JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 96, NO.

BS, PAGES 8441-8452, MAY 10, 1991

On the Mechanicsof EarthquakeAfterslip

CmuSJ. MARO•E•
Departmentof Geology,Universityof Melbourne,Parkville, Victoria
and Division of Geomechanics,
CSIRO, Mt. Waverly,Victoria, Australia

C.H. SCHOLTZ

Lamont-DohertyGeologicalObservatoryand Departmentof GeologicalSciences,ColumbiaUniversity,Palisades,New York

ROGER BILHAM

CooperativeInstitutefor Researchin EnvironmentalSciences,Universityof Colorado,Boulder

We proposea modelfor earthquakeafterslipbasedon rate and statevariablefrictionlaws. In the model,

afterslipis attributedto the interactionof a velocity-weakening regionat depth(within which earthquakes
nucleate)with an upperregion of velocity-strengthening frictional behavior.The existenceof this upper
region is supportedby independentseismologicobservationsand the results of laboratory friction
experiments. In our model,afterslipis the resultof relaxationof a stressperturbationwithin the velocity-
strengthening region,which ariseswhenan earthquake propagates into that regionfrom below. We derive
the stressperturbation andits decayfrom the frictionconstitutive law usinga simple,1 degree-of-freedom
approximationfor the elastic interactionbetweenthe fault and its surroundings.This approximationis
basedon thickness-averaged displacements and slip velocitieswithin the velocity-strengthening region,
which is assumedto slip as a rigid block. Coseismicand postseismicslip are coupled through the
thickness-averaged stiffness k of the velocity-strengtheningregion. We assume k to be inversely
proportionalto the thicknessof this region, which meansthat thicker velocity strengtheningregions
have a greatertendencyto arrest coseismicslip. We model the afterslip-timehistoriesof the 1966
Parkfieldand 1987 Superstition Hills earthquakesand relatethe modelparametersto physicalparameters
which may govern the rheologic behavior of the faults. In accord with field observations,our model
predicts(1) that afterslipon somefaults scaleswith the thicknessof the (unconsolidated)sedimentary
cover and (2) that proportionallymore afterslipoccursfor earthquakes in which coseismicsurfaceslip is
smallcomparedwith coseismicslip at depth.Velocity-strengthening frictionalbehavioris to be expected
for faults within poorly consolidatedsedimentsand for thosethat containsignificantgougezones(about
>500 m) within their shallowregions(<3-5 km). Combiningour resultswith thoseof recentlaboratory
friction studiesindicatesthat relatively young faults with little accumulatedfault gouge should exhibit
little afterslip.

INTRODUCTION longer time scale, afterslip accumulatesroughly linearly with

log time. This behavior is remarkablyconsistentfor different
Surface slip from large earthquakes often continues
faults suggesting that afterslip represents a fundamental
subsequentto the earthquakefor a period of a year or more. physicalprocessor propertyof fault zones.
Such postseismicslip, or afterslip, was first documentedin
From the point of view of earthquakemechanics,we would
California following the 1966 Parkfieldearthquake[Smithand like to understandthe afterslip mechanismas well as its
Wyss, 1968] and has since been measuredfollowing several
significancein a broadercontext,for example,its relationship
other large earthquakes(1968 Borrego Mountain [Burford,
to the mechanismsof rupturenucleationand/orits influenceon
1972], 1975 Lake Oroville [Clark et al., 1976], 1976
the relativeproportionsof coseismicand postseismic slip for
Guatemala [Bucknam et al., 1978], 1979 Imperial Valley
a givenearthquake.A numberof modelsfor afterslipandfault
[Harsh, 1982; Sharp et al., 1982], and 1987 SuperstitionHills
creep events have been proposed[e.g., Scholz et al., 1969;
[Bilham, 1989; Sharp et al., 1989; Williams and Magistrale,
Nason and Weertman, 1973; Crough and Burford, 1977;
1989]). In detail, these measurements show that afterslip
Wesson,1987, 1988; Sharp and Saxton,1989]. Most of these
accumulates in discretepulses(so calledcreepevents)of a few
have focusedon creepevents,showingthat their steeponset
to 10 mm in amplitude,which decayin frequencyfollowingthe
and gradual decay is consistentwith a power law or quasi-
mainshock[e.g., Smith and Wyss,1968; Bilham, 1989]. On a
plasticrheology[e.g., Crough and Burford, 1977]. With the
exceptionof Wesson's[1987] work, afterslipcurveshavebeen
describedby empirical relations,with little attentionpaid to
1NowatDepartment of GeologyandGeophysics, University of the underlyingphysical mechanisms.
California, Berkeley. Wesson[1987] extendedan approachusedto modelcreep
eventsand showedthat the overall time dependenceof afterslip
Copyright1991 by the AmericanGeophysicalUnion.
couldbe explainedby a "two-phase"fault composed of locked
Papernumber91JB00275. patchesand patcheswith a quasi-plasticrheology.As with the
148-0227/91/91JB-00275505.00 models for fault creep, this rheology was assumedto result
8441
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8442 MARONEET AL.: MECHANICSOF EARTHQUAKEAFTERSLIP

from clay-richfault gouge[e.g.,Nason and Weertman,1973]. velocity-strengthening

characteristics
of such sediments,
AlthoughWesson'smodel fits afterslipcurves,it involvesad which was also the basis of their explanationfor the upper
hoc assumptions about the theology and the spatial seismicitycutoff.
distributionof coseismicand postseismicslip, and it doesnot Becausethis velocity strengthening
regionis a fundamental
explicitly couple coseismicand postseismicslip. The fault aspectof thepresentmodel,we summarize the laboratory
and
zoneis assumedto obeyone constitutivelaw duringcoseismic field observationsin support of it. (1) Both background
slip and a different law during postseismicslip. These microseismicityin Imperial Valley and aftershocksof the
limitationsmay be overcomeby usinga rate andstatevariable 1979 earthquake showthat seismicity is confinedbelow4-5
constitutiverelation for the fault zone [e.g., Rice and Gu, km [Doser and Kanamori,1986]. This depthcorresponds to a
1983]. In this case, coseismicand postseismicphenomenaare transitionfrom poorly consolidatedsedimentsto basement
directlycoupledthroughtransientstressing associatedwith the rocks [Fuis et al., 1982], indicating that seismicity is
earthquake. restricted to the basement rocks. Marone and Scholz [1988]
The purposeof this paper is to proposea mechanismfor summarizeseveralotherexamplesin whichan uppercutoffin
afterslipbasedon the rate and statevariablefrictionallaws. seismicity is observedalong faults that contain a well-
The model is motivatedby (1) the successof the rate/state developedgouge/brecciazone or are overlain by thick
variable constitutive laws in describing a wide range of sediments.(2) Coseismicslip duringthe 1979 ImperialValley
observations from laboratory friction experiments [e.g., earthquake decreased dramaticallyat about5 km [Archuleta,
Dieterich, 1979, 1981; Tullis and Weeks,1986; Marone et al., 1984], suchthat little to no coseismicslip occurredwithinthe
1990;Wongand Zhao, 1990] andnaturalfaults[e.g.,Rice and sedimentsabove 4-5 km. Although this combinationof
Gu, 1983; Tse and Rice, 1986; Marone and Scholz, 1988; detailedfault kinematicdataand crustalstructureis unique,such
Stuart, 1988; Okubo, 1989] and (2) by field and laboratory an upwarddecreasein coseismicslip is also indicatedby
observations(summarized below), which indicate that the "buriedslip", as commonlyinferredfrom comparisons of
upperregions(< 3-5 km) of manywell-developed naturalfaults seismicmoment and coseismicsurfaceoffset [e.g., Scholzet
exhibit velocity strengthening frictional behavior, i.e., al., 1969; Scholz, 1972;Agnewand Wyatt, 1989;McNally et
frictional resistanceincreaseswith slip velocity. al., 1989]. (3) Recent laboratory friction experimentson
simulatedfault gouge (unconsolidated sand) showvelocity
strengthening frictionalbehavior[Maroneet al., 1990],which
CHARAC'I'ERISTICS
OFEARTHQUAKE AFTERSLIP impliesinherentlystable,aseismic slip [e.g.,Gu et al., 1984].
Theseresultsare in contrastto experimentsshowingvelocity
Scholzet al. [1969] notedthat afterslipof the 1966 Parkfield weakeningbehaviorfor barerock surfaces or highlyindurated
earthquake wasrestrictedto the upper4 km andsuggested that gouge[e.g.,Dieterich,1979,1981;Biegelet al., 1989;Tullis
it was the stable (aseismic) equivalent of aftershocks,each et al., 1989]. However, the experimentson unconsolidated
arisingfrom brittle fractureof stressconcentrations produced gougeare a moreappropriate analogfor poorlyconsolidated
duringthe mainshock.Whereasaftershocks were concentrated sediments andfault gougewithinthe shallowregionsof active
at the endsof the ruptureandbelow 3-5 km, afterslipoccurred fault zones.

