Authors: Cerri, Jacopo; Bužan, Elena;Pokorny, Boštjan;

Wildlife-vehicle collisions jeopardize wildlife conservation and human safety. Including environmental covariates affecting the spatial behavior of wildlife can improve collision forecasts, and a few studies have already focused on how moon illumination affects wildlife-vehicle collisions. However, no study to date has quantified the interplay between moon illumination and cloud coverage, an atmospheric phenomenon decreasing the amount of moonlight reaching terrestrial ecosystems. We modeled the synergistic effect of moonlight and cloud coverage over the number of collisions between vehicles and European roe deer (Capreolus capreolus), in Slovenia. Data included complete nation-wide set of daily collisions (n = 49,259) collected between 2010 and 2019 by citizen scientists. Moonlight was measured both as quarters and as the fraction of illuminated moon. Data were modeled through bivariate Gaussian Processes with a Negative Binomial distribution. Treating moonlight as a continuous covariate makes its interaction with cloud coverage more interpretable than using moon quarters, and it also reduced prediction errors. However, the analysis of the NDVI suggests that at some times of the year it could be hard to disentangle the effect of cloud coverage from that of changes in animal behavior driven by vegetation development, as these two components co-vary in time.