Updip Fluid Flow in the Crust of the Northeastern Noto Peninsula, Japan, Triggered the 2023 Mw 6.2 Suzu Earthquake During Swarm Activity

An Mw 6.2 earthquake occurred in Suzu, northeastern Noto Peninsula, Japan, on 5 May 2023, followed by many aftershocks. Before this mainshock‐aftershock sequence, an intense earthquake swarm lasted in the vicinity for 2.5 years. Here, we estimated the rupture process of the Mw 6.2 mainshock and relocated >20,000 surrounding small earthquakes. The results show that systematic upward migration occurred via a complex network of faults in the preceding swarm period and that the mainshock rupture was initiated near the shallow end of the swarm earthquakes. The mainshock rupture propagated farther updip, followed by many aftershocks in the shallow extension. Upward fluid movement likely caused systematic upward earthquake migration from a depth of 18–5 km. The present results indicate the importance of monitoring swarm events since large (M > 6) and dangerous earthquakes can occur during such swarms.

• An M w 6.2 event initiated near the shallow end of a fault where the preceding earthquake swarm showed systematic upward migration • The M w 6.2 rupture propagated farther updip, followed by many shallow aftershocks on the fault • Upward fluid migration can trigger earthquakes as large as M w > 6 Supporting Information: Supporting Information may be found in the online version of this article.
fault network (Yoshida, Uno, et al., 2023).Previous studies have suggested that fluid movement at depth and the accompanying aseismic deformation are involved in this swarm (Amezawa et al., 2023;Nakajima, 2022;Nishimura et al., 2023;Yoshida, Uno, et al., 2023).The presence of a deeper low-seismic-velocity region and a distinct seismic reflector led to the inference that the upward migration of slab-derived fluids triggered this swarm (Yoshida, Uno, et al., 2023).
The 2023 M w 6.2 earthquake occurred at the northern end of the preceding swarm region and caused damage to surrounding residents and structures.As in this case, large earthquakes (M w > 6) generally cause more harm than earthquake swarms.Several past studies have suggested that the characteristics of in situ conditions may differ between swarms and large earthquakes (Mogi, 1963(Mogi, , 1967;;Utsu, 2002;Zaliapin & Ben-Zion, 2013); however, the present M w 6.2 event occurred adjacent to the preceding swarm.Clarifying the details of the occurrence of the M w 6.2 event may help elucidate the occurrence mechanisms of large earthquakes in general.One important factor is the spatial relationship between the source faults of the preceding swarm and the mainshock.To this end, we relocated the hypocenters of more than 20,000 M JMA > 1 earthquakes in this region before and after the mainshock and estimated the rupture process.

Earthquake Relocation
We relocated the hypocenters of 21,941 M JMA ≥ 1 earthquakes in the study region-listed in the Japan Meteorological Agency (JMA) unified catalog-from 1 March 2003 to 10 July 2023 (Figure S1 in Supporting Informa- tion S1), using the double-difference relocation method (Waldhauser & Ellsworth, 2000).This process is the same as that by Yoshida and Hasegawa (2018a).We assumed the 1-D velocity model proposed by Ueno et al. (2002).We used 491,960 P-wave and 490,032 S-wave differential arrival time data from the arrival time data in the JMA unified catalog.We also used 50.5 million P-wave and 65.8 million S-wave differential arrival time data derived from the waveform correlation analysis.Waveform data were obtained from Hi-net (NIED, 2019b) and F-net (NIED, 2019a) of the National Research Institute for Earth Science and Disaster Resilience (NIED) and the JMA (Figure 1a).

Rupture Process Inversion
We estimated the rupture process and coseismic slip distribution of the M w 6.2 event using the same procedure as Yoshida, Taira, et al. (2020).We first determined the apparent moment rate functions (AMRFs) of the mainshock by waveform deconvolution and then inverted them to retrieve the spatiotemporal distributions of the coseismic slip.
We estimated AMRFs using the time-domain deconvolution method of Ligorría and Ammon (1999), which employs the method of Kikuchi and Kanamori (1982).The data were velocity-type strong-motion waveforms (transverse components of S-waves) from NIED F-net (NIED, 2019a).We applied a low-pass filter (fourth-order Butterworth type) with a cutoff frequency of 0.7 Hz.Using a positivity constraint, we derived 21 AMRFs that explained more than 80% of the observed waveforms in variance (Figure S2 in Supporting Information S1).We adopted the waveforms of the March 2022 M w 4.7 earthquake as empirical Green's functions (EGFs) as it occurred close to the target event with a similar focal mechanism (pink beach-ball in Figure 1a).This EGF event was the only event that yielded more than 10 AMRFs.
The rupture process and slip distribution of the M w 6.2 event were derived from the AMRFs using the methods of Hartzell and Heaton (1983) and Ross et al. (2018).The assumptions are as follows.(a) The rupture initiates at the center of a rectangular model fault with a strike of 41° and dip angle of 40°.This geometry is based on the mainshock focal mechanism (Figure 1a) and relocated hypocenter distribution.(b) The rupture front propagates over the fault at a constant velocity V r = 0.8V s = 2.8 km/s, where V s = 3.5 km/s is the S-wave velocity.We set the fault length and width to 34.5 km and divided it into 25 × 25 subfaults.The local moment-rate function at each subfault is represented by the superposition of five triangular submoment rate functions.We determined the half-duration of this as t h = 0.7 s, based on the cutoff frequency of the low-pass filter (0.7 Hz).We assumed the same values for the damping factor and smoothing factor, and determined the specific values of λ = 2 based on the trade-off curve.We used the non-negative least-squares algorithm of Lawson and Hanson (1977) to obtain the solutions.

