VOL. 84, NO.

B5 JOURNAL OF GEOPHYSICAL RESEARCH MAY 10, 1979

b Values and c0-* SeismicSourceModels: Implicationsfor Tectonic

StressVariations Along Active Crustal Fault Zones and the
Estimation of High-Frequency StrongGround Motion
THOMAS C. HANKS

U.S. GeologicalSurvey,Menlo Park, California 94025

In this studythe tectonicstressalongactivecrustalfault zonesis taken to be of the form •(y) + Aa,(x,

y), where •(y) is the averagetectonic stressat depth y and Aa,(x, y) is a seismologicallyobservable,
essentiallyrandom functionof both fault plane coordinates;the stressdifferencesarisingin the courseof
crustal faulting are derived from Aa,(x, y). Empirically known frequencyof occurrencestatistics,
moment-magnituderelationships,and the constancyof earthquakestressdropsmay be usedto infer that
the numberof earthquakesN of dimension>r is of the form N • 1/? and that the spectralcomposition
of Aao(x,
y) isoftheformI'œ•.(k)l• •/k•, where
A•%(k)
isthetwo-dimensional
Fourier
transform
of
Aa,(x, y) expressedin radial wave numberk. The 3/= 2 model of the far-field shearwave displacement
spectrum
is consistent
withthespectral
composition
I•,(k)l • l/k s,provided
thatthenumber
of
contributionsto the spectralrepresentation of the radiatedfield at frequencyf goesas(k/ko)•, consistent
with the quasi-staticfrequencyof occurrencerelationN • 1/?; k0 is a referencewave numberassociated
with the reciprocalsourcedimension.Separately,a variety of seismologicobservationssuggests that the 3/
= 2 model is the one generally,althoughcertainlynot always,applicableto the high-frequencyspectral
decay of the far-field radiation of earthquakes.In this framework, then, b valuesnear 1, the general
validity of the 3/ = 2 model,and the constancyof earthquakestressdropsindependentof sizeare all
relatedto theaverage
spectral
composition
of Aa,(x,y), I • l/kL Shouldoneof thesechange
as
a resultof premonitoryeffectsleadingto failure, as has beenspecificallyproposedfor b values,it seems
likely that one or all of the other characteristicswill change as well from their normative values.
Irrespectiveof theseassociations,the far-field, high-frequencyshearradiation for the 3/= 2 model in the
presenceof anelasticattenuation may be interpreted as band-limited, finite duration white noise in
acceleration.Its rmsvalue,a.... isgivenby the expression arms-' 0.8512x/•(2*r)•/106]
(Aa/pR)(fmax/fo)•/a,
whereAa is the earthquakestressdrop, p is density,R is hypocentraldistance,f0 is the spectralcorner
frequency,andfmaxis determinedby R and specificattenuationI/Q. For severalreasons,one of whichis
that it may be estimatedin the absenceof empirically definedground motion correlations,armsholds
considerablepromiseas a measureof high-frequencystrongground motion for engineeringpurposes.

INTRODUCTION theseis the averagetectonicstressoperativeto causefailure on

Very little is known about the heterogeneitiesin material the fault zone in the first place; whether this value is of the
propertiesand tectonicstressthat existalongactivecrustal order of 100bars(or perhapssomewhatless)or of the order of
fault zones,yet suchheterogeneities
arelikelyto playa central a kilobar (or perhapssomewhatgreater) is as yet unresolved
role in earthquakemechanics.It is now known that crustal [e.g., Hanks, 1977]. In the first caseone may anticipate that
variations in tectonic stress must be of the order of 100% of the
earthquakestressdrops,in theiraveragevalueAa, are several
tens of bars and that this value is independentof source averagevalue, but in the secondcasethey needonly be a small
strengthover 12 ordersof magnitudein seismicmoment[e.g., fraction of the averagevalue (although they could be larger).
Aki, 1972;ThatcherandHanks, 1973;KanamoriandAnderson, A seconddifficulty is that variations in the stressesdriving
1975;Hanks, 1977]. Becauseearthquakesare generallyepi- relative motions, in the self-stress[Andrews, 1978] resulting
sodicfunctionsof spaceand time along even the most well from pastfaulting episodes(which may be quite nonuniformif
developedcrustalfault zones,it may then be inferredthat nonuniform faulting displacementsare commonin the caseof
stressvariations of at least Aa commonly exist along active individual earthquakes),and in material propertiesalong the
crustalfault zones(althoughthey mightbe greatlyreduced,if fault zone all contributeto inhomogeneityin tectonicstresses
not eliminated,at the time and placeof throughgoingearth- along the fault zone of interest.Even if the amplitude-wave-
quakefaulting,givingrise,for example,to the notablyaseis- lengthcontentof actualstressvariationsalongfaultswasknown,
mic section of the San Andreas fault that broke in the great which it is not, a difficult problem would remain in separat-
earthquakeof 1857).But because it may alsobe inferredthat ing out the causativeprocesses to which it shouldbe related.
suchfaults,in general,can be no further away from repeated Similarly, there is accumulatingevidencethat the dynamic
failure than stressreaccumulationcomparableto Aa, it seems faulting displacementsand associatedstressdifferencescan be
plain that tectonicstressheterogeneities
of the orderof Aa highlyinhomogeneous in the courseof crustalfaulting,and it
mustplaya centralrolein determiningwhya particularearth- is natural to suspect(but difficult to prove) that these in-
quakeoccursat a particularpoint in spaceand time and homogeneities arisefrom variationsof the preexistingtectonic
thereforein any rationalcapabilitythat purportsto predict stresses acrossthe incipientrupture surface.For both the San
earthquakes. Fernando (for example, Hanks [1974] and Bouchon[1978],
Beyondthesetruisms,analysisof the natureand extentof among many suchinvestigations)and the Borrego Mountain
stressheterogeneities
along seismicallyactive faults is com- [Burdick and Mellman, 1976;Heaton and Helmberger, 1977]
earthquakes,there is considerableevidencethat faulting was
plicatedbyimportantbutpoorlyunderstood
problems.
Oneof
initiated with localized but massivefaulting with associated
This paperis not subjectto U.S. copyright.Publishedin 1979by the stressdifferencesnot at all representativeof thoseinferred for
American Geophysical Union. the entire faultingprocess.In a similarmannerthe largerpeak
Paper number 9B0052. , 2235
2236 HANKS: FAULT MECHANICS

