4.1 Location and spatial extent of structural anomalies
First, we evaluated how the structural heterogeneities, estimated by previous studies, were reconstructed by our velocity model. In the overriding Amur Plate, high-density and high-seismic velocity body called the Shionomisaki igneous complex (SIC) exists just beneath the toe of the Kii Peninsula (Honda and Kono 2005; Kodaira et al. 2006; Qin et al. 2021). In both the average P- and S-velocity perturbations within a 5-km-thick layer just above the plate interface (Fig. 9a), the high-velocity zone is clearly imaged. In particular, the center of the high-velocity zone (A in Fig. 9a), located just north of the epicenter of the 1946 Nankai earthquake, was attributed to the highest gravity anomaly (Honda and Kono 2005). This finding leads to the conclusion that this high-velocity region represents the spatial extent of the SIC.
Based on the interpretation of Kimura et al (2014), this high-velocity region reaches the vicinity of the wedge mantle. The high-velocity zone between 30 and 40 km of the depth contour of the plate interface (B in Fig. 9a; 34° N, 135.5° E) corresponds to the landward mantle because the depth of Moho discontinuity in this area is shallower than 30 km (Matsubara et al. 2017). Although areas A and B were seemingly separated in the S-wave velocity model, their boundary was unclear in the P-wave velocity model (Fig. 9a). Arnulf et al. (2022) also investigated the spatial extent of SIC (called the “Kumano Pluton” in their paper) based on the high (> 6.5 km/s) P-wave velocity zone. The P-wave velocity contour of 6.5 km/s from our model at 15 km depth (green line in Fig. 10a) and the spatial extent of the Vp > 6.5 km/s areas within the overriding plate along Y-axis direction (− 30 ≤ X ≤ 40 km; black lines in Fig. 10a) are consistent with their result. Note that the Vp > 6.5 km/s areas of our model located south of the overlapped area with Arnulf et al. (2022) correspond to the oceanic crust of the subducting plate. Based on this consistency, we determined that the southern limit of the SIC was imaged by our model. However, our result is inconsistent with theirs at north of 34° N (Fig. 10a). In addition, it is difficult to estimate the spatial extent of Vp > 6.5 km/s areas along the northern side (X ≤ −40) owing to the extent of the wedge mantle. Therefore, the north extent of the SIC was not determined in this study.
In certain subduction zones, the relationship between the structural heterogeneity of the upper plate and the seismic slip distribution has been previously investigated, thereby indicating that the high-velocity zone corresponds to the large coseismic slip zone (e.g., Zhao et al. 2011). For the area of interest in this study, we compared the structure with the distribution of interplate coupling strength (Yokota et al. 2016) and did not find a significant relationship, thereby agreeing with Yamamoto et al. (2017). Note there are various models for the slip distribution of the 1944 Tonankai and 1946 Nankai earthquakes whose characteristics were not similar. For example, the largest slip areas of the 1946 Nankai earthquake in Baba and Cummins (2005) do not overlap with those reported by Murotani et al. (2015). Thus, the comparison between our velocity structure and the slip distribution of past megathrusts was not conducted in this study.
Within the slab, the existence of subducted topographic anomalies, such as seamounts (Kodaira et al. 2000) and the Paleo-Zenith Ridge (Park et al. 2004), has been previously reported. We attempted to identify their location from our velocity model. As these features are characterized by thick crust, they are illustrated as low-velocity regions in the deep structure of the slab relative to the surrounding mantle. Figure 9b shows the average velocity perturbation within a 5-km-thick layer between 10 and 15 km beneath the plate interface. We identified two low-velocity zones (C and D in Fig. 9b). Here, low-velocity zone C corresponded to the subducted seamount (Kodaira et al. 2000), and its spatial extent was nearly the same as that from Yamamoto et al. (2017). Low-velocity zone D was located close to the area where the Paleo-Zenith Ridge subducted (Park et al. 2004), thereby indicating no overlap with the resolved area from Yamamoto et al. (2017). Although the location of low-velocity zone D from P-wave perturbations was slightly south of that of the S-wave perturbations, this low-velocity zone was attributed to the existence of the Paleo-Zenith Ridge. Although smaller-size subducted seamounts than our grid spacing can theoretically exist, we could not find other similar low-velocity zones in the study area. Thus, it is reasonable to suggest that there was seemingly no other subducted seamount or ridge as large as them.
Arnulf et al. (2022) observe a significant (Vp = 6.5–7.5 km/s) low-velocity anomaly within the subducting mantle around the aftershock region of the 2004 off-Kii earthquake and interpreted it as a combination of serpentinization and enhanced porosity from bending stress. Although their model has the lowest point of P-wave velocity at 30 km depth, there is no such low-velocity zone in our model (Fig. 10b). Note that the low-velocity zones imaged between the 20 and 30 km contour of the plate interface depth correspond to the oceanic crust. Instead, the location of the significant low P-wave area (Vp < 6.5 km/s) from their model was spatially close to the subducted Paleo-Zenith Ridge, where we found the low-velocity anomaly 10–15 km beneath the plate interface (Fig. 9b). That is, the depth of the low-velocity anomaly obtained in our study was ~ 10 km shallower than that of Arnulf et al. (2022). We consider this difference to be due to the poor hypocenter location accuracy in the dataset of Arnulf et al. (2022). As shown in Fig. 6, there is a large difference in hypocenter depth of offshore seismicity between this study and the JMA unified catalog, which Arnulf et al. (2022) applied. Because their analysis was performed by earthquakes whose calculated depths were more than 10 km deeper than their actual depth, the low-velocity region shown in Arnulf et al. (2022) was imaged at a deeper position than the actual location.
4.2 Spatial relationship among interplate earthquakes, interplate coupling, and slow earthquakes
We confirmed the occurrence of interplate earthquakes in the study area (Fig. 8). The comparison with the thermal structure (Oleskevich et al. 1999) revealed that most interplate earthquakes occurred between 150 and 350 °C, thereby resonating with the range of seismogenic zone depth (Hyndman et al. 1997). In addition, they were seemingly located outside of high Vp/Vs (> 1.9) zone except for two areas, where a subducted topographic high was identified (Fig. 8c). The downdip limit of the Vp/Vs > 1.9 zone was located ~ 100 km from the trough axis in line with the downdip limit of the dehydration occurred within the sediment layer (Hyndman and Peacock 2003). The high Vp/Vs ratio around the plate interface normally indicates the existence of pore fluids and is attributed to the weak coupling zone as pore fluids decrease the effective normal stress (Moreno et al. 2014). Thus, interplate seismicity can be interpreted as occurring in areas with relatively less pore fluid and not fully creeping zones. However, the high Vp/Vs zone in this study area contained at least one strong interplate coupling area (Yokota et al. 2016), indicating that the relationship between Vp/Vs and coupling strength along the plate boundary proposed by Moreno et al. (2014) cannot be applied to the Nankai Trough. Thus, the existence of interplate microearthquakes cannot be simply attributed to the Vp/Vs distribution. Rather, there are other potential factors, such as topographic anomalies in the subducted plate (Nakamura et al. 2022) or thickness variation of the subducted sediment layer (Akuhara et al. 2017), that may control the occurrence of interplate seismicity in the Nankai trough.
To identify the areas where interplate earthquakes prevailed, we selected the earthquakes within 1 km of the plate interface (Nakanishi et al. 2018). On this basis, we defined 14 areas as active areas of interplate earthquakes (Fig. 11). These active areas were generally located outside the especially strong coupling zones (> 5 cm/yr; Yokota et al. 2016), except Area 11, where the Mw 5.9 off-Mie earthquake occurred on April 1, 2016 (Nakano et al. 2018b) (Fig. 11a). Two areas in the vicinity of the trough axis (Areas 13 and 14) corresponded to the aftershock area of the 2004 off southeastern Kii earthquakes.
At the next step, we examined the relationship between these 14 active areas and slow earthquakes. We did not find overlap with the SSE region in the Kii Channel (Kobayashi 2014). The segregation of slow earthquakes and regular earthquakes at the plate boundary has been previously demonstrated in the off Boso (Ito et al. 2019), Ryukyu Trench (Yamamoto et al. 2018; 2020), Hikurangi (Bartlow et al. 2014), and other areas. However, in the offshore, there was an area of overlap with the fault location of the SSE between 2017 and 2018 (Yokota and Ishikawa 2020) (Area 9; Fig. 11a). Moreover, three areas (Areas 8, 9 and 12) corresponded to the area where shallow VLFE was identified based on the offshore observation (Sugioka et al. 2012; Nakano et al. 2018a; Toh et al. 2020; Yamamoto et al. 2022). The overlapping areas were all located at the area, where the plate interface was located at the depth of ~ 7 km and was attributed to the subduction position of a seamount or its vicinity (Fig. 11a). The source location accuracy of VLFE was found to be lower (e.g., ~ 0.03° in horizontal and 1.3 km in depth for VLFE in Yamamoto et al. 2022) than that of the relocated hypocenters in this study (~ 0.3 km in both the horizontal and vertical direction in this study). Thus, a fine-scale analysis is challenging with the current methodology, but it is reasonable to suggest that several of the active areas of interplate earthquakes in the Nankai Trough subduction zone overlapped with the source areas of the slow earthquakes. Compared with the estimation of the VLFE location based on onshore data (Takemura et al. 2019), the active areas in Areas 6, 7, and 13 seemingly overlapped with VLFE (Fig. 11b). However, as the VLFE location, estimated by Takemura et al. (2019), was plagued by large uncertainty in the across-trough direction (see Fig. S1 of Takemura et al. 2019), their result may not reveal spatial differences between Areas 6 and 8, Areas 7 and 9, and Areas 12 and 13.
