China is deploying domestically developed AI models at the forefront of its typhoon monitoring and forecasting infrastructure, marking a new step in the country's broader push to become a global AI leader. The national weather agency and affiliated research institutes are now running these homegrown systems in operational settings, with plans to extend them to developing countries across the Asia-Pacific and beyond. The effort comes as AI-based weather prediction is emerging as a competitive frontier, with models from both Chinese institutions and Western companies like Google DeepMind challenging the traditional numerical methods that have dominated global forecasting for decades.

An AI Weather System Built for Global Use

China was the first adopter of its own AI weather laboratories, but the country has signaled ambitions to make its systems an international standard. According to a Xinhua report carried by the State Council Information Office, the China Meteorological Administration (CMA) unveiled the Fenghe weather services large language model at this year's World Artificial Intelligence Conference in Shanghai, presenting it as a global open-source project.

Fenghe delivers personalized weather information and risk alerts to Chinese users. An international version was provided to the United Nations' Early Warnings for All initiative — a program aimed at ensuring universal hazard protection by 2027 — offering consultations in both Chinese and English and enabling developers to build applications on top of the model.

Countries Already Running Chinese AI Weather Models

Seven developing countries — Pakistan, Djibouti, Ethiopia, Jordan, Mongolia, the Solomon Islands, and Sri Lanka — have already integrated Chinese AI weather models directly into their daily disaster operations. Many of these nations sit in regions highly exposed to tropical cyclones, flooding, and other weather-driven hazards, yet have historically lacked the supercomputing infrastructure required for high-resolution numerical weather prediction. They are not using Fenghe, however. Instead, they operate Mazu (Multi-hazard Alert, Zero-gap and Universal), another CMA system that is further along in its international rollout.

Pan Jinjun, a CMA chief engineer, described Mazu as an "international public good" that provides "tailored early warning services for developing nations." An additional 40 Global South countries are currently experimenting with a faster public cloud version of the system, which is named after the Chinese sea goddess.

Research Shows Measurable Gains in Typhoon Forecasting

On the research front, Chinese scientists are reporting concrete improvements. Researchers at the Institute of Atmospheric Physics under the Chinese Academy of Sciences, collaborating with Fudan University, combined China's FuXi AI weather model with a technique called Orthogonal Conditional Nonlinear Optimal Perturbations (O-CNOPs) to map the potential range of storm trajectories. PreventionWeb reported the work on August 1.

Across 62 typhoon cases and 91 comparative experiments, the combined FuXi-CNOPs system matched the performance of the best global operational ensembles at the 24-hour mark and outperformed them from 24 to 120 hours. The system reduced maximum track errors by up to 32.33% and tightened uncertainty estimates by as much as 29.2%, according to the research team led by fellows Duan Wansuo and Li Hao. The findings were published in Advances in Atmospheric Sciences.

Duan told the Global Times that the approach requires no additional training of large models, which keeps computing costs down and facilitates deployment in routine operations — a practical advantage for weather agencies with limited computational budgets.

A Growing Stack of Chinese Forecasting Models

These tools build on a foundation of several Chinese forecasting models that have attracted international attention. FengWu, developed by Shanghai-based researchers, generates global forecasts at 0.25-degree resolution. In a study published in Communications Earth & Environment, FengWu outperformed the European Centre for Medium-Range Weather Forecasts' high-resolution model — long regarded as the global benchmark for medium-range prediction — as well as Pangu-Weather and GraphCast. It extended skillful forecasts beyond the 10-day threshold and reduced five-day tropical cyclone track error to 201 kilometers for 2022.

Huawei's Pangu-Weather, another domestically developed system, produces forecasts significantly faster than traditional numerical prediction methods. Huawei has stated that both the ECMWF and the CMA's National Meteorological Center have confirmed its performance advantages.

Together, these models provide Beijing with a domestic technology base as it promotes its weather AI systems internationally, positioning China not only as a user of advanced forecasting but as a supplier to nations seeking affordable alternatives to Western-run prediction infrastructure.