1

Modelling of seismic guided waves at the Dead Sea Transform

ChristianHaberland�, Amotz Agnon

�, RadwanEl-Kelani

�, Nils Maercklin

�, Issam

Qabbani�, Georg Rumpker

�, TrondRyberg

�, FrankScherbaum

�, andMichaelWeber

�

Shorttitle: GUIDED WAVESAT THE DEAD SEATRANSFORM

�GeoForschungsZentrumPotsdam,Germany

�Hebrew University, Jerusalem,Israel

�An-NajahNationalUniversity, Nablus,PalestineTerritories

�NaturalResourcesAuthority, Amman,Jordan

�Instituteof Geosciences,UniversityPotsdam,Germany

2

Abstract.

On severalrecordingsof linearseismometerarrayscrossingtheAravaFault (AF) in

theMiddle East,we seeprominentwave trainsemerging from in-fault explosionswhich

we interpretaswavesbeingguidedby a fault-zonerelatedlow-velocity layer. TheAF

is locatedin theArava Valley andis consideredtheprincipalactive fault of themainly

N-S striking DeadSeaTransformSystemin this section.Observationsof thesewave

trainsareconfinedto certainsegmentsof thereceiver linesandoccuronly for particular

shotlocations.They exhibit largeamplitudesandarealmostmonochromatic.We model

themby a 2D analyticalsolutionfor the scalarwavefield in modelswith a vertical

waveguideembeddedin 2 quarterspaces.A hybrid searchschemecombininggenetic

algorithmanda local randomsearchis employedto explorethemultimodalparameter

space.Resolutionis investigatedby synthetictests.Theobservationsareadequatelyfit

by modelswith a narrow, only 3 to 12 m wide waveguidewith S-wave velocity reduced

by 10 to 60%of thesurroundingrock. We relatethis verticallow-velocity layerwith the

damagezoneof theAF sincethelocationof receiversobservingandof shotsgenerating

theguidedwaves,respectively, matchwith thesurfacetraceof thefault. Thethicknessof

thedamagezoneof theAF, at leastat shallow depths,seemsto bemuchsmallerthanin

othermajorfault zones.Thiscouldbedueto lesstotal slip on this fault.

3

1. Introduction

In thelast few yearstheanalysisof trappedseismicwaveshasbecomea powerful

tool to studythe structureof lithosphericshearzones[e.g.,Houghet al., 1994;Li

and Leary, 1990;Li et al., 1998,1999,1994;Malin et al., 1996;Lou et al., 1997].

Similar to light beingtrappedandvery efficiently guidedby a glassfibre (owing to its

higherrefractive index), seismicwavescaughtby a low-velocity layer cantravel as

characteristicwave trainsover long distances.Due to strongvelocity contraststhey

arefrequentlyobserved in coal seams[e.g.,DresenandRuter, 1994],oil-reservoirs

[Chonet al., 1996],andalsoin tectonicallydominatedsettingssuchassubductionzones

(low-velocityoceaniccrustdescendinginto theuppermantle[Fukaoet al., 1983;Abers,

2000]). In lithosphericfault zones,thestructuraldamagezonerelatedto thedeformation

is consideredto form a low-velocity layerin whichguidedwavescandevelop.

Becausetrappedwavespropagatein theselayersfor considerableportionsof the

propagationpath,they containvaluableinformationon propertiesof thenarrow zones

itself, namelygeometry(connectivity, width, spatialorientation)andphysicalproperties

(velocitiesandattenuation).Dueto thenarrownessof thestructureson theonehand

andthe integrative characterof seismictraveltimesandattenuationon theotherhand,

this informationis usuallynot obtainedby conventionalseismicinvestigations(seismic

tomography, refraction).

The fault zoneguidedwaves(FZGW) studiesup to now includedthe pure

identificationandgeometricalmappingof waveguides,exemplarystudiesof principal

wave propagationin 2D [Ben-ZionandAki, 1990;Ben-Zion, 1998]or 3D structures

4

[Igel et al., 1997;Li andVidale, 1996;Huanget al., 1995; Igel et al., 2001;Jahnke

et al., 2002],andthemodellingof observedtrappedwaveforms[e.g.Li et al., 2000].For

theanalysisbothearthquakesandexplosions[Li et al., 1997b,1998,1999]areusedas

sources.

Most quantitative studiesattemptto derive subsurfacemodelsby waveformor

dispersioncurve matchingof guidedwave observationswith syntheticdata.Many of

themsearchthemodelspacesystematically, andsomestudiesquantify thegoodness-

of-fit of themodelsandpresentequivalentmodels[e.g.Penget al., 2000]. In orderto

accountfor thecomplex influenceof certainparameterson theappearanceof guided

waves,Michaeland Ben-Zion[1998b] proposedthe useof geneticalgorithm(GA)

to searchthe large parameterspace.GA is a very robust global searchalgorithm

[e.g. Goldberg, 1989] that proved to be useful in waveform fitting [Sambridge and

Drijkoningen, 1992;SenandStoffa, 1992;LomaxandSnieder, 1995;Levin andPark,

1997].

In this studywe presentdataof explosiongeneratedhigh-frequency guidedwaves

at theDeadSeaTransform(DST),a 1000km long,prominentshearzonein theMiddle

Eastexhibiting a total slip of � 100 km during the last 20 Myr. Theanalysisis part

of the interdisciplinaryresearcheffort DESERT (Dead Sea Rift Transect)in which

severalgeophysicalmethods(reflectionandrefractionseismics,gravimetry, magnetics,

electromagnetics)togetherwith geologicalstudiesandmodellingaim to resolve the

structureanddynamicsof thetransformat differentscales.Backboneof thestudyof the

small-scalefault structurewas2D and3D controlledsourcehigh-resolutiontomography

andreflectionseismics.In orderto derive subsurfacemodelsthat bestexplain the

5

observations,we modeltheguidedwavesby ananalyticalsolution[Ben-ZionandAki,

1990]andsearchthemodelspaceusinga hybrid schemecombiningGA anda local

randomsearch.

2. Geological setting

TheSSW-NNE striking DST separatestheArabianplatefrom theSinaimicroplate

(seeFigure1). It stretchesfor approximately1000km from theRedSeaRift to the

Taurus-Zagroscollision zone.Formedin theMioceneandrelatedto thebrake up of the

Afro-Arabiancontinentit accommodatesthelateralmovementbetweenthetwo plates.

The total amountof left-lateraldisplacementis � 100 km, recentrelative motion is

between3 and4 mm yr � � [Klinger etal., 2000]. Figure 1.

BetweentheGulf of Aqaba/EilatandtheDeadSea,thesinistralstrike-slipArava

Fault (AF) constitutesthemajorbranchof theDST [Atallah, 1992;Garfunkel et al.,

1981], taking mostof the slip. In the centralArava Valley the (straight)fault trace

(striking 15 - 20�E) is outlinedby scarps,pressureridges,small rhombgrabens,and

waterholes.Recentactivity is indicatedby offsetgulliesandalluvial fans[Klinger etal.,

2000]. Four stronghistoricearthquakesreportedlyhit theAF: 1068A.D., 1212A.D.,

1293A.D., and1458A.D. [Ambraseyset al., 1994;Amit et al., 2002;Klinger et al.,

2000].Theseeventswerecorroboratedin sedimentaryrecords[Ken-Tor etal., 2001].By

contrast,thecurrentseismicactivity alongthesouthernsectionof theDST is rathersmall

[Salamonet al., 1996].Theregion eastof theAF is segmentedinto numerousblocksby

additional(W-E andNW-SEstriking) faults(someof themnormalfaults). In thestudy

area,Neogenemarl, Cretaceouslimestone,andMioceneconglomerates(in thenorth

6

Precambriangraniteoutcrops)arein partcoveredby youngalluviumandaeoliansands.

3. Data and observations

In a specificallydesignedcontrolledsourceexperimentwe deployed5 seismiclines

crossingtheAF at about30�30’N (seeFigure1). Theroughlywest-eaststriking lines

1, 2 and3, which wereapproximately5 km apart,hadlengthsof about9 km. Spacing

betweenthe4.5 Hz geophonegroups(SM-6, verticalcomponent)on lines1, 2, and3

was100m. On theselong linessignalswererecordedby a SUMMIT datalogger(line 1,

250samplespersecond)andPDAS-100dataloggers(lines2 and3, 200Hz persecond).

