Abstract
The existing global tomographic methods result in different models due to different parametrization, scale resolution and theoretical approach. To test how current imaging techniques are limited by approximations in theory and by the inadequacy of data quality and coverage, it is necessary to perform a global-scale benchmark to understand the resolving properties of each specific imaging algorithm. In the framework of the Seismic wave Propagation and Imaging in Complex media: a European network (SPICE) project, it was decided to perform a benchmark experiment of global inversion algorithms. First, a preliminary benchmark with a simple isotropic model is carried out to check the feasibility in terms of acquisition geometry and numerical accuracy. Then, to fully validate tomographic schemes with a challenging synthetic data set, we constructed one complex anisotropic global model, which is characterized by 21 elastic constants and includes 3-D heterogeneities in velocity, anisotropy (radial and azimuthal anisotropy), attenuation, density, as well as surface topography and bathymetry. The intermediate-period (>32 s), high fidelity anisotropic modelling was performed by using state-of-the-art anisotropic anelastic modelling code, that is, coupled spectral element method (CSEM), on modern massively parallel computing resources. The benchmark data set consists of 29 events and three-component seismograms are recorded by 256 stations. Because of the limitation of the available computing power, synthetic seismograms have a minimum period of 32 s and a length of 10 500 s. The inversion of the benchmark data set demonstrates several well-known problems of classical surface wave tomography, such as the importance of crustal correction to recover the shallow structures, the loss of resolution with depth, the smearing effect, both horizontal and vertical, the inaccuracy of amplitude of isotropic S-wave velocity variation, the difficulty of retrieving the magnitude of azimuthal anisotropy and so on. The synthetic data set can be used to validate and calibrate new processing methodologies and has been made available to the scientific community at the Institut de Physique du Globe de Paris (IPGP) website (www.ipgp.jussieu.fr/~qyl). Any group wishing to test their tomographic algorithm is encouraged to download the synthetic data.
Highlights
As a result of the deployment of many digital broad-band networks over the past two decades, there has been a tremendous growth in the volume of acquired seismic data
Since the input model is very simple, the retrieved image matches the input model very well. This preliminary benchmark demonstrates that numerical accuracy and acquisition geometry is suitable for global tomographic test, and illustrates some interesting well-known problems of tomographic inversions: (1) generally, the tomographic images show smoother velocity variations compared with the input model; (2) the shape of heterogeneities can be deformed by the inhomogeneous path distribution; (3) the effects of horizontal smearing can result in merging of some separated anomalies into a single one and (4) the small-scale, weak anomalies are not robust
To validate the meshes and the time step used in the coupled spectral element method (CSEM), synthetics seismograms for a 1-D spherically symmetric model are computed and compared with the solution obtained by quasi-analytical normal-mode solution
Summary
As a result of the deployment of many digital broad-band networks over the past two decades, there has been a tremendous growth in the volume of acquired seismic data. In the framework of the European training network Seismic Wave Propagation and Imaging in Complex media: a European Network (SPICE), it has been decided to carry out a global-scale benchmark, testing global inversion algorithms through the inversion of the same high-accuracy synthetic data set. The goal of this benchmark data set is to compare different inversion methods in terms of resolution to assess their limits and advantages, and to enable seismologists to analyse how current imaging techniques are limited or not by approximations in theory, the inadequacy of data quality and coverage (Qin et al 2006). We discuss the construction of the input model, the generation of the synthetic data set and the corresponding inversion
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have