SwinRDM: Integrate SwinRNN with Diffusion Model Towards High-Resolution and High-Quality Weather Forecasting

Association for Artificial Intelligence 2023, Lei Chen, Du, Fei, Hu, Yuan, Fan Wang, Zhibin Wang

arXiv preprint · 11 January 2023 · Alibaba (DAMO Academy)

In one sentence. SwinRDM couples a SwinRNN forecaster with a diffusion super-resolution step to reach 0.25° forecasts that beat IFS up to 5 days on Z500, T850, T2M and precipitation.

Abstract

Data-driven medium-range weather forecasting has attracted much attention in recent years. However, the forecasting accuracy at high resolution is unsatisfactory currently. Pursuing high-resolution and high-quality weather forecasting, we develop a data-driven model SwinRDM which integrates an improved version of SwinRNN with a diffusion model. SwinRDM performs predictions at $0.25^\circ$ resolution and achieves superior forecasting accuracy to IFS (Integrated Forecast System), the state-of-the-art operational NWP model, on representative atmospheric variables including 500 hPa geopotential (Z500), 850 hPa temperature (T850), 2-m temperature (T2M), and total precipitation (TP), at lead times of up to 5 days. We propose to leverage a two-step strategy to achieve high-resolution predictions at $0.25^\circ$ considering the trade-off between computation memory and forecasting accuracy. Recurrent predictions for future atmospheric fields are firstly performed at $1.40625^\circ$ resolution, and then a diffusion-based super-resolution model is leveraged to recover the high spatial resolution and finer-scale atmospheric details. SwinRDM pushes forward the performance and potential of data-driven models for a large margin towards operational applications.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.