Google DeepMind's WeatherNext AI Gives Forecasters an Extra Day to Prepare for Cyclones

Google DeepMind has open-sourced WeatherNext, an AI model that predicts cyclone tracks and intensity with a single 15-day forecast — generated in under a minute on a single TPU.

Google DeepMind and Google Research have unveiled WeatherNext 2, an AI weather prediction model that can forecast cyclone tracks and intensity up to 15 days out — giving meteorologists and emergency planners roughly an extra day of warning time compared to traditional forecasting methods.

The model is a single unified system, trained on nearly 20 terabytes of global atmospheric data and historical cyclone records from the International Best Track Archive for Climate Stewardship. Previously, predicting a storm's path required coarse global models, while intensity demanded separate, specialized local models. WeatherNext does both at once.

A 15-day forecast runs in under a minute on a single Google TPU, allowing forecasters to rapidly explore probability distributions and tail-risk scenarios that were computationally impractical before.

The research, published in Nature, was developed in collaboration with the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, the UK Met Office, and other weather agencies. Both the model weights and code have been released as open source on GitHub, making the technology freely available to meteorological agencies worldwide.

The timing is significant. As ocean temperatures continue rising, tropical cyclones are intensifying faster and behaving less predictably. An extra day of reliable warning can mean the difference between a successful evacuation and a humanitarian disaster.

Sources: Google Blog, Engadget, Nature