SYMP 23-4 - Taming 21st century problems by integrating modeling with field and lab research: Examples from pesticide risk assessment for pollinators and symposium synthesis

Understanding pesticide effects on pollinators is among the most urgent and difficult ecological management problems. Current risk assessment of pesticides in Europe and North America is based on toxicity measurements on individuals under controlled conditions in the laboratory. However, the protection goal is populations in the field. To evaluate effects on populations under realistic environmental conditions, ecological models have been increasingly developed. The honey bee model BEEHAVE was launched in 2014 to consider both processes within a hive (egg-laying, brood cycles, disease-carrying mites) and foraging in a landscape characterized by the availability of nectar and pollen. BEEHAVE has been used in more than 20 studies addressing a wide range of questions in stylized or realistic scenarios, and extended to take into account specific toxicity data and exposure pathways within the hive. EFSA, the European agency responsible for providing guidelines for pesticide risk assessment, has used BEEHAVE to quantify natural variation in colony size for healthy, unstressed colonies across Europe. The success of BEEHAVE is based on several factors: it is based on first principles, i.e. food collection and consumption, and can therefore be applied to any landscape; it incorporates the main environmental factors, i.e. weather, land use and hence food supply, and beekeeping practices; and it is rich enough in structure to provide outputs that can be compared with independent data, for example from Germany's automated hive-weighing network, bee flight data, or citizen science data on the distribution and phenology of flowering plant species. BEEHAVE is well documented, open source and implemented in a free software platform that allowed users to easily define their own scenarios or add extensions for specific purposes. BEEHAVE was also used to propose experiments and field studies, or to re-analyze existing data. Despite its success, a major limitation is the lack of resources to continuously update and develop the model, its implementation and the data workflows needed to run, test and recalibrate the model. Models for modern problems, although designed to be as simple as possible, need to be complex and rely on multiple data representing patterns and processes at different spatial and temporal scales. Such models are therefore no less big science, requiring manpower and long-term funding, than big science projects in other disciplines. The current hype around 'digital twins' recognizes this, but the twin concept needs to be carefully defined and implemented to avoid false expectations and mis-developments.