Background

The Niigata–Kobe Tectonic Zone (NKTZ) is a geodetically derived high-strain-rate zone that extends for ~500 km (NE–SW) across central Japan (Sagiya et al. 2000) (purple shading in Fig. 1). The NKTZ is contracting in the WNW–ESE direction at a rate of ~107 year−1, and the contraction rates are a few times higher than those observed in surrounding regions. The high strain rates were first interpreted based on kinematic models, such as a detachment model (Hirahara et al. 1998) and collision models (e.g., Shimazaki and Zhao 2000; Heki and Miyazaki 2001), but these models are unlikely to produce the NKTZ due to the implausible physics required to produce the combination of the observed stress field, inferred mantle flow, and observed fault movement (Iio et al. 2002). Iio et al. (2002, 2004) proposed an alternative model whereby deformation in the weak, ductile lower crust facilitates the high strain rates along the NKTZ. However, geodetic observations with dense GNSS arrays have revealed regional variations in the width of the NKTZ, ranging from tens kilometers to the east of the Itoigawa–Shizuoka Tectonic Line (ISTL) to 50–100 km to the west of the ISTL (e.g., Sagiya et al. 2004; Ohzono et al. 2009; Nishimura et al. 2012). These observations suggest that underlying physical mechanisms generating surface deformation are different between the west and east of the ISTL.

Fig. 1
figure 1

Tectonic setting of central Japan. The Niigata–Kobe Tectonic Zone (NKTZ) (Sagiya et al. 2000) is denoted by purple shading. Black triangles represent active volcanoes, and orange stars denote epicenters of earthquakes (M ≥ 6.5) that occurred at depths of ≤20 km during 1923–2015. Gray lines denote active faults. The Itoigawa–Shizuoka Tectonic Line (ISTL) is shown by green curve. Names of regions discussed in the text are labeled

Full size image

Nakajima and Hasegawa (2007) discussed the origin and segmentation of the NKTZ on the basis of seismic velocity structures and suggested that the crustal structure west of the ISTL differs from that east of the ISTL. Jin and Aki (2005) reported a narrow, conspicuous low-coda attenuation zone west of the ISTL at frequencies less than 4 Hz, whose existence was later corroborated by Carcole and Sato (2010) and Hiramatsu et al. (2013). Shear-wave splitting analysis of the upper crust suggests that the high strain rates at the surface west of the ISTL are caused by high deformation rates below the brittle–ductile transition zone (Hiramatsu et al. 2010). These seismological observations obtained in the last decade all support the model by Iio et al. (2002, 2004), suggesting that concentrated deformation in the weakened lower crust plays a crucial role in the generation of high strain rates west of the ISTL.

Three-dimensional (3D) seismic velocity heterogeneities have enhanced our understanding of spatial variations in the elastic properties beneath high-strain-rate zones (e.g., Nakajima and Hasegawa 2007; Okada et al. 2014). However, the anelasticity associated with such zones is poorly understood. Anelasticity derived from intrinsic seismic attenuation is a key parameter for evaluating ongoing viscoelastic deformation in the lower crust below high-strain-rate zones and mature fault systems (e.g., Eberhart-Phillips et al. 2014; Eberhart-Phillips 2016). Here we estimate the 3D P-wave attenuation (Q −1p ) structure in central Japan using waveform data recorded by a dense nationwide seismic network. We use a three-step approach developed by Nakajima et al. (2013) to delineate the 3D Q −1p structure, which can minimize the potential trade-offs among unknown parameters. The corner frequency (f c) of each earthquake is first estimated by the S-coda spectral ratio method. Next, we perform a joint inversion for the attenuation term \(t^{*}\) and site amplification factors. Finally, the values of \(t^{*}\) are inverted for the 3D Q −1p structure (e.g., Rietbrock 2001). The obtained attenuation model is discussed in the context of the deformation patterns observed within the NKTZ.

Data and methods

Corner frequency

We used velocity waveform data from 8300 earthquakes (2 ≤ M ≤ 5) with focal depths shallower than 350 km that occurred between October 2006 and February 2015 (Fig. 2a), and calculated spectral amplitudes of S-wave codas in 10 s time window taken at twice the theoretical S-wave travel time for a one-dimensional seismic velocity model (Hasegawa et al. 1978). Spectral ratios were calculated at common stations for earthquake pairs with a magnitude difference ≥0.5 and an inter-event distance ≤40 km. We then fitted an ω 2 source model (Brune 1970) to the averaged spectral ratio in the frequency range of 1–32 Hz and determined values of f c for the earthquake pair by a grid search with f c steps of 0.2 Hz. Figure 3b shows an example of f c estimation procedure for an earthquake pair. Values of f c were estimated to be 5.2 and 15.4 Hz for M3.5 and M2.4 earthquakes, respectively. Standard errors of f c evaluated by a bootstrap method with 1000 bootstrap samples were 0.27 Hz for the M3.5 earthquake and 1.12 Hz for the M2.4 earthquake (Fig. 3c). We applied this procedure to available earthquake pairs and determined a pair of f c values. When five or more f c values were obtained from different pairs for an earthquake, we calculated the average f c and used the value in the subsequent analysis. The total number of the earthquakes with f c estimates was 4718 (Fig. 2a).

Fig. 2
figure 2

a Distribution of hypocenters. Gray circles denote 8300 earthquakes used for estimates of corner frequencies, while colored circles represent 4718 out of 8300 earthquakes for which the corner frequency could be estimated. Circle colors correspond to earthquake focal depths. b Distribution of seismograph stations (white squares) and horizontal grid nodes (blue crosses)

Full size image
Fig. 3
figure 3

Examples of estimates of corner frequency (f c), attenuation term (\(t^{*}\)), and site amplification factors. a Map showing the locations of two earthquakes (red stars) and three stations (blue squares). EQ1 is an earthquake with M3.5 and a focal depth of 62 km, and EQ2 is an earthquake with M2.4 and a focal depth of 46 km. b S-coda spectral amplitude ratios for a pair of EQ1 and EQ2. Gray and blue lines denote spectral ratios for various common stations and the averaged spectral ratio, respectively. The red line denotes the theoretical spectral ratio that best fits the observations. Values of f c estimated for the two earthquakes are indicated by green lines with f c values. Inter-event distances (D) and the number of common stations (N) are shown in the panel. c Uncertainties in f c estimates evaluated by a bootstrap method. Histograms of f c for (left) EQ1 and (right) EQ2 are plotted from 1000 bootstrap samples, and standard errors of f c are 0.27 Hz for EQ1 and 1.12 Hz for EQ2. d Spectral amplitudes of observed P waves (blue curves) and noise (dashed curves). Red line represents the theoretical spectral amplitude calculated from the optimal value of \(t_{0}^{*}\), site-amplification factors, and the ω 2 source model. e Site amplification factors for the three stations

Full size image

Attenuation term and site amplification factor

For the earthquakes with f c estimates, we calculated P-wave and noise spectral amplitudes from vertical component of velocity seismograms with a window length of 2.56 s after and before the P-wave onset, respectively. We limited our analysis to a frequency range with signal-to-noise ratios of ≥3. Frequency-dependent attenuation was assumed to obey the relationship

