Choice of climate data affects the performance and interpretation of species distribution models

Climate is a core aspect of a species’ niche that is frequently incorporated into species distribution models (SDMs). The proliferation of readily accessible climate data is stimulating the increasing use of climate-based species distribution models (CSDMs) in conservation planning and management. However, it is uncertain how sensitive CSDMs are to the choice of climate data used. We compared the performance of maximum entropy-based CSDMs calibrated with seven different climate data sets (WorldClim, Chelsa, TerraClimate, Climate Research Unit Time-Series [CRU-TS], PRISM, StreamCat, and EarthEnv). For each species (four amphibians and two reptiles in California), we used a standardized, objective procedure to select climate predictors from each data set. We then assessed performance with standard metrics, calculated variable importance scores, and compared predicted distributions to historical range maps of the target species. The climate data set used affected CSDM performance, with models based on some data sets overpredicting and others underpredicting distributions compared with historical range maps of each of the species. Critically, for the six species we tested, no one climate data set always performed worst or best. These results indicate that the effects of different climate data sets should be incorporated into model evaluation when developing, applying, and interpreting climate-based SDMs.