Generative artificial intelligence improves projections of climate extremes

TL;DR

This paper presents FuXi-CMIPAlign, a deep learning framework that enhances climate model projections by improving the resolution and realism of extreme weather event simulations, addressing limitations of traditional models.

Contribution

It introduces a novel generative deep learning approach combining Flow Matching and domain adaptation to improve downscaling of climate extremes from coarse GCM outputs.

Findings

Enhanced accuracy, stability, and generalization in climate extreme projections.

Effective simulation of compound extremes like tropical cyclones.

Improved spatial, temporal, and multivariate downscaling capabilities.

Abstract

Climate change is amplifying extreme events, posing escalating risks to biodiversity, human health, and food security. GCMs are essential for projecting future climate, yet their coarse resolution and high computational costs constrain their ability to represent extremes. Here, we introduce FuXi-CMIPAlign, a generative deep learning framework for downscaling CMIP outputs. The model integrates Flow Matching for generative modeling with domain adaptation via MMD loss to align feature distributions between training data and inference data, thereby mitigating input discrepancies and improving accuracy, stability, and generalization across emission scenarios. FuXi-CMIPAlign performs spatial, temporal, and multivariate downscaling, enabling more realistic simulation of compound extremes such as TCs.