MimoUDiff: A Unified Multi-Source Data Fusion Framework via MIMO Unet and Refined Diffusion for Precipitation Nowcasting

Chunlei Shi, Han Xu, Yongchao Feng, Dan Niu

Keywords: Precipitation Nowcasting; Multi-Source Data Fusion; Vision Mamba; Diffusion;

Abstract:

Precipitation nowcasting is a vital spatio-temporal prediction task essential for various meteorological applications, but it faces significant challenges due to the chaotic property of precipitation systems. Mainstream methods primarily rely on radar data for echo extrapolation, but over longer lead times, radar echoes mainly exhibit translation, failing to capture precipitation generation and dissipation processes. This results in blurry predictions, attenuation of high-value echoes, and positional inaccuracies issues. In the other hand, deterministic models using MSE loss often produce blurry forecasts, while probabilistic models struggle with localization accuracy. To address these challenges, we propose a multi-source data fusion framework, which integrates satellite and radar data, with former effectively complementing limitations of latter. In this framework, we leverages global motion fields to capture echo dynamics and introduces a residual diffusion mechanism to reduce memory usage by non-residual features. Various spatio-temporal models (e.g. RNN-based, CNN-based, and ConvRNN-based models) can seamlessly integrated into this framework. Extensive experiments on a Jiangsu dataset demonstrates significant improvements over state-of-the-art methods, particularly in short-term forecasts. The code and models will be released.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4429