Bulletin of the Seismological Society of America

Aftershock sequence and statistics of the 2017 MW 5.5 Pohang earthquake, SouthKorea: implication of fault heterogeneity and post-seismic relaxation

--Manuscript Draft--

Manuscript Number:

Article Type: Article

Section/Category: Regular Issue

Full Title: Aftershock sequence and statistics of the 2017 MW 5.5 Pohang earthquake, SouthKorea: implication of fault heterogeneity and post-seismic relaxation

Corresponding Author: Junkee Rhie, Ph.D.Seoul National UniversitySeoul, KOREA, REPUBLIC OF

Corresponding Author's Institution: Seoul National University

Corresponding Author E-Mail: rhie@snu.ac.kr

Order of Authors: Jeong-Ung Woo

Minook Kim

Junkee Rhie, Ph.D.

Tae-Seob Kang, Ph.D.

Abstract: The sequence of foreshocks, mainshock, and aftershocks associated with a faultrupture are the result of interactions of complex fault systems, the tectonic stress field,and fluid movement. Analysis of shock sequences can aid our understanding of thespatial distribution and magnitude of these factors, as well as providing a seismichazard assessment. The 2017, M W 5.5 Pohang earthquake sequence occurredfollowing fluid-induced seismic activity at a nearby enhanced geothermal system siteand is an example of reactivation of a critically stressed fault system in the PohangBasin, South Korea. We created an earthquake catalog based on unsupervised data-mining and measuring the energy ratio between short- and long-window seismogramsrecorded by a temporary seismic network. The spatial distribution of approximately4,000 relocated aftershocks revealed four fault segments striking southwestward. Wealso determined that the three largest earthquakes ( M > 4) were located at theboundary of two fault segments. We infer that locally concentrated stress at thejunctions of the faults caused such large earthquakes and that their ruptures onmultiple segments can explain the high proportion of non-double couple components.The area affected by aftershocks expands to the southwest and northeast by 0.5 and 1km decade -1 , respectively, which may result from post-seismic deformation orsequentially transferred static Coulomb stress. The b -values of the Gutenberg-Richter relationship temporarily increased during the aftershock period, suggesting thatthe stress field was perturbed. The b -values were generally low (< 1) and locallyvariable throughout the aftershock area, which may be due to the complex faultstructures and material properties. Furthermore, the mapped p -values of the Omorilaw vary along strike, which may indicate anisotropic expansion speeds in theaftershock region.

Author Comments: The first author, Jeong-Ung Woo will receive a Ph.D degree in February 2020. In theauthor affiliation, I fill the blank for the current state.

Suggested Reviewers: Kwang-Hee KimPusan National Universitykwanghee@pusan.ac.krHe is the first author of the paper "Assessing whether the 2017 Mw 5.4 Pohangearthquake in South Korea was an induced event" published at Science.

Francesco GrigoliETHfrancesco.grigoli@sed.ethz.chHe is the first author of the paper "The November 2017 Mw 5.5 Pohang earthquake: A

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation

possible case of induced seismicity in South Korea" published at Science.

Grzegorz KwiatekGFZkwiatek@gfz-potsdam.deHe is the first author of the paper "Controlling fluid-induced seismicity during a 6.1-km-deep geothermal stimulation in Finland" published at Science Advances.

Chang-Soo ChoKorea Institute for Geosciences and Mineral Resourcesnemex@kigam.re.krHe is one of the authors of "Surface Deformations and Rupture Processes Associatedwith the 2017 Mw 5.4 Pohang, Korea, Earthquake" published at BSSA and he did lotsworks on relocated seismicity.

Opposed Reviewers:

Additional Information:

Question Response

<b>Key Point #1: </b><br><i>Key Pointsare now mandatory for BSSA, and willappear at the front of articles starting in2020. Please submit three COMPLETEsentences addressing the following: 1)what problem did you address?; 2) whatconclusions did you come to?; and 3)what are the implications of your findings?Each point must be 110 characters or less(including spaces).

Three largest M > 4 earthquakes of the 2017 MW 5.5 Pohang sequence was located atjunctions of fault segments.

Key Point #2: Along-strike expansion of aftershock area was observed, implying afterslips orCoulomb stress transfer.

Key Point #3: Generally low b-values (<1) and variations in p-values along northeast direction weremapped.

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation

04 February 2020

Betty Schiefelbein

Manuscript Coordinator

Bulletin of the Seismlogical Society of America

Dear Editor:

I wish to submit an article for publication in the Bulletin of the Seismological Society of America titled

“Aftershock sequences and statistics of the 2017 MW 5.5 Pohang earthquake, South Korea: implication of

fault heterogeneity and post-seismic relaxation.” The paper was coauthored by Jeong-Ung Woo, Minook

Kim, Junkee Rhie, and Tae-Seob Kang.

This study analyzed over 4,000 earthquakes that were part of the foreshock-mainshock-aftershock

sequence associated with the main 2017 MW 5.5 Pohang earthquake. It was of interest to us that a high

proportion of non-double-couple events in the large magnitude earthquakes were recorded as part of the

Pohang event. Analysis of these data using the double-difference method confirmed the presence of a

complex network of four faults in the immediate vicinity of the Pohang enhanced geothermal site. Further

analysis of the Gutenberg-Richter b-values and the Omori p-values for the mainshock and aftershock

sequences indicated that they varied both temporally and/or spatially in the 2017 Mw 5.5 Pohang sequence

indicating a certain degree of subsurface heterogeneity than previously understood. We believe that our

study makes a significant contribution to the literature because, although our paper essentially presents a

case study located on the Korean Peninsula, the methods of analysis of b- and p-values that we describe

are of wider relevance to seismic hazard assessment and prediction in general.

Further, we believe that this paper will be of interest to the readership of your journal because it falls

within the remit of the BSSA to report on seismology and seismic hazard analyses. Specifically, we

believe that it fits well within the category of characterization of fault systems and seismotectonics

relevant to understanding seismicity and carrying out seismic hazard assessments, as described by Pratts.

Please consider, as potential referees, Kwang-Hee Kim (Pusan National University;

kwanghee@pusan.ac.kr), Francesco Grigoli (ETH; francesco.grigoli@sed.ethz.ch), Grzegorz Kwiatek

(GFZ; kwiatek@gfz-potsdam.de), and Chang-Soo Cho (Korea Institute for Geosciences and Mineral

Resources; nemex@kigam.re.kr).

This manuscript has not been published or presented elsewhere in part or in entirety and is not under

consideration by another journal. We have read and understood your journal’s policies, and we believe

that neither the manuscript nor the study violates any of these. There are no conflicts of interest to declare.

Thank you for your kind consideration of this manuscript. I look forward to hearing from you.

Sincerely,

Junkee Rhie

School of Earth and Environmental Sciences, Seoul National University

08826

+82-2-880-6731

+82-2-871-3269

rhie@snu.ac.kr

Letter to editor Click here to access/download;Letter to Editor;letter.doc

1

Aftershock sequence and statistics of the 2017 MW 5.5 Pohang earthquake, South Korea: 1

implication of fault heterogeneity and post-seismic relaxation 2

3

Jeong-Ung Woo 1, Minook Kim2a, Junkee Rhie 1*, Tae-Seob Kang 3 4

Corresponding author: Junkee Rhie (rhie@snu.ac.kr) 5

School of Earth and Environmental Sciences, Seoul National University, 1 Gwanak-ro, Gwanak-6

gu, Seoul 08826, Republic of Korea 7

Phone: +82-2-880-6731 8

Fax: +82-2-871-3269 9

1 School of Earth and Environmental Sciences, Seoul National University, Seoul 08826, Republic 10

of Korea 11

2 Department of Structural Systems and Site Evaluation, Korea Institute of Nuclear Safety, 12

Daejeon 34412, South Korea 13

3 Division of Earth Environmental System Science, Pukyong National University, Busan 48513, 14

Republic of Korea 15

16

Key points 17

18

1. Three largest M > 4 earthquakes of the 2017 MW 5.5 Pohang sequence was located at 19

junctions of fault segments. 20

a also at Division of Earth Environmental System Science, Pukyong National University, Busan

48513.

Manuscript Click here to access/download;Manuscript;manu.docx

2

2. Along-strike expansion of aftershock area was observed, implying afterslips or Coulomb 21

stress transfer. 22

3. Generally low b-values (< 1) and variations in p-values along northeast direction were 23

mapped. 24

25

Abstract 26

27

The sequence of foreshocks, mainshock, and aftershocks associated with a fault rupture are the 28

result of interactions of complex fault systems, the tectonic stress field, and fluid movement. 29

Analysis of shock sequences can aid our understanding of the spatial distribution and magnitude 30

of these factors, as well as providing a seismic hazard assessment. The 2017, MW 5.5 Pohang 31

earthquake sequence occurred following fluid-induced seismic activity at a nearby enhanced 32

geothermal system site and is an example of reactivation of a critically stressed fault system in 33

the Pohang Basin, South Korea. We created an earthquake catalog based on unsupervised data-34

mining and measuring the energy ratio between short- and long-window seismograms recorded 35

by a temporary seismic network. The spatial distribution of approximately 4,000 relocated 36

aftershocks revealed four fault segments striking southwestward. We also determined that the 37

three largest earthquakes (M > 4) were located at the boundary of two fault segments. We infer 38

that locally concentrated stress at the junctions of the faults caused such large earthquakes and 39

that their ruptures on multiple segments can explain the high proportion of non-double couple 40

components. The area affected by aftershocks expands to the southwest and northeast by 0.5 and 41

1 km decade-1, respectively, which may result from post-seismic deformation or sequentially 42

transferred static Coulomb stress. The b-values of the Gutenberg-Richter relationship 43

3

temporarily increased during the aftershock period, suggesting that the stress field was perturbed. 44

The b-values were generally low (< 1) and locally variable throughout the aftershock area, which 45

may be due to the complex fault structures and material properties. Furthermore, the mapped p-46

values of the Omori law vary along strike, which may indicate anisotropic expansion speeds in 47

the aftershock region. 48

49

4

Text 50

51

INTRODUCTION 52

On 15 November 2017, a moderate-sized earthquake of moment magnitude (MW) 5.5 or local 53

magnitude (ML) 5.4 struck the city of Pohang, located in the southeastern part of the Korean 54

Peninsula, which damaged infrastructure, injured 90 people, and made 1500 homeless (Kim et al., 55

2018a). The earthquake (hereafter referred to as the mainshock) was the second-largest 56

earthquake event among earthquakes recorded instrumentally in South Korea since 1978, 57

according to the catalog of the Korea Meteorological Administration (KMA). A close 58

examination of the seismic source characteristics of such a rarely observed moderate-sized 59

earthquake and its foreshock-mainshock-aftershock sequence is necessary not only to evaluate 60

the current stress field (Zoback, 1992; Soh et al., 2018) and fault properties, but also to 61

understand aftershock triggering mechanisms (King et al., 1994; Kilb et al., 2000). Estimation of 62

statistical parameters (i.e., the Gutenberg-Richter b-value and the Omori law p-value) from a 63

large number of microearthquakes in conjunction with the seismic source properties of 64

aftershocks can give information on fault heterogeneities, such as crack density, slip distribution, 65

applied shear stress, viscoelastic properties, and heat flow (Wiemer and Katsumata, 1999; Murru 66

et al., 2007). 67

One important point to note is that the mainshock occurred near an enhanced geothermal 68

system (EGS) site (Grigoli et al., 2018; Kim et al., 2018b; Lee et al., 2019). A body of evidence 69

supports the claim that the mainshock was triggered by five fluid-injection experiments as well 70

as an associated loss of heavy drilling muds and released tectonic energy on a critically stressed 71

fault (Ellsworth et al., 2019; Woo et al., 2019a). The periods of stimulation experiments 72

5

conducted on two hydraulic wells (PX-1 and PX-2) were closely correlated with microseismicity 73

observed near the wells. Induced seismicity mapped in the vicinity of the EGS indicated the 74

presence of a previously unmapped fault. Microseismicity triggered on this fault migrated to the 75

location of the mainshock. A breakout was observed in the PX-2 well at intervals corresponding 76

to the assumed fault. The groundwater levels of PX-1 and PX-2 decreased abruptly by 121 m and 77

793 m, respectively, immediately after the mainshock but gradually recovered by 0.078 m/day 78

and 0.198 m/day, respectively (Lee, 2019). 79

Previous studies of aftershock distributions in the Pohang Basin determined the presence of 80

complex fault geometries (Hong et al., 2018; Kim et al., 2019). Separately, Grigoli et al. (2018) 81

reported that obtaining a significant non-double-couple (non-DC) component when inverting the 82

moment tensor for a mainshock can be attributed to the complexity of the rupture process in a 83

multi-fault system. The spatial pattern of early aftershocks associated with two 2016 Gyeongju 84

earthquakes (ML 5.1 and ML 5.8), which occurred on two sub-parallel faults approximately 40 km 85

away from the Pohang mainshock, is differentiated from the presence of two or three fault 86

segments with varying strikes and dips for the early aftershocks associated with the 2017 Pohang 87

earthquakes (Uchide and Song, 2017; Son et al., 2017; Woo et al., 2019b). 88

In this study, we created an earthquake catalog for the foreshock-mainshock-aftershock 89

sequence from data recorded by local permanent seismic networks, temporary seismometers 90

deployed as part of the aftershock monitoring system, and the temporary Pohang EGS 91

monitoring system. Earthquakes were detected using a machine-learning data mining technique 92

for data obtained during the first ten days and a conventional automatic detection algorithm was 93

employed for the aftershock monitoring system as a whole. Each detected earthquake was 94

located by manual picking and visual inspection and then precisely relocated by the double-95

6

difference method (Waldhauser and Ellsworth, 2000). Using the spatial distribution of over 4000 96

earthquakes, we modeled fault systems as a series of multiple fault segments by mapping the 97

spatio-temporal distribution of the statistical parameters b and p. 98

Mapping the distribution of earthquake magnitudes provides an independent analysis of the 99

characteristics of aftershock activities and can be used to analyze spatial heterogeneities of 100

material properties, such as stress state, level of asperities, and heat flow rate (Scholz, 1968; 101

Wiemer and Katsumata, 1999; Wiemer and Wyss 2000; Ávila-Barrientos et al., 2008); assess 102

seismic hazards via epidemic-type aftershock sequence modeling (ETAS; Ogata, 1998); and 103

conduct probabilistic seismic hazard analysis (PSHA; Cornell, 1968). In this study, we evaluated 104

the relative magnitude of each earthquake by using amplitude ratios relative to earthquakes of 105

known ML. 106

107

DATA AND METHOD 108

Seismic Networks 109

Continuous seismic waveform data used to detect and analyze seismic source parameters were 110

collected from four different networks (Figure 1). The first data set was obtained from a 111

combined permanent seismic network operated by KMA, the Korea Institute of Geoscience and 112

Mineral Resources (KIGAM) and the Korea Hydro and Nuclear Power (KHNP). The permanent 113

seismic networks of KMA, KIGAM, and KHNP are named KS, KG, and KN, respectively. The 114

second set of continuous waveform data were recorded by nine borehole seismometers installed 115

at depths of between 100 and 150 m, which operated to monitor microseismic events for the 116

Pohang EGS project. Three of the temporary borehole seismometers recorded the mainshock, 117

while the operation of the other borehole seismometers started within the next 2 days; all of them 118

7

operated until the end of November 2017. The third continuous waveform data set was collected 119

by 37 temporary broad-band seismometers installed after the mainshock by the university 120

consortium (Pukyong National University and Seoul National University) and KIGAM 121

independently. The first seismometer installed temporarily for monitoring aftershocks started its 122

operation approximately 1 h after the onset of the mainshock. Lastly, we used waveforms of 214 123

early aftershocks, occurred within four hours from the mainshock, recorded at eight short-period 124

temporary seismometers deployed by Pusan National University (Kim et al., 2018b). The 125

seismograph stations of these temporary networks were densely spaced and located within the 126

radius of 20 km from the EGS site, respectively (Figure 1). 127

128

Detection and Hypocenter Determination 129

Since stabilizing temporary seismometers for aftershock monitoring can take many hours, 130

conventional algorithms for earthquake detection, such as STA/LTA (Withers et al., 1998; 131

Trnkoczy 2002), are of limited use for locating early aftershocks because of the incompleteness 132

of the local seismometer network. In this study, we utilized the Fingerprint and Similarity 133

Threshold (FAST) data-mining algorithm that uses waveform similarity to detect such early 134

aftershock sequences (Yoon et al., 2015; Yoon et al., 2017; Bergen et al., 2018) with a 135

conventional energy-based algorithm for the period for aftershock monitoring system. The FAST 136

algorithm finds pairs of waveforms having similar spectrograms without any prior information, 137

allowing us to obtain pairs of earthquake candidates with correlative signals. The performance of 138

the FAST algorithm to discriminate true events from earthquake candidates can be improved by 139

measuring similarity at multiple stations (Bergen et al., 2018). 140

8

We applied the FAST method to ten days of continuous seismograms recorded between 14 Nov 141

2017 and 23 Nov 2017 to cover the period of operation of the aftershock monitoring system. We 142

used three-component seismograms obtained from two short-period (PHA2 and DKJ) and one 143

broadband (CHS) seismometers, which are located within 30 km from the mainshock. The three 144

borehole seismometers of the EGS monitoring system that were operational at the onset of the 145

mainshock were not used in detection due to the high level of ambient background noise and 146

regularly observed pulse-like signals. The sampling rate of the seismograms was fixed at 100 Hz 147

and the frequency range of the bandpass filter was set to 2 – 20 Hz. All parameters employed in 148

the FAST algorithm routines were either determined manually from performance trials and or 149

were previously applied values (Yoon et al., 2017; Yoon et al., 2019a) and are summarized in 150

Table A1 and A2. 151

We detected 1580 candidate events via the FAST search, leading to a subset of 1357 locatable 152

earthquakes from visual inspection. This is comparable to the number of events in the earthquake 153

catalog published by Kim et al. (2018b), which utilized eight local seismographs deployed within 154

3 km of the EGS site for earthquake detection: the FAST algorithm successfully detected 155

169/174 or 97% of earthquakes for the same period. 156

While the aftershock monitoring network was operational (i.e., from 15 November 2017 to 28 157

February 2018), we applied an automatic algorithm to detect and locate microseismic 158

earthquakes (Sabbione and Velis, 2013). Continuous waveforms were transformed into 159

characteristic functions for measuring the ratio between the short-term average (STA) and the 160

long-term average (LTA). We declared candidate earthquakes when the STA/LTA ratio 161

exceeded 5 for a given time window of 4 s at more than three stations. For each triggered time 162

window, the normalized squared envelope functions of Baer and Kradolfer (1987) were 163

9

calculated to determine the time at which to maximize the function value (hereafter referred to as 164

the BK function). Since the BK function can be maximized for the arrivals of either the P-wave 165

or the S-wave, the maximum value of the BK function was tested to discriminate whether the 166

measured local maximum corresponded to the first arrival. If we observed a local high BK 167

function value before the maximum of the BK function in a given time window, we set two 168

consecutive time samples as the arrivals of the P- and S-waves. Otherwise, we searched for other 169

local maximum after the triggered time window and set the maxima as the P- and S-wave phase 170

arrivals when a secondary maximum was available. The phase arrivals determined in this way 171

were visually confirmed by using a Wadati plot (Wadati, 1933). 172

We determined the initial hypocenters of the detected earthquakes via Hypoellipse (Lahr, 1999), 173

with phase arrival times being determined by manual inspection and a 1-D layered seismic 174

velocity model for the Pohang EGS site (Woo et al., 2019a; Table 1). In this procedure, we 175

combined the earthquakes detected from either the FAST algorithm or the STA/LTA method 176

with events with ML > 2.0 listed in the KMA and Kim et al. (2018b) event catalogs. Earthquakes 177

with an onset difference of less than 2 s were regarded as duplicate events. Station corrections 178

were calculated based on a comparison of the theoretical arrival times for five immediate 179

foreshocks reported by Woo et al. (2019a) and their picked arrival times. 180

Initial hypocenters were relocated with hypoDD (Waldhauser and Ellsworth, 2000) by using 181

travel time differences obtained from waveform cross-correlation measurements as well as 182

picked phase times as inputs to the double-difference algorithm. The 1-D velocity model of Woo 183

et al. (2019a) was applied for the relocation procedure, again (Table 1). All relocated events were 184

shifted by 39 m, 28 m, and 96 m in eastwards, northwards, and downwards, respectively, to 185

match the centroid of the five immediate foreshocks with the results of Woo et al. (2019a), of 186

10

which recordings at 17 PX-2 borehole chains were applied to obtain accurate hypocenters. We 187

resampled waveforms to 1,000 Hz with a cubic spline after first having applied a 2–10 Hz 188

bandpass filter. Each seismogram was reduced to a 1 s time window centered at each phase 189

arrival time. We allow a time shift up to 0.1 s for the cross-correlation measurements. Time 190

shifts that maximized the cross-correlation coefficient (CC) between two pairs of waveforms 191

were used only if the maximum CCs were greater than 0.85. The squared maximum CCs were 192

used to weight the measurements. The relative locations were calculated by least-squares fitting 193

of the data and the location uncertainties were evaluated by using bootstrapping analysis 194

(Waldhauser and Ellsworth, 2000). Synthetic travel time differences between paired events were 195

reconstructed by random selection of a set of residuals and relative locations for these synthetic 196

travel times were calculated 200 times. 197

198

Magnitude Estimation and Statistical Analysis 199

Waveform similarity can be assessed to estimate the relative magnitudes of earthquakes (Shelly 200

et al., 2016; Yoon et al., 2019b). We adopted a simple magnitude-amplitude relationship 201

modified from the equation of Shelly et al. (2016) that considers the differences in hypocentral 202

distance between two earthquakes: 203

dm = clog10(a/r), (1) 204

where dm, a, and r represent the ratios of magnitude, amplitude, and hypocenteral distance, and c 205

is a coefficient for the magnitude-amplitude relationship (Shelly et al. 2016). The coefficient c in 206

Equation 1 varies with the earthquake magnitude scale that is used: for example, Shelly et al. 207

(2016) reported that c = 1 for ML and c = 2/3 for MW. In this study, we used a set of MLs of 208

aftershocks and Equation (1) to estimate the coefficient c, following the method of Woo et al. 209

11

(2019b). If the CC of a waveform pair was greater than 0.85, we calculated the amplitude ratio as 210

the slope of the eigenvector for the largest eigenvalue of the covariance matrix of the two 211

waveforms (Shelly et al., 2016). Thus, for earthquakes with known values of ML, we were able to 212

estimate the parameter c. 213

We can also determine relative magnitudes of earthquakes by using our estimated value of c in 214

Equation (1). Estimated relative magnitudes (MRel) were arithmetically averaged to produce a 215

representative value and uncertainties were obtained from their standard deviations. 216

The Gutenberg-Richter law (G-R law) describes the relationship between earthquake frequency 217

and magnitude. Its statistical properties are widely accepted and applied to the investigation of 218

seismo-tectonic properties in a specific region over a certain time period. Examples of 219

application of the G-R law include work on aftershock sequences by Wiemer and Katsumata 220

(1999) and Woo et al. (2019b), on earthquake swarms by Farrell et al. (2009), on induced 221

seismicity by Shapiro (2007), and in laboratory experiments by Scholz (1968). The earthquake 222

frequency distribution with magnitude can be written as: 223

log10 N(≥ M) = a - bM, (2) 224

where N is the number of earthquakes equal to or greater than a magnitude M, and a and b are 225

scaling constants. a is proportional to the overall seismicity in a given spatio-temporal interval, 226

whereas b represents the ratio of the number of large earthquakes to small earthquakes. The 227

behavior of b-values has been attributed to crack density (Mogi, 1962), stress drop (Wyss, 1973), 228

and tectonic stresses (Schorlemmer et al., 2005, Scholz, 1968), and slip distribution (Wiemer and 229

Katsumata, 1999). 230

We determined the magnitude of completeness (MC) for 3,521 magnitudes based on a modified 231

goodness-of-fit method of Wiemer and Wyss (2000), following Woo et al. (2019b). Then, we 232

12

evaluated the b-value for a set of magnitudes using the maximum likelihood method of Aki 233

(1965) with a magnitude bin of 0.1. The uncertainty of b-values was estimated with the method 234

of Shi and Bolt (1982). 235

Omori’s law describes the decay rate of aftershocks. Its parameters are also broadly applied to 236

interpret regional seismic and tectonic properties (Omori, 1894; Utsu, 1961). The extended form 237

of Omori’s law can be written as: 238

R(t) = K(t+c)-p, (3) 239

where K, c, and p are the scaling coefficients that describe the aftershock decay rates in a given 240

region. p, which represents the power of the aftershock decay rates, has a range of 0.6 to 1.8 and 241

is considered to be a function of stress and temperature in the crust (Utsu and Ogata, 1995; 242

Wiemer and Katsumura 1999). We mapped the spatial variation of p-values by binning 250 243

magnitudes and by selecting magnitudes greater than MC. The three parameters and their 244

associated uncertainties were determined following the maximum likelihood method presented 245

by Ogata (1983). 246

247

RESULTS 248

Of the 4,446 earthquakes with initial locations, we relocated seven foreshocks, the mainshock, 249

and 3,938 aftershocks using hypoDD (Waldhauser and Ellsworth, 2000), having excluded 250

earthquakes with fewer than seven traveltime difference measurements. Uncertainties of relative 251

locations to within two standard deviations were estimated as 25 m in the eastwest direction, 18 252

m in the northsouth direction, and 37m vertically. Figure 2 presents the spatial distribution of 253

aftershocks, both in plan view and cross-sections, four in the dip direction (A1-A2, B1-B2, C1-C2, 254

and D1-D2) and one in the strike direction (E1-E2). From the map, we determined the apparent 255

13

strike of aftershocks (crossline of E1-E2) to be 210°, which corresponds to the azimuth of the first 256

principal vector obtained from two-dimensional principal component analysis (PCA) (Jollifle, 257

2002). From cross-sections in the dip direction (A1-A2 to D1-D2), we observed that the spatial 258

distribution of aftershocks delineates at least four different fault segments (Figure 2). In the most 259

northeastern part of the study area, a sub-vertical fault was identified from the aftershock 260

distribution. An ML 3.5 earthquake on 21:05:15, 19 November 2017; UTC with a focal 261

mechanism (strike: 234°, dip: 85°, rake: -174°) is consistent with the inferred fault. Among the 262

relocated earthquakes, the first observed event on the fault plane occurred within 72 s of the 263

onset of the mainshock (Figure A1), which indicates that reactivation of the fault segment was 264

initiated by the mainshock rupture or soon afterward. Two slightly different fault geometries, 265

both dipping northwestward, are distinguished in the middle of sections B1-B2 and C1-C2 from 266

the spatial distribution of the aftershocks. The aftershock distribution along B1-B2 has a wider 267

range of focal depths, a shallower dip, and a strike closer to north-south than that of C1-C2. Both 268

the mainshock and the ML 4.3 aftershock are located adjacent to a virtual boundary of B1-B2 and 269

C1-C2 and their focal mechanisms are consistent with the observed fault geometry. Earthquakes 270

in the southwestern part of D1-D2 occurred after the largest aftershock (ML 4.6) (Figure A1) and 271

their focal depths deepened to the south-east, dipping in the opposite direction to the three other 272

fault segments observed on A1-A2, B1-B2 and C1-C2. Such a conjugate fault geometry is matched 273

with one nodal plane of the focal mechanism (strike: 34°, dip: 52°, rake: 136°) of the largest ML 274

4.6 aftershock. 275

From the complex fault geometry delineated by the four cross-sections, we constructed a 276

simplified fault model to describe the observed aftershock distribution. For the three segments 277

that reactivated with the occurrence of the mainshock, we described their geometry using the 278

14

aftershocks that occurred within one day of the mainshock. Because the mainshock was situated 279

on a virtual boundary between two faults (F2 and F3) with slightly different strikes and dips, we 280

divided the aftershock area based on the hypocenter of the mainshock and an apparent strike of 281

210°, which we estimated from PCA of data in map view. The aftershocks on the most 282

northeasterly fault segment (F1) were de-clustered from the adjoining fault (F2) using the simple 283

assumption that the Heunghae Fault (i.e., Song et al., 2015; Yun et al., 1991) vertically intersects 284

them both. Earthquakes that occurred up to 1 day after the largest ML 4.6 aftershock were used to 285

investigate the most southwesterly fault segment (F4). Faults F1F4 were used to divide the 286

study area into four regions and earthquakes were assigned to a region on the basis of the 287

location of their hypocenter. We applied PCA analysis with bootstrapping to earthquakes that 288

were resampled 200 times to estimate strike, dip, fault length, and fault width. The fault length 289

and width were determined as the difference between the 2.5th and 97.5th percentile of the strike 290

and dip components. The resulting fault geometry is summarized in Table 2. 291

We determined c using the 266 relocated earthquakes with known ML. We evaluated c as 0.85 292

by PCA (Figure 3), which is larger than the case for the MW magnitude scale (c = 2/3) scale but 293

smaller than the case for the ML magnitude scale (c = 1). The difference in c implies that the ML 294

magnitude does not naturally match MW for earthquakes within the range of magnitudes included 295

in this study, filtered to a frequency range of 2 – 10 Hz. The observed value of c is relatively 296

high compared with 0.7 that was estimation using the MLs of the Gyeongju aftershock sequences 297

(Woo et al., 2019), which may be the result of systematic differences between ML and KMA 298

magnitude. 299

We estimated the magnitudes of 3,521 earthquakes with measurements ≥ 5. Figures 3b and 3c 300

illustrate the comparison of MRel with ML and the variations of MRel with time. Since MRel is 301

15

exactly proportional to ML without any scaling parameters, we propose that MRel can replace ML 302

as the magnitude scale for subsequent analysis. 303

We examined temporal variations of seismic b-values by binning 600 earthquake magnitudes 304

(MRel) into a set (Figure 4a). There was an overlap of four hundred earthquakes between two 305

consecutive bins. The MC decreased from 0.8 to 0.2 during the first 3 days of the early aftershock 306

sequence, which is indicative of a decrease in the background noise level for that period. The b-307

value for the first bin was evaluated as 0.66, which is consistent with b-values for earthquakes 308

detected during fluid injection into the Pohang EGS site before the occurrence of the mainshock 309

(Woo et al., 2019). The b-value increased with time for the first three days up to a maximum of 310

0.98 and fluctuated during a month. After a month, it decreased to 0.77 until the largest 311

aftershock of ML 4.6 occurred. We tested the temporal changes of b-values with a fixed MC of 312

0.8, corresponding to the maximum values over the whole period, to investigate whether the 313

observed temporal variations of b-values were biased by the choice of MC (gray dots of Figure 4a) 314

and confirmed that the main features were not significantly changed. Figure 4b illustrates the 315

magnitude-frequency distributions of three data sets highlighted in Figure 4a. 316

The spatial variation of b-values was investigated for the vertical cross-section along the 317

apparent strike of 210°. Earthquakes within 1.5 km of each 0.5 0.5 km grid cell on the cross-318

section were binned into that cell. We analyzed the b-value only if each bin contained at least 319

250 earthquakes. Figure 4c illustrates the spatial distribution of b-values on the vertical cross-320

section. The estimated b-values are between 0.63 and 0.86, all of which are lower than the 321

typically assumed b-value of 1 (Wyss, 1973). Since ML is approximated by MRel, such low b-322

values can be interpreted as an increase in applied shear stress and effective stress (Scholz, 1968; 323

Wyss 1973), low material heterogeneity (Mogi, 1962), or a high stress drop (Wyss, 1973). 324

16

Considering that the slip tendency of the mainshock is indicative of a critically stressed fault 325

(Chang et al., 2020) and the stress drop of 5.6 MPa for the mainshock is not higher than that of 326

other earthquakes in South Korea (Rhee and Sheen 2016; Woo et al., 2019a), our preferred 327

interpretation is that the generally low b-values in the aftershock area may result from high 328

applied stress in this region. We estimated a b-value of 0.69 near the hypocenter of the 329

mainshock, which is comparable to the values observed for the earthquakes during the fluid 330

injection (= 0.66). 331

The significance of temporal and spatial differences in b-values can be verified by Utsu’s test 332

(Utsu, 1992), in which the probability that the b-values between two sets of earthquakes are the 333

same is defined via Akaike Information Criterion (Akaike, 1974). We first tested the statistical 334

significance of the temporal differences of b-values among early (< 1 day), intermediate (~ 3 335

days), and late aftershocks (~ 80 days), which are highlighted in green, red, and blue, 336

respectively, in Figures 4a and 4b. The probability that the b-value for the intermediate period is 337

not significantly higher than those of the early and late aftershocks was estimated as 2.6×10-7 and 338

1.0×10-3, respectively, indicating that the temporal increase and decrease of b-values are 339

statistically reasonable with a significance level of 5%. Similar variations of b-values with time 340

can be found for the 2016 Gyeongju earthquake (Woo et al., 2019b) and other cases (Smith, 341

1981; Chan et al., 2012; Gulia et al., 2018), which can be interpreted as local stress changes due 342

to the mainshock rupture or a mixed effect of a changing spatial distribution of b-value and a 343

heterogeneous population of aftershocks with time (Figure 4c). 344

We also applied Utsu’s test for all pairs of spatially varying b-values for which the difference is 345

statistically significant with a significance level of 5% if Δb > 0.1 for half the cases and Δb > 346

0.135 for all cases (Figure 4d). Therefore, we roughly divided the aftershock area into three sub-347

17

regions: R1 with relatively low b-values; R2 with high b-values and Δb > 0.1; and R3 with high 348

b-values and Δb > 0.135 (Figure 4c). The ML 4.3 and ML 4.6 earthquakes are located near R2 and 349

R3 and have high b-values relative to the values of the hypocenter area (R1), which can be 350

interpreted as indicating material heterogeneity with respect to the conjugate fault system 351

(Figures 2c and 2e). Alternatively, spatial variations of pore pressure or applied stress may 352

contribute to b-value heterogeneity. 353

The p-values that describe the power law decay rate of aftershocks were estimated for two data 354

sets: (1) period A, between the onset of the mainshock and the ML 4.6 aftershock; and (2) period 355

B, after the onset of ML 4.6 aftershock. This grouping was chosen because the occurrence of the 356

largest aftershock resulted in increased seismicity, which resets the decay rate for the mainshock 357

(Figure 3c). For each data set, we estimated the p-value that represents the whole data set and the 358

spatial variation of p-values at the cross-sections along the apparent strike of 210°, with the same 359

bins used for estimating the spatial variations of b-values. The p-value of period A was estimated 360

as 1.10, which is larger than the value for period B (= 0.78). Such a difference may result from 361

differing initial stress levels for periods A and B with respect to the stress perturbation of the 362

mainshock sequence, spatial heterogeneity of the internal structure for the conjugate fault system 363

(Figure 2e; Wiemer and Katsumata, 1999), or just an insufficient number of earthquakes in the 364

calculation of p-values for period B. With the exception of p-values for period B, the p-values of 365

the period A were higher in the southwestern region than those in the northeastern region. This 366

could be indicative of a spatial variation of heat flow (Kisslinger and Jones, 1991), heterogeneity 367

of fault strength (Mikumo and Miyatake, 1979) or an insufficient number of aftershocks to allow 368

accurate fitting of the aftershock power decay law for the southwestern aftershock region prior to 369

the occurrence of the ML 4.6 aftershock. 370

18

371

DISCUSSION 372

Expansion of aftershock areas with time 373

Expansion of early aftershock sequences is widely observed (Tajima and Kanamori 1985; 374

Fukumaya and Ellsworth, 2000; Peng and Zhao, 2009; Kato et al., 2014). Some temporal 375

evolution of aftershock areas have been interpreted to be the result of afterslip or post-seismic 376

deformation (Helmstetter and Shaw 2009; Peng and Zhao 2009; Perfettini et al., 2017; Ross et al., 377

2017). Speeds of along-strike expansion of the aftershock zone were measured on a logarithmic 378

time scale and showed that propagating aftershock can cause the expansion of aftershocks (Peng 379

and Zhao 2009; Frank et al., 2017; Perfettini et al., 2017; Ross et al., 2017). In the present study, 380

we examined the spatio-temporal distribution of aftershocks on a logarithmic time scale to 381

consider possible post-seismic deformation following the mainshock (Figure 6). In a map view, 382

we observed that the aftershock zone has roughly expanded along the apparent strike direction, 383

especially during the first day (Figures 6a and c), whereas no clear trends were observed in a 384

vertical sense (Figure 6b). 385

The speed of virtual aftershock migration fronts for the bilateral expansion along the strike 386

direction were ~1 km decade-1 northeastward and ~0.5 km decade-1 southwestward (Figure 6c), 387

which may indicate post-seismic deformations related to aseismic afterslip (Peng and Zhao 2009; 388

Perfettini et al., 2018). The difference in the migration speeds can be attributed to different rate-389

and-state parameters described by Dieterich (1994) following the equations published by 390

Perfettini et al. (2018). However, in our case, we also observe a significant p-value variation in 391

the northeastern and southwestern parts of the study area (Figure 5a). Such variations of p-value 392

require a different model rather than the rate-dependent friction law (Helmstetter and Shaw 2003; 393

19

Mignan 2015). Assuming that the power-law rheology governs post-seismic velocity which is 394

proportional to (1+t/t*)-p (Montési, 2004), where t* is a characteristic time of the aftershock, the 395

slip velocity or the aftershock occurrence rate decays with time as a power of p. For regions with 396

low p-value, the slip velocity decreases relatively slowly and the accumulated post-seismic 397

displacement required to rupture asperities can takes short time compared to that of the regions 398

with high p-value. Therefore, the p-value variation observed for the aftershock area during the 399

period A may be related to the different seismic migration speeds (Figures 5a and 6c). We did 400

not further compare p values and the migration speed in this study, since it may require more 401

complex analysis than a simplified form of the Omori’s law (Narteau et al., 2002). Furthermore, 402

there is an absence of data for very early (« 1 day) or late (> 100 days) aftershock rates. 403

The expansion of the aftershock zone can also be explained by a cascade of sequentially 404

triggered aftershocks in terms of changes to the static Coulomb stress (Ellsworth and Bulut, 405

2018). Since no clear evidence of post-seismic deformation was observed from differential 406

InSAR analysis (Song and Lee, 2019), the observed expansion of aftershocks during a single day 407

could possibly be attributed to changes to the static stress field caused by the aftershock 408

sequences rather than a result of aseismic deformation. However, the descending image of 409

differential InSAR reveals surface deformation during the first day after the mainshock, whereas 410

the ascending image reveals deformation during the next 19 days. This implies that the co-411

seismic deformation associated with afterslips related to the expansion of aftershock zone that 412

occurred within 20 days of the mainshock might be captured by the InSAR image. Therefore, the 413

possibility of afterslip-driven aftershocks cannot be discounted, even without the observation of 414

post seismic deformation. 415

20

High percentage of non-DC components observed for the mainshock and two largest 416

aftershocks 417

The moment tensor solutions of the mainshock and two largest earthquakes have high 418

percentages (> 30%) of non-DC components (Grigoli et al., 2018; Hong et al., 2018), in contrast 419

to the normally observed moment tensor solutions in South Korea. Such high non-DC 420

components of the moment tensor solutions of the three largest earthquakes can result from 421

complex shear faulting of multiple DCs, tensile opening/closing, and shear faulting in anisotropic 422

and heterogeneous media (Miller et al., 1998). It has already been established that the spatial 423

distribution of the Pohang earthquake sequence indicates that multiple fault segments were 424

reactivated in a complex fault system and the faulting types of the focal mechanism vary 425

throughout the aftershock area (i.e., Kim et al., 2019; Chang et al., 2020). Hence, a combination 426

of multiple DC moment tensor solutions with varying senses of slip motion could be one of the 427

causes of the three largest earthquakes having high non-DC components. 428

We propose the following sequence of events to explain the mainshock and major aftershock 429

sequence associated with the MW 5.5 Pohang earthquake. We infer that the nucleation of the 430

mainshock rupture was initiated at the junction between F2 and F3 and that the rupture 431

propagated along F2 and F3 with possible intervention of F1. Later, the ML 4.3 earthquake was 432

initiated between two adjacent conjugate faults dipping southwestward and northeastward in the 433

deeper aftershock region below the mainshock (Figure 2c). Finally, the ML 4.6 earthquake 434

nucleated at the southwestern tip of the aftershock area and subsequent aftershocks occurred on a 435

previously unrecorded southeastward dipping fault, suggesting that the rupture of the ML 4.6 436

earthquake sequences was initiated at the intersection of conjugate faults F3 and F4. 437

21

Although the three earthquakes were located at the intersection of multiple fault planes, it is hard 438

to envisage that all the earthquakes located in the surrounding area ruptured on multiple fault 439

planes. Some MRel 3 3.6 earthquakes without non-DC components were located in the vicinity 440

of the interconnecting faults (Choi et al., 2018), which may suggest that a certain amount of 441

seismic energy is required for the simultaneous movement of multiple fault segments. The fault 442

dimensions for the three largest earthquakes are inferred to be greater than 1 km, based on the 443

assumption of a constant stress drop of 5.6 MPa on a circular crack (i.e., Figure 2f), leading us to 444

propose that a kilometer rupture scale is the threshold to rupture multiple fault planes. Low b 445

values observed throughout the aftershock area can be considered as stress concentrations within 446

areas of high asperities (Wimmer and Wyss, 2000). High asperities in the regions adjoining two 447

or more fault segments may concentrate tectonic energy either as an earthquake nucleation point 448

or as barriers to rupture propagation. This may explain why only ML > 4 non-DC component 449

earthquakes were observed. The sonic log data of the PX-2 borehole recorded the existence of 450

anisotropic structures in the Pohang Basin (Ellsworth et al., 2019). Such anisotropic materials 451

can also cause earthquakes with high non-DC components. However, it is our preferred 452

interpretation that non-DC components in the three largest earthquakes result from the fault 453

complexity because low, non-DC earthquakes for ML 3 – 3.6 earthquakes were also observed. 454

455

Comparison between aftershock activities and induced seismicity at the EGS site during 456

stimulation. 457

The seismicity recorded during the five hydraulic stimulation experiments at the Pohang EGS 458

site and the inferred focal mechanisms revealed a fault plane located near the PX-2 well (Woo et 459

al., 2019a). PX-2 seismicity was clustered on a plane with a strike of 214° and a dip of 43° and 460

22

migrated southwestward, heading toward the location of the mainshock (Woo et al., 2019a). 461

However, the fault geometry for the induced earthquakes related to the PX-2 well has a 20° 462

shallower dip angle than the moment tensor solution of the mainshock and aftershocks. It 463

suggests that complex fault segments exist locally throughout the aftershock region and that a 464

simple fault plane does not explain the detailed fault structures. The ML 4.3 earthquakes have 465

deeper focal depths and their focal mechanism has steeper dips than that of the mainshock, which 466

can also be regarded as a result of complex fault geometry. Observation of various types of focal 467

mechanisms in aftershock sequences (Kim et al., 2019; Chang et al., 2020) are also a 468

manifestation of the complex geometry, which is in contrast to the nearly identical focal 469

mechanisms for the PX-2 seismicity (Woo et al., 2019a). 470

The b-values observed during the Pohang EGS project have insignificant variations, with an 471

average value of 0.66 (Woo et al., 2019a, Langenbruch et al., 2020); whereas, the b-values 472

estimated for the early aftershock sequences are statistically different from the b-values for a bin 473

of approximately 3 days after the mainshock (Figure 4). If we assume that b-values act as a 474

stress-meter (Scholz, 2015; Rigo et al., 2018; Woo et al., 2019b) and temporal variation of b-475

values during the aftershock period represents the level of stress state, the invariant b-values 476

observed during the stimulation period suggest that stress perturbations caused by fluid injection 477

may be far lower than the accumulated tectonic stress, indicating the existence of a critically 478

stressed fault system before the mainshock. 479

480

Reactivation of a multi-segment fault system and spatial variations of b-values and p-values 481

The complexity of the Pohang aftershock distributions was modeled as four fault segments, 482

following the approach of Hong et al. (2018) and Kim et al. (2018b, 2019) (Figure 7). The 483

23

seismicity along a subvertical fault, F1, in the northeastern of the study area clearly represents 484

migration of the aftershock front northeastward during the first day of the aftershock sequences 485

(Figure 6). Although this fault plane is located about 3 km away from the mainshock hypocenter, 486

it may have been reactivated as a part of the mainshock rupture process. Alternatively, it may 487

have been dynamically triggered by the mainshock considering circumstantial evidence that 488

aftershock activity on the fault segments was initiated within just 2 min (Figure A1) and the slip 489

distribution of the mainshock calculated from the static deformations with InSAR data is largest 490

in the northeastern part of the fault model (Song and Lee, 2019). Aftershocks on F1 are bounded 491

by the Heunghae Fault, which has surface expression (Figure 1), detaching F1 from F2 and F3. 492

Therefore, in either case, the reactivation of F1 may require a certain stress threshold to be 493

ruptured preferentially to F2 and F3. 494

Two slightly different geometries of F2 and F3 are suggested by Hong et al. (2018), reflecting a 495

complex fault system near the Pohang EGS site. While the b values vary slightly on F2, the 496

observed p values were higher for F3, at least until the occurrence of the ML 4.6 event. The 497

different behaviors of the two statistical parameters imply that the two fault segments exist under 498

different physical conditions, such as: differential stress states (Scholz, 1968), local 499

heterogeneity of the rock matrix that may interact with viscous materials (Wyss, 1973; Bayrak et 500

al., 2013), or variable spatial distribution of heat flow (Kisslinger and Jones, 1991). 501

The b values decreased to ~0.7 when fault segment F4 was reactivated by the ML 4.6 aftershock. 502

The lower b-values may indicate F4 was already highly stressed when the ML 4.6 earthquake was 503

triggered. The observed p-values for period B were generally much lower than those for period A, 504

which may be the result of using short time periods for analysis during period B or just uneven 505

seismicity observed for periods A and B. 506

24

507

CONCLUSIONS 508

In this study, we detected over 4,000 earthquakes related to the MW 5.5 (ML 5.4) Pohang 509

earthquake by using both unsupervised data-mining and a conventional automatic earthquake 510

detection method. From the spatio-temporal distribution of relocated seismicity, we observed 511

that four fault segments were responsible for the aftershocks. All the faults strike 512

northeastsouthwest, but have different dip angles and dip directions. The three largest 513

earthquakes are located at the boundaries of two adjoining fault segments, which may have 514

focused the stress released by multiple faults, resulting in high, non-DC earthquake mechanisms. 515

By measuring amplitude ratios between two similar earthquakes, we estimated relative 516

magnitudes of earthquakes to infer the statistical parameters related to earthquake frequency and 517

magnitude. The observed spatio-temporal distribution of b-value indicates that they were 518

spatially variable, but generally as low as ~0.7, and temporarily increased with time. The 519

observed p values were different for the northeastern and southwestern parts of the study area, 520

implying that heterogeneities in material properties such as frictional heat can lead to two 521

different speeds of aftershock expansion rate with logarithmic time. The complexity of faulting 522

in the aftershock zone will influence the duration and magnitude of seismic activity that is 523

caused by the locally perturbed stress field that is a result of the mainshock. We hope that our 524

findings can be applied to an interpretation of aftershock mechanisms under the general complex 525

fault systems and can be utilized to perform a seismic hazard assessment lowering the epistemic 526

uncertainty about the characteristics of the fault sources and their contemporary seismic activity. 527

528

Data and Resources 529

25

530

The unpublished manuscript by Chang et al. (2020), “Stress state and fault slip susceptibility 531

prior to the 2017 MW 5.5 Pohang earthquake triggered by enhanced geothermal system 532

stimulation” has been submitted to Geochemistry, Geophysics, Geosystems. The earthquake 533

catalog used in this study will be released at zenodo website (with doi number) when it is 534

published in journal. 535

536

Acknowledgments 537

538

We thank the Korea Institute of Geoscience and Mineral resources (KIGAM), the Korea 539

Meteorological Administration (KMA), the Korea Hydro & Nuclear Power (KHNP), and the 540

K.‐ H. Kim for providing seismic data used in this study. We appreciate C. E. Yoon and G. C. 541

Beroza for comments on FAST usage, W. L. Ellsworth for advice on visualizing the seismicity, J. 542

Song for discussion on non-DC earthquakes. This work was conducted during the Korean 543

Government Commission (KGC) on the relations between the 2017 Pohang earthquake and EGS 544

project, funded by the Korea Institute of Energy Technology Evaluation and Planning (KETEP) 545

grant from the South Korean government (MOTIE) (no. 2018‐ 3010111860). This work was 546

supported by the Nuclear Safety Research Program through the Korea Foundation of Nuclear 547

Safety (KoFONS) using the financial resource granted by the Nuclear Safety and Security 548

Commision (NSSC) of the Republic of Korea (No. 1705010). 549

550

References 551

552

26

Akaike, H. (1974). A new look at the statistical model identification, IEEE Trans. 553

Automat. Control 19, 716–723. 554

Aki K. (1965). Maximum likelihood estimate of b in the formula logN=a−bM and its 555

confidence limits, Bull. Earthq. Res. Inst. Univ. Tokyo 43, 237–239. 556

Ávila-Barrientos, L., Zúñiga, F. R., Rodríguez-Pérez, Q., and Guzmán-Speziale, M. 557

(2015). Variation of b and p values from aftershocks sequences along the Mexican subduction 558

zone and their relation to plate characteristics. J. S. Am. Earth Sci. 63, 162–171. 559

Baer, M., and Kradolfer, U. (1987). An automatic phase picker for local and teleseismic 560

events. Bull. Seismol. Soc. Am. 77, 1437–1445. 561

Bayrak, Y., Yadav, R. B. S., Kalafat, D., Tsapanos, T. M., Çınar, H., Singh, A. P., Bayrak 562

E., Yılmaz, Ş., Öcal, F, and Koravos, G. (2013). Seismogenesis and earthquake triggering during 563

the Van (Turkey) 2011 seismic sequence. Tectonophysics 601, 163–176. 564

Bergen, K. J., and Beroza, G. C. (2018). Detecting earthquakes over a seismic network 565

using single-station similarity measures. Geophys. J. Int. 213, 1984–1998. 566

Chan, C. H., Wu, Y. M., Tseng, T. L., Lin, T. L., and Chen, C. C. (2012). Spatial and 567

temporal evolution of b-values before large earthquakes in Taiwan. Tectonophysics 532, 215–222. 568

Choi, J. H., Ko, K., Gihm, Y. S., Cho, C. S., Lee, H., Song, S. G., Bang, E.-S., Lee, H.-J., 569

Bae, H.-K., Kim S. W., Choi S.-J., Lee, S. S., and Lee, S. R. (2019). Surface deformations and 570

rupture processes associated with the 2017 M w 5.4 Pohang, Korea, earthquake. Bull. Seismol. 571

Soc. Am. 109, 756–769. 572

Cornell, C. A. (1968). Engineering seismic risk analysis. Bull. Seismol. Soc. Am. 58, 573

1583–1606. 574

27

Dieterich, J. (1994). A constitutive law for rate of earthquake production and its 575

application to earthquake clustering. J. Geophys. Res.: Solid Earth 99, 2601–2618. 576

Ellsworth, W. L., and Bulut, F. (2018). Nucleation of the 1999 Izmit earthquake by a 577

triggered cascade of foreshocks. Nat. Geosci. 11, 531–535. 578

Ellsworth, W. L., Giardini, D., Townend, J., Ge, S., and Shimamoto, T. (2019). 579

Triggering of the Pohang, Korea, Earthquake (MW5.5) by Enhanced Geothermal System 580

Stimulation. Seismol. Res. Lett. 90, 1844–1858. 581

Farrell, J., Husen, S., and Smith, R. B. (2009). Earthquake swarm and b-value 582

characterization of the Yellowstone volcano-tectonic system. J. Volcanol. Geotherm. Res. 188, 583

260–276. 584

Frank, W. B., Poli, P., and Perfettini, H. (2017). Mapping the rheology of the Central 585

Chile subduction zone with aftershocks. Geophys. Res. Lett. 44, 5374–5382. 586

Fukuyama, E., Ellsworth, W. L., Waldhauser, F., and Kubo, A. (2003). Detailed fault 587

structure of the 2000 western Tottori, Japan, earthquake sequence. Bull. Seismol. Soc. Am. 93, 588

1468–1478. 589

Grigoli, F., Cesca, S., Rinaldi, A. P., Manconi, A., Lopez-Comino, J. A., Clinton, J. F., 590

Westaway, R., Cauzzi, C., Dahm, T., Wiemer, S. (2018). The November 2017 Mw 5.5 Pohang 591

earthquake: A possible case of induced seismicity in South Korea. Science 360, 1003–1006. 592

Gulia, L., Rinaldi, A. P., Tormann, T., Vannucci, G., Enescu, B., and Wiemer, S. (2018). 593

The Effect of a Mainshock on the Size Distribution of the Aftershocks. Geophys. Res. Lett. 45, 594

13–277. 595

Helmstetter, A., and Shaw, B. E. (2009). Afterslip and aftershocks in the rate‐ and‐ state 596

friction law. J. Geophys. Res.: Solid Earth 114, B01308 597

28

Hong, T. K., Lee, J., Park, S., and Kim, W. (2018). Time-advanced occurrence of 598

moderate-size earthquakes in a stable intraplate region after a megathrust earthquake and their 599

seismic properties. Sci. Rep. 8, 1–8. 600

Jolliffe I. T. (2002). Principal Component Analysis, Second Ed., Springer, New York. 601

Kato, A., and Obara, K. (2014). Step‐ like migration of early aftershocks following the 602

2007 Mw 6.7 Noto‐ Hanto earthquake, Japan. Geophys. Res. Lett. 41, 3864–3869. 603

Kilb, D., Gomberg, J., and Bodin, P. (2000). Triggering of earthquake aftershocks by 604

dynamic stresses. Nature 408, 570–574. 605

Kim, H. S., Sun, C. G., and Cho, H. I. (2018a). Geospatial assessment of the post-606

earthquake hazard of the 2017 Pohang earthquake considering seismic site effects. ISPRS Int. J. 607

Geo-Inf. 7, 375. 608

Kim, K. H., Ree, J. H., Kim, Y., Kim, S., Kang, S. Y., and Seo, W. (2018b). Assessing 609

whether the 2017 MW 5.4 Pohang earthquake in South Korea was an induced event. Science 360, 610

1007–1009. 611

Kim, K. H., Seo, W., Han, J., Kwon, J., Kang, S. Y., Ree, J. H., Kim, S., and Liu, K. 612

(2019). The 2017 ML 5.4 Pohang earthquake sequence, Korea, recorded by a dense seismic 613

network. Tectonophysics 774, 228306. 614

King, G. C., Stein, R. S., and Lin, J. (1994). Static stress changes and the triggering of 615

earthquakes. Bull. Seismol. Soc. Am. 84, 935–953. 616

Kisslinger, C., and Jones, L. M. (1991). Properties of aftershock sequences in southern 617

California. J. Geophys. Res.: Solid Earth 96, 11947–11958. 618

29

Lahr J. C.1999. HYPOELLIPSE: A computer program for determining local earthquake 619

hypocentral parameters, magnitude and first‐ motion pattern (Y2K compliant version), version 620

1.0, U.S. Geol. Surv. Open‐ File Rept. 99‐ 23, On‐ Line Edition. 621

Langenbruch, C., Ellsworth, W. L., Woo, J. U., and Wald, D. J. (2019). Value at Induced 622

Risk: Injection‐ induced seismic risk from low‐ probability, high‐ impact events. Geophys. Res. 623

Lett. 47, c2019GL085878. 624

Lee, K.-K. (2019). Final Report of the Korean Government Commission on Relations 625

between the 2017 Pohang Earthquake and EGS Project. Geological Society of Korea. 626

https://doi.org/10.22719/KETEP-2019043001. 627

Lee, K. K., Ellsworth, W. L., Giardini, D., Townend, J., Ge, S., Shimamoto, T., Yeo, I.-628

W., Kang, T.-S., Rhie, J., Sheen, D.-H., Chang, C., Woo, J.-U., Langenbruch, C. (2019). 629

Managing injection-induced seismic risks. Science 364, 730–732. 630

Mignan, A. (2015). Modeling aftershocks as a stretched exponential relaxation. Geophys. 631

Res. Lett. 42, 9726–9732. 632

Mikumo, T., and Miyatake, T. (1979). Earthquake sequences on a frictional fault model 633

with non-uniform strengths and relaxation times. Geophys. J. Int. 59, 497-522. 634

Miller, A. D., Foulger, G. R., and Julian, B. R. (1998). Non‐ double‐ couple earthquakes 635

2. Observations. Rev. Geophys. 36, 551–568. 636

Mogi, K. (1962). Study of the elastic shocks caused by the fracture of heterogeneous 637

materials and its relation to earthquakes phenomena. Bull. Earthq. Res. Inst. 40, 125–173. 638

Montési, L. G. (2004). Controls of shear zone rheology and tectonic loading on 639

postseismic creep J. Geophys. Res.: Solid Earth 109, B10404. 640

Murru, M., Console, R., Falcone, G., Montuori, C., and Sgroi, T. (2007). Spatial 641

30

mapping of the b value at Mount Etna, Italy, using earthquake data recorded from 1999 to 2005. 642

J. Geophys. Res.: Solid Earth 112, B12303 643

Narteau, C., Shebalin, P., and Holschneider, M. (2002). Temporal limits of the power law 644

aftershock decay rate. J. Geophys. Res.: Solid Earth 107, ESE–12. 645

Ogata, Y. (1998). Space-time point-process models for earthquake occurrences. Ann. Inst. 646

Stati. Math. 50, 379–402. 647

Ogata, Y. (1983). Estimation of the parameters in the modified Omori formula for 648

aftershock frequencies by the maximum likelihood procedure. J. Phys. Earth 31, 115–124. 649

Omori F. (1894). On after‐ shocks of earthquakes, J. Coll. Sci. Imp. Univ. Tokyo 7, 111–650

200. 651

Peng, Z., and Zhao, P. (2009). Migration of early aftershocks following the 2004 652

Parkfield earthquake. Nat. Geosci. 2, 877. 653

Perfettini, H., Frank, W. B., Marsan, D., and Bouchon, M. (2018). A model of aftershock 654

migration driven by afterslip. Geophys. Res. Lett. 45, 2283–2293. 655

Rhee, H. M., and Sheen, D. H. (2016). Lateral variation in the source parameters of 656

earthquakes in the Korean Peninsula. Bull. Seismol. Soc. Am. 106, 2266–2274. 657

Rigo, A., Souriau, A., and Sylvander, M. (2018). Spatial variations of b-value and crustal 658

stress in the Pyrenees. J. Seismol. 22, 337–352. 659

Rong, K., Yoon, C. E., Bergen, K. J., Elezabi, H., Bailis, P., Levis, P., and Beroza, G. C. 660

(2018). Locality-sensitive hashing for earthquake detection: a case study of scaling data-driven 661

science. Proceedings of the VLDB Endowment, 11, 1674-1687. 662

31

Ross, Z. E., Rollins, C., Cochran, E. S., Hauksson, E., Avouac, J. P., and Ben‐ Zion, Y. 663

(2017). Aftershocks driven by afterslip and fluid pressure sweeping through a fault‐ fracture 664

mesh. Geophys. Res. Lett. 44, 8260–8267. 665

Shapiro, S. A., Dinske, C., and Kummerow, J. (2007). Probability of a given‐ magnitude 666

earthquake induced by a fluid injection. Geophys. Res. Lett. 34, L22314. 667

Shi, Y., and Bolt, B. A. (1982). The standard error of the magnitude-frequency b value. 668

Bull. Seismol. Soc. Am. 72, 1677–1687. 669

Smith, W. D. (1981). The b-value as an earthquake precursor. Nature 289, 136–139. 670

Song, C. W., Son M., Sohn, Y. K., Han R., Shinn Y. J., and Kim J.-C. (2015). A study on 671

potential geologic facility sites for carbon dioxide storage in the Miocene Pohang Basin, SE 672

Korea. J. Geol. Soc. Korea 51, 53–66 (in Korean). 673

Song, S. G., and Lee, H. (2019). Static slip model of the 2017 Mw 5.4 Pohang, South 674

Korea, earthquake constrained by the InSAR data. Seismol. Res. Lett. 90, 140–148. 675

Sabbione, J. I., and Velis, D. R. (2013). A robust method for microseismic event 676

detection based on automatic phase pickers. J. Appl. Geophys. 99, 42-50. 677

Schorlemmer, D., Wiemer, S., and Wyss, M. (2005). Variations in earthquake-size 678

distribution across different stress regimes. Nature 437, 539–542. 679

Scholz, C. H. (1968). The frequency-magnitude relation of microfracturing in rock and 680

its relation to earthquakes. Bull. Seismol. Soc. Am. 58, 399–415. 681

Scholz, C. H. (2015). On the stress dependence of the earthquake b value. Geophys. Res. 682

Lett. 42, 1399–1402. 683

Shelly, D. R., Hardebeck, J. L., Ellsworth, W. L., and Hill, D. P. (2016). A new strategy 684

for earthquake focal mechanisms using waveform‐ correlation‐ derived relative polarities and 685

32

cluster analysis: Application to the 2014 Long Valley Caldera earthquake swarm. J. Geophys. 686

Res.: Solid Earth 121, 8622–8641. 687

Soh, I., Chang, C., Lee, J., Hong, T. K., and Park, E. S. (2018). Tectonic stress 688

orientations and magnitudes, and friction of faults, deduced from earthquake focal mechanism 689

inversions over the Korean Peninsula. Geophys. J. Int. 213, 1360–1373. 690

Son, M., Cho, C. S., Shin, J. S., Rhee, H. M., and Sheen, D. H. (2018). Spatiotemporal 691

Distribution of Events during the First Three Months of the 2016 Gyeongju, Korea, Earthquake 692

Sequence. Bull. Seismol. Soc. Am. 108, 210–217. 693

Tajima, F., and Kanamori, H. (1985). Global survey of aftershock area expansion 694

patterns. Phys. Earth Planet. Inter. 40, 77–134. 695

Trnkoczy, A. (1999). Topic Understanding and parameter setting of STA/LTA trigger 696

algorithm. New manual of seismological observatory practice, 2. 697

Trnkoczy A. (2002). Understanding and parameter setting of STA/LTA trigger algorithm, 698

in IASPEI New Manual of Seismological Observatory Practice (NMSOP), Bormann P. (Editor), 699

Vol. 2, GeoForschungsZentrum, Potsdam, Germany, 1–19. 700

Uchide, T., and Song, S. G. (2018). Fault rupture model of the 2016 Gyeongju, South 701

Korea, earthquake and its implication for the underground fault system. Geophys. Res. Lett. 45, 702

2257–2264. 703

Utsu, T. (1961). A statistical study on the occurrence of aftershocks, Geophys. Mag. 30, 704

521–605. 705

Utsu, T. (1992), On Seismicity, Report of the Joint Research Insititute for Statistical 706

Mathematics, Inst. Stat. Math., Tokyo 34, 139–157. 707

33

Utsu, T., Y. Ogata, and Matsu’ura, R. S. (1995). The centenary of the Omori formula for 708

a decay law of aftershock activity, J. Phys. Earth 43, 1–33. 709

Wadati, K., and Oki, S. (1933). On the travel time of earthquake waves, Part II, Geophys. 710

Mag. 7, 101–111. 711

Waldhauser, F., and Ellsworth, W. L. (2000). A double-difference earthquake location 712

algorithm: Method and application to the northern Hayward fault, California. Bull. Seismol. Soc. 713

Am., 90, 1353–1368. 714

Wiemer, S., and Katsumata, K. (1999). Spatial variability of seismicity parameters in 715

aftershock zones. J. Geophys. Res.: Solid Earth 104, 13135–13151. 716

Wiemer, S., and Wyss, M. (2000). Minimum magnitude of completeness in earthquake 717

catalogs: Examples from Alaska, the western United States, and Japan. Bull. Seismol. Soc. 718

Am. 90, 859–869. 719

Withers, M., Aster, R., Young, C., Beiriger, J., Harris, M., Moore, S., and Trujillo, J. 720

(1998). A comparison of select trigger algorithms for automated global seismic phase and event 721

detection. Bull. Seismol. Soc. Am. 88, 95–106. 722

Woo, J. U., Kim, M., Sheen, D. H., Kang, T. S., Rhie, J., Grigoli, F., Ellsworth, W., L., 723

and Giardini, D. (2019). An in‐ depth seismological analysis revealing a causal link between the 724

2017 MW 5.5 Pohang earthquake and EGS project. J. Geophys. Res.: Solid Earth. 124, 13060–725

13078. 726

Woo, J. U., Rhie, J., Kim, S., Kang, T. S., Kim, K. H., and Kim, Y. (2019). The 2016 727

Gyeongju earthquake sequence revisited: aftershock interactions within a complex fault system. 728

Geophys. J. Int., 217, 58–74. 729

34

Wyss, M. (1973). Towards a physical understanding of the earthquake frequency 730

distribution. Geophys. J. R. Astron. Soc., 31, 341–359. 731

Yoon, C. E., Bergen, K. J., Rong, K., Elezabi, H., Ellsworth, W. L., Beroza, G. C., Bailis 732

P. and Levis, P. (2019a). Unsupervised Large‐ Scale Search for Similar Earthquake Signals. Bull. 733

Seismol. Soc. Am. 109, 1451–1468. 734

Yoon, C. E., O’Reilly, O., Bergen, K. J., and Beroza, G. C. (2015). Earthquake detection 735

through computationally efficient similarity search. Sci. Adv. 1, e1501057. 736

Yoon, C. E., Huang, Y., Ellsworth, W. L., and Beroza, G. C. (2017). Seismicity during 737

the initial stages of the Guy‐ Greenbrier, Arkansas, earthquake sequence. J. Geophys. Res.: Solid 738

Earth 122, 9253–9274. 739

Yoon, C. E., Yoshimitsu, N., Ellsworth, W. L., and Beroza, G. C. (2019b). Foreshocks 740

and Mainshock Nucleation of the 1999 M w 7.1 Hector Mine, California, Earthquake. J. 741

Geophys. Res.: Solid Earth 124, 1569–1582. 742

Yun H., Min, K. D., Moon, H.-S., Lee H. K., and Yi, S. S. (1991). Biostratigraphic, 743

Chemostratigraphic, Paleomagnetostratigraphic, and Tertiary formations in southern part of 744

Korea: regional tectonics and its stratigraphical implication in the Pohang basin. J. Paleont. Soc. 745

Korea 7, 1–12 (in Korean). 746

Zoback, M. L. (1992). First‐ and second‐ order patterns of stress in the lithosphere: The 747

World Stress Map Project. J. Geophys. Res.: Solid Earth 97, 11703–11728. 748

749

Full mailing address for each author 750

1 School of Earth and Environmental Sciences, Seoul National University, Seoul 08826, 751

Republic of Korea 752

35

J.-U.W., woong140@snu.ac.kr 753

J.R., rhie@snu.ac.kr 754

2 Department of Structural Systems and Site Evaluation, Korea Institute of Nuclear Safety, 755

Daejeon 34412, South Korea 756

M.K., kimmw0419@kins.re.kr 757

3 Division of Earth Environmental System Science, Pukyong National University, Busan 48513, 758

Republic of Korea 759

T.-S.K., tskang@pknu.ac.kr 760

761

Tables 762

763

Table 1. The 1-D layered seismic velocity structure for the Pohang EGS site. 764

Depth to the top of the layer (km) P-wave velocity (kms-1) S-wave velocity (kms-1)

0.0 1.67 0.48

0.203 4.01 2.21

0.67 5.08 3.03

2.4 5.45 3.07

3.4 5.85 3.31

7.7 5.91 3.51

12 6.44 3.70

34 8.05 4.60

765

Table 2. Parameters of the faults involved in the aftershock sequences. 766

36

Properties Fault 1 (F1) Fault 2 (F2) Fault 3 (F3) Fault 4 (F4)

Strike (°)

Median 222.7 207.4 223.1 25.8

Median

absolute

deviation

1.1 1.2 0.4 3.9

Dip (°)

Median 77.4 59.8 61.2 68.2

Median

absolute

deviation

2.0 1.3 0.6 5.0

Fault length (km) 2.8 2.4 3.4 2.1

Fault width (km) 1.9 3.5 2.9 1.4

Fault thickness (km) 0.9 0.8 0.7 0.8

767

37

List of figure captions 768

769

Figure 1. Map of (a) temporary and (b) permanent seismic stations used for analysis of source 770

parameters, geologic lineaments, faults, and relocated hypocenters. Three surface ruptures near 771

the study area are illustrated in (a) (Song et al., 2015; Yun et al., 1991). The focal mechanism of 772

the mainshock that was determined from the polarity of first arrivals is illustrated in (b). (c) 773

shows the location of the Gyeongsang Basin (GB) and the Yeonil Basin (YB) where many 774

NENNE sinistral strike-slip surface ruptures and NW transfer faults have developed. The red 775

boxes in (b) and (c) represent the domain of (a) and (b), respectively. 776

777

Figure 2. (a) Distribution of the 3946 epicenters relocated via hypoDD (Waldhauser and 778

Ellsworth, 2000) by using traveltime differences. The earthquakes projected onto each of the 779

cross-sections A1-A2 to E1-E2 shown in (b) to (f) fall within the rectangles denoted by dashed 780

black lines in (a). The trajectory of two stimulation wells PX-1 and PX-2 are illustrated as gray 781

lines in (c) with open sections colored in blue and red. The mainshock and two largest 782

aftershocks (ML 4.3 and ML 4.6) are denoted as red, blue, and green stars, respectively. (b–f) 783

Depth distribution of the relocated hypocenters along the cross-sections of A1-A2 to E1-E2. 784

Possible interpretations for delineated faults from the aftershock distribution are marked as gray 785

lines in (b), (c), (d), and (e). The circles in (f) represent the rupture radii of earthquakes with MRel 786

> 1.5, assuming a stress drop of 5.6 MPa, which corresponds to an approximated value for the 787

mainshock estimated by the spectral ratio method (Woo et al., 2019a). The red, blue, and green 788

circles in (f) indicate the rupture size of the three largest earthquakes with ML 5.4, 4.3, and 4.6, 789

respectively. 790

38

791

Figure 3. (a) Determination of the scaling parameter c in Equation (1) from known ML 792

magnitudes. The amplitude ratios measured from two similar waveforms observed at a station 793

are measured and counted to estimate the scaling parameter for given MLs. The red line indicates 794

the scaling parameter c of 0.74 calculated from the slope of the first principal components 795

between magnitude differences and the ratio of amplitude divided by hypocenteral distances. (b) 796

Comparison between ML and MRel. The red line indicates identity relation. (c) The distribution of 797

earthquake magnitudes with their origin time. The three largest earthquakes (ML 5.4, 4.3, and 4.6) 798

are denoted as red, blue, and green stars, respectively, with their MLs. The microearthquakes of 799

which magnitudes cannot be measured from Equation (1) are denoted as square symbols at the 800

bottom of the graph. 801

802

Figure 4. (a) Temporal variations of seismic b-values and MC for each bin of each time period. 803

We combined MRels obtained from Equation (1) with MLs of the three largest earthquakes. A set 804

of 600 earthquakes constitute a bin for measuring b-values and there is an overlap of 400 805

earthquakes between two consecutive bins. The standard deviations of each of the magnitude 806

bins are represented as vertical and horizontal error bars. The black, red, and blue dots indicate 807

three typical bins for the evaluation of b-values in (b). The gray dots and error bars represent b-808

values and their standard errors calculated based on a maximum MC of 1 for all bins. (b) Four 809

examples of the curves used to estimate the two scaling parameters of the G-R law. For each case, 810

the color used for plotting the data and the equation of G-R law corresponds to a specific bin of 811

