- Nvidia Earth-2 accelerates weather forecasting and significantly reduces computational costs
- Earth-2 includes CorrDiff, FourCastNet3, Medium Range, Nowcasting, Global Data Assimilation and the PhysicsNeMo framework
- Energy companies rely on Earth-2 to improve grid reliability and photovoltaic predictions
Nvidia has unveiled its new Earth-2 family of open AI models that it says could transform weather forecasting and climate prediction as we know it.
The Nvidia Earth-2 family includes CorrDiff, FourCastNet3, Medium Range, Nowcasting, Global Data Assimilation, and the PhysicsNeMo framework for training and fine-tuning AI physics models.
These models integrate high-resolution data from satellites, radars and weather stations to provide continuous estimates of atmospheric conditions.
High-resolution modeling for fast forecasting
Earth-2 uses generative artificial intelligence to accelerate all stages of forecasting, from processing observational data to generating global and localized storm predictions.
CorrDiff uses a generative AI architecture to downscale coarse continental predictions to high-resolution regional forecasts, producing results up to 500 times faster than traditional methods.
FourCastNet3 delivers accurate forecasts of wind, temperature and humidity that outperform conventional ensemble models, while predictions are up to 60 times faster.
The system also integrates models from the European Center for Medium-Range Weather Forecasts, Microsoft and Google, enabling users to combine multiple approaches within a single framework.
Nvidia’s PhysicsNeMo allows AI physics models to be trained and fine-tuned at scale, providing flexibility for both operational forecasting and scientific research.
Earth-2 Global Data Assimilation produces initial atmospheric conditions in seconds on GPUs instead of hours on supercomputers, enabling faster integration into downstream models.
Organizations across the research, energy and government sectors are already using these AI tools to improve forecast accuracy and reduce computational costs.
The Israel Meteorological Service already uses CorrDiff and plans to deploy Nowcasting for high-resolution forecasts up to eight times daily.
Energy companies such as TotalEnergies, Eni and GCL are testing Earth-2 to improve grid operation, short-term risk awareness and photovoltaic forecasting.
Brightband and meteorologists in Taiwan use Earth-2 CorrDiff and Medium Range to provide accurate global and local forecasts, and The Weather Company is now evaluating Nowcasting for ultra-short-term local storm forecasting.
These AI tools reduce computational demand, with some models reporting a 90% reduction in computation time compared to classical methods on CPU clusters.
The open source availability of Earth-2 on platforms such as Hugging Face and GitHub allows researchers, companies and startups to fine-tune forecasts to local conditions.
By combining multiple models and AI tools, organizations can generate probabilistic and actionable insights that inform decisions in agriculture, energy, disaster preparedness and insurance risk assessment.
“Philosophically, scientifically, it’s a return to simplicity…We’re moving away from hand-tailored niche AI architectures and leaning into the future of simple, scalable transformer architectures,” said Mike Pritchard, director of climate simulation at Nvidia.
“This provides the basic building blocks used by everyone in the ecosystem — national meteorological services, financial services companies, energy companies — anyone who wants to build and refine weather forecasting models.”
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