mainlywithin the centralregionandwasrestricted to the upper The velocitystrengthening characteristics

4 km. Crook [1984] made a similar observation for slip
subsequent to the 1979ImperialValley earthquake. He inverted
geodeticmeasurements [e.g., Crook et al., 1982] and found
that afterslip occurred only within the upper 4-5 km. In
Imperial Valley, this depthcorresponds to the transitionfrom
poorly consolidatedsediments(above) to basementrocks
below [Fuis et al., 1982], indicating that afterslip was
restricted to the sediments. A similar relationship has been
notedby Burford [1972] andWilliams and Magistrale[1989],
who reported that afterslip of the Borrego Mountain and
SuperstitionHills earthquakesscaledwith the local sediment
thickness.Furthermore,Burford's [1972] measurements and
those of Bucknam et al. [1978] show an inverse relationship containthick gougezones.Intraplatefaults,whichtendnot to
between coseismicand postseismicslip, with less afterslip
occurringin areasof larger coseismicslip. Taken together,
these observations imply a causal relationship between
afterslipand the presenceof poorly consolidated
sediments,as
originally suggestedby Smith and Wyss[1968].
Althoughseveralauthorshave sincenotedthis relationship,
[Burford, 1972; Sharpet al., 1989; Williams and Magistrale,
1989], only recently has there been an explanationfor its
underlyingcause and its significancefor the mechanicsof
afterslip.In connectionwith a modelto explainthe paucityof
very shallowseismicityon somefaults,Marone and Scholz
[1988] pointed out that the correlationbetweenafterslipand
poorly consolidatedsedimentscould be explained by the
of unconsolidated
material offer an explanationfor the above observations
[Marone and Scholz,1988]. Specifically,the uppercutoff in
seismicitycan be explainedas a transitionfrom unstable,
velocity-weakening frictional behaviorat depth to stable,
velocity-strengtheningbehaviorwithin the upper3-5 km.
This is to be expected for faults overlain by poorly
consolidated sediments and for those that contain thick
unconsolidatedgougezonesat shallowdepths.Alternatively,
this upper stability transition could be explained by low
normal stress at shallow depths [e.g., Brace and Byerlee,
1970]; however,the transitionis only observedin association
with faults within unconsolidated sediments or those that
have significantgougezones,do exhibit shallowseismicity
[Marone and Scholz, 1988]. In addition, the correlation
betweenthe upperseismicitycutoff in ImperialValley and the
contactbetweenbasementrocks and poorly lithified sediments
arguesfor a theologicalexplanationof the stabilitytransition.
Finally, because velocity strengtheningimplies negative
stressdrop, poorly consolidatedmaterialswill tend to arrest
dynamicrupture,in agreement with Archuleta's[1984] results
showinga rapid reductionin coseismicslip at the baseof the
sediments in Imperial Valley. The process of rupture
retardation results in a stress increase, due to velocity
strengthening.As we showhere,afterslipcanbe explainedby
relaxationof this stressperturbation.
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARONEETAL.:MECHANICS
OFEARTHQUAKE
AFTERSLIP 8443

TI-m MODEL

In accord with the above cited laboratory and modeling

studies,a fundamentalpremiseof the presentmodelis that the
fault rocksof the uppercrustexhibitrate and statevariable
frictional behavior.Thus, their steadystatefrictional strength
xss varieswith slidingvelocityV andcanbe written[seeRice
and Gu, 1983]

xss= x, + (A-B) In(V/V,). (1)

Here, A-B is the frictionrate parameterand x, is the strength

(a)
for steadystateslidingat velocityV,. Equation(1) is derived
from the constitutiverelationsgiven by Rice and Gu [1983],
which are compactversionsof thoseproposedby Dieterich
[1979, 1981] and Ruina [1983] to model laboratoryfriction
data. The constitutive relations are

In(V/V,) + B•, (2)

d• V
• = - D--•
[•q+In(V/V,)], (3)
wherev is a parameter characterizing
theevolvingstateof the
slidingsurfaceandD c is thecharacteristic
distance
overwhich
v evolves. At steady state • = -In[V/V,), which upon
substitution into (2) leads to (1). Transient stressing (b)
associated with ruptureof a fault whichobeysrate and state
Fig. 1. Two layer modelof the elasticlithospherein whicha velocity-
variable friction can be evaluatedby combining(2) and (3) strengtheningregion (A-B > 0) of thicknessh overlies a velocity
with an equationdescribingelasticinteractionsbetweenthe weakeningregion.The upperregionis thoughtto be 3-5 km thickand
fault andits surroundings[e.g.,Rice and Gu, 1983]. representsthe behavior of unconsolidated fault gougeforeccia or
In our model, we consider an elastic lithosphere which unconsolidated sediments.Earthquakescan only nucleatewithin the
containsa fault zone with a velocity strengthening (A-B > O)lower region. (a) Coseismicslip distributionshowingthickness-
averagedslipwithintheupperregion.(b) Crosssectionalongthefault
regionof thickness h overlyinga velocityweakening region plane; dashed lines show schematic contours of coseismic
(Figure1). Afterslipis theresultof stress
relaxation
withinthe displacement for an earthquake at depth.Note that coseismic slip is
velocitystrengthening arising suppressed
region;the stressperturbation within the velocity-strengthening region.
from the suddenincreasein slip velocity(from the preseismic
sliprate, Vo, to the coseismicsliprate,Vcs)associated with
seismicrupturepropagation into thisregionfrom below. where Vo is the load point velocity and k is the thickness-
To evaluatethe stressperturbationand relaxationprocess,we averaged stiffness for the velocity strengtheningregion
considera single-degree-of-freedom approximationfor the (Figure 1).
elastic interactionbetween the fault zone and its surroundings. The coupled equations(2)-(4) are solved numericallywith
From (1), the stressjump (Ax) producedby the increasein slip V•=V, to derive slip along the fault within the velocity-
velocity is Ax = ln(Vcs/Vo)(A-B). We assumethat stress strengthening region. This treatment neglects the stress
concentration within the upper region associated with
relaxation,and thus afterslip,is governedby the constitutive
approachof the ruptureto the free surface(as in a modelwith a
properties of the faultzones'upperregion(Figure1) andelastic
dislocationbelow a locked region). However, this is perhaps
interactionsbetween the fault and this region. The velocity-
justified if the velocity-strengthening portion of the plate
strengtheningregion is treated as a single block, and a
slips coseismically, at least in a thickness-averaged sense
thickness-averaged stressjump within thisblockis relatedto a
(e.g., Figure la).
thickness-averaged displacementthrougha stiffnessk=G/h,
where G is the shear modulus of the velocity-strengthening Although the simplistic approach used to derive k has
limitations,it (1) describescorrectlythe increasedcapacityto
regionandh is its thickness. Afterslipis thusdrivenby elastic
strainswithin the velocity-strengthening portion of the plate arrestcoseismicslip of thicker velocity strengthening regions
(Figure lb), and (2) yields an inverse relationshipbetween
(Figure 2). By using a thickness-averaged stiffness and
coseismicplate stiffnessand h, as would be predictedby an
consideringonly thickness-averaged slip within the velocity
elastic crack model in which, prior to rupture, shear stress
strengthening region,the problemis reducedto that of a one
within the plate is supportedprimarily alongthe portionof the
dimensionalspring-slider,with springstiffnessk, subjectto
the coseismicvelocity changeat the load point (Figure 2). fault that slips coseismically.
Elastic interactionsbetween the fault and its surroundingscan An alternativeapproachwould be to use the stiffnessgiven
be written by Tse et al. [1985] for partially lockedplate margins.In this
case, treating the velocity strengtheningregion h (Figure la)
dx as a slippedand unstressed regionoverlyinga lockedregionto
dt = k (gl-g) (4) a depthH gives k = (•G/2H) / ln[2/(l+cos(•h/H))] prior to
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8444 MARONE
ETAL.:MECHANICS
OFEARTHQUAKE
AFrERS•

