Introduction

It is well known that elastic anisotropy in the Earth’s mantle is closely related to mantle dynamics. Sources of seismic anisotropy in the crust and mantle have been extensively investigated. Two widely accepted models of such anisotropy are olivine crystal alignment in the upper mantle (Hess 1964; Francis 1969; Fuchs 1977; Ando et al. 1983) and fracture alignment in the shallow crust (Nur and Simmons 1969; Gupta 1973; Crampin 1978; Crampin et al. 1980). The most widely supported model of mantle anisotropy is olivine alignment. The most common mantle constituent is olivine, and olivine crystals are thought to align in the direction of mantle flow, thereby generating anisotropy via the preferred orientations of the crystals. Shear-wave splitting analysis is one of the most commonly used tools to detect anisotropy and infer mantle flow direction (e.g., Silver and Chan 1991). Beneath Japan, the Pacific and Philippine Sea plates are subducting beneath the Eurasian plate to the west and northwest, respectively. Complex mantle flow with different flow directions is therefore expected, and the flow directions can be estimated from observed directions of mantle anisotropy polarization.

Fracture alignment is widely invoked to explain observations of seismic anisotropy in the crust (e.g., Kaneshima 1990). In this model, shear-wave splitting in the crust is caused by the orientations of faults or cracks, and it is thought that propagating cracks are preferentially aligned subparallel to the orientation of the axis of maximum stress. This, in turn, means that the “fast” direction of shear waves should also be oriented subparallel to the axis of maximum stress (Crampin 1978). Heterogeneous structures in the upper mantle can generate anisotropy by a similar fracture alignment process (e.g., Iidaka and Obara 1995). Iidaka and Obara (1995) detected anisotropic regions in the mantle wedge beneath central Japan. The anisotropic regions correspond to zones of low seismic velocity and low Q; this region is therefore thought to result from heterogeneous structures formed by fluid rising from the mantle wedge.

Data from a high-density Global Navigation Satellite System (GNSS) array study, operated by the Geospatial Information Authority of Japan (GSI), indicate a zone of high strain rate along the Japan Sea coastline (e.g., Sagiya et al. 2000). This zone has been termed the Niigata–Kobe Tectonic Zone (NKTZ). Many large historic earthquakes have occurred in the NKTZ, including the magnitude 8 Nobi earthquake of October 28, 1891. This event is Japan’s strongest known intraplate earthquake. The maximum size of typical intraplate earthquakes in Japan is ~7; however, during the Nobi earthquake the rupture propagated along three different faults: the Nukumi, Neodani, and Umehara faults.

The faults of the 1891 Nobi earthquake are all located in the NKTZ. Intraplate earthquakes occur as a result of stress concentration and strain accumulation on crustal faults. Understanding the tectonic setting of large intraplate earthquakes is important in earthquake hazard mitigation. Recently, it has been suggested that the presence of crustal fluid is closely related to the triggering of intraplate earthquakes. Iio et al. (2002) proposed a model that explains the cause of some intraplate earthquakes, whereby a weak zone in the lower crust deforms relatively easily, resulting in a high concentration of stress above it.

Many previous studies have investigated how the presence of fluid in the lower crust affects intraplate earthquakes (Eberhart-Phillips and Michael 1993; Johnson and McEvilly 1995; Zhao et al. 1996; Hasegawa et al. 2009). Fluid released from the upper mantle and lower crust is closely related to the mechanical state of the crust, as fluid plays an important role in crustal weakening. Zones of low resistivity are generally located in the deeper part of an active fault (e.g., Ogawa et al. 2001; Becken and Ritter 2012), and conductivity in the crust is sensitive to interconnected fluids in rock. Fluids under high pressures reduce the strain thresholds of rocks and facilitate brittle failure within the upper crust (e.g., Byerlee 1990). Thus, fluids in interconnected porosity networks can strongly influence deformation along fault zones.

The Research Group for the Joint Seismic Observations at the Nobi Area deployed 43 temporary seismic stations in an 80 × 80 km area, including the source faults of the 1891 Nobi earthquake, to better understand the tectonic setting of such a large intraplate event. Shear-wave splitting was investigated using both temporary and permanent seismic stations.

