Harnessing machine learning for quantifying and attributing compound heatwave changes in metropolis

Compound Heatwaves (CoHots), characterized by persistent day-night combined high temperatures, have intensified in recent decades, posing growing threats to human health, productive activities, and socioeconomic systems. Although much research has focused on the evolution of CoHots, high-resolution mapping of their changes in large metropolitan areas remains limited by sparse observational networks and coarse-resolution reanalysis data. Additionally, the influence of urbanization on the onset timing of CoHots has received little attention.

This study compares the start dates of CoHots across more than 700 urban–rural station pairs worldwide, revealing a significantly earlier onset in urban areas. Using machine learning and SHAP interpretability analysis, we demonstrate that this effect is primarily driven by urban building volume and height, rather than by the fraction of impervious surfaces. The influence is further amplified in climates with warm nights and strong daytime solar radiation.

To quantify urbanization's impact at a spatially and temporally continuous scale, we developed the Urban-informed Heatwave Ensemble AI Downscaling (U-HEAD) framework. This model integrates dynamic urbanization factors through an ensemble machine learning approach to downscale 0.25° ERA5 reanalysis data to 1 km resolution. Compared to the original product, U-HEAD substantially improves the simulation of spatiotemporal patterns and long-term trends of compound heat events. The framework can also be integrated with statistical downscaling methods to generate future high-resolution projections of CoHot evolution under combined climate change and urbanization scenarios. This research provides a robust, high-resolution modeling tool to quantify urbanization’s role in shaping compound heat extremes. Future work will focus on applying U-HEAD to project CoHot risks under various climate and urban development pathways, and to inform climate-resilient urban planning and heat adaptation strategies.