Google DeepMind said it has open-sourced the code and model weights for WeatherNext to support earlier and more accurate tropical cyclone forecasts.
According to the developers, WeatherNext’s three-day forecast is as accurate as the two-day forecasts produced by previous models. The system predicts a cyclone’s track, intensity and wind structure at the same time.
The model was trained on nearly 20 TB of global atmospheric data and the IBTrACS database covering almost 5,000 historical storms. WeatherNext can calculate 1,000 scenarios for each cyclone, including rare cases of rapid intensification.
The system produces a single 15-day forecast in under a minute on a TPU. It requires input data at a resolution of 28 by 28 kilometres, about 100 times coarser than traditional high-resolution models.
During the 2025 hurricane season, the model helped the US National Hurricane Center forecast Hurricane Melissa’s rapid intensification and landfall in Jamaica. This gave emergency teams additional time to prepare.
Google DeepMind released the code and weights for WeatherNext Cyclones, WeatherNext 2 and the compact WeatherNext 2-mini. The company stressed that official weather warnings should come from national or local meteorological agencies.
