China develops artificial intelligence satellite system to nowcast storms four hours ahead
Researchers in China developed a real-time storm nowcasting system that uses meteorological satellite observations and artificial intelligence (AI) to forecast severe convective weather up to four hours in advance. The system is designed to detect and predict the evolution of convective cloud clusters before they intensify into damaging weather, improving short-term warning capability.
“Nowcasting” refers to using current atmospheric observations to produce rapid, short-term forecasts for the near future. The approach is widely used in early warning systems because it targets fast-developing hazards such as thunderstorms, intense rain, and other severe convective events.
The research team built the system around continuous satellite monitoring from the Fengyun-4 (FY-4) series, which provides repeated, wide-area observation suitable for tracking complex, rapidly changing cloud systems. The model was described as a deep diffusion approach trained on satellite brightness temperature and related imagery to learn spatiotemporal patterns in convective development.
Coverage was reported at roughly 20 million square kilometers, spanning much of China and nearby regions. The system produces high-frequency predictions every 15 minutes and generates forecasts for convective evolution across the next four hours.
The work was published in the Proceedings of the National Academy of Sciences (PNAS) and described as improving performance for thunderstorm and convection nowcasting compared with existing approaches used for similar short-term forecasting tasks.
The rollout of Fengyun-4 03 (FY-4 03), launched in late December 2025, was also linked to higher-capability observation and faster data handling. The satellite was described as supporting inter-satellite coordination, high-speed data transmission, and broadcast functions, enabling three FY-4 spacecraft to operate as a more integrated observation network.
