Testing predictive models of landscape-scale pesticide exposure for California bumblebees

Predicting pesticide exposure to bees is a critical component of agricultural risk assessment and mitigation. However, we lack spatially explicit models of pesticide exposure that are (1) temporally dynamic, (2) capable of integrating across pesticides that vary in toxicity and decay rates, and (3) validated against field-realistic pesticide loads. We integrate fine-scale pesticide reporting datasets with models of pollinator foraging to develop maps of potential pesticide exposure risk in agricultural landscapes across three counties in California’s Central Valley. These exposure maps take into account not only individual pesticide load, but also chemical decay rates by projecting daily estimates of pesticide residues based on compound-specific half-lives. We then evaluate model predictions with pesticide residues in pollen collected and assayed from 28 Bombus vosnesenskii colonies at 14 agricultural sites. From 36 chemicals assayed, we found significant amounts of eight different pesticides in the pollen of bumblebee colonies. These pesticides include commonly applied herbicides, fungicides (e.g. ergosterol inhibitors), and insecticides (e.g. pyrethroids and organophosphates). Both predicted and observed pesticide loads varied over the study period, with compound-specific application peaks. Incorporating decay rates into the model generated notably different landscape-level pesticide exposure outcomes and these were used in subsequent model validation. Correspondence between predicted pesticide exposure and observed colony residue levels was pesticide-dependent and did not vary between chemical classes. Our results are a step forward in developing and validating predictive models of pesticide risk, with the ultimate goal of guiding scenario planning for reduced risk pesticide practices in a major agricultural region.