Weather forecast via AI – wissenschaft.de

Until now, weather forecasting was a complex challenge that required maximum performance from mainframe computers. Two research teams have now independently developed methods that use AI to make predictions that are as accurate as classical models – and with significantly less computational effort. One model forecasts the weather up to a week in advance, and the other specializes in short-term forecasts of heavy precipitation. Experts warn that extreme weather events in particular could push AI to its limits.

Air pressure, temperature, wind speed, water vapor: These and many other physical factors have been considered so far when meteorologists use complex algorithms on large computers to calculate how the weather is likely to develop in the days ahead. However, the predictions of so-called numerical models are time consuming and require a great deal of computational effort. As an alternative, different research teams are working on using artificial intelligence to predict the weather. However, earlier models were too imprecise for practical use.

Accurate at multiple speeds

Now two research teams have independently developed AI systems that can rival classical models in accuracy and far exceed them in speed. The first model, Pangu-Weather, comes from a team led by Kaifeng Bi of Huawei Cloud in Shenzhen, China. “By training models on 39 years of global weather data, Pangu-Weather predicts better than the world’s best numerical system and much faster at the same time,” the research team says.

Unlike previous models, Pangu-Weather does not include physical parameters. “Instead of making predictions based on physical knowledge, AI predicts statistically plausible weather patterns based on historical measurements,” explain Amy Ebert-Above and Kyle Hilburn of Colorado State University, who were not involved in the study. Accompanying commentary, also published in Nature. “The AI ​​model provides predictions about 10,000 times faster than numerical models with the same spatial resolution and comparable accuracy.” Since Pangu-Weather includes a 3D model, it also provides reliable values ​​for different heights.

Heavy rain forecast

The second model, NowcastNet, was developed by a team led by Yuchen Zhang of Tsinghua University in Beijing and focuses on an area where classical weather forecast models previously had problems: the short-term prediction of extreme precipitation. “Heavy rainfall contributes significantly to meteorological disasters, and there is a great need to mitigate their social and economic impacts through skilled forecasts with high accuracy, long time duration, and local detail,” the team wrote. “Existing methods are very error-prone: physical-based numerical methods struggle to capture the chaotic dynamics involved in these events, and previous data-based learning methods do not follow the laws of physics.”

For NowcastNet, the research team combined physical equations with machine learning methods. “Based on radar observations from the USA and China, our model generates physically reasonable predictions of precipitation with timescales of up to three hours,” the researchers explain. The model was tested by 62 meteorologists from across China. NowcastNet outperformed previously leading methods in 71 percent of cases.

Opportunities and risks

From the point of view of Ebert-Uphoff and Hilburn, new AI models offer great opportunities on the one hand. “These methods are so promising that they could lead to a paradigm shift, in which artificial intelligence models completely replace numerical methods,” they wrote. But at the same time, they urge caution. Because AI is trained using historical data, it may not be accurate enough if man-made climate change leads to an unprecedented build-up of extreme weather events.

“Given the potential benefits and risks associated with AI models for weather forecasting, it is time for meteorologists to get involved to ensure that AI weather forecast models are well-suited to their tasks,” said Ebert-Uvoff and Hilborn. Moreover, meteorologists need to learn how to interpret their forecasts, because AI models behave differently than physical models, so understanding their forecasts takes special training.

Sources: Kaifeng Bi (Huawei Cloud, Shenzhen, China) et al., Nature, doi: 10.1038/s41586-023-06185-3; Yuchen Zhang (Tsinghua University, Beijing, China) et al., Nature, Available Here. doi: 10.1038/s41586-023-06184-4

© wissenschaft.de – Elena Bernard