Semi-global inversion of vp to vs ratio for elastic wavefield inversion. (20th September 2018)

• Record Type: Journal Article
• Authors: da Silva, Nuno V; Yao, Gang; Warner, Michael
• Is Part Of: Inverse problems. Volume 34:Number 11(2018:Nov.)
• Publication Date: 2018-09-20
• Subjects: semi-global inversion -- elastic inversion -- full waveform inversion -- global inversion -- inverse problems; Inverse problems (Differential equations) -- Periodicals; 515.357
• DOI: 10.1088/1361-6420/aade1f
• Languages: English
• ISSNs: 0266-5611

Abstract

Abstract: We introduce an approach to estimate the ratio between P- and S-wave velocities, v p / v s, in the scope of elastic full waveform inversion (FWI). Elastic FWI is generally implemented with local optimization methods relying on initial estimates of the long wavelengths of P- and S-wave models. However, successful inversions can be hindered if an accurate enough relation between v p and v s velocities is not used as a constraint. This relation can be estimated from empirical relations. Herein, we introduce an alternative approach based upon a semi-global inversion scheme. We observe that for a large number of cases, and particularly in the context of FWI, v p / v s can be represented on a sparse basis. This sparse basis has a much smaller dimension than that of the typical model space in elastic FWI. This creates the possibility of using global optimization methods. The optimal estimate of v p / v s is obtained with quantum particle swarm optimization (QPSO). This method probes a population of possible models. The assessment of each model of v p / v s in the population is obtained with nested local iterations updating for v p only. Conventional elastic FWI is then carried out for jointly estimating high-resolution models of v p and v s . We demonstrate with synthetic examples that the estimates of v p are relatively robust to errors in the estimated v p / v s, and that effectively a sparse representation of the model of v p / v s is feasible for the reconstruction of a model of v s . We also demonstrate that the proposed approach performs better than constraining elastic FWI with an empirical relation between v p and v s, leading to improved estimates of models of v p and v s from seismic data.