Volume 46, Issue 15 p. 8763-8771
Research Letter
Free Access

Migration of Very Long Period Seismicity at Aso Volcano, Japan, Associated With the 2016 Kumamoto Earthquake

Andri Hendriyana

Andri Hendriyana

International Institute for Carbon-Neutral Energy Research, Kyushu University, Fukuoka, Japan

Exploration and Engineering Seismology Research Group, Bandung Institute of Technology, Bandung, Indonesia

Search for more papers by this author
Takeshi Tsuji

Corresponding Author

Takeshi Tsuji

International Institute for Carbon-Neutral Energy Research, Kyushu University, Fukuoka, Japan

Department of Earth Resources Engineering, Kyushu University, Fukuoka, Japan

Correspondence to: T. Tsuji,

[email protected]

Search for more papers by this author
First published: 29 July 2019
Citations: 11

Abstract

Earthquakes are known to precede volcanic activity, including long-period or very long period (VLP) volcanic seismicity. However, the relationships among earthquakes, VLP seismicity, and volcanic eruptions are not well understood. Here we present the locations of VLP seismicity at Aso volcano, Japan, between January 2015 and December 2016, a period that includes the Mw 7.0 Kumamoto earthquake and phreatomagmatic eruptions. By using a differential-time backprojection method developed in this study to accurately locate VLP events, we clearly identified two distinct VLP clusters. Whereas the eastern cluster was active during eruptions, the western cluster displayed intense VLP seismicity only for a few months after the earthquake. The western cluster may be associated with opening of new fractures during the earthquake. This study explores the mechanisms that can relate earthquake to volcanic activities and provides a new approach to analyze the dynamic behaviors inside volcanoes that may yield useful information for hazard evaluation.

Key Points

  • Backprojection technique based on differential-time and phase weighted stacking is used to map clusters of very long period seismicity
  • Clusters of very long period seismicity indicate clear relationship between distant earthquake and volcanic activities
  • The 2016 Kumamoto earthquake suddenly induced very long period seismicity at west of Aso Caldera lasting for five months

Plain Language Summary

The relationship between earthquake and volcanic seismicity is studied by analyzing continuous seismic data recorded between January 2015 and December 2016, a period that includes the Mw 7.0 Kumamoto earthquake and volcanic eruptions at Aso. This study focuses on analyzing very long period (VLP) seismic event with a period of 15 s that can be explained by pressure fluctuation within hydrothermal systems. The intense activity of long-period event could be an indication of impending eruptions. By using our developed localization method, we revealed two clusters of VLP sources, namely, eastern and western clusters. The eastern cluster was always active during an active period of Aso volcano. However, the western cluster showed intense VLP seismicity only for a few months after the earthquake. The western cluster could be associated with opening of new cracks triggered by the Kumamoto earthquake. This study suggests a strong relationship between distant large earthquake with volcanic seismicity that could be useful for disaster mitigation. Moreover, monitoring of temporal and spatial evolution of VLP sources may be considered as a tool to study the response of volcanic activity to ground shaking from large earthquake.

1 Introduction

Many large, densely populated cities are situated near active volcanoes prone to explosive eruptions. For such cities, eruption forecasting is essential to avoid or minimize disasters. Understanding the conditions inside volcanoes that presage eruptions can contribute to reliable eruption forecasting. Volcanoes are commonly activated after remote earthquakes (Jousset et al., 2013; Surono et al., 2012; Walter et al., 2007; Walter & Amelung, 2007), raising the threat of combined seismic and volcanic disasters. Therefore, it is important to explore the relationship between earthquakes and eruptions, which may also further our understanding of the geological processes occurring before eruptions.

The 2016 Kumamoto earthquake sequence began on 14 April with a Mw 6.2 event on the Hinagu fault (Uchide et al., 2016; Yagi et al., 2016) and culminated on 16 April in a Mw 7.0 main shock on the Futagawa fault that caused more than 200 fatalities (Figure 1). The main source mechanism of these events was right-lateral motion, and their rupture depths were shallower than 10 km. The maximum displacement along the fault plane was ~6 m, and the coseismic rupture terminated just west of Aso caldera (Yagi et al., 2016; Figure 1a). Some studies (Miyakawa et al., 2016) have suggested that the rupture halted at Aso because the rocks there were ductile owing to the high temperatures around the magma body.

