1 Introduction

Taiwan is one of the most active orogenic belts around the world that primarily results from the oblique arc-continental collision between the passive margin of the Eurasian Plate (EP) and the overriding Luzon Arc on the Philippine Sea Plate (PSP) (Fig. 1) (e.g., Suppe 1981, 1984; Teng 1987, 1990; Shyu et al. 2005a). The PSP is moving in the northwestern direction of ~ 306° at the rate of > 8 cm/year toward the EP’s passive continental margin (e.g., Yu et al. 1997; Hsu et al. 2009). Along the western boundary of PSP, it collides to the Central Range of Taiwan along the Longitudinal Valley at eastern Taiwan and subducts along the Ryukyu subduction zone at the northern edge of PSP offshore northeastern Taiwan (e.g., Teng 1990). The colliding segment of the Luzon Arc forms the Coastal Range in eastern Taiwan against the accretionary wedge to the west (e.g., Lundberg et al. 1997). In between the Coastal Range (CoR) and the Central Range (CeR) is the Longitudinal Valley, the important suture zone between the PSP and EP. The Longitudinal Valley Suture consumes nearly half of the PSP-EP convergence rate (~ 30–40 mm/year) (e.g. Hsu et al. 2003, 2009). This large shortening rate contributed to the high frequency of seismicity and high potential of damaging earthquakes, such as the 1951 ML 7.3 Hualien earthquake sequences (Hsu 1962; Shyu et al. 2007; Chen et al. 2008) and the recent 2018 ML 6.4 Hualien earthquake (Huang and Huang 2018; Lee et al. 2018).

Fig. 1
figure 1

(Active fault map is modified from Shyu et al. (2005a, b), Shyu et al. (2020), velocity of PSP is from Yu et al. (1997); the velocity of Ryukyu Arc is from Lallemand and Liu (1998))

General tectonic map of Taiwan showing major active tectonic elements near Eastern Taiwan. a Eastern Taiwan is experiencing tectonic collisions between the volcanic arc (Coastal Range/Luzon arc) and the basement-involved accretionary wedge (Central Range) at the western edge of the Philippine Sea Plate. More than half of the active convergence between the Philippine Sea Plate and the Eurasian Plate is absorbed by the Longitudinal Valley Suture system (LVS) and its associated structure. In the north, both the coastal range and the Longitudinal Valley Suture intersects to the Ryukyu Trench, with the Okinawa Trough and Ryukyu Forearc Ridge formed above the Ryukyu subduction zone. The opening and tectonic rotation near the LVS-Ryukyu trench junction enhance the N–S convergence along the Ryukyu Trench. b Active tectonic map along the Longitudinal Valley shows three major structures from the base of the Central Range to the offshore eastern coast. These structures on the surface include the Central Range Fault, The Longitudinal Valley Fault, and the Chimei Canyon Thrust as the deformation front of the Offshore Eastern Taiwan Thrust Belt

Full size image

The active Longitudinal Valley Suture includes two major fault systems: the east-dipping Longitudinal Valley Fault (LVF) at the western flank of the Coastal Range and the west-dipping Central Range Fault (CRF) at the foot of the metamorphic Central Range (e.g., Shyu et al. 2005b; Shyu et al. 2006a, b). These two faults are accompanied by the east-verging Offshore East Taiwan Thrust Belt offshore Coastal Range, where the active Chimei Canyon Thrust Fault (CCF) cropped out the seafloor as the easternmost frontal thrust (Malavieille et al. 2002; Hsieh et al. 2020). All these three fault systems are active at the latitude of Ruisui (Malavieille et al. 2002; Shyu et al. 2006a, b, 2007; Hsieh et al. 2020). Based on geological and geomorphological investigations, Shyu et al. (2006a) suggest the west-dipping Central Range Fault cropped out on the surface at the central-western part of the Longitudinal Valley, where the Wuhe conglomerates are uplifted by the west-dipping reverse fault, with an estimated fault slip rate of ~ 7.3 mm/yr (Shyu et al. 2020). A few kilometers to the east, the east-dipping Longitudinal Valley Fault appears near the base of the Coastal Range and is responsible for the second M ~ 6.9 earthquake in the 1951-Nov earthquake sequence (Shyu et al. 2007). Both geological and geodetic analysis suggest the slip rate of the Longitudinal Valley Fault reaches 20 to 30 mm/yr (e.g., Shyu et al. 2006b; Huang et al. 2010). The fold-and-thrust-belt offshore eastern Taiwan is also likely accommodating a significant amount of PSP-EP motion, with an estimated geological slip rate reaching ~ 20 mm/yr since ~ 1-Ma at the frontal Chimei Canyon Thrust (Hsieh et al. 2020).

The close-spacing between these rapid-slip active faults allows these structures to develop a complex structural system along the valley. Just north of Ruisui, the Central Range Fault is likely overridden by the Longitudinal Valley Fault based on geological and seismological observations (e.g. Shyu et al. 2005b, 2006a; Chuang et al. 2014; Lee et al. 2014) (Fig. 1b). Although the Central Range Fault is concealed by the east-dipping Longitudinal Valley Fault north of Ruisui, it remains seismic capable. The best evidence showing the Central Range Fault’s activity is the 2013 ML 6.4 Ruisui Earthquake (Chuang et al. 2014; Lee et al. 2014). Both the mainshock slip inversion result and the aftershock distribution suggest the 2013 Ruisui earthquake rupture was on the concealed Central Range Fault, where the fault tip stopped ~ 10 km beneath the western part of the Coastal Range.

