Geophysical Journal International

Transcription

1 The Gyeongju earthquake sequence revisited: Aftershock interaction within a complex fault system Abbreviated Title: The Gyeongju earthquake sequence in a complex fault system Journal: Manuscript ID Manuscript Type: Date Submitted by the Author: Complete List of Authors: Keywords: GJI-S--0.R Research Paper 0-Nov- Woo, Jeong-Ung; Seoul National University, School of Earth and Environmental Sciences Rhie, Junkee; Seoul National University, School of Earth and Environmental Sciences; Kim, Seongryong; Chungnam National University, Geology and Earth Environmental Sciences Kang, Tae-Seob; Pukyong National University, Department of Earth Environmental Sciences Kim, Kwang-Hee; Pusan National University, Geological Sciences Kim, YoungHee; Seoul National University, School of Earth and Environmental Sciences Earthquake source observations < SEISMOLOGY, Dynamics and mechanics of faulting < TECTONOPHYSICS, Dynamics: seismotectonics < TECTONOPHYSICS, Continental neotectonics < TECTONOPHYSICS, Earthquake interaction, forecasting, and prediction < SEISMOLOGY

2 Page of Title: The Gyeongju earthquake sequence revisited: Aftershock interaction within a complex fault system Abbreviated Title: The Gyeongju earthquake sequence in a complex fault system Jeong-Ung Woo, Junkee Rhie *, Seongryong Kim, Tae-Seob Kang, Kwang-Hee Kim, YoungHee Kim School of Earth and Environmental Sciences, Seoul National University, Seoul 0, Korea Department of Geology and Earth Environmental Sciences, Chungnam National University, Daejeon, Korea Department of Earth and Environmental Sciences, Pukyong National University, Busan, Korea Department of Geological Sciences, Pusan National University, Busan, Korea * Corresponding author rhie@snu.ac.kr Phone: Fax: +--- Key words: Earthquake source observation; Dynamics and mechanics of faulting; Seismotectonics; Continental neotectonics; Earthquake interaction, forecasting, and prediction SUMMARY On September, a moderate earthquake (ML.) occurred in Gyeongju, South Korea, located hundreds of kilometres away from the nearest plate boundaries. The earthquake, the largest instrumentally recorded event in South Korea, occurred in a sequence of thousands of earthquakes, including a ML. event 0 minutes before and a ML. event a week later. As a case study, we analyse the source parameters of the

3 Page of Gyeongju earthquake sequence: precise relocations, fault structures, focal mechanisms, stress tensor analysis, and Coulomb stress changes. To determine high resolution hypocentres and focal mechanisms, we employ our temporary seismic network for aftershock monitoring as well as regional permanent seismic networks. Spatiotemporal distribution of events and inverted moment tensors indicate that the ML. event and the ML. event occurred on two parallel dextral faults striking NNE-SSW at a depth of - km and the ML. event lie on their conjugate fault with sinistral displacements. Seismicity on the fault for the ML. event abruptly decreased as soon as the ML. event occurs. It is not solely explained by the Coulomb stress change, requiring more complex processes to explain it. The tectonic stress field obtained from inverted focal mechanisms suggests that the heterogeneity between the intermediate and minimum principal stresses exists along 0 NNE-SSW and vertical directions. The Coulomb stress changes imparted from the ML. event and the ML. event are matched with off-fault seismicity including the ML. event. Multifaceted observations such as Coulomb stress interactions between parallel or conjugate faults and the heterogeneity of the tectonic stress field in the aftershock area may reflect the reactivation processes of a complex fault system. This study offers a distinctive case study to understand the general characteristics of intraplate earthquakes in multi-fault systems. INTRODUCTION Intraplate earthquakes pose certain seismic hazards and risks as natural disasters, although they host only % of the world s seismic moment release (Wang 0; Johnston 0 ; Durá-Gómez & Talwani 0; Chen et al. ; Rastogi, 0). In spite of their importance in hazard analysis, their general characteristics are not well understood (Gangopadhyay & Talwani 0; Talwani ). It is widely accepted that intraplate earthquakes are caused by the tectonic forces transmitted from plate boundaries and located in preexisting zones of weakness such as failed rifts (Bodin & Horton 0; Sykes ; Talwani & Rajendran ; Tavakoli et al. ; Zoback ). However, this hypothesis has not been clearly verified because some studies have reported that they

4 Page of occurred regardless of faults with surface expressions (Assumpção et al. 0; Ferreira et al. ). Furthermore, the rarity of the surface extension of fault ruptures makes it hard to observe the causality (Adams ). Our understanding of intraplate earthquakes has 0 advanced as a lot of seismographs are installed in intraplate regions and diverse approaches for array networks are developed (Bianchi et al. ; Matos et al. ; Ross et al. ; Soto Cordero et al. ; Talwani ). The Korean peninsula is an intraplate region located ~00 km away from the Ryukyu and Nankai trough. By an event catalogue published by the Korea Meteorological Administration (KMA), only earthquakes with ML were instrumentally recorded since (So et al. ). The largest amongst them, a ML. earthquake in Gyeongju, South Korea, on September produced strong coseismic ground shaking, which was sufficient to be felt throughout South Korea (Hong et al. ). The event is located in the Gyeongsang Basin where there are several systematic fault sets with surface expression, such as the Yangsan 0 Fault System and the Gaum Fault System (Figs. and a, Chang ; Hwang et al. 0; Han et al. ). The Gyeongju earthquake sequence started from a ML. event that ruptured 0 minutes before the ML. event and thousands of earthquakes including a ML. aftershock of September occurred in the sequence (Kim et al. b). Hereafter, for convenience, we refer to the three events of ML., ML., and ML. as E, E (main shock), and E, respectively. From the distribution of hypocentres and inverted moment tensors of the three events, it has been demonstrated that these earthquakes occurred on a deep seated fault system at a depth range of - km (Hong et al. ; Kim et al. a; Kim et al. a; Kim et al. b; Lee et al. ; Son et al. ). In particular, Son et al. () delineated two distinct parallel dextral faults striking to the NNE-SSW direction from 0 relocated aftershocks and Uchide & Song () observed that the inverted finite fault slips of E and E propagated to SSW and NNE directions, respectively. Possible correlation between the Yangsan Fault and the Gyeongju earthquakes has been raised because epicentres are closely located to the fault with km of dextral displacement (Kim et al. b; Kim et al. c; Lee et al. ; Kyung 0). However, it is difficult to prove

5 Page of whether the deep seated fault system is extended to the surface or not because focal depths exceed km and no surface deformation due to E has been reported (Park et al. ). In this study, we investigated the source parameters of the Gyeongju earthquakes using a larger data set obtained from both permanent and temporary seismic networks. Our basic goal is to create a more complete catalogue of earthquakes in the vicinity of the mainshock 0 from to. We located earthquakes detected by an automatic algorithm and resolved the detailed fault segments based on the Double-Difference method. The relative magnitudes of events were estimated from the amplitude ratios with reference events. Long period waveforms and the first motions of P-wave polarities were used for moment tensor inversion and focal mechanism determination, respectively. Based on our catalogue, we conducted further analysis including statistical parameters, tectonic stress fields, and the role of stress transfer induced by E and E. Aforementioned multifaceted approaches will contribute to a better understanding for the reactivation process of interacting complex faults in the aftershock region and the generalised characteristics of intraplate earthquakes within complex fault systems. 0 DATA ACQUISITION Data used for the detection of events and the determination of their hypocentres, magnitudes and focal mechanisms is divided into two parts: the continuous waveforms recorded at permanent seismic stations and those of temporary seismic stations installed for monitoring aftershocks and analysing detailed source parameters (Fig. ). Three years of waveform data from January to December are gathered from six broadband and two short-period seismometers within 0 km from the epicentre of E. For moment tensor determination of selected events, waveforms recorded at broadband stations with epicentral distances greater than 0 km are additionally collected and data from five 0 broadband stations (inset of Fig. ) are finally used. The temporary network of broadband instruments started its operation within two hours of E and we assembled the waveform data by the end of (Fig. ). For the epicentre of E, the data coverage of the eight

6 Page of permanent stations yields the azimuthal gap of 0 and the minimum epicentral distance of km. However, by adding data from the temporary instruments, the two parameters could be reduced to and 0. km, respectively. Therefore, it is expected that not only additional events with small magnitudes can be detected, but also the source parameters of the earthquakes can be analysed at high resolution scales using both the permanent stations and the temporary stations (see details about the temporary seismic network of Kim et al. a). 0 HYPOCENTRE DETERMINATION. Method of analysis To analyse not only aftershocks within a few kilometres from the epicentre of E but also the seismicity changes before and after E, we applied an automatic earthquake detection method to continuous waveform data and determined their absolute locations using the single difference method (HYPOELLIPSE, Lahr ). In order to resolve the spatial pattern of aftershocks, the earthquakes near E were relocated using the Double-Difference algorithm (HypoDD, Waldhauser & Ellsworth 00). P- and S-wave phase arrival times were automatically measured from the recursive short-term average to long-term average ratio 0 (STA/LTA) method (Grigoli et al. ; Withers et al. ; Woo et al. ). Then, incorrectly measured phase arrival times were manually inspected to discriminate overlapping events and remove false detections. A regional one-dimensional velocity model for the Gyeongsang Basin (e.g., Kim et al. ) was applied throughout the hypocentre determination. In the process of relocation, the travel time differences calculated from,0 picked arrival time pairs as well as those measured from,0, waveform cross-correlation pairs were applied. We tested whether the data obtained from the picked arrival times distort the relocation results or not and confirmed insignificant changes in the results. For the waveform cross-correlation between two events, time windows centred at measured arrival 0 times were interpolated at a sampling rate of 00 and the maximum cross-correlation

7 Page of coefficient (maxcc) of s time windows were measured by allowing up to ±. s time shifts (e.g., Hauksson & Shearer 0). All the seismograms were bandpass-filtered from to Hz before the cross-correlation and the travel time differences between two waveforms are used only if the maxcc is above 0.. For the bootstrapping to test the relocation uncertainties, the events were relocated by resampling the differential travel times for each event pair. The mean of standard deviations of the trials corresponds to m horizontally and m vertically, respectively.. Absolute location 0 During the study period, earthquake candidates were automatically detected and earthquakes within 0 km of the epicentre of E were selected through the manual inspection of waveforms (Fig. ). Only events (.%) occurred before the onset of E. These events are mainly observed in offshore regions or along several local fault structures (Fig. a). However, only two events were located within km of the epicentre of E during this period, and thus it is considered that the background seismicity near the aftershock region is insignificant. After the onset of E, the number of earthquakes located outside the radius of km from the epicentre of E is (.%) and their locations are similar to the background seismicity before E. However, some regions have experienced spatial changes in seismic activities as of the occurrence of E. For example, the earthquakes where the 0 Ulsan Fault and the Dongrae Fault intersect each other have shifted km to the south (see the blue circled area of Fig. a). The earthquakes near the Miryang Fault have migrated to the north by a few kilometres (see the magenta circled area of Fig. a). These observations suggest that the seismicity has changed due to the Gyeongju earthquakes, although it is hard to validate how long these changes are maintained. Starting from E, events (.%) occurred within km from the epicentre of E and the distribution of the hypocentres showed a trend along the NNE-SSW direction (Fig. b). Although their epicentres are mainly located between the two surface fault traces of the Yangsan Fault and Deokcheon Fault striking to NNE-SSW or N-S direction (Kim et al.

8 Page of c), the spatial distribution of the earthquakes does not exactly coincide with the surface 0 geological features (Fig. b).. Double-difference relocation Including the three largest events, events are relocated within km from the epicentre of E (Fig. a). This number is ~ times greater than that of the relocated events of previous studies, which only used data of permanent station networks for a similar period (Son et al., ). This result illustrates the importance of aftershock monitoring networks for observations of microseismic events (Kim et al., a; Kim et al., a). In the map view, aftershocks are primarily aligned along the N E and extend km in the fault-parallel direction and km in the fault-normal direction. The focal depths of the earthquakes range 0 from km to km (Figs. c-d). Comparing the spatial pattern of the relocated hypocentres with that obtained from the single difference method, we can clearly observe the distinctive fault geometries that cannot be resolved in the initial distribution. In particular, it seems that the largest events (E and E) occur independently on two distinct sub-parallel faults with an offset of m, while E occurred on a conjugate fault plane with a strike of ESE-WNW (Fig. a). For convenience, hereafter, we refer the three fault planes overlapping with the locations of E, E, and E as F, F, and F, respectively. The size and shape of each fault plane are estimated from the hypocentre distributions. The F strikes at N E and deepens towards the southeast direction with a dip angle of in the depth range of km. The F is km long and. km wide, and comprises a 0 square-like fault plane (Fig. c). Located to the east of F, F has the same strike as F, but the dip angle of the fault plane varies with the depth range, and can be separated as: () Fa with a dip of and a depth range of.. km and () Fb with a dip of and a depth range of. km (Figs. c-d). E is located on Fb whose dip angle is matched with that of one nodal plane of the calculated moment tensor solution of E. The widths of Fa and Fb correspond to. km and. km, respectively, while the lengths of Fa and Fb are the same at. km. The strike of F is set to N E, since the earthquakes near

9 Page of E are aligned perpendicular to the strike of F and F. However, the number of earthquakes distributed over the streak is not enough to infer the dip angle. Therefore, we modeled F as a vertical fault plane based on the moment tensor inversion of E (Fig. a). 0 Using the spatial distribution of the aftershocks for an hour from the onset of E, the length and width of F were set to km and 0. km, respectively. The parameters of each fault plane are summarised in Table and the schematic diagram is illustrated in Fig.. E and E are located at the northeastern part of F and southwestern part of F compared with the spatial distribution of aftershocks (Figs. c and d). The asymmetric aftershock distributions with respect to hypocentres may represent that E and E ruptured toward southwest and northeast, respectively (Mendoza & Hatzell, ). This is consistent with the rupture directivity inferred from finite fault slip models of E and E (Uchide & Song ). 0. AFTERSHOCK STATISTICS.. Magnitude estimation Among the various methods to estimate earthquake magnitude, we use the scaling relationship between magnitude differences and amplitude ratios. If two events have similar ray paths, focal mechanisms and source time histories, the magnitude difference (dm) between them is expressed as: dm = clog α () where α is the amplitude ratio of one event to the other event, and c is a linear scaling parameter between the amplitude ratio and the magnitude difference (Shelly et al. ). In this study, the linear relationship and the scaling parameter ( c) of Eq. () were investigated 0 using aftershocks in the KMA catalogue and the magnitudes of aftershocks detected in this study were relatively determined from linear scaling with KMA catalogue events. For the cross-correlation measurements applied in the procedure of relocation, the amplitude ratios between two waveforms aligned by their phase arrival times were measured

10 Page of as the slope of the principal component obtained from the principal component analysis (PCA) for the data vectors (e.g., amplitude relation between two aligned waveforms on a specific channel, see the details in Figure of Shelly et al. ). The PCA sequentially finds principal components, which show the relationships among variables (Jolliffe, ). Compared to measuring peak-amplitude ratios, this method is known to give reliable amplitude ratios because it applies full waveforms unlike the method that uses only the 0 differences in peak amplitude values (Shelly et al. ). Figure illustrates that logarithms of amplitude ratios are approximately linearly proportional to the magnitude differences for the events of the KMA catalogue that are also listed in our event catalogue. The c values estimated from the least square method and the PCA are not significantly different (0. and 0., respectively). Of the two methods, c value is taken from the PCA, since it is a more appropriate approach for extracting patterns among variables without assigning dependent and independent variables (Jolliffe, ). For the whole set of aftershocks, we calculated the amplitude ratios between target events and KMA catalogue events in the same manner for the estimation of the c value. Each measured amplitude ratio for a target event is translated to a relative magnitude by using Eq. () and the c value of 0.. To increase the reliability of the results, the magnitudes of target events were estimated by averaging the magnitudes obtained by more than measured amplitude ratios to KMA catalogue events. Here, we define the estimated magnitudes as Mrel since Eq. () can be valid for a limited range of magnitudes. From the procedure, the magnitudes of 0 earthquakes were relatively estimated.. Spatial and temporal characteristics of aftershock activity As characteristics for aftershock occurrences, the frequency-size distribution and the aftershock decay rate have been applied in many cases (Abdelfattah et al. ; Aktar et al. 0; Ansari ; Enescu et al. ; De Gori et al. ; Wiemer & Katsumata ; 0 Zhao & Wu 0). The former characteristic can be expressed in the following way: logn = a bm ()

11 Page of where N is the cumulative number of earthquakes with magnitude equal to or greater than M. The a value and b value, which are generally treated as constants, reflect the characteristics of earthquakes in a particular region and period. However, they can change with time or location. In particular, the b value can be affected by various physical properties such as asperities, effective stress, material heterogeneity, crack density, thermal gradient, and tectonic regimes (Goebel et al. ; Raub et al. ; Schorlemmer et al. 0; Tormann et al. ; Westerhaus et al. 0; Wiemer & Katsumata ). The frequency of aftershocks generally decreases in inverse proportion to time and the 0 tendency can be expressed using Omori's formula: R(t) = K(t + c) p () where K, c, and p are constants to be determined. In particular, the p value that represents a measure of the decay rate of the seismic activities, is known to vary from 0. to. (Wiemer & Katsumata ), and many factors such as structural heterogeneity, stress and temperature influence the variation in p value, including structural heterogeneity, stress and temperature in the crust (Utsu & Ogata ). In this study, we reasoned that the minimum magnitude of completeness (Mc) that holds Eq. () changes over time due to expansion of the temporary observation network, affecting the estimation of b values and p values. Therefore, firstly we analysed the temporal variations 0 in b for every set of 00 events in chronological order by allowing duplication of 00 events, considering the time-variant Mc (Figs. a-b). We then estimated b values and p values for all aftershocks, and two event subsets on F and F for the maximum value of Mc (Figs. c- d). The b value for each event set is evaluated from the maximum likelihood method (MLM) by Aki () with a magnitude bin size (ΔM) of 0. and the uncertainty of each estimated b value is calculated by using the equation of Shi & Bolt (). The Mc is calculated from a modified goodness-of-fit method of Wiemer & Wyss (00). After the calculation of the goodness-of-fit value for every Mc with a 0. interval, we find the minimum Mc (Mcmin) of which the goodness-of-fit value is greater than 0. and take the Mc that yields the maximum goodness-of-fit value in the range of [Mcmin, Mcmin+ΔM]. The three constants and their

12 Page of standard deviations in Eq. () for each event set with magnitudes greater than Mc are estimated from the maximum likelihood procedure of Ogata (). The results of time-variant Mc and b values are presented in Figs. a-b. For the first three days, Mc decreases from. to ~ and remains at ~ for the subsequent period (Fig. a). The b value, on the other hand, is 0. ± 0.0 at the initial stage and increases to. ± 0.0 during the three days after the onset of E (Fig. b). The b values of the total data set, the events on F and F are approximately the same at 0. ± 0.0, 0. ± 0.0, and 0. ± 0.0, respectively, but the p values of the three event sets are.0 ± 0.0, 0. ± 0.0, and. ± 0.0, respectively, which are slightly different from each other (Figs. c-d). For each event set, the Mc for calculating b values and p values is fixed to., which is the 0 maximum value over time. The initial three days of decreasing Mc correspond exactly to the period when the temporary observation network is being expanded. The decrease in Mc during the early earthquake sequences may come from the high level of background noises. However, we checked that the temporal change in Mc does not appear in the result from the earthquakes detected by using only the permanent stations. The b value temporarily increased from 0. to. during the initial three day period in which the seismic observation network was being expanded. We tested that the increasing trend still appears even if Mc is fixed to the highest values (i.e.,.) and confirmed that the choice between various Mc and fixed Mc does not affect the increase of b value. The two 0 event segments marked as solid red dots are offset by three days and their b value difference is estimated as 0. (Fig. b). To determine whether the difference between the two b values is statistically significant or not, we applied the Utsu s test (Utsu ) and the probability that there would be no difference in the b value between the two sets of populations are estimated as - %, confirming that the difference in the b values of the two sets of data would be statistically significant. The tendency for b values to increase after the onset of an earthquake has been observed in other cases: the 0 Ms.0 Parkfield earthquake (Shcherbakov et al. 0) and

13 Page of September Mw. event between Izmit epicentre and Lake Sapanca (Raub et al. ). Raub et al. () estimated that the increase in b values is related not only to the 0 release of stress, but also to the unclamping of faults. Hosono & Yoshida (0) reported that the expected number of relatively large earthquakes lacks in later earthquake sequences compared with that calculated from the modified Omori formula for an early time period, and this effect may result in the relatively small b values at the initial stages. The estimated b values for all data sets, E, and E are all close to and consistent with the estimation of.0 for the KMA catalogue events (Hong et al. ). This b value of also corresponds to the commonly observed value in other earthquake sequences (Wu et al. ; Ansari ). The p value of.0 for the entire data set is considerably lower than the. of Hong et al. (), which uses the KMA catalogue events. The disparity can be attributed to the use of different fitting ranges in Omori's Law, or the additional consideration of Mc for determination of the p values.. MOMENT TENSOR SOLUTIONS AND FOCAL MECHANISMS. Method of analysis In this study, the deviatoric moment tensor solutions were determined using the ISOLA software of Sokos & Zahradnik (0) for selected earthquakes with ML. After several trials have been attempted for the appropriate stations and filtering ranges, the moment tensor solutions of selected earthquakes were determined using the waveforms recorded at four or five stations, and two different frequency ranges of Hz for the three largest events and Hz for the rest. The optimal focal depths were determined from moment tensor inversion by testing different depth ranges with a km bin. The same regional one- dimensional velocity model used in the location procedure (e.g., Kim et al. ) was used to construct Green s functions as well. For aftershocks with more than P-wave arrival times, the P-wave polarities at each station are manually measured and we applied the FOCMEC software of Snoke (0) to calculate their focal mechanisms. The focal mechanism analysis was performed only if an

14 Page of earthquake has at least polarity measurements. We sorted out the candidate focal mechanisms with minimum measurement errors for each event and exclude events from the analysis if candidate solutions contain polarity errors larger than. Among the possible solutions for each event, we selected one optimal solution of which the sum of the distances from its P, T, and B axes to all other solutions on the focal sphere is minimised. All solutions, including the selected optimal solution, are visually inspected to confirm similarity in mechanisms. We obtained reliable focal mechanisms in this procedure.. Results The detailed moment tensor inversion results of the selected events are given in Fig. and Table. Although some earthquakes have a considerable amount of compensated linear vector dipole components (~%), double-couple components are generally dominant. All inverted moment tensor solutions are classified as strike-slip events of which one nodal plane is parallel to F and F, which is consistent with previous researches (Kim et al. a, 0 Son et al. ). Assuming that the fault rupture of each earthquake propagates on F or F, the true fault plane has a NNE-SSW strike direction with a right-lateral movement. We classified the focal mechanisms estimated by P-wave first motions according to the ternary diagram of Frohlich (). The number of earthquakes that correspond to strike-slip events and thrust events were (%) and (%), respectively. The remaining events were classified as odd faulting types, while no normal faulting events were observed. From the spatial distribution of the strike-slip events and thrust events, which account for % of the calculated focal mechanisms, we can observe that both types of focal mechanisms occur on both F and F (Fig. ). The trend and plunge of the P-axis of the strike-slip events are ± and ±, 0 respectively, and those of the thrust events are ± and ±, respectively, indicating that the trend of the maximum principal stress axis is homogeneous at 0 0. However, despite coexistence of the two faulting types on both F and F, the detailed

15 Page of spatial distributions of the strike-slip and thrust events are quite different from each other, implying that the other principal stresses are locally heterogeneous. Reliability of the determined faulting types is verified by visually comparing the observed and expected P-wave amplitudes. For example, P-wave amplitudes recorded near the epicentres of strike slip events should be weak, but they should be positive and large for thrust events. In order to cross-check the reliability of the inverted fault parameters calculated by the methods implemented in the FOCMEC and ISOLA software, the focal 0 mechanisms of the four earthquakes inverted from both methods were compared and no significant differences were observed. STRESS INVERSION As mentioned in the previous section, inverted focal mechanisms of the events are mainly divided into strike-slip and thrust events under the homogeneous P-axis and their spatial distributions are different from each other. One key observation is that thrust earthquakes rarely occur at depths deeper than ~ km, whereas strike-slip earthquakes occur at greater depths (e.g., the sub-region of C in Figs. b, c, e, and f). Another important observation is that thrust events are predominantly scattered in the area with 0 depths shallower than. km and horizontal distances greater than km on the A-A profile (e.g., the sub-regions of C in Figs. b and e). The spatial heterogeneity of the fault types is indicative of the spatial heterogeneities observed in the tectonic stress field. To examine the heterogeneity in the stress field of the aftershock region, the study area is divided into five sub-regions where the spatial distribution of strike-slip and thrust events varies and the stress inversion was performed for each sub-region (Fig. ). The inversion was applied via the MSATSI software, which was redesigned from the conventional SATSI algorithm for use in the MATLAB environment (Hardebeck & Michael 0; Martínez-Garzón et al. ). No regularization for the spatial difference of the stress field is applied and one fault plane from two conjugated planes is randomly selected by considering the fault plane 0 ambiguity. Detailed inversion parameters, such as the number of fault planes used and the

16 Page of direction of each principal axis are presented in Fig.. Trends and plunges of the maximum stress components (σ ) in all sub-regions are estimated as ~0 N and ~, respectively. It represents the maximum horizontal stress (σ Hmax ) is homogeneous in the study area. However, the intermediate stress field (σ ) and the minimum stress field (σ ) are not uniform in the five sub-regions. In the depth range less than. km, σ in C and C corresponds to the minimum horizontal stress (σ hmin ) and the lithostatic stress (σ v ). At intermediate depths (. km), the amplitudes of σ and σ are so close to each other that they cannot be well constrained (Fig. ). For C at depths < km, all the principal stress orientations are well constrained due to the dominance of strike- 00 slip events. The R values defined as (σ -σ )/(σ -σ ) are equal to or greater than 0. 0 throughout all sub-regions, suggesting that the aftershock region is in the transpressional 0 regime (Bohnhoff et al. 0). 0 In C, where σ and σ are clearly distinguished as the vertical stress and minimum 0 horizontal stresses, the principal stresses normalised by vertical stress can be derived from 0 the Coulomb Friction Law and the R value, with the assumption that excessive differential 0 stresses would be released by slip along optimally oriented fault planes (Jaeger et al. 0; 0 Soh et al. ). Based on the frictional coefficient (μ) of 0. and the hydrostatic pore 0 pressure, the relative minimum horizontal stresses (σ hmin /σ v ) and the relative maximum 0 horizontal stresses (σ Hmax /σ v ) correspond to 0.0 and., respectively, which are consistent with the results of Soh et al. (). If μ varies from 0. to following Byerlee s Law (Byerlee ), σ hmin /σ v and σ Hmax /σ v have a range of and..0, respectively. If we assume a unit weight of granitic rocks for the continental crusts (. knm -, Soh et al. ), the values of σ Hmax, σ v, and σ hmin at a depth of ~ km are estimated as,, and MPa, respectively. STRESS TRANSFER ANALYSIS The 0 minute gap of the origin times of the E and E events, and the m offset of the two faults (F and F) may indicate the static stress interaction in a multi-fault system. In

17 Page of other words, the perturbed stress field caused by the seismic activity on F may have affected the spatial and temporal distribution of earthquakes on F, and vice versa. One of the favourable approaches in regards to the triggering of earthquake aftershocks is the Coulomb hypothesis, which explains the increased (or decreased) seismicity with regional static stress changes. For a specific receiver fault, the relationships amongst the Coulomb stress change (Δσ f ), shear stress change (Δτ), normal stress change (Δσ), and pore pressure change (ΔP) on this fault are given as: Δσ f = Δτ + μ(δσ + ΔP) () where μ is the coefficient of internal friction and unclamping is defined as positive for the normal stress (Lay and Wallace, ). If we use the apparent frictional coefficient μ instead of μ, which involves both the effects of pore pressure changes and normal stress changes, then Eq. () can be rewritten as (King et al. ; Reasenberg & Simpson ): Δσ f = Δτ + μ Δσ () We applied two approaches for the Coulomb stress calculations via the Coulomb. software (Lin & Stein 0; Toda et al. 0). Firstly, the Coulomb stress changes imparted by both E and E were analysed for optimally oriented strike-slip faults to resolve the temporal variation of the off-fault seismicity. The epicentres of detected events determined by the single difference method in Fig. were compared with the Coulomb stress changes on the optimally oriented receiver fault for the maximum stress field of the N E direction with a strength of bars (Hong et al. ). We then calculated the Coulomb stress changes on F, F, and F from the E and (or) E slip models to check whether the static 0 stress change of each fault plane correlates with off-fault seismicity. Following the scheme of Hong et al. (), the apparent frictional coefficient μ, Young s modulus, and Poisson s ratio were set to 0., 0 GPa, and 0., respectively. The finite rupture models of E and E of Uchide & Song () were used to configure the slip amounts on the faults. Fig. illustrates the Coulomb stress changes on optimally oriented faults at a depth of km generated by both E and E with the epicentre of all located earthquakes. In the region

18 Page of far from F and F (i.e., the area outside of the black box in Fig. ), the seismicity increases within the area of positive Coulomb stress changes and decreases within the negative lobes. More specifically, the effects of the Coulomb stress changes were inversely proportional to the epicentral distances up to ~0 km (see the upper right panels of Fig. ). 0 The calculated Coulomb stress change on Fb for the fault model of E gives a negative value at the loci of E. However, an elliptical patch with positive Coulomb stress changes was observed near E, not more than 00 m away (Fig. a). If the uncertainties of the finite slip model and relocated hypocentre are taken into account, the Coulomb stress changes by E may have advanced the occurrence of E, although dynamic stress changes, fault geometries, and the regional stress state can affect the result (Gomberg et al. 0). The fault model of E yields positive Coulomb stress changes on an inverted T-shaped area (Fig. b). The seismic activity on F for hours after E overlaps with the area, implying that the Coulomb stress change caused by the occurrence of E may also have affected the seismicity on F. Slips from both finite fault models of E and E result in positive Coulomb 0 stress changes on the sinistral slip of F (~ bar), which may indicate that the Coulomb stress changes imparted by the two largest events trigger the sinistral slip of E on a conjugate fault striking to ESE-WNW direction. DISCUSSION. Fault system complexity The detailed spatial distribution of the Gyeongju earthquakes represents that E and E occur on the two parallel fault planes of F and F with an offset of m. Complex fault systems with two or more parallel faults can also be found in other regions (Durá- Gómez & Talwani 0; Rabak et al. ; Yano & Matsubara ). Yano & Matsubara 0 () reported that part of the Kumamoto earthquake sequence is aligned on two vertical fault planes in the northeastern area of Mt. Aso with two moderate-sized aftershocks (M ) located on either side of the fault plane. This is consistent with our results in that major events in a sequence separately occurred on visually identified faults.

