NVIDIA Earth-2 CorrDiff AI Weather Model Deployed at Taiwan NCDR for Typhoon Forecasting

Taiwan's National Science and Technology Center for Disaster Reduction (NCDR) deployed NVIDIA's CorrDiff generative AI weather model for typhoon forecasts and other disaster alerts, with an estimated gigawatt-hour of energy savings due to the energy efficiency of CorrDiff running on the NVIDIA AI platform. The context was that to better prepare communities for extreme weather, forecasters need to see exactly where it'll land. Weather agencies and climate scientists around the world are harnessing NVIDIA CorrDiff, a generative AI weather model that enables kilometer-scale forecasts of wind, temperature, and precipitation type and amount, as part of the NVIDIA Earth-2 platform for simulating weather and climate conditions. Technically, CorrDiff uses generative AI to sharpen the precision of coarse-resolution weather models — resolving atmospheric data from 25-kilometer scale down to 2 kilometers using diffusion modeling. The optimized CorrDiff NIM microservice for U.S. data is 500x faster and 10,000x more energy-efficient than traditional high-resolution numerical weather prediction using CPUs. CorrDiff was trained on the Weather Research and Forecasting (WRF) model's numerical simulations to generate weather patterns at 12x higher resolution. The initial CorrDiff model was described in a paper published in Communications Earth and Environment, optimized on Taiwan weather data in collaboration with Taiwan's Central Weather Administration, before being scaled to cover the continental United States as an NVIDIA NIM microservice. At scale, CorrDiff predictions are embedded in NCDR's disaster monitoring site, helping Taiwan forecasters better prepare for typhoons. The model is also being used by AXA for extreme weather risk simulations, and NoteSquare (Korean startup) has modified CorrDiff for regional weather data from the Korea Meteorological Administration. NVIDIA researchers have released additional generative AI diffusion models showing how the model could be enhanced to more robustly resolve small-scale details in different environments and better capture rare or extreme weather events. The model continues to be applied in regional forecasting, renewable energy management, agricultural planning, and downwash prediction in urban areas.