Analysing and predicting wildlife–vehicle collision hotspots for the Swiss road network

Abstract Context Wildlife–vehicle collisions (WVCs) are a significant threat for many species, cause financial loss and pose serious risk to motorist safety. Objectives We used spatial data science on regional collision from Switzerland with the objectives of identifying key environmental factors modelling WVC nationwide scale. Methods 43,000 records roe deer, red wild boar, chamois 2010 2015 both midlands mountainous landscape types. compared fixed-length road segmentation approach segments based Kernel Density Estimation, data-driven method. The segments’ properties were derived land-cover geodata using novel neighbourhood operations. Multivariate logistic regression random forest classifiers identify rank relevant predict in areas without data. Results hotspots sinuosity, two composite browsing/forage availability traffic noise—a proxy flow. Our best models achieved sensitivities 82.5% 88.6%, misclassifications 20.14% 27.03%, respectively. predictions better forested revealed limitations open due lack up-to-date annual crop changes. Conclusions illustrate added value fine-grained modelling, show how such detailed information can be annotated functions. Finally, we recommend inclusion improving modelling.