Google Open-Sources AI Model That Warns of Hurricanes
Google/DeepMind
Google DeepMind and Google Research have developed WeatherNext, an AI model that predicts cyclone dynamics. The company has released the model weights and source code on GitHub, and published research in Nature. The model can forecast both the track and intensity of cyclones, generating a 15-day forecast in under a minute on a TPU.
Researchers from Google DeepMind and Google Research, with participation from US and UK agencies, trained an AI model called WeatherNext that provides accurate forecasts of cyclone dynamics. Google has published the model weights and source code on GitHub, and the research paper is published in Nature. Tropical cyclones (hurricanes and typhoons) pose a unique challenge because atmospheric flows are traditionally analyzed with coarse global models, while storm intensity is better predicted by specialized local models that assess thermodynamic processes in the core. WeatherNext was trained on over 20 terabytes of global atmospheric data and historical information, allowing it to predict both the track and intensity of a cyclone with a single model. The researchers state that it can generate a 15-day forecast in less than a minute on Google's TPU, enabling forecasters to quickly assess probability distributions for potentially destructive threats.
- Abbreviations
- TPU = Tensor Processing Unit — тензорный процессор
Source: 3DNews —
original