19 Page of The third largest event (E) located at the southwestern tip of the aftershock region and a streak of earthquakes following E are aligned on F, which is perpendicular to F and F (Fig. a). Conjugate fault systems can be found in many previous studies ( Sumatra earthquake, Meng et al. ; 00 Wharton Basin earthquake, Robinson et al. 0; 00 Western Tottori earthquake, Fukuyama et al. 0). In particular, the 00 Western Tottori earthquake has characteristics similar to the Gyeongju earthquake in that relatively small 0 conjugate faults ruptured at the tip of the main faults. The dip of F estimated from relocated earthquakes varies at a depth of ~. km (Fig. d). Therefore, we suggest that the fault plane is bent or two distinct faults having different dips are crossing at the region. To check that an inaccurate velocity model affects the fault structure, various velocity models were tested and any significant changes were not observed.. Interactions between complex faults The first couple of hours of spatio-temporal variations in seismicity on F and F are very characteristic. In the period between E and E, most earthquakes are confined on F at a 0 depth range of - km. Seismic activities on F abruptly diminished after the occurrence of E for at least hours (Fig. b). Instead, most of earthquakes following E are located at a depth shallower than km. This is in contrast to the deep seismic activity on F before E. The decreased seismicity on F starting from the occurrence of E cannot be fully explained by static stress changes because the Coulomb stress changes on F imparted by the finite slip model of E are positive in some areas (Fig. b). Therefore, other factors such as irregular fault geometry, heterogeneous elastic modulus, and complex stress states are required to explain it.. A possible microscale heterogeneity in the complex fault system 00 The b values for F and F are almost unity, but the p value for F is estimated to be 0 significantly less than that of F. The difference in p values may result from the decreased

20 Page of seismicity of F after the occurrence of E. To check the effect of initial earthquake 0 sequences, the p value for F was re-estimated after removing the earthquakes before E. 0 We confirmed that it remains at 0. ± 0.0, which is still less than those for all data sets and 0 events on F. Therefore, it is reasonable that the decay rate of the seismicity on F is 0 generally slower than that on F. Still, many factors can attribute to the spatial changes in 0 the p values: stress, fault heterogeneity, and crustal heat flow (Enescu et al. ; Wiemer 0 & Katsumata ), requiring further analysis on this parameter in order to pinpoint the exact 0 cause. The maximum Mrel of the reverse fault earthquakes determined by focal mechanism inversion corresponds to.. None of the fault types calculated from the moment tensor inversion has a thrust regime. Indeed, from the hypocentre distribution, no characteristic structure is identified along the NNE-SSW direction, the strike direction of the reverse faults. Therefore, the reverse fault events occur on smaller fault planes than those of the strike-slip events, and microcracks that cannot be identified by the hypocentre distribution may develop throughout the study area, generating the thrust events that occur in response to the heterogeneous stress field at each location.. Model Implication Background seismicity around the study area is concentrated along mapped faults, which indicates that earthquakes occurred on preexisting faults where local stress had accumulated rather than at newly formed faults (Fig. a; Lee & Jin ; Han et al. ; Kim et al. b). The Gyeongju earthquakes ruptured beneath the Yangsan fault with a rupture length of 0 0 km and a horizontal offset of ~ km (Fig. a; Kyung 0; Kim et al. b), but the general trend identified from the aftershocks is rotated clockwise by º from two mapped faults near epicentres. Kim et al. (a) interpreted the angular difference as Riedel shears in a strike-slip duplex. However, we additionally identified F and thrust faults striking in different directions from the general trends. Therefore, a more complicated structure rather than simple Riedel shears are needed to explain minor faults. The

21 Page of hypothesis proposed by Uchide & Song () suggested that the fault rupture models for E and E occurred in a fault jog of extensional oversteps and the pull-apart stress between them would produce normal faults in the aftershock area. However, normal faulting events were not observed by us; we suspect that the faulting types of the Gyeongju aftershocks are more likely to be affected by local stress fields and preexisting faults, rather than the directivity of rupture propagations. CONCLUSIONS In this study, the source parameters of Gyeongju earthquakes from three years of seismicity near the aftershock zone were analysed using data from a temporary seismic network as 0 well as two permanent seismic networks. For the three largest earthquakes with ML >, distinct fault planes at a depth of km were identified. The two largest events (E and E) occurred on two sub-parallel faults (F and F) striking in the NNE-SSW direction, whereas the third largest event (E) occurred on a vertical fault plane (F) perpendicular to F and F. The focal mechanisms estimated from the first motions of the P-waves are composed of strike-slip events, thrust events, and their intermediate types of events, which are not matched with an extensional fault jog and require modifications in the duplex strike- slip model. The strike-slip events are in accordance with the inverted moment-tensors of the selected events. The difference of the spatial distribution between the strike-slip events and thrust events indicates the heterogeneity between the minimum horizontal stress and 0 lithostatic stress. In particular, reverse fault events are hardly observed deeper than km, implying that the minimum horizontal stress is no longer greater than the overburden pressure below a depth of km. From the magnitudes of the earthquakes estimated by scaling the amplitude ratios and the magnitude differences, the temporal variations of the b values and Mc were analysed over time. The b values increased over the initial three days, and did not vary significantly afterward. If the b values are assumed to be a stress-metre indicator, the increase of b values over time seems to reflect the stress releasing process. The b values for both fault

22 Page of planes of F and F were close to, whereas the p value for F is smaller than those for the whole data or events on F. The Coulomb stress changes caused by the two largest 0 events are correlated with the spatial change of the seismic activities off the aftershock region. It is possible that the occurrence of E is facilitated by the Coulomb stress changes imparted by E and the Coulomb stress changes caused by E affects the seismicity of F, vice versa. However, a sudden reduction in the seismic activity on F following the onset of E could not be explained by static stress changes, indicating that complex fault properties are required to resolve it. The positive Coulomb stress changes for the sinistral slip of F generated by E and E may promote the occurrence of E at the southeast tip of the aftershocks. We suggest that our observations, such as the complicated spatio-temporal distribution of seismicity, heterogeneous stress field, and the p value difference between the two parallel faults partially represents the complex fault properties of the Gyeongju 0 earthquakes. More detailed models for fault geometry, local stress conditions, and heterogeneous rock materials are required to fully understand the fault reactivation process of the complex fault system in the intraplate region. Acknowledgements We thank Korea Meteorological Administration (KMA) and Korea Institute of Geosciences and Mineral Resources (KIGAM) for sharing continuous waveform data for this study. We are grateful to not only all landowners and organizations who allowed us to install and operate temporary seismic stations on their properties but also graduate students in the Gyeongju earthquake research group for maintaining the temporary seismic network. This 0 work was supported by the Nuclear Safety Research Program through the Korea Foundation Of Nuclear Safety (KoFONS) using the financial resource granted by the Nuclear Safety and Security Commission (NSSC) of the Republic of Korea (No. 00). References Abdelfattah, A.K., Mogren, S. & Mukhopadhyay, M.,. Mapping b value for 0 Harrat

23 Page of Lunayyir earthquake swarm, western Saudi Arabia and Coulomb stress for its mainshock, J. Volcanol. Geoth. Res.,, -. Adams, J., Wetmiller, R.J., Hasegawa, H.S. & Drysdale, J.,. The first surface faulting from a historical intraplate earthquake in North America, Nature,, -. 0 Aki, K.,. Maximum likelihood estimate of b in the formula logn=a-bm and its confidence limits, Bull. Earthq. Res. Inst.,, -. Aktar, M., Özalaybey, S., Ergin, M., Karabulut, H., Bouin, M.-P., Tapırdamaz, C., Bicmen, F., Yoruk, A. & Bouchon, M., 0. Spatial variation of aftershock activity across the rupture zone of the August Izmit earthquake, Turkey, Tectonophysics,, -. Ansari, S.,. Aftershocks properties of the Shonbe Mw. earthquake, central Zagros, Iran, J. Asian Earth Sci.,, -. Assumpção, M., Schimmel, M., Escalante, C., Barbosa, J.R., Rocha, M. & Barros, L.V., 0. Intraplate seismicity in SE Brazil: stress concentration in lithospheric thin spots, Geophys. J. Int.,, Bianchi, M.B., Assumpção, M., Rocha, M.P., Carvalho, J.M., Azevedo, P.A., Fontes, S.L., 0 Dias, F.L., Ferreira, J.M., Nascimento, A.F., Ferreira, M.V. & Costa, I.S.L.,. The 0 Brazilian Seismographic Network (RSBR): Improving seismic monitoring in Brazil, Seismol. 0 Res. Lett.,, -. 0 Bodin, P. & Horton, S., 0. Source parameters and tectonic implications of aftershocks of 0 the M w. Bhuj earthquake of January 0, Bull. Seismol. Soc. Am.,, -. 0 Bohnhoff, M., Grosser, H. & Dresen, G., 0. Strain partitioning and stress rotation at the 0 North Anatolian fault zone from aftershock focal mechanisms of the Izmit M w =. 0 earthquake, Geophys. J. Int.,, -. 0 Byerlee, J.,. Friction of rocks, Pure appl. Geophys.,, -. Chang, K.H.,. Late Mesozoic stratigraphy, sedimentation, and tectonics of southeastern Korea, J. Geol. Soc. Korea,, 0 (in Korean). Chen, Y., Tsoi, K.L., Chen, F., Gao, Z.H., Zou, Q.J. & Chen, Z.L.,. The great Tangshan earthquake of : an anatomy of disaster, p., Pergamon Press, New York.