Further to the north, the surface expression of the Longitudinal Valley Fault also becomes less clear close to the latitude of Hualien, as the Luzon Arc and PSP lithosphere are gradually submerged before subducting into the Ryukyu subduction zone in the north (Wu et al. 2009; Huang et al. 2012; Shyu et al. 2016a). In addition to the submerged overriding Luzon Arc and the Longitudinal Valley Suture system, other active structures such as the Milun Fault and its associated hanging wall structures also appear near the vicinity of Hualien, forming a separate group of active structures (Fig. 1b). Unlike the Central Range Fault and the Longitudinal Valley Fault that transect through almost the entire valley, this group of young and active structures are relatively short and shallow, and do not directly connect to any major structures in the valley (Shyu et al. 2016a). The recent 6 February 2018 ML 6.3 Hualien earthquake occurred underneath these shallow active structures, on a west-dipping blind fault (Lee et al. 2018). This earthquake also triggered the slips on the overlaying shallow faults, showing a complex structural pattern and mutual interaction at the northernmost Longitudinal Valley, close to the junction between the Longitudinal Valley Suture and the Ryukyu Subduction zone. (e.g., Shyu et al. 2016a; Lee et al. 2018; Huang and Huang 2018; Kuo et al. 2018; Yen et al. 2018).

Yet, the earthquake activity has not eased after the 2018 Hualien earthquake. On 19 April 2019 (ML 6.3), 15 February 2020 (ML 5.7), 18 April 2021 (ML 5.8), and 7 July 2021 (ML 5.5), a series of M5.5+ earthquakes occurred in the northern Longitudinal Valley (Fig. 2). In particular, the two 2021 Hualien earthquakes that occurred subsequently in half a year and rattled the area with prolonged aftershock activity (or called swarm) have revoked people’s attention and urged an investigation into local seismogenic structures. While previous studies for the 2018 earthquake already gained valuable insight into interaction between a deep west-dipping blind fault and the shallow east-dipping Milun fault at the northern tip of the Longitudinal Valley and the Coastal Range (e.g., Yen et al. 2018; Kuo-Chen et al. 2018), how these ruptured faults related to the major structures (i.e. Longitudinal Valley Fault) at the surface, and how far they extend to the south remains unclear. This series of M5.5+ Hualien earthquake sequences unlocked different fault structures on both sides of the valley and provided a unique opportunity to illuminate the geometry of subsurface structures in a more complete picture (Fig. 2). To this end, we combine the seismological observations including travel-time tomography, earthquake relocation, and focal mechanism analyses with published surface and subsurface geological observations (e.g. active fault mapping; balance structural profiles, etc.) to assess the structural geometry and their relationship at the northern half of the Longitudinal Valley Suture system.

Fig. 2
figure 2

Earthquake distribution of the 2018–2021 earthquake clusters near Hualien city. Colored dots represent earthquakes in different clusters, with mainshock events denoted by the star. Red lines represent the Longitudinal Valley Fault in the south and the Milun Fault in the north. Blue triangles in the lower-right inserted map shows the seismic station used in this study

Full size image

2 Data and methods

In this study, we collect the arrival times and P-wave polarity data from the Central Weather Bureau earthquake catalog from January 1991 to August 2021, which is a long-term effort compiling analysts’ picks from the permanent Central Weather Bureau Seismic Network (CWBSN) (Shin 1992) and Taiwan Strong-Motion Instrumentation Program (TSMIP) (Shin et al. 2003), with a total of 991 stations. We select the earthquakes based on the criteria: (1) locations within Longitude 121.0°–122.2° and Latitude 22.4°–24.6°, (2) focal depths within 0–60 km, (3) picking quality from level 0 to 3 (CWB picking quality ranking from 0 to 4. Level 0 is the best), (4) minimum recording stations of 6, and (5) gap angle smaller than 180. This results in a total of 186,738 events.

We then apply a double-link declustering method (Wu and Chiao 2006) to extract earthquake clusters from background seismicity. The double-link declustering method is similar to the single-link clustering method of Davis and Frohlich (1991) but uses not only space distance but also time distance between consecutive two nearest events into the linking process. For a sequence of n events, the linking process will start from Event 1 and search Event 2 to n sequentially. If there are m events falling within the designated threshold distances of space and time from Event 1, they will be grouped into Cluster 1 and continue the linking process from each of them to their following events to grow Cluster 1 until no events are further linked. Then, the whole process will restart for the events yet grouped into Cluster 1 to form Cluster 2, and iterate. We use the empirical suggestions of 3 day and 5 km as time and space threshold distances from Wu and Chiao (2006). These values have also been tested in a number of studies and worked generally well for extracting compact clusters (Wu and Chen 2007; Chen et al. 2012; Wu et al. 2018; Hsu et al. 2020, 2021).

A 3-D double-difference relocation algorithm, 3D-DD (Hsu et al. 2020) is used to relocate the extracted clusters with a 3-D crustal velocity model of Huang et al. (2014). The inter-event distance is tested to be 2 km to pair events to a common station for differential travel time calculation. This relocation algorithm differs from hypoDD (Waldhauser and Ellsworth 2000) by two enhancements, which are (1) capable of incorporating 3-D velocity models to better take care of velocity heterogeneity near sources and (2) retaining isolated events by combining the single-event location algorithm using absolute travel time data and the double-difference location algorithm using differential travel time data into one inversion scheme, as

$$W\left[ {\begin{array}{*{20}c} {G_{DD} } \\ {G_{SE} } \\ \end{array} } \right]m = W\left[ {\begin{array}{*{20}c} {d_{DD} } \\ {d_{SE} } \\ \end{array} } \right]$$

where the G matrices are the partial derivatives of travel time with respect to model vector m for the adjustments of the event location and origin time. d is the data vector. The subscripts DD and SE represent the double-difference part and the single event part, respectively. W is the weighting matrix that considers the picking data quality. We weigh the double-difference data 10 time the single-event data here.

