Ten years ago, the National Oceanic and Atmospheric Administration (NOAA) began exploring how AI could help in monitoring the health of marine ecosystems. Since then, NOAA has plunged into research and development.
“We are now well-positioned to begin making the transition from research to operation in certain domains,” said Benjamin L. Richards, the Chair of NOAA’s Artificial Intelligence Executive Committee, in an interview with Techstrong AI. “We’re in a position to continually make strides along that trajectory. Now is a good time, we have new data sets in hand, more training data, to seek input from industry.”
That transition begins with a “Request for Information” (RFI). NOAA is looking for vendors to help it develop AI-powered tools capable of monitoring marine ecosystems, everything from operating autonomous craft capable of measuring and mapping underwater geographic features, to studying protected species with images and acoustic recordings, to reviewing aerial and underwater surveys to assess fish populations. NOAA operates six regional science centers for fisheries, with locations from Seattle to Miami. Through the Department of Commerce, NOAA’s Fisheries Optic Strategy Initiative (OSI) Working Group is seeking vendors to conduct large-scale image collection, processing and storage. OSI is scheduled to meet with prospective vendors in November 2024, and intends to fund multiple projects thereafter, allotting between $250,000 and $500,000 each.
“Over the past ten years, we’ve done a lot of internal development, with existing partners identified through an RFI back in 2015,” Mr. Richards said. “We’ve made a lot of progress, and since that time, the industry has evolved, we’ve evolved, models have evolved, so there’s a lot more industry, a lot more vendors in the playing field, and we feel it’s the right time to go back out to industry and say, OK, we’ve been doing this over the past five to ten years, what are we missing, what don’t we know about? As part of the federal government, funded by taxpayers, we want to make sure that we are upholding the public trust, being efficient with taxpayer dollars; we want to make sure we are kicking the bushes, and being able to leverage the best, brightest, most efficient methods out there, and that’s what this OSI workshop in November aims to do.”
With the first RFI in 2015, NOAA brought in some of the machine learning techniques that had been developed for object detection and optical imagery for the military, and civilian applications, namely biomedical research and driverless cars.
“Realizing that a lot of the background deep learning models really don’t care whether they’re looking at images of fighter planes or missiles or people with backpacks, or tumors, it’s just what you train them on, so we leveraged some of that development to start training deep learning models on specific fish targets of interest, primarily in the Pacific Islands region, the Southeast, Gulf, Caribbean, South Atlantic, New England, and the Pacific Northwest,” Mr. Richards said.
NOAA partnered with Kitware, based in Clifton Park, NY., and with the University of California, San Diego, to develop two initial systems, for fish and for coral reef ecosystems. Out of the partnerships, they developed VIAME, an open source software toolkit for training deep learning models for object detection, that can count and classify specified targets. It is currently in use at the Southeast Fisheries Science Center in Miami, to document Red Snapper and Gray Triggerfish, and at the Northeast Fisheries Science Center in Woods Hole, MA., to conduct scallop surveys.
AI will likely lift many graduate students, technicians and analysts, from tedious and time-consuming work, such as counting fish populations from thousands of hours of video footage, so that they can concentrate on more academic research endeavors. The cameras record hours of footage in which fish don’t appear, so the first thing that AI can do is eliminate that wasted time, “so that we can optimize the use of our human capital and have them only look at imagery that AI has identified as something to look at,” Mr. Richards said.
But AI won’t replace humans entirely in that task. In critical assessments, humans will still review low-confidence results flagged by AI.
“OSI envisions a transformative suite of products that enable end-to end automation of optical sampling from acquisition to analysis,” said Matthew D. Campbell, Gulf and Caribbean Reef Fish Branch Chief, Population and Ecosystem Monitoring Division. “The creation of fully integrated hardware-to-software optical data pipelines, embedded into onboard processing systems where possible. We aim to use end-to-end automated approaches to reduce survey days and hands-on processing time, while increasing our temporal coverage and spatial range of data acquisition, diversity of targets, and quality of management-relevant metrics. This transformation will reduce our reliance on white-ships, diversify our portfolio of sampling strategies, and avail monitoring technology to under-served regions.”
NOAA’s fish stock assessments are crucial for managing U.S. marine resources, which, according to NOAA’S 2024 Annual Report to Congress, supported 2.3 million jobs and generated $321 billion in sales in 2022. Sustainable fisheries ensure not only healthy marine ecosystems but also thriving coastal communities and a robust economy.
AI is not just transforming NOAA’s operations. Other organizations, like the Wild Salmon Center in Portland, Oregon, are employing AI to monitor salmon populations. The center’s AI tool, “Salmon Vision,” can distinguish between types of salmon, assess their health, and even identify whether they are wild or from a hatchery.
“We need information on how many salmon are returning everywhere that we’re fishing for salmon,” said Will Atlas, Wild Salmon Center Senior Watershed Scientist. “You can’t tell me with a straight face that you’re having a sustainable fishery if you don’t know how many fish you have coming back. And that’s a problem right around the Pacific Rim.”
NOAA’s advancements in AI-driven research promise a future where scientists can focus on big-picture challenges while AI-powered machines and devices handle the labor-intensive details. The Administration has used AI for years to help with weather predictions. As Mr. Richards noted, the agency is now leveraging artificial intelligence “from the surface of the sun to the bottom of the sea.”