Load
Point

vø
l
Slider

Fig. 3. Schematicdisplacement historyfor the seismiccycle as

predictedby thespring-slider
model(Figure2). Theloadpointvelocity
Fig. 2. Plan view showingthickness-averaged coseismicslip and increasesfromVo (long-term creeprate)to Vcs(coseismic
sliprate)at
thickness-averaged stiffnessk within the velocitystrengthening timeto for duration
tr, by analogy,
producing coseismic
slipat depth
region.
Displacement
ULcrepresents
coseismic
displacement
at the ULc. Theslider,whichrepresents
thethickness-averaged
surface
baseof the velocity-strengtheningregion(or at a point sufficiently response,
undergoes
coseismic
slipUScbeginning
attimet' > tr, and
remotefromthefaultplane).Thecoseismic surface displacement along
the fault is givenby the thickness-averaged
coseismic displacement postseismic
slipUp.Seetext
forfurther
discussion.
USc.To derivecoseismic
slip and afterslipfor the velocity-
strengthening regionwe usea single-degree-of-freedom spring-slider
system. A masslessslideris loadedthrougha springof stiffnessk by a to Vcsfor a periodtr (by analogy,the rise time or coseismic
pointthatmovesat velocityVo. The sliderexhibitsrate/state variable slip duration) resultingin slip of the loadpointULc (the
frictional behaviorwith velocity strengthening. By analogy,an coseismicdisplacement at depth). The slider undergoes
increase in loadpointvelocityrepresentsan earthquake andcoseismic coseismic displacementUSc(coseismicsurface offset)
slip at depth,whereasmotionof the sliderrepresents the (thickness-
averaged)displacement alongthe velocity-strengthening portionof
beginning at t'> to, whereUScis a function
of tr andbothUSe
[hefault.Because thesliderexhibits
velocity
strengthening,energyis andt' arefunctions
of thestiffness
k andA-B;thatis,USc
may
stored within the spring during coseismicslip and releasedvia be--0 for a sufficiently
compliantspringor smallvalueof tr.
afterslip.In our model,k scalesinverselywith the thicknessof the At t = to+tr,the loadpointvelocityreturnsto Vo (coseismic
velocity-strengtheningregion.
slip ends)andthe elasticstrainstoredwithinthe springis
releasedvia afterslip,
Up.Thisstrainarises
fromtheincreased
frictionalresistanceassociated
with velocitystrengthening
coseismicslip [Tseet al., 1985].With thisrelation,(1) k is (e.g.,Figure4), asindicated
by equation
(1). Because
themodel
inversely
proportional
to h and(2) thepresence
of theslipped
regionnearthetopof theplatetendsto arrestdynamic rupture.
ignores
inertia
andcontains
nodissipative
terms,
U•,+USc-->
However, Quin's [1990] dynamicmodelingof the 1979 ULc ast -->oo.Coseismic andpostseismic sliparecoupled and
ImperialValleyearthquake
shows thata negative stressdropis inversely related.
For a sufficiently
stiff spring,t' -->to, and
requiredin the upperregion.The coseismic displacementsthesliderundergoes coseismicdisplacement USc=ULc . From
above5 km are overestimatedby at leasta factorof 2 if the our relationfor k, this is equivalentto a vanishinglythin
upperregionhaszero stressdropas it wouldfor the slipped
region envisaged by Tse et al. [1985]. Furthermore, the
stiffness given by Tse et al. [1985] differs from that we
proposeaboveby only a factor of 2-3 for h of 3-5 km and vo
H=15 km and thus given its other limitations does not
representa significantadvantageover the simplerform.
Clearly,a moredetailedmodelis neededto fully accountfor !
along-strikevariationsin slip and three-dimensional
elastic
interactions
betweenthe fault andits surroundings.
However,
thisis beyondthe scopeof the presentwork.

Spring-Slider
Systemas an Analogfor theSeismic
Cycle Displacement
We may regardmotionof the spring-slider systemas an
Fig.4. Variationin frictionalresistancewithslipvelocityfor velocity
analogto slip duringthe postearthquake
portionof the seismic
strengthening.
A•:is the steadystateincrease in frictionfor a velocity
cycle,with the loadpointrepresentingthevelocityweakening jump from Vo to Vcs.D c is the characteristic distanceover which
regionand the slider.representingthe velocity-strengtheningfrictiondecaysfollowinga change in loadpointvelocity.tr is the
region.Considerthe displacement historyshownin Figure3. coseismic slip duration,andt' is theonsetof coseismic slipfor the
At timeto,theslipvelocityof theloadpointincreases
fromVo slider(velocity-strengthening
region).SeeFigure3.
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARONE ET AL.' MECHANICSOF EARTHQUAKEAFtERSLIP 8445

velocity-strengthening region, in agreement with the Figure 6, which showsthe relative proportionsof coseismic
observationof large coseismicand small postseismicslip in and postseismicslip versush for differentvaluesof tr anda-b.
areaswhere the alluvial coveris thin [e.g., Burford, 1972]. ConsiderFigure 6a, the total slip of the slider 1 year after the
To illustratethe behaviorof afterslipproducedby the model, mainshock (by analogy, coseismic surface offset plus
we show a model afterslip curve along with data from two afterslip) is composedalmost entirely of coseismicslip for
earthquakes(Figure 5). The numerical afterslip curve is not small h, since this correspondsto a very stiff spring in the
intendedas a direct simulationof the data. Rather we present spring-slider model. With increasing h, coseismic slip
the curve to show its generalform for sometypical parameters. diminishes and afterslip increases,in agreementwith field
The model was calculated using equations (2)-(4)and the observationsof reduced coseismicsurface slip and increased
following values: h=3 km, G=10 GPa, effective normal stress afterslipwheresedimentsare thicker[e.g.,Burford, 1972].
gradient(•n') =15 MP•, Vo=2 mm/yr, Vcs=0.2m/s, tr=0.5
s, D c=10 mm, and a-b=O.005 where a-b=(A-B)/on'. The tr = 1.o s, a-b = 0.005
:i?---------------•--':-----•ñ.
.. -•--::•••_
_-=e•__ (b)
preseismicslip rate Vo is characteristicof central or southern
California where V o is generallyone tenth or lessof the long-
term plate velocity [e.g., Schulz et al., 1982; Louie et al.,
1985]. The valuesof Vcsandtr are consistent with Archuleta's
[1984] modeling of the 1979 Imperial Valley event. The
constitutiveparametersare the laboratoryvaluesof Marone et
al. [1990], except for Dc, which previousmodeling indicates i
0
i
1
i
2 3
i I
4 5

must be much larger for natural faults than laboratorysamples tr = 1.0 s, a-b = 0.01
[e.g., Tse and Rice, 1986; Scholz, 1988; Stuart, 1988; !-::-----:--•----•-=--•=:••=_
':--'
'- _=_-=-___-____ (d)
Lorenzetti and Tullis, 1989]. In general, the model afterslip
curve is similar to the measureddata, showinga steeponset
followed by gradual decay 100-300 days after the mainshock.
Varying individual parameters shows that afterslip is
insensitiveto D c (postseismicdisplacementafter 1 year varies
by < 5% for Dc in the range 10 I.tm to 10 mm if coseismicslip
0 1 2 3 4 5
UScis greaterthan 2-3 timesD c) but dependsfairly stronglyon
k, tr, a-b, and the ratio Vcs/Vo.
It is instructiveto considervariationsin model afterslipwith Fig. 6. Model calculationsof the relative contributionsof coseismic
parameters such as coseismic duration, the constitutive and postseismic
slip to total surfaceoffset(Us) 1 yearafteran
parametera-b, and the thicknessof the velocity strengthening earthquake.
Usisnormalized
bytheslipatdepth
(uL).(a)and(b)The
region (or k, sincek=G/h). For given values of a-b, k, and effectof coseismicduration(tr) for a givena-b. (c) and(d) The effectof
increasing a-b. In each case, thinner velocity-strengthening regions
Vcs/Vo
theratioof coseismic
surface
offsetto afterslip
(USc/Up) produce a greaterproportionof coseismicslip relative to afterslip.For
should
increasewithrupture
duration
or ULc(roughly speaking,
a given value of a-b, coseismicslip increaseswith ruptureduration
earthquakestressdrop), sinceoncethe sliderreachesa steady (Figures 6a and 6b). Coseismicslip and the total slip within 1 year
state frictional resistanceduring coseismicslip its velocity is decreasefor greatervelocity strengthening(Figures6a and 6d).
Vcsandthusoffset(USc)accumulates rapidly,while the amount
of afterslipis fixed by a-b, k, andVcs/Vo.This is illustratedin
For the model parameters of Figure 6a, velocity
strengtheningregions thicker than 1.6 km do not experience
800 coseismicsurfaceoffset.For large h, afterslipdecreases and the
7OO ß '•'--•6 Parkfield total slip of the slider at 1 year is significantly less than that
of the load point (by analogy,slip at depth). This is simply
600

