Collision is among the important processes for the growth of continents, and the way in which subduction becomes a collision is still an active research topic. Here, I examine the seismogenic structures of southern and central Taiwan where the subduction along the Manila Trench has terminated and given way to an arc-continent collision on land. Based on focal mechanisms and seven finite-fault slip models, coseismic tectonic extrusion is active in this region, in which the basement highs on the incoming passive margin are acting as indentors and strongly modifying the seismic moment release patterns in the collision zone. At least three magnitude 7 earthquakes have ruptured both north and south of an indentor called the Peikang high in the past hundred years. After examination, the basement highs show little global positioning system (GPS)–recorded relative motion with respect to the incoming passive margin; high Bouguer gravity anomalies associated with denser materials of the basement; and low heat flow due to less dewatering and exhumation. With regard to seismogenic structures, faster GPS relative motions, lower Bouguer gravity anomalies, and higher heat flow characterize the regions surrounding the indentors. Similar processes might be operating in other arc-continent collision zones. For other regions where there are fewer seismic instruments to monitor earthquakes, it might be helpful to combine a geological survey with gravity and other geophysical data sets to help identify such potential seismogenic structures.


Crustal deformation styles change when subduction terminates and becomes a collision. Some of these structural changes have been successfully documented by studying surface geology and other data sets (Silver et al., 1983; Teng, 1990; Abers and McCaffrey, 1994; Wu et al., 1997; Brown et al., 1998; Draut and Clift, 2001; Kao et al., 2001; Malavieille et al., 2002; Dilek et al., 2010). In the study of many mature collision zones, the map-view tectonic extrusion model is a leading hypothesis (e.g., Peltzer and Tapponnier, 1988; Houseman and England, 1993) used to explain observed collisional features; however, it is still unclear whethvner and how such process can occur coseismically, and whether smaller-scale (tens of kilometers) indentors can cause map-view extrusion in a style similar to that of the continental scale. It is also not clear if there are geophysical features that can be correlated with these potential earthquake rupture zones. These questions are important for seismic hazard analysis purposes. Here, recently published earthquake waveform modeling results are used to interpret seismogenic structural patterns of an arc-continent collision in the Taiwan region where the subduction zone terminates as a result of the arrival of the thick and buoyant continental passive margin at the convergence zone.

The incoming passive margin basement is composed of horsts and grabens, which were formed by many strike-slip and normal faults. These steep preexisting, or subvertical weak zones appear to have been reactivated in the collision zone, thus complicating the seismogenic structures due to the preexisting weak fault surfaces. In particular, the basement highs can act as indentors that influence the style of coseismic energy release. Also included will be a discussion about the ways in which global positioning system (GPS), gravity, and heat-flow data sets can be used to help interpret these high seismic hazard zones in other collision zones.


Taiwan is located along the boundary between the Eurasian and Philippine Sea plates. To the south, the oceanic lithosphere of the South China Sea is subducting eastward beneath the Philippine Sea plate (Bowin et al., 1978) with a high rate of 7–9 cm/yr (e.g., Yu et al., 1997). To the north, the subduction terminates and changes into arc-continent collision where the Chinese passive margin enters into the convergence zone (McIntosh et al., 2005). Thus, in map view, Taiwan captures an along-strike transition from subduction of oceanic lithosphere to collision of a passive margin with the same convergence zone, with the resultant termination of subduction (Fig. 1). Currently, the rate of southward propagation of the Taiwan collision zone along the convergent boundary ranges from 55 to 90 mm/yr (Suppe, 1984; Byrne and Liu, 2002; Willett et al., 2003).

Along the subduction zone, the mid-ocean ridge of the South China Sea plate used to spread in a NW-SE direction (Briais et al., 1993). To the north in the collision zone, the Chinese passive margin has NE-SW–striking normal faults dipping to the SE and NW (Liu et al., 1997; Hsu et al., 2004). Some of the normal faults are linked by NNW-SSE–striking transfer faults. Some of the normal faults are deep rooted (Nissen et al., 1995; Wang et al., 2006). Based on gravity modeling, Nissen et al. (1995) documented Moho-cutting fault structures along a passive margin transect west of Taiwan. Near Taiwan, the continental shelf shows more complex structures, including horsts and grabens that generate basement highs and thick sedimentary strata. There are two basement highs on the incoming plate west of Taiwan: the Peikang high and Kuanyin high. Lin et al. (2003) published a basement map, showing these basement highs. Lu et al. (2002) also interpreted the influence of the basement highs on the curvatures of the fault traces in the overriding plate in this region (Fig. 2). Recently, Byrne et al. (2011) discussed in detail how such basement highs affect mountain-building processes. They proposed that the basement highs and the basins surrounding them are results of spreading direction changes. These structures on the incoming plate can influence the deformation on the overriding plate.

