Junco hyemalis subspecies analysis of song, genetic, and geographic data

The dark-eyed junco (<i>Junco hyemalis</i>) has experienced rapid phenotypic diversification, resulting in subspecies with distinct plumage that reside in partially overlapping regions. The learned songs of these subspecies might gradually accumulate changes, but if song is used for subspecies recognition during mate selection, song differences may facilitate reproductive isolation. Here, we use community-science recordings to explore subspecies-level song variation. We build a machine learning classifier to assess song distinguishability between subspecies. We find that the subspecies have significant song-feature differences. Notably, subspecies pairs with adjacent ranges that do not hybridize have more distinguishable songs and more evidence for genetic differentiation than pairs that do hybridize. Thus, song distinguishability appears to have some predictive power about which subspecies have reported sightings of hybrids, suggesting the possibility that song might reinforce certain subspecies boundaries more than others. Finally, we compare song variation to genetic and geographic data to characterize the evolutionary landscape of the dark-eyed junco. Weak geographic signal in the song and genetic data indicates that individuals who share a range might be more likely to share song characteristics and be genetically similar. Our machine learning analyses of dark-eyed junco songs demonstrates that song differences might be informative about the strength of reproductive boundaries between subspecies, since greater song distinguishability correlates with observed hybridization events and genetic differentiation.