A locally-developed artificial-intelligence-powered storm forecast system is being tested by the national and Hong Kong weather authorities, with the system able to improve the forecasting of thunderstorms or sudden rainfall from the current maximum of two hours to four hours ahead.
The system, developed by a research team from Hong Kong University of Science and Technology (HKUST), can boost the accuracy by over 15 percent at a 48-kilometer spatial scale.
During a media briefing on Wednesday, the research team said that the breakthrough strengthens the nation’s forecasting accuracy and transforms early warnings systems for vulnerable communities across Asia, providing more time for emergency services and the public to prepare for extreme weather hazards.
The China Meteorological Administration and the Hong Kong Observatory are testing the warning system, the team said.
A key feature of the new AI computational framework — the Deep Diffusion Model of Satellite Data — is that its training is based on data collected by China's FengYun-4A satellite between 2018 and 2021, and it incorporates professional meteorological domain expertise. With satellite sequence data, it accurately captures the evolution of convective cloud structures, and allows for observation of changes on the Earth's surface, hours or days in advance — much earlier than using data from radar or automatic weather station. This also makes it particularly valuable for predicting extreme weather in areas with limited radar or automatic weather stations.
The study on the forecasting model was published in Dec 2025 in the Proceedings of the National Academy of Sciences.
ALSO READ: Cross-boundary data sharing to benefit 300,000 HK patients
Dai Kuai, the first author of the paper, explained that conventional weather forecasting models rely mainly on ground-based radar, but radar signals are easily affected by terrain and the composition of precipitation and often detect changes only after convective clouds have already formed. By leveraging satellite data that monitor cloud evolution from space, the new AI model can detect signs of convective development much earlier, enabling more timely warnings.
Dai added that the new model, with faster and more accurate forecasts, will help with strengthening regional disaster preparedness and response. He highlighted the increasing frequency of extreme weather in Asia, citing the four black rainstorm warnings — with heavy rainfall exceeding 70 millimeters in an hour — that were issued in Hong Kong within a period of eight days last summer.
Su Hui, who leads the Climate Change and Extreme Weather direction for the State Key Laboratory of Climate Resilience for Coastal Cities, said that the model is now undergoing trial by the China Meteorological Administration, which also provides satellite data for research, and the Hong Kong Observatory. The Observatory and the HKUST team plan to combine it with traditional radar‑based forecasting to improve regional accuracy and provide more complete weather outlooks.
Meanwhile, a startup from HKUST is commercializing the technology through custom solutions for corporate clients. Organizations and companies worldwide, including insurers, power stations and a Vietnamese ride‑hailing firm, have already shown interest in collaborating.
Contact the writer at amberwu@chinadailyhk.com
