Speaker: Zoi-Heleni Michalopoulou (New Jersey Institue of Technology, NJIT)
Title: Inverse problems in ocean acoustics: particle filtering and linearization for parameter estimation
Abstract: Solving the inverse problem in ocean acoustics, that is, obtaining from received acoustic fields parameters that are related to source location and the environment is a complex non-linear problem. Efficiency and accuracy depend on the nature of the available data and the forward model employed in the inversion. Full-field methods (such as Matched Field Processing) have been used in the past to solve such problems but those can be prone to errors due to the parameterization of the propagation medium and are often computationally inefficient. To circumvent the use of full fields, we develop a method that extracts ray path arrival times from received fields; particle filtering is the foundation of the approach. Probability density functions of arrival times computed via particle filtering are propagated backward through the proposed inversion process, which is conducted after linearizing the forward model. We discuss the computation of the Jacobian matrix, which is necessary for the linearization process. With the sequential filter providing full probability density functions of arrival times, we are able to obtain full probability densities of the unknown parameter, which enable uncertainty quantification in the inversion process.