Nvidia has unveiled a new suite of advanced AI weather forecasting models, collectively known as Earth-2, promising to deliver significantly faster and more accurate predictions. The announcement, made at the American Meteorological Society meeting in Houston, comes as a major winter storm impacts the U.S., highlighting the critical need for improved meteorological tools. These innovative AI models aim to revolutionize weather prediction, making it more precise and accessible globally.
At the forefront of this release is the Earth-2 Medium Range model, which Nvidia asserts outperforms Google DeepMind's GenCast across more than 70 variables. GenCast, released by Google in December 2024, had already set a new benchmark by offering significantly more accurate forecasts up to 15 days out compared to previous systems. Nvidia's latest development signals a rapid advancement in AI-driven weather prediction capabilities.
"Philosophically, scientifically, it's a return to simplicity," said Mike Pritchard, director of climate simulation at Nvidia, speaking to reporters ahead of the meeting. "We're moving away from hand-tailored niche AI architectures and leaning into the future of simple, scalable, transformer architectures."
Unlike traditional weather forecasting, which heavily relies on physics simulations, these new AI models represent a significant shift. The Earth-2 Medium Range model, specifically, is built upon Nvidia's new Atlas architecture, with further details expected soon.
Beyond the Medium Range model, Nvidia's comprehensive Earth-2 suite also introduces two other crucial tools: the Nowcasting model and the Global Data Assimilation model.
Nowcasting for Immediate Impact
The Nowcasting model specializes in ultra-short-term predictions, ranging from zero to six hours. Its primary goal is to assist meteorologists in forecasting the immediate impacts of severe weather and storms. Pritchard emphasized its global applicability:
"Because this model is trained directly on globally available geostationary satellite observations, rather than region-specific physics model outputs, Nowcasting's approach can be adapted anywhere on the planet with good satellite coverage."
This capability is particularly beneficial for governments and smaller nations, enabling them to better anticipate and prepare for localized severe weather events.
Global Data Assimilation: Efficiency and Accuracy
The Global Data Assimilation model plays a foundational role by compiling continuous snapshots of weather conditions from thousands of global locations, utilizing data from diverse sources such as weather stations and balloons. These snapshots serve as essential starting points for subsequent weather predictions. Historically, generating these initial data sets consumed significant computing resources, often accounting for half of traditional supercomputing loads for weather forecasting. Pritchard highlighted the efficiency gains:
"This model can do that in minutes on GPUs instead of hours on supercomputers."
This drastic reduction in processing time makes advanced forecasting more accessible.
These three new additions complement Nvidia's existing AI weather models, CorrDiff and FourCastNet3. CorrDiff is designed for generating rapid, high-resolution predictions from coarse-grained forecasts, while FourCastNet3 focuses on modeling individual weather variables such as temperature, wind, and humidity.
Democratizing Advanced Weather Insights
A key benefit of Nvidia's Earth-2 suite is its potential to democratize access to sophisticated weather forecasting tools. Historically, such capabilities were largely restricted to affluent nations and major corporations due to the prohibitive cost of supercomputing time. Pritchard explained:
"This provides the fundamental building blocks used by everyone in the ecosystem – national meteorological services, financial service firms, energy companies – anyone who wants to build and refine weather forecasting models."
Some of these tools are already seeing practical application; meteorologists in Israel and Taiwan are utilizing Earth-2 CorrDiff, while The Weather Company and Total Energies are actively evaluating the Nowcasting model. Pritchard underscored the strategic importance of these advancements, particularly for national sovereignty:
"For some users, it makes sense to subscribe to an enterprise centralized weather forecasting system. But for others like countries, sovereignty matters. Weather is a national security issue, and sovereignty and weather are inseparable."








