Hong Kong scientists create AI weather system

January 29, 2026

Scientists in Hong Kong have created an advanced artificial intelligence system capable of predicting severe storms and heavy rainfall up to four hours in advance—significantly extending current forecasting limits of 20 minutes to two hours.

The breakthrough technology, developed by researchers at the Hong Kong University of Science and Technology (HKUST), is designed to strengthen disaster preparedness and improve emergency response as climate change increases the frequency of extreme weather events.

Project leader Professor Su Hui, Chair of Civil and Environmental Engineering at HKUST, explained that the system combines artificial intelligence with satellite imagery to deliver more accurate and timely weather predictions. “Our goal is to use AI and space-based data to enhance early warnings for severe weather, helping communities stay safer,” he said.

The new forecasting model focuses on predicting intense rainfall and thunderstorm activity. It uses generative AI methods that introduce controlled noise into training datasets, allowing the system to learn how to refine and reconstruct weather patterns for improved precision.

Developed in partnership with China’s meteorological agencies, the platform updates forecasts every 15 minutes and has increased prediction accuracy by over 15 percent, according to the research team. The findings were published in the Proceedings of the National Academy of Sciences in December.

The innovation comes at a critical time, as southern China and Hong Kong experienced unusually high levels of typhoons and extreme rainfall in 2025. The Hong Kong Observatory reported record-breaking weather alerts, including five highest-level rainstorm warnings and sixteen second-highest alerts during the year.

Both the China Meteorological Administration and the Hong Kong Observatory are preparing to integrate the AI model into their operational forecasting systems.

Known as the Deep Diffusion Model based on Satellite Data (DDMS), the framework was trained using infrared brightness temperature measurements collected by China’s Fengyun-4 satellite between 2018 and 2021. These satellite observations enable earlier detection of cloud development than traditional radar-based methods.

Researchers combined the satellite data with meteorological expertise to track the evolution of storm systems. The model was later tested using seasonal data from 2022 and 2023, confirming its reliability and forecasting performance.

With its ability to deliver faster and more accurate storm predictions, the AI-powered system represents a major step forward in climate resilience and disaster risk management across the region.