Sunteți pe pagina 1din 10

Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.

org/

From Markov Layer Matrices to Equivalent-Media Systems


Joe Dellinger

ABSTRACT

Schoenbergand Muir (1989) show how to generalizethe layer-averagingmethodof Backus


(1962) to the moregeneralconceptof a "layer group,"in which layersare not only averagedto find
an equivalenthomogeneous medium,but also added,subtracted,and negated.Dellinger and Muir
(1993) showhow to recastthe Dix equations(Dix, 1955) into paraxiallayer-groupform.
Backusaveragingis a low-frequencyapproximation;the Dix equationsare a high-frequency
paraxial(small ray parameter)approximation.Layer matrices,whichextrapolatea monochromatic
wavefieldacrossa layer,areexactfor all frequenciesandanglesof propagation.In thispaperI show
how by suitablemanipulationof layer matricesit is possibleto exactly calculatethe ensemble-
averagewavefieldemergingfrom a statisticallydefinedlayer stack.This resultis usedto define an
equivalent-mediumsystembasedon layer matrices.
If the layerprobabilitiesare independent,
the theoryis particularlysimpleanddirectlyleadsto
an obviouslayer-groupformulation.The methodalso worksfor layersdefinedvia Markov chains,
althoughit hasnot yet beenprovenwhetherthereis a (noncommutative) layer-groupformulation
for this case.

INTRODUCTION

Equivalent-medium theoriesshowushowto calculatethehomogeneous mediumthatis"equiva-


lent" in its bulk propertiesto a specifiedheterogeneous one. Many resultshavebeenfoundfor the
importantspecialcaseof stratifiedmedia,startingfrom the pioneeringwork of Backus(1962) and
O' Doherty and Anstey ( 1971).
Schoenberg(1983) and later Schoenbergand Muir (1989) generalizedBackus'resultsto the
caseof anarbitrarystackof generallyelasticlayers.Becausetheirmethodcalculatestheresponse of
the layer stackto a staticload, it can only find the zero-frequencylimiting behavior.Shapiroand
Hubral(1996) addressthis limitationby makinguseof the propertiesof "self-averaged"wavefield
variables.Their methodis valid for all frequencies,but breaksdown in the presenceof "large"
(greaterthan 30%) variationsin the propertiesof the medium,as doesO'Doherty and Anstey's
originalformula(Banik et al., 1985). Frazer(1994) presentsthreealternativestrategiesthat avoid
this limitationfor the specialcaseof a binary sediment(two discretelayer typesof varyingthick-
nessin alternation).For thegenerallayeredacousticcasethereis an extensivebodyof mathematical
literature;Aschet al. (1991) providea goodreview.
If themodelis specifiedstatistically,
thenit maymakesenseto definethehomogeneous equiva-
lent mediumvia the "ensembleaverage"wavefield,i.e., the wavefieldafter passagethroughsome
finite numberof layersaveragedoverall possiblerealizationsof the model.Sucha definitionimme-

BP Amoco, 200 Westlake Park Blvd., WL1, Room 1019, Houston, IX 77079, USA

99
lOO Layer-Matrix Equivalent Media Systems

diately raisesthe difficult questionof whether the ensembleaveragehas any relation to reality,
where we observethe seismogramresultingfrom the single real earth, not an ensembleaverage
(Resnicket al., 1986). I do not believe this is necessarilya fatal flaw for all ensemble-average
methods.After all, if each realization(with probabilityone) is tendingtowardsthe samelimiting
behavioras the numberof layersbecomeslarge,then so shouldthe ensembleaverage.Some care
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

mustbe takenwhenchoosingwhat quantitiesto average,however,to avoidnonphysicalsmoothing