Details are in the caption following the image
Map of the central part of Kyushu Island, Japan, including Aso volcano. (a) The inset shows the location of the main map. The epicentral location and centroid moment tensor solution of the 2016 Kumamoto earthquake main shock (Mw 7.0) are shown from the Global CMT project (Dziewonski et al., 1981; Ekstrom et al., 2012). White dots are epicenters of foreshocks and aftershocks. Inverted triangles are locations of the F-net and V-net seismic network stations used in this study, and the red triangle marks Nakadake crater. The black rectangle indicates the area shown in (b). (b) Blue and red small dots are epicenters of VLP sources. Red dots indicate VLP events occurred between 16 April (main shock) and 7 September 2016. The black rectangle marks the area shown in Figure 3c, and the orange oval marks the inferred location of the magma chamber (Sudo & Kong, 2001).

Aso volcano, one of the largest active volcanoes in the world, erupted approximately 7 hr after the main shock and again on 1 May 2016 (JMA, 2016), and large eruptions occurred on 7 and 8 October 2016 (JMA, 2016). During the October eruptions, the volcano ejected cloud rich in SO2 gas that reached an altitude of ~12 km (Sato et al., 2018). A causal relationship has been proposed between these eruptions and the Kumamoto earthquake (Ozawa et al., 2016; Uchide et al., 2016), although Aso produces frequent eruptions. Indeed, magmatic eruptions occurred at Aso from November 2014 to May 2015, a year before the Kumamoto earthquake (Minami et al., 2018). The relationship between earthquakes and eruptions has been studied worldwide at other volcanoes (Nishimura, 2017). Pressure changes leading to eruptions can be triggered by transient stresses brought by passing waves from distant large earthquake (Hill et al., 2002). Stress changes from a remote large earthquake (Mw 6.3) appear to have triggered the 2006 eruption of Merapi volcano, Indonesia (Walter et al., 2007). However, clear evidence of the relationship between Kumamoto earthquake and Aso eruption has not been reported.

Volcanic eruptions result from elevated pressures within a shallow magma reservoir due to the ascent of fluids from a deep-seated magma chamber (Shapiro et al., 2017; Surono et al., 2012; Walter, 2007). Phreatic eruptions occur when high-temperature volcanic fluids (gases or magma) come into contact with and vaporize shallow groundwater in a closed system beneath low-permeability rocks. These vaporized fluids fill cracks, which in turn increases the formation pressure. This physical process leads to disturbances in hydrothermal systems and intense seismic activity. Pressure fluctuations within a hydrothermal system can trigger and intensify volcanic long-period (LP) earthquakes in a pattern known to precede volcanic eruptions (Chouet, 1996; Jolly et al., 2012; Kaneshima et al., 1996; Lokmer et al., 2008; Neuberg, 2000; Shapiro et al., 2017). Swarms of intense LP signals with a dominant period of ~0.5 s were a factor in issuing an advanced warning of explosions before an eruption of Redoubt volcano, Alaska (Chouet et al., 1994).

At Aso, the interaction between volcanic fluids and shallow aquifers is manifested as very long periods (VLPs) tremor with periods of about 15, 7.5, 5, and 3 s (Yamamoto et al., 1999). VLP seismicity with a period of 15 s occurs at Aso during both periods of quiescence and unrest (Kaneshima et al., 1996; Kawakatsu et al., 2000). These signals originate 1.1–1.5 km below Nakadake crater and are associated with the response of resonating fluid-filled cracks to pressure fluctuations within aquifers or hydrothermal systems (Kaneshima et al., 1996; Kawakatsu et al., 2000; Legrand et al., 2000). The continuous nature of VLP seismicity at Aso can be explained by the stable supplies of heat from the deep magma reservoir and water from the crater lake (Kaneshima et al., 1996).

