NVIDIA Earth-2

NVIDIA Earth-2 provides a comprehensive family of open models, libraries, and frameworks, offering a fully open software stack to democratize access to professional-grade weather and climate AI.

Utilizes the Atlas architecture to deliver high-accuracy predictions for over 70 weather variables, providing forecasts up to 15 days in advance and outperforming leading open models.

Leverages generative AI to predict satellite and radar imagery, delivering high-fidelity, zero- to six-hour hazardous weather forecasts in minutes, and is the first AI model to outperform traditional physics-based systems in simulating storm dynamics.

Generates precise initial atmospheric conditions in seconds on GPUs, significantly faster than traditional supercomputers, and enhances forecasting skill when coupled with Earth-2 Medium Range.

Enables generative AI downscaling 500x faster with a 10,000x improvement in energy efficiency, augmenting current applications and workflows to generate more datasets for better probabilistic distributions of weather events.

Accelerates AI-based global weather forecasting, allowing enterprises to develop solutions using datasets up to 20x larger and capture extreme weather events while maintaining or boosting energy efficiency.

Provides a platform to build, fine-tune, and deploy Earth-2 open models on users' own infrastructure, facilitating rapid development of AI weather forecasts.