Reciprocal extrapolation of species distribution models between two islands - Specialists perform better than generalists and geological data reduces prediction accuracy

This study aims to test the extrapolation effects of species distribution models (SDM) using three groups of predictor variables: climate, relief and geology (bedrock type). We highlight potential ecological differences for selected taxa, regarding both generalists and specialists in terms of edaphic conditions. We used distributional data of 12 woody species shared by two large Mediterranean islands (Crete and Sicily) to calibrate Maxent models of their potential distribution. We trained models with data from Crete and extrapolated to Sicily and vice versa. We tested ten proxies for the three variable groups and compared AUC values as a measure of model performance. Extrapolation of SDMs worked fairly well across species and islands for those models including climatic and relief data, while those including geological information performed worse for the tested species. Edaphic generalists performed less well compared to edaphic specialists. The latter performed best with climate data included in models and more poorly with only geological data. This may be due to differences in bedrock type affinities of species between the two islands.