After 3D-DD relocation, we determine focal mechanisms of the relocated earthquake clusters using a P-wave first-motion genetic algorithm (Wu et al. 2008), which is an efficient global search method that better avoids the local minimums in the nonlinear optimization process for first-motion polarity determination. Wu et al. (2008) also defined a quality index, Qfp, combining the gap angle, data number, polarity misfit, and proportion between up/down polarities, for solution evaluation. They suggested Qfp > 1 as a good solution. We therefore only used the solutions with Qfp > 1 in the study.

It may be worth noting that the locality of first-motion polarities (ray take-off angles) on a focal sphere are not only affected by the earthquake location but also the lateral velocity variations. Both of which are handled in the 3D-DD, using the 3-D velocity model and 3-D ray tracing (Huang et al. 2014). In contrast, while the hypoDD works well in improving the event relative location, the ray-tracing with 1-D velocity models may still yield less accurate take-off angles and bias the first-motion determination (Hsu et al. 2020, Additional file 1).

3 Results

3.1 Earthquake declustering, relocation, and focal mechanism redetermination

Figure 3 shows 29 extracted earthquake clusters (colored dots) and background seismicity (gray dots) within the study area after double-link declustering. The color denotes the event occurrence time. The wide-spread cluster in blue are aftershocks induced by the 1999 ChiChi earthquakes. Excluding that, 28 clusters will be used in later analyses. Most of the earthquake clusters in the northern Longitudinal Valley occurred after ~ 2010 and are spatially distributed in two groups in a map view, both trending NNE–SSW along two sides of the valley. The recent 2018–2021 Hualien earthquake clusters (in orange to red after 2018) mainly ruptured the faults north of 23.8° N. South of this latitude, the 2013 Ruisui earthquake and the 2014 Fonglin earthquake occurred on the western side of the valley (Lee et al. 2014) and the earthquake clusters on 6 February 2012 and 2 August 2019 occurred on the eastern side. The seismicity is relatively quiet further south of the latitude 23.6° N.

Fig. 3
figure 3

Distribution of background seismicity (grey-colored dots) and extracted earthquake clusters (colored dots) from January 1991 to August 2021. Locations for the profiles shown in Figs. 6, 7 and 8 are indicated as black lines here

Full size image

All of these earthquake clusters are distributed within the top ~ 25 km of the crust, except one located east offshore at deep depths of 30–45 km (Fig. 4a). The depth of the clusters beneath the Central Range are systematically deeper than those beneath the Coastal Range. After relocating the clusters (Fig. 4b), the concentration in earthquake distribution is clearly visible in several places. For instance, the 2018 Hualien earthquake cluster shows a systematic shift toward the east and better aligns with the fault trace, due to strong lateral velocity variations in this tectonic-complex region. The average location uncertainties are about 200 and 150 m in horizontal and vertical directions using the approximate covariance matrix estimation (Waldhauser and Ellsworth 2000).

Fig. 4
figure 4

Distribution of earthquake clusters before and after relocation and redetermined first-motion focal mechanisms. a The epicenter of the seismic cluster from the original earthquake catalog. b The epicenter of the seismic cluster after 3D-DD relocation. The black dashed circles indicate the fault segments remaining locked. c The first-motion focal mechanism from this study. The color represents the focal depth (km) of these events

Full size image

Figure 4c shows the redetermination of the focal mechanisms after relocation. The focal mechanisms in western-group clusters show consistently reverse and left-lateral faulting while those in eastern-group clusters show primarily reverse component with greater diversity toward the northern offshore of Hualien (Fig. 4c). Using the quality index in the algorithm (Wu et al. 2008), we show that the solution quality of the focal mechanisms overall improves after the 3D-DD relocation (Fig. 5).

Fig. 5
figure 5

Improvement on qualify factor of first-motion focal mechanism determination after 3D-DD relocation. Blue and red curves show the distribution of qualify factor before and after the relocation

Full size image

3.2 Velocity profiles with relocated earthquake clusters

We examine a series of E–W and N–S velocity profiles of Huang et al. (2014) with newly relocated earthquake clusters and redetermined focal mechanisms in the northern Longitudinal Valley (Figs. 6, 7 and 8). The background seismicity and the AutoBATS focal mechanism catalog (Jian et al. 2018) are also used for comparison.

Fig. 6
figure 6

Selected approximately E–W direction velocity profiles transversing the Longitudinal Valley. All the E–W profiles are shown in the supplementary. The #1 denotes the “western-group clusters” between 25 and 15 km deep. The #2 denotes the “eastern-group clusters”. The #3 denotes the clusters likely produced by the seismic activity of the deeper part of the LVF, whereas the #4 denotes the clusters likely associated with the rupture of the shallow part of the LVF. See maintext for detailed discussion. The gray circles and gray beachballs are the relocated hypocenter of background seismicity and focal mechanisms from AutoBATS (Jian et al. 2018). The black circles and purple beachballs are the relocated hypocenter and redetermined first-motion focal mechanism of the 28 earthquake clusters. Structural interpretation is delineated by translucent lines and arrows in the upper panels. LVF Longitudinal Valley Fault, MLF Milun Fault. The profile location is shown in Fig. 3

Full size image
Fig. 7
figure 7

Approximately N–S direction velocity profiles along the Longitudinal Valley. The strike of these profiles is paralleling to major faults’ orientations. See maintext for detailed discussion. The gray circles and gray beachballs are the relocated hypocenter of background seismicity and focal mechanisms from AutoBATS (Jian et al. 2018). The black circles and purple beachballs are the relocated hypocenter and redetermined first-motion focal mechanism of the 28 earthquake clusters. The profile location is shown in Fig. 3

