With the winter storm currently hitting the US, weather forecasts for some areas are all over the map, with varying snow predictions.
Nvidia could not specify a release date for the improved Earth-2 weather forecast model. Or, accurately that the company claims a new model, maybe they know what we don’t?
New AI models promise to make weather forecasts faster and more accurate. Nvidia claims that one model in particular, Earth-2 Medium Range, beats Google DeepMind’s AI weather model, GenCast, in more than 70 variables. GenCast, which Google released in December 2024itself is more accurate than existing weather models that can produce forecasts of up to 15 days.
Nvidia announced the new tool Friday at the American Meteorological Society meeting in Houston.
“Philosophically, scientifically, it comes back to simplicity,” Mike Pritchard, director of climate simulation at Nvidia, told reporters on the phone before the meeting. “We’re moving away from niche AI architectures that are hand-designed and lean toward a future of simple, scalable, and scalable transformer architectures.”
Traditionally, most weather forecasts have relied on simulating physics as observed in the real world. AI models are a relatively recent addition. The Earth-2 Medium Range model is based on Nvidia’s new architecture called Atlas, about which the company will release more details on Monday.
In addition to Medium Range, Nvidia’s Earth-2 suite includes a Nowcasting model and a Global Data Assimilation model.
Techcrunch event
San Francisco
|
13-15 October 2026
Nowcasting produces short-term predictions from zero to six hours into the future, and aims to help meteorologists predict the impact of storms and other dangerous weather.
“Because the model is trained directly on globally available geostationary satellite observations, rather than region-specific physics model output, the Nowcasting approach can be adapted anywhere on the planet with good satellite coverage,” Pritchard said. This should help state governments and small states understand how severe weather systems could affect their regions.
The Global Data Assimilation Model uses data from sources like weather stations and balloons to produce a continuous snapshot of weather conditions at thousands of locations around the world. These images are then used as a launching point for weather models to make predictions.
Typically, these images require a significant amount of computing power before the forecasting project can begin. “It uses about 50% of the total load of a traditional weather (forecast) supercomputer,” Pritchard said. “This model can run in minutes on a GPU instead of hours on a supercomputer.”
Three new models join two existing models: CorrDiff, which uses rough forecasts to generate fast, high-resolution predictions, and FourCastNet3, which models individual weather variables like temperature, wind, and humidity.
Pritchard said the new model should give more users access to powerful weather forecasting tools, which have historically been the domain of wealthy countries and large corporations, which have the funds to pay for expensive supercomputer time.
“It provides the basic building blocks that are used by everyone in the ecosystem — the national meteorological service, financial services companies, energy companies — anyone who wants to build and refine weather forecasting models,” Pritchard said. Some devices have been used. Meteorologists in Israel and Taiwan are already 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 subscribe to the company’s centralized weather forecast system. But for others like the country, sovereignty matters,” said Pritchard. “Weather is a matter of national security, and sovereignty and weather are inseparable.”