Scholarly article on topic 'How to use mechanistic effect models in environmental risk assessment of pesticides: Case studies and recommendations from the SETAC workshop MODELINK'

How to use mechanistic effect models in environmental risk assessment of pesticides: Case studies and recommendations from the SETAC workshop MODELINK Academic research paper on "Environmental engineering"

0
0
Share paper
Academic journal
Integr Environ Assess Manag
OECD Field of science
Keywords
{""}

Academic research paper on topic "How to use mechanistic effect models in environmental risk assessment of pesticides: Case studies and recommendations from the SETAC workshop MODELINK"

Integrated Environmental Assessment and Management — Volume 12, Number 1—pp. 21-31

© 2015 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals, Inc. on behalf of SETAC 21

How to Use Mechanistic Effect Models in Environmental Risk Assessment of Pesticides: Case Studies and Recommendations from the SETAC Workshop MODELINK

Udo Hommen,*f Valery Forbes, f§ Volker Grimm, k Thomas G Preuss,#ff Pernille Thorbek,ff and Virginie Ducrot§§kk

fFraunhofer Institute for Molecular Biology and Applied Ecology IME, Schmallenberg, Germany

fSchool of Biological Sciences, University of Nebraska, Lincoln, Nebraska, USA

§Present address: College of Biological Sciences, University of Minnesota, St. Paul, Minnesota, USA

¡¡Helmholtz Centre for Environmental Research—UFZ, Leipzig, Germany

#RWTH Aachen University, Institute of Environmental Research, Aachen, Germany

ffPresent address: Bayer CropScience AG, Monheim am Rhein, Germany

ffSyngenta Limited, Product Safety, Jealott's Hill International Research Centre, United Kingdom §§INRA, Rennes, France

kkPresent address: Bayer CropScience AG, Monheim am Rhein, Germany (Submitted 16 May 2015; Accepted 23 June 2015)

EDITOR'S NOTE:

This is 1 of 6 articles reporting on the results of a SETAC technical workshop entitled "MODELINK: How to use ecological effect models to link ecotoxicological tests to protection goals," held in Le Croisic, France, in October 2012 and in Monschau, Germany, in April 2013. The main objective of the workshop was to provide case studies and recommendations relating to the application of mechanistic effects models in environmental risk assessment of pesticides. Models, species, and criteria used in MODELINK should be viewed as examples serving the purpose of illustrating how such models could be used for solving specific risk assessment issues.

ABSTRACT

Mechanistic effect models (MEMs) are useful tools for ecological risk assessment of chemicals to complement experimentation. However, currently no recommendations exist for how to use them in risk assessments. Therefore, the Society of Environmental Toxicology and Chemistry (SETAC) MODELINK workshop aimed at providing guidance for when and how to apply MEMs in regulatory risk assessments. The workshop focused on risk assessment of plant protection products under Regulation (EC) No 1107/2009 using MEMs at the organism and population levels. Realistic applications of MEMs were demonstrated in 6 case studies covering assessments for plants, invertebrates, and vertebrates in aquatic and terrestrial habitats. From the case studies and their evaluation, 12 recommendations on the future use of MEMs were formulated, addressing the issues of how to translate specific protection goals into workable questions, how to select species and scenarios to be modeled, and where and how to fit MEMs into current and future risk assessment schemes. The most important recommendations are that protection goals should be made more quantitative; the species to be modeled must be vulnerable not only regarding toxic effects but also regarding their life history and dispersal traits; the models should be as realistic as possible for a specific risk assessment question, and the level of conservatism required for a specific risk assessment should be reached by designing appropriately conservative environmental and exposure scenarios; scenarios should include different regions of the European Union (EU) and different crops; in the long run, generic MEMs covering relevant species based on representative scenarios should be developed, which will require EU-level joint initiatives of all stakeholders involved. The main conclusion from the MODELINK workshop is that the considerable effort required for making MEMs an integral part of environmental risk assessment of pesticides is worthwhile, because it will make risk assessments not only more ecologically relevant and less uncertain but also more comprehensive, coherent, and cost effective. Integr Environ Assess Manag 2016;12:21-31. © 2015 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals, Inc. on behalf of SETAC.

Keywords: Ecological risk assessment Plant protection products Mechanistic effect model Population model TK-TD model

* Address correspondence to udo.hommen@ime.fraunhofer.de Published online 6 October 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/ieam.1704

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

INTRODUCTION

To gain authorization of plant protection products, EC Regulation 1107/2009 requires an environmental risk assessment that demonstrates that the use of the product following good agricultural practices will not cause any unacceptable impacts on individuals (for some taxa) or populations or functions of nontarget species in the field. Currently, risk

assessments mainly rely on predicted exposures and ecotox-icological data generated for birds and mammals, terrestrial invertebrates (foliar and soil), and plants, as well as fish, aquatic invertebrates, macrophytes, and algae.

Despite this broad coverage provided by experimental protocols, not every situation can be tested experimentally. Mechanistic effect models (MEMs) have therefore been suggested as useful tools to complement experimentation in higher-tier risk assessments (e.g., Brock et al. 2009; Forbes et al. 2009; Galic et al. 2010; Hommen et al. 2010; Schmolke et al. 2010; EFSA 2010; DG SANCO 2012; EFSA 2013a; EFSA 2013b; Fischer and Moriarty 2014). The MEMs comprise ecological models, such as population models, and organism-level effect models, such as toxicokinetic-toxicodynamic (TK-TD) models.

From a practical point of view, MEMs offer the possibility to cover issues that cannot be dealt with in experiments in the laboratory or the field. Indeed, a number of questions may arise during the course of a risk assessment, with regard to, for instance, the temporal scale of effects, extrapolation to untested species, or exploration of factors identified as sources of uncertainties. Uncertainty related to these issues is 1 of the main reasons why models are currently proposed for risk assessments. Furthermore, the ability of ecological modeling to extrapolate from laboratory conditions and tests addressing individual organisms to the population or community level under field conditions presents an opportunity to improve the link between standard ecotoxicological data and protection goals, thus providing better risk assessments. Therefore, the relevance and usefulness of ecological models have been recognized by academia, industry, and regulators, as reflected in the opinion on "Addressing the New Challenges for Risk Assessment" (DG SANCO 2012), which stressed the importance of mechanistic effect modeling.

Yet, in contrast to exposure modeling (e.g., FOCUS 2001), effect modeling has not yet been extensively used by applicants and has often not been accepted in risk assessments by the regulatory authorities (Wogram 2009, Schmolke et al. 2010). Because the use of modeling does not necessarily increase the likelihood of getting a dossier accepted, modeling still has a high cost/benefit ratio for the applicants. Conversely, even though regulators are interested in MEMs as tools, they have not yet accepted them as stand-alone evidence in a risk assessment, only as weight of evidence at best. This results in a situation in which investment in and submission of MEMs to regulatory authorities are limited by a lack of confidence that they will be accepted. The MODELINK workshop set out to improve this situation by demonstrating, in 6 case studies targeting different groups of organisms and regulatory contexts, how MEMs can be integrated into current and future risk assessment schemes.

The aim of this article is to present the motivation for the MODELINK workshop and its rationale, and to provide a summary of its case studies and recommendations. Therefore, we first give a short overview of MEMs, refer to activities linked to the use of such models, and describe the specific objectives and scope of the workshop. We provide short summaries of the 6 case studies, 5 of which are described in greater detail in the other articles of this special series. Finally, we list the general recommendations of the workshop on when and how MEMs should be used in environmental risk assessments.

A SHORT OVERVIEW OF MEMs

In this article, we use the term MEMs to refer to abstractions of real systems that represent biological, physical, andchemical processes and their consequences within and across levels of biological organization in a mechanistic way (Grimm and Martin 2013). Because models on the level of the organism are included, we avoid the term "ecological models" here, which implies models at population and higher levels of biological organization.

At the organism level, TK-TD models simulate survival or sublethal effects (e.g., Ashauer et al. 2011) over time. The hypothesis underlying TK-TD models is that the concentration at the target site, and not the external exposure concentration, triggers the effect of a toxicant. The TK-TD models represent the dynamics of the uptake and elimination of toxicants into an organism (toxicokinetics) and in a second step the effect over time on the organism (toxicodynamics) as a function of the internal concentration. The TK-TD models have been shown to deliver useful insights and predictions, and they have been applied to many different species and compounds with different modes of action, which has led to a diversity of approaches used, each having its own terminology, assumptions, symbols, and equations. However, most of the published aquatic TK-TD models to simulate survival over time have now been unified in a single coherent framework (Jager et al. 2011). Only a few toxicodynamic approaches for sublethal effects are available (Ashauer et al. 2011), and most applications are based on Dynamic Energy Budget (DEB) theory (Jager et al. 2006; Martin et al. 2014).

Population models allow extrapolation of lethal and sublethal effects from the individual to the population level (Pastorok et al. 2002, Schmolke et al. 2010). The effects of pesticides on populations of nontarget organisms do not only depend on exposure and toxicity at the organism level, but also on ecological factors that are relevant at the population level, such as traits (e.g. dispersal abilities, generation time, fecundity), population structure, density dependence, timing of exposure, landscape structure, community structure, and occurrence of other stressors.

Population models range from very simple generic models, such as exponential or logistic growth (Barnthouse 2004), over-matrix models (Caswell 2001; Stark et al. 2004; Ibrahim et al. 2014), which differentiate age classes or developmental states, to individual-based models (sometimes called agent-based models) taking into account variability among individuals, their local interactions, and their adaptive behaviour (e.g., Van den Brink et al. 2007; Preuss et al. 2010). More complex individual-based models also may take into account resource dynamics and spatial landscape patterns (e.g., Topping et al. 2003; Wang and Grimm 2010; Topping et al. 2014).

The simplest generic population models are already used in environmental risk assessment to analyze the data from biotests conducted at the population level. The estimation of growth rates in algae or Lemna sp. test protocols is based on the exponential growth model, as well as the calculation of the intrinsic rate of increase proposed in the Daphnia magna reproduction test (OECD 2012). In ecology, conservation biology, and natural resource management, population models, in particular matrix- and individual-based models, are widely used and accepted, but they are not yet regularly used in environmental risk assessment.

Community and ecosystem models allow risk characterization within a community or ecosystem context (Pastorok

et al. 2002). These models consider biotic interactions (e.g., predator-prey relationships, competition) and (in the case of ecosystem models) also abiotic factors (e.g., light, nutrients, temperature). Examples of models of aquatic ecosystems in the context of risk assessment of chemicals can be found in Hommen et al. (1993); Traas et al. (2004); Park and Clough (2012); and Bartell et al. (2013). Models exist also for terrestrial ecosystems, but Pastorok et al. (2002) found only 1 model that considered toxic effects, that is, effects of sulfur dioxide on a grassland community. The use of ecosystem models in ecology as well as ecotoxicology goes back to the early 1980s (e.g., O'Neill et al. 1982). Because of the high number of species and interactions simulated, ecosystem models are usually differential equation models with each population described by just 1 summary state variable in which the populations are understood as compartments. In MODELINK, we did not include any case studies using ecosystem models, but they were the topic of a keynote lecture, and their potential use was discussed.

WORKSHOP OBJECTIVES AND SCOPE

In recent years several initiatives and projects aimed at exploring the potential for using MEMs in ecological risk assessments of chemicals. The LEMTOX workshop (Forbes et al. 2009) identified opportunities and obstacles for the use of MEMs in pesticide risk assessment. Shortly later, the US Environmental Protection Agency (USEPA) organized an international workshop on population-level ecological risk assessment, also covering population models (USEPA 2009). The SETAC Europe advisory group MEMORISK (Preuss et al. 2009) was established to provide a forum for communication through, for example, the organization of group meetings and sessions at SETAC conferences as well as an advisory group for regulators and model end-users willing to use models in risk assessment. In parallel, the EU-funded CREAM Initial Training Network (http://cream-itn.eu) offered a unique platform for educating young scientists in the field of MEMs, that is, model development, testing, and documentation. CREAM (Grimm et al. 2009) also delivered models and tools that can be applied in regulatory risk assessments (see special issues Grimm and Thorbek [2014] and Galic and Forbes [2014]).

The focus of CREAM was on formulating sound guidance on good modeling practice and model documentation (Grimm et al. 2014). Recently, the European Food Safety Authority (EFSA) has also published a scientific opinion on good modeling practice (EFSA 2014). However, neither of these

initiatives and resulting documents provides guidance on how to actually use MEMs in environmental risk assessment within the European risk assessment scheme for plant protection products. Therefore a need exists for guidance on how MEMs can be used to address specific risk assessment questions, and thus, the general aim of the MODELINK workshop was to provide guidance for when and how to apply MEMs to regulatory risk assessments of pesticides and to address the overarching questions listed in Box 1.

The workshop focused on the risk assessment of plant protection products, because information on exposure patterns and the relevant groups of organisms is more developed than in other regulatory frameworks for chemicals. Also, specific protection goals have so far only been suggested for pesticides (EFSA 2010). The groups of organisms considered in the workshop were thus the taxa covered in EC Regulation 1107/2009, that is, algae, macrophytes, aquatic invertebrates, fish, soil invertebrates, nontarget arthropods, birds, and mammals. Honeybees and other pollinators were not addressed because another SETAC workshop covered modeling of pollinators (Fischer & Moriarty 2014), and new EFSA guidance was being produced in parallel to that workshop (EFSA 2013a). Whereas the MODELINK workshop focused on prospective pesticide risk assessments under European regulations, CropLife America recently organized a Science Forum with the aim to gain a better understanding of the current status of population models and how they could be used in ecological risk assessments for threatened and endangered species potentially exposed to pesticides in the United States (Forbes et al. 2015).

The specific protection goals defined in EFSA opinion 1821 (EFSA 2010) were used as the basis to develop concrete/ workable risk assessment questions that were answered in the case studies. These specific protection goals define 6 dimensions: ecological entity (individuals, [meta-] populations, or functional groups), attribute(s) of that entity (behavior, survival or growth, abundance or biomass, processes, biodiversity), magnitude of (tolerable) effect, spatial scale of effect, temporal scale of effect, and degree of certainty that should be achieved by a risk assessment. The levels of biological organization addressed in MODELINK ranged from the organism-level to population-level. Organism-level models (TK-TD) were used to improve understanding and prediction of effects at the organism level from the processes of pesticide uptake, metabolism, and elimination, whereas population models were used to extrapolate individual-level effects to effects on populations, including their recovery.

Box 1. The overarching questions addressed in the MODELINK workshop

1. How to translate protection goals (taking the new specific protection goals based on ecosystem services as suggested in EFSA [2010] as a working example) into concrete and workable problem formulations, including, for example, refined risk assessment questions and corresponding model outputs?

2. How to define relevant scenarios that cover risk assessment questions in terms of species choice (e.g., focal, surrogate, or indicator species) and spatial and temporal scales? Such criteria are important for choice of models, including checking whether existing models are suitable or whether new ones need to be developed.

3. How to define criteria for:

a. Deciding when the use of MEMs can improve risk assessments, that is, when standard data cannot answer the risk assessment questions but MEMs can?

b. Choosing the model type to be used to link standard and higher tier test data to the abovementioned model outputs?