Results
We obtained relocation results for 21,693 events out of the 21,941 events.The relocation algorithm removed the remaining 248 events because their hypocenters were located above the ground surface or contained outliers in the differential arrival time data.The uncertainty in earthquake locations was evaluated via 50 bootstrap resamplings of differential arrival time data.The mean lengths of the 95% confidence intervals were 0.0014° in longitude, 0.0011° in latitude, and 0.17 km in depth.
Figure 2 shows the spatial distribution of the relocated hypocenters, indicating that the earthquake swarm preceding the 2023 M w 6.2 earthquake (red to green) occurred in multiple clusters and planes and exhibited migration behavior (Figure 2e-2i).This result is consistent with previous studies by Yoshida, Uno, et al. (2023), who analyzed earthquakes before the mainshock (between 2003 and September 2022).Figure 3 shows the relationship between earthquake occurrence time and depth for the five major earthquake clusters.(An enlarged view after the M w 6.2 event is shown in Figure S3 in Supporting Information S1.) Four of the cluster regions are identical to those of Yoshida, Uno, et al. (2023), with the swarm starting first in cluster S, followed by clusters W, N, and E (Figures 3b-3e).They are distributed separately in space.The N and E clusters were connected on the map after the June 2022 M w 5.2 earthquake, but are still separated in their three-dimensional locations.The fifth region (V) is north of them and contains the majority of aftershocks of the M w 6.2 mainshock, but there were few earth-quakes before the mainshock (Figure 3g).Most events before the mainshock were deeper than 10 km, showing upward migration in each cluster (Figures 3c-3f).
The hypocenter of the 2023 M w 6.2 mainshock is located in the northernmost part of the preceding swarm (Figure 2a) and near the shallowest end of the swarm earthquakes (Figure 2e).Earthquakes after the mainshock (blue in Figure 2) occurred further north of the mainshock hypocenter.These aftershock hypocenters were concentrated in a planar structure that became shallower toward the northwest, delineating the mainshock fault.This fault appears continuous with a fault zone of the previous earthquake swarm (Figure 2e; Yoshida, Uno, et al., 2023).Immediately after the mainshock, shallower earthquakes (depth <10 km) became more active.The early aftershock region expanded almost instantaneously (Figures S4c and S5 in Supporting Information S1; measured from the mainshock hypocenter), possibly reflecting afterslip propagation or pore pressure diffusion.
Figure 4; Figure S6 in Supporting Information S1 show the estimated coseismic slip distribution of the 2023 mainshock on the map and fault coordinate system, respectively.The results reproduced the observed AMRFs, with a variance reduction (VR) between the observed and theoretical AMRFs of 93.1%.The peak of the coseismic slip amount was located on the shallower northern side of the hypocenter.The rupture continued for approximately 10 s and extended updip from the hypocenter (Figure S7 in Supporting Information S1).Large aftershocks with M JMA > 5 occurred near the edge of the slip area.During the inversion process, we assumed V r / V s = 0.8.Changing the assumed V r /V s in the range of 0.7-0.9 made minor difference in the value of VR, indicating that the value of V r is not well constrained.However, using different values of V r /V s in this reasonable range does not significantly affect the relationship between the hypocenter and the slip peak (Figure S8 in Supporting Information S1).
Overall, the present sequence shows a persistent tendency for earthquakes to migrate upward during the preceding swarm and the mainshock-aftershock sequence.Although several earthquakes occurred within 1 km of the mainshock hypocenter within a day before, no clear change in the migration pattern or seismic activity level was observed before the mainshock (Figure S9 in Supporting Information S1).In addition, no clear anomalies were observed immediately before the mainshock (Figure S10 in Supporting Information S1) at a nearby GEONET GNSS station (ID:0253; cross in Figure 3a).

