The eastern black rail, a subspecies of bird that’s scattered in clusters along American coastlines from Texas to North Carolina, is one of the most elusive creatures a wildlife biologist can choose to study. Adults are the size of a human fist. Chicks are the size of golf balls. Both are dark and shy, the better to hide among the dense marsh grasses under which they skulk and scurry. The birds are maddeningly difficult to lay eyes on. Search YouTube, which has hours of video of pretty much everything, and you’ll find just eight videos of eastern black rails, the longest of which is barely over a minute.

Hardcore birders try in vain to check this subspecies (Laterallus jamaicensis jamaicensis) off their life lists; those who do usually hear the bird rather than see it. Professionals do little better: Christy Hand, a biologist with the state of South Carolina who’s been studying the eastern black rail for nearly 10 years, has seen only a handful, each for just a few seconds.

The secretive birds can cast a sort of spell. Bernd Heinrich centered a family memoir, The Snoring Bird, on his specimen-collecting father’s yearlong quest to capture a closely related species, the vanishingly rare Celebes rail (Aramidopsis plateni). Heinrich calls rails “an embodiment of the wild and inaccessible nature hidden from human senses…. They are otherworldly birds, because they live where few people venture, and where those who do seldom see or hear them…. There is always more than is seen, and they come to dwell in the imagination.”

Hand agrees. For the past decade, she’s pursued these minute birds throughout her study area in the Ashepoo, Combahee, and Edisto (ACE) basin (named after the three rivers that meet there), one of the largest undeveloped estuaries along the U.S. Atlantic coast. She is still unsure exactly how many eastern black rails live there—a few score? a few hundred?—but they’ve inhabited her mind.

“People working with rails,” she says, “we tend to be a bit quirky. I dream about the rails. I don’t dream about wood storks so much because I can just look at them. But with the rails, I’m always wondering what’s going on. I get these little glimpses and I think my brain’s trying to fill in the gaps.” In her mind she hears their distinctive kickee-doo call.

The call is a great boon to Hand’s work, as it’s the easiest way to locate the birds. Yet taking census of rails by sound is problematic, because the birds don’t seem to have anything like a set schedule of when they are most vocal. Day, night, dawn, dusk? They seem to favor no hour. This makes it tricky to design a traditional point-count study, in which trained birders detect bird presence by going out to look and listen for short periods at a precise time of day in a particular location. In addition, the rails often move from place to place as water levels in their marsh habitat change, so you can easily miss them even in a spot you know a family of rails frequents. 

To get a handle on these birds, Hand has had to try new ways to find them and learn how they live. For a while, she would go into their habitat and play a recording of a black rail, then listen for replies. She’d get some. But she wished for something better and less intrusive, partly out of concern that the artificial calls might disrupt the birds’ normal social behaviors.

About five years ago, the falling prices of autonomous recording units, or ARUs (also known as passive acoustic monitors, or PAMs), made it practical to leave recording devices in the field to capture all the racket near a given spot.

Then the problem became how to go through all that audio. Listening to it in real time would require hundreds of hours of painstaking work by people skilled at identifying birds by call—not really an option for a small team in a modestly funded state operation. So Hand tried a couple of computerized sound-analysis systems known as call recognizers. These artificial intelligence tools examine sonograms of field recordings to try to distinguish, amid the layered and jumbled visual representation of an outdoor soundscape, the sonographic shape of a certain bird’s call. In Hand’s case, that was the eastern black rail’s distinctive kickee-doo, caught amidst the clangor not just of other birds but all the bugs, frogs, cars, and planes that fill the marsh air with noise.

The first call recognizer she tried showed promise, but in trials it produced a lot of false positives (that is, it misidentified many bird calls that were on the tape) and required a convoluted workflow that had her team double-checking results. She needed something better.

Meanwhile, in late 2020, Hand heard of a new call-recognition app called BirdNET, created by a young programmer named Stefan Kahl of the Cornell Lab of Ornithology. The app creates a so-called deep neural net—an artificial intelligence algorithm that mimics the human brain’s decision-making routines. It excels at discerning, in complex sonograms like those in Hand’s recordings, the distinctive shapes of individual species’ songs, calls, grunts, buzzes, croaks, barks, and hisses. It streamlined into one process the multiple steps of Hand’s previous toolset and could analyze hundreds of hours of sound in just a few days.

