 52–54 buffer zone treatment levels, 52 empty patch, 54 results, exported effects, untreated population density, 52
Index state change scheduling, 50 Matrix models, 20–22 advantages, 22 age-structured models, 20, 21 disadvantages, 22 linear difference equations, 20 population density, 22 population growth rate, 21 Matrix population models, development of for estimation of toxicant effects, 33–45 ability to estimate risks of chemical exposure to populations, 35 acephate, 42 advantages of matrix models over other models, 39 conclusions, 43–45 daphnid life spans, 41 deterministic matrix models, 38 deterministic population projection of hypothetical species, 37 disadvantages of matrix models compared to other models, 39–40 fruit fly species, toxicity and hazard ratio of acephate and, 42, 44 insect species exposed to pesticide, 42–43 Leslie matrix, 36 matrix models, 36–37 model construction and interpretation, 33 quotient method, 35 RAMAS model, 38 rat–elephant phenomenon, 40–42 reason for using matrix models over other models, 36 risk quotient method of risk assessment and its limitations, 34–35 species dose–response curves, 42 stochastic matrix models, 38–39 survival rates, 37 transition rate, 36 Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides, see MASTEOP Microtus agrestis, 59 Mind models, 75 Model(s), see also Matrix population models, development of for estimation of toxicant effects advantage of stochastic matrix models over deterministic models, 38 agent-based models, 12 ALMaSS, 5, 58–60 application examples, 60–64 process overview and scheduling, 59 purpose, 59 state variables and scales, 59 assumptions, 84
Index
beech forest model, 80 Bembidon, 61 challenges in risk assessment, 7 development, standard scenarios, 109 diversity, 3, 84 FOCUS, 5 forest, 16 fox tapeworm, 73 individual-based models, 12 landscape, 59 MASTEP, 5 measures of toxicity, 81 mind, 75 rabies, 74 RAMAS, 38 silviculture, 78 suspicion of, 95 validation, 25 verification, 79
n NetLogo, 94 NOEC, see No observable effect concentration Nontarget organisms, effects of crop protection products on, 11 No observable effect concentration (NOEC), 35
o ODD protocol, see Overview–Design concepts– Detail protocol Oedothorax fuscus, 63 Ordinary differential equation, 19 Overview–Design concepts–Detail (ODD) protocol, 85 IBMs, and, 22 industry perspective, 108 model based on, 58
p Parasitic cycle modeling, 72 PASCAL database, 80, 81 Pattern-oriented modeling, 66 PDF, see Probability density function PECs, see Predicted environmental concentrations Perdix perdix, 59 Pesticide application, species reproduction and, 91 Pesticide authorization, see Population modeling, potential role of in pesticide authorization Pesticide registration decision making, 84 difference equations, 20 discrepancy of model use, 78 use of ecological models to support, 12
125 Pesticide risk assessment aim of, 90 current state of EMs in, 4 peer review unit (PRAPeR), 97 problem with ecological modeling of, 16 protocols, weakness in, 94 role of ecological modeling in, 12, 75 role of IBMs for, 24 Pesticide risk assessment tool barriers of using ecological modeling as, 7–8 black box syndrome, 8 challenges, 7 EM development challenges, 7 guidance documents, 7 problem formulation, 7 risk assessment model challenges, 7–8 benefits of using ecological modeling as, 5–6 data analysis, 6 ecology, 5 extrapolation, 6 life cycle characteristics, 6 prediction of effects, 6 Pesticides review, EU Directive 91/414/EEC, 98 Physics-envy syndrome, 58 Plant protection products (PPPs), 27, 105 application procedures of in Germany, 30 authorization, 28 effects of on nontarget organisms, 106 EU Directive 91/414/EEC, 95, 102 extrapolations, 106 potential for recovery, 106 pulsed exposure, 107 purpose, 105 sublethal effects of, 107 toxicity, 105 Plant protection products, regulatory challenges, 27–31 application procedures of in Germany, 30 authorization, 28 effects observed in toxicological tests, 30 EU active substances program, 28 EU Directive 91/414 EEC, 27 extrapolation of mesocosm results, 29 lessons learned, 30–31 past experiences with population models, 30–31 population recovery potential, determinant of, 29 regulatory challenges, 28–30 regulatory framework, 27–28 tiered approach, 28 toxicological sensitivity, 31 ultima ratio, 29 univoltine life cycles, aquatic invertebrates, 29 voltinism of test species, 28
