Overview
- Earth-2 Medium Range and Nowcasting are now available as open-source models via Earth2Studio, Hugging Face and GitHub, with the Global Data Assimilation model scheduled for release later this year.
- The Medium Range model, built on the Atlas architecture, provides 15-day global forecasts across more than 70 variables, while the StormScope-based Nowcasting model generates kilometer-scale, zero‑to‑six‑hour storm predictions from satellite and radar data.
- Nvidia says the trained models run up to 1,000 times faster than traditional physics-based simulations and cost less to operate, enabling very large forecast ensembles for scenarios like insurance risk analysis.
- The forthcoming HealDA data assimilation component is designed to create atmospheric initial conditions in seconds on GPUs instead of hours on supercomputers, removing a major computational bottleneck.
- Early users named by Nvidia include the Israel Meteorological Service, Taiwan’s Central Weather Administration, The Weather Company, the U.S. National Weather Service, and energy and finance firms, and the launch was announced at the American Meteorological Society meeting in Houston.