Introduction

The M JMA 7.3 (Mw 7.0) Kumamoto earthquake occurred at 1:25 JST (Japan Standard Time) on April 16, 2016, which followed the nearby M JMA 6.5 (Mw 6.2) earthquake at 21:26 JST on April 14, 2016. The dextral strike-slip earthquake rupture propagated mainly in an ENE direction from the hypocenter to the west of Aso volcano along the Futagawa fault (e.g., Asano and Iwata 2016; Kobayashi et al. 2016) (Fig. 1). One prominent feature of this earthquake sequence is the spatial distribution of the events: These are not limited to aftershocks in areas around the rupture zone, but include triggered seismic events located 50–100 km from the mainshock (Shimizu et al. 2016). The active seismicity can be divided into three regions: Seis. 1, around the main faults; Seis. 2, the northern part of Aso volcano; and Seis. 3, the region around the three volcanoes, Yufu, Tsurumi, and Garan (Fig. 1). Since the earthquakes around Seis. 2 and Seis. 3 began immediately after the Kumamoto earthquake, earthquakes in these regions are considered to be triggered by the mainshock of Seis. 1. Subsequent earthquakes not only were small earthquakes, but also included moderate earthquakes (>M JMA 3.5) that totaled 230 events by May 8, 2016. The concern for future earthquakes has been derived from the fact that the NE–SW-trending line along which the earthquakes occur corresponds to the possible western extension of the Median Tectonic Line (MTL), which is the longest and most active arc-parallel, right-lateral, strike-slip fault system in the Japan arc (Fig. 1). The oblique subduction of Philippine Sea Plate (Seno et al. 1993) induces a shear stress in the vicinity of the MTL and separates the fore-arc sliver from the crust of the arc at depth (Kamata and Kodama 1994; Miyazaki and Heki 2001; Tabei et al. 2002). Along the western extension of the MTL in Kyushu, the shear stress is found to be partly released by deformation (Nishimura and Hashimoto 2006; Wallace et al. 2009; Matsumoto et al. 2015). In 1975, M JMA 6.1 and M JMA 6.4 earthquakes occurred within 3 months in the region of Seis. 2, and between Seis. 2 and Seis. 3, respectively (Yamashina and Murai 1975). Along the MTL, three M7 class earthquakes occurred within 4 days in 1596 (Kanaori et al. 1994; Toda et al. 2015). Therefore, it is reasonable to be concerned about the occurrence of large earthquakes around the MTL and its western extension.

Fig. 1
figure 1

Broadband magnetotelluric (MT) and telluric observation sites with hypocenters one month after the M JMA 6.5 earthquake of 21:25 JST, April 14, 2016. The hypocenters are relocated by the manual reading of P-wave arrival times (Shimizu et al. 2016). Inset shows the location of studied area (solid rectangle), active volcanoes, plate boundary, and Median Tectonic Line (MTL). Seis. 1, Seis. 2, and Seis. 3 represent the seismicity noted in the text. The three stars represent hypocenters of events with M JMA > 6.0. The largest star represents the hypocenter of the Kumamoto earthquake (32.7552°N 130.7661°E, 12.9 km depth), which occurred at 01:25 JST, April 16, 2016

Full size image

In addition to the shear stress along the western extension of the MTL, the island of Kyushu is considered to be influenced by tectonics stress associated with back-arc spreading (e.g., Seno 1999) (Fig. 1). Geomagnetic depth sounding research supports a region of increased electrical conductivity in the mantle below the back-arc side of Kyushu (Handa et al. 1992; Shimoizumi et al. 1997). Furthermore, the series of Kumamoto earthquakes, including those that were triggered by the initial series, occurred in a graben structure (Matsumoto 1979; Kamata 1989; Handa 2005) and around the active volcanoes. Such complex tectonic settings suggest complex subsurface structures that may be related to the characteristic seismicity. Because electrical resistivity is sensitive to the presence of fluids, and subsequently the elasticity of the media, it is important to investigate how resistivity structures relate to earthquake generation (Ogawa et al. 2001; Fujinawa et al. 2002; Goto et al. 2005; Guerer and Bayrak 2007; Wannamaker et al. 2009; Yoshimura et al. 2009; Ichihara et al. 2011, 2014, 2016; Ogawa et al. 2014; Kaya et al. 2013). To investigate the relationship between earthquakes and electrical resistivity structure, we gathered and analyzed the broadband (typically 0.003–10,000 s) magnetotelluric (MT) and telluric data, which resolve the resistivity structure from the surface to the depth of the upper mantle.

