Estimation of environmental parameters with machine learning using a compact tetrahedral array and sources of opportunity

Abstract

In a previous paper, we showed that we could localize sound sources using a compact tetrahedral hydrophone array in a continental shelf environment south of Block Island, Rhode Island. The tetrahedral array of phones, 0.5 m on a side, was deployed to monitor the construction and operation of the first offshore wind farm in the United States. Directions of arrival (DOAs) for a number of ships were computed using a time difference of arrival technique. Given the DOAs, ranges are estimated using supervised machine learning techniques. Here, we extend this work to estimate a number of environmental parameters including water depth and sediment composition. Training sets of range-dependent ocean waveguides and sediment sound speeds were generated using a propagation model for a neural network. Data from the tetrahedral array were processed by the neural network, which provided estimates of the water depth and sediment parameters such as sound speed and density. These estimates are compared to bathymetric data and core data collected as part of the site characterization for the wind farm. [Work supported by the Office of Naval Research and the Bureau of Ocean Energy Management.]