Diffusion transformers improve short term rain forecasts accuracy
NowcastDiT: Diffusion Transformers are Effective Precipitation Nowcasters
Machine LearningArtificial IntelligenceComputer Vision and Pattern Recognition
Summary
Short-term weather forecasts for rain are hard because rain can change quickly in time and space. The authors found that a standard type of AI model called a Diffusion Transformer can predict these rain changes well, without needing complex special tweaks. They improved the model by adding ways to make predictions smoother over time and reward better weather-relevant results. Tests on known weather datasets showed their model gave better rain forecasts than previous methods.
What this means in practice
- •For meteorological agencies: Improve short-term precipitation forecasts with a new diffusion transformer model boosting accuracy and realism in weather nowcasting.
- •For emergency response teams: Use more accurate rain nowcasts for better preparation and risk management during sudden weather events.
Authors
Haoran Xu, Xingzhuo Guo, Yuchen Zhang, Jincheng Zhong, Jianmin Wang, Mingsheng Long
Abstract
Precipitation nowcasting demands accurate short-term forecasts under strong spatiotemporal variability. Diffusion models are well suited to modeling complex precipitation distributions, yet existing approaches often introduce increasingly specialized designs, leaving the capability of a standard diffusion architecture underexplored. We show that a standard Diffusion Transformer already provides a simple and scalable foundation for precipitation nowcasting, with domain-specific requirements accommodated naturally within its design space. Based on this principle, we develop NowcastDiT and instantiate this flexibility through two complementary adaptations: a dynamics-aware noise prior for temporally coherent forecasts, and end-to-end reinforcement learning with timestep-aware rewards for meteorological skill. Experiments on SEVIR and MRMS benchmarks show that NowcastDiT achieves state-of-the-art performance in both perceptual quality and meteorological skill. These results suggest that standard DiT can serve as an effective foundation for precipitation nowcasting.