Name: NVIDIA Earth-2: The Future of AI Weather Forecasting Is Open
Uploaded: 2026-02-07T21:01:20.000Z
Duration: 2 min 20 s
Description: Thousands of YouTube videos with English-Chinese subtitles! Now you can learn to understand native speakers, expand your vocabulary, and improve your pronunciation...

Weather has a profound impact on our daily lives.

Predicting how it will affect agriculture, energy,

aviation, and emergency response is critical.

But the atmosphere is chaotic, and forecasting is complex.

It accelerates the AI revolution in forecasting

by offering developers and researchers an open set of tools that lowers the barrier to AI-enabled weather simulation.

Building on the impact of models like ForeCastNet and CoreDiff,

NVIDIA Earth-2 is now a complete portfolio of open models.

Introducing Earth-2 Global Data Assimilation,

To initialize predictions, Earth-2 Global Data Assimilation models turn observations from diverse sources

— like balloons, satellites, and stations — into atmospheric states in seconds,

a task that previously took hours on high-performance computing clusters.

Earth-2 Nowcasting models offer severe weather prediction at continental scale from zero to six hours.

They represent the first time AI has surpassed the accuracy of physics-based forecasting of storm evolution.

Earth-2 Medium Range models deliver 15-day global forecasts with accuracy that outperforms other open models on standard benchmarks across more than 70 variables.

Earth-2’s models span the complete forecasting pipeline.

When they’re combined, weather can be simulated at speeds previously impossible,

And because Earth-2 is completely open — not just model weights, but inference tools and training frameworks too —

developers can customize and fine-tune simulations to specific needs,

using their own data and infrastructure to build sovereign weather and climate predictions with full control.

See into the future of weather with NVIDIA Earth-2.