NVIDIA Earth-2
VerifiedOpen family of AI models and tools for weather forecasting and climate simulation
Product preview
NVIDIA Earth-2
What it is
NVIDIA Earth-2 is a collection of open models, libraries and frameworks for building AI-based weather and climate forecasting systems. The model family spans the full pipeline: global data assimilation to generate initial atmospheric conditions, medium-range forecasts for dozens of weather variables up to 15 days ahead, generative nowcasting of storms in the zero to six hour window, and CorrDiff for high-resolution downscaling, alongside FourCastNet 3. Developers use the Earth2Studio toolkit, published on GitHub, and model weights on Hugging Face to run, fine-tune and deploy forecasts on their own GPU infrastructure. It targets researchers, weather agencies, startups and industries such as energy and finance.
Key features
- Medium-range global forecasts up to 15 days
- Generative storm nowcasting
- CorrDiff high-resolution downscaling
- Earth2Studio toolkit for fine-tuning
What we liked
- + Open models and tooling
- + Covers the full forecasting pipeline
- + Runs on your own infrastructure
Worth noting
- · Requires GPU hardware and technical expertise
- · Not a ready-made consumer weather app
Editor's notes
More context from AI BirdView
Weather and climate forecasting has traditionally depended on expensive physics simulations run on supercomputers. NVIDIA Earth-2 offers an alternative path with an open, end-to-end set of AI models, so organizations can build their own forecasting systems instead of relying solely on outside providers. NVIDIA says its models compare well with other leading open weather models on key benchmarks.
- Generates initial atmospheric conditions in seconds on GPUs through global data assimilation
- Forecasts dozens of weather variables up to 15 days ahead with the medium-range model
- Predicts hazardous weather in the zero to six hour window using generative nowcasting
- Downscales coarse data to high resolution with CorrDiff for local detail
- Ships open weights on Hugging Face and the Earth2Studio toolkit on GitHub
NVIDIA Earth-2 is best for meteorologists, climate researchers, national weather services and companies in energy, insurance or agriculture with in-house data science teams. A practical tip: start with Earth2Studio's example workflows on a single GPU to reproduce a known forecast before fine-tuning any model on your own regional data.
Reader reviews
From people who've used NVIDIA Earth-2. Emails are never shown publicly.
No reviews yet. Be the first to share your experience.
Alternatives to NVIDIA Earth-2
Other tools we like in this space.