of the result(Stankeand Burridge, 1993).
One popularmethodof statisticallydefining layeredgeologiesin Geophysicshas been via
Markov chains(Godfrey et al., 1980). In this chapter,I show how to explicitly calculatethe en-
sembleaveragewavefieldthrougha Markov layer stackusingpropagatormatrices(Gilbert and
Backus,1966). Note, this is notexactlythe sameresultas performinga reflectivity-methodcalcula-
tion for each possiblerealizationand averagingthe outputreflectedand transmittedwavefields,
becauseof the nonlinearinteractionbetweenthe bulk propagatormatrixandthe standardboundary
conditions(i.e., a unit incidentwavefieldfrom above,no wavefieldcomingup from below).
If the ergodicityassumption(that the ensembleaverageis equivalentto the spatialaveragewe
actuallyobserveif thereare enoughlayers)is valid, no furtherassumptions or approximations are
required.In particular,this methodremainsvalid even if thereare large impedancecontrastsbe-
tween layers.By analyticallyexpandingout the elementsof the ensemble-average propagatorma-
trix in powerseriesin frequency,low-frequencyequivalentmediumsystemsto anydesiredorderof
accuracycan be found.In principle,the ensemble-average propagatormatrixcan be explicitlycal-
culatednumerically,and directly usedin reflectivitycalculations.In practice,instabilitiesdue to
poorlyconditionedmatricesand resonances often makenumericalfrequency-domain calculations
usingensemble-average propagatorsproblematicalaway from small frequencies.More work re-
mainsto be doneto implementthe techniquesfrom the mathematicalliteraturefor stabilizingsuch
calculations(Geist et al., 1990).

COMMUTATIVE EQUIVALENT-MEDIUM SYSTEMS

The Schoenberg-Muir layer calculus


Schoenbergand Muir (1989) introduceda "layer calculus"reformulationof Backus'(1962)
originalderivation.They showthat the problemconsiderablysimplifiesif the relevantparameters
are sorted into three classes:

1) Layer parametersdefining the individuallayer properties.(In their example,thickness,


density,and elasticconstants.)

2) Constantparameters,which havethe samevaluefor all layersin the stack(e.g., normal


stressand tangentialstrain).

3) Additiveparameters,whichsumthroughthestack(e.g.,thickness,densitytimesthickness,
normal straintimes thickness,and tangentialstresstimesthickness).

Oncethis hasbeendone,the problemreducesto writing the physicsof the layersin the form:

Additive = Function(Layer) x Constant ()

Becausethe additiveparameterssum,andthe constantparametersare the samein all the layers,the


coefficientFunction(Layer)must also sum throughthe stack.The terms in Ftmction(Layer)thus
providean alternative"layer group"parameterization
of the elasticconstants.Expressedin these
Dellinger 101

terms,theparameters for thehomogeneous equivalentmediumfor thestackcanbefoundby simply


summingthe layer parametersof the individuallayers.
DellingerandMuir (1993) showthattheDix equations canbederivedusingthesamemethod-
ologyof constant,additive,andlayerparameters.(In the Dix case,theconstantparameteris the ray
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

parameterp, the additiveparameteris traveltimealongthe ray,and the layerparametersare zero-


offsettraveltimeand moveoutvelocity.)In the nextsection,I showhow this methodologycan also
be appliedto ensembleaveragesof reflectivitymodels,if the layersare independently
chosen.

Propagator matrices
Layermatricesoperatein the Fourierdomainto extrapolatea monochromatic planewavefield
acrossa layer.In practicethey are usuallywritten in termsof upgoingand downgoingwavefield
variablesbecausethatis preferredfor computationalpurposes(Kennettand Kerry, 1979). Unfortu-
nately,a layer matrix thusdefineddependson boththe propertiesof the layer beingextrapolated
through(to extrapolatethe wavesacrossthe layer) andthe followinglayeras well (to accountfor
reflectionsand transmissionsat the interface)(Claerbout, 1976).
We requirelayermatricesthatcanbe permutedin anyorderandstill havea physicalmeaning.
Forexample,
if isthelayermatrixthatextrapolates
a wavefield
across
a layerof typeA, andB_
doesthe samefor a layerof typeB, thenBA shouldrepresentextrapolationacrossa layerof typeA
overa layerof type B, AB, a layer of type B over a layer of typeA, A.A, a double-thicklayer of
typeA, etc.This propertycanbe ensuredby usingpropagatormatrices,layermatricesthatoperate
on variablesthatareautomaticallycontinuousat layerboundaries(Gilbert and Backus,1966). (It is
also possibleto formulatethe problemin termsof up- and downgoingwave variablesby simply
sandwiching eachlayerof interestbetweeninfinitelythin layersof an arbitraryreferencemedium
R, whichrepresents the propertiesof the spacethe layer stackis embeddedwithin.)
Nownotewhathappens
whenwe expandout(for example)
(+ _B)
2,where_AandB_are
squarematricesof the samedimensions:

