NOAA is shifting several critical systems to cloud platforms in an effort to improve weather-forecast accuracy and address growing data-management challenges, agency leaders said this week.
The service currently relies on two on-premises supercomputers that power the Weather and Climate Operational Supercomputing System (WCOSS). Each machine runs at about 14.5 petaflops and was designed for a five-year service life, but officials note the hardware is nearing its capacity limits.
Moving model output from the supercomputers to the agency’s 122 local forecast offices will require data reduction. Three-dimensional weather models contain many variables across atmospheric layers, and some of those layers may be omitted to fit bandwidth constraints, the agency explained.
“Trapped data is actually probably not my biggest problem. It’s actually data that we’ve deleted and will never get back,” Neil Jacobs said at the Google Public Sector Summit in Washington, D.C.
The agency plans to transition high-performance computing operations to the cloud. It aims to have a “full-scale system” ready by the end of 2026 and a go-live date by the end of 2027, according to the project lead.
During a learning phase, it is testing how its weather models run on cloud resources and deciding on the optimal configuration.
Google Cloud has been selected as the primary provider for the WCOSS modernization effort.
“For the NOAA workforce, the move to cloud [high-performance computing] provides a more streamlined, flexible environment, enabling meteorologists and researchers to quickly adapt future modeling innovations,” Brent Mitchell wrote in a public-sector blog post.
Separately, the agency announced that the Advanced Weather Interactive Processing System (AWIPS) will also move to cloud infrastructure by early 2028. Today, National Weather Service forecasters download model data to Linux workstations on site, a process that can stall during office evacuations or when higher-resolution models are released.
By relocating AWIPS, forecasters will be able to retrieve forecasts and warnings from remote locations, improving resilience and ensuring timely access to the latest model outputs across the service’s network.
