Stand in the Nouabalé-Ndoki forest in northern Congo and your view ends a few tree-lengths away. Past that it is trunks and green and the assumption of more trunks. The air is loud — insects, hornbills, the constant tick of falling debris — and none of that noise is telling you the one thing you most want to know, which is what is standing just out of sight.
This is the practical problem at the heart of forest elephant conservation, and it is more absolute than it sounds. A closed canopy hides everything beneath it from above, and a wall of vegetation hides everything from the side. Counting animals you cannot see, across thousands of square kilometres you cannot drive through, is not a harder version of the savanna problem. It is a different problem.
Meanwhile the animals themselves were disappearing at a rate that made the measurement problem urgent. In March 2021 the IUCN assessed the African forest elephant separately from its savanna cousin for the first time, on the strength of new genetic evidence, and listed it as Critically Endangered: a decline of more than 86% over 31 years. The Elephant Listening Project puts the ongoing toll at more than 12,000 killed each year for ivory.
You cannot protect what you cannot count, and you cannot count what you cannot see. So the question became: what else does an elephant give off?
The bioacoustics researcher Katy Payne stood watching Asian elephants at the Washington Park Zoo in Portland, Oregon, and noticed something she could not quite account for: a thrumming vibration in the air. Not a sound she was hearing. A sensation she was feeling — in her chest, in the room.
Her guess was that the elephants were talking below the range of her ears, and that what reached her was the part of it that travels through the body rather than the eardrum. It is a strange kind of hypothesis to act on — that an animal in front of you is making a noise you have no organ for.
Work at the zoo with William Langbauer Jr. and Elizabeth Thomas confirmed that elephants were producing infrasonic calls. Playback experiments on wild African elephants, with Russell Charif, Lisa Rapaport and Ferrel Osborn, confirmed the animals were using them — to coordinate group movement and to locate mates across distances at which no visual or ordinary acoustic contact was possible.
The animals had not been quiet. We had been deaf.
Human hearing gives out somewhere around 20 hertz. People with excellent ears can just detect a tone there; below it, the sensation stops being a pitch and becomes pressure. A man's speaking voice sits between roughly 70 and 200 Hz, a woman's between 140 and 400. The fundamental frequency of a forest elephant rumble is underneath all of that — beneath the floor, in a register we have to borrow instruments to visit.
Playback speed 1 times. Fundamental 15 hertz. Below the floor of human hearing.
There is a reason evolution put elephant conversation down in the basement. Low-frequency sound is far better at getting through cluttered space. High frequencies scatter and are absorbed by leaves, trunks and undergrowth; long wavelengths bend around them. If you are a large animal trying to stay in contact with your family through several kilometres of vegetation, you build your voice out of the frequencies that survive the trip.
On the savanna, the numbers are spectacular. Playback experiments showed elephants responding to each other at 2 km, and because it is hard to reproduce a call as loudly as an elephant makes it, researchers estimated true detection range at around 4 km — meaning a single rumble reaches family anywhere in roughly 50 square kilometres.
Then there is the evening effect, which is the most quietly astonishing fact in this story. As the savanna cools, a temperature inversion forms and acts like a ceiling, bouncing sound back down toward the ground. The listening area swells from about 30 km² at midday to about 300 km² in the evening — a tenfold expansion of the social world, arriving on schedule every day. Savanna elephants make most of their loud low calls during exactly those hours. Whether that is instinct or opportunism nobody can yet say.
And then ELP went and measured the forest, and the forest did not cooperate. Using an acoustic array around a clearing in Gabon to pinpoint where each call came from, they worked out how much energy a rumble loses on its way through Central African rainforest. The average detection distance came out at roughly 800 metres — not four kilometres. When the forest fell quiet, a rumble might reach beyond 3 km. But on an ordinary day, 800 metres.
