Geological structure-guided hybrid MCMC and Bayesian linearized inversion methodology

Abstract

Seismic inversion is a common method for hydrocarbon reservoir characterization, as it consists of a proven and effective approach to derive elastic properties from reflectivity seismic data. Markov Chain Monte Carlo (MCMC) based seismic inversion approach is a suitable choice to numerically evaluate the posterior uncertainties associated with the inverse solution without assuming linear forward operators, Gaussian, or generalized Gaussian prior models. However, the existing MCMC based seismic inversion approaches are mostly performed trace-by-trace, which means that the spatial coupling of model parameters is not considered. When the results of trace-by-trace based inversion are combined to generate a 2D profile, the final results will be laterally discontinuous. Moreover, the large dimension of the model space causes low convergence efficiency of MCMC-based seismic inversion. To overcome these issues, a geological structure-guided hybrid MCMC and Bayesian linearized inversion (BLI) methodology for seismic inversion is implemented. The geological structure information obtained using plane wave destruction (PWD) is incorporated to the MCMC based inversion algorithm in the form of dips yields more geologically meaningful results. The hybrid MCMC and BLI strategy, which takes advantage of BLI's high efficiency to provide initial configuration for MCMC, is used to improve the convergence of MCMC-based inversion. Additionally, the block coordinate descent (BCD) algorithm is introduced to replace the large-scale matrix solution in geological structure-guided, and consequently reduce memory consumption and time cost. This methodology is validated on a synthetic seismic dataset, as well as on a real case. It has proven to be a reliable approach to obtain acoustic impedance (AI) from post-stack seismic data in an efficient way. It also addresses the uncertainty related with the ill-posed characteristics of the inversion methodology itself.