Deepmind's WeatherNext AI Model Revolutionizes Cyclone Forecasting

Google Deepmind's WeatherNext (WN-C) is an AI-powered weather model that predicts cyclone tracks and intensity simultaneously. Developed in collaboration with the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, and the UK Met Office, WN-C has been running live on Google's Weather Lab since June 2025.

According to Deepmind, WN-C solves a decades-old tradeoff in cyclone forecasting by handling both track prediction and intensity in a single system. This is achieved through the use of Functional Generative Networks (FGN), which are eight times faster than the diffusion method used in GenCast.

The model has shown impressive results, with an estimated storm center position error of 230 kilometers for a five-day forecast, compared to 370 kilometers for ECMWF's ensemble system. WN-C also outperforms specialized regional models like NOAA's Hurricane Analysis and Forecast System (HAFS) in terms of intensity forecasts.

Deepmind has made the code and weights for WeatherNext Cyclones publicly available on GitHub, releasing a compact variant that runs on a single TPU and can be used in a free Colab notebook. The model is intended to support forecasters, not replace them, and traditional numerical models are still considered important for intensity forecasts.

The development of WN-C comes at a time when Google's track record with forecasting has been questioned, particularly after its Android Earthquake Alerts system severely underestimated the 2023 Turkey earthquakes.