2.1 Hypocenter relocations
We relocated 21.102 events listed in the JMA unified catalogue for the southern Kagoshima Bay region for the period from March 1, 2003 to April 8, 2018 using the Double-Difference (DD) method (Waldhauser & Ellsworth, 2000). This relative relocation method minimizes the residuals between the observed and theoretical travel time differences for adjacent earthquake pairs at each station. We applied the DD method to differential arrival time data, which were estimated from the waveform cross-correlation, and those listed in the JMA unified catalog. The procedure was identical to that reported in Yoshida and Hasegawa (2018a, b), which can be briefly described as follows.
First, we obtained precise differential arrival time data using waveform cross-correlations. We used the waveform data observed at 20 permanent seismic stations surrounding the focal area (Fig. 1a; green stations). At each station, the ground velocity was measured using three-component short-period seismometers (natural period of 1s) and a sampling rate of 100 Hz. We applied a 5–12 Hz Butterworth filter to the waveforms of each target event. We used 2.8 and 4.3 s time windows for the P- and S-waves, respectively, starting 0.3 s before their arrival. The arrival times were obtained from the JMA unified catalogue. If arrival times were not available, they were estimated using the one-dimensional JMA2001velocity model (Ueno et al., 2002) and the hypocenters, and origin times listed in the JMA unified catalogue. We calculated the waveform cross-correlations of event pairs with hypocenters within 3 km from each other and obtained differential arrival times when the cross-correlation coefficients were greater than 0.8. In total, we acquired 17.332.318 P-wave differential arrival time data points and 27.738.043 S-wave data points. We also derived the differential arrival data from the arrival time data listed in the JMA unified catalog: 474.670 data for P-waves and 543.226 data for S-waves. For the mainshock, only data derived from the JMA unified catalog were used because of its long duration.
Second, we applied the hypo-DD algorithm (Waldhauser & Ellsworth, 2000) to the differential arrival time data. We used a spherical shell two-layer model (Aki, 1965) for the hypocenter relocation. In this model, the seismic velocities in each layer are proportional to the power of the distance from the center of the Earth (Figure S3). The medium parameters were determined for consistency with the seismic tomography results obtained in the Kyushu region (Saiga et al., 2010). We used the hypocenters listed in the JMA unified catalogue for the initial locations for the relocation. Figures 2a and 3a show the distribution of these initial hypocenters. Differential arrival time data were weighted with respect to the square root of the cross-correlation coefficient. The hypocenters were updated after 50 iterations of the relocation procedure. During the first ten iterations, a higher weight was assigned to the catalogue data to constrain the relative locations of large-scale features. In the latter 40 iterations, a higher weight was assigned to the data derived by the cross-correlations to delineate shorter-scale features. We evaluated the uncertainty in the relative hypocenter locations by recalculating the hypocenters 200 times based on bootstrap resampling of differential arrival time data.