Statistical-Physical Adversarial Learning From Data and Models for Downscaling Rainfall Extremes

Abstract

Quantifying the risk from extreme weather events in a changing climate is essential for developing effective adaptation and mitigation strategies. Climate models capturing different scenarios are often the starting point for physical risk. However, accurate risk assessment for mitigation and adaptation often demands a level of detail they typically cannot resolve. Here, we develop a dynamic data-driven downscaling (super-resolution) method that incorporates physics and statistics in a generative framework to learn the fine-scale spatial details of rainfall. Our approach transforms coarse-resolution (0.25°) climate model outputs into high-resolution (0.01°) rainfall fields while efficaciously quantifying the hazard and its uncertainty. The downscaled rainfall fields closely match observed spatial fields and their distributions. Contrary to conventional thinking, our results suggest that coupling simple statistics and physics to learning improves the efficacy of downscaling midlatitude rainfall extremes from climate models.