Fault geometry, rupture dynamics and ground motion from potential earthquakes on the North

17 Pages·2012·4.29 MB·English

Geophysical Journal International
Geophys.J.Int. (2012)188,1071–1087 doi:10.1111/j.1365-246X.2011.05289.x
Fault geometry, rupture dynamics and ground motion from potential
earthquakes on the North Anatolian Fault under the Sea of Marmara
David D. Oglesby1 and P. Martin Mai2
1DepartmentofEarthSciences,UniversityofCalifornia,Riverside,CA92521,USA.E-mail:[email protected]
2DivisionofPhysicalSciences&Engineering,KingAbdullahUniversityofScienceandTechnology,Thuwal23955–6900,SaudiArabia
Accepted2011November3.Received2011August29;inoriginalform2011February22
D
o
w
n
SUMMARY lo
a
Using the 3-D finite-element method, we develop dynamic spontaneous rupture models of d
e
d
earthquakesontheNorthAnatolianFaultsystemintheSeaofMarmara,Turkey,considering fro
the geometrical complexity of the fault system in this region. We find that the earthquake m
h
size, rupture propagation pattern and ground motion all strongly depend on the interplay ttp
betweentheinitial(static)regionalpre-stressfieldandthedynamicstressfieldradiatedbythe s://a
propagating rupture. By testing several nucleation locations, we observe that those far from c
a
anobliquenormalfaultstepoversegment(nearIstanbul)leadtolargethrough-goingrupture de
m
ontheentirefaultsystem,whereasnucleationlocationsclosertothestepoversegmenttendto ic
producerupturesthatdieoutinthestepover.However,thispatterncanchangedrasticallywith .ou
onlya10◦rotationoftheregionalstressfield.Oursimulationsalsorevealthatwhiledynamic p.c
o
unclampingnearfaultbendscanproduceanewmodeofsupershearrupturepropagation,this m
/g
unclampinghasamuchsmallereffectonthespeedofthepeakinslipvelocityalongthefault. ji/a
Finally, we find that the complex fault geometry leads to a very complex and asymmetric rtic
patternofnear-faultgroundmotion,includinggreatlyamplifiedgroundmotionontheinsides le-a
offaultbends.Theground-motionpatterncanchangesignificantlywithdifferenthypocentres, bs
evenbeyondthetypicaleffectsofdirectivity.Theresultsofthisstudymayhaveimplicationsfor tra
c
seismichazardinthisregion,forthedynamicsandgroundmotionofgeometricallycomplex t/1
8
faults,andfortheinterpretationofkinematicinverserupturemodels. gy8/3
o/1
Key words: Earthquake dynamics; Earthquake ground motions; Earthquake interaction, ol07
m1
forecasting,andprediction;Computationalseismology;Continentalmargins:transform;Dy- /6
s8
namicsandmechanicsoffaulting. Sei473
0
JI b
Gy g
u
e
s
1 INTRODUCTION wthieth∼in15th0e0MkmarlmonargaNSAeaFitshtahtehraesfonroettrhuepotunrleydsiingnthifiecpaansttp2o3r0tioyenaorsf t on 0
TheNorthAnatolianFault(NAF),runningacrossnorthernTurkey (Reilingeretal.2000),andalargeearthquakeislikelytooccuron 7 A
hinazaanrdeianstt–hwisesrtegdiiornec,taiosnmfaonrif∼es1te5d00inktmhe, sdeoqmuiennacteesofthmeosdeeisrmateic- thiUspsianrgtopfrothbeabNiAlisFtic(Pmarestohnosdes,tathl.e2s0e0is0m;Picarhsaoznasrd20in04I)s.tanbulhas pril 2
0
to-large earthquakes since 1939 (12 earthquakes with M > 6.7; beenestimatedinrecentstudiesafterthe1999earthquakes(Atakan 19
e.g. Ambraseys 1970; Stein et al. 1997). The most recent events, et al. 2002; Erdik et al. 2004). Specific ground-motion scenarios
the M 7.4 Izmit and the M 7.2 Du¨zce earthquakes, ruptured the basedonkinematicmodelshavealsobeengenerated(Pulidoetal.
westernportionoftheNAFin1999asitenterstheBayofIzmit,the 2004; Sørensen et al. 2007). However, the relevant question for
easternmostpartoftheMarmaraSea.Thewesternterminationofthe seismichazardassessmentsisnotwhetherornotanearthquakewill
Izmitearthquakeislocatedonlyabout40kmsoutheastofIstanbul occur,butwhereandhowlargethenextdamagingearthquakewill
(Fig. 1). Towards the western end of the Marmara Sea, the 1912 be.Hubert-Ferrarietal.(2000)suggesttheoccurrenceoftwoevents
Sarosearthquake(M 7.3)representsthemostrecentmajorevent, ofequalorlargermagnitudethanthe1999Izmitearthquakewithin
withitseasternextensionbeingstilluncertain(Armijoetal.2002; the next few decades. In contrast, Le Pichon et al. (1999) argue
see Fig. 1). Historical earthquakes are reported by Ambraseys & for the possibility of single-event rupture along the entire length
Jackson(2000)tohaveoccurredin1509and1776,withmagnitudes of the submarine Marmara Sea fault system. In this context, the
intherange7.2–7.4,mostlikelylocatedonthenorthernbranchof detailedfaultgeometryoftheNAFintheMarmaraSeaneedstobe
theNAFasittraversestheMarmaraSeaandbetweentheendpoints considered.TheNAFsplitsintoseveralfaultbranchesclosetothe
ofthe1912Sarosandthe1999Izmitearthquakes.Thefaultsystem Du¨zcerupturezone(Armijoetal.2002):thesouthern,central,and
(cid:4)C 2012TheAuthors 1071
GeophysicalJournalInternational(cid:4)C 2012RAS
1072 D.D.OglesbyandP.M.Mai
D
o
w
n
Figure1. MapviewoftheNorthAnatolianFault(NAF)intheMarmaraSea,withpiecewiseplanarfaultgeometryoverlain.Arrowsinupperleftdenotethe lo
orientationoftheprimaryregionalstressfield(afterArmijoetal.2005). ad
e
d
northernNAF.IntheMarmaraSeaitself,threegeneralfaultconfig- Table1. Geometricalandstressparameters. from
u1Ar9ra9mt9ioi;jnoOsekotafaycl.eu2trr0ae0ln.2t2)ly.0T0a0hct;eisvIememrfeaonudletestlshafale.vae2t0ub0ree1e;neiLtphereoPrpiaocshseoindng(lePeta(arcklo.en2et0itn0au1l-;. Segment Le(nkgmth)L (Sdtergikreeeϕs) NoσrNma(MlSPtrae)ss ShσeSar(MStPreas)s StrengthS https://a
