On the choice of material parameters for elastic waveform inversion

Ryan Modrak, Yanhua Yuan, Jeroen Tromp

Research output: Contribution to journalConference article

5 Scopus citations

Abstract

The material parameterization used to compute derivatives, search directions, and model updates in elastic waveform inversion can have a significant effect on the robustness and efficiency of the overall nonlinear optimization procedure. For isotropic media, conventional wisdom holds that it is better to work with compressional- and shear-wave speeds a and b than Laḿe parameters l and μ. Conventional wisdom further holds that, given their improved scaling and reduced covariance, bulk and shear moduli k and μ, or "wave speed-like" parameters qkr and qμ r, are an even better choice. Numerical tests involving hundreds of inversions, six subsurfaces models, and three full-waveform misfit functions, however, reveal a more complicated picture. Working in the time domain and using numerically safeguarded quasi-Newton model updates, we find that the relative performance of material parameterizations is strongly problem dependent. Despite their physical relevance, a and b rank last among all parameterizations in both efficiency and robustness. In a large number of cases, l and μ outperform all competitors. These results make sense in light of some additional observations, namely, that (1) relative performance of material parameterizations correlates with nonlinearity, which can be affordably quantified in terms of deviation of the misfit function from a quadratic form, and (2) wave-speed like parameters perform well for phase-based inversions, but not for waveform- or envelope-based inversions. Since a direct relation between wave-speed perturbations and data residuals exists for phase misfit functions, but not for envelope- or waveform-difference misfit, good performance of Laḿe parameters relative to wave speed-like parameters in the latter cases is perhaps not surprising.

Original languageEnglish (US)
Pages (from-to)1115-1119
Number of pages5
JournalSEG Technical Program Expanded Abstracts
Volume35
DOIs
StatePublished - Jan 1 2016
EventSEG International Exposition and 86th Annual Meeting, SEG 2016 - Dallas, United States
Duration: Oct 16 2011Oct 21 2011

All Science Journal Classification (ASJC) codes

  • Geotechnical Engineering and Engineering Geology
  • Geophysics

Fingerprint Dive into the research topics of 'On the choice of material parameters for elastic waveform inversion'. Together they form a unique fingerprint.

  • Cite this