China deploys AI weather models to combat intensifying extreme weather

China is aggressively integrating artificial intelligence into its meteorological infrastructure to mitigate the risks of increasingly frequent extreme weather. By deploying a suite of AI-based forecasting models, the country aims to provide faster and more precise warnings to protect its population, agriculture, and critical infrastructure.

At the center of this push are three primary models: Fengwu, developed by the Shanghai AI Laboratory; Pangu, created by Huawei; and Fuxi, from Fudan University. Unlike traditional numerical weather prediction, which relies on supercomputers to simulate complex atmospheric physics, these AI systems learn patterns from vast archives of historical weather data. This fundamental shift in approach allows for significantly faster forecast generation, reducing the time between data collection and actionable warnings.

A shift in forecasting power

The transition to AI represents a new competitive arena in global meteorology. For decades, the industry relied almost exclusively on numerical models. Now, China's efforts place it alongside other global leaders developing similar technology, including Google with its GraphCast and GenCast models, Nvidia's FourCastNet, and the European Centre for Medium-Range Weather Forecasts with AIFS.

Technical evaluations indicate that these new tools are reaching a high level of maturity. Specifically, the Fengwu model has outperformed Google's GraphCast across approximately 80% of 880 evaluated weather variables. Furthermore, Fengwu has extended the window for skillful global medium-range forecasts beyond 10 days, reaching a threshold of 10.75 days.

Implications for disaster management

The ability to generate rapid, accurate predictions is a critical requirement for disaster management in East Asia, a region frequently battered by typhoons that cause widespread flooding and transport disruptions. The speed of AI models allows for a more agile response from both state and private actors.

"With more extreme weather, people need information to make decisions, both local governments, the national government, also the average person, farmers and fisherman," said Sun Zhi, CTO of Techwind. For these groups, the difference between a 48-hour and a five-day warning can be the difference between total crop loss and a successful harvest, or between a timely evacuation and a catastrophe.

The road ahead

Despite these gains in speed and medium-range tracking, AI is not yet a total replacement for traditional meteorology. The current strategy in China involves using AI models alongside traditional numerical systems to create a hybrid forecasting environment.

While AI excels at pattern recognition and path prediction, the industry continues to monitor how these models handle the nuances of storm intensity and long-term climate events. The next phase of development will likely focus on closing these gaps to ensure that the speed of AI is matched by the physical precision of traditional atmospheric science.