accelerations(>•0.1g) at closedistances(R -• 10 km) almost quake magnitudefrequencyof occurrencestatisticsreduceto a

certainly representlocalized,dynamic stressdifferencesmany simplematter of geometricalscalingin termsof the reciprocal
timesgreaterthan the averageearthquakestressdrop [Hanks faulting area.
and Johnson,1976]. In a very real sense,of course,the ideas Equation (1), however, is also satisfied(with a different a
presentedin the studiescited above are simply scaled-down value) by the density distribution of the number of earth-
versions(in spatial dimensionand wave period) of the com- quakeswith respectto M, -dN/dM [Richter,1958,p. 359]. In
plex, multiple-event interpretation of large and great earth- this interpretation the density distribution of the number of
quakes(a numberof suchinvestigations are referencedby Das earthquakeswith respectto r is proportional to r -a, a result
and A ki [1977], who presenttheir barrier model of the earth- anticipated in the more complicated but essentiallysimilar
quake mechanismpartially in this context). model of Caputo [1976].
In any event, considerableinterest has developedaround To interpret (4), imagine a planar fault surface large in
theseobservationsand ideas, for at leasttwo important rea- comparisonto any earthquake sourcedimensionof interest,
sons.First, higher-qualityrecordingsand more detailedanaly- and a populationof incipientearthquakesto occurupon it; the
sisof suchearthquakesmay providea clearerunderstandingof earthquakepopulation is characterizedby the frequencyof
the nature and extent of tectonicstressheterogeneities along occurrencerelation (4) and averagestressdrops equal to Aa
active crustal fault zones.Second,reliable estimatesof high- but with scatterabout this value comparableto that observed
frequencystrongground motion and their usein the aseismic in the availablestressdrop data. Beforeany of the earthquakes
designof high-frequencystructuresdependquite stronglyon occur, all of the stressdifferencesthat will be realized at the
the nature and extentof theselocalizeddynamicstressdiffer- time of occurrencefor eachand everyeventexiston the fault
encesthat developin the courseof crustalfaulting. surfacein 'potential' form; we denotethis distributionin both
In this study the problem of tectonic stressvariations is spatialdimensionsasthe stressdrop potentialfunctionA%(x,
addressedthrough interpretations, developed herein, of b- y). As a matter of convenience, we assignzero meanto A%(x,
value data and the high-frequencyspectralcharacteristicsof y) and denotethe averageshearstresson the fault as •(y). In
the radiatedfield (w-* models)for crustalearthquakes.In fact, the earth, •(y) is the averagetectonicstressand is presumably
however, the quantitiesbeing investigatedare stressdiffer- a functionof depth;it is not, however,sampledby earthquake
encesavailableto be releasedat the time of faulting, to which stressdrops or any other measure of the radiated field of
may be addeda stressfunctionof whicheventhe averagevalue earthquakes.
is unspecifiedin this studyand which, in the absenceof addi- Within this framework we can expectthe stressdrop to be
tional information, is unknown. Following the ideas of An- realized acrossan area of incipient rupture A to be derived
drews [1978], one may infer but cannot prove that this un- from the root-mean-square(rms) value of Aa•,(x, y) over A,
known stress function is intrinsically smooth, .arising where we understand,purely as a formality, that only those
fundamentally from large-scalestressesdriving relative motion regionsof mostlypositiveAa•,(x,y) are candidatesfor rupture.
across the fault, and that the actual variations in tectonic The mean-squarevalue of A%(x, y) acrossA is
stressesalong active crustalfault zonesare, to a first approxi-
mation, reasonablyestimatedthrough the ideasdevelopedin
this study.
(a%'-): lff [a%(x,y)]'-dxary (5)
b VALUESAND EARTHQUAKESTRESSDROPS Since
A%(x,
y)pro,•_q4ces
earthquakes
which,
ontheaverage,
havestressdrops•Aa and satisfythe frequencyof occurrence
Hanks [1977] showedthat the relationsbetweenfrequency relation (4), (5) must be constantindependentofA. Becauseof
of occurrenceN of earthquakesof magnitude_>M, Rayleigh's theorem,
logN = a- bM (1)
between seismic moment M0 and M, •-1fo
©[IA%(k)l
'• ]•.
kdk (6)

log M0 = cM + d (2) must also be constant,independentof A, whereA%(k) is the

two-dimensionalFourier transformof Aa•,(x,y) expressed in
and between sourceradius r, earthquake stressdrop Aa, and terms of radial wave number k; wavelength3, is 2•-/k.
M0, It is difficult to be generalabout the circumstances
in which
(6) will be constantfor any A, but we can arrangea special
Mo = kAa? (3)
case
byassuming
I(k)l • k-• anda bandlimitation
for
can be combined to obtain, using b = I and c = 1.5, Aa•,(k) betweenkminand kmax.Physically,the k -'• dependence
of Iœ%(k)lcanberationalized
onthebasisof similarity,
and
N = const/('ha)wa? (4)
the band limitationmeans,for example,that • greaterthan the
N in (4), consistentwith N in (1), is the numberof earthquakes seismogenicdepth or lessthan a grain sizedo not contribute
with dimension _>r, and Aa has been usedfor Aa in (3). It is materiallyto Aa•,(x, y). Now, for n > 1 and kmax>> kmm,the
empirically known that b is generally, but not always, very constancyof (6) requiresthat
nearly equal to l, irrespectiveof the choiceof region and time
Akmman-a= const (7)
interval in which earthquakesare counted.Also, c is empiri-
cally known to be 1.5 whether local magnitudeML [Thatcher which, dimensionally,can only be arrangedby taking n = 2
and Hanks, 1973] or surface wave magnitude Ms [Kanamori and kmin • A-•/u.
andAnderson,1975]is usedin (2), althoughseriousdepartures Physically,this meansthat the only significantspectralcon-
from (2) with c = 1.5 begin to developfor Ms >• 7•. Equation tributionsto (Aa•,•') occur at X comparableto the sourcedi-
(4) impliesthat if the earthquakesof the countedsampleshare mension of incipient rupture. For the k-•' dependenceof
the same Aa, as they do on the averagefor all samplesfor I<k)l it is,of course, clearthattheshorter-wavelength
which the Aa havebeen determined[e.g.,Hanks, 1977],earth- contributionswill be negligible,but for the samereasonthe
 HANKS: FAULT MECHANICS 2237