Lines4 and5 wereshorter(ca. 200m) andhadreceiver spacingsof 10 m. They were

equippedwith REFTEK dataloggersrunningwith 200samplesper secondandwith

Mark L4-C-3D 1 Hz threecomponentseismometers[DESERTTeam, 2000;Maercklin

et al., 2000].

The fault traceis clearly visible in satelliteimagesover long portions,andit

is feasibleto identify it in the field at certainsegmentswith an accuracy of better

than100m. However, to enhancetheprobabilityof hitting the fault (andto generate

guidedwaves)we placed4 groupsof 3 individual shotseachat locationswherewe had

indicationsof the fault tracefrom geologicalsetting,satelliteimages,or topography.

Distancebetweenshotswithin a groupwasbetween20 and50 m. At eachshotpoint

45 kg of chemicalexplosivesweredetonatedin 20 m deepboreholes.Dueto theuseof

differentialGPS,all positionsandheightsof shotsandreceiverscouldbedetermined

with anaccuracy of � 1 m.

Weobserveprominenthigh-amplitude,high-frequency wave trainson receiver lines

7

2, 3, 4, and5 for two shots,101and102. Theobservationson line 4 and5 areshown

in�

Figure2. They arebestdevelopedon line 4 for shots101and102,andon line 5 for

shot101. In generaltheobservationsshow a seriesof characteristics,giving evidenceof

beingguidedwaves.Thesewavesoccuronly at certainreceiversfor certainshots.The

correspondingreceiversareconfinedto narrow sectionsof thelines(for positionsrefer

to Figure1 andTables1 and2). Shots101and102andtheobservingreceiverson line 4

andline 5 match(within thegivenaccuracy of independentgeologicalinformation)with

thesurfacetraceof theAF. All othershotsdid notgeneratesuchphases.As indicatedby

barsin Figure2 theguidedwavesshow very high amplitudes,which areup to 10 times

larger thanthesignalsat similar timeson otherreceivers.Offsetsbetweensourcesand

receiverswere2.3 (line 4) and1.1 km (line 5). Guidedwavesobservedat line 5 show

higherfrequenciesandashorterduration(ca.0.2s insteadof 0.3s)ason line 4, which is

expectedfor shorteroffsets. Table 1.

Table 2.

Figure 2.Comparedto otherreportedFZGW theguidedwavesat theAF show ratherhigh

frequencies(between20 and50 Hz). However, thedominantfrequenciesarecontrolled

by thegeometryandphysicalproperties,andsimilar high-frequency guidedwavesare

known for examplefrom coalseams[e.g.DresenandRuter, 1994]. As clearlyvisible

in thetime series,thewave trainsherearealmostmonochromatic.Lines4 and5 were

equippedwith 3 componentsensorswhich allowedthedeterminationof thepolarization.

Thephasesareverticallypolarized,thusonly verticalcomponentdatais shown in Figure

2. This factsuggeststhatthesephasesareLove-typechannelwavesasdescribede.g. in

DresenandRuter [1994].

8

4. Modelling and Inversion

Although influencedby many factors,guidedwavesareprimarily controlledby

a 2D structure:a waveguideof ’some’extentwith a reducedseismicvelocity, anda

sourcewithin or closeto thewaveguide.Factorsinfluencingdevelopmentandshapeof

guidedwavesaretheamountof velocity reductionwithin thewaveguide,thewidth of

thewaveguide,attenuation,andthepropagationlength.Furthermore,thedevelopmentof

guidedwavesdependson thepositionof theseismicsourcewithin thewaveguide.3D

variationsof the2D waveguidestructure,which mayconcernthewaveguidegeometry

(flexure,branching,bending,interruption,shape,heterogeneities)aswell asphysical

propertiesof bothhostrock andwaveguide(velocity, gradients),arealsoknown to alter

theappearanceof guidedwaves[Jahnke et al., 2002;Li andVidale, 1996;Igel et al.,

1997,2001].

In additionto puremappingof thewaveguidewe intendto derive quantitatively

someof themodel-describingparametersfrom theobservedwaveforms.Accordingly,

wecalculatesyntheticseismogramsfor variousmodels,comparethemwith theobserved

waveformsandinfer themostprobablemodelfrom thebestfitting dataset. Bearing

in mind well-known trade-offs amongcertainparameters[Ben-Zion, 1998] andthe

limitation of our dataset,we restrictourselvesto 2D models.This allows theestimation

of effectiveparametersaveragedover thewholepropagationlength.

We calculatedsyntheticswith a 2D analyticalsolutionfor thepropagationof waves

generatedby a � line sourcein a single layer �� �� which is embeddedwithin

two quarterspaces�� �� and �� �� [Ben-ZionandAki, 1990;Ben-Zion, 1998]. The

9

modelsaredefinedby thewidth of thewaveguide � andseismicvelocities ��� , seismic

attenuationfactors � � ���� � anddensity � of all threelayers. Sincewe intendto model

Love-type(shear)guidedwaves, � � areshearvelocities(i.e., � ��� � ). As in Ben-Zion

[1998] attenuationwasincorporatedfollowing Aki andRichards [1980]. Thereceiver

canbeplacedanywherein themodel,andthesourceanywherewithin thesinglelayer.

Thechosendepthof thesourceis usedasa proxy for theoverall propagationlength.

Syntheticvelocity seismogramswereband-passfiltered(1 or 4.5 to 90 Hz, respectively)

accordingto instrumentresponseandsamplingrate.

Thegoodness-of-fitbetweenobservedandsyntheticwaveformswasestimatedby

usingtheweightedmeanof thesemblance[NeidellandTaner, 1971]of windowsof 0.4s

containingtheprospective guidedwave train of eachobservedandsynthetictracepair

(for a givenparametercombination).We invertedsimultaneouslyfor up to 11 tracesof

a shotgather. Theuseof semblanceandthe inversionfor many tracessimultaneously

considerthedistribution of guidedwave energy acrossthewaveguideandin thehost

rock (leaking)asa majorattributeof theobservations(in additionto theamplitudeand

frequency behaviour of thewaveforms),thusstabilizingtheinversion.SeeAppendixA

for detailsof theappliedobjective function.

To efficiently searchthis multimodalparameterspaceandto derive anacceptable

model(or models)we employ a geneticalgorithm(GA). Imitating biologicalevolution,

GA is considereda sturdyoptimizationtechniquewith thepotentialof finding global

extremaandto follow-up multiple maxima[e.g. Goldberg, 1989]. It proved very

successfulin many optimizationproblems[e.g.Carroll, 1996;GibsonandCharbonneau,

1998],especiallyin seismic(waveform)inversion[SambridgeandDrijkoningen, 1992;

10

SenandStoffa, 1992;LomaxandSnieder, 1995;Levin andPark, 1997].SeeAppendixB

for!

detailsof basicGA.

As in othergeophysicalapplications(with noisepresent)we areinterestedin

(ideally: all) modelsequallywell explainingthedata(within a givenaccuracy), which

would allow thespecificationof valid parameterbounds.Accordingly, we tunedtheGA

for a broadsamplingof theparameterspaceratherthanfor fastconvergence:we used

a largepopulationsizeof 50, appliedniching(by phenotypesharing)[seefor example

Goldberg, 1989]andturnedoff theforcedreproductionof thebestmodel(no elitism).

Furthermore,we combinedGA andconsecutive local randomsearchesaroundthebest

individualsof eachgenerationforming a hybrid scheme.Thepurposeof this phenotype

manipulationis two fold: Assumingoptimain theobjective functionstretchingover

severalneighboringparameterbins,thereis a goodchanceof finding bettermodelsin

thedirectvicinity of analreadyspottedgoodmodel. Thesecondbenefitconcernsthe

enhancedexplorationof regionswith large(however, potentiallysub-optimal)valuesof

theobjective function(i.e. well fitting models).In eachgenerationwetestfour randomly

chosenmodels(that is 8% of thepopulation)in thevicinity (within a region of 5% of

theallowedparameterbounds)of 10%of thebestfitting models,subsequentlyreplacing

less-fittingindividualswith betterones.

5. Resolution

In orderto studytheperformanceof thesearchandto exploretheresolutionof the

obtainedmodelswe conductedsynthetictests.Dataweregeneratedfor thesamesource

andreceiver geometry(samestationspacingetc.) asin our experiment,andparameter

11

settingsasshown in Table3. We addedrandomtime shiftsof at most " 0.005s and

random# noise(maximumsignal-to-noiseratio of 5) to thesynthetics,which we then

invertedin thesamewayastherealdata.Synthetictimeseriesareshown in Figure3. Table 3.

