NVIDIA Earth-2

The Atlas architecture delivers high-accuracy predictions for over 70 weather variables up to 15 days in advance, outperforming leading open models like GenCast on key industry benchmarks.

Using generative AI to predict satellite and radar imagery, Earth-2 Nowcasting delivers high-fidelity, zero- to six-hour hazardous weather forecasts in minutes. It’s the first AI model to outperform traditional physics-based systems in simulating storm dynamics.

The Earth-2 Global Data Assimilations model generates precise initial atmospheric conditions in seconds on GPUs rather than hours on supercomputers. When coupled with Earth-2 Medium Range, it delivers the most skillful forecasting predictions produced by an open, entirely AI pipeline.

CorrDiff enables generative AI downscaling 500x faster with a 10,000x improvement in energy efficiency. This augments current applications and workflows, allowing enterprises to generate more datasets for better probabilistic distributions of weather events​.​

FourCastNet 3 accelerates AI-based global weather forecasting. With it, enterprises can develop solutions using datasets that are up to 20x larger, capture extreme weather events while maintaining or boosting energy efficiency.

City-scale simulation data is now incorporated  into the Earth-2 planetary digital twin. In this video, we demonstrate the combination of high-resolution simulation data from ICON, WRF, and PALM with Google Photoreal Tiles served by Cesium using the Earth-2 visualization service to help address questions in the urban environment.

Researchers are using interactive visualization of AI weather forecasts, simulation data, and archived data to analyze extreme weather.

By accelerating analysis 700,000X, NVIDIA Omniverse and PhysicsNeMo can help engineers with planning and operating carbon capture and storage, ensuring safe operation and long-term storage and reducing the amount of carbon dioxide released into our atmosphere.

By running FourCastNet in NVIDIA PhysicsNeMo, we were able to generate 21-day weather trajectories of 1,000 ensemble members in a tenth of the time it previously took to do a single ensemble—and with 1,000X less energy consumption.

Built on NVIDIA Omniverse and the OpenUSD 3D framework, the Earth-2 platform enables aggregation and visualization of diverse, global-scale climate simulation and geospatial datasets. Made possible with cloud-native technology, visualizations can be explored by anyone around the globe.