-3
(where h is the seismogenicdepth and/i is the fault dip), the
two-dimensionalcharacterof the fault surfacecollapsesessen-
(ll
!
o tially to one, and it can be expectedthat the ideaspresented
above will no longer hold. For h -• 15 km and a vertical
_ transform fault, one may estimateroughly that this will occur
whenL •> 30 km or, equivalently,whenM8 •> 6•. In particular,
(3) then takes the form
Mo = k'AaL• (8)
Moreover, (2) with c = 1.5 beginsto fail at slightlylargerMs,
7-7i. Finally, as is well known, M• becomesan increasingly
poorer measureof sourcestrengthfor M0 •> l0:7 dyn cm, or
M• >• 7i [e.g.,Kanamori, 1977].As such,presentuncertainties
in estimatingboth c and 'magnitude' at large magnitudepre-
clude, at the presenttime, an extensionof theseresultsto the
Io•j frequency
more nearly one-dimensionalcharacter of large and great
Fig. l. Spectralrepresentationof the o•-: and o•-8 sourcemodels earthquakes.But thesedifficultiesin no way changethe argu-
for two constantstressdrop earthquakesobservedat the samedis- mentsgivenabovefor M• •< 6} earthquakesfor whichr •< w.
tance R in a uniform, elastic,isotropicfull space.
EVIDENCE FOR AND INTERPRETATION OF
THE o•-z SOURCE MODEL
longer•, contributionsare plainly a problem.But if •, >> (A)•/:
made a significantcontributionto (A•rp:)acrossour chosenA, In spectral form the far-field radiation emanating from
then it is most likely that (A•rp:)acrossa larger A' would also simple seismicsource models [e.g., ,4ki, 1967; Brune, 1970,
be --•(Atr):; that is, in such an eventuality the rupture of A' 1971] is generallycharacterizedby a long-periodlevel f•0 pro-
would be the event of interest, having incorporated in its portionalto M0, a cornerfrequency f0 proportionalto r-x, and
rupture the smallerarea A. In other words, limiting the rup- a high-frequency spectraldecayof the form (f/fo) -• (in the
ture area to some A must mean that A >> A •/: cannot contrib- followingdiscussion, frequencyis denotedby f in hertzrather
ute significantlyto (A•rp:)acrossA; otherwise,a larger area than •vin radiansper second).The cornerfrequencyf0, funda-
would rupture. As such,the frequencyof occurrencerelation mentally,is closelyallied with the reciprocaldurationof fault-
(4) may be written as N ,-• (A/A0)-: "• (k/ko):, whereA is the ing Ta-•, but it is well known that several'faulting durations'
wavelengthcorrespondingto any earthquake sourcedimen- can be defined, in particular those associatedwith the fault
sion of interest and A0 is some reference source dimension. length, fault width, and the rise time of a propagating dis-
In this context,then, a spectralcompositionof A•rp(X,y) placementdiscontinuity.Dependingon the faulting geometry
of the form ]•rp(k)] "• k-: will guarantee constantstress and rise time characteristics,the associatedcorner frequencies
drop earthquakesindependentof the size of the rupture sur- can be well separated, leading to more complicated high-
face and that the frequency of occurrencerelation will be frequencyspectralamplitude decay(that is, 7 is a function of
satisfied.It is worth emphasizing,however,that this represen- frequency). Moreover, by making the displacementdisconti-
tation can well be nonuniqueand need not be correct, even nuity a smooth enough function of time, 7 can becomearbi-
though a different representationthat will guaranteethe con- trarily large at high enough frequencies.Whether in fact a
stancyof (6) for any A is not obvious.We shallfind, however, generally applicablesourcerepresentationof high-frequency
that the dynamic field radiated by earthquakesin the caseof spectralcharacteristics existswithin the infinity of possibilities
the 3' = 2 modelis consistent with I•ap(k)l '• k-: and is as yet theoreticallycontroversialand observationallyunre-
providesseparatesupportfor this representation. solved. M ore as a matter of convenience than a matter of hard
In supposingthat theseideasare relevantto currentlyactive fact, high-frequencyspectralcharacteristicsof seismicsources
crustal fault zones, some additional points should be made. are generallydiscussed
i
in terms of •0 and f0 relatedby the
First, stressdrops both higher and lower than A• will occur constantstressdrop assumption(•0f0a = constin the context
with certain probabilitiesdetermined by the distribution of of the •0-f0 relationsof HanksandThatcher[ 1972])and7 = 2
/x•rp(x,y) about its rms value. Existingstressdrop data are (the •v-squaremodel) or 7 = 3 (the •v-cube model, in the
mostly in the range of severalbars to severalhundredbars, terminologyof ,4ki [ 1967]).
allowing for likely biasingto lower valuesin the caseof many Figure 1 schematicallyillustrates the 7 = 2 and 7 = 3
of the smallerearthquakes[ThatcherandHanks, 1973;Hanks, seismic source models in terms of two idealized far-field shear
1977].These determinationssuggesta log normal distribution wave displacementspectralamplitudesat the samedistanceR.
about a logarithmicmean of approximately30 bars(Kanamori In both the 7 = 2 and 7 = 3 casesthe two earthquakeshave
and Anderson[1975] have suggested/x•r= 60 bars on the basis been assignedthe sameA•r, so the corner frequencieslie on a
of arithmetic averaging),one logarithmicstandarddeviation line of slope-3 in theselog-logplots. In both casesthe larger
beingabout 0.5. Thus while the area-independentrms value of event (event 1) has •0 and M0 3 orders of magnitudelarger
/x•rp(x,y) is determinedby/x•r •- 30 bars,it may vary,at least than the smallerevent(event 2), and f0• is 10 timessmaller
occasionally,to severalhundredbars.It is interestingthat this than f0•:•.
latter value is approximatelythe sameas the variation about At frequencies greaterthan f0•:•, spectralamplitudesare 10
the mean of the frictional strengthof commoncrustalrocksat times greater for event 1 than for event 2 in the 3' = 2 casebut
constantpressureand temperature[Byedee, 1978]. are the samein the 3' = 3 models.How do we interpret these
Second,activecrustalfault zonesare plainly not infinite in modelsin terms of time-domain amplitudes,recognizingthat
both spatial dimensions.For those earthquakeswith fault Ta• •- 10Ta•:•?Figure 2 presentsthe extremeinterpretations.
length L sufficientlygreater than fault width w -• h/sin • Here, for purposesof illustrationwe have taken f0• = 0.05
2238 HANKS: FAULT MECHANICS