Data

The data used in this experiment were recorded by temporary seismic stations deployed by the Research Group for the Joint Seismic Observations at the Nobi Area and permanent Hi-net stations operated by the National Research Institute for Earth Science and Disaster Prevention (NIED; Okada et al. 2004). A station map is shown in Fig. 1. Most seismic stations had three-component sensors with a natural frequency of 1 Hz.

Fig. 1
figure 1

Location map of seismic stations and earthquakes. Red circles denote epicenters of deep earthquakes. Black triangles indicate permanent seismic stations operated by Hi-net and universities. Blue squares indicate temporary seismic stations operated during this study. The lightly shaded area denotes the NKTZ. Yellow lines are faults that ruptured during the 1891 Nobi earthquake. The location of the research area (black square) is shown in the inset

Full size image

Our data set consisted of earthquakes located deeper than 230 km that occurred between January 1, 2009, and December 31, 2014 (Fig. 1; Table 1). Only seismic rays with incidence angles of <30° were used, to avoid waveform distortion due to the phase shift between the SH and SV components (e.g., Nuttli 1961; Mendiguren 1969), as the use of such waves could lead to false detections of shear-wave anisotropy.

Table 1 List of earthquakes
Full size table

Methods

Shear-wave splitting is usually expressed in terms of two parameters: fast polarization azimuth, ϕ (in degrees), and time lag, τ (in seconds), which is the delay between the fast and slow arrivals. We investigate shear-wave splitting using techniques based on Fukao (1984). A band-pass filter with a frequency window of 0.5–8 Hz is applied to all waveform data. We pick S waves manually from individual seismograms. We then calculate the cross-correlations of the two horizontal components of motion over the ranges ϕ = −90° to +90° and a window of 4 s, in increments of 1° and 0.01 s, respectively. We choose the time lag τ and fast polarization azimuth ϕ that yield the maximum correlation (Fig. 2). To confirm the resultant splitting parameters, we calculate anisotropy-corrected seismograms. The anisotropy-corrected particle motion is linear (Fig. 2d), as expected for a double-couple source mechanism. Low-quality waveform data with poor signal-to-noise ratios (SNR) are removed using the data with maximum cross-correlation coefficients of >0.75. The linear polarizations obtained from anisotropy-corrected data suggest that observed waveform splitting is caused by an anisotropic medium, rather than seismic noise.

Fig. 2
figure 2

Example of waveform data. a Original waveform data from the N–S and E–W components. b The rotated waveforms with the maximum cross-correlation coefficient. c Original particle motions. d Anisotropy-compensated particle motions

Full size image

Results and discussion

Heterogeneous mantle anisotropy and its origin

Based on spatially distinct groups of earthquake hypocenters, the research area is divided into four regions denoted as a–d in Fig. 3. Shear-wave splitting data in each subfigure were obtained from earthquakes in the plotted region.

Fig. 3
figure 3

Polarization data from the research area are shown in four geographic regions (a–d). The polarization azimuths of the maximum-velocity phase and the time delay between the maximum- and minimum-velocity phases are indicated by the direction and length of each bar, respectively. Bars are overlain on locations of seismic stations. Stars denote epicenters of deep earthquakes

Full size image

Region A is located in the northeastern part of the study area (Fig. 3a). The depths of the earthquakes are 230–290 km. The polarization direction data show significant lateral variations: Polarization directions align NE–SW in the northern part of the region and ESE–WNW in the southern part. The polarization directions from region B also show lateral variations: Using earthquakes with depths of 250–320 km, the polarization directions align ESE–WNW in the southern part of the region, with a maximum time delay of ~2 s (Fig. 3b). Notably, the faults ruptured by the 1891 Nobi earthquake are located in this area. Results for region C, in the southern part of the research area, are shown in Fig. 3c. Here, the depths of the earthquakes used are 320–370 km, and the most common polarization direction is ENE–WSW. Polarization data from region D again suggest large lateral variations (Fig. 3d). Using earthquakes with focal depths of 340–355 km, we find that polarizations from the central part of region D align ENE–WSW, consistent with region C; yet, in the southeastern and northeastern parts of the region, polarizations align N–S and delay times are small.

