Diffusion Distillation for Efficient Weather Ensembles
Yiming Yang, Valentin Brekke, James Briant, Serge Guillas
cs.LG stat.AP
Abstract
Diffusion models generate skillful weather ensembles but require costly iterative sampling. We introduce a supervised energy-distance distillation method that compresses a multi-step diffusion teacher into a single-step student by aligning student forecasts with teacher samples and ground-truth observations. Experiments on global forecasting and typhoon-track prediction show that our student outperforms existing distillation methods and preserves skill for extreme events. It matches or surpasses the teacher across key metrics using only one neural function evaluation per autoregressive step.
Topics
Classified with taxonomy v2 on Sat, 5 Sept 2026.