Efficient Kilometer‐Scale Precipitation Downscaling With Conditional Wavelet Diffusion

Abstract Precipitation products such as Integrated Multi‐satellitE Retrievals have coarse resolution ( km), which limits their application in hydrological modeling and extreme weather analysis. We propose the Wavelet Diffusion Model (WDM), a fast generative framework for high‐quality precipitation downscaling trained on multi‐radar multi‐sensor radar data ( km). WDM employs a conditional diffusion model to learn multiscale precipitation structures in the wavelet domain and to reconstruct high resolution precipitation from low resolution data. Compared with pixel space baselines, our approach produces more realistic fields with fewer artifacts and offers 9× faster inference to kilometer scale resolution. Our results demonstrate that WDM addresses both accuracy and speed, providing a promising method for enhancing hydrological forecasts and other geoscience downscaling tasks.