Using reports of bee mortality in the field to calibrate laboratory-derived pesticide risk indices

Mounting evidence suggests that pollinators worldwide are experiencing dramatic population declines, and exposure to pesticides is one of the factors that can account for this. By making use of a database containing more than two decades of honey bee (Apis mellifera) hive poisoning incidents from the United Kingdom (Wildlife Incident Investigation Scheme WIIS]) and corresponding pesticide use surveys, we attempted to explain honey bee poisoning incidents in the field using models derived from pesticide use information, laboratory-generated bee toxicity data (defined as a hazard ratio; application rate divided by LD(50)), and physico-chemical properties of the applied pesticides. Logistic regression analyses were used to assess the relationship between honey bee poisoning incidents in the field and these parameters. In analyzing models with multiple dimensions, we selected the best model by the best subset method, an iterative method based on maximum likelihood estimation, and Akaike's information criterion. Results suggested that the size of the area treated and hazard ratios calculated from application rates and oral or contact toxicity (but the latter especially) can be used to predict the likelihood that honey bee mortality will occur. Model predictions also suggest that some insecticides carry an extreme risk for bees, despite the lack of documented incidents.