Estimation of geoacoustic parameters using machine learning techniques

Modal dispersion of broadband acoustic data has been used extensively to estimate the geoacoustic parameters using perturbation techniques (Rajan et al., 1987), non-linear inversion techniques (Potty et al., 2000) and Bayesian inference (Warner et al., 2015). The Airy phase region corresponding the minimum group velocity of the normal modes are extremely sensitive to bottom properties in comparison with properties of the water column. This talk will explore the sensitivity of the group speed minima and the associated frequencies of the acoustic normal modes to bottom parameters. The group speed minima and the associated frequency data will be used to train a machine-learning algorithm. Synthetic data will be generated for various bottoms and it will be used as the training pool. Data collected from New England Bight, East China Sea, and New England Mud patch will be used to test the algorithm. The value added to the a priori bottom parameter information using the algorithm will be discussed in the context of computational cost associated with the algorithm. [Work supported by the Office of Naval Research.]Modal dispersion of broadband acoustic data has been used extensively to estimate the geoacoustic parameters using perturbation techniques (Rajan et al., 1987), non-linear inversion techniques (Potty et al., 2000) and Bayesian inference (Warner et al., 2015). The Airy phase region corresponding the minimum group velocity of the normal modes are extremely sensitive to bottom properties in comparison with properties of the water column. This talk will explore the sensitivity of the group speed minima and the associated frequencies of the acoustic normal modes to bottom parameters. The group speed minima and the associated frequency data will be used to train a machine-learning algorithm. Synthetic data will be generated for various bottoms and it will be used as the training pool. Data collected from New England Bight, East China Sea, and New England Mud patch will be used to test the algorithm. The value added to the a priori bottom parameter information using the algorithm will be discussed in the context...