Preliminary Results of a Large Offset Seismic Survey West of Hengchun Peninsula, Southern Taiwan

As a part of the U .S.-R.O.C. collaborative deep seismic imaging experi­ ment, six ocean-bottom seismographs (OBSs) were deployed along the 12°12'N parallel west of Hengchun Peninsula and wide-offset seismic sig­ nals from a large air-gun array shot over the line were recorded. A pre­ liminary analysis of these data together with those from another OBS to the east of the peninsula reveals that a 11-km thick crustal layer, inter­ preted to be an extended continental crust of the southern margin of the Chinese mainland, lies under most of the line and dips toward the east at low angle. A complex margin wedge is observed at depth immediate west and underneath the peninsula, which may have resulted from the thick sedimentary layer, quite likely including the basement complex, pushed against the strong backstop provided by the Philippine Sea plate converg­ ing from the east. (


INTRODUCTION
The area in and around Taiwan is one of the tectonically most active areas on the surface of the earth. However, our understanding of the complex process of collision between two lithospheric plates, the Eurasian plate and the Philippine Sea plate, in the area is still incom plete despite extensive studies involving surface geology (e.g., Ho, 1986), geodesy (e.g., Yu, streamer, ocean-bottom seismographs (OBSs) and land stations.
The analysis of the vast quantity of acquire data has not yet been completed, pending combined interpretation of the entire data set from several diverse groups that participated in the experiment. This paper, therefore, is an attempt to present preliminary results from a set of OBS stations along a single seismic line, TAICRUST Line 33, lying west of Hengchun Penin sula at the southern tip of the island (Figure 1 ). We anticipate that publication of these results, together with similar papers in this volume covering other lines and line segments, will facili tate future interpretation of the combined data set.

TECTONIC BACKGROUND
Because of the oblique nature of the collision between the Eurasian plate and the Philip pine Sea plate (Seno, 1977), the focus of intense deformation in and around Taiwan, caused mainly by the collision of the Luzon Arc on the Philippine Sea plate to the eastern continental margin of the Chinese mainland, has been progressing generally from north to south in the region (e.g. Huang, et al., 1997). In this tectonic setting, the Hengchun Peninsula is located in the area of early collisional deformation, with highly deformed island of Taiwan to the north and pre-collision, simple subduction of the South China Sea along the Manila Trench to the south.
The T AICRUST Line 33 west of the peninsula and its companion line, Line 29 to the east of the peninsula, cross this area of incipient collision (Figure 1). Within this area, the Luzon Arc, with its forearc basin and accretionary prism, resulting from the subduction of the South China Sea under the Philippine Sea plate, is beginning to collide with the extended continental crust of the Chinese mainland. The study reported here is an attempt to delineate the deep crustal structure in the area to assist in understanding of the process of the collision.

Field Experiment
The data reported in this and its companion papers in this volume were acquired using ocean-bottom seismographs (OBSs) of the University of Texas Institute for Geophysics (UTIG) and National Taiwan Ocean University (NTOU). They were the upgraded versions of the digital OBSs described in Nakamura et al. (1987) and Nakamura and Garmany (1991). The instruments were deployed along each seismic line from RIV Ocean Researcher I ahead of shooting with a large air-gun array of RIV Maurice Ewing, and were retrieved after the shoot ing.
A total of 39 deployments were made along six seismic lines during the experiment, re covering all with full data. The locations of the instruments used in this paper are listed in Table 1. Each UTIG instrument recorded data in four components -gimbal-mounted three-

