Whenever Alyssa Bohart heard a voice from her computer repeatedly chiming—status alert, status alert—the search was on. The warnings came from a radar device installed in Churchill, Manitoba—a modified military system programed with artificial intelligence (A.I.) and trained to detect polar bears. Bohart’s job was to remotely operate a camera and visually confirm that the AI was making the right call.
Over the past four years, the nonprofit Polar Bears International (PBI) has led radar, or “bear-dar,” tests to hopefully help northern communities stay safe. Polar bears are powerful predators, and for people in the Arctic an essential part of coexisting with bears is maintaining a respectful distance. But with climate change diminishing sea ice, polar bears are spending more time on land. The chances of polar bears and humans interacting are increasing, which can have devastating consequences for both parties.
“We would just like to help manage those human interactions and make sure there are more nonlethal options,” says Alysa McCall, a staff scientist with PBI.
The project was sparked when PBI was looking for new ways to prevent human-polar bear conflicts. A fortuitous meeting with SpotterRF, a company that makes military radar devices, gave them an idea. Instead of detecting approaching drones or vehicles, maybe they could harness the radar system to detect polar bears heading toward towns and send out an early warning.
In their initial tests, the researchers discovered that “the technology works great,” says Geoff York, PBI’s senior director of conservation. “If anything, it detects too much.” Although to the unfamiliar the tundra might seem barren, the radar flagged thousands of motion alerts. So, in 2020, the team turned to A.I. to try to narrow down the hits.
“By incorporating this A.I., we’re truly teaching this radar to learn what a polar bear is,” McCall says. “It’s learning to take away the humans and dogs, and quads and caribou, and only alert us when there’s a polar bear.”
This past fall, with Bohart’s help, the team positively identified 28 polar bears and had 89 alerts that didn’t turn out to be bears. McCall says their goal is to have more positive identifications than negative, but they also want to err on the side of caution with the calls the A.I. makes. In rare events, the team noticed polar bears, either on the camera or in person, that the bear-dar didn’t identify. In those cases, the two most confusing variables for the bear-dar were variations in the bears’ sizes, and the direction the bears were walking in relation to the radar. “We’re not quite there yet. But the results are promising enough that we have a good direction of where to go with the A.I.,” says McCall.
Douglas Clark, an expert in polar bear-human conflict at the University of Saskatchewan who has lived and worked with polar bears in Churchill and who is not involved in PBI’s project, notes that detection devices come with challenges. Arctic conditions are tough on technology, he says. “The challenge with detection technology is you’re asking people to trust their lives to it,” he says. “This is not a critique of the radar application, and it’s not a critique of the efforts. It’s just this is a field in which the bar is high.”
McCall agrees that the bear-dar would only be a piece of the safety puzzle. “Even if we get this perfect, it will not be the only polar-bear-safe tool someone would need,” she says. In Churchill, bear-dar could supplement the well-trained Polar Bear Alert team that patrols town and deters intruding bears.
Todd Atwood, a polar bear ecologist with the United States Geological Survey who also isn’t involved in the bear-dar project, thinks small communities without well-funded programs like Churchill’s—particularly in Alaska, where he lives—could benefit from the help of bear-dar. “But could they afford a system like that? Do they have folks in the community that can maintain and troubleshoot that system?” he says.
York believes that with more work bear-dar could be accessible to more communities. “We just have to work with partners and find a system that is economically sustainable for communities to actually purchase and maintain.”
Luckily, the team may be on the cusp of rapid progress toward that goal. When word got out about PBI’s progress with bear-dar, other researchers reached out with possible alternate devices. One, developed at Utah’s Brigham Young University, is a lidar rather than radar system that has less range but is cheaper and more portable. This coming fall, PBI hopes to test this less expensive device and do a trial deployment of the SpotterRF bear-dar.
With more tools and support, McCall says northern towns could tailor the bear-dar’s early warnings to their needs. “With more warning, you have more options,” says McCall, “and that’s safer for the bears and the people.”
This article is from Hakai Magazine, an online publication about science and society in coastal ecosystems. Read more stories like this at hakaimagazine.com.
Related stories from Hakai Magazine: