Crustal Structure of the Northeastern Taiwan Area From Seismic Refraction Data and Its Tectonic Implications

Seismic refraction data from onshore and offshore experiments in the eastern-northeastern Taiwan region \\'ere used to study the velocity struc­ ture by the two-dimensional ray-t.racing method. In the \1elocity model, a structural fault boundary located beneath the Longitudinal Valley "ras used to separate the northern Coastal Range (CR) on the eastern side from the eastern flank of the Central Range (EFCR) on the western side. The P­ 'vave 'relocities from the surface to the dept. h of 1 2-15 km varied from 3.9 to 5.8 km/s beneath the CR and from 4.8 to 6.1 km/s beneath the EFCR. Com­ paring the velocity structures along various latitudes, it \\'as found that the CR extends northward to 24.2 <>N. The velocity structures of the CR, the Hsinchen Ridge (HR) and the Yaeyama Ridge (YR) indicate that the HR and the YR both belong to t. he same type of tectonic unit as the CR. To the north of 24.2<•N, the velocit)' structure of the Ilan Ridge (IR), located be­ t,veen the EFCR and the southwest. ern end of the Ryukyu arc, is similar to that. of the EFCR; hence, probably indicating it is the northeastern exten­ sion of the EFCR. This suggests that the EFCR bends eastward and be­ longs to the same tectonic unit as the southwestern Ryukyu arc. From a comparison of the velocity structures of the CR, EFCR and of other typical continental arcs, orogens and oceanic arcs in the literature, it can be con­ cluded that the northern CR b�longs to an oceanic arc and that the EFCR is a continental arc. Further more, from the analysis of the velocity struc­ tures beneath the CR and EFCR, it is believed that the upper crust of the CR is weaker in strength than the EFCR, which means that the arc-conti­ nent collision is not an appropriate model for the formation of Tai\\ran is­ land. (

