Seismological Structures on Bimodal Distribution of Deep Tectonic Tremor

Deep tectonic tremors occur at the downdip extent of the seismogenic zone due to fluid processes. Beneath the northeastern Kii Peninsula, southwestern Japan, there is an along‐dip bimodal distribution of tremor. However, no constraint exists on the structures controlling that distribution. We extract detailed seismological structures from multi‐band receiver functions and evaluate conditional differences in the distribution. To achieve high resolution images along the plate interface, we utilize records of regional deep‐focus earthquakes from the Pacific slab. Cross‐section images show the subducting oceanic plate with depth‐dependent phases along the bimodal distribution, revealing a conspicuous plate interface at the updip portion and an inconspicuous interface below the mantle wedge at the downdip portion. This indicates that episodic tremors occur in the high pore‐fluid plate interface below the impermeable forearc crust, and that continual tremors occur at the permeable mantle wedge corner, owing to continuous fluid supply from the oceanic crust.

pore-fluid pressure is high in the subducting crust (e.g., Kodaira et al., 2004), and LFEs were located beneath the mantle wedge corner due to the fluid infiltration into the wedge corner (Kato et al., 2010). In Cascadia, however, some LFEs and tremors are located shallower than the forearc Moho (e.g., Kao et al., 2009;Plourde et al., 2015;Royer & Bostock, 2014) and are thought to occur around the forearc crust by updip fluid transport with quartz precipitation (Hyndman et al., 2015;McCrory et al., 2014). In both cases, fluid transport properties are key factors of slow earthquakes around the deep plate interface (e.g., Peacock et al., 2011).
In Nankai, the oceanic Philippine Sea (PHS) plate is subducting beneath the continental Eurasian plate along the Nankai trough (Figure 1), for which multiple models of the PHS plate have been proposed using SAWAKI ET AL.   (NE-Kii). The boxed region K1 denotes the study area used for creating cross-section receiver functions (RF) images. Purple dots are the epicenters of tectonic tremor Obara et al., 2010). The black circles denote epicenters of very-low-frequency earthquake (VLFE) estimated by Ito et al. (2009). The gray crosses are epicenters of low-frequency earthquake (LFE) (Katsumata & Kamaya, 2003). We define the episodic tremor as occurring in the region within the green box and the continual tremor in the region within the orange circle . Philippine Sea (PHS) and EU correspond to the PHS plate and the Eurasian plate, respectively. Black dashed lines are the depth contours of the PHS-EU boundary (Hirose et al., 2008). active or passive sources (e.g., Baba et al., 2002;Hirose et al., 2008;Matsubara et al., 2008). One of the most powerful methods to identify the geometry of the subducting plate is using receiver functions (RF; Ammon, 1991), which can extract P-to-S converted waves on subsurface velocity boundaries, and identify oceanic slab geometry by identifying the oceanic Mohorovičić discontinuity (Moho) or the subducting plate interface.  calculated low-frequency RFs (<0.6 Hz) and constrained the oceanic Moho shape in Nankai and found that most of the tremor epicenters were distributed along isodepth contours of the plate interface beneath the Kii Peninsula and Shikoku island at a depth of 33 ± 3 km (Figure 1). This implies that tectonic tremor would occur under similar pressure conditions across the entire tremor source area, assuming that all tremors occurred along the subducting plate interface (e.g., Shelly et al., 2006).
Recent studies of high-frequency RFs have identified a low-velocity layer in the upper oceanic crust in Nankai, which is in a highly hydrous state (e.g., Akuhara et al., 2017;Akuhara & Mochizuki, 2015). Beneath the Kii Peninsula, the basaltic oceanic crust subducts to a depth of 50 km and is a low-velocity layer (Hori et al., 1985). Akuhara et al. (2017) determined the thickness of the low-velocity layer to be less than 1-2 km at depths of around 20 km and revealed the presence of a fluid-rich sediment layer beneath the overriding plate. In Cascadia, Audet et al. (2009) showed a strong negative velocity contrast at the top of the oceanic crust extending downdip to the region of ETS beneath a lower continental crust. Besides, mafic gabbro rocks that constitute the lower forearc crust have a permeability of around two orders of magnitude lower than that of the serpentinite that is altered from the mantle peridotite (Katayama et al., 2012) and prevent geofluid from migrating upward (Audet et al., 2009;Peacock et al., 2011). Thus, the strong velocity contrast at the subducting interface is also representative of a low-permeability seal of overlying crust capable of raising pore-fluid pressures to near-lithostatic conditions along the plate interface, which may enable ETS at depths up to 35 km (Peacock et al., 2011).
A significant finding of tremor distribution has been made by Obara et al. (2010) in that an along-dip bimodal distribution of tectonic tremor was identified in the northeastern Kii Peninsula (NE-Kii), namely: an updip episodic tremor and a downdip continual tremor along the plate interface ( Figure 1). The updip tremor is laterally aligned along the isodepth plate interface and comprises bursts of tremor accompanied by geodetically detectable SSEs (episodic tremor). In contrast, continual downdip tremor activity is distributed in isolated small clusters at the downdip portion from the observed area of episodic events along the plate interface. Most of the epicenters for VLFEs and LFEs are located in the same area as that of episodic tremor at the updip and few are in the area of continual tremor at the downdip (Ito et al., 2009;Katsumata & Kamaya, 2003). Obara et al. (2010) also implied that continuous stress accumulation at the boundary between the free-sliding and the transition zones could be the cause of continual tremor, However, no constraint of seismological structures on bimodal tremor activities has been achieved.
Seismological structures capable of hosting slow earthquakes are still under debate, and structural conditions of bimodal tremor activities in NE-Kii are not well understood in that it is unclear as to which seismological structures host the tremors-the interface between oceanic and continental crusts, or the interface between the oceanic crust and mantle wedge material. Estimating the seismological structure around the source region of slow earthquakes will be highly beneficial toward constraining the generation mechanisms of such slow earthquakes and megathrust earthquakes. Therefore, we herein perform a multi-band RF analysis including regional deep-focus event records and show the seismological structure in high detail by targeting the source regions of ETS to estimate the characteristics of bimodal tremor emerging around the plate boundary.