Today you can download to your smartphone a similar piece of software as part of the birding app Merlin, also developed by the Cornell Lab. Its “Sound ID” feature is built on the same foundation as BirdNET. You can watch it work while holding your phone in your hand. Find a quiet place with a few birds, hit the “Sound ID” button, and across the top half of your screen scrolls a sonogram of the surrounding soundscape, while on the bottom half of your screen appear the IDs of any birds currently calling within earshot.

In the woods around my home in Vermont, I’m likely to get red-eyed vireos, song sparrows, hairy woodpeckers, common yellowthroats, perhaps an ovenbird. In, say, Monterey, California, you might get an acorn woodpecker, a chestnut-backed chickadee, a Bewick’s wren, an oak titmouse, or hawks from high above, betrayed by their keening. The app recognizes multiple typical songs and calls—some as brief as a quick but distinctive chip—for each species it’s trained in. Sometimes the screen fills with the identifications of several birds at once, for the app can tell even overlapping songs and calls from one another. Sometimes the app identifies birds you can’t even quite hear.

BirdNET turned out to be just the fast and easy bird-ID tool Hand was looking for. With Kahl’s help, she has used her existing recordings of eastern black rails to train the app to recognize those calls, and she can now measure black-rail presence in a fast, efficient, and low-impact way.

This past spring she acquired 16 ARUs that her team arranged across six wetland areas in the ACE basin, where habitat management for rails was already ongoing or planned. The battery-powered units, each the size of a small paperback book, record several hours a day in the morning and evening. The sound goes onto SD cards that Hand pulls once per month to harvest the data, which is backed up and then sent to Kahl at BirdNET for processing. And for the first time, Hand is gathering enough data from enough points that she’s getting a better feel for the rail’s distribution. The data is still not broad enough to enable her to estimate the population. But it’s rich enough to help her evaluate the effects of the department’s management tools, such as adjusting water levels in impoundments and strategically burning vegetation to create rail-friendly ground cover. The data will only become richer as she collects it in subsequent years.

This conservation bioacoustic work ranges from the closely focused, like Hand’s rails, to the expansive. Research teams are using recorders to study the dynamics of aggressive loon calls (who knew?); monitor shipping lanes for whales so vessels can reroute around them; observe how fish and coral larvae locate their home reef by perceiving its distinctive sound; detect the presence of crop-devastating insects before they proliferate; and monitor protected lands for the sounds of illegal logging and hunting. They can even perform a sort of quickie health exam on some ecosystems. In a simple but powerful discovery, researchers conducting a study for the Nature Conservancy in Indonesia found that they could estimate the diversity of species in a Bornean forest just by analyzing how much of the sound-frequency spectrum was occupied by animal vocalizations.

The granularity of this type of data—its density in both space and time—is the key to its new power. The ability to keep tabs on species’ presence in many places constantly, rather than occasionally, essentially creates a giant new instrument exquisitely sensitive to change at both population and ecosystem levels. It’s as if a large collection of lenses were merged into one giant lens that could see things the smaller ones could not.

One day this past June, I took a walk with ecologist Aaron Weed around his workplace, the steeply hilled, sweetly forested 600 acres of Marsh-Billings-Rockefeller National Historic Park in Woodstock, Vermont. This park’s woodlands, managed sustainably for well over a century, are part of a nearly contiguous forest stretching across some 30 million acres in northern New England and New York State—an area known here as the Northern Forest. In the 1800s, settlers and sheep farmers cut this huge forest hard, so that a landscape once around 85 percent forested became 80 percent cleared. Since then, partly through care and partly through benign neglect (as the farmers moved west to better farmland), the woods have rebounded. They now cover some 90 percent of the terrain.

