Google Open-Sources WeatherNext AI, Boosting Cyclone Forecast Lead by a Day
Google has released the code and weights for its WeatherNext AI model, which gives forecasters an extra day of lead time on cyclones and predicted Hurricane Melissa's path five days ahead with 80% confidence.
This article was drafted with AI assistance from multiple sources and was reviewed and approved by a human editor before publication.
Google has made the code and model weights of its WeatherNext AI available on GitHub, opening the forecasting tool to researchers and agencies worldwide. The release follows a study published in the journal Nature on Thursday, detailing the model's architecture and performance.
WeatherNext is designed to predict both the track and intensity of cyclones using a single model, a departure from traditional systems that often rely on separate components. The model was trained on nearly 20 terabytes of global atmospheric data, supplemented by historical records from the International Best Track Archive for Climate Stewardship.
According to Google, WeatherNext can generate a complete 15-day forecast in under a minute on a TPU, making it significantly faster than conventional numerical weather prediction models. This speed, combined with its accuracy, gives forecasters on average one day more lead time than existing models; predictions three days out are as accurate as previous models' two-day forecasts.
The model's capabilities were demonstrated in October 2025, when it predicted that Hurricane Melissa would strike Jamaica as a Category 5 storm with 80% confidence five days before landfall. The hurricane subsequently caused flooding and landslides across the island, underscoring the potential value of such early warnings.
WeatherNext is the second generation of Google's weather AI, following the introduction of the first version last year. Its development involved contributions from the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, the UK Met Office, and other weather agencies, reflecting a collaborative effort to improve tropical cyclone forecasting.
By open-sourcing the code and weights, Google aims to enable broader adoption and further refinement of the model. Researchers and meteorological services can now run their own forecasts, potentially improving preparedness for extreme weather events. The Nature paper provides detailed technical documentation, while the GitHub repository offers the tools needed to deploy WeatherNext in different operational settings.
The release marks a significant step in applying artificial intelligence to meteorology, with the potential to save lives and reduce economic losses from cyclones. As climate change intensifies storms, tools like WeatherNext could become increasingly vital for early warning systems worldwide.