Pre-stack Bayesian cascade AVA inversion in complex-Laplace domain and its application to the broadband data acquired at East China

Abstract

AVA (amplitude variation with angle) inversion plays an important role in the extraction of lithology and pore fluid information of underground media. To make full use of the limited seismic data and recover much richer low-frequency background of elastic parameters, Bayesian inversion framework, linear model regularization and Laplace mixed-domain forward solver are jointed together to put forward the complex-Laplace mixed domain AVA cascade inversion in this paper. Firstly, the Laplace mixed-domain convolution model is deduced and the magnification phenomenon of low frequency corresponding to seismic data in Laplace mixed domain is analyzed detailedly. Besides, the explicit formulations of pre-stack AVA inversion are constructed based on Laplace mixed-domain operator and Aki-Richard AVA approximation. Then, the objective function in Laplace mixed domain based on Bayesian framework is deduced with the linear initial models of P-wave velocity, S-wave velocity and density. It is worth noting that the proposed algorithm can be separated into two stages: (1) the recovery of much richer low-frequency information with complex-Laplace domain AVA inversion and (2) the estimation of final elastic parameters with pure frequency domain AVA inversion approach. The second stage of cascade AVA inversion is restricted with the low-frequency estimation results in the first stage, which can improve the convergence accuracy of the elastic parameters estimation. Finally, the feasibility of the proposed AVA inversion and the reliability of the fluid discrimination are verified by numerous model tests and one field broadband application in China.