126 Population modeling, potential role of in pesticide authorization, 97–103 advantage of population modeling, 97 aquatic organisms, 99–101 FOCUS step 4 calculations, 101 guidance document for, 102 mesocosm data, 100 rapidly reproducing species, 100 disadvantages of population modeling, 101 introduction, 97–98 major problem in population modeling, 102 nontarget arthropods, 99 options for risk refinement, 98 population model hurdles, 103 regulatory acceptance and outlook, 101–103 terrestrial vertebrates, 98–99 toxicity exposure ratio, 101 Population recovery potential, determinant of, 29 PPPs, see Plant protection products PRAPeR, see Pesticide risk assessment peer review unit Predicted environmental concentrations (PECs), 52 Probability density function (PDF), 50
r Rabies model, 74 RAMAS model, 38 Rat–elephant phenomenon, 40–42 Ring vaccination, 71 Risk quotient, development of, 35
s Scientific innovation, basis of, 28 Silviculture, models for, 78 Small fox tapeworm, 72 Species dose–response curves, matrix models, 42 Species reproduction, pesticide application and, 91 State of the art, 77–87 ALMaSS framework, 83 appendix, 85–87 beech forest model, 80 BIOSIS database, 80, 81 CAB Abstracts database, 80, 81 cellular automata, 81 DEBtox, 82 developmental stages, 80 difference equations, 83 discussion, 83–85 good modeling practice, 85 introduction, 78–79 measures of toxicity, 81 methods, 80–81 model assumptions, 84
Index
model diversity, 84 PASCAL database, 80, 81 population model types, 80 quantification of risk, 84 results, 81–83 silviculture, models for, 78 testing, verification, calibration, and validation, 79–80 toxicity endpoints, 82 Stochastic matrix models, advantage of over deterministic models, 38
t Terrestrial ecotoxicology, guidance document, 102 Terrestrial vertebrates, 98–99 TERs, see Toxicity exposure ratios Test species voltinism, PPP regulatory challenges, 28 Tiered approach, regulatory risk assessments, 28 Toxicity exposure ratios (TERs), 11, 34–35, 101, 105 Typhlodromus pyri, 99
u Ultima ratio, 29 Unexpected events, 91 Univoltine life cycles, aquatic invertebrates, 29 US Environmental Protection Agency (USEPA), 34 levels of concern, 35 population-level assessments, 34 USEPA, see US Environmental Protection Agency
v Verification, 79 Vinclozolin, 64, 65 Voltinism, test species, 28
w Wildlife diseases, ecological models, 67–75 alternative management concept, 70 antirabies vaccination, 72 comparison of characteristic aspects of controlling wildlife pathogens, 68 discussion, 74–75 disease agents, 68 ecological models substituting field studies, 69–70 background, 69 conceptual model, 69 ecological model, 69–70 problem, 69 results, 70
127
Index
ecological models supporting design of field application, 72–74 background, 72 conceptual model, 72 ecological model, 72 problem, 72 results, 74 epidemics, 67 examples, 69–74 “fencing the outbreak” strategy, 69 introduction, 67–68
management decisions, 75 management strategies, 68 mind models, 75 parasitic cycle modeling, 72 predator–prey relationship, 72 rabies model, 74 red fox, parasite of, 72 ring vaccination, 71 small fox tapeworm, 72 Wildlife diseases, risk assessment of,
other Titles from the Society of environmental Toxicology and Chemistry (SeTAC) Freshwater Bivalve Ecotoxicology Farris, Van Hassel, editors 2006 Estrogens and Xenoestrogens in the Aquatic Environment: An Integrated Approach for Field Monitoring and Effect Assessment Vethaak, Schrap, de Voogt, editors 2006 Assessing the Hazard of Metals and Inorganic Metal Substances in Aquatic and Terrestrial Systems Adams, Chapman, editors 2006 Perchlorate Ecotoxicology Kendall, Smith, editors 2006 Natural Attenuation of Trace Element Availability in Soils Hamon, McLaughlin, Stevens, editors 2006 Mercury Cycling in a Wetland-Dominated Ecosystem: A Multidisciplinary Study O’Driscoll, Rencz, Lean 2005 Atrazine in North American Surface Waters: A Probabilistic Aquatic Ecological Risk Assessment Giddings, editor 2005 Effects of Pesticides in the Field Liess, Brown, Dohmen, Duquesne, Hart, Heimbach, Kreuger, Lagadic, Maund, Reinert, Streloke, Tarazona 2005 Human Pharmaceuticals: Assessing the Impacts on Aquatic Ecosystems Williams, editor 2005
141
Toxicity of Dietborne Metals to Aquatic Organisms Meyer, Adams, Brix, Luoma, Stubblefield, Wood, editors 2005 Toxicity Reduction and Toxicity Identification Evaluations for Effluents, Ambient Waters, and Other Aqueous Media Norberg-King, Ausley, Burton, Goodfellow, Miller, Waller, editors 2005 Use of Sediment Quality Guidelines and Related Tools for the Assessment of Contaminated Sediments Wenning, Batley, Ingersoll, Moore, editors 2005 Life-Cycle Assessment of Metals Dubreuil, editor 2005 Working Environment in Life-Cycle Assessment Poulsen, Jensen, editors 2005 Life-Cycle Management Hunkeler, Saur, Rebitzer, Finkbeiner, Schmidt, Jensen, Stranddorf, Christiansen 2004 Scenarios in Life-Cycle Assessment Rebitzer, Ekvall, editors Ecological Assessment of Aquatic Resources: Linking Science to Decision-Making Barbour, Norton, Preston, Thornton, editors 2004 Life-Cycle Assessment and SETAC: 1991–1999 15 LCA publications on CD-ROM 2003 Amphibian Decline: An Integrated Analysis of Multiple Stressor Effects Greg Linder, Sherry K. Krest, Donald W. Sparling 2003 Metals in Aquatic Systems: A Review of Exposure, Bioaccumulation, and Toxicity Models Paquin, Farley, Santore, Kavvadas, Mooney, Winfield, Wu, Di Toro 2003
Silver: Environmental Transport, Fate, Effects, and Models: Papers from Environmental Toxicology and Chemistry, 1983 to 2002 Gorusch, Kramer, La Point 2003 Code of Life-Cycle Inventory Practice de Beaufort-Langeveld, Bretz, van Hoof, Hischier, Jean, Tanner, Huijbregts, editors 2003 Contaminated Soils: From Soil–Chemical Interactions to Ecosystem Management Lanno, editor 2003 Life-Cycle Assessment in Building and Construction Kotaji, Edwards, Shuurmans, editors 2003 Porewater Toxicity Testing: Biological, Chemical, and Ecological Considerations Carr, Nipper, editors 2003 Community-Level Aquatic System Studies — Interpretation Criteria (CLASSIC) Giddings, Brock, Heger, Heimbach, Maund, Norman, Ratte, Schäfers, Streloke, editors 2002 Interconnections between Human Health and Ecological Variability Di Giulio, Benson, editors 2002 Life-Cycle Impact Assessment: Striving towards Best Practice Udo de Haes, Finnveden, Goedkoop, Hauschild, Hertwich, Hofstetter, Jolliet, Klöpffer, Krewitt, Lindeijer, Müller-Wenk, Olsen, Pennington, Potting, Steen, editors 2002 Silver in the Environment: Transport, Fate, and Effects Andren, Bober, editors 2002 Test Methods to Determine Hazards for Sparingly Soluble Metal Compounds in Soils Fairbrother, Glazebrook, van Straalen, Tararzona, editors 2002
2002 Test Methods to Determine Hazards for Sparingly Soluble Metal Compounds in Soils Fairbrother, Glazebrook, van Straalen, Tarazona, editors 2002 9th LCA Case Studies Symposium 2001 Avian Effects Assessment: A Framework for Contaminants Studies Hart, Balluff, Barfknecht, Chapman, Hawkes, Joermann, Leopold, Luttik, editors 2001 Bioavailability of Metals in Terrestrial Ecosystems: Importance of Partitioning for Bioavailability to Invertebrates, Microbes, and Plants Allen, editor 2001 Ecological Variability: Separating Natural from Anthropogenic Causes of Ecosystem Impairment Baird and Burton, editors 2001 Guidance Document on Regulatory Testing and Risk Assessment Procedures for Protection Products with Non-Target Arthropods (ESCORT 2) Candolfi, Barrett, Campbell, Forster, Grady, Huet, Lewis, Schmuck, Vogt, editors 2001 Impacts of Low-Dose, High-Potency Herbicides on Nontarget and Unintended Plant Species Ferenc, editorToxicology and Chemistry (SETaC) 402 Society of Environmental 2001 Risk Management: Ecological Risk-Based Decision-Making Stahl, Bachman, Barton, Clark, deFur, Ells, Pittinger, Slimak, Wentsel, editors 2001 8th LCA Case Studies Symposium 2000 Development of Methods for Effects-Driven Cumulative Effects Assessment Using8/28/09 Fish Populations: Moose River Project Munkittrick, McMaster, Van Der Kraak, Portt, Gibbons, Farwell, Gray, 2000
K10692_Book.indb 401
Ecotoxicology of Amphibians and Reptiles Sparling, Linder, Bishop, editors 2000 Environmental Contaminants and Terrestrial Vertebrates: Effects on Populations, Communities, and Ecosystems Albers, Heinz, Ohlendorf, editors 2000 Evaluation of Persistence and Long-Range Transport of Organic Chemicals in the Environment Klecka, Boethling, Franklin, Grady, Graham, Howard, Kannan, Larson, Mackay, Muir, van de Meent, editors 2000 Multiple Stressors in Ecological Risk and Impact Assessment: Approaches to Risk
8:57:24 PM
2000 Environmental Contaminants and Terrestrial Vertebrates: Effects on Populations, Communities, and Ecosystems Albers, Heinz, Ohlendorf, editors 2000 Evaluation of Persistence and Long-Range Transport of Organic Chemicals in the Environment Klecka, Boethling, Franklin, Grady, Graham, Howard, Kannan, Larson, Mackay, Muir, van de Meent, editors 2000 Multiple Stressors in Ecological Risk and Impact Assessment: Approaches to Risk Estimation Ferenc and Foran, editors 2000 Natural Remediation of Environmental Contaminants: Its Role in Ecological Risk Assessment and Risk Management Swindoll, Stahl, Ells, editors 2000 7th LCA Case Studies Symposium 1999 Evaluating and Communicating Subsistence Seafood Safety in a CrossCultural Context: Lessons Learned from the Exxon Valdez Oil Spill Field, Fall, Nighswander, Peacock, Varanasi, editors 1999 Ecotoxicology and Risk Assessment for Wetlands Lewis, Mayer, Powell, Nelson, Klaine, Henry, Dickson, editors Society of Environmental Toxicology and Chemistry (SETaC) 1999
403
Endocrine Disruption in Invertebrates: Endocrinology, Testing, and Assessment DeFur, Crane, Ingersoll, Tattersfield, editors 1999 Guidance Document on Higher-Tier Aquatic Risk Assessment for Pesticides (HARAP) Campbell, Arnold, Brock, Grandy, Heger, Heimbach, Maund, Streloke, editors 1999
K10692_Book.indb 402
Linkage of Effects to Tissue Residues: Development of a Comprehensive Database for Aquatic Organisms Exposed to Inorganic and Organic Chemicals Jarvinen and Ankley, editors 1999 Multiple Stressors in Ecological Risk and Impact Assessment Foran and Ferenc, editors 1999 Reproductive and Developmental Effects of Contaminants in Oviparous Vertebrates Di Giulio and Tillitt, editors 1999 Restoration of Lost Human Uses of the Environment Grayson Cecil, editor 1999 6th LCA Case Studies Symposium
8/28/09 8:57:24 PM
Foran and Ferenc, editors 1999 Reproductive and Developmental Effects of Contaminants in Oviparous Vertebrates Di Giulio and Tillitt, editors 1999 Restoration of Lost Human Uses of the Environment Grayson Cecil, editor 1999 6th LCA Case Studies Symposium 1998 Advances in Earthworm Ecotoxicology Sheppard, Bembridge, Holmstrup, Posthuma, editors 1998 Ecological Risk Assessment: A Meeting of Policy and Science Peyster and Day, Editors 1998 Ecological Risk Assessment Decision-Support System: A Conceptual Design Reinert, Bartell, Biddinger, editors 1998 Ecotoxicological Risk Assessment of the Chlorinated Organic Chemicals Carey, Cook, Giesy, Hodson, Muir, Owens, Solomon, editors 1998 Principles and Processes for Evaluating Endocrine Disruption in Wildlife Kendall, Dickerson, Giesy, Suk, editors 404 Society of Environmental Toxicology and Chemistry (SETaC) 1998 Radiotelemetry Applications for Wildlife Toxicology Field Studies Brewer and Fagerstone, editors 1998 Sustainable Environmental Management Barnthouse, Biddinger, Cooper, Fava, Gillett, Holland, Yosie, editors 1998 K10692_Book.indb 403
Uncertainty Analysis in Ecological Risk Assessment Warren-Hicks and Moore, 8/28/09 editors 1998 5th LCA Case Studies Symposium 1997 Atmospheric Deposition of Contaminants to the Great Lakes and Coastal Waters Baker, editor 1997
Biodegradation Kinetics: Generation and Use of Data for Regulatory DecisionMaking Hales, Feijtel, King, Fox, Verstraete, editors 1997 Biotransformation in Environmental Risk Assessment Sijm, de Bruijn, de Voogt, de Wolf, editors 1997
8:57:24 PM
