Assessment of roadkill likelihood methods: the use of single occurences versus hotspots for different taxa

Abstract

Roads are responsible for a massive mortality of wildlife annually. In order to define measures to reduce roadkill risk many studies have analyzed the spatial environmental variables that explain roadkill likelihood. There are several approaches to conduct this type of analysis such as modelling the roadkill presence-only (high quantity and low quality of the data) or the incidence of roadkill (hereafter hotspots) (low quantity and high quality of the data). However, there is no consensus on which one is the best to generate the most accurate and robust results. We aimed to compare which type of records (only roadkill, spatial hotspots, or spatio-temporal hotspots) generate better model performance and the variables that better explain the roadkill likelihood. We analyzed roadkill records of two species in each class of terrestrial vertebrates collected in Brazil: amphibians (Leptodactylus latrans and Rhinella icterica), reptiles (Philodryas patagoniensis and Helicops infrataeniatus), birds (Volatinia jacarina and Nothura maculosa), and mammals (Didelphis albiventris and Myocastor coypus). We used the Siriema software to identity roadkill hotspots, and the MaxEnt software to analyze the relationship between the three types of records with the environmental variables. All models had a high performance (AUC > 0.7). Our findings suggest that MaxEnt models performance were better with hotspots. In general, the environmental variables that explained roadkill were consistent with species habitats and no differences were found between habitat specialist and generalist species Therefore, we recommend the use of hotspots for modeling the roadkill risk for either habitat specialist and generalist species.