Niche-based species distribution models (SDMs) are commonly used to predict impacts of global change on biodiversity, but the reliability of these predictions in space and time depends on their transferability. We tested how the strategy used to choose predictors impacts the transferability of SDMs at a cross-continental scale.