New software puts researchers “face to face” with bears

Photo: courtesy of the BearID Project
Along the coast of British Columbia, the BearID Project utilizes facial recognition software to track brown bears for scientific research in a non-invasive manner.

By Kayla Heinze, Communications Specialist

Last month on the blog we turned our imagination to ancient forests as we explored the time-tested path of coexistence forged by Indigenous peoples and bears. Today, I’m bringing you a story of how that same traditional knowledge is being used in conjunction with ground-breaking modern technology to illuminate novel insights on bear behavior.

We travel again to the brackish estuaries of Knight Inlet in western British Columbia. Here, bear research biologist Dr. Melanie Clapham is working with Silicon Valley software developers and members of the Nanwakolas Council, a group of six First Nations from Vancouver Island and the mainland coast that manages Knight Inlet Lodge, to modify facial recognition software for conservation purposes.

Known as the BearID Project, Dr. Clapham and her team are training their facial recognition program to identify individual bears. This software offers biologists a new, less invasive tool for tracking behavior and population dynamics. At Knight Inlet, First Nations Guardians like Harold Glendale (Da’naxda’xw/Awaetlala) believe this tracking technology can help with stewardship efforts, supplementing the intimacy guardians have with local grizzly populations from being out on the land day-to-day. 

Still being used solely in the Great Bear Rainforest, the BearID Project’s software has potential implications for wildlife studies far beyond. From research on grizzlies in other ecosystems to other species entirely (Dr. Clapham has said she has talked with conservationists about applying it to sun bears, Asiatic bears, wolves, and more), the future of this software is unwritten but promising. 

Facial recognition for bears

Grizzlies around Knight Inlet are the first subjects for a new facial recognition software that could expand scientists’ ability to track individual bears. (Photo by Matt Hart)

Similar in design to artificial intelligence (AI) programs that track people, the bear recognition software pinpoints unique and distinguishable facial features — eyes, nose, ears, etc. — and compares it to other data points to classify individuals. Through deep learning, the software attempts to mimic human recognition capabilities. With an accuracy rate of about 84 percent, according to Dr. Clapham’s 2020 paper published in Ecology and Evolution, the recognition program is proving to be a faster, and automated, alternative to researchers identifying bears themselves. 

Bears can be difficult to differentiate, though people who interact with them regularly are often able to recognize individuals. They lack features that immediately distinguish them, like the pattern of a giraffe’s spots or a zebra’s stripes, and can change considerably in appearance throughout the year. Anyone who’s looked at Katmai National Park and Preserve’s before and after photos of their famous fat bears knows that bulking up for hibernation dramatically alters a grizzly’s features.

As a result of many hours spent exploring Knight Inlet, Dr. Clapham has developed the ability to identify some bears unaided. In an interview with The New York Times, she said that the bears she often confuses were the same as those that proved challenging for the recognition program. This seems to indicate, she said, that the software is successfully adopting patterns of data analysis similar to those of our own neural networks. 

This kind of AI deep learning is ongoing. Raw data comes in the form of trail camera photos and the program becomes increasingly accurate as it receives more inputs.

The automation is a big time saver for researchers, allowing them to comb through larger data sets. Facial recognition from camera footage is also far less invasive than other tracking tools, such as radio collars, which require researchers to traverse backcountry, capture and tranquilize a bear to attach its collar. While it wasn’t designed to replace other tracking methods, the facial recognition software — a shiny, new tool in the box — is expanding the possibilities in this corner of science. 

Technology informing conservation

Trail camera footage provides the raw data for facial recognition software.

In the long term, those involved in the project hope to see it inform bear management and land use plans. It will take at least a few more years for the project to accumulate enough data for a solid baseline. But already the technology is offering novel insights into the life of individual bears, deepening the existing familiarity guardians and researchers have with them. 

Take the Knight Inlet bear called Flora, also known as bear F016. Through the trail cameras, researchers observed her feeding on a sea lion carcass and displaying defensive behavior. With the recognition software, they can easily tell when Flora is observed in future camera footage and analyze her behavioral tendencies, answering questions like, “Does Flora act defensively more often than other bears in the study population?”

A rare and often solitary species, it is extremely pertinent to conservation efforts to understand bears on the individual level. Especially in ecosystems like the Lower 48’s Cabinet-Yaak and Selkirk, where grizzly numbers remain well below 100 bears, the chances of long-term population survival are significantly swayed by the fate of each bear. Identifying movement areas used by multiple bears, or by a particular breeding female, for example, can help identify key pieces of private land as conservation priorities for programs like Vital Ground’s One Landscape Initiative.

Individual behavior tracking could also help prevent bear-human conflicts by informing preventive actions. Over time, researchers could get a sense of what triggers Flora’s defensive behavior, for example, and make choices to minimize her odds of getting into trouble.

But the intimacy of advanced facial recognition doesn’t stop at conflict prevention. Insights about the lives of bears, their relationships with each other, and with the rest of the ecosystem are useful broadly in conservation efforts. They also serve to bolster our care for wildlife. As author Richard Louv so eloquently phrased his maxim of love and familiarity, “We cannot protect something we do not love, we cannot love what we do not know and we cannot know what we do not see.”

With the digital eyes of facial recognition software helping us to see and know, what remains ours is the work of love.

Learn more about Vital Ground’s conservation vision…

Share this post:

Posts by Category

Posts by Month