MW-Nowcast is a six-hour ensemble radar nowcasting model that splits the prediction task into a deterministic part for large-scale storm structure and a generative part for local, uncertain variations. It was tested on independent data from the US, Europe, and China, where it outperformed leading methods for heavy and extreme precipitation across the full six-hour horizon. For the most intense rainfall, it doubled the warning time compared to the best existing generative baseline, achieving six-hour skill that was previously only possible at three hours.
Joint deterministic-generative model extends reliable extreme-precipitation nowcasting to six hours
MW-Nowcast achieves detection skill at 6 h previously limited to 3 h for leading methods, across the United States, Europe and China.
Big Tech
Ning Wang · Zuliang Fang · Weixin Jin · Zhongjian Lv · Shuang Qin · Pengcheng Zhao · +16 more
Microsoft Corporation · City University of Hong Kong · University of Cambridge
Research Digest··3 min read
In one line
A radar nowcasting model extends skilful extreme precipitation forecasts to 6 hours by separating predictable storm structure from uncertain local details.
Before you read this
An idea this paper builds onRead it as
The authors present MW-Nowcast, a six-hour ensemble radar nowcasting model that jointly learns a deterministic predictor for organized precipitation structure and a generative flow-matching model for local residuals.
The authors designed a conditional flow-matching generator that produces diverse ensemble members around a shared deterministic prediction of storm-scale structure, treating local growth, decay and initiation as stochastic residuals.
How it works
stage by stageThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.