Nvidia’s Advanced AI Weather Models May Have Anticipated Recent U.S. Storm
As the U.S. is currently grappling with a significant winter storm, Nvidia’s newly unveiled Earth-2 weather forecasting models appear to have perfectly timed their release. These AI-driven models are designed to enhance the speed and accuracy of weather predictions, potentially offering insights into evolving weather patterns ahead of time.
At a recent gathering of meteorological experts in Houston, Nvidia introduced its Earth-2 Medium Range model, which the company claims surpasses Google DeepMind’s GenCast in accuracy across more than 70 variables. Released in December 2024, GenCast marked a significant leap in predictive capabilities, covering forecasts up to 15 days in advance.
Mike Pritchard, Nvidia’s director of climate simulation, highlighted the simplicity and scalability of the new models, moving away from complex custom architectures. “We’re aiming for a return to foundational principles while leveraging cutting-edge technology,” Pritchard remarked in a pre-meeting briefing.
The Earth-2 suite consists of three primary models: Medium Range, Nowcasting, and Global Data Assimilation. The Nowcasting model, which forecasts weather conditions from zero to six hours in the future, is particularly aimed at assisting meteorologists in understanding the immediate impact of severe weather events. “This model, trained on globally available satellite data, can be applied anywhere with good satellite coverage,” Pritchard explained.
Meanwhile, the Global Data Assimilation model employs data from various sources, including weather stations and balloons, to generate continuous weather snapshots worldwide. Traditionally, such processes have been computationally intensive, consuming significant supercomputing resources. However, Nvidia’s new approach drastically reduces this time from hours to mere minutes using GPUs.
Additionally, Nvidia continues to support its existing models, CorrDiff and FourCastNet3, which specialize in producing high-resolution forecasts and modeling individual weather parameters like temperature and humidity.
These innovations aim to democratize access to powerful weather forecasting tools, which have largely benefited affluent nations and corporations. “Our goal is to provide basic weather forecasting frameworks accessible to national meteorological services, financial institutions, and energy companies alike,” stated Pritchard.
Some of these models are already in operation, with meteorologists in Israel and Taiwan employing Earth-2 CorDiff, and The Weather Company alongside Total Energies testing the Nowcasting model. Pritchard emphasized the importance of sovereignty in weather forecasting, recognizing that for many nations, the implications of weather prediction extend into national security realms.