of finite duration (with singularparticle accelerations),Geller

[1976] followed Haskell [1964] to obtain 3' = 3 at high fre-
I sec quencies.His justification of this model with existingrno-Ms
data is not correct, however, becausehe assumedthat rn0 and
Ms faithfully represent spectral amplitudes at 1- and 20-s
• 2 20 periods, respectively,acrossthe entire range of magnitudes
observableat teleseismicdistances.Geller [1976] notesthat 'it
is not exactlycorrect'to do this;quite generally,it is not at all
correctto do this, exceptfor the smallerearthquakes(M >• 5)
for whichfo >• 1 Hz. In thelattercase,bothrnoandMs become
long-period measurements,proportional to Mo, b.ut then, of
course,rno-Ms data carry no information at all about high-
frequencyspectralcharacteristicsof earthquake sources.
Fig. 2. Time domain interpretation of the w-2 and w-8 source
There are, in addition, several other observations that are in
models:(a) co -2 modelwhen 1-sradiation arrivescontinuouslyacross
Ta (2 s on the left-handsideand 20 s on the right-handside,of which general accordwith the high-frequencyspectralcharacteristics
only 10 s are shown in the figure), (b) co-• model when 1-s radiation of the co-•' model. First, the difference of a factor of 20 in the
arrivesin the first 1-sinterval, (c) co
-8 modelwhen 1-sradiation arrives maximum rnoof •7.0 and maximum Ms of •8.3 is 'exactly'
continuouslyacrossTa, and (d) co -a model when 1-sradiation arrives
in the first 1-sinterval.Relative 1-samplitudesaregivenin two groups
predictedby the 3' = 2 model(becauseof the periodshiftin the
of four, one for the co -• model interpretationand one for the w-a' amplitudemeasurement from 1 to 20 s) providedthat rno-•
model interpretation;the choiceof 1 in the upper left-hand cornerof Ms at •- 7. In the'latest'form of the linearrelationsbetweenrn0
each squareis arbitrary. and Ms, rno= Ms at 6.75 [Richter, 1958, p. 348]. Second,the
sameargumentsusedaboveto justify the 3' = 2 model in terms
Hz, fo© = 0.5 Hz, Ta(x)= 20 scc,and Ta© = 2 scc,and we are of rnoand Ms data, and the upper limit to each, may also be
investigatingpossibleinterpretationsof 1-s time domain am- used to explain why peak accelerationdata at a fixed, close
plitudes, those usedin determiningmo. distanceare such a weak function of magnitude,especially
Figure 2a is the interpretationfor the 3' - 2 earthquakesif above ML -• 4•t-5 [Hanks and Johnson,1976].
the 1-s energy arrives more or less continuously over the Third, the high-frequencyspectralcharacteristicsof the San
complete faulting duration. In this case, l-s spectralampli- Fernando earthquakeare very well known [Berrill, 1975],even
tudes for the larger event are 10 times greater than for the at frequencies 2 ordersof magnitudegreaterthan fo = 0.1 Hz
smallerevent,but the 1-stime domain amplitudesare the same [WyssandHanks, 1972],becauseof the largenumberof strong
for both events--they have the samemo.If all the 1-senergy motion accelerogramsthat recordedthis earthquake at local
arrivesat the sametime, however,the 1-stime domain ampli- distances.With allowance for radiation pattern effectsand
tudesand m0of the larger eventare 10 timeslarger(Figure 2b). anelasticattenuation, the 3' = 2 model of Brune [1970, 1971],
For the 3' - 3 earthquakes,1-sspectralamplitudesmust be parameterizedby Mo = l0•6 dyn cm and r = l0 km, is the
the same.In Figure 2c this is achievedin a manner analogous simplestpossibleinterpretation of the data, although more
to that in Figure 2a, but now 1-stime domain amplitudesfor complicatedinterpretationsare possibleand, in view of the
event 1 are 10 times smaller than for event 2; that is, rn0 must highly inhomogeneouscharacter of faulting for this earth-
decreasewith Mo. Figure 2d is the analogueto Figure 2b; here quake, perhapswarranted.
1-s time domain amplitudes for the two earthquakesare the Finally, a great numbei' of spectral determinations have
same; they have the same mo. been made in the courseof numeroussourceparameter stud-
The interpretationin Figure 2c is certainlyunacceptable: ies,althoughthe great bulk of theseare single-stationmeasure-
doesnot systematicallydecreasewith increasingMo. Neither, ments (in addition to those cited by Hanks [1977], seealso
however, does rn0 increasebeyond rn0 •- 66-7, and the inter- Trifunac [ 1972a,b], JohnsonandMcEvilly [ 1974],Bakun et al.
pretation in Figure 2b is also inappropriate,at least for Mo [1976], and Hartzell and Brune[1977]). Of the three parame-
10•6 dyn cm. One's preferencefor the interpretationin Figure ters•2o,fo, and 3', 3' is almostalwaysthe leastwell determined.
2a or 2d and thus one'spreferencefor 3' - 2 or 3' - 3 seismic Even so, 3' = 2 is the value most often recovered,althoughthe
sourcemodelsthen dependson whether one believesthat all same observationsclearly demonstratethat 3' is not precisely
(or most) of the 1-s energy arrives more or lesscontinuously 2, or even particularly close ot it, for each and every earth-
through Ta (Ta > 1 s) or arrivesmore or lessimpulsivelyin a quake. Still and all, the severalsetsof observationssumma-
•, l-s window (and in the caseof m0, the first one or two such rized in this section leave little alternative to the conclusion
windows)no matter what the value of Ta. It is appropriateto that the 3' = 2 model is the one generally, if certainly not
recall now that both possibilitiesare extreme, and grossly always, operative.
simplified,interpretations;the truth, in most cases,shouldlie Figure 3 presentsthe accelerationspectralamplitudes,in the
somewhere in between. Even so, when Ta >> 1 s in the case of presenceof anelasticattenuation for the two 3' = 2 events
the larger earthquakes(Ms • 6{), it is clear that Figure 2d is whosedisplacementspectralamplitudesare givenin Figure 1.
much more the exceptionthan the rule, as almost all short- In the frequencyband fo -< f -< fmax,accelerationspectral
period seismogramsof large and great earthquakesreveal. amplitudesare constant,fmaxbeingdeterminedby settingthe
Thus I conclude,as Aki [1967] did more than a decadeago, argumentof the exponent in the expression e-•rfR/Q•equalto
that rnt,-Msdata supportthe 3' - 2 model, in the interpretation 1. Then one interpretation, again nonunique, of Figure 3 is
of Figure 2a, as the one generally(but certainly not always) that the correspondingaccelerationtime historiesare band-
applicable to the representationof high-frequencyspectral limited (fo -< f -< fmax),finite duration(0 _<t - R/i• -< Ta)
characteristics. white noise. The whitenessarisesfrom the constant spectral
With the assumptionsthat (1) fault propagation in both amplitudesequal to flofo •' in the band fo -< f -< fma,,.The
coordinatesof the fault planeis equallyphasecoherentand (2) randomnesshas simply been assumed,but in view of the
the sourcedisplacementtime function is a propagatingramp generallychaotic nature of strongmotion accelerogramsfor
 HANKS:FAULTMECHANICS 2239

