Remote sensing variables improve species distribution models for alpine plant species

Abstract Species distribution models (SDMs) are cost-effective, transparent and flexible planning tools to support various areas in nature conservation. Variables taken from remote sensing (RS) are broadly applicable to biodiversity studies. In our study, we combined RS-variables (normalized differenced vegetation index and land surface temperatures), with topographic and geological variables to produce detailed SDMs in the context of a seed collection campaign of the Alpine Seed Conservation and Research Project. To identify effective predictor variable combinations we compiled three different variable sets and compared the predictive model performance. The full model, that combines all types of variables, slightly outperforms (average values for TSS: 0.91, AUC: 0.98, Kappa: 0.7) models that use topo-climatic variables (average values for TSS: 0.91, AUC: 0.98, Kappa: 0.68) or NDVI (average values for TSS: 0.85, AUC: 0.96, Kappa: 0.54) alone. We also produced ensemble models that performed slightly better compared to the different model algorithms used in our approach. We identified the temperature of the coldest month, mean NDVI and bedrock as important variables that determine the distribution of alpine plant species. Our full models show high accordance with actual species distribution ranges and are highly relevant for efforts to identify special areas for either in-situ or ex-situ conservation.