HR: 16:40h
AN: NS34A-02    [Abstracts]
TI: An unsplit Convolutional perfectly matched layer technique improved at grazing incidence for the differential anisotropic elastic wave equation: application to 3D heterogeneous near surface slices.
AU: * Martin, R
EM: roland.martin@univ-pau.fr
AF: Magique3D-INRIA, Laboratoire de Modélisation et Imagerie en Géosciences de Pau. Groupe Magiqe3D-INRIA Futurs. Avenue de l'Université. Bāt IPRA. BP 1155., Pau, 64 64013, France
AU: Komatitsch, D
EM: dimitri.komatitsch@univ-pau.fr
AF: Magique3D-INRIA, Laboratoire de Modélisation et Imagerie en Géosciences de Pau. Groupe Magiqe3D-INRIA Futurs. Avenue de l'Université. Bāt IPRA. BP 1155., Pau, 64 64013, France
AB: In geophysical exploration, high computational cost of full waveform inverse problem can be drastically reduced by implementing efficient boundary conditions. In many regions of interest for the oil industry or geophysical exploration, nearly tabular geological structures can be handled and analyzed by setting receivers in wells or/and at large offset. Then, the numerical modelling of waves travelling in thin slices along wells and near surface structures can provide very fast responses if highly accurate absorbing conditions around the slice are introduced in the wave propagation modelling. Here we propose then a Convolutional version of the well known Perfectly Matched layer technique. This optimized version allows the generation of seismic waves travelling close to the boundary layer at almost grazing incidence, which allows the treatment of thin 3D slices. The Perfectly Matched Layer (PML) technique, introduced in 1994 by Bérenger for Maxwell's equations, has become classical in the context of numerical simulations in electromagnetics, in particular for 3D finite difference in the time domain (FDTD) calculations. One of the most attractive properties of a PML model is that no reflection occurs at the interface between the physical domain and the absorbing layer before truncation to a finite-size layer and discretization by a numerical scheme. Therefore, the absorbing layer does not send spurious energy back into the medium. This property holds for any frequency and angle of incidence. However, the layer must be truncated in order to be able to perform numerical simulations, and such truncation creates a reflected wave whose amplitude is amplified by the discretization process. In 2001, Collino and Tsogka introduced a PML model for the elastodynamics equation written as a first-order system in velocity and stress with split unknowns, and discretized it based on the standard 2D staggered-grid finite-difference scheme of Virieux (1986). Then in 2001 and 2004, Zheng applied this technique to Biot poroelastic systems. Unfortunately, this standard PML suffers from two drawbacks: the fact that the unknowns are split adds to the memory cost of the simulations because additional arrays must be used to store all the split components . After numerical discretization, the numerical reflection coefficient between the physical domain and the PML region becomes large at grazing incidence and therefore the efficiency of the absorbing layer is poor. In 2000, Roden and Gedney introduced an implementation of the PML for Maxwell's equations based on the original (unsplit) components of the wave field and optimized for grazing incidence using an analytical integration of the convolution term. This formulation, which is commonly known as the Convolution- Perfectly Matched Layer (C-PML), overcomes the two main drawbacks of the classical PML formulation mentioned above. The PMLs are tested here for near surface complex structures involving highly dispersive weathered zones and salt domes which can be dealt with such 3D PMLs.
UR: http:migp.univ-pau.fr
DE: 0902 Computational methods: seismic
DE: 0935 Seismic methods (3025, 7294)
DE: 3285 Wave propagation (0689, 2487, 4275, 4455, 6934)
DE: 7290 Computational seismology
SC: Near-Surface Geophysics [NS]
MN: 2007 Joint Assembly