Meteorological Triggers of Regionally Extreme Atmospheric Evaporative Demand Across Global Land: Attribution With Adversarial Random Forests

Abstract

Atmospheric evaporative demand (AED) strongly influences drought intensification, vegetation stress, and wildfire risk. However, the meteorological triggers of regionally extreme AED remain poorly understood at the global scale. Here, we develop a physically constrained counterfactual attribution framework that combines the FAO56 Penman–Monteith equation with adversarial random forests (ARF) to quantify the relative contributions of the main meteorological drivers to regionally extreme AED across global land regions. This framework reduces bias in contribution estimates by preserving the dependence among correlated meteorological drivers. Applied to 40 years (1986–2025) of ERA5‐Land daily data on equal‐area grids, our analysis reveals that dominant trigger mechanisms vary systematically across hydroclimatic regimes. In humid tropical regions, extreme AED is primarily triggered by enhanced shortwave radiation under anomalously clear‐sky conditions, whereas wind speed dominates across many arid and semi‐arid regions where large background vapor pressure deficits amplify aerodynamic transport. In contrast, complementary Shapley‐value attribution shows that long‐term trends in mean AED are dominated by rising air temperature across most regions. This slow warming‐driven background “push” often differs from the more regionally diverse meteorological “triggers” that drive AED beyond its extreme threshold. More broadly, these findings reveal the complex and regionally varying hydroclimatic controls on atmospheric evaporative demand, while demonstrating the potential of the ARF‐based framework as a generalizable approach for quantifying variable contributions to extreme events.