ous)strike-slipfaultsystem,oramorecomplicatedgeometrywith A 6.5 96.6 13.03 4.596 0.979 ca
B 13.0 79.4 15.971 4.905 1.414 de
accotnivtreolpsutlhl-eapaabritlibtyasifnosr.rTuhpetudreetatoilepdroMpaargmataeraalSoenagfamuultltgipeloemfeaturlyt CD 199..10 18085..00 1114..748767 34..893703 11..202524 mic.o
segments(Wesnousky1988;Harris&Day1999;Muller&Aydin E 39.8 77.8 16.243 4.843 1.523 up
2004),tothusgeneratelargethrough-goingruptures,andtherefore F 2.4 114 10.760 2.650 2.419 .co
m
criticallyaffectstheseismichazardofthecityofIstanbul. G 36.1 88.6 14.373 4.961 1.040 /g
ForrealisticdynamicrupturesimulationsontheNAF,theprop- H 33.6 119.0 10.906 2.306 3.487 ji/a
ertiesoftheregionalstressfieldneedtobeconsideredsothatone I 27.5 88.6 14.373 4.961 1.040 rtic
mayresolvethestresscomponentsontothedifferentfaultsegments. le
-a
Pinaretal.(2003)analysedmomenttensorsofsmall-to-moderate posed by Armijo et al. (2002) that best reproduces the observed bs
sizeearthquakesintheMarmaraSeatodiscriminatebetweendif- morphologyandverticalstructuraloffsetsofakeyhorizonwithin tra
c
ferentseismotectonicmodelsofthearea,aswellastoestimatea theMarmaraSea(Muller&Aydin2005).Tosimplifythespatially t/1
regionalstresstensor.Thedistributionoffault-planesolutionssug- varyingregionalstressfield,weadoptauniformregionalstressfield 88
geststhatthestressfieldintheeasternpartoftheSeaofMarmara (σ1 at130◦,σ3 at220◦,σ2 vertical)thatisresolvedintoallfault /3/1
isrelativelyhomogenouscomparedtothewesternpart.Pinaretal. segments (Fig. 1, Table 1). To account for the spatial changes in 07
1
(2003) identify several shear zones whose locations and sense of theregionalstressesandrespectiveuncertainties,weexaminethe /6
8
motionareexplainedbyasimpledeformationmodelthatrequires effectswhenrotatingthehorizontalprincipalstressesby10◦clock- 47
amajorE–Wstrikingright-lateralstrike-slipfault(theNAF).Their wise and counter-clockwise. From our suite of dynamic rupture 30
inversion of the orientation of P- andT -axes of fault-plane solu- calculations,wetheninvestigatetheconditionsforandproperties by
tionsresultsinaregionalstressfieldwithmaximumcompression, of dynamic ruptures on this geometrically complex fault system, gu
Nσ1E,–iSnWthedNireWct–ioSnE.Hdiorewcetivoenr,ainndthmeiwniemstuemrncpoamrtporfestshieonS,eσa3o,fiMntahre- tphoesieralikseelviehroeotdhrteoatgetonetrhaeteciltayrgoeftIhstraonubguhl-,gaonindgtheearrtehsquultaiknegslothwa-t est on
mara,amoreheterogeneousstressfield,asindicatedbyfault-plane frequencyground-motionpatterns. 07
solutions, may result from the change in strike of the NAF from ThefaultconfigurationoftheNAFwithintheMarmaraSeaalso Ap
nearlyE–WtoWSW.Thecorrespondingdeformationmodelofan isinterestingfromamoretheoreticalpointofview,inthatitisan ril 2
ENE–WSWstrikingright-lateralstrike-slipfaultrequiresastress exampleofageometricallycomplexfaultsystemthatmayexem- 01
fieldwithanESE-orientedmaximumprincipalstressaxis,σ1,and plifymanyoftheattributesthathavebeenstudiedpreviouslywith 9
anNNE-orientedminimumprincipalstressaxis,σ3.Insummary, moregenericdynamicfaultingmodels.Manyresearchers(Segall&
theNAFintheMarmaraSeaexperiencesanalmosthorizontalmax- Pollard 1980; Harris et al. 1991; Harris & Day 1993, 1999;
imumcompressiveaxisσ1inthewesternpart,whichisrotatedby Yamashita & Umeda 1994; Kase & Kuge 1998, 2001; Duan &
16◦ incounter-clockwisesenseintheeasternpart.Theσ2-axisis Oglesby2006;Oglesby2008;Oglesbyetal.2008)havemodelled
almostverticalintheeast(puresheartectonics),whereasitplunges the behaviour of fault systems with unlinked stepovers, and have
at36◦inthewest(transpressivetectonics).Pinaretal.(2003)also foundthattheabilityofrupturetopropagateacrossastepoverde-
reportchangesintheσ3-axis,beingalmosthorizontalintheeast pendson(1)thegeometricalpropertiesofthestepover(overlapor
anddipping49◦inthewest. gap along strike; extensional or compressional stepover), and (2)
Inthisstudy,weinvestigaterupturedynamics,sourcesizesand thedetailsofthestressandslippatternontheedgesofbothfault
ground-motionpatternsforpotentialearthquakesonthefaultsys- segments. Other work has focused on the effects of fault bends
tem of the NAF in the Marmara Sea using spontaneous dynamic onrupturedynamics(Bouchon&Streiff1997;Aochietal.2000,
rupturesimulations.Weconsiderthe3-Dfaultconfigurationpro- 2002, 2005; Aochi & Fukuyama 2002; Harris et al. 2002; Aochi
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
RupturedynamicsontheNorthAnatolianFault 1073
&Madaraiga2003;Oglesby&Archuleta2003;Duan&Oglesby rupture timing). Our study is fundamentally different, in that we
2005).Theresultsofthesestudiesindicatethatrupturepropagation seektomodelthebasicphysicsoftheearthquakeprocess.Rather
atafaultbenddependsonaninteractionbetweentheinitialapplied thanbeingspecifiedapriori,theearthquakesize,slipdistribution,
stressfieldandthedynamicstressfieldradiatedbytherupturefront; rupturetimingandtheradiatedseismicwavesareaproductofthe
faultbendsmayserveasbarrierstoruptureinsomecases,whereas simulations. As such, our work extends previous ground-motion
inothercasestheymayallowthrough-goingrupture. simulationworkintheregionaswebuildphysicallyself-consistent
TheNAFintheMarmaraSea,however,ischaracterizedbyalarge models.