(6 + - (b + ?)(b + - bb + + + (2)

(_A4-_B)
2is thesumof all possible
permutations
of twochoices
drawnfromtheset{A, B} multi-
pliedtogether.
If andB_arepropagator
matrices,
then(_A+ B_)
2hasa physical
interpretation'
it is
the ensemblesumof the wavefielddownwardcontinuedthrougha two-layerstack,with eachlayer
independentlychosento be of typeA or B with equalprobability.Youcaneasilyverify thatthistrick
generalizes
to the sumof M matricesraisedto the Nth power(i.e., M differentkindsof layerar-
rangedin a stackN layerstall).
If we multiplyeachpropagator
matrixby its probabilityof beingchosenwe calculatea prob-
ability-weighted
sum.Letpxbetheprobability
of choosing
a layerof typeX. Then,forthecaseof
threetypesof layers,

(pA_+ psB_+ pcq)v ()

calculatesthe ensemble-average
propagatormatrix for a stackN layersthick. Examiningequation
(3), it is clear that

_E-- p,4A_+ psB_+ Pc (4)

thusmustbe the propagatormatrixfor a singlelayerof the averagedmedium.


We are now readyto build an ensemble-average equivalent-mediumsystemusingpropagator
matrices.The layer parameterscan be layer characteristics
like velocityand density.The constant
102 Layer-Matrix Equivalent Media Systems

parameters
arethetemporal
frequency
roandhorizontal
wavenumbers
k, andk,.of themonochro-
matic plane waves.As discoveredin equation(4), it is the probability-weighted
elementsof the
propagatormatricesthatadd.
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ACOUSTIC LAYERS AT NORMAL INCIDENCE

I now demonstratehow thisworksfor thecaseof independentlychosenrandomacousticlayers


at normalincidence.(For a laboratoryexampleof sucha model,seeRio et al. (1996).) The continu-
ous wavefield variablesfor this problemare pressureand verticalparticle velocity.The temporal
frequencyro is constantacrossall layers.The layer propertiesare specifiedby their velocityv,
densityp, andthicknessh. The additiveparametersare theelementsof the propagatormatrix.
To constructthe propagatormatrix for a layer, convertfrom pressureP and verticalparticle
velocityW to up anddowngoingwavefieldvariables,phaseshiftacrossthe layer,thenconvertback
to the original variablesagain(Claerbout,1976):

(5)
- - t +__texp
h 0 +'i- 1 +pv W
Multiplying the threematricestogether,we obtain

(6)
W +t i sin(7-)
COS
(-5-)
hca w

The elementsof this matrixdefinethe additiveparametersfor the problem.Note thereis no simple


lineardependency of the elementsof this matrixon the constantvariableco.If therewere, an en-
semble-averaged stackof acousticlayerswouldbehaveexactlythe sameasa singleacousticlayer.
A stackof acousticlayersdoeshoweverlook like a singleacousticlayer in the low-frequency
(static) limit. We can see this by expandingthe elementsof the matrix in equation(6) in a power
seriesin frequencyto. The coefficientson eachpowerof codefinethe additivelayer parameteriza-
tion. (This is exactlyanalogousto the powerseriesfor ray parameterp for the Dix-equationscase
discussedby Dellinger and Muir (1993).) The coefficientson the co termsare the only onesthat
matterin the low-frequency(static)limit.
Expandingequation(6) to first orderin rowe obtain:

- (7)