M ,-, 5 earthquakesat R •< 10 km, in finite time windowsand

countedsample.In particular,whenb < 1, thereis a relative
frequency
bands,thisassumption
doesnotseemunreasonable. excessof largerearthquakes to smallerearthquakes, which
Indeed,the ideathat groundaccelerationtime historiescan be maybeinterpreted
in termsof I(k)l deficient
in short-
treated as band-limited, finite duration white noise has been wavelength
amplitudesrelativeto a normative
k-2 depen-
the basisfor considerablework in the analysisof existing dence;then3' > 2 wouldbe expected
if thedynamicstress
accelerogramsand in the computationof syntheticaccelero- differences
arisefromthesame
tectonic
stress
fields.
Presently
gramsfor morethan30 yearsin the engineering community availabledata, unfortunately,
are not suitedfor a criticalex-
[e.g.,Housner,1947;Hudson,1956;Bycroft,1960;Housnerand amination
of thishypothesis
principally
because
of thepoor
Jennings,1964;Jenningset al., 1968]. control on 3'.
HanksandJohnson [1976]developed the followingrelation Correlations
ofwell-determined b-valueand.3'datamaybe
between theamplitude/iof anyaccelerationpulseatR andthe particularly importantto developin viewof the suggestion
dynamicstressdifferenceaa giving rise to it in the source thatb decreases
priorto largerearthquakesandis thereforea
region: possible
meansof earthquake prediction[Scholzet al., 1973;
1 o'tt
Wyss and Lee, 1973;Rikitake, 1975],
although the data are
a-
p R
(9) hardlyconclusive on thismatter[LahrandPomeroy, 1970].
With respectto the ideaspresented
here,severalpointsare
wherep is density.In thisframeworktheinterpretation
of the worthmakingaboutthispossibility.First,if b = 1 in a certain
3, = 2 model given above in terms of band-limited,finite regionovera longenoughperiodof time,thenb valuesesti-
durationwhitenoisein groundacceleration translatesdirectly matedovera shorterperiodof timethat explicitly excludes a
into a white,randomdistributionof dynamicstressdifferences largerearthquake(i.e.,theoneto be predicted) will be natu-
(at wavelengths lessthan r), but only with respectto the rally biasedto valuesthat are greaterthan one, not lessthan
essentiallyone-dimensional
configuration of Figure3. That is, one.Thusthoseareaswithb < 1 in a timeintervaljustbefore
of Figure3 is a one-dimensional an earthquakelargerthan any memberof the setcountedto
looselyspeaking,the abscissa
wavelength spectrumof a two-dimensional faultingprocess. determineb are especiallyinteresting.Second,as discussed
Now,if theradiated
fieldisdrawnfromI • (k/ko)-' previously,
oneinterpretation
of b < 1 is that I.(k)l is
andif thenumberof contributions to thespectralrepresenta- relativelydeficientin short-wavelengthamplitudes; thedevel-
tion of the radiatedfield at frequency f, wheref ,-, k, is opment of longer-wavelength stressconcentrationswould
proportionalto (k/ko)2, as suggested fre- seemto be a nautralpreludeto the occurrence
by the quasi-static of a larger
quencyof occurrence relation(4), thenthewavenumberspec- earthquake. Third,if b < 1 is accompanied by 3' > 2 for those
trumoftheradiated dynamic stressdifferences
willbeconstant'earthquakes thatarecounted to defineb, onemayproceed on
for k • 1/r. Through(9) this impliesa white acceleration an earthquake-by-earthquake basisratherthan waitingthe
spectrum and thus the 3' = 2 model of the far-field shear muchlongerperiodof timeforenough earthquakes to yielda
displacementspectrum. well-determined
b.Another
possibility,
andanimportant
one,
DISCUSSION
isthatb different
from1 maybeaccompanied
byearthquake
stressdrops that are not independentof sourcesize. More
The consistency
of the spectralcompositioninferredfor generally,if changesin any one of the normativesituationsof
A•,(x, y) and the 3' = 2 modelof the high-frequency ra- b valuesnear 1, the generalvalidityof the 3' = 2 model,the
diatedfieldis notable,in viewof the grosslydifferingtime constancy of earthquakestressdropsindependent of size,and
and dimension with themindividually(as I
scalesassociated I k occur asa result ofprocesses premonitory to
long as severaldecadesand hundredsof kilometersin the case larger-scalefaulting,it seemsreasonablethat at leastone and
of Aa•,(x, y) and as short as fractionsof secondsand tensof perhapsall of theotherphenomena will changeaswell.
metersin thecaseof the 3' = 2 modelfor smallearthquakes). Finally,it is worthnotingthat if, as in the interpretation
Becauseof the varietyof uncertaintiesand unknownsassoci- here,the frequencyof occurrence statistics,
or b values,are
ated with this coincidence, it is probablyprematureto make governedby the spectralcomposition of Aa•,(x,y), thenthe
too muchof it or reachtoo far for its physicalsignificance. overallrateof seismicity, or a value,is presumably controlled
Even so, it would follow quite naturallyif the tectonicstress by •(y). At leastalongmajorplatemargins,•(y) increases
alongactivecrustalfaultzoneswasof theform•(y) + Aa•,(x, slowlyon a time scaleof hundredsof yearsuntil the areaof
y), with the previouslydescribedcharacteristics for each. interestis rupturedby throughgoing faulting,at whichtime
If thisis thecase,b valuesdifferentfrom 1 mightbe accom- •(y) precipitously decreases
bytheearthquake stressdrop.It is
paniedby 3' differentfrom 2 for thoseearthquakes of the well knownthat theSanFranciscoBayarea,to a considerable
distance awayfromtheSanAndreasfault,wasconsiderably
moreseismic at theM >•6 levelin the ,-,70yearspriorto the
1906earthquake than it hasbeenin the ,-,70years,
since
fMAX
=f(R,O) [Tocher,1959],excluding theimmediateaftershock
sequence,
-o-"'
(fo"')' andit seemsreasonable thata stress
dropof approximately
100 bars along the San Andreasfault at the time of the earth-
.,-
o /
,- •o
f;'
,-•(2)t.
(2}
ß2 •---'
tToJ __. quakeplayeda centralrolein thegreatlyreduced
seismicity
rate. At the sametime, however,this situation underscoresthe
potentialambiguity
at all wavelengths>•hbetween thelong-
wavelengthcharacter
of Aa•,(x,y) and•(y) variablealongthe
fault length.
I • ß I I I I
HIGH-FREQUENCY STRONGGROUND MOTION
log frequency
FiB. 3. Accelerationamplitudespectraat R for the •-• earth- Whetheror not b valuesand3' arerelatedthrougha com-
quakesof FJSurc2, with attenuationexplicitlyshown. monoriginin a tectonicstress
of the form •(y) + Aa•,(x,y)
2240 HANKS: FAULT MECHANICS

