Latest News

Hong Kong scientists launch AI-based model to better predict extreme climate

Hong Kong scientists launch AI-based model to better predict extreme climate
Hong Kong scientists launch AI-based model to better predict extreme climate

Hong Kong scientists have developed a system that uses artificial intelligence to forecast thunderstorms and heavy downpours four hours in advance, as opposed to the current range of 20 minutes to 2 hours.

The system will allow emergency services and governments to respond more efficiently to the increasingly extreme weather conditions linked to climate changes, said the team of Hong Kong University of Science and Technology on Wednesday.

Su Hui, professor and chair of the civil and environmental engineering department at the University, led the project.

Su said at a press conference that the system was designed to predict heavy rains. The work had been published in the Proceedings of the National Academy of Sciences, in December.

The model uses generative AI, which injects noise into the training data to teach the system to reverse the process in order to make more accurate forecasts.

The team claims that the system, developed in collaboration with China’s weather authorities and updated every 15 minutes, has increased accuracy by over 15%.

Scientists said that such work is vital because the number and intensity of typhoons, and wet weather episodes in Hong Kong and southern China in 2025 will be far above the seasonal norm.

The observatory reported that the city issued its most severe rainstorm warnings five times in a single year and its second-highest 16 times.

The China Meteorological Administration and Hong Kong Observatory are both working on incorporating the model into their forecasts.

The Deep Diffusion Model based Satellite Data (DDMS) was trained using infrared brightness temperatures collected by China's Fengyun-4 Satellite between 2018 and 2020.

Su said that satellites can detect clouds earlier than other forecasting systems, such as radar.

Data was combined with expert meteorological knowledge to capture the evolution and development of convective clouds systems. This was then validated by spring and summer samples taken in 2022 and 2023. (Reporting and writing by Joyce Zhou, Farah master; editing by Clarence Fernandez).

(source: Reuters)