Multi-Band Receiver Function Analysis
The RF method has been widely used to study the velocity discontinuity structures beneath seismic stations. Gaussian low-pass filters are usually applied to avoid numerical instability on the deconvolution operation, but the depth resolution with RF is attributed to the frequency contents of incident waves because high-frequency waveforms with shorter S-wave wavelengths  S will reveal smaller-scale structures (e.g.

10.1029/2020GL092183
propagating P-to-S converted waves from horizontal boundaries as , for the ratio of P and S-wave velocities   1.75. Assuming the general corner frequency of 0.6 Hz, the separation is around 3.6 km for the S-wave velocity of 3.7 km/s. We note that the resolution of 3.6 km is not sufficiently high to investigate detailed seismological structures around ETS activities considering the thrusting on the plate interface (Shelly et al., 2006) with confined shear zones, although it is sufficient to obtain phases of the oceanic Moho discontinuity and the plate interface. Therefore, we calculate RFs with two different corner frequencies of 0.6 and 2 Hz and refer to this approach as multi-band RF analysis. We use the low-frequency RFs to identify primary structures and the broader-band RFs for detailed structures. RFs are calculated by the extended-time multitaper ( Figure S1) technique (Helffrich, 2006;Shibutani et al., 2008), which was designed to carry out impartial estimation in RF amplitudes for the entire duration of the analyzed waveform (supporting information Text S1).
We investigate seismological structures of the subducting PHS plate and the overriding plate using seismic recordings of P-coda waves from teleseisms and regional deep-focus earthquakes. The direct P and the P-coda waves are recorded in the first portion of a teleseismic seismogram with the epicentral distance between 30° and 90°, mainly in the vertical component due to the near-vertical incidence ( Figure 2a and Figure S2).

