To cross or not to cross: modeling wildlife road crossings as a binary response variable with contextual predictors

Roads are significant barriers to landscape‐scale movements of individuals or populations of many wildlife taxa. The decision by an animal near a road to either cross or not cross may be influenced by characteristics of the road, environmental conditions, traits of the individual animal, and other aspects of the context within which the decision is made. We considered such factors in a mixed‐effects logistic regression model describing the nightly road crossing probabilities of invasive nocturnal Brown Treesnakes (Boiga irregularis) through short‐term radiotracking of 691 snakes within close proximity to 50 road segments across the island of Guam. All measures of road magnitude (traffic volume, gap width, surface type, etc.) were significantly negatively correlated with crossing probabilities. Snake body size was the only intrinsic factor associated with crossing rates, with larger snakes crossing roads more frequently. Humidity was the only environmental variable affecting crossing rate. The distance of the snake from the road at the start of nightly movement trials was the most significant predictor of crossings. The presence of snake traps with live mouse lures during a portion of the trials indicated that localized prey cues reduced the probability of a snake crossing the road away from the traps, suggesting that a snake's decision to cross roads is influenced by local foraging opportunities. Per capita road crossing rates of Brown Treesnakes were very low, and comparisons to historical records suggest that crossing rates have declined in the 60+ yr since introduction to Guam. We report a simplified model that will allow managers to predict road crossing rates based on snake, road, and contextual characteristics. Road crossing simulations based on actual snake size distributions demonstrate that populations with size distributions skewed toward larger snakes will result in a higher number of road crossings. Our method of modeling per capita road crossing probabilities as a binary response variable, influenced by contextual factors, may be useful for describing or predicting road crossings by individuals of other taxa provided that appropriate spatial and temporal resolution can be achieved and that potentially influential covariate data can be obtained.