along activecrustalfault zones,as discussed in the last section, The analysisbeginswith Parseval'stheorem,
the generalvalidity of the •, = 2 modelhasimportantimplica-
tions for new approachesto the estimationof high-frequency
stronggroundmotionfor aseismicdesignpurposes. One possi-
la(t)l
' dt• • la>l'
bility that suggests itself immediatelyis developedbelow, in where a(t) is the accelerationtime history and •(co) is its
comparisonwith the existingapproach.The alternatepoint of Fourier amplitudespectrum.For •(co)we take the •, = 2 model
view, that strongmotion accelerogramswritten at closedis- of Figure 3 and note that for large earthquakesat closedis-
tances(R -• 10 km) for potentiallydamagingearthquakesare tances,fm,x >> f0, SOthat contributionsto the right-handside
important data for investigatingin more detail the validity of of (10) for f _< f0 are small. We further assumethat the
the •, = 2 model, shall be left as beingimplicit. significantmotion is confinedto the shearwave arrival win-
Since the first strong motion accelerogramswere written dow 0 • t - R/• < Ta and anelastic attenuation cuts the
morethan 40 yearsago,peak acceleration hasb•eenthe most spectralamplitudes
off sharplyat f • fm,•. Then(10) maybe
commonlyused singleindex of strongground motion. It has, written
however, been known for some time that peak acceleration
need not be, and too often cannot be, a uniformly valid mea-
sure of strongground motion over the entire frequencyband
and amplituderangeof engineeringinterest.The very charac- The rms acceleration is
ter of the peak accelerationdatum as a short-period,time
domainamplitudemeasurement is the principalreasonfor two 1
important limitations on its value as a measure of,strong armsm [a(t)l:dt (12)
ground motion. First, for M >• 5 earthquakesat close dis-
tances, taken here as a rough thresholdof potentially dam- Equations(11) and (12), togetherwith
aging groundmotion, the period of this phaseis much shorter
•(w) = •2•f0) a b. • f • fmax (13)
than the faulting duration. Thus the peak accelerationsimply
cannot measuregross source properties of potentially dam- and the approximation
agingand destructiveearthquakes,evenif suchdata may, in a
large enoughset of observations,indicatelimiting conditions
b = 1/T• (14)
on the failure processin very localized regions of the fault result in
surface.Second,this samecharacteristicof the peak accelera-
tion datum makes precisecorrectionsfor wave propagation arms= 2•"(2')'•ob'(fm,db) •" (1•)
effects,including anelasticattenuation and elastic scattering, for fm,• >> f0. Finally,for the B•ne [1970,1971]scaling,
impossibleexcept under very unusual conditions. Both of
theseproblems,but especiallythe second,are in turn respon- Aa = 106vR•0f0
a (16)
siblefor the notoriouslylarge scatterin peak accelerationdata,
[Hanksand Thatcher,1972],which, upon substitutionin (15),
eventhroughvery smallvariationsof magnitude,distance,and
gives
site conditions.It is this last problem that limits the utility of
peak accelerationevenas a measureof high-frequencystrong
ground motion.
Thesedifficultiesin interpreting,manipulating,and extrapo-
arms
=0.85
21/•2•)a
106 (17)
lating peak accelerationdata are widely acknowledgedby The factor of 0.85 introducedin (17) accountsfor free surface
engineersand seismologists alike, and recentlyacquiredpeak amplificationof SH waves(2.0), vectorialpartition onto two
accelerationdata for 3 •< M •< 5 earthquakesat R •- 10 km horizontal componentsof equal amplitude (1/2•/a), and the
have accentuatedthem [Hanks andJohnson,1976;Seekinsand rms value of the shearwave radiation pattern (0.6) [Thatcher
Hanks, 1978]. But if peak accelerationis not a reliable mea- and Hanks, 1973].
sure of high-frequencystrongground motion, as is gener- Table 1 compares armsvalues estimated from (17) with
ally agreedto be the case,then what is? observed, whole record values for the San Fernando earth-
One such measurethat is almost certainly better is the rms quake at PacoimaDam and the Kern County earthquakeat
acceleration,arms.First, sincethe time integral of the square Taft and with 'observed'valuescorrectedfor (record length/
ground accelerationis proportionalto the work per unit mass Ta)•/• to estimate the armsvalue that occursin the time inter-
done on a set of linear, viscouslydamped, single-degree-of- val of the S wave arrival plus T•. For the San Fernando
freedom oscillatorswith natural frequenciesbetween0 and •o earthquake the 'observed'value is 70% greater than the esti-
[Arias, 1970], armsis then of considerableengineeringimpor- matedvalue at PacoimaDam; in the caseof the Kern County
tance (to the extent that actual structuresmay be approxi- earthquakeat Taft the 'observed'value is only 30% greater
mated by suchoscillators)with respectto the designcapabili- than the estimatedvalue. By conventionalseismologicalstan-
ties of the rate of dissipation of this energy. Second, as a dardsin estimatinghigh-frequencyamplitudes,this agreement
broadbandintegral measure,it almostcertainlywill be a more is remarkable.
stable measureof high-frequencystrongground motion than Thesecomparisonsare, on the one hand, encouragingwith
individual high-frequencytime domain amplitude measure- respectto the useof armsaSa measureof high-frequency strong
ments.Finally, as describedbelow,armscan be directlyrelated ground motion and, on the other hand, further evidencefor
to a very few parametersof the earthquakesourceand source- the general validity of the • = 2 model. In both respects,
station propagationpath and thus can be estimatedin the however,further examinationof existingdata is required,and
absenceof strong ground motion observationsor empirical strongmotion accelerograms at R • 10 km are a particularly
correlations derived from them. valuable setof observationsfor theseanalyses.
 HANKS:
FAULTMECHANICS 2241