Discussion and Conclusion
The estimated spatiotemporal earthquake distribution and mainshock rupture process showed a persistent trend of upward migration throughout the sequence, as summarized in Figure 5. (a) The seismicity rate increased at the end of 2020, forming an earthquake swarm.The swarm migrated to a shallow depth via a complex fault network (depth from 20 to 12 km; Figure 3b), with the largest event being the June 2022 M w 5.2 event (black star in Figure 5).(b) Near the northern end of the fault zone of the 2022 M w 5.2 event, the M w 6.2 mainshock occurred.The mainshock hypocenter is located near the updip end of the southeast-dipping plane, and the rupture propagated further updip (depth from 12 to 9 km).(c) Aftershocks occurred mainly at shallower depths on the fault (depth from 10 to 5 km).
A southeast-dipping active fault (Suzu-Oki segment; Inoue & Okamura, 2010) exists near the aftershock area (Figure 5a).The aftershocks immediately below the surface trace were as deep as 6 km (Figure 5b), suggesting that the mainshock and swarm fault differs from the Suzu-Oki fault.The earthquake distribution shows that some events occurred on planes other than the mainshock fault, suggesting that multiple faults exist in the vicinity.Further north in this aftershock area, an M w 6.3 (M JMA 6.6) earthquake occurred in 1993, but the focal depth was not well determined because of its offshore location.Based on the JMA unified catalog, the hypocentral depth was approximately 2 km, suggesting that the earthquake may have occurred on an extension of the 2023 M w 6.2 mainshock fault (Figure 5b).However, the hypocenter of the 1993 event determined by Tsukuda et al. (1994) was approximately 15 km deep; thus, it is difficult to discuss the details of the relationship between the 1993 and 2023 events using current data.Yoshida, Uno, et al. ( 2023) suggested that preceding swarm of the 2023 M w 6.2 earthquake was caused by upward fluid movement.Their basis is (a) the upward migration of earthquakes via multiple planes, the presence of (b) a deeper low-seismic-velocity region, and (c) a distinct seismic reflector near the initiation point of the swarm activity.Earthquake migration itself can be explained by effects other than pore pressure and fluid migration, such as aseismic slips and earthquake interactions (Helmstetter & Sornette, 2002;Im & Avouac, 2023).Geodetic data analysis suggests that aseismic deformation occurred during this swarm (Nishimura et al., 2023), and the aseismic slip occurrence during fluid migration is consistent with predictions from numerical simulations of the response of a fault under fluid intrusion (Bhattacharya & Viesca, 2019;Eyre et al., 2019;Wynants-Morel et al., 2020;Yoshida et al., 2021).Recent observations support the occurrence of aseismic slip during fluid intrusion for both natural and fluid-injection-induced seismicity (Cornet et al., 1997;Danré et al., 2022;De Barros et al., 2020;Guglielmi et al., 2015;Hatch et al., 2020;Wei et al., 2015;Yoshida & Hasegawa, 2018b;Yukutake et al., 2022).Although these multiple influences likely contributed to the generation and migration of the present sequence, the fluid supply probably played a central role, given the above observations taken together.
The rising fluids and accelerated aseismic slip during the swarm probably contributed to the occurrence of the 2023 M w 6.2 earthquake near the shallow end of the swarm.Earthquake migration occurred not only in the updip direction but also in the fault strike direction (Figure 5a; Figure S11 in Supporting Information S1).Small earthquakes occurred near the mainshock hypocenter approximately 300 days before (Figure 5; Figure S4b in Supporting Information S1) but did not migrate toward the shallow side before the mainshock.This may indicate that the shallow side of the fault was less permeable and fluids and earthquakes could not migrate upward, similar to the fault-valve model (Sibson, 1992).The continued fluid supply may have increased the pore pressure, and the induced aseismic slip may have loaded the source area during this period, leading to the mainshock.The mainshock stress change and afterslip caused some aftershocks.Additionally, the mainshock rupture likely enhanced the permeability within the fault zone, promoting the upward discharge of fluids from deeper portions of the fault.
Increased seismicity rate and migration behavior are sometimes reported before a large earthquake, and these events are broadly referred to as foreshocks (Bouchon et al., 2013;Dodge et al., 1995;Kato & Ben-Zion, 2020;Matsumoto et al., 2021;Ruiz et al., 2017;Socquet et al., 2017;Yoshida, Taira, et al., 2020).Some foreshocks can be explained solely by the interactions between earthquakes (Ellsworth & Bulut, 2018), whereas others are accompanied by surface deformation that cannot be explained by earthquakes alone (Ruiz et al., 2017;Socquet et al., 2017).The effects of aseismic slip-nucleation process or independent-or occasional fluid diffusion have been proposed to be involved in them.This depiction is similar to that of the present Suzu sequence, but the present sequence is unique in that the earthquake migration before the mainshock occurred in an intricate network of faults.
Earthquake migration in intraplate swarm activity is often interpreted to represent fluid involvement, as is common with fluid-injection-induced seismicity, and is well explained by fluid diffusion models (Ohtake, 1974;Parotidis et al. (2003)).Fluids may also play an important role in some mainshock-aftershock sequences (De Barros et al., 2019;Miller et al., 2004;Nur & Booker, 1972;Ross et al., 2017;Yoshida, Uchida, et al., 2020).However, there are fewer clear examples of fluid contributions to mainshocks and aftershocks compared to swarm activity because afterslip and aftershock interactions are also strong candidates for the cause of aftershock migration (Cattania et al., 2015;Helmstetter & Sornette, 2002;Im & Avouac, 2023).The present observations show that the fluid supply, probably in combination with interseismic interaction and aseismic deformation, can trigger M > 6 earthquakes.This suggests the importance of monitoring swarm activities and fluid migration for time-dependent earthquake hazard evaluations.

