China's First Fully Homegrown AI Weather Forecasting System 'Fengqing' Enters Operation
China has begun national operations of Fengqing, its first fully homegrown AI ensemble weather forecasting system. The V1.0 model passed a business-readiness review on October 17 and has officially entered service across the country's meteorological network.
Developed under the leadership of the National Meteorological Center, together with multiple universities and research institutes, the system was trained on a 100,000-GPU AI4S supercomputing cluster at the National Supercomputing Center in Tianjin. It combines vision Transformer architecture with meteorological physics constraints to represent atmospheric uncertainty, improving probabilistic forecasts of rainstorms, typhoons, and severe convective weather.
According to the development team, the system's core computing power, underlying algorithms, and data sources are all domestic. It delivers an effective global forecast lead time of 11 days and outperforms the latest AI ensemble forecasting system from the European Centre for Medium-Range Weather Forecasts (ECMWF) on several core metrics, probabilistic forecast precision, and spread-error control. Compared with traditional numerical weather prediction, Fengqing can complete global-scale weather simulations within minutes with the support of Tianjin's supercomputing resources.
The system will serve meteorological disaster prevention and mitigation, transportation, energy scheduling, ecological protection, and public services. The research team plans to continue improving forecast accuracy and lead time to further strengthen China's homegrown intelligent weather forecasting system.
