Who and where: predicting road mortality risks using trait models

Wildlife-vehicle collisions are a major cause of mortality for many species. Empirical estimates of road mortality show that some species are more likely to be killed than others but to what extend this variation can be explained and predicted using intrinsic characteristics remains poorly understood. This study aims to identify general patterns associated to road mortality and generate spatial and species-level predictions of risks. We fitted trait-based random forest regression models (controlling for survey characteristics) to explain 783 empirical road mortality rates from Brazil, representing 170 bird and 73 mammalian species. Fitted models were then used to make spatial and species-level prediction of road mortality risk in Brazil considering all 1831 birds and 623 mammals which occur within the country’s continental boundaries. Survey frequency and geographic location were key predictors of observed rates, but mortality was also explained by species’ traits including body size, reproductive speed, and ecological specialization. Spatial predictions revealed high potential road mortality risk in Amazonia for both birds and mammals, and additionally high risk in Southern Brazil for mammals. Predicted rates for all Brazilian endotherm uncovered potential vulnerability to road mortality of several understudied species which are currently listed as threatened by the IUCN. With a fast expanding global road network, there is an urgent need to develop improved approaches to assess and predict road-related impacts. This study illustrates the potential of trait-based models as assessment tools to better understand correlates of vulnerability to road mortality across species, and as predictive tools for difficult to sample or understudied species and areas.peerReviewe