Figure 3.We invertedfor 4 freeparameters:waveguidewidth � , velocity ratio � ��� �%$ � ��� � , host

rock velocities � �&� � and � �&� � (which we setidenticalexceptfor a smallperturbationto

avoid unrealisticsymmetryeffectsin thesynthetics)andanoverall attenuationfactor

( �'� �� (� � �� �� � �� (� � �� ). The lateralsourcepositionwassetfix to oneinterface

betweenwaveguideandhostrock. Furthermore,thepositionof thereceiver observing

thelargestguidedwaveamplitudewasfixedto thecentreof thewaveguide.

Figure4 A - D shows valuesof thefitting function( ) ) of all in theGA run tested

modelsasa functionof thefour freeparameters.In our implementationof GA search

we ran 300 generations(approximately15,000testedmodels)until we achieved a

satisfactoryfit andabroadsamplingof theparameterspace.As expected,theGA search

revealsasuiteof modelswith anacceptablefit ( )+*-,/.1032 ). Thebesttestedmodelreaches

)4 +,/.1065 , which is thesamevalueasthevalueassociatedwith theoriginal synthetic

model(however, it is not thesamemodel).Thedistributionof acceptableF-valuesgives

estimateson the resolutionof the particularparameter. The thresholdof acceptable

F-valueswasdeterminedby visual inspectionof the fit for somemodels. Table3

summarizestheserangesof theindividualparametersandthebestretrievedmodel.The

parameterof thesyntheticinput model(blackstarsin Figure4 A-E) plot well within

theseallowed-parameterlimits indicatingthat thesyntheticmodelis well recovered.

However, for particularparameters(i.e. � �&� �%$ � ��� � or � � �� ) theselimits areratherlarge

suggestingassociatedlow resolutionand/orstrongtrade-offs. Largestsemblancevalues

12

of individual tracesreach� 7,/.8539 . Figure 4.

An alternative representationof theacceptablemodelsanda crossplotshowing all

well fitting modelsasa functionof � and � ��� �%$ � ��� � , unveils a strongtrade-off between

thesetwo parameters(Figure4 E andF). Thewaveformscanbeequallywell fitted by

certainmodelswith narrow waveguidesandlargevelocitycontrastsor wider waveguides

andsmallervelocity contrasts.All otherparametercombinationsexhibit no such

dependenciesbut ratherrandomscatterin our synthetictest. Independentconstraining

informationononeof thesetwo parameters(for examplefrom independentseismological

or geologicalstudies)coulddrasticallyreduceambiguities.It is noteworthy thatgentle

variationsof � imply (petrophysically)significantvariationsof � �&� �%$ � ��� � .In therecoverytestpresentedaboveweinvertedfor four freeparameters,andseveral

parameterswerenot consideredin theinversion.Furthertestswith anindividual �'� �� of

thewaveguide(both in thesyntheticmodelandasa free inversionparameter)showed

thatthis parameteris not resolved(for thegeneralmodelandobservationgeometryused

in our test). For syntheticmodelsexhibiting anelevatedwaveguide-� � �� (what is for

examplediscussedfor damagezonesof faults)theGA searchfoundevenwell fitting

modelswith reducedwaveguide-�'� �� .

Furthermore,thelateralsourceposition :<; within thewaveguide,which generally

influencestheshapeanddevelopmentof guidedwaves,wasnot includedin thesearch.

Sourcesoffsetfrom thewaveguideinterfaceresultin generallylesstrappedwaveenergy

andproducea characteristichigh-frequency phasesuperposedon the low-frequency

guidedwave mode[Ben-Zion, 1998,Figure5 (left)]. Sincethelatter is not observedin

theseismogramswe skippedthis parameterin the inversion,reducingcomputational

13

burdenandavoiding furtherambiguities.

6. Results

Basedon theexperiencewith thesyntheticdataset,we invertedtheobserveddata

for thesame4 parameters.Figures5 through7 show thebestfitting modelsthatresulted

from theGA searchesof datafrom shots101and102observedalongreceiver line 4,

andfrom shot101observedalongreceiver line 5. Dueto weaksignal,dataof shot102

observedon line 5 werenotassessedquantitatively. Figure 5.

Figure 6.

Figure 7.

The bestmodelsreachfitting valuesaround )= >,/.@?62 . However, singletraces

reachfrequentlymuchhighersemblancevalues( A�B,C.10D2 ). Table4 summarizesall

derivedparametersandpermittedparameterlimits. Theestimateof accuracy wasderived

from thedistribution of all modelswith F-valueslarger than98%of the ) -valueof the

bestmodel.Remarkableis thenarrow waveguidewidth of between3 and30 m, which

appearsto bewell resolved. Thedensityplot (Figure8) shows thatmostwell fitting

models(for observationson line 4) arecharacterizedby a4 to 6 m widewaveguide.Very

similar to thesynthetictest,we noticea pronouncedtrade-off between� and � ��� �E$ � �&� �(Figures5 through7 E). Althoughlessresolved, � � is for all measurementslarger than

50. Table 4.

Figure 8.Theresultsof bothshotsfor line 4 andalsobetweenthetwo linesareveryconsistent.

For shot102many goodmodelswith a larger � ��� � above1800m/sareobserved.For the

observationsalongline 5 we derivea slightly smaller � ��� � which wouldbein accordance

with thesmallersource-receiver offsetand,in turn,a smallerpenetrationdepth(smaller

velocities). It seemsthat the observationson line 5 arebetterexplainedby smaller

14

velocitycontrasts(above0.6).Nevertheless,weshouldkeepin mind thattheobservation

alongline 5 have only poorsignal-to-noiseratio. Figure9 shows thesynthetictraces

calculatedwith arbitrary”best” modelsoverlaidon theobserved time seriesof shots

101and102observedat line 4, andof shot101observedat line 5, showing thegood

agreementbetweenobservationsandsynthetics. Figure 9.

Our syntheticsarecalculatedfor a � linesource.In orderto checkthewarrantyof

this assumptionwe appliedanapproximate2D to 3D transformationto thesynthetics

for someof theinversionsby convolving thetime serieswith � $GF H with H beingthetime

[Vidale et al., 1985;Igel et al., 2001;Ben-Zionet al., 2002]. However, we noteonly

minor changesin thewell fitting modelsandtheestimatesof valid parameterranges.

On line 3 we observedsimilar high-frequency wavetrains(not shown) for shots101

and102on a ratherbroad( �JID,6,LK ) receiver line segmentslightly offset from theAF

(seeFigure1). Theseobservationsaremostlikely associatedwith a SW-NE running

fault branch.However, we couldnot modeltheseobservationssatisfactorily with 2D

analyticalsynthetics.In principal, variationsof our simplemodelgeometrysuchas

low-velocity layerson top of a verticalwaveguideor wideningof thewaveguidetoward

thesurfacecouldexplain a spatiallybroaderradiationof high-frequency guidedwave

energy (generalstudiesof theseeffectsarediscussedin Jahnke et al. [2002], Igel et al.

[2001], Igel et al. [1997], andLi andVidale [1996]). In fact,shallow high resolution

tomographicimagesof the studyarearevealeda numberof superficiallow-velocity

featuresassociatedwith sanddunesandmarl [Maercklin et al., 2000]. However, the

requirednumerical(3D) modellingof theseobservationsis beyond the scopeof the

presentpaperandwill bepartof futureresearch.

15

7. Discussion

We interpretthehigh-energy latearrivals in our seismicrecordingsaswavesthat

have beentrappedin a sub-vertical low-velocity channelassociatedwith theAF. The

strongestargumentsupportingthis interpretationis thefactthatwe observe thesewaves

only at distinct receivers,andthat thesewavesaregeneratedonly by certainsources.

Sowe canrule out puresourceor receiver siteeffectsasvalid explanations.Recently,

Rovelli et al. [2002] andBen-Zionet al. [2002] reportedon shallow low-velocity

structures(maybebetterreferredto asreceiver sitestructures)insteadof deepcoherent

waveguideswhich canefficiently trapseismicwaves.

Thefact,thatmostpositionsof thesourcesandreceiversgeneratingandobserving

guidedwaves,respectively, matchwith the surfacetraceof theAF, suggest,that the

waveguideis formedby the damagezoneof the fault. Studieson exhumedfaults

discriminatethreemajorentitieswithin themesoscopicstructureof brittle strike-slip

faults[e.g.Chesteret al., 1993;SchulzandEvans, 2000]: the (undeformed)protolith

(host-rock),thedamagezone,andthefaultcore.Thefaultcore(centimetresto decimetres

thick) with highly deformedrock is theplacewheremostof theslip is accommodated.

