Search for a command to run...
Noise from calving icebergs, cracking ice, and melting ice dominate the underwater soundscape of glacierized fjords. The combination of these sources results in one of the loudest recorded ambient ocean environments. Until now, the combined symphony of sounds has made identifying specific sources difficult, limiting its value to provide insights into the physical processes influencing ice-ocean interactions. Here, we show that unsupervised machine learning can separate the signal into five dominant sound profiles related to glacier activity. For this work, we deployed an array of hydrophones approximately 400 m from the terminus of Xeitl Sít’ (LeConte Glacier) in Southeast Alaska and recorded sound at regular intervals between October 2016 and May 2017. Using the k-means clustering algorithm, we cluster spectral shapes of 10,440 background acoustic spectra, defined as the 25th-percentile spectral level of each recording. We identify five distinct acoustic clusters and associate their temporal occurrence with environmental time series including ice movement, local meteorological conditions, and oceanographic data. We further associate the spectral shape and audio signals to known glacier sources such as calving and ice melt. Our analysis reveals that these acoustic clusters correspond more closely with glacier and ice-mélange activity than other environmental variables, confirming the dominance of glacier behavior on fjord soundscapes. This research demonstrates the straightforward application and effectiveness of clustering passive acoustic data and sets a foundation for using soundscapes to monitor and detect environmental changes in glacierized fjords.