Machine learning identifies specific habitats associated with genetic connectivity in Hyla squirella.

The goal of this study was to identify and differentiate the influence of multiple habitat types that span a spectrum of suitability for Hyla squirella, a widespread frog species that occurs in a broad range of habitat types. We collected microsatellite data from 675 samples representing 20 localities from the southeastern USA and used machine-learning methodologies to identify significant habitat features associated with genetic structure. In simulation, we confirm that our machine-learning algorithm can successfully identify landscape features responsible for generating between-population genetic differentiation, suggesting that it can be a useful hypothesis-generating tool for landscape genetics. In our study system, we found that H. squirella were spatially structured and models including specific habitat types (i.e. upland oak forest and urbanization) consistently explained more variation in genetic distance (median pR(2)  = 47.78) than spatial distance alone (median pR(2)  = 23.81). Moreover, we estimate the relative importance that spatial distance, upland oak and urbanized habitat have in explaining genetic structure of H. squirella. We discuss how these habitat types may mechanistically facilitate dispersal in H. squirella. This study provides empirical support for the hypothesis that habitat-use can be an informative correlate of genetic differentiation, even for species that occur in a wide range of habitats.