Thedamagezonesurroundingthecoreconsistsof anincreasedconcentrationof faults,

fissuresandveins. Largefault zonesmight form a complex network of faultsof many

sizes[WallaceandMorris, 1986].

It waspossibleto modelmostof theobservationsquantitatively. The inversion

especiallyof thesehigh-frequency phasesis highly non-uniqueandcanonly beapplied

undercertain(restrictive) assumptions(seediscussionon constraintsabove). The

16

uppermostvelocity structure(weatheringlayeretc.) cancauseseveredistortionsof the

highM

frequency waveforms,which cannotbeaccountedfor by our simple2D models.

This holdsalsofor general3D variationswhich couldbiasbothour estimatesof thebest

fitting modelsandof theacceptableranges.Moreover, wenotethattheobjectivefunction

asrevealedby thesystematicGA searchis characterizedby many broadandnarrow

local maximaof similar amplitudeandcomplex shape[seealsoMichaelandBen-Zion,

1998a].This meansthattheobserved(noisy)datacanbealmostequallywell explained

by many differentmodels[seee.g.MichaelandBen-Zion, 1998a]thusyielding in part

quitelargeconfidencelimits of certainparameters.Webelievethattheseambiguitiesand

limits arestronglydependentontheexperimentgeometry, predominantfrequencies,used

frequency band,etc. in relationto thewaveguidegeometryandproperties.Nevertheless,

theparametersearchesgavea narrow effective waveguidewidth of about3 to 30 m (for

theobservationson line 4). We favour a very narrow waveguide(3 to 12 m) sincein

theaccompanying high-resolutioncontrolledsourcetomographyexperimentwe found

no evidencefor a wider subverticallow-velocity structureat shallow depth.Moreover,

trappingefficiency is strongerfor largervelocitycontrasts,andtheseare,accordingto the

trade-off curves(Figures5 to 7 E), allowedfor narrow waveguides.Finally, thesearches

of all independentmeasurementsendedup with bestmodelsfeaturing � between3 and

12 m.

The apparentnarrownessof the AF waveguide(andthe deduceddamagezone

width) putsit in markedcontrastto othermajor fault zones,at which thewaveguide

width hadbeendeducedby FZGW. For example,Li et al. [1990] andLi et al. [1997a]

reportwidthsof waveguidesof theSanAndreasfault (SAF),California,of between100

17

and170m. In theNorth AnatolianFault System,Ben-Zionet al. [2002] derivedfault

zonewidthson theorderof 100m. Thelow-velocity zoneof theNojima fault, Japan,

haswidth of 30 - 60 m [Li et al., 1998]. In general,thedamagezonethicknessseems

to beproportionalto thetotal slip alonga fault [e.g.WallaceandMorris, 1986],with a

constantof proportionalityof between�N,O� � and �P,O� � [Scholz, 1987].

Althoughthewholesystemof theDSTexperienced� 100km of cumulativerelative

movementsincethelast20Myr, thereareindicationsthatseveralsub-parallelfaultswere

active duringthattime, thusdistributing thetotal movement.Eyal et al. [1981] showed

that in easternSinai thetransformsplaysto numerousparallelstrands,someof which

hadslippedseveralkilometresandareinactivenow. They haveestimatedthathalf of the

total 100km aredistributedin easternSinai. It is conceivablethattheremaining50 km

aredistributedbetweentheactive AF andparallelstrandssuchasZofar andor Quaira

faults.At thesurfaceandin satelliteimagesseveralsub-parallellineamentsarevisible in

thestudyarea,andit is possiblethatwe might have imagedonly oneof severalstrands.

However, dueto thegivensourceandreceiver spacingcoveringa largerangeof scales

(meterto kilometre),we rule out theexistenceof low-frequency guidedwaves,and,in

turn, largerwaveguidewidthsona largerscalefor thestudyarea.

Anotherreasonfor a narrow shallow damagezonecould lie in thepeculiaritiesof

thetwo geologicalunitsatbothsidesof thestudiedAF segment.While thedeeperpartof

theeasternblock is mainly formedby PrecambriangranitesandCretaceouslimestones,

thewesternunit (down to a depthof severalkilometres)is mainly formedby sediments

of theHazevaformationandDanaconglomerates,whichwere- in partsyn-kinematically

- depositedduring theMiocene. Also the easternunit is partly coveredby layersof

18

Hazeva formationstrataandyoungersediments.However, structuraldetailsarestill not

knoQ

wn to date.Owing to thefactthatwe sampledtheuppermostpartof theAF segment

by FZGW (correspondingbodyP wavespenetrateto approximately300m depth),we

mighthave imagedthesuperficialdamagezone(severalhundredsof metersdepth)of the

AF thatexperienced- dueto youngerageof the involvedshallow geologicalunitsand

syn-depositionalfaulting- a reducedtotal slip.

Theinversionprovidesvaluesfor theeffectiveshallow hostrockshearwavevelocity

of between1740and1850m/s.They areslightly smallerfor theshort-offsetobservation

at line 5 (1570to 1780m/s). Togetherwith theobservedP wavearrival timesthis yields

a realistic � R $ � � velocity ratio of theshallow rocksof about1.8. We noticea wide range

of possiblevelocity reductionwithin thewaveguideof about10 to 60%of thevelocity

of the hostrock. We think that smallerratiosaremorerealistic,sincethe trapping

efficiency is larger for thesevalues(seealsonoteabove). Similar large valueswere

obtainedfor examplein a conventionalseismicexperimentat theexhumedSanGregorio

Fault (California),wherebrecciaandfoliatedrock form thedamagezone[Gettemyetal.,

2001]. FZGW analysisat othermajorstrike-slipfaultssuggestedvelocity reductionsin

theorderof 25 to 50%relativeto thesurroundingrock [e.g.Li etal., 1997a,1994;Li and

Vernon, 2001].Accordingly, thereducedvelocitiescouldbeinterpretedby damagedhost

rock materialwithin thefault zone.Salinewatersarepresentat thesurfacetraceof the

AF (isolatedwaterholes),andwe speculatethataqueousfluids arealsopresentat greater

depthof thefault zone.Thesefluidscouldalsocontributeto thevelocity reductionof the

waveguide. Nevertheless,magnetotelluricstudiessuggestthat thefault planeis rather

impermeablesinceit separatesdifferently resistive domainsat differentdepthlevels

19

[Ritter et al., 2001].

Thesizeof thedamagezoneis determinedby a competitionbetweenlocalization

anddelocalizationprocesses.Theformerarerelatedto thebrittle instabilityof damagable

media[Lyakhovsky et al., 1997],whereasthe lattermayberelatedto the long term

dynamicsof the fault zone[Lyakhovsky et al., 2001]. During the organizationof a

systemof faultsto form a planarfault zone,somefault strandsdominateandtake over

theslip at theexpenseof othersub-parallelstrands.If therateof healingof distributed

damageis high comparedto the loadingrate,theorganizationis slow. Furthermore,if

healingsaturatesover theperiodof a seismiccycle, theseismicityis clustered,andno

characteristicearthquakedevelops.Geologicstudiesof theDST indeedsuggestongoing

organization[Rotsteinet al., 1992]. Thehistoricalseismicityis clusteredasapparent

from the quiescencesince1458,the yearof a fourth earthquake in four centuries.

Paleoseismicstudiesindicatefurtherclusteringof activity over periodsof thousands

[Marco andAgnon, 2003] to tensof thousandsyears[Marco et al., 1996;Amit et al.,

2002]. High resolutionarchaeoseismicreconstructionof two historicaldevastating

earthquakesshow strongvariability of slip, at oddswith thecharacteristicearthquake

model[Ellenblumet al., 1998]. By contrast,theSAF shows characteristicearthquake

behaviour [SiehandJahns, 1984]thatis consistentwith low healingto loadingratio. The

SAF zonedoesnot healcompletelyon thetime scaleof theseismiccycle,andruptures

tendto repeaton thesamesmoothtrend[Wesnousky, 1994;Stirling etal., 1996].TheAF

seemsto healover thelongerseismiccycle, sothetraceis not assmooth.It remainsto

beseenwhetherthenarrow gougezoneindicatedhereis a relic of thelastearthquake,a

productof ongoingcreep,or aprecursorof thenext one.

