Chinese scientists have developed an AI-driven optimized ensemble forecast system for typhoons. The system combines China’s domestically developed FuXi AI weather model with the Orthogonal Conditional Nonlinear Optimal Perturbations (O-CNOPs) method, which generates physically interpretable initial perturbations to predict multiple possible typhoon tracks, the Global Times learned from the research team on Wednesday.

Tests against a wide range of typhoon cases, the system has demonstrated substantial performance improvements over leading global operational ensemble forecasting systems, marking a significant step toward more interpretable and reliable AI-based weather forecasting.

The research, conducted jointly by the research teams led by Duan Wansuo, a research fellow from the Institute of Atmospheric Physics, Chinese Academy of Sciences, and Li Hao, a research fellow from Fudan University, was recently published in the latest issue of the academic journal Advances in Atmospheric Sciences.

Due to the atmosphere’s chaotic nature, even tiny changes in initial conditions can significantly alter a typhoon’s path. Therefore, forecasters must predict not only a typhoon’s most likely track but also its possible deviations – a core challenge that ensemble typhoon track forecasting seeks to address, Duan told the Global Times on Wednesday.

AI-based weather models have developed rapidly in recent years and run much faster than traditional numerical models. However, forecasting unpredictable typhoon tracks requires more than speed. To become truly useful in operational forecasting, AI must also evaluate subtle differences in initial conditions, much like experienced forecasters do, and generate a range of credible track scenarios.

Researchers analyzed 62 typhoon cases and conducted 91 comparative ensemble forecasting experiments. Compared with the world’s leading operational ensemble forecasting systems, the system performs on par with leading global operational ensemble forecasting systems in 24-hour forecasts, but outperforms them over 24 to 120 hours, reducing maximum track errors by up to 32.33 percent.

The FuXi-CNOPs system also more consistently predicts typhoon’s speed and direction, and provides more reliable quantification of forecast uncertainty, improving uncertainty quantification accuracy by up to 29.2 percent.

The system has improved typhoon track forecasts by precisely identifying sensitive regions, providing a key foundation of physical interpretability for reliable AI-driven ensemble forecasting, researchers said.

The system offers a new AI-based ensemble forecasting approach that overcomes the limitations of purely data-driven models. Unlike mainstream generative AI-based ensemble forecasting approaches, it requires no additional training of massive models, reducing computing demands and making it more suitable for operational deployment.

Building on the current progress, the research team will continue to optimize and upgrade the system, aiming to improve typhoon track forecasts and eventually enhance intensity predictions. The framework is also expected to be extended to ensemble forecasting for other high-impact weather events, Duan said.