Assessment of environmental variables for species distribution modelling: insights from the mosaic distribution of red- and yellow-bellied toads

Species distribution modelling can possibly be improved through the preferential use of explanatory variables that reflect the natural history features of the species being modelled. Red- and yellow-bellied toads (genus Bombina) engage in an intricate mosaic distribution across Europe. Analysing new atlas data on these species mutual distribution in Hungary with principal coordinate analysis we identified their differential ecological preferences as forested, hilly and mountainous for B. variegata and open lowland for B. bombina. These locally operating parameters we consider to be good proxies for the essential species difference which resides in breeding in ephemeral puddles (B. variegata) versus larger permanent ponds (B. bombina). With two-species distribution modelling - in which the presence of one species is contrasted with the presence of the counterpart species - we obtained excellent model fit (AUC) for climate and elevation / land cover datasets alike (AUC=0.98 versus 0.95). For both models fit values dropped upon transference to surrounding countries, yet the latter model kept significantly higher predictive power (AUC=0.91) than the climate model (AUC=0.79). Swapping elevation for hilliness as suggested in the literature had a significant negative effect on model performance. We conclude that an informed parameter selection enhances model transferability, therewith improving our understanding of species-habitat associations.