Full size image
Fig. 8
figure 8

Enlarged profile R–R′ and S–S′ across a the northernmost LVF and b the Milun fault area, respectively. See maintext for detailed discussion. The gray circles and gray beachballs are the relocated hypocenter of background seismicity and focal mechanisms from AutoBATS (Jian et al. 2018). The black circles and purple beachballs are the relocated hypocenter and redetermined first-motion focal mechanism of the 28 earthquake clusters. The profile location is shown in Fig. 3

Full size image

Figure 6 shows a series of velocity profiles from south to north, extending from Ruisui in the south to Hualien in the north. Several features are visible within these E–W running profiles at the northern half of the Longitudinal Valley. To the west, a cluster of west-dipping seismicity exists in most our E–W running profiles beneath the Central Range (western-group clusters; #1 in Fig. 6). These well-defined western-group clusters are distributed between ~ 25 and ~ 15 km deep beneath the Central Range and coincide with the seismic activity of the previously suggested Central Range Fault (Lee et al. 2014; Shyu et al. 2016a). The dip angle of these western-group clusters can be as steep as 70° from our relocated seismicity and consistent with the focal mechanism solutions in this study (Figs. 4c and 6). East of its updip-ward projection, another group of shallow seismicity appears between ~ 13 and ~ 5 km deep beneath the Longitudinal Valley and the Coastal Range (eastern-group clusters; #2 in Fig. 6). These shallow eastern-group clusters reveal a much gentler west-dipping fault geometry (~ 30° westward) than that of the western-group clusters. Most focal mechanisms within this cluster also show a shallower geometry than those within the western-group at the greater depth. Its geometry of the narrower distribution and a gentle westward dip after relocation indicates a possible connection with the deeper western-group clusters, which can be found most prominently between the profile D–D′ and H–H′ and north of the profile M–M′ (Fig. 6 and Additional file 1: Fig. S1).

The transition between the deep and the shallow clusters is not smooth and continued within these profiles. Rather, the relocated earthquake clusters including the 2018–2021 Hualien earthquake sequences suggests a sharp and abrupt dip-angle change at ~ 13 km deep right beneath the eastern margin of the Central Range (Fig. 6b, c). Such feature can be easily observed from profile D–D′ to H–H′ where seismic activities were evenly distributed along these two clusters. While this dip-angle change is not directly observed by seismicity distribution in other profiles, the focal mechanisms that show different patterns between the shallow and deep clusters support this change. This abrupt change suggests these two clusters may involve more than one rupture fault plane, in which the shallow and gentle fault plane is branched out from the previously recognized steeply dipping Central Range Fault at ~ 12 to ~ 13 km deep.

The dense earthquake hypocenters of these two west-dipping fault planes also follow a sharp west-dipping boundary at the lower edge of a high-Vp block and the higher/western edge of a Vp/Vs ratio anomaly beneath the Longitudinal Valley and Coastal Range (Fig. 6). This boundary can be found in most of our E–W profiles (Additional file 1: Fig. S1) and is likely representing the structural boundary of the exhumed metamorphic block under the Central Range and the northern Longitudinal Valley. Part of this structure boundary has been interpreted as the Central Range Fault as mentioned in the previous section, and the tomography results suggest this structural system extends further eastward under the Longitudinal Valley. Above this interpreted metamorphic rocks within the northern Longitudinal Valley, only a very thin layer (< ~ 3 km) of low-Vp patch found in these profiles (e.g., Fig. 6b, c). This low-Vp patch is mainly constrained by the near-surface logging data and could be slightly extrapolated laterally (Huang et al. 2014). Nonetheless, its distribution and geometry should correspond to the late-Cenozoic Longitudinal Valley sediments and suggest a relatively limited deposit within the valley where the depositional center lies close to the eastern side of the valley. East of the interpreted metamorphic rocks and low-velocity basin sediments, another low-Vp area appears beneath the eastern part of the Coastal Range. This low-Vp patch is likely associated with the formations of sedimentary and volcaniclastic rocks exposed in the northern part of the Coastal Range (e.g. Wang and Chen 1993).

It is also worth noting that the interpreted metamorphic rocks almost extend to the area beneath the Coastal Range at the northernmost Longitudinal Valley (Profile F–F′ and H–H′; Fig. 6). The extension of the metamorphic rocks matches well to the eastern limits of the CRF-related seismic cluster where both the background seismicity and the relocated earthquake clusters end near the upper-eastern corner of this high-velocity patch. All of our profiles suggest the entire CRF-related system is mantled by the low-velocity Coastal Range sedimentary rocks and Valley sediment sediments east of the Coastal Range.

In addition to the P-wave velocity profiles, the Vp/Vs ratio profiles at the northern Longitudinal Valley also show clear signals of the west-dipping fault system under the valley (Fig. 6). Compared to the surrounding area, the CRF-related seismicity mostly falls along western edge of high Vp/Vs ratio. The high Vp/Vs ratio may indicate the existence of fluid near the CRF structure and within the fractured Longitudinal Valley suture in general, as observed in the source areas of the 2003 Chengkung and 2018 Hualien earthquakes (Wu et al. 2007; Wen et al. 2019; Toyokuni et al. 2021). This can be seen clearly from Fig. 7 that the high Vp/Vs ratio anomalies are distributed along the Longitudinal Valley suture and follow the subducting trend of the PSP. The fault-related fracture system may provide high permeability to allow groundwater to flow along the fault. The subduction and dehydration of the forearc of the Philippine Sea Plate could also be an alternative origin of the fluid. Such a fluid-rich environment provides an ideal condition for the occurrence of earthquake swarms for instance the 2021/07/07 Hualien earthquake sequence and those observed in the Hualien region (e.g., Shyu et al. 2016a; Peng et al. 2021).