Iision suture bet\\i•een the Eurasian and the Philippine Sea pl ates is generally believ•ed to be the Longitudinal Valley (L V) (Figu1•es l and • 2), which sepctrate.s the Coastal •Range (CR) in the.east frc.)m the re.st of Taiwan island.•To the east of the LY is the narro\v CR, made up of Nliocene t() Pliocene andesitic v•olcanic bodies and )'O unger agglo1ne1•ate and turbidite dep<)sits (e.g., Ho, 1986;Lundbe1•g and Dorsey, 19 88).To the west of the L V is the 1nain body ot• Tc1iwan, composed ot • a pre-Tertiary• metamorphic.basement ov•erlain b)' Paleogene low-grade rneta1norphosed sedin1e.nts�Neogene-f ' olded and thrusted sedimentary rock layers and Quater na1Ay allu\1ia1 deposits (e.g., Ho, 1986).The Tanana() Schist (TS) (Figure 2. ), prop()Sed by Yen  ( 1954), is a pre-Tertiary metamorphic complex and t � or1ns the oldest geologic tectonic element at• Taiwan which constituting the easte.rn flank of the Central Range (.EFCR) gec)Iogic prc)v•- Some geophysical studies have show n that there are actuall)' distinct dif'f' erences in tl1e crt1st structure between the CR and the.EFCR.The seismic experiment by • Tsai et ell.( 1974)  ctnd the earthquake travel ti1ne analysis by L. ee et czl.( 1986) pr()Vided some int' ormation as t() the c1•ustal velocity structure ot' the CR and EFCR.F1•()m the gra\1ity, magnetic, seismic and geologic data of eastern Tai wa11 and its of'f'shore areas, Hu Ll nd Chen ( 1986) showed that a tectonic break separates the CR t' rorn the EFCR.Ho\\1ever, the nature ()f� the north\\1ard e. xten Sl()n of the CR and EFCR has 11ot been unde1•stO()d.To the \V•est and south ot' the Nana() Basin (NB) .. there exists a ENE (_N70 <) ) and then SE (N 130 () ) trending ridge that includes the Hsinchen Ridge (HR) and the Yaeya1n<1 Ridge.(YR) (Figure 1 ).Alth()Ugh the topography shows that the.CR , HR and YR are in fact a continuous 1•idges (_ Figure I), whether 01• n()t they belong to the sar11e geologic<ll unit is an issue which still needs t' urther study.
In this study, with the use of seismic ret' raction d�lta fr()tn th1•ee seis111ic experin1ents con ducted separately in 1985, 1993 and I 995 and the two-di1nensional ray-tracing methl)d, the velocity prot'i les in the easte.rnTai\\ian area \\i•ere constructed and then used to investigate the northward and east\vard extensions of the CR and the EFCR.The re lationship between the CR� the HR and the YR and between the EFCR and Ryuk:y•u arc (Rii\) in terms ()f. the velocity structures in the prot' iles are also discussed.
": ...         used for that of the flan Ridge (IR; Yu and Hong� 1992;Song, 1994) and the Okinawa Trough (OT) and to compare them with the v• elocity structure beneath the EFCR and the RA (Figure 1 ).The 1993 experiment was carried out in the Hualien area (e.g., \\lang et al., 1995).In the sea, the controlled sources were detonated at approximate])' 150 m in depth.Seismic signals \V ere recorded by the Central Weathe.r Bureau Seismographic Ne.tvvork (CWBSN) and a por table teleme.tricseismic array• (PANDA-2) dev•eloped by Memphis State University (Chiu et c1l., 199 1) and de.p loyed b)' the Institute of Earth Sciences, Academia Sinica in Taipei, Taiwan .The construction of the ve.locity structures beneath the CR and the EFCR along L93-1 was based on the seismic records of 8 explosion shots obtained by land stations (Figure 2).The ''elocity structures along L9 3-2. to L93-5 \vere based on seismograms of the ex plosions re corded by• three PANDA-2 stations and the CWBS� station WHF (Figure 2).In a similar way, the velocity• structures along L93-6 and L93-7 \\'ere based on seismograms of explosive sources recorded by four CWBSN stati ons, namely TWC, ENT, NSK and NST (Figure 2).All of these velocity profiles \:Vere used to discuss the latitudinal variations ot, th e velocity structure be neath the CR and the EFCR.
In August and September ot, 1995, a comprehensi\i•e deep seismic imaging program (the T AICRUST program) was carri ed out in the offshore areas of T�tiwan ( : Liu, 1995).Profi les L95l and L95-2 (Figure 2) made use of the T\\irD and EHY stations of" the CWBSN and the controlled sources released f14om a 20-airgun array towed b)1 the R/V Mau14 ice E�1ing of Co lumbia University ot" the U.S.A.These t\vo profiles were used to construct the de.tailed veloco () ity structure of the CR at 23.5 N and 24 N.

DATA ANAL-YSIS AND INTERPRETATION
In this stt1dy1, it is as sumed that the velocit)l in each lay• er increases \\1i th a certain gradient, and a t�10-dimensional forV\•ard modeling technique based on the asymptotic ray theo f)' (Cerv1en ) ' et al., 1977) was used to obtain the t14avel time t"or a given model.The adoption of the velocit)' gradient model is bel ieved to better re.present the actual velocity-depth profile than the alterna ti\1e homogeneous layer model s (e.g.Kennett, 1977;Whitmarsh, 1978).The program RAY A1'11P-PC (Clossle)', 1987) \¥as adopted tor interacti ve computi ng.This program allowed for the rapid testing of t\\'o-dimensional , laterally hete.r()gene.ous\1elocit)1 models for trav'el time consistency \\1ith the data.For dift'erent source-rec ei ver distances, the tra\tel ti me curves of� the.f'irst arrivals of seismic wav•es e1nitted by the explosions and ai14 -gun sources corre sponded to different ra)' paths in the v• elocity model.In order to make it better understood ho\\1 the ra)1-tracing method \\1as used to construct the ve]<)City model, the ray paths of the fi rst arrivals are briefl)' described here.Later arrivals ne.ed not be described, howe\1er, because the ray-tracing technique is similar.The t"irst arrivals can be the refracted phases t�rom the crust (ret"erred to as the Pr phase), the reflected phases from interfaces in the crust (referred to as the PiP phase) or t'rom the Moho di sc.ontinuity (referred to as the PmP phase).In Figure 3, the Pr phase is designated as rl, r2, ... , and the PiP phase as i I, i2., ... , depending on the deepest lay•er they reached.Profile L93-I (Figure 3) is taken as an example.The record sections for shots SI� S4, S7 and S8 are shown in Figure 4.A remarkable phase i2, appea14ing at offse.ts of 30-50 km in Figure 4a, is interpreted as representative of the reflec ted \v ave.s from the bottom ot� the 5. 2-5.8 km/s layer in Figure 3b.The upper cr•ust of the CR is charac.terizedb)' a del ay• ed arri\1al branch rl which can be explained as the refracted arri\1als from the 3.9-4.5 km/s layer in Figure 3b.()