linkedfaultstepover,just∼20kmsouthofIstanbul(Fig.1;stepover
insegmentsG,HandI).Afewstudieshaveexaminedsuchlinked
stepovers. Nielsen & Knopoff (1998) modelled linked stepovers
2 METHOD
over multiple earthquake cycles using a quasi-static method, and
foundthattheseareascouldserveasbothbarrierstoruptureand OurmodellingmethodsandphysicalsetupfollowthoseofOglesby
preferred zones of rupture nucleation. Magistrale & Day (1999) etal.(2008).Weusea3-Dexplicitfinite-elementmethod(Whirley
modelledthedynamicsofthrustfaultsthatarelinkedbyunloaded &Engelmann1993;Oglesby1999)tomodelthespontaneous,dy-
strike-sliptearfaults,andfoundthatthepresenceofalinkingfault namicruptureofthenon-planarNAF.Thismethodhasbeenusedto D
o
greatlyincreasedthepotentialrupturedistancebetweenthethrust reproducerealisticrupturebehaviourinpastearthquakes(Oglesby w
n
faults.Oglesby(2005)performeddynamicmodelsofearthquakes &Day2001;Oglesbyetal.2004),andhasbeenvalidatedaspartof loa
d
on strike-slip faults with linking dip-slip faults; this general con- theSouthernCaliforniaEarthquakeCenter/U.S.GeologicalSurvey e
d
figuration more closely matches the NAF geometry in this work. DynamicRuptureCodeVerificationExercise(Harrisetal.2009). fro
Hefoundthattheresultsdependonwhetherthestepoverwascom- Thenumericalmethoddoesnotallowfaultopening,butwhenthe m
pressional or extensional: slip on the strike-slip segments in an effectivenormalstressgoestozeroorbecomestensile,theshear http
extensional stepover tended to unclamp the linking normal fault, stressissettozero. s
leadingtothrough-goingrupture,whereasthesameeffecttended Ourassumedfaultgeometryfollowsasimplifiedversionofthat ://a
c
a
toclampthelinkingthrustfaultinacompressionalstepover,thus inferredbyArmijoetal.(2005),showninmapviewinFig.1;the3- d
e
promoting rupture arrest. He also found that due to the complex Dfaultparametrization(averagelengthofanelementedge=500m) m
ic
dynamic stress interaction between the different segments during isportrayedinFig.2.Wenotethatforcomputationalease,theco- .o
the rupture process, the final size of the earthquake is often de- ordinatesystemofour3-Dfaultgeometryisrotated1.4◦ counter- up
.c
termined by the location of the hypocentre, and the sequence in clockwise from the geographic coordinate system. Our computa- o
m
which individual segments rupture. Recently, Lozos et al. (2011) tionaltimestepis8.9×10−3s.Directinspectionofslip-weakening /g
performeda2-Ddynamicparameterstudyofstrike-slipfaultswith curvesindicatesthatourgridspacingallowsustoresolvethephysi- ji/a
stepoversandlinkingstrike-slipsegments,andfoundthatforlarge calprocessoffaultstrengthbreakdown.Inaddition,asdetailedlater rtic
stepoverwidthstheabilityofrupturetopropagatethroughalinked inthiswork,weperformedsmallerscalemodelswithhigherspatial le-a
stepover was dominated by the static stress field in the stepover resolution(gridsize=200m)andfoundthatourresultsarerobust bs
region,whereasforshortstepoverwidthsitwasdominatedbythe withrespecttogridsize.Thus,weverifythatournumericalmethod tra
c
dynamicstressfieldradiatedbyfaultslip. isresolvingthephysicalprocessesofinterest.Withthelowestwave t/1
TheNAFisanidealtestcasetoseeifsomeofthedynamicef- speedbeing1.8616kms–1 inourmodelling,weestimatethatour 88
/3
fectsdiscoveredinthestudiesaboveapplyinanaturalfaultsystem, maximumresolvedwavefrequencyisaround0.4Hz,assuming10 /1
0
and whether they have important implications for ground motion gridsperwavelength.Faultdipsareallvertical,exceptthatofthe 7
1
in this region. Previously, Oglesby et al. (2008) investigated the oblique normal segment H, which has a dip of 70◦. Experiments /6
8
effectofnucleationlocationonearthquakesizeinthisregion,and withverticaldipforsegmentHimplythattheresultsarenotvery 47
3
notedthatanewmodeofsupershearrupturepropagationmayre- sensitivetodip.Thefaultsystemisembeddedina3-Dsolidwith 0
b
sultfromslipongeometricallycomplexfaultsystems.Thecurrent boundariesfarenoughfromthefaulttoavoidedgereflectionsarriv- y
g
workgreatlyextendsthisinitialworkbyinvestigatingindetail(1) inginthedomainofinterestduringthemodelperiod.Wenotethat u
e
theeffectofstressorientationonearthquakesize,(2)theeffectof innature,shortsegmentsDandFarelikelytobezonesofcomplex st o
non-planarfaultgeometryongroundmotionand(3)thenewmech- deformation,butweparametrizethemasshortfaultsegmentsfor n
0
anism of supershear rupture propagation on non-planar faults. In simplicity.Thefaultsareembeddedinalinearlyelastic,isotropic, 7
A
ourstudy,weusedynamicmodelstoinvestigatehowdifferencesin layeredmaterial(Oglesbyetal.2008). p
earthquakesizes,faultslipdistributions,rupturepropagationpat- Wenotethatourcornersareeffectivelysmearedoutoverroughly ril 2
0
terns and ground motion arise from specific assumptions on the oneelement(500m)width,sostressbuild-upsatthefaultcorners 1
9
regionalstressfieldandtherupturenucleationlocation.Although arenotsingular.Thedynamicnormalstressinourmodelseventually
thefaultgeometry,stressfieldandotherinputparametersremain reaches zero within one to two elements from the fault edges on
uncertain,experimentswithdifferentparametrizationsallowusto segmentsD,FandH,butthiseffectsimplyleadstotheshearstress
bracketsomeofthepossiblefaultingandground-motionbehaviour on these points smoothly approaching zero; it does not result in
