AI model achieves breakthrough in forecasting cyclones — Google DeepMind
Google DeepMind and partners open-source WeatherNext AI models, delivering a one-day improvement in cyclone forecasting accuracy, equivalent to a decade of progress, to aid disaster preparedness.
The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.
Google DeepMind, in collaboration with Google Research and meteorological agencies including the National Hurricane Center and UK Met Office, today announced WeatherNext, an AI model that improves cyclone forecasting accuracy by an average of one full day. The model predicts cyclone track, intensity, and wind structure with state-of-the-art precision, enabling earlier warnings for communities at risk. This advancement addresses a longstanding challenge where timely, accurate forecasts can significantly reduce loss of life and economic damage from tropical cyclones.
The WeatherNext Cyclones model was co-trained on nearly 20 terabytes of global atmospheric data and historical cyclone observations from the IBTrACS database, covering nearly 5,000 storms. It uses Functional Generative Networks to generate 1,000 forecast scenarios per cyclone, capturing rare but critical events such as rapid intensification. The model operates efficiently at a resolution of 28x28km, 100 times coarser than traditional high-resolution models, yet maintains high accuracy, surprising researchers and prompting further investigation into its predictive mechanisms.
During the 2025 hurricane season, WeatherNext Cyclones contributed to a historic forecast for Hurricane Melissa, predicting its rapid intensification and landfall in Jamaica. This enabled the National Hurricane Center to issue advance warnings, providing critical preparation time for emergency teams. The model’s ensemble approach allows forecasters to assess the probability of extreme scenarios, improving decision-making for disaster response and resource allocation.
Google DeepMind is open-sourcing the WeatherNext 2, WeatherNext Cyclones, and WeatherNext 2-mini models to accelerate research and operational forecasting worldwide. The models are available for academic use, operational deployment, and localized adaptation. Forecasts can be explored via Google’s Weather Lab, which now includes global weather predictions alongside cyclone tracks, offering a unified view of temperature, precipitation, and wind speed data.