Training at the Zoo and Looking Out for Squirrels

This polar bear season in Churchill, Manitoba, will see Polar Bears International (PBI) continuing trials of “bear-dar,” a unique application of a radar system to warn Arctic communities and remote camps of the presence of polar bears. The system was originally designed by our partner, Spotter Global, for security purposes. When complete and fully functional, it will combine a radar, a camera, and AI software to identify any nearby bears and alert responders. 

For now, it is in its trial phase, as PBI staff and volunteers work to train the software to distinguish between polar bears and other animals or objects. To help with that process, the system spent the summer pointing at the polar bear enclosure at Assiniboine Park Zoo in Winnipeg. It is presently back in Churchill, adorning the Tundra Buggy Lodge for the next round of testing. Here, Kieran McIver, PBI’s director of Churchill field operations, recaps the latest progress.

Perhaps the biggest remaining obstacle with bear-dar is training its software to reliably identify when a potential target is or is not a polar bear. Was that the goal behind moving it to Assiniboine Park Zoo for the summer?

Yes, that was the idea. The software requires quite a number of data points and it's really difficult to get as many as were required with wild polar bears here in Churchill. So we determined that if we could get it down 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, so we reasoned we could easily acquire these data points. Now that we have done that and added all those data points into the AI system, we are attaching the equipment to the Tundra Buggy® Lodge for the duration of bear season, to put it back in front of wild bears.

So how did everything work when it was in Winnipeg?

It wasn’t as simple as teaching it what a polar bear looked like. We also had to teach it what a polar bear doesn’t look like, and to differentiate between a polar bear and a person or another type of animal like a caribou, or a vehicle, different things like that.   

There are lots of people at the zoo, obviously. So we were able to get lots of data points on people. And then when staff were in the polar bear enclosure — say, mowing the grass –- or when after-hours security were driving around the zoo, we took that as an opportunity to train the radar to start learning what a vehicle looks like.

How it works is that the system will pick up a track of a moving object and record it. The radar system is connected to a PTZ [pan-tilt-zoom] camera, so once it detects movement, and if it thinks it’s something that we might be interested in, it will take control of the camera and actually start tracking that target visually. We created a schedule so that staff and volunteers could log in remotely multiple times a week to see what the system had saved and marked as possible polar bears. We discarded the ones without video footage right away, because we were unable to confirm if it was a bear or not. And then we would review all the ones that had video and it was as simple as going through each track and saying, ‘Yes, that is a polar bear. No, that’s not a polar bear.’ And we would feed that information into the AI. It was a lot of hands-on work for quite a few people who were very generous with their time.

Did it work? Is the AI getting better, or does that remain to be seen?

I think we’re gonna have to wait a little bit and see how things look at the end of the season. It seems pretty promising: At the zoo, we were getting a high percentage of success when it thought it was identifying a polar bear, but we really won’t know until we have a season of it set up here in a wild setting. One of the things we’re on the lookout for is how the local landscape — rock, snow, water, ice, etc. — is possibly going to have an impact on how the radar behaves.

Isn’t one of the challenges the fact that polar bears in Churchill tend not to move a lot and spend a lot of time just plopped down on the tundra waiting for the ice to form? Was that also a consideration in moving the equipment to the zoo for the summer?

The bears in the zoo like to lounge around as well. I think that’s just natural behavior. And it’s kind of a funny problem to have, you know? I don’t think anyone would have expected that to be an issue going in. But yeah, the bears will plop down, and they’ll sit there for hours or a day or whatever, and won’t move a muscle. And because the radar mainly senses movement, it would just lose interest in the target. And so then the camera wouldn’t necessarily stay focused on the bear, and off it would go. So that could still be a potential issue. But I think as long as the system is able to identify a possible polar bear and send out an alert, then people can get a visual confirmation and choose how to respond.

I understand that there were some problems with a much smaller mammal in Winnipeg.

When I arrived at the zoo at the end of summer, we were planning to dismantle the tower on which the radar was mounted and get it ready for travel back up north. I was standing next to the tower, and I looked up and noticed apparent wear on one of the wires. And I mean, this tower is pretty new, so I wouldn’t have expected that we’d already be going to have this type of wear and tear happening. But as I got closer, I realized something had been gnawing on the cable. Oh well, I thought, it’s just one cable. But as I did my walk around the tower, I realized that squirrels had been going to town on it all summer long. They even climbed the tower and started chewing on the cables that were connected to the solar panels. And then we had a 200-foot section of cable that was plugged into an outlet at the zoo to keep the batteries topped off. And all throughout that cable, there were a bunch of gnaw marks. It was unfortunate because it put me into panic mode, and I was scrambling trying to find new cables before we shipped everything back north.

Squirrels and other rodents aren’t exactly uncommon in the Arctic. So we learned a lesson, and we’ll have to plan for that type of thing in the future.

It’s just one of those things that mean you have to pivot and adapt and, you know, go with the flow. And that’s just how it goes with field work in general, and especially the type of work we do here in the North.

Kieran Mulvaney is the author of The Great White Bear: A Natural and Unnatural History of the Polar Bear. He is a frequent contributor to the Polar Bears International website and also writes for publications including National Geographic, Smithsonian, and The Guardian.