Skip to Main content Skip to Navigation

Imagerie sismique à deux dimensions des milieux visco-élastiques par inversion des formes d'ondes : développements méthodologiques et applications

Abstract : The knowledge of Earth intern structures at different scales is of major interest for economy, humans, environment and science. Several methods have been developed for Earth imaging using seismic wave information. The full waveform inversion attempts to build quantitative high-resolution images of the subsurface physical parameters using the full wavefield, solved as an optimization procedure. In this thesis, I present application of two-dimensional frequency-domain full waveform inversion for imaging visco-elastic parameters from large offsets data. In a first time, methodologies and algorithms are presented. The frequency-domain P-SV waves propagation modelling, the forward problem of the inversion process, is solved with a finite element discontinuous Galerkin method. This method allows a flexible choice of interpolation orders and the use of triangular unstructured meshes. The inverse problem is linearized in order to limit the number of forward problem simulations, and solved with the quasi-Newton L-BFGS algorithm in order to exploit information contained in the estimated Hessian matrix. The full imaging process is implemented in a massively-parallel algorithm for the distributed-memory architectures of modern computing centres. In a second time, algorithms are applied to several case studies. Applications are performed in realistic synthetic models, representative of onshore and offshore environments. These studies show the difficulties associated to elastic parameters imaging from complex data including converted waves, multiples and surfaces waves. Several multi-scale hierarchical procedures are proposed in order to mitigate non-linearity of the inverse problem and to improve convergence toward the global minimum. Finally, a sensitivity study is performed to analyse the behaviour of several minimization norms and criteria, when noisy data are inverted. A first application in realistic synthetic models is presented before an acoustic application to field data. These tests show some limits of the classic L2 norm in the data space, while the L1 norm appears to be a robust alternative for frequency-domain inversion of decimated data.
Document type :
Complete list of metadatas
Contributor : Romain Brossier <>
Submitted on : Thursday, January 28, 2010 - 12:57:36 PM
Last modification on : Wednesday, December 9, 2020 - 3:08:03 PM
Long-term archiving on: : Thursday, October 18, 2012 - 1:35:11 PM


  • HAL Id : tel-00451138, version 1


Romain Brossier. Imagerie sismique à deux dimensions des milieux visco-élastiques par inversion des formes d'ondes : développements méthodologiques et applications. Planète et Univers [physics]. Université Nice Sophia Antipolis, 2009. Français. ⟨tel-00451138⟩



Record views


Files downloads