Pickingoff the coefficientson ro terms,anddiscardingconstantmultplicativefactors(which are


irrelevant),we find the layer-groupparameterization in the low-frequencylimit to be: {hp, h/pv2}.
Sincewe alsoexpectlayerthicknessto add,we alsoincludethe additionalelement{h }. If the only
availablelayer parametersare h, p, and v, thereare no morefree parametersleft to fit any higher-
order terms in
The three layer-groupparametersfor the low-frequencylimiting caseare all familiar:
h: layer thicknessadds
hp: massadds
h/pv2:thicknesstimescomplianceadds(the "springsin series"result)
Dellinger 103

To solvethe equivalent-mediumproblem,

1)Convertthelayerparametersforeachlayeri fromtheparameterization
{h,p, v} to thelayer-
groupparameterization
{hi, (hp)i, (h/pv2)i}
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

2) Sumthe layer-groupparameterization
overi to find
h h
--
3) Convertthesummed
resultbackto theusualparameterization
to find{h,tack,
P,tack'
Vstack
}'

This is exactlythe Schoenberg-Muir(1989) layer-groupcalculusspecializedto the acousticverti-


cal-incidence case.

Fitting to higher orders


To fit higherpowersof co,we mustextendourdefinitionof a "layer"to allow itsbehaviorto vary
as a functionof frequency.This will give us more free parameters,enablinga fit over a broader
spectrum of frequencies.
Fromthefundamental propertiesof theproblemwe knowthatstackingthe
samelayeroveritselfN times(andthusraisingitspropagator to theN'hpower)shouldgivethesame
resultasa singlelayerN timesasthick.This suggests lookingfor solutionsof the form d'(m"".
In
particular,noteequation(6) is exactly(Claerbout,1976)

- exp ih (8)
0 W
W k+1 pv k

The matrixinsidethe exponentialcomesdirectlyfrom the associated differentialequation.


It is alsopossibleto usethe calculatedensemble-average propagatormatrixdirectly.For low
frequencies this reproducesthe behaviorof the Backus-average result.At slightlyhigherfrequen-
cies deviationsfrom the Backus-average resultstartto becomedetectable.Figure 1 comparesthe
resultsof theensemble-average propagator matrixtechnique(bottom)andBackusaveraging(middle)
versusanensembleaverageover50 realizationsof thereflectedwavefieldfroma 50-layerGoupillaud
(constanttraveltime-thickness) stack(top). 40% of the layershave thickness.4, velocity .8, and
density3.; the other60% of the layershavethickness.6, velocity 1.2, anddensity2.6. The stackis
embeddedin a spacewith unit velocityanddensity.The predominantfrequencyin the sourcewave-
let is .2; the wavelengthis thusonly about2.5 timesthe averagelayer thickness.The ensemble-
averagepropagator matrixmethodcorrectlypredictsthephaseshiftin thewavereflectedbackfrom
the bottomof the stack.At higherfrequenciesnumericalstabilityand resonanceproblemscan be-
comeoverwhelming;morework remainsto be doneto discoversolutionsto theseproblems.

PUTTING LAYER ORDER BACK IN

Up to this point [ havepurposely"scrambledout" layer order,to forcethe layer-groupto be


commutative.If the layer-orderstatisticsaredefinedMarkov-style,i.e., the probabilityof eachtype
of layer occurringdependsonly on the previouslayer, it is possibleto augmentthe propagator
matrixto account
forthelayerstatistics.
LetPzrbetheprobability
of layertypeY occurring
given
that the previouslayer wasof type X. The Markov transitionmatrix for the two-layertype system
then looks like

p4,4
ps,4
).
P.4s Pss
(9)
104 Layer-Matrix Equivalent Media Systems

By makingp andPmbothsmall,we construct


a binarymediumwith exponential
probability
distribution/'unctionsfor the layer thicknesses.
We now mustcombinetheMarkov transitionmatrixin equation(9) with the propagatormatri-
cesof thesortpreviouslydefined[for example,in equation(6)]. I will demonstrate
howto do thisby
constructing the augmentedmatrixfor the samecasewe examinedin the previoussection,e.g., two
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

acousticlayer types.Label the elementsof the propagatormatrix for a singlelayer of type A as


follows:

Art
At2)
A2 A22
(lO)

(andsimilarlyfor layertype B). We requirean augmentedmatrixthat simultaneously carriesalong


both the probabilitiesof that set of layer-typetransitionsoccurringand the associatedwavefield
variables.
This can be accomplishedby usingaugmentedvariablesthat carry alongboth wavefield infor-
mationandmemoryof the previouslayertype.Extrapolatingacrossa layero[ typeA thenbecomes:

0 20 40 60 80 100 120

FIG. 1. Top: The ensembleaverage of the wavefield reflected from 50 different realizations of a
Goupi!!aud layer stack 50 layers thick. The direct reflection off the top of the stack arrives at
time 10; the reflection from the bottom of the stack arrives at time 60. All three plots are
drawn to the same absolute scale. Middle- The result of replacing the layer stack with its
Backus average. Bottom' The result of replacing the layer stack with its ensemble-average
propagator matrix equivalent medium. Compare the phase shift in the reflected arrival at
time 60 on the three traces.
De!linger 105
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

(11)

Thesubvector
() gives
the
pressure
andvelocity
(orwhatevercontinuous
variables
aresuited
to
theproblem)
iftheprevious
layer
wasoftype
A;() isthepressure
and
velocity
iftheprevious
layer was of type B. (For any given realization,the previouslayer will havebeeneitherA or B, so
exactlyone of thesesubvectorswill be nonzero.)As you caneasily verify,the augmentedA matrix
in equation(l l) multipliesby the properprobabilityof this transitionoccurring,extrapolatesthe
wavefieldvariablesacrossa layerof typeA, andoutputsthenewwavefieldvariablesin theA subvector
slot(sinceasseenfromthenextlayerthepreviouslayerwill havebeenof typeA). We cansimilarly
extrapolateacrossa B layer matrix:

(12)

We nowhavewhatwe need:an arbitrarystringof matrixmultiplications suchasAABAABB


extrapolatesthe wavefieldthroughthe corresponding stackof layers,and simultaneouslyweights
the resultby theprobabilityof thatparticularstackrealizationoccurring.We cannowsumthe layer
matricesjust aswe did in equation(2) to averageoverall possiblerealizations.Summingthepropa-
gatormatricesin equations(11) and (12), we find that the ensembleaveragewavefieldemerging
from the bottomof a stackof N layerswith Markov layer statisticsis'

N
PA ?4

PVA

(13)

Here we assumethatthereareN layers,just beforethe firstof whichwe hada layerof typeA. If we


assumethe lastlayeris alsoof typeA, thentheensemble-average wavefieldmeasuredat thebottom
will havepressure
PandverticalparticlevelocityW(exceptthatthesewill bescaledby theprob-
ability thatthe last layer wasof type A).
106 Layer-Matrix Equivalent Media Systems

To calculatethe equivalentmediumwe mustperforman eigenvector-eigenvalue decomposi-


tionof theaugmentedmatrix.(We cancertainlydo thisnumericallyfor particularvaluesof thelayer
constants
co,k.,,andk,.,evenif wecan'tfindananalytical
formula
forpowerseries
ofthese.)
Findthe
largesteigenvalue(or eigenvalues).
As N becomeslarge,the largesteigenvalue(s)will dominatethe
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

behaviorof the system,andthustheirassociated eigenvector(s)


mustdeterminethe limiting behav-
ior of the equivalentmedium.Any smallereigenvaluesmustcorrespondto components of the en-
semble-average solutionthat exponentiallydecayas the numberof layersincreases.Althoughnot
yet demonstrated, the magnitudeof the smallereigenvaluesthusshouldprovidea directmeasureof
how quickly the solutioncan be expectedto convergeto the limiting behavioras the numberof
layersincreases.

