Extended Data Fig. 2: Improvement on data-fitting through inversion.
Figures in the left column (a, c, and e) show the distribution of correlation coefficient between synthetic and observed seismograms before (red line) and after (blue bars) the 206 iterations. Figures in the right column (b, d, and f) are illustrated in the same way, but are for the time shift between synthetic and observed data. The first row is for all P motion related body wave segments (for example, P, PP, etc.); the second row is for S motion related segments (for example, S, SS, etc.); and the third row is for surface wave segments (Rayleigh and Love waves). To be consistent, only the 110 earthquakes that are included in the fine-grid stage are used here, and only segments with signal-to-noise ratio16 > 8 are counted. Waveforms used for distributions before inversion for the coarse-grid stage (red lines) were filtered at lower frequencies, which are supposed to be easier for data-fitting than ones after inversion (blue bars).
