Teaching radar how to see polar bears

As climate change causes sea ice to retreat, polar bears may venture inland more frequently in search of food—and these outings can bring them into conflict with humans. As KIERAN MULVANEY writes, Polar Bears International (PBI) is collaborating with Spotter Global to create an early alert radar system—dubbed “Bear-dar”—to alert communities and remote camps when a bear is approaching.

The idea behind the initiative is to create a system that comprises two parts: a radar to detect approaching objects, and artificial intelligence (AI) software to determine whether any of those objects is a polar bear.

Somewhat surprisingly, the first step—sourcing a radar and related hardware—was relatively straightforward. After trialling a few systems, PBI settled on the Spotter Global version that it uses now.

Teaching AI to recognize polar bears

Most radar excels at detection, especially when it comes to moving objects. But the identification piece—where the AI software comes in—is more challenging, especially when the target is a furry animal. Using radar to detect a moving object is one thing, but determining which of many objects might be a polar bear is another thing entirely. The AI requires human intervention to teach it to recognize potential polar bears and ignore everything else.

That is why PBI tested the radar near Churchill, Manitoba, Canada. Every fall, polar bears in the region wander past—and sometimes through—the town on their way to the shores of Hudson Bay, where they wait for ice to form on the water’s surface. Unfortunately, those polar bears have proven a tricky subject to track because while they are waiting to head out onto the ice, they tend not to be particularly active.

“They might flop down for 10 minutes, or they might flop down for several days,” explains Geoff York, PBI’s senior director of research and policy. “And when they go still, the radar drops their signal.”

That’s why the system also spent a summer pointed at the polar bear enclosure at Assiniboine Park Zoo in Winnipeg under the direction of PBI’s Kieran McIver.

“The software requires quite a number of data points, and it’s really difficult to get the number required with wild polar bears here in Churchill,” says McIver. “So we determined that if we could get the system to Winnipeg, we’d have these bears at the zoo, and we would know where they were—and we knew there were quite a few of them. We reasoned that we could easily acquire the data points. Now that we’ve done that, we’ve put the system back in front of wild bears.”

Focusing on coexistence

The added advantage of the zoo is that it is full of people, vehicles and other animals, so in addition to teaching the AI which signals are polar bears, McIver and his team could help it learn which ones weren’t.

Equipped with those extra data points, the “bear-dar” is once again being tested in the wild. Ultimately, PBI hopes to deploy the devices in interested Arctic communities.

“I think we as conservation organizations have a responsibility—especially when we’re talking about conserving a large mammal that’s a predator—to the people who live among them,” says York. “That’s one of the reasons we are focusing more and more on coexistence and trying to make sure communities have resources that enable them to feel safer.”