4. How to practically use MEMS and their outputs in the current regulatory risk assessment framework?

The suite of models used in the workshop represents the current state-of-the-art of MEMs, but not all existing types of mechanistic effect models were compared. Models used in the workshop were chosen based on their relevance for ecotox-icology and risk assessment issues, and availability of code and data to be used in case studies. The aim of the workshop was not to promote 1 or a few modeling approaches over others, but to demonstrate how to use MEMs as tools for risk assessment. Therefore, mechanistic models that can be useful in ecological risk assessment are not restricted to the models used during the workshop. Furthermore, model development, documentation, and model testing were not part of the workshop. Thus, the validation status of the models addressed in MODELINK was variable, from little to comprehensive validation (see Augusiak et al. [2014] for an in-depth discussion of model evaluation and validation, or "evaluda-tion") and documentation (see, e.g., Grimm et al. [2014] for recommendations on model documentation). The workshop thus did not address the validation status ofthe models. For the case studies, it was assumed that the models fulfilled the requirements for use in risk assessment (e.g., EFSA 2014) and that potential users of MEMs were familiar with the modeling rationale (e.g., modeling cycle) and with the requirements for using MEMs in risk assessment (as recently summarized by EFSA [2014]). Rather, participants explored how such models could be used to improve risk assessment, as if—for the time being—they were generally "evaludated" and accepted by regulatory authorities. The motivation for this was the strong feedback between the envisaged use of models to support decision making and model "evaludation."

WORKSHOP ORGANIZATION

Before the workshop, suitable and "ready to use" models, that is, models for which executable program, user guide, and documentation, preferably following the TRACE (transparent and comprehensive ecological modeling) recommendations (Schmolke et al. 2010; Grimm et al. 2014), were identified, as well as case-study datasets for model applications. The case studies were fully designed before the workshop. Sixty participants were invited and included experts in modeling, ecotoxicology, and risk assessment (regulatory scientists and policy makers). These experts were selected based on their expertise and to provide sector (academic, government, and private sector), geographical, and sex balance. The workshop was divided into 2 meetings separated by a homework period. During the first meeting (Le Croisic, France, 22-25 October 2012), participants were introduced to ecological modeling in general (and to the selected models in particular), and to the challenges in using the models in the risk assessment of pesticides. Participants were gathered in working groups related to groups of organisms and related guidance documents. Each working group gathered approximately 10 experts with different backgrounds (modeling, aquatic or terrestrial ecotoxicology, risk assessment) to answer a specific risk assessment question (i.e., case study). Participants were introduced to the data, models, and questions or issues to be used or addressed in their case study, including types of model output preferred by modelers and accepted by regulators; translation of protection goals into specific and workable problem formulations; and definition of modeling scenarios, including typical temporal and spatial scales. Modelers trained other case study participants so that they could understand the relevant models in the context of their case study and were able to use the models.

The case studies were designed so that the first tier of the risk assessment failed, and refinement of the risk assessment was needed. Participants were thus asked to identify the problem to be solved and to phrase it in terms of specific protection goals. Different possibilities for refining the risk assessment were discussed, based on guidance relevant to each case study. During the workshop and homework period, participants were invited to conduct refinement of the risk assessment using 1 (or several) of the available models. Results were presented in case study reports, which also presented feedback on the participants' experience with using the models, for example, added value of this approach compared with other refinement approaches, and pending issues in the context of risk assessment. The reports also included recommendations on why, when, and how to use the models.

The outcome of each case study was presented in the second meeting (Monschau, Germany, 22-25 April 2013), with particular emphasis on feedback from participants on the application of the models selected, interpretation of the results, and their use in risk assessment. Recommendations to be provided by the workshop were then discussed in breakout groups, which proposed pragmatic, operational solutions for translating specific protection goals into workable questions, choosing the species to be modeled, designing the ecological scenarios to be modeled, and fitting the results of MEMs into current risk assessment schemes.

SUMMARY OF CASE STUDIES

Here we briefly summarize the 6 case studies of MODELINK, 5 of which are described in detail in the papers of this special series. Table 1 provides an overview of focal species, specific risk assessment questions, and dimensions of specific protection goals addressed in the case studies.

Case Study 1: Using TK-TD Modeling as an Acute Risk Assessment Refinement Approach for Vertebrates

Case study 1 (Ducrot et al. this issue) focused on acute risk assessment in vertebrates. The task was to evaluate the risk from application of a single hypothetical pesticide, via seed treatment in winter cereals, for small granivorous birds (skylarks) and small omnivorous mammals (wood mouse); moreover, to evaluate the risk from another hypothetical pesticide, applied in a pulsed treatment, for fish (fathead minnow). According to the relevant guidance documents (EFSA 2009; EFSA 2013b), the protection goal is to make "any mortality effect unlikely." The tier 1 risk assessments for these hypothetical scenarios failed to indicate acceptable risks when based on the lethal dose or lethal concentration for 50% of the individuals. The EFSA (2009, 2013b) lists several possible options for risk assessment refinement. Cases in which the use of TK or TK-TD models should be preferred over other existing refinement approaches were highlighted.

Toxicokinetic-toxicodynamic modeling was then used to assess whether refining the exposure and effect assessments would yield acceptable risk. The specific objectives of this case study were to address the influence of feeding behavior on body weight-normalized doses of the active substance (and thus its toxic effects) in the skylark and the wood mouse and to realistically address the temporal pattern of exposure and effects in the fathead minnow, using long-term pulsed exposure scenarios, as defined by FOCUS surface water models (FOCUS 2001).

Table 1. Overview of the 6 case studies of MODELINK, including dimensions of specific protection goals as defined in EFSA opinion 1821 (EFSA 2010), problem addressed, focal species, type of model used, and model output

No Ecosystem services keydriver Problem Focal species Entity Attribute Spatial scale Temporal scale Model type Model output

1 Vertebrates Effects of time variable feeding Skylark, wood mouse Organism Survival In-crop 1d TK Maximum internal concentration

Effects of time variable exposure on survival Fathead minnow Organism Survival Edge-of-field water body 485 d TK-TD Survival after 485 d

2 Vertebrates Extrapolation to population-level effects Common vole, field vole, wood mouse Population Abundance 25, 40, or 10 000 ha 10 y exposure 20 y post exposure Spatially explicit IBM Maximum predicted effect on Jan 1

3 Soil invertebrates Effects of application scenarios on populations Eisenia fetida, Folsomia candida Population Abundance & biomass In-crop 1y Spatially explicit IBM Time to recovery

4 Terrestrial invertebrates Recovery of populations Linyphyiid spiders, carabid beetle Population Abundance In-field / off-field A few years Spatially explicit IBM Time to recovery

5 Aquatic invertebrates Effects of time variable exposure on populations Gammarus, Chaoborus, Daphnia Population Abundance Edge of field water body 1 to a few years TK-TD & IBM Magnitude and duration of effect

6 Aquatic macrophytes Effects of time variable exposure on biomass Lemna sp., Myriophyllum spicatum Population Biomass Edge of field water body 1 y TK-TD & biomass growth model Magnitude and duration of effect

Comparison of the results of classical risk assessment with those from the model-based approach showed that TK and TK-TD models constitute the most relevant way to directly connect a range of realistic exposure patterns to effects. They represent realistic exposure scenarios and simulate relevant mechanisms of effects, including delayed toxicity and recovery. The paper provides recommendations on how to properly use TK and TK-TD models in acute risk assessment for vertebrates, with a particular emphasis on the endpoints to be used as a basis for the risk assessment. Recommendations were also provided with respect to the use of models for extrapolation between species, chemicals, and exposure scenarios; for example, extrapolation may require reparame-terization and recalibration of the model (e.g., when TK or TD properties differ between species and chemicals). An example checklist for extrapolation of TK models among pesticides was provided. Toxicodynamic assumptions influence the choice of the best-suited endpoint for risk assessment, which needs to be checked when extrapolating to another pesticide.