20

Theapparentnarrownessof thedamagezonegivesevidencethat thedeformation

alongthis largestrike-slipfault is confinedto a narrow region. This is furtherevidenced

by conventionalseismicexperiments.Thesestudiesrevealedablockedvelocitystructure

beneaththestudyareashowing theAF separatinga westernlow-velocity block andan

easternhigh-velocityblock by ratherlargehorizontalgradients,andshowedindications

for reflectionsfrom sub-vertical structures[Maercklin et al., 2000]. The localized

deformationalongtheDST asimagedby this studycouldsupportthesuppositionthat

theDSThasoriginatedin (andis cuttingthrough)a relatively coldandstablecontinental

lithosphereasproposedfor exampleby DESERTTeam[2002].

We revealeda waveguidealonga 4 km segmentof theAF. We werenot successful

in observingguidedwavesat largeroffsetsandfor 9 othershotsfiredwithin theassumed

fault trace.After all, this is not surprisingwhenkeepingin mind thenarrownessof the

waveguideandthegivenaccuracy of a priori informationonthefault trace.Furthermore,

low-Q in thenarrow waveguideattenuatesthehigh-frequency wavesquickly. However,

this fact andtheslight differencesbetweentheobservationson line 4 and5 suggest

a variablestructure(positionof waveguides,continuity, etc.) alongthe studiedAF

segment.

8. Conclusions

With a specificallydesignedactiveseismicexperimentin theArava Valley, Jordan,

we wereableto recordwave trainswhich weretrappedin a sub-vertical, low-velocity

waveguide. The locationsof sourcesgeneratingandreceiversobservingtheseswaves

correspondto thesurfacetraceof theArava Fault, implying that this waveguideis the

21

damagezoneof this prominentfault.

Themodellingof theguidedwaveswith a2D analyticalsolutionin conjunctionwith

thesystematicGA searchfor thebestmodels(includingproperevaluation)provedto be

successfulin obtainingimportantgeometricalandpetrophysicalparametersof theAF.

We deriveda rathernarrow shallow waveguidewidth of between3 and12 m. Although

lessconstrained,thereductionof thewaveguidevelocity relative to thehostrockvelocity

is in therangeof between10 and60%.

Thevelocity reductionwithin thewaveguidecouldbecausedby damagedrock

materialandsaturationwith fluids which characterizethe wearzoneof this fault

accumulatedover time. Accordingto scalinglaws, the apparentnarrownessof the

damagezone(closeto thesurface)pointstowardatotalslip of severalhundredsof meters

to several tensof kilometresalongtheAF fault at theshallow depthscale.Regarding

the total DST slip of about100 km, this canbe understoodconsideringa possible

distribution of theslip on severalfaultsin time andspace(QuairaandZofar faultsetc.).

Furthermore,rockscloseto thesurfaceeastandwestof theAF might havebeenformed

syn-kinematicallyand,asa result,experiencedlessrelativeslip.

Theverynarrow low-velocitystructurerevealedby this studycouldnot bedetected

directly by otherseismicexperimentscarriedout in the samearea(reflectionand

refraction).In orderto derive someof theparametersin similar experimentswith high

accuracy, wesuggestto usea largenumberof sourcesandreceivers(in-fault,cross-fault)

andto useabroadfrequency range(e.g.,highsamplerate).

22

Appendix A: Objective function

Our fitnessfunction )TSR , which we maximizein theinversion,is calculatedfrom a

weightedmeanof individual semblancevaluesU<SR � V [NeidellandTaner, 1971]of each

observedandsynthetictracepair W ( X is numberof traces)for a givensetof model

describingparametersYZ \[ Z �%] .^._. Za`cb ( d is numberof parameters):

)TSRe fgVih � � V UaSR � V . (A1)

IndividualsemblancesUaSR � V arecalculatedfrom all samplesj within thetimewindow

of interestof observedandsynthetictraces( k V and lGSR � V , respectively) of tracepair W :

UaSR � V m1n [ok V^� nqp l Vr� SR � n b �

� m1n [s[ok V^� n b � p [tl Vr� SR � n b � b . (A2)

Semblanceis 1 if tracesarecorrelated,0.5 if uncorrelated,and0 if anti-correlated.

Theweight � V increasestheinfluenceof traceswith highsignalenergy in theinversion:

� V m n [tk Vr� n b �fm� h � m1n [tku� � n b � . (A3)

In orderto accomplishsmalldeviationsof the true3D subsurfacestructurefrom

thesimple2D modelswe allowedtime shiftsof " 1 sample( " 2 samplefor dataof

line 5/shot101)betweenneighboringsynthetictraces.Beforecalculatingtheobjective

function,we applieda uniform tracescaleaccordingto themaximalsignalamplitudein

thecorrespondingshotgather. Calculationswerealsoconductedfor reversedsignof the

syntheticamplitudes.

23

Appendix B: Genetic Algorithm

Developedby JohnH. Holland, [e.g.Holland, 1975],geneticalgorithms(GAs)

searchamultimodalparameterspacefor (global)maximaby usingmechanicsof natural

selectionandnaturalgenetics[Goldberg, 1989]. Model describingparametersof a

particularproblemarefirst codedinto binarystrings.In a standardGA, a populationof

binarystrings(that is a randomlychosensetof models)is modifiedby threeoperators,

namelyreproduction(selection),crossover andmutation. In the reproductionstep

individual stringsarecopiedto thenext generationaccordingto their objective function

valuesin a sensethathighervalueshave higherprobabilitiesto becopied. In thenext

step,offspringsarecreatedfrom this matingpool by assemblingsubstringsof parent

binarystrings.Substringscanbecut at oneor morecrossoverpositions.Finally, random

mutationsof singlebits areperformedon thepopulationwith achosen(low) probability.

This procedureis appliedthroughmany generationsyielding a broadsamplingof

the parametersetwhile revealingoneor moremaxima. Sincethe inventionof GA,

modificationsof andadditionsto this principal GA outline formedpowerful search

algorithms[e.g.Goldberg, 1989]. We useda populationsizeof 50,mutationprobability

of 0.02,a (uniform) crossover probabilityof 0.5,anda creepmutationprobabilityof

0.01.For in-depthdiscussionof GA andthesettingspleasereferto Goldberg [1989] and

Carroll [1996].

24

Acknowledgments.

For thisanalysisweuseddataof theControlledSourceArray (CSA)experimentaspartof the

DeadSeaResearchProject(DESERT, http://www.gfz-potsdam.de/pb2/pb22/projects/deadsea/ds-

home.html). DESERT wasfinancedby the DeutscheForschungsgemeinschaft(DFG), the

GeoForschungsZentrumPotsdam(GFZ) andtheMinerva Foundation.TheNaturalResources

Authority, Jordan,provided substantiallogistical support.We thankK. Abu-Ayyashfor his

efforts. Furthermorewe thanktheSiteGroup(Jordan)andChemical& Mining (Jordan)for

their efforts with drilling andexplosives. We gratefulacknowledgethework of all field groups,

in particularR. Ribhy, K.H. Jaeckel, S. Grunewald andG. Haim. Thanksto J. Mechiefor

supervisingpart of thedrilling activities, andJ. Bribachfor technicalassistance.Thanksto

YossiBartov for fruitful discussionson thegeologyaroundthestudiedDST segment.We are

indebtedto theAssociateEditorandreviewersYehudaBen-ZionandJohnVidalefor theircritical

remarksandhelpful suggestionson themanuscript.We usedinstrumentsfrom theGeophysical

InstrumentPoolPotsdam(GFZ) andtheFR Geophysics(FreeUniversityof Berlin). All figures

weremadeusingthegenericmappingtool (GMT) [WesselandSmith, 1995,1998]. Thanksto

David L. Carroll for makinghis GA codeavailable.We took advantageof theCWP/SUseismic

dataprocessingpacket [CohenandStockwell Jr., 1999].

25

References

Abers,v

G. A., Hydratedsubductedcrustat 100-250km depth,Earth Planet.Sci.Lett., 176,

323–330,2000.

Aki, K., andP. G. Richards,QuantitativeSeismology, W.H. FreemanandCo.,USA, 1980.

Ambraseys, N., C. Melville, andR. Adams,TheSeismicityof Egypt,Arabia and theRedSea,

CambridgeUniversityPress,Cambridge,1994.

Amit, R., E. Zilberman,Y. Enzel,et al., Paleoseismicevidencefor time dependency of seismic

responseon a fault systemin thesouthernArava Valley, DeadSearift, Israel,Geol.Soc.

Am.Bull., 114, 192–206,2002.

Atallah,M., Tectonicevolution of thenorthernWadi Arava,Jordan,Tectonophysics, 204, 17–26,

1992.

