Predicting spatial and seasonal patterns of wildlife–vehicle collisions in high-risk areas

Context Vehicle collisions with wildlife can injure or kill animals, threaten human safety, and the viability of rare species. This has led to a focus in road-ecology research on identifying key predictors ‘road-kill’ risk, goal guiding management mitigate its impact. However, because complex context-dependent nature causes risk exposure, modelling road-kill data ways that yield consistent recommendations proven challenging. Aim Here we used multi-model machine-learning approach identify spatio-temporal predictors, such as traffic volume, road shape, surrounding vegetation distance settlements, associated risk. Methods We collected location, identity body size each mortality across four seasons along eight roads southern Tasmania, ‘road-kill hotspot’ concern. focused three large-bodied frequently affected crepuscular Australian marsupial herbivore species, rufous-bellied pademelon (Thylogale billardierii), Bennett’s wallaby (Macropus rufogriseus) bare-nosed wombat (Vombatus ursinus). fit point-location using ‘lasso-regularisation’ logistic generalised linear model (LL-GLM) out-of-bag optimisation decision-tree-based ‘random forests’ (RF) algorithm for optimised predictions. Results The RF model, high-level feature interactions, yielded superior out-of-sample prediction results additive classification accuracy 84.8% 871 observations true skill statistic 0.708, compared 61.2% 0.205 LL-GLM. lasso rejected visibility density, ranking roadside type presence barrier fencing most influential locality. Conclusions Forested areas no fence curved sections posed highest animals. Seasonally, frequency wildlife–vehicle increased notably females during oestrus, when they were more dispersive so had higher encounter rate roads. Implications These findings illustrate value combination attributive predictive machine learning rank interpret complexity possible well offering guide practical interventions road-related hazards.