#AI and Accelerated Computing: Transforming #Climate Research and Predictions / #DigitalTwins
AI and accelerated computing are poised to revolutionize climate research, according to NVIDIA founder and CEO Jensen Huang. Speaking at the Berlin Summit for the Earth Virtualization Engines (EVE) initiative, Huang emphasized the significance of climate modeling and its impact on policymakers, researchers, and the industry. The summit, which brings together global participants, focuses on leveraging AI and high-performance computing to enhance climate prediction.
During his keynote, Huang highlighted three pivotal breakthroughs that climate researchers must achieve to advance their goals. First and foremost, there is a need for faster and higher-resolution climate simulations, reaching down to a couple of square kilometers. Secondly, the ability to pre-compute vast amounts of data is crucial. Lastly, interactive visualization of this data using NVIDIA Omniverse is vital to empower policymakers, businesses, companies, and researchers.
The EVE initiative, in collaboration with Earth-2, aims to provide easily accessible kilometer-scale climate information for sustainable planet management. Huang pointed out that the convergence of Earth-2 and EVE was opportune, as Earth-2 was based on three fundamental breakthroughs. The initiative advocates for coordinated climate projections with a resolution of 2.5 kilometers, building upon substantial advancements made over the past 25 years.
Accelerated computing has already made a significant impact on various applications, including ICON, IFS, NEMO, MPAS, WRF-G, and more. Further computing power is on the horizon for such applications, with the NVIDIA GH200 Grace Hopper Superchip leading the way. Designed specifically for AI and HPC applications at a massive scale, this breakthrough accelerated CPU offers up to 10 times higher performance for processing terabytes of data.
To harness vast amounts of data efficiently and gain valuable insights, Huang introduced NVIDIA Modulus, an open-source framework for developing and training physics-based machine learning models, and FourCastNet, a data-driven weather forecasting model. By learning physics from real-world data, FourCastNet can accurately predict weather patterns such as the path of Hurricane Harvey by incorporating the Coriolis force, which accounts for the Earth's rotation.
Combining these models with traditional simulations allows for more detailed and long-range forecasts. Huang demonstrated how running the FourCastNet ensemble on NVIDIA GPUs anticipated an unprecedented North African heatwave. Utilizing Modulus, NVIDIA achieved faster generation of 21-day weather trajectories with reduced energy consumption compared to previous methods.
Furthermore, NVIDIA's technologies are making knowledge more accessible through digital twins, enabling interactive models of complex systems, from Amazon warehouses to the propagation of 5G signals in dense urban environments. Huang showcased an impressive high-resolution interactive visualization of global-scale climate data in the cloud, zooming in from a global view to a detailed view of Berlin. This approach holds promise for predicting climate and weather patterns in diverse locations such as Tokyo, Buenos Aires, and beyond.
Huang emphasized NVIDIA's commitment to developing more powerful systems for training AI models, simulating physical problems, and enabling interactive visualization. These cutting-edge supercomputers are just beginning to come online, representing the forefront of computing technology. Huang concluded his talk by acknowledging the contributions of key researchers and playfully envisioning a mission statement for EVE: to push the limits of computing in climate modeling, explore new methods and technologies, and boldly navigate Earth's tomorrow by understanding the impact of mitigation and adaptation on the global climate.