Same Birds, Different Song

Machine learning uncovers the impact of climate and geography on birdsong.

This one species of bird is probably the single most well-studied bird in terms of how it sounds, but we don't know how or why its song changes through time and space.

Language changes all the time: splitting into new dialects, morphing into slang and spreading ways of speaking from one place to another. This phenomenon holds true for animals as well as humans, as Kaiya Provost, PhD, assistant professor of biology at Adelphi, knows well from her research into the evolutionary biology of birds.

Birdsong is an incredibly complex mode of communication. In scientific terms, the song of the White-crowned Sparrow includes whistles, trills, buzzes and a category for "other" sounds known as "special notes." Individual songs help birds find mates and signal their location to other birds.

Two closely related subspecies of the White-crowned Sparrow-a bird that Dr. Provost calls "incredibly cool"-have retained distinct dialects of their own despite living side by side in the Pacific Northwest. "This one species of bird is probably the single most well-studied bird in terms of how it sounds," she said. "But we don't know how or why its song changes through time and space."

Dr. Provost's enthusiasm for the White-crowned Sparrow rubbed off on Jiaying Yang, an undergraduate from The Ohio State University, now a PhD student at Vanderbilt University, with an interest in bioacoustics. Together, the two developed a project that applied machine learning technologies to vast datasets of White-crowned Sparrow recordings, hoping to prove a theory: that changes in birdsong are the products of a changing environment.

Their findings, which used recordings held by The Ohio State University's Borror Laboratory of Bioacoustics Database, were published as a paper in the April 2024 issue of Ornithology: "Machine learning reveals that climate, geography, and cultural drift all predict bird song variation in coastal Zonotrichia leucophrys."¹

White-crowned Sparrows can be found all over North America, with some variants breeding in the Arctic and wintering in Mexico. Other variants, like some of the Pacific Northwest subspecies explored in Dr. Provost's paper, stay roughly where they are year- round. While most subspecies keep to their own territory, two subspecies in this particular part of the Pacific Northwest frequently overlap. This overlap, which occurs in what is called a "hybrid zone," allowed Dr. Provost to pinpoint the causes behind changing birdsong.

Though previous methods of collecting and cataloguing birdsong worked well, they were laborious, requiring researchers to manually check and label data. This meant that immense amounts of data stored in databases like the Borror Lab too often went underutilized. But with machine learning tools that can be given detailed instructions about what to find and learn, smaller teams can be far more efficient. "Now we can process these things much faster and in much bigger batches than we used to," Dr. Provost said.

Out of a dataset of 20,000 syllables of whistles, trills, buzzes and special notes of White-crowned Sparrow, Dr. Provost and Yang manually annotated 2,000, then had the machine learning model run the rest. Once the songs were tagged, separating out the syllables, the two could cross-reference that data with information about climate, geography and time. Their conclusion, as the paper notes, indicates a correlation between changes in song and "[cultural drift], geographic distance, and climatic differences, but the response is subspecies- and season-specific."

One subspecies of White-crowned Sparrow, Z. l. nuttalli, which lives in Southern California and does not migrate, showed a limited influence from climate-related factors on its birdsong. However, the other subspecies, Z. l. pugetensis, which is migratory, showed a much larger influence. "And when you group the two together, the impact of climate goes through the roof," Dr. Provost added. Varying climate up the West Coast likely plays a major role in the relationship between the two subspecies-which, Dr. Provost says, might explain why they have held on to their identities as separate groups over time, even though they interbreed in the hybrid zone between their habitats.

Dr. Provost is looking forward to using the same machine learning tools on a much larger scale. Her current project deals with 137 species of sparrows, instead of just one. "Machine learning gives us a powerful way to get a lot of data very accurately and very quickly," she said. "Doing it by hand would probably be more accurate in the end, but it would take us 20 years."

Read more in the 2025 issue of Academic & Creative Research Magazine, where we highlight the innovation and imagination shaping Adelphi's academic community.

1 Yang, J., Carstens, B. C., & Provost, K. L. (2023). Machine learning reveals that climate, geography, and cultural drift all predict bird song variation in coastal Zonotrichia leucophrys. Ornithology, 141(2). https://doi.org/10.1093/ornithology/ukad062

About Kaiya Provost, PhD

Kaiya Provost, PhD, is an assistant professor of biology in the College of Arts and Sciences. Her primary specialties are evolutionary biology of birds, with a particular focus on phylogenomics, singing behavior and bioinformatics.