LP or VLP seismicity can be used as an indicator of impending eruptions (Budi-Santoso et al., 2013; Burlini et al., 2007; Jousset et al., 2013; Surono et al., 2012). Thus, close monitoring of LP or VLP earthquakes may be useful for detecting upcoming volcanic eruptions. Previous studies have investigated the locations and source mechanisms of VLP seismic events at Aso (Kaneshima et al., 1996; Kawakatsu et al., 2000; Legrand et al., 2000; Yamamoto et al., 1999). However, the temporal and spatial variations of VLP sources during magmatic and phreatomagmatic eruptions as well as during the 2016 Kumamoto earthquake are poorly known.

Backprojection technique has demonstrated good performance in localizing crustal earthquakes (Hendriyana et al., 2018) as well as low-frequency volcanic earthquakes (Haney, 2013). While conventional backprojection techniques are applied directly to waveforms from arrays of stations, we applied it to the cross-correlation functions among the station pairs. The location of a seismic event was estimated from the coherency section calculated based on differential time between pair-wise stations.

The paper presents VLP source locations accurately determined by our developed differential-time back projection technique. We interpret that the located VLP sources clearly emphasize a strong correlation between the 2016 Kumamoto earthquake and pressurization within hydrothermal systems as indicated by intense VLP seismicity.

2 Data and Methods

We analyzed VLP seismicity data from January 2015 to December 2016, a period that included large Kumamoto earthquakes and Aso eruptions. VLP signals were extracted from velocity seismograms recorded by five broadband seismometers of F-net and V-net (Figure 1a). Since we used VLP events with period of 15 s, then the band-pass filter between 0.05 and 0.1 Hz was applied. Our analysis of VLP seismic data included detection and collection of VLP signals from continuous seismograms, and determination of their source locations. Because repeated VLP events with highly similar source mechanisms are likely to have highly similar waveforms, we used methods based on cross correlations to perform automated detection of VLP events (Shapiro et al., 2017). A localization method based on differential times made use of the similarity of waveforms from different seismometers. Here we discuss the methods of detection, collection, and localization of VLP seismic events.

2.1 Detection of Initial VLP Events

Recognition of VLP events in a continuous seismic data set relies on two sequential steps: (1) detection of potential VLP events and (2) classification of VLP events based on a cross-correlation matrix (Green & Neuberg, 2006; Jousset et al., 2013). Detection of VLP events began by transforming seismograms into a characteristic function so as to highlight the presence of VLP events. The characteristic function we chose was based on the ratio between the short-term amplitude average (STA) and the long-term amplitude average (LTA; Allen, 1978) owing to the assumption that STA/LTA functions are comparable to the signal-to-noise ratios of seismic arrivals. Since VLP signals are mainly consisted of Rayleigh wave (Kawakatsu et al., 1994), we applied the STA/LTA method to the vertical component of velocity seismograms recorded at station N.ASHV, the closest station to Nakadake crater, to ensure signals of high quality (Figure 1a). By considering the frequency of the filtered signals (0.05–0.1 Hz), window sizes of 10 and 100 s were used to determine STA and LTA, respectively. This method detected 2,201 potential VLP events from January to November 2015.

The detected VLP events were classified on the basis of waveform similarity, as quantified by waveform cross correlations for every possible pair of VLP events. The maximum values of cross-correlation functions were used to construct the cross-correlation matrix, CCm (Figure S1 in the supporting information). This matrix displays clusters of seismic events that may represent families of VLP events. An average value of the CCm element, CCave, was computed for each row, and all VLP events associated with the highest value of CCave were considered as one VLP family. We considered only waveform pairs with CCm values higher than 0.9, leaving around 1,660 waveforms to be considered as multiplet members of this VLP family. These waveforms were aligned and stacked, and the resulting waveform was used as the template waveform (Figure S2 in the supporting information). The same procedure was followed to derive template waveforms for the other two horizontal components of the seismogram from station N.ASHV and the seismograms of stations N.TKDF, N.TMCF, N.INNF, and N.ASIV (Figure S3 in the supporting information). These template waveforms were used as filters to compile the final set of VLP events (Frank et al., 2014).

