DeepMind's WeatherNext AI predicts hurricanes earlier

A new artificial‑intelligence system called WeatherNext, which will be released as open‑source software, has demonstrated the ability to forecast a storm’s trajectory and strength with high accuracy despite being fed lower‑resolution atmospheric data. The model, developed by a team of meteorologists and computer scientists, outperformed several established forecasting tools in recent validation tests, correctly predicting both the path and intensity of multiple tropical cyclones several days in advance.

Although the results are promising, the underlying mechanisms that enable WeatherNext to extract detailed predictive signals from coarse inputs remain unclear to the researchers. Ongoing investigations aim to decode the model’s internal representations and assess its reliability across diverse weather scenarios, with the goal of integrating the technology into broader forecasting workflows once its operation is fully understood.