Case Study 2: Population-level Risk Assessments for Small Mammals Using Individual-Based Population Models

Case study 2 (Schmitt et al. this issue) presents a chronic risk assessment for small mammals that may be exposed to residues from a fictitious fungicide after spray applications in orchards or cereals, or used as a cereal seed treatment. According to the relevant guidance document (EFSA 2009), the focus of the

long-term risk assessment is on reproductive toxicity as measured in toxicological studies in rodents. A scenario was chosen in which the standard risk assessment failed (i.e., a toxicity exposure ratio < 5) when based on the toxicological NOAEL (No Observed Adversed Effect Level) whereas it passed when considering the LOAEL (Low Observed Adverse Effect Level) instead. Therefore, population modeling was employed to assess whether small effects in reproductive toxicological endpoints (LOAEL level) have considerable influence on population abundance.

The specific objective of this case study was to examine various scenarios covering a range of exposures around the LOAEL for reproductive endpoints, using different population models that were designed for the common vole, the field vole, and the wood mouse, respectively. Results indicated in all 3 cases very low population sensitivity unless the hypothetical pesticide was applied at unrealistically high rates (i.e., 10x intended application rate = 2x LOAEL). Even then, recovery of local population impacts was fast and occurred within 1 y. Only when simulating intentionally high acute toxic effects as a positive control, and when considering large complex landscapes in which local populations were extinguished, were recovery periods of concern. In these cases, recovery was only possible by migration and was not fully achieved within the simulated period 20 y after end of application. Recommendations for population model-based risk assessment include the selection of simulated exposure levels, duration of

simulations, statistically robust number of replicated simulation runs, and endpoints to report. Overall, the application of the population models provided multiple advantages for higher-tier ecological risk assessments for small mammals, including consistent and transparent direct links to specific protection goals, and the consideration of more realistic exposure and environmental scenarios.

Case Study 3: Risk Assessment for Soil Invertebrates Using Spatially Explicit Individual-based Population Modeling

Case study 3 (Reed et al. this issue) explored acute and chronic risks for soil invertebrates that may be exposed to a fungicide applied to potatoes at planting using 3 different application methods: soil treatment in furrow, overall treatment (i.e., full-field spray onto the soil surface to 5 cm depth), and an incorporated treatment (i.e., mixed to 20 cm depth). Terrestrial nonarthropod invertebrates have been identified as a key driver of ecosystem services, and specific protection goals have been developed for this group (EFSA 2010). The relevant protection goal for the problem formulation in this case study was to protect soil function and populations of invertebrate species. Standard nontarget arthropod assessment was not relevant here, because the fungicide was applied directly to the soil. The "in-furrow band spray" method led to high soil exposure because the fungicide was applied directly to the soil, and to difficulties in assessing the impact of treatment in bands because the standard risk assessment assumes incorporation across the entire field. After conducting the standard risk assessment for earthworms, collembolans, and predatory mites, the toxicity exposure ratio values were above the trigger values (as defined by SANCO 2002) for all uses for the acute risk to earthworms and the chronic risk to mites, and for all groups from the preplanting incorporated use. However, toxicity exposure ratios were below the trigger values (indicating unacceptable risk) for chronic risk to earthworms and collembolans from the "in furrow" and "overall" treatments. Therefore, the specific objective of this case study was to examine how MEMs could be used as a refinement option for earthworms and collembolans, with particular emphasis on linking variable spatial exposure to effects on and recovery of earthworm and collembolan populations. Both models are spatially explicit individual-based models (IBM), incorporating individual and landscape variability. The population models suggest that earthworms and springtail populations would likely recover within 1 y after pesticide application regardless of application method. The population modeling also illustrated that a lower predicted average environmental concentration in soil could potentially lead to greater effects at the population level, depending on the spatial heterogeneity of the pesticide and the behavior of the soil organisms. The recommendations regarding the use of MEMs for risk assessment include choosing model outputs that are closely related to specific protection goals, using available toxicity data and accepted fate models to the extent possible in parameterizing models to minimize additional data needs, and testing, evaluating, and documenting models following recent guidance documents.

Case Study 4: Population-level Effects in Nontarget Arthropods Using Landscape-Based IBMs

Case study 4 (workgroup led by P. Thorbek) addressed off-crop risk assessments for nontarget arthropods exposed to insecticide spray application in cereals. Population-level effects

and recovery were addressed based on foliar DT50 (Time for 50% dissipation) and LR50 (Rate for 50% Lethality) for 2 nontarget species (carabid, linyphiid spider), using 2 models with different levels of complexity. A particular emphasis was given to the habitat characteristics, spatial scale, and their representation in the scenarios. The models were used to assess risk both in off-crop habitats and in-field, and the necessary levels of mitigation in the form of unsprayed areas were explored. The specific objective of this case study was to examine how to develop scenarios for simulations from the habitat (i.e., landscape characteristics, size of landscape, number of landscapes), exposure, and species (e.g., in terms of species dispersal abilities) perspectives. Therefore, a suite of scenarios was developed and used for the modeling to cover both a landscape-scale risk assessment and 1-field scenario, the latter being in line with the current risk assessment scheme.

Because nontarget arthropods are a very diverse group, the choice of focal species for both in-crop and off-crop risk assessments needs careful consideration. Because the disposal abilities and life histories of nontarget arthropods are so diverse, the landscape structure relevant for the risk assessment also needs to be carefully considered. In the current risk assessment scheme, lethal and sublethal effects up to 50% would be acceptable if the potential of population recovery at least within a year can be demonstrated. In our modeling scenarios, recovery was defined as reaching 90% of control abundance, which means that in some cases the use of modeling may be more conservative than lower-tier risk assessments. The background for this is that in setting the trigger value for the current risk assessment scheme, a calibration of laboratory to field study effects is used. Often a large difference is found between effects in laboratory studies and field studies, which in part is caused by differences in exposure (e.g., glass plate vs dense plant structures in hedgerows) and especially for off-crop risk assessments additional work is needed to find out what realistic exposures would be.

Case Study 5: Population-level Effects and Recovery of Aquatic Invertebrates After Multiple Applications of an Insecticide Using a Combination of TK-TD and IBM

Case study 5 (Dohmen et al. this issue) explored risk of a rapidly dissipating insecticide (pyrethroid type), which was applied several times during the growing season, to 3 representative aquatic invertebrates. As exposure was pulsed, populations may recover after pulses, and therefore the Ecological Threshold as well as the Ecological Recovery Option (EFSA 2013b) were addressed: effects on population densities should be negligible or should last no longer than days to weeks at the edge of fields and up to several days in protected areas and overall watersheds. The hypothetical experimental data covered 3 tiers: from tier 1 acute and chronic toxicity to daphnids over tier 2 (acute toxicity to a suite of 12 invertebrate species), and derivation of the geometric mean and a species sensitivity distribution to tier 3 (mesocosm study).

The data were designed to indicate unacceptable risk to aquatic invertebrates, and therefore the specific objectives of this case study were to assess latency of effects and population recovery for sensitive species with different recovery patterns as observed in mesocosms (i.e., Gammarus pulex, Chaoboruscrystallinus, andDaphniamagna). Therefore,

TK-TD modeling was used to check whether a 7-d interval was sufficient to consider the peaks in mesocosm studies as toxicologically independent exposure events. Coupling to population models allowed the simulation of effects of fluctuating exposures on and recovery of the populations.

