Parsimonious Green function data bases for global centroid moment tensor inversions

The calculation of synthetic seismograms for global centroid moment tensor (GCMT) inversions relies on advanced 3-D Earth models. However, use of the path-average approximation for mode summation and surface-wave ray theory limits the method's accuracy. This can cause incorrect predictions of ground motion amplitude and polarization, and other unaccounted-for effects, which can bias the estimated earthquake parameters. To address this issue, we have developed a new and efficient way to calculate, store and access high-fidelity, long-period synthetic seismograms for state-of-the-art 3-D tomographic Earth models. We adapted the spectral-element wave-equation solver SPECFEM3D_GLOBE to generate a data base of Green functions on a global, sparse spectral-element grid of hypocenters for a large set of 180 station locations, using source-receiver reciprocity to speed up the calculation. The seismograms are organized and stored in a format that facilitates rapid access to a particular source region and stations of the Global Seismographic Network. Seismograms for any centroid location can be calculated efficiently via spatial interpolation without losing accuracy compared to full forward calculation. As a proof-of-concept, we perform 9000 CMT inversions using the Sawade et al. approach, with GCMT solutions as starting models and without restriction on the number of iterations. Although the location updates are consistent with Sawade et al., we find a reduction in non-double-couple components in all types of events except for shallow strike-slip events. Given these encouraging results for future routine implementation, we present a first test and an outlook for routine 3-D GCMT analysis.