Factors influencing transferability in species distribution models

Photo by M. G. Betts.

By Josée S. Rousseau and Matthew G. Betts

Predicting the abundance of birds across large geographical areas is essential for sound conservation planning. However, as the saying goes “all models are wrong, but some are useful”. The trick is determining why predictions about species abundances are sometimes wrong, under what conditions they fail, and which circumstances they succeed. Using >100 species, we found that species with large distributions, short life spans, and inhabiting regions with lower topographic variation are more likely to have abundance distribution models that fail when extrapolating to new areas. Long geographic distances between where models are built and where they are applied is also problematic.

These findings are important because many conservation efforts, such as translocating endangered species, forecasting biological invasions, and prioritizing the locations of reserves, rely at least in part on the accuracy of transferred models. Models often assume that a species is correlated with the same habitat across space, an assumption called “stationarity”, and often ignores that the environment across locations is different, thus resulting extrapolation problems.

To increase prediction accuracy across space, we recommend using models that account for non-stationarity. We also recommend limiting extrapolation whenever possible, by using similar environments between regions. Lastly, it is imperative that we consistently and continuously monitor our environment and its biodiversity in order to appropriately account for the impact of changes in habitat and climate on species abundances. This will improve species distribution models and increase the success of conservation efforts.

Comments