Nvidia unveils AI models for faster, cheaper weather forecasts

By Stephen Nellis SAN FRANCISCO, Jan 26 (Reuters) – Nvidia on Monday released three open-source artificial intelligence models aimed at helping create better weather forecasts, faster. The models, which the AI chip firm announced at the American Meteorological Society’s annual meeting in Houston, are part of a broader push by the company to provide open-source software, powered by its chips, for everything from chatbots to self-driving vehicles. In the case of weather forecasting, Nvidia is aiming to replace expensive and time-consuming conventional weather simulations with AI-driven versions that the company said can rival or exceed the accuracy of older methods. The AI models, once trained, are also faster and cost less to run. Mike Pritchard, the director of climate simulation research for Nvidia and a professor of earth system sciences at the University of California, Irvine, said that one of the practical business applications of the new weather models will be in the insurance industry. Insurance companies often want to understand extreme outlier events, such as massive floods or hurricanes. But predicting such events in detail has historically been expensive, because weather forecasting is performed in "ensembles," or groups of individual "member" predictions about how a weather event might play out from a given starting point. To find possible outlier events, the ensembles must contain many members, but calculating each one in precise detail to see whether a particular property might flood is slow. "The tension is gone, because once trained, AI is 1,000 times faster," Pritchard said in an interview. "So you're free to run massive ensembles. And insurance companies are running like 10,000-member ensembles." Nvidia's "Earth-2" models introduced on Monday include one aimed at making 15-day weather forecasts, one that specializes in forecasts of up to six hours for severe storms over the U.S., and one that can be used to integrate disparate data streams from a variety of weather sensors to make them a more useful starting point for other forecasting technology.  (Reporting by Stephen Nellis in San Francisco; Editing by Ethan Smith)

(The article has been published through a syndicated feed. Except for the headline, the content has been published verbatim. Liability lies with original publisher.)