Species distributions models (SDMs) are an important tool for predicting a species' distributional extent through use of e.g. climatic similarities. Stacking models of several taxa, moreover, can predict species richness and community composition, although the approach often leads to overpredictions.
Exploring a novel approach for improving stacked SDMs, this study suggests trimming the area used for calibration to a hypothesized accessible area, before the modelling is done. This involves visual inspection of the species occurrences and using natural breaks-e.g large rivers or mountain ranges–in habitats to limit theoretical distribution area.
The authors used GBIF-mediated occurrences of all hummingbird species (Aves: Trochilidae) to test the performance of stacked models–with or without constraining accessible area. By statistical comparison, the constrained model had significantly higher prediction success and also provided better estimates of richness and community composition.
In conclusion, limiting theoretical distribution area–effectively incorporating dispersal limitations–can improve efficacy of stacked species distributions models.