Quantifying Contributions of Sonar Systems on Marine Soundscapes: A Statistical and Machine Learning Analysis

Soundscapes are used to better understand marine environments. They are characterized by both natural and anthropogenic sounds. Unlike sonar systems used for military purposes, the potential contributions of ocean-mapping sonar systems on marine soundscapes are insufficiently understood. This research studies the contributions of the Kongsberg TOPAS PS40 sub-bottom profiler and the RESON SeaBat T51-R multibeam echosounder to soundscape metrics, including amplitude, impulsivity, uniformity, and periodicity of the soundscape to compare during exposure to pre-exposure and post-exposure periods across different frequency ranges. Statistical analyses were used to determine if there were any significant changes in the soundscape between exposure periods. Additionally, machine learning applications, including a convolutional neural network and a transformer, were evaluated to determine their ability to predict the most recent time since exposure using soundscape metrics as inputs. Both the statistical and machine-learning analyses were helpful in quantifying potential contributions of the ocean-mapping sonar systems on the environment.