,• 500-
f'• ..............._..•_-
1976
Guate•la becausethe stiffnessbecomesvery low; the remainingslip
would occur as postseismicslip after 1 year. For longer rise
timetheratioUSc/Up
increases
for smaller
h, and,in general,
a
a, 400-
greaterproportionof the total slip occurswithin the first year
• 300- after the mainshock (Figure 6b). Increased velocity
strengthening(larger a-b) resultsin a smaller amountof both
200-
coseismicand postseismicslip (Figures 6c and 6d). Thus the
100' total slip at 1 year decreaseswith the degree of velocity
strengthening.
o
0 1•0 2•0 3•0 4•10 500
Days After Mainshock Analytical Approximationof the Afterslip Model

Fig. 5. Relativeafterslipversustimefor thenumericalmodelandtwo Although the spring-slider system reproduces measured

earthquakes.The model is calculatedusing typical earthquake afterslip and is useful for examining the mechanicsof our
parameters (seetext)andlaboratory-derived valuesfor the constitutive model, it requires a numerical solution. This is troublesome
parameter(a-b). The generalcharacteristics of the modelcurveagree from the point of view of fitting the model to data and
well with the data,eachshowinga steeponsetfollowedby a gradual
decayin slip rate.Parkfielddataare from TaylorRanch[Smithand recoveringparametersthat might be usedto evaluatethe model
Wyss,1968].GuatemaladataarefromZacapa[Bucknam et al., 1978]. (e.g., the constitutive parameters,which may be compared
The line fit to the Guatemaladatais thatgivenby Bucknamet al. with laboratoryvalues,or the depthh, which may be compared
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8446 MARONE ET AL.: MECHANICS OF EARTHQUAKEAFTERSLIP

with geologicor seismicallyderivedestimates).We therefore postseismicslip rate to reachthe long-termcreeprate [Nason,

introduce the following closed-form solution for afterslip 1973; Schulz et al., 1982; Wesson, 1987], so at least in this
(given by Scholz [1990])' casethe omissionof the Vot term is not significant.We may
conclude,then, that from the point of view of fitting available

Up=A-Bl•lkV•s}t+
1]+V
ot.
k L• A-B
afterslip data, the numerical model and equation (5) are
(5) essentially identical.
A derivationof (5) is givenin the appendix;the symbolshave COMPARISON OF THE MODEL WITH DATA
the samemeaningas abovewith the exceptionof VScs, whichis
the thickness-averagedcoseismic slip velocity within the One way to evaluate our afterslip model is to treat the
velocity strengthening region.Note that VScsis expectedto be parametersin equation(5) as free variablesand determinehow
less than Vcs [Archuleta, 1984]. As in the numericalmodel, well the relation fits afterslip data. However, in addition,
equation(5) is basedon rate/statevariablefriction and the idea because(5) is derived from a physical model, we can also
that afterslipis the resultof relaxationof a stressperturbation comparethe values so obtainedwith independentestimates;
producedby dynamicrupture propagationinto a velocity- i.e., laboratorymeasurements in the caseof a-b or seismicdata
strengtheningregion. This relation is valid if the velocity- in the caseof Vcs and h. If the model is to be considered
strengtheningregion reaches a steady state friction level successful, the valuesderivedfrom a leastsquaresfit to thedata
during coseismic slip (i.e., for a thickness-averaged slip must agree with the constraintsimposed by independent
greaterthanDc, asshownin Figure4). measurements.

We may evaluatelimitationsof the analyticalapproximation We fit afterslip data from two earthquakesusing a "two-
by comparingequation(5) with the numericalmodelfor some variable"versionof equation(5):
typical parameters(Figure 7). To simplify the comparison,we
make use of the following: (1) much of the availableafterslip
datafall within the range 1 day to 1 year, (2) the measurements Up= ct t+l
are often relative to some fixed points installed after the
mainshock,and (3) due to creepevents,afterslipmeasurement whereo•=(A-B)/kand[•=VScs. We firstconsider afterslip
datafor
can only be consideredaccurateto +4 mm for the purposesof the 1966 Parkfield earthquake.Data from five siteswere fit, in
fitting the overall behavior;that is, a creepevent couldhave a leastsquaressense,usingan iterativ.e nonlinearcurve-fitting
occurredjust prior to or just after the measurement. Given these technique(Table 2). The data and site locationsare from Smith
limitations, we need only compare equation (5) and the and Wyss[1968]. The sitesare listed in Table 2 from southto
numericalsolutionfor t <1 year and for relativeafterslip.(Note north; Calx Ranch is 15 km north of the southernmostextent
that the primary reason for this restrictionis to simplify the of surfacerupture, and ClassenRanch is at the northernmost
comparison;the analyticalapproximationagreeswell with the end of surfacerupture(21 km northof CalxRanch).The errors
numerical solution after 1 year and in an absolute sense.) indicatereasonably certainvaluesfor c• andlesscertaintyin [•.
Figure 7a shows that equation (5) reproducesthe numerical The large relative uncertaintyin the values.forClassenmay be
model well.
attributedto having fewer data points there (7) than for the
Similar comparisonswere made for a rangeof the parameter other sites (10-12) and becausemeasurements beganthere 50
values a-b, D c, h, etc. (Table 1). An exampleis shownin days after the mainshock,comparedto 2-10 daysfor the other
Figure 7b, which showsthe effectof varying(a-b) from 0.005 sites. Curves generatedusing equation(5) and the data from
to 0.015. The curveshave been offset and plottedversuslog Table 2 showgoodagreementwith the afterslipdata (Figure8),
time to allow comparisonof the afterslip rate. As with the and thus at the least, equation (5) is capable of describing
lower a-b value, the closed form solution reproducesthe measuredafterslipdata.
We may now comparethe model parameters(Table 2) with
numerical solution well. For the range of a-b values
independentestimates.From the aboverelation,ct=C•n'(a-b)/k,
considered,the maximumdifferencein relativeafterslipfor t <
which for a given lithostatic gradient results in a relation
1 year is 5 mm; thisoccursfor a-b=0.005 andrepresents about
between a-b and k or the thickness of the velocity
2% of the slip that occurredbetween1 day and 1 year. Table 1
also showsthat varying the characteristic friction distanceD c strengthening region (Figure 9). The a-b valuesare plottedon
does not degradethe agreementbetweenthe two models.The a log scale to expand the region 0.001-0.005, which
only case in which significantdisagreementarisesis for small representsthe range of relevant laboratory measurements
values of h, and this is due to the lack of a term for continued [Marone et al., 1990]. For siteswithin the Parkfield area, the c•
fault creep[in additionto afterslip]in equation(5). That is, for value for the laboratoryrange of a-b indicatesa 2-4 km thick
thin velocity-strengtheningregions, the thickness-averaged velocity strengtheningregion, in agreementwith estimatesof
stiffness (k=G/h) becomes large, resulting in very little the depth distribution of afterslip (= 4 km [Scholz et al.,
afterslip. In the numerical model, this meansthat the slider 1969]) and an estimate of h based on the upper cutoff in
resumesslipping at the long-termcreeprate within a few tens seismicity(3-4 km [Marone and Scholz,1988]).
of days, whereas(5) showsessentiallyno slip after this time. The modelingalso providesan estimateof the VScs, via [•,
The differencebecomessignificantfor h < about1 km (using which is the thickness-averaged coseismicslip velocity within
the other values indicated in Table 1) and thus it may be the velocity-strengtheningregion. However, evaluation of
appropriateto includea term Vot in equation(5) for fault zones equation (5) shows that afterslip is only sensitiveto VScs
with little overlying sediment or which contain little duringthe first few hoursto 1/2 day following the mainshock.
unconsolidatedgougeforeccia.We note that for the central After 1 day the relative afterslipis insensitiveto VScs . Thus,
sectionof the San Andreasfault it takesabout1-2 yearsfor the unless the absolute offset is known (which it is not for the
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARONEET AL.: MECHANICSOFEARTHQUAKE
AYIERSLIP 8447

TABLE2. Afterslip
Modeling;
1966Parkfield
Earthquake
0•, [5,
450- Numerical
M"odel.••••••
5__ _
CarrRanch
mm
27.64
mm/day
190.6
400- - Peacock Ranch
+ 1.56
38.03
+ 53.5
91.5
+ 0.79 + 8.9
350 - Taylor Ranch 39.38 162.7
+ 0.51 +9.9
Parkfield City 36.94 107.4
:• 300- Classen Ranch
+ 1.06
50.08
_+ 13.9
7.8
25O - - + 5.64 + 2.2
(a-b)=0.005 Valuesfor least squaresfit of equation(5) to afterslipdata;
200 i I I uncertainties
are 1 s.d.,o•=(A-B)/kand13=VScs.
DataarefromSmith
0 100 200 300 4O0 and Wyss [ 1968].

Days After Mainshock Figure 4, ax representsthe excess shear stress arising from
I I
increased slip velocity and velocity strengthening. In
900
connectionwith the modelingof Quin, At arisesas a negative
800 stressdrop within the upper 4-5 km, which is required to
(a-b)=0.015 reproducethe slip distributionof Archuleta [1984]. That is, in
700
600
300I , , , •
500 I • Taylor
400
-" (a-b)=0.005
300 200 -

100 ..m 150 -

0 I I

1 10 100 .• 100
log(DaysAfter Mainshock)
Fig. 7. Comparison of the numerical and analytic solutionsfor
afterslip.(a) The relativeafterslipfrom 1 dayto 1 yearis plottedand 0 rio 360
showsgoodagreement(seeTable 1 for theparametervaluesused).(b)
The datafrom Figure7a are offsetto facilitatecomparison andplotted Days After Mainshock
versuslog time. Increasingthe value of a-b does not degradethe Fig. 8. Comparisonof afterslipmeasurementswith curvesgenerated
agreementbetween the two solutions. usingequation(5) andthe parametersgivenin Table2. Errorbarsare
_+4mm. Data from Smithand Wyss[1968].

Parkfield event) or relative measurementsbegin immediately

1966 Parkfield Earthquake
after the mainshock,the 13valuesare unreliable.The valuesin
Table 2 are significantlylower than thosegiven by laboratory J. C3.assen
experimentsor seismic modeling,which are 0.01-0.5 m/s -- Parkfield

(8.64x102-4.32x104mm/d)[Johnson
and Scholz,1976;
Okubo and Dieterich, 1984; Archuleta, 1984]. Nonetheless,it
is worth noting that VScsprovides an estimateof ax, via
equation(1), which may be comparedwith that derivedfrom
dynamicrupturemodeling[e.g., Quin, 1990]. As indicatedin

TABLE 1. Comparison
of NumericalandAnalytical Solutionsfor
RelativeAfterslip
L, Vo, h, (a-b), o , 2 ; ' ; ' ;
mm mml}rr km Thickness (h) km
Range
Studied 10-50 0.2-20 0.5*-5.0 0.005-0.015 Fig. 9. Plotsof the constitutive
parameter
a-b versusthickness
of the
Max. Misfit 3 mm, <1% 2 mm, <1% 1 mm, <1% 5 mm, 2%
velocitystrengtheningregion.The curvesaredefinedby theparameter
Misfit is maximum differencebetweenequation(5) and numerical 0•,whichis derivedfromfittingequation(5) to afterslipdata(seeTable
modelfor times< lyear, this differenceis alsogivenas a percentageof
relative slip between 1 day and 1 year. Each parameterwas varied 2). Laboratorymeasurements of a-b are in the range0.001-0.005
independently, with the othersheld at L=I 0 ram, Vo=2 mm/yr,Vcs=O.1 (dashedlines), and thus the curvesindicatea velocity strengthening
m/s,h=2.5kin,a-b---0.005,On'=15MPa/km. region2-4 km thick, in goodagreementwith independent estimates.
*Forh < lkmthecomparison wasmadeovertheafterslip duration The plot is made by substitutingk=GIh in the definitionof 0• (see
(and not 1 year), sinceafterslipgives way to (long-term)fault creep text),whichafterrearrangingyieldsa-b = (o•G)I(on' h); we useon' =
before 1 year. 15 MPa/km.
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8448 MARONE ET AL.' MECHANICS OF EARTHQUAKEAFYERSLIP

orderto reproducethe sharpupwarddecreasein coseismicslip,
the dynamic modelingrequiresa retardationof rupture,i.e., a
negative stressdrop, which Quin's modelingindicatesis 2-3
MPa. This can be comparedwith the initial driving force for
afterslip ax arising from the stressincreaseassociatedwith
rupture into a velocity strengthening region.Using the VScs
valuesgivenby our modeling(Table 2) andthe a-b valuesfrom
Figure 9 and assumingVo= 2 mm/d, we obtaina relationship
between Ax and h (Figure 10). Our Ax estimatesare about a
factor of 2 lower than thoseof Quin [1990].
A similar analysiswas madefor slip subsequent to the 1987
SuperstitionHills earthquake(Table 3). Unlike at Parkfield,
reliable measurements of absolute surface offset are available
from several sites, and investigators combined these
measurementswith measurementsof continuing afterslip
[Sharp et al., 1989; Williams and Magistrale, 1989]. Because
of the large offsets measuredonly a few hours after the
mainshock, equation (5) was modified to allow for the
possibilityof coseismicsurfacerupture.Thus, in additionto
fitting the data with equation (5), we fit the data with the
relation

1966 Parkfield Earthquake

-- Cart Up-UcS
+ 1 . t+ l ,
---O-- p• (6)
ParldieM
whereUScis thecoseismic surfaceruptureand0•' and[•' havethe
Taylor
samemeaningas 0• andI• (notethe distinction, sincein fitting
thedatawith USc,•t, and[• thelatterparametersareexpected to
differ from those in equation(5)). As above,the parameters
were obtained from a best fit to the data and the values were
comparedwith independentmeasurements.
Comparisonof the parametersfor fits with and without the
coseismicslip term (compare0•', [•' with 0• andI•, Table 3)
. , ! ß ! ,. ! . shows that the c• term changesvery little but that [• is
1 2 3 4 5
significantlysmallerwhen the coseismicslip term is included.
Thickness (h) Km This is in accordwith the finding that [• is mostsensitiveto
with dynamicrupture slip during the first 1/2 day or so, and thus controls the
Fig. 10. The stressperturbation(Ax) associated
into a velocity strengthening
regionis plottedversusthe thicknessof absolutelevel of the afterslip curve whereas0• controlsthe
that region. The plots are made using the relationAx = a-b On' time dependenceof slip after about 1 day. Including the
ln(VScs/Vo).Thevalues of VScsanda-b aregivenbyfittingequation coseismicslip term substantiallyimprovesthe fit to the data
(5)to afterslip
measurements, notethatVScs= 13(seeTable2).Weused (Figure 11), however,in somecases(e.g., site 2U, Figure 11)
thea-b valuesfromFigure9 andVo = 2 mm/d. significantmisfitremains.As discussed below,thismaybe the

TABLE3. Afterslip
Modelinly;
1987Superstition
HillsEarthcluake
Site or, I•, or', I•', ffc, Distance
(N),
mm mm/day mm mm/day mm km
124' 18.14 5.85x103 22.30 8.0 108.4 1.5
+ 0.83 + 2.22x103 + 0.70 + 1.3 + 2.2
156' 54.25 3.29x104 53.45 302.2 240.8 7.2
+ 0.70 + 3.93x103 + 1.08 + 264.0 + 42.3
168' 63.06 1.14x105 64.93 224.3 383.8 9.5
+ 0.80 + 1.43x104 + 0.74 + 42.0 + 11.9
1Rt 60.45 6.86x104 62.77 623.6 271.6 11.8
+ 3.61 + 3.83x104 + 2.44 + 508.4 + 42.2
2Mr 60.71 4.98x103 65.70 140.1 194.1 15.5
+ 1.40 + 7.93x102 + 1.80 + 47.3 + 15.8
2T9 61.55 3.46x104 65.69 518.1 237.1 18.6
+ 0.67 + 2.67x103 + 1.25 + 115.3 + 11.3
270* 59.69 3.54x104 64.60 156.9 303.1 18.6
+ 1.17 + 5.75x103 + 0.88 + 26.0 + 7.9
271' 73.46 1.05x104 78.73 107.7 316.1 19.0
+ 5.42 + 5.23x103 _+2.12 + 38.0 + 21.6
2U• 65.39 6.84x103 73.67 249.2 181.5 19.1
+ 1.63 + 9.53x102 + 3.29 + 105.0 + 20.2
277* 35.07 1.05x104 37.53 52.7 176.3 21.5
+ 0.96 + 2.25x103 +0.43 + 10.4 + 6.3
284* 21.57 1.96x103 21.72 294.2 40.1 22.5
+ 0.33 + 2.06x102 +0.77 + 1127.4 + 79.4
Valuesfor leastsquaresfit of equations(5) (0•and13)and(6) (off,13',andUSc)to afterslipdata;uncertainties
are 1 s.d.Thedistance
fromthenorth
endof the surfaceruptureis givenfor eachsite;totalsurfacerupturelengthis 24.5 km [Sharpet al., 1989].
*Data from Sharpet at. [1989].
pData from Williamsand Magistrate [ 1989].
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARONEET AL.: MECHANICSOF EARTHQUAKEAFrERSLIP 8449