24 Page of De Gori, P., Akinci, A., Pio Lucente, F. & Kılıç, T.,. Spatial and temporal variations of aftershock activity of the October M w. Van, Turkey, earthquake, Bull. Seismol. Soc. Am.,, -. Durá-Gómez, I. & Talwani, P., 0. Finding faults in the Charleston area, South Carolina:. Seismological data, Seismol. Res. Lett., 0, -00. Enescu, B., Enescu, D. & Ito, K.,. Values of b and p: their variations and relation to physical processes for earthquakes in Japan and Romania, Rom. J. Phys,, 0-0. Ferreira, J.M., Oliveira, T., Takeya, M.K. & Assumpção, M.,. Superposition of local and regional stresses in northeast Brazil: evidence from focal mechanisms around the Potiguar marginal basin, Geophys. J. Int.,, -. Frohlich, C.,. Triangle diagrams: ternary graphs to display similarity and diversity of earthquake focal mechanisms, Phys. Earth and Planet. Inter.,, -. Fukuyama, E., Ellsworth, W.L., Waldhauser, F. & Kubo, A., 0. Detailed fault structure of the 00 western Tottori, Japan, earthquake sequence, Bull. Seismol. Soc. Am.,, -. Gangopadhyay, A. & Talwani, P., 0. Symptomatic features of intraplate earthquakes, Seismol. Res. Lett.,, -. Goebel, T.H.W., Becker, T.W., Schorlemmer, D., Stanchits, S., Sammis, C., Rybacki, E. & Dresen, G.,. Identifying fault heterogeneity through mapping spatial anomalies in acoustic emission statistics, J. Geophys. Res.,, B0, doi:./jb00. Gomberg, J., Bodin, P. & Reasenberg, P.A., 0. Observing earthquakes triggered in the near field by dynamic deformations, Bull. Seismol. Soc. Am.,, -. Grigoli, F., Cesca, S., Amoroso, O., Emolo, A., Zollo, A. & Dahm, T.,. Automated seismic event location by waveform coherence analysis, Geophys. J. Int.,, -. Han, M., Kim, K.H., Son, M. & Kang, S.Y.,. Current microseismicity and generating faults in the Gyeongju area, southeastern Korea, Tectonophysics,, -. 0 Hardebeck, J.L. & Michael, A.J., 0. Damped regional-scale stress inversions: Methodology and examples for southern California and the Coalinga aftershock sequence, J.

25 Page of Geophys. Res.,, B0, doi:./0jb00. Hauksson, E. & Shearer, P., 0. Southern California hypocenter relocation with waveform cross-correlation, Part : Results using the double-difference method, Bull. Seismol. Soc. Am.,, -0. Hong, T.-K., Lee, J., Kim, W., Hahm, I.-K., Woo, N.C. & Park, S.,. The September M L. midcrustal earthquake in the Korean Peninsula and its seismic implications, Geophys. Res. Lett.,, -. Hosono, K. & Yoshida, A., 0. Do large aftershocks decrease similarly to smaller ones? 0 Geophysical research letters,, Hwang, B.H., Son, M., Yang, K., Yoon, J. & Ernst, W.G., 0. Tectonic evolution of the Gyeongsang Basin, southeastern Korea from 0 Ma to the present, based on a strike-slip and block rotation tectonic model, Int. Geol. Rev., 0, -. Jaeger, J.C., Cook, N.G.W. & Zimmerman, R., 0. Fundamentals of Rock Mechanics, p., John Wiley & Sons. Johnston, A.C.,. The seismicity of stable continental interiors, in Earthquakes at North-Atlantic Passive Margins: Neotectonics and Postglacial Rebound, pp. -, ed. Gregersen, S. & Basham, P.W., Springer, Dordrecht. Jolliffe, I.,. Principal component analysis, in International encyclopedia of statistical 0 science, pp. -, Springer, Berlin, Heidelberg. Kim, S., Rhie, J. & Kim, G.,. Forward waveform modelling procedure for -D crustal velocity structure and its application to the southern Korean Peninsula, Geophys. J. Int.,, -. Kim, K.-H., Kang, T.-S., Rhie, J., Kim, Y.-H., Park, Y., Kang, S.-Y., Han, M.-H., Kim, J.M., Park, J., Kim, M., Kong, C.-H., Heo, D.., Lee, H.-Y., Park, E., Park, H.J., Lee, S.-J., Cho, S.W., Woo, J.-U., Lee, S.-H. & Kim J.W.,. The September Gyeongju earthquakes:. Temporary seismic network for monitoring aftershocks, Geosci. J.,, -. Kim, Y.-H., Rhie, J., Kang, T.-S., Kim, K.-H., Kim, M. & Lee, S.-J.,. The September

26 Page of Gyeongju earthquakes:. Observation and remaining questions, Geosci. J.,, -. Kim, K., Kim, J.-M., Han, M.-H., Kang, Y.S., Son, M., Kang, T.-S., Rhie, J.-K., Kim, Y.-H., Park, Y.-C., Kim, H.-J., You, Q.Y. & Hao, T.-Y.,. Deep fault plane revealed by high- precision locations of early aftershocks following the September ML. Gyeongju, Korea, earthquake, Bull. Seismol. Soc. Am., -. Kim, Y., He, X., Ni, S., Lim, H. & Park, S.-C.,. Earthquake source mechanism and rupture directivity of the September M w. Gyeongju, South Korea, earthquake, Bull. Seismol. Soc. Am.,, -. Kim, Y.-S., Kim, T., Kyung, J.B., Cho, C.S., Choi, J.-H. & Choi C.U.,, Preliminary study 0 on rupture mechanism of the. Gyeongju earthquake, J. Geol. Soc. Korea,, 0- (in Korean). King, G.C., Stein, R.S. & Lin, J.,. Static stress changes and the triggering of earthquakes, Bull. Seismol. Soc. Am.,, -. Kyung, J.B., 0. Paleoseismology of the Yangsan fault, southeastern part of the Korean peninsula, Ann. Geophys.,, -. Lahr J.C.,. HYPOELLIPSE: a computer program for determining local earthquake hypocentral parameters, magnitude, and first-motion pattern (YK compliant version), version.0, pp., Open-File Report, U.S. Geol. Surv., Reston, VA. Lay, T. & Wallace, T.C.,. Modern global seismology, p., Academic Press, London. 0 Lee, K. & Jin, Y.G.,. Segmentation of the Yangsan Fault System: Geophysical studies on major faults in the Kyeongsang Basin, J. Geol. Soc. Korea,, -. Lee, J., Ryoo, Y., Park, S.C., Ham, Y.M., Park, J.S., Kim, M.S., Park, S.M., Cho, H.G., Lee, K.S., Kim, I.S., Kim, H.S. & Bae, S.H.,. Seismicity of the M L. Gyeongju earthquake and aftershocks in South Korea, Geosci. J.,, -. Lin, J. & Stein, R.S., 0. Stress triggering in thrust and subduction earthquakes and stress interaction between the southern San Andreas and nearby thrust and strike slip faults, J. Geophys. Res.,, B00, doi:./0jb000.

