Darius Baruo
Sep 19, 2025 20:39
NVIDIA, in collaboration with Berkeley Lab, introduces HENS, a machine learning tool for predicting extreme weather, offering supercomputer-class forecasting with reduced computational power and cost.
NVIDIA, in partnership with Lawrence Berkeley National Laboratory, has unveiled a groundbreaking machine learning tool named Huge Ensembles (HENS) designed to predict extreme weather events with the efficiency of supercomputers, but without their exorbitant costs and power requirements. As reported by NVIDIA’s official blog, this tool is poised to transform how climate scientists, city officials, and emergency managers prepare for and respond to severe weather scenarios.
Revolutionary Forecasting Capabilities
HENS offers an open-source solution, available as code or a ready-to-run model, capable of forecasting low-likelihood, high-impact events such as prolonged heat waves or centennial hurricanes. Unlike traditional models, HENS can predict these events from six hours up to 14 days in advance, with a resolution of 15 miles (25 kilometers), providing critical lead time for preparation and response.
The tool’s development is detailed in a two-part study published in the journal Geoscientific Model Development, showcasing HENS as one of the most extensive and reliable ensembles of weather and climate simulations, producing data equivalent to 27,000 years of weather patterns.
Advanced AI-Driven Methodology
Utilizing NVIDIA’s PhysicsNeMo and Makani frameworks, HENS refines weather prediction by training AI models on 40 years of high-quality atmospheric data. This approach not only enhances accuracy but also reduces the energy and computational resources typically required for such large-scale simulations.
According to Ankur Mahesh, a graduate researcher at Berkeley Lab, the extensive dataset generated by HENS serves as a treasure trove for analyzing the statistics and drivers of extreme weather events, a scale previously unattainable with traditional methods.
Boosting Prediction Accuracy
HENS significantly outperforms conventional weather models by generating thousands of ensemble members, far surpassing the limits of standard models which typically produce only 50. This increase in ensemble members allows for a more comprehensive exploration of potential weather outcomes, enhancing the ability to predict rare but severe events.
During rigorous testing, HENS demonstrated an ability to capture 96% of extreme weather events, with prediction uncertainties over ten times smaller than those of traditional models, establishing it as a highly reliable tool for climate research and disaster preparedness.
Future Developments
Looking ahead, the team plans to delve deeper into the 27,000-year simulations to unearth new insights into the causes of catastrophic events like heat waves and hurricanes. Additionally, there is an ongoing effort to further streamline HENS’s computational requirements, making it even more accessible for widespread use.
Image source: Shutterstock
