China leads AI weather forecasting as extreme weather intensifies, Fengwu predicts typhoon within 30 minutes

China Emerges as AI Weather Forecasting Leader

China has positioned itself at the forefront of AI weather forecasting as extreme weather events intensify globally. Three Chinese-developed systems—Fengwu from Shanghai AI Laboratory, Huawei Pangu from Huawei, and Fuxi from Fudan University—are now working alongside traditional numerical weather prediction models to deliver faster, more cost-efficient forecasts1

1

2

2

. The deployment of these AI-driven weather forecasting systems during typhoon season in East Asia marks a shift in how meteorological agencies approach disaster preparedness.

The rise of AI weather forecasting has created intense competition among technology companies, research institutes, and meteorological agencies worldwide. While Google GraphCast and GenCast, Nvidia-backed FourCastNet, and the European Centre for Medium-Range Weather Forecasts' AI Forecasting System (AIFS) represent Western efforts, China's coordinated push involving state labs, tech champions, and universities signals a deliberate national strategy2

2

3

3

.

Fengwu Outperforms Google GraphCast on Multiple Metrics

Fengwu has attracted significant attention after developers reported it outperformed Google GraphCast across roughly 80% of evaluated weather variables and extended skillful global medium-range forecasts beyond 10 days1

1

5

5

. This performance represents a narrowing gap between Chinese and Western AI capabilities in meteorology. The system's ability to push useful forecasts past the traditional 10-day threshold could fundamentally change how governments and communities prepare for weather events.

Sun Zhi, CTO of Techwind, the company responsible for Fengwu's industrial applications, emphasized the practical urgency driving this work: "With more extreme weather, people need information to make decisions, both local governments, the national government, also the average person, farmers and fisherman. So we want to help provide better information so people can make decisions"1

1

5

5

.

Typhoon Dolphin Demonstrates AI Precision

The recent tracking of Typhoon Dolphin provided a real-world test of these AI systems' capabilities. Five days before the storm made landfall on mainland China, Fengwu predicted the time and place of impact to within 30 minutes and 30 km (19 miles)1

1

2

2

. This level of precision transforms evacuation planning from guesswork into actionable logistics, giving communities critical additional time to move people and assets out of harm's way.

During typhoon season, even small improvements in track forecasts help authorities better prepare for flooding, organize evacuations, and manage potential transport disruptions1

1

3

3

. For a country managing weather-exposed agriculture and extensive coastlines, this speed and accuracy carry direct economic and safety implications.

Speed and Cost-Efficiency Advantages Over Traditional Systems

For decades, weather prediction has relied on traditional numerical weather prediction models running on supercomputers that simulate atmospheric physics. AI models instead learn patterns from vast archives of historical weather data and can produce forecasts in a fraction of the time1

1

3

3

. Chinese researchers emphasize that faster forecasts enable running more scenarios, updating warnings more frequently, and issuing earlier alerts for typhoons, floods, and heatwaves4

4

.

The cost-efficiency advantage is equally significant. These AI-driven weather forecasting systems operate at a fraction of the computing cost required by supercomputers, making advanced meteorology more accessible to resource-constrained agencies1

1

4

4

. This efficiency could democratize access to high-quality weather forecasting globally.

Limitations in Storm Intensity and Climate Change Predictions

Despite impressive advances in tracking, AI models still lag conventional weather forecasts in predicting storm intensity—a critical gap that determines whether a storm is manageable or catastrophic1

1

2

2

. The systems also remain untested in predicting major long-term climate events like El Niño patterns or sea surface temperature changes affecting ecosystems.

Sun Zhi acknowledged these constraints candidly: "If we predict a climate change event 18 months in advance, people won't believe it. They need to know it's reliable. We need to do years of scientific research before people trust us when we say there will be an El Nino event or we say the changing temperature on the sea's surface will affect the breeding cycle of fish"1

1

5

5

. Building public trust requires extensive validation against real-world outcomes.

Hybrid Approach Likely to Continue

While AI systems are becoming an increasingly important complement to conventional forecasting, they are unlikely to fully replace traditional weather models in the near future1

1

3

3

. A hybrid approach leveraging both AI speed and traditional physics-based precision appears optimal for the foreseeable future.

Experts note that older models still perform better with storm intensity prediction and rare, record-breaking events. Many of the performance claims from AI systems still require independent, peer-reviewed testing at scale4

4

. As climate change continues serving up conditions outside historical training data, the complementary strengths of both approaches become increasingly valuable for comprehensive disaster preparedness.