Regional Deep-Focus Earthquakes in the Pacific Slab
Beneath southwestern Japan, two oceanic plates are presently subducting below the Eurasian plate. One is the PHS plate, subducting just below the Eurasian plate and causing a relatively large number of ordinary earthquakes and several types of slow earthquakes by interacting with the overriding continental plate. The other is the Pacific (PAC) plate, subducting further below the PHS plate at a greater angle. Many deep-focus intraplate earthquakes are occurring in the PAC at the Izu-Bonin subduction zone ( Figure 2b) and showing the double seismic zone (e.g., Iidaka & Furukawa, 1994), while epicenters of deep-focus earthquakes are aligned linearly from northwest to southeast. Generally, the regional-distance earthquakes should be excluded from the RF analysis because their seismic signals contain reverberation phases (e.g., Park & Levin, 2001). However, since rays from regional deep-focus earthquakes propagate in more vertical fashions to the surface, ray-paths of regional deep events can more closely resemble those of teleseismic events, as used in RF analysis.
Southwestern Japan is a suitable region with available seismic records of regional deep-focus earthquakes for application to the multi-band RF analysis. A large number of deep-focus earthquakes in the PAC slab regularly occur below the Kii Peninsula at depths of more than 200 km ( Figure 2b). Ray slowness values of regional deep earthquakes are not very different from those of ordinary teleseismic events ( Figure S2). In fact, seismograms of radial and vertical components resemble each other ( Figure S4), which demonstrates that we can calculate RFs from regional deep earthquakes as teleseismic RFs. Furthermore, the regional deep events tend to occur closer to seismic stations than teleseismic events. Thus, waveforms suffer less energy loss by the attenuation effect at higher frequencies. With higher signal-to-noise ratios than teleseismic seismograms at the same magnitude, we can raise the number of traces used in RF analysis, owing to the regional deep-event records with smaller magnitudes. Notably, deep-focus earthquakes occurring under the Kii Peninsula are distributed linearly from the northwest to the southeast, which is similar to the dip azimuth of the PHS plate (Hirose et al., 2008) (Figures 1 and 2b). Thus, we took advantage of the linear distribution of such deep earthquakes for the two-dimensional (2D) mapping along the dip direction of the PHS plate.

Data
We used three-component seismograms of regional deep-focus events and teleseisms, observed at Hi-net boreholes. In this study, a regional deep-focus earthquake was defined by an event that occurred in the PAC slab deeper than 200 km and within the epicentral distance of 10°. Based on this definition, we selected 74 regional deep-focus events with epicentral distances up to 10° and magnitudes over 5.3 from February 2005 to December 2018 (Figure 2b), following selection procedures (supporting information Text S2). We also selected 78 teleseisms located at the epicentral distance from 30° to 90° with magnitudes over 6.6 from January 2006 to July 2015 ( Figure 2a). Events that occurred far below the Sea of Okhotsk within the epicentral distance of ∼23° were included in teleseisms because multiple phases for P waves did not appear when the source depth was near or over 500 km depth and the epicentral distance was greater than 22° ( Figures S3d-S3f). We applied a bandpass filter from 0.05 to 50 Hz when removing the instrumental response and corrected the horizontal components using sensor orientation values estimated by NIED.

