China's Meteorological Administration Unveils Three Advanced AI Weather Modeling Systems

On June 18, the China Meteorological Administration (CMA) unveiled three cutting-edge artificial intelligence models: the "Fengqing" global medium- and short-term forecasting system, the "Fenglei" near-term forecasting system, and the "Fengshun" global subseasonal-seasonal prediction system. Developed in partnership with Tsinghua University, these models signify CMA's commitment to enhancing meteorological predictions in line with global standards.

The "Fengqing" model combines advanced atmospheric physics with interpretability to deliver efficient computations grounded in solid physical principles. By automatically revealing the intrinsic evolution of weather systems, it significantly improves long-term forecasting accuracy. Utilizing a scalable multi-time optimization strategy, "Fengqing" enhances multi-day forecast effectiveness, extending prediction capabilities while continuously refining short- to medium-term forecasts. Tests show that it can reliably project weather conditions up to 10.5 days ahead, outperforming leading weather prediction models in Europe and North America, especially for longer forecasts.

To address challenges in near-term forecasting, the CMA and Tsinghua University developed the "Fenglei" model, which integrates data-driven and physics-based methodologies. This approach greatly enhances forecast accuracy for radar echoes at kilometer scales within the first 0 to 3 hours. The "Fenglei" model employs mesoscale physical modeling alongside AI-driven convection predictions, achieving significant improvements in detail and precision. A robust data-computation platform enables rapid radar echo extrapolation every 6 minutes, boosting strong echo prediction skills by 25%.

To manage uncertainties in climate predictions extending beyond 15 days, the CMA collaborated with Fudan University and the Shanghai Institute of Science and Intelligent Technology to launch the "Fengshun" model. This groundbreaking model utilizes flow-dependent ensemble perturbation techniques to better account for uncertainties in climate system changes and incorporates vital air-sea interaction processes, enhancing predictions of the Madden-Julian Oscillation (MJO).

These innovative models have been deployed on CMA's intelligent computing platform, producing deterministic and probabilistic forecast products for global elements and extreme events over the next 60 days. This deployment has notably improved forecasting accuracy for global precipitation.

The "Fengqing," "Fenglei," and "Fengshun" models have been rigorously trained and evaluated using China's global atmospheric reanalysis data (CRA-40), radar observations, and satellite remote sensing data. This advancement has significantly minimized reliance on international reanalysis data for mainstream weather prediction.

On May 24, the CMA announced an artificial intelligence weather forecasting model demonstration program at the Seventh Digital China Construction Summit and Digital Meteorology Subforum. This initiative aims to inspire collaboration across sectors, promoting innovation in AI technology development for meteorological applications. Furthermore, the CMA released its fifth batch of open-access meteorological data directories to facilitate the training and evaluation of AI models across various industries.