Elastic full waveform inversion based on the homogenization method: theoretical framework and 2-D numerical illustrations. Issue 2 (5th February 2018)

• Record Type: Journal Article
• Authors: Capdeville, Yann; Métivier, Ludovic
• Is Part Of: Geophysical journal international. Volume 213:Issue 2(2018:May)
• Page Start: 1093
• Page End: 1112
• Publication Date: 2018-02-05
• Subjects: Inverse theory -- Numerical solutions -- Computational seismology -- Seismic tomography -- Wave propagation; Geophysics -- Periodicals; 550
• DOI: 10.1093/gji/ggy039
• Languages: English
• ISSNs: 0956-540X

Abstract

SUMMARY: Seismic imaging is an efficient tool to investigate the Earth interior. Many of the different imaging techniques currently used, including the so-called full waveform inversion (FWI), are based on limited frequency band data. Such data are not sensitive to the true earth model, but to a smooth version of it. This smooth version can be related to the true model by the homogenization technique. Homogenization for wave propagation in deterministic media with no scale separation, such as geological media, has been recently developed. With such an asymptotic theory, it is possible to compute an effective medium valid for a given frequency band such that effective waveforms and true waveforms are the same up to a controlled error. In this work we make the link between limited frequency band inversion, mainly FWI, and homogenization. We establish the relation between a true model and an FWI result model. This relation is important for a proper interpretation of FWI images. We numerically illustrate, in the 2-D case, that an FWI result is at best the homogenized version of the true model. Moreover, it appears that the homogenized FWI model is quite independent of the FWI parametrization, as long as it has enough degrees of freedom. In particular, inverting for the full elastic tensor is, in each of our tests, always a good choice. We show how the homogenization can help to understand FWI behaviour and help to improve its robustness and convergence by efficiently constraining the solution space of the inverse problem.