The use and misuse of SDMs to predict biodiversity responses to climate change

Background/Question/Methods

To prevent further losses of biodiversity due to climate change ecologists have invested significant effort developing tools to project future patterns of biodiversity. Presence-only species distribution models, and in particular MaxEnt, have become the most widely used tools for mapping the distribution of species and projecting how they will respond to climate change. However, these model predictions often fail to align with the observed responses of biodiversity to climate change. In this talk we investigate how the commonly assumed default prevalence value in MaxEnt models or an assumed threshold of habitat suitability for species occurrence overestimate biodiversity responses to climate change, and we provide 3 solutions for generating results that are more aligned with observations. We do this by linking MaxEnt models derived from presence-only records with known presences and absences from the Breeding Bird Survey for 284 bird species.

Results/Conclusions

These species-specific tau parameters were highly right-skewed with a mean of 0.271 and a median of 0.219, much less than the standard 0.5 that is default in MaxEnt models. Assuming either the default uniform species prevalence in MaxEnt models or a threshold for occurrence can greatly overestimate climate change effects on species richness and range size, underestimate its effect on rare species, and misidentify of the most important areas for conservation. We show that MaxEnt models can be brought more in line with real-world observations by (1) reporting proportional rather than absolute change or change using thresholds, (2) using a uniform, biologically informed prevalence value, or (3) using globally available species attributes to approximate species-specific default prevalence values.