Animal Tracking Is Getting a Makeover
4 min read
This article was originally published by Hakai Magazine.
Some wild animals are relatively easy to study. Certain penguin populations, for instance, are so unaccustomed to large predators that they barely fear humans and will often wander right up to scientists lurking nearby. Namibia’s brown hyenas are the opposite. These roughly one-meter-long mammals—more closely related to mongooses than dogs—live in small clans but usually travel and scavenge alone. They roam mainly at night and tend to skirt even the most cunningly placed camera traps. That’s if, like the hyena cubs that devoured the pair of cameras that the hyena researcher Marie Lemerle had positioned outside their den, they don’t destroy them outright. “They managed to open the metal case and then chewed on the camera, so even the SD card was finished,” says Lemerle, a researcher with the Brown Hyena Research Project.
So when staff from the U.S.-based nonprofit WildTrack reached out earlier this year to find out if Lemerle would be interested in collaborating on the development of a new automated hyena-identification system, she was enthused.
Zoe Jewell, a British conservationist, has spent the past 13 years helping WildTrack develop an artificial-intelligence-powered system to identify animals from pictures of their footprints. The work was inspired by Jewell’s experiences working alongside Zimbabweans tracking black rhinoceroses. So far, the AI tool can identify 17 animals, including leopards, lions, and rhinos. But the WildTrack team’s goal is to produce more fine-grained assessments—teaching their machine-learning system to identify which individual animal left which print.
For the past five months, Lemerle has been building up a reference library of hyena tracks for WildTrack’s training data sets. Each time she finds a clear hyena footprint at Baker’s Bay, a breeding ground for Cape fur seals on Namibia’s Atlantic coast, where brown hyenas come to hunt, Lemerle reaches for the 30-centimeter ruler in her backpack, lays it on the sand beside the print, and takes a photograph with her smartphone.
Then the WildTrack team, headquartered at North Carolina’s Duke University, analyzes the footprint’s size and shape in intricate detail. They break each print into 120 different measurements, which the machine-learning software can compare with others in the database to look for a match. Sometimes, Jewell says, all they need to tell hyenas apart are subtle differences in the angles between their toes.
Although innate physiological differences set hyena tracks apart, so too do the scars of life. Like Hunger Games tributes trying to reach the Cornucopia, brown hyenas wanting to reach the seal colony in Baker’s Bay during daylight hours have to run a gantlet of other hyenas and mobs of black-backed jackals intent on stealing their prey. They receive grisly injuries: shredded ears, gashed necks, and occasionally a severed foot. Some hyenas limp with broken legs. “If each individual has a different limp, that probably has to show somehow on their tracks,” Lemerle says.
The AI-powered tool should, one day, be a huge complement to more traditional study methods, Lemerle adds. “It would be very nice in the early morning if I take photos of the tracks and see who was there,” she says.
The tool, Jewell says, should give Lemerle a better idea of where individual hyenas are going and how they’re using their environment, without necessarily having to see them.
Wesley Gush, a graduate student at the University of Pretoria, in South Africa, who was not involved in the research, has studied brown hyenas using camera traps at the Bubye Valley Conservancy, an expansive wildlife reserve in southern Zimbabwe. “Brown hyenas are one of Africa’s more cryptic large carnivores,” Gush says, adding that their elusive nature can belie their true numbers.
“The development of an automated tool would have significant potential for assisting wildlife researchers and managers,” he says. “It would be amazing if it works.”
Beyond aiding field researchers, the team at WildTrack hopes the system will help protect wild brown hyenas and other imperiled species.
Fewer than 3,000 adult brown hyenas reside in Namibia, out of fewer than 10,000 across southern Africa. The animals are considered near threatened; the species suffers from collisions with vehicles and revenge killings by livestock farmers. Jewell says WildTrack’s machine-learning system and associated smartphone app could be used, for example, to prove that tracks found near farms aren’t those of a brown hyena, which could reduce the number of retaliatory attacks.
“The model that we develop for [Lemerle] could be used anywhere to help protect brown hyenas,” says Jewell. “That’s the hope.”
