The National Oceanic and Atmospheric Administration and Google linked up to cooperatively steer artificial intelligence and machine learning-driven pilot projects that could advance the agency’s environmental monitoring, weather predicting, and climate research capabilities.
Through a three-year deal unveiled Tuesday, the two will collectively study and develop small-scale AI systems—and eventually wider-ranging prototypes—that might be infused across NOAA’s broader enterprise. Such resources would be of good use to the water and climate-focused agency, which is responsible for providing raw data for weather forecasts nationwide, including those used to predict hurricanes and other natural disasters.
And as the research and technology deployments unfold, both entities aim to simultaneously offer their personnel unique, novel hands-on experience and training opportunities, coupling the agency’s leading environmental science expertise with the company’s command of AI.
In joining forces, each expects “to make significant progress, more efficiently, and instill new expertise in the partners’ workforces,” Sid Boukabara, principal scientist in NOAA’s Center for Satellite Applications and Research told Nextgov Tuesday.
Google spokesperson Jane Khodos on Tuesday also noted to Nextgov that if the effort is successful, “this has the potential to be a significant leap in NOAA’s ability to leverage the enormous volume and diversity of environmental data in order to enhance prediction, including for extreme weather events such as hurricanes and tornadoes.”
An other transaction authority agreement signed between the tech giant and NOAA’s Satellite and Information Service underpins the work. Born jointly by Google and NOAA, the total cost of that OTA is around $10 million, Boukabara confirmed.
“This includes leveraging researchers and existing projects aligned with the objectives of the OTA agreement,” he said.
Initial work will hone in on creating the smaller scale AI and ML systems—then, NOAA and Google Cloud insiders will hone in on executing “full-scale prototypes” that the agency can ultimately operationalize across its organization, according to the release detailing the effort. The hope is that they can tap the tech solutions to drive forward NOAA’s use of satellite and environmental data.
Previously, Google AI teamed up with NOAA’s Pacific Island Fisheries Science Center to train an AI model using their massive collection of underwater recordings and harness new insights about whales’ behavioral and migratory patterns. And through this latest effort, the company “is especially interested in platforms that use AI to increase the utilization of weather satellite observations,” Khodos said.
The work will also build on priorities laid out in NOAA’s new AI strategy, released earlier this year.
Boukabara, who also serves as the agency’s project manager of the agreement, said both Google and NOAA have kickstarted the planning phase of their partnership, and as a result of it in about a month down the line, they’ll have a firmer idea of specific joint AI and ML-centered projects they’ll collaborate on.
“But overall, I can say that these AI-based systems will aim at assimilating a higher rate of environmental data from a diverse set of sources,” he explained.
Among others, those sources could include satellites, crowd-sourced data, and balloons. The ultimate intent, Boukabara said, is to generate—with a much faster speed of execution—a richer environmental analysis at higher spatial, vertical and temporal resolutions. And if it works out, “this will have the potential to significantly improve the experience of” citizens, forecasters, governments and other users of environmental data, he noted.
One of NOAA’s primary missions is to save lives and property by providing accurate warnings, forecasts, and other environmental monitoring information, Boukabara further explained, adding that officials do so by taking advantage of the observations made by sensors on satellites, balloons, ground platforms and elsewhere. Employing an AI-based approach can boost the rate and types of data that can be used, and a number of projects have already demonstrated AI’s promise “to make a leap in our ability to use environmental data for weather prediction” and “do the same thing [they] do now but better and significantly faster,” he said.
It’s better because researchers can actually exploit information in the data captured that is usually difficult to put to use with regular techniques. This includes the step called ‘data assimilation,’ which Boukabara noted is key in producing an accurate numerical weather prediction, or NWP.
“We have a number of things to develop, tailor and to test, such as ensuring the analysis we obtain with AI is scientifically consistent with the physics of the forecast model used for NWP,” he said. “Recent and new developments in the AI and machine learning field such as explainable AI [and] trustworthy AI, should help us in this endeavor—and we plan to fully take advantage of them.”
Down the line, the work could also help the federal agency “increase its medium-range weather forecast skills and reclaim global leadership,” he added.