Integrated risk factors for vertebrate roadkill in southern Ontario

Road mortality of animals (roadkill) threatens public safety and wildlife populations. As mitigation tools, predictive models of roadkill are becoming more common in the published literature; however, few models generalize across multiple taxa, and thus are less useful for management scenarios that account for multiple target species. Using a dataset of 653 vertebrate roadkills collected from 2 parks in southern Ontario, we constructed generalized linear mixed models to determine the simultaneous risk factors for bird, frog, mammal, five-lined skink (Eumeces fasciatus), snake, toad, and turtle hatchling roadkills from among a set of 8 potential predictor variables. Posted road speed limit was the dominant roadkill predictor (positive coefficient), followed by maximum daily temperature (positive), habitat diversity (positive), and distance from wetlands (negative). All else being equal, as road speed limits increase from 20?km/hr to 50?km/hr, the model predicted the season's mean roadkill probability for a given location to increase from less than 0.1 to 0.75. Conversely, roadkill probability declined from 0.55 to 0.29 as distance from wetland edges increases from 0?km to 1?km. Model diagnostics calculated from randomly resampled cross-validation datasets indicated that the best model formulation had an averaged predictive accuracy of 67.5% and an area under the curve (AUC) of 0.867. The best model also reflected seasonal patterns of animal behavior, including late-summer frog movements and fall turtle hatching events. The best model also compared favorably to single-taxon equivalent models. To reduce the incidence of vertebrate roadkill, we recommend that managers lower road speed limits, especially on roads near diverse habitats and near wetlands, and on warmer days if temporary signage is used. (c) 2012 The Wildlife Society.