Circadian signals in the deep: A machine learning approach to exploring deep-sea choruses

The deep sea has traditionally been considered a quiet and arrhythmic environment. However, recent studies have uncovered unexpected acoustic diversity and phenological patterns in deep-sea soundscapes. Among the most intriguing phenomena are evening choruses—continuous low-frequency sounds that emerge shortly after sunset. These choruses are likely produced by midwater fishes undertaking diel vertical migrations and may function as circadian cues for deep-sea organisms capable of acoustic perception. Despite their potential ecological significance, studies of deep-sea choruses remain largely descriptive, limiting comparisons across regions and depth zones. In this study, we present a machine learning-based approach for the detection and analysis of evening choruses. By using median power spectral density as the acoustic representation and applying a source separation model, we performed unsupervised feature learning and temporal encoding to isolate periodic biological signals. Using this method, we successfully identified recurring evening choruses at multiple sites across the Pacific Ocean, at depths ranging from 277–5500 m. Given their potential role in the chronobiology of deep-sea ecosystems, we advocate for sustained global monitoring of deep-sea soundscapes to investigate long-term changes and understand the role of acoustic cues in structuring deep-sea biodiversity.