27 Page of Martínez-Garzón, P., Kwiatek, G., Ickrath, M. & Bohnhoff, M.,. MSATSI: A MATLAB package for stress inversion combining solid classic methodology, a new simplified user- 00 handling, and a visualization tool, Seismol. Res. Lett.,, Matos, C., Custódio, S., Batlló, J., Zahradník, J., Arroucau, P., Silveira, G. & Heimann, S., 0. An active seismic zone in intraplate west Iberia inferred from high resolution 0 geophysical data, J. Geophys. Res.,, Mendoza, C., & Hartzell, S. H.,. Aftershock patterns and main shock faulting, Bull. 0 Seismol. Soc. Am.,, -. 0 Meng, L., Ampuero, J.P., Stock, J., Duputel, Z., Luo, Y. & Tsai, V.C.,. Earthquake in a 0 maze: Compressional rupture branching during the Mw. Sumatra earthquake, 0 Science,, -. 0 Ogata, Y.,. Estimation of the parameters in the modified Omori formula for aftershock frequencies by the maximum likelihood procedure, J. Phys. Earth,, -. Park, S.C., Yang, H., Lee, D.K., Park, E.H. & Lee, W.J.,. Did the September Gyeongju, South Korea earthquake cause surface deformation? Geosci. J.,, -. Rabak, I., Langston, C., Bodin, P., Horton, S., Withers, M. & Powell, C.,. The Enola, Arkansas, intraplate swarm of 0, Seismol. Res. Lett.,, -. Raub, C., Martínez-Garzón, P., Kwiatek, G., Bohnhoff, M. & Dresen, G.,. Variations of seismic b value at different stages of the seismic cycle along the North Anatolian Fault Zone in northwestern Turkey, Tectonophysics,, -. Rastogi, B.K., Gupta, H.K., Mandal, P., Satyanarayana, H.V.S., Kousalya, M., Raghavan, R., Jain, R., Sarma, A.N.S., Kumar, N. & Satyamurty, C., 0. The deadliest stable continental region earthquake occurred near Bhuj on January 0, J. Seismol.,, 0-. Reasenberg, P.A. & Simpson, R.W.,. Response of regional seismicity to the static stress change produced by the Loma Prieta earthquake, Science,, -0. Robinson, D.P., Henry, C., Das, S. & Woodhouse, J.H., 0. Simultaneous rupture along two conjugate planes of the Wharton Basin earthquake, Science,, -. Ross, Z.E., Kanamori, H., Hauksson, E. & Aso, N.,. Dissipative intraplate faulting

28 Page of during the Mw. Tottori, Japan earthquake, J. Geophys. Res.,, -. Schorlemmer, D., Wiemer, S. & Wyss, M., 0. Variations in earthquake-size distribution across different stress regimes, Nature,, -. Shcherbakov, R., Turcotte, D.L., & Rundle, J.B., 0. Scaling properties of the Parkfield aftershock sequence, Bull. Seismol. Soc. Am.,, S-S. Shelly, D.R., Ellsworth, W.L. & Hill, D.P.,. Fluid-faulting evolution in high definition: Connecting fault structure and frequency-magnitude variations during the Long Valley Caldera, California, earthquake swarm, J. Geophys. Res.,, -. Shi, Y. & Bolt, B.A.,. The standard error of the magnitude-frequency b value, Bull. Seismol. Soc. Am.,, -. Snoke, J.A., 0. FOCMEC: Focal mechanism determinations, in International Handbook of Earthquake and Engineering Seismology, Chapter., pp. -, eds Lee, W.H.K., Kanamori, H., Jennings, P.C. & Kisslinger, C., Academic Press, San Diego, CA. So, M.M.L., Mote, T. & Pappin, J.W.,. Seismic hazard assessment of South Korea. 0 Japan. Geotech, Soc. Spec. Pub.,, -0. Soh, I., Chang, C., Lee, J., Hong, T.K. & Park, E.S.,. Tectonic stress orientations and magnitudes, and friction of faults, deduced from earthquake focal mechanism inversions over the Korean Peninsula, Geophys. J. Int.,, 0-. Son, M., Cho, C.S., Shin, J.S., Rhee, H.M. & Sheen, D.H.,. Spatiotemporal distribution of events during the first three months of the Gyeongju, Korea, earthquake sequence. Bull. Seis. Soc. Am.,, 0-. Sokos, E.N. & Zahradnik, J., 0. ISOLA a Fortran code and a Matlab GUI to perform multiple-point source inversion of seismic data, Comput. Geosci.,, -. Soto-Cordero, L., Meltzer, A. & Stachnik, J.C.,. Crustal structure, intraplate seismicity, 0 and seismic hazard in the Mid-Atlantic United States, Seismol. Res. Lett.,, -. Sykes, L.R.,. Intraplate seismicity, reactivation of preexisting zones of weakness, alkaline magmatism, and other tectonism postdating continental fragmentation, Rev. Geophys.,, -.

29 Page of Talwani, P. & Rajendran, K.,. Some seismological and geometric features of intraplate earthquakes, Tectonophysics,, -. Talwani, P.,. Unified model for intraplate earthquakes, in Intraplate Earthquakes, pp. -, ed. Talwani. P., Cambridge University Press, New York. Tavakoli, B., Pezeshk, S. & Cox, R.T.,. Seismicity of the New Madrid seismic zone derived from a deep-seated strike-slip fault, Bull. Seismol. Soc. Am., 0, -. 0 Toda, S., Stein, R.S., Richards-Dinger, K. & Bozkurt, S.B., 0. Forecasting the evolution of seismicity in southern California: Animations built on earthquake stress transfer, J. Geophys. Res., 0, B0S, doi:./0jb00. Tormann, T., Wiemer, S. & Mignan, A.,. Systematic survey of high-resolution b value imaging along Californian faults: Inference on asperities, J. Geophys. Res.,, -. Uchide, T. & Song, S.G.,. Fault rupture model of the Gyeongju, South Korea, earthquake and its implication for the underground fault system, Geophys. Res. Lett.,, -. Utsu, T.,. On seismicity, in mathematical seismology (VII), Coop. Res. Rep. Inst. Stat. Math., Tokyo, Utsu, T. & Ogata, Y.,. The centenary of the Omori formula for a decay law of aftershock activity, J. Phys. Earth,, -. Waldhauser, F. & Ellsworth, W.L., 00. A double-difference earthquake location algorithm: Method and application to the northern Hayward Fault, California, Bull. Seismol. Soc. Am., 0, -. Wang, Z., 0. Seismic hazard and risk assessment in the intraplate environment: The New Madrid seismic zone of the central United States. Geol. Soc. Am.,, -. Westerhaus, M., Wyss, M., Yilmaz, R. & Zschau, J., 0. Correlating variations of b values and crustal deformations during the 0s may have pinpointed the rupture initiation of the M w=. Izmit earthquake of August, Geophys. J. Int.,, -. 0 Wiemer, S. & Katsumata, K.,. Spatial variability of seismicity parameters in aftershock zones, J. of Geophys. Res.,, -.

30 Page of Wiemer, S. & Wyss, M., 00. Minimum magnitude of completeness in earthquake catalogs: Examples from Alaska, the western United States, and Japan, Bull. Seismol. Soc. Am., 0, -. Withers, M., Aster, R., Young, C., Beiriger, J., Harris, M., Moore, S. & Trujillo, J.,. A comparison of select trigger algorithms for automated global seismic phase and event detection, Bull. Seismol. Soc. Am.,, -. Woo, J.U., Kim, J., Rhie, J. & Kang, T.S.,. Characteristics in hypocenters of microseismic events due to hydraulic fracturing and natural faults: a case study in the Horn 0 River Basin, Canada, Geosci. J.,, -. Wu, Q., Chapman, M.C. & Beale, J.N.,. The aftershock sequence of the Mineral, Virginia, earthquake: Temporal and spatial distribution, focal mechanisms, regional stress, and the role of Coulomb stress transfer, Bull. Seismol. Soc. Am.,, -. Yano, T.E. & Matsubara, M.,. Effect of newly refined hypocenter locations on the seismic activity recorded during the Kumamoto Earthquake sequence, Earth, Planets and Space,, doi:./s Zhao, Y.Z. & Wu, Z.L., 0. Mapping the b values along the Longmenshan fault zone before and after the May 0, Wenchuan, China, Ms.0 earthquake, Nat. Hazards Earth Syst. Sci.,, Zoback, M.L.,. First- and second-order patterns of stress in the lithosphere: The world 0 stress map project, J. Geophys. Res.,, 0-.

