Migrating tremor off southern Kyushu as evidence for slow slip of a shallow subduction interface

See allHide authors and affiliations

Science  08 May 2015:
Vol. 348, Issue 6235, pp. 676-679
DOI: 10.1126/science.aaa4242

Silent slip events get shallow

Clues to help better predict the likelihood of devastating earthquakes and tsunamis may be embedded in a more gentle type of rumbling. Using oceanbottom seismometers, Yamashita et al. report rare observations of migrating tremors in the shallow part of a subduction zone off southern Kyushu, Japan. The tremors appear to be linked to a very low-frequency earthquake and seem to migrate to the region where big earthquakes are generated. The tremors may be tracing how and where stress gets concentrated onto the earthquake-producing portion of the fault.

Science, this issue p. 676


Detection of shallow slow earthquakes offers insight into the near-trench part of the subduction interface, an important region in the development of great earthquake ruptures and tsunami generation. Ocean-bottom monitoring of offshore seismicity off southern Kyushu, Japan, recorded a complete episode of low-frequency tremor, lasting for 1 month, that was associated with very-low-frequency earthquake (VLFE) activity in the shallow plate interface. The shallow tremor episode exhibited two migration modes reminiscent of deep tremor down-dip of the seismogenic zone in some other subduction zones: a large-scale slower propagation mode and a rapid reversal mode. These similarities in migration properties and the association with VLFEs strongly suggest that both the shallow and deep tremor and VLFE may be triggered by the migration of episodic slow slip events.

Slow earthquakes are a range of seismic phenomena with longer durations than ordinary earthquakes. They are considered key to understanding the generation of megathrust earthquakes because their activity affects stress accumulation on the locked zone. Slow earthquakes are well documented in the deep transition zone between the seismogenic zone and stable sliding zone. Episodic tremor and slip (ETS) is the coupled phenomena of nonvolcanic deep low-frequency tremor (1) and slow slip events (SSEs) (2) with a specific recurrence interval. Observations of ETS have been documented in Cascadia (3) and southwestern Japan (4). ETS events also coincide with deep very-low-frequency earthquake (VLFE) activity (5). These events share a common mechanism as shear slip on the plate interface (6). Because the spatiotemporal evolution of deep tremor and VLFEs reflects the rupture process of SSEs (7), tremor activity is useful for assessing interplate aseismic slip in the deep transition zone. Many observations of deep tremor suggest that SSEs are common on the deep side of the seismogenic zone.

The shallow subducting plate interface near the trench was thought to be a conditionally stable zone (8). Understanding the slip potential of this region is important for evaluating the development of great earthquake ruptures, such as the 2011 Tohoku-Oki earthquake, and the associated potential for tsunami generation. Therefore, the nature of slow earthquakes (particularly SSEs, which account for the greatest moment release of slow earthquakes) has critical implications for the interplate slip and moment budget of the shallow plate interface. Shallow VLFE activity has been observed in Japan (911) and Costa Rica (12), but low-frequency tremor observations near the trench are scarce (13). The previous studies have not indicated spatiotemporal variation of activities for shallow tremor and VLFEs. Although SSEs have recently been detected at both deep and shallow plate interface in southwest Japan by Global Navigation Satellite System (GNSS) analysis (14), the spatiotemporal relationship with seismic shallow slow events is still unclear.

The region east off the east coast of southern Kyushu (the western part of Nankai Trough) is one of the most active areas of ordinary earthquake seismicity (fig. S1) and shallow VLFE seismicity in southwestern Japan (9, 10). Estimation of interplate coupling from long-term ground deformation using GNSS data (15) and the existence of small repeating earthquakes (16) suggest that the subduction megathrust of this area is creeping (low interplate coupling) while producing small and moderate-sized interplate earthquakes. The locked Nankai subduction zone to the northeast provides a sharp contrast to this region. We expected a phenomenon like ETS to occur in the shallow plate interface because shallow VLFE seismicity is seen repeatedly (fig. S2), although no tremor has been previously reported in the area. We deployed 12 short-period ocean bottom seismometers (OBSs) for ~80 days to investigate shallow tremor activity of this area (Fig. 1) (17).

Fig. 1 Shallow low-frequency tremor distribution and tectonic setting.

Location map (inset) and bathymetry of the study area, showing spatiotemporal distribution of shallow tremor (colored circles) [time zone is Japan standard time (JST), or universal time coordinated + 09:00], locations and numbers of OBS stations (yellow squares), epicenters of small repeating earthquakes (orange circles) (16), the coseismic slip area of M7-class interplate earthquakes (dark gray) (26, 27), and the outer edge of the subducted KPR (dark green bold) (24). The convergence direction of the Philippine Sea Plate (30) is denoted by the red arrow, the trench axis is indicated by the dashed blue line, and bathymetric contours are shown at 100-m intervals. KUSM is a land-based seismic station, which used in Fig. 2A.

Fortuitously, the first major episode of shallow VLFE activity in 3 years started just after we deployed the OBSs. This VLFE episode occurred in the same area as previous activity, according to array analysis of data from land-based broadband seismic stations (10) (fig. S2). The episode lasted for 1 month and included three active stages (Fig. 2C). During the same time period, our OBS network recorded a complete, month-long episode of shallow tremor activity. The dominant frequency of the tremor waveforms, ranging from 0.5 to 8 Hz, was lower than that of ordinary earthquakes (Fig. 3) (17). Each peak of tremor in the OBS observations almost corresponded to each wave train recorded by land-based broadband seismographs, after allowing for the difference of ~1 min in travel time between the two sets of stations (Fig. 2, A and B).

Fig. 2 Comparison with shallow tremor and VLFE activity.

(A) Example of shallow tremor in 1-hour continuous root mean square (RMS) envelopes from the OBS network and land-based station KUSM (locations in Fig. 1). Envelope traces were processed with a 2- to 8-Hz band-pass filter and smoothed with a 5-s moving window. Note that the gain of the trace from KUSM is 20 times as high as that of the OBS stations. The event at 08:46 is an ordinary small earthquake near OBS station 11. (B) Example of shallow VLFE in 1-hour continuous waveforms from F-net land-based broadband stations operated by the National Research Institute for Earth Science and Disaster Prevention. Waveform traces were processed with a 10- to 50-s band-pass filter. UD, up-down (vertical) component. (C) Comparison of spatial (left) and temporal (right) distribution of shallow tremor (red) and shallow VLFEs (gray). Note that the shallow tremor and VLFEs were detected every 1 min and 15 s, respectively.

Fig. 3 Characteristics of shallow tremor in time and frequency domain.

(A) Example waveforms of shallow tremor recorded at OBS stations. A vertical component waveform (red) is superimposed on a horizontal component (black). Each trace is normalized by the maximum amplitude (Amp.) of a horizontal component. More example waveforms are shown in fig. S3. (B) Example of power spectra of shallow tremor, ordinary earthquake, and background noise observed at OBS station 8. The ordinary earthquake spectrum is calculated from the S-wave record and its coda using a horizontal channel. The selected event occurred within several kilometers of the tremor and had approximately the same amplitude. The time-window length is 10.24 s, starting 2 s before the S-wave arrival. The tremor spectrum and background noise are stacked and from the same horizontal channel, using a 10.24-s window over a 102.4-s time period. The instrument response was corrected. An example of deep tremor and background noise spectra from west Shikoku, Japan, is shown recorded at the land-based seismic station also shown in fig. S4. PSD, power spectral density.

The shallow tremor episode started on 28 May 2013, had major activity on 10 to 14 June and 16 to 21 June, and was essentially finished on 27 June 2013 (Fig. 2C). The time sequence of the tremor episode is similar to that of VLFEs. The source locations of the shallow tremor were estimated using the envelope correlation method (1), assuming an S-wave velocity of 3.5 km/s (17). The envelope functions are from combining the horizontal channels, and the tremor locations come from 2-min windows. Horizontal locations were constrained to less than ±5 km within the OBS network (fig. S5), and they roughly overlapped the VLFE focal area (Fig. 2) (17). Although this method has poor depth resolution, well-constrained shallow tremors were concentrated at an estimated depth of ~15 km (fig. S6), consistent with the depth of the plate interface in this region. The shallow tremor was distributed in a narrow belt parallel to the trench (Fig. 1), matching the up-dip limit of interplate small repeating earthquakes in this region (16), which reflects the up-dip limit of the zone where the ordinary interplate earthquakes occur. Background seismicity is also nonactive in the focal region of shallow tremor (fig. S1). Thus, the shallow tremor and ordinary interplate earthquakes appear to be spatially distinct.

We detected two distinct sequences of migrating tremor events, separated by an aseismic period lasting a few days. In the first sequence, events began off the east coast of Tanegashima Island in early June 2013. They then migrated generally northward with a slight change to the north-northwest around OBS station 8, reaching the area of OBS stations 6 and 7 on 12 to 14 June (fig. S7). In the second sequence, events resumed south of OBS station 8 on 16 June. They migrated north-northwestward to the vicinity of OBS station 7, then turned sharply east to the area around OBS station 9 (fig. S8). Compared with the location of subducted Kyushu-Palau Ridge (KPR), the first migration stopped within the ridge but the second migration overrode and reached the outside of the ridge’s edge.

In the first and second sequences, tremor migrated over a length of 90 km in 3 days and 1.5 days, respectively (Fig. 4). The tremor generally propagated along strike (before turning up-dip on June 18) at speeds of approximately 30 to 60 km/day, similar to or somewhat higher than the deep tremor at Nankai and Cascadia (18, 19). The migration front of first sequence typically shows a piecewise linear pattern, as in the deep ETS event (20). In addition, the very short duration in which tremor migrated in the reverse direction at speeds of 100 to 200 km/day occurred during the later part and after the propagation of first migration sequence (Fig. 4). This fast and backward migration is, to some extent, similar to the rapid tremor reversal (RTR) observed in deep ETS activity (18, 19).

Fig. 4 Spatiotemporal change of shallow tremor activity.

(Top) Hourly event counts (N, number of events) and (bottom) space-time plot of shallow tremor along line N–S in Fig. 1. Yellow polygons show the location and number of OBS stations. The gray shaded area indicates lower precision for the area outside the OBS network. The first and second migrations propagated north-northwestward at 30 to 60 km/day, and three RTR events (red shading) propagated in the reverse direction at 100 to 200 km/day.

The migration property of shallow tremor and its activity associated with VLFEs basically resembles the pattern of deep tremor during ETS events. Some theoretical studies have explained this tremor migration as the successive failure of small fault patches within the brittle-ductile transition zone, as SSEs propagating on the plate interface increase the stress around zones of high slip rate (21, 22). Therefore, the migration of shallow tremor may typify episodic SSEs in the near-trench shallow plate interface, analogous to the coupled phenomena of deep tremor, VLFEs, and episodic SSEs at the down-dip side of the locked zone (7). Although this study did not detect shallow SSEs associated with tremor and/or VLFE activity through geodetic evidence, the migration of shallow tremor is consistent with the migrating rupture front of SSE. These findings suggest that the both deeper and shallower sides of a subduction plate interface have similar frictional properties. The coupled phenomena of tremor, VLFEs, and SSEs probably occur in both plate interfaces.

The shallow tremor that we documented has some features in common with deep tremors, which suggests that there is a common generation process between shallow and deep tremors, as expected from processes like dehydration (1). The undulating surface structure of the KPR may play an important role in effective transportation of the clay minerals and fluid at depth and may thus be the key to the generation of shallow tremor in this area. On the other hand, there are some differences between both tremors in the details. The migration fronts of the two sequences of shallow tremor moved slightly faster than deep tremor sequences, which migrate at speeds ranging from a few to a few tens of kilometers per day (18, 19). The shallow tremor migrated over almost the same area after a very short interval (2 days), and the second migration was faster than the first migration (fig. S9). We lack reports of large-scale and along-strike multiple migrations during a deep tremor episode, although RTR and very fast tremor streaks that migrate along the dip direction at 30 to 200 km/hour (23) are known to occur repeatedly, even within an episode. Therefore, the properties of large-scale and along-strike migration of shallow tremor are similar but not identical to those of deep tremor. The rapid reverse propagation is also not the same as the deep RTR reported for the Cascadia deep tremor that occurs during the linear propagation of the overall tremor front (19). Rheological differences between shallow and deep tremor zones may explain the difference in the behavior between both tremors.

The subducted KPR is thought to be a segment boundary of coseismic slip (24). Meanwhile, the migration episodes of tremor propagated obliquely across the KPR, which suggests that, this ridge did not work as a strong segment boundary for the slow rupture. This behavior is consistent with the argument that subducted seamounts provide favorable conditions for aseismic creep and small earthquakes but unfavorable conditions for generation and propagation of large ruptures (25).

Shallow tremor stopped migrating further in the northward direction and turned sharply eastward around OBS station 7. This area is located in the up-dip side where large [moment magnitude (Mw) 6.5 to 7.5] megathrust events occur (26, 27). Therefore, the locking or creeping state of deeper megathrust changed along-strike, and the migration to the northward direction was blocked by the locked zone. In addition, the migration did not finally reach to the up-dip side of the rupture area of the Mw 7.5 megathrust event. These observations suggest that the migration of shallow tremor occurred in a plate interface, which is the shallower extension of the creeping megathrust. In other words, the shallow tremor migration may be a response to the creeping of the deeper megathrust. Whether it can also occur up-dip of a locked megathrust such as at Nankai and Cascadia awaits future observations.

As the occurrence of the megathrust event approaches, the state of interplate coupling will become progressively weakened from the surrounding area, even despite the locked megathrust. As a result, a change of spatiotemporal activity of shallow slow earthquakes is expected in the up-dip side of the locked zone, similar to the case for deep slow earthquakes, as predicted by some computer simulation studies (28).

The worst-case scenario is for shallow slow earthquakes to precede a large seismic rupture, such as the slow slip observed just before the Tohoku-Oki earthquake (29). Therefore, monitoring the spatiotemporal changes of shallow slow earthquakes is important for evaluating the slip of the shallow plate interface offshore. This aids assessment of the potential hazard of tsunamigenic earthquakes. Long-term ocean-bottom observations of many subduction zones and geodetic observations to confirm the suspected shallow SSEs for this region will further clarify the relationship between shallow slow earthquakes and the frictional behavior of megathrusts.

Supplementary Materials

Materials and Methods

Figs. S1 to S9

References (3135)

Data S1

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank the crew of the T/S Nagasaki-maru (Faculty of Fisheries, Nagasaki University) for their skillful work and K. Creager, H. Houston, M. Vallee, K. Goto, Y. Yamanaka, K. Mochizuki, the members of SEVO of Kyushu University, and the ERI of The University of Tokyo for valuable discussions and comments. We also thank the anonymous reviewers for providing suggestions and comments that improved the manuscript. This study was supported by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan under its Observation and Research Program for Prediction of Earthquakes and Volcanic Eruptions. Y.Y. received support as a Japan Society for the Promotion of Science research fellow (grant 24-3726). Shallow tremor catalog data are available in the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article