Public geospatial datasets as an approach to maximizing effi ciency in the collection of site covariates in wildlifeâ•fivehicle collision studies

Wildlife–vehicle collisions (WVCs) are a major research focus because of increasing human health and safety concerns and the potential for biological impacts on wildlife. A key component of both understanding the causes of WVCs and designing mitigation measures is the collection and analysis of environmental and roadway data at WVC sites. However, collecting these site data can be logistically challenging and potentially dangerous to researchers. We studied the feasibility and accuracy of using public geospatial datasets, particularly Google Earth and Street View, as an alternative approach to assessing WVC onsite covariates. We randomly selected 50 sites from a larger WVC study and measured the topography, habitat type, width of the road median, and presence of fencing at each site as representatives of typical WVC site covariates. We compared the measurements recorded in the fi eld to estimates obtained from public geospatial datasets in the lab. We determined that median topography had the lowest overall accuracy (60%), followed by presence of fencing with accuracy at 75% of sites. By contrast, median habitat type was identifi ed correctly in almost all comparisons (96% overall accuracy). The root mean squared error for median width was 1.15 m overall. Our results suggest that Google platforms may serve as viable alternatives to fi eld data collection for site covariates related to coarse measures of habitat type and some characteristics of road topography, thus reducing time requirements and potential safety risks to researchers in the fi eld. However, there are several crucial caveats to consider when using geospatial platforms, particularly as they relate to 3-dimensional depictions of roadway features. Thus, we urge caution when attempting to use digital platforms to collect data on these covariates.