AI Weather Model Predicts Cyclones Up To A Day Earlier

Google DeepMind has developed an AI weather model called WeatherNext that can predict cyclones more than a day earlier than traditional forecasting systems. The model, which was published in a paper in Nature, achieved state-of-the-art accuracy in predicting a cyclone's track, intensity, and wind structure.

The researchers trained the model on nearly 20 terabytes of atmospheric data and historical records from nearly 5,000 storms. One surprising finding is that WeatherNext performs well even with lower-resolution weather data, which is roughly 100 times coarser than traditional high-resolution models used for intensity forecasting.

The model's biggest real-world test came with Hurricane Melissa in October 2025, where it predicted with 80% confidence that the storm would explode into a Category 5 and strike Jamaica five days before landfall. However, experts note that WeatherNext is not a crystal ball and its skill at predicting rapid intensification remains limited.

Google DeepMind released the WeatherNext model as open-source software, making it available to researchers, weather agencies, and nonprofits. The company also notes that the model's real value lies in buying time for forecasters, giving them about 24 additional hours to prepare for dangerous storms.