The eastern portion of the CRF-related cluster seems to override another group of earthquakes, which appears to be a steep east-dipping cluster below 10 km deep (#3 in Fig. 6). This steep east-dipping cluster contains less seismicity than those in the CRF-related cluster in our relocated earthquake catalog; however, it aligns well to the background seismicity and focal mechanisms from the AutoBATS solutions (Fig. 6b, c). This cluster is likely related to the deeper section of the Longitudinal Valley Fault, or an active fault within the same fault system. Both of them were well delineated in the southern part of the Longitudinal Valley (Shyu et al. 2006a, 2011; Kuo-Chen et al. 2004). This LVF-related cluster shows clear northward warping in the N–S running profiles (Profile P–P′ in Fig. 7), suggesting the Longitudinal Valley Fault is subducting northward together with the northern Coastal Range and the Philippine Sea Plate (Huang et al. 2012; Shyu et al. 2016a).

The E–W running profiles from south to north show different features on the Longitudinal Fault at depth (Fig. 6 and Additional file 1: Fig. S1). Just north of Ruisui, seismicity and focal mechanism prior to the 2018–2021 clusters suggest most of the LVF-related earthquakes occurred at 20 to 30 km deep, with the fault plane roughly dips ≥ 60° eastward. Although only a handful of earthquakes occurred at the shallow depth, the upward projection of the LVF seismicity clearly intersects with the up-dip termination of the CRF-related cluster (#1 in Fig. 6a) (e.g. Lee et al. 2014). This feature suggests the CRF system and the LVF system form a structural contact at the latitude just north of 23.5° N, as pointed out by earlier seismological and geological studies (Shyu et al. 2006a; Chuang et al. 2014; Lee et al. 2014). Above this structural contact, we do not notice any clear pattern of seismic activity associated with the LVF in the past decade (Fig. 6a). Considering the surface trace of this LVF section ruptured during the Nov-1951 earthquake (Shyu et al. 2007), the lack of LVF seismicity at the shallow depth near 23.5° N suggests the LVF near Ruisui is currently locked. Such interpretation is supported by the geodetic analysis in this region which shows the LVF’s shallow section near Ruisui is currently highly coupled (e.g. Thomas et al. 2014).

North of the 23.6° N, the relationship between the west-dipping seismicity (i.e. CRF-related) and the east-dipping seismicity (i.e. LVF-related) can be clearly delineated in our relocated seismicity profile (Profile F–F′ to J–J′ in Fig. 6). The LVF-associated seismicity seems to be concealed by the east-dipping CRF-related seismicity, with nearly no shallow earthquake occurred on the east-dipping LVF near the surface. The only exception is a small, steeply east-dipping cluster at ~ 23.7° N (#4 in Fig. 6 and the enlarged Profile R–R′; Fig. 8). This earthquake sequence on 2 August 2019 occurred above the west-dipping CRF cluster, extending from ~ 1 km deep to ~ 7 km deep beneath the western flank of the Coastal Range. Along with its up-dip projection, the surface projection of this earthquake cluster well matches the LVF’s surface location from Shyu et al. (2020), suggesting this steep sequence could present the near-surface geometry of the LVF (~ 50° eastward). In an enlarged profile near this location (Profile R–R′, Fig. 8a), it seems that a second east-dipping cluster developed parallel to this shallow LVF seismicity, forming an imbricate thrusting geometry that is consistent with the structural pattern from the balanced structural section in Hsieh et al. (2020) at the west part of the Coastal Range. Before the 2018–2021 Hualien earthquake sequences, the seismicity in this region is relatively scattered and difficult to define such meaningful structures.

The downward projection of this shallow LVF-related cluster, however, does not directly connect to the LVF-related seismicity below the CRF-related seismicity (Figs. 6b, 8a). Instead, a small offset between the projection and the relocated seismicity is found at the Profile F–F′ (Figs. 6b and 8a), where earthquakes from the deeper section of LVF (#3 in Fig. 6b) are slightly west of the downward projection of the shallow LVF-related seismicity (#4 in Fig. 6b). This west-ward shifting of deep LVF-related seismicity seems to be enhanced further north, where we could clearly see the westward migration of LVF-related seismicity along the northern-most Longitudinal Valley below the CRF-related earthquake cluster (Fig. 6c, d). Such feature from relocated seismicity implies the Longitudinal Valley Fault is likely truncated by the Central Range fault system north of ~ 23.7° N, and the deeper portion of the LVF keeps moving northwestward with the PSP underneath the Central Range Block.

Only a few deep LVF-related seismicity can be found in the relocated 2018–2021 earthquake sequences near the latitude of 24° N, just east of Hualien (Fig. 6e). Most of the recent 2018–2021 sequences here are either associated with the west-dipping fault or associated with a sub-horizontal fault system offshore Hualien at ~ 6–7 km deep (#2 in Fig. 6e). Here, although the Milun fault ruptured during the 2018 Hualien earthquake, we do not observe seismicity that is clearly related to the shallow Milun Fault from our 3D relocated seismicity (Figs. 6e and 8b). The relocated seismic cluster at ~ 8 km also shows no preferred orientation near Hualien, either. The lack of well-defined seismicity beneath Hualien may indicate this cluster is sourced from multiple close-spaced faults at shallow depth, rather than sourced from a single, well-defined fault plane, such as the Milun Fault.

