TAIPEI (Taiwan News) — Nvidia announced StormCast, a new generative AI model providing greater accuracy and precision in weather prediction, in a press release on Monday (Aug. 19).
StormCast comes on the heels of a previously released AI weather prediction model named CorrDiff, a collaboration with Taiwan’s Central Weather Administration (CWA) beginning in 2022. StormCast adds hourly autoregressive prediction functions suited to extreme weather, potentially helping mitigate future weather-related emergencies, per CNA.
Nvidia said it worked with Taiwan’s National Science and Technology Center for Disaster Reduction to upgrade the AI weather prediction model from a 25-kilometer scale down to 2 kilometers. It has 1,000 times faster speed and 3,000 times less energy for a single inference when compared to traditional methods, significantly reducing costs.
NVIDIA said the new weather modeling can predict strong wind downwash, helping limit building damage and promoting pedestrian safety, especially in urban areas.
StormCast is in its early stages, but when used for predicting rainfall, it offers 5–10% improved accuracy in two to four-hour forecasts when compared to the US National Oceanic and Atmospheric Administration’s (NOAA) High-Resolution Rapid Refresh (HRRR).
Nvidia noted that extreme weather phenomena threaten lives and homes each year. In the US each year, weather causes more than $150 billion (NT$4.85 trillion) in losses. As natural disasters such as hurricanes and tornadoes increase in frequency and intensity, it is more important to use new technologies to improve weather forecasts.
StormCast offers greater efficiency, no longer requiring millions of US dollars spent on CPUs. A single system costs just US$60,000 utilizing an NVIDIA H100 Tensor Core GPU to run the program, per NVIDIA press release.
The massive savings are just one example of how generative AI can accelerate computing efficiency and lower costs, making better weather prediction models more accessible to governments and private companies.
