Informational Geometry to Quantify Information Content of Seabed Parameterizations
Abstract
The proposed work will quantify the reliability, robustness, and uncertainty associated with inferring ocean properties from acoustical data using tools developed for the fields of optimal experimental design and information geometry. We propose to use information geometry to investigate how inference problems can be tailored to align with the information content of the acoustical data. Thiswork will complement ongoing work in the Ocean Acoustics and Task Force Ocean (TFO) programs by informing design of experiments andoptimizations to obtain seabed properties and improve source detection and ranging. Specifically, the information geometry approach will be used to compare the information content for different seabed parameterizations, types of input data, and forward models toidentify parameter sensitivity and parameter coupling. Our previous work has focused on relatively low frequencies (50-400 Hz) using transmission loss computed using normal mode sound propagation model assuming a fluid parameterization for a single sediment layer in a range-independent environment. In the proposed work, we will investigate additional seabed parameterizations over a wider band of frequencies and apply this approach to bottom loss curves as function of grazing angle. These studies will also attack persistent questions in ocean acoustics regarding the inherent uncertainty in parameter estimates. The optimal design of inference problems in ocean acoustics will take advantage of the structure of model manifolds to obtain reduced-order models that minimize uncertainty in inferred parameters without spending unnecessary time and resources trying to identify parameters that are irrelevant for explaining acoustical data. The proposed work can inform future updates to the low and high-frequency bottom loss catalogs used to categorize seabeds across the world#s oceans. The approach of using information geometry to reveal minimal models with high information content will lead to guidelines about optimal experimental designs for maximizing information content subject to real-world experimental constraints and guide the design of more robust machine and deep learning models for Naval applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 09, 2024
- Source ID
- N000142412566
Entities
People
- Traci Neilsen
Organizations
- Brigham Young University
- Office of Naval Research
- United States Navy