$$t^{*} (t) = t_{0}^{*} f^{ - \alpha } ,$$

where \(t_{0}^{*}\) is the attenuation term at 1 Hz (Stachnik et al. 2004). We used α = 0.27 because α = 0.2–0.3 has been estimated for the crust and the uppermost mantle beneath the Japanese Islands (Nakajima et al. 2013; Nakajima 2014; Saita et al. 2015) and this value range is the most appropriate value for the mantle (e.g., Jackson et al. 2002). We constructed a set of observation equations for multiple earthquakes at a single station and carried out a simultaneous inversion for \(t_{0}^{*}\) and relative site amplification factors (for details, see Nakajima et al. 2013). As values of f c estimated by the coda spectral ratio technique were for S waves, we calculated values of f c for P waves by assuming the relationship f cp = 1.33 × f cs (Uchida et al. 2007). An example of spectral amplitudes observed at three stations for a single earthquake (EQ1) is shown in Fig. 3d. Observed spectral amplitudes (blue curves in Fig. 3d) are reproduced by theoretical spectral amplitudes (red curve) that incorporated the estimated \(t_{0}^{*}\), site amplification factors, and the ω 2 source model. Relative site amplifications estimated for the three stations are shown in Fig. 3e.

Tomographic inversion

The 3D Q −1p structure was determined by the inversion of 247,595 \(t_{0}^{*}\) values observed at 455 stations. It is noted that Q −1p values estimated in this study correspond to those at 1 Hz. We calculated ray paths and travel times based on the 3D P-wave velocity model of Nakajima and Hasegawa (2007) with the ray tracing technique of Zhao et al. (1992a). In the inversion, we used a model parameterization identical to that in Nakajima and Hasegawa (2007), where crustal discontinuities (Zhao et al. 1992b) and the upper boundary of the Pacific slab (Nakajima et al. 2009) were considered, adjacent horizontal grid nodes were spaced at intervals of 0.2°, and vertical nodes were spaced at intervals of 5–50 km (Fig. 2b). We set initial values of Q −1p to be 0.0033 for the crust and mantle wedge, and 0.001 for the Pacific slab, and selected a damping parameter of 100 based on a trade-off curve between the root mean square (RMS) residuals of \(t_{0}^{*}\) and model variance. Final results were obtained after four iterations; \(t_{0}^{*}\) RMS residuals decreased from 0.036 s in the initial model to 0.015 s.

We carried out two checkerboard resolution tests to assess the reliability of tomographic results and confirmed that the obtained results are reliable beneath the land area at a depth of <40 km (see Additional file 1). We hereafter discuss the tomographic results in well-resolved, unmasked areas in Fig. 4, where recovery rates are greater than 20%.

Fig. 4
figure 4

a Map showing Q −1p at depths of 5, 10, 25, and 40 km. The color scale is shown in the right bottom panel. Black triangles represent active volcanoes. Gray dots at a depth of 10 km represent earthquakes (M ≥ 1) in a depth range of 5–15 km in 2003–2015, and stars at depths of 10 and 25 km denote earthquakes (M ≥ 6.5) at depths of ≤20 km from 1923 to 2015. Deep low-frequency earthquakes that occur within 2.5 km of each depth slice are indicated by white circles. The black dashed line shows the 40 km iso-depth contour of the upper surface of the Philippine Sea slab (Hirose et al. 2008; Nakajima et al. 2009). Areas with recovery rates of <20% for the two-grid-model of checkerboard resolution test are shaded in gray. b Vertical cross section of Q −1p along line A–B shown in the right bottom panel in a. Dashed lines indicate the Moho (Zhao et al. 1992b) and the upper surface of the Philippine Sea slab. The intersection of line A–B with the ISTL is indicated by a green arrow. Black triangles at the top denote active volcanoes. Gray dots and white circles represent earthquakes (M ≥ 1) and low-frequency earthquakes, respectively, that occurred within 5 km of line A–B. c Vertical cross section of P-wave velocity perturbations along line A–B (Nakajima and Hasegawa 2007). Other symbols are the same as in b

Full size image

Tomography results

The results show marked variations in Q −1p with depth (Fig. 4a). At a depth of 5 km, high-attenuation areas are distributed mainly in Kinki and Tokai districts, and around active volcanoes. In contrast, areas of moderate to low attenuation are dominant at a depth of 10 km, even though isolated high-attenuation areas are observed in Kanto, Tokai, and Kinki districts. Seismicity in the upper crust and earthquakes with M ≥ 6.5 (white stars in Fig. 4) tend to occur in areas of low to moderate attenuation. The lower crust, at a depth of 25 km, shows high attenuation in Kinki, southeast of Biwa Lake, and around volcanic areas in the northern part of the study area. On the whole, the NKTZ shows high attenuation west of the ISTL, but there is a moderate- to low-attenuation area northeast of Biwa Lake, surrounded by high-attenuation areas.