The deformation styles of the overriding plate also vary from subduction to collision. The subduction zone includes a well-developed accretionary prism, forearc basin, and arc. The accretionary prism can be classified into three domains (Reed et al., 1992). The lower slope domain near the trench shows typical fold-and-thrust structures. The upper slope domain is transparent in seismic profiles, suggesting intense deformation or subvertical strata. The back-thrust domain lies along the boundary between the accretionary prism and the forearc basin where a tectonic wedging process is active (Chi et al., 2003). The forearc strata are subparallel, some with mass-wasting deposits (Yen and Lundberg, 2006), at least at shallow depths. The arc is subparallel to the trench with a constant distance of ∼120 km in the subduction zone, but the distance decreases where the subduction zone terminates near Taiwan, and the forearc basin is consumed in the collision zone.

The island of Taiwan is located in a mature collision zone. From west (foreland) to east (hinterland), there are six geological units: Coastal Plains, Western Foothills, Hsuehshan Range, Central Range, Longitudinal Valley, and Coastal Range. The Coastal Plain contains passive margin structures overlain by a vast amount of sediments eroded from the east since the time the collisional mountain-building process has been active. In the Western Foothills, the thrusts show curvatures with wavelengths of ∼100 km. Some of the thrusts are connected by transfer faults. The Hsuehshan Range is interpreted as a pop-up structure from reactivation of normal faults bounded by a previous graben (Teng, 1990). The Central Range is the northern extension of the upper slope domain of the offshore Taiwan accretionary prism. The Longitudinal Valley is the northern extension of the consumed forearc basin, and it has taken up at least half of the current 7–9 cm/yr interseismic shortening across this convergence zone (Yu et al., 1997). The Coastal Range is an accreted volcanic arc. Based on paleomagnetic data, the arc was segmented and rotated clockwise when it entered the collision zone (Lee et al., 1991). There is no morphological evidence of a forearc basin in the collision zone (Lundberg et al., 1997), where the collided arc is juxtaposed directly against the Central Range of Taiwan.

Due to the oblique plate-boundary configuration in the Taiwan region, different stages of subduction and collision occur simultaneously along the strike of the convergent boundary. As a result, the evolution of seismogenic structures can be studied based upon evidence from the younger subduction zone to the south to the collision zone to the north. In particular, we are interested in determining if there are unique seismogenic structures that are associated with the termination of the subduction in the mature collision zone in Central Taiwan, where the basement highs along the passive margin have entered into the convergence zone as indentors.


In this study, I will mainly use two seismic data sets derived from our group. The first data set was published recently (Chang et al., 2011) and is based upon a new procedure we developed to invert moment tensor solutions using Taiwan’s excellent strong motion waveform data. The second data set is from our previous work on finite-fault inversions of the Chi-Chi main shock and its large aftershocks (Chi et al., 2001; Chi and Dreger, 2002; Chi and Dreger, 2004). The reliability of these two data sets has been carefully documented and rigorously reviewed in our previous work. In addition, another moment tensor catalog (e.g., Kao et al., 2001) derived from data from a regional broadband network (BATS) will also be discussed briefly. Because these solutions are constrained by full waveforms, these solutions are usually considered of higher reliability. In this work, I will focus on geological and geophysical interpretation of these data sets and correlate them with other available geophysical data in order to study the seismogenic structures where the subduction terminates.

We (Chang et al., 2011) inverted strong motion waveforms to derive moment tensors of many Mw >4.8 and greater earthquakes since 1993. In the 2011 paper, a new procedure used to derive the solutions was documented. Here, I will discuss the interpretation of these solutions. This catalog covers a large portion of the seismic moment release in this region (Fig. 3). Overall, the maximum compressive stress (sigma 1) direction is subparallel with the plate convergence direction, except in regions where the arc starts to dock and rotate along the passive margin.