inthisarea.Forinstance,changesintheregionalstressdirectionof extraseismicradiationorothernon-physicalbehaviourofthenodes
±10◦ maysignificantlyaltertheruptureproperties,whichinturn in question. A more realistic model would incorporate smoother
affectsthegroundmotions,whicharealsosensitivetotheoverall cornersorzonesofoff-faultdamagearoundthefaultcorners.We
faultgeometry.Wealsofindthattheverydefinitionofruptureve- leaveunansweredthequestionofwhetherfaultopeningmayhappen
locity may need closer examination if dynamic models are to be innature,butduetorockfailureandthecuttingofsecondaryfaults
usedinconjunctionwithkinematicmodels.Notethatotherstud- itisunlikelytobecommon;webelievethatthelackoffaultopening
ies(e.g.Pulidoetal.2004;Sørensenetal.2007)havepreviously inourmodelsdoesnotaffectourresultssignificantly.
modelledgroundmotioninthisregionwithkinematicsourcemod- OurnumericalmodelsemployaCoulombfrictionlawτ ≤μσ ,
n
els(i.e.withassumedearthquakesize,faultslipdistributionsand where τ is the frictional stress on the fault, μ is the frictional
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
1074 D.D.OglesbyandP.M.Mai
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
Ffoirguwrheic2h.s3y-nDthmeteicshslvipie-rwatoeftifmauelthgisetoomrieestrayr;efpolrovttiesduailnctlhaerietylecthtreodniisccsruetpipzaletimonenstizaeresmhoawrknediswdiotuhbtlreiatnhgeleasc.tualcomputationaldiscretizationsize.Locations s://ac
a
d
e
m
coefficient and σn is the normal stress (positive in compression) 5.5-km nucleation size on segment B imply that the results be- ic.o
acrossthefault.Weuseaslip-weakeningmodelfortheevolution yondthenucleationpatcharenotverysensitivetothenucleation up
ofμ(Ida1972;Palmer&Rice1973;Andrews1976a)withaspa- patchsize. .co
m
tially invariant slip-weakening distance of 0.4 m, a static friction We apply a regional stress field inferred from moment tensor /g
coefficientof0.6andaslidingfrictioncoefficientof0.1.Thelarge inversionsofsmall-tomedium-sizedearthquakes(Pinaretal.2003; ji/a
drop in frictional coefficient and centimetre-scale slip-weakening Oglesby et al. 2008), with an orientation depicted in Fig. 1. This rtic
distancearemotivatedbyrecentexperimental/observational(Tsut- regionalstressfieldisthenresolvedintostrike-slip,down-dipand le
-a
sumi & Shimamoto 1997; Goldsby & Tullis 2002; Di Toro et al. fault-normalstresscomponentsonfaultsegmentsA–Iasafunction b
s
2004;Hanetal.2007)andtheoretical/numerical(Andrews2002; ofeachsegment’sorientation(valuesofshearandnormalstressfor tra
c
Rice,2006;Suzuki&Yamashita2006;Beeleretal.2008;Bizzarri, eachsegmentaregiveninTable1).Thesevaluestaperlinearlyto t/1
2011)workthatimplysignificantfaultweakeningathighsliprates 0.1oftheirambientvaluesbetweenadepthof3kmandthefree 88
duetoeffectssuchasmelting,flashheatingandporefluidpressur- surface. We assume a relatively constant effective normal stress /3/1
ization.Ourchoiceoftheslip-weakeningdistanceisconsistentwith (normal stress minus pore fluid pressure) with depth as in many 07
1
theworkcitedearlierandalsoismotivatedbycomputationalstabil- previous faulting models (e.g. Tse & Rice 1986; Rice 1993) to /6
8
ityconcerns.A50percentlargerslip-weakeningdistanceresults avoid a systematically higher stress drop on deeper parts of the 4
7
inquiteminorchangesinthemodelledruptureandslippatterns, fault (which is not typically observed in real-world earthquakes). 30
whereas a 50 per cent smaller slip-weakening distance results in Thispatternofeffectivenormalstresscanbeattributedtothepore by
pervasivesupershearrupturepropagation,whichweattributetoa pressureincreasingatlithostaticratherthanhydrostraticratesbelow gu
e
lackofresolutionoftheslip-weakeningprocessinsuchacase. acertainthresholddepth. s
Changes in the compressive normal stress (typical in cases of In this regional stress field, segments A–G and I are loaded in t on
complex fault geometry) are immediately reflected in this model right-lateral strike-slip direction, whereas segment H is loaded in 07
aschangesinthefrictionalstress.Thisinterplaybetweenthetwo anobliquenormal/right-lateralmanner.However,thisloadingdoes A
p
components of stress will have important implications for the re- notconstituteanaprioriconstraintinthedirectionofslip;slipis ril 2
sults in this work. The large stress drop with respect to the ab- allowed in any on-fault direction and its direction is a calculated 0
1
9
solute stresses implied by the above parametrization results in resultofthemodels.Wealsoinvestigatetheeffectsonsourcedy-
strongdynamiceffectsintheresults,aswillbeshownlaterinthis namics when rotating the regional stress field by 10◦ clockwise
work. Rupture nucleation at the selected hypocentre is achieved and counter-clockwise from the above orientation to test the ro-
in our models by artificially increasing the shear stress to the bustnessofourresultstodifferentassumptionsaboutthetectonic
failure level in an expanding circle (rupture velocity = 3.0 km loading.Itisimportanttonotethatduetothedifferentfaultsegment
s–1) up to a radius of 7 km; beyond this radius, the rupture time orientations,individualfaultsegmentshavesignificantlydifferent