30
Distance (km) H07 HOS H28 H29 H17 H27 H09                            TAO, V(J/.7, N(J. 4, Del:errzbe1� 1996 The sei smic velocity models presented in this paper represent the simplest structures that best fit the tra\le] times of phases identified in the shot records.In mcJst cases, the calculated travel times agree with the data t() within 0. 1 s.l'\owhere, in fact, do they div•erge by more than 0.3 s.Before modeling, a static correction (w•ith \1 elocit)' 4.0 km/s) was app lied to each seis mogram f"or receiver elevation.These corre.ctionsappeared to be adequate, as indicated by the general!), good agreement between the calculated t"irst arrival times and the observed data.In modeling, 1.5 km/s was taken to be the velocity of" the water layer and the source \\1as located at the same depth beneath sea Ie,1el .The amplitudes ot� the seismograrns shO\\'n in this study (except•for the record sections fo r L.95-1 and L95-2) �'ere normalized individually to the maxi mum amplitude in the first I 0 seC()nds of the record� so they could not be strictly compared between traces.

Profile L86-GM
In order to recognize the existe11ce and location of' the \'ertical boundary (L V fault) de scribed here earlier, the grav•it)1 and magnetic se.ctions L86-GM \Vere used (Figure 2) alc)ng with the analytical signal analysis technique ot' Hsu et c1l.( 1996).
Figures 5a and 5b show the gravity and magnetic profi les along line.L86-G�1 in Figure 2 . .Figures 5c and 5d show• the amplitude curves of the analytic signal (Hsu et c1l., 1996) t�or the gravity and magnetic profiles, 14 especti vely.The boundar)' beneath the L V \\1as determined (b 1 in .Figure 5e) by picking the maximum amplitude in Figures 5c and 5d.Th is boundary \\1as used in the-follo\\1ing analysis of' the velocity models.