Broadband magnetotelluric data

We conducted the MT and telluric data surveys during 2014–2015 in the vicinity of Aso caldera, in the region of triggered seismicity (Seis. 2 and Seis. 3), and to the south of the city of Oita. In addition, we used previously published MT and telluric data from the region of the main shock of the 2016 Kumamoto earthquake (Asaue et al. 2004, 2007, 2012) and around Aso volcano (Takakura et al. 2000; Asaue et al. 2006; Hata et al. 2016). The number of sites used in this study amounted to 247, including 94 unpublished new data.

For the 2015 MT survey around Aso volcano, the MT data were measured by the Phoenix MTU5 systems (telluric and geomagnetic field observations). The MT response functions were calculated using the SSMT2000 program (Phoenix Geophysics Ltd). Typically, recording duration was 2–3 nights. We also employed remote-reference processing (Gamble et al. 1979) using MT data recorded at the Esashi Magnetic Observatory, which is located about 1000 km northeast of Aso volcano.

For the 2014–2015 MT surveys around Seis. 2 and Seis. 3, and south of Oita, the MT data were measured with a Metronix ADU07e system (telluric and geomagnetic field observations) and the NT System Design ELOG1K (telluric only observations). Typically, recording duration was 10 days. MT response functions were calculated using a robust estimation code (Chave and Thomson 2004). At the telluric observation sites, geomagnetic data from the nearest sites were used for calculations. In all calculations, notch filtering was applied to the time series data to reduce anthropogenic 60 Hz noise and its odd-order overtones (Aizawa et al. 2013). We employed remote-reference processing (Gamble et al. 1979) for periods <10 s, using MT data recorded at other MT sites. For the periods >10 s, the 1-Hz-sampled geomagnetic data recorded at the Kakioka Magnetic Observatory (located about 1000 km east-northeast of Kyushu) were used for remote-reference processing. Using these approaches, we obtained MT response functions across a broad (0.005–3000 s) range of periods. The periods of the MT response functions vary slightly among datasets due to differences in sampling frequencies. We interpolated the MT response functions and errors in the frequency domain using a cubic spline function. The MT response functions were then defined for specific frequencies.

Resistivity structure determined by one-dimensional analysis

Recent development of 3-D MT inversion codes by finite difference methods (e.g., Siripunvaraporn and Egbert 2009; Kelbert et al. 2014) allows us to deduce three-dimensional (3-D) resistivity structure. In our dataset, the sites are mainly located along five lines across the Futagawa and Hinagu faults, with the overall region located along an elongated NE–SW region (Fig. 1). Applying 3-D inversion codes to such uneven site locations requires the construction of a huge horizontal mesh and subsequently has extensive memory and computational time requirements, even with a high-end workstation. In this study, as an alternative, we have adapted a one-dimensional (1-D) inversion routine for the data at each site. The apparent resistivity and the phase of the sum of the squared elements (ssq) invariant impedance (Szarka and Menvielle 1997; Rung-Arunwan et al. 2016) are inverted with Occam’s algorithm (Constable et al. 1987). The ssq impedance (Zssq) is defined as,

$$ Z{\text{ssq}} = \sqrt {\frac{{Zxx^{2} + Zxy^{2} + Zyx^{2} + Zyy^{2} }}{2}} , $$

where Zxx, Zxy, Zyx, and Zyy are the components of the impedance tensor. Commonly used determinant impedances are generally biased downward by the presence of galvanic distortion, while ssq impedances are robust with respect to distortion and are therefore suitable for obtaining a first-order approximation of the regional structure (Rung-Arunwan et al. 2016).

In the 1-D inversion, we assigned error of ±10% to each ssq impedance (equivalent to ±0.0434 in log apparent resistivity and ±2.85° in phase), with the exception of the dead-band data from 6.4 to 25.6 s, which were assigned larger error of ±30%. In addition, outliers from smooth sounding curves were judged visually and were assigned errors larger than ±30%. Using these procedures, we estimated the smoothest resistivity structure in which the model response fit the data to an RMS tolerance of 1.0. Figures 2 and 3 show the comparison of the apparent resistivity and phase maps of the observed data (obs) with the calculated response (clc). Overall, the observed data are well explained by the estimated resistivity structure.