Both slow earthquakes and microearthquakes along the fault have been previously thought to occur in the transition zone between the strong coupling seismogenic zone and fully creeping zone (e.g., Sholtz 2002; Obara and Kato 2016). The magnitude of completeness for microearthquakes in this study is less than 2 (Fig. 6), which is much lower than that for VLFE (~ Mw = 3; Nakano et al. 2018a, b; Yamamoto et al. 2022). In addition, the improvement of the accuracy in the hypocenter relocation using the offshore seismic network with a 3-D velocity model enabled the distinction between interplate and intraplate earthquakes. As both the detection capability and location accuracy of regular earthquakes were superior to those of shallow slow earthquakes, a comprehensive detection of interplate microearthquakes is valuable for monitoring the current stick–slip status of the strong coupling zone in the Nankai Trough.
4.3 Temporal variation of interplate earthquake activities
For each active zone, we examined the temporal changes in seismic activity and the number of integrated earthquakes at the plate boundary ± 25 km by using the entire relocation result from this study (Fig. 12). Because the network-based observations exhibited temporal variability, we focused on the temporal distribution after March 2015 at the area close to the DONET2 network (Areas 1–10) and after 2012 at the area within the DONET1 network (Areas 11–14), respectively. No remarkable changes were identified in Areas 1–5, which were close to land, or in Areas 10, 13, and 14, which were close to the trough axis. However, in other areas, characteristic activity was identified, even while considering the periods when the dataset was incomplete (gray masked periods in Fig. 12).
In Areas 6, 8, and 9, the cumulative number of earthquakes increased sharply in 2018. Seafloor crustal displacement observations in this area suggest that SSEs occurred between 2017 and 2018 (Yokota and Ishikawa 2020). As the activations in these areas during the SSE period seemingly occurred between − 5 and 5 km in relative depth (Fig. 12), we examined the temporal variation of five areas close to this SSE region. Therefore, we narrowed the time window to two years and the epicenter to ± 5 km from the plate interface (Fig. 13a). As a result, swarm-like activation of interplate seismicity in Areas 8 and 9 was identified in March and May 2018, respectively. Although detailed discussion is not possible owing to the lack of an available VLFE catalog from offshore observation corresponding to this SSE, VLFE activity has been previously reported by Takemura et al. (2019), with most activity near the western side (Area 8) in March and the eastern side (Area 9) in May (black line in Fig. 13a).
However, the activation of interplate seismicity seen in Area 6 occurred approximately six months later than the timing of the VLFE activity observed by Takemura et al. (2019). Moreover, VLFE activity was not accompanied by the interplate seismicity in Area 7, and no swarm-like seismicity or VLFE activity was identified in Area 10. The onshore network-based VLFE activity periods were the same between Areas 6 and 8 and between Areas 7 and 9. This finding might indicate the previously indicated large location uncertainty in across-trough direction. Thu, it is necessary to determine the source locations of the VLFE by using the seafloor observation network during this activity period to determine the difference in activity between regular earthquakes and VLFE and to understand the state of stick–slip behavior in the shallow part of the subduction zone.
In Area 11, a sharp increase in the cumulative number of earthquakes related to the 2016 off-Mie earthquake was identified (Fig. 12). Unlike the other areas, this area was located within the strongly coupled zone (Yokota et al. 2016), where we also observed several seismic activities occurring near the plate boundary before the 2016 earthquake (e.g., in 2013 and 2015). This indicates that the 2016 off-Mie earthquake did not occur at the seismic gap but in an area of background seismicity. The investigation of the location of background interplate seismicity within the strongly coupled area can unravel a potential nucleation point for moderate-size or, even large interplate earthquakes.