The entirety of the polarization data set (obtained from all four regions) is plotted in Fig. 4a. Large lateral variations are clearly visible, and some remarkable characteristics can be seen. The polarizations are aligned NE–SW in the northeastern part of the study area, ESE–WNW in the central part, and ENE–WSW in the southwestern part (Fig. 4a). Near the boundaries of each polarization group, the directions are less uniform. At the border area of the polarization groups, the rays propagate both of the anisotropic regions. We suppose that the data are scattered and less uniform. We now discuss the potential causes of anisotropy in each area. The earthquakes used here are all deep earthquakes in the subducting Pacific plate, with focal depths of 230–370 km. Their ray paths thus propagate through the mantle wedge and crust; we must therefore consider both crustal and mantle anisotropy as possible sources of the observed anisotropy.

Fig. 4
figure 4

a Polarization data from all regions with iso-depth contours of the Pacific and Philippine Sea slabs. The iso-depth lines of the subducting Philippine Sea and Pacific plates are indicated by red and blue broken lines, respectively (Nakajima and Hasegawa 2007). Numbers indicate line depths in kilometers. The north end of the subducting Philippine Sea plate, estimated from seismic tomography, is indicated by a thick broken black line. The yellow lines are the locations of faults that ruptured during the 1891 Nobi earthquake. The rose diagrams show the number of the polarization directions for each part. The boundaries of each part are shown by green dashed line. b Schematic figure of subducting oceanic plates in the Japan region. The NE–SW polarization directions observed in the northeastern part of a and ENE–WSW directions observed in the southwestern part are consistent with the movement directions of the subducting oceanic slabs. c Schematic figure of the region beneath the 1891 Nobi earthquake. The heterogeneous structures were inferred from magnetic conductivity studies. The inferred fluid-rich area is shaded blue. The brown square indicates the region of the 1891 Nobi earthquake. Heterogeneous structures appear to correspond to anisotropic regions in the mantle wedge

Full size image

Most of the observed time delay values are larger than 0.5 s. Crustal anisotropy in Japan has been described in many studies (e.g., Kaneshima 1990; Hiramatsu et al. 2015). It has been suggested that the maximum possible time delay of shear-wave splitting due to crustal anisotropy is <0.2 s for Japan (Kaneshima 1990). Hiramatsu et al. (2015) studied crustal anisotropy in the area of the 1891 Nobi earthquake using the same temporary seismic network as in this study and suggested that the maximum time delay in this area was <0.12 s. The observed time delays in this study are much larger than those values; we therefore conclude that the observed splitting is caused mainly by mantle anisotropy.

We consider two possible causes for the observed anisotropy: olivine alignment (e.g., Silver and Chan 1991) and structural heterogeneities (e.g., Iidaka and Obara 1995). Iso-depth lines of the Philippine Sea and Pacific plates are shown in Fig. 4a, adapted from Nakajima and Hasegawa (2007) and Zhao and Hasegawa (1993), respectively. NE–SW polarization, which dominates in the northeastern part of the research area, is highly consistent with the subduction direction of the Philippine Sea slab. Anisotropy in this region had been studied by many researchers (e.g., Ando et al. 1983; Iidaka et al. 2009; Hiramatsu et al. 1998). In this area, the geometry of the subducting Philippine Sea plate has been extensively studied (e.g., Nakajima and Hasegawa 2007): It is widely accepted that the dip direction of the Philippine Sea slab in this region is northeast. The plate motion direction of the Philippine Sea plate has been studied by many scientists (e.g., Demets et al. 2010). The direction of the Philippine Sea plate was estimated to NW at the Nankai Trough. But, the driving force of the plate is considered as the negative buoyancy of the cold slab (e.g., Forsyth and Uyeda 1975). The mantle flow is caused by the subducting slab. The direction of the mantle flow should be parallel to the subduction direction.

The causes of anisotropy in mantle are well studied (e.g., Hess 1964; Francis 1969; Fuchs 1977) and usually interpreted as a result of olivine crystal alignment. Olivine is a major mantle constituent, and it becomes anisotropic with only modest amounts of crystal alignment; crystals align in the direction of mantle flow, and the resultant polarization direction is parallel to the flow direction.

