Reliable weather forecasts can save lives, and the National Oceanic and Atmospheric Administration (NOAA) is looking to cutting-edge technologies such as AI and cloud computing to improve its forecasting system and models.
At a House Science, Space, and Technology subcommittee hearing on March 6, one former NOAA official told members of Congress that “cloud is the answer” to addressing the growing volumes of new weather data.
“As files get larger, transfer speeds get slower. The mission of the National Weather Service is to protect life and property where often seconds matter. Increasing the latency of data availability, even if it is more accurate, is counterproductive,” explained Neil Evans, a former acting NOAA administrator and the current chief science advisor for the Unified Forecast System.
“Cloud is the answer,” Evans stressed. “Not only does cloud computing eliminate the latency problem, the elasticity of on-demand commercial cloud also eliminates the bottleneck of limited access to compute research for research and development.”
Evans said that forecasting efficiencies can also be gained from AI capabilities, adding that he is encouraged by NOAA’s “rapid adoption” of various AI and deep learning techniques.
“I think there’s a tremendous opportunity for AI all across numerical weather prediction, even down to potentially replicating the entire modeling system,” he said. “So, right now the current state of the art is physics-based models. AI can replicate a lot of what those algorithms do. And the advantage, I think, with AI is that you can run this on a tiny fraction of the amount of compute.”
However, Evans emphasized that AI will never replace those physics-based models, because you can’t predict the weather without “observing the atmosphere.” Instead, he said, AI will simply optimize the speed and efficiency of which the physics-based models predict the weather.
Scott Weaver, the chief executive officer of CLIMET Consulting, added that “the forecast can be done much quicker because you’ve trained it already.” Therefore, once you train the system, you can run it on a laptop because it doesn’t require a “massive amount of compute.”
“We’re at a technological revolution here,” said Weaver, who once served as a research meteorologist for both NASA and NOAA. “So, very exciting time.”
