Effects of variable selection on modelling habitat and potential distribution of the Andean bear in Bolivia

Abstract

Species distribution models are used in ecology and conservation biology to draw inferences about the drivers of species' ranges. However, poor conceptual background, environmental variable selection, and algorithm selection can contribute to misleading model predictions. We assessed the effects of environment variable selection and compared statistical performance and output maps of correlative resource- and biotope-based models for estimating the habitat and potential distribution of the Andean bear (Tremarctos ornatus) in Bolivia's Tropical Andes. The resource-based approach estimated bear habitat using 7 resources associated with 3 ecological functions: feeding, shelter, and access to water. In contrast, the biotope model described the habitat by applying 11 environmental predictors related to topography, vegetation, and human activities. Both models performed equally well overall and better than random, with shelter as the most influential variable for the resource model and Yunga forest for the biotope model. However, discrepancies in the extent and arrangement of predicted bear distribution between models differed and emphasized the effect of variable selection, which could influence the delineation of conservation areas for this species. We suggest using a resource-based approach when modelling species distribution because of the more direct relationship to the species investigated and greater ease of interpreting results.