BirdNET Turns a Million Hours of Forest Sound Into Conservation Data

Knowing which birds live where is the foundation of conservation, and it is astonishingly hard to do. A traditional survey means a trained biologist hiking out before dawn, listening for a few minutes at a handful of spots, and writing down what they hear. An area the size of a national forest would need centuries of expert labor to survey properly. BirdNET, a free and open source AI developed by the Cornell Lab of Ornithology and Chemnitz University of Technology, collapses that problem. It listens to audio recordings and identifies the bird species singing, at a scale no human team could ever match.

That matters because birds are the most visible indicator of ecosystem health. When forest habitat degrades, sensitive species vanish from the soundscape years before they vanish from satellite imagery. BirdNET turns cheap microphones into a continent scale sensor network for biodiversity, and it is already reshaping how land managers and researchers monitor the natural world.

From Owl Monitoring to a Million Hours of Audio

The breakthrough came from an unlikely starting point: spotted owls. In 2017, ecologist Connor Wood began tracking the declining nocturnal owls across California's Sierra Nevada, a region roughly the size of Vermont. He and University of Wisconsin Madison ecologist Zach Peery deployed what eventually became a network of 1,600 autonomous recording units across the landscape. The recorders ran through the morning dawn chorus, capturing far more sound than anyone could use.

In 2020, that archive became the first large scale test of BirdNET. The AI, trained on thousands of hours of curated animal vocalizations, processed the audio and detected more than 90 species across 1 million hours of recordings. What began as a single species monitoring project had become, almost by accident, one of the most detailed portraits of a bird community ever assembled. Wood, now on the research faculty at Cornell's K. Lisa Yang Center for Conservation Bioacoustics, has said the technology lets ecologists dream big, because a nuanced view of an entire ecosystem is finally within reach.

How the AI Listens

BirdNET is a deep neural network that converts sound into a long string of numbers, effectively a barcode, and matches that pattern against its training data. The freely available model can identify more than 6,500 avian species from a few seconds of audio. The approach is not limited to birds. Because every sound gets a barcode, the same underlying technique has been adapted to track gray wolves, frogs, and toads in the Sierra Nevada.

In September 2026, Cornell released BirdNET Live, a mobile app built for field researchers and conservationists that performs real time species identification on a smartphone. BirdNET Live currently recognizes 8,927 bird species and can identify the sounds of 268 mammals, 254 insects, and 340 amphibians as they happen. A biologist standing in a rainforest at dusk can now know instantly which of dozens of species are calling around them, without a radio, a lab, or an expert on the phone.

From Data to Decisions

The Sierra Nevada work has since evolved into something with direct policy consequences. With funding from NASA, Wood and Peery combine BirdNET acoustic data with satellite imagery that reveals canopy cover and tree height, and feed both into a decision support tool for the U.S. Forest Service in California. The software predicts how a management action, such as a prescribed fire, will affect bird populations of concern, from warblers to woodpeckers. Land managers can compare options in hours instead of seasons, which matters when forests across the American West urgently need restoration treatment.

Conservation groups are applying the same playbook in the tropics. In Panama's Parita Bay, Audubon researchers used acoustic monitors and AI analysis of the recordings and found 27 species previously unknown in the mangrove ecosystem, evidence that helped tie bird protection to carbon rich habitat. With a 2 million dollar grant from the Bezos Earth Fund, the Escucha Aves project is now training community members in Colombia, with planned expansion in Peru and Bolivia, to deploy recorders and run the AI analysis themselves. In one early result, the system detected a Tawny faced Quail more than 60 miles beyond its known range, a discovery local groups now use to promote birding tourism.

Citizens and Scientists on the Same Team

BirdNET also powers the identification features behind birding apps that have introduced millions of people to the birds around them. Cornell's Merlin Bird ID, which shares the Lab's sound recognition technology, has passed 50 million cumulative downloads, and the Lab's eBird platform has now amassed more than 2 billion bird observations from over 1 million users. Every phone that learns to recognize a song is a potential sensor, and every casual listener is a potential contributor to a global biodiversity dataset that scientists peer review and publish on.

The next frontier is whole ecosystem listening. BirdNET Pantanal, a collaboration with the Wildlife Conservation Society, is in development to monitor every vocal species in the Pantanal, the world's largest tropical wetland, from singing birds and croaking frogs to roaring jaguars. It is an ambition that would have been absurd a decade ago, and that open source AI has made ordinary: put a microphone in a forest, come back with an ecosystem.