Figure 4b shows a vertical cross section of Q −1p along the NKTZ (line A–B shown in the panel for 40 km depth), and characterizes attenuation features in the crust of the NKTZ. Marked high-attenuation areas exist in the lower crust at horizontal distances of 0–120 km and 200–300 km, while the lower crust shows low attenuation east of the ISTL at distances of >370 km. In volcanic areas located at horizontal distances of 300–370 km, high-attenuation areas extend continuously from the lower crust to the surface. The resolution of the uppermost mantle is not sufficient to discuss the observed attenuation structure, as a result of an insufficient number of rays propagating at depths >60 km, where the recovery of the checkerboard resolution tests is not good (Fig. 4b; Additional file 1: Figure S1).

Discussion

One of the most striking features revealed by this study is the segmentation of the NKTZ derived from the attenuation structure in the crust. Nakajima and Hasegawa (2007) proposed that the NKTZ can be divided into a non-volcanic area (region A), a volcanic area (region B), and an area east of the ISTL (region C) (Fig. 4c). The distribution of Q −1p along the NKTZ correlates well with that of P-wave velocity perturbations (dVp) in the sense that high-attenuation areas correspond to low-velocity areas and vice versa (Fig. 4b, c). The correlation coefficient between Q −1p and dVp at a depth of 25 km is calculated to be −0.47 with a probability value of ~10−7, indicating a moderately weak but statistically significant negative relationship.

To the east of the ISTL (region C), the upper crust shows high attenuation and low velocity, while the lower crust shows low attenuation and moderate to high velocity (see dark symbols in Fig. 5), which differentiates seismic characteristics of the lower crust west of the ISTL. The high-attenuation, low-velocity upper crust probably reflects a thick (~6 km) accumulated basin fill (e.g., Sato 1994). Seismic anisotropy derived from shear-wave splitting analysis reveals that the anisotropy east of the ISTL is structurally induced, which can be explained by intense faulting and folding in the sedimentary basin (Hiramatsu et al. 2010).

Fig. 5
figure 5

a Relationship between P-wave attenuation values obtained in this study and P-wave velocity perturbations (Nakajima and Hasegawa 2007) at a depth of 25 km for each grid node located beneath the NKTZ (purple shading in Fig. 1). The number of grid nodes shown is 109. The cross-correlation coefficient for the two data sets is −0.47. Colors represent the distance along the NKTZ calculated from point A in line A–B (Fig. 4). b Relationship between P-wave attenuation values obtained in this study and P-wave velocity perturbations that are averaged at every 20 km along line A–B. Bars denote 1-sigma uncertainties

Full size image

The volcanic area (region B) shows high attenuation and low velocity throughout the crust. High attenuation in this area is also suggested from S-wave attenuation studies (e.g., Carcole and Sato 2010; Liu and Zhao 2015) and low-frequency (1–4 Hz) coda-wave attenuation (e.g., Jin and Aki 2005; Hiramatsu et al. 2013). As Vp/Vs values in the lower crust are high (~1.85; Nakajima and Hasegawa 2007) and a low-resistivity body has been reported at depths of 20–30 km (e.g., Ogawa and Honkura 2004), we infer that the existence of magmatic fluids is responsible for the high-attenuation and low-velocity anomalies. The magmatic fluids probably originate from dehydration-related fluids in both the Philippine Sea and Pacific slabs (Nakamura et al. 2008). The possibility of the upward migration of slab-derived fluids is supported by high 3He/4He ratios (~8 Ra) (Umeda et al. 2013). The absence of apparent high-attenuation areas beneath the Moho is probably due to the insufficient number of rays propagating through the uppermost mantle, as a consequence of small number of deep-focus earthquakes with reliable f c estimates; this inference is supported by the poor recovery rates of the checkerboard patterns at depths of >60 km (Fig. 4b).

