In this paper we apply Full waveform tomography (FWT) based on the Adjoint-Wavefield (AW) method to iteratively invert a 3-D geophysical velocity model for the Canterbury region (Lee, 2017) from a simple initial model. The seismic wavefields was generated using numerical solution of the 3-D elastodynamic/ visco- elastodynamic equations (EMOD3D was adopted (Graves, 1996)), and through the AW method, gradients of model parameters (compression and shear wave velocity) were computed by implementing the cross-adjoint of forward and backward wavefields. The reversed-in-time displacement residual was utilized as the adjoint source. For inversion, we also account for the near source/ station effects, gradient precondition, smoothening (Gaussian filter in spatial domain) and optimal step length. Simulation-to-observation misfit measurements based on 191 sources at 78 seismic stations in the Canterbury region (Figure 1) were used into our inversion. The inversion process includes multiple frequency bands, starting from 0-0.05Hz, and advancing to higher frequency bands (0-0.1Hz and 0-0.2Hz). Each frequency band was used for up to 10 iterations or no optimal step length found. After 3 FWT inversion runs, the simulated seismograms computed using our final model show a good matching with the observed seismograms at frequencies from 0 - 0.2 Hz and the normalized least-squared misfit error has been significantly reduced. Over all, the synthetic study of FWT shows a good application to improve the crustal velocity models from the existed geological models and the seismic data of the different earthquake events happened in the Canterbury region.
This poster presents work to date on ground motion simulation validation and inversion for the Canterbury, New Zealand region. Recent developments have focused on the collection of different earthquake sources and the verification of the SPECFEM3D software package in forward and inverse simulations. SPECFEM3D is an open source software package which simulates seismic wave propagation and performs adjoint tomography based upon the spectral-element method. Figure 2: Fence diagrams of shear wave velocities highlighting the salient features of the (a) 1D Canterbury velocity model, and (b) 3D Canterbury velocity model. Figure 5: Seismic sources and strong motion stations in the South Island of New Zealand, and corresponding ray paths of observed ground motions. Figure 3: Domain used for the 19th October 2010 Mw 4.8 case study event including the location of the seismic source and strong motion stations. By understanding the predictive and inversion capabilities of SPECFEM3D, the current 3D Canterbury Velocity Model can be iteratively improved to better predict the observed ground motions. This is achieved by minimizing the misfit between observed and simulated ground motions using the built-in optimization algorithm. Figure 1 shows the Canterbury Velocity Model domain considered including the locations of small-to-moderate Mw events [3-4.5], strong motion stations, and ray paths of observed ground motions. The area covered by the ray paths essentially indicates the area of the model which will be most affected by the waveform inversion. The seismic sources used in the ground motion simulations are centroid moment tensor solutions obtained from GeoNet. All earthquake ruptures are modelled as point sources with a Gaussian source time function. The minimum Mw limit is enforced to ensure good signal-to-noise ratio and well constrained source parameters. The maximum Mw limit is enforced to ensure the point source approximation is valid and to minimize off-fault nonlinear effects.
