A European weather center is adding an AI model developed by Huawei to its forecasting work. The move reflects growing interest in machine learning for weather prediction, while also recognizing that traditional forecasting methods can still perform better on some details.
What Pangu-Weather brings to forecasting
The European Centre for Medium-Range Weather Forecasts (ECMWF) is working with the Chinese technology company Huawei to deploy AI-based forecasting models. Huawei’s Pangu-Weather model appeared on the European center’s website at the end of July.
The model can produce forecasts from one hour to seven days ahead. Its outputs cover conditions including temperature, humidity, wind and sea level pressure. Huawei’s AI chief scientist, Tian Qi, said it can provide global weather forecasts every second.
A paper in Nature reported that Pangu-Weather was more accurate than conventional numerical forecasting methods and produced predictions 10,000 times faster. Those reported capabilities help explain the interest in applying AI to forecasting, where speed and accuracy both matter.
Testing showed strengths, with limits
The center’s decision followed extensive comparative testing conducted between April and July of this year. In those tests, Pangu-Weather showed advantages on several accuracy indicators, including in extreme weather forecasts.
That does not mean the AI model performs best in every situation. Dr. Florian Pappenberger, head of forecasting at the European Meteorological Center, said Pangu-Weather is often more accurate on general trends, while traditional models can be better at specific details such as typhoon trajectories.
The distinction matters because a forecast can be useful at different levels: identifying a broad weather pattern and predicting the finer details of a particular event are related but separate challenges. The source describes AI and traditional models as having complementary strengths, rather than presenting AI as a complete replacement for established numerical methods.
Why researchers see potential in AI
Tian Qi said AI models may find patterns in large volumes of atmospheric data that people cannot detect. If those patterns help describe how the atmosphere develops, they could improve forecasts and address accuracy limits in numerical models.
Pappenberger described AI as a game changer for meteorology and said it would play an increasingly important role in the future. That view sits alongside the center’s recognition that conventional models can remain more capable for some specific predictions.
The promise, then, is not simply faster forecasts. AI systems could offer another way to learn from weather data, potentially improving the overall picture that forecasters can use. Their value will depend on how well they perform across different forecast measures and conditions.
A wider shift in weather and climate research
AI has been gaining a role in weather and climate research beyond this European center’s work with Huawei. The article points to Google unveiling a model in 2020 to predict weather data in real time, Stanford University presenting climate predictions for the next decade based on an AI model, and IBM and NASA publishing AI models for climate research.
Huawei also said it was in talks with the China Meteorological Administration about cooperation. The broader examples suggest that AI applications are being explored across both weather forecasting and climate research, although the source does not claim that these projects use the same methods or have the same goals.
For ECMWF, Pangu-Weather’s deployment marks a practical step: putting an AI forecast model alongside existing approaches. The comparison between them will remain central, especially where broad patterns and precise event details call for different strengths.