Data Processing and Analysis
The acquired OBS data were first processed through standard processing steps using a software package OBSTOOL (Christeson, 1995). These steps include: (1) computing precise location and orientation of each instrument on the ocean floor using water-wave arrivals (Nakamura et al., 1987), (2) merging with the navigation data, (3) rotating the horizontal axes to radial and tangential directions, and (4) reformatting the data to produce data files in stan dard SEG-Y format. We used differential GPS (DGPS) navigation data as processed after the cruise by the Institute of Oceanography of National Taiwan University, and applied a correc tion for the setback of the center of the air-gun array from the GPS antenna to compute the precise location of each shot.
A set of SEG-Y format data tapes thus generated is archived as a permanent record, and used in all subsequent analyses. Generally, the data were further processed through several standard waveform shaping steps such as deconvolution, band-pass filtering and automatic gain control, and were plotted as standard record sections for interpretation. For the analysis reported in this paper, we used a processing package SIOSEIS provided by Paul Henkart of Scripps Institution of Oceanography.
Our analysis of the data to date was mainly to derive a two-dimensional P-wave velocity model of the vertical cross-section below the seismic line. We picked arrival times on the record sections and used forward modeling and inversion to derive a velocity model using the 2-D ray tracing/inversion program of Zelt and Smith (1992). The starting model for this exer cise was constructed first by combining the results of layer solutions based on reciprocal arriv als at each pair of stations and arrivals at each station in a split-spread profile configuration, together with the very shallow structures estimated from the coincident multichannel seismic (MCS) record section ( Figure 2). Both first and later arrivals were used in this exercise to ascertain inclusion of prominent refractors in the model even when they did not appear as first arrivals.
Although the Zelt and Smith scheme allows one to formally invert arrival times to com pute the v' elocity structure of the model, such inversions as applied to a large region of the model turned out to be unstable because of the complexity of the structure in this area. In particular, large lateral variations, both in geometry of interfaces and velocities within each layer, coupled with relatively large station spacing caused problems. For this reason, the use of the inversion scheme was limited only to certain local regions of the model where such inversion was stable.
As stated above, the analysis is still incomplete. We are currently in the process of apply ing certain other analysis techniques, including synthetic seismograms and tomographic in version of first arrival times, but we must defer their results to later papers. Also, the results presented here are based only on vertical geophone data. Analyses of other components,    Figure 4 shows a P-wave velocity model constructed to fit these arrivals following the procedure described above. As usual, not all model parameters are equally well constrained. As is clear from the ray diagram of Figure 5a, the deep structure at either end of the model is completely unconstrained. Furthermore, the layer thicknesses and velocities are less certain west of station 45 because the model parameters there are based on unreversed seismic arriv als. The same is true with the structure below 6 km depth east of station 40.  3 structural heterogeneity of smaller scales than the model parameterization intervals. Note also that rms travel time error is within normally acceptable limits for crustal scale modeling. It is quite possible that the velocity variations perpendicular to the seismic line (the 3-D effect) is also contributing to these high X 2 values. A possible cause of the high X 2 values for phases 61, 71 and 81 will be discussed later.

4.RESULTS
In the model, the following main structural components can be identified: 1. The topmost layers of total thickness up to about 8 km in which the P-wave velocity gener ally increases from about 1.6 km/s just below the sea floor to about 3.6 km/s near the bot tom. These layers can be interpreted to be sedimentary layers with the degree of compac tion increasing with depth. These layers are thinner near the peninsula. 2. A layer (two layers in the model), about 11 km thick, lying below the sedimentary section with the P-wave velocity generally increasing with depth from about 6 km/s to 7 km/s and dipping g e, ntly from west to east. The velocity contrast from the sediment above is distinct. From the seismic velocity and thickness of the layer, it can be interpreted to be extended continental crust which occupied the position between the Chinese mainland and the oce anic basin of the South China Sea. This thickness is within the range of crustal thicknesses The dip of this slab appears to increase from about 2° under the western half of the profile to more than 8° as it approaches the peninsula. However, the velocity at the top of the slab becomes indistinct from that of the material above as it dips below the peninsula.
3. A wedge of highly complex structure that lies between the sediments and the slab east of station 43 with average P-wave velocity intermediate to those of the sediments and the slab. This wedge of material has a geometry similar to an accretionary prism associated with subduction of the slab from west to east. Within this wedge, there is a general increase of average velocity from approximately 4 km/s to 5 km/s from west to east However, the details of velocity variation appear to be far more complex than the velocity variation repre sented in the model. A comparison of seismic record sections across the wedge shows that arrivals from a series of shots west of the peninsula (Line-33 shots) as observed at a station east of the peninsula are highly irregular with wavy first arr ivals, while those from a series of shots east of the peninsula (Line-29 shots) as observed at stations west of the peninsula are regular with smooth first arrivals (compare arrivals at Stations 39 and 40 in the model distance range of 120-140 km with those at Stations 40 and 41 in the 200-220 km range in Figure 3a). Furthermore, the character of refracted arrivals from this layer (Pa of Figure 3b) is far more complex than those of other arrivals: Numerous closely spaced arrivals follow the first arrival, in contrast to relatively transparent nature of seismic traces following other first arr ivals ( Figure 6). What is shown in Figure 4 is a result of a relatively simple approach using an irregular interface to model these arrivals. However, despite countless geometrical combinations, all interface patterns we tried resulted in rather high rms errors and large X 2 values for arrivals 61, 71 and 81 ( Table 2), suggesting that another solution may be re quired. One way to explain this behavior is to introduce some form of alternating high and low velocity layers steeply dipping from west to east inside the wedge, as schematically drawn in Figure 7. To find out the details of this structure requires more closely spaced .OBS stations than were used in the experiment, but preliminary synthetic seismograms (work in progress) computed for a structure similar to that of Figure 7 show similar charac teristics, reproducing both the contrasting (wavy vs. smooth) shapes of the first arrivals and the enhanced energy following the first arrivals.

