The Potential for the Use of Agent-Based Models in Ecotoxicology

The Animal Landscape and Man Simulation System (ALMaSS): a history, design, and philosophy

This article is the first article in the new topical RIO journal collection for ALMaSS. This editorial introduces ALMaSS, its history, component parts and philosophy, and forms a first access point for those interested in knowing more. It is written from my own personal perspective as the instigator and main developer for the system, effectively as the ‘father’ of ALMaSS.

Mechanistic modeling of pesticide exposure: The missing keystone of honey bee toxicology

The role of pesticides in recent honey bee losses is controversial, partly because field studies often fail to detect effects predicted by laboratory studies. This dissonance highlights a critical gap in the field of honey bee toxicology: there exists little mechanistic understanding of the patterns and processes of exposure that link honey bees to pesticides in their environment. The authors submit that 2 key processes underlie honey bee pesticide exposure: 1) the acquisition of pesticide by foraging bees, and 2) the in-hive distribution of pesticide returned by foragers. The acquisition of pesticide by foraging bees must be understood as the spatiotemporal intersection between environmental contamination and honey bee foraging activity. This implies that exposure is distributional, not discrete, and that a subset of foragers may acquire harmful doses of pesticide while the mean colony exposure would appear safe. The in-hive distribution of pesticide is a complex process driven principally by food transfer interactions between colony members, and this process differs importantly between pollen and nectar. High priority should be placed on applying the extensive literature on honey bee biology to the development of more rigorously mechanistic models of honey bee pesticide exposure. In combination with mechanistic effects modeling, mechanistic exposure modeling has the potential to integrate the field of honey bee toxicology, advancing both risk assessment and basic research. Environ Toxicol Chem 2017;36:871-881. © 2016 SETAC.

SEARCH: Spatially Explicit Animal Response to Composition of Habitat

Complex decisions dramatically affect animal dispersal and space use. Dispersing individuals respond to a combination of fine-scale environmental stimuli and internal attributes. Individual-based modeling offers a valuable approach for the investigation of such interactions because it combines the heterogeneity of animal behaviors with spatial detail. Most individual-based models (IBMs), however, vastly oversimplify animal behavior and such behavioral minimalism diminishes the value of these models. We present program SEARCH (Spatially Explicit Animal Response to Composition of Habitat), a spatially explicit, individual-based, population model of animal dispersal through realistic landscapes. SEARCH uses values in Geographic Information System (GIS) maps to apply rules that animals follow during dispersal, thus allowing virtual animals to respond to fine-scale features of the landscape and maintain a detailed memory of areas sensed during movement. SEARCH also incorporates temporally dynamic landscapes so that the environment to which virtual animals respond can change during the course of a simulation. Animals in SEARCH are behaviorally dynamic and able to respond to stimuli based upon their individual experiences. Therefore, SEARCH is able to model behavioral traits of dispersing animals at fine scales and with many dynamic aspects. Such added complexity allows investigation of unique ecological questions. To illustrate SEARCH's capabilities, we simulated case studies using three mammals. We examined the impact of seasonally variable food resources on the weight distribution of dispersing raccoons (Procyon lotor), the effect of temporally dynamic mortality pressure in combination with various levels of behavioral responsiveness in eastern chipmunks (Tamias striatus), and the impact of behavioral plasticity and home range selection on disperser mortality and weight change in virtual American martens (Martes americana). These simulations highlight the relevance of SEARCH for a variety of applications and illustrate benefits it can provide for conservation planning.

