Identification of areas of endemism from species distribution models: threshold selection and Nearctic mammals

We evaluated the relevance of threshold selection in species distribution models on the delimitation of areas of endemism, using as case study the North American mammals. We modeled 40 species of endemic mammals of the Nearctic region with Maxent, and transformed these models to binary maps using four different thresholds: minimum training presence, tenth percentile training presence, equal training sensitivity and specificity, and 0.5 logistic probability. We analyzed the binary maps with the optimality method in order to identify areas of endemism and compare our results regarding previous analyses. The majority of the species tend to have very low values for the minimum training presence, whereas most of the species have a value of the tenth percentile training presence around 0.5, and the equal training sensitivity and specificity was around 0.3. Only with the tenth percentile threshold we recovered three out of the four patterns of endemism identified in North America, and detected more endemic species.The best identification of areas of endemism was obtained using the tenth percentile training presence threshold, which seems to recover better the distributional area of the mammals analyzed.