NVIDIA is excited to announce three new open-source models as part of the NVIDIA Earth-2 family, making it easier than ever to build weather forecasting capabilities across the weather stack, including tasks such as data assimilation, forecasting, nowcasting, downscaling and more. In addition, developers can quickly get started building weather and climate simulations by using NVIDIA open source software: Earth2Studio for creating inference pipelines and Physics Nemo for training models.
NVIDIA Earth-2 comprises a set of accelerated tools and models which enables developers to bring together typically disparate weather and climate AI capabilities. Because Earth-2 is completely open, developers can customize and fine-tune their simulations to their specific needs, using their own data and their own infrastructure to build sovereign weather and climate predictions they fully own and control. Earth-2:
- Is a suite of leading open weather and climate models
- Is easy-to-use thanks to an ecosystem of open source software
- Enables you to create your own sovereign capabilities
Earth-2 Nowcasting: Kilometer-Scale Severe Weather Prediction
Out now on Hugging Face: Earth-2 Nowcasting, powered by a new model architecture called StormScope, using generative AI to make country-scale forecasts into kilometer‑resolution, zero- to six-hour predictions of local storms and hazardous weather in just minutes. Earth-2 Nowcasting can generate the first predictions that outperform traditional, physics-based weather-prediction models on short-term precipitation forecasting by simulating storm dynamics directly. It harnesses AI to directly predict satellite and radar data.
This version is trained directly on globally available geostationary satellite observations (GOES) over the contiguous US (CONUS). However, this method could be applied to train versions of the model over other regions with similar satellite coverage.
Research Paper: Learning Accurate Storm-Scale Evolution from Observations
Earth-2 Medium Range: Highly accurate 15-Day Global Forecasts
Out now on Hugging Face: Earth-2 Medium Range, powered by a new model architecture called Atlas, enabling high-accuracy weather prediction for medium-range forecasts — or forecasts of up to 15 days in advance — across 70+ weather variables including temperature, pressure, wind and humidity. It uses a latent diffusion transformer architecture to predict incremental changes in the atmosphere so as to preserve critical atmospheric structures and reduce forecasting errors. On standard benchmarks, it outperforms leading open models such as GenCast on the most common forecasting variables measured by the industry.
Research Paper: Demystifying Data-Driven Probabilistic Medium-Range Weather Forecasting
Earth-2 Global Data Assimilation: An End-to-End AI Pipeline
Coming soon to Hugging Face: Earth-2 Global Data Assimilation, powered by a new model architecture called HealDA, which produces initial conditions for weather prediction — snapshots of the current atmosphere, including the temperature, wind speed, humidity and air pressure, at thousands of locations around the globe. Earth-2 Global Data Assimilation can generate initial conditions in seconds on GPUs instead of hours on supercomputers. When coupled with Earth-2 Medium Range, this results in the most skillful forecasting predictions produced by an open, entirely AI pipeline.
These models join established open NVIDIA weather and climate models such as FourcastNet3, CorrDiff, cBottle, DLESym and more.
NVIDIA Earth2Studio is an open-source Python ecosystem for quickly creating powerful AI weather and climate simulations. It provides all the necessary inference tools to get started with the new model checkpoints on Hugging Face. It’s as easy as:
Hugging Face Package for Earth-2 Nowcasting
Research Paper: Learning Accurate Storm-Scale Evolution from Observations
Hugging Face Package for Earth-2 Medium-Range
Research Paper: Demystifying Data-Driven Probabilistic Medium-Range Weather Forecasting