The culprit was not the trees. It was the noise. A rainforest is acoustically crowded in a way a savanna is not — insects, frogs, primates, rain, all of it competing for the same channel. The limiting factor on an elephant's voice is not how far it travels but how far it stays distinguishable from everything else. That is a genuinely humbling result for a project whose entire method is listening, and ELP published it about their own technique.
The machinery has a wonderfully unglamorous history. ELP's first terrestrial recorders were built into PVC sewer pipe, with a laptop hard drive for storage and circuit boards writing sound straight to binary. Each unit in the project's first major Gabon deployment needed 16 kilograms of batteries to run for three months. Every one of those kilos had to be carried in on someone's back.
The current recorder, the SWIFT, does the same three months on one kilogram. In a landscape with no roads that is not a specification, it is the difference between a project being possible and not. ELP's single biggest expense is not hardware — it is sending a team in to service the grid and collect the data.
Which is worth picturing properly. A 2013 survey in northern Congo took three separate missions to lay out its recorders, each starting from the village of Liouesso, each involving one to three days down the Lingoue River by pirogue — a dugout canoe — before finding somewhere along the west bank dry enough to climb out. Then walking west through the forest, following elephant paths where they existed and cutting through where they did not. The longest mission ran 22 days: one researcher, a chef du mission, and five porters carrying food, camp and the recorders.
What that labour has built is the largest archive of Central African sound in existence — more than 900,000 hours by ELP's own count on its research pages, and described as passing a million on the page for its public database. Birds, primates, insects, frogs, thunder. If it makes a sound, it is in there.
The flagship deployment is a grid: 50 acoustic units placed by stratified random sampling across cells of 25 km², covering 1,250 km² of Nouabalé-Ndoki National Park and the logging concession next door. Slightly more than half the sites sit inside the park. Recordings run 24 hours a day at 8 kHz — a deliberately low sample rate, because the quarry lives at the bottom of the spectrum and low rates buy you months of runtime.
One detail on that public database says everything about the context this work happens in. ELP publishes the grid's outer coordinates but withholds the exact position of every recorder, releasing them only to researchers on request. The reason given is plain: poaching in the area is ongoing.
No recorders placed. 0 square kilometres within detection range, 0 percent of the landscape.
Fifty recorders across 1,250 square kilometres puts something like 3.7% of the landscape inside detection range at any moment. That number tends to land badly, and it should not. This was never surveillance. It is sampling — the acoustic equivalent of a survey transect, and the 5.5 km spacing between units is a negotiated settlement between how much forest you want to hear and how far a team can realistically walk.
The other half of the answer is that the elephants move and the recorder does not. A unit sitting in one spot for four months is not covering 0.9 km²; it is intercepting everything that walks through 0.9 km² for a third of a year. Given enough sites and enough months — 28,786 recorder-days in the flagship study — you can build a picture of how animals use a landscape that no aircraft, and no amount of counting dung, would ever give you.
A microphone left running in a forest for four months does not get to choose what it records. Set one out to catch elephants and it will also catch rain, thunder, chainsaws, and — with terrible clarity — gunfire.
Between December 2017 and April 2021, across that 50-unit grid, ELP and its Congolese colleagues logged 322 gunshots. The distribution is the finding: 98% of them fell in the first two years. Gunshot events declined strongly and consistently across the study period.
On a spectrogram the two signals could not look less alike, and the contrast is almost too neat to be true. An elephant rumble is warm and horizontal — a stack of parallel bands sitting low in the frame, holding for a second or more, its harmonics ruled above the fundamental like staves. A gunshot is a broadband impulse: a single cold vertical slash from the bottom of the frame to the top, over in milliseconds.
The animals are horizontal. The violence is vertical. You can learn to read it in about ninety seconds.
Which is roughly what the panel below asks you to do — the same scanning task an analyst performs, minus the four months of it.
