Machine learning reveals relationships between song, climate, and migration in coastalZonotrichia leucophrys

Abstract

Vocalization behavior in birds, especially songs, strongly affects reproduction, with differences in vocalizations between populations often interpreted as evidence of divergence. Previous work has demonstrated that the White-crowned Sparrow (Zonotrichia leucophrys) has multiple dialects, including between neighboring (and genetically distinct) subspecies nuttalli and pugetensis. However, it is unknown whether the divergence in their songs correlates to environmental or geographical factors. Previous work has relied on limited amounts of data because traditional methods to study bird songs rely on the manual annotation of song spectrograms into individual syllables. Here we explore the performance of automated machine learning methods of song annotation, which can process large datasets more efficiently, paying particular attention to the question of subspecies differences. We utilize a recently published artificial neural network to automatically annotate hundreds of White-crowned Sparrow vocalizations across two subspecies. By analyzing differences in syllable usage and composition, we find that nuttalli and pugetensis have significantly different dialects. Our results are consistent with the interpretation that these differences are caused by the reduction in the number of shared syllables in the White-crowned Sparrow repertoire, as reported by some previous research. Our large sample size enables us to demonstrate that divergence in dialects is correlated with environmental difference, but not with geographical distance, and is also associated with migratory status. Our findings support the hypothesis that the evolution of vocalization behavior is affected by both biotic and abiotic environments, in addition to population structure.