Optimal search strategies for nonlinear, multi-parameter geoacoustic inverse problems

Abstract

A mode travel time based technique for geoacoustic inversions has been developed using broadband explosive sources [Potty et al., J. Acoust. Soc. Am. 108, 973–986 (2000)]. In this inversion scheme we search for various parameters including water depth, sound speed in the water column represented by Empirical Orthogonal Function (EOFs), compressional wave speed in the sediment and sub-base, thickness of the sediment and range. The sensitivity of the objective function to perturbations in these parameters is different, i.e., different parameters have different sensitivities. The most sensitive parameters can be estimated with higher accuracy compared to parameters with low sensitivities. In this study we aim to link the search process to parameter sensitivities by matching the fineness of search to sensitivities. The highly sensitive parameters will be searched with a fine scale sampling of the model space with a coarser sampling adopted for less sensitive model parameters. This condition will be incorporated into the search tool (Genetic Algorithm) used for the inversion scheme. The objective of this approach is to utilize the computational resources in an optimum manner by giving more emphasis to parameters which are more sensitive as opposed to a uniform weighting for all parameters. This approach will be applied to data from the ASIAEX-2001 and Shelf Break Primer data sets. [Work supported by ONR.]

Long Range Sediment Tomography in the East China Sea

“…One alternative then is to use the model sensitivity as a proxy to resolution and use it to adjust the model sampling [2,3]. But resolution of model parameters is not often available a priori for most of the practical problems.…”

“…Based on this sensitivity study and the data from geophysical surveys (Miller et al [3]) the sediment was modeled as indicated in Figure 1(b). This sediment model was used in the inversion scheme to estimate the compressional wave speed profile.…”