Improving communication and validation of ecological models : a case study on the dispersal of aquatic macroinvertebrates

In recent years, ecological effect models have been put forward as tools for supporting environmental decision-making. Often they are the only way to take the relevant spatial and temporal scales and the multitude of processes characteristic to ecological systems into account. Particularly for environmental risk assessments of pesticides the potential benefits of including modelling studies were recognized and a dialogue between different stakeholder groups was opened. Representatives from academia, pesticide-producing industries, and regulators are nowadays discussing their needs, possibilities, and ways of implementation for improving the use and usefulness of such models. However, it quickly became evident that not all involved parties possess the same background knowledge in regards to modelling terminology and model quality understanding. Proper communication of a given model's structure, robustness, and soundness is crucial to render a model of real use to the decision-making. Doubts about a model's quality and mode of operation may lead to an immediate rejection of the conclusions drawn from its estimations. In this thesis, we addressed this point of concern, and performed a literature review focusing on aspects surrounding quality assessments, validation, and communication of models. "Validation" was identified as a catch-all term, which is thus useless for any practical purpose. Based on the review, we developed a framework that splits the seemingly blurry process into associated components and introduce the term ‘evaludation’, a fusion of ‘evaluation’ and ‘validation’, to describe the entire process of assessing a model's quality and reliability. Considering the iterative nature of model development, the modelling cycle, we identified six essential elements of evaludation: (i) ‘data evaluation’ for scrutinising the quality of numerical and qualitative data used for model development and testing; (ii) ‘conceptual model evaluation’ for examining the simplifying assumptions underlying a model's design; (iii) ‘implementation verification’ for testing the model's implementation in equations and as a computer programme; (iv) ‘model output verification’ for comparing model output to data and patterns that guided model design and were possibly used for calibration; (v) ‘model analysis’ for exploring the model's sensitivity to changes in parameters and process formulations to make sure that the mechanistic basis of main behaviours of the model has been well understood; and (vi) ‘model output corroboration’ for comparing model output to new data and patterns that were not used for model development and parameterisation. In a subsequent step, we used the evaludation framework to re-evaluate and adjust the documentation framework TRACE (TRAnsparent and Comprehensive Eco- logical modelling; Schmolke et al. 2010), a general framework for documenting a model's rationale, design, and testing. TRACE documents should provide convincing evidence that a model was thoughtfully designed, correctly implemented, thoroughly tested, well understood, and appropriately used for its intended purpose. TRACE documents link the science underlying a model to its application, thereby also linking modellers and model users, for example stakeholders, decision makers, and developers of policies. TRACE thus becomes a tool for planning, documenting, and assessing model evaludation, which includes understanding the rationale behind a model and its envisaged use. To provide an example of the measures that can be taken to increase general trust in a model's design and output, we chose MASTEP (Metapopulation model for Assessing Spatial and Temporal Effects of Pesticides) for a case study. MASTEP is an individual-based model used to describe the effects on and recovery of the water louse Asellus aquaticus after exposure to an insecticide in pond, ditch, and stream scenarios. The model includes processes of mortality of A. aquaticus, life history, random walk between cells and density dependence of population regulation. One of the submodels receiving particular criticism was the random walk procedure and the uncertainty attached to the parameters used. The parameters were estimated based on experimental studies performed under very limiting conditions. We designed and performed experiments to derive more precise parameters and to better understand the movement behaviour of this freshwater isopod. The experimental procedure that we developed employed video tracking of marked individuals that were introduced alone or as part of a group of unmarked individuals into arenas of approximately 1m2 in size. We recorded the paths of the marked individuals under a set of different conditions, i.e. presence or absence of food or shelter, population density, and after sublethal exposure to chlorpyrifos and imidacloprid. Based on the experimental findings, we refined the movement modelling procedure used in MASTEP to derive more realistic dispersal estimates, with which we revisited a modelling study performed previously by Galic et al. (2012). In this study, the effects of pesticide application timing on population dynamics and recovery times were tested and compared to outcomes from previous versions. It was furthermore possible to integrate an increased level of environmental complexity that could not be addressed before due to a lack of data. Compared to former versions of the population model, recovery times did not change significantly when the same movement parameters were applied to all simulated individuals. This indicates that the previous assumptions already yielded robust estimations. Accounting for life stage dependent movement restraints, though, delayed recovery when exposure occurred shortly before a reproduction cycle. Based on these findings, it was concluded that an increase of ever more realism and environmental complexity in modelling studies needs to be done carefully on a case-by-case basis. Increased realism in models can introduce an unwarranted increase in model complexity and uncertainty, which is not always supporting an improved credibility level of a model. Despite the need for basic ecological research for more comprehensive ecological models, we further argue that a modelling study in general can benefit greatly from an improved plan that considers communication needs from the start. Considering such needs early on can help develop a time- and cost-saving strategy for model testing and data collection, while providing a thorough understanding of a model's underlying mechanisms across several layers of stakeholder groups.