Data covariance estimation for geoacoustic inversion

Abstract

The estimation of data-error correlations to form a full covariance matrix that is explicitly incorporated into the inversion procedure, is presented. Matched-field geoacoustic inversion is based on estimating a model of seabed geoacoustic parameters by matching complex acoustic pressure fields. Neglecting significant error correlations represents the data as more informative than they actually are and leads to under-estimating parameter uncertainties. Incorporating full covariance matrices in the inversion leads to essentially uncorrelated standardized residuals.