Fig. 3. The number of group velocity dispersion curves at different periods.
We adopted the data processing procedures following the method of Bensen et al. (2007) and Fang et al. (2009). Data are processed one day at a time for each station after being decimated to 1 Hz. Other parts involved instrument response removal, clock synchronization, time-domain normalization, bandpass filtering (4−50 s period), and spectral whitening. Following this, the day-long waveform at each station is correlated with other seismic stations and the daily results are stacked to produce the final cross-correlation results.
The resulting cross-correlations contain surface wave signals coming from opposite directions along the path linking the stations. The cross-correlations are often asymmetrical due to the inhomogeneous distribution of ambient noise sources. To simplify data analysis and enhance the signal-to-noise ratio (SNR) of the surface waves, we separated each cross-correlation into positive and negative lag components and then added the two components to form the so-called symmetric component. The following analysis was done on the symmetric signals exclusively.
We use the CPS (Computer Programs in Seismology) software developed by Herrmann and manually picked up the group velocity dispersion curve based on the multiple filtering technology to (Dziewonski et al., 1969; Levshin et al., 1992,Herrmann, 1973). If there are n stations, then the empirical Green’s function on n(n-1)/2 paths can be calculated. In order to ensure reliable results, a quality control of the dispersion curve was carried out.
An empirical Green’s function is accepted if its signal to noise ratio is greater than 10 and the inter-station distance is at least 3 times of wavelength at a given period (Yao et al., 2006; Bensen et al., 2007). Furthermore, we excluded paths that are shorter than 120 km because of the lack of adequate condition related to 3 times of the wavelength. Finally, a total of 1883 dispersion curves of the station pairs meeting the above requirements were extracted from the 3916 Rayleigh wave waveform data (Fig. 2). We show in Fig. 3 the number of ray paths used for surface wave imaging at different periods, and confirm that the number of rays in most periods is relatively uniform.
Rayleigh wave velocity and S-wave velocity inversion
For the surface wave tomography, a generalized 2-D-linear inversion procedure developed by Ditmar and Yanovskaya (1987) and Yanovskaya and Ditmar (1990) was applied to construct the group velocity inversion, which is a generalization to 2-D inferred from the classical 1-D method of Backus and Gilbert (1968). In this study, we designed a 0.5° × 0.5° grid lateral. The damping parameter (α) controls the trade-off between the fit to the data and the smoothness of the resulting group velocity maps. We use the value of α = 0.2, which yields relatively smooth maps with small fit error.
From the Rayleigh wave group velocity obtained by the above inversion approach, we extracted the dispersion curves of group velocity at each grid node. We then inverted for the 1-D shear wave velocity structure under each grid node following the method of Herrmann and Ammon (2004). The velocities in between the nodes are interpolated linearly. In this way, a 3-D shear wave velocity structure was constructed (Fig. 4).
The initial model has a constant shear-wave velocity of 4.48 km/s from the surface to 90 km depth that is divided into 2 km layers. By starting with an overestimated velocity model, we ensured that no artificial low-velocity zone or layer boundary was introduced as a consequence of the nonlinear of the inversion. A fixed Vp/Vs ratio of 1.732 was used and the density was calculated from the P-wave velocity (Zanjani et al., 2019).