The preferred orientation of the olivine crystal is expected to be parallel to the plate subduction. Thus, the observed NE–SW polarizations are consistent with the subduction direction of the Philippine Sea slab (Fig. 4a). We can therefore assume that the direction of mantle flow corresponds to the direction of shear-wave splitting polarization; it follows that the NE–SW-aligned polarizations can be explained by mantle flow related to the subducting Philippine Sea slab. A high-density network of temporary seismometers was deployed in this area by The Japanese University Group of the Joint Seismic Observations at NKTZ (2005). Iidaka et al. (2009) studied shear-wave splitting in the region and found clear fine-scale lateral variations in polarization directions, consistent with the subduction direction of the Philippine Sea slab. Thus, our findings are consistent with the results of Iidaka et al. (2009).

Next, we discuss the possible source of the ENE–WSW polarizations observed in the southwestern part of the research area (Fig. 3c). Hiramatsu et al. (1998) studied seismic anisotropy in central Japan, including the southern part of the target area of the present study, and estimated depth variations in anisotropy using a dense set of ray paths. The anisotropic region found in their study was consistent with a low-velocity region obtained from seismic tomography (Hirahara et al. 1989; Sekiguchi 1991); thus, they concluded that the cause of the anisotropy was related to the heterogeneous structures created by fluid-filled cracks. Nakamura et al. (2008) found velocity perturbations that suggested a low-Vp region at depths of ~100–150 km beneath the eastern part of our study area; however, there was no corresponding low-Vs region at the same depths. If the cause of the anisotropy is truly fluid-filled cracks, then the area should be characterized by low-Vs and low-Q regions (e.g., Iidaka and Obara 1995; Hiramatsu et al. 1998). It follows that the anisotropy found in the present study is unlikely to result from structural heterogeneities associated with fluid-filled cracks. Instead, the well-resolved direction of subduction of the Pacific plate (e.g., Zhao and Hasegawa 1993) is consistent with the alignment of the observed polarizations. We therefore suggest that the anisotropy in these regions could be caused by the alignment of olivine crystals in the mantle wedge. The crystal alignment is expected to result from mantle flow related to subduction of the Pacific plate.

The NE–SW and ENE–WSW orientations can be explained by anisotropy caused by olivine alignment related to the subducting oceanic slabs (Fig. 4b). However, this does not adequately explain the observed ESE–WNW polarizations in the central part of the research area (Fig. 4a). Beneath this area, two oceanic plates are descending with different directions. The Pacific plate is descending to WSW. The dip direction of the Philippine Sea slab is NE. The effects caused by the subducting two slabs have to be considered. However, the depth of the Philippine Sea slab is almost similar to that of the Moho boundary beneath the faults of Nobi earthquake (Nakajima et al. 2015). The thickness of the mantle wedge between the continental Moho and top of the Philippine Sea slab seems to be about 5–10 km at the area. The contribution to the shear-wave splitting caused by the flow related to Philippine Sea slab seems to be small.

Instead, we must consider other possible causes. Other explanations of the anisotropy in this area have been proposed by, for example, Iidaka and Obara (1995) and Hiramatsu et al. (1998). Iidaka and Obara (1995) studied seismic anisotropy in the mantle wedge in central Japan and found remarkable lateral variations: The magnitudes of shear-wave splitting time delays differed between the fore-arc and back-arc sides of the volcanic front, and the anisotropic region observed at the back-arc side of the volcanic front was characterized by low seismic velocity and low Q. Their interpretation was that the anisotropic region was caused by heterogeneous structure generated by fluid rising from the mantle wedge.

Relationship between fluids from the mantle and the generation of the 1891 Nobi earthquake