4 Discussion

4.1 The fault interception revealed by the recent Hualien earthquake sequences

The east-dipping LVF system and west-dipping CRF system are two major fault systems that have been geologically and geomorphologically identified throughout the Longitudinal Valley (Wang and Chen 1993; Shyu et al. 2005b, 2006a). The 2006 Taitung earthquake sequence and the 2003 Chengkung earthquake sequence are the two best examples delineating subsurface geometry of these two fault structures in the southern part of the valley (Wu et al. 2006; Kuo-Chen et al. 2007; Shyu et al. 2011), where the CRF shows a sub-vertical west-dipping fault plane with dominantly left-lateral faulting and the LVF shows a listric east-dipping fault geometry with thrust motions. However, the seismological expression of the CRF and LVF becomes obscured in the central and northern segments of the valley. Not until the occurrence of the 2013 Ruisui earthquake and the 2014 Fonglin earthquake unveiled the geometry of the CRF at the latitude 23.5° N–23.7° N. These two earthquake sequences showed a gentler CRF fault geometry that shall intercept with the LVF at ~ 10 km if the subsurface projection of the LVF is made from its surface location with estimated dip angle (Chuang et al. 2014; Lee et al. 2014). Since the aftershocks of the Ruisui earthquake do not extend across this interception and surface expression of the LVF in the northern Longitudinal Valley is still evident from geomorphic studies (e.g. Shyu et al. 2005b), the CRF is inferred to be concealed by the LVF despite the LVF-related seismicity being relatively diffuse and unclear (Chuang et al. 2014; Lee et al. 2014). In fact, the long-term pattern of background seismicity shows a clear west-dipping trend in the northern Longitudinal Valley and an east-dipping trend in the south. This polarity reversal of the plate boundary was discussed and still poorly resolved (Kim et al. 2006).

The Hualien earthquake sequences from 2018 to 2021 have progressively provided critical information about the subsurface structures in the northern Longitudinal Valley (Fig. 2). The 2018 Hualien earthquake is a complex event involving multiple fault ruptures and their mutual triggering (e.g., Huang and Huang 2018). Despite different fault settings in previous studies (Huang and Huang 2018; Lee et al. 2018; Lo et al. 2019), the primary slips were determined on a west-dipping blind fault and the east-dipping Milun fault at 60°–70° (e.g. Kuo et al. 2018). Our detailed relocation analysis shows that this west-dipping fault geometry is most clear offshore Hualien city within 5–15 km depths (Fig. 6f). When entering Hualien city, the fault dip becomes gentle but more scattered beneath the Milun fault (Fig. 6e). The downdip extension of this blind fault was soon after revealed by the 2019 Hualien earthquake (Lee et al. 2020). With a steeper dip and left-lateral faulting, this downdip portion of the fault is consistent with the CRF revealed by Ruisui and Fonglin earthquakes (Fig. 6b, f), despite a separation of 30 km (from Latitude 23.7° N to 24.0° N) in between.

Do the seismic activity of the 2013 Ruisui and the 2019 Hualien earthquakes represent the same CRF-related fault structure? The subsequent Hualien earthquake sequences in 2020–2021 fill in the gap and give the answer (Fig. 2). The fault revealed by the 2020/02/15 and 2021/04/18 sequences shows remarkably similar geometry with the blind fault for 2018–2019 Hualien earthquakes, where a steep west-dipping fault plane at the depths of 15–25 km and a gentle west-dipping fault plane extending toward the shallow depths and east beneath the Coastal Range (Fig. 6c, f). The 2021/07/07 Hualien earthquake sequence then occurred in the north of the 2021/04/18 sequence and continued this gentle part of the fault plane to the clusters beneath the Milun fault (Fig. 7a). Although this gentle and shallow portion of the seismicity at ~ 10-km depth was previously interpreted as a detachment surface separating the submerging Coastal Range and its overlaying buoyant sediments (Shyu et al. 2016a), The earthquake sequence between 2020 and 2021 extend this cluster further southward to the latitude 23.7° N, where this gentle, west-dipping connects to the CRF-LVF intersections beneath the Coastal Range (Fig. 7). Thus, this primary west-dipping fault, presumably to be the shallow fault branched out from the previously known CRF, shall cut through the LVF in the northern Longitudinal Valley. This hypothesis is further supported by the east-dipping distribution of background seismicity at the northern Coastal Range area (lower panels, Fig. 6c). One deep earthquake cluster in our declustering analysis also shows clear east-dipping distribution at the depths of 30–45 km, although the inconsistent focal mechanisms with those of background seismicity need further investigations (Figs. 4a and 6c). The up-dip projection of these LVF clusters cannot directly meet the surface trace of LVF at the base of the Coastal Range (Fig. 6b–d), especially after taking into account the ~ 50° east-dipping shallow fault geometry revealed by the geological cross-sections (e.g., Shyu et al. 2006b; Hsieh et al. 2020) and seismic clusters. The mismatch between the earthquake cluster and the surface trace of LVF, plus only a few east-dipping aftershocks present in cross-section F–F′, H–H′, and J–J′ were triggered in 2018 and 2021 Hualien earthquakes and always below the CRF, lead us to suggest the northernmost Longitudinal Valley Fault is likely to be truncated by the eastward propagation of the Central Range fault system. This structural development also seems reasonable to accommodate not only the northward but also westward gradual subduction of the PSP and forearc fragment when approaching the Ryukyu trench.