From home range dynamics to population cycles: validation and realism of a common vole population model for pesticide risk assessment

Despite various attempts to establish population models as standard tools in pesticide risk assessment, population models still receive limited acceptance by risk assessors and authorities in Europe. A main criticism of risk assessors is that population models are often not, or not sufficiently, validated. Hence the realism of population-level risk assessments conducted with such models remains uncertain. We therefore developed an individual-based population model for the common vole, Microtus arvalis, and demonstrate how population models can be validated in great detail based on published data. The model is developed for application in pesticide risk assessment, therefore, the validation covers all areas of the biology of the common vole that are relevant for the analysis of potential effects and recovery after application of pesticides. Our results indicate that reproduction, survival, age structure, spatial behavior, and population dynamics reproduced from the model are comparable to field observations. Also interannual population cycles, which are frequently observed in field studies of small mammals, emerge from the population model. These cycles were shown to be caused by the home range behavior and dispersal. As observed previously in the field, population cycles in the model were also stronger for longer breeding season length. Our results show how validation can help to evaluate the realism of population models, and we discuss the importance of taking field methodology and resulting bias into account. Our results also demonstrate how population models can help to test or understand biological mechanisms in population ecology.

Population models in pesticide risk assessment: lessons for assessing population-level effects, recovery, and alternative exposure scenarios from modeling a small mammal

In the last few years, the interest in using ecological population models as a tool for pesticide risk assessment has increased rapidly. Practical guidance, however, on how to perform a risk assessment with a population model is still lacking. It is still unclear which endpoint (population density, population growth, etc.) is the most sensitive indicator of population-level effects and how risk can be evaluated at the population level. Moreover, a main added value of model-based risk assessments, which is an understanding of the mechanisms involved in alternative exposure scenarios, so far has received little attention. We therefore used an example model to compare commonly used endpoints and alternative exposure scenarios. The model is a structurally realistic, but relatively simple, individual-based, spatially explicit model for the common shrew (Sorex araneus), which was selected because it has been tested and validated extensively. We show that population density is more sensitive for detecting population-level effects in the short term (months) than population growth rate. Population viability measured by extinction risk can also be a relevant endpoint, because it is especially sensitive for small populations. We show that landscape structure and the timing of pesticide application (population structure at the time of application) can have a great impact on population recovery, and we analyze statistical tests for use in population-level risk assessments. Our results demonstrate which factors and insights should be taken into account in population-level risk assessments.

Internet resources for agent-based modelling

The use of agent-based models (ABMs) is steadily increasing in all the disciplines including environmental chemistry and toxicology. This growth is mainly driven by their ability to address problems that conventional modelling techniques cannot, such as the change of scale or the emergence of unanticipated phenomena resulting from interactions between their constitutive goal-directed agents. After a brief introduction on the basic principles of agent-based modelling and the presentation of selected case studies, the main software resources available on the Internet are presented. An attempt is made to estimate the complexity of these tools versus their potentialities and flexibility.

Ecological models and pesticide risk assessment: current modeling practice

Ecological risk assessments of pesticides usually focus on risk at the level of individuals, and are carried out by comparing exposure and toxicological endpoints. However, in most cases the protection goal is populations rather than individuals. On the population level, effects of pesticides depend not only on exposure and toxicity, but also on factors such as life history characteristics, population structure, timing of application, presence of refuges in time and space, and landscape structure. Ecological models can integrate such factors and have the potential to become important tools for the prediction of population-level effects of exposure to pesticides, thus allowing extrapolations, for example, from laboratory to field. Indeed, a broad range of ecological models have been applied to chemical risk assessment in the scientific literature, but so far such models have only rarely been used to support regulatory risk assessments of pesticides. To better understand the reasons for this situation, the current modeling practice in this field was assessed in the present study. The scientific literature was searched for relevant models and assessed according to nine characteristics: model type, model complexity, toxicity measure, exposure pattern, other factors, taxonomic group, risk assessment endpoint, parameterization, and model evaluation. The present study found that, although most models were of a high scientific standard, many of them would need modification before they are suitable for regulatory risk assessments. The main shortcomings of currently available models in the context of regulatory pesticide risk assessments were identified. When ecological models are applied to regulatory risk assessments, we recommend reviewing these models according to the nine characteristics evaluated here.