The November 2017 Mw 5.5 Pohang earthquake: A possible case of induced seismicity in South Korea

See allHide authors and affiliations

Science  01 Jun 2018:
Vol. 360, Issue 6392, pp. 1003-1006
DOI: 10.1126/science.aat2010

Triggering quakes in a geothermal space

Enhanced geothermal systems (EGSs) provide a potentially clean and abundant energy source. However, two magnitude-5 earthquakes recently occurred in South Korea during EGS site development. Grigoli et al. and Kim et al. present seismic and geophysical evidence that may implicate the second of these earthquakes, which occurred in Pohang, as an induced event. The combination of data from a local seismometer network, well logs, satellite observations, teleseismic waveform analysis, and stress modeling leads to the assessment that the earthquake was probably or almost certainly anthropogenically induced. The possibility remains that the earthquake occurred coincidentally at the EGS site location, but the aftershock distribution and other lines of evidence are concerning for future development of this geothermal resource.

Science, this issue p. 1003, p. 1007


The moment magnitude (Mw) 5.5 earthquake that struck South Korea in November 2017 was one of the largest and most damaging events in that country over the past century. Its proximity to an enhanced geothermal system site, where high-pressure hydraulic injection had been performed during the previous 2 years, raises the possibility that this earthquake was anthropogenic. We have combined seismological and geodetic analyses to characterize the mainshock and its largest aftershocks, constrain the geometry of this seismic sequence, and shed light on its causal factors. According to our analysis, it seems plausible that the occurrence of this earthquake was influenced by the aforementioned industrial activities. Finally, we found that the earthquake transferred static stress to larger nearby faults, potentially increasing the seismic hazard in the area.

Deep geothermal resources can provide a valuable contribution to the production of renewable energy. Through enhanced geothermal systems (EGSs), geothermal energy production is no longer confined to volcanic or hydrothermal regions. Unlike the conventional geothermal systems, EGS technologies exploit geothermal resources through hydraulic stimulation—which involves the injection of high-pressure cold water to increase the permeability of the target formation at a few kilometers of depth—by creating new fractures or enhancing existing ones. Although the potential for deep geothermal energy is indisputably large, in urban areas the problem of induced seismicity associated with such operations is often not adequately addressed.

On 15 November 2017, a moment magnitude (Mw) 5.5 earthquake struck South Korea, injuring ~70 people and causing extensive damage in and around the city of Pohang. This earthquake was preceded by the Mw 5.5 Gyeongju event of 12 September 2016, which occurred ~30 km farther south on a major right-lateral fault, the Yangsan fault, which continues northward through the Pohang area (1, 2) (Fig. 1, A and B). These earthquakes are the largest recorded in South Korea since instrumental monitoring of seismicity began in 1903 (2). The proximity of the 2017 Pohang earthquake to an EGS site (Fig. 1B), where hydraulic stimulation operations had recently taken place, has led to a public debate in South Korea regarding the potential anthropogenic origin of this event. At this EGS site between early 2016 and September 2017, many thousands of cubic meters of water were injected under pressure into wells reaching ~4 km of depth (3). Although an investigation by the South Korean government is currently ongoing, here we present observations that suggest a causal connection between the EGS activity and the most recent large earthquake.

Fig. 1 The 2016 and 2017 Mw 5.5 earthquakes in South Korea.

(A) Regional map showing locations of the Gyeongju and Pohang earthquakes, the Yangsan fault, and the available open seismic stations. (B) Map of the study area showing the main faults of the area, the distribution of seismicity with respect to the EGS site, and the mechanisms of the largest events. A more detailed map of the area of study (outlined by the yellow square) is shown in Fig. 2A. UTC, universal time coordinated.


The Korean Peninsula is generally considered stable with low to moderate intraplate seismic activity, but the historical seismicity of this region indicates large long-term variations in earthquake rate and energy release (4, 5). The relatively low rate of activity since 1904 was preceded by much higher activity between the 15th to 18th centuries, with a peak of 1000 reported historical earthquakes between 1500 and 1600 CE (5). The largest events reached magnitude ~7 (4, 5). The historic earthquakes were likely associated with the major fault systems of the area, such as the Yangsan fault, and highlight that these structures are active (4, 5). In principle, given the historically varying rates of seismicity and prevalence of faults in this region, the increase in seismic activity represented by the 2016 Gyeongju and 2017 Pohang earthquakes is not completely inconsistent with the historically varying rates of seismicity. This line of reasoning preserves the possibility that the occurrence of earthquakes close to the EGS site is a coincidence.

We applied full-waveform seismological methods to regional and teleseismic data (Fig. 1A) (6), as we do not have access to open data from a local seismic network (with the exception of two accelerometers deployed in the epicentral area). We analyzed 15 days of continuous waveform data spanning from 15 to 30 November. We detected and relocated 46 events, most with magnitude M >2. The trend of these 46 epicenters indicates a west-southwest (WSW)–to–east-northeast (ENE) strike of the fault that ruptured in the mainshock (Fig. 2A). We determined 3- to 7-km hypocentral depths for most of these events (Fig. 2, A and B). These depths are shallower than the depths of typical seismic events in the area (~12 to 15 km) (Fig. 2C) (2, 4). For the Pohang earthquake, we determined the depth of both the mainshock and the largest aftershock to be 4.0 to 4.5 km (Fig. 2B). Depth is a critical parameter for discrimination between natural and induced seismicity (7), so we obtained independent estimates (by using array analysis at teleseismic distances and two accelerometers located in the epicentral region) that confirm the shallow depth (6). Our moment tensor inversion indicated that the mainshock had a reverse-to-oblique double-couple (DC) mechanism, with a WSW-to-ENE striking nodal plane, subparallel to the aftershock zone and dipping north-northwest at ~66°. The full moment tensor had a large non-DC term. Conversely, the Mw 4.3 aftershock indicated reverse faulting on a WSW-to-ENE striking fault (Figs. 1B and 2A). On the basis of many previous analyses [e.g., (8, 9)], we inferred that the non-DC component (Fig. 2A) is caused by a complex rupture process that includes the (near-)simultaneous activation of differently oriented faults. By mapping the azimuthal distribution of the apparent source durations, maximum energy peaks, and centroid time delays, we detected a common pattern, which we interpret as the failure of two subevents at close origin times and separated by a short distance along the azimuth of rupture directivity. We estimated a distance of 3.5 to 4.0 km between the two subevents, indicative of a dynamic triggering process (6). We thus hypothesize that the earthquake involved the failure of two different faults with slightly different orientations, which might, in principle, explain the non-DC term of the moment tensor (9), as well as the complexity of P-wave signals for the mainshock (6). A potential alternative source model, characterized by a complex rupture along a single fault, with heterogeneous slip directions, cannot explain the non-DC component of the moment tensor and results in a substantially worse fit for the pattern of the relative hypocentral centroid location (6).

Fig. 2 Spatial distribution of the 2017 Pohang seismic sequence.

(A) Detailed map showing the epicentral distribution of seismicity and (B) two cross sections displaying the depth distribution of seismicity (including location uncertainties) and the fault as inferred by geodetic analysis. The EGS site is located at 36°06′23.34′′N, 129°22′46.08′′E and includes the two injection wells that reach depths of 4127 and 4348 m (3). (C) Focal depth distribution of earthquakes in the study region and comparison with the 2016 Gyeongju seismic sequence.

We also used satellite radar interferometry [differential interferometric synthetic aperture radar (DInSAR)] to map and measure the coseismic surface deformation associated with the mainshock and to independently model its source geometry (6, 10). Our InSAR analysis indicated a maximum surface deformation of ~5 cm, with a fault location and area consistent with the seismological analysis (Fig. 3A). Despite the moderate event size and complex rupture resolved by seismological data, we can fit the observed deformation with a simple elastic dislocation model (11) based on a single fault plane with shear displacement and opening (Fig. 3B). The fault dimensions (length ~5 km and width ~1.6 km) and the apparent slip, derived from this inversion of geodetic data, are compatible with a magnitude of Mw 5.5 (12) and consistent with our seismological analysis. However, DInSAR data cannot resolve the small-scale complexities associated with the progressive rupture of adjacent fault patches. Nonetheless, our geodetic analysis confirmed that the earthquake nucleation and main slip on the fault occurred at very shallow depth (4 to 5 km), on a reverse fault striking WSW-to-ENE and with a ~75° dip toward the NW (Fig. 2B). The average residuals are ~0.05 cm and within the measurement accuracy (Fig. 3C). The DInSAR results are in good agreement with the aftershock locations and the focal mechanisms of the largest events, placing strong independent constraints on the location and extent of a previously unmapped fault system.

Fig. 3 DInSAR data and model.

(A) Surface deformation (satellite line-of-sight displacements) obtained with InSAR. Seismicity and the extrapolated fault trace are indicated by black circles and a dashed line, respectively. (B) Modeled surface deformation using a rectangular fault plane with the following parameters: latitude = 36.100° ± 0.005°, longitude = 129.383° ± 0.003° (center of the rectangular fault), depth = 4.3 ± 0.3 km (upper edge of the fault), strike (from north) = 225° ± 12°, dip (from horizon) = 75° ± 11°, length = 5.0 ± 0.7 km, width = 1.6 ± 0.4 km, slip = 1 ± 0.22 m, and rake = 123° ± 35°. (C) Difference between InSAR data and model. The standard deviation is <0.5 cm, which is below the accuracy threshold of the measurements.

Natural and induced earthquakes are not distinguishable by their waveform characteristics. It is generally necessary to build a convincing chain of evidence to differentiate induced earthquakes from naturally occurring events (7, 1316). For instance, the hypothesis that the Pohang earthquake sequence is anthropogenic is supported by the spatial correlation between the mainshock and its aftershocks and the injections. Our seismological and geodetic analyses indicate that the activated fault passes directly beneath the EGS site (Fig. 2B), within ~1 km of the termination of the injection wells (Fig. 2B). The combined evidence from the hypocentral locations, the DInSAR data inversion, the observed interval between P- and S-wave arrival times at the borehole and surface stations, and the observed strong motion and damage patterns is consistent with this interpretation (6). Another piece of evidence supporting that this seismicity is induced is our relocation of the local magnitude 3.1 earthquake on 15 April 2017, which occurred during hydraulic stimulation operations at the EGS site: This event was very close to the 15 November mainshock (Fig. 2A). No stimulation activities occurred in the 2 months preceding this mainshock, but induced earthquakes can be delayed by days, weeks, or even months after the start or end of injection (14). Our techniques (using regional and teleseismic data from the public domain, thus restricted to earthquakes of magnitude ≥~2) demonstrate the extent to which a candidate case study of induced seismicity can be investigated using public data, without proprietary data from site operators.

We also studied the relation between the seismic moment (Mo ~ 1.7 × 1017 N·m) of the mainshock (6) and the volume (V) of fluid injected. Assuming that unreported injections are similar to initial stimulation (3), we estimated an upper bound to the total volume of ~10,000 m3. The limitation of water supply and storage capacity of the EGS justifies our assumption. We approximated an upper bound to Mo by using the product of net injected volume and the shear modulus (17). For V = 10,000 m3, this relation implies an upper bound of Mo ~ 2 × 1014 N·m, or Mw ~ 3.5. On this basis, an induced earthquake of Mw 5.5 would require three orders of magnitude more volume (V ~ 107 m3). However, several counterexamples to this scaling relation exist [(18, 19) and references therein], and the Pohang earthquake provides much clearer evidence that the use of this relation should be subject to caution.

Although none of our observations exclude the possibility that the Pohang earthquake was induced by industrial activity at the EGS site, our seismological and geodetic analyses (Figs. 2 and 3) rule out a reactivation of the Yangsan fault. Our hypocenter is <1 km southeast of the injection point, at the same depth as the injection. The coseismic deformation and source model derived from the DInSAR analysis confirm this location (Figs. 2 and 3). Aftershock locations and focal mechanisms consistently indicate activation of a previously unmapped fault system. This case thus highlights the importance of a preliminary seismotectonic assessment of the area surrounding any future EGS project site, aimed at identifying potentially active faults. Along with real-time analysis and control systems (20), this evaluation may mitigate the risk associated with induced seismicity.

We also investigated whether the September 2016 Mw 5.5 Gyeongju earthquake might have contributed to triggering the 2017 Pohang event. Coulomb stress modeling (6) indicates that the Gyeongju earthquake caused a slight (~0.0005 MPa) increase in static stress on the fault that ruptured in the subsequent Pohang earthquake. This value is small, but it is feasible to propose the occurrence of a clock advance on the subsequent fault. In turn, the Mw 5.5 Pohang earthquake transferred static stress of 0.015 MPa onto the northern part of the Yangsan fault (Fig. 4B), potentially increasing the seismic hazard in this area.

Fig. 4 Coulomb stress modeling.

(A) Static coulomb failure stress showing the effect of the 2016 Gyeongju earthquake at 4 km of depth on the fault activated by the 2017 Pohang event and (B) the effect of this latter event on the Yangsan fault at 14 km of depth. Receiver faults are denoted by red dot-dashed lines. The triangle represents the local accelerometer, and the squares denote the injection wells.

In accordance with our findings, it is plausible that the occurrence of the 2017 Pohang earthquake was influenced by the nearby stimulation activities. If so, the Pohang event was the largest and most damaging earthquake ever to have been associated with an EGS, making it a potential “game changer” for the geothermal industry worldwide.

Supplementary Materials

Supplementary Text

Figs. S1 to S12

Waveform and Local Accelerometer Data

References (2139)

References and Notes

  1. See supplementary materials.
Acknowledgments: We thank F. Bethmann, T. Kraft, and D. Giardini for information, comments, and suggestions that helped to improve the paper. Funding: This work was funded by the EU projects DESTRESS (EU H2020 research and innovation program, grant agreement 691728) and SHEER (EU H2020 research and innovation program, grant agreement 640896). A.P.R. is currently funded by a Swiss National Science Foundation, Ambizione Energy grant (PZENP2-160555). Author contributions: F.G. performed the analysis of the seismic sequence. S.C. performed moment tensor inversion. A.P.R. performed coulomb stress failure analysis. A.M. processed and inverted geodetic data. J.A.L.-C. performed the rupture directivity analysis. J.F.C. and C.C. analyzed the local strong motion data. F.G., R.W., T.D., and S.W. interpreted the results. All the authors contributed to writing and reviewing the paper. Competing interests: The authors declare no competing interests. Data and materials availability: The waveform data used in this study are publicly available and can be downloaded from the NIED (; for the Japanese network) and IRIS (; for the Korean network) websites. The data from the local accelerometer are available in the supplementary materials. Sentinel-1 radar data can be downloaded from the ESA website (

Stay Connected to Science

Navigate This Article