Fig. 2
figure 2

Apparent resistivity map for comparison of observed data (obs) with calculated response (clc) from the resistivity structure (Fig. 4). Apparent resistivity was calculated from the rotationally invariant ssq impedances in Eq. (1) (Rung-Arunwan et al. 2016). The white line represents the rim of Aso caldera. Observational data with errors of >0.3 in log10Ωm are not shown. The notations are the same as in Fig. 1

Full size image
Fig. 3
figure 3

Phase map for comparison of observed data (obs) with calculated response (clc). Observational data with errors of >20° are not shown. Other notations are the same as in Fig. 2

Full size image

Figure 4 shows the estimated resistivity structure. Strictly speaking, it is difficult to evaluate how well 1-D analysis using ssq impedance approximates regional structure. In this study, we calculated the skew angle (β) of the magnetotelluric phase tensor (Caldwell et al. 2004) (Fig. 5) to check for the presence of strong 3-D features produced by the true resistivity structure: Large |β| indicates the presence of a 3-D resistivity structure around a site. Although large |β| values are present locally, large parts of the studied area show |β| < 10° to periods of 25.6 s, which correspond to a 12.5 km skin depth in a 30-Ωm half-space. At longer periods, to 1024 s, areas of |β| > 10° are present, especially at the north flank of Kuju volcano. It should be noted here that the zones of large |β| approximately correspond to zones where the estimated resistivity significantly changes horizontally. This implies that the 1-D analysis might approximate the 3-D structure, even in a zone of large |β|. Full 3-D analysis will be the subject of future work.

Fig. 4
figure 4

Resistivity structure estimated from the 1-D inversion. Other notations are the same as in Fig. 1. These maps were produced by interpolating the resistivities of 1-D structure at each site

Full size image
Fig. 5
figure 5

Skew angle (β) of magnetotelluric phase tensor (Caldwell et al. 2004). Large |β| indicates 3-D feature around a site

Full size image

In a broad sense, obtained resistivity structure (Fig. 4) shows a structural boundary across the Hinagu and Futagawa faults at depths of 1 and 4 km (i.e., the northwestern parts of the faults are conductive, while the southeastern parts are resistive), as stated in Asaue et al. (2012). A similar feature is also found between Kuju volcano and the city of Oita, where a NE–SW-trending structural boundary exists. Between such structural boundaries, Aso volcano exists with the local conductor. At depths of 8 and 12 km, the zone around the Futagawa and Hinagu faults (the Seis. 1 zone) becomes approximately resistive, although the location of hypocenter of the Kumamoto earthquake was relatively conductive. The Aso and Kuju volcanoes are imaged as conductive zones. The Yufu, Tsurumi, and Garan volcanoes do not show a dominant conductor beneath them, while their southern part is imaged as a conductor. At a depth of 20 km, the northwestern parts of the Futagawa and Hinagu faults gradually become conductive again. This deep conductive zone is located beneath the region of Seis. 1, which extends to ~16 km depth. The conductive zone beneath Aso volcano continues to depth, whereas the one beneath Kuju volcano moves to the west.

Discussion

To investigate the relationship between the earthquakes and the resistivity structure in the study area, the hypocenters, which were relocated by manual readings of P-wave arrival times (Shimizu et al. 2016), are plotted on the resistivity sections at depths of 8 and 12 km (Fig. 6). Overall, the aftershocks (Seis. 1) were distributed mainly in the resistive zone (100–1000 Ωm) at depths of 4–16 km beneath which the relatively conductive zones exist, especially to the northwest of Hinagu and Futagawa faults (Figs. 4, 6). The triggered earthquakes of Seis. 2 also occur in the resistive zone or the resistive-conductive transition zone. In 1975, an M JMA 6.1 earthquake occurred at approximately the same location, although its hypocentral depth is poorly determined (Yamashina and Murai 1975) (Fig. 6). In the region of Seis. 3, the triggered earthquakes also occurred in the relatively resistive zones (>100 Ωm) west of Yufu volcano and east of Tsurumi volcano, or within the resistive-conductive transition zone, and in general avoid the surrounding conductive (<100 Ωm) zones. Between Seis. 2 and Seis. 3, the hypocenter of the 1975 M JMA 6.4 earthquake (Oita-Chubu earthquake) is located in a relatively resistive zone, although its hypocentral depth is poorly determined (Yamashina and Murai 1975; Fukuoka District Meteorological Observatory 1976) (Fig. 6). In general, the earthquakes occur in electrically resistive zones adjacent to conductive zones or resistive-conductive transition zones, and seismicity is low in conductive zones.

