https://hal-insu.archives-ouvertes.fr/insu-01396878Capdeville, BBCapdevilleIPGP - Institut de Physique du Globe de Paris - INSU - CNRS - Institut national des sciences de l'Univers - UPD7 - Université Paris Diderot - Paris 7 - UR - Université de La Réunion - IPG Paris - Institut de Physique du Globe de Paris - CNRS - Centre National de la Recherche ScientifiqueBerkeley Seismological LaboratoryGung, BBGungBerkeley Seismological LaboratoryRomanowicz, B.B.RomanowiczBerkeley Seismological LaboratoryTowards global earth tomography using the spectral element method: a technique based on source stackingHAL CCSD2005global seismologyinverse problemspectral element methodtomography[SDU.STU.GP] Sciences of the Universe [physics]/Earth Sciences/Geophysics [physics.geo-ph]Fareau, Eva2016-11-15 10:30:062023-02-04 03:10:572016-11-15 10:51:32enJournal articleshttps://hal-insu.archives-ouvertes.fr/insu-01396878/document10.1111/j.1365-246X.2005.02689.xapplication/pdf1We present a new tomographic method based on the non-linear least-squares inversion of seismograms using the spectral elements method (SEM). The SEM is used for the forward modelling and to compute partial derivatives of seismograms with respect to the model parameters. The main idea of the method is to use a special data reduction scheme to overcome the prohibitive numerical cost of such an inversion. The SEM allows us to trigger several sources at the same time within one simulation with no incremental numerical cost. Doing so, the resulting synthetic seismograms are the sum of seismograms due to each individual source for a common receiver and a common origin time, with no possibility to separate them afterward. These summed synthetics are not directly comparable to data, but using the linearity of the problem with respect to the seismic sources, we can sum all data for a common station and a common zero time, and we perform the same operation on synthetics. Using this data reduction scheme, we can then model the whole data set using a single SEM run, rather than a number of runs equal to the number of events considered, allowing this type of inversion to be feasible on a reasonable size computer. In this paper we present tests that show the feasibility of the method. It appears that this approach can work owing to the combination of two factors: the off-path sensitivity of the long-period waveforms and the presence of multiple-scattering, which compensate for the loss of information in the summation process. We discuss the advantages and drawbacks of such a scheme.