The recent tracking of Typhoon Dolphin’s path toward China showcased the use of advanced artificial intelligence in weather prediction. Alongside traditional methods, new AI models are gaining prominence, underlining China’s growing role in enhancing weather forecasting accuracy.
Chinese-developed systems such as Fengwu by Shanghai AI Laboratory, Huawei’s Pangu, and Fudan University’s Fuxi, are leading examples. These AI models produce forecasts faster than conventional methods and deliver comparable or superior accuracy. Traditionally, weather predictions depend on numerical models on supercomputers simulating atmospheric physics. In contrast, AI models analyze historical weather data, offering quicker forecasts.
The technology is especially important during East Asia’s typhoon season. Even slight improvements in predicting typhoon tracks aid in organizing evacuations, preparing for flooding, and managing transportation disruptions. This has spurred global competition among tech companies, research institutions, and meteorological agencies, with China emerging prominently.
“With more extreme weather, people need information to make decisions,” said Sun Zhi, CTO of Techwind. “We want to help provide better information so people can make decisions.”
Globally known AI systems include Google’s GraphCast and GenCast, Nvidia-supported FourCastNet, and the European Centre for Medium-Range Weather Forecasts’ AI system, AIFS. Fengwu gained attention for outperforming GraphCast in 80% of evaluated weather variables, extending accurate forecasts beyond 10 days.
Although AI models demonstrate substantial advantages, full replacement of traditional weather models is not imminent. For instance, while Fengwu accurately predicted Typhoon Dolphin’s landfall within 30 minutes and 30 km five days in advance, AI forecasts still fall short in predicting storm intensity and untested climate phenomena.
Sun from Techwind acknowledges the challenge in gaining public trust for AI predictions, citing the need for extensive scientific research. As both AI and traditional methods develop, their concurrent use will likely persist.