Technical Paper: Sparse, Accurate Seismic-Tomographic Residual-Covariance Matrices Using Orthogonal Wavelets

Society: SEG
Paper Number: 2012-1518
Presentation Date: 2012


We treat seismic tomography in the guise of high-dimensional Bayesian inference with associated prior and posterior probability distributions. Given significant advances in recent years in the reliability of the model prior covariance-matrix estimate and the model-to-data map (tomography operator), it is incumbent to improve the residual covariance matrix R as well. More realistic R is essential in well-constrained tomography, to contribute to a more accurate model-posterior covariance-matrix estimate for uncertainty quantification, scenario experiments etc. (Osypov et al., 2011). It will also enable better model building in situations where there is so much data heterogeneity that a single overall residual-covariance value, or even variable variances of many separate picks (i.e., diagonal covariance), neither sufficiently represent correlation between data sources. We present and evaluate the benefit of a new approach in seismic tomographic uncertainty quantification.

Request More Information