2.2 Collection of VLP Events by Template-Matching Filter

A template-matching filter (e.g., Frank et al., 2014;Figure S4 in the supporting information) was used to extract VLP events from the three-component seismograms from stations near Aso volcano. Normalized cross-correlation values between template waveforms τ(t) and seismic recordings s(t) were calculated from
(1)
where Δt is the sampling rate, Ns is the number of seismic stations and Nt is the number of sample within a cross-correlation window. We selected VLP seismic events by defining a threshold cross-correlation value of 8 (from the maximum value of 15). The records from January to November 2015 yielded 9,738 VLP events, and the records for all of 2016 yielded 6,351 VLP events.

2.3 Localization of VLP Events

We determined VLP source locations by the differential-time backprojection technique. The similarity of VLP waveforms observed at different stations was used to improve the accuracy of the resulting locations. For this purpose, we designed a differential-time backprojection technique in which the VLP source location that was estimated on the basis of coherency values calculated from all potential source locations. Since main component of VLP signals is Rayleigh wave (Kawakatsu et al., 1994), differential times were attributed to lateral propagation delays from source to receivers and were estimated by waveform cross correlation. With an assumption that the source mechanisms of VLP events were dominated by isotropic component (Kawakatsu et al., 2000; Legrand et al., 2000), seismic stations with different azimuth from the source will register waveforms having similar phases (Figure S3 in the supporting information). This assumption allows us to accurately predict differential times by using waveform cross correlation.

Potential source locations were defined using a two-dimensional grid of points spaced 0.0004° (<50 m) apart. The area surrounding the VLP source location appeared as an area of high coherency, as measured by the value of semblance defined as follows. If the cross-correlation function between two vertical-component seismograms, s(t), recorded at stations i and j is defined as ρij(δ) = si(t) ⊗ sj(t), then the semblance at each grid point, , is represented by
(2)
where M = Ns(Ns+1)/2. The maximum cross-correlation value incorporates the time difference (δ) between the arrivals of the VLP event at stations i and j. To increase the spatial resolution of the coherency (semblance) cube, we added a phase-weighted stacking function (Schimmel & Paulssen, 1997) to the formulation of semblance in which the phase of the cross-correlation function is defined by ϕ = tan−1(ℏ{ρ}/ρ), where represents a Hilbert transform.

By considering the wavelength of the VLP signals and the coverage of the seismic network, the heterogeneity of the subsurface model can be ignored, and hence, a homogeneous velocity model is sufficient. After testing a range of velocities from 2,500 to 3,500 m/s, we selected a homogeneous velocity model of 3,000 m/s, which resulted in the highest maximum coherencies and best focused images. A previous study at Aso used a velocity of 3,500 m/s (Sandanbata et al., 2015). An example of the resulting coherency sections for single event is shown in Figure 2. The locations of VLP seismic events and the corresponding uncertainties were calculated by the method described by Anikiev et al. (2014).

Details are in the caption following the image
Estimation of a VLP source location on the 2-D coherency section. VLP source location (blue dot) and its uncertainties (white lines) are estimated from the probability density function (Anikiev et al., 2014).

3 Results

3.1 Detection and Intensity of VLP Seismicity

We detected approximately 18,571 reliable VLP events with periods of 15 s in this data set by using a template-matching filter search. These VLP events and their RMS amplitudes were analyzed to reveal the interplay among VLP seismicity, magma inflation, and eruptions (Syahbana et al., 2014). The intensity of VLP seismicity (as indicated by the daily number of events) could be used as a proxy to monitor pressure changes within the hydrothermal systems of Aso. In general, this measure of intensity (Figure 3a) agreed well with the RMS amplitude of VLP waveforms (Figure 3b), an indication that both of these results are derived from the same physical process in a shallow reservoir. The waveform amplitudes represent magnitude (Ishimoto & Iida, 1939) or the amount of energy released by the event (Sandanbata et al., 2015). Peaks of VLP intensity and RMS amplitude were preceded by episodes of magma inflation during March and November 2015 and September 2016 (Figures 3a and 3b).