Profile L93-1
As sho\vn in Figure 3b, the velocity model t'o 11 prot'i le.L93-l is represented b)' a t�10dimensional seismic velocit)' model that is composed ()f.tWC) discrete and independent!)' de rived "blocks'' which are connected by a near ve.rtical boundary, the L-Y boundary.The depth of the base of the sedimentary layer (:2.0 km/s layer) under the shot points is at the depth of about 5. 5 km.It decreases in depth and thickness from shot point SI to station H0 9.The thickness of the sedimentary laye1 � is about 2 km beneath point S8 and tapers northwest\vard toward the station H09.Beneath the onland 2.52 km/s sedimentary• \v edge (Chen and Wang, 1994) in the LY, the 3.9-4.5 km/s and 5.2-5.8km/s layers are separated from the 4.8-5.4km/s and 5.7-6.l layers by a vertical boundar)' (Figure 3b).These layers are underlain by two plan ner layers with velocities of 6.2-6.9 km/s and 7 .2-7.8km/s.
In Figures 4a-4d, the travel time .c. urves predicted by the model are supe-rimposed on the observed vertical-component 14 eco1�ct sections.The arriv•al phase rl at the stations H09 and H27 (located on the CR and LY, respectively) is delay•ed about 0.3 to 0.5 s compared to the i2 phase observed at other stations located () n the E-FCR in the S 1 and S4 sections (Figures 4a and 4b).In the S7 and S8 sections (Figures 4c and 4d), this delay is not clear since the increasi ng ot, fset makes the rays penetrate.to deeper layers in Figure 3b.The wa\1eforms also reflect the larger background noise which appeared at H09 and H27 compared to that at the other stations.This can be due to the site effect: For the shot point S7 (Figure 4c ), high energy is recognized in an offset range of 45-57 km (denoted by• r2) and 60-73 km (denoted by i3).These phases are 15 Fig. 5. Geologica] boundaries determined from gravity and magnetic data using the analytical signal method developed by Hsu et al. ( 1996).interpreted as having ret'1• acted t•1•01n the 5.2-5.8kn1/s layer and reflected t'rom the.bottom ot• the 6.2-6.9 km/s layer, respecti vel)1• H()\\1ever, fl) r the sh(lt p()int S8 (Figure 4d), clear a1•riv,11s are observ•ed at offsets greate1• than 7() kin, \:vhich are interpreted as the PmP (M()hl) reflec tic .1n).Thus, the Moho depth C()uld be abl)Ut 25 kin beneath prt)file L93-1, but this is an(lthe1• issue which still 1• equires furthe1• investigation.
TA O_ , VlJ l. 7, N(J. 4, Dece111her 1996 3.9-4.5 km/s la)1er and a 5.2-5.8km/s la1• er.The thickness of' the fo rmer ranges from 4.5 km under the shot points to 5 km beneath station TW O, while the latter, similar to that described in the model for L93-1 (Figure 3b), is about 5 km in the center of the shot points but diminishes to ab()Ut 3.5 km under stati(ln TWD.The i I and i2 phases are interpreted here as wide-angle reflections from the top and the bottom of this layer, respecti vel)'.A series of' strongest later anivals MT appear at 9.5-15 km ot'fset (Figu14e 7a), and they1 a1•e interpreted as multiples (see als (l Cheng and Vv' ang, 1995), \v hich n1eans that their 1• ay paths include one or mot�e reflections betw een the free surface and the ocean fl oor. •

Profile L95-2
The \1elocity• model of profi le L95-2 is sho\\1n in Figure 8b.Its eastern part of' is similar to that of L93-l (Figure 3b ).Profi le L95-2 contains a 3.9-4.4km/s layer and a 5.1-5.7 km/s layer.Ho\\i•ever, in the western part of the model, the 3.9-4.4km/s layer comes in contact with a 2.5-2.9 km/s sedimentary layer and the 5.1-5.7 km/s lay•er �1ith a 4.8-5 .3km/s layer.The ampli tude of the i2 phase is stronge.rthan the r2 phase at the 31-35 kn1 offset (_ Figure 8a), but these two phases cannot be sepa1Aated at an ot't'set greater than 38 kn1.The travel times of th' e se two phases w•ere calculated using the retlected and re.f'racted paths in the model (Figure 8b ) .

Profile L85-1
The model for L85-l is characterized by an area l)f' d()Wnw•arping and crustal thickening structures (Figure 9b).The.dramatic feature ot' this 1nodel is in the thickness of� its low-\relocity sedimentary layer which ranges from about 3 km just south of the OBS-C (0 km offse.t) to about 4-5 km half\\1ay bet\\1 een the OBS-C and the southe1•n most shot (60 km offset) where it then progressi,1ely decreases tl) about 1.5 km unde1Aneath.The sediment structure was mainly determined from the first arri\1al s observed at an offset less than I ()-20 km (Figure 9a, i 1 ) .The thickness and depth ot' the 3.9-4.5,5.2-5.8 and 6.2-6.8 km/s la)1ers under the se.dimentary layer beneath the three shot pl)ints (()ffset 35-60 km) are mainly• similar to those in the eastern part ot' L9 3-1 ( Figure 3b ).