Ben-Zion,Y., Propertiesof seismicfault zonewavesandtheir utility for imaginglow velocity

structures,J. Geophys.Res., 103, 12,567–12,585,1998.

Ben-Zion,Y., andK. Aki, Seismicradiationfrom an wyx line sourcein a laterallyheterogeneous

planarfault zone,Bull. Seismol.Soc.Am., 80, 971–994,1990.

Ben-Zion,Y., Z. Peng,D. Okaya,L. Seeber, J.G. Armbruster, N. Ozer, A. J.Michael,S.Baris,

andM. Aktar, A shallow fault zonestructureilluminatedby trappedwavesin the

Karadere-Duzcabranchof theNorth AnatolianFault,westernTurkey, Geophys.J. Int.,

2002,submitted.

Carroll,D. L., Chemicallasermodelingwith geneticalgorithm,AmericanInstituteof Aeronautics

andAstronauticsJournal, 34, 338–346,1996.

Chester, F. M., J.P. Evans,andR. L. Biegel, Internalstructureandweakeningmechanismsof the

SanAndreasFault,J. Geophys.Res., 98, 771–786,1993.

26

Chon,Y.-T., W. Turpening,S.Rutherford,andM. Parchman,Reservoir continuity loggingusing

connectivity mappingwhile drilling, TheLeadingEdge, March, 203–208,1996.

Cohen,J. K., andJ. W. StockwellJr., CWP/SU:SeismicUnix Release33: a freepackage for

seismicresearch andprocessing, Centerfor WavePhenomena,ColoradoSchoolof Mines,

1999.

DESERT Team,Multinationalgeoscientificeffort kicks off in theMiddle East,EOSTrans.AGU,

81, 609,616–617,2000.

DESERT Team,Structureanddynamicsof theDeadSeaTransform,Geology, 2002,submitted.

Dresen,L., andH. Ruter, SeismicCoal Exploration Part B: In-SeamSeismics, Elsevier Scientific

PublishingCompany, New York, 1994.

Ellenblum,R., S. Marco,A. Agnon,T. Rockwell,andA. Boas,Crusadercastletorn apartby

earthquake at dawn, Geology, 26, 303–306,1998.

Eyal,M., Y. Eyal,Y. Bartov, andG. Steinitz,Thetectonicdevelopmentof thewesternmargin of

theGulf of Elat Aqabarift, Tectonophysics, 79, 28 – 55,1981.

Fukao,Y., S. Hori, andM. Ukawa, A seismologicalconstrainton thedepthof basalt-eclogite

transitionin asubductingoceaniccrust,Nature, 303, 413–415,1983.

Garfunkel, Z., I. Zak, andR. Freund,Active faulting in theDeadSeaRift, Tectonophysics, 80,

1–26,1981.

Gettemy, G., H. Tobin, J. Hole, andA. Sayed,Towardexplaining scale-dependentvelocity

structureacrossanexposedbrittle fault zone,EOSTrans.AGU, 82, S41A–0583,2001.

Gibson,S.,andP. Charbonneau,Empiricalmodelingof thesolarcoronausinggeneticalgorithms,

J. Geophys.Res., 103, 14,511–14,521,1998.

27

Goldberg, D. E., Geneticalgorithmsin search, optimization,andmachine learning, Addison

Wesley, Boston,1989.

Holland,J.,Adaptionin natural andartificial systems, Ann Arbor: TheUniversityof Michigan

Press,1975.

Hough,S.E., Y. Ben-Zion,andP. C. Leary, Fault-zonewavesobserved at thesouthernJoshua

Treeearthquake rupturezone,Bull. Seismol.Soc.Am., 84, 761–767,1994.

Huang,B.-S.,T. Teng,andY. T. Yeh,Numericalmodelingof fault-zonetrappedwaves:Acoustic

case,Bull. Seismol.Soc.Am., 85, 1711–1717,1995.

Igel, H., Y. Ben-Zion,andP. C. Leary, Simulationof wyx�z and {|z}w�~�z wave propagationin

fault zones,Geophys.J. Int., 128, 533–546,1997.

Igel, H., G. Jahnke,andZ. Ben-Zion,Numericalsimulationof fault zoneguidedwaves:accuracy

and3-D effects,PureandAppliedGeophysics, 2001,in press.

Jahnke,G., H. Igel, andZ. Ben-Zion,Three-dimensionalcalculationsof seismicfault zonewaves

in modestlyirregularstructures,Geophys.J. Int., 2002,accepted.

Ken-Tor, R., A. Agnon,Y. Enzel,M. Stein,S. Marco,andJ. F. Negendank,High-resolution

geologicalrecordof historicearthquakesin theDeadSeabasin,J. Geophys.Res., 106,

2221–2234,2001.

Klinger, Y., J.Avouac,L. Dorbath,N. A. Karaki,andN. Tisnerat,Seismicbehaviour of theDead

Seafault alongArabavalley, Jordan,Geophys.J. Int., 142, 769–782,2000.

Levin, V., andJ. Park, P-SHconversionsin a flat-layeredmediumwith anisotropy of arbitrary

orientation,Geophys.J. Int., 131, 253–266,1997.

Li, Y.-G., andP. C. Leary, Fault zonetrappedwaves,Bull. Seismol.Soc.Am., 80, 1245–1271,

1990.

28

Li, Y.-G.,andF. Vernon,Characterizationof theSanJacintofault zonenearAnza,California,by

fault zonetrappedwaves,J. Geophys.Res., 106, 999–999,2001.

Li, Y.-G., andJ.E. Vidale,Low-velocity fault-zoneguidedwaves: Numericalinvestigationsof

trappingefficiency, Bull. Seismol.Soc.Am., 86, 371–378,1996.

Li, Y.-G.,P. Leary, K. Aki, andP. Malin, Seismictrappedmodesin theOroville andSanAndreas

fault zones,Science, 249, 763–766,1990.

Li, Y.-G., K. Aki, D. Adams,A. Hasemi,andW. H. Lee,Seismicguidedwavestrappedin

the fault zoneof the Landers,California, earthquake of 1992,J. Geophys.Res., 99,

11,705–11,722,1994.

Li, Y.-G., K. Aki, andF. Vernon,SanJacintofault zoneguidedwaves: A discriminationfor

recentlyactive fault strandsnearAnza,California,J. Geophys.Res., 102, 11,689–11,701,

1997a.

Li, Y.-G.,W. L. Ellsworth,C. H. Thurber, P. E. Malin, andK. Aki, Fault-zoneguidedwavesfrom

explosionsin theSanAndreasfault at ParkfieldandCienegaValley, California,Bull.

Seismol.Soc.Am., 87, 210–221,1997b.

Li, Y.-G.,K. Aki, J.E. Vidale,andM. G. Alvarez,A delineationof theNojima fault rupturedin

the � 7.2Kobe,Japan,earthquake of 1995usingfault zonetrappedwaves,J. Geophys.

Res., 103, 7247–7263,1998.

Li, Y.-G., K. Aki, J. E. Vidale,andF. Xu, Shallow structureof theLandersfault zonefrom

explosion-generatedtrappedwaves,J. Geophys.Res., 104, 20,257–20,275,1999.

Li, Y.-G.,J.E. Vidale,S.M. Day, F. Xu, E. Cochran,andD. Oglesby, Characterizationof rupture

zoneof theHectorMine andLanders,California,earthquakesusingfault-zoneguided

waves,EOSTrans.AGU, 81, F858–F859,2000.

29

Lomax,A., andR. Snieder, Thecontrastin uppermantleshear-wave velocity betweentheEast

EuropeanPlatformandtectonicEuropeobtainedwith geneticalgorithminversionof

Rayleigh-wave groupdispersion,Geophys.J. Int., 123, 169–182,1995.

Lou, M., J.A. Rial, andP. E. Malin, Modelingfault-zoneguidedwavesof microearthquakesin a

geothermalreservoir, Geophysics, 62, 1278–1284,1997.

Lyakhovsky, V., Y. Ben-Zion,andA. Agnon, Distributed damage,faulting, andfriction,

J. Geophys.Res., 102, 27,635–27,649,1997.

Lyakhovsky, V., Y. Ben-Zion,andA. Agnon, Earthquake cycle, fault zones,andseismicity

patternsin a rheologicallylayeredlithosphere,J. Geophys.Res., 106, 4103–4120,2001.

