Earth and Planetary Science Letters 500 (2018) 1–14
Contents lists available atScienceDirect
Earth and Planetary Science Letters
www.elsevier.com/locate/epsl
Dual control of fault intersections on stop-start rupture in the 2016 Central Italy seismic sequence
R.J. Waltersa,∗,L.C. Gregoryb,L.N.J. Wedmoreb,1, T.J. Craigb, K. McCaffreyc, M. Wilkinsond, J. Chene,Z. Lie, J.R. Elliottb,H. Goodallb,F. Iezzif, F. Liviog, A.M. Michettig,G. Robertsf,E. Vittorih
aCOMET,DepartmentofEarthSciences,DurhamUniversity,Durham,UK bCOMET,SchoolofEarth&Environment,UniversityofLeeds,Leeds,UK cDepartmentofEarthSciences,DurhamUniversity,Durham,UK
dGeospatialResearchLtd.,DepartmentofEarthSciences,DurhamUniversity,Durham,UK eCOMET,SchoolofEngineering,NewcastleUniversity,Newcastle,UK
fDepartmentofEarthandPlanetarySciences,Birkbeck,UniversityofLondon,London,UK gDipartimentodiScienzaeAltatecnologia,UniversitàdegliStudidell’Insubria,Como,Italy hServizioGeologicod’Italia,ISPRA,Rome,Italy
a r t i c l e i n f o a b s t ra c t
Articlehistory:
Received21February2018 Receivedinrevisedform23July2018 Accepted29July2018
Availableonlinexxxx Editor:J.-P.Avouac
Keywords:
earthquakes InSAR
seismicsequence faultsegmentation fluiddiffusion
Largecontinentalearthquakesnecessarilyinvolvefailureofmultiplefaultsorsegments.Butthesesame critically-stressedsystemssometimesfailindrawn-outsequencesofsmallerearthquakesoverdaysor yearsinstead.Thesetwomodesoffailurehavevastlydifferentimplicationsforseismichazardanditis notknownwhyfaultsystemssometimesfailinonemodeortheother,orwhatcontrolsthetermination and reinitiation ofslipinprotracted seismicsequences. A paucityofmodern observations ofseismic sequences has hamperedour understanding to-date, buta seriesof three Mw>6 earthquakes from AugusttoNovember2016inCentralItalyrepresentsauniquelywell-observedexample.Hereweexploit awealth ofgeodetic, seismologicaland fielddata tounderstand thespatio-temporalevolutionof the sequence.Ourresultssuggestthatintersectionsbetweenmajorandsubsidiaryfaultscontrolledtheextent andterminationofruptureineacheventinthesequence,andthatfluiddiffusion,channelledalongthese samefaultintersections,mayhavealsodeterminedthetimingofrupturereinitiation.Thisdualcontrol ofsubsurfacestructureonthestop-startrupture inseismicsequencesmay becommon;future efforts shouldfocusoninvestigatingitsprevalence.
©2018TheAuthor(s).PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense (http://creativecommons.org/licenses/by/4.0/).
1. Introduction
In regions of distributed continentalfaulting, networksof ac- tivefaultsarecommonlysegmentedonlengthscalesof10–25km, approximately equal to the seismogenic thickness of the Earth’s crust (Scholz, 1997; Stock and Smith, 2000; Klinger, 2010). This intrinsicmaximumfault size limitsthe magnitudeofcontinental earthquakesthatruptureasingle faultorsegmentto<Mw∼6–7 (Pachecoetal., 1992; Triep andSykes,1997), depending onlocal seismogenicthickness andfault geometry. Therefore,large conti- nentalearthquakesabovethisthreshold(Scholz,1997) suchasthe 2010M7.2El Mayor–Cucapah,Mexico, 2016M7.8Kaikoura, New
* Correspondingauthor.
E-mailaddress:[email protected](R.J. Walters).
1 Nowat:COMET,SchoolofEarthSciences,UniversityofBristol,Bristol,UK.
Zealand, and 1980 M6.9 Irpinia, Italy events (Wei et al., 2011;
Hamlingetal.,2017; WestawayandJackson, 1987) necessarilyin- volve failure ofmultiple faults orsegments. Multi-fault failure in seismic sequences, spanning a longer period of hours to years, is also common, with static stress transfer invoked as the ma- jor cause forthisspatio-temporal clustering oflarge earthquakes withinasmallfractionoftheirestimatedrecurrenceintervals(e.g.
