Known Unknowns: Does Culvert Passability Relate to Landscape Characteristics?

Despite the potential for road-stream crossings to fragment habitat and obstruct fish movement, they are often overlooked during studies on hydrologic connectivity.  Region-wide field surveys are desirable, but prohibitive given the quantity of crossings to survey. We used a geographic information system (GIS) and two modeling approaches to determine if local- and watershed-scale variables could predict passability at sample locations. We conducted surveys during 2013 and 2014 on approximately 450 road crossings in the Chipola, Etowah, and Nolichucky River systems in the southeastern United States. Using a passability score calculated from field measurements, we used logistic regression and random forest classification tree analysis to model passability as a function of landscape variables at two spatial scales. Both modeling approaches indicated catchment size was important to predict passability. Our analysis also found that watershed-scale variables, including land cover type, and some local-scale variables were predictors of culvert passability. The random forests algorithm was a stronger predictor, likely because this non-parametric approach can account for the variability in conditions leading to impassibility in a heterogeneous landscape. Our models can be used to identify regions with an increased likelihood of habitat fragmentation, allowing managers to target those areas for focused surveys and remediation.