Improving performance and transferability of small mammal species distribution models

Abstract

ABSTRACT In theory, interpretation and transferability of species distribution models (SDMs) should be improved by including abiotic and biotic factors that directly influence a species’ fundamental niche. We investigated whether adding topographic, soil and vegetation variables to a climate-only model improved model performance and predictive capacity for four coastal small mammal species. Adding landscape variables improved the structural goodness of fit for all four species (e.g. 2.6–47.6% increase in deviance explained), and the information-theoretic rankings (based on AICc, BIC and DIC) for the wet-heath specialist (Muridae, Rattus lutreolus lutreolus) and peramelid (Peramelidae, Isoodon obesulus obesulus). For the latter species, improved model performance successfully coincided with improved predictive capacity in the out-of-region validation (increase in the area under the curve, AUC). However, this result was poorly supported by trends in the successful classification of absences (specificity) indicating a modelling bias caused by low prevalence of species occurrence. Across all SDMs, additional abiotic and biotic landscape variables contributed between 3.7 and 14.9% of accumulative deviance explained. Our results illustrate increased model fit and transferability for select species, highlighting the potential for landscape variables that represent resources to better represent the fundamental niche in SDMs.