Model predictions were largely in line with field observations or the results of a mesocosm study, providing additional evidence of the suitability and reliability of the models for risk assessment purposes. Because of their flexibility, models can predict the magnitude and duration of unacceptable effects on population-level endpoints, based on previously defined protection goals, for a range of insecticide exposure scenarios and freshwater habitats. Especially the coupling of TK-TD models with population models offers the option to link dynamic exposure to effects on the relevant biological level as defined in the specific protection goals.

Case Study 6: Pesticide Risk Assessment for Aquatic Macrophytes Using TK-TD and Population Models

Case study 6 (Hommen et al. this issue) addressed the risk of an herbicide spray application on aquatic macrophytes in edge-of-field water bodies. Specific protection goals for aquatic macrophytes refer to abundance or biomass of populations and are differentiated for the ecological threshold option (i.e., effects should be negligible) and the ecological recovery option (i.e., small to medium effects of short duration, e.g., weeks to a few months) (EFSA 2013b). Assessments based on laboratory tests with macrophytes were chosen that indicated unacceptable risks for some exposure scenarios. Therefore, this case study aimed at demonstrating how population growth models of 2 macrophyte species (Lemna sp. and Myriophyllum spicatum) coupled to TK-TD models could be used to extrapolate from experimentally tested exposure situations to the diverse and complex patterns predicted by the exposure models (FOCUS 2001) and how the seasonal variability of macrophyte growth in the field could be considered in the risk assessment.

Biomass dynamics of both species were predicted for the unexposed control situation, the originally predicted exposure patterns, and the exposure patterns resulting from multiplying the original concentrations with increasing factors. For the case study, the specific protection goals were preliminarily quantified to derive example decision criteria. Based on a discussion of the uncertainties related to the experimental and modeling approaches, example assessment factors were used to derive regulatory acceptable concentrations based on the models' predictions. The main recommendations from this case study were to develop ecological scenarios and to agree on ecological quantitative protection goals for macrophytes.

WORKSHOP RECOMMENDATIONS

Based on the experiences with the case studies, the second meeting of MODELINK was used to develop general recommendations on the use of mechanistic effect models in pesticide risk assessment under Regulation (EC) No 1107/ 2009. The 12 recommendations described in the following are grouped according to general recommendations, specific recommendations on how to translate specific protection goals to workable questions, how to select species and scenarios to be modeled, and where and how to fit MEMs into current and future risk assessment schemes.

General Recommendations on Mechanistic Effect Models in Environmental Risk Assessment

1. Mechanistic effect models should be used more often to provide added value to refined environmental risk assessments by adding higher ecological realism. The MEMs can use more information from ecotoxicological studies. For instance, they allow optimal use of results from laboratory studies and the possibility to pool results from a suite of independent studies with the same species and chemicals (same or different test protocols); for example, results from absorption-distribution-metabolism-elimination studies and acute toxicity tests with birds and mammals can be considered together, which also holds true for bioaccumulation studies and acute-chronic (e.g., early life stage) studies with fish. The MEMs thereby increase our understanding of potential effects on individuals, with direct application to the risk refinement of tier 1 and tier 2 studies.

Mechanistic effect models also can address biological levels of organization that are difficult to test in the laboratory or in the field, such as populations and communities (tier 3 studies). They are particularly well designed to address population recovery. Thus, they facilitate the implementation of recent guidance where the ecological recovery option is mentioned (e.g., EFSA 2013b). They also can tackle temporal and spatial scales that cannot be easily addressed experimentally, if at all (e.g., long-term effects of repeated exposures in the field, or various effect patterns across EU regions). Because MEMs can be spatially explicit, the influence of habitat and/or landscape characteristics can be investigated in detail, thereby facilitating a priori evaluation of the relative efficacy of different mitigation measures. Mechanistic effect models allow explicit linking of exposure and effects, which is of special importance for the scenario-specific and time-variable exposure of nontarget organisms resulting from pesticide use. They are particularly well suited to forecasting risks from a suite of scenarios covering different exposure patterns and regions of concern. Therefore, MEMs can help to define good agricultural practices that are in line with specific protection goals. Although the MODELINK case studies covered only assessments on the organism and population levels, community and ecosystem models should also be considered among the MEMs for pesticide risk assessment going forward. For example, the analysis and prediction of indirect effects becomes relevant under the ecological recovery option, where some direct effects are considered acceptable, as long as these do not lead to unacceptable indirect effects.

Based on these observations, MEMs have the potential to provide more biological and ecological realism at every step of the current risk assessment scheme, while reducing animal testing. We thus recommend that MEMs are used more often in risk assessments of pesticides to reduce uncertainty.

2. Shared expertise and resources for using MEMs for risk assessment have to be established. The participants articulated the need to clarify the risk assessment issues that modeling is needed to address. Concrete examples were provided in the MODELINK workshop, which did not yet cover the full range of risk assessment issues in which modeling could be successfully used. In this regard, keeping track (e.g., in a data base or authority website) of risk assessments in which models have been submitted and why, as well as their acceptance by regulators, would be interesting. This would

help to build confidence in the use of modeling to support decision making.

Different modeling questions and approaches have different data needs. Data generated in the context of data requirements for the registration of pesticides are often not sufficient to conduct modeling. Participants highlighted the need to better identify the minimal required data set specific to each model. Mechanisms have to be established to share the necessary resources (e.g., data, software), expertise, and feedback among model developers and evaluators on a regular basis.

How to Translate Specific Protection Goals into Workable Questions?

3. The outputs of ecotoxicological tests should be more mechanistically linked to the specific protection goals, such as by the use of MEMs. Specific protection goals are related to organisms (for vertebrates), populations or metapopulations, functions, and communities of nontarget organisms providing ecosystem services in agricultural landscapes (EFSA 2010, 2013b). In contrast to this, effect assessments are often based on organismlevel effects in surrogate species, sometimes refined by highertier testing with populations and communities in the (semi-) field. Mechanistic effect models provide a mechanistic link from measured endpoints to assessment endpoints relevant for the ecosystem services to be protected. For example, in the current guidance document for pollinators, a bee model was used to link mortality of foragers to effects on bee colony size, which is considered to be directly proportional to the ecosystem service of pollination (EFSA 2013a). Therefore, specific protection goals should always be used as a basis for problem formulation (i.e., definition of the question to be solved with the model and of model features: spatial and temporal scales, attributes to be modeled, model outputs, and so forth).

4. Quantitative specific protection goals with respect to magnitude, temporal, and spatial scale of acceptable effects should be developed to facilitate and harmonize interpretation of modeling results and assessing risks. Although the protection goals laid down in the regulation are relatively vague, EFSA (2010) has formulated specific protection goals for ecosystem service providers potentially affected by the use of plant protection products. However, these specific protection goals are still semiquantitative, using terms such as negligible, small, medium, and large for magnitude and days, weeks, or months for duration of effects. In the evaluation of population- or community-level experiments (e.g., mesocosm studies, earthworm or collem-bola field tests), the regulatory acceptable concentration used for risk assessment is usually driven by the evaluation of statistically significant (and ecologically relevant) deviations from the control test systems. For population model results, this is less feasible, because there is either no variability in the results (for deterministic models) or the statistical power is much larger than in experiments because of the much higher number of possible replicates (here: runs of a stochastic model or Monte-Carlo runs). Thus, when the ecological entity (e.g., a specific population) and its attribute for an assessment have been agreed on, decision criteria have to be defined on magnitude, duration, and spatial scale of acceptable effects. Examples of such quantitative criteria are given for reduction of honeybee colony sizes (EFSA 2013a) or in Hommen et al. (this issue) for reduction of aquatic macrophyte biomass. The definition of such quantitative protection goals for the most

representative ecosystem service providers has to be done by groups comprising different experts and stakeholders. Models, for example, landscape models, can be used to define the appropriate spatial scale for an assessment (e.g., Topping et al. 2014).

