Axios Pro Exclusive Content

Nvidia powers up AI weather forecasting

A simulation of weather in Boston on the Weatherverse Sim by The Weather Company

A simulation of weather in Boston visualized within its Weatherverse Sim by The Weather Company. Image courtesy of the Weather Company

AI chip juggernaut Nvidia is looking to make weather forecasting more timely and more local using a new simulation and generative AI.

Why it matters: Better weather forecasting can help predict extreme weather events with more accuracy, helping people, power grids, and a range of industries.

Driving the news: Nvidia CEO Jensen Huang addressed the company's weather ambitions at its developer conference in San Jose, Calif.

  • During his 2-hour keynote from the city's massive SAP Center, Huang spent time announcing a new simulated Earth model, or a "digital twin," called Earth-2 as well as a generative AI model called CorrDiff.
  • Nvidia says CorrDiff can create higher-resolution images more quickly and more efficiently and can synthesize metrics.

Zoom in: Taiwan's Central Weather Administration plans to use the simulation and model to more accurately forecast where typhoons will reach landfall.

  • Companies including The Weather Company, Meteomatics and Spire are working with Nvidia to build hyper-local weather forecasting systems using the tech.
  • The Weather Company CEO Sheri Bachstein told Axios that using Nvidia's tech, it plans to build the first forecasting model that can deliver predictions down to the mile range and on demand every 10 to 15 minutes.

Catch up quick: Weather AI models are trained on historical data to learn complex systems and generate projections.

  • Before AI emerged, the traditional approach is to build numerical models that use physics equations and weather observations to produce simulations of future conditions.
  • AI is changing the practice of weather forecasting by leveraging a lot more data and doing it much more quickly than the traditional physics-based methods.
  • Industries that commonly use weather forecasts include solar and wind generation, utilities, aviation and agriculture.

Yes, but: There's some skepticism in the weather and climate communities around AI-driven weather predictions, Generate's Andrew Freedman writes.

  • Researchers point out that AI-driven weather predictions lack transparency and also question whether training models using past weather patterns will be accurate enough with record-breaking warming.
  • Physics models have also steadily become more accurate.

Zoom out: Nvidia has provided its tech for weather forecasting for years, but as its chips become more powerful, it can offer faster processing of more data for its customers.

  • Other companies that offer AI models used for weather forecasting include Google, IBM, Tomorrow.io, and government agencies NASA and NOAA.

The big picture: AI models and simulations will transform weather forecasting.

  • As chips and computing become more powerful, companies will be able to build systems that deliver more accurate, faster and better predictions at a lower cost and with more efficiency.
Go deeper