Despite a decade of behavioural research on grizzly bears in B.C.'s Knight Inlet, Melanie Clapham still has trouble telling some individual bears apart.
Brown bears, which include grizzly bears, can change dramatically in their appearance during their younger years and, unlike other wildlife that has spots or stripes, they lack distinguishing markings on their bodies. Ms. Clapham, a conservation biologist and postdoctoral research fellow at the University of Victoria, dreamed of technology that could help her individually identify these furry mammals.
While she was looking for a tech team to make that idea possible, south of the border, Ed Miller and Mary Nguyen, two Silicon Valley engineers who are also outdoor and wildlife enthusiasts, had started a project to develop machine-learning models that could be adapted to grizzly bears.
The three connected in 2017 through a conservation technology network. Their research, published in the open-access journal, Ecology and Evolution, earlier this month, has culminated in a system that uses deep learning, a method of artificial intelligence, to detect and identify brown bears from photographs.
Researchers say the new technology, termed BearID, created a “non-invasive” technique for them to study the animals.
“It enables monitoring these bears without having to capture them and tag them or get genetic information from them,” Mr. Miller said.
BearID detects a bear in an image, rotates and extracts the face, creates an encoding for the face that will eventually help classify and identify the individual.
The research team trained and tested the application using more than 4,600 images of 132 known grizzly bears located in Knight Inlet, as well as Brooks River, in Katmai National Park, Alaska.
With the help of this software, the team identified individual bears with an 84-per-cent accuracy.
Ms. Clapham said not being able to recognize individuals in a consistent manner limits the scope of the type of research that can be conducted on them, but the new system is helping remove those barriers.
By combining BearID and remote cameras, researchers will be able to conduct population assessments on grizzly bears, which have been historically difficult to do with accuracy.
“To be able to manage species and populations appropriately, we really need accurate estimates of the population numbers, and so this can hopefully help us to get there for grizzly bears,” she said.
For Mr. Miller and Ms. Nguyen, who could only work on this project outside their regular work hours, developing the software wasn’t an easy task.
Machine-learning models for human face recognition are trained on millions of images, however, for wild bears, Mr. Miller said, there are hardly enough photographs.
“It would be very difficult to collect photographs of 300,000 different bears, let’s say, and keep them each individually identified, so that’s what makes it very difficult,” he said, adding they had to tailor the algorithm to work with a smaller set of data on a particular group of bears.
At the early stage of the project, Mr. Miller said they found a machine-learning software that could detect the face of a dog, so the team fine-tuned the model for bear faces, which became the first crucial step of the bear-identification process.
In the future, this software can potentially be applied to other species such as wolves, Mr. Miller added.
Knight Inlet is a prime habitat for black bear and grizzly bear populations in B.C. and is located in the traditional territories of the Da’naxda’xw-Awaetlala First Nation. The nation’s leader called the technology a good fit for his community.
“We wanted to make sure that we could understand the impact we’re having on the bears as well. So this technology just helps us really pinpoint which bears are in which areas and how habitual that is, and what is their range like,” said Dallas Smith, president of Nanwakolas Council that comprises five First Nations, including Da’naxda’xw-Awaetlala.
The new technology also helps the communities deal with bear-human conflicts, which have become more prevalent, he said.
These are “important to us because, as First Nations, grizzly bears are iconic to our culture,” Mr. Dallas noted.
Ms. Clapham said her team has been working with these communities by using BearID in combination with camera trapping to observe and recognize individual bears within their territories.
As salmon populations have been in steady decline over the past few years, Ms. Clapham said, her team is interested in continuing work with these First Nations to look at how the bears' movements are affected by reductions in salmon returns.
Last year, as part of the study, Mr. Miller visited Knight Inlet.
“It was so fun and exciting to be able to see these bears out in the wilderness living their lives, not realizing that they’re celebrities from our standpoint,” he said.
We have a weekly Western Canada newsletter written by our B.C. and Alberta bureau chiefs, providing a comprehensive package of the news you need to know about the region and its place in the issues facing Canada. Sign up today.