TABLE 1. Comparisonsof Estimated and 'Observed' arm, Values groundmotionfor aseismic designpurposes.Alternatively,
thesesameideas,togetherwith strongmotionaccelerograms
SanFernando Kern County
Earthquake Earthquake written at closedistances
for potentiallydamagingearth-
at Pacoima Dam at Taft quakes,maybe usedfor investigating
in moredetailthe'y = 2
model.
Aa, bars 50* 60* •4cknowledgments.I have enjoyed the critical remarks of D. J.
f0, Hz 0.1t 0.05:]: Andrews,A. McGarr, and W. Thatcher in developingthe ideasof this
R, km • 10 •40 manuscriptand usefulconversationswith M. Caputo regardingfre-
fmax,Hz '-•30õ • 8õ quencyof occurrencestatistics.I deeplyappreciatethe effortsof E. A.
arms, cm/s • Flinn in his capacityof editor and an unknown associateeditor in
Estimated 140 30 havingme evaluatecriticallyan original,and erroneous,guessthat the
Observed
II 120,110 26,27 spectralcompositionof tiaa(x, y) was white, a guesswritten into the
'Observed' 240 39 work of Hanks (1977) without consequence. Carol Sullivanpatiently
typed this manuscriptseveraltimes. This researchwas supportedin
*Kanamoriand•4nderson[1975]. part by National Science Foundation grant ENV76-81816. Pub-
•Hanks and Wyss[1972]. lication approvedby the Director, U.S. GeologicalSurvey.
:]:Estimated
for L = 50 kin, r = L/2, andf0 = 2.34/•/2•-r.
{}From •rfmaxR/Ql•= 1 with/• = 3.2km/sandQ = 300. REFERENCES
IIForbothhorizontal
components,
fromBrady etal. [1971]
for Aki, K., Scalinglaw of seismicspectrum,J. Geophys.
Res.,72, 1212-
PacoimaDam andHudsonandBrady[1969]for Taft. 1231, 1967.
Aki, K., Earthquakemechanism,Tectonophysics, 13, 423-446, 1972.
SUMMARY AND CONCLUSIONS Andrews,D. J., Couplingof energybetweentectonicprocesses and
earthquakes,J. Geophys.Res.,83, 2259-2264, 1978.
The nature and magnitudeof variationsin the deviatoric Arias, A., A measureof earthquakeintensity,in SeismicDesignfor
stresses
existingalong activecrustalfault zonesare centralto a Nuclear Power Plants, edited by R. J. Hanson, MIT Press,Cam-
full understanding of thecauseandeffectof earthquakes, but bridge, Mass., 1970.
thesevariationsare at the presenttime only poorly under- Bakun,W. H., C. G. Bufe,andR. M. Stewart,Body-wave spectraof
central California earthquakes,Bull. Seismol.Soc. •4mer.,66, 363-
stood.Thesestresses maybewrittenin theform•(y) q-Aap(x, 384, 1976.
y), where •(y) is the averagetectonicstressat depthy and Berrill,J. B., A studyof high-frequency stronggroundmotionfrom
Aap(x,y) is the seismologicallyobservable stressdroppoten- the San Fernandoearthquake,Ph.D. thesis,Calif. Inst. of Technol.,
tial function.The constancy of earthquake stressdropsinde- Pasadena, 1975.
pendentof sourcedimension suggeststhatthespectral compo- Bouchon, M., A dynamicsourcemodel for the San Fernando earth-
quake,Bull.Seismol. Soc.•4mer.,68, 1555-1576, 1978.
sitionof Aa•(x,y) isof theformI A%•(k)[• k-'. IndependentBrady, A. G., D. E. Hudson,
andM.J.Trifunac, Earthquake •4ccelero-
supportfor this representationexistsin the generalvalidityof grams, vol.1,partC,Earthquake EngineeringResearch Laboratory,
the 'y = 2 model of the far-field shearwave displacement CaliforniaInstituteof Technology, Pasadena, 1971.
spectrum,underthe reasonableassumptions that the radiated Brune,J. N., Tectonicstressandthespectraof seismic
shearwaves,J.
Geophys.Res., 75, 4997-5009, 1970.
field of earthquakesis alsodrawn from the stressdifferencesof Brune,J. N., Correction,J. Geophys.
Res., 76, 5002, 1971.
A%(x, y) and that the numberof contributionsto the radiated Burdick,L. J., andG. R. Mellman,Inversion
of thebodywavesfrom
fieldat frequency f goesas(k/ko)2,consistent with the quasi- the BorregoMountainearthquaketo the sourcemechanism, Bull.
staticfrequencyof occurrence relationN • l/r •. Separately,a Seismol.Soc. •4mer., 66, 1485-1499, 1976.
variety of seismologicobservations suggeststhat the '• = 2 Bycroft,G. N., White noiserepresentation of earthquakes,
J. Eng.
Mech. Div. •4mer.Soc. Civil Eng., 86, 1-16, 1960.
model is the one generally,althoughcertainlynot always, Byerlee,J., Frictionof rocks,Pure•4ppl.Geophys., 116,615-626,1978.
applicableto the high-frequencyspectraldecayof the far-field Caputo, M., Model and observedseismicityrepresentedin a two-
radiation of earthquakes. dimensionalspace,•4nn.Geofis.,29, 277-288, 1976.
That theconstancy of earthquakestressdrops,b valuesnear Das,S.,andK. Aki, Faultplanewithbarriers:
A versatile
earthquake
model,J. Geophys.Res., 82, 5658-5670, 1977.
1, and the generalvalidity of the 'y = 2 model may all be Geller,R. J., Scalingrelationsfor earthquake sourceparameters
and
relatedto the samespectralcompositionof Aav(x,y) is a magnitudes, Bull. Seismol.Soc.•4mer.,66, 1501-1523,1976.
notable result, although there is as yet considerableuncer- Hanks, T. C., The faulting mechanismof the San Fernandoearth-
tainty, especiallyin an observational
sense,in relatingthese quake,J. Geophys.Res., 79, 1215-1229,1974.
Hanks,T. C., Earthquakestressdrops.ambienttectonicstresses,
and
phenomena
to a common
physical
origin,namely,I • stresses
that driveplatemotions,Pure•4ppl.Geophys.,
115,441-458,
k -:. If, however,thesephenomenaindeedshare a common 1977.
explanationin a tectonicstressof b(y) + Aav(x,y) existing Hanks,T. C., and D. A. Johnson,Geophysical assessment of peak
alongactivecrustalfault zones,whereAav(x,y) hasspectral accelerations, Bull. Seismol.Soc.•4mer.,66, 959-968, 1976.
composition
I k-', a possible isthatb Hanks,T. C., and W. R. Thatcher,A graphicalrepresentation
consequence of
seismicsourceparameters,J. Geophys.
Res., 77, 4393-4405,1972.
valuesdifferentfrom I would be accompaniedby '7 different
Hartzell,S. H., andJ. N. Brune,Sourceparametersfor theJanuary
from 2 for thoseearthquakescountedto determineb. Another 1975 Brawley-ImperialValley earthquakeswarm,Pure •4ppl.
is that changesin b valuesfrom 1 may be accompanied by Geophys.,115, 333-355, 1977.
earthquake stressdrops different from Aa and/or different Haskell,N. A., Total energyand energyspectraldensityof elastic
from the normativearea independence. waveradiationfrom propagatingfaults,Bull. Seismol.Soc.,4mer.,
54, 1811-1841, 1964.
Irrespectiveof thesepossibleassociations,the '7 = 2 model Heaton,T. H., andD. V. Helmberger,A studyof the strongground
in the presenceof anelasticattenuationsuggests that high- motion of the BorregoMountain, California, earthquake,Bull.
frequencystronggroundmotion hasa straightforwardinter- Seismol. Soc. •4mer., 67, 315-330, 1977.
pretation as band-limited, finite duration white noisein accel- Housner,G. W., Characteristics
of strong-motion
earthquakes,
Bull.
Seismol.Soc. ,4mer., 37, 19-31, 1947. •
eration. An estimateof armsis easily constructedfrom this Housner, G. W., and P. C. Jennings,Generationof artificial earth-
interpretation, and for several reasonsit appearsto be of quakes,J. Eng.Mech.Div. •4mer.Soc.CivilEng.,90, 113-150,1964.
potential importanceas a measureof high-frequencystrong Hudson,D. E., Response
spectrumtechniques
in engineering
seis-
2242 HANKS: FAULT MECHANICS