Figure 1 .
Figure 1.(a) Study region (rectangle).Gray dots show the hypocenters of shallow earthquakes (depth <40 km) with the Japan Meteorological Agency magnitude M JMA ≥ 2.0 from 1 January 2003, to 10 July 2023.Beach-balls show the moment tensor solutions by F-net(Kubo et al., 2002) of earthquakes with M w > 5, with the green and pink beach-balls denoting the 2023 M w 6.2 event and the empirical Green's function event, respectively.The yellow beach-ball indicates the 1993 M w 6.3 earthquake from the GCMT catalog(Ekström et al., 2012).Blue crosses show the seismic stations used for relocation.The red line approximates the surface trace of the Suzu-Oki fault segment(Inoue & Okamura, 2010).(b) Area surrounding (a) (rectangle).The triangles show the F-net station used for rupture process inversion.(c) Magnitude-time diagrams in the study region.Blue and red circles with gray vertical bars indicate the magnitudes for M JMA ≥ 1 events before and after the mainshock, respectively.The black line denotes the cumulative number of earthquakes with M JMA ≥ 2 daily.

Figure 2 .
Figure 2. Cross-sectional view of the relocated hypocenters.The upper left panel shows the locations of vertical cross-sections for the other eight panels.Circles represent hypocenters and their sizes correspond to the fault diameter assuming a stress drop of 3 MPa.The color scale shows the relative occurrence time of each earthquake, with the horizontal line indicating the timing of the M w 6.2 event.Green star shows the M w 6.2 mainshock.

Figure 3 .
Figure 3. Temporal changes in earthquake depth.Five different colors represent the five clusters.Green star shows the 2023 M w 6.2 event.(a) The earthquakes in the five clusters.Cross indicates a nearby GEONET GNSS station (ID: 0253).(b) Comparison of time and earthquake depth.(c-g) Comparisons of time and earthquake depth for each cluster.Curve represents the cumulative number of earthquakes.

Figure 4 .
Figure 4. Results of the slip inversion of the 2023 M w 6.2 earthquake.(a) Map projection of the slip distribution.The green beach-ball and black star represent the focal mechanism and hypocenter of the M w 6.2 event, respectively.The black beachball represents the focal mechanism of the eGF event.Blue and red circles indicate earthquakes with M JMA > 5 before and after the 2023 M w 6.2 event, respectively.Dots represent subfaults and thick line represents the shallowest edge of the model fault.(b) Moment-rate function.(c) The fitting of observed (black) and synthetic (red).

Figure 5 .
Figure 5. Spatiotemporal distribution of the hypocenters near the fault zone of the M w 6.2 earthquake (within 1.0 km from the line in b).(a) Map view, and (b) cross-section.The color scale shows the relative occurrence time of each earthquake.Gray indicates earthquakes outside the fault zone.The thick black line with triangles in (a) and the black triangle in (b) show the surface trace of the Suzu-Oki active fault segment (Inoue & Okamura, 2010).Green contour lines in (a) show the mainshock coseismic slip distribution, and the bold line in (b) indicates the segment with a large slip (>0.18 m).The black star and contour lines represent the hypocenter and slip area of the 2022 M w 5.4 event, respectively.The red and yellow stars represent the hypocenters of the 1993 M w 6.3 event reported by the Japan Meteorological Agency and Tsukuda et al. (1994), respectively.(c) Earthquake depths compared to the occurrence times.