Maercklin,N., R. El-Kelani,C. Haberland,T. Ryberg, F. Scherbaum,M. Weber, andDESERT

2000Group,DESERT 2000- CSA: Seismicstudiesof theArava fault, DeadSeaRift

Transform,Jordan,EOSTrans.AGU, 81, F1203,2000.

Malin, P. E., M. Lou, andJ.A. Rial, Fr waves:A secondfault-guidedmodewith implicationsfor

fault propertystudies,Geophys.Res.Lett., 23, 3547–3550,1996.

Marco,S.,andA. Agnon,Repeatedearthquake faultingrevealedby high-resolutionstratigraphy,

Tectonophysics, p. in press,2003.

Marco,S.,M. Stein,A. Agnon,andH. Ron,Long-termearthquake clustering:A 50,000-year

paleoseismicrecordin theDeadSeaGraben,J. Geophys.Res., 101, 6179–6191,1996.

Michael,A. J.,andY. Ben-Zion,Challengesin inverting fault zonetrappedwavesto determine

structuralproperties,Journal of ConferenceAbstracts, 3, 3547–3550,1998a,conference

onMathematicalGeophysics,July12-17,1998.

Michael,A. J.,andY. Ben-Zion,Invertingfault zonetrappedwaveswith geneticalgorithms,EOS

Trans.AGU, 79, S11C–16,1998b.

30

Neidell, N., andM. T. Taner, Semblanceandothercoherency measuresfor multichanneldata,

Geophysics, 36, 482–497,1971.

Peng,Z., Y. Ben-Zion,andA. Michael,Quantitative inversionof seismicfault zonewaveforms

in therupturezoneof the1992Landersearthquake for structuralpropertiesatdepth,EOS

Trans.AGU, 82, S32E–09,2000.

Rabb’a,I., TheGeologyof theAl Qurayqira(JabalHamraFaddan),Map SheetNo. 3051II),

in 1:50.000Geological MappingSeries, GeologicalBulletin No. 28, TheHashemite

Kingdom of Jordan,Ministry of Energy andMineral Resources,NaturalResources

Authority, GeologyDirectorate,GeologicalMappingDivision,Amman,1994.

Ritter, O.,J.Schmidt,U. Weckmann,H. Thoss,A. Abueladas,andV. Haak,A 3D magnetotelluric

studyof theDeadSeaTransformfault in Jordan,EOSTrans.AGU, 82, S41A–0582,2001.

Rotstein,Y., Y. Bartov, andU. Frieslander, Evidencefor local shifting of themain fault and

changesin the structuralsetting,Kinarot basin,Dead-Seatransform,Geology, 20,

251–254,1992.

Rovelli, A., A. Caserta,F. Marra,andV. Ruggiero,Canseismicwavesbe trappedinsidean

inactive fault zone?Thecasestudyof NoceraUmbra,centralItaly, Bull. Seismol.Soc.

Am., 2002,accepted.

Salamon,A., A. Hofstetter, Z. Garfunkel, andH. Ron,Seismicityof theeasternmediteranean

region: Perspective from theSinaisubplate,Tectonophysics, 263, 293–305,1996.

Sambridge,M., andG. Drijkoningen,Geneticalgorithmin seismicwaveforminversion,Geophys.

J. Int., 109, 323–342,1992.

Scholz,C. H., Wearandgaugeformationin brittle faulting,Geology, 15, 493–495,1987.

Schulz,S.E., andJ.P. Evans,Mesoscopicstructureof thePunchbowl Fault,SouthernCalifornia

31

andthegeologicandgeophysicalstructureof activestrike-slipfaults,Journalof Structural

Geology, 22, 913–930,2000.

Sen,M. K., andP. L. Stoffa,Rapidsamplingof modelspaceusinggeneticalgorithms:examples

from seismicwaveforminversion,Geophys.J. Int., 108, 281–292,1992.

Sieh,K. E., andR. H. Jahns,Holoceneactivity of theSanAndreasFault at WallaceCreek,

California,Geol.Soc.Am.Bull., 95, 883–896,1984.

Stirling, M. W., S.G. Wesnousky, andK. Shimazaki,Fault tracecomplexity, cumulative slip, and

theshapeof themagnitude-frequency distribution for strike-slip faults:A globalsurvey,

Geophys.J. Int., 124, 833–868,1996.

Vidale,J.,D. V. Helmberger, andR. W. Clayton,Finite-differenceseismogramsfor ��� waves,

Bull. Seismol.Soc.Am., 75, 1765–1782,1985.

Wallace,R., andH. Morris, Characteristicsof faultsandshearzonesasseenin deepmines,Pure

andAppliedGeophysics, 124, 107–126,1986.

Wesnousky, S. G., TheGutenberg-Richteror characteristicearthquake distribution, which is it,

Bull. Seismol.Soc.Am., 84, 1940–1959,1994.

Wessel,P., andW. Smith,New versionof theGenericMappingToolsreleased,EosTrans.AGU,

76, 329,1995.

Wessel,P., andW. Smith,New, improvedversionof theGenericMappingTools released,Eos

Trans.AGU, 79, 579,1998.

Ch.Haberland,N. Maercklin,G. Rumpker, T. Ryberg, M. Weber, GeoForschungsZen-

trum Potsdam,Telegrafenberg E254,14473Potsdam,Germany, haber@gfz-potsdam.de,

32

nils@gfz-potsdam.de,rumpker@gfz-potsdam.de,trond@gfz-potsdam.de,mhw@gfz-

potsdam.de�A. Agnon, Instituteof EarthSciences,The Hebrew University, Jerusalem91904,

Israel,amotz@cc.huji.ac.il

R. El-Kelani,An-NajahNationalUniversity, EarthSciences& SeismicEngineering

Centre,P.O.Box707,Nablus,PalestineTerritories,radwan@najah.edu

I. Qabbani,NaturalResourcesAuthority, P.O. Box 7, 11118Amman,Jordan

F. Scherbaum,Instituteof Geosciences,University Potsdam,PO 601553,14415

Potsdam,Germany, fs@geo.uni-potsdam.de

Received-; revised-; accepted-.

To appearin theJournalof GeophysicalResearch

This manuscriptwaspreparedwith AGU’s LATEX macrosv5, with the extension

package‘AGU�a� ’ by P. W. Daly, version1.6bfrom 1999/08/19.

33

Figure Captions

Figure 1. (left) Sketchshowing the tectonicsettingof the Middle East. (middle) Site map

of the active seismicexperimentat the Arava Fault (AF) aspart of the DeadSeaTransform

(DST).Fault trace(darkgrey line) asinferredbeneathsuperficialdeposits.Shotsaredenoted

by stars.Verticalsensorsonlines2 and3 (spacing100m) areshownby triangles,3-component

sensorsalong lines 4 and 5 (spacing10 m) are shown by squares. (right) Close-upof the

region displayedin themiddlepanel. Filled symbols(stars,triangles,andsquaresasbefore)

indicateshotsgeneratingandreceiversobservingguidedwaves,opensymbolsindicateshots

and receivers without suchgenerationor observation, respectively. Geologyadoptedfrom

Rabb’a[1994] (NL: Na’ur limestone,LM: Neogenemarl).

Figure 2. Observationson receiver line 4 (left) andline 5 (right) for thethreeshots101(top),

102 (middle), and103 (bottom). For eachgather, normalizedmaximumenergy (bars,top,

within time windows of 1.2 to 1.6 and0.6 to 0.8 s, respectively), time series(middle) and

spectra(bottom)for eachtraceareshown. Notethehigh-frequency wavetrainsfrom shots101

and102 on easterntracesof both lines between1.3 and1.6 s (line 4, left) and0.6 and0.8 s

(line5, right), but no suchphasefor shot103(andall othershots;seetext). For positionsrefer

to Figure1. Shown areunfilteredverticalcomponentdatacontainingpresumedguidedwaves.

Pwaveonsets(notshown) arriveat around0.7s (line 4, left) and0.4(line 5, right).

Figure 3. Waveformscalculatedfor syntheticmodel (for parametersrefer to Table3). We

addedrandomstaticshiftsof at most " 0.005s andrandomnoise(maximumsignal-to-noise

ratio of 5) to the synthetics.Shown areraw synthetics(thin line) andsyntheticswith noise

(thick line) whichweusedin therecovery test.

