Google open-sources AI model designed for earlier hurricane warnings
What's the story
Google DeepMind and Google Research have developed an advanced weather prediction model, WeatherNext, to improve cyclone warnings. The National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, and the UK Met Office were among other global weather agencies that contributed to this project. Now, both the code and model weights of WeatherNext have been made open source on GitHub for other scientists to use.
Innovative methodology
WeatherNext uses nearly 20TB of global atmospheric data
Tropical cyclones are notoriously difficult to predict due to global atmospheric currents that determine a storm's path.
These have traditionally been best analyzed by coarse global models.
However, the intensity of a storm is best predicted by specialized local models assessing thermodynamic processes at its core.
WeatherNext uses nearly 20TB of global atmospheric data and historic information from the International Best Track Archive for Climate Stewardship to predict both a cyclone's track and intensity with one model.
Enhanced forecasting
Google launched the 2nd generation of WeatherNext last year
The WeatherNext researchers have claimed that their model can generate a single 15-day forecast in less than a minute on a Tensor Processing Unit (TPU).
This capability allows forecasters to quickly evaluate the probability distribution of potentially devastating tail risks.
Google launched the second generation of WeatherNext last year, furthering its commitment to using AI for improved weather predictions.