BEIJING: Chinese scientists have developed a new artificial intelligence-powered forecasting system that significantly improves the accuracy of typhoon track predictions by combining advanced AI with established principles of atmospheric physics, according to a newly published scientific study.
The research, published in the journal Advances in Atmospheric Sciences, addresses one of the greatest challenges in weather forecasting: the atmosphere’s chaotic nature, where even the slightest changes in initial conditions can develop into major variations in a storm’s path.
This phenomenon has long limited the reliability of medium- and long-range typhoon forecasts.
While AI has become increasingly important in modern meteorology due to its ability to process vast amounts of weather data quickly and efficiently, many existing forecasting models rely primarily on recognising patterns from historical datasets.
Researchers say such approaches often lack a solid grounding in atmospheric physics, reducing their reliability over longer forecasting periods.
To overcome this limitation, a team led by Professor Duan Wansuo of the Institute of Atmospheric Physics at the Chinese Academy of Sciences, working alongside Professor Li Hao’s team at Fudan University, integrated nonlinear dynamics algorithms into China’s domestically developed FuXi meteorological large model. The result is a new forecasting platform known as FuXi-CNOPs.
According to the researchers, the system can identify the most sensitive atmospheric regions that influence a typhoon’s movement and detect the initial weather disturbances most likely to grow into significant forecasting errors.
By incorporating the physical laws governing atmospheric behaviour, the model generates predictions that are both more stable and scientifically robust.
The research team evaluated the system using data from 62 representative typhoons and conducted 91 comparative forecasting experiments.
The results showed that FuXi-CNOPs performs on a level comparable with leading international operational forecasting systems for 24-hour predictions.
Its strongest performance, however, was recorded in forecasts covering periods between 24 and 120 hours, where it consistently produced more accurate and dependable track predictions than conventional AI-based approaches.
The new system also offers notable computational advantages. Researchers said that while many established forecasting systems require 51 sets of computational data to complete a forecast, FuXi-CNOPs achieves improved accuracy and stability using only 31 datasets.
This reduction in computational workload lowers resource consumption while enhancing operational efficiency.
The researchers believe the technology could strengthen disaster preparedness by providing earlier and more dependable warnings of typhoon movements, helping authorities and emergency responders make better-informed decisions before severe weather strikes.
