Generative climate downscaling enables high-resolution compound risk assessment by preserving multivariate dependencies

2026-05-12

AI Summary

Traditional climate downscaling methods struggle to preserve the complex inter-variable dependencies while enhancing spatial resolution, thereby limiting the accuracy of compound hazard risk assessments. This study introduces diffusion generative models to climate downscaling for the first time, proposing a novel generative framework that jointly models multiple meteorological variables—specifically temperature, precipitation, and three others—while integrating bias correction. Evaluated over Japan at a 50-fold increase in linear resolution, the method reduces inter-variable correlation errors by more than fourfold compared to existing approaches. It also achieves substantial improvements in univariate accuracy, fidelity of spatial detail, and detection capability for extreme drought events, demonstrating clear superiority over current state-of-the-art techniques.

Abstract

Physics-based climate projections using general circulation models are essential for assessing future risks, but their coarse resolution limits regional decision-making. Statistical downscaling can efficiently add detail, yet many methods treat variables independently, degrading inter-variable relationships that govern compound hazards such as heat stress, drought, and wildfire. Here we show that a diffusion-based multivariate generative framework, combined with bias correction, recovers degraded inter-variable correlations even under a 50$\times$ increase in linear resolution. When applied to five meteorological variables over Japan, the framework reduces inter-variable correlation errors by more than fourfold relative to existing baselines while improving both univariate and spatial accuracy, leading to more accurate detection of severe drought. These results demonstrate that multivariate generative downscaling improves the reliability of compound risk assessment under large resolution gaps.

Problem

• climate downscaling
• compound risk
• multivariate dependencies
• inter-variable correlations
• high-resolution

Innovation

• generative downscaling
• diffusion model
• multivariate dependencies
• climate risk assessment
• bias correction