In the foreland region, including the Coastal Plains, Western Foothills, Hsuehshan Range, and western flank of the Central Range, there are diverse focal mechanisms from shallow depths down to the lower crust. Most of them reflect thrusts or reverse mechanisms, but there are also strike-slip faults. With the exception of the Chi-Chi main shock and some of its aftershocks, which we will discuss later, most of the dip-slip events are of relatively high angles and at greater depths, some as deep as 40 km. Lin and Roecker (1993) have also proposed a cluster of seismicity down to 60 km at latitude 24°N.

In the hinterland region, particularly the eastern flank of the Central Range, there are two groups of particular focal mechanisms. In the south, between 23°N and 23.5°N, where the arc starts to collide with the passive margin, there are mostly strike-slip faults in the shallow to midcrust area. As the collision becomes more mature at 24°N, we see mostly normal faults in the hinterland.

This catalog also includes a 1999 earthquake sequence in Meishan in the foreland region under the Coastal Plains. This earthquake sequence, which occurred one month after the Chi-Chi main shock, is unique because there were two M 7 earthquakes that occurred in this region in 1906 (ML 7.1) and in 1941 (ML 7.1), causing substantial damage. A similar event now would cause much more damage due to the current dense population in this region. Several very different and conflicting seismogenic structures have been interpreted, including right-lateral, left-lateral strike-slip faults, and thrusts. From the catalog (Fig. 4), we found that this sequence is dominated by a NE-SW–trending strike-slip fault with reverse faults at the ends of the fault trace. Shyu et al. (2005) also interpreted the strike-slip fault as a transfer fault. This earthquake sequence ruptured the basement and occurred along the southern boundary of the Peikang high (Fig. 2). A month prior to this earthquake sequence, the Chi-Chi earthquake sequence ruptured to the north of the Peikang high.

We inverted finite-fault source models of the Chi-Chi earthquake and its large aftershocks (Chi et al., 2001; Chi and Dreger, 2002; Chi and Dreger, 2004). The main shock ruptured within a triangular region that is bounded by the Peikang and Kuanyin basement highs. During the earthquake sequence, two transfer faults were also identified (Kao and Chen, 2000; Chi and Dreger, 2004). From our finite-fault model, we found that at least one of the strike-slip faults ruptured in the basement at depth with a small moment release; thereafter, most of the moment release occurred above the décollement (Fig. 5).

A cross-section view (Fig. 6) shows that the Chi-Chi main shock ruptured on the ramp and flat of the décollement. There was another aftershock that ruptured on the flat. We identified a back thrust that ruptured above the décollement, and a reverse fault that ruptured from the décollement to the basement. As mentioned in the previous paragraph, the earthquake nucleated on a vertical strike-slip fault in the basement, but then ruptured upward, with most of the moment release occurring at shallow depth on the vertical transfer fault above the décollement. Kao and Chen (2000) interpreted this strike-slip fault as a transfer fault, similar to the lateral ramp of the thrusting on the décollement.


The steeply dipping reverse faulting and subvertical strike-slip faulting in the mid- or lower crust suggest that these are reactivations of preexisting structures from the passive margin (Fig. 6). Some of the normal faults might have cut the lower crust as suggested by Nissen et al. (1995). In addition, there are new thrusts that developed in shallow sedimentary strata deposited during the collisional processes. However, even the shallow earthquake events are affected by the passive margin structures. For example, the strike-slip fault at a shallow depth is collocated with the strike-slip fault plane at basement depth. Also, the Chi-Chi main shock ruptured between the basement highs.

There are debates about whether seamounts (or other indentors) can act as asperities or barriers for earthquake ruptures (e.g., Tsumura et al., 2009; Wang and Bilek, 2011). In the case of Chi-Chi main shock, the basement highs acted as barriers that stopped the rupture of the main shock. The distance between the basement high thus defines the length of the earthquake rupture surface. Based on scaling relations, we can roughly estimate the size of the earthquake using the length of the earthquake rupture surface, i.e., distance between the basement highs, for seismic hazard estimates or earthquake simulation scenarios.