Fig. 6
figure 6

Relationship between resistivity structure and earthquakes. Hypocenters 1 month after the M JMA 6.5 earthquake of 21:25 JST, April 14, 2016, are plotted on the resistivity profiles at depths of 8 and 12 km with a depth range of ±1.5 km. Hypocenters of the M JMA 6.1 and M JMA 6.4 earthquakes in 1975 (Yamashina and Murai 1975) are also overlain on the profiles. Because the hypocentral depths of these two earthquakes were not determined, they are plotted on both profiles. Other notations are the same as in Fig. 1

Full size image

Because the conductive zones are located in the middle crust in the vicinity of active volcanoes (Kuju, Aso, Tsurumi, Garan, and Yufu) or in the lower crust beneath the Futagawa and Hinagu faults, we interpret that the deep conductors represent high-temperature ductile or low-rigidity zones due to the presence of fluids such as magma or saline water. In contrast, we interpret the resistive zones as relatively cold brittle zones with a fluid deficit. We hypothesize that the conductive zone preferentially deforms, such that the static stress over Kyushu (Matsumoto et al. 2015; Savage et al. 2016) accumulates preferentially in proximal brittle resistive zones and subsequently causes large earthquakes. The concept of local stress accumulation has been proposed based on the results of previous magnetotelluric studies (e.g., Ogawa et al. 2001; Ichihara et al. 2008, 2014; Wannamaker et al. 2009). The concept is similar to the hypothesis of Iio et al. (2002) who assumed that the lower crust had a deformable weak zone. In addition, our results suggest that fluids supplied from conductive zones to nearby resistive zones can promote earthquake occurrences (e.g., Mitsuhata et al. 2001; Ogawa et al. 2001; Yoshimura et al. 2009) by increasing pore pressure and decreasing the effective normal stress. Although it is usually difficult to uniquely determine the triggering mechanism, the effect of fluids on earthquake generation is especially plausible in the region of Seis. 3, which is approximately 100 km away from the hypocenter of the M JMA 7.3 earthquake. Between the hypocenter of the main shock and Seis. 3 lies a remarkably conductive zone. In this case, the static stress change (e.g., Hardebeck et al. 1998) caused by the 2016 Kumamoto earthquake was not effectively transferred at distance, but instead, the dynamic effects of seismic shaking were considered to be reasonable earthquake triggers (Miyazawa 2016). Because Seis. 3 occurs beneath active volcanoes, gas-rich hydrothermal water and/or gas bubbles exist at depth beneath the region, and their upwelling could have been facilitated by ground motion leading to earthquakes (e.g., Hill and Prejean 2005; Aizawa et al. 2016) in the Seis. 3 region.

This study shows that the seismogenic zones correspond approximately to resistive zones lying adjacent to conductive zones, or to the conductive-resistive transition zone. These results are consistent with previous magnetotelluric studies conducted across the epicenters of large (>M 6) inland earthquakes (Mitsuhata et al. 2001; Ogawa et al. 2001; Tank et al. 2003, 2005; Kasaya and Oshiman 2004; Ichihara et al. 2008, 2014; Yoshimura et al. 2008; Kaya et al. 2009; Umeda et al. 2011, 2014; Chandrasekhar et al. 2012) with the exception that aftershocks occur in a thick sedimentary layer (Uyeshima et al. 2005). Note here that the dense magnetotelluric observations occasionally image localized subvertical conductors beneath the active faults (e.g., Unsworth et al. 1997; Wannamaker et al. 2002; Becken et al. 2008; Ikeda et al. 2013; Sass et al. 2014). These local conductors, which are termed fault zone conductors, were interpreted to be damaged zones characterized by a fluid filled fracture network and altered clay materials. Previous two-dimensional (2-D) inversions in the region of our dataset have imaged the vertical conductors with a width of 1–4 km beneath the Futagawa fault (Asaue et al. 2004). The 1-D inversion of this study also shows the local conductor along the Futagawa fault at a depth of 1 km (FZC in Fig. 4). In addition, Fig. 4 shows the local conductor with a resistivity of around 100 Ωm at the hypocenter of the main shock. To confirm the presence of such small-scale conductors and their relationship to the earthquakes, the collection of more magnetotelluric observations and 2-D and/or 3-D inversions is necessary.