Area 12 seemingly had no significant fluctuations during this period, except a slight increase in the cumulative number of earthquakes in the first half of 2018 (Fig. 12). However, a deeper investigation of Area 12 suggests that while intraslab earthquakes (> 5 km deeper than plate interface) generally prevail, earthquakes occur at relatively shallow depths close to the plate boundary (within 5 km from the plate interface) in four episodes during the observation period: the first half of 2014 (episode-1), the second half of 2015 (episode-2), the first half of 2017 (episode-3), and the first half of 2018 (episode 4) (Fig. 13b). The first and third episodes exhibited fewer earthquakes, while the second and fourth episodes experienced more earthquakes. Based on borehole pore pressure changes, recurrent SSEs were shown to occur in the area adjacent to the western region (Araki et al. 2017; Ariyoshi et al. 2021) (Fig. 11a). Of these events, SSEs with large amounts of slip were reported during episodes 2 and 4. VLFE activity was also confirmed by the offshore observation data during episode 2 (Nakano et al. 2018a), although no VLFE was observed during episode 4.
This study proved the existence of interplate earthquake swarms, synchronized with SSEs at Areas 8, 9, and 12 in the Nankai Trough. Such earthquake swarms, associated with SSE, have been previously reported in subduction zones such as off Boso (Hirose et al. 2014; Fukuda 2018), Ecuador (Vaca et al. 2018), and Hikurangi (Bartlow et al. 2014), where slow-slip stress loading and stress triggering outside the SSE region were the main drivers of earthquake swarms. In this study, the seismic swarm activity in Areas 8, 9, and 12 may have been activated by the SSE, because the timing of the SSE and VLFE activity was the same.
Near the subducted seamount in the Hikurangi forearc, similar regular earthquake swarms collocated with slow earthquakes were reported, but they were located within the overriding plate (Shaddox and Schwartz 2019). The observed waveforms near the swarm areas generally indicate an increasing difference between S- and P-wave arrivals with respect to hypocenter depth (Fig. 14). Considering this result and the accuracy of our hypocenter depth in this study (~ 0.3 km), we suggest that seismic swarms observed in this study are located within the slab and across the upper plate and include seismic activity at the plate boundary, rather than being located solely within the overriding plate as in Hikurangi. The focal mechanisms of low angle thrust solutions should be a strong piece of evidence for interplate seismicity. Although we could estimate the focal mechanism for the 2016 off-Mie earthquake based on P-wave polarity data by applying the FOCMEC package (Snoke 2003) as a low angle thrust (Fig. 14b), focal mechanism solutions could not be obtained for swarm activities collocated with slow earthquakes owing to their small magnitudes. Further study by adding the information of the S-phase waveforms may allow us to estimate a number of their focal mechanisms.
In Area 8, seismic activity swarms occurred twice with an interval of ~ 10 days (Fig. 13a). This finding suggests that monitoring of seismic swarm activity around the plate boundary may lay the foundation for estimating the state of interplate stick–slip changes with higher temporal resolution. Furthermore, episode 4 in Area 12 (with the largest number of interplate seismic activity but composed of smaller magnitude than that in episode 2) was not accompanied by any VLFE (Fig. 13b). Based on the relationship between the seismic energy release and magnitude (Gutenberg and Richter 1956), the summation of the seismic energy release for the swarm activity during episode 2 was estimated as 2.53 × 108 N m, much larger than that during episode 4 (6.77 × 107 N m), respectively. This, in turn, indicates that the total energy release during swarm activity was one of the indicators of the magnitude of SSEs.
Note that in most previous studies, the SSE areas and seismic swarm activity areas did not spatially overlap but were rather adjacent to each other. In this study, the occurrence of swarm earthquakes within the area of SSE and/or VLFE activity was also confirmed, but large uncertainties were identified in the location of shallow slow earthquakes. Hence, the detection of seismic swarm activity along the plate interface can strengthen the accuracy, thereby improving the estimation of slow-slip fault models. Furthermore, if the number of detections can be increased, as in the case of Nicaragua (Thorwart et al. 2013), one can retrieve physical parameters on the fault, such as diffusion coefficients and porosity based on the swarm seismicity catalog. As the minimum magnitude of the earthquake swarms obtained in this study was very small, compared to previous studies, it is essential to take advantage of the offshore seismic networks. The use of seafloor observation networks and hypocenter determination by using 3-D velocity model should extend the prospects for estimating spatiotemporal changes of b-values focused on the plate interface in the Nankai megathrust seismogenic zone, which has only previously been discussed with regard to spatial changes based on the JMA catalog without distinction between interplate and intraplate earthquakes (Nanjo and Yoshida 2018).