Eavesdropping on Animals: Can Bioacoustics Help Save Species?
--
From mountaintops to the ocean floor, scientists are using sound to study the living world
The first time Vijay Ramesh saw the Western Ghats, he was a child. His family would travel hours from their home in Bangalore, one of India’s largest cities, to spend the weekend on the edge of the lushly forested mountain range.
Years later, Ramesh returned to study the region’s birds. The diversity of habitats and native bird species made the Ghats an ideal place to study how climate change and land use have affected the feathered locals. But Ramesh had a problem.
“In a lot of the forests I work in, it’s very hard to see the birds. You often hear them before you see them,” says Ramesh, who recently completed his doctoral studies at Columbia University. “Why not use microphones instead of binoculars?”
Ramesh is one of a growing number of scientists who use sound to study the natural world. Researchers are eavesdropping on frogs and fish, elephants and earthworms, and many hope what they hear can inform and inspire conservation action around the world.
Sound-based surveys can help answer basic questions: What animals are here, and why? Long-term monitoring might be able to track shifts in ecosystems from climate change or other pressures. And bioacoustic studies could help evaluate conservation efforts as they happen: It may be possible to hear a habitat being restored.
In recent years, recording devices have become cheaper and readily available, making it easier to capture vast amounts of sound. Advances in machine learning — artificial intelligence — are helping scientists make sense of it, although what to do with hundreds or even thousands of hours of sonic data remains a major obstacle.
Bioacoustics made its first big splash over 50 years ago, when a 34-minute vinyl LP became a surprise hit. Released in 1970, the record introduced the public to a sound few had ever heard: the songs of humpback whales.
The album was an instant classic. It was even included, seven years later, on the Golden Record aboard NASA’s Voyager spacecraft. Back on Earth, the album spurred huge interest in whales, helping kickstart the U.S. environmental movement — and the field of bioacoustics. But the first recordings of humpbacks were an accident.
“It pretty much all started during World War I and II,” says Holger Klinck, director of the K. Lisa Yang Center for Conservation Bioacoustics at Cornell University.
The first people who used electronics to listen to the ocean were naval sonar operators. They weren’t hoping to hear whales, but enemy submarines. Songs of the Humpback Whales featured recordings made later by biologists Katy Payne and Roger Payne with the help of a navy engineer.
Klinck has a copy of the album in his office, along with the flag of his home country, Germany, and a license plate from Oregon, where he holds a dual appointment at Oregon State University’s Marine Mammal Institute. A bookshelf hosts a colony of figurine seals — the animal that launched his career in bioacoustics.
In 2003, a leopard seal killed a marine biologist in the waters of Antarctica. After the attack, researchers wanted a better way to keep track of the animals, which can grow to over 10 feet and 1,000 pounds.
“I built a system that divers could use to detect leopard seal calls in real time. If there’s a leopard seal around, you may want to reconsider whether or not you go diving,” says Klinck — though he notes the 2003 incident is the only known fatal attack by the species.
For decades, scientists’ microphones — hydrophones — stayed in the sea, in part because sound travels more efficiently in water than in air: It takes many more recorders to cover the same area in a forest than it does in the ocean. But Klinck says technological improvements in the last decade have made recording on land much more affordable, enabling researchers to launch bioacoustics projects everywhere from deserts to the jungle.
In the Western Ghats, humans live in close proximity to wildlife — including tigers, leopards and elephants — as well as hundreds of other species, many of which live nowhere else. Plantations for growing tea and other crops have fragmented the forest, reducing habitats but providing income for people.
Conservation efforts have been underway for decades. Ramesh collaborated with an Indian environmental organization that has been restoring forest in the Ghats for over 20 years, one plot at a time. These sites now host many of the plants that define the local rainforest. Ramesh wanted to see if rainforest animals had also returned.
To find out, Ramesh identified 43 sites, which included undisturbed forested areas as benchmarks, actively restored plots, and formerly deforested areas that had been left to naturally regrow without intentional restoration. Ramesh placed audio recorders at each of the sites, then waited to hear what they captured. Ramesh knew he would focus on birds. Highly vocal, their sounds are easy to capture, and the presence of birds that live only in rainforest is a useful proxy for habitat restoration. But when Ramesh first got into ecology, he says, “I was never really interested in birds.”
He’s come around — a shift he credits in part to the Malabar whistling thrush, an electric blue bird about the size of a crow, also called “the whistling schoolboy” for its distinctive call.
“It’s a wonderful species. I have no bad things to say about this bird,” Ramesh says — except maybe that they begin whistling before dawn, when he’s trying to sleep before a day of field work.
Using audio recorders instead of conducting surveys manually allowed Ramesh to monitor a far larger area for far longer than he otherwise could have managed — and to be less disruptive.
“I don’t need to be disturbing bird species. I don’t need to make my presence felt,” he says. “I can just leave a recorder out there and collect data for hours on end.”
Ramesh captured over 5,000 hours of audio — a strategy he won’t repeat.
“It was so much,” he says, making it difficult even to store the data. But to learn anything from the sounds, he needed to analyze them, identifying which of hundreds of native birds could be heard at any given moment.
In the end, Ramesh sampled 70 hours of his audio, chosen randomly from recordings made during the “dawn chorus,” when birds are most vocal. Ramesh’s research assistant, Akshay Anand, listened to it all, carefully noting every species.