Details are in the caption following the image
Temporal variation of VLP events in the study period. (a) Daily VLP number (black dots), 14-day running average of VLP numbers (red curve), and magma inflation episodes (red vertical arrows). Red dashed arrow indicates magma inflation period from July 2016. (b) RMS amplitudes of all detected VLP events (black points), seven-day running average of daily average VLP amplitudes (red curve), occurrence of the Kumamoto main shock (red vertical line), and Aso eruptions (blue arrows). (c) Spatial distribution of VLP sources. The map area is outlined in Figure 1. Black and red dots indicate the VLP source locations. Red dots indicate VLP events occurred between 16 April (main shock) and 7 September 2016. (d) Temporal evolution of VLP sources. The VLP source migrated after the 2016 Kumamoto earthquake. We considered only those VLP location with longitude uncertainties less than 0.013° (see Figure S5 in the supporting information). Aso eruptions and magma inflation episodes are from JMA.

3.2 Localization Results and Migration of VLP Seismicity

Using our developed differential-time backprojection method, we resolved the VLP source distribution into two distinct clusters (Figures 1b, 3c, and 3d), one southeast of Nakadake crater (the eastern cluster) and the other, southwest of Nakadake (the western cluster). During most of our 2015–2016 observation period, activity was almost entirely confined to the eastern cluster (black dots in Figures 3c and 3d). The eastern source locations are same as those identified in previous works (Sandanbata et al., 2015). After the 2016 Kumamoto earthquake, however, VLP activity shifted abruptly to the western cluster for about five months (red dots in Figure 3d), suggesting that static and dynamic stresses associated with the earthquake influenced the magmatic plumbing system of Aso. In September 2016, one month before the largest eruption, VLP events migrated to the original eastern cluster. After the large eruption on 8 October, VLP seismicity temporarily stopped. These observations clearly show that VLP events are influenced by the earthquake and are related to the volcanic eruptions.

4 Discussion and Conclusion

The pressure field as indicated by intense VLP seismicity (Figure 3) changed with the ascent of hot volcanic fluids into the shallow conduit (Shapiro et al., 2017). During magma cooling, volcanic gases exsolved from the melt generated bubbles that gradually increased the pressure within the conduit. As cracks began to open due to high fluid pressure, volcanic gases and other magmatic materials started to migrate to the surface, eventually causing eruptions. Afterward, the pressure decreased when a crack became established as a conduit between the hydrothermal reservoir and the surface, which in turn led to the lowering of both VLP intensity and RMS amplitude. This interpretation is supported by the observation of magma deflation at the end of March 2015 (JMA, 2015), after which VLP intensity and RMS amplitude remained low until the next episode of magma ascent during October 2015. These observations suggest that VLP seismicity is directly related to the pressure variation associated with the magma activity.

We interpret the migration of VLP events after the earthquake, as a response to permeability enhancement (Manga et al., 2012) or fractures opening owing to the extension associated with the Kumamoto earthquake (Figure 4). The Kumamoto earthquake produced a large amount of rupture in the area just west of Aso (Uchide et al., 2016;Figure 1). Because the western part of Aso volcano is located in a dilatational area associated with strike-slip faulting (Kato et al., 2016), strain above the magma chamber could be released (Figure 4b). The extensional stress in the western side of crater perturbed by the earthquake could be large enough to have opened preexisting vertical cracks and allowed magmatic fluid to ascend through them (Ozawa et al., 2016). Indeed, the static stress change was estimated as 3.5 MPa (Ozawa et al., 2016) and was higher than the peak stress change that triggered seismicity (Hill & Prejean, 2007). The generation of a crack due to strain release was also indicated by the seismic velocity reduction that occurred after the earthquake (Nimiya et al., 2017). The fractures opened during the earthquake could work as a new fluid pathway beneath the area of the western cluster, which led to VLP events there after the 2016 earthquake. In addition to the static stress changes, we estimated dynamic stress changes in the western cluster as ~3.08 MPa (Text S3 in the supporting information). Because the static and dynamic stresses changes in western cluster are similar, both dynamic and static stresses could influence magma plumbing system (i.e., VLP cluster migration).

