Ambient acoustics as indicator of environmental change in the Beaufort Sea: experiments & methods for analysis

• Main Author: Chen, Rui
• Other Authors: Schmidt, Henrik
• Format: Thesis
• Language: English
• Published: Massachusetts Institute of Technology and Woods Hole Oceanographic Institution 2021
• Subjects: Ambient noise; Arctic change; Noise analysis; Arctic; Arctic Ocean; Beaufort Sea
• Online Access: https://hdl.handle.net/1912/27035

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution June 2021. The Arctic Ocean is a vital component of Earth’s climate system experiencing dramatic environmental changes. The changes are reflected in its underwater ambient soundscape as its origin and propagation are primarily dependent on properties of the ice cover and water column. The first component of this work examines the effect on ambient noise characteristics due to changes to the Beaufort Sea sound speed profile (SSP) and ice cover. Specifically, the emergence of a warm water intrusion near 70 m depth has altered the historical Arctic SSP while the ice cover has become thinner and younger due to the rise in average global temperature. Hypothesized shifts to the ambient soundscape and surface noise generation due to these changes are verified by comparing the measured noise data during two experiments to modeled results. These changes include a broadside notch in noise vertical directionality as well as a shift from uniform surface noise generation to discrete generation at specific ranges. Motivated by our data analyses, the second component presents several tools to facilitate ambient noise characterization and generation monitoring. One is a convolutional neural network (CNN) approach to noise range estimation. Its robustness to SSP and bottom depth mismatch is compared with conventional matched field processing. We further explore how the CNN approach achieves its performance by examining its intermediate outputs. Another tool is a frequency domain, transient event detection algorithm that leverages image processing and hierarchical clustering to identify and categorize noise transients in data spectrograms. The spectral content retained by this method enables insight into the generation mechanism of the detected events by the ice cover. Lastly, we present the deployment of a seismo-acoustic system to localize ...