GEOPHYSICALRESEARCHLETTERS, VOL. 24, NO. 24, PAGES 3317-3320,DECEMBER 15, 1997
Syntheticseismogramsthrough syntheticFranciscan:Insights
into factors affecting large-aperture seismicdata
ChristofLendl•'2AnneM. Trthu• JohnA. Goff3 AlanR. Levander
n,BruceC. Beaudoin
5'6
increaseduring the past decadein the spacingof traceswithin
individual record sections(due in large part to acquisitionof
instrumentsby the IRIS PASSCAL programonshoreand to the
developmentof large-volume,tuned airgun arrays offshore)
resultsin our improvedability to identifylow amplitudeand/or
Abstract. In spiteof an orderof magnitudeincreaseoverthepast
15 yearsin spatialsamplingof the wavefield,a majoruncertainty
in the analysis of active source seismic data remains phase
identification. This uncertaintyresultsin part from the wide
range of spatial scalesof velocity heterogeneityin the crust.
Smaller scalevariationsthan thosewhich can be deterministically
resolvedgiven the designof a particularseismicexperimentcan
be modeled statistically using geologic constraints.Here we
presentsyntheticseismograms
generatedfrom severaldifferent
realizations of a stochastic model describing the velocity
heterogeneityof Franciscanterranerocks. We comparethe
resultsto observeddata and to syntheticseismogramsgenerated
for a model derived from tomographicinversionof the data in
orderto obtainqualitativeinsightsinto therelativeimportance
of
largeandsmallscalevelocityheterogeneity.
Not surprisingly,
the
secondaryphases. Decreasedspacingof overlappingrecord
sections(due to increasedshotspacingonshoreand to increased
numbersof seafloorrecordersoffshore)resultsin improved
resolution of lateral velocity variations. Nonetheless,
identification
of phasesremainsa majorsources
of uncertainty,
and the uncertainlyincreasesas the lengthof the raypathand
geologicalcomplexityincrease. Adequatelyincorporating
the
additionalinformationon crustalstructure
providedby amplitude
variations
in thedata alsoremainselusivefor all butverysimple
models. In fact, our understandingof how seismicwaves with
synthetic
datafor thetom0graphic
modelbestreproduce
observed wavelengthson the order of hundredsof meterspropagatefor
small-scalevariationsin first arrival time, which only occurfor hundredsof kilometersthroughthe geologicallycomplicated
particularrealizationsof the stochasticmodel. The synthetic lithosphere
remainsquitepoor. Observedseismicdatagenerally
seismograms
generatedfor the stochastic
modelsbestreproduce contain a complicatedcoda, which is generally attributedto
the level of signal-generated
noiseandsuggestthatthe amplitude scattering[eg. Menke and Chen, 1984; Frankel and Clayton,
of velocity variation locally within the Franciscan is 1986] andis notpredictedby modelsderivedfrom travel-times.
approximately1 krn/s.They alsoillustratetheeffectof a strongly
Several recent studieshave attemptedto characterizethe
heterogeneous
upperandmid-cruston theamplitude-versus-offsetelasticor acousticpropertiesof geologicallycomplexmaterialsin
patternof arrivalsfrom the lower crustand uppermantle.These orderto generatesyntheticseismograms
that reproducesomeof
effectsmay sometimes
be interpreted
deterministically,
leadingto the complexityobservedin data. Maps of surfacegeology
biasedmodelsor an overly optimisticestimateof lower crustal indicatethatspatialvariationsin theelasticproperties
of thecrust
resolution.
typically span length scalesof several orders of magnitude
[Levanderand Holliger, 1992]. This variabilitycanbe described
statisticallyusing a small number of parameters,the valuesof
Introduction
which can be derivedfrom geologicalmapsand rock physics
Most effortsto modellarge-apertureseismicdataare basedon data. Details of the appropriatestatisticalmodeldependon the
matchingthe travel times of observedarrivals to travel times local geology. This approachhas successfullyreproduced
predictedby a model. In the pastdecade,considerable
effort has characteristics
of the codain regionswherethelarge-scale
crustal
beenexpendedto improvemethodsfor calculatingtravel-times velocitystructureis relativelysimple.