We found that the areas north and south of the Peikang high are capable of generating Mw 7 or greater earthquakes: Mw 7.6, 1999 Chi-Chi earthquake to the north, and M 7.1, 17 March 1906 Meishan earthquake and M 7.1, 17 December 1941 Chungpu earthquake to the south of the Peikang high. These large events can rupture on the strike-slip transfer faults or on the thrust fault plane. In other words, in the case of Taiwan, the three-dimensional structures of the tectonic extrusion include the bounding strike-slip faults (or subvertical reverse faults) and a very shallow dipping thrust in between.

It is worth noting that the time span of these results is less than 100 yr and may not represent the total coseismic release patterns in this region over a longer time scale. However, the particular seismic release patterns reported in this study are related to the termination of the subduction and show at least part of the diverse deformation styles existing in such a tectonic setting.


Both the Chi-Chi sequence and 22 October 1999 Meishan sequence are related to the Peikang high, which acts as an indentor to cause the coseismic tectonic extrusion. Next, this paper will explore hints in other geophysical data sets that can help to identify such indentors and the associated high seismic risk zones.

Above the basement highs that act as indentors, the GPS velocity is relatively small compared with the other parts of Taiwan (Fig. 7). The basement highs (Fig. 2) are rooted in the passive margin and thus are relatively stationary with respect to the stable Asian continent. As a result, there are very small strain rates in the basement high regions, but there are relatively large interseismic strain rates along the thrusts right next to the basement highs (Hu et al., 1997; Hsu et al., 2009).

The basement highs are composed of denser materials than the sedimentary strata, and this should be reflected by positive Bouguer gravity anomalies. We found that the two basement highs are clearly visible in the Taiwan Bouguer gravity anomaly map (Fig. 8). As a result, the Chi-Chi main shock is identified as having ruptured in the low-gravity region (blue color) between the two gravity highs.

Heat flow is low above the indentors, but high in front (east) of the indentors (Fig. 9). The higher heat flow might be related to shortening and exhumation or it might be related to the more active dewatering in these zones. There is an active upward gas- and fluid-migration process east of the indentors (e.g., Yang et al., 2006). Similar active deformation is also operating in the fold-and-thrust belt between the indentors.

We found that combined analyses of GPS, gravity, and heat-flow data sets can help to identify the basement highs that act as indentors. For other collision zones around the world, it might be interesting to study regions with similar geophysical features to better explore their seismic risks. In the collision zone, the rapid erosion on the mountains and sedimentation on the continental shelf and slope might have buried some young structures near the frontal thrust region. As a result, these high seismic risk zones might be located near where the geomorphic features are relatively flat and show less surface expression. Seismology, GPS, and gravity observations need to be combined with field observations to better understand the seismic risks in such regions.


The dense population in Taiwan makes dense seismic instrumentation necessary. We used the excellent seismic waveform data sets to invert for seismogenic structures using some state-of-the-art methods. The technical aspects of these methods and initial results have recently been published in earthquake seismology literature. Here, I use the results from these studies to study the seismogenic structures of this region. This is among the most closely seismic-monitored collisional regions with the most comprehensive coseismic energy release data sets in the world. In addition, the oblique convergence in Taiwan region makes subduction and collisional processes occur simultaneously along the convergence zone. We can thus study the range of seismogenic behavior where subduction terminates.

In the collision zone where the “typical” subduction process terminates, the irregular basement topography along the passive margin influences the deformation style (Fig. 10). There is more strike-slip faulting in the collision zone. The horsts on the passive margin that enter into the convergence zone can act as indentors and cause coseismic tectonic extrusion. Here, I have documented coseismic tectonic extrusion deformation on both sides of one indentor, which has generated very high seismic risks near these extrusion regions. I also documented characteristics of GPS, gravity, and heat-flow patterns near these seismogenic zones. The lessons learned from this region might be applied to other ongoing or matured arc-continent collision zones where seismogenic structures are difficult to study.

I thank John Wakabayashi and Science Editor Raymond M. Russo for the great editorial work. I also appreciate two reviewers for the helpful comments. I wish to thank many of my colleagues who made this work possible. This research is partially funded by National Science Council of Taiwan under grant 101-2116-M-001-024 to WCC. This is Contribution Number IESAS1719 of the Institute of Earth Sciences. This is Taiwan Earthquake Center contribution 00085.