Hubertetal.,1996; KingandCocco,2001; Wedmoreetal.,2017).
Both large earthquakes and seismic sequences require that all component faults are near-critically stressed,a condition that is thought likely to occur commonly in nature through stress- synchronisation offaults (Scholz,2010). In addition, recentwork suggests that as in seismic sequences, the final magnitude of a multi-fault earthquake cannot be predicted from the initial rup- tureprocess(Weietal., 2011).Thissimilarityininitialconditions means that multi-faultearthquakes andseismic sequences begin in thesame wayanddiffer accordingto when rupturestops: ei- https://doi.org/10.1016/j.epsl.2018.07.043
0012-821X/©2018TheAuthor(s).PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/).
Fig. 1. Overviewofepicentralregionandfaultgeometryusedinthisstudy.(a)Regionaltectonicmap,showingmappedactivenormalfaults(magenta,modifiedfromRoberts andMichetti,2004),up-dipsurfaceprojectionofmodelfaultsdisplayedinFigs.6–9(colouredtomatch(c)andFigs.3,4and6),andbodywavefocalmechanismsforeach earthquake(seeFig.2).Whitedashedlineshowsinferredeast-dippingfaultfromFig.7,andrelocatedaftershocksfromChiaraluceetal. (2017) areshowninblack.Locations ofshort-baselineGNSSinstrumentsareshownbybluetriangles.BlackboxshowsextentofFig.3c.(b)Regionalmapshowingthelocationof(a)anddirectionofregional crustalextension.(c)3Dcartoonofmodelfaultgeometryadoptedinthisstudy.Thickcolouredlinesshowthesurfaceprojectionofeachfaultandcorrespondtothecoloured faultsin(a)andFigs.3,4and6.(Forinterpretationofthecoloursinthisfigureandotherfigures,thereaderisreferredtothewebversionofthisarticle.)
therdynamic andstaticstress transfer causecascadingfailure of multiplecritically-stressed faults orrupture isarrested before all thesefaults have failed. In thelatter casethe start of rupture in subsequent subevents determines the temporal evolution of the seismic sequence.Large earthquakes andseismic sequences have vastly different implications for seismic hazard: high hazard in a single event, or moderatehazard spanning years orpotentially decades. Butour understanding of what controlswhethermulti- faultruptureoccursoverdaystoyearsorinseconds,andofwhat controls the spatio-temporal evolution of seismic sequences, has beenseverelylimitedbyapaucityofhigh-resolutionobservations ofmodernseismicsequences.
Combined analysisofgeodetic andseismological datacan im- agestop-start rupturebehaviour andaddress these questions,by disentangling the spatial pattern and temporal evolution of slip inseismicsequences athighresolution.Asequence of3 Mw>6 earthquakesfromAugustto November2016intheCentralApen- ninemountains,Italy(Fig.1)presentsararechancetoinvestigate aseismicsequencewithmoderndatasetsandhereweexploitseis-
mologicalandfieldobservations,aswellasgeodeticdata,toimage the kinematicsofthesequence,andtounderstandstructuraland dynamic controlson itsevolution. Ourresultssuggest thatstruc- tural complexity, namely the intersections between two sets of obliquefaults,mayhaveplayedanimportantdualroleintheCen- tralItalyseismicsequence:firstbylimitingtheextentofindividual rupturesandsecondby channellingfluidflowandcontrollingthe timingofsubsequentfailurethroughoutthesequence.
2. Seismologicalconstraintonearthquakesourcemechanisms
TheCentralItalyseismicsequencestartedwithanM∼6 earth- quakeonthe24thAugust2016,andwasfollowedbytensofthou- sandsofaftershocks,includingtwolargeM>6 eventsonthe26th and30thOctober(Chiaraluceetal.,2017,Fig.1).Werefertothese threemajorearthquakes astheAmatrice,VissoandNorciaevents respectively. Theseismicsequencecontinuedinto2017,withsev- eral earthquakes M<5.7 onJanuary 18th, butherewefocus on thethreelargesteventsonly.
R.J. Walters et al. / Earth and Planetary Science Letters 500 (2018) 1–14 3
Fig. 2. Summaryofseismologicalresults.Leftcolumnshowsthebest-fitseismologicalfocalmechanism(black)determinedusingMT5softwarepackage(Zwicketal.,1994).
ForeachoftheAmatrice,Visso,andNorciaearthquakes,thecompositegeodeticmechanismfromthegeodeticsolutionisshown,asboththefullmomenttensor(dashed greenline)andbestdoublecouple(solidgreenline).Best-fitseismologicalmechanismparametersandsourcetimefunctionarealsogiven.Centralpanelsshowdepth-misfit curves.Eachpointonthecurveisdeterminedbyfixingthecentroiddepthtothegivenvalue,andinvertingthewaveformdatatodeterminethebestfitmechanismand source-timefunctionatthatdepth.Verticalgreenbarsshowthecentroidforthegeodetically-derivedslipdistributionsshowninFig.6fortheAmatrice,VissoandNorcia earthquakes,forcomparison.Rightpanelsshowdip-misfitplots.Oneach,redindicatestheSW-dippingplane,bluetheNE-dippingplane.Waveformsandbest-fitsynthetics foreachearthquakeareshowninSupplementaryFigs. S1–S3.
Foreachofthesethreeearthquakes,weinvertteleseismiclong- periodbodywavesforthebest-fitfocalmechanism(Fig.2,Supple- mentaryFigs. S1–S3).Wetreateachearthquakeasafinite-duration point-source centroid, with a moment-release function parame- terised by a series of 1 s triangular elements. We invert P and S H waveformstodeterminethemoment-ratefunction,acentroid depth,totalmomentandafocalmechanism(strike,dip,rake),us- ingaleast-squaresapproach(e.g.seeWaltersetal.,2009).
WeestimateaseismologicalMwof6.2,6.1and6.6fortheAm- atrice,VissoandNorciaearthquakes respectively,andfindsimilar normal-faultingmechanismsonNNW–SSE strikingfaults foreach event. Our seismological results suggest shallow centroid depths of ∼4 km and relatively shallow dips for all three earthquakes (<∼45◦fortheVissoandNorciaearthquakes,<∼50◦fortheAma- triceearthquake;Fig.2).Thecentroiddepthsestimatedhereagree wellwithprevious seismological estimates(e.g.,Chiaraluce etal., 2017; Pizzi et al., 2017). All these seismological results are con- sistent inindicating that the majorityof momentrelease iscon- centrated within the upper ∼8 km of the crust, with centroids from3–6kmdepthforallthreeofthelargestevents.Comparison
between thefocal mechanismsestimated here, and the compos- ite focal mechanisms resultingfrom finite-fault inversionof near source data shows a good agreement in strike and dip, with a slight (∼10◦) difference in rake that likely results from collaps- ingadistributedpatternofslipacrossseveralfaultsintoasimple compositemechanism(Pizzietal.,2017).
3. Fieldmeasurementofsurfaceruptures
Thenormalfaulting mechanismfromourseismological results is consistent with the NE–SW extension that characterises the Apennine mountain belt (D’Agostino et al., 2011, Fig. 1b). The earthquakesoccurredintheregionoftheSW-dippingLagaandMt.
Vettore(hereafterreferred to as‘Vettore’) normalfaults.Ofthese twomajorfaults,theLagafaultwasthoughttohavelastruptured in a major earthquake in 1639 A.D. (Rovida et al., 2016), whilst theVettorefaultwasonlyknowntobeactivefrompalaeoseismic investigations(GaladiniandGalli,2003).
Followingeach ofthethreemajorevents,we startedmapping thesurface rupturesthedayaftertheearthquake,contributingto
the compilation of measurements collated by the openEMERGEO internationalworkinggroup(Civicoetal.,2018).Measurementsof throw,heave,netslipandslipvectortogetherwithfaultstrikeand dip were collected in the region of the Vettore and Laga faults, using a handheld compass clinometer anda ruler. Forthe Ama- triceearthquake,measurements were collected overthe1 month following the earthquake. For the Visso earthquake, due to the shorttime betweenthisevent andthe subsequent Norciaearth- quake, the identification and mapping of the ruptures is likely incompleteandwewereonlyabletocollectmeasurementsofthe slip.