Comparison of Stacked Receiver Function Traces
We calculated teleseismic and regional deep-event RFs for a set of locations on NE-Kii (Figures 2c-2h and Figures S5-S10). Teleseismic RFs show two significant phases: positive phases around 4.5-5.5 s and negative phases around 4 s of lapsed time from initial P-wave arrivals (Figures 2c and 2f), which correspond to depths of around 36-44 and 32 km, respectively. The depths at the P-S converting point for positive and negative phases were consistent with those of the oceanic Moho  and the plate interface (Hirose et al., 2008), respectively. The positive phases around 3.5-4 s (28-32 km depth) suggest a continental Moho (Figures S5, S6, and S8). RFs from regional deep events only (hereafter deep-event RFs) were clustered in narrow ranges of back azimuth (Figures 2d and 2g, Figure S2). The deep-event RFs from 132° to 168° in back azimuth show clear negative phases around 3.7 s, similar to teleseismic RFs, indicating the presence of the PHS interface. However, the deep-event RFs from 324° to 348° in back azimuth that were characteristic of deep-focus events exhibited an unidentified split in negative phases around 4-5 s and very deep oceanic Moho phases around 6 s. RFs including both the teleseismic and deep-focus events (hereafter mixed RFs) improved the coverage of back azimuth, and the oceanic Moho phases with positive amplitudes show back-azimuthal dependency of delay time and amplitude (Figures 2e and 2h). Shiomi and Park (2008) verified the relationship between the back azimuth dependency and a dipping structure of the conversion plane. Our result implies a north-dipping oceanic Moho, which represents northward dipping when compared with the oceanic Moho geometry by   (Figure S11). Furthermore, broader-band RFs (2 Hz) exhibited many sharp phases compared with the low-frequency RFs (0.6 Hz). At downdip for back azimuth from 252° to 36°, the oceanic Moho phases split into two or three peaks, while the PHS boundary phases merely sharpened. The time separation of two peaks on the high-frequency RFs corresponds to a depth difference of ∼1 km. Thus, the simple sharp phases from the PHS boundary represent a sharp velocity transition on a thin subduction interface (<1 km thickness), while splitting on the oceanic Moho phases implies a gradual downward velocity increase with a thick oceanic Moho (2-3 km thickness; Nedimović et al., 2005).

Cross-Section Image
We estimated subsurface structures from a two-dimensional cross-section of multi-band RFs along a survey line, which crosses the source regions of episodic tremor and continual tremor ( Figure 3). We applied common-conversion-point (CCP) stacking of RFs to obtain cross-section images-a back-projection method in which RF amplitudes are properly plotted along a depth-converted raypath. We used the JMA2001 one-dimensional velocity model (Ueno et al., 2002) for the depth conversion. The corner frequencies of 0.6 Hz (low frequency) and 2 Hz (high frequency) were selected for comparing the cross-section derived from mixed RFs to that from teleseismic RFs. The conventional low-frequency image from teleseismic RFs ( Figure 3b) shows a positive signal of the continental Moho at 30 km depth (      20 km X ) and a continuous positive phase dipping in the northwest direction at 30-50 km depth, which is over 5 km deeper than the depth contour of the plate interface (Hirose et al., 2008). This demonstrates the velocity increase at the oceanic Moho from the crust to the mantle. A negative phase dipping to the northwest at 20-45 km depth shows the plate interface. However, this phase is discontinuous. The narrow range of incident angles for teleseismic rays led to the lateral discontinuity of target phases.
In contrast, low-frequency images, including deep-event RFs (Figure 3c), show more continuous and conspicuous phases than the ordinary RFs using only teleseismic data (Figure 3b). The negative dipping phase is the PHS interface, which is consistent with the isodepth contours of the plate interface (Hirose et al., 2008). Deep-event RFs with a wider range of slowness and an aligned distribution of events contribute to the clarification of phases by widening the range of trace packets under each seismic station and filling more grids during the CCP stacking procedure. A well-imaged oceanic Moho with positive amplitudes was confirmed at −50 < X [km] < 50 by the improved continuity of RF phases showing the oceanic plate.
High-frequency RFs (Figure 3d)  ). The interpretation of each phase is shown in Figure 3e. The depth-dependent plate interface with negative amplitudes (blue) is observed along with the oceanic or continental Moho with positive amplitudes (red). In general, RF amplitudes are dependent on the shear-wave velocity contrast (Audet et al., 2009) and the sharpness of the boundary thickness (Levin et al., 2016;Li et al., 2017). These SAWAKI ET AL.  The vertical axis denotes the depth conversion of each trace from the delay time by 2D stacking method. The purple dots denote epicenters of tectonic tremor from the NIED catalog Obara et al., 2010). The black circles denote epicenters of very-low-frequency earthquake (VLFE) estimated by Ito et al. (2009). The gray crosses are epicenters of LFE (Katsumata & Kamaya, 2003). ET and CT are abbreviations of the episodic tremor and the continual tremor. The JMA2001 1D velocity model (Ueno et al., 2002) is applied in depth conversion. If a seismic station is over 10 km distant from the Ps conversion point, the RF amplitude was attenuated in response to the distance. The grid size is 0.5  0.5 km, and the Gaussian smoothing parameter  is 1.5 km. results demonstrate a strong velocity contrast in a narrow region at the updip interface and a gradual velocity transition in the mantle wedge from the downdip interface to the continental Moho.
Interestingly, the depth-dependent image of the plate interface demarcates active regions of updip episodic events and downdip continual tremor (Figure 3e). The plate interface with strong negative amplitudes lies beneath most of the epicenters for episodic tremor-LFEs and VLFEs (        5 km 15 X ). However, the cross-section distance for epicenters of continual tremor corresponded to the entry point into the unclear interface around the horizontal distance of 20-30 km, indicating the mantle wedge.

