Chapter 3

Surfacedeformation and tectonicsettingof Taiwan inferr ed fr om aGPSvelocity field.

3.1�

Intr oduction

The�

islandof Taiwan is a zoneof active continentaldeformationlocatedin anexcep-tional�

tectonicsettingwithin theplateboundaryzonebetweentheEurasianPlate(EUP)and� the Phillipine SeaPlate(PSP)(figure3.1). At Taiwan thePSPis moving towardthe�

EUPat a rateof 70-80mm/yr in thedirectionN306� E (Senoet al., 1993,insetoffigure�

3.1). Thecomplexity of Taiwan’s tectonicsettingarisesfrom thefactthatat theRyuk�

yu Trenchthe PSPsubductsnorthnorthwestward underneaththe EUP, whereasat� the Manila Trenchthe PSPoverridesthe EUP in a westward direction. Taiwan islocatedat the transferzonebetweensubductionandoverriding of the PSP. The 150km�

long, NNE-trendingLongitudinalValley Fault (LVF) on the islandof Taiwan isgenerally� consideredasthesuturezonebetweenthetwo plates(BarrierandAngelier,1986;Biq, 1972)accountingfor 25-30% of thetotalplateconvergence(Angelieretal.,2000).

The�

still ongoingcollision betweentheLuzonvolcanicarcandtheChineseconti-nental margin startedat least8Myr ago(Ho, 1988;Kao et al., 1998;Lallemandet al.,2001;Teng,1990)therebycreatingandbuilding theTaiwanorogen.Dueto theobliqueorientation� of thestrike of thearcrelative to thestrike of thepassive margin, thecolli-sion� at Taiwanhasmigratedsouthward,incorporatingever new portionsof theLuzonarc� (Lewis andHayes,1983;Suppe,1981). Detailsof thegeodynamicevolution andpresent-day tectonicsettingof Taiwan areby no meansresolved. This becomesevi-

Thischapteris in pressfor publicationin J�. Geoph.Res.as:A.G. Bos,W. SpakmanandM.C.J.Nyst,

Surfacedeformationandtectonicsettingof Taiwaninferredfrom aGPSvelocity field.

47

48 Chapter3

dent�

from thevarietyof contradictingmodelsproposedin theliterature(Angelieretal.,1990;Chemendaetal., 1997;Lu andMalavieille, 1994;Suppe,1981;Teng,1990;Wuet� al., 1997).Suppe,1981;Teng,1990;Wu etal., 1997].

These�

modelsfocuson two closelyrelatedprocesses:thegeometryanddynamicsof� thetransitionbetweenthetwo subductionzonesandtheevolutionof theTaiwanoro-gen.� Thehypothesisof the”thin-skinned”or ”critical taper”model(Barr andDahlen,1990;Dahlen,1990)for actively deformingfold-and-thrustbeltshasbeenchallengedby�

modelsincluding thebasementof thecrustin themountainbuilding process(Ell-w� ood et al., 1996;HwangandWang,1993;Wu et al., 1997). In the ”thin-skinned”models� activedeformationis confinedto a trapezoidalsegmentof awedgeoverlyingaplanar decollementfault,whereasthe”basementinvolved” modelsrequirethediscreteincorporationof autochthonousbasementmaterialinto theshallow partsof theorogen.A modelof crustalsubductionfollowed by exhumation(Chemendaet al., 2001;Lin,1998;Lin et al., 1998)hasalsobeenproposedto explain therapiduplift andhigh heatflo�

w in the CentralRangeof Taiwan (figure 3.1). Lallemandet al. (1997,2001)andTengetal. (2000)proposedetachmentof theEurasianslabbeneathTaiwanasamecha-nismto createspacefor thesubductionof thePSPslabalongthewestwardpropagatingRyuk�

yu trench. Analoguemodelinghasshown the possibility of subductionreversalalong� aboundarycomparableto theLVF of Taiwanin asimilarsettingof two contrarysubduction� zones,aswell asprovidedinsight in themechanismsbehindtheprocessesof� exhumationandlateralextrusion(Chemendaetal., 1997,2001;Lu andMalavieille,1994).Theseandothermodelsstill requirefurthertestingagainstavailablegeological,geophysical� and,in particular, geodeticdataacquiredin thepastdecade.

In 1989the”TaiwanGPSNetwork” wasestablishedby theInstituteof EarthSci-ences,� AcademicaSinica(IESAS).Basedon the 1990-1995dataof the network, Yuet� al. (1997)derivedthepresent-dayvelocity field of theTaiwanarea.Sucha velocityfield providessignificantinformationon the kinematicsof the crustaldeformation.Iapply� theinversionmethodof SpakmanandNyst(2002)to theGPSvelocityfield of Yuet� al. (1997)in orderto determinethekinematicpropertiesof thesurfacedeformation.This methodsolvessimultaneouslyfor the velocity gradientfield andfault slip rate.TheSpakmanandNyst (2002)methodutilizesaphysicalrelationbetweenthevelocitygradient� field, faultmotionandtherelativevelocitydatawhichis significantlydifferentfrom�

similarstudiesoncomplicatedregionswhicharebasedonspatialinterpolationofthe�

data(e.g.Beavan andHaines,2001;Wdowinski et al., 2001). The incorporationof� fault motion in a joint inversionwith the velocity gradientfield is a uniquechar-acteristic� of the method. In contrastto for instanceelasticdislocationmodeling(e.g.Bennett�

et al., 1996),my methodrequiresno assumptionson crustalor fault dynam-ics (seeSpakmanandNyst (2002)for furtherdiscussion).Fromthevelocity gradienttensor�

I easilyobtainthestrainrateandrotationratetensorfields. In conjunctionwithgeological� observationsin thearea,I interprettheobtainedsurfacedeformationmodelin termsof thekinematicsof crustalprocesses.By combiningthesurfacedeformation

3.1�

Introduction 49

Figure�

3.1: T�ectonicsettingof Taiwan showingmajor faults as usedin this study: LVF =

LongitudinalValley Fault; LF = LishanFault; CF = Chuchin Fault; CHF = Chaochou-ChishanFault; CKF = Chukou Fault; DF = DeformationFront, and themaingeological provinces: I= CoastalPlain; II = WesternFoothills (WF); IIIa = westernCentral Range (WCR); IIIb =eastern� Central Range(ECR);IV = CoastalRange (CoR);IJ = Ishigaki-Jima;IP = Ilan Plainand� COB= ContinentalOceanBoundary.

50�

Chapter3

modelwith seismicitydataandseismictomographyI am ableto proposea coherentmodelfor thepresent-daytectonicactivity at theislandof Taiwan.

3.2�

Geologicsetting

Geologically�

, Taiwancanbedividedinto four N-NE trendingprovinceswhicharesep-arated� by active faults(figure3.1). Theprovincescomposea west-vergentcollisionalprism involving both theChinesecontinentalmargin andtheLuzonArc. In theeast-ern� CentralRange(ECR) Pre-Tertiary high-grademetamorphicrocksof the Chinesemargin areexposed,while theCoastalRange(CoR)is composedof Neogeneandesiticv� olcanicunitsof thenorthernLuzonArc (Yu etal., 1997;Hu etal., 1996).

The generalstructuraltrendsof the Taiwan mountainbelt show an elongatedS-shape� (figure 3.1). SouthernTaiwan is dominatedby the on-shoreextensionof theManila�

accretionarywedge,representinga particularzoneof weaknessrelatedto thenorthern part of the Manila subductionsystem. To the north of this region, the ac-cretionary wedgeis terminatedby the ChishanTransferFault zone(CTFZ), whichappears� asa majorstructural,seismologicalandkinematicboundarytrendingN130� E(Lacombe!

et al., 2001). Another major wrenchfault zonecutting acrossTaiwan isthe�

Sanyi-PakuaTransferFaultZone(S-PTFZ).ThisN140� E" trendingleft-lateralfaultzoneis alsoaccompaniedby highseismicityandoffsetsseveralmajorstructures(Def-fontaines�

etal.,1997).TheprominentPeikangandKuanyin High representtheshallowpre-Cretaceous Chinesecontinentalbasement.They areaccompaniedby significantBougueranomalies(HsiehandHu, 1972)andarecharacterizedby tectonicstability.Thedeformationfront (DF) is locatedalongthewesternedgeof theaccretionarywedgein#

thesouthandprogressesnorththroughtheCoastalPlain(CP)alongtheeasternedgesof� thebasementhighs.Generally, all deformationrelatedto theconvergenceof thePSPand� EUPis consideredto beaccommodatedeastof theDF (figure3.1).

3.3�

GPSdata

I$

usethedatasetof GPSmotionvectorspublishedby Yu etal. (1997).Thesevectorsaredetermined�

for 131stationsof the”TaiwanGPSNetwork” thatweresurveyed4-6timesbetween�

1990and1995,four continuouslyrecordingstationsandfivesemi-permanentstations� (figure3.2). In thecalculationof thestationvelocities,Yu et al. (1997)chosethe�

Paishastation(S01R),situatedat therelatively stableChinesecontinentalmargin,as� thefixedreferencestationandfixedtheazimuthfrom Paishato Taipei (at 52.1� )% toresolve thetranslationalandrotationalambiguitiesof thewholenetwork in theestima-tion.�

I adoptthis referencestationfor plotting purposes,but notethat my analysisisindependentof referenceframe.

3.4�

TheSpakman-Nystmethod 51

Figure�

3.2: GPS&

velocityvectors of Taiwan [Yu et al., 1997] asusedin this studywith their95%'

confidenceerror ellipses.

Se(

veralearthquakeswith magnitude5.0or moreoccurredin or neartheareaof ob-serv� ationduringthetimeof observation.However, sincethey did nothaveasignificantef� fecton thelengthchangesof thebaselines,steadystatemotionis assumedduringtheperiod of observation(Yu etal., 1997).

3.4�

The Spakman-Nystmethod

I usetheinversionmethodof SpakmanandNyst (2002)to determinethesurfacedefor-mationof Taiwanin termsof discretemovementsalongfaults(creep/slip)andcontin-uous) deformationin crustalblocksfrom relative motion data. The kinematicmethoddoes�

not requireany assumptionsaboutthe dynamicscausingthe deformationor the

52�

Chapter3

rheologyof thecrust. For stationarydeformationthemethodrelatestheobservedrel-ati� ve motion( *,+.-0/ )% betweenstation1 and� station2 to

�thevelocity gradientfield ( 34+ )

%and� fault slip rate( 576 )% on fault segment 8 (figure

!3.3):

*,+9-0/;:<>=@?ACB ? D9EFHG 3I+KJMLON@PRQSLUT

<6 B ?WV 6X576YJ[Z

6-0/ N (3.1)!

where� \ A-0/ is theintegrationpathconnectingstations1 and� 2 . ] denotes�

thenumberofcrossed fault segments,V 6^:`_,a depends

�on the fault orientationwith respectto the

direction�

of integrationalongpath \ A-0/ , 576 is#

theslip rateon fault 8 at� the intersectionZ 6-0/ between� \ A-0/ and� thefault and 34+KJbLON is thevelocitygradienttensor.The�

first part of the right handside of equation3.1 denotesthe contribution ofcontinuous deformationto the relative motion betweenthe two stations. The inte-gration� over the 34+ is donein parts,sincethe 3I+ can be discontinuousacrossslip-ping/creeping faults.Thesymmetricpartof thevelocitygradienttensorconstitutesthestrain� ratefield, wherethe antisymmetricpart constitutesthe rotationratefield. Thesecond� part representsthemotiondueto fault slip on ] faultscrossedby a path \ A-0/ .Equation"

3.1 is purely linear in the unknown quantities3I+ and� 576 and� offers a com-plete descriptionof the relative crustalmotion *;cR-0/ resultingd from stationarycrustaldeformation.�

A setof e geodetic� relative motionobservationsyield at least efJgeihjaRNgkml v� ec-tor�

equations3.1coupledthrough 34+ and� 576 . I divide thefaultsinto segmentsandforeach� segmentthe relative fault slip rate 5 6 is parameterizedasa (segmentdependent)constant rate.Thestudyregion is subdividedinto a network of n nodesconnectedbytriangulation�

with therestrictionthattrianglescannotintersectwith faults.I adoptalin-ear� behavior of thevelocitygradient(quadraticdisplacement)onthespatialcoordinates

Figure3.3: Theforward problemof relatingrelativevelocityto thevelocitygradientfieldandfaulto

slip. Thecurvedpath p9qsr connectst two observationsites u and� v withw relativemotionxzy q0r .{

3.4�

TheSpakman-Nystmethod 53

in eachtriangle(SpakmanandNyst, 2002). This modelparameterizationis basicallythe�

only assumptionenteringthe observation equation(3.1) andis usedto arrive at alinear|

matrix systemof equations.Thedensityof thestationdistribution in themodelarea� is usedasa guidefor local densificationof thegrid. I includethesurfacetracesof� ”major”, active faults.The”minor” faultsareimplicitly representedby thevelocitygradient� field. I do not attemptto incorporateall known faults,sincethecurrentdatadensity�

is insufficient to independentlyresolvethevelocitygradientfield andslip/creepratesd onnumerous”minor” faultsin suchamoredetailedparameterization.

Substitution(

of theparameterizationin 3.1yieldsanordinarysetof coupledequa-tions,�

which canbeassembledin a matrix-vectorform (seeSpakmanandNyst (2002)for details). Sincethe relative motion between2 points is independentof the choiceof� \ A-0/ , the matrix systemcanbe extendedwith moredataequationsusing the sameobserv� ation set *Ucm-0/ . Closedintegrationpaths( \ A-0/ h \ A/b- )% betweenstations1 and� 2will� alwaysrenderzerorelative motion, therefore,the closedpathswhich only sam-ple continuouslydeformingcrustareeffectively replacedby the local constraintthat3~}�34+�:�� . Additional paths \ A-0/ may� be requiredacrossfaultsto ensureinternalconsistenc y betweenfaultslip rateandthevelocitygradientfield. Thiseffectively leadsto�

anextendedsetof dataequations.Dataerrors,non-uniquenessandill-conditioningof� the observation matrix causemy matrix-vectorsystemto be an inconsistentsetofequations.� To dealwith problemsassociatedwith this, I adoptan inversionschemewhich� selectsa solutionthatfits thedatain a leastsquaressenseandat thesametimeminimizessomemodel norm. Basedon the characteristicsof the model area(e.g.station� density)I adopta combinationof amplitudedampingon the boundaryof themodelingareaand,in themodelinterior, amplitudedampingcombinedwith spatiallyv� aryingfirst derivative regularization.Themodelminimizing thedataresidualandtheadopted� model norm (definedby the regularization)is given by Spakmanand Nyst(2002):!

� :�JM������� ?� ��T V������ T V

� ? ��� T V���� � ? � ? N � ? ������� ?���� (3.2)

!with� a posteriorimodelcovariancegiven by �i:�JM� � � � ?� ��T V

�� � � T V� ?���� T

V�� � � ?�� ? N � ? and� modelresolutionkernel ��:`� � � ��� ?� � . HereA representsthe

coef ficient matrix, �� is#

theobservedvelocity datavectorincluding the 3¡}¢34+£:¤�constraints, and� � denotes

�thedatacovariancematrix. Sincemy datavector �� consists

of� all efJge¥h�aRNgkml relative velocities,my datacovariancematrix � � w� ould obtaindimensions� efJge¦h§aRNgkml by

� e¨Jge¦h�aRN�kml . Inversionof this matrix would be veryinpractible,#

if not impossible. ThereforeI only utilize the datastandarddeviations,which� throughstandarderror propagationresultsin the standarddeviation vectorforthe�

datavector �� . ��� and� ��� are� identitymatricesfor theboundaryandgeneraldamping,respectid vely, and � ? stands� for the first derivative operator. V � , V ? and� V � are� theweighting� factorsof theamplitudedampingandfirst derivativeregularizationequations

54�

Chapter3

and� control the trade-off betweenfitting thedataandminimizing theweightedmodelnorm. Theinversiondependson four tuningparameters:theweight ( V9© )

%attributedto

the�

surfaceconstraint3~}�3I+f:`� , the weight V � for�

the boundarynodedamping,the�

weight V ? for�

the amplitudedampingin the interior of the modelandthe weight

V � attrib� utedto thefirst derivative operator. I investigatea rangeof solutionsobtainedwith� differentcombinationsof thetuningparameters.

For theanalysisof theinversionresultsI definethemodelnormalizedª � per degreeof� freedom:

ª � « : ac¬- B ?