seTac A Professional Society for Environmental Scientists and Engineers and Related Disciplines Concerned with Environmental Quality The Society of Environmental Toxicology and Chemistry (SETAC), with offices currently in North America and Europe, is a nonprofit, professional society established to provide a forum for individuals and institutions engaged in the study of environmental problems, management and regulation of natural resources, education, research and development, and manufacturing and distribution. Specific goals of the society are • Promote research, education, and training in the environmental sciences. • Promote the systematic application of all relevant scientific disciplines to the evaluation of chemical hazards. • Participate in the scientific interpretation of issues concerned with hazard assessment and risk analysis. • Support the development of ecologically acceptable practices and principles. • Provide a forum (meetings and publications) for communication among professionals in government, business, academia, and other segments of society involved in the use, protection, and management of our environment. These goals are pursued through the conduct of numerous activities, which include: • Hold annual meetings with study and workshop sessions, platform and poster papers, and achievement and merit awards. • Sponsor a monthly scientific journal, a newsletter, and special technical publications. • Provide funds for education and training through the SETAC Scholarship/Fellowship Program. • Organize and sponsor chapters to provide a forum for the presentation of scientific data and for the interchange and study of information about local concerns. • Provide advice and counsel to technical and nontechnical persons through a number of standing and ad hoc committees. SETAC membership currently is composed of more than 5000 individuals from government, academia, business, and public-interest groups with technical backgrounds in chemistry, toxicology, biology, ecology, atmospheric sciences, health sciences, earth sciences, and engineering. If you have training in these or related disciplines and are engaged in the study, use, or management of environmental resources, SETAC can fulfill your professional affiliation needs. All members receive a newsletter highlighting environmental topics and SETAC activities and reduced fees for the Annual Meeting and SETAC special publications. All members except Students and Senior Active Members receive monthly issues of Environmental Toxicology and Chemistry (ET&C) and Integrated Environmental Assessment and Management (IEAM), peer-reviewed journals of the Society. Student and Senior Active Members may subscribe to the journal. Members may hold office and, with the Emeritus Members, constitute the voting membership. If you desire further information, contact the appropriate SETAC Office. 1010 North 12th Avenue Pensacola, Florida 32501-3367 USA T 850 469 1500 F 850 469 9778 E
[email protected]
Avenue de la Toison d’Or 67 B-1060 Brussels, Belgium T 32 2 772 72 81 F 32 2 770 53 86 E
[email protected]
www.setac.org Environmental Quality Through Science®
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 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 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.
Arable fields Building Coniferous forest Coniferous forest Mixed forest Parkland/Recreational grass Permanent pasture Scrub Unmangaged grassland Water Young forest plantation
Figure 7.1 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.
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 (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.
Figure 7.3 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.
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 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.
Figure 8.2 Screenshot of the model landscape of the rabies model, applied to compare alternative emergency control options. Pixels represent fox families that are either completely healthy (green), contain infected animals (red), contain at least 1 contagious animal (black), contain immunized animals (light blue), or are died off (white).
Figure 8.4 Schematic representation of the fox tapeworm model. Grid cells represent vole families; cylinders symbolize individual faces either loaded with eggs (red) or not containing eggs (black). Accordingly, if a vole family lives inside a defecated grid cell, it switches to infected for several weeks (pink vole). While the modeled individual foxes prey on the grid, they might randomly catch an infected vole from a grid cell where infestation was assigned formerly. The worms are developing during the next time steps and the cycle starts again after the next defecation (after Hansen 2001).