Ocean Acoustic Propagation Modeling Using Scientific Machine Learning

Abstract

Ocean acoustic propagation models support a wide range of oceanic applications. Conventional approaches to modeling of acoustic propagation in an ocean environment usually require full environmental knowledge. Unfortunately, such knowledge is very difficult to accurately acquire. Measuring an adequate amount of accurate acoustic data required by conventional data-driven techniques is also difficult and expensive. We propose an ocean acoustic propagation modeling approach that requires very limited environmental knowledge, and small amount of acoustic measurements. By applying the concept of scientific machine learning, we can embed known scientific domain knowledge into data-driven machine learning techniques. One can efficiently learn to predict acoustic fields from much lesser training data compared with conventional data-driven techniques.