­ � -® �- (3.3)!

where� n are� thenumberof dataequations,e the�

numberof modelparameters,c¢:n�h¯e the�

degreesof freedomof themodel, ­ -°:¤Q²± �7³- hjQ²´µ©·¶- is#

thepredictionerrorof� themodeland ® �- is the a ® data

�variance.For modelswith ahighdegreeof freedom

a� properdatafit will result in a ª � of� aroundone. A ª ��¸>¸ a indicatesthat I amunable) to fit thedatawith thecurrentmodel,whereasa ª �º¹>¹ a identifies

#anattempt

to�

modelthedataerror, andmay imply moremodeldetail thansupportedby thedatasignal.� The ª � determination

�is, however, completelydependenton thedataerror: An

o� verestimationof suchanerrormayresultin a ª � which� is significantlysmallerthan1 andvice versa.Further, I use »®m¼ : ?½ ½- B ?.¾ ¿ -0- as� an averageindicationof themodelerror.

3.5�

Inversions

TheSpakmanandNyst (2002)methodprovidesthepossibility to solve for threedif-ferent�

typesof modelsfor surfacedeformation. The first application(I) only solvesfor�

the fault slip ratevectors,henceassumingthecrustaldeformationfield to beonlycontrolled by motion/creepon unlocked,freely slippingfaults;this correspondsto as-suming� rigid motionof largecrustalblocks.Thesecondapplication(II) solvesfor thev� elocitygradienttensoronly andthusignoresfault contributionsto therelativemotionfield i.e. assumesthatall faultshave beenlockedat thesurfaceduringtheobservationperiod. Finally, thethird (III) is thejoint approachwhich solvesfor bothfault slip rateand� thevelocitygradientfield. I invert for all threemodeltypesandprovideananalysisof� theresultsfor comparison.

The�

densityof my datasetandmy aim to modeltheregionaldeformationfield re-strict� theestimationof thevelocity gradientandfault slip rateto relatively substantialareas� andfault segments,respectively. Around the LVF the stationdensityallows asignificantly� denserparameterizationcomparedto therestof themodelarea.I includethe�

surfacetracesof the six major geologicalfaults(Hu et al., 1996,1997;Yu et al.,1997)in my parameterization.Thelocationof thefault traceshavebeendigitizedfrom

3.5�

Inversions 55

Figure�

3.4: FÀ

inal parameterizationof the modelsof Taiwan. Thick lines indicatefault seg-mentation,Á black dotsare the triangle nodesandgrey dotsare thesitepositions.Notethat inmyÁ choicefor the triangle nodesI are not restrictedto the locationsof the observationsites.T�rianglesdo not intersectfaults. Nodesat thefault are doubledto allow thevelocitygradient

fieldÂ

to bediscontinuousacrossfaults.

YÃ

u et al. (1997) to insurethe station-to-fault relative positions. This approachleadsto�

a total of 290 triangles,spannedby 192 nodesand57 fault segmentswith 57 slipratevectors(figure3.4). Thoughfault slip rateparameterdampingis possible,inver-sion� I provedto beoverdeterminedwith acceptablemodelcovarianceandrequirednodamping.�

For inversionsII andIII I aim for asolutionwith minimal regularization,ac-ceptable covarianceandresolutionanda ª �IÄ aXÅ Æ . However, a larger ª � (2.89

!for my

preferred model)andslightly reducedspatialresolutionareacceptedto avoid spuriousv� aluestypicalof anunderdampedsolution.NotethatI implementthe a ® standard� devi-ations� in my inversionprocedure.Utilizing a Ç ® standard� deviationwouldresultin a ª �of� 0.48.Thefirst derivativeregularizationis allowedto varyspatiallydependingonthenumberof integrationpathscrossingeachtriangle(hitcountof thetriangle):Triangleswith� lesshits than aXkmaRl of� the maximumhitcountobtainan increasedregularizationweight� ( V �ÉÈ and� V ��Ê , respectively). Thefinal tuningparametersobtainedfor bothinver-sion� II andIII were: V9© :ËÇÌÅ ÆÍP·a�ÆRÎ (or

!astandarddeviationof ® © :�aXÅ ÏÐPÑa�Æ �ÓÒ JÕÔ�ÖOZÌN �

?)%

for thecurl constraintsand V � :�aRŠƺPYa�ÆR× for theboundarydamping,V ? :£ÏÌÅ ÆØPYa�ÆÚÙfor�

theamplitudedampingand, V � È :�ÛRÅ Æ4Pma�Æ? �

and� V � Ê :�lXÅ Æ4Pma�Æ?�?

for�

thespatiallydependent�

first derivative operatoron high hitcount (¸ ?? ��Ü ÔÞÝÉß )

%and low hitcount

56�

Chapter3

1 à@á à@â ] ª � »Z ¼ »®�ã¼ »®�ä¼? �SåRæç ©¼è¼ç ©

I - - 114 24.4 1.0 - 0.13é

II$

155 290

- 4.96ê

0.65é

1.25 -III 192 290 114 2.89 0.62

é1.46 0.22

éT�able3.1: Aspects

ëof theinversionparameterizationandaverageresultsfor inversionsI, II and

III.ì

Key: u ,í inversion; îÚï ,í numberof modelnodes;îÚð ,í numberof triangles; ñ ,í numberof faultseò gments; óôÉõ�ö ÷ø øqCù ÷Yú qsq ,í the average resolution,with ú q0q theû diagonal elementsof the

resolutionmatrixand ü theû

numberof modelparameters; óýmþõ ö ÷øUÿ øUÿq ù ÷ � � qsq ,í theaveragestandarò d deviation for the componentsof ��� ,í with ü þ ö�� î ï theû numberof componentsof��� ;� óýõ ö ÷ø�� ø��qCù ÷ � � q0q ,í the average standard deviation for the componentsof �� ,í withü ö�� ñ theû numberof slip components.

(! ¹ ?? � Ü ÔÞÝÉß )

%regions,respectively. The fault slip ratevectorsin inversionsI andIII

were� never subjectedto damping.Table3.1 providesanoverview of someaspectsofthe�

inversionparameterizationandaverageresultsfor inversionsI, II andIII.

3.5.1�

I: Inversion for fault slip rate only

In$

an attemptto fit the databy purefault motion, i.e. assumingrigid crustalblocks,I$

only considerthe 6 major geologicalfaultson Taiwan. In the model the faultsarerepresentedby 57slip ratevectors,eachconsistingof afaultnormalandafaultparallelcomponent, resultingin a total of 114modelparameters.The inverseproblemprovesto�

be over-determined(perfectresolution)but the solution,with small modelcovari-ance,� shows largedatamisfitswhichcannotbeaccommodatedby the95%confidenceerror� ellipses(ª � :�l�ÏÌÅ Ï ; figure3.5a).Usingamoredetailedfault parameterizationofallo� wing twice theamountof slip modelparametersperfault doesnot leadto a signif-icant#

reductionof thedatamisfit nor rendersa significantlydifferentslip ratesolution.A comparisonwith neotectonicobservationsof fault slip ratesalsofails the test. Forinstance,thesolutionfor slip rateon theLongitudinalValley fault comprisesa combi-nation of thrustandleft-lateralmotion,consistentwith geologicalobservations(figure3.5b).�

However, themagnitudeof theslip rateis significantlylargerthantheobservedslip� ratesdeterminedfrom geologicalandcreepmetermeasurements(Angelier et al.,1997;Lee et al., 2001). The Lishanfault accommodatessignificantthrustmovementon� its northernsegments,wheregeologicalevidenceshows left-lateralmovementinthe�

northwith increasingthrustmovementtowardsthesouth(Leeet al., 1997).Whilethe�

modeledleft-lateralmotion on the Chaochou-Chishan-Chuchinfault is supportedby�

geologicalobservations,noevidenceof thenormalcomponenthasbeenfound(La-combe et al., 2001).Therelative increasein thrustmovementon theChukou complieswith� geologicalobservations(Yu etal., 1997).

3.5�

Inversions 57

Figure3.5: Solving�

for fault slip rate only: a) Data misfits: Sincemy data are the relativevelocities� betweenevery pair of stations,I determinethe misfit at each stationby taking thedif�

ferencebetweenthe predictedvelocityand the data velocity for each path connectingthisstationò and determinethe average. I compare it with the average error ellipseof the station,whicw h is alsodeterminedfromtheerror ellipsesof all thesepaths.b) Fault slip ratesolution.

Since(

I have only parameterizedthesix majoractive faultsin theTaiwanregion, Iackno� wledgethepossibilityof animproveddatafit should”minor” faultsystemsbein-cluded. Here,I only testwhetherrigid motionof largecrustalblocks(implicitly definedby�

adoptingthemajorfault systems)canfit thedata.In inversionIII the”minor” faultsand� elasticloadingareimplicitly representedby thevelocitygradientfield. Onaccountof� the poor datafit andthe poor comparisonto neotectonicobservations,I concludethat�

thehypothesisof purerigid blockmotioncannotexplain theGPSdata.