How to Select and Define Species and Scenarios to be Modeled?

5. The species to be modeled must be relevant for the specific protection goals; they can be standard test species or representative vulnerable species. Not only the sensitivity to the toxicant but also the ecological sensitivity has to be considered. A standard test species might be relevant for modeling if the main question is to extrapolate the effects measured in standard toxicity tests to other exposure conditions or to longer exposure durations. An example is given in Ducrot et al. (this issue) for extrapolating lethal effects on fish from standardized tests to prolonged or time-variable exposure conditions by TK-TD modeling. Such a use of models is comparable to the experimental tier of modified exposure studies (EFSA 2013b), where other uncertainties, such as regarding species-to-species extrapolation or population-level effects, remain. Although experimentally only a few exposure profiles can be tested (and used for calibration and testing of the model), the model can be applied to a large set of scenarios if needed.

A standard test species also can be a relevant species for modeling if it can be considered a representative of vulnerable species in the field, which means it has to be 1) likely to be exposed in the field because of the assessed use of the product, 2) sensitive to direct effects of the stressor, and 3) ecologically sensitive because of its life cycle or dispersal traits. Therefore, if tests with additional species (geometric mean or species sensitivity distribution approach; EFSA 2013b) indicate that the standard species is very sensitive and if its life cycle traits lead to high vulnerability or it is shown that more vulnerable taxa are significantly less sensitive, a model of, for example, Daphnia magna or the rainbow trout can be relevant for the risk assessment.

Generic species, which are supposed to represent a specific group of species, such as small mammals, are used in the risk assessment for birds and mammals (EFSA 2009) and might also be used for mechanistic modeling as long as they do not result in overconservative combinations of several worst-case assumptions. However, models of generic species are difficult to test (validate) against data, and therefore the workshop participants recommended modeling real species. Different approaches exist to identify representative species for vulnerable ecosystem service providers (focal species; see also previous discussion). A workable approach is to start with the physiological or toxicant sensitivity. Within the sensitive group, the traits affecting exposure (seasonality, dispersal) and ecological sensitivity should be identified and used for species selection. However, the availability of data for model development and testing also has to be considered. Examples of identification of vulnerable species are given in Ibrahim et al. (2013, 2014) for fish.

6. The models should be realistic but the scenarios should be sufficiently protective. The workshop differentiated between the models, for example, describing the dynamics of a population depending on environmental factors, and the scenarios defining environmental factors such as temperature,

habitat quality, food supply, landscape structure, agricultural practices, and so forth. Because MEMs are supposed to support risk assessment for a wide range of environmental scenarios and different toxic substances, they should be as realistic as possible for their specific purpose. The workshop participants agreed that the appropriate level of conservatism for a given risk assessment should be obtained through the design of the environmental scenario. This recommendation has been highlighted as well in the EFSA scientific opinion on good modeling practice for the risk assessment of plant protection products (EFSA 2014). Noticeably, modelers should follow the concept of decreasing conservatism and increasing realism that underlies the tiered risk assessment approach and apply this to the model-based risk assessment. This requires focusing particular attention on the development and choice of the various scenarios to be simulated with the model, as highlighted in case study 4. Scenario selection should first be driven by exposure: Where do we expect exposure of nontarget organisms and which conditions (e.g., climate, habitat structure) are appropriately realistic or conservative for exposure and effects?

7. Scenarios should be different for lower and higher tiers and represent different regions in the EU and different crops. At lower tiers, scenarios can be less differentiated and more conservative than at higher tiers. For example, the spatial scale can vary from part of a field over field and edge-of-field up to catchment and landscape scale. For the aquatic risk assessment, the FOCUS exposure scenarios (e.g., its weather data, FOCUS 2001) are considered a good starting point to develop ecological scenarios.

8. A set of standard effect models and scenarios should be developed. The workshop participants agreed that a set of standard effect models and scenarios—similar to the FOCUS models and scenarios for aquatic exposure assessments (FOCUS 2001)—would facilitate model use and increase acceptance of model results. The starting point for a list of standard model species can be the key drivers (groups providing ecosystem services) listed in the EFSA opinion on specific protection goals (EFSA 2010), such as microbes, algae, nontarget aquatic and terrestrial plants, aquatic invertebrates, nontarget arthropods, including honeybees, nontarget non-arthropod terrestrial invertebrates, and aquatic and terrestrial vertebrates. Because different models have to be used for different purposes, it might not be desirable or possible to use only 1 model per key driver as a basis for effect assessment. However, identifying some models or model combinations that best address a range of modeling questions may be possible. Unified modeling frameworks (e.g., general unified threshold model for survival; Jager et al. 2011) should be used when available. Effect modeling frameworks are not yet available for population and community models. The models used as support for risk assessment should be well documented and tested (i.e., according to good modeling practice (EFSA 2014, Grimm et al. 2014).

Where and How Do Mechanistic Effect Models Fit in the Current Risk Assessment?

9. Mechanistic effect models should be considered as refinement option at all higher tiers of environmental risk assessment of chemicals. Mechanistic effect models are not considered appropriate at tier 1, which is based on the core data sets

according to the data requirements but can be used at all higher tiers; for example, organism-level models (TK-TD models) are considered as a tier 2 tool comparable to modified exposure tests, and population and community models are considered as a tier 3 tool corresponding to population- and community-level tests (e.g., mesocosm studies, field tests).

There was consensus during the workshop that MEMs can be used for both vertebrates and invertebrates to answer specific questions that have arisen after standard and highertier studies. For example, they can be used to better describe exposure and effects of organisms (see case study 1, Ducrot et al. this issue) or they can be used to extrapolate effects and recovery of vulnerable populations from a community-level study (e.g., a mesocosm study) to other habitat types or landscapes (e.g., case studies 4 and 5, Dohmen et al. this issue).

In the best case, MEMs can even replace higher-tier testing or additional animal testing, which would be especially useful for vertebrate risk assessment. An example question to be answered could be whether an effect on 1 parameter in a bird reproductive study affects the population.

10. For the use of mechanistic effect models in the risk assessment of pesticides, natural variability should be indicated in the model, and it should be demonstrated that the model can show an effect. The models should preferably be able to simulate individual variability and, where needed, the effects of variable environmental factors. For building confidence in a model and its results, demonstrating that the model is able to predict effects is also important. This can be done by simulating the effects of a known positive control or by a dose-response approach modeling different exposure levels extending from slight to clear effects.

11. Model results should not be evaluated by statistical tests solely but based on confidence or credibility intervals, diagrams, and tables indicating magnitude and duration of effects. Model predictions should be accompanied with confidence (or credibility) intervals, which allow determining the significance of predicted effects with regard to the control situation. Because the statistical significance of endpoint values generated via Monte-Carlo (i.e., stochastic) simulations can easily be influenced by the number of simulation runs, comparing means or medians of treatment and control runs is recommended. Providing different diagrams and tables indicating magnitude and duration of the effects is also recommended. Options are plots of the population dynamics at different exposure levels (including control), plots of abundance or biomass relative to control, or percent deviation from control over time. Dose-response curves can be used to summarize effects on endpoints such as maximum deviation from control, abundance at specific times (e.g., before the start of the reproduction period), annual production, time to recovery, or area under the curve. In addition, tables with magnitude and duration of effects for different exposure levels were found to be useful in the case studies. Plotting safety margins for a suite of simulated scenarios also may be helpful when comparing relative risks across scenarios.

