As extreme weather events become increasingly frequent, intense and long-lasting, a new artificial intelligence system developed in Hong Kong promises to significantly improve early warnings for severe storms and torrential rainfall.
A team of scientists has created an AI-based forecasting model capable of predicting heavy rain and thunderstorms up to four hours before they occur—far exceeding current forecasting windows, which typically range from 20 minutes to two hours. The system uses generative artificial intelligence techniques, deliberately adding noise to training data so the model learns to reverse the process and generate more accurate forecasts.
Developed in collaboration with China’s meteorological authorities, the model updates its predictions every 15 minutes and has improved forecasting accuracy by more than 15 percent, according to the research team. The technology comes at a critical time.
Researchers noted that in 2025, Hong Kong and large parts of southern China experienced an unusually high number of typhoons and high-humidity weather systems, far above seasonal norms.
According to the Hong Kong Observatory, the city issued its highest-level storm warning five times last year and the second-highest warning 16 times—both record figures.
China’s national meteorological service and the Hong Kong Observatory are now working to integrate the new model, known as the Deep Diffusion Model Satellite Data (DDMS), into operational forecasting systems.
The research team from the Hong Kong University of Science and Technology said the system could help governments and emergency response agencies better prepare for the growing risks associated with climate change.
“We hope to combine artificial intelligence with satellite data to improve extreme weather forecasting so that we can be better prepared,” said Xu Hui, professor in the Department of Civil and Environmental Engineering at the university and lead researcher on the project. “This system is designed specifically to predict intense rainfall events.”