For decades, predicting the future weather has been a real challenge requiring teams of meteorologists to use raw supercomputing and pure physics. Today, artificial intelligence is rewriting those rules, as models based on machine learning are no longer a laboratory promise, but are working in real time, providing support to these professionals.
The story of Isaiah. Proving the power of this technology in meteorology, X user Packy Young shared how a DeepMind model predicted the appearance of Hurricane Isaiah with almost magical accuracy. In particular, the Weather Labs FNV3 model is one of these great promises.
As for the trajectory, the AI nailed it. According to HuriLab’s analysis, the AI model placed the moment of landfall with an error margin of a few hours to more than three days, and it was the model that came closest to the actual point of impact, which was in Florida. Meanwhile, traditional systems, both European and American, have drifted far in the West.
But with subtlety. Although it worked quite well, the director of the National Hurricane Center himself clarified that the official NHC forecast was 20 to 40% more efficient than AI at predicting hurricane strength. Here, undoubtedly, classical models like HWRF and HMON were the best in predicting long-term intensity.
a revolution. Beyond this very odd case, there is plenty of scientific literature that points us to the benefits of using these AI models. For example, an article is published the nature That same year, the Weather Next Cyclones model showed that, after it was evaluated with cyclones between 2023 and 2025, it was able to deliver an average expected benefit of a day or more compared to the main models working now.
But the most important evidence to us is provided by the NHC in their 2025 report, the year they incorporated AI models into their real-time operations. Here the DeepMind model outperforms the NHC forecast on track by 12 to 72 hours, achieving comparable efficiency in intensity. The result is that it is now perfectly integrated into everyday activities.
Institutional change. During Isaiah’s time, American meteorologists adjusted their official forecasts to resemble the DeepMind model, relegating historical giants like the European model or the American GFS to the background. This level of confidence also explains what happened with Hurricane Melissa, where the NHC relied on AI to predict its rapid intensification about three days before it reached Jamaica.
For its part, the European Center for Medium-Range Weather Forecasts has already fully implemented AI with its AIFS model ready with the IFS physical model since February 25 last year. The main difference here lies in its efficiency, as the AI predictably consumes about 1,000 times less power.
in spain, AEMET is not far behind, but maintains a more realistic approach with its progress within the European framework with the ANEMOI-ML model. Here the public agency points out that AI’s results today are worse than those of traditional models, with the main problem being the smoothing of meteorological extremes, which is an important risk factor when issuing warnings for torrential rainfall.
Meteorologists are needed A priori one might think they are not going to be replaced by AI. Currently, most models require traditional meteorology and a support system to capture all the information, including complex physical calculations. This does not mean that, when AI handles the entire process of taking data from sensors to weather maps, traditional physical models may become obsolete.
Picture | Josius Selinus
In Xataka | In 2023, twenty trucks crossed the Netherlands carrying 400 tons of Wadden Marsh. Its destination was a channel capable of creating storms
