A 40-year, continent-wide, multispecies assessment of relevant climate predictors for species distribution modelling

AimSpecies distribution models (SDMs) are increasingly used to address numerous questions in ecology, biogeography, conservation biology and evolution. Surprisingly, the crucial step of selecting the most relevant variables has received little attention, despite its direct implications for model transferability and uncertainty. Here, we aim to address this with a continent-wide, evaluation of which climate predictors provided the most accurate SDMs for bird distributions. LocationConterminous United States. MethodsFor 243 species, we used yearly data since 1971 (from the North American Breeding Bird Survey) to run SDMs (six different algorithms) with combinations of six relatively uncorrelated climate predictors (selected from 22 widely used climate variables). We then estimated the importance of each predictor - both spatially and over a 40-year time period - by comparing the accuracy of the model obtained with or without a given predictor. ResultsThree temperature-related variables (annual potential evapotranspiration, mean annual temperature and growing degree days) produced significantly more accurate SDMs than any other predictors. Among precipitation predictors, annual precipitation provided the most accurate results. Albeit only rarely used in SDMs, the moisture index performed similarly strongly. Interestingly, predictors that summarize average annual climate produced more accurate distributions than seasonal predictors, despite distinct seasonal movements in most species considered. Encouragingly, spatial and temporal (over 40years) evaluation of variables yielded very similar results. Main conclusionsThe approach presented here allowed us to identify the statistically most relevant predictors for birds in the USA and can be applied to other taxa and/or in different parts of the world. Appropriately selecting the most relevant predictors of species distributions at large spatial scale is vital to identifying ecologically meaningful relationships that provide the most accurate predictions under climate change or biological invasions.