34

Figure 4. A-D: Fit of all in the GA searchtestedmodels(syntheticrecovery test)displayed

for the 4 free parameters.Grey dotsandblack crossesindicate ) ��,/.80D2 and ) *�,/.1032 ,respectively. Theparametersof thesyntheticinput modeltogetherwith its associated) value

aredenotedby black star. The dashedline marksthe ) level of the bestmodel found. E:

Cross-plotof all acceptablemodels( ) colour coded;red denotehigh values,purpledenote

low fit values)asa function of � and � ��� �E$ � ��� � illustrating thestrongtrade-off betweenthese

two parameters(syntheticrecoverytest).Thesyntheticinputmodelis representedby theblack

star. Dataweresortedaccordingto ) valueprior to plotting. F: Cross-waveguidevelocity

distribution of acceptablemodelsfor thesyntheticrecovery test. Grey (black) linesrepresent

modelswith ,/.10���)���,/.80D2 ( )�*�,C.10D2 ). Thesyntheticinput modelis representedby thered

line.

Figure 5. Searchresultsfor shot101observedat line 4. Notationsasin Figure4 exceptthat

grey dotsandblackcrossesindicate )J��,/.1560L�u)���� ��� and )+*-,/.1560��P)���� ��� , respectively.

Figure 6. Searchresultsfor shot102observedat line 4. Notationsasin Figure4 exceptthat

grey dotsandblackcrossesindicate )J��,/.1560L�u)���� ��� and )+*-,/.1560��P)���� ��� , respectively.

Figure 7. Searchresultsfor shot101observedat line 5. Notationsasin Figure4 exceptthat

grey dotsandblackcrossesindicate )J��,/.1560L�u)���� ��� and )+*-,/.1560��P)���� ��� , respectively.

Figure 8. Cross-waveguide velocity distribution of acceptablemodels( ) * ,/.1?6� ) of the

observationson line 4 (shots101and102). Dif ferentshadingindicateshow oftena particular

modelcomponentwasfound in the searches(from light to dark increasingfrequency). The

distinctblackvertical linesshow thatmostwell-fitting modelsarecharacterizedby a 4 to 6 m

wide waveguide.

35

Figure 9. Syntheticsof thebestmodels(thin line) andcorrespondingdata(thick line) of shot

101(top)and102(middle),line 4, andof shot101observedat line 5 (bottom).Observedtraces

havebeenshifted( " 1 or 2 samplesrelative to theneighboringtraces)accordingto thevalues

derivedin theinversion.

36

Tables

37

Table 1. Positionsof shotsgeneratingguided

waves

id latitude(� d ) longitude(

�E�) altitude(m)

s101 30.57800 35.33440 33.0

s102 30.57800 35.33311 21.7

38

Table 2. Positionsof receiver line segments

recording# guidedwaves

id latitude(� d ) longitude(

���) altitude(m)

412 30.55746 35.32787 31.0

413 30.55745 35.32798 30.9

414 30.55743 35.32808 31.3

415 30.55742 35.32818 31.4

416 30.55741 35.32828 31.6

504 30.58718 35.33677 36.3

505 30.58724 35.33669 36.7

506 30.58730 35.33660 36.7

507 30.58734 35.33652 36.9

508 30.58738 35.33638 34.2

509 30.58742 35.33629 33.8

510 30.58747 35.33621 33.2

39

Table 3. Parametersusedfor synthetictestand

searchresults

Parameter valueusedin rangeof values bestvaluefound

forwardcal- revealedby the in search

culation search

� ��� � (m/s) 1750 1710- 1775 1713

� ��� �%$ � ��� � 0.6 0.4- 0.9 0.43

� (m) 5 3 - 18 3.31

�'� �� 0.005 0 - 0.013 0.011

� (kg/m�) 2500 - -

: � at interface - -

Table 4. Parametersderivedby inversionof datarecordedalongline 4 and5. For each

parameterrangesandbestvaluesaregiven.

Parameter shot101/line4 shot102/line4 shot101/line5

range bestvalue range bestvalue range bestvalue

� ��� � (m/s) 1685- 1870 1742 1694- 1850 1847 1570- 1780 1710

� ��� �%$ � ��� � 0.4- 0.95 0.43 0.4- 0.62 0.44 0.45- 0.95 0.84

� (m) 3 - 35 3.70 4 - 7 4.47 2 - 27 6.02

�'� �� 0.005- 0.025 0.013 0.005- 0.015 0.006 0 - 0.015 0.0014

100 km 1 km 1 km

30N3430N

Taurus

35N

DS

T

SeaDead

ArabianPlate

35E

Line 5

Line 4

Line 3

30N34

Line 2

LM

35E20

Line 1

Line 3

NL

Line 5

AF

LM

35E20

S 103

S 101

S 102

SinaiPlate

SeaRed

Med

iter

ran

ean

Sea

AF

Line 2

Line 4

1.2

1.4

1.6

0

40

800 50 100 150 200

1.2

1.4

1.6

0

40

800 50 100 150 200

1.2

1.4

1.6

0

40

800 50 100 150 200

0.6

0.8

1.0

0

40

800 50 100 150 200

0.6

0.8

1.0

0

40

800 50 100 150 200

0.6

0.8

1.0

0

40

800 50 100 150 200

time

(s)

time

(s)

time

(s)

f (H

z)f (

Hz)

f (H

z)

line 4 (m) line 5 (m)

Fig. 2

shot 102shot 102

shot 103 shot 103

shot 101 shot 101

W EEW

1.2

1.4

1.6

1.8

time

(s)

60 80 100 120 140 160 180 200profile (m)

Fig. 3

0.70

0.75

0.80

0.85

fit

1650 1700 1750 1800v1(m/s)

A

0.70

0.75

0.80

0.85

fit

0.4 0.5 0.6 0.7 0.8 0.9 1.0v2/v1

B

0.70

0.75

0.80

0.85

fit

0.00 0.01 0.02 0.03Q1

-1

C

0.70

0.75

0.80

0.85

fit

0 10 20 30 40 50width (m)

D

0.4

0.5

0.6

0.7

0.8

0.9

1.0

v 2/v

1

0 10 20 30 40 50width (m)

E 0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0v s

(km

/s)

-20 -10 0 10 20x-fault position (m)

F

Fig. 4

0.60

0.65

0.70

0.75

0.80

fit

1650 1700 1750 1800v1(m/s)

A

0.60

0.65

0.70

0.75

0.80

fit

0.4 0.5 0.6 0.7 0.8 0.9 1.0v2/v1

B

0.60

0.65

0.70

0.75

0.80

fit

0.00 0.01 0.02 0.03Q1

-1

C

0.60

0.65

0.70

0.75

0.80

fit

0 10 20 30 40 50width (m)

D

0.4

0.5

0.6

0.7

0.8

0.9

1.0

v 2/v

1

0 10 20 30 40 50width (m)

E 0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0v s

(km

/s)

-20 -10 0 10 20x-fault position (m)

F

Fig. 5

0.60

0.65

0.70

0.75

0.80

fit

1650 1700 1750 1800v1(m/s)

A

0.60

0.65

0.70

0.75

0.80

fit

0.4 0.5 0.6 0.7 0.8 0.9 1.0v2/v1

B

0.60

0.65

0.70

0.75

0.80

fit

0.00 0.01 0.02 0.03Q1

-1

C

0.60

0.65

0.70

0.75

0.80

fit

0 10 20 30 40 50width (m)

D

0.4

0.5

0.6

0.7

0.8

0.9

1.0

v 2/v

1

0 10 20 30 40 50width (m)

E 0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0v s

(km

/s)

-20 -10 0 10 20x-fault position (m)

F

Fig. 6

0.60

0.65

0.70

0.75

0.80

fit

1550 1600 1650 1700v1(m/s)

A

0.60

0.65

0.70

0.75

0.80

fit

0.4 0.5 0.6 0.7 0.8 0.9 1.0v2/v1

B

0.60

0.65

0.70

0.75

0.80

fit

0.00 0.01 0.02 0.03Q1

-1

C

0.60

0.65

0.70

0.75

0.80

fit

0 10 20 30 40 50width (m)

D

0.4

0.5

0.6

0.7

0.8

0.9

1.0

v 2/v

1

0 10 20 30 40 50width (m)

E 0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0v s

(km

/s)

-20 -10 0 10 20x-fault position (m)

F

Fig. 7

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

v s (

m/s

)

-20 -10 0 10 20x-fault-position (m)

Fig. 8

1.2

1.4

1.6

1.8

time

(s)

60 80 100 120 140 160 180 200line 4 (m)

1.2

1.4

1.6

1.8

time

(s)

60 80 100 120 140 160 180 200line 4 (m)

0.6

0.8

1.0

time

(s)

60 80 100 120 140 160 180 200line 5 (m)

Fig. 9

shot 101

shot 102

shot 101