Remotely sensed temperature and precipitation data improve species distribution modelling in the tropics

AimSpecies distribution modelling typically relies completely or partially on climatic variables as predictors, overlooking the fact that these are themselves predictions with associated uncertainties. This is particularly critical when such predictors are interpolated between sparse station data, such as in the tropics. The goal of this study is to provide a new set of satellite-based climatic predictor data and to evaluate its potential to improve modelled species-climate associations and transferability to novel geographical regions. LocationRain forests areas of Central Africa, the Western Ghats of India and South America. MethodsWe compared models calibrated on the widely used WorldClim station-interpolated climatic data with models where either temperature or precipitation data from WorldClim were replaced by data from CRU, MODIS, TRMM and CHIRPS. Each predictor set was used to model 451 plant species distributions. To test for chance associations, we devised a null model with which to compare the accuracy metric obtained for every species. ResultsFewer than half of the studied rain forest species distributions matched the climatic pattern better than did random distributions. The inclusion of MODIS temperature and CHIRPS precipitation estimates derived from remote sensing each allowed for a better than random fit for respectively 40% and 22% more species than models calibrated on WorldClim. Furthermore, their inclusion was positively related to a better transferability of models to novel regions. Main conclusionsWe provide a newly assembled dataset of ecologically meaningful variables derived from MODIS and CHIRPS for download, and provide a basis for choosing among the plethora of available climate datasets. We emphasize the need to consider the method used in the production of climate data when working on a region with sparse meteorological station data. In this context, remote sensing data should be the preferred choice, particularly when model transferability to novel climates or inferences on causality are invoked.