3.5.2�

II: Inversion for continuousdeformation only

In this applicationI do not solve for fault motion contributions to the velocity field.The parameterizationconsistsof 290 triangles,spannedby 155 modelnodes. Eachnode generates4 componentsof thevelocitygradienttensor, which resultsin a totalof620�

modelparameters.Thedatamisfit (ª � : ÏÌÅ���� ; figure3.6b)is significantlybetter

58�

Chapter3

than�

obtainedfor modelI. Largestmisfitsarefoundin theLongitudinalValley region,where� many GPSstationsarecloseto eithersideof theLVF (figure3.6b).Fromfigure3.2�

I infer thattheGPSvelocitychangesstronglywhencrossingthefault asa resultoff�aultmotionnotmodeledin this inversion.

Most�

of the convergencebetweenthe EUP andthe PSPis absorbedin the Lon-gitudinal� Valley region. Extensionis found in the southernCentralRange(CR) andthe�

Ilan Plain (IP). Large contractioncan also be seenjust eastof the DeformationFront (figure 3.6a). Lee andAngelier (1993)determinedthe strainratesfor the cen-tral�

part of the Longitudinal Valley from trilaterationdataassumingthat the crustaldeformation�

shouldbe modeledby continuouscrustalflow. They show contractionratesat Juisui,Yuli andChihshangof (all expressedin ÖÓZ � ? )% lXÅ Ç }Þa�Æ �ÓÒ , ÇÌÅ aº}Þa�Æ �ÓÒand� lXÅ a } a�Æ �ÓÒ , respectively, with a consistentN132± E trend. I find strain ratesofaRÅ Ï��U_�ÆÌÅ ÆÚlØ}�a�Æ �ÓÒ in

#N140± E" at Juisui, ÆÌÅ����;_�ÆÌÅ ÆÚl }¢a�Æ �YÒ in

#N137± E" at Yuli andaRÅ ÇÚÛÚ_4ÆÌÅ ÆÚlÍ}Ia�Æ �ÓÒ in N133± E atChihshang,whichareconsistentlysmaller. Thiscould

Figure�

3.6: Solving�

for continuousdeformationonly: a) Data misfits,b) Strain ratefield. Thecontouringt denotestheeffectivestrain rate(= � �"! �$# q0r # qsr&% ÷(' ) % .{ Thearrowsdenotetheprincipalstrò ain rates: contraction (black) and extension(white). For a color version of this figure seefigurÂ

eC.7of appendixC.1.

3.5�

Inversions 59

be�

dueto the differencesin scalebetweentheir densetriangulationnetworks andmysparser� GPSnetwork andmy coarserparameterization(Kahleetal.,2000;Nyst,2001).YÃ

u andChen(1998)performeda studysimilar to thatof LeeandAngelier (1993)forthe�

southernpartof theChukou fault. They usedthedataof 5 yearsof GPScampaigns(1993-1997)!

to studyspatialandtemporalvariationsin crustalstrain.Their calculatedcontraction ratesof ÆÌÅ Ï��z} a�Æ �ÓÒ ÖÓZ � ? in N119± E at thecentralpartof theChukoufaultincrease#

to lXÅ Æ9aØ}Ëa�Æ �ÓÒ ÖÓZ � ? in#

N109± E" further to thesouth. HereI find an increasefrom� ÆÌÅ Ï�* _ ÆÌÅ ÆÚl } a�Æ �ÓÒ ÖÓZ � ? in

#N119± E" to aXÅ Æ�� _ ÆÌÅ ÆÚl } a�Æ �ÓÒ ÖÓZ � ? in

#N96± E.

"They

also� determinedextensionratesin thesouthof ÆÌÅ�*4} a�Æ �ÓÒ ÖÓZ � ? in N18± E. This agreesreasonablywell with my extensionratesof ÆÌÅ Û�ÆØ_¢ÆÌÅ ÆÚlØ}¢a�Æ �ÓÒ ÖÓZ � ? in N27± E in thesame� area.

Though�

this applicationclearly offers a betterdatafit thanthe inversionfor faultslip� rateonly (I), it remainsinconsistentwith neotectonicobservations. Further, thepresence of major, active faultsin this region is undisputed.

3.5.3�

III: Inversion for fault slip rate and continuousdeformation

The�

joint inversionutilizesthesametriangulargrid asinversionII. On thefault tracesduplicate�

modelnodesareusedto allow decouplingacrossthefault. Thejoint inversionis thusbasedon thesame290trianglesspannedby 192modelnodesandanadditional57�

slip ratevectors,yielding a total of 882 (768 for 34+ and� 114 for 5M6 )% modelpa-rameters.d Someof the datamisfits for this joint inversionareslightly larger thanthea� verage95% confidenceellipsesleadingto an averagedatamisfit ª � of� 2.89 (figure3.8a).�

I observe thatthefew misfit vectorsthatexceedthelimits of thedataconfidenceellipses� arestill primarily locatedin the LongitudinalValley region, wheredeforma-tion�

is strongest.Comparisonwith the data(figure 3.2) shows that the misfit vectorsdenote�

local differenceswithin the trendof thedata.Our adoptedparameterizationistoo�

coarseto modelthesestronglocal variationsin relativemotion.Beforeanalyzingthe fault slip rate(figure 3.8b) andthe strain-androtationrate

fields�

(figure3.9a,b)I will discussthecovarianceandresolutionof thesolution.

Co+

varianceand resolution

Fromthemodelcovariancematrix ¿ I extract thestandarddeviationsin theestimatesof� my model parameters( ® -�: ¾ ¿ -0- ).% For the velocity gradientfield I plot the ® -corresponding to a95%confidencelevel astwo contourplots:onefor the , -derivatives(longitude)!

(figure3.7a)andonefor the - -derivatives(latitude)(figure3.7b).Thelargererrors� (

Ä Û } a�Æ � ׺ÖÓZ � ? )% occur along the southwesternextent of the DeformationFront andin thenorthnorthwesternregion of themodelarea. In regionswith a smallamplitude� (

Ä a�} a�Æ � × ÖÓZ � ? )% the solutionbecomesinsignificant. The errorsof thefault slip rate parameters,plotted as 95% confidenceerror vectorsin the - - and , -

60�

Chapter3

Figure3.7: Co.

varianceandResolutionof thejoint inversionsolution.Thecovarianceis rep-r/ esentedasthemodelstandard deviations( ý õ ö � � qsq ).0 A) contourplot of the95%confidenceý q on1 � y32 ,í B) contourplot of the95%confidenceý q on1 � y34 ,í C) 95%confidenceý q on1 (5 plot-6tedû aserror vectors in the 7 -directionand 8 -direction,D) contourplot of theresolutionof � y 2 ,íE) contourplot of theresolutionof � y 4 .{

3.5�

Inversions 61

direction�

(figure3.7c),area factor10 smallerthantheaveragemagnitudeof fault sliprateobtainedin thesolution.

Figure�

3.7dand3.7eshow contourplotsfor thediagonalelementsof theresolutionk�ernelof thevelocity gradientfield. The interior of thestudyregion is relatively well

resolvedwith diagonalelementsexceeding0.75.Thedifferencebetweenresolutioninlatitudeandlongitudecanbeunderstoodby realizingthatdueto thebetterNS spreadof� observations(comparedto EW) latitudinal variationsarebetterresolved. Reducedresolutiond attheboundaryof my modelingareais dueto theinfluenceof theregulariza-tion.�

ThereducedresolutionoffshoreTaiwanin thenorthwesternpartof my modelingarea� canbe attributedto the poor datacoveragein this part of the model. In the in-terior�

of themodelthe resolutionis reduceddueto a trade-off betweenfault slip rateand� the velocity gradient. This canbe observed betweenthe DeformationFront andthe�

Chukou fault, around24± N9 betweentheChukou fault andtheChaochou-Chishanf�ault, andbetweentheChaochou-Chishanfault andsouthernLishanfault. I notethat

the�

fault slip ratethusmodeledmay representa combinationof fault creepandinter-seismic� signal,dependingon thetrade-off betweenthevelocitygradientfield andfaultmotion. This trade-off occursin caseof lack of datato seperatefault motion fromthe�

velocity gradientfield in a kinematicinversion. (Nyst, 2001;SpakmanandNyst,2002).

