On a cool morning in early summer, Zhongqi Miao stands behind a tree in the Asian small-clawed otter exhibit at Seattle’s Woodland Park Zoo tinkering with a small camera hidden among the leaves.
The Microsoft AI for Good Lab scientist’s focus is on Little Frei and Connor, the two furry residents clambering over rocks and swimming in a pool, because they’re similar in size and shape to Pacific martens — an animal that has perplexed conservationists for years.
The zoo is piloting a new wildlife monitoring system with built-in AI that could eventually help determine why the Pacific marten, a small tree-climbing creature in the weasel family, has apparently nearly vanished from Washington’s Olympic Peninsula.
Woodland Park Zoo is among 20 organizations to receive a grant earlier this year from Microsoft’s AI for Good Lab to support innovative AI-based projects in Washington state. The zoo is testing SPARROW (Solar-Powered Acoustic and Remote Recording Observation Watch), an AI-powered system the lab developed that collects wildlife data from cameras and acoustic sensors and transmits it directly to the cloud via satellite, enabling researchers to access — and act on — real-time insights.
Designed to enable biodiversity monitoring in remote locations around the world, the open-source platform, announced in December 2024, can be built and used by anyone. The zoo, under its Living Northwest conservation program, hopes to use SPARROW to find and evaluate Pacific martens in isolated areas of the Olympic Peninsula.
The technology is being deployed in other projects in several countries, but the zoo pilot is its first application in Washington, where the technology was developed on Microsoft’s Redmond campus.
“Having innovative partners like Woodland Park Zoo provides us with early insights into how the technology is performing and what could be improved,” says Juan M. Lavista Ferres, chief data scientist and director of the Microsoft AI for Good Lab.
“And starting in our own backyard allows for quicker adjustments in a more controlled setting before SPARROW is deployed in more remote, harsher conditions.”
Real-time detection
The zoo originally planned to test SPARROW to understand how animals live and interact with humans in urban areas, but conversationists realized the technology could be more valuable for the zoo’s Olympic Marten Project, which aims to determine why marten populations on the Olympic Peninsula remain small and how they might be restored.
Those efforts are challenging in part because biologists working on the project retrieve data from wildlife cameras only during summer, given the region’s rugged terrain and winter conditions. If a camera detects one of the elusive martens over the winter, researchers often don’t find out until months later.
“We only get data once a year from the cameras we’ve currently put out,” says Robert Long, senior conservation scientist and director of the Living Northwest program.
“So for us to have feedback about how a population’s doing or whether an animal even occurs in an area, we usually have to wait almost a full year between visits.”
If SPARROW detects martens on the Olympic Peninsula, biologists could quickly go to those sites and set devices to collect genetic samples that can help determine population size and viability. Having an array of remote cameras providing real-time alerts of detections could be key to helping save Pacific martens in the Olympic Mountains.
“Because martens are so rare in the Olympics, this technology would allow us, if we have a detection, to really intensify our efforts to collect a DNA sample,” says Paula MacKay, a carnivore conservation specialist with Living Northwest.
“The potential conservation value really lies in that real-time detection for rare species,” she says. “That allows researchers to take the next step, whatever that next step would be, toward better informing the scientific knowledge about that particular species.”
Accelerating wildlife research
Located on the northwestern tip of Washington state, the Olympic Peninsula is a 3,600-square mile expanse encompassing glacier-capped peaks, mossy rainforests and a rugged Pacific coastline. Martens were once widespread on the peninsula, inhabiting both forests and high-elevation subalpine zones.
But by the 1960s, martens had become rare due to habitat loss, trapping, climate change and other pressures. During the 1990s, there were no Pacific martens sighted on the Olympic Peninsula at all. Then in 2008, a young female marten was found dead on a peninsula trail — the first confirmation of the species in the area in almost 20 years. There were just a few sightings over the ensuing years.
In 2015, Long and a colleague from Idaho Fish and Game collaborated with a Microsoft engineer to develop an automated scent dispenser that could operate through the winter and replace the bait lures traditionally used for wolverine monitoring, which need to be rebaited every few weeks. But rebaiting isn’t always feasible in winter, when remote mountainous areas are often avalanche-prone or inaccessible due to snow.
By contrast, the battery-powered scent dispensers, placed in trees with cameras facing them, could be programmed to emit a small amount of liquid scent lure daily for a year or more without any maintenance.
“It was super successful,” Long says. “It bumped up our wolverine detection rate by three to four times what we had experienced previously.”
The new scent dispenser approach also led to 13 marten detections in Olympic National Park between 2017 and 2019, more than the total number recorded in the previous half-century. With the help of volunteers, Woodland Park Zoo started building the scent dispensers and selling them to other conservation biologists interested in attracting a variety of carnivore species.
Now, Long and his colleagues hope to build on their earlier success by pairing scent dispensers with SPARROW technology. They plan to test SPARROW in Washington’s Cascade Range, where martens are prevalent, then deploy the technology to look for martens on the Olympic Peninsula.
The solution’s AI can be trained to recognize particular species, sifting through millions of images and isolating only those showing a specific animal. That ability, Long says, could make SPARROW useful for the zoo’s monitoring of other vulnerable populations in the Cascade Range, like wolverines and Canada lynx.
“We could program it for the six or seven species we’re most interested in and move our data analysis process far ahead,” he says, “instead of waiting to get a huge dump of data once a year that we then have to analyze.”
Otters and tigers and bears
Back at the zoo, Miao and Rahul Dodhia, deputy director of the AI for Good Lab, installed two additional cameras outside the Malaysian tiger exhibit and the nearby sloth bear exhibit, using zip ties and poles to rig the cameras in locations not accessible by the animals.
The goal, Dodhia explained, is to have multiple cameras communicating with a single SPARROW unit to gauge if the cameras are communicating properly, the system is sufficiently powered and the satellite signal to the Wi-Fi is functioning. The system is programmed to identify only animals, leaving out any images of people the cameras might capture.
“The idea is, how does this SPARROW system perform when it’s going 24 hours a day, every day of the week, because we want it to be operational without any human intervention for at least a year,” Dodhia says. “This is a safe simulation of a wild environment.”
Tools like the SPARROW system and scent dispenser are crucial to the zoo’s emphasis on using non-invasive monitoring methods to study rare and threatened species, says Katie Remine, Living Northwest’s conservation manager.
“This type of technology is so important,” she says. “It allows you to learn about the animals without stressing them.”
‘Transformative potential’
The catalyst for SPARROW emerged through conversations Lavista Ferres had with biodiversity experts at a conference in May 2024.
In talking with those experts, Lavista Ferres realized that the main challenge in collecting biodiversity data wasn’t access to AI models, which many conservationists were already using, but the slow process of manually collecting data. Seven months later, the lab launched SPARROW to enable conservationists to place devices in the most remote regions of the world without needing to physically retrieve data.
“By removing this barrier, SPARROW reduces the data cycle from months to days, or even hours,” Lavista Ferres says. “When you combine this with AI, we unlock transformative potential — not just accelerating the pace of conservation work, but expanding the very kinds of questions and research these experts can now pursue.”
Long and MacKay hope that by providing real-time data, SPARROW might help conservationists determine how Pacific martens can thrive on the Olympic Peninsula as they once did.
“It would probably allow us to do something that we really won’t be able to do otherwise,” Long says, “so it’s pretty important.”
Lead photo: A Pacific marten in the North Cascades. (Photo by Daniel Harrington )