The area of ESE–WNW-aligned polarizations in Fig. 4a includes the faults that ruptured during the 1891 Nobi earthquake. Investigations of seismic velocity structure in the area have revealed the faults responsible for the earthquake. The relationship between crustal fluids and earthquakes has been studied previously, including studies of the controls of such fluids on intraplate earthquakes. A close relationship between the earthquake’s fault zone and a subsurface low-velocity zone has been established (Eberhart-Phillips and Michael 1993; Johnson and McEvilly 1995; Zhao et al. 1996). A review of intraplate earthquakes in Japan found clear imaging of low-velocity zones and/or high-Vp/Vs zones near the rupture sources of many major earthquakes (Hasegawa et al. 2009), presumably related to the presence of lower crustal fluid; however, velocity structure alone is not sufficient to infer the presence of fluids. Geomagnetic surveys have found regions of low resistivity near many active faults, which could relate to the presence of fluids in the source area (Gupta et al. 1996; Tank et al. 2005; Wannamaker et al. 2009; Becken et al. 2011; Becken and Ritter 2012). In Japan, some close relationships between low-resistivity regions and earthquake faults have been proposed (Mitsuhata et al. 2001; Yamaguchi et al. 2001; Uyeshima et al. 2005; Yoshimura et al. 2008). The characteristics of the relationships between faults and resistivity structures are summarized by Ogawa et al. (2001) and Becken and Ritter (2012). Low-resistivity zones are generally located in the deeper parts of active faults and are thought to be caused by fluids. Fluids released from the upper mantle are closely related to the mechanical state of the crust.

In the source region of the 1891 Nobi earthquake, seismic tomography reveals fine-scale P- and S-wave velocity structures in the crust and uppermost mantle underlying the faults of the earthquake (Nakajima et al. 2015). The lower crust beneath the Nukumi fault, corresponding to the northern extent of the Nobi earthquake rupture, contains a low-velocity zone. On the other hand, inferred seismic velocities beneath the Umehara fault, at the southern terminus of the rupture, are only moderate. In the uppermost mantle beneath the Nobi earthquake regions have seismic velocities ~15 % lower than the velocities expected for mantle peridotites. Across-fault variations in seismic velocity include a body of low P-wave velocity in the lower crust, partly underlying the southwestern region of the Nukumi fault. These results suggest fluids with volume fractions of 2–3 % are present in the local area, extending from the uppermost mantle to the middle crust beneath the Nukumi fault.

Resistivity is highly sensitive to the amount of crustal fluid present. The existence of a region of low resistivity near an active fault can be interpreted as evidence of crustal fluid in the source area. Local resistivity structure suggests the presence of crustal fluid (Uyeshima et al. 2013), including a region of low resistivity beneath the Nobi earthquake fault system. In this region, the depth to the Philippine Sea slab is ~50 km and the Pacific slab is located at depths of ~300 km. The region of low resistivity extends from the lower crust to the mantle wedge just above the Pacific slab and includes the Philippine Sea slab. Figure 4c shows a conceptual model of this interpretation: Fluid is released from the subducting Pacific slab by dehydration reactions and rises buoyantly from the mantle wedge. Additional fluid is contributed by the Philippine Sea slab; the resultant aggregation of fluid is large enough to reach the upper crust.

In this model, heterogeneous structures caused by rising fluid create the anisotropic medium. We consider the anisotropic region to be a result of fluid in the mantle, generated by dehydration of the subducting oceanic slabs. Fluid plays an important role in crustal weakening, and crustal conductivity is sensitive to interconnected fluids in rocks. Fluids at high pressures have the ability to reduce the strain thresholds of rocks and facilitate brittle failure within the upper crust (e.g., Byerlee 1990), while fluids in interconnected porosity networks reduce the shear moduli, which can strongly influence deformation in fault zones.

Conclusions

We used seismic data from permanent networks and a temporary deployment of three-component sensors to investigate anisotropy in the rupture area of the 1891 Nobi earthquake, the largest known intraplate earthquake in Japan. Large lateral variations in azimuthal directions of shear-wave splitting were detected. In the northeastern Chubu Region, polarization aligns NE–SW; this can be explained by mantle flow related to subduction of the Philippine Sea plate. Polarization directions in the southwestern part of the Chubu Region align ENE–WSW, consistent with mantle flow related to the subducting Pacific plate. In central Chubu, however, polarizations align ESE–WNW, inconsistent with the subduction of either oceanic slab. This area is located just beneath the fault zone of the 1891 Nobi earthquake and overlies a mantle wedge corresponding to low seismic velocity and low electric resistivity. Anisotropy here is best explained by mantle heterogeneities related to fluid upwelling from the subducting slabs. Because these fluids drive crustal weakening, this explanation is consistent with fluid-weakened lower crust as the cause of the Nobi earthquake.