Wildlife-vehicle collisions are a major form of human-wildlife conflicts. Predictive animal-vehicle collision models have been developed to identify collision hotspots and guide mitigation strategies. However, most current models are static and unable to produce dynamic forecasts that incorporate changing climate and weather. The goal of our study was to develop a predictive and dynamic model of animal-vehicle collisions in Maine, USA. More than 6,700 moose-vehicle collisions (MVC) occurred from 2003 to 2017 in Maine, raising road safety, socio-economic, and wildlife conservation concerns. We sought to identify factors that contribute to a higher probability of MVCs by comparing two methodological approaches. We obtained 14 years of moose-vehicle collision data from Maine Department of Transportation. We developed a multiple logistic regression MVC model using static spatial data. We then imported temporal data in a Maximum Entropy (MaxEnt) model and created dynamic hourly MVC forecasts.
Results/Conclusions
Our models show that MVCs in Maine are more likely to happen on roads with intermediate to high speed limits and volumes, in or near forest cover, and close to wetlands. Sunlight, snow depth, humidity, and soil moisture are also significantly associated with higher MVC risk. The result of this study suggests that predictive and dynamic MVC models can be developed to inform drivers of crash likelihood hotspots in near-real time. Effectively applying these models allows for a more proactive, timely, and diagnostic response to MVCs and provides a novel approach to more comprehensively understand and predict human-wildlife conflicts.