Fine-scale roadkill risk models: understanding the intersection of wildlife and roads

Abstract

Robust, spatially explicit approaches accounting for ecological drivers are needed to identify environmental correlates of roadkill and set conservation priorities. We predicted wildlife road mortality across a nationwide road network using species distribution models with environmental covariates. We applied MaxEnt to a citizen science database of > 60,000 roadkill records to predict roadkill probability. Twenty-eight environmental covariates at 50 m spatial resolution were included, such as road type and land cover composition. We focused on ecological guilds and endangered species: common venomous snakes (CVS), semiaquatic and aquatic snakes (SAS), turtles, and the Maki’s keelback snake (Hebius miyajimae, HM). All predictive models performed well with AUCs > 0.7. Projected roadkill risks for CVS, SAS, turtles, and HM were highest in montane regions, coastal lowlands, the southwestern coast, and parts of central Taiwan, respectively. Roadkill projection models performed well across ecological levels and scales. Road-type strongly influenced roadkill risk. As predictions and variable importance differed across guild and species models, individual models need to be produced for each group of interest. Additionally, the project emphasizes the importance of systematic collection of roadkill data, which contributes to both informing conservation action and engaging the public in wildlife education. We discovered novel findings on predicted high- and low-risk areas for groups with conservation need and produced interactive roadkill risk maps as a conservation tool for managers and practitioners. Importantly, this methodology is not limited to Taiwan; it can be applied anywhere with sufficient roadkill and environmental data and is scalable to address the ecological question of interest.