Accurate and efficient velocity estimation using Transmission matrix formalism based on the domain decomposition method. (1st February 2017)

• Record Type: Journal Article
• Authors: Wang, Benfeng; Jakobsen, Morten; Wu, Ru-Shan; Lu, Wenkai; Chen, Xiaohong
• Is Part Of: Inverse problems. Volume 33:Number 3(2017:Mar.)
• Publication Date: 2017-02-01
• Subjects: full waveform inversion -- inverse thin-slab propagator -- domain decomposition method -- full T-matrix inversion method -- non-linear estimation; Inverse problems (Differential equations) -- Periodicals; 515.357
• DOI: 10.1088/1361-6420/aa5998
• Languages: English
• ISSNs: 0266-5611

Abstract

Abstract: Full waveform inversion (FWI) has been regarded as an effective tool to build the velocity model for the following pre-stack depth migration. Traditional inversion methods are built on Born approximation and are initial model dependent, while this problem can be avoided by introducing Transmission matrix (T-matrix), because the T-matrix includes all orders of scattering effects. The T-matrix can be estimated from the spatial aperture and frequency bandwidth limited seismic data using linear optimization methods. However the full T-matrix inversion method (FTIM) is always required in order to estimate velocity perturbations, which is very time consuming. The efficiency can be improved using the previously proposed inverse thin-slab propagator (ITSP) method, especially for large scale models. However, the ITSP method is currently designed for smooth media, therefore the estimation results are unsatisfactory when the velocity perturbation is relatively large. In this paper, we propose a domain decomposition method (DDM) to improve the efficiency of the velocity estimation for models with large perturbations, as well as guarantee the estimation accuracy. Numerical examples for smooth Gaussian ball models and a reservoir model with sharp boundaries are performed using the ITSP method, the proposed DDM and the FTIM. The estimated velocity distributions, the relative errors and the elapsed time all demonstrate the validity of the proposed DDM.