Anisotropic Bayesian linearized stochastic seismic inversion with multi-parameter decoupling

Abstract

Shale oil reservoir emerges as a significant unconventional energy source, commonly predicted by anisotropic seismic inversion. Considering the intricate nature of shale oil reservoirs, it becomes imperative to consider uncertainties during anisotropic inversion. An effective approach to address this involves stochastic inversion, specifically the anisotropic Bayesian linearized inversion (ABLI), which characterizes statistical and spatial correlations of subsurface parameters through a crucial multivariate correlation matrix constructed through geostatistics. However, an inevitable challenge in stochastic inversion arises from interference during the calibration of statistical and spatial correlations of subsurface parameters. This challenge becomes particularly pronounced in anisotropic inversion, heightened by the multitude of involved model parameters. Existing decorrelation approaches primarily address statistical correlation, neglecting the impact of spatial correlation. To tackle this issue, a novel multi-parameter decoupling strategy is proposed, formulating decoupling anisotropic Bayesian linearized inversion (D-ABLI). D-ABLI introduces an advanced decorrelation approach, and uses principal component analysis (PCA) to simultaneously eliminate impact of statistical and spatial correlations on ABLI. The decoupling enhances the inversion accuracy of model parameters in ABLI, particularly for density and anisotropic parameters. The theoretical underpinnings of the decoupling strategy are demonstrated to be reasonable, and the effectiveness of D-ABLI is proved through a theoretical data test and a field data test regarding shale oil reservoirs. The D-ABLI results offer the capability to estimate fracture density accurately and unveil the distribution of shale oil.