Artificial intelligence has the potential to be a “transformative tool” in understanding the Chesapeake Bay and improving restoration efforts, according to a recent report by Bay scientists.
AI helps sort and interpret immense amounts of data, allowing products like high-resolution maps to track land use change in detail never before possible, according to the state-federal Chesapeake Bay Program’s Scientific and Technical Advisory Committee (STAC). It has the potential to more precisely manage nutrients on farmlands, assess cleanup progress and help prioritize efforts.
But AI has triggered increased construction of massive data centers that have potentially negative consequences on the environment. They often rely on power sources that spew pollutants into the air, draw billions of gallons of water from the region’s aquifers and rivers, and create massive concrete footprints that can spell doom for local stream health.
Those tradeoffs were apparent at the committee’s virtual meeting in April, focused on the opportunities for AI to help the Bay restoration.
The committee, which issued a report about the transformative potential of AI for the Bay last year, began its April meeting with a disclaimer: “STAC acknowledges the many concerns about ethical and environmental impacts of AI usage, including water and energy usage, land use and potential community impacts.”
The statement, as much as anything, highlights the mixed-bag AI creates. It is already used effectively to improve forecasts of storm surges, surface waves, saltwater intrusion, oxygen levels in the Bay and harmful algae blooms.
It’s essential in sorting through the huge amounts of satellite imagery used to create high-resolution watershed maps, which are used to assess land use changes and predict future ones — all of which can aid decision making and accelerate environmental efforts.
The Bay is uniquely situated to benefit from the new technologies. It is one of the best-studied and best-monitored coastal regions in the country.
“Here in the Chesapeake, we have long-term monitoring, and we have data from all different sources,” said Qian Zhang, an associate research scientist with the University of Maryland Center for Environmental Science (UMCES), who chaired last year’s STAC workshop and spoke at the April meeting.
The sheer amount of accumulated data poses challenges, though, causing some of their potential to go untapped. It’s also not just one big pile of information: It’s data from many different sources, collected for many different reasons. “AI can both help us leverage that data and also integrate different data sources,” Zhang said.
Victoria Coles, an UMCES professor, touted the potential of AI to help with mundane jobs, such as identifying phytoplankton species in water quality samples. But developing AI models and “training” them with datasets is an energy-intensive task. Researchers, Coles said, need to consider whether the information gained is always worth the cost.
“We should be thinking about, what is the incremental improvement in our modeling capacity and what is the energetic cost,” she said. “There’s no debate about whether we should be using AI, but we also need to be justifying [whether] a 1% in model accuracy is sufficient to justify large training costs.”
AI does not eliminate the need for labor-intensive tasks such as collecting real-world monitoring data needed to feed models, or to “ground truth” the results they produce.
Also, although AI can produce information quickly, it’s often unclear how AI models actually work. Scientists often describe them as “black boxes” and have raised concerns that, in some cases, they could produce results that look plausible — but for the wrong reasons.
Because of that, scientists warn that there should be limits on the ways that AI is used, especially in management and decision-making.
Those ethical questions were highlighted by Brian Erickson, a social scientist at Oregon State University who surveyed dozens of state and federal agency officials involved in reviewing public comments related to environmental rules.
Some felt AI would be helpful with the otherwise tedious job of reviewing the thousands — even hundreds of thousands — of comments received about major rules. Some said it had the potential to remove bias, Erickson reported.
But because it’s not always clear what’s happening “under the hood” with AI models, others worried about accuracy and about losing nuance and context in comments. Agencies are typically required to review all substantive comments, and failure to properly manage comments could lead to lawsuits.
One told Erickson: “When we’re in the business of working with people’s concerns and feelings that impact their everyday life or their business or their community, I’m not really eager or interested in giving up my human interaction with that to save work time.”
But streamlining and accelerating some Bay conservation work might be more straightforward. For instance, to meet requirements of the state-federal Bay Program, conservation district staff spend large amounts of time verifying that farm and urban stormwater best management practices, like stream buffers and cover crops, are in place.
Using satellites, drones and other remote sensing technology along with AI could save time ensuring those measures are on the ground. That would free conservation districts to work with farmers and other landowners to implement new practices, rather than counting old ones.
“A lot of the professional conservation community is being asked to do the job of verification and it, in a lot of ways, can take away from tasks that they’re doing to put new conservation on the ground,” said Matt Royer, director of the agriculture and environment center at Penn State University.
One example of how AI could accelerate implementation is the “Conservation Concierge” under development by a nonprofit called the Commons, which uses high tech services to promote conservation efforts.
The tool uses satellite imagery, watershed information and other data to assess conservation practices that might be suitable for a particular farm. It also analyzes where those practices should be used and what state, federal or other funding sources could support it.
And AI can do it quickly. “It takes it less than two minutes to generate this kind of information, whereas if you’re doing it the old-fashioned way, it might be a couple of months,” said Alex Echols, the agriculture program strategist with the Campbell Foundation which is funding the effort.
It doesn’t replace the need for conservation professionals, but it can allow them to work with more farmers, reduce red tape and get better results, he said.
Echols described himself as a bit of an AI skeptic and a “dinosaur,” but he said the technology can transform the current “clunky” system, which is critical if restoration goals are to ever be met. “To have this dinosaur, of all people, talking about AI is funny,” Echols noted.