Project Abstract
|
In this study, we iteratively improve CVM-S4 through F3DT based on a combination of the scattering-integral (SI) method and the adjoint-wavefield (AW) method. At the current stage, we have carried out 26 iterations with the first 2 iterations and the latest 5 iterations based on SI and the 18 iterations in the middle based on AW. The use of SI in our F3DT allowed us to achieve faster convergence and to resolve smaller-scale structural features. The data used in our F3DT include more than 38,000 seismograms, which were generated by 160 local small to medium-sized earthquakes and recorded at more than 200 broadband three-component California Integrated Seismic Network (CISN) stations. In addition to earthquake recordings, we also included more than 12,500 ambient-noise Green’s functions in our F3DT, which significantly improved the spatial resolution of our structure model for most areas in Southern California. Our latest model CVM-S4.26 provides substantially better fit to the observed earthquake seismograms and ambient-noise Green’s functions for frequencies up to 0.2 Hz than our starting model CVM-S4. It reveals strong crustal heterogeneities throughout Southern California, some of which are completely missing in CVM-S4 and CVM-H11.9.1 but exist in models obtained from some crustal-scale active-source seismic surveys. The correlation between CVM-S4.26 and CVM-H11.9.1 at shallow depths is relatively high. However, there also exist substantial differences between these two models, especially at large depths. In this paper, we evaluate the qualities of CVM-S4.26 and CVM-H11.9.1 based on waveform misfits between synthetics computed using these two models and their corresponding observed seismograms for frequencies up to 0.2 Hz. The waveform misfit is quantified using two different criteria, the time-frequency domain goodness-of-fit (GOF) criterion proposed by Kristekova et al. (2009) and the time-domain misfit criterion defined in Zhu & Kanamori (1996), which was also used in Tape et al. (2009, 2010) for quantifying the improvements of their tomographically improved model CVM-H6.0 with respect to their starting model CVM-H5.5. For the more than 38,000 earthquake recordings used in our inversion, CVM-S4.26 provides substantially better fit than CVM-H11.9.1 based on either misfit criterion. |
