Diffusion Distillation for Efficient Weather Ensembles

2026-08-27

AI Summary

为解决扩散模型生成天气集合预报需高昂迭代采样成本的问题，提出一种监督能量距离蒸馏方法，通过单步学生模型模拟多步教师模型。

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.

Problem

• Diffusion Models
• Weather Ensembles
• Efficiency
• Iterative Sampling

Innovation

• supervised energy-distance distillation
• multi-step to single-step compression
• weather ensembles
• extreme events