1987 SuperstitionItills consistentwith reduced slip at the ends of the rupture. The
I I
800 difference between • and •' and the errors in each are small
relative to differences between sites, and thus including the
coseismicslip term does not significantlyalter the values of
a-b or h (Figure 12).
As indicatedabove, the afterslipcurvesare very sensitiveto
[5within the first few hoursto day, and thuswe expectthat
• 400' includingUScresultsin a significantdecrease in [5 (Table 3).
ß• 300- ß IR The thickness-averaged coseismicslip velocitiesgiven by [5
(withoutthe UScterm) are in goodagreementwith Archuleta's
200' [1984] kinematic modeling for the nearby 1979 Imperial
Valley earthquakeand laboratorymeasurements [Johnsonand
1oo
0.1 015 $•0 $(•.0 500.0 Scholz, 1976; Okubo and Dieterich, 1984]. However, given
Days After Mainshock the data in supportof coseismicsurfacerupture [Williams and
Magistrale, 1989; Agnew and Wyatt, 1989] and the
Fig. 11. Comparisonof afterslipmeasurements with curvesgenerated
usingequations(5) (solidline) and (6) (dashedline) andtheparameters improvement our fits showwhenUScis included(Figure11), we
given in Table 3. Note that the fit is improved by including the favor using[5for VScs,even thoughthe valuesare somewhat
coseismic
sliptermUSc(dashed
line).Errorbarsforthedataareonthe lower than those of Archuleta [1984]. As above, these values
orderof the symbolsize.Data (and sitereferences)from Williamsand can be usedto estimateA• (Figure 13). The data for h-2-5 km
Magistrale [ 1989].
show good agreement with those given by Quin's [1990]
dynamic modeling.
result of slip triggeredby the Elmore Ranch earthquakein
additionto afterslipfrom the SuperstitionHills event.
Figure 12 showsthe a-b valuesimpliedby the ct values for 1987 Superstition Hills Earthquake
the Superstition Hills event (Table 3). For the range of 156
laboratory a-b values, the modeling indicates a velocity
strengthening
region2.5-5km thickfor siteswithinthe 4 168

central portion of the surfacebreak, in excellent agreement 2M

with the seismic refraction work of Fuis et al. [1982], which
271
shows a 3-5 km thick layer of sediments overlying the
277
Superstition
Hills
fault
and
the
area
tothe
east.
The
• values
• 3
are lower for sites near the end of the surface break, which is

2
1987 Superstition Hills Earthquake

156

ß i ß ! ß I ß

1 2 3 4 5

Thickness (h) Km
.001 Fig. 13. The stressperturbation(Ax) associated with dynamicrapture
into a velocity strengtheningregion is plotted versusthe thicknessof
that region. The plots are made using the relation Ax = a-b On'
ß ! ß s '. ! ß ' ! ß s ln(Vcs/Vo).Thevaluesof Vscsanda-b aregivenby fittingequation
0 1 2 3 4 5 (5)to afterslip
measurements,
notethatVScs= [5(seeTable3). Weused
Thickness (h) Km the a-b valuesfrom Figure11 andthe preseismic
slip rateVo measured
by Louie et al. [1985], 0.5 mm/yr.
Fig. 12. Plotsof the constitutive parametera-b versusthicknessof
the velocity strengtheningregion. The curves are defined by the
parameter o;,whichis derivedfromfittingequation (5) to afterslipdata DISCUSSION
(seeTable 3). Laboratorymeasurements of a-b are in the range0.001-
0.005 (dashed lines), and thus the curves indicate a velocity Any relation to describethe natureof afterslipaccumulation
strengtheningregion 2.5-5 km thick, in good agreementwith with time has to derive from some form of viscoustheology
independent estimates. The plot is madeby substituting k=G/h in the [e.g., Wesson, 1988]. In the present modeling we employ
definition of o; (see text), which after rearrangingyields: a-b = [o• rate/state variable constitutive laws, on the basis of their
G)/(on' h); weuseon'= 15MPa/km. successin describing laboratory and field observationsof
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8450 MARONE ET AL.: MECHANICS OF EARTHQUAKEAFtERSLIP

frictional phenomena. In the context of this class of The modeling also provides an estimateof the thickness-
constitutive laws, it is important to note that afterslip can averagedcoseismicslip velocity and offset (Tables2 and 3).
only occur on faults that contain a region of velocity- The values of coseismicoffset obtainedfrom fitting data with
strengtheningbehavior. For a fault zone that exhibits only equation(6) (Table 3) are in good agreementwith thoseof
velocity-weakening behavior, the steady state frictional Agnew and Wyatt [1989], which were derivedfrom coseismic
resistancedecreaseswith slip velocity, eliminatingthe stress strain measurements, and the earliest field measurements
transientneededto drive afterslip.One could arguethat stress [Williamsand Magistrale, 1989]. The VScsvaluesare somewhat
transients could be produced with velocity weakening if morecomplicatedbecausetheytradeoff with USc(equations (5)
coseismicslip were small comparedwith D c (i.e., if slip ceased and (6) and Table 3) and becausedata at very early times are
during the initial stressincreaseupon an increasein velocity, unavailable.Differentiating equation(5) with respectto time
Figure 4), however,D cis thoughtto be on the orderof 10 mm, and setting t=0 shows that, as expected, the afterslip rate
whereascoseismicoffsets are at least an order of magnitude shouldequalthe coseismicslip rate at t=0. With or withoutthe
larger for earthquakesof magnitude> 6. Changesin normal coseismicslip term, our th.•ckness-averaged valuesof VScsare
stressalong a fault duringcoseismicslip might alsocontribute considerablylower than thoseof Archuleta [1984]. This may
to afterslip; however, this would require a very specific be due to the very shortdurationof coseismicslip at the surface
ambient stress field and fault orientation, which seems or to inertial effects not taken into accountby our model.
unlikely in view of the generalityof afterslipbehavioramong Becauseof the proximity of the magnitude6.2 Elmore Ranch
different faults. earthquake,only 12 hoursprior to the Superstition Hills event,
The model presented here is supportedby its ability to we consideredthe possibilitythat the misfit for site 2U (Figure
explain commonobservationsrelating to afterslip.One such 11) was the resultof slip triggeredby the ElmoreRanchevent.
observationis that of buried slip, or delayed afterslip,suchas Modeling with a more complex relation to account for two
reportedfor the 1966 Parkfield earthquake[Smithand Wyss, overlappingepisodesof afterslip showedimprovedfits, but
1968; Scholz et al., 1969]. The results shown in Figure 6 not sufficientlybetter to warrantdetailedstudy.
indicate that for typical values of rise time and the constitutive
parameter a-b, coseismic surface slip is not expected for Creep Events
velocity strengtheningregions thicker than 1-2 km. Thus, Although we have focused on the overall behavior of
since our modeling indicates that h = 2-4 km, it is expected
afterslip, in detail, afterslip is composedof discreet creep
that coseismic slip will not break the surface, resulting in events.While the presentmodel cannot addresssuchfeatures
buried slip. directly, it is worth askingwhetherthey can be understoodin
A secondobservationis the relationshipbetween coseismic the context of rate and state variable friction laws. The
slip and afterslip and the correlation between the depth to questionhasbeen addressed in the work of Rice and Gu [ 1983,
basementand afterslipmagnitude,suchas reportedby Burford p. 209], who in referring to "..transientslip motion, which
[1972], Bucknam et al. [1978], and Williams and Magistrale speeds up, relieves stress, and gradually relaxes without
[1989]; Williams and Magistrale also review a number of seismic expression.." report that "..this sort of transiently
earlier examplesin southernCalifornia. This is predictedby accelerated but aseismic fault slip is the only type of
the model, as shown in Figure 6. Areas of locally lower h interestingaftereffectpossibleon fault segmentswith velocity
experiencegreatercoseismicslip and lesserafterslip,for a strengthening..". Rice and Gu's [1983] Figure 9b showsan
given earthquake.In southernCalifornia,h is controlledby the acceleratedslip rate following a suddenincreasein stress,such
thickness of the unconsolidated sediments, which overlie the as producedby the passingof a rupturefront within a velocity
basement,and Williams and Magistrale'sdata combinedwith strengtheningregion.
the crustal structure of Fuis et al. [1982] show a good In connectionwith creep events, a similar initial increasein
correlationbetween afterslip and h. In otherregions,suchas stress/sliprate would result from localized failure within the
central California, the velocity strengthening region consists velocity strengtheningregion, producinga small incrementof
of a zone of unconsolidatedgouge within the fault. In these aseismicslip. The magnitudeof suchcreepeventsis controlled
cases, along-strikevariations in h may be controlledby the by the size of the failed region, which dependsupon the local
pinchandswellof the gougezonedueto roughness of thefault stiffnessand the constitutiveparametersa-b andDc. Bilham's
[e.g., Scholz and Aviles, 1986; Power et al., 1987], since [1989] data show that the average creep event size remains
narrower areas may be expectedto undergoindurationand constant with time after the mainshock. Thus, the overall
consolidationmore quickly, leading to velocity weakening decay in afterslip with time is causedby an increasein the
behavior [Dieterich, 1981; Tullis et al., 1989]. interevent time. This is to be expected if the stiffness and
constitutiveparametersof the fault zone material are constant
Model Parameters throughtime, since the stress[or slip rate] increaseassociated
with each localized failure episode will then be constant,
The parametersgiven by fitting our aftersliprelation to data producinga characteristiccreepevent size.
compare well with independentestimates.In the case of a-b,
the values are fairly tightly constrained by laboratory CONCLUSIONS
measurements [Marone et al., 1990], giving values for the
depth of the velocity strengtheningregion of 2-4 km for The model presentedhere indicates that in accord with a
Parkfield and 2.5-5 km for SuperstitionHills (Figures9 and number of other observations of earthquake mechanics,
11). In both cases, these values agree with those given by afterslip can be explained as a manifestation of rate/state
independent estimates, providing a check on the variable friction behavior within fault zones. Our model is
appropriatenessof using laboratoryvaluesfor this parameter. based on a fault zone in which a region of velocity

