Nvidia’s new AI weather models may have seen this storm coming weeks ago

During the current storm in the US, the weather forecast for some areas was all over the map, and the snow forecast was very different.

Nvidia couldn’t time the release of its new Earth-2 weather forecasting models. Or, based on how accurate the company claims the new models are, maybe they knew something we didn’t?

New AI models promise to make weather forecasting faster and more accurate. Nvidia says one model in particular, Earth-2 Medium Range, beats Google DeepMind’s AI model, GenCast, by more than 70 different models. which Google released in December 2024itself was more accurate than existing weather models that could produce forecasts up to 15 days.

Nvidia announced the new hardware on Monday at the American Meteorological Society conference in Houston.

“Intellectually, scientifically, it’s a return to simplicity,” Mike Pritchard, head of climate testing at Nvidia, told reporters at a press conference ahead of the conference. “We’re moving away from hand-crafted niche AI ​​and toward a future of simple, scalable, scalable architectures.”

Traditionally, most weather forecasts rely on physics simulations as they occur in the real world. AI models are a recent addition. The Earth-2 Medium Range model is based on Nvidia’s new Atlas design, which the company said it will release details on Monday.

On the Medium Range side, Nvidia’s Earth-2 suite includes a Nowcasting model and a Global Data Assimilation model.

Nowcasting produces short-term forecasts from zero to six hours ahead, and aims to help meteorologists predict the impact of hurricanes and other severe weather.

“Because the model is trained directly on global geostationary satellite observations, rather than based on physics, the Nowcasting method can be adapted anywhere in the world with a good satellite,” said Pritchard. This should help national and local governments understand how severe weather can affect their communities.

The Global Data Assimilation model uses data from sources such as weather stations and balloons to create a continuous picture of the weather for thousands of locations around the world. These images are used as a starting point for weather forecasting.

Traditionally, these simulations require a lot of computing power before the prediction process begins. “It eats up about 50% of the normal climate (forecast),” Pritchard said. “This model can do this in minutes on GPUs instead of hours on supercomputers.”

The three new models join two existing models: CorrDiff, which uses visual forecasts to produce faster, more accurate forecasts, and FourCastNet3, which features weather variables such as temperature, wind, and humidity.

Pritchard said the new models should give more users access to powerful weather forecasting tools, which in the past have been the domain of rich countries and large corporations, who have the money to pay for expensive weather forecasting.

“This provides the building blocks that are used by everyone in the environment – global weather services, finance companies, energy companies – anyone who wants to make and improve weather forecasts,” Pritchard said. Some tools are already in use. Meteorologists in Israel and Taiwan have been using Earth-2 CorrDiff, for example, while The Weather Company and Total Energies are evaluating Nowcasting, Nvidia said.

“For some users, it makes sense to register weather forecasts between companies. But for others, like countries, freedom is important,” Pritchard said. “Weather is a matter of national security, and governance and weather cannot be separated.”