Optimization and Bayesian approaches to geoacoustic inversion
Date
2002
Authors
Lapinski, Anna-Liesa Salome
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis presents the results of applying two geoacoustic inversion methods, an hybrid optimization algorithm and a Bayesian sampling algorithm, to data from environments with range independent and range dependent geophysical and geometric properties. The hybrid algorithm combines the local method of downhill simplex with the global method of simulated annealing, in an adaptive algorithm. The Bayesian inversion algorithm uses a Gibbs sampler to estimate properties of the posterior probability density (PPD), such as mean and maximum a posteriori parameter estimates, marginal probability distributions, highest posterior density intervals , and the model covariance matrix can be calculated. The methods were applied to noise-free and noisy benchmark data for several shallow ocean environments. An appropriate model parameterization is unknown for many of the environments, which increases the difficulty of the problems. An under-parameterized approach was applied to determine the optimal parameterization. The model solutions were estimated well given the varying sensitivities of the parameters. The Bayesian inversion method provided complete solutions including quantitative uncertainty estimates to the inversion problems, while the hybrid inversion method provided parameter estimates in a fraction of the computation time.