The following settings were used: 100 bootstrap permutations to capture model variability, 20% of random occurrences for model testing, and logistic output as the suitability index. The median of the 100 models was used as the best hypothesis of the potential distribution of U. thibetanus. From the suitability index,we generated a binary map using as a threshold
the highest value that included 95% of calibration points (i.e., E=5% omission error) (Peterson et al., 2008, 2011).