Nvidia Unveils Earth-2 Platform as Open AI Weather Forecasting Stack
This week, Nvidia showed off new Earth-2 features that make it easier for people to use AI to predict the weather and climate. Earth-2 is the first fully open accelerated weather AI software stack, according to the company. The platform has libraries and frameworks for models that support scientists, developers, businesses, and governments all over the world.
Nvidia said that Earth-2 makes advanced weather modeling easier to use without needing traditional supercomputing infrastructure. Users can speed up the processes of making forecasts and processing observations by using pretrained models. The system can handle everything from local analysis to global forecasting for 2 weeks.
Source: Nvidia Blog/Website
Earth-2 Shifts Forecasting Away From Traditional Supercomputer Reliance
In the past, weather forecasts were based on physics-based models that ran on powerful, centralized supercomputers. Nvidia said that using artificial intelligence to make predictions cuts down on both the time and money needed to run the system. This change makes it possible for more countries and businesses to set up their own custom forecasting systems.
Earth-2 makes advanced climate and weather intelligence tools available to more people by lowering the cost of hardware. Companies can run application-specific models without having to keep a lot of computing power on hand. Nvidia said that this method was necessary for making forecasting available to more people around the world.
Nvidia Releases 3 New Earth-2 Forecasting Models Publicly
Nvidia announced that 3 new models are now available, which will give users more options with the Earth-2 toolset. Earth-2 Medium Range, Earth-2 Nowcasting, and Earth-2 Global Data Assimilation are some of these. Each model has its own goals for forecasting and operational needs.
The new features work well with the current Earth-2 models CorrDiff and FourCastNet3. Together, they improve the accuracy of forecasts and the resolution of reports. Nvidia stressed that modular adoption should depend on the needs of the user and the infrastructure.
Recommended Article: China Prepares H200 AI Chip Purchases as Nvidia Visits Shanghai
Earth-2 Medium Range Delivers 15 Day High Accuracy Forecasts
Earth-2 Medium Range uses Nvidia Atlas architecture to make 15 day forecasts. The model can handle more than 70 weather variables, such as temperature, pressure, wind, and humidity. It aims for medium-range predictions with better accuracy in space and time.
This feature lets planners and researchers reliably look at changing weather patterns. Medium-range forecasting helps with energy logistics, agriculture planning, and making decisions about how to prepare for disasters. Nvidia pointed out that their method is more accurate than traditional forecasting methods.
Earth-2 Nowcasting Focuses on Local Storms and Hazardous Weather
StormScope, a generative AI model, is used by Earth-2 Nowcasting to make short-term predictions. It makes predictions about local storms and dangerous weather events that are accurate to within 1 kilometer. This helps with quick responses when minutes and local effects are important.
The model makes it easier for emergency management and energy systems to understand what’s going on. Short-term accuracy lowers operational risk and makes safety outcomes better. Nvidia said that nowcasting is important for environments where decisions have to be made in real time.
Global Data Assimilation Accelerates Atmospheric Condition Generation
Earth-2 Global Data Assimilation creates the initial conditions in the atmosphere that are needed for accurate forecasts. It is based on the HealDA model and uses GPUs instead of supercomputers. In just a few seconds, the system can take pictures of the whole world.
When used with other Earth-2 models, this speed boost makes forecasting faster and more accurate. Rapid assimilation makes it possible to update forecasts more often. Nvidia talked about how operational forecasting pipelines have become more efficient.
Early Adopters Span Government Energy Finance and Weather Services
Some of the first users were national weather agencies, startups, energy companies, and financial risk firms. Organizations are based in places like Israel, Taiwan, the United States, and around the world. These deployments confirm that Earth-2 works in a variety of operational settings.
Industry leaders said that Earth-2 would change the way scalable weather intelligence is used. Nvidia said that models will be available for both business and personal use. GitHub and Hugging Face will both have the full family available.