12. Mechanistic effect models can address some of the uncertainties that cannot be fully addressed by laboratory or field experiments, thereby reducing the remaining uncertainty in a risk assessment. Therefore, providing a qualitative assessment of the remaining uncertainty in the risk assessment after ecological

modeling has been conducted is recommended. Mechanistic effects models can bring more realism to risk assessment by accounting for, for example, realistic exposure patterns of animals or plants, habitat or landscape characteristics, or factors that influence the dynamics of populations or communities. Mechanistic effect models also can extrapolate from laboratory to field conditions, from organism-level to population- or community-level effects, between different species and between different exposure or environmental scenarios, thereby providing additional, ecologically relevant insight to risk assessment. In this respect, MEMs can investigate some of the common uncertainties that are inherent to the risk assessment process and make them more explicit. Workshop participants thus recommend providing a table that lists the risk assessment uncertainties covered by the modeling and those that are potentially left uncovered. This table should address the uncertainty that is inherent to the model (e.g., attributable to model parameter input values) and the uncertainties that emerge from the modeled species and scenarios (e.g., uncertainty attributable to the previously mentioned extrapolations). This table should help the risk assessors to evaluate the overall remaining uncertainty in a model-based risk assessment. An example of such a table has been recently provided in the EFSA opinion on good modeling practice for the risk assessment of plant protection products (EFSA 2014). This overall assessment might be used as a basis to identify whether some specific points of concern remain in the risk assessment, which cannot be covered by safety factors or mitigation measures and require further consideration.

CONCLUSIONS

A number of scientific papers and workshops have demonstrated the utility ofmechanistic modeling in ecological risk assessment, which is supported by recent guidance and scientific opinions of stakeholders and authorities. However, a practical appraisal of when and how to use modeling in real risk assessments has been missing. The MODELINK workshop aimed at addressing the "when" and "how" issues in an operational manner, based on an array of case studies that addressed typical risk assessment issues for pesticides. Closing this operational gap is a condition for a wider use of MEMs in risk assessment. It requires a better link between available data from ecotoxicological tests, model outputs, and protection goals. This could be practically achieved in all case studies by deriving quantitative specific protection goals with respect to magnitude, temporal and spatial scale of acceptable effects for interpreting modeling results, and assessing risks.

Providing some general recommendations on when to use modeling approaches in ecological risk assessment ofpesticides was possible. Conversely, providing a general recommendation on how to fit MEMs into pesticide risk assessment schemes was not possible, because specific risk assessment schemes are complex and differ considerably in rationale and protocol for different groups of organisms. The MODELINK workshop provided some concrete examples of how this could be done. Experience from the case studies of MODELINK indicates that MEMs, specific protection goals, and risk assessment schemes should be closely linked and should influence each other; for example, MEMs can help to develop specific protection goals that are actually specific and at the same time generally applicable. For putting the full potential of MEMs into practice and thereby making environmental risk assessment more comprehensive and ecologically relevant,

coordinated actions across, for example, the EU and stakeholders in academia, industry, and regulatory agencies are required. In this respect, sharing experience and ensuring some feedback between modelers and risk assessors is encouraged.

Acknowledgment—We are very grateful to the colleagues in the organizing team (Anne Alix, Domenica Auteri, Patrice Carpentier, Peter Dohmen, Walter Schmitt, and Lina Wendt-Rasch). We would also like to thank all participants for the valuable contributions during the workshop meetings and the elaboration of the case studies. Financial support for the workshop was provided by BASF, Bayer CropScience, CEFIC LRI, INRA, Makhteshim Agan, Sumitomo Chemical, and Syngenta.

REFERENCES

Ashauer R, Agatz A, Albert C, DucrotV, Galic N, Hendriks J, JagerT, Kretschmann A, O'Connor I, Rubach MN, et al. 2011. Toxicokinetic-toxicodynamic modeling of quantal and graded sublethal endpoints: A brief discussion of concepts. Environ Toxicol Chem 30:2519-2524. AugusiakJ, Van den Brink PJ, Grimm V. 2014. Merging validation and evaluation of ecological models to 'evaludation': A review of terminology and a practical approach. Ecological Modelling 280:117-128. Barnthouse LW. 2004. Quantifying population recovery rates for ecological risk

assessment. Environ Toxicol Chem 23:500-508. Bartell SM, Brain RA, Hendley P, Nair SK. 2013. Modeling the potential effects of atrazineon aquatic communities in midwestern streams. Environ Toxicol Chem 32:2402-2411.

BrockTCM, AlixA, Brown C, Capri E, Gottesbueren B, Heimbach F, Lythgo C, Schulz R, Streloke M, editors. 2009. Linking aquatic exposure and effects: Risk assessment of pesticides. Pensacola (FL): SETAC. 440 p. Caswell H. 2001. In Sinauer Associates, editors. Matrix population models: Construction, analysis, and interpretation. 2nd ed. Sunderland (MA): Sinauer Associates. 722 p.

DG SANCO [SCENIHR, SCHER, SCCS; Scientific Committee on Emerging and Newly Identified Health Risks, Scientific Committee on Health and Environmental Risks and Scientific Committee on Consumer Safety]. 2012. Addressing the new challenges for risk assessment. ISBN 978-92-79-XX. Parma (IT): DG SANCO. 157 p.

Dohmen P, Preuss TG, Hamer M, Galic N, Strauss T, Van den Brink PJ, De Laender F, Bopp S. 2016. Population-level effects and recovery of aquatic invertebrates after multiple applications of an insecticide. Integr Environ Assess Manag 12:67-81. DucrotV, Ashauer R, Bednarska AJ, Hinarejos S, ThorbekP, Weyman G. 2016. Using toxicokinetic-toxicodynamic modeling as an acute risk assessment refinement approach in vertebrate ecological risk assessment. Integr Environ Assess Manag 12:32-45.

[EFSA] European Food Safety Authority. Panel on Plant Protection Products and their Residues (PPR). 2010. Scientific Opinion on the development of specific protection goal options for environmental risk assessment of pesticides, in particular in relation to the revision of the Guidance Documents on Aquatic and Terrestrial Ecotoxicology (SANCO/3268/2001 and SANCO/10329/2002). EFSA Journal 8:1821-1875. [EFSA] European Food Safety Authority. 2009. Guidance document on risk assessment for birds & mammals on request from EFSA. EFSA Journal 7:1438.358 p.

[EFSA] European Food Safety Authority. 2013a. Guidance document on the risk assessment of plant protection products on bees (Apis mellifera, Bombus spp. and solitary bees). EFSA Journal 11:3295.268 p. [EFSA] European Food Safety Authority. 2013b. Scientific opinion: Guidance on tiered risk assessment for plant protection products for aquatic organisms in edge-of-field surface waters. EFSA Journal 11:3290.186 p. [EFSA] PPR Panel (EFSA Panel on Plant Protection Products and their Residues). 2014. Scientific opinion on good modelling practice in the context of mechanistic effect models for risk assessment of plant protection products. EFSA Journal 12:3589.92 p. Fischer D, MoriartyT. 2014. Pesticide riskassessment for pollinators. Hoboken (NJ): Wiley-Blackwell. 248 p.

FOCUS (Forum for the co-ordination of pesticide fate models and their use). 2001. FOCUS surface water scenarios in EU evaluation process under 91/414/EEC. Report of the FOCUS Working Group on Surface Water Scenarios. EU Document Reference SANCO/4802/2001-rev2. 245 p.

Forbes VE, Hommen U, Thorbek P, Heimbach F, van den Brink PJ, Wogram J, Thulke HH, Grimm V. 2009. Ecological models in support of regulatory risk assessments of pesticides: developing a strategy for the future. Integr Environ Assess Manag 5:167-172.

