Sei sulla pagina 1di 2

176

Seismic 13

FIG.6. Reconstruction
with depthcorrection.100percentcontrast,
Cr = 2000 misec.

All of the perturbativeapproachesto multidimensionalwave


equationprocessing.for example. wave equationmigration(see,
e.g., Claerbout,1971;French, 1975:Schneider,1978;Stolt, 1978;
Sattleggeret al, 1980), or Born approximationinversion (see,
e.g., Cohen and Bleistein, 1979; Raz, 1981: Clayton and Stolt,
1981) require some input velocity information. In the Born approximation to inversion, a reference or backgroundvelocity is
chosenand a perturbationaboutthis velocity is determined.Similarly, a velocity model is a required input to all wave equation
migrationtechniques.
The purposeof this paper is to examine the sensitivityof the
Born approximationto this input information. One importantaspect of this questionis whether the perturbation,as given by the
Born inversion, will be of a corrective nature. In this context
corrective means that an improvementof, or correctionto, the
estimatedbackgroundvelocity takesplace after applyingthe Born
inversion.Specifically,one would like resultsof the Born approximation for the velocity to be closerto the actualvelocity than the
backgroundvelocity is. A secondaspectof this issueis the tradeoff betweenvelocity determinationand reflectormapping.That is,
one would hope that it would be possibleto determinecorrectly
both the location of a reflector and the velocity below that reflector. We illustrate, by meansof a simple example, these two
aspectsof the sensitivityissue.
Considera I-D acousticmedium where the propagationof the
wave field P (z ,r) is governedby:

ity contrasts(50 and 100 percent)in Figures5 and 7, distortionin


the reconstructions
is quite serious.The improvementof the depth
correctionshown in Figures 4, 6, and 8 is apparent.

Future work
We intend to examine the effects of the two nonlinearcorrections on different scatteringgeometrieswhich include single and
multiple scatterers.When this is done successfully,we intend to
apply the inversionalgorithmto real seismicreflectiondata.

Sensitivity of Born Inversion to the Choice

S13.6

of Reference Velocity: A Simple Example


A. B. Weglein, Cities Service; und S. H. Gray,
Amoco Production Research

We examine the sensitivity of the Born model to the input


backgroundvelocity. We use a one-dimensionalanalyticexample
to point out the difference between a corrective procedureand
merely a perturbativeone. We examine various aspectsof the
sensitivityissue,includingthe trade-off betweenvelocity determination and mappingof reflector location. Lnparticular,we show
that one choicefor the background,or reference,velocityCx leads
to an accuratedeterminationof the locationof the first reflectorbut
an inaccurateestimationof the velocity below that reflector. A
secondchoice for CR can reverse this situation, accuratelyestimating the velocity but not the reflector location. Also, there is a
rangeof choicesfor C, for which the resultsof an inversionmay
actuallyyield lessaccuratevelocityestimatesthanthe estimateCR
itself. Althoughthis problem is discussedwithin the contextof the
Born model, it is an issuecommon to all perturbativemethods
(e.g., migrationmethods)which transformsurfacereflectiondata
into a map of subsurfacereflectors.

(1)

wherec(:) is the local acousticvelocity. Characterizethe acoustic


velocityc (: ) in termsof a homogeneous
referencevelocityCK and
a variation in the index of refractionu(z) as follows:
1
-=$I
c*(2)

WZ)].

In equation (2), the number C, is an input into the model; it is


chosen(or guessed)before the inversionis carriedout.
For simplicity, we assumethat an impulsive sourceis used to
probethe medium, and that this sourceis locatedat; = 0, within a
half-spacewhere the acousticvelocity has the constantvalueC,.
Thus the incident field is given by P,(;,,) = 6(t -z/C,,). The
total field is the sum of the incident field and a scatteredfield
P,(z,t). Under these circumstances,Gray and Bleistein (1980)
showedthat, within the Born approximation,the variationin the
index of refractiona(:) can be expressedin termsof the scattered
field PJz ,t) at ; = 0 by
nz:r,
a(:) = -4

I 0

P,(o,t)c~r.

(3)

From equations(2) and (3), the reconstructedvelocity depends


explicitly on the choice of referencevelocity.
To simplify our example further, we assumethat the medium
consistsof two half-spacesin contact, with acousticvelocitiesC0
and C , and the interfacebetweenthem locatedat a depthZ0 (see
Figure 1). Then the reflectedfield returnedto a receiverat z = 0 is
given by

4v(O,r) = =

S(t - 2Z,,/C,J.

C, +c,,

For our example we set C,, = 1, C, = 917, and Z,, = 1, and find
from equation(4) that

Downloaded 19 Jan 2010 to 129.7.53.87. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Seismic 13
-fz=

0
t

FIG.

