Google DeepMind Open-Sources WeatherNext AI for Cyclone Forecasting

According to the paper published in Nature, the prediction accuracy for three days ahead has reached the level that previous models achieved for two days ahead. This is reported to be a level of progress equivalent to approximately 10 years of advancement in meteorology.

During the 2025 hurricane season, the National Hurricane Center (NHC) in the United States utilized this model. It is reported that the NHC predicted the rapid intensification and landfall of Hurricane Melissa in Jamaica in advance and issued warnings, securing critical lead time for on-site preparations.

The company stated that it has resolved the trade-off between path and intensity in cyclone forecasting. A single AI model simultaneously predicts atmospheric flow and local thermodynamic processes with high precision. In evaluations targeting historical data from 2023 to 2024, an improvement in lead time of 24 hours or more was confirmed.

Along with this announcement, "WeatherNext 2" and "WeatherNext Cyclones" have been open-sourced. The goal is to provide tools to the research community and regional forecasters, as well as to support renewable energy. The system generates 1,000 scenarios for each cyclone to provide probability maps.