In this poster, we determine centroid moment tensor (CMT) solutions by two different approaches: iterative inversion based on the adjoint-wavefield method and direct inversion from the strain Green’s tensor and observed data. Since the heterogeneity of a 3-D velocity model can introduce error in source-parameter inversions based on 1-D models, it is necessary for source inversion along with structure inversion in regions with complicated crustal structure, such as the upper South Island, New Zealand. Synthetic waveforms are simulated through solving the 3-D visco-elastic wave equations (Graves, 1996) in the software emod3d and the construction of the strain Green’s tensors and seismograms from reciprocal principles (Graves, 2001). We perform the ground-motion simulation for a domain of 352x352x160 km utilizing an inverted velocity model (m10) refined to 2 km grid (Figure 1) from full waveform tomography study (Nguyen, 2020).

Synthetic tests are carried out to verify each of the inversion methods. For broadband data inversions, the adjoint CMT inversions are in good agreement with the direct CMT inversion using 3-D Green’s functions. The revised CMT solutions then were included for further tomography inversion to improve the 3-D velocity model.