Regional 3D tomography of the upper mantle using a summed source approach

Global and large scale regional waveform tomography has so far been based on approximate expressions of the wavefield in the framework of normal mode perturbation theory. As powerful numerical methods, such as the Spectral Element Method (SEM), are progressively adapted to waveform tomography and replace approximate computations, we gain significantly in the accuracy with which the effects of strong lateral heterogeneity of different scales can be represented. However, the main drawback is the large increase in computational time, which limits the frequency range as well as the number of iterations that can be performed. Capdeville et al. (2005, GJI) introduced an approach in which the computational time is significantly reduced, by simultaneously computing the wavefield for a large number of events, and comparing it to the correspondingly aligned and summed seismograms. This approach was tested on a global synthetic dataset with added noise, and showed considerable promise. We have implemented and tested this approach in the case of regional tomography, using a regional version of the SEM, RegSEM (Cupillard, 2008), developed for spherical geometry, which uses Perfectly Matched Layers (PML) at the border of the region and includes general 3D anisotropy, Moho and surface topography, ocean bathymetry, attenuation and ellipticity. We show the results of tests of this approach against the standard "single source - single station" waveform inversion and discuss various challenges encountered. We present a model of upper mantle structure in southeast Asia, using a dataset of over 100 events observed at 6+ stations, with epicentral distances between 5-40 degrees, and in the period range 300-32 s.