and to develop inversion algorithms to automate the
Here, we present the results of generating synthetic
determination of the best-fitting model. These efforts have seismogramsthrough a model of Franciscan accretionary
greatlyincreasedthe efficiency with which data can be modeled complex rocks. Motivation was provided by the desire to
andhaveimprovedourunderstanding
of theresolvingcapacityof understand
how passagethroughvery heterogeneous
Franciscan
the data. The resultant deterministicvelocity models contain accretionary
complexrocksaffectsthe wavefieldandour ability
large blocksor layerswithin which velocityvarieson a scalethat to resolvedetailsof underlyinglithospheric
structure
in theregion
is comparableto the larger of the sourceor receiver spacings. of the Mendocinotriplejunction(MTJ). A majoractivesource
Although details of solutions depend somewhat on model seismicexperiment(figure 1) was recently conductedin this
parameterization,differences are decreasingbetween crustal region in order to image the subsurfacegeometryof the three
modelsobtainedindependently
by differentgroupsusingdifferent lithospheric
platesandprovideconstraints
on geodynamic
models
methods to model the same data.
of the effectof triplejunctionmigration[Trehuet al., 1995].
Much of this increasedsimilarity between models is due to
increasesin data acquisitioncapability. An order of magnitude
•Collegeof OceanicandAtmos.Sci.,OregonStateUn., Corvallis,OR
2Senckenberg
Institute,
Wilhelmshaven,
Germany
3Un.of TexasInstitute
forGeophysics,
Austin,
TX
4Dept.of Geology
andGeophysics,
RiceUn.,Houston,
TX
5Deptof Geophysics,
Stanford
Un.,Stanford,
CA
6IRIS PASSCALInstrument
Center,Stanford,CA
Copyright
1997bytheAmerican
Geophysical
Union.
Papernumber97GL03262.
0094-8534/97/97GL-03262505.00
StochasticModel Description
The Franciscanterraneof the MTJ regionconsistsof turbidite
and argillaceoussandstones,
intermixedwith numerousmelange
unitsthat originatedpartly from large scalesubmarinemassflow
depositsand partly from tectonicmixing in the subductionzone.
The melangeunitscontainblocksof graywacke,radiolarianchert,
recrystallized limestone, blueschistfacies metamorphicrock,
greenstone,
andplutonicrock of ophioliticaffinity [Aalto, 1994].
All thesepartsare stronglyintermixedandform a highlycomplex
structure. Goff and Levander [1996] derived a statistical
descriptionof Franciscanrocksthat approximatesthe two main
units of Franciscanrocks, sandstone(66%) and melange(34%),
3317
3318
LENDL ET AL.: SYNTHETIC SEISMOGRAMS THROUGH SYNTHETIC FRANCISCAN
125ø
42 ø
4V
39øN
124ø
123ø
refracted through the Franciscan rocks. Allowing for twodimensionalvelocity variationin a tomographicinversionof first
arrivalsleadsto a structurecontaininglarge "blobs"of higheror
lowervelocityin theupper10 km [Beaudoinet al., 1996].
Three differentrealizationsof the stochasticP-wave velocity
model are shown in figure 2a, as are the simple and
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
•11 \\%// %11•11 \\%//, •\11•11 \\%
tomographically-derived deterministic models. For each
i•:..:i•-::.!•!:!!!!:!!!::i:::i:!•::•::•.!::•:•:•:•:•:•::•:•::•:•::•
•
•1,
,,• II •• II• •x
...........
'•Z?•..•.•.•
.........
