Introduction
As climate change intensifies the frequency and severity of extreme weather events, the need for precise, fast, and accessible meteorological forecasting has become critical. NVIDIA has responded to this challenge by launching the Earth-2 family of open models—the world’s first fully open, accelerated set of models and tools for AI weather intelligence.
NVIDIA Earth-2 makes professional-grade weather AI accessible worldwide at every stage, from processing initial observation data to generating long-term global forecasts. By moving from closed, compute-heavy traditional models to open, GPU-accelerated AI pipelines, NVIDIA is empowering an entirely new ecosystem of climate-tech innovators.
The Earth-2 Open Model Family
The latest announcement introduces three groundbreaking model architectures that cover the entire spectrum of weather prediction:
1. Earth-2 Medium Range (Atlas)
Powered by the Atlas architecture, this model enables high-accuracy weather prediction for forecasts up to 15 days in advance. It tracks over 70 weather variables, including temperature, pressure, wind, and humidity. Atlas has already shown it can outperform leading open models on standard industry benchmarks.
2. Earth-2 Nowcasting (StormScope)
The StormScope architecture utilizes generative AI to provide kilometer-resolution predictions of local storms and hazardous weather. While traditional models struggle with the computational intensity of short-term precipitation forecasting, StormScope can generate these "nowcasts" for zero to six hours in just minutes, simulating storm dynamics directly from satellite and radar imagery.
3. Earth-2 Global Data Assimilation (HealDA)
Accurate forecasts depend on accurate "initial conditions"—a snapshot of the current state of the atmosphere. The HealDA architecture (Global Data Assimilation) can generate these snapshots in seconds on GPUs. Traditionally, this process required hours on the world's most powerful supercomputers. When paired with Atlas, HealDA completes an entirely AI-driven forecasting pipeline.
Accelerating Weather Intelligence
These new models join an existing stack of powerful tools designed to bring unprecedented speed and resolution to weather data:
- Earth-2 CorrDiff: A generative AI architecture used to downscale coarse continental-scale predictions into high-resolution regional weather fields. It provides local forecasting up to 500x faster than classic numerical methods.
- Earth-2 FourCastNet3: Delivers high accuracy for variables like wind and temperature, rivaling top diffusion-based methods while operating up to 60x faster.
- PhysicsNeMo: An open-source Python framework that allows researchers to train and fine-tune AI-physics models at scale.
Industry Adoption and Real-World Impact
The "open" nature of the Earth-2 family is already accelerating innovation across various sectors:
- Weather Forecasting: Startups like Brightband are already running Earth-2 Medium Range operationally to issue daily global forecasts.
- Government Agencies: The Israel Meteorological Service is using CorrDiff to reduce compute time by 90% while improving the accuracy of precipitation verification.
- Energy Operations: Global energy giants like TotalEnergies and Eni are testing Earth-2 models to improve short-term risk awareness for energy systems and gas demand forecasting.
- Financial and Insurance: AXA is using FourCastNet to simulate thousands of hypothetical hurricane scenarios to better assess risk and develop new benchmarking techniques.
Conclusion
NVIDIA's Earth-2 represents a paradigm shift in how we understand and prepare for the Earth's changing climate. By providing an open ecosystem of high-performance models, NVIDIA is not just releasing code; they are fostering a global community of weather intelligence.
Whether it's a farmer needing to protect crops from a sudden storm or a grid operator balancing renewable energy sources, the speed and accuracy of Earth-2 will save lives and protect environments. The future of weather forecasting isn't just in the clouds—it's in the silicon that helps us predict them.