MARONEETAL.:MECHANICS
OFEARTHQUAKE
strengtheningfrictional behavioroverliesa region of velocity
weakening(within which earthquakesnucleate).Afterslip is
driven by a stress perturbation within the velocity
strengtheningregion, which arisesfrom the increasedslip rate
associatedwith rupture propagation from below. Thus the
model provides an explanation for the coupling between
afterslipand coseismicslip suchas indicatedby buriedslip and
modeling showingrapid upward decreasesin coseismicslip.
That is, afterslip is driven by a stressperturbationthat builds
as ruptureis arrested;both afterslipand the stressperturbation
arising from velocity-strengtheningfrictional behavior.
We derive an afterslip relation basedon this model and a 1
degree-of-freedom approximation to elastic interactions
betweenthe fault and its surroundings and find that the relation
satisfactorilyfits afterslip measurementsusing typical values
for the relevant constitutiveand seismicparameters.We also
fit the model to afterslipmeasurements to derive bestfit values
of the constitutive parameter a-b, the thickness of the
velocity-strengtheningregion, and the coseismic slip and
velocity. The values so obtainedare in good agreementwith
independent estimates from laboratory experiments and
seismicmodeling.Our modelingindicatesthat the depthof the
velocity-strengtheningregion is 2-4 km for the Parkfield
region and 2.5-5 km for the SuperstitionHills region. The
results of recent laboratory friction experiments[Marone et
al., 1990] and seismic observations[Crook, 1984; Doser and
Kanamori, 1986, Marone and Scholz, 1988] indicate that
velocity strengthening is expected within unconsolidated
Acknowledgments.We thank Fred Chester,Shamita Das, Simon
sedimentsand fault gouge/breccia. In the contextof the present Cox, Jack Boatwright, and an anonymous reviewer for critical
model, faults that have no shallow velocity strengthening comments.C.H.S. was supported
region (those that do not cut throughunconsolidated alluvium G1668 duringthe courseof this work. LDGO contribution4767.
or containa significant(-- >500 m) gouge/breccia zone) should
not exhibit appreciableafterslip.
APPENDIX
A derivation of equation (5) is given here, based on that
presentedby Scholz [1990]. The basisfor (5) is the idea that
afterslip results from relaxation of a stresspertubationthat
arises when an earthquakepropagatesup into a region of
velocity-strengtheningfrictional behavior. The driving force
for afterslip derives from the slip rate dependenceof friction
(Figure 4). We make the assumption that a steadystatefriction
level is achieved during coseismicslip, which for the one-
dimensionalmodel presentedabove,is equivalentto requiring
a thickness-averaged coseismicdisplacement> D c within the
velocity-strengtheningregion. The driving force for afterslip
is given by Ax (Figure 4, equation(1)).
Afterslip is assumedto obey rate and state variable friction
and thus from the definition of the friction rate parameter,we
may write an equation relating the postseismicslip velocity
Vpthepreseismic
slipvelocity
Vo andthedriving
forceAx
Ax= (A-B)ln(Vp/Vo). (A1)
Notethattheinitialvalue(attheendofcoseismic
slip)ofVp
is VScs
. ThedrivingforceAx decreases
witheachincrement
of
afterslip,which may be written:
(A2)
wherexc is the initial valueof the drivingforce(equation(A1))
and k is the springconstantas definedabove.Using (A1) to
eliminate Ax and rearrangingyields
21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
AFIERSLIP 8451
dt = Vøexp [Xc-Up
A-Bk} '
(A3)
which uponseparationof variablesandintegrationleadsto
equation(5):
U•,=A-B1 t+ 1 ,
k L!A-B ! (A4)
If long-termcreepcontinuesduring[in additionto] afterslipit
may be appropriateto addthe termVo t:
LIA-B ! (A5)
Note addedin proof. The strong-motion waveform-inversion
resultsof Wald et al., [1990] for the 1987 SuperstitionHills
earthquake showsubstantial reduction in coseismic
slip at 3-5
km, with little coseismicslip occurring abovethis depth.As
do modeling results of the 1979 Imperial Valley event
[Archuleta, 1984; Quin, 1990], Wald et al.'s results indicate
the existenceof a 3-5 km thickvelocitystrengthening region
in the Superstition Hills area. Their resultssupport1) the
generalmodelwe usein whichafterslipis causedby rupture
propagationinto a shallowvelocity strengthening region
(Figure1) and2) the resultsof our afterslipmodeling,which
indicatea velocity strengthening region3-5 km thick in the
vicinity of Superstition
Hills (Figure12).
in part by USGS grant 14-08-0001-
REFERF2qCES
Agnew,D.C., and F. K. Wyatt, The 1987 SuperstitionHills earthquake
sequence:Strainsand tilts at PitionFlat observatory,
Bull. Seismol.
Soc. Am., 79, 480-492, 1989.
Archuleta,R. J., A faulting model for the 1979 Imperial Valley
earthquake,J. Geophys.Res.,89, 4559-4585, 1984.
Biegel, R. L., C. G. Sammis, and J. H. Dieterich, The frictional
properties
of a simulated
gougewith a fractalparticledistribution,
J.
Struct. Geol., 11,827-846, 1989.
Bilham, R., Surface slip subsequentto the 24 November 1987
SuperstitionHills, California, earthquakemonitoredby digital
creepmeters,Bull. Seismol.Soc.Am., 79, 424-450, 1989.
Brace,W. F., and J. D. Byerlee,Californiaearthquakes: Why only
shallow focus?, Science,168, 1573-1575, 1970.
Bucknam,R. C., G. Plafker, and R. V. Sharp, Fault movement
[afterslip]following the Guatemalaearthquake
of February4, 1976,
Geology, 6, 170-173, 1978.
Burford,R. O., Continuedslip on the CoyoteCreek fault after the
BorregoMountainearthquake,in The BorregoMountainEarthauake
of April 9, 1968, U.S. Geol. Surv.Prof. Pap., 787, 105-111, 1972.
Clark, M. M., R. V. Sharp,R. O. Castle,and P. W. Harsh,Surface
faulting near Lake Oroville, California, in August 1975, Bull.
Seismol. Soc. Am., 66, 1101-1110, 1976.
Crook, C.N., Geodetic measurementsof horizontal crustal deformation
associatedwith the October 15, 1979 Imperial Valley California
earthquake,Ph.D. thesis,240 pp., Univ. of London,1984.
Crook, C.N., R. G. Mason, and P. R. Wood, Geodetic measurementsof
horizontaldeformationon the Imperialfault, inThe ImperialValley
Earthquakeof October 15, 1979, U.S. Geol. Surv.Prof. Pap.,1254,
183-191, 1982.
Crough,S. T., andR. O. Burford,Empiricallaw for fault-creep
events,
Tectonophysics,42, T53-T59, 1977.
Dieterich,J. H., Modelingof rockfriction,1, Experimental
resultsand
constitutiveequations,J. Geophys.Res., 84, 2161-2168, 1979.
Dieterich, J. H., Constitutivepropertiesof faults with simulated
gouge,in MechanicalBehavior of CrustalRocks,Geophys.Monogr.
 21562202b, 1991, B5, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/91JB00275 by University Of Southern Queensland, Wiley Online Library on [11/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8452 MARONEET AL.: MECHANICSOF EARTHQUAKE
AFtERSLIP