mology, in World Conferenceon EarthquakeEngineering,Earth- Seekins,L. C., and T. C. Hanks, Strong motion accelerograms of the
quake EngineeringResearchInstitute, Berkeley,Calif. 1956. Oroville aftershocksand peak accelerationdata, Bull. Seismol.Soc.
Hudson, D. E., and A. G. Brady, Strong Motion Earthquake Amer., 68, 677-689, 1978.
Accelerograms,vol. 2, part A, Earthquake EngineeringResearch Thatcher, W., and T. C. Hanks, Sourceparametersof southernCali-
Laboratory, California Institute of Technology,Pasadena,1969. fornia earthquakes,J. Geophys.Res., 78, 8547-8576, 1973.
Jennings,P. C., G. W. Housner,and N. C. Tsai, SimulatedEarthquake Tocher, D., Seismichistory of the San Franciscoregion, San Fran-
Motions, Earthquake EngineeringResearchLaboratory, California cisco Earthquakesof March 1957, Calif. Div. Mines Geol. Spec.
Institute of Technology,Pasadena,Calif., 1968. Rep., 57, 1959.
Johnson,L. R., and T. V. McEvilly, Near-field observationsand Trifunac, M.D., Stressestimatesfor the San Fernandoearthquakeof
sourceparametersof centralCalifornia earthquakes,Bull. Seismol. February 9, 1971:Main event and thirteen aftershocks,Bull. Seis-
$oc. Amer., 64, 1855-1886, 1974. mol. Soc. Amer., 62, 721-750, 1972a.
J. Geophys. Trifunac, M.D.,
Kanamori, H., The energyreleasein greatearthquakes, Tectonic stress and the source mechanism of the
Res., 82, 2981-2987, 1977. Imperial Valley, California, earthquakeof 1940, Bull. Seismol.Soc.
Kanamori,H., andD. L. Anderson,Theoreticalbasisof someempiri- Amer., 62, 1283-1302, 1972b.
cal relationsin seismology,
Bull. Seismol.Soc.Amer.,65, 1073- Wyss, M., and T. C. Hanks, The sourceparametersof the San Fer-
1096, 1975. nando earthquakeinferred from teleseismicbody waves,Bull. Seis-
Lahr, J., and P. W. Pomeroy,The foreshock-aftershock
sequence
of tool. Soc. Amer., 62, 591-602, 1972.
the March 20, 1966earthquakein the Republicof Congo, Bull. Wyss, M., and W. H. K. Lee, Time variations of the averageearth-
Seismol. Soc. Amer., 60, 1245-1258, 1970. quakemagnitudein centralCalifornia, in Proceedings of theConfer-
Richter,C. F., ElementarySeisinology,
768 pp., W. H. Freeman,San ence on 7kctonicProblemsof the San AndreasFault System,Stan-
Francisco, Calif., 1958. ford University Publications,Stanford, Calif., 1973.
Rikitake, T., Earthquakeprecursors,Bull. Seismol.Soc. Amer., 65,
1133-1162, 1975. (Received May 4, 1978;
Scholz,C. H., L. R. Sykes,and Y. P. Aggarwal, Earthquakepredic- revised November 3, 1978;
tion: A physicalbasis,Science,181, 803-810, 1973. acceptedJanuary 1, 1979.)