is a calculated result of the models. We note that such a large- initial shear stress and strength. One measure of the differences
size nucleation patch is dictated by our assumed stress field: a in stress between the fault segments is the relative fault strength
critical nucleation size for spontaneous nucleation (e.g. Andrews parameter S = τyield−τ0 (Andrews 1976b). Under the simplifying
τ−τ
1976a; Day 1982) applies on each fault segment, and we find assumptionthatτ0 fin=al μ σ ,wecalculatethevalueof S for
final sliding n
that a 7-km large patch is needed to ensure rupture propagation each segment (listed in Table 1). By this measure, we note that
inourmodels,especiallythosewithnucleationonless-favourable segments such as A, C, G and I are rather favourable for rupture
segments. However, additional numerical experiments with a (S < 1.1), whereas segments B, E, F and H are less favourable
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
RupturedynamicsontheNorthAnatolianFault 1075
(S >1.2).Astaticcalculationof S mayhelptoexplainsomeas-
pects of dynamic rupture behaviour. However, slip on one fault
segmentcanstronglyaffectthestressfieldofothersegments,sothe
effectivevalueof S oneachfaultsegmentchangeswithtimeasa
resultoftheinterplaybetweenthetectonicpre-stressfieldandthe
dynamicstresses.
3 RESULTS
3.1 Stressincrementsduetoslip
Oglesbyetal.(2008)exploredtheeffectofnucleationlocationon
theabilityofrupturetopropagateacrossthestepoversegmentH.
Theyfoundthatwhenrupturenucleatedfarfromthestepoverseg-
D
ment(i.e.nucleationonsegmentEorfartherwest),rupturecould ow
propagate across the statically unfavourable oblique segment H. nlo
Rupturesthatnucleatednearertooronthestepoversegment(i.e. ad
e
segmentsG,HorI)arrestedprematurelyonsegmentH.Therea- d
sonforthecomplexrelationshipbetweennucleationlocationand fro
m
final earthquake size is the interaction between the tectonic pre- h
stressfieldandthedynamicstressesduringtheearthquakerupture ttp
s
process.Deeperinsightintothisinteractionmaybegainedbyex- ://a
aminingthechangeinshearandnormalstressonthefaultsegments ca
d
duetoathrough-goingrupture.FornucleationonsegmentB,the e
m
finalstaticstressdrop(i.e.finalminusinitialshearstress)indicates ic
adropinshearstressovertheentirefaultsystem(Fig.3,toppanel) .ou
p
as expected for a through-going rupture. More interesting is the .c
o
normalstresschange(Fig.3,bottompanel)showingstrongposi- m
tive(clamping)andnegative(unclamping)stressvariationsclosely /gji/a
associatedwithdiscontinuitiesinthefaultgeometry.Forexample, rtic
right-lateralsliponsegmentsGandIhasunclampedtheedgesof le
segmentH,makingitmorefavourableforrupturethanifitwereto -a
b
s
rsuopmtueroefinthiesosltaritkioen-s.liTphsisegumncelnatmspminaygpexropdlauicnesrwuhpytunreuscltehaattiopnroopn- Fonigfuaruelt3s.ysItnecmredmueentotsainnesahretharqu(taokpepnauncelel)aatenddonnosremgamle(nbtoBtto.Smtrpoanngecl)hastnrgeesss tract/1
agate across segment H, even though the tectonic pre-stress (i.e. 8
innormalstressareconcentratedattheedgesofthefaultsegments.Red 8
its S value) would imply that segment H is a barrier to rupture. /3
denotesclampingandbluedenotesunclampingofthefault. /1
Conversely,right-lateralsliponsegmentHhasclampeddownthe 0
7
adjacentedgesofsegmentsGandI.Asimilarpatterncanbeseen 1
/6
on (and adjacent to) stepover segments D and F. These segments 8
aredynamicallyunclamped,buttendtohavehigherinitialS-values fieldisnotveryaccuratelydetermined,weexaminehowvariations 473
intheregionalstressfieldaffectourmodelling.Weconstructtwo 0
duetotheirorientation(S=1.22andS=2.42,respectively);rup- b
additionalsetsofmodelswithdifferentregionalstressorientation y
turepropagationonthesesegmentsisessentiallyacontestbetween g
butthe sameregional stressamplitude: thefirstiswith thestress u
the tectonic regional stress, which favours rupture slowing down e
andterminating,anddynamicunclamping,whichfavoursthrough- sfitereldssrootraietendtactioounn,taenr-dcltohcekswecisoendbyis1w0◦ithwtihthesretrsepsescfitetoldtrhoetaotreidgi1n0a◦l st on
going rupture. This result corroborates findings of Oglesby et al. 0
clockwise.Finalslippatternsfortheseregionalstressorientations 7
(2008): for the largest and least favourable stepover segment H, A
aredisplayedinFig.4,withthenon-rotatedresultsreproducedfrom p
nbuucillde-autiponoffadritrheecrtiavwitayyafnrodmdythneamsteicposvtreersssewgmaveenstathllaonwnsuacgleraetaitoenr OgTlehsebyefefteaclt.o(2f0r0o8ta)tiinngthtehSeusptpreosrstifingelIdnfcoorumnatetiro-cnlofocrkcwoimsepacrainsobne. ril 201
closetosegmentH.Thus,onlyearthquakesthatnucleatefarfrom 9
predictedreasonablywellbyexaminingFig.1.Inthisrotatedstress
segment H will unclamp segment H enough to generate through-
field, segment H becomes almost parallel to the maximum com-
goingrupture;earthquakesthatnucleateadjacentto(oron)segment
pressivestressσ ,thusexperiencesaverylowresolvedshearstress,
Hdonotexperienceenoughofthis‘dynamicdirectivity’effectto 1
andisthereforequiteunfavourableforslip.Thedynamicrupture
overcometheunfavourabletectonicpre-stressfield.Intheselatter
computationscorroboratethisconjecture,asshowninthefinalslip
cases, rupture terminates quickly when it propagates outside the
distributions(Fig.4):allsimulations,includingtheBandEnucle-
regionofsignificantdynamicunclampingonsegmentH.
