PEEC Nature Club members who developed a 3-year ‘PumaGuard’ project combining wildlife biology, AI, and field engineering, from left, Gavin Bent, Seb Koglin, Phoebe Reid, Celia Pesiri, Adis Bock, Aditya Viswanathan, Suchi Jha, and Tate Plohr. Not pictured is Zoe Bent, who graduated in 2025. Courtesy/PEEC
By MARK MACINNIS
Los Alamos Daily Post
At a recent evening presentation at the Los Alamos Nature Center, a group of Los Alamos High School (LAHS) students in the Pajarito Environmental Education Center’s (PEEC) Nature Club unveiled something far beyond a typical high school science club project: a working artificial intelligence surveillance system already evaluated to reduce conflicts between humans and predators such as pumas.
The students shared the presentation with an audience that clearly marveled at their progress. Developed over the last three years, “PumaGuard” combines motion-sensitive cameras, AI machine learning, and automated deterrents to identify and respond to mountain lions in real time.
The system was prototyped locally, field-tested with researchers at the University of California, Davis, and has drawn interest internationally and from New Mexico cattle ranchers seeking solutions to livestock losses. As one student said, “We are using innovative AI in a niche area—helping humans and wildlife coexist.”
Coexistence, Not Conflict
The project began with a growing local issue. Los Alamos, situated along the edge of the Jemez Mountains, has seen increasing encounters between residents and mountain lions—known interchangeably as pumas or cougars. Livestock losses, particularly goats at the North Mesa stables, have become a recurring concern.
Efforts by the New Mexico Game and Fish Department in capturing and euthanizing individual animals have proven ineffective. As the presenters said, “If you remove a dominant puma in its territory, a younger one quickly takes its place—so the problem doesn’t go away.” Inward migration of pumas as well as resident mountain lions on the mesas and canyons continues the cycle.
Students framed the problem differently: how might non-lethal deterrence of predators direct them away from humans and their livestock or pets? There are other prey animals around Los Alamos, including the ubiquitous mule deer, of which every driver is wary.
PumaGuard System
At its core, PumaGuard is a simple but powerful workflow sequence: a motion-triggered camera captures an image, an AI trained vision model determines whether the subject is a puma, and if confirmed with a Raspberry Pi computer, it activates a deterrent such as sound, light, or other stimuli.
Workflow of PumaGuard Response. Workflow diagram showing the PumaGuard detection and deterrence sequence using cameras, AI classification, and Raspberry Pi control hardware. Courtesy/PEEC
Behind the system is a sophisticated vision learning approach. Students implemented a two-stage workflow: first detecting whether an image contained an animal, then classifying whether that animal was a puma.
This approach reduced false alarms from snow, vegetation, or other animals tremendously. “We trained our AI-model on thousands of trail camera puma images, provided by a collaborator at New Mexico State University (NMSU), which is why it performs so well.”
The Nature group shipped the entire prototype system to UC Davis researchers for field testing, at road-kill baited prey sites (state permitted). The field testing achieved about 92 percent correct Puma identification in 800 imaged encounters, with no false positives reported. The deterrent system—using sounds and lights controlled by a Raspberry Pi computer—responded incredibly fast: “From motion detection to identifying a puma and deploying a response—it all happens in about six seconds,” students said.
PumaGuard Validated Puma Behavior gives a surprise
In Los Alamos, the prototype system successfully detected and deterred wildlife, including mountain lions and ring-tailed cats. At a backyard pond, pumas drawn to drink water, which the system sensed and activated a playback of puma mating calls. In an electric moment captured on video the recorded calls prompted a puma to approach the speaker and respond with vocal mating responses.
The mating call interaction was attempted since it took about 6 months of nightly surveillance to observe pumas and their responses at this pond. More observations showed that a puma fled when deterrent sounds broadcast and retreated even faster when the same puma (identified by facial scars) returned to the pond, which triggered the deterrent noises. These studies provided proof of concept of the “bait and deter” approach.
International Recognition & Applications
The project has gained international recognition. LAHS Students were among four project winners (of more than 350) in a global competition at the NeurIPS AI conference in 2024. The conference drew researchers from major universities and companies, including Google, Meta, and Apple. The group and collaborators later published their work in the peer-reviewed journal, Sensors. Interest is expanding, including outreach from a New Mexico cattle ranch exploring the PumaGuard use during the spring calving season.
Vision
Plans include expanding into other species that are apex predators in various locations (wolves for example) and adding mobile Wi-Fi alerts to livestock owners for real-time notifications. The remarkable achievement of these students was not simply a case of them using AI. They built, trained, validated, and deployed a practical monitoring and deterrence system. More broadly, PumaGuard demonstrates how artificial intelligence might enable better coexistence between humans and predator wildlife worldwide.