Informational Geometry for Optimal Design of Geoacoustic Inversions
Abstract
The dependence of passive SONAR performance on the ocean environment is of Naval relevance. In both source localization efforts and geoacoustic inversions, a stronger measure is needed of the information content in the data about individual parameters. Fundamenta lly, this study will uncover the effective physics that govern sound propagation in underwater environments using optimal experiment al design and information geometry. This technique will allow for the design of geoacoustic inversions that do not spend time and r esources trying to identify parameters that are irrelevant for explaining acoustical data. The application of the information geome try to geoacoustic inversions will indicate the types of input data (including frequency range) that will provide reliable estimates and low uncertainties for parameters of interest. The proposed work will use tools developed for the field of optimal experimen tal design and information geometry to quantify the reliability, robustness, and uncertainty of geoacoustic inversions. The expe rimental design, in this context, refers to the combination of input data, environmental parameterization, and the cost or error fu nction used in an inversion. We propose to augment traditional optimal experimental design, based on the Fisher information, with t ools of information geometry, i.e., statistical techniques that combine information theory with differential geometry. Information geometry generalizes local methods based on the Fisher information matrix, i.e., the Cramer-Rao lower bounds, using tools such as ge odesics, likelihood profiles, and topological analyses. Combining these techniques will 1) yield new insights into the global struc ture of the inversion search space, 2) identify the optimal type of input data and model parameterization to obtain environmental pr operties of interest, and 3) provide improved uncertainty estimates in the inferred values. This systematic evaluation of geoacoust ic inversions will advance understanding of how the ocean environment affects ocean acoustics and how acoustics can be used to deter mine properties of the ocean environment. Three primary tasks will be accomplished. First, a Fisher information approach will be a pplied to quantify the information content in ocean seabeds using both acousto-eleastic and visco-elastic grain shearing parameteriz ations. This approach will be used to compare the information content for different experimental designs for geoacoustic inversions . Second, the Fisher information approach will be extended to calculate geodesics on the model manifold. The resulting manifold bou ndary approximations will used to obtain minimal models for the seabed parameterizations. Third, the optimal experimental design te chniques and the minimal models will be applied to data from the Seabed Characterization Experiments in the New England Mud Patch. The use of information geometry to reveal minimal models with high information content can guide the design of future geoacoustic in versions, experiments, and machine and deep learning technologies for Naval applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 20, 2021
- Source ID
- N000142112761
Entities
People
- Traci Neilsen
Organizations
- Brigham Young University
- Office of Naval Research
- United States Navy