Can we mitigate animal-vehicle accidents using predictive models?

Abstract

1. Vehicle collisions with wild animals are a serious problem that justifies the widespread application of mitigation measures such as road fencing and provision of crossing structures. Models that predict the best location for mitigation measures can improve wildlife survival and road safety. 2. A database of 2067 records of animal-vehicle collisions was used to create two data sets at different spatial scales. The first comprised records of road sections of 1 km length with high rates of collision in combination with road sections with a low number of collisions. The second comprised records of collision and no collision incidence at points on the road system at a 0.1-km scale. Logistic regression was used to investigate the relationship between incidence of collision and measured habitat features in each data set. The models were validated with a subset of the original data not used in developing the models. 3. Road sections with high collision rates were associated with areas having high forest cover, low crop cover, low numbers of buildings and high habitat diversity. The fitted model achieved a significant predictive success during validation (χ2 = 4.82, 1 d.f., P = 0.028), with more than 70% correct classification of cases. 4. Specific collision points typically had no guard-rails or lateral embankments, were not near underpasses, crossroads or buildings, and featured hedges or woodland near the road. The fitted model also showed a significant predictive power in validation (64% correct classification, χ2 = 9.51, 1 d.f., P = 0.002) and accurately predicted 85.1% of collision points. 5. Synthesis and applications. Predictive models of animal-vehicle collision locations should be used at both a landscape level and a local scale during the process of road design and implementation of mitigation measures. Modelling of collision risk could inform decisions on road alignment and on the exact location of crossing structures for mammals, to improve wildlife survival and road safety. This is the first study integrating both landscape and local scales of analysis for the variables associated with animal-vehicle collisions.