source: Google DeepMind: WeatherNext: AI model achieves breakthrough in forecasting cyclones
level: research
Google DeepMind published a Nature paper showing its WeatherNext AI model achieves state-of-the-art accuracy in predicting cyclone tracks, intensity, and wind structure. The model provides an extra day of lead time: its three-day forecasts match the accuracy of prior models at two days. This gain is roughly equivalent to a decade of meteorological progress. The model already aided the National Hurricane Center during the 2025 hurricane season, including a historic forecast for Hurricane Melissa.
WeatherNext Cyclones was evaluated on historical cyclones from 2023 to 2024, gaining more than 24 hours of lead time advantage over other top models. It was co-trained on nearly 20 terabytes of global atmospheric data and the IBTrACS database of nearly 5,000 storms. The model operates at a 28x28 km resolution, 100 times coarser than traditional models, yet still delivers strong intensity forecasts. A smaller version, WeatherNext 2-mini, runs at 111x111 km resolution and also performs well.
The model uses Functional Generative Networks to produce ensembles of predictions, capturing uncertainty. It can generate a 15-day forecast in under a minute on a TPU. During the 2025 season, the ensemble size was scaled to 1,000 members to capture rare rapid intensification events. Google DeepMind open-sourced the code and model weights for WeatherNext 2 and WeatherNext Cyclones, making them freely available for research and operational use. Forecasts can be explored on the updated Weather Lab platform.
why it matters: An extra day of accurate cyclone warning can give communities critical time to prepare, potentially reducing deaths and economic losses from extreme weather.
source: Google DeepMind: WeatherNext: AI model achieves breakthrough in forecasting cyclones