Profile L85-2
The model t'or L85-2 (Figure I Ob) is similar to that ot• prot'i le L93-7 (Figure 6t�) .Charac terized b)' a thin sedimentar)1 layer and high velocity sublayers under the shot points, it has a 4.8 km/s layer thickness of' ab()Ut 3.5 km., and a 5.4-6.0 km/s layer thickness ot' about 11 km.

TECTONIC INTERPRET A TIO NS
In this study, the model represents the simpl est structure to fit the data \v hereas the true structure of the region is probably somewhat more complicated.There are few �1el l-de.finedlate.rphases, meaning that the model is primarily based on the travel times of the first arrivals.The wide spacing of both the shots and stations does not allow• that the detailed surface geol og)' be taken into account i,i the fitting of tra\1el times.Nevertheless, the models re,real the.main features of the velocity structure in the northeastern Taiwan area.
.... .. 11 shows the \r elocity models constructed in this stud)' together with the a\,erage crustal structure of the continental arcs and the orogens (Christensen and Mooney, 1995) and typical oceanic arc (e.g. the Ky•ushu-Palau Ridge) (Ludwig et al., 1973).For the convenience of comparison be.tween the velocity columns, all the sedimentary ]ayers were removed in each velocity column in Figure 11 (where there \\las a sedimentary layer in the original velocity column, the base of the sedimentary lay•er w • as taken as zero depth ).Based on the results of this study, to the \\' est of the LV t'ault (Figure l la), all mode.lshave a common basement (4.8-5.4 km/s layer) and a subbasement (5.6-6.4 km/s lay•er).The thickness of' the.basement is about 9

Cl. Q.) Cl
• km near L93-l (24 °N), bL1t de.creases to about 3 km near L93-5 (Tailuko) and then increases to about 5 km at L93-7 (llan Plain) and to about 4 km at L85-2 (llan Ridge).In contrast� the thickness of the subbasement progressivel)' increases northwards and then thins nea14 the Ilan Plain and Ilan Ridge.The.eight models on the right in Figt1re 11 a can be divided into t\�o groups, such that Group A represents the.va1�iati()n in the velocity structure beneath the Tananao Schist (the EFCR), while Group B represents the crust l)f. the southern and south\vestern end of the Ryuk:y•u arc.The similarit)1 in the velocity• structures shown in Figure 11 a suggests that the EFCR bends eastward near 24.5 ( . ) N and should belong to the same tectonic unit as the south western Ryuk)1U arc (Figure 12).This east\\i•ard bending featt1re has alS(1 been revealed in seismic tomographic studies (e.g.� Chen, 19 95 � Ma, et c1 l., 1996) and in the geological study ot' pre.-Miocene (Kizaki, 1986). 2.4

6.2
---6.9 7.2 7.8      N and should belong to the.sa1ne tectonic unit as the south\\1estern R)iukyu arc (RA).The velocity structure of the Hsinchen Ridge (HR) and the Y aeyamcl Ridge (YR) could be the sa1ne t)-'pe ()f tectonic Lt nit as the Coastal Ra11ge (CR . ). OT represents the Okina\\1a Trough; NB is the CJ1eng et al.
In direct oppositic)n, when the resultant velocity strt1ctures ot• this stL1dy are compared \\1 ith th()Se from othe.r tec.to11ic units (Figure I 1 ) , the velc)city struc.ture of the CR \v hich repre sents the tectonic unit () f' the Luzon ai•c seems simila1• to the Ky•ushu-Palau Ridge in terms ot• its velocity f" e atu1•e s.In additio11, the velocity structure ()f. the EFCR is similar to the continental a1•c .Obv•iouslj1, the st1•ength ()f. the material <)f' the EFCR should be stronger and heavier than that ot' the.CR.This implies that the CR playred a weaker r<lle in its collision with the EFCR.Hence, the resultant velocity st1•ucture C)f. this stud)' suggests that the mechc1nism of' the colli sion bet\\1een the EFCR and CR exhibits drL1matic evidence C()ntradicting the notion of a typi cal arc-continental margin collision.Beside this, it alsl) suggests that our rest1lts are consistent with the arc-arc co1lision 1nodel (Hsu and Sibuet, 1995 ) in the northeaste1n Ta iwan area .
(3) The Ilan Ridge, located at the south\vestern end ()f� Ry1ukyt1 e: 1 rc, is the northeastern extensi()Il () f. the EFCR.This suggests that the.EFCR probably bends eastwards and belongs to the s�t111e te.ctonic unit as the southwestern Ryt1kyu arc.( 4) Based on the v• elocity structttre, the CR is lighter and \\leaker than the EFCR.This cont1•adicts with the arc-contine11tal CC)llision mode.I which suggests that the island of Tai\\1an is the re sult ()f. the collision of • the stronger Luzon arc against the \v eaker .L � sian continental 1nargin.