Spectrograms — visual representations of sound — make the work easier, but Ramesh says to identify each bird in a 10-second clip can take as long as 15 minutes. Sometimes the difficulty is the number of different birds calling at once. Ramesh has recorded as many as 20 species in a 10-second clip. But individual identifications can also be a challenge.
“You’re constantly playing the clip over and over again, trying to see if this particular whistle is a dove or a mountain imperial pigeon,” Ramesh says.
Kristin Brunk, a postdoctoral researcher at Cornell University, can relate. “It quickly gets unmanageable,” she says.
Brunk is working on a project to monitor birds in the Sierra Nevada to see how drought and wildfires could affect them in years to come. It’s a massive undertaking, using 2,000 recorders spanning the entire mountain range, deployed from May to July, capturing 10 hours of audio a day.
That’s much more than a human could feasibly process. Luckily, Brunk and her colleagues have powerful computers and an algorithm custom-built to identify birds by the sounds they make: BirdNET, available to the public as a free smartphone app, which can currently recognize over 3,000 species of birds.
Brunk and her team chose species of particular interest to represent distinct habitats: the fox sparrow for shrubby areas, the acorn woodpecker for stands of oak, the iconic California spotted owl for old-growth forest. A favorite of Brunk’s is the mountain quail, which favors steep, rocky areas a few years after fire.
“They’re very secretive,” but quite vocal, she says. “Bioacoustics is the perfect way to study them.”
Brunk’s team runs the algorithm on the million or so hours they recorded over the summer, then checks the computer’s work. It usually does very well, but some birds present special challenges. Confirming the cries of goshawks, for example, has proved difficult.
“Steller’s jays are really good mimics,” Brunk explains.
The jays produce pitch-perfect imitations of the birds of prey. So perfect, the team ultimately had to catalog many cries as “possible” goshawks.
Brunk has encountered a variation of the phenomenon before, while doing her Ph.D. research near Santa Cruz. In that region, Steller’s jays make calls of red-shouldered hawks. It’s not fully understood what purpose the jays’ mimicry serves. It could be to scare other birds or to attract a mate. Brunk says even for very well-studied birds, it’s hard to say for certain why they do what they do.
“Maybe that’s just the nature of wildlife research,” she says. “What makes it so interesting is that we never really know what they’re thinking.”
For now, Brunk’s goal is relatively modest: to determine which birds are living where. She hopes with more years of recordings and other environmental data, she’ll get a clearer picture of why certain birds prefer particular habitats. And it’s a safe bet that long-term monitoring in the Sierra Nevada will give her the chance to study the effects of drought and wildfire.
But Brunk sees plenty of limitations, beyond copycat corvids muddying the data. The recordings can only tell her if a species is present — whether it’s a single individual or dozens. A computer program that can take head counts from sound alone is a tall order, but Brunk says it would be a boon for conservationists. She also wants an algorithm that can distinguish adult birds from their offspring.
“You can know all the habitats where a bird is living, but if you can get an idea of where it’s breeding more successfully, you can really focus in on those habitats for protection,” Brunk says.
Ramesh’s work in the Western Ghats shows thoughtful conservation can make a difference for vulnerable species. After the laborious process of manually identifying the birds in his recordings, Ramesh found clear evidence that the birds most suited to rainforest habitats — the species most threatened by deforestation — were more likely to be found in actively restored sites than in plots left to naturally regrow. That suggests the restoration efforts are succeeding at providing needed habitat. Still, after more than 20 years of maintenance, several rainforest birds are missing. Played back-to-back, the soundscape of a restored site is noticeably less dense than that of undisturbed forest.
And when Ramesh compared his recordings visually, he found a surprise: Actively restored and naturally regrowing sites look strikingly similar. Neither type of site has much activity above 10 kHz — toward the upper range of what’s audible to humans — while in undisturbed rainforest those frequencies are richly filled.
Sounds that high-pitched aren’t coming from birds — but insects. It’s a discovery Ramesh never would have made without the recordings
“Maybe insects are much more sensitive to changes in habitat structure and microclimate,” he says. The return of rainforest birds is encouraging, but a more precise measure of habitat restoration could come from a different creature entirely.
Ramesh hopes one day his work will help identify areas most important to protect. But processing recordings by hand is too slow — to be useful to policymakers and land mangers, he’ll likely need machine learning. Unfortunately, it won’t be as simple as plugging his recordings into BirdNET.
Existing algorithms don’t work well in most rainforests, partly because the recordings used to build the models came largely from North America and Europe. Tropical habitats also tend to have more species — and the noises they produce — which make the soundscapes harder to parse.
Still, Ramesh is hopeful. He recently began a postdoctoral position at Cornell, where he joins Klinck, Brunk and dozens of other researchers working on bioacoustics.
The technology is getting better all the time. Brunk sees a future where scientists can ask more complicated questions of the sounds of tomorrow — and today. New advancements could glean more from audio already captured and archived.
“These recordings have been made, and we have them now,” Brunk says. “As technology improves, we can go back and revisit them and get more and more information out of them.”
Perhaps more cynically, Klinck sees their recordings as museum specimens, relics of a vanishing world: A child born 30 years from today may never see an old-growth Douglas fir forest, or the spotted owls that call it home, but if they want to, they can hear them — along with recordings from habitats around the world.
“We’re conserving them for future generations,” Klinck says. “These soundscapes are being altered for good.”
