NVIDIA's open family of AI weather and climate models, covering forecasting, nowcasting, and high-resolution downscaling.
Earth-2 is NVIDIA's set of open models, libraries, and frameworks for weather and climate simulation rather than a single app. Its pieces include the Atlas architecture for medium-range forecasts of 70+ variables up to 15 days out, nowcasting models that generate short-term satellite and radar imagery for 0-6 hour severe weather prediction, a data-assimilation component that reconstructs atmospheric state on GPUs in seconds, and CorrDiff, a generative model that sharpens forecast resolution far faster than traditional downscaling methods. Users can download the Earth2Studio framework from GitHub and Hugging Face to fine-tune and run these models on their own infrastructure, or call specific models like CorrDiff through NVIDIA's NIM microservices. It is aimed at meteorologists, climate researchers, and companies building weather products, not a consumer weather app.
Core models and the Earth2Studio framework are open source and free to self-host; hosted access through NVIDIA's NIM microservices follows NVIDIA's standard cloud/API consumption pricing, which is not disclosed on the product page.
Use tool βEarth-2 is a research-grade toolkit for weather and climate AI, not a polished self-serve product - its value is giving people building weather products access to models like CorrDiff and Atlas without training them from scratch. It requires real infrastructure and domain expertise to put to use; there is no simple sign-up-and-go path for a non-technical user.
No reviews yet β be the first to review NVIDIA Earth-2.
Reviews are tied to your EffectHub account β one review per tool, so the rating for NVIDIA Earth-2 reflects real users.
No questions yet β ask the first one about NVIDIA Earth-2.
Questions about NVIDIA Earth-2 are tied to your EffectHub account, so answers can reach you and the section stays free of spam.