CONCLUSIONS

I haveshownthat it is possibleto exactlycalculatethe ensemble-average propagatormatrix for


layered-medium systemswith Markov-stylelayerstatistics, permittingan equivalent-medium to be
calculatedfor any stratified-mediumextrapolationproblem.Unfortunately,no similartrick hasyet
beenfoundfor exactlycalculatingtheensemble-average reflectedandtransmittedwavefields.From
the preliminaryresultspresentedhere,it doesappearthatthe ensemble-average propagatormatrix
methodproducesthe desiredbehavior.More work remainsto be doneto implementmethodsfor
stabilizingthe requiredfrequency-domain calculationsat high frequencies(Geistet al., 1990).

ACKNOWLEDGMENTS

I wish to thank FrancisMuir, Nell Frazer, Bertram Nolte, Mike Schoenbergand especially
JohnScalesfor many usefulconversations aboutequivalentmediaand reflectivitymethods.This is
Universityof Hawaii at Manoa Schoolof Oceanand EarthScienceand Technologycontribution
number 4497.
Dellinger 107

REFERENCES

Asch, M., Kohler,W., Papanicolaou,G., Postel,M., and White, B., 1991, Frequencycontentof
randomlyscatteredsignals:SIAM Review,33, 519-625.
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Backus,G. E., 1962, Long-waveelasticanisotropyproducedby horizontallayering:J. Geophys.


Res., 67, 4427-4440.
Banik, N. C., Lerche, I., and Shuey,R. T., 1985, Stratigraphicfiltering part I-derivation of the
O'Doherty-Anstey formula: Geophysics,50, 2768-2774.
Claerbout,J. F., 1976, Fundamentalsof geophysicaldata processing: McGraw-Hill.
Dellinger,J., and Muir, F., 1993, Dix revisited:a formalismfor raysin layeredmedia:J. Can. Soc.
Expl. Geophys.(5IWSA meetingspecialissue),29, 93-97.
Dix, C. H., 1955, Seismicvelocitiesfrom surfacemeasurements: Geophysics,20, 68-86.
Frazer,L. N., 1994,A pulsein a binary sediment:Geophys.J. Internat.,118, 75-93.
Geist, K., Parlitz, U., and Lauterborn,W., 1990, Comparisonof differentmethodsfor computing
Lyapunovexponents:Progressof TheoreticalPhysics,83, 875-893.
Gilbert, F., and Backus,G. E., 1966, Propagatormatricesin elasticwaveand vibrationproblems:
Geophysics,31, 326-332.
Godfrey,R., Muir, F., andRocca,F., 1980,Modelingseismicimpedancewith Markov chains:Geo-
physics,45, 1351-1372.
Kennett,B. L. N., and Kerry,N.J., 1979,Seismicwavesin a stratifiedhalf space:Geophs.J. Roy.
Astr. Soc., 57, 557-583.
O'Doherty,R. F., andAnstey,N. A., 1971, Reflectionson amplitudes:Geophys.Prosp.,19, 430-
458.

Resnick,J. R., Lerche,I., and Shuey,R. T., 1986, Reflection,transmission,


and the generalized
primary wave: Geophys.J. Internat.,87, 349-377.
Rio, P.,Mukerji, T, Mavko, G., andMarion, D., 1996,Velocitydispersionandupscalingin a labo-
ratory-simulatedVSP: Geophysics,61, 584-593.
Schoenberg, M., andMuir, F., 1989,A calculusfor finely layeredanisotropicmedia:Geophysics,
54, 581-589.
Schoenberg, M., 1983, Reflectionof elasticwavesfrom periodicallystratifiedmediawith interfa-
cial slip: Geophys.Prosp.,31,265-292.
Shapiro,S. A., and Hubral, P., 1996, Elasticwavesin finely layeredsediments: The equivalent
mediumandgeneralizedO'Doherty-Ansteyformulas:Geophysics,61, 1282-1300.
Stanke,F. E., andBurridge,R., 1993,Spatialversusensembleaveragingfor modelingwavepropa-
gation in finely layeredmedia:J. Acoust.Soc. Am., 93, 36-41.
Downloaded 12/06/12 to 192.159.106.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

S-ar putea să vă placă și