NOAA bolsters its AI forecasting capabilities
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Natalie Peeples/Axios
The NOAA is adding to its high-performance computing resources in a bid to bolster its uses of AI and machine learning for weather and climate.
Why it matters: The agency's $100 million investment comes as it faces rapid advances in forecasting techniques, and new entrants build accurate weather models trained on historical data.
- These competitors include big tech companies like Nvidia and Google, as well as private forecasting startups like tomorrow.io and WindBorne, plus NOAA's peer agencies abroad.
- The money comes from the 2021 infrastructure package and 2022 climate laws.
Zoom in: According to the atmosphere and oceans agency, the money will be used to install a new high-performance computer from General Dynamic Information Technology.
- The new machine, called Rhea, after the Greek goddess and mother of gods, will be located at a modular data center in West Virginia.
- It would both benefit the agency's physics-based models and provide better training data for its AI/machine learning efforts, the agency told Axios.
- It would also help to "get more data into our forecast models."
Between the lines: NOAA, particularly its National Weather Service, has consistently fallen behind its peer agencies when it comes to computing power and model accuracy, although that has been alleviated to some extent.
- The advent of AI-based models increases the pressure on the agency to bolster its computing resources even faster.
- "NOAA has had a persistent need for additional computing power," Michael Morgan, assistant secretary of commerce for environmental observation and prediction, tells Axios via email. "With this recent investment, we are positioned to more ably meet the existing demand. AI/ML is a driver of this need."
By the numbers: Rhea would add about 8 petaflops of computing speed and capacity to NOAA's existing capacity of 35 petaflops.
- One petaflop is one quadrillion operations per second.
- Another machine is also in the pipeline for that facility, which NOAA says would bring its total research and development computing power to 48 petaflops.
The bottom line: This move is a step towards a more AI-centric NOAA since the modular data center approach could accommodate additional machines.