The Amatrice and Visso earthquakes each generated semi- continuoussurface ruptureswith∼10–20cmsliponpre-existing bedrock scarps, towards the southern and northern ends of the mappedVettorefaultrespectively(Figs.3,4).Incontrast,theNor- cia earthquake ruptured portions of the Vettore fault along its entire length, generating offsets up to 2.3 m along the central portionsofthe fault,andre-rupturing someofthe samesections that failed in the earlier earthquakes (Figs. 3, 4). These sections re-rupturedwiththesamekinematicsbutapproximatelyan order ofmagnitude greater slip. During the Norcia earthquake,numer- ous smaller faults in the hangingwall of the Vettore fault were alsoactivated,includingmetre-scaledisplacementofanantithetic structure∼2kmSWofthemainfault (Fig.4c).Thesefield mea- surementsareconsistentwithourseismologicalsolutionsandthe regionalextensiondirection.
4. Geodeticestimationofslipinthesequence 4.1. Geodeticdatasets
We measurethe surface displacements during theseismic se- quence with eighteen separate geodetic datasets: one regional GNSS dataset for each of the three earthquakes (INGV, 2016a, 2016b),oneshort-baselineGNSSdatasetfortheNorciaearthquake (Wilkinson et al., 2017) and fourteen InSAR datasets from the Sentinel-1 andALOS-2 satellites,each constraining the coseismic displacement fields from one or more of the three earthquakes (Fig.5g,SupplementaryFig. S4).
We processed Sentinel-1 Synthetic Aperture Radar interfero- grams using the GAMMA software package (http://www.gamma- rs.ch) andunwrappedthem usingthe MCF algorithm(Costantini, 1998). ALOS-2 interferograms were generated using the JPL/Cal- tech/Stanford ISCE package (https://winsar.unavco.org/isce.html) and unwrapped using the SNAPHU method (Chen and Zebker, 2002). Orbital effects were corrected using precise orbits from the European and Japanese Space Agencies respectively, and to- pographic effects were removed using 1-arcsec topographic data from the Shuttle Radar Topographic Mission (Farr et al., 2007).
Unwrapping errors were manually checked and corrected. Inter- ferograms were resampled in preparation for modelling using a nesteduniform sampling approach (e.g. Floyd etal., 2016), with higherdensity in thenearfield and lower density inthe farfield, to obtain about 1500 line-of-sight data-points per interferogram (SupplementaryFig. S4).
4.2. Modelfaultgeometry
In order to relate our geodetic measurements of surface dis- placement to slip on faults in the subsurface, we first define a simplified array ofnine rectangular modelfaults (Fig. 1a, c,Sup- plementaryTable1). Thislevel ofcomplexity insource geometry is commonly required for geodetic modelling of multi-segment, moderate-magnitude events like the Norcia earthquake (e.g. the S. Napa,CaliforniaandDarfield,NZearthquakes;Floydetal.,2016;
Elliott et al., 2012), and requires that fault geometries are fixed
prior to inversion, often on the basis of additional geological or geophysicalconstraints.FortheCentralItalyseismicsequence,we use a wealth of such additional information to define fault ge- ometriesatthesurfaceandatdepth,including:1)discontinuities and low-coherence regions in our InSAR data (e.g. Fig. 5a, c,e), which are commonly indicative of surface or near-surface fault- ing; 2) our field mapping of surface ruptures (Fig. 3; Civico et al., 2018);3) relocatedaftershock clouds(Chiaraluce etal., 2017;
e.g. Supplementary Fig. S5) and;4) our body-wavefocal mecha- nisms.
Ourmodelgeometryprimarilyconsistsoffourmajorfaultseg- ments:threesegmentsforthemainVettorefault,andoneforthe northern Laga fault. The locations, strikes and dips ofthese four faultsarewellconstrainedbytheabovedatasets.
We placethree additionalminorfaultswithin thehangingwall of the Vettore fault; one antithetic and two synthetic (Fig. 1c).
These minor faults are necessary to explain important near-fault complexity both in the geodetic displacements for the Norcia earthquake (e.g.theregion betweenthecentral Vettorefaultand the minor antithetic fault in Fig. 5c) and the complex array of decimetricsurface ruptureswe mappedinthefield (Fig.3). Tests showingthe increasedlocalmisfitto thesedatawhen theminor faults are each removed from the model are shown in Supple- mentaryFigs. S13–S24andsummarisedinSupplementaryTable2.
