ePO 63 – Deep-learning downscaling of human heat stress over iberia at high spatial and sub-daily resolution

Heat stress is a critical climate-related risk to human health, exacerbated by the increasing frequency, intensity, and duration of heat extremes. The Wet Bulb Globe Temperature (WBGT) is a robust heat stress metric that integrates air temperature, humidity, wind speed, and solar radiation, providing a physically meaningful measure of human thermal strain. However, WBGT projections at scales relevant for health impact assessment are limited by coarse, biased global climate models (GCMs) and the lack of high-frequency outputs. Here, we present a deep-learning (DL) based framework to generate high-resolution WBGT projections over the Iberian Peninsula (IP) under CMIP6 scenarios. A U-Net convolutional neural network is trained using ERA5, ERA5-Land and CERRA data to learn nonlinear mapping from coarse-resolution (≈1.5° – 1°) GCM predictors to fine-scale (0.1°- 0.05°) WBGT fields, overcoming a key limitation of conventional downscaling approaches. Historical evaluations indicate that the DL approach substantially improves the spatial structure, distribution, and extremes of WBGT relative to raw GCM outputs, yielding clear distributional added value, particularly for high-end heat stress conditions. Future projections reveal large, robust, and spatially coherent increases in mean and extreme WBGT across the IP, with intensification relating strongly with emissions scenario and time horizon. The strongest increases emerge in inland and central regions, while coastal areas show more moderate but still significant changes. Sub-daily results indicate a pronounced amplification of daytime heat stress extremes, underscoring the growing risk of heat-related illness under a warming climate. This study demonstrates the potential of DL to deliver physically consistent, high-resolution, sub-daily WBGT projections, providing a critical foundation for climate-health impact assessments, early-warning systems, and adaptation planning in heat-vulnerable regions.