Learning how to listen to the ocean

Shallow ocean environments are diverse. Variations in water depth,sound speed, and the ocean floor affect how sound propagates making it difficult to identify where sound originates. Models for sound propagation are based on ray tracing, normal modes, and parabolic equations. These models predict the frequency-dependent sound field for a set of environmental parameters when the location of the sound source is known. The inverse problem is actually of more interest: finding the location of a sound source when you don’t know the exact ocean environment. Optimization algorithms have tackled this problem for many years. The future, however, lies with machine learning and deep learning. Machine learning will likely be sufficient for source classification when a fairly good estimate of the environment is known, but deep learning will be needed to address complexities associated with uncertain ocean properties.