Google open-sources an AI model it says can help with earlier hurricane warnings

Google has opened-source an AI model called WeatherNext, capable of providing a 15-day forecast to predict hurricane tracks and intensity. This advanced model was developed by researchers from Google DeepMind, Google Research, and other global weather agencies including the National Hurricane Center, UK Met Office, and the Cooperative Institute for Research in the Atmosphere. The model's code and weights are now available on GitHub for wider scientific use.

Published in Nature and also detailed in a Google blog post, WeatherNext merges global atmospheric data with local storm intensity predictions to offer a comprehensive cyclone forecast. Traditionally, global atmospheric currents were analyzed using coarse global models, while cyclone intensity was assessed by specialized local models.

WeatherNext, trained on nearly 20 terabytes of data, can deliver this single 15-day forecast in under a minute on a Tensor Processing Unit (TPU), enabling forecasters to swiftly assess the probability distribution of potential catastrophic outcomes.

Google's WeatherNext model has been improved since its initial release last year, and the company has previously used AI to aid in predicting flash floods.