Ser. vol. 24, editedby N.L. Carter,M. Friedman,J. M. Logan,and D. Scholz, C.H., and C. A. Aviles, The fractal geometryof faults and
W. Sterns,pp. 103-120, AGU, Washington,D.C., 1981. faulting, in Earthquake SourceMechanics,Geophys.Monogr. Ser.
Doser, D. I., and H. Kanamori, Depth of seismicityin the Imperial vol. 37, editedby S. Das, J. Boatwright,and C. H. Scholz,pp. 147-
Valley region 1977-1983 and its relationshipto heat flow, crustal 156, 1986.
structure,and the October 15, 1979, earthquake,J. Geophys.Res., Schulz, S.S., G. M. Mavko, R. O. Burford, and W. D. Stuart,Long-
91, 675-688, 1986. term fault creepobservations in centralCalifornia,J. Geophys.Res.,
Fuis, G. S., W. D. Mooney, J. H. Healey, G. A. McMechan,andW. J. 87, 6977-6982, 1982.
Lutter, Crustal structure of the Imperial Valley region, in The Sharp,R. V., and J. L. Saxton,Three-dimensional recordsof surface
Imperial Valley Earthquakeof October15, 1979, U.S. Geol. Surv. displacement on the SuperstitionHills FaultZoneassociated with the
Prof. Pap,, 1254, 25-50, 1982. earthquakesof 24 November 1987, Bull. Seismol.Soc. Am., 79,
Gu, J.-C., J. R. Rice, A. L. Ruina, and S. T. Tse, Slip motionand 376-389, 1989.
stability of a single degreeof freedomelasticsystemwith rate and Sharp,R. V., et al., Surfacefaultingin the centralImperialValley, in
statedependentfriciton. J. Mech. Phys.Solids,32, 167-196, 1984. The Imperial Valley Earthquakeof October 15, 1979, U.S. Geol.
Harsh,P. W., Distributionof afterslipalongthe imperialfault, in The Surv. Prof. Pap., 1254, 119-144, 1982.
Imperial Valley Earthquakeof October15, 1979, U.S. Geol. Surv. Sharp,R. V., et al., Surfacefaultingalongthe Superstition Hills fault
Prof. Pap., 1254, 193-203, 1982. zone and nearby faults associatedwith the earthquakesof 24
Johnson,T. L., and C. H. Scholz, Dynamic propertiesof stick-slip November 1987, Bull. Seismol. Soc. Am., 79, 252-281, 1989.
friction of rock, J. Geophys.Res.,81, 881-888, 1976. Smith, S. W., and M. Wyss, Displacementon the San Andreasfault
Lorenzetti,E., and T. E. Tullis, Geodeticpredictionsof a strike-slip subsequentto the 1966 Parkfield earthquake,Bull. Seismol.Soc.
fault model:Implicationsfor intermediate-and short-termearthquake Am., 58, 1955-1973, 1968.
prediction,J. Geophys.Res.,94, 12,343-!2,361, 1989. Stuart, W. D., Forecastmodel for great earthquakesat the Nankai
Louie, J. N., C. L. Allen, D.C. Johnson,P. C. Haase, and S. N. Cohn, troughsubductionzone,Pure Appl. Geophys.,126, 619-641, 1988.
Fault slip in southernCalifornia,Bull. Seismol.Soc. Am., 75, 811- Tse, S. T., and J. R. Rice, Crustalearthquake instabilityin relationto
833, 1985. the depthvariationof frictional slip properties,J. Geophys.Res.,
McNally, K. C., T. Lay, M. Protti-Quesada,G. Valensise,D. Orange, 91, 9452-9472, 1986.
and R. S. Anderson,SantaCruz Mountains[LomaPrieta]earthquake, Tse, S. T., R. Dmowska,and J. R. Rice, Stressingof lockedpatches
Eos Trans. AGU., 70, 1463, 1989. alonga creepingfault,Bull. Seismol.Soc.Am.,75, 709-736,1985.
Marone, C., and C. H. Scholz,The depthof seismicfaultingand the Tullis, T. E., and J. D. Weeks, Constitutivebehaviorand stabilityof
uppertransitionfrom stableto unstableslip regimes,Geophys.Res. frictional sliding of granite, Pure Appl. Geophys.,124, 383-414,
Lett., 15, 621-624, 1988. 1986.
Marone, C., C. B. Raleigh, and C. H. Scholz, Frictionalbehaviorand Tullis, T. E., M. L. Blanpied, and J. D. Weeks, The velocity
constitutivemodeling of simulatedfault gouge,J. Geophys.Res., dependence of granitefrictionwith andwithoutsimulated
gouge,Eos
95, 7007-7025, 1990. Trans. AGU, 70, 1302, 1989.
Nason,R.D., Fault creepand earthquakeson the San Andreasfault, in Wald, D. J., D. V. Helmberger,and S. H. Hartzell,Ruptureprocessof
Proc. Conf. Tectonic Probs. San Andreas Fault System.,Edited by the 1987 superstition hills earthquakefrom the inversionof strong-
R.L. Kovach, A. Nur, Stanford Univ. Publ. Geol. Sci., 13, 275-285, motion data, Bull. Seismol. Soc. Am., 80, 1079-1098, 1990.
1973.
Nason, R. and J. Weenman, A dislocationtheory analysisof fault Wesson,R. L., Modelling aftershockmigration and afterslipof the
creepevents,J. Geophys.Res., 78, 7745-7751, 1973. San Juan Bautista, California, earthquakeof October 3, 1972,
Okubo, P. G., and J. H. Dieterich,Effectsof physicalfault properties Tectonophysics,144, 215-229, 1987.
on frictional instabilitiesproducedon simulatdfaults, J. Geophys. Wesson,R. L., Dynamicsof fault creep,J. Geophys.Res.,93, 8929-
Res., 89, 5815-5827, 1984. 8951, 1988.
Okubo, P. G., Dynamic rupture modeling with laboratory-derived Williams, P. L., and H. W. Magistrale,Slip along the Superstition
constitutiverelations,J. Geot•hys.Res., 94, 12,321-12,335, 1989. Hills fault associatedwith the 24 Novermber 1987 Superstition
Power, W. L., Tullis, T. E., S. R. Brown, G. N. Boitnott, and C. H.
Hills, California,earthquake,Bull. Seismol.Soc.Am., 79, 390-410,
1989.
Scholz, Roughnessof naturalfault surfaces,Geophys.Res.Lett., 14,
29-32, 1987.
Wong, T.-f., and Y. Zhao, Effectsof load point velocityon frictional
instabilitybehavior, Tectonophysics, 175, 177-195, 1990.
Quin,H., Dynamicstressdropandrupturedynamicsof theOctober15,
1979 Imperial Valley earthquake,Tectonophysics,175, 93-117,
1990. R. Bilham, CIRES, Universityof Colorado,Boulder,CO 80302.
Rice, J. R., andJ.-C. Gu, Earthquakeaftereffectsandtriggeredseismic C. J. Marone, Departmentof Geologyand Geophysics,Universityof
phenomena, Pure Appl. Geophys., 121, 187-219,1983. California, Berkeley, CA 94720
Ruina,A., Slip instabilityand statevariablefrictionlaws,J. Geophys. C. H. Scholz, Lamont-Doherty Geological Observatory,Palisades,
Res., 88, 10,359-10,370, 1983. NY 10964.
Scholz,C.H., The critical slip distancefor seismicfaulting,Nature,
336, 761-763, 1988.
Scholz,C.H., The Mechanicsof Earthquakesand Faulting, Cambridge
UniversityPress,New York, 1990. (ReceivedJanuary8, 1990;
Scholz,C.H., M. Wyss,and S. W. Smith,Seismicandasismicslip on revisedSepte•nber28, 1990;
the San Andreasfault, J. Geophys.Res.,74, 2049-2069, 1969. acceptedNovember13, 1990.)