Forbes V, Brain R, Edwards D, Galic N, Hal T, Honegger J, Meyer C, Moore D, Nacci D, Pastorok R, et al. 2015. Assessing pesticide risks to threatened and endangered species using population models: Findings and recommendations from a CropLife America Science Forum. Integr Environ Assess Manag 11:348-354.

Galic N, Forbes V. 2014. Ecological models in ecotoxicology and ecological risk assessment: An introduction to the special section. Environ Toxicol Chem 33:1446-1448.

Galic N, Hommen U, Baveco JM, van den Brink PJ. 2010. Potential application of population models in the European ecological risk assessment of chemicals. II. Review of models and their potential to address environmental protection aims. Integr Environ Assess Manag 6:338-360.

Grimm V, Ashauer R, Forbes V, Hommen U, PreussTG, Schmidt A, van den Brink PJ, WogramJ,ThorbekP. 2009. CREAM: A European project on mechanistic effect models for ecological risk assessment of chemicals. Environ Sci Pollut Res 16:614-617.

Grimm V, AugusiakJ, Focks A, Frank B, Gabsi F, Johnston ASA, Liu C, Martin BT, Meli M, Radchuk V, et al. 2014. Towards better modelling and decision support: Documenting model development, testing, and analysis using TRACE. Ecol Model 280:129-139.

Grimm V, Martin BT. 2013. Mechanistic effect modeling for ecological risk assessment: Where to go from here? Integr Environ Assess Manag 9:58-63.

Grimm V, Thorbek P. 2014. Population models for ecological risk assessment of chemicals: Short introduction and summary of a special issue. Ecol Model 280:1-4.

Hommen U, Baveco JM, Galic N, van den Brink PJ. 2010. Potential application of ecological models in the European environmental risk assessment of chemicals. I. Review of protection goals in EU directives and regulations. Integr Environ Assess Manag 6:325-337.

Hommen U, Poethke HJ, Dülmer U, Ratte HT. 1993. Simulation models to predict ecological risks of toxins in freshwater systems. ICES J Marine Sci 50:337-347.

Hommen U, Schmitt W, Heine S, BrockTCM, Duquesne S, Manson P, Meregalli G, Ochoa-Acuna H, van Vliet P, Arts G. 2016. How TK-TD and population models for aquatic macrophytes could support the risk 1 assessment for plant protection products. Integr Environ Assess Manag 12:82-95.

Ibrahim L, Preuss TG, Ratte HT, Hommen U. 2013. A list of fish species that are potentially exposed to pesticides in edge-of-field water bodies in the European Union: A first step towards identifying vulnerable representatives for risk assessment. Environ Sci Pollut Res 20:2679-2687.

Ibrahim L, Preuss TG, Schaeffer A, Hommen U. 2014. A contribution to the identification of representative vulnerable fish species for pesticide risk assessment in Europe: A comparison of population resilience using matrix models. Ecol Model 280:65-75.

Jager T, Albert C, Preuss TG, Ashauer R. 2011. General unified threshold model of survival: A toxicokinetic-toxicodynamic framework for ecotoxicology. Environ Sci Technol 45:2529-2540.

Jager T, Heugens EHW, Kooijman SALM. 2006. Making sense of ecotoxicological test results: Towards application of process-based models. Ecotoxicology 15:305-314.

Martin B, Jager T, Nisbet RM, Preuss TG, Grimm V. 2014. Limitations of extrapolating toxic effects on reproduction to the population level. Ecol Appl 24:1972-1983.

[OECD] Organisation for Economic Co-operation and Development. 2012. Test No. 211: Daphnia magna reproduction test, OECD Guidelines for the Testing of Chemicals, Section 2. Paris (FR): OECD Publishing.

O'Neill RV, Gardner RH, Barnthouse LW, Suter GW, Hildebrand SG, Gehrs CW. 1982. Ecosystem risk analysis: A new methodology. Environ Toxicol Chem 1:167-177.

Park RA, Clough JS. 2012. AQUATOX (release 3.1): Modeling environmental fate and ecological effects in aquatic ecosystems. Vol 2. Technical documentation. Washington (DC): USEPA Office of Water, Office of Science and Technology. EPA 823/R-09/004.

Pastorok RA, Bartell SM, Ferson S. 2002. Ecological modeling in risk assessment: chemical effects on populations, ecosystems, and landscapes. Boca Raton (FL): Lewis. 328 p.

Preuss TG, Hommen U, Alix A, Ashauer A, van den Brink PJ, Chapman P, Ducrot V, Forbes V, Grimm V, Schafer D, et al. 2009. Mechanistic effect models for ecological risk assessment of chemicals (MEMoRisk): A new SETAC-Europe Advisory Group. Environ Sci Pol Res 16:250-252.

Preuss TG, Hammers-Wirtz M, Ratte H. 2010. The potential of individual based population models to extrapolate effects measured at standardized test conditions to relevant environmental conditions: An example for 3, 4-dichloroaniline on Daphnia magna. J Environ Monit 12:2070-2079.

Reed M, AlvarezT, Chelinho S, Forbes VE, Johnston ASA, Meli M,VossF, Pastorok R. 2015. A risk assessment example for soil invertebrates using spatially explicit agent-based models 1 (ABMs). Integr Environ Assess Manag 12:58-66.

SANCO. 2002. DRAFT Working Document Guidance Document on Terrestrial Ecotoxicology Under Council 32 Directive 91/414/EEC. Brussels (BE): Sante des Consommateurs. European Commission, Health and Consumer Protection 33 Directorate-General, SANCO/10329/2002 rev 2 (final).

Schmitt W, Auteri D, Bastiansen F, Ebeling M, Liu C, Luttik R, Mastitsky S, Nacci D, Topping CJ, Wang M. 2015. An example of population-level risk assessments for small mammals using individual-based population models. Integr Environ Assess Manag 12:46-57.

Schmolke A, Thorbek P, Chapman P, Grimm V. 2010. Ecological models and pesticide risk assessment: current modeling practice. Environ Toxicol Chem 29:1006-1012.

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 U SA 101:732-736.

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, Kjsr LJ, Hommen U, HoyeTT, Preuss TG, Sibly RM, van Vliet P. 2014. Recovery based on plot experiments is a poor predictor of landscape-level population impacts of agricultural pesticides. Environ Toxicol Chem 33:1499-1507.

Traas TP, Janse JH, Van den Brink PJ, BrockTCM, 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.

[USEPA] United States Environmental Protection Agency. 2009. Summary report: Risk assessment forum technical workshop on population-level ecological risk assessment. Risk Assessment Forum. Washington (DC): US Environmental Protection Agency. 71 p. [cited 2015 Oct 18]. Available from: http://nepis.epa. gov/Exe/ZyPDF.cgi/P1005KXE.PDF?Dockey=P1005KXE.PDF

Van den Brink PJ, Baveco JM, Verboom J, Heimbach F. 2007. An individual-based approach to model spatial population dynamics of invertebrates in aquatic ecosystems after pesticide contamination. Environ Toxicol Chem 26:22262236.

Wang M, Grimm V. 2010. Population models in pesticide risk assessment: Lessons for assessing population-level effects, recovery, and alternative exposure scenarios from modeling a small mammal. Environ Toxicol Chem 29:12921300.

Wogram J. 2009. Regulatory challenges for the potential use of ecological models in risk assessments of plant protection products. In: Thorbek P, Forbes VE, Heimbach F, Hommen U, Thulke HH, v. d. Brink PJ, Wogram J, Grimm V. Ecological models for regulatory risk assessments of pesticides: Developing a strategy for the future. Boca Raton (FL): CRC Press. p. 27-32.