Considering the relationship between resistivity structure and seismicity, we suggest that two zones (Zones A and B in Fig. 6) have similar structures to the zones of Seis. 1–Seis. 3. Zone A corresponds to the southern part of Hinagu fault, which is 10–50 km from the hypocenter of the main shock. Zone A includes a zone of high radon-222 concentration in soil gas, which suggests large gas ascent velocities caused by frequently induced strain along the Hinagu fault (Koike et al. 2014). Further, a conductive zone like Seis. 1 is suggested at depths of ~20 km beneath Zone A, and therefore, this region is considered to have a relatively high potential of earthquake generation. Indeed in 1619, a M 6.2 earthquake occurred around Zone A (Usami 1967).

Zone B is located 70–100 km east-northeast of the mainshock hypocenter. Zone B has been classified as seismically inactive for the last 20 years (Matsumoto et al. 2015), and no earthquake was triggered by the 2016 Kumamoto earthquake (Figs. 1, 6). However, Zone B corresponds to the possible western extension of the MTL where a possibly active fault (the Imahata-Shiraie fault) is located (Research Group for Active Faults in Japan 1991). The resistivity structure show the structural boundary in Zone B at depths of 1–12 km (Figs. 4, 6). To investigate the origin of the structural boundary, we compared the shallow resistivity structure with the gravity data. Figure 7 shows the resistivity structure at a depth of 1 km and the Bouguer anomaly, assuming a terrain density of 2.63 g/cm3 (Geological Survey of Japan AIST 2013). The structural boundary of Zone B is well correlated to a zone with a steep gravity gradient, which is interpreted to be the southern rim of a graben structure (Kamata 1989; Kamata and Kodama 1994). Recent studies suggest that the graben is a pull-apart basin related to the MTL dextral movements partly with volcanic depressions (Itoh et al. 1998; Saiga et al. 2010; Itoh et al. 2014). The altered volcano-clastic rock and hydrothermal water filling the graben are considered to be the cause of the low resistivity at shallow levels. Although Itoh et al. (1998) and Itoh et al. (2014) suggest that Zone B is inactive at the present time, GPS data support active shear around this region (Nishimura and Hashimoto 2006; Wallace et al. 2009). The stress field estimated from earthquakes also supports aseismic slip at the shear zone (Matsumoto et al. 2015). Furthermore, Zone B corresponds to the edge of the conductor at depths of 8 and 12 km. Therefore, Zone B may have a high potential of occurrence of large earthquakes.

Fig. 7
figure 7

Comparison between resistivity structure and Bouguer gravity anomaly assuming a terrain density of 2.63 g/cm3 [copyright for the basemaps of the gravity maps belongs to the Geospatial Information Authority of Japan (GSI); the map is licensed under the Creative Commons Attribution 2.1 Japan License]. Bouguer anomaly map was drawn using a database edited by AIST (Geological Survey of Japan AIST 2013)

Full size image

Conclusions

1-D analysis of the resistivity structure constrained by the 247 broadband MT and telluric observation sites has clarified that the aftershocks and triggered earthquakes of the 2016 Kumamoto earthquake occurred on electrically resistive zones adjacent to conductive zones or resistive-conductive transition zones. Seismicity was found to be quite low in the electrically conductive zones that are interpreted to be fluidized. This relationship is consistent with previous MT studies of other seismogenic zones around the world. Therefore, we conclude that seismic hazard assessments may be improved by considering the resistivity structure.

We interpret the difference in resistivity to represent a difference in elastic properties. The release of stress that had accumulated within the resistive region in the vicinity of the resistive-conductive boundary, probably led to the series of earthquakes. Increases in pore pressure from fluids supplied from the conductive zone may have been an additional cause of the earthquakes. Future dense MT observations made around the Futagawa and Hinagu faults and 3-D inversion will contribute to improving the sharpness of resistivity structure images and will characterize seismicity more clearly from the viewpoint of its relationship with resistivity structure.