Google DeepMind open-sources AI model for cyclone forecasts

Google DeepMind and Google Research have released an open-source artificial intelligence model designed to significantly improve the forecasting of tropical cyclones, a development that could extend critical warning times for communities in a storm's path. The model, named WeatherNext, was made publicly available on GitHub this week, providing its code and pre-trained weights for researchers and forecasting agencies worldwide.

The system is engineered to predict both the track and intensity of cyclones up to 15 days in advance, generating a single probabilistic forecast in less than a minute on specialized hardware. This speed and extended forecast window contrast with traditional physics-based models, which can require days of supercomputer time. A key innovation is the model's ability to produce these detailed predictions using lower-resolution global atmospheric data, with grid cells measuring 28 square kilometers, which is substantially less granular than conventional methods.

During the 2025 Atlantic hurricane season, the National Hurricane Center tested a second-generation version of the model, known as WeatherNext 2. In those evaluations, the AI system reportedly...