Contemporary climate (temperature, precipitation etc.) strongly affects species distributions, which is why there are a plethora of modeling exercises that build statistical relations between species occurrences and long-term (e.g., 30- or 50-year) climate means. Yet, while ecologists increasingly recognize the dynamic nature of geographic ranges, species distribution models continue to be based on climate covariates that are essentially static in the short term. This is unfortunate, because the distributions of species – especially at their periphery – are likely dynamic, and may respond more to short-term weather events rather than long-term climate. We tested this hypothesis by comparing the use of short-term weather versus long-term climate means as predictor variables in species distribution models (SDMs) for 432 breeding bird species in the contiguous US. In addition, we evaluated how weather data affects predictions of bird distributions and if differences in model predictions were related to life-history traits. We validate the importance of weather with an independent dataset.
Results/Conclusions
Models based on weather outperformed those based on climate in model training (average increase in AUC of 0.043, p<0.0001), and more accurately predicted the independent testing data for the majority of species (increase in AUC of 0.023, p<0.0001). Similarly, weather data predicted species ranges better, particularly for long-distance migrant birds, species with large range size, and for specific foraging guilds such as raptors and upper-canopy insectivores. These results highlight the finding that long-term climate averages may mask the dynamic nature of the relationship between species and their environment and thus, misrepresent our interpretation of potential impacts of future climate change on species distributions.