A less data demanding ecophysiological niche modeling approach for mammals with comparison to conventional correlative niche modeling

Ecophysiological models are more data demanding and, consequently, less used than correlative ecological niche models to predict species' distribution under climate change, especially for endotherms. Hybrid models that integrate both approaches are even less used, and several aspects about their predictions (e.g. accuracy, geographic extent and uncertainty) have been poorly explored. We developed a hybrid model for mammals using hours of activity and hours of heat stress as mechanistic variables, fitted using macroclimatic data and applied to conventional correlative modeling. We then compared the outputs from conventional correlative models with our hybrid model for 58 tropical mammals in term of accuracy, uncertainty, and predicted geographic distribution under climate change. We expected that hybrid models to have higher accuracy than correlative ones, with difference in predicted geographic distribution extent. We found no substantial differences between correlative and hybrid predictions for accuracy, uncertainty, and extent. Although the area predicted as suitable did not differ in extent, they differ in location, with lower congruence between models for future prediction. This result challenged the widespread assumption that hybrid models are more accurate. The ecophysiological model approach proposed here ease ecophysiological data requirements. We propose, therefore, choosing model approach based on study's objective, rather than on data requirements or the assumption that hybrid models have better predictions. The main advantage of the hybrid model is in providing a more complete view of the species response, as proximal (causal) and distal (environment) aspects are combined.

Automated summary

The study found that hybrid models did not show higher accuracy in predicting species distribution compared to correlative models. While the extent of suitable areas predicted did not differ significantly, there were differences in the location of potential suitable areas between the two types of models. Therefore, the choice of model approach should be based on the research objective rather than assumptions about data requirements or the superiority of hybrid models.