Where the toad crosses the road: Multi-method Texas herpetofauna roadkill modeling for conservation planning

Abstract

Road mortality is a significant contributor to the global decline in herpetofauna, making regional roadkill models important for conservation planning. We used the iNaturalist Roadkills of Texas community science volunteered geographic information database to develop the first regional herpetofauna roadkill models in North America for: (1) all Texas herpetofauna combined; (2) Texas Horned Lizards (Phrynosoma cornutum), a Texas state threatened species; and (3) Gulf Coast Toads (Incilius nebulifer), a common generalist species. We novelly compare the utility of extrapolative feature-selected MaxEnt models to interpolative kernel density estimate models (KDEMs) of roadkill to inform conservation planning for sensitive Texas herpetofauna. A random subset feature selection algorithm was used to identify subsets of five to seven of 55 variables optimizing Maxent roadkill model performance and for ranking variables. MaxEnt roadkill models had acceptable to excellent accuracy, outperforming KDEMs for Texas herpetofauna combined and the Texas Horned Lizard. Both model types had poor accuracy in projecting Gulf Coast Toads roadkill, however, KDEMs had much lower complexity. Roadkill models of both types for various taxa revealed roadkill risk areas for partially sympatric sensitive species having insufficient data for individual models, including the Texas Tortoise (Gopherus berlandieri) and Houston Toad (Anaxyrus houstonensis). Top ranked variables in feature-selected MaxEnt roadkill models included distance to residential roads, human population density, percent cover developed open space, and spatial density of non-roadkill herpetofauna occurrences. Identified high roadkill risk areas should be further surveyed to refine roadkill projections and inform mitigation planning to support sensitive Texas herpetofauna conservation efforts.