Predicting species assemblages at wildlife crossing structures using multivariate regression of principal coordinates

Wildlife populations are in decline due to human threats, including highways. Strategies for reducing road impacts on wildlife include wildlife fencing which keep animals off roads and wildlife crossing structures (WCSs) which provide safe passage across roads. Wildlife crossing structures are diverse and transportation managers are often interested in identifying which WCS designs are effective for target species so a model that predicts target species usage of WCSs is likely to be beneficial to managers and biologists. Wildlife crossing structures are typically built for select species but are utilized by other species, so it may be beneficial to examine WCS use at the community level. We used camera trap data to develop a predictive model of mammal community composition at WCSs built for ocelots (Leopardus pardalis) to predict total detections, successful crossings, and failed crossings using spatial, temporal, structural, environmental, and anthropogenic characteristics. During the first-year after construction of WCSs, structural and anthropogenic characteristics of the WCSs were more important than the environmental characteristics although we expect environmental characteristics to become more important with time. Our models reasonably predicted total detections but were less effective at predicting successful and failed crossings, likely due to potential finer-scale, more dynamic effects like noise or microclimate conditions that may drive an animal's decision to use a WCS. While our study focused on WCSs built for ocelots, to our knowledge, our model is the first model of WCS effectiveness for mammal communities and provide a generalized framework for predicting WCS use which can be applied anywhere where WCSs are being built.