ation,terminateonsegmentHunderthecounter-clockwiserotated
stressfield.IfnucleatedonsegmentsGandI,theruptureproceeds
an even shorter distance across segment H than in our primary
3.2 Effectofregionalstressorientation
modelsbeforedyingout.
Asnotedearlier,theresultsofoursimulationsreflecttheinterplay A 10◦ clockwise rotation of the regional stress field results in
between the regional tectonic stress field and the dynamic stress morecomplicatedrupturebehaviourthana10◦ counter-clockwise
fieldradiatedbythepropagatingrupture.Sincetheregionalstress rotation. Under this assumption, oblique segment H changes to
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
1076 D.D.OglesbyandP.M.Mai
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/g
ji/a
rtic
le
-a
b
s
tra
c
t/1
8
8
/3
/1
0
7
1
/6
8
4
7
3
0
b
y
g
u
e
s
t o
n
0
7
A
Fcoiguuntreer-4c.loFciknwalisseli1p0◦pa(dtteenrnostefdoCrCeaWrt,htqoupapkaensenl)uacnledactelodcoknwifsaeu1lt0s◦e(gdmeneonttesdBC(Wa),,bEott(obm),pGan(ecl)),wHit(hdr)esapnedctIt(oet)h.eRpersiumltasryarsetrsehssowfienldf.oTrhsetreosrsiefintealtdisonrootfattehde pril 2
0
regionalstressfieldhasaverystrongeffectonthefinalearthquakesize,andaffectstheearthquakesdifferentlydependingonhypocentrelocation. 1
9
becomemorefavourableforrupture,whereasthemainstrike-slip I.NucleationonsegmentsHandIproducesrupturethatpropagates
segments A, B, C, E, G and I are less favourable. Nucleation on acrosstheentirefaultsystem.Thus,byrotatingtheregionalstress
segment B still produces through-going rupture, but the slip dis- fieldonly10◦ (avaluewithintheuncertaintyofthestressorienta-
tribution has changed qualitatively with respect to the pattern in tionestimates),thebehaviourofthefaultsystemduringdynamic
Oglesby et al. (2008), with higher slip on segment H and signif- ruptureisdrasticallychanged.Theroleofabarrierisnowtakenby
icantlylowerdisplacementsonsegmentE(whichisunfavourable segmentE.NucleationonsegmentsHandI,whichunderourinitial
forruptureinthisregionalstressfield).NucleationonsegmentE, assumptionsinOglesbyetal.(2008)producedsmallerevents,now
under our nucleation assumptions, does not produce spontaneous producewhole-systemrupture.Conversely,nucleationonsegment
rupture.NucleationonsegmentGproducesrupturethatdiesouton E, which previously led to a whole-system event, does not even
segmentsFandE,butpropagatesthroughsegmentH,ontosegment produceapropagatingrupture.
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
RupturedynamicsontheNorthAnatolianFault 1077
cleationonsegmentB,itslowsdownandalmoststopsonsegment
3.3 Temporalevolutionofrupture
FfornucleationonsegmentE.Thus,thesamesegmenthasoppo-
We now return to an in-depth discussion of our dynamic rupture siteeffectsonrupturepropagationfordifferentnucleationlocations
calculationsforthebasic(unrotated)regionalstressfield.Morein- withthesamedirectivity.Thisphenomenonmaybeunderstoodas
sight into the temporal evolution of rupture propagation, and the acombinationofstaticanddynamiceffects:segmentFisstatically
dynamiceffectsthatinfluenceit,maybegleanedbyexaminingslip unfavourableforrupture,andintheEnucleationcasetherupture
velocity pulses at hypocentral depth (Fig. 5). Nucleation on seg- fronthasless‘dynamicdirectivity’thanintheBnucleationcase.
ment B (Fig. 5a) produces largely unilateral rupture that appears Thus,forEnucleationtheshearstressincreaseandthedynamicun-
topropagateataroughlyconstantrupturevelocityuntilitreaches clampingarediminished,andrupturepropagationacrosssegment
segmentF,whereittemporarilyspeedsup;onsegmentItherupture F is more difficult. As in the B nucleation case, slip on segment
transitions permanently to supershear speed. The sudden rupture Ftends toclampsegmentG,further inhibitingcontinued rupture
acceleration on segment F is due to the dynamic unclamping of propagation. After a delay of approximately 3 s, rupture restarts
thissegment,whichisotherwiseorientedunfavourablyforrupture spontaneously on segment G, and then propagates across the en-
in this stress regime. We note that rupture onto segment G is de- tire fault system. Nucleation on segment G (Fig. 5c) produces a
layed by approximately 0.5 s because slip on segment F tends to similarly complex rupture pattern. For the same reasons as the E Do
clamp segment G. The acceleration to supershear speed on seg- nucleationcase,rupturealmostdiesoutattheintersectionbetween wn
ment I follows the standard Burridge–Andrews mechanism, with segments G and F (although travelling in the opposite direction), loa
d
thissegmenthavinganSvalueoflessthan1.77(Andrews1976b). whilethemagnitudeofdynamicunclampingisinsufficientforrup- e
d
NucleationonsegmentE(Fig.5b)producesacomplexpatternin turetopropagateacrosssegmentH.NucleationonsegmentsHand fro
whichtherupturealmostdiesoutattheintersectionsofsegmentsF IbothproducerupturesthatterminateonsegmentH(notshownfor m
andG.WenotethatwhileruptureacceleratesonsegmentFfornu- brevity). http
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/g
ji/a
rtic
le
-a
b
s
tra
c
t/1
8
8
/3
/1
0
7
1
/6
8
4
7
3
0
b
y
g
u
e
s
t o
n
0
7
A
p
ril 2
0
1
9
Figure5. Syntheticslipvelocityamplitude(incorporatingbothstrike-slipanddip-slipcomponents)timehistoriesathypocentraldepth(Z=8km)forpoints
alongstrike.Differentfaultsegmentsarelabelledandcolouredalternatelyredandblueforclarity.Rupturespeedsupandslowsdownondifferentfault
segments,dependingonthenucleationlocation.TheblacklineslabelledVpandVsindicatethemoveoutoftherupturefrontpropagatingattheP-andS-wave
speed,respectively,atthatdepth.
