KATHMANDU — Bhadai Tharu says he still remembers the day a tiger attacked him. It was around 1 p.m. on Jan. 6, 2004. The encounter left him blind in his left eye.
More than two decades later, the 59-year-old doesn’t speak ill of the animal that attacked him.
“When incidents, like the one I was involved in, happen, people get angry,” says Tharu, who is now chair of the Gauri Community Forest on the fringes of Bardiya National Park, a global hotspot for the endangered big cats.
“Tigers have their role in the ecosystem, and both people and tigers should coexist,” Tharu says.
His perspective comes as Nepal’s tiger population reaches a record 429 individuals, with Bardiya alone hosting 112. To avoid human-tiger encounters such as the one Tharu experienced, researchers turned to a survival skill long practiced by local people: listening to the forest. And for a while, at least, it seemed to have worked.
The pilot project, documented in a recently published study, tested an artificial intelligence-powered system that “listens” to alarm calls from chital deer (Axis axis) and maps areas where tigers (Panthera tigris) or leopards (Panthera pardus) may be present and that humans should avoid.
A forest that speaks
Long before researchers arrived with sensors and machine-learning models, Indigenous communities in Bardiya had already been listening to the forest.
“Footprints, scat, and alarm calls from deer, langurs, red monkeys all tell us that a tiger may be nearby,” Tharu says.
In his experience, the warning often begins with rhesus macaques (Macaca mulatta), known locally as red monkeys. “They become restless first, and then the deer respond,” Tharu says.
According to study co-author Ramjan Chaudhary, a field biologist with the Bardia Nature Conservation Club, a local community organization, interpreting alarm calls has long been common among nature guides, forest guards, and residents living near wildlife habitat.
Co-author Casey Anderson, a U.S.-based wildlife filmmaker and conservationist who led the deployment of the project, says the team sought to automate knowledge that communities already make use of.
“We weren’t teaching the forest or the people who live there something new,” he says. “In many ways, we were teaching a computer to recognize something those people already understood.”
The researchers deployed acoustic monitoring devices throughout the Dalla Community Forest neighboring Bardiya National Park, and carried out their predator-response experiments there and in the Khata Community forest, including parts of the Khata International Biological Corridor linking Nepal and India.
Turning alarm calls into data
Rather than tracking tigers directly, the researchers focused on how prey species respond to them.
“Sometimes the best way to find a predator isn’t to look for the predator itself. It’s to listen to everything around it,” Anderson says.
Using a tiger model, the team tested whether prey species produced alarm calls when confronted with a predator. Chital deer, rhesus macaques, and Tarai gray langurs (Semnopithecus hector) consistently responded with alarm calls during encounters with the model, suggesting that those vocalizations could act as indicators of predator risk. Researchers also noted that prey don’t always vocalize when predators are present, meaning alarm calls may not capture every encounter.
The team then trained a lightweight AI model to recognize chital deer alarm calls on solar-powered monitoring devices, which cost about $60 each.
When an alarm call was detected, the system transmitted information through a wireless network and generated a map highlighting areas where tigers or leopards might be present. The researchers deployed 10 of the monitoring units throughout Dalla Community Forest for three months.
“The ultimate objective was never simply to locate tigers. It was to reduce dangerous encounters between people and predators,” Anderson says.
Shared landscapes
These forests are used by both wildlife and people.
Residents collect firewood, harvest fiddlehead ferns known locally as niuro, and other forest products while grazing livestock in areas also used by tigers, leopards, elephants (Elephas maximus), and rhinos (Rhinoceros unicornis).
“People use the forest for grass and fodder,” Tharu says. “They keep pigs and goats, and their livelihoods depend on the forest.”
Human-tiger conflict remains a persistent challenge. A recent study found that tigers attacked 245 people in Nepal between 2010 and 2024, killing 141 of them, while another documented 18 deaths and 10 injuries around Bardiya National Park between 2019 and 2023.
Communities hail promising technology
For local forest users, the technology made immediate sense as it reflected knowledge they were already using. Thenghchu Tharu, a 60-year-old forest guard at the Dalla Community Forest who helped monitor the pilot project, says he’s relied for years on alarm calls from chital deer to gauge predator presence.