DISCUSSION
Although the results reported here are still preliminary, the OBS data have revealed sev eral interesting features under this seismic line. In this section, we will discuss likely geologi cal significance of the main features listed in the preceding section.
The area as a whole is characterized by an extended continental crust of the southern margin of the Chinese mainland subducting under and colliding with the western front of the Philippine Sea plate, which lies east of the study area (see the accompanying paper by Chen and Nakamura in this volume for details of the structure east of Hengchun Peninsula). This collision is a relatively recent phenomenon, or even in its incipient stage, in this part of the plate boundary. As evidenced by the existence of arc-fore-arc-basin morphology to the south east of the study area, active eastward subduction must be involved in shaping the area, at least in the past if not fully active at present. This process, including earlier subduction of normal oceanic crust, has produced a margin wedge above the subducting slab. As the subducting slab changed its character from pure oceanic to more continental type as the Philippine Sea plate moved northwestward, the subduction may have been impeded somewhat. However, the distribution of earthquake hypocenters and their focal mechanisms (Tsai, 1986;Wu, et al., 1991) suggest that the eastward subduction of the slab under this area is still active, and the present study also shows that the extended continental crust maintains its downward dip to ward the east.
Looking at the structure in more detail, the subducting slab is covered with a sedimentary section, about 7 km thick. This compares with the 2-5 km thickness of the syn-and post-rift sediments, mostly from the Chinese mainland, measured along the south China margin to the southwest (Nissen, et al., 1995). The sedimentary section under Line 33 is thicker, which suggests that in addition to the mainland-derived sediments this sedimentary section also in cludes a significant contribution derived from erosion of rapidly uplifting island of Taiwan as this area lies on the southern edge of the Kaoping slope to the north.
Although the pre-collision structure likely included a sedimentary accretionary prism, the margin wedge built over the subducting slab appears to be more complex than a simple se quence of folds and thrust faults involving the presently observed sedimentary section. The OBS data suggest that the wedge consists of material with an average velocity intermediate to those of the sediments and the slab. Our more detailed, but still tentative, model suggests that eastward dipping layers of alternately high and low velocity may best characterize this wedge.
This structural configuration is consistent with development through underplating and/or ma jor out-of-sequence thrusting and which quite likely includes basement complex accreted from the top of the transitional subducting crust.
The material in the wedge was heavily compressed and deformed as it approached the backstop (Luzon Arc), showing relatively high seismic velocity under Hengchun Peninsula even near the surface. The bottom of the wedge is now seismically indistinguishable from the top of the slab. At its eastern extreme, it becomes of sufficiently high velocity that it is com parable to continental crust in the inteipretation of seismic data to the east (Chen and Nakamura, 1998).
Details of the structure under the peninsula, the nature of the backstop and the boundary between the wedge and the advancing front of the Philippine Sea plate are yet to be determined with a combined analysis of the results of this paper together with those from land stations on the peninsula and from other OBSs to the east of the peninsula. However, to find out the detailed internal structure of the wedge may require OBS receiver spacing much closer than that used in the present experiment. The hypothetical structure depicted in Figure 7 should show anisotropy in macroscopic scale, with waves traveling in the lower right-upper left di rection, parallel to the layering, exhibiting faster speed than those traveling in the lower left upper right direction across the layers. In fact, there is a suggestion in the data that this is indeed observed. The arrivals that cross the offshore portion of the wedge, where the ray coverage is more omni-directional than under the peninsula (Figure 5a), tend to be late when the ray crosses the wedge in the lower west-upper east direction and early when the ray crosses the wedge in the upper west-lower east direction (Figure 5b). An experiment may be devised specifically to confirm/refute this observation and thus to test the hypothesis.

CONCLUSIONS
A preliminary analysis of the OBS data from a seismic line west of Hengchun Peninsula at the southern tip of Taiwan has revealed that the area is underlain by an 11 km thick extended continental crust subducting at relatively low angle to the east. The thick sediments on top of the subducting slab together with the strong backstop provided by the colliding Philippine Sea plate to the east of the profile are producing a complex margin wedge at depth. The wedge appears to contain high velocity material from deep in the sedimentary section on top of the subducting slab; however, the true nature of the wedge and the backstop is yet to be deter mined from combined analysis of all the available data.