4.2 A seismogenic model of the northern Longitudinal Valley suture

The 2018–2021 earthquake sequence provides us a unique opportunity to review the active seismogenic system at the northernmost Longitudinal Valley. Within this area, the northern section of LVF is considered as a highly coupled fault, with a continuous listric shape of the fault plane from the depth of ~ 30 km deep to the surface, based on the space-geodetic analysis and modelling results (e.g., Thomas et al. 2014). The reliability of such a model relies on a reasonable regional fault geometry either derived from the geological or seismological analysis, and different boundary condition settings in these models may greatly alter the fault coupling results (e.g., Hsu et al. 2012, 2016 for the Manila Trench). While the northernmost LVF is likely highly coupled with strain accumulated on its fault plane, its main fault plane did not rupture during the 2018 Hualien earthquake, even though the 2018 Hualien earthquake sequence promoted the Coulomb Stress on the northernmost LVF (Chan et al. 2019). Only a very shallow and short patch on the Milun Fault and the northernmost LVF (i.e. Lingding Fault) were induced during the mainshock (Central Geological Survey 2018; Lee et al. 2018; Yen et al. 2018; Huang et al. 2019). A quickly arrested 2018 surface rupture on the northernmost LVF system and the limited aftershocks distribution above the depth of 10 km hint the heterogeneous pattern of coupling on the LVF fault plane, or the structural complexity creates seismic barriers on the shallow section of the northernmost LVF to stop the induced rupture.

We suggest the structural interaction between the active CRF and the active LVF at the northernmost part of the Valley creates the unique geological environment to allow the seismic barriers developed on the coupled fault plane, and create segmented seismogenic fault patches at the northern part of the Longitudinal Valley. Figure 9 shows a schematic 3D structural model at the northern Longitudinal Valley based on our closely-spaced velocity and seismicity profiles and published interpretations. For the geometry of the Central Range Fault, we first adopt the CRF fault geometry from Shyu et al. (2016b) (~ 45°) for the Ruisui area and subsequently modify its geometry using the distribution of background seismicity and relocated earthquake clusters from 1991 to 2021. It is notable that the CRF becomes much steeper north of Ruisui (≥ 60°) after adapting the geometry from the relocated seismic clusters. This results in a fault plane geometry similar to that suggested by Lee et al. (2014), but steeper than the fault plane (~ 42°) suggested by Chuang et al. (2014). For the geometry of the Longitudinal Valley Fault, we follow similar “listric” fault geometry adapted from Shyu et al. (2006b) for the shallow section of the Longitudinal Valley Fault near the surface. We also trace the deeper part of the LVF using background seismicity and the relocated earthquake clusters.

Fig. 9
figure 9

A schematic structural model of the northern Longitudinal Valley showing the truncation relationship between the active Central Range Fault system and the active Longitudinal Valley Fault system. Black lines are the simplified seismogenic fault of the Central Range Fault system, with blue sub-vertical splay faults developed within the Central Range blocks. The Red lines are faults within the Longitudinal Valley Fault system, including the Longitudinal Valley Fault (LVF) and the Chimei Fault. Gray lines represent the location and approximate geometry of the Takangkou-Height Thrust from Hsieh et al. (2020), while the Green lines show the Milun fault (M.L.F.) and its associated faults offshore Hualien

Full size image

Unlike the previous regional model that shows the LVF as a listric and continuous east-dipping fault running along the northern half of the Coastal Range (e.g., Thomas et al. 2014), our observation suggests north of ~ 23.7° N, a young west-dipping fault that plausibly branched out from the CRF truncates the east-dipping, listric LVF beneath the Coastal Range and continuously propagates eastward. This gentle west-dipping fault eventually connects to the 2018 Hualien mainshock rupture patch and separates the LVF into a shallow-rooted fault segment near the surface and a deep fault segment. The structural separation and overriding may therefore prevent the steep and shallow LVF segment from accommodating tectonic convergent strain on its fault plane since the strain from the PSP-EP motion would be much easier to be accommodated by the gentler CRF-related fault plane. Instead, the shallow northern LVF may evolve to be a transpressional strike-slip fault system as the secondary fault structure to accommodate the structural–parallel horizontal motion from the strain partitioning of the oblique PSP-EP collision. In this context, the Milun Fault, which shares similar steep east-dipping geometry as the northernmost LVF, and the Milun Tableland on its hanging wall could then be a product of the restraining pop-up ridge and play a passive role to be easily triggered by the nearby fault structures, such as the cases in the recent 2018 and the 1951 Hualien earthquakes, with an anomalous short rupture cycle of 67 years (Chan et al. 2020; Shyu et al. 2016b, 2020). This strain-partitioning interpretation matches the geomorphic observations from Shyu et al. (2005b), where north of Kuangfu (~ 23.6° N), the surface expression of LVF is dominated by landforms associated with a series of left-lateral active faults along the northwestern side of the Coastal Range.

4.3 Implications for seismic potentials

The new seismogenic model could also reshape the seismic potentials of the northern Longitudinal Valley Suture system. The limited width of shallow LVF (fault width ~ 8 km) means even if the fault remains to be a capable fault, the potential magnitude of sole shallow-LVF rupture would be limited to M ~ 6 to 6.2 based on the Fault Rupture Width-Magnitude empirical relationship (Wells and Coppersmith 1994). This structural separation may also explain the narrow and shallow triggered fault ruptures on the northernmost LVF during the 2018 Hualien earthquake, as the fault rupture may be stopped by the shallow-dipping CRF-related fault plane beneath this truncated fault segment.

As for the deeper segment of the LVF north of Ruisui, we expect it remains a major seismogenic structure since its fault plane still connects to the untruncated LVF in the south. This deeper segment also contains sufficient fault width (> 10 km) to accommodate tectonic strain from PSP-EP plate motion. The last fault rupture on the untruncated LVF adjacent to this deeper LVF section is the surface rupture of the Nov-1951 earthquake, where 1 to 2 m high coseismic fault scarp can extend to ~ 23.6° N based on the eyewitness accounts and written records (Shyu et al. 2007). Once again, the northern termination of the 1951 rupture coincides with the CRF-LVF truncation area (~ 23.7° N), which implies the change of structural geometry at ~ 23.7° N plays an important role in controlling the coseismic fault ruptures propagation during major LVF earthquake events.