This forest, however, today faces a century in which a different human-made force, climate change, will alter the species composition and health of this forest in broad, fundamental, and sometimes troubling ways. Weed is part of a team creating a regional system of bioacoustic observations to track—and help predict and manage for—these coming changes. This venture, the Northeast Temperate Inventory and Monitoring Network, coordinates studies and data for 13 National Park Service holdings in the Northeast, from Acadia in the far north to a small holding in Morristown, New Jersey.

Perhaps the busiest user of these bioacoustic tools and techniques is the Cornell Lab of Ornithology’s Center for Conservation Bioacoustics, a 30-year-old department that has long helped lead the field. The Lab today is rapidly expanding thanks to a $24 million endowment from philanthropist Lisa Yang in 2021 (which changed the Center’s name to the K. Lisa Yang Center for Conservation Bioacoustics). Plenty of bioacoustics conservation projects have been and are being done without Cornell’s involvement. But its experience, funding, and tool development have made the Center one of the biggest players in the field.

And its largest project so far stands out as a prime demonstration of the discipline’s growing power and potential. This project began in 2017 high in California’s Sierra Nevada with a simple question: Was the barred owl, an invasive species, a major factor in the ongoing decline of the native California spotted owl? The project would help answer that question decisively—and also lay the groundwork for a much broader investigation in the Sierras.

The question of whether barred owls threatened spotted owls in general was not a new one. Earlier in the 2010s, teams in both the Cascades of Oregon and the coastal mountains of northern California had shown that barred owls—which were originally limited to eastern North America until changing land use and other factors allowed them to expand westward beginning in the 1800s—were bullying and driving out northern spotted owls, a subspecies already famous for losing old-growth habitat to logging. One team of researchers, led by the late Lowell Diller, who was a biologist with the Green Diamond Resource Company, a timberland owner, had shown in 2016 that if you removed barred owls from the disputed area (by shooting them), spotted owls would soon reoccupy that area and resume successful breeding. But that first experiment, the paper concluded, needed to be confirmed by additional studies; if that happened, the authors predicted, “this could be the foundation for development of a long-term conservation strategy for northern spotted owls.”

Having heard from the Washington team about the Center for Conservation Bioacoustics, Peery and Wood contacted director Holger Klinck to see what he would recommend. Klinck said he might have just the thing: The lab had recently started making lightweight, easy-to-deploy, relatively inexpensive autonomous recorders (which the lab would soon brand Swift recorders) that could be left in the field for weeks. Made in a large shed on the grounds of the Cornell Lab of Ornithology, these units cost around $250—about 70 percent less than similar ARUs had just a few years before. Perhaps more important in making the project viable, the Center had also created a software package, called Raven, that could be calibrated to recognize specific bird songs and calls amidst the forest soundscape.

Having reviewed data on typical spotted and barred owl densities, the sizes of their territories, and available habitat, as well as project budget, the researchers concluded that obtaining a representative sampling of the birds’ distributions would require some 100 units rotated through 167 sites. The plan was to have Wood and five field technicians take the Swifts, strap them to trees scattered in a randomized pattern around the spotted owl’s critical habitat over two consecutive summers, and see if the barred owls were indeed taking over the spotted owl’s terrain. If so, the uncertainty would be resolved and the bloody but necessary large-scale removal of an invasive species could be carried out.

Despite the early state of the technology, Klinck (who was also new, at least to the Center) felt confident it would work. His only real concern, which he kept mostly to himself, was logistics—whether, in short, the team could deploy and manage so many ARUs at once. Not to mention the fact that the 28-year-old Wood had never worked with audio equipment or audio-processing software, managed a field crew, or, for that matter, studied owls.

What Wood and his team lacked in experience, they made up for with energy and determination. For weeks, up several thousand feet in the Lassen and Plumas National Forests in northern California, they set up the Swifts; deployed them; visited them a week later to switch out their batteries, remove their SD cards, and swap in fresh ones; then moved them to new sites. By the end of the first summer, the team returned to Madison having collected some 49,000 hours of forest noise, which Wood then ran through an audio processing tool he developed in Raven.