Tremor in the Crust-Crust Zone
The broader-band RF image (Figure 3e) shows the negative phases for the plate interface, which is clear and sharp up to 35 km depth ( km 15 X along the cross-section) and less clear at the deeper portions (  15 X [km]). The distinctively conspicuous phase of the updip interface confirms that the upper oceanic crust is in a highly hydrous state, as previous studies have noted (e.g., Akuhara & Mochizuki, 2015;Kato et al., 2010). The sharper phase with a strong negative amplitude at the shallower portion shows that the seismic velocity steeply drops from forearc to oceanic crusts within a narrow area, which indicates a low-velocity layer in a hydrous state at the uppermost oceanic crust. Sediments would likely undergo compaction at depths less than 20 km (Hyndman & Peacock, 2003). Therefore, the sharp phase of the plate interface may not be associated with the possible existence of the fluid-rich sediment layer (Akuhara et al., 2017). A more plausible cause of the sharp phase is the basaltic oceanic crust (Hori et al., 1985;Hyndman & Peacock, 2003). If the potentially large velocity contrast at the thin subducting interface represents the presence of near-lithostatic pore-fluid pressures (Audet et al., 2009), the over-pressured basaltic oceanic crust should likely extend to depths of 35 km depth beneath NE-Kii as well as beneath Cascadia (Peacock et al., 2011). The depth variation of high-frequency RFs can be interpreted as the difference in the permeability along the plate interface, and it is likely that the strong negative phase at the updip portion represents a low-permeability interface (e.g., Audet et al., 2009). The low-permeability seal probably causes an increase in pore-fluid pressures on the plate interface, which could lead to the high level of activity of short-term SSEs or LFEs (Nakajima & Hasegawa, 2016;Nishimura et al., 2013) in NE-Kii. Hypocenters of LFEs are well mapped at the upper oceanic crust (Figure 4), which is nearly consistent with the assumption that LFEs occur near plate interfaces (Shelly et al., 2006). Episodic tremor also occurs around the interface between the continental crust and oceanic crust under the condition that sufficient geofluid is confined, thereby assuming that the tremor comprises a swarm of LFEs (Shelly et al., 2007). High tremor energy rates and the large slip amount of SSEs in NE-Kii  may support the hypothesis that frequent short-term SSEs break the permeability seal and allow for abundant built-up fluid along the interface to migrate upward, leading to tremor generation (e.g., Gosselin et al., 2020;Nakajima & Uchida, 2018).