“When chital repeatedly call out, we know there may be a tiger or leopard nearby,” he says.
During the pilot project, he compared the system’s alerts with signs observed on the ground: “It matched exactly.”
When alerts suggested tiger presence in a particular area, residents could be warned to stay away temporarily.
“We would tell people there is a tiger there and not to go because it could be dangerous,” Thenghchu Tharu says.
He adds he believes the system could reduce risky encounters for people entering the forest to collect fodder, gather firewood, or graze their livestock.
“If people are informed beforehand, they do not need to enter that area,” he says.
Study co-author Chaudhary says communities responded positively to the project and were “enthusiastic about seeing this new technology in action.”
Bhadai Tharu, the tiger-attack survivor, shares that view. “We were happy to see it. It was very good.”
He says he believes the technology could save lives. “It can save people from attacks. When people are informed beforehand, incidents can be reduced.”
Why the pilot ended
Yet despite the positive feedback from communities and encouraging results, the system was never expanded in Nepal.
“We expected it would still be implemented, but it never happened,” Bhadai Tharu says.
Several community members interviewed for this story expressed disappointment that the project ended after the trial phase. Anderson says the problem was not a rejection of the technology itself.
“I wouldn’t characterize it as Nepal deciding not to pursue the technology or as a failure of the concept,” he says.
Instead, he says, the project struggled to secure the staffing, funding, and local institutional support needed to move beyond a proof of concept.
“As the initial project’s resources began to run out, we weren’t able to establish the local staffing and university involvement necessary to keep it operating,” Anderson says.
The equipment was later moved to India, where researchers had an opportunity to continue testing and refining the concept.
“There was an opportunity to continue exploring and developing the concept in India, so moving the equipment allowed the work to continue rather than simply ending when that particular phase in Nepal concluded,” Anderson says.
Officials have taken a more cautious view. Sabnam Pathak, information officer at Nepal’s Department of Forests and Soil Conservation, says new conservation technologies need long-term evaluation before wider adoption, adding they should only be deployed “after a thorough study.”
The study itself acknowledged several limitations, including false detections, equipment failures, and overheating problems. Alarm calls can indicate predator risk, but can’t reliably distinguish whether a tiger or a leopard triggered them. In some cases, deer fled without vocalizing, meaning predator presence could go undetected.
Shyam Kumar Thapa, program manager at the Nepal office of the Zoological Society of London (ZSL), says the challenge goes beyond protected areas.
“The most serious concern is the escalation of these interactions into human fatalities and retaliatory killing of tigers,” he says.
He adds the technology could help shift the management of human-wildlife conflict from a reactive approach to a more proactive one by helping authorities and communities anticipate risk before dangerous encounters occur.
Anderson says he believes technology can help amplify those efforts.
“It isn’t artificial intelligence replacing human knowledge,” he says. “It’s artificial intelligence helping us listen more effectively to the natural intelligence that has always been there.”
Banner image: A tiger in Bardiya National Park, 2021. Image by AP Photo/Niranjan Shrestha.
Nepal plans park for ‘problem’ tigers as attacks raise concerns
Citations:
Kershenbaum, A., Markham, A., Root‐Gutteridge, H., Smith, B., Anderson, C., McClaughry, R., … Dassow, A. (2026). An autonomous network of acoustic detectors to map tiger risk by eavesdropping on prey alarm calls. Remote Sensing in Ecology and Conservation, 12(4), 661-672. doi:10.1002/rse2.70061
Dhakal, B. K., Lamichhane, B. R., Kandel, S., Subedi, N., Low, M., & Silwal, T. (2026). Tigers in the neighborhood: Factors associated with fatal tiger attacks in a human‐dominated landscape, Nepal. Ecology and Evolution, 16(8). doi:10.1002/ece3.74206
Shrestha, P., Kadariya, R., Subedi, N., Paudel, U., & Bahadur K. C., R. (2025). Claws and consequences: Human–tiger conflict and community responses in Bardiya National Park, Nepal. Frontiers in Conservation Science, 6. doi:10.3389/fcosc.2025.1623056
Cite this article
Praveen Kumar Yadav (2026). In Nepal, AI builds on local knowledge to help prevent tiger attacks. Mongabay Conservation news. DOI: https://doi.org/10.66709/news-333197
