NVIDIA’s StormCast AI model powers weather forecasting and climate simulations.

As hurricanes, tornadoes, and other extreme weather events increase in frequency and severity, it is becoming increasingly important to use cutting-edge technologies to improve and accelerate climate research and forecasting. According to the NVIDIA Blog, amid the peak of the current Atlantic hurricane season, NVIDIA Research has announced StormCast, a groundbreaking generative AI model for high-fidelity atmospheric dynamics emulation.

Advanced features of StormCast

StormCast enables reliable weather forecasting at the mesoscale, which is larger than a storm but smaller than a cyclone, and is essential for disaster planning and mitigation. This advancement comes at a time when extreme weather events continue to take lives, destroy homes, and cause more than $150 billion in damage each year in the United States alone.

StormCast, detailed in a paper written in collaboration with Lawrence Berkeley National Laboratory and the University of Washington, represents a significant advance in generative AI applications for climate research and actionable extreme weather forecasts. These AI models can help scientists solve high-stakes challenges like saving lives and protecting infrastructure.

Integrated with NVIDIA Earth-2

NVIDIA Earth-2, a digital twin cloud platform that combines AI, physics simulation, and computer graphics, enables simulation and visualization of weather and climate predictions on a global scale with unprecedented accuracy and speed. For example, Taiwan’s National Disaster Reduction Science and Technology Center uses CorrDiff, an NVIDIA-generated AI model provided as part of Earth-2, to predict the fine details of typhoons.

CorrDiff can analyze 25km of atmospheric data at 2km resolution, 12.5x faster. It is 1,000x faster than existing methods and uses 3,000x less energy for a single inference. This efficiency significantly reduces costs, making life-saving work more affordable.

From local to global influence

Global climate studies often start at the local level, where weather and the physical risks of climate change can vary dramatically. Reliable numerical weather predictions at this level are computationally expensive due to the high spatial resolution required to represent mesoscale fluid dynamics.

Convection-permitting models (CAMs) are useful for tracking storm evolution and structure and understanding weather-related physical hazards at the infrastructure level. These models traditionally have to trade off resolution, ensemble size, and cost-effectiveness. However, machine learning models trained on global data have emerged as useful emulators of numerical weather prediction models, improving early warning systems for severe events.

StormCast, leveraging generative diffusion, now enables weather forecasts at 3 km per hour. When applied with precipitation radar, the model provides forecasts with a lead time of up to 6 hours, which is up to 10% more accurate than the National Oceanic and Atmospheric Administration’s (NOAA) state-of-the-art 3 km operational CAM.

Scientific cooperation and future prospects

NVIDIA researchers trained StormCast on about three and a half years of NOAA climate data from the central United States, using NVIDIA accelerated computing to speed up calculations. The model’s output shows physically realistic heat and humidity dynamics and can predict over 100 variables, allowing scientists to see a realistic 3D evolution of the storm’s buoyancy.

“Given the tremendous impact of organized thunderstorms and winter precipitation, and the major challenges in predicting them with confidence, producing computationally tractable storm-scale ensemble weather forecasts is one of the greatest challenges in numerical weather prediction,” said Tom Hamill, director of innovation at The Weather Company. “StormCast is a remarkable model that addresses this challenge, and The Weather Company is excited to work with NVIDIA to develop, evaluate, and potentially deploy this deep learning forecast model.”

“Developing high-resolution weather models requires AI algorithms that can resolve convection, which is a formidable challenge,” said Imme Ebert-Uphoff, machine learning director at Colorado State University’s Cooperative Institute for Atmospheric Research. “The new NVIDIA research explores the potential to achieve this using a dispersive model like StormCast, representing an important step toward developing future AI models for high-resolution weather forecasting.”

NVIDIA Earth-2 is ushering in a new and critical era of climate research by accelerating and visualizing physically accurate climate simulations, demonstrating the importance of generative AI in solving global climate challenges.

Image source: Shutterstock