Decorrelated linearized seismic-petrophysics inversion

Abstract

Seismic model and rock-physics models (RPMs) are of significance in petrophysics inversion to estimate petrophysical properties from seismic data. Nevertheless, the prediction of petrophysical properties is commonly restricted by two issues. One is that computationally intensive algorithms are required to invert petrophysical parameters coupled in most nonlinear RPMs. Another is that statistical error caused by correlation characterization between different petrophysical parameters can affect the accuracy of inverted results. To solve these two issues, a decorrelated linearized seismic-petrophysics inversion (DLSPI) based on a linearized seismic-petrophysics model is proposed. First, a novel linearized seismic-petrophysics model is constructed based on the Taylor first-order approximate RPMs including Nur's critical porosity model and Gassmann's equation. Next, a linearized seismic-petrophysics inversion (LSPI) is proposed under the linear Bayesian framework. This approach can estimate petrophysical parameters from seismic data directly and efficiently. Then, a decorrelation strategy based on principal component analysis (PCA) is introduced to the LSPI to form the DLSPI, which can avoid the effect of the statistical correlation between different petrophysical parameters and improve inversion accuracy. Finally, the proposed method is validated both on synthetic seismic data and real seismic data from a sandstone reservoir. The inverted results indicate the efficiency and stability of the proposed method.