Because the calculation of the 3D Born sensitivity kernels is very expensive computationally, we have to select a target subregion (longitude 75 to 150 degrees and latitude 0 to 45 degrees) (Figure 2.50). In order to further reduce the intensity of the computation, we require that both events and stations must be in the large region, and only the ray paths along the minor arcs are selected. We calculated 3D Born sensitivity kernels for 162 events using the computing facilities (Jacquard) of the National Energy Research Scientific Computing Center (www.nersc.gov). When we generate synthetics for the N-Born inversion, we use N-Born (Panning et al., 2007) in the subregion and NACT outside of the subregion.

Our starting model is the NACT model in the large region. We expand its radially anisotropic () model from spherical spline Level 5 to Level 6, to conduct the conformal laterial parameterization for both isotropic and anisotropic inversion. We apply the N-Born approximation in the forward modeling part and calculate linear 3D Born kernels in the inverse part, and the adopted damping scheme for isotropic and radially anisotropic models is the same as that used in the NACT inversion.

Our N-Born shear velocity isotropic and anisotropic models are shown in Figure 2.51 and 2.52, respectively. Both N-Born and NACT derived models can fit waveforms very well, with up to variance reduction (depending on the choice of damping). While the models agree in general, there are some notable differences between them in detail. For example, beneath the Tibetan plateau, the N-Born model shows a stronger fast velocity anomaly in the depth range 150 km to 250 km, which disappears at greater depth. This indicates that there is no delamination of lithosphere beneath the plateau, as has been suggested by some authors. More importantly, the N-Born anisotropic model can recover well the downwelling structure associated with subducted slabs (e.g., around Phillippine plate) (Figure 2.51). Beneath the Tibet plateau, radial anisotropy shows (Figure 2.52) at depths of 300 km to 400 km, which implies horizontal rather than vertical flow and may help us to distinguish between end member models of the tectonics of Tibet.

In order to refine the velocity structure beneath the region of our study, we perform forward waveform modeling with the method of frequency-wave number integration (FKI). Broadband seismograms are downloaded from IRIS and corrected to absolute ground velocity (cm/sec). We show 2 event locations of events 2000256 (9/12/2000, Mw6.1) and 2003107 (4/17/2003, Mw6.3), and the IRIS station distributions (Figure 2.53). We start with the 2000256 event, for which the continental ray paths are dominant, to obtain the 1D velocity structure between the source and each receiver. Broadband data are bandpass filtered at 0.005-0.05Hz. We used the Harvard CMT solution for the source parameters, and the starting model is a 1D layered average crustal velocity structure derived from CRUST2.0. Using the best velocity model we can obtain (Figure 2.54), we compute Green's functions and perform the moment tensor analysis for two ranges of frequency (0.01-0.05Hz and 0.005-0.03Hz). Then, we select the event 2003107, for which we have similar ray paths as for event 2000256, to perform the moment tensor analysis using Green's function obtained from our 1D simulation. We find a moment tensor solution in good agreement with the CMT solution, whereas the solution obtained using the PREM reference model is very poor. While this example was chosen because we expect that we can use Harvard CMT solutions for M6 events as good references, this indicates that the additional regional modeling effort is worthwhile and will lead to better moment tensor solutions for smaller events in the area when we extend the modeling to higher frequencies (0.02-0.05Hz).

Berkeley Seismological Laboratory

215 McCone Hall, UC Berkeley, Berkeley, CA 94720-4760

Questions or comments? Send e-mail: www@seismo.berkeley.edu

© 2007, The Regents of the University of California