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
1078 D.D.OglesbyandP.M.Mai
Byanalysingrupturetimeandrupturevelocityforeachpointon (Bernard&Madariaga1984;Spudich&Frazer1984),andalsohas
thefaultsystem,weexamineindetailtheconnectionbetweenfault beennotedinsome2-Dfaultingmodels(Bouchon&Streiff1997;
geometryandvariationsinrupturepropagation.Fig.6showsrup- Duan & Oglesby 2005; Adda-Bedia & Madariaga 2008). These
turetimecontoursandrupturevelocityfornucleationonsegments lobesareparticularlyvisibleatsegmentintersectionsE–F,F–G,and
B,EandG.Rupturevelocityiscalculatedbasedonthetimeatwhich G–H.
slipvelocityexceeds1mms–1 atagivenpointonthefault.The Anotherstrikingfeatureistheasymmetryofgroundmotionwith
reciprocal of the spatial gradient of this rupture time distribution respect to the northern and southern sides of the fault. In con-
isequaltothelocalrupturevelocity.Notethatthisvalueneednot trasttothesymmetricpatternexpectedforsimpleplanarstrike-slip
correspondtoanyphysicalquantityinthematerial,butrathercon- faults,theinsidecornersofthefaulttendtohaveamplifiedmotion
stitutesanapparentrupturespeedoverthefaultplane.Ourresults compared to the outside corners. This effect has been noted for
arenotaffectedbychangingtheslipvelocitythresholdbyafactor faultwalldisplacementsin2-Dfaultingmodels(Davis&Knopoff
oftwo(smallerorlarger),implyingthatthechosenthresholdvalue 1991; Duan & Oglesby 2005), and this work confirms that it is
doesnotproduceartefactsintherupturevelocitydistribution.The also a strong effect for 3-D ground motion. In more detail, the
‘checkerboard’ noise inthe rupture velocity plots occurs because east–west motion (approximately fault-parallel) and north–south
particlevelocitiesarerecordedatanintervalof0.01s,leadingto motion(fault-normal)fornucleationonBrevealthatasymmetric D
o
an apparent quantizing of the rupture velocity; the actual rupture groundmotionaredifferentforthedifferentcomponentsofmotion: wn
propagationissmooth. theeast–westcomponentexhibitslargergroundmotiononthesouth loa
d
AnalysingrupturevelocityfornucleationonB(Fig.6a)reveals sideofsegmentEandthenortheastsideofsegmentH,whereasthe e
d
aninterestingpattern:asnotedinOglesbyetal.(2008),supershear opposite is true for the north–south component of motion. The fro
ruptureoccursonthewesternedgesofsegmentsDandF,andthe asymmetryofgroundmotionaroundsegmentHisatleastpartly m
westernfreesurfaceofH.Anewfeatureinoursimulationsarises attributabletothenon-verticaldipangleofthissegment(e.g.Davis http
fornucleationonsegmentsE(Fig.6b)andG(Fig.6c):therupture &Knopoff1991;Oglesbyetal.1998),whereastheasymmetryon s
velocity pattern on segment D is spatially reversed (compared to the other segments arises from non-planarity of the fault system ://a
c
a
nucleation on B). In these cases, supershear rupture is confined andcorrespondingseismicradiationatfaultkinks(Adda-Bedia& d
e
to the eastern edge of segment D. This pattern reversal can be Madariaga2008). m
ic
explained if one attributes the supershear rupture propagation to Theneteffectofdirectivity,cornerphasesradiatedatthefault .o
u
thedynamicunclampingofthesegment.Whenrupturepropagates bends and asymmetric particle motion is to produce a complex p
.c
towards segment D from either side, it unclamps the edge of D near-fault peak-motion pattern that becomes more homogeneous o
m
closesttothepreviouslyslippingsegment.Theunclampedregion fartherfromthefaultduetoanaveragingeffectonradiationfrom /g
hassignificantlylowerstrength,andthusfavourssupershearrupture largerareasonthefaultsystem.WhencomparingtheGnucleation ji/a
propagation.However,asrupturetraversessegmentD,itthenexits casetotheBnucleationcase(Figs7banda),weseesomesignif- rtic
le
the region of significant unclamping, and reverts to its original icant differences. The obvious effects are the opposite directions -a
slower velocity. This effect can be thought of as an extreme case of amplification due to directivity and the lack of strong ground bs
oftherupture-frontaccelerationseeninbentfaultmodelsofKase motion near segments H and I, because of rupture dying out on tra
c
&Day(2006).Thismechanismfortransitiontosupershearrupture H in the G nucleation case. Another potentially important effect t/1
8
velocityanditsimplicationswillbeexaminedinmoredetailinthe is the spatial variation of ground-motion asymmetry for different 8
/3
Discussionsection. directionsofrupturepropagation.Thiseffectisparticularlynotable /1
0
nearthesmallstepoversegmentsDandF.Wewillexaminethecase 7
1
ofsegmentDinmoredetail(betweenapproximately35and50km /6
8
3.4 Patternofgroundmotion alongstrike),asshownintheFigs7(c)and(d),whichdisplaythe 473
peak horizontal ground velocity (the magnitude of the horizontal 0
b
The simulated ground motion from potential earthquakes on this velocityvector)fortheBandGnucleationcases,respectively.In y
g
sectionoftheNAFdisplaysmanyfeaturesrelatedtothefaultgeom- the B nucleation case (Fig. 7c), rupture approaches from the left u
