Macrohabitat models of occurrence for the threatened CheatMountain salamander, Plethodon nettingi

Abstract

The federally threatened Cheat Mountain salamander (Plethodon nettingi; hereafter CMS) is known to occur at approximately 70 small, scattered sites in the Allegheny Mountains of eastern West Virginia. We used a comparative modeling approach to explain the landscape-level distribution and habitat relationships of CMS in relation to a suite of biotic and abiotic habitat variables measured across the species' range. We collected data on 13 explanatory macrohabitat variables at CMS-occupied (n = 180) and random (n = 180) sites. We then examined CMS-macrohabitat relationships using a priori, logistic regression models with information-theoretic model selection, classification tree modeling, and discriminant function analysis. Among logistic regression models, a model containing the variables elevation, aspect, slope, and lithology received the strongest empirical support, although a model containing these variables and current vegetation type also received limited support. Variable selection within our classification tree and discriminant function modeling was consistent with logistic regression results. Common variables in all three approaches indicated that the probability of finding CMS across the species' range increased in areas at higher elevations and underlain by sandstone. Validation of models with empirical support using reserved data indicated that classification accuracy was ≥80% for all three analytical methods. Finally, we linked model outputs from all three methods to GIS coverage maps that predicted CMS occupancy within the study area. Our results indicate that geophysical and ecological characteristics measured at large spatial scales may be useful for quantifying salamander habitat relationships in forested landscapes, and more specifically increase the capacity of managers to locate and plan for the continued persistence and recovery of CMS.