Explainable artificial intelligence reveals spatially divergent effects of global change on mammals

Abstract Understanding how global change reshapes species distributions remains a central challenge in ecology, yet most global assessments treat environmental drivers as static, uniform, or aggregate influences. This obscures where, why, and how specific drivers generate ecological risk or opportunity. Here, we apply explainable machine learning to assess how climate averages, variability, extremes, and land cover are projected to reshape future distributions of 1,992 terrestrial mammals. Using Shapley Additive Explanations with species distribution models, we quantify the directional contribution of 16 environmental drivers and track how these contributions change over space and time. We propose two ecological diagnostics, directional turnovers and favorability shifts, that reveal where drivers switch from favorable to unfavorable, or the reverse, and where their influence strengthens or weakens without sign reversal. While not designed to infer causation, our framework assesses whether projected spatial patterns align with ecological expectations. We show that climate extremes impose intense but spatially restricted impacts, temperature-related drivers dominate exposure patterns, particularly for specialist species, and the same driver can impose risk in one region while creating opportunity in another. By decomposing projected biodiversity change into driver-specific signatures, our framework advances biodiversity forecasting toward a more actionable understanding of ecological risk and opportunity.