A newly deployed artificial intelligence model is significantly improving the accuracy of typhoon forecasts in Hong Kong and mainland China, addressing one of the most challenging aspects of meteorology: predicting rapid intensification. The system, developed by researchers at the Shenzhen Institutes of Advanced Technology (SIAT) under the Chinese Academy of Sciences, has been integrated into the operations of the Hong Kong Observatory and China’s National Meteorological Center.
According to the South China Morning Post, the AI model was installed approximately three weeks ago and has already demonstrated its capabilities by providing real-time updates on the progression of Typhoon Jangmi, which formed in late May and made landfall in Japan on June 3, causing significant disruptions and forcing airlines, including Cathay Pacific and Hong Kong Airlines to cancel or reschedule flights.
Tackling the Hardest Problem in Tropical Meteorology
The primary focus of the new AI model is predicting rapid intensification, defined as a tropical cyclone’s maximum sustained winds increasing by 15 meters per second within 24 hours or by 10 meters per second within 12 hours. This phenomenon is notoriously difficult to forecast using traditional numerical weather prediction methods, and failures to predict it have repeatedly led to inadequate preparation and avoidable casualties.
Li Qinglan, a professor at SIAT who leads the project, explained that rapid intensification is highly unpredictable, often leading to delayed preventive measures and emergency responses. The new model addresses this by integrating four machine learning algorithms to analyze complex patterns in atmospheric data, sea surface temperatures, and upper-level wind shear, which traditional models struggle to process simultaneously. The research team began its work on typhoon intensity prediction in 2013, giving it over a decade of accumulated expertise.
Outperforming the US National Hurricane Center
When tested against historical data, the AI model achieved a higher hit rate and a lower false alarm rate for 24-hour rapid intensification events in the North Atlantic Ocean between 2016 and 2020 compared to forecasts from the US National Hurricane Centre (NHC). This benchmark comparison is significant: the NHC is widely regarded as the global standard for tropical cyclone forecasting.
(Related: China’s State Council Issues ‘Employment-First’ Plan Targeting AI Job Creation)
Wong Wai-kin, a senior scientific officer at the Hong Kong Observatory, confirmed that the team will continue evaluating the model’s performance using real-time data from the current typhoon season, which meteorologists forecast will produce four to seven typhoons affecting Hong Kong. The observatory already utilizes various AI tools to support weather forecasting and makes some AI-powered forecast products available to the public through its Earth Weather web portal.
China’s Growing Strength in AI Meteorology
The deployment of this model is part of a broader pattern of Chinese advances in AI-driven weather forecasting. In 2024, a consortium including the Shanghai AI Laboratory and the Hong Kong University of Science and Technology launched FengWu-GHR, an advanced AI-driven global high-resolution weather forecasting model that extended the effective forecast lead time by 12 hours to 11.25 days. Separately, Huawei’s Pangu-Weather model, published in Nature in 2023, demonstrated that AI could outperform traditional numerical weather prediction models on medium-range forecasts.
Real-World Stakes: Why Forecast Accuracy Matters
The practical consequences of improved typhoon forecasting are significant. Hong Kong’s economy is highly sensitive to typhoon signals — a No. 8 signal triggers widespread business closures, flight cancellations, and public transport shutdowns, costing the city hundreds of millions of Hong Kong dollars per event. More accurate predictions of rapid intensification would allow authorities to issue warnings earlier and with greater confidence, giving residents, businesses, and emergency services more time to prepare.
For mainland China, the stakes are even higher. Coastal provinces like Guangdong, Fujian, and Zhejiang are regularly struck by typhoons that cause billions of yuan in agricultural and infrastructure damage. The National Meteorological Center’s adoption of the SIAT model reflects a recognition that AI-driven forecasting can deliver material improvements in preparedness and response, translating directly into reduced economic losses and, more importantly, saved lives.
Looking ahead, the SIAT research team is extending its AI forecasting work beyond intensity prediction to cover strong winds, heavy rainfall, and storm surge, the three hazards that cause the most damage and casualties when a typhoon makes landfall. These are areas where traditional numerical weather prediction models have historically struggled with spatial resolution and lead time. If the AI approach proves as effective for these hazards as it has for rapid intensification, it could fundamentally change how coastal communities in China and Hong Kong prepare for extreme weather events, providing emergency managers with longer and more reliable windows for evacuation, infrastructure protection, and resource pre-positioning.
The SIAT model’s deployment at two of the region’s most important meteorological institutions suggests that Chinese weather agencies are moving beyond pilot projects and beginning to integrate AI forecasting into their core operational workflows, a transition that, if successful, will make AI an indispensable part of China’s climate resilience infrastructure.