Welcome to the MambaRain Project! This initiative aims to develop an efficient and accurate long-term precipitation nowcasting framework. Through an innovative hybrid architecture and loss function design, we overcome the performance bottlenecks of traditional approaches for 0-3 hour forecasts and deliver more reliable technical support for meteorological decision‑making. Watch the following promotional video to quickly discover the strengths and features of MambaRain.
Click the image below to view the promotional video:
Note: GitHub does not support direct MP4 video playback within the README. Clicking the link above will redirect you to the playback page, where you can watch the video.
- [Multi‑Scale Spatio‑Temporal Modeling:Integrates Mamba’s long‑range temporal memory with self‑attention mechanisms to simultaneously capture the global evolution of storm systems and local convective details.]
- [Terrain-aware Forecasting:Enhances precipitation prediction accuracy in complex terrains (e.g., mountains, coastlines) by incorporating DEM geographic data.]
- [Cross‑Region Generalization:Validated on datasets from both the arid Xinjiang region and the humid southeast of China, achieving an average CSI improvement of over 12 %.]
@article{shi2025MambaRain,
title={MambaRain: Multi-Scale Mamba-Attention Framework for 0-3 Hour Precipitation Nowcasting},
author={xx, xx, xx, xx},
journal={preprint xxxxx},
year={2025}
}
If you have questions about this repo, please contact Shi ([email protected]).