Since the 1960s, forecasting the weather had been left to the government’s supercomputers. These computers ran complicated physics-based calculations. NVIDIA’s release of has broken down that barrier. Earth-2 open family models and tools AI climate and weather prediction is now available to everyone, from startups to national agencies.
A move to democratize climate science NVIDIA unveiled 3 groundbreaking new models Powered by new architectures Atlas, StormScopeThen, HealDA. These tools will accelerate the forecasting process by several orders of magnitude and deliver accuracy equal to or greater than that achieved with traditional methods.
Weather Intelligence – Democratization
Running a weather model with high accuracy required an expensive infrastructure, which only few countries were able to afford. NVIDIA’s Earth-2 changes the calculus by offering an ‘open stack’, a collection of pretrained models, inference libraries, and customization recipes available on platforms like GitHub and Hugging Face.
Mike Pritchard is the Director of Climate Simulation for NVIDIA. He stressed that NVIDIA will not be a provider of weather services. They are instead building the “foundational building blocks” This allows nations and businesses to create their own forecasting systems.
“Sovereignty matters. Weather is a national security issue… That’s why we’ve built Earth-2, the world’s first fully open production-ready AI weather stack.” – Mike Pritchard of NVIDIA
HealDA, Atlas and StormScope are the New heavyweights.
This announcement presents 3 models that are designed to address the different stages in forecasting, from analyzing messy data up until predicting storms several weeks ahead.
1. Earth-2 Medium Range (Powered By Atlas)
This model targets the 15-day window of forecasting and uses a newly developed architecture known as Atlas. The system predicts with high accuracy over 70 variables of weather, such as wind, pressure and humidity.
- Performance: Atlas outperforms standard benchmarks in the industry GenCastThe current most popular open model is the one that covers the majority of variables.
- The Shift This is a return of the traditional. “simple, scalable Transformer architectures,” Moving away from niche AI-based designs.
2. Earth-2 Nowcasting StormScope powered
The game has changed for the immediate response to disasters. Powered by StormScopeThis generative AI model is focused on the 0-to-6 hour window and provides kilometer scale resolution of local weather.
- What it means: The first AI-based model has outperformed traditional methods based on physics for forecasting short-term rainfall.
- Speed: This tool generates dangerous weather forecasts in just minutes. It gives emergency responders crucial time to react.
- Sovereignty: This system can be implemented by any country that has good satellite coverage.
3. Earth-2 Global Data Assimilation HealDA – Powered by HealDA
Sometimes the unsung heroes of forecasting. “data assimilation” This is a process for combining satellite data and balloon data to create a cohesive snapshot of the atmospheric conditions in order to make a weather forecast.
- Breakthrough Assimilation is a traditional method that consumes about 50% of the supercomputer cycles. NVIDIA’s HealDA Architecture accomplishes the task. Minutes on GPUs Instead of spending hours in front of supercomputers.
- Result: It produces the most accurate predictions of any AI pipeline when combined with the Medium Range models.
Solar Power and Hurricane Risk: Real World Impact
Earth-2 is used by global major players and is a proof that AI-based weather forecasting has reached its commercial prime.
- Renewable Energy TotalEnergies You can also find out more about the following: GCL Earth-2 has been used to help a leading solar material manufacturer predict the solar and winds variability. Cloud cover predictions can have a significant impact on energy trading for solar farms.
- Israel Meteorological Service The Use of CorrDiff The Earth-2 model is a part of their family. Reduced compute time by 90% While generating high resolution forecasts as often as eight times per day.
- Insurance & Risk: AXA You can also find out more about the following: S&P Global Energy Earth-2 can run at the speed of thousands “counterfactual” scenarios. They can understand climate changes that are rare and high impact by simulating hypothetical data over thousands of years.
- Everyday Operations BrightbandEarth-2 Medium Range, a provider of AI weather tools, has already been integrated by to provide daily global forecasts.
Bottom line
NVIDIA’s Earth-2 upgrade is more than a technological improvement; it represents a paradigm shift for how people interact with climate. NVIDIA’s goal is to enable hyper-local and high-accuracy forecasting in the future by lowering entry barriers.
Tools like StormScope You can also find out more about the following: Atlas It is likely that governments and industry will use this infrastructure as a vital part of their operations.
Earth-2 medium range and Nowcasting is available at GitHub, NVIDIA, and Hugging Face Earth2Studio. Earth-2 Global Data Assimilation should be available by the end of this year.
Developers can learn more by visiting the NVIDIA Earth-2 technical blog. Earth-2 Medium Range [Read the research paper], Earth-2 Nowcasting [Read the research paper]Then, Earth-2 Global Data Assimilation [Read the research paper].
Jean-marc has been a highly successful AI executive. He leads and accelerates the growth of AI solutions. In 2006, he founded a company that specializes in computer vision. A Stanford MBA, he is an experienced speaker and a well-known AI conference presenter.
