Item

Abstract

The most extreme heat waves are intensifying faster than moderate events across many mid-latitude regions. Climate models underestimate these trends, suggesting an incomplete representation and understanding of the underlying physical processes. Here, we use an explainable machine-learning framework to investigate interactions among atmospheric circulation, surface soil moisture, and temperature, and their role in recent heat extremes. We show that a three-way interaction among these components becomes active once a critical temperature threshold is exceeded, contributing, on average, ∼10% to peak temperature anomalies during the most extreme heat events across ∼28% of humid mid-latitude regions. We further reveal that this interaction is systematically underestimated in CMIP6 and CESM2 Large Ensemble simulations. Soils in climate models are too dry at the initiation of heat waves, therefore suppressing subsequent soil-moisture depletion and inhibiting feedback activation. These findings help explain regional model biases in upper-tail warming and highlight avenues for improving future projections of extreme heat events.