Density reconstruction in multiparameter elastic full-waveform inversion. (8th November 2017)

• Record Type: Journal Article
• Authors: Sun, Min'ao; Yang, Jizhong; Dong, Liangguo; Liu, Yuzhu; Huang, Chao
• Is Part Of: Journal of geophysics and engineering. Volume 14:Number 6(2017:Dec.)
• Page Start: 1445
• Page End: 1462
• Publication Date: 2017-11-08
• Subjects: density -- elastic full-waveform inversion -- multiparameter -- Hessian -- scattering-integral algorithm -- truncated Gauss–Newton; Geophysics -- Periodicals; Prospecting -- Geophysical methods -- Periodicals; Engineering -- Periodicals; 622.1505
• DOI: 10.1088/1742-2140/aa93b0
• Languages: English
• ISSNs: 1742-2132

Abstract

Abstract: Elastic full-waveform inversion (EFWI) is a quantitative data fitting procedure that recovers multiple subsurface parameters from multicomponent seismic data. As density is involved in addition to P - and S -wave velocities, the multiparameter EFWI suffers from more serious tradeoffs. In addition, compared with P - and S -wave velocities, the misfit function is less sensitive to density perturbation. Thus, a robust density reconstruction remains a difficult problem in multiparameter EFWI. In this paper, we develop an improved scattering-integral-based truncated Gauss–Newton method to simultaneously recover P - and S -wave velocities and density in EFWI. In this method, the inverse Gauss–Newton Hessian has been estimated by iteratively solving the Gauss–Newton equation with a matrix-free conjugate gradient algorithm. Therefore, it is able to properly handle the parameter tradeoffs. To give a detailed illustration of the tradeoffs between P - and S -wave velocities and density in EFWI, wavefield-separated sensitivity kernels and the Gauss–Newton Hessian are numerically computed, and their distribution characteristics are analyzed. Numerical experiments on a canonical inclusion model and a modified SEG/EAGE Overthrust model have demonstrated that the proposed method can effectively mitigate the tradeoff effects, and improve multiparameter gradients. Thus, a high convergence rate and an accurate density reconstruction can be achieved.