Blind Inversion of Multichannel Nonstationary Seismic Data for Acoustic Impedance and Wavelet

Seismic data inversion for acoustic impedance (AI) has received increasing attention over the past few decades for its key role in estimating reservoir properties. The calculation of the AI from band-limited post-stack data is a type of nonlinear inverse problem. When the wavelet is unknown, it will be a blind inverse problem. We propose an efficient algorithm for multichannel AI blind inversion from post-stack seismic data. To eliminate the discontinuity in the inversion with the existing single-trace processing strategy, we construct a multichannel blind inversion framework which can obtain the AI model and the wavelet simultaneously. For the AI inversion, Tikhonov and total variation (TV) regularization methods are mostly used. However, traditional Tikhonov regularization assumes smooth variation in the AI model, whereas TV regularization creates a staircase effect where AI changes gradually. A new kind of weighted isotropic total variation (WITV) regularization scheme is proposed to preserve the texture of the data and suppress the noise when solving the ill-posed AI and nonstationary wavelet blind inversion within an acceptable time and memory consumption. We develop an optimization method which helps efficiently refine the inverse procedure based on the alternating-direction method of multipliers (ADMM) iterative scheme. Experimental tests show that the new iterative algorithm efficiently converges to a solution of the nonlinear minimization problem. We illustrate the performance and optimality of our inversion algorithm with simulated and field data, which indicates that our algorithm performs well in both the AI inversion and the wavelet inversion, offering an efficient method for processing large-scale nonstationary multichannel data.