Wenote thatwhilstthegeodeticandfield dataconstrainthesur- face locationofthesethreefaults andslipintheshallowsubsur- face,thegeodeticdataareinsensitivetotheirgeometryatdepths greaterthan1–2kmduetostrongtrade-offswithsliponthemain Vettorefault.
Twomorefaultsrepresent:a12-kmlongENEdippingstructure antithetic to the Vettore fault that we call the Norcia Antithetic fault; andaNE–SW striking14-km longnormalfault wecall the Pian Piccolo fault that cross-cuts the Castelluccio plain between the Vettorefault andNorciafault systemto theSW (Figs.1,3c).
Theintersectionsbetweenmodelfaultsatdeptharesupportedby alignments ofrelocated aftershocks(fromChiaraluce etal., 2017) when projectedontoourmodelfault planes(Figs.6,7). ThePian Piccolofaultisrequiredtoexplain astrongNE–SWalignedsignal intheInSAR datathatcoversthe Norciaearthquake(e.g.Fig.5e).
In addition,thelocation andstrikeofthisfault are supportedby thegeomorphologyoftheCastelluccio plain(Fig.3d),previousge- ologicalmapping(ColtortiandFarabollini,1995) andby relocated aftershocks (Fig.7), thelatterofwhich alsoconstrains thedipat
∼40◦.Testsremoving thisstructure requiremajor (>2 m)slipin theNorciaearthquakeontheVettorefaultatdepthsgreater than 10km,makingthegeodeticcentroiddepthincompatible withthe centroiddepthsobtainedfrombody-waveseismology(Fig.2,Sup- plementary Figs. S13,S15). The NorciaAntitheticfault isstrongly delineated in the aftershock data (Chiaraluce etal., 2017) so we includethisstructureinourmodelgeometry,withthisfaulttrun- catedatits southernendby thePianPiccolofault. Thisstructure islesswell constrainedby thegeodetic datathantheother eight faults, andwe ranseveraltestswith:the PianPiccolofault trun- catedinsteadbytheNorciaAntitheticfault;thetwofaultscrossing and neither truncating the other;and theNorcia Antithetic fault removedcompletely.WhilstourpreferredgeometryfortheNorcia Antitheticfault(Fig.1c)doesimprovethefittoboththeInSARand GNSSdatafortheNorciaearthquake,thesealternativegeometries only resultinmarginallyhigher misfittothe data.However, itis important to note that inall of thesealternative geometries, the distributionofslipontheVettore–Lagafaultsystem,andtherefore alsothemajorfindingsofthisstudy,remainthesame.
R.J. Walters et al. / Earth and Planetary Science Letters 500 (2018) 1–14 5
Fig. 3. Surfacefaultingandfieldmapping.(a)OverlappingslipmeasuredonsurfacerupturesoftheM6.1VissoearthquakeandtheM6.6Norciaearthquake.Duetotheshort timebetweenearthquakes,fieldmappingofthiseventislikelyincomplete.(b)Overlappingthrow,heave,strikeandslip-vectorazimuthoftheM6.2Amatriceearthquake andtheM6.6Norciaearthquake.(c)Fieldmappingofthe2016centralItalyearthquakesequenceshowingthelocationsofallrecordedsurfaceruptures(Civicoetal.,2018) fromallthreemajorearthquakesandthelocationsofthesimplifiedfaultsusedintheinversionofgeodetic.Base-mapshowssunshadeddigitaltopography(Tarquinietal., 2007).DarkgreydashedlinewithtrianglesshowsthesurfacetraceoftheSibillinithrust,andblack-dashedlineshowsthesurfaceprojectionoftheinferredeast-dipping faultfromFig.7.(d)Zoom-inoftherhomb-shapedPianoGrandebasin(highlightedwithdashedpolygon),includingbreaks-in-slopeusedtoinferactivefaulting.(e)Oblique trendsinseismicity.RelocatedaftershocksfromChiaraluceetal. (2017) arecolouredbydepth,andshowlinearalignmentscloserto015◦thanto040◦.Blackboxshowsthe extentsof(c).