•::.:'
.•_It •' II• ,,,,• II •
stochasticmodel, the peak-to-peak velocity difference in the
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
- •'•
////• • * ' ////• • *. ' ////•a:'•Z//
stochasticgrid is 1.0 km/s. For the lowermostcrustand upper
mantle, a model representingthe crust of the subductedGorda
plate, which dipsto the southand endsnearkm 80, wasincluded
in all models based on preliminary analysis of wide-angle
reflections; south of km 80, a lower crust representinga
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
• -•.'?
tectonically or magmatically underplated mafic layer was
:::::...•g•:::•:•:?•:•::[:::•::::?::
.•g:.•...
included. This particularlower crustalmodel is not meant to
representa "final" model for line 9; it is includedto illustratethe
effect of the overlying stochasticallyheterogenouscrust on
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::•::•]•::::::[::•:::•:•[•:::•[•:
- . .:: e•'..g:•' •
signals from underlying structures. Although a general
thickeningof the crust in the central portion of line 9 is well
::::::::::::::::::::::::
•:::
::•::::••f•f•
•[::•4•?:•:•?:[::?::•:?:[:[•?:•::::•?:•:•:?::
: •:- -e'.½
• ':•- .•:..-e'.
determinedfrom the data, detailsof the lower crust and upper
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
' • • ...
mantlein thisregionare the subjectof ongoinginvestigation.
Calculation of Synthetic Seismograms
Syntheticseismogramswere calculatedusing an isotropic,
elastic,2-dimensional,staggered-grid,
finite-differencealgorithm
to solve the wave equation. Three-dimensionalversionsof this
Figure 1. Simplifiedgeologicmapof noahwestern
C•ifornia algorithmhave been discussedby Rodriguesand Mora [1992]
showingthe locationof the seismicprofilethat inspiredthis and Igel et aL [1995]. The algorithm,which calculateseighthmodelingeffoa. Shotpoints
areshownastfi•gles; theshotpoint orderspatialderivativesusingthe methodof Holberg[1987] and
modeledin thispaper(911)is labeled.SAF - San•dreas fault, forth-ordertemporalderivativesthroughTaylor seriestruncation,
MAF - Maacama fault, BSF - Bartlett Springs fault, CSZ - has been implemented on a ConnectionsMachine CM-5 at
Oregon State University to permit calculation of synthetic
Cascadiasubductionzone, MFZ - Mendocino fracturezone.
seismograms
and visualizationof wavefieldpropagation
through
complicated 2-dimensional models derived from travel-time
analysis of data from large-aperture active-source seismic
asa bimodal,approximately
fractal,sinuously
connected
velocity experiments[Lendl, 1996]..
distribution, with model parametersderived from digitized
In figure2b, theobserved
datafroma 2270kg explosiveshot
geological
maps.The Franciscan
is particularly
interesting
in that [Godfrey et al., 1995] locatedat the northernend of line 9 are
it is characterized
by two distinctlengthscalesandtransverse comparedto syntheticseismogramsfor sevendifferent models:
anistropy.Characteristic
verticallengthscalesareapproximately thesimpledeterministic
model,thetomographic
model,modelR1
1.3 and 5.5 km, with horizontalscalelargerby a factorof 2.75. for peak-to-peak
stochastic
velocityvariationsof 0.5, 1.0 and 1.5
We present synthetic seismograms for several different km/s, and modelsR2 and R3 for a peak-to-peakstochastic
realizationsof this velocitymodelandcomparethemto synthetic velocityvariation1 km/s. ModelsR1, R2 and R3 represent
different realizations of the same stochastic model.
All
seismograms
for deterministicvelocitymodels.
Each realizationof the stochasticmodelis represented
by a seismograms
are displayedwith a reductionvelocityof 6.5 km/s
binarygrid in which valuesof + 1 representsandstone
and values and preserverelative amplitudevariationsas a functionof both
of-1 identify melange units. This is admittedly a gross time and offset.
simplificationof velocity structureof the Franciscanterrane,as
For the calculations,
the densityandelasticproperties
of the
the melangeis itself a very complex,heterogeneous
medium. A material,
derivedfromP andS wavevelocities,
mustbespecified.
grid spacingof 100 m was chosento permit calculationof P-velocitygridswerecreatedasdiscussed
above.S-velocityand
seismic waves with a frequencycontent of up to 5 Hz for densitygridswerederivedfromtheP-velocitygridsassuming
a
velocitiesgreaterthan 3 km/s. The valuesin the grid werethen Poisson'sratio of 0.25 and a fifth order polynomialfit to a
scaledto producea specifiedmaximumvelocity differencethat compilationof laboratoryvelocity/densitymeasurements
[Zelt
rangedfrom 0.5 to 1.5 km/s for differentmodels.After scaling, andSmith,1992]. Resultsarenot very sensitive
to uncertainties
the averagevelocity of the stochasticvariationswas removed in S-wavevelocityor density[Lendl,1996].
