WeatherNext: AI model achieves breakthrough in forecasting cyclones

Google researchers announced a breakthrough in cyclone forecasting using an artificial intelligence model called WeatherNext that delivers more than a full day of additional lead time. The advance is detailed in a Nature paper published Thursday. The model uses Functional Generative Networks to generate ensembles of 1,000 predictions in less than a minute on a tensor processing unit. Researchers said the system captures rare but consequential scenarios such as rapid intensification events like Hurricane Melissa in 2025. The model operates at a resolution of 28 by 28 kilometers, which is 100 times coarser than traditional physics-based models. A smaller version called WeatherNext 2-mini runs at 111 by 111 kilometers and also shows strong performance. Google is open sourcing the code and model weights for WeatherNext Cyclones, WeatherNext 2 and WeatherNext 2-mini. The models are available for academic research, operational forecasting and specialized local applications. Forecasts can be explored on Weather Lab, a refreshed interface that now includes global weather predictions alongside cyclone tracks. Both Weather Lab and WeatherNext are part of Google Earth AI.