e
etryandassociateddynamiceffects.Fig.7comparesthesimulated (west) edge, and its propagation from segments C to D produces st o
peakgroundvelocity(filteredto0.25Hz)ofthethreecomponents stronger particle motion on the inside of the corner and forward n
0
ofmotionfornucleationonsegmentsB(Fig.7a)andG(Fig.7b). alongstrike,thatis,onthesouthernsideofsegmentD.Asimilar 7
A
The B nucleation case is useful as a general illustration. As ex- effectoccursontheD–Ecorner,exceptwiththeoppositedirection p
pected, ground motion generally increases to the east, reflecting of bend. The net effect is to produce stronger ground shaking on ril 2
0
theeastwarddirectivityoftherupturefront.Thenorth–southcom- thesouthernsideofsegmentDovermostofitslength,andhigher 1
9
ponent of motion, normal to the predominant strike of the fault ground motion on the northern side of segment E. A similar ef-
system, measures the largest ground motion, as expected in the fect occurs on segment F (between approximately 85 and 90 km
caseofprimarilysubshearrupture.However,thisfamiliarpattern along strike). When rupture approaches these segments from the
is strongly modulated by the fault geometry. Segment H exhibits opposite direction, such as in the case of G nucleation (Fig. 7d),
smaller ground motion than segment G, even though it is farther thiseffectreversesitself.WhentheruptureapproachessegmentD
along strike. Because of the slow rupture propagation around the fromsegmentE,ithashighdirectivityandradiatesmorestrongly
corner from segments H to I, segment I does not experience a on the E/D corner, producing higher ground motion on the north
strongdirectivityeffect,leadingtosmallernear-faultgroundmo- sideofsegmentD.AsimilareffectisseenontheD/Ccorner,with
tion than on segment G, which has identical values of pre-stress. highergroundmotiononthesouthsideofsegmentC.Theground-
Additionally,changesinfaultstrike,whichcausetherupturefront motionasymmetryalsoflipssidesonsegmentF.Thiseffectismost
toaccelerateviaachangeinspeedand/ordirection,areassociated significant for small segments, where the radiation effects from
withstronglobesofradiation.This‘cornerphase’effectisconsis- oneofthecornersissignificantovertheentirelengthoftheseg-
tentwithisochroneaccelerationleadingtostrongseismicradiation ment.Thistypeofground-motionvariabilityandasymmetrydueto
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
RupturedynamicsontheNorthAnatolianFault 1079
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/g
ji/a
rtic
le
-a
b
s
tra
c
t/1
8
8
/3
/1
0
7
1
/6
8
4
7
3
0
b
y
g
u
e
s
t o
n
0
7
A
p
ril 2
0
1
9
Figure6. Rupturetimecontours(toppanel)andrupturevelocity(bottompanel)forearthquakesnucleatedonsegmentB(a),E(b)andG(c).Thelocationof
supershearrupturepropagationonsegmentDchangesdependingonthenucleationlocationandhencethedirectionofrupturepropagation.
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS
1080 D.D.OglesbyandP.M.Mai
D
o
w
n
lo
a
d
e
d
fro
m
h
ttp
s
://a
c
a
d
e
m
ic
.o
u
p
.c
o
m
/g
ji/a
rtic
le
-a
b
s
tra
c
t/1
8
8
/3
/1
0
7
1
/6
8
4
7
3
0
b
y
g
u
e
s
t o
n
0
7
A
p
sFeiggmureent7B. P(eaa)kanpdarGtic(lbe).veNloocrtihty–Saomupthlitduidreecitnioenasist–pweerpsten(tdoipcuplaarnetol),thNeoprtrhim–Saoryutfhau(mltisdtdrilkee,paannedl)thaunsdduisppdloawysngdrieraetcetrioenffse(cbtsototofmdirpeacntievli)tyf.oTrhneucnloeant-ipolnanoanr ril 20
1
faultgeometryproducesnumerouslocalhigh-amplitudeareasintheground-motionfield,whichchangesignificantlyfordifferenthypocentrelocations.(c) 9
PeakhorizontalparticlevelocityforsegmentsC,DandEfortheBnucleationcase.(d)PeakhorizontalparticlevelocityforsegmentsC,DandEfortheG
nucleationcase.Trianglesindicatelocationsforsyntheticvelocitytimehistoryplotsinelectronicsupplement.Notethattheparticlemotionasymmetryflips
inthenorth–southdirectionfordifferentrupturedirectivity.
complexityinfaultgeometryisparticularlypronouncedinthenear- IintheBnucleationmodel(Fig.7a),wherea‘true’(long-lasting)
field region, and hence is important for earthquake engineering supersheartransitiontakesplace.However,thereisnonotableMach
practiceinground-motionprediction. cone;thesupershearrupturemanifestsitselfonlyasstrongerstrike-
Finally, we note that typical signatures of supershear rupture parallel motion than strike-perpendicular. The lack of supershear
propagation(i.e.aMachconeandfault-parallelmotionthatishigher ground-motionattributesovermostoftheourmodelspaceislikely
than fault-normal motion) are not obvious in our ground-motion duetothesmallareasoverwhichsupershearruptureoccurs;italso
maps, even though supershear rupture propagation occurs in our mayimplythatthepropagationspeedofthepeakslip-ratepulseis
simulations.Theonlyexceptionisthegroundmotionnearsegment moreimportantforground-motiongenerationthanthepropagation
(cid:4)C 2012TheAuthors,GJI,188,1071–1087
GeophysicalJournalInternational(cid:4)C 2012RAS