Finally, after manually reviewing all apparent detections, Wood and his team were able to produce a precise map of every “occupancy,” or sonic appearance, of a spotted or barred owl. The map showed that within this enormous study area, which included most of the California spotted owl’s critical habitat, barred owls had encroached upon spotted owls in about 8 percent of viable forest.

The next year, 2018, the research team expanded its operation with more surveys and equipment. They surveyed every site three times to obtain more data, returning that fall with an additional 145,000 hours of audio. That produced another map of the owls’ territories that Wood and his team could compare to the one from the previous year. The picture was not pretty: The second map showed that the barred owl had expanded its share of the spotted owl’s preferred habitat to 21 percent—a 163-percent increase in a single year.

That’s not necessarily the end of the spotted owl story, of course. Studies have shown that barred owls sometimes return to areas where they’ve previously been removed, and most experts agree that the species is in the Pacific Northwest to stay, making repeated removals necessary. And some biologists worry that barred owls are simply so aggressive that such measures will fail in the long run.

In the meantime, though, the two-year study by Wood and colleagues showed that the new bioacoustic tools can gather actionable data at a large scale in a short period of time, and capture real-time dynamic events like the displacement of one species by another.

In the summers since, the team has expanded its Sierra Nevada project. Working with even larger crews, Wood and his collaborators have installed Swifts in a gridwork spread across not just the former study area in northern California but most of the Sierra Nevada range—an area roughly 400 miles long and from 50 to 80 miles across. They’ve deployed some 2,000 Swifts, four times as many as in the original project.

The team is also broadening the scope of its research: This time they will monitor some 80 bird species, as well as wolves and the Yosemite toad, which breeds only in the spring snowmelt high in the Sierras. The project should enable them to track not just species of particular concern but also how various animal populations and ecosystems respond to the impacts of climate change—both gradual and sudden—as well as our efforts to manage habitats and mitigate those impacts through such measures as forest thinning and prescribed burns.

The growing ease and falling cost of bioacoustic technologies are creating a wildlife and ecosystem science that’s not just faster and more powerful but more inclusive. Several Yang Center programs, for instance, are providing training and equipment for local communities at remote sites in Asia, Central America, Africa, and other biodiverse spots so that researchers there can do work driven by indigenous interests and concerns. These investigations range from the welfare of key species to the enforcement of laws regulating logging, construction, and hunting.

At all of these locations, Yang Center researchers are now turning some of their most remote projects into learning centers to give Indigenous research communities deep grounding in the operations and possibilities of these tools. The Yang Center calls this “capacity building,” and it’s one of the three pillars of the Center’s mission, along with research and technology development. For a long time, says director Klinck, this capacity-building pillar was underfunded. The $24 million gift from Yang in 2021 solved that problem.

“If we really want to have a global impact,” says Klinck, “we cannot be the bottleneck. It’s not scalable if people can’t do it for themselves.” Klinck and his Yang Center colleagues want to enable researchers of all origins and paths to pursue the many kinds of investigations the newer, more powerful bioacoustics tools make possible. To this end, Yang Center biologists are creating yearlong programs in the use of ARUs, BirdNET, and other bioacoustic tools, as well as study design, so that local collaborators are able to both use those tools themselves and teach others to do so. The goal is to not just develop collaborators but to mentor a new generation of independent researchers.

In Africa, for instance, the Yang Center’s long-running Elephant Listening Project uses ARUs to eavesdrop on elusive forest elephants and identify gunshots of poachers. It has helped train an independent team in Republic of Congo that is running crucial aspects of a 50-site acoustic grid in the 1,514-square-mile Nouabalé-Ndoki National Park, which is connected to other protected areas. Working with the Wildlife Conservation Society, the Elephant Listening Project is also establishing an “analysis and training hub” in central Africa with solar power systems and high-performance computers to process the acoustic data, as well as staff that can run the algorithms to detect elephants and gunshots. This hub will serve as a central processing and training center for other field teams running bioacoustic projects around the region, and will invite students from the region to work on independent research projects there. Here again, the idea is to replace the Yang Center as the hub, moving analysis capacity into the hands and lands of local people. Yang Center teams in Central America and Indonesia are launching similar efforts, with investment in equipment and long-term training programs.