from the grid to producea meanvalue of 0, and the grid was
Stability
andconvergence
of finitedifference
calculations
masked to outline a region of Franciscanrocks in the model. depend
on theproperchoiceof timestepAt andgridspacing
Ax,
BeneaththeFranciscan,
thevalueof the stochastic
gridwas0.
which dependin turn on the lowestvelocityin the model,the
The stochasticvelocity grid was then addedto a simple highestfrequencyin the source,and the algorithmsusedfor
deterministicvelocitygrid determinedfrom travel-timeanalysis spatial and temporal differentiation [Levander, 1989].
of the data from line 9 of the Mendocinotriplejunctionseismic Theoreticalconsiderations
[Rodrigues
andMora, 1992;Igel et
experiment. Line 9 is located entirely within the Franciscan al., 1995;Lendl, 1996]indicatethata grid spacing
of at least3
Coast Rangesof northwesternCalifornia (figure 1). For this grid points/wavelength
and (cp At)/(Ax) = 1ß205, wherec is the
study,the velocitywithintheFranciscan
rockswasrepresented
by
a velocity of 3.0-4.0 km/s at the surfaceand of 6.5 km/s at the the algorithmusedhere. A seriesof simulationswasdoneto test
baseof the Franciscan.This simplevelocitystructurerepresents the domainof validity and determinean acceptablelevel of
the one-dimensionalvelocitymodel that providesthe bestfit to dispersion.The final choicewas a grid spacingof 100 m and
the observed first arrivals interpreted to representP-waves time stepsof 0.01 s. The maximumfrequencyin the seismic
largest
P-wave
velocity
inthe
model,
are
adequate
forsta•ility
of
LENDL ET AL.: SYNTHETIC
S DISTANCE
ALONG
MODEL
(km)N
0
50
100
150
200
250
0
I--
SEISMOGRAMS
THROUGH SYNTHETIC
S
N
shot
911- data
/•' .•'1"h•.•_•_;:
....; '[t,•',"•;';
-' t.• •.' ..•-'•,T'i
FRANCISCAN
S
3319
N
10 tomography
10
lower crust
LIJ
'------
ugper zrtle
C340 ,•mple
rncclel
o
250
E
125
to•m
25O
o
sim.lemodel
1- 20
10
125
R1 - 0.5 krn/s
10
r
0
3:20
0
,,
25O
.hat
• rand I R I
125
0
R2- 1.0krn/s
125
250
10
0
R1 - 1 0 krn/s
10
0
c3 40
'h
•O.
IRZ
0
25O
125
0
125
250
0
0
o
R1 - 1 5 km/s
10
1- 20
l•.,ha
,,n
elF'3
P-WAVEVELOCITY (knVs)
0
250
125
OFFSET (km)
Figure2 a. Velocitymodels.
0
!
250
125
0
OFFSET (km)
Figure2b.SynthelJc
seismograms
forthemodels
of figure2a compared
todataforshot911.
Figure 2a. Velocitymodels.Positions
of shotpoints
alongtheseismicprofilearealsoshown.Fromtopto bottom,themodels
are:thesimpledeterministic
model;thetomographic
inversion
of firstarrivaltimes[Beaudoin
et al., 1996],modifiedto include
the samelower crustas in the other models;three differentrealizationsof the stochasticmodel, generatedby using different
randomnumberseeds(seetext). The amplitudeof velocityvariationfor the stochastic
modelsis 1 km/s.
Figure 2b. Observeddatafor shotpoint911 comparedto syntheticseismograms
for the modelsshownin figure 2b and for
differentamplitudes
of stochastic
velocityvariation.Severalimportant
phases
havebeenlabeledonthesynthetics
for thesimple
deterministic
model:
Pg- crustal
diving
wave;
Pn- upper
mantle
interface
head
wave;
PiP- reflection
fromthetopofthelower
crustlayer;PmP- reflectionfromthebaseof thelowercrustallayer. Seetext for furtherexplanation.
source,
whichis defined
bya Gaussian
function,
is 5 Hz. The noise
mustbesignal-generated.
