Google has open-sourced WeatherNext, an artificial intelligence weather model designed to improve hurricane warnings with 15-day forecasts of storm track and intensity.
Google DeepMind and Google Research helped train the model, with contributions from the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere, the UK Met Office and other weather agencies worldwide.
Google is releasing both the code and model weights for WeatherNext on GitHub, allowing other scientists to use the system.
Tropical cyclones remain difficult to predict because storm paths have traditionally been analyzed with coarser global models, while storm intensity has relied on specialized local models focused on thermodynamic processes in the cyclone core.
WeatherNext was trained on nearly 20 terabytes of global atmospheric data and historical information from the International Best Track Archive for Climate Stewardship to predict both storm track and intensity with a single model.
“We can now generate a single 15-day forecast in less than a minute on a TPU, empowering forecasters to quickly evaluate the probability distribution of potentially devastating tail-risks,” the WeatherNext researchers wrote.
Google introduced the second generation of WeatherNext last year. Google’s research teams have also worked on using AI to help predict flash floods.
Featured image credit
