Spatially biased versus extent of occurrence records in distribution modelling predictions: a study case with South American anurans

Ecological Niche Modelling (ENM) is used to estimate potential species distributions through the association of general climate data with precise geographic occurrence records. Occurrence data are mainly obtained from museums or other natural history collections. However, these data are usually incomplete and spatially biased compared to actual geographic species’ distribution. Here, we compared predictions of occurrence for 13 widely distributed South American anuran species generated from two series of distribution data: a) original (and biased) point records and b) random distribution points within the extent of occurrence of the species. We compared the distribution predictions for baseline and 2050 climate change scenarios. By using six modelling algorithms, we found that the accuracy measure AUC (Area Under the Curve) of three algorithms (ED, OM-GARP and SVM) presented higher AUC values when the ENMs were generated from the original point records, whereas the other algorithms presented similar AUC values between the ENMs generated from different sets of occurrence data. The size of the predicted areas is larger when the ENMs are generated by random occurrence records (except for the algorithms BIOCLIM and ED), both in the baseline and future climate scenario projections.