On the whole, the lower crust beneath region A shows a marked low-velocity and high-attenuation anomaly (Fig. 4b, c). Since heat flow measurements do not suggest high-temperature conditions within the NKTZ (Tanaka et al. 2004), the presence of aqueous fluids probably results in reduced seismic velocities and enhanced seismic attenuation (Toksöz et al. 1979). However, a low-attenuation area exists at horizontal distances of 130–180 km, where the Philippine Sea slab is in contact with the continental Moho (Fig. 4b). It is known that a temperature decrease by ~100 °C reduces seismic velocity by only ~1% (e.g., Duffy and Anderson 1989), but attenuation in olivine could be halved under upper mantle conditions (e.g., Jackson et al. 2002). Although few systematic experiments have investigated the temperature dependence of seismic attenuation in lower crustal materials, we infer that the lower crust locally cooled by shallow-angle subduction of the Philippine Sea slab is the primary cause of the low to moderate seismic attenuation values observed in this part of the crust. The inferred low-temperature conditions in the crust northeast of Biwa Lake are supported by deep cutoff depths of crustal earthquakes (Omuralieva et al. 2012).

The depth variations in the anelastic structures can qualitatively account for regional variations in the width of the high-strain-rate zone. A narrow (25–40 km wide) zone of surface deformation on the Echigo plain (Nishimura et al. 2012), east of the ISTL, is probably due to the shallow deformation of the thick, anelastically weakened sediment. In contrast, the NKTZ west of the ISTL, which has a width of ~100 km, is caused by high rates of anelastic deformation in the lower crust, as proposed by Iio et al. (2002, 2004). It is likely that long-term regional tectonic stresses promote anelastic deformation in the weakened lower crust and that strain rates enhanced by the deformation reduce grain size of minerals (Platt and Behr 2011). The highly attenuated lower crust observed beneath the NKTZ west of the ISTL is probably affected by grain size reduction due to localized, high-strain-rate deformation, because grain size reduction enhances seismic attenuation (Jackson et al. 2002). One important implication of our attenuation model is that surface deformation northeast of Biwa Lake, where the lower crust shows moderate to low attenuation, would differ from that of other parts of the NKTZ, because marked anelastic deformation of the lower crust is unlikely to occur in this region.

Spatial variations in the weakness of the lower crust would cause spatial gradients in anelastic deformation rates, and stress could be consequently concentrated in the overlying brittle layer. The occurrence of many M ≥ 6.5 earthquakes in the upper crust, below which anelastic properties in the lower crust vary sharply (Fig. 4a), may be one of the manifestations of high stress concentration in the brittle upper crust. While it is considered that reduction in the effective normal stress on a fault as a result of the supply of overpressurized fluids plays key roles in facilitating crustal earthquakes (e.g., Sibson 1992; Hasegawa et al. 2005; Yoshida et al. 2014), our observations add growing body of evidence that anelastic deformation in the lower crust is equally important in seismogenesis in the upper crust.

Conclusions

This study characterizes seismic attenuation structures in central Japan and discusses spatial relationship between attenuation structures and surface deformation along the NKTZ. The anelastically weakened lower crust west of the ISTL promotes surface contraction over a region ~100 km wide, while anelastic deformation in the shallow, thick sedimentary layer west of the ISTL causes a narrow region of surface deformation. Our observations can explain qualitatively the regional variations in the width of the high-strain-rate zone across the ISTL. The attenuation structure revealed in this study provides practical and important constraints on quantifying subsurface deformation, which is essential to constructing realistic models of seismogenesis in high-strain-rate zones.