A source parameter estimation algorithm that incorporates Green's functions for 3D heterogeneous viscoelastic anisotropic media is developed. The algorithm is based on the construction of synthetic seismograms for a double-couple point source by linear combination of the five fundamental sources that correspond to the moment tensor components. Construction of the Green's functions for such complex media is performed numerically by eighth-order finite differencing. For the source inversion, the model search procedure of Zhao and Helmberger (1994) is used. Results for a synthetic event in a realistic 3D model of the San Fernando basin show that incorporation of the 3D structure reduces the residual errors of the waveform fitting by more than 50% compared to those for a 1D layered model; azimuthal anisotropy produced by vertical cracks reduces the errors about 30%; and attenuation accounts for the other 20%.
The algorithm is applied to real data from two 1994 Northridge aftershocks. The source parameters obtained from high-frequency seismograms at local distances are similar to those obtained using traditional approaches at regional distances. In the source estimations from real data, the largest improvement in the residuals is observed in going from a 1D to the 3D model. The most significant contribution of the 3D structure is a considerable improvement in the uniqueness of solution. Incorporation of anisotropy and attenuation to the 3D model into estimations do not improve the solutions for the real data and implies that these aspects of the model are not yet sufficiently well defined to contribute to improving source-parameter estimations.