Details are in the caption following the image
Schematic diagram of the magma plumbing system at Aso volcano. The magma plumbing system consists of a shallow magma chamber (Sudo & Kong, 2001), a sill (Tsutsui & Sudo, 2004), a magma conduit (Hata et al., 2016), and shallow hydrothermal reservoirs (Kaneshima et al., 1996; Kawakatsu et al., 2000; Yamamoto et al., 1999). The two clusters of VLP seismicity were revealed in this study. (a) The eastern VLP cluster is part of an established magma conduit. (b) The western VLP cluster was triggered by the Kumamoto earthquake. (c) The cracks feeding the western cluster closed, and fluid migration resumed in the eastern conduit by the time of the October eruptions.

Furthermore, magma chamber volume has been shown to increase (Ozawa et al., 2016), owing to dynamic stress transfer by the propagation of seismic waves of the Kumamoto earthquake (Manga & Brodsky, 2006). Magma inflation has been detected at the surface since July 2016 (red dashed line in Figure 3d). Bubbles might have been dislodged from base and wall of magma chamber and rose to the roof of chamber leading to advective overpressure (Hill et al., 2002;Figure 4b). The increases in VLP events in western cluster in June and July 2016 (Figure 3a) could be an indication that rising hot magmatic fluid perturbed pressure state west of the caldera. Exsolution of gases during magma ascent may be the reason that large amounts of SO2 (about 15,000 t) were discharged during the 8 October 2016 eruption. The evidence thus supports our interpretation that volcanic fluids were able to easily infiltrate the newly opened cracks and induce VLP seismicity.

In September 2016, one month before the largest eruption, the fractures in the western cluster could be closed. The temporally enhanced permeability could return to preearthquake condition (Manga et al., 2012). Recovery of permeability to its original value may cause VLP event migration to the eastern cluster (Figure 4c). In addition, the earthquakes related to upward magma movement started to occur in mid-September, when magma inflation was detected beneath the western cluster (JMA, 2016). The magma upward motion may have increased the stress field in the overlying rocks and caused the previously opened cracks to close. Furthermore, the magma inflation after July 2016 that closed the open fractures in western cluster could increase fluid pressure, leading to the October eruptions.

The activity of VLP seismicity decreased after the 8 October eruption, indicating that the development of the shallow conduit had reduced the pressure gradient between the hydrothermal systems and the surface. However, the VLP source locations after the eruption are not to be changed (still in eastern cluster). This suggests that the earthquake generating cracks has a great potential to influence the magmatic plumbing system inside volcanoes by opening new cracks in the western cluster (Figure 4).

By offering accurate VLP source locations, our study explores the role of VLP seismicity in the interactive relationship between earthquakes and volcanic eruptions at Aso. The new information derived from our monitoring approach to VLP seismicity could reveal new details of the dynamic behavior within Aso and other volcanoes after the earthquake, and provide useful information for disaster prevention.

Acknowledgments

We used the F-net and V-net data obtained from the NIED server. Figures were generated by using Generic Mapping Tools (Wessel & Smith, 1998). We gratefully acknowledge the support of I2CNER, sponsored by the World Premier International Research Centre Initiative, MEXT, Japan. We thank M. Haney, P. Jousset, and Associate Editor M. Denolle for a careful review and helpful suggestions. T.T. was supported by the Japan Society for the Promotion of Science (JSPS) through a Grant-in-Aid for Scientific Research on Innovative Areas (JP17H05318). All authors analyzed the data, interpreted the results, and described the manuscript. A.H. mainly performed data analysis. T.T. conceived this study. The authors declare no competing interests. The F-net and V-net data can be downloaded from the NIED server.