Google DeepMind Releases Open-Source AI Model That Predicts Tropical Cyclones a Day Earlier Than Current Methods

Aug 6, 2026

Google DeepMind and Google Research have developed and open-sourced WeatherNext, an AI weather model that predicts tropical cyclone tracks, intensity, and wind structure with state-of-the-art accuracy. Published in Nature on August 6, 2026, the model provides an average of 24 hours of extra lead time, equivalent to a decade of meteorological progress. WeatherNext demonstrated its capabilities during the 2025 hurricane season by predicting Hurricane Melissa's rapid intensification and landfall in Jamaica five days in advance.

WeatherNext cyclone forecasting breakthrough

▪WeatherNext was co-developed in collaboration with the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, and the UK Met Office.
▪WeatherNext achieved state-of-the-art accuracy in predicting tropical cyclones, representing roughly a decade of meteorological progress in a single model.
▪Google DeepMind and Google Research developed WeatherNext, an AI model that predicts a tropical cyclone's track, intensity, and wind structure.

AI model training methodology

▪WeatherNext generates up to 1,000-member ensembles to capture rare but consequential scenarios like rapid intensification events.
▪WeatherNext uses Functional Generative Networks to efficiently produce ensembles of different predictions to capture weather uncertainty.
▪WeatherNext operates on data with a resolution of 28x28km, which is 100 times coarser than traditional regional weather models.
▪WeatherNext was co-trained on nearly 20 terabytes of global atmospheric data and the historical IBTrACS database spanning nearly 5,000 historical storms.

Hurricane Melissa prediction success

▪The prediction of Hurricane Melissa marked the first time an AI weather model successfully identified a Category 5 hurricane from a weak initial tropical depression.
▪WeatherNext predicted Hurricane Melissa's landfall in Jamaica with 80% confidence five days in advance, which rose to nearly 100% three days before impact.
▪During the 2025 hurricane season, WeatherNext predicted Hurricane Melissa's rapid intensification and landfall in Jamaica five days in advance.

Extra day lead time

▪WeatherNext gives forecasters an average of an extra day of predictive lead time for cyclone tracks, intensity, and wind structure.
▪WeatherNext's three-day forecasts achieve the same level of accuracy that prior models could only provide for a two-day forecast.

Open-source model release

▪Google DeepMind released WeatherNext 2-mini, a compact version of the model designed to run on a single TPU in a free public Colab notebook.
▪Google DeepMind open-sourced the code and model weights for WeatherNext 2 and WeatherNext Cyclones on GitHub.

Tropical cyclone forecasting challenges

▪Tropical cyclones have caused more than 700,000 deaths and $1.4 trillion in economic losses globally over the past 50 years.
▪Traditional cyclone forecasting requires a trade-off between coarser global models for storm tracks and specialized, high-resolution local models for storm intensity.