Daniele Falcinelli, University of Rome "La Sapienza"

Human-caused mortality is a primary threat to large mammal conservation worldwide, and hunting represents one of its most pervasive sources. Beyond direct demographic impacts, hunting influences animal behaviour and space use. Despite its relevance, modelling of hunting-mortality risk remains underexplored, particularly regarding the influence of different modelling strategies on spatial prediction accuracy. Here, we compared four alternative approaches for modelling hunting-mortality risk in a European brown bear Ursus arctos population in central Finland, using long-term (2002–2014) harvest and telemetry data. We developed resource selection functions contrasting hunting-mortality locations with (i) random background locations, (ii) GPS-telemetry locations of live bears, (iii) random locations weighted by a species distribution model specific for the hunting season, and (iv) random locations weighted by a bear-density probability surface. Predictive performance was evaluated with independent harvest data (2015–2017) using the Boyce index. Across all approaches, hunting-mortality risk increased with bear density, forest and shrubland cover, while decreasing with hunter density. Although the spatial distribution of predicted risk was broadly consistent among approaches, the extent and intensity of high-risk areas varied markedly. The random-background approach identified the most extensive high-risk zones and achieved the highest predictive performance; other approaches predicted more localised and clustered risk hotspots, performing comparatively worse. These results indicate that the random-background approach may better capture potential risk across landscapes, suggesting that harvest reflects landscape accessibility and human-use patterns rather than strictly animal space use. Integrating this approach with SDMs within habitat-based frameworks can help identify potential refugia and ecological traps, promoting the conservation of threatened large mammals.