Google DeepMind researchers report in Nature that their WeatherNext AI model can forecast a tropical cyclone's track, intensity and wind structure with state-of-the-art accuracy — giving forecasters roughly an extra day of warning compared with existing models. The improvement, achieved by a single model trained on nearly 20 terabytes of global atmospheric data and a database of nearly 5,000 historical storms, is roughly equivalent to a decade of meteorological progress.
During the 2025 hurricane season the model helped the US National Hurricane Center issue an early warning for Hurricane Melissa's rapid intensification and landfall in Jamaica. The team is now generating 1,000 forecast scenarios per cyclone to support decision-making. WeatherNext even runs at a 100x coarser resolution than traditional intensity models, an unexpected result scientists say they are still working to understand.
Alongside the paper, the models — WeatherNext Cyclones, WeatherNext 2 and a compact mini version that runs on a single TPU — are being open sourced, with code and weights freely available. Weather agencies including the UK Met Office and the National Hurricane Center collaborated on the work.
