China’s “Fengqing” Weather AI Enters National Operations With an 11-Day Forecast Range

October 17 — According to Hong Kong’s Wen Wei Po, the “Fengqing” AI global short- and medium-range ensemble weather forecasting system V1.0 has recently passed its operational review and formally entered national weather operations. The system was trained and iterated end to end on the 100,000-card AI4S supercomputing cluster at the core node of the National Supercomputing Center in Tianjin. It is China’s first operational-grade weather AI ensemble forecasting model with fully self-controlled data, models and computing power, marking important progress in Tianjin’s supercomputing support for the country’s weather AI technology.

The Fengqing system relies on the computing infrastructure at the National Supercomputing Center in Tianjin, using a large-scale, highly parallel AI4S scientific computing cluster to train on vast amounts of weather data, optimise the model and iterate on accuracy. Its core computing power, underlying algorithms and data sources are all domestically produced, building a self-controlled technical system for intelligent weather forecasting and filling a gap in China’s global medium-range ensemble intelligent forecasting.

The Fengqing system was developed under the leadership of the National Meteorological Center together with universities and research institutes. It combines a vision Transformer architecture with meteorological physics constraints, so it can characterise the uncertainty of atmospheric motion and improve probabilistic forecasts of hazardous weather such as heavy rain, typhoons and severe convection. Its effective global forecast range reaches 11 days, and several core forecast metrics, probability forecast accuracy and the ability to control spread error are all better than the latest AI ensemble forecasting system of the European Centre for Medium-Range Weather Forecasts (ECMWF).

Compared with traditional numerical weather prediction models, the system can use Tianjin’s supercomputing power to complete a global, large-scale inference of weather elements within minutes, balancing forecast speed, accuracy and stability, and improving the ability to anticipate extreme weather and medium- to long-term trends.

Once in operation, the system will serve national weather disaster prevention and mitigation, transport, energy scheduling, ecological protection and public welfare. The development team will keep improving the model’s accuracy and forecast range, further refining the domestic intelligent weather forecasting system.

In recent years Tianjin has continued to promote the integration of supercomputing, AI and meteorological science, drawing on the computing advantages of the National Supercomputing Center in Tianjin to support research in atmospheric science, Earth system modelling and intelligent weather forecasting. The Fengqing system entering operation is a result of Tianjin’s supercomputing supporting national-level meteorological science and technology innovation, and will support China’s independent innovation in meteorological technology and the global development of intelligent weather forecasting.