Alsocharacteristic
arethelarge
synthetic
seismograms
forthesimple
deterministic
model
(figurevariations
intheamplitude
ofthesignal,
withstrong
secondary
2b)givea qualitative
ideaofthelevelofnumerical
noise
with arrivals
observed
locally
thatcannot
bereadily
identified.
These
these
parameters.
Whilethedatacontain
higher
frequencies,
features
arenotobserved
in thesynthetic
seismograms
forthe
calculating
synthetic
datawithsuch
highfrequencies
formodelssimple
deterministic
model.
Whilethesynthetic
seismograms
for
thesizerequired
isbeyond
ourcurrent
computing
capabilities. themodel
derived
deterministically
viatomographic
inversion
of
Absorbing
boundary
conditions
areimplemented
onthesides the first arrivalsaccurately
reproduce
the variableapparent
andbottomof themodelby extending
themodelandmultiplyingvelocityof the first arrivals,the focussing
anddefocussing
of
displacement
amplitudes
withinthisregionby an absorbtion
energy
in thePg arrivalandin thefollowing
oneortwoseconds
factorthatdecreases
exponentially
withinthisboundary
region is exaggerated.
Thismodelalsodoesnotreproduce
theobserved
[Cerjanet al., 1985].Forthesynthetics
shown
here,thetopof levelofbackground
signal-generated
noise.
the model acts as free surface. Because the near-surface
Comparing the seismogramsfor the simple deterministic
velocitiesin the model are considerablyhigher than velocities model to thosefor stochasticmodel R1 with different peak-tomeasurednear the surface along line 9 (again becauseof peakstochastic
velocityvariations,we seethat,not surprisingly,
computational
limitations),
andbecause
intrinsicattenuation
is the seismograms
for the stochastic
modelsshowconsiderable
not included,strongsurfacewavesappearin the synthetics
that energyfollowingthefirstarrivals.For velocityvariations
of 0.5
arenot apparentin thedata.
km/s, the background scattering level is distributed
homogeneously along the whole profile. With increasing
Discussion
stochastic
velocitycontrast,
strongfrequency-dependent
One featureof the data (figure 2b), whichis generally attenuation
is observed
withincreasing
offset. Othereffectsof
characteristicof data collectedwithin the Franciscanterrane,is the increasedscatteringare that weakerarrivalsdisappearinto the
the strong"background"
noisefollowingthefirst arrivals. This "background"
signal-generated
noiseandthatthedetailsof travel-
3320
LENDLETAL.: SYNTHETIC
SEISMOGRAMS
THROUGH
SYNTHETIC
FRANCISCAN
time triplications are obscured. The model with a 1.0 km/s
modeling
algorithm,
andtoMarkAbbott
andtheNASAEarth
Observing
stochastic
velocity
variation
produces
synthetic
seismograms
that, System
program
forproviding
theCM-5supercomputer
thatmadethe
qualitatively,
mostresemble
theobservations.
However,
because calculations
possible.
TomBrocher
andananonymous
reviewer
provided
the sourcefor the syntheticseismograms
doesnot contain helpfulcomments.
Figures
wereconstructed
usingGMT [Wessel
and
frequencies
above5 Hz whereastherealsourcedoesandbecause Smith,1995].Thisproject
wasfunded
by theContinental
Dynamics
scattering
in themodelistwo-dimensional
whereas
scattering
in programof the NationalScienceFoundation
through
grantsEAR-
the earthis three-dimensional,
quantitative
estimations
of the
9219870 and EAR-9527001.
amplitudeof velocityvariationand attenuation
in the Franciscan
terranefromthesesimulations
arenotjustified.
References
We alsocompared
synthetic
seismograms
forseveral
different
Aalto,K. R., LateCenozoic
sediment
provenance
andtectonic
evolution
realizationsof the samestochastic
model(i.e. modelsderived
of thenorthernmost
CoastRanges,
California,
Penrose
Conference,
using the same backgrounddeterministicmodel and the same
Eureka,
CA, Guidebook
Penrose
FieMTrip,73-89,1994.
statisticalparameters
to definevelocityvariationwithin the Beaudoin,
B.C.,N.J.Godfrey,
S.L. Klemperer,
C.Lendl,
A.M. Trthu,
Franciscan
terrane).
Theseismic
sections
generated
in thisway T.J.Henstock,
A.R.Levander,
J.E.Holl,A.S.Meltzer,
J.H.Luetgert,
Resultsfrom
showsimilarities
anddifferences.
Forexample,
thecharacter
of andW. D. Mooney,Thetransitionfromslabto slabless:
TripleJunction
seismic
experiment,
Geology,
24,
thecrustal
divingwave(Pg)is differentfor thedifferent the1993Mendocino
195-199, 1996.
realizations.
Theseismic
sections
forR1andR2show
thePg
C., D. Kosloff,
R. Kosloff
andM. Reshef,
A non-reflecting
arrivalcontinuing
smoothlywith an approximately
constant Cerjan,
boundary
condition
fordiscrete
acoustic
andelastic
waveequations,
apparent
velocity
fromoffsets
of 25-110
km,where
Pg is Geophysics,50,705-708, 1985.
overtakenby the Pn arrival. Amplitude modulationand Frankel,A., andR. W. Clayton,Finitedifference
simulations
of seismic
secondary
arrivals
thatclosely
follow
Pgaresomewhat
different scattering:
implications
for thepropagation
of short-period
seismic
forthese
tworealizations.
Incontrast,
thePgarrival
forR3shows waves
inthecrust
andmodels
ofcrustal
heterogeneity,
J.Geophys.
Res.,
largervariations
of apparent
velocityandgreateramplitude 91, 6465-6489, 1986.
modulation,
similarto the observations
andto the syntheticGodfrey,
J.J.,B.C.Beaudoin,
C.Lendl,
A.Meltzer,
andJ.Luetgert,
Data
seismograms
generated
forthetomographic
inversion
model.
reportfor the 1993Mendocino
triplejunction
seismic
experiment,
Report
95-275,
U.S.Geological
Survey,
83pp.,1995.
The apparent
amplitudeversusoffset(AVO) behaviorof the Open-File
Incorporating
"sinuous
connectivity"
into
wide-angle
reflectionfromthebaseof thecrust(PmP)alsois Goff,J.A.,andA.R.Levander,
stochastic
models
ofcrustal
heterogeneity:
Examples
fromtheLewisian
differentfor the threerealizations.For R1 andR3, PmPis
gneiss
complex,
Scotland,
theFranciscian
formation,
California,
andthe
observed
foroffsets
of 60-210km,butthecharacter
ofthisphase Hafafitgneiss
complex,
Egypt.,J. Geophys.
Res.,101,8489-8502,
is different. R1 showstwo large"shingles"
at 140-160km 1996.
offsets,
whereas
R3 shows
smaller"shingles"
atoffsets
of 75-120 Holberg,
O., Computational
aspects
of thechoiceof operator
and
km. R2 doesnotshow"shingles"
andPmPis delayed
beyond sampling
interval
fornumerical
differentiation
in large-scale
simulation
Geophys.
Prospecting,
35,629-655,1987.
offsets
of about160km. In addition,
a strong
phase
precedes ofwavephenomena,
propagation
through
finitePmPat offsetsof 80-110km onR2 (andin thedata),butis not Igel,H., P. Mora,B. Riollet,Anisotropic
difference
grids,Geophysics,
60, 1203-1216,
1995.
visible on R1 and R3.
wavefield
modeling
of large-aperture
data
The differencesamong the syntheticseismicsections Lendl,C., Finitedifference
triplejunction
seismic
experiment,
M.S.
generatedfor three different realizationsof the samestochastic fromthe 1993Mendocino
thesis,163pp.,OregonStateUn., 1996.
model obviouslydependon the spatialdistributionof the
Levander,
A. R.,Finite-difference
forward
modeling
in seismology,
in
stochastic
velocity
variations.
In areas
of dominantly
large-scale TheEncyclopedia
ofSolid
EarthGeophysics,
edited
byD. James,
pp.
1989.
velocityvariations,
thewavefronts
arebroken
intoseparately 410-431,Van-Nostrand-Reinhold,
distinguishable
wavefronts,
andstrongtraveltime
variations
and Levander,
A. R.,andK. Holliger,
Small
scale
heterogeneity
andlargeamplitude
modulation
result.Areasof dominantly
small-scale scale
velocity
structure
of thecontinental
crust,
J. Geophys.
Res.,97,
variations,on the otherhand,scatterthe wavefrontandcreatean
8797-8804, 1992.
envelope
ofenergy.
Wavefronts
forR3travelalmost
exclusivelyMenke,W., andR. Chen,Numericalstudiesof thecodafall-offrateof
wavesin randomly
layeredmedia,Bull.Seis.Soc.
throughareaswith large-scale
velocityvariations,whereas multiply-scattered
Am., 74, 1079-1081, 1984.
wavefronts
for R1, travelprimarilythrough
a regionof small- Pullammanappallil,
S.,A. Levander,
andS. P.Larkin,Determination
of
scalevariationsfor the raypathsof phasesdiscussed
above.It
crustal
stochastic
parameters
fromseismic
exploration
data,
J.Geophys.
should
benotedthatthesamevariations
wouldbegenerated
for Res.,102, 15269-15286,1997.
differentshotpoint
locations
anda singlemodelrealization,Rodrigues,D., andP. Mora,Analysisof a finitedifferencesolutionto the
perhaps
explaining
observed
AVOdifferences
fora givenphase three-dimensional
elasticwaveequation,
62ndAnn.Int. Mtg., Soc.
from differentshotpoints
in activesourceseismicdata. Such
Expl. Geophys.,ExpandedAbstracts,124%1249, 1992.
WorkingGroup,Pullingtherugoutfrom
variations
aresometimes
interpreted
asthesignature
of lower Tr6hu,A.M., andMendocino
crustand uppermantlestructurewhereastheseresultsshowthat
theymaybe produced
by amplitude
variations
withintheupper
and mid-crust.
underCalifornia:Seismicimagesof the Mendocino
TripleJunction
region,Eos TransAGU, 76 (38), 369, 1995.
Wessel,P., andW. H. F. Smith,New versionof thegeneric
mapping
toolsreleased,EosTransAGU, 76 (33), 329, 1995.
Weconclude
thatthesuccess
ofthi•initial
effort
atsimulating
Zelt, C. A., and R. B. Smith, Seismictraveltime inversionfor 2-D crustal
wave propogationfor distancesof 100s of km through
complicated
crustalmodelssuggests
thatfutureeffortsto extract
velocitystructure,Geophys.
J. Int., 108, 16-34,1992.
information
onsmall-scale
crustalheterogeneity
directlyfromthe
data [Pullammanappallilet al., 1997] shouldbe fruitful. This
C. Lendl, Senckenberg Institute, Schleusenstr. 39A, D-26382,
informationcan then be usedto compensate
for small-scale Wilhelmshaven,Gemany.
author),Collegeof OceanicandAtmospheric
effectswhenmodelingtravelingtimesto derivelarge-scale A.M. Trthu (corresponding
Sciences,
OregonStateUniversity,104OceanAdminBldg.,Corvallis,
lithosphericstructure.
Acknowledgements.
Many thanksto thosewho made the
Mendocino
triplejunctionseismic
experiment
possible,
including
the
OR 97331-5503 (e-mail: trehu@oce.orst.edu).
J. A. Goff, Un. of TexasInstitutefor Geophysics,
Austin,TX.
A. R. Levander,Dept.of GeologyandGeophysics,
RiceUn., Houston,TX.
B.C. Beaudoin,IRIS PASSCALInstrumentCenter,Stanford,CA.
many instrumentdeployersand the IRIS/PASSCAL,USGS, and
Lithoprobeinstrumentcenters,whichfurnishedinstruments
for the data
(Received
March18,1997;revisedOctober
3, 1997;
acquisition.
Thanks
alsoto PeterMoraforproviding
thefinitedifference acceptedOctober29, 1997 ).