31 Page of 0 0 TABLES 0 Table. Fault parameters for the three major earthquakes. Fault ID Width (km) Length (km) Max Strike / Dip (º, Strike / Dip (º, Min Depth Number of Depth Area (km ) aftershock Moment tensor (km) Earthquake (km) distribution) inversion) F.. / / 0 Fa..... / / 0 Fb.... / F / NA / 0 0

32 Page of 0 0 Table. Source parameters of the ten earthquakes used in the moment tensor analysis. Event Origin Event ID Time (UTC) (yy/mm/dd, Latitude (ºN) Longitude (ºE) hh:mm:ss.ss) /0/, (E) :::0.. /0/, (E) ::... /0/, ::..0. /0/, ::.0.0. /0/, ::..0.0 /0/, (E) ::...0 /0/, ::... Depth (DD; MTI, km) Magnitude(Mw) Strike (º) Dip (º) Rake (º).; ;. / / - / -.;. / 0 / /.;.0 / 0 / 0-0 / -.;. / / - / -.00;. / 0 / /.00; / ; / ; - / - ;. / 0 0 / 0 0 /.; / ; / ; - / - ;. / 0 / / //0,...;. / ; / ; / ;

33 Page of 0 ::0. / 0 / / //,.; / ; / ; / ;..0. ::. / / / //,.;.. 0::.. / / - / - 0 The depth of each event is calculated from both Double-Difference (DD) relocation and the moment tensor inversion (MTI). 0 If we obtain the focal mechanism of an event using first P-wave motions, we ordinarily describe the fault parameters obtained from the 0 waveform inversion and first P-wave motions.

34 Page of FIGURES Figure. The distribution of the stations used for the detection of events and determination their locations, magnitudes, and focal mechanisms. The three major events (E, E, and E) are shown with their moment tensors obtained from the low frequency waveform inversion method. The two faults with surface expressions near the main shock are denoted as green lines. Other faults and lineaments with surface expression are shown as dark grey lines. Six broadband sensors operated by Korean Meteorological Administration (KMA) and Korea Institute of Geoscience and Mineral Resources (KIGAM) are illustrated as green and blue triangles and the temporary broadband sensors are represented by yellow triangles. Two short-period sensors from KMA are denoted as blue squares. The red box corresponds to the regions in Figs. b and a. Major geological units are separated with black lines in the inset: PB (Pyungnam Basin), OB (Ongjin Basin), IB (Imjingang Belt), GM (Gyeonggi Massif), OFB (Okcheon Fold Belt), YM (Youngnam Massif), GB (Gyeongsang Basin), and YB (Yeonil

35 Page of Basin). The box with red lines in the inset indicates the region of the main figure. The five broadband seismometers shown in the inset are used to determine the moment tensor solutions of ten selected events. Figure. (a) Distribution of the hypocentres calculated from the visually inspected P-phase and S-phase arrival times and (b) a zoomed in veiw of the region of interest. The seismicity before and after the onset of E are shown with red and black dots, respectively. The name of major surface ruptures are presented with abbreviations: GUF (Gaum Fault), MiRF (Miryang Fault), MoRF (Moryang Fault), YSF (Yangsan Fault), DRF (Dongrae Fault), USF (Ulsan Fault) and DCF (Deokcheon Fault). The blue circled area and magenta circled area have experienced temporal changes in seismicity (see the details in the Section.). The red box in (a) corresponds to the zoomed in region of (b). The green, red, and blue stars in (b) are the locations of E, E, and E, respectively. The other symbols are matched with those in Fig.. 0

36 Page of Figure. Distribution of the relocated hypocentres in the aftershock region. (a) Distribution of the epicentres of the Gyeongju earthquakes. (b) Five-hour seismicity at the early stage of aftershocks as a function of focal depth. Four different symbols are used for the events in four selected periods. (c-d) The depth distributions along A-A (oriented NºE) and B-B (oriented NºE). The three moment tensor solutions of E, E, and E are distinguished as three different colors of compressional quadrants: green (E), red (E),

37 Page of and blue (E). The blue dots in (a), (c), and (d) represent one hour of seismic activities following E, and estimated rupture propagation directions of E and E by Uchide and 0 Song () are denoted as green and red arrows, respectively. Fault geometries of F, Fa, and Fb are illustrated as green, yellow, and red lines, respectively. The other symbols are matched with those in Fig.. Figure. A three-dimensional schematic diagram to illustrate the spatial distribution of the aftershocks and the geometry of the four faults (F, Fa, Fb, and F). Detailed information for each fault is summarised in Table.

38 Page of Figure. Determination of the scaling parameter c in Eq. () by using KMA catalogue events. For an event pair, the logarithm of the amplitude ratio measured from waveform data and the magnitude differences between two events correspond to the abscissa and the ordinate of the graph. Two methods were tested for estimation of the c value: the least squares method (red line; c = 0.) and the principal component analysis method (blue line; c = 0.).

39 Page of Figure. (a) Event magnitudes as a function of time. Temporal variation in Mc is represented as a green line. (b) The b value variation as a function of time. The horizontal and vertical error bars indicate one standard deviation of event origin times and b values, 0 respectively. The b value difference between two event sets denoted as solid red dots was tested by Utsu s test (see section.). (c) The frequency-magnitude distribution of the three event clusters: all events (blue dots), F (green dots), and F (red dots). A grey line represents the Mc used for estimation of b values. (d) The occurrence rate of the three event clusters: all events (blue dots), F (green dots), and F (red dots). Three lines with different colors in (c) and (d) represent the obtained G-R law and modified Omori s law, respectively.

40 Page of Figure. The spatial distribution of the selected events with their deviatoric moment tensors. Each number above the moment tensor in (a) represents the moment magnitude of the event. The compressive quadrants of the three largest events in (b) and (c) are colored in green (E), red (E), and blue (E). All the relocated hypocentres are denoted as grey dots on map (a) and cross-sections (b) and (c). The other symbols are matched with those in Fig..

41 Page 0 of Figure. The hypocentre distribution of the strike-slip events (a-c) and thrust events (d-f) classified by Frohlich () and the histograms for the number of strike-slip and thrust 0 events as a function of focal depth (right panels of c and f). The moment tensor solutions of E, E, and E are illustrated in (a-c) with green, red, and blue compressional quadrants.

42 Page of The subregion of C-C in (b) and (e) divided by blue lines represent the domains for the stress inversion (Fig. ). The other symbols are matched with those in Fig..

43 Page of Figure. Stress inversion results for the events in the different blocks (C-C). Firstly, the study area is divided into the left section (C, C, and C) and the right section (C and C) at the km point on the A-A profile. Then, each section is split into two or three subsections with different depth ranges. For each subplot, the depth ranges, relative stress 00 magnitude (R), and number of focal mechanisms used (N) are shown below the inversion 0 results. Each subplot is illustrated with a pie-chart representing the relative portions of strike- 0 slip events (red), thrust events (blue), and other types of events (green). The three principal 0 stress components (σ > σ > σ ) are represented with 00 bootstrapping results. The 0 maximum principal stress (σ ) is well resolved with low uncertainty for all the subsections. 0 Except for C, the other principal stress components are not constrained due to the mixed 0 population of strike-slip events and thrust events. 0

44 Page of Figure. Coulomb stress changes on optimally oriented receiver faults at a depth of km. Each sub-fault element is assumed to be an area of km. The seismicity before and after the onset of E illustrated in Fig. a are shown with red and green dots, respectively. The off-fault seismic activities are regarded as the earthquakes that occur outside the black box. For each epicentral distance bin, the percentage difference between the off-fault earthquake densities of the areas with increased Coulomb stress changes and those with

45 Page of decreased Coulomb stress changes is illustrated in the inset histogram. The other symbols are the same as those in Fig.. Figure. Coulomb stress changes imparted by finite slip models of Uchide and Song () on the sub-fault elements of other faults. (a) Coulomb stress changes on Fb caused by the finite slip model for E. The seismic activities on F between the onsets of E and E are denoted as yellow dots. (b) Coulomb stress changes on F caused by the finite slip model for E. Five hours of seismicity following E is shown as white dots. The green and red stars correspond to the locations of E and E, respectively.