This dissertation applies a Bayesian framework for making quantitative statistical inferences about geoacoustic properties from ocean acoustic data using matched-field processing techniques. Data acquired during the ASIAEX 2001 East China Sea experiment are used to infer the geoacoustic properties.; In a Bayesian approach, information and uncertainty regarding model parameters obtained from the measurements are summarized in the posterior probability distribution. This posterior distribution is proportional to the product of a prior distribution (which incorporates information on model parameters before the measurements) and of a likelihood function (which quantifies how well a model fits the measurements). From this posterior distribution of model parameters, we obtain all information about the model parameters, such as maximum a posteriori estimate (best-fit model), mean as well as standard deviation.; The quality of the best-fit model is checked using matched-field processing for source localization. In the less than 1 kHz frequency band, the effect of environmental mismatch on source tracking can be reduced by using inversion techniques to estimate geoacoustic parameters, resulting in improved source localization performance. The parameter uncertainty (in terms of mean and standard deviation) given by the Bayesian approach is validated by comparing the variabilities of the estimated parameters inverted from multiple independent data sets.; A Bayesian approach to inverse problems requires estimation of the uncertainties in the data. An extension of the Bayesian parameter uncertainty analysis to include the uncertainty of data errors is carried out. Following a full Bayesian methodology, we derive the analytic expressions for the posterior probability distribution of the model parameters for both single and multi-frequency data.; The impact of uncertainty embedded in the geoacoustic inversion results on the estimation of transmission loss is investigated. An approach for estimating the statistical properties of transmission loss is developed using information on the model parameters obtained from the inversion. The utility of this approach is that one can compute the probability distributions of transmission loss at all frequencies, ranges and depths. Examples demonstrate the use of transmission loss probability density functions to extract characteristic features such as median and lower/upper percentiles.
展开▼