Two things usually happen when people try that. The gunshots are easy — that is why the detection algorithms work and why this signal is useful at all. The rumbles are harder than expected, which is exactly why ELP automates the first pass and then hand-verifies the output. In the 2024 study every gunshot detection was checked by a human and the false positives thrown out, and detector output was manually reviewed across 28,786 recorder-days.
The other thing that happens is you drag the year slider to 3 and the marks nearly vanish, and you feel a small surge of triumph — which is the point at which this story has to be careful with you.
The most valuable thing to come out of the grid was not a headcount. It was a behaviour.
Under pressure, the elephants went nocturnal. In areas where they had reason to perceive risk, activity shifted into the hours of darkness — and crucially, that shift showed up before any change in population numbers could have been measured. In the actively logged concession, nocturnal activity held roughly steady for three years and then jumped in the fourth.
This is the actual product of listening, and it is worth being precise about why it matters. A population estimate tells you what has already happened; by the time a count drops, the animals are dead. A behavioural signal tells you that something is happening now, in a specific place, while there is still something to protect. Elephants shifting their day is an early-warning system that no carcass survey can replicate — and it only exists because something was recording continuously when nobody was there.
ELP is candid that the picture is incomplete. The same study reports seasonal and forest-type variation in nocturnal activity for which, in the authors' own words, they have no explanation, and notes that their advance predictions were too simple for what the data actually did. That is what an honest result looks like at the edge of a method.
The Elephant Listening Project sits inside the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology — part of Cornell University, which holds a four-star rating from Charity Navigator. Katy Payne co-founded it; Daniela Hedwig has directed it since 2021. The science is peer-reviewed, the archive is public, and the limitations below are mostly ones ELP published about itself. Four are worth holding up to the light.
Batteries and data are collected roughly every four months by teams walking in. A gunshot recorded in January may not be read until May. This is the single most important thing to understand about the method, and it is the opposite of what "real-time detection" marketing in this sector usually implies. Acoustic grids of this kind tell you where to patrol next season. They do not interrupt anything happening today.
Turning rumbles into a population figure requires knowing how often an average elephant calls — and that rate can only be measured where you can see the animals, which means at forest clearings where they gather in unusually large, unusually social groups. The authors flag the problem themselves: calling rates "might often be lower when elephants are moving through the forest," in small family groups, with less to say. The conversion factor is doing a lot of quiet work.
That collapse in gunshots is genuinely encouraging, and it is not clean evidence of anything on its own. It coincides with heavy investment by the Fondation Nouabalé-Ndoki in professionalising the anti-poaching programme — so it may reflect better enforcement, or poachers displaced somewhere less watched, or fewer elephants left to shoot. The paper does not claim to separate these, and neither should we.
Detection data guides armed patrols, and in the Congo Basin that raises a question this story cannot dodge. Eco-guards in the region have faced sustained and credible allegations of violence against Indigenous Baka and Bayaka people, including beatings documented by a UN Development Programme investigation. No such allegation attaches to ELP, whose role is scientific — but its partner WCS manages Nouabalé-Ndoki, and better information about where people go in a forest is not automatically neutral. Who holds it, and what they are permitted to do with it, is a separate question from whether the microphones work.
One more discomfort, offered plainly. The strongest argument now made for forest elephants is carbon: their browsing thins small stems, easing competition so surviving trees grow larger and denser, and a 2019 Nature Geoscience study estimated their extinction would cut aboveground biomass in central African forests by about 7% — losses valued near US$43 billion. It is a good argument and it works on people who are unmoved by extinction. It also quietly concedes the frame, pricing an animal by what its absence would cost us. Worth noticing that we reached for it, and worth noticing that a creature whose voice we could not hear for most of our shared history is now being defended in the one language we reliably listen to.
This is independent editorial content — we are not affiliated with the Elephant Listening Project, the Cornell Lab of Ornithology or Cornell University, and we receive nothing from the link below. If you'd like to support the work described, you can give directly.
Donate directly to the Elephant Listening Project100% of donations go to the charity. This site is supported by advertising only.