(a). The filled circles indicate the cutoff magnitudes of MC that honor Equation (1) for larger 812

magnitudes. (c) Two dimensional spatial variations of b-values at a vertical profile along both 813

39

the apparent strike of 210°. Hypocenters of three largest earthquakes are denoted as red, blue, 814

and green stars, respectively. Dashed lines and solid lines indicate the significant difference level 815

of 5% with median and conservative thresholds. (d) Two-sigma interval distribution of 816

probabilities that differences of paired b-values (Δb) in (c) is insignificant. The Δbs are binned by 817

0.05. The difference is statistically significant with a significance level of 5% if Δb > 0.1 for half 818

the cases and Δb > 0.135 for all cases. 819

820

Figure 5. (a) Two dimensional spatial variations of p-values for a cross-section along the 821

apparent strike 210° for earthquake sequences before the occurrence of the largest aftershock 822

(ML 4.6). (b) Two dimensional spatial variations of p-values for the same depth profile shown in 823

(a), but for seismic sequences after the onset of the ML4.6 earthquake. (c and d) aftershock decay 824

rates and their corresponding Omori’s law plots with the estimated parameters obtained from the 825

whole data set used for mapping p-values in (a) and (b), respectively. 826

827

Figure 6. Temporal distribution of seismicity presented in (a) plan view, (b) a depth profile along 828

the apparent strike of 210°, and (c) a unidirectional projection along the apparent strike. We 829

illustrate the radius of earthquakes with MRel ≥ 1.5 by assuming a circular crack rupture and a 830

stress drop of 5.6 MPa from Woo et al. (2019a). In (a) and (b), the location of the mainshock and 831

the two largest consecutive aftershocks of ML 4.6 and ML 5.4 are represented as red, blue and 832

green stars, respectively. The rupture radii of the three largest earthquakes are displayed in (c) 833

with colors to match the star symbols in (a) and (b). The trajectory of two stimulation wells PX-1 834

and PX-2 are illustrated as gray lines in (a) and (b). The thick gray lines in (c) represent the 835

40

linear scaling relationship between the along-strike expansion of earthquake sequences with 836

logarithmic time scale. 837

838

Figure 7. Schematic diagram that illustrates four fault segments inferred from the hydraulic 839

stimulation wells of PX-1 and PX-2 and the distribution of aftershocks. The three cubes colored 840

in red, blue, and green show the hypocenters of the mainshock and the two largest aftershocks of 841

ML 4.3 and 4.6, respectively. The occurrence of the mainshock triggered seismicity on fault 842

segments F2 and F3, and possibly affected the re-activation of F1. Fault segment F4, located to 843

the southwest of F3, was not delineated until the largest aftershock (ML 4.6) occurred. 844

845

846

41

Figures 847

848

849

Figure 1. Map of (a) temporary and (b) permanent seismic stations used for analysis of source 850

parameters, geologic lineaments, faults, and relocated hypocenters. Three surface ruptures near 851

the study area are illustrated in (a) (Song et al., 2015; Yun et al., 1991). The focal mechanism of 852

the mainshock that was determined from the polarity of first arrivals is illustrated in (b). (c) 853

shows the location of the Gyeongsang Basin (GB) and the Yeonil Basin (YB) where many 854

NENNE sinistral strike-slip surface ruptures and NW transfer faults have developed. The red 855

boxes in (b) and (c) represent the domain of (a) and (b), respectively. 856

42

857

858

Figure 2. (a) Distribution of the 3946 epicenters relocated via hypoDD (Waldhauser and 859

Ellsworth, 2000) by using traveltime differences. The earthquakes projected onto each of the 860

cross-sections A1-A2 to E1-E2 shown in (b) to (f) fall within the rectangles denoted by dashed 861

black lines in (a). The trajectory of two stimulation wells PX-1 and PX-2 are illustrated as gray 862

lines in (c) with open sections colored in blue and red. The mainshock and two largest 863

aftershocks (ML 4.3 and ML 4.6) are denoted as red, blue, and green stars, respectively. (b–f) 864

Depth distribution of the relocated hypocenters along the cross-sections of A1-A2 to E1-E2. 865

Possible interpretations for delineated faults from the aftershock distribution are marked as gray 866

43

lines in (b), (c), (d), and (e). The circles in (f) represent the rupture radii of earthquakes with MRel 867

> 1.5, assuming a stress drop of 5.6 MPa, which corresponds to an approximated value for the 868

mainshock estimated by the spectral ratio method (Woo et al., 2019a). The red, blue, and green 869

circles in (f) indicate the rupture size of the three largest earthquakes with ML 5.4, 4.3, and 4.6, 870

respectively. 871

872

44

873

874

Figure 3. (a) Determination of the scaling parameter c in Equation (1) from known ML 875

magnitudes. The amplitude ratios measured from two similar waveforms observed at a station 876

are measured and counted to estimate the scaling parameter for given MLs. The red line indicates 877

the scaling parameter c of 0.74 calculated from the slope of the first principal components 878

between magnitude differences and the ratio of amplitude divided by hypocenteral distances. (b) 879

Comparison between ML and MRel. The red line indicates identity relation. (c) The distribution of 880

earthquake magnitudes with their origin time. The three largest earthquakes (ML 5.4, 4.3, and 4.6) 881

are denoted as red, blue, and green stars, respectively, with their MLs. The microearthquakes of 882

which magnitudes cannot be measured from Equation (1) are denoted as square symbols at the 883

bottom of the graph. 884

885

45

886

887

Figure 4. (a) Temporal variations of seismic b-values and MC for each bin of each time period. 888

We combined MRels obtained from Equation (1) with MLs of the three largest earthquakes. A set 889

of 600 earthquakes constitute a bin for measuring b-values and there is an overlap of 400 890

earthquakes between two consecutive bins. The standard deviations of each of the magnitude 891

bins are represented as vertical and horizontal error bars. The black, red, and blue dots indicate 892

three typical bins for the evaluation of b-values in (b). The gray dots and error bars represent b-893

values and their standard errors calculated based on a maximum MC of 1 for all bins. (b) Four 894

examples of the curves used to estimate the two scaling parameters of the G-R law. For each case, 895

the color used for plotting the data and the equation of G-R law corresponds to a specific bin of 896

(a). The filled circles indicate the cutoff magnitudes of MC that honor Equation (1) for larger 897

magnitudes. (c) Two dimensional spatial variations of b-values at a vertical profile along both 898

the apparent strike of 210°. Hypocenters of three largest earthquakes are denoted as red, blue, 899

46

and green stars, respectively. Dashed lines and solid lines indicate the significant difference level 900

of 5% with median and conservative thresholds. (d) Two-sigma interval distribution of 901

probabilities that differences of paired b-values (Δb) in (c) is insignificant. The Δbs are binned by 902

0.05. The difference is statistically significant with a significance level of 5% if Δb > 0.1 for half 903

the cases and Δb > 0.135 for all cases. 904

47

905

906

Figure 5. (a) Two dimensional spatial variations of p-values for a cross-section along the 907

apparent strike 210° for earthquake sequences before the occurrence of the largest aftershock 908

(ML 4.6). (b) Two dimensional spatial variations of p-values for the same depth profile shown in 909

(a), but for seismic sequences after the onset of the ML 4.6 earthquake. (c and d) aftershock decay 910

rates and their corresponding Omori’s law plots with the estimated parameters obtained from the 911

whole data set used for mapping p-values in (a) and (b), respectively. 912

913

48

914

Figure 6. Temporal distribution of seismicity presented in (a) plan view, (b) a depth profile along 915

the apparent strike of 210°, and (c) a unidirectional projection along the apparent strike. We 916

illustrate the radius of earthquakes with MRel ≥ 1.5 by assuming a circular crack rupture and a 917

stress drop of 5.6 MPa from Woo et al. (2019a). In (a) and (b), the location of the mainshock and 918

the two largest consecutive aftershocks of ML 4.6 and ML 5.4 are represented as red, blue and 919

green stars, respectively. The rupture radii of the three largest earthquakes are displayed in (c) 920

with colors to match the star symbols in (a) and (b). The trajectory of two stimulation wells PX-1 921

and PX-2 are illustrated as gray lines in (a) and (b). The thick gray lines in (c) represent the 922

linear scaling relationship between the along-strike expansion of earthquake sequences with 923

logarithmic time scale. 924

49

925

926

Figure 7. Schematic diagram that illustrates four fault segments inferred from the hydraulic 927

stimulation wells of PX-1 and PX-2 and the distribution of aftershocks. The three cubes colored 928

in red, blue, and green show the hypocenters of the mainshock and the two largest aftershocks of 929

ML 4.3 and 4.6, respectively. The occurrence of the mainshock triggered seismicity on fault 930

segments F2 and F3, and possibly affected the re-activation of F1. Fault segment F4, located to 931

the southwest of F3, was not delineated until the largest aftershock (ML 4.6) occurred. 932

933

934

50

Appendices 935

936

Table A1. Fingerprint extraction parameters for the FAST algorithm to detect earthquakes with 937

waveform similarity. 938

Fingerprint extraction parameter Value

Time-series window length of generated spectrogram 6.0 s

Time-series window lag of generated spectrogram 0.1 s

Spectral image window length 64

Spectral image window lag 10

Fingerprint sparsity 400

Final spectral image width 32

Number of hash functions per hash table 4

Number of hash tables 100

Number of votes 2

Near-repeat exclusion parameters 5

939

Table A2. Input parameters for the network detection in the FAST algorithm (Bergen et al., 2016; 940

Bergen and Beroza, 2018; Rong et al., 2018) 941

Event-pair extraction, pruning, and network detection parameters values

Time gap along diagonal 3 s

Time gap adjacent diagonal 3 s

Adjacent diagonal merge iteration 2

51

Number of votes 10

Minimum fingerprint pairs 3

Maximum bounding-box width 5 s

Minimum number of stations for detection 1

Arrival time constraint: maximum time gap 5

942

943

944

945

52

Figure A1. Snapshots of the distribution maps of aftershocks at (a) 10 min, (b) 1 h, (c) 5 h, (d) 1 946

day, (e) 30 day, and (f) 4mo from the onset of the mainshock. For each subfigure, earthquakes 947

until Red and gray dots 948

949