Tremor in the Crust-Mantle Area
In contrast, downdip continual tremor corresponds to the mantle wedge region with the broad image of the plate interface ( Figure 3). The decrease in amplitude of the broad phases of the slab interface at the downdip portion shows that the velocity transition is gradual at the mantle wedge corner. When taking into account that the permeability of the mantle peridotite is higher than that of the crustal gabbro (Katayama et al., 2012), the higher permeability of the slab interface is facilitated (e.g., Audet et al., 2009). This indicates the mantle wedge is in a hydrous state due to fluid drainage from the plate boundary and produces serpentinite (e.g., Hacker et al., 2003). Thus, we believe that a possible source region of the continual tremor is the partially serpentinized mantle wedge corner. Obara et al. (2010) mentioned that the continuous stress concentration and weak frictional strength at the boundary between the free-sliding and the transition zones at the wedge corner may cause activation of continual tremor. Assuming the fluid source is the eclogitization of the oceanic crust at the deeper portion (e.g., Hacker et al., 2003), the slab-derived fluid is stably supplied into the shallower part of the mantle wedge. Thus, the continuous supply of fluid to the wedge corner could cause continual tremor activities. If a large volume of mantle wedge can be hydrated, the pore-fluid pressure is unlikely to rise suddenly, which would account for minor and continual activities.

Comparison with Cascadia and Prospects for Spatio-Temporal Variations
In Cascadia, Wech and Creager (2011) showed gradual transitions of tremor activity from episodic to continual at greater depths, and Idehara et al. (2014) revealed some clusters of continual tremor at downdip of the tremor band. The forearc mantle corner passes through the center of the tremor band (McCrory et al., 2014), suggesting that continual tremor occurs at the wedge corner and that some of this episodic tremor could occur beneath the forearc crust in Cascadia, as well as in NE-Kii (from our results). This is because the peak occurrence of episodic tremor coincides with the mantle wedge corner (Audet et al., 2010). Unlike Tokai (Kato et al., 2010) and NE-Kii, the inverted Moho has been reported in Cascadia (e.g., Bostock et al., 2002), which may suggest that hydrous conditions at the downdip portion differ. Further observations and analyses are required to clarify the differences in geophysical properties. Importantly, our high-frequency analysis with regional deep events contributes to the detection of detailed imagery of the over-pressured oceanic crust, without requiring dense-network temporal observation data. Therefore, this method may also be applied to other regions in the Nankai subduction zone, such as Tokai and Shikoku, to assess regionality. However, it should be noted that the cross-section image presents an averaged overview of the longer-term situation, rather than periodical characteristics of slow earthquake activities. Gosselin et al. (2020) illustrated the temporal variation in seismic velocity during ETS in Cascadia. However, additional teleseisms and regional deep events would be required to assess the temporal variation in structural conditions along the interface during the occurrence of slow earthquakes. In this manner, considering the use of temporal observation data may help enhance the spatial coverage. Investigating the spatio-temporal variation of RF phases is crucial for evaluating pore-fluid conditions and the occurrence of slow earthquakes or megathrusts. SAWAKI ET AL.   (Katsumata & Kamaya, 2003). The green ellipse denotes assumed regions where episodic tremor and very-low-frequency earthquakes (VLFEs) occurred. The orange ellipse denotes assumed regions of continual tremor. The black rectangle shows the region of the largest slip amount for short-term slow-slip events (SSEs) (Nishimura et al., 2013).

Conclusions
We performed multi-band receiver function analysis using Hi-net seismograph records on NE-Kii, in southwestern Japan. Regional deep earthquakes that occurred in the PAC slab were used to calculate high-frequency RFs. The wider range of the rays made the cross-section images more continuous. Cross-section images of RFs revealed the subducting oceanic crust over a positive phase of the oceanic Moho, thereby illustrating a conspicuous negative phase of the plate interface in the shallower portion and an obscure structure of the mantle wedge in the deeper portion. The conspicuousness of the updip interface confirmed that the upper oceanic crust is under fluid-saturated conditions below the impermeable forearc crust. The along-dip difference of the interface phase corresponded well with the along-dip bimodal distribution of tectonic tremor observed by Obara et al. (2010). The present results imply that updip episodic tremor, LFEs, and VLFEs occur around the high pore-fluid plate interface below the impermeable continental crust, and also show that the downdip continual tremor occurs at the hydrous mantle wedge corner. This implication provides new insights into the structural diversity of slow earthquake source regions in the Nankai subduction zone. Further studies are required to assess the heterogeneity and temporal variations in seismological structures around the plate interface in response to the occurrence of short-term or long-term SSEs and episodic slow events.