Coupled Bayesian Data Assimilation and Learning for Underwater Acoustics

Accurate acoustic propagation relies on the knowledge of properties of the ocean-seabed environment, such as sound speed, bathymetry, and geoacoustic parameters. Yet, these properties are often difficult to directly measure, thus requiring the use of efficient inversion algorithms that extract all the pertinent information from commonly sparse and interdisciplinary ocean measurements. In this work, we present a partial differential equation (PDE)-based technique for Bayesian nonlinear data assimilation and learning to assimilate ocean-acoustic measurements and jointly infer the unknown properties of the ocean environment, the bathymetry and geoacoustics, the acoustic field, and even the model parameterizations themselves. Idealized test cases and realistic examples are showcased, in which we highlight the principled joint nonlinear inversion of the ocean physics-acoustics-bathymetry-seabed properties.