Modeling exposure of mammalian predators to anticoagulant rodenticides

Abstract

Anticoagulant rodenticides (AR) are a widespread and effective method of rodent control but there is concern about the impact these may have on non-target organisms, in particular secondary poisoning of rodent predators. Incidence and concentration of AR in free-living predators in Denmark is very high. We postulate that this is caused by widespread exposure due to widespread use of AR in Denmark in and around buildings. To investigate this theory a spatio-temporal model of AR use and mammalian predator distribution was created. This model was supported by data from an experimental study of mice as vectors of AR, and was used to evaluate likely impacts of restrictions imposed on AR use in Denmark banning the use of rodenticides for plant protection in woodlands and tree-crops. The model uses input based on frequencies and timings of baiting for rodent control for urban, rural and woodland locations and creates an exposure map based on spatio-temporal modeling of movement of mice-vectored AR (based on Apodemus flavicollis). Simulated predator territories were super-imposed over this exposure map to create an exposure index. Predictions from the model concur with field studies of AR prevalence both before and after the change in AR use. In most cases, incidence of exposure to AR is predicted to be <90%, although cessation of use in woodlots and Christmas tree plantations should reduce mean exposure concentrations. Model results suggest that the driver of high AR incidence in non-target small mammal predators is likely to be the pattern of use and not the distance AR is vectored. Reducing baiting frequency by 75% had different effects depending on the landscape simulated, but having a maximum of 12% reduction in exposure incidence, and in one landscape a maximum reduction of <2%. We discuss sources of uncertainty in the model and directions for future development of predictive models for environmental impact assessment of rodenticides. The majority of model assumptions and uncertainties err on the side of reducing the exposure index, hence we believe the predictions to be robust and to indicate that the scale of the problem may be large.