Fig. 2 .
Fig. 2. L()Cations (solid line.s) of the ve]()City profiles C()nsr1�ucted in this study.Two broken solid lines indicate the locations ()f.shot lines of the R/V Maurice Ewing in 1995.The heav•y S()lid li ne south C)t• Hual ien city• (S()]id square) indicates the location ot • the gravity and magnetic prof'i les taken from Hu and Chen ( 1986).CR represents the C<. )ast .a] Range; LY is the Longitudinal Val le)'.The bathymetric cont<)Ur interval is 50() 1neters.The solid triangles l)n the land side are the PANDA-2 stations.Solid circles are the CWBSN stati ons.Solid stars are the explosion sources.The in ve1•ted triangles on lines L85l and L. 85 -2 are the.OBS-C and OBS-TS shown in Figures 9 and I 0.
-----------, • .. -. .-. . . . . . ... .. .. .. -: 7 (a) T()p()graphic and (b) P-wave velocity structures of' prof"ile L93-l.The ray paths of' the i4 ef .lected (Pi P phase) and ret•racted (Pr phase) \:\laves used in this study are also shown.The ve1ocities at the t()p and bottom c)f.each layer are presented in kilometers per second.The m<: 1jor t�eature of the in odel is a prominent discontinuity, the Longitudinal Valle)1 Fault, in the upper crust withi n the Longitudinal Valley (:LV) .Areas of' g(1od \le locity control for the upper to lower crust (inside the dashed heaV)' line) \Vere derived from the examination of the ray-tracings fo14 each shot point and are indicated.TS is the Tananao Schist; CR is the Coastal Range; and HC is the Hualien Canyon.'v' ertical exagge14atio11 is taken as 2.6: 1 flJr (a) and 1.2:I for (b ).

Fig. 7 .
Fig. 7. (a) Observed seismograms recorded by• the TWD station and calculated travel time curves along profile.L95-l.Strong wavetrains appearing at offset distances of' I 0-l S km a14e interpreted as n1ultiples (MT).(b) Ray diagram and velocity structu14e.Vertical exaggeration is taken as 2.6: J for (b).

Fig. 11 .
Fig. 11.Comparison between the average crustal v-elocity structures of orogens, c.ontinental arcs and the Kyushu-Palau Ridge (the three on the left) and the velocity models t() the \\'est (a) and east (b) of the L V fault con structed in this study (: \\lithout wate-r and sedimentary laye-r).The shade.d parts sho\\' the .distribution of the \1elocit)1 structure ranging fro m about 5.7 km/s to 6.4 km/s.See text for explanation.

Fig. 12 .
Fig. 12. Scheme ()f the tectonic pr()Vinces (sh,1ded Z()nes) and boundaries based on the resttlts ()f the modeling in this stt1dy.The St)lid squares indicate the locati()ns ot• the velocity colum11s shc)wn in Figure 11.Note the east ern flank ()f the Cent14al Range (EFCR) h<-1s bent and e. xtends eastward at 24.S ()N and should belong to the.sa1ne tectonic unit as the south\\1estern R)iukyu arc (RA).The velocity structure of the Hsinchen Ridge (HR) and the Y aeyamcl Ridge (YR) could be the sa1ne t)-'pe ()f tectonic Lt nit as the Coastal Ra11ge (CR .).OT represents the Okina\\1a Trough; NB is the