Thetrade-off is bestresolvedin areaswith relativemotionobservationsclosetothe�

fault zones.In areaswherethis is not thecase,aninterseismicsignaldueto blockmotion may be introducedin the slip ratesolution,whereasthe fault may in fact belock|

ed.This trade-off alsoaffectstheamplitudeof thevelocitygradientestimatein theproximity of thefault. I notethata solutionfor only continuousdeformation(solutionII of this study)implicitly reflectspureelasticloadingof locked faults. I furthernotethat�

thetrade-off problemis nota featureof theSpakmanandNyst (2002)method,butis#

a problemfor any interpretationmethodof relative motiondata.I referto Spakmanand� Nyst (2002)for a furtherdiscussionon this topic.

F:

ault slip rate contribution

In the joint solutionI find significantlyreducedfault slip ratescomparedto inversionI,$

thoughthesenseof motionis reasonablyconsistentbetweenthetwo solutions.Thef�aultslip ratecontributionof thejoint solution(figure3.8b)showsrelatively smallhor-

izontalslip ratesof theorderofÄ l mm/yr on mostfault segments.Largerhorizontal

slip� ratesareobservedon thesouthernsegmentsof theChukou faultandon theLongi-tudinal�

Valley fault. Largevertical rateshave beendeterminedon Taiwan(Lin, 1998;YÃ

u andKuo, 2001),but arenot reproducedby my modelsinceI only considerhori-zontalvelocities/rates.Slip rateson theLongitudinalValley fault vary from about19mm/yr with anazimuthof N124± E at Taitungto 25 mm/yr in N136± E at Yuli and7.8mm/yr� in N151± E" atHualien.Yu etal. (1990)showedthattheLongitudinalValley faultis averyweaklylockedor almostfreelycreepingfault. At Tapo,Angelieretal. (1997)

62�

Chapter3

Figure�

3.8: J;oint solution: a) Data misfits,b) Fault slip ratesolution.For further explanation

seeò figure3.5

determined�

present-dayratesbasedon detailedsurveys of faultedconcretestructures.Their rateof 22 mm/yr in N143± E comparesreasonablywith my valueof 26.84_ 0.22

émm/yr� in N136± E.

"The differencemight be explainedby the scaledifferencesof the

v� ery localobservationsandmy moreregionalparameterization.TheLishanfault shows left-lateralmovementat thenorthernsegmentchangingto

thrust�

movementmoreto thesouth.This variationin movementis in agreementwithgeological� observationsdoneby Lee et al. (1997). Yu et al. (1997) found a signifi-cant increasein shorteningratestowardsthe southalongthe Chukou fault, which isin supportof the increaseof thrustmovementI observe in my model. I further findthat�

the DeformationFront actsasa right-lateralthrust,which is consistentwith theobserv� ationsof Lacombeetal. (2001).

Strain<

rate contribution

The parameterizationof faults in applicationIII allows the velocity gradientfield tobe�

discontinuousacrossactive faults. The implementationof the faultssignificantlyreducesthemagnitudesof thestrainandrotationratescomparedto applicationII, es-

3.5�

Inversions 63

pecially aroundthe LongitudinalValley fault andthe Chukou fault. The magnitudesobtained� for thestrainratefield still remainamongstthelargestestimateson Earthforinterseismic#

deformation.The�

strainratefield (figure3.9a)shows a generalbehavior of NW-SEcontractionin the LongitudinalValley of easternTaiwan. In southwesternTaiwan I observe ananticlockwise� rotationof theprincipalaxisdirectionto E-W contraction.ThePeikangHigh=

is foundto haverelatively smallstrainrates.Comparingthis trendwith thestressfield�

modelingresultsof Hu et al. (1996)I find a generalagreement.However, differ-ences� betweentheirresultsandmy solutionareobserved,especiallyaroundtheIP. Thismay� bedueto ignoringtheopeningof theOkinawaTrough(OT) in theirelastic-plasticmodel,� whichcouldhaveasignificanteffecton thestrainratesin northeasternTaiwan.Anothercausefor discrepanciescanbe that stressandstraindirectionsdo not neces-sarily� align in caseof elastic-plasticrheology, which hampersa comparisonbetweentheir�

stressdirectionsandmy straindirections.Themagnitudeof thestrainratefielddecreases�

towardsthenorth.

Figure 3.9: J;oint solution: a) Strain rate field, b) Rotationrate field in degrees/Myr. The

number> s refer to local averagesof relativelystrong rotation ratesand their � ý err� ors. For acolort versionof this figureseefigureC.8of appendixC.1.

64�

Chapter3

Eastof theLongitudinalValley fault relatively largeNW-SEcontractiondominatesthe�

deformation. In contrast,the centralandsouthernCR show a dominantNE-SWto�

E-W extension.Geologically, this extensionis inferredfrom normalfaulting in theorogen� (Crespiet al., 1996;Kosugaet al., 1988),thermalobservations(Crespiet al.,1990)andnormalfault typeearthquake swarms(Lin andTsai,1981;Rau,1996).TheW?

esternFoothills (WF) aresubjectedto a WNW-ESEcontractionaccompaniedby asignificantly� southward increasingNNE-SSWextension. This trend is in agreementwith� Pleistocenepaleostresspatternsin southwesternTaiwan (Lacombeet al., 1999;Rocheret al., 1996).In theIP I find sugnificantNW-SEextension,consistentwith theinferredextensionaldirection in the Okinawa Trough(Hermanet al., 1978;Kimura,1985;Sibuetetal., 1987).

Rotation@

rate contribution

The rotation rate tensoris given by the anti-symmetricpart of the velocity gradienttensor�

(figure3.9b).Strictly speaking,theobtainedrelativerotationratesapplyto rigidrotationsd of small equidimensionalblocks. Interpretationof regional (anti) clockwiserotationscanbemadeif on averagetheobtainedrotationratesare(negative) positive.Since(

I am not able to solve for a net uniform rotation of the entire study region acomparison with paleomagneticblock rotationsis not straightforward. This is furthercomplicated by thefactthatalmostall paleomagneticobservationsareolderthan8Ma(Lee,!

1993;Miki et al., 1993),henceoccurredprior to the last collisionalphase(Ho,1988;Kao et al., 1998;Teng,1990). I noticethat many large regionswith the samesense� of rotationrateareboundedby thefaultswhich is indicativeof faultmotion.

Model III (figure3.9b)showstwo regionswith largeanticlockwiserotationratesinthe�

CoR,where,just offshorethenorthernsectionof theCoR,equallylargeclockwiserotationsd areobtained.This offshoreareaandthe areaaroundTaitungare,however,poorly resolvedandcareshouldbetakenwheninterpretingtheserotationrates.

The WF aredivided in threeblockswith opposingrotations;NorthernTaiwan isdominated�

by anticlockwiserotations,thecentralpart experiencesclockwiserotationa� veraging10.1_ 0.8

é ±�k e ÖÓZ as� the southernWF aresubjectedto 12.5_ 0.7é ±�k e ÖÓZ of�

anticlockwise� rotationonaverage.TheCRshowsasimilardivision: Thenorthernsec-tion�

undergoesaveragely14.8_ 0.8é ± k e ÖÓZ of� clockwiserotation,themiddlesectionis

subjected� to anticlockwiserotationaveraging9.2_ 0.8é ±Sk e ÖOZ , andthesouthernsection

sho� ws reducedanticlockwiserotation in the westandclockwiserotation in the east.The intersectionsof theoppositelyrotatingblockscoincidewith the two major trans-fer fault zoneson Taiwan. Only thesouthernregion of theCoastalPlainundergoesaclockwise rotationof a few Q ­BA Z ­�­DC k�Ô�ÖÓZ . TheIlan Plainis rotatingclockwisewith arateof 10.3_ 0.7

é ± k e ÖÓZ on� average.