Other than the LVF-related fault system, the eastward termination of the CRF-related seismic cluster also raised an important question regarding its interaction with pre-existing structures offshore Coastal Range. Our seismic data does not provide direct evidence of the eastward termination of the gentle, west-dipping, CRF-related structure offshore Hualien. The seismic cluster associated with this structure follows the lower margin of the low-velocity patch, presumably near the base of the Coastal Range’s sedimentary and volcaniclastic rocks, and terminates close to a bathymetric slope break adjacent to the Philippine Sea’s deep ocean floor (Hsieh et al. 2020) suggest this submarine slope change is corresponding to the active Takangkou-High Thrust within the Offshore East Taiwan Thrust Belt (Figs. 1 and 9). According to their structural interpretation, the Takangkou-High Thrust is an active west-dipping splay fault within this thrust belt, associated with the change of Chimei Submarine Canyon’s morphological pattern but without a clear fault scarp developed on the canyon floor (Hsieh et al. 2020). The bathymetric slope break at the Takangkou-High Thrust becomes much more clear north of 23.6° N from the bathymetric data, and its location roughly coincides with the eastward termination of the CRF-related seismic cluster (e.g., Profile L–L′; Fig. 6). Thus, we cannot exclude the possibility that after the young CRF-related fault truncates the LVF, this younger fault system connects and reactivates pre-existing structures within the Offshore East Taiwan Thrust Belt.

While the reactivation of the structure within the Offshore East Taiwan Thrust Belt is plausible from our seismic observation, we do not see that the entire basal detachment beneath the Offshore East Taiwan Thrust Belt is seismically active. Even though the minimum depth of the basal detachment from the balanced cross section is ~ 7 km at the latitude of Ruisui (Hsieh et al. 2020), similar to the hypocenter depth of the easternmost CRF-related seismic cluster, our relocated seismic cluster does not show the earthquake hypocenters propagate through this interpreted basal detachment. The lack of seismic activity at the other part of the basal detachment, and the high geological slip rate inferred to the active Chimei Canyon Thrust (Malavieille et al. 2002; Hsieh et al. 2020) lead us to believe that the remaining part of the basal detachment is likely locked. Further investigations would be thus important to understand its future seismic potentials.

Lastly, from a broader view it seems to be apparent that the series of the Hualien earthquake sequences since 2018 is rupturing in different deep and shallow fault segments in turn of the proposed CRF-related west-dipping fault system (Fig. 4b). There have been two segments remaining un-ruptured in this series (Black dashed circle, Fig. 4b). The southern one had previously ruptured in 1990 Hualien earthquake sequence (Lee et al. 2011), where the relocated aftershocks also showed a consistent west-dipping distribution from 3 to 12 km. But further attention may be called for the northern one. The ruptures on the adjacent segments could have imparted stress to the locked segments and advance their recurrence cycle.

5 Conclusion

The northern Longitudinal Valley has been the most seismically active and tectonically complex area subject to the transition from the oblique collision to subduction in Taiwan. Motivated by a recent seismic crisis of the 2018–2021 Hualien earthquake series, we combined the 3-D velocity models, the relocated earthquake clusters, and the redetermined focal mechanisms to re-access the seismogenic structure with surface geological observations in the region. The results show that the recent occurrence of the 2018–2021 Hualien earthquake sequences help reveal a primary west-dipping fault system more clearly than the past, with a steep and deep (25–15 km) fault plane beneath the eastern flank of the Central Range and a gentle and shallow (15–8 km) fault plane extending eastward beneath the valley and the Coastal Range (Fig. 6). The geometry and the reverse-sinistral nature of the deep fault plane is consistent with those fault planes illuminated by the 2013 Ruisui earthquake in the central part and 2006 Taitung earthquake in the southern part of the Longitudinal Valley, which evidently delineate the northern part of the CRF between the latitudes 23.5° to 24.2°.

The gentle and shallow fault plane is, however, a structure that only becomes clearer after the 2018–2021 Hualien earthquake sequences. It was previously identified as a detachment only beneath the Hualien and northern offshore areas but now shows a southward continuation down to the latitude of 23.7° N by the 2020/02/15 and 2021/07/07 earthquake sequences. We suggest that this fault is a branching-out structure from the deep CRF and could possibly link to the Offshore East Taiwan Thrust Belt. The geometrical analysis through the closely-spaced velocity and seismicity profiles also indicates that this shallow portion of the CRF may truncate the east-dipping LVF at the depth of around 10 km or less.

While further investigations for the structural transition from the southern to the northern Longitudinal Valley are needed, our model provides a new view of the seismogenic structures and their relationships beneath the northernmost valley. The truncation of the LVF has important implications for regional tectonics and future earthquake potentials. We infer the truncated shallow LVF segments in the northern Longitudinal Valley were evolved to be a transpressional strike-slip fault system that mainly accommodates the LVF-parallel motions. Part of the LVF-normal motions are likely consumed by the reactivated structures in the Offshore East Taiwan Thrust Belt offshore Hualien coast. These reactivated structures are likely connected to the shallow and gentle portion of the CRF identified in this study, facilitated by the northwestward subduction of the Coastal Range and the forearc with the PSP. The rest of LVF-normal motions may still accumulate on the deep LVF segment while the shallow truncated segment of the LVF may only serve as a secondary fault system that is diffusive, passive, and is easier to be triggered locally by nearby earthquakes, such as the repetitive shallow ruptures on the Milun Fault in the 2018 and 1951 Hualien earthquakes.