Optimal transport for full waveform seabed property inversion

Processing waveforms measured at one or more hydrophones in the ocean provides us with a means for conducting geoacoustic inversion. By comparing these waveforms to replica waveforms calculated with a sound propagation model and minimizing the distance between real and replica data, one can identify the most suitable property values for the propagation medium. Typically, distance measures such as the Euclidean distance are employed. However, such measures present us with multiple local minima in addition to the sought after global minimum. Search approaches are challenging to implement, as “trapping” in regions of local minima can hinder global optimization. Optimal transport, relying on the minimization of different similarity measures, provides us with a function that is easier to navigate, being characterized by different properties from those of the Euclidean distance. In this work, we illustrate how optimal transport can facilitate inversion for seabed parameter estimation and we compare results employing optimal transport to those obtained from conventional approaches. [Work supported by ONR.]