+Z

I. Two half-spaceswith acousticvelocitiesC,, and C

P,(O,f) = *

S(1~ 2) = $6(f - 2),

which is independentof the reference velocity C,<. Now we can


calculatethe Born CX(:)from equation(3):

From this result and equation(2). it is clear that


(,(,)

c,f_

v2

; cc,
c,, : > c,,

The superscriptB indicates a Born inversion velocity CH(:) =


C,( 1 + (Y)~ given by equation(3).
Now we examinethe effects of different choices,or guesses,for
CK on the quality of the inversion. lf we chooseCK = 1, then the
locationof the reflectorwill be correctlydeterminedbut the calculated velocity in the region : > 1 will be too large. On the other
hand, for a different choice ofC,( (i.e.. C,? = 9/7,\/? = ,908).
the velocity below the reflector is correctly determined but the
location of the reflector is incorrect. We see from this simple
exercise that it is impossibleto simultaneouslyfind the correct
locationand size ofa singlereflectorin the Born approximation.It
is also impossibleto simultaneouslypredictthe correctvelocity in
both the half-spaces.
Next, we showthat a certain rangeofchoices
for CH can lead to
velocity estimatesin the lower layer [via equation(5)] which are
actually further from the true velocity than CK is. Thus, a certain
rangeof values for C,! can lead to incersions which are not corrective. Moreover, it is true in generalthat the rangeof valuesfor CH
can contain valuesvery close to C,, (for this example the range IS
1.065< C,(). In fact, it can be shown that the closerC, is to C,,
(i.e.. the smallerthe computed(Yis in the secondhalf-space),the
closerthis noncorrectiveset of valuesfor C,, comesto containing
the numberC,,. This is dur to the fact that the bdundarybetween
corrective and noncorrectivechoices for reference velocity lies
betweenC,, andC , This itnpliesthat even thoughthe true valueof
N can be arbitrarilysmall for; > C,, , the reconstructedvelocity(,
can be a less accurateestimate for : >CCK than the reference
velocityC,<.This is a somewhatsurpriarngresult, in that the amallnessof the cyis usuallyusedto justify the Born approximation.In
this sense.the Born inverxionis extremely sensitiveto the choice
of referencevelocity.

177

Thus, this simple exampledemonstrateshow the choiceof referencevelocity in the Born methoddependson what is considered
the principalobjectiveof the inversion.We have shownthat if the
primary aim is to determinethe locationof the first reflectorthen
the optimal referencevelocity shouldbe chosento be the velocity
of the first layer. However, we have shownthat this is not the best
referencevelocity if the principalaim is to determinethe velocity
of the lower half space.In situationswherethe velocitystructureis
more complex, the choice of referencevelocity will againdepend
on the objective of the inversion. If, for example, the velocity
increasesor decreasessteadilywith depth, the choiceC,< = C,, will
lead to lessand lessaccuratemappingof reflectorswith depth(see,
e.g., Clayton and Stolt, 1981).
In conclusion, the sensitivity of Born inversion to the input
quantity CR leads to a trade-off between reflector location and
velocity determinationin the simplest possible example. Further
work will be requiredto evaluatemore complicatedmodelswhere,
perhaps,globalcorrectivemeasuresshouldbe defined. The choice
of a referencevelocity to optimize somejudiciouscombinationof
velocity determinationand reflector mapping could be useful.
These issuesalso need to be investigatedfor nonconstantbackgroundBorn inversion(see e.g., Clayton and Stolt, 1981; Weglein, 1982).
Explorationistsalreadyhave a qualitativeawarenessof someof
the issuesraisedin thispaperin relationto the analogousmigration
methods. This simple example, using the Born model, helps US
begin to quantify this understanding.
References
mapping:
Geophysics,
v. 36. p, 467-481.
Clayton, R.W., andStolt, R. H., 1981. A Born-WKBJ inversion method
for acoustic
reflectiondata:Geophysics,
v. 46. p. 1559-1567.
Cohen. I. K., and Bleistein, N., 1979, Velocity inversion for acoustw
Claerbout, J. F.. 1971, Toward a unified theory of reilector

waves: Geophysics, v. 44, p. 1077-1087.


French. W. S., 1974. Two-dimensional and three-dltnenaional migration
of model-experiment reflection profiles: Geophysics, v. 39, p. 265287
Gray, S. H., and Bleistein, N., 1980, One-dimensional velocity inwrslon
for acoustic waves: Numerical results: J. Acoust. Sot. Am. 6714). p.
1141-1144.
Raz. S., 1981, Three-dimensional velocity profile mversmn from tinite
offset scattering data: Geophysics. Y. 46, p. 837-842.
Sattlegger. I. W., Stiller, P. K.. Echterhoff, J. A.. and Hentwhkc, M K .
1980, Common offset plane migration: Geophys. Prosp.. v. 28, p. X59871.
Schneider, W. A., 1978. Integral formulation for migration in two and
three dimensