Habitat Suitability Based Models for Ungulate Roadkill Prognosis

SIMPLE SUMMARY: Red deer, roe deer and wild boar movements and crossings of the two highways in Lithuania were modeled. Validity of obtained models was tested by comparing the pathway predictions to the real roadkill and roadkill cluster locations in 2002–2009 (at the time the models were created) and in 2010–2017 (testing the prognostic value of these models). Across both periods and on both highways, the roe deer roadkill locations were significantly closer to the model-predicted pathways than to randomly selected points. The prediction of roadkill locations was also good for wild boar. The roe deer roadkill clusters and multi-species clusters were significantly better represented by the model than by random distribution. Thus, habitat suitability-based models of ungulate movement are recommended as an additional tool for planning wildlife-vehicle collision mitigation measures. ABSTRACT: Roads do not only have a detrimental effect on nature (fragmenting habitats, isolating populations and threatening biodiversity), but the increasing numbers of wildlife-vehicle collisions are also a direct threat to humans and property. Therefore, mitigation measures should be placed with respect to animal distribution and movements across the roads. We simulated red deer, roe deer and wild boar movements in Lithuania, focusing on the two main highways A1 and A2. Using regional habitat suitability and linkage models, we calculated movement pathways and the most probable crossing zones in 2009. The prognostic value of these models was tested by comparing the pathway predictions to the real roadkill and roadkill cluster locations in 2002–2009 and 2010–2017. Across both periods and on both highways, the roe deer roadkill locations were significantly closer to the model-predicted pathways than to randomly selected points. The prediction of roadkill locations was also good for wild boar. The roe deer roadkill clusters and multi-species clusters were significantly better represented by the model than by random distribution. On both highways, the biggest differences in distance from the predicted locations were near big cities. We recommended wildlife movement models as an additional tool for planning wildlife-vehicle collision mitigation measures and we advise measures for increasing their predicting power.