2023_programme: Seabed separation using the model manifold of absolute and relative transmission loss

• Session: 04. Inverse Problems in Acoustical Oceanography
  Organiser(s): Julien Bonnel and Stan Dosso
• Lecture: Seabed separation using the model manifold of absolute and relative transmission loss [invited]
  Paper ID: 2057
  Author(s): Neilsen Tracianne, Spendlove Jay, Transtrum Mark
  Presenter: Neilsen Traci
  Abstract: Inferences of seabed properties using geoacoustic inversions depend on both the assumed parameterization in the sound propagation model and the data information content. The ability of an inversion to obtain reasonable seabed properties can be evaluated by using techniques of information geometry. In this formalism, a model is intepreted as a geometric object representing outputs of the forward model for all physically reasonable parameter values. For the current work, forward model computes transmission loss (TL) for a range-independent normal-mode model at a vertical array of receivers. The unknown parameters are sound speed and density of a single sediment layer over a hard basement. Values of the parameters are mapped by the model into a high-dimensional space of predictions. The set of all such predictions form a two-dimensional surface known as the model manifold. The geometry of the model manifold directly connects to the information content of data to the unknown model parameters. Results show how different sediment types parameterized by sound speed and density map to a ribbon-like 2d manifold. The TL model manifolds are then compared to model manifolds for relative TL. The relative TL model manifolds are more compact and the distance along the model manifold between different sediment types is much smaller, indicating that relative TL has less information about the properties of the seabed. The model manifold approach can be used to evaluate information content for various experimental designs and to find appropriate reduced-order models to characterize the impact of the seabed on sound propagation.
• Corresponding author: Dr Tracianne Neilsen
  Affiliation: Brigham Young University
  Country: United States
  e-mail: