Invited Talk
Advancing Global Weather Prediction: an AI-driven Data Assimilation Framework and the Fuxi Weather System
Wei HAN, China Meteorological Administration

Biography

Dr. Han Wei is the Deputy Chief Engineer at the China Meteorological Administration's Earth System Modeling and Prediction Center and the Deputy Chief Designer of the Fengyun Meteorological Satellite Engineering Application System. He is a member of the International Radiation Commission and serves on the editorial boards of Monthly Weather Review, Advances in Atmospheric Sciences, and Advances in Earth Science (Chinese).

Dr. Han's research interests include satellite radiance assimilation, operational data assimilation system development, and the integration of artificial intelligence with numerical weather prediction. He made important contributions to the Constrained Bias Correction and Constrained VarBC (CVarBC) schemes, which incorporate prior uncertainty constraints from satellite calibration and radiative transfer models to prevent model bias drift, significantly improving assimilation accuracy. The CVarBC scheme has been operationally adopted by ECMWF since 2018.

His work on assimilating next-generation geostationary hyperspectral infrared radiances earned him the Best Oral Presentation Award at the International TOVS Study Conference (ITSC) in 2021. His current AI research focuses on integrating AI with data assimilation, developing AI-enhanced assimilation frameworks, creating deep learning-based observation operators, and building an end-to-end AI weather forecasting system.

Traditional numerical weather prediction (NWP) and emerging deep learning (DL)-based forecasting models rely on initial conditions generated by conventional data assimilation (DA) systems. Recent breakthroughs in DL-based models have demonstrated competitive performance against leading operational NWP systems, inspiring the development of a novel DL-based DA framework. This study introduces AI-DA, a generalized DL framework designed to assimilate multi-source satellite observations by leveraging its multimodal data integration capabilities, mirroring the core principles of traditional DA systems. Experiments utilizing data from the Advanced Geosynchronous Radiation Imager (AGRI) aboard Fengyun-4B and five microwave sounders on polar-orbiting satellites, combined with Global Navigation Satellite System radio occultation data, validate AI-DA’s robust assimilation performance and physical consistency. Integrating AI-DA with the DL-based forecasting model FuXi, we establish FuXi Weather—the first end-to-end AI-driven assimilation-forecast system. Operating on a 6-hourly cycle, FuXi Weather generates 10-day global forecasts at 0.25° spatial resolution, achieving superior predictability compared to the European Centre for Medium-Range Weather Forecasts’ high-resolution forecasts (HRES). Notably, it extends the skillful prediction window for key variables (e.g., 500 hPa geopotential height) from 9.25 to 9.5 days while utilizing fewer observational inputs. Despite these advancements, short-term forecast accuracy requires further refinement. Future efforts will focus on enhancing system performance through expanded data assimilation and model optimization. This work underscores the transformative potential of AI in revolutionizing weather prediction frameworks.ReferenceLi, Z., Han, W., Xu, X., Sun, X., Li, H., 2024. All-Sky Microwave Radiance Observation Operator Based on Deep Learning With Physical Constraints. Journal of Geophysical Research: Atmospheres 129, e2024JD042436. https://doi.org/10.1029/2024JD042436Li, Y., Han, W., Li, H., Duan, W., Chen,L., Zhong, X., et al. (2024). Fuxi‐en4dvar:An assimilation system based on machine learning weather forecasting model ensuring physical constraints. Geophysical Research Letters, 51, e2024GL111136. https://doi.org/10.1029/2024GL111136