3.6�

Geometryof thesubductionsystem 65

Comparison+

to lithospheric stressdata

I$

comparetheprincipalaxesof my strainratefield with theprincipalstressaxesfromearthquak� e focal mechanisms.Since I only model the two-dimensionalhorizontalstrain� rate tensor, I compareit to the horizontalcomponentsof the seismicmomenttensor�

. Thesefour componentswereextractedfrom the Harvard momenttensorso-lution|

of eachearthquake within themodelingareawith magnitudelarger than4 anddepth�

lessthan50 km. In my comparisonI make a distinctionbetweenshallow events(! ¹

20km) andthe deeperevents. ThoughI am awarethat the deeperseismicitymaybe�

relatedto lithosphericdeformation(subductingslab),I includethedeepereventstoallo� w for comparisonwith eventsin thecrustalrootof theorogenunderneaththeCen-tral�

Range.Theprincipalaxesof seismicmomenttensorsdonotneedto coincidewiththose�

of thesurfacestrainratessince(i) theseismicmomenttensorsreflectlocalreleaseof� stress(which doesnot necessarilycoincidewith regionalstrainratedirections)and(ii)!

thesurfacestrainratesdonotnecessarilyreflectdeformationatdepth.In$

theWesternFoothills I find a reasonableagreementin thedirectionsof theprin-cipal axesfor all events. In all otherareasa consistentmisfit exists for both shallowand� deepevents. In the southernCentralRangethe E-W compressionat depthis incomplete contrastto the E-W extensionof my model. I believe that thesedisagree-ments� shouldbe attributedmostly to the fact that thosesurfacevelocitieswhich arethe�

responseto deepercrustalprocessesnot necessarilycontainall informationon thedeformation�

occurringin the deepercrust. Therefore,a combinationwith crustalde-formation�

data(e.g. seismicity)is usefulto studycrustalprocessesdriving thesurfacedeformation.�

3.6�

Geometryof the subductionsystem

OfE

basicimportancefor understandingthe determinedstrain-and rotation ratesandfault slip rateis thegeometryof thetwo subductionsystemsin themantle.Numericalmodeling� predictsthatsubductionof acontinent-oceanboundary(COB)leadsto favor-able� conditionsfor detachmentof theoceaniclithosphere(VandeZeddeandWortel,2001;Wong A Ton andWortel, 1997). Several recentstudieshave adoptedthe hy-pothesis of detachmentof theEUPslabundernorthernandcentralTaiwanin ordertoe� xplain the ongoingsubductionof the PSPslab(Lallemandet al., 1997,2001;Tenget� al., 2000).Analoguemodels(Chemendaet al., 2001)demonstratethepossiblecon-junctionF

of EUPslabdetachmentandincipientwestwardsubductionof thePSPunderTaiwan.

The questionremainswhetherslabdetachmentis actuallyoccurring. Lallemandet� al. (2001) basetheir interpretationof slab detachmenton the global tomographicmantle� modelof Bijwaardetal. (1998),whichshowspredominantlylow P-wavespeedsin the upper100 km of the EUP subduction.However, sincethe non-subductedpart

66�

Chapter3

Figure�

3.10: LayerG

cuts from the global tomographic mantlemodelof Bijwaard and Spak-manÁ [2000]. Theleft columndisplaystheP-wavevelocityanomaliesbetween95kmand380km(depthsH

indicated) in percentages of the following referencevelocities(from top to bottom)8.047,I

8.119, 8.275, 8.482, 8.701, and 8.913 km/s. White dots denoteintermediatedepthhypocenterJ

swithin 25kmof each layerdepth.Thegrid of black linesarelongitudeandlatitudelinesK

in stepsof 2 degrees(seeupperleft panelfor 2 values).Theright columndisplaysthere-sultsò of a sensitivitytestfor estimatesof spatialresolutionin which theimagerecoveryis testedof1 a syntheticvelocitymodelconsistingof isolatedblocks(circular outlines)of 1.2 degreesinsizeò anda thicknessof about50 km. Theamplitudesare of alternatingsignandtheblocksareshiftedò in depth.Thecolors denotetherecoveryof thesyntheticmodel.Mild smearingeffectsbetweenL

thesyntheticblocksarevisibleat all depths,but thesyntheticblocksaregenerally wellrecoverable, particularly in the upper200 km underTaiwan. Belowthis depthresolutionre-duces�

gradually. I concludefromthis(andothersensitivitytests)that thetomographicimageoftheû actualEarth (left column)is interpretableat lengthscalesof 50-60km.For a color versionof1 this figureseefigureC.9fo appendixC.1.

3.7�

Model interpretation 67

of� the Chinesecontinentallithosphereis imagedwith similar low wave speeds,it isdif�

ficult to discriminatebetweenadetachedslabandsubductedcontinentallithosphereparticularly in theupper100-150km of themantle. Below this depthrangetheEUPslab� is visible in the mantletomography. Tenget al. (2000) invoke slabdetachmentas� the driving force behindthe high uplift ratesand large heatflow observed in theeastern� CentralRange.Themaximumuplift ratesandlargestheatflows arecurrentlyobserv� edbetween23± N9 and23.5± N,

9whichwouldsuggestthatthetip of thesouthward

propagating detachmenttearhasadvancedsignificantlysouthof this location(Worteland� Spakman,1992,2000).However, bothseismicity(figure3.12;cross-sectionA) andseismic� tomography(figure3.10)contradictthis suggestionby showing a continuousslab� up to 23± N.

9The�

assumptionof slabdetachmentis requiredif the northward extensionof theEUP"

slabbeneathTaiwan intersectswith the PSPslab. Figure3.10shows cutsfromthe�

tomographicmodelof BijwaardandSpakman(2000),which differs from earlierw� ork (Bijwaardet al., 1998)asa resultof 3D ray tracing to correctfor seismicraybending�

effectscausedby the3D velocity structureof themantle.For theTaiwanre-gion� this leadsto animproveddefinitionof theslabgeometry. Therelatively fast(blue)anomalies� north of 24± N9 andeastof 122± E imagethe northnorthwestward plungingPSPM

subductionzonealongtheRyukyu trench.Similarly, theeastwarddippingEUPisvisible� southof 24± N9 andat longitudescenteredaround122± E.

"For depthsabove 145

km the EUP slabis bestdelineatedby intermediateseismicity. Below this depththeslab� reaches24± N9 atmost(thelatitudeof Hualien).I assumethatthePSPslabandtheEUP"

slabareat the point of makingcontactwith eachotherat Hualien. On accountof� theseimages,slabdetachmentof the EUP below north to centralTaiwan doesnotat� presentseema strongrequirementto createspacefor thePSPto subductin a north-westw� arddirection.Theimageassociatedwith theEUPslabdoesnotshow significantlateral|

displacementwith depth,which attestsof its steepdip. This couldbeattributedto�

resistanceof theChinesecontinentalmargin to engagein thewestwardroll-backofthe�

EUPslab(viewedin aEurasianfixedreferenceframe).

3.7�

Model interpretation

3.7.1�

Southern Taiwan: southward extrusion

In southernTaiwanmy surfacedeformationmodelexhibits strainratesof almostpureE-W"

contraction,whichis consistentwith ongoingcollisionalshorteningin theorogen.The�

contractionalratesareaccompaniedby southwardincreasing,predominantlyN-Soriented� extensionalstrainrates(figure3.9a).I find right-lateralthrustmotionof

Ä a�Ïmm/yr in a N54± W? directionon theDeformationFrontandleft-lateralnormalmotionof� 6-13 mm/yr in direction N110± -130± E" (increasingsouthward) on the Chaochou-ChishanN

fault (figure 3.8b). The block boundedby thesefaultsis subjectedto an an-

68�

Chapter3

Figure�

3.11:Sc�

hematicrepresentationof thetectonicinterpretationsmadein thisstudy:Grav-itationalO

collapse(thin black arrows)andopeningof theOkinawaTrough(dark grey arrows)inO

northernTaiwan,ongoingcollision with crustal failure, accretionandexhumationof crustinO

theCentral Range of thecentral block (light grey arrows),lateral extrusionof thesedimen-taryû cover to theManila accretionarywedge in southwestTaiwan(wideblack arrows),andthede�

velopmentof a tear fault with incipientnorthwestward subductionat thetransitionbetweentheû two trench systemsat Hualien.OT denotesOkinawaTrough.

ticlockwise�

rotationrateofÄ amlXÅ Û ± /Myr

P, wherethe areason eithersideof this block

are� experiencingclockwiserotationratesor significantlysmalleranticlockwiserotationrates(southernCentralRange;figure3.9b). Thesepatternsof surfacedeformationareindicati#

veof lateralextrusiontowardstheManilaaccretionarywedgesouthof Taiwan.T�o the north the extrusion is terminatedby the ChishanTransferFault Zone (figure

3.11).�

The analoguemodelsof Lu andMalavieille (1994)alsosuggesta southwardlateralextrusionof thesedimentarycoverof thesouthernWesternFoothills in responseto�

the obliquecollision betweenthe PSPand the EUP. The southernTaiwan areaisfurther�

characterizedby low shallow seismicitywith a significantincreaseat thebaseof� the crust(figure 3.12; cross-sectionA) andthe resultsof geodeticre-triangulationwithin� theorogen(Chen,1984)demonstratethatin thisareasubsidenceprevails. Theseobserv� ationsarein supportof my interpretationof my surfacedeformationmodel insouthern� Taiwan.

3.7�

Model interpretation 69

3.7.2�

Central Taiwan: oblique collision, shorteningand exhumation

In$

thecentralsectionof theWesternFoothills andwesternCentralRange(figure3.1)the�

modelsurfacestrainratefield is dominatedby WNW-ESEcontractionandclock-wise� ratesof

Ä a�ÆÌÅ a ± /MyrP

(figure 3.9aand3.9b). I obtain right-lateralmotion withan� increasingthrustcomponentfor theDeformationFront,left-lateralthrustmotiononthe�

Chukou fault, insignificantmotionon theChaochou-Chishanfault andminor left-lateralthrustmotionontheLishanfault (figure3.8b).Thesemodelingresultsrepresentthe�

surfaceresponseto theongoingobliquecollision causingshorteningandshearingin#

theorogen.However, in theeasternCentralRange(figure3.1) I obtaina dominante� xtensionalprincipalstrainrate. Theprincipaldirectionof theextensionrotatesfromNE-SW9

in thecentral-northerneasternCentralRangeto E-W in thesouth,in agreementwith� geologicalobservationsgatheredin thearea(Crespiet al., 1996;Lu et al., 1998).The�

extensionis accompaniedby ananticlockwiserotationrateofÄ �XÅ l ± /Myr

P. Recent

reportsof levelingmeasurementsacrosstheCentralRangedescribeuplift ratesashighas� 36-42mm/yr for thepastdecade(Liu, 1995a)especiallybetween23± N9 and23.5± N9(K!

osugaet al., 1988). A feasibleexplanationof my surfacedeformationpatterns,thehighuplift rates,andthehigh-grademetamorphiccorecomplexesobservedin theeast-ern� CentralRange(section3.2)couldbetheexhumationof crustalmaterial.Regionaltomography�

of RauandWu (1995,figure3.13)shows a thickeningof thecrustunderthe�

highelevationsof theCentralRange,wherethedeepestpartis offsetto theeast.Inthe�

norththelow velocityrootextendsdown to adepthof 35km (c ´ :Q�XÅ ÆÌ8SÔÞk C ).% Thisis in agreementwith a Moho depthof 33 km inferredfrom gravity Bougueranomalymodeling� (Yenet al., 1998).In bothtomographiccross-sectionsthevelocity in thetop15km undertheCentralRangeis relatively highcomparedto thevelocitiesunderneaththe�

WesternFoothills andtheCoastalRange.This coincideswith a zoneof relativelylittle seismicityfoundunderneaththeeasternCentralRange(figure3.12). Chemendaet� al. (2001),Lin (1998)andLin et al. (1998)proposea modelof crustalsubductionfollowedby exhumationin a orogen-normaldirectionof theeasternCentralRangetoaccount� for the observations. In this model the continentalcrustof the SouthChinaSea(

is subductedto mid-crustallevelswhereit becomesdetachedfrom themantlepartof� the lithosphereandexhumes.However, in this modeltheobliquity andsouthwardpropagation of the collision is not taken into account.Sincecollision commencedinthe�

north andis progressively moving southward, the subductingcontinentalcrust inthe�

north would reachcritical conditionsneededfor its detachmentfrom the mantlewell� beforeits southernequivalent. Therefore,it is reasonableto assumea southwardpropagation of the exhumationin conjunctionwith the collision (figure 3.11). Thissouthw� ard propagationof the exhumationmay causethe anticlockwiserotation ob-serv� ed in the easternCentralRangeas well as divert the principal directionsof thee� xtensionto amoreNE-SWorientationin theregionswith maximumexhumation.

70R

Chapter3

Figure�

3.12:LocalG

seismicityascatalogedbytheCentral WeatherBureauof Taiwan,betweenMayS

1900andOctober2000.Light grey dotsin cross-sections:0 T MS T 3,

UBlack dotsin cross-

sections:ò üWV�X .{ Eventscontainedin dottedcircle correspondto the incipient subductionoutlined1 in sectionC.

3.7�

Model interpretation 71

3.7.3�

Northern Taiwan: crustal extensionand orogeniccollapse

Most�

of northernTaiwan (north of the Sanyi-PakuaTransferfault zone)is subjectedto�

relatively smallanticlockwiserotationrates(0± -5± /Myr;P

figure3.9b)andleft-lateralfault slip rates(figure3.8b).Strongclockwiserotationratesof

Ä a�ÆÌÅ Æ ± /MyrP

arefoundin#

theIlan Plain. Thestrainratefield is characterizedby relatively smallNW-SEcon-traction�

correspondingto geologicalobservationsin thearea(Crespietal.,1996;Teng,1996).Basedon my modelandthegeologicalobservations,thenorthernextentof themountainbelt is no longersubjectedto active crustalshortening.However, theexten-sional� regimedueto theopeningof theOkinawaTrough(Liu, 1995b;Yehetal., 1989)inducescrustalstretchingandgravitationalcollapse.

The back-arcextensionin the trough is relatedto the southwestward migrationand� rotationof the Ryukyu trench(Liu, 1995b;Yeh et al., 1989),which commenced2Ma

ago (Lee et al., 1991; Lee, 1993; Miki et al., 1993). The significantNW-SEe� xtensionobtainedin my modelin theIlan Plainclearlyportraystheon-landextensionof� the Okinawa Trough(figure 3.11). Shallow seismicityin northernTaiwan is low,whereas� theIlanPlainis characterizedbyhigh,verylocalizedshallow seismicityandanincreasedheatflow, which would beexpectedin a back-arcbasin.Thus,theOkinawaTrough hasacquiredthe ability to propagateinto Taiwan along a pre-existing weakzone,Y theLishanfault. In thenortherneastern-CentralRangetheWNW-ESEextensionmay� bedueto a combinationof thegravitationalcollapseandthecrustalexhumation,as� identifiedin thecentralandsoutherneasternCentralRange(figure3.11).

Figure�

3.13: ReZ

gional tomographyadaptedfrom Rauand Wu [1995]. A well resolvedhighvelocity� featurecomparableto theonefoundfor thePSPat theRyukyuTrench canbeobservedunderneath[ theCoastalRange andEasternCentral Range. TheCentral Range is underlainbya� regionof highvelocity. For a color versionof this figureseefigureC.10of appendixC.1.

72R

Chapter3

3.7.4�

The CoastalRange: incipient northwestward subductionof the PSP

In$

theCoastalRange,themodeledcreepmotion on theLongitudinalValley fault de-creases significantlynorthwardfrom

Ä a�� mm/yr� with anazimuthof N124± E" atTaitungto� Ä *XÅ�� mm/yr in N151± E at Hualien(figure3.8b). My strainratefield shows strongNE-SW9

contractionaccompaniedby large anticlockwiserotationratesin the CoastalRange�

andcomparableclockwiserotationratesjustoffshorebetween23.5± N9 and24± N9(figure!

3.9aand 3.9b). Southof 23.5± N9 the NE-SW contractionis more moderate(though!

significant)and smoothand predominantlyanticlockwiserotation ratesareobtained.� Somecareshouldbetakenwheninterpretingtheseresultsdueto poorerres-olution� in this areaespeciallyin the , -derivatives. However, basedon thedecreaseinmodeledfault motionsignificantcontractionalstrainratesor a significantdecreaseinPSPmotionareto beexpected.Sincethereareno indicationsto suspecta changeinPSPM

velocity at 23.5± N,9

40-50mm/yr of PSPmotionhasto beaccommodatedeastofthe�

CoastalRange.The focal mechanismsof the earthquakesin this region denoteafault with an averagestrike of 23± N9 andan averagedip of 75± locatedjust offshorethe�

islandof Taiwan(figure3.14). At Hualien(23.9± N),9

I plottedtheseismicityalonga� profile striking perpendicularto this fault (figure3.12;cross-sectionD). Theprofilesho� ws deep(up to 50 km), localizedseismicityaroundthefault andindicatesa north-westw� arddippingWadati-Benioff zone.Seismically, theslabextendsto approximately75R

km. Thefault locationis alsotiedwith asignificantgravity low observedin HualienCanN

yon (Yen et al., 1998). Further, regional tomography(RauandWu, 1995,figure3.13)�

shows awell definedhighvelocity zone(c ´¸ �XÅ ÆÌ8SÔÞk C )% underneaththeCoastal

RangeandCentralRange,which comparesin sizeto the Wadati-Benioff zoneof thePSPM

at theRyukyu trench.BasedonanaloguemodelingChemendaet al. (1997,2001)propose the possibleoccurrenceof incipient subductionof the PSPat the latitudeofHualien. Suchincipientsubductionwould accommodatemostof theconvergencebe-tween�

the PSPandEurasia. The hypothesisexplainsmy surfacedeformationmodeland� is in fact confirmedby the seismicitydataand the regional tomographicmodel(figure!

3.11). Thefault indicatedby thefocal mechanismsof eventsin theareacouldrepresentthetop of thePSPslabdevelopingnorthwestwardsubductionbelow easternT�aiwan.

3.8�

Geodynamicevolution

T�o provide anunderstandingfor thepresent-daytectonicsettingof Taiwan I consider

the�

geodynamicdevelopmentof the region. Eastof present-dayTaiwan, the subduc-tion�

of thePSPbelow theChinesecontinentalmargin in thepastmillion yearshasbeencharacterized by tearingof thecontinental-oceanboundary(COB)in roughlyWSWdi-rection.d Thetearconsistentlyfollowedthegeometryof theCOB,therebyprogressivelycreating thewestRyukyu Trench(Lallemandetal.,2001).Thiswestwardmigrationof

3.8�

Geodynamicevolution 73

Figure3.14: Focal mechanismsfrom the Harvard CMT catalog (M V 4, depth\ 50km).]

TheblacL

k mechanismsall denotea fault with a strikeof 23̂ N_

anddip of 73̂ withw thehorizontal.Thegr` ey vectors denotetheGPSvelocityvectors of Yu et al. [1997]. Thewhitevectorrepresentstheû PSPmotion. Barbedlines representthe Ryukyu trench where the segmentcentral in thefigurÂ

edenotesmyproposedsouthward continuation.

incipientPSPsubductionis causedby theobliquenessof thePSPmotionrelativeto thestrik� eof theEurasianplate.Trenchcreationthusadvancedwith anaveragespeedcom-parable to thecomponentof PSPmotionalongthestrike of theCOB. Thepositionofthe�

teartip coincideswith thenorthernextentof the(moreor lessN-Sstriking)Manilatrench,�

which accommodatesEUP subductionup to this latitude (Lallemandet al.,2001).

In the past3-5 My the strike of the Ryukyu trenchhasalignedwith the rela-ti�ve motionof thePSP. Whenviewedin a Eurasianfixedreferenceframe,theManilla

trench�

hasbeenrolling backto thewestwhile thecomponentof thePSPmotionalongthe�

strikeof theCOB(or normalto theManila trench)providestheoverridingmotion.Asa

longasroll-backspeedandoverridingmotionareequalthereis nospecialreasonforstrong� orogenicactivity. TheTaiwanorogenitself mustthenresultfrom achangein thedif�

ferencebetweenroll-backspeedof thetrenchandthewestwardcomponentof PSPadv� ance.I believe that theTaiwanorogenis a direct resultof a strongsouthwardturnof� thestrikeof theChinesecontinentalmargin, whichexistedeastof Taiwan.Ongoingroll-backof theManila trenchnow bringstheCOB in thepositionof beingoverriddenby�

the PSP(Chai, 1972;Suppe,1981,1984;Tsai, 1986). Resistanceof the Chinesecontinental lithosphereto subductionhasdecreasedthe roll-back motion causingthePSPto indenttheChinesecontinentalmargin thuscreatinganaccretionarywedge,the

74R

Chapter3

Taiwan orogen. The decreasein slabroll-back alsoled to a steepeningof the dip ofthe�

subductingEUP, asobservedin theglobaltomography(figure3.10).Subsequentlypart of thecontinentalcrustsubductedto mid-crustallevelsbeforebuoyancy forcesini-tiated�

failureandthickeningof thesubductingcrustalongtwo conjugatethrustfaultsresultingin theexhumation(squeezingout) of a crustalslice(Chemendaet al., 2001)as� I have deducedfor the easternCentralRange.Further, the Ryukyu trenchhasnotpropagated into the Chinesecontinentalmargin (henceis not tearingthe continentase� xemplifiedby the”T1-tear” of Lallemandet al. (2001)).Instead,I proposethesouth-w� ardpropagation(moreor lessalongtheeastcoastof Taiwan)of theRyukyu trench,thus�

overcomingthejog in theCOB geometrybelow present-dayTaiwan. In conjunc-tion�

with this southwardbendingof the trenchincipientnorthwestwardsubductionofthe�

PSPunderTaiwanis now in progress.

3.9�

Summary and conclusions

Themethodof SpakmanandNyst (2002)hasbeenappliedto derive thesurfacedefor-mationfrom GPSmotion vectors,whereI exploreddifferentmodelingoptions. Thef�ault-slip-rate-onlysolutionof inversionI resultsin large datamisfits andthe agree-

ment� with neotectonicfaultobservationsis poor. Thesolutionin termsof 34+ generates�smaller� datamisfits. This modelgenerallyagreeswith othercontinuous-deformation-only� studies. However, the presenceof large, active faultson Taiwan cannot be ig-nored, thusthekinematicsof Taiwanshouldbestudiedin termsof fault slip rateand3I+ .

In$

versionIII providesanacceptabledatamisfit, covarianceandspatialresolution.In this model the methodof SpakmanandNyst (2002)exposesa trade-off problembetween�

fault slip rate and the velocity gradientfield that can only be resolved byplacing both stationscloseto the faultsandstationsin the interior of crustalblocks.ThemodelshowssignificantmotionontheLongitudinalValley fault,southernChukoufault andthesouthernDeformationFront. Strainratesrotatefrom NW-SEcontractionin#

the CoastalRangeto E-W contractionin the southernWesternFoothills andNW-SE(

extensionin northernTaiwan. The rotation rate field shows several blocks withconsistent (anti) clockwiserotation.

The interpretationof my surfacedeformationmodel,combinedwith the seismic-ity data,gravity dataandtomography, leadsto a coherentmodel for the present-daytectonic�

activity of Taiwan(figure3.11). I divide Taiwanin 4 distinctdomains:south-ern� Taiwan,centralTaiwan,northernTaiwanandtheCoastalRange.Thedomainsarebounded�

by thetwo majortransferfault zonesandtheLongitudinalValley fault. I de-duce�

activeconvergencebetweentheEUPandthePSPin bothcentralandsouthernTai-w� an. In southernTaiwanthealmostE-W collisionhasresultedin N-S lateralextrusionof� theweaksedimentarycovertowardstheManilaaccretionarywedge.To thenorththe

3.9�

Summaryandconclusions 75

e� xtrusionis terminatedby theChishanTransferFault zone.In my deformationmodelthe�

easternCentralRangeof centralTaiwanshowsalmostorogen-perpendicularexten-sion,� while thewesternCentralRangeandtheWesternFoothills show predominantlyNW9

-SEcollisioninducedcontraction.In conjunctionwith Chemendaetal. (2001),Lin(1998)!

andLin et al. (1998)I relatethis featureto active exhumationof a crustalslice.However, I believe thattheexhumationis propagatingsouthwardalongwith thecolli-sion,� causingtheprincipaldirectionsof theextensionto rotateto amoreNE-SWorien-tation.�

NorthernTaiwanis transferringinto astateof gravitationalcollapseinducedbythe�

inlandpropagationof theopeningOkinawa Trough(Liu, 1995b;Yehet al., 1989)along� apre-existingweakzone,theLishanfault. Thoughmy modelshowsthatsouthof23.5 ± N9 averysignificantpartof theconvergencebetweenthePSPandtheSouthChinaSea(

is accommodatedby slip ontheLongitudinalValley fault,whereasnorthof 23.5± N9about� 40-50mm/yr needsto beaccommodatedeastof theislandof Taiwan. I observea� clearnorthwestwarddippingWadati-Benioff zonein boththeseismicitydataandthered gional tomographicmodelof RauandWu (1995). I deducethat in thetransferzonebetween�

the two contrarysubductionzonesa southwardpropagatingcrustaltearfaulthasdevelopedeastof Taiwan. Thetearfault is thecrustalresponseto incipientnorth-westw� ard subductionof the PSPbelow easternTaiwan. Thus, the Ryukyu trenchisbending�

southwardbecomingalmostperpendicularto theconvergencedirection,whilesubduction� of the PSPcontinues.Slab-slabinteractionbetweenthe PSPslabandtheEurasianslabmayoccuranddetachmentof theEurasianslabmaycommence.In thissetting� asuddenrapidsouthwardpropagationof incipientsubductionis conceivable.

76R

Chapter3