Bayesian matched-field geoacoustic inversion

Stan E. Dosso


A Bayesian approach to matched-field geoacoustic inversion, with emphasis on rigorous uncertainty estimation Bayesian formulation, is discussed. Integration can be carried out using Markov-chain Monte Carlo importance sampling methods, such as fast Gibbs sampling. The data uncertainty distribution, which defines the likelihood function, must include both measurement errors and theory errors. Wide bounds were applied for the geoacoustic parameters to limit the inversion to physically reasonable values but allow the acoustic data to determine the solution. The KS test indicated no significant evidence against the assumption of Gaussian-distributed error processes, while the runs test suggested that the estimated data covariance matrices accounted for much, but not all, of the data error correlations.


Acoustics; Covariance matrix; Functions; Monte Carlo methods; Parameter estimation; Uncertainty analysis; Geoacoustic inversion; Likelihood functions

Full Text:



  • There are currently no refbacks.