China’s Fengwu AI Forecasted Typhoon Landfall Five Days Ahead

A typhoon gave China’s AI weather models a real-world test — and one system reportedly narrowed down the storm’s landfall days in advance.

Chinese researchers are using AI models to respond to the country’s current weather conditions proactively.

According to Reuters, Shanghai AI Lab’s Fengwu predicted the typhoon’s landfall location and timing five days ahead to within about 30 kilometers and 30 minutes. The result highlights how AI-based forecasting systems could complement conventional weather models by producing useful predictions quickly and with less computing power.

Fengwu is one of several AI weather systems being developed in China, alongside Huawei’s Pangu and Fudan University’s Fuxi, as meteorologists test whether machine-learning models can improve forecasting without replacing traditional physics-based systems or human expertise.

According to The Independent, its developers reported it outperformed Google’s GraphCast on 80% of evaluated weather variables in tests, extending skillful global medium-range forecasting beyond 10 days. That does not mean Fengwu is 80% more accurate overall. The figure refers specifically to the proportion of evaluated variables where it performed better.

Huawei has developed Pangu, while Fudan University has developed Fuxi, giving Chinese meteorologists multiple AI-based systems for forecasting atmospheric conditions.

The models work differently from traditional numerical weather prediction. Conventional systems use supercomputers to solve mathematical equations that represent atmospheric physics. AI models, on the other hand, are trained on vast datasets of historical weather data, enabling them to learn patterns and predict how atmospheric conditions will change.

Once trained, these models can produce forecasts with far less computation.

Why AI weather models are getting attention

The bigger story is not that AI can forecast weather; it is that prediction itself is becoming one of AI’s most practical uses.

The same basic approach can also be applied to problems such as wildfire risk and industrial equipment failures, where identifying a likely future event early can be more valuable than simply describing what is happening now.

Beyond these, there’s also the use of AI in prediction markets like sports and stocks.

Taken together, these give predictive AI a different role from the generative systems that have dominated public attention. Instead of asking an AI to create something, the user is effectively asking it to assess the available evidence and estimate the future.

AI systems aren’t replacing human experts

AI systems can be very good at learning atmospheric patterns from historical data, but weather is a chaotic physical system. Some events are difficult to predict because small differences in atmospheric conditions can produce markedly different outcomes, making it hard for AI models to forecast every weather event accurately.

Besides, AI systems tend to have certain weaknesses, particularly when dealing with edge cases, which are very common in weather forecasting. An unusual combination of weather conditions or a rapidly developing storm can create situations the model may not have encountered often enough during training to make an accurate prediction.

That means AI systems are not replacing the human experts who interpret forecasts. These human experts bring the unique human touch with years of experience to assess uncertainty and identify when an AI prediction may not tell the full story.

What this means for everyone

The clearest advantage of AI forecasting may be speed. Once trained, models such as Fengwu can produce forecasts with much less computation than conventional numerical systems.

If those systems can also maintain useful accuracy farther in advance, that could give emergency officials more time to issue warnings, prepare infrastructure and move people out of harm’s way. For now, the likely future is not AI replacing conventional forecasting, but AI becoming another increasingly important tool alongside physics-based models and human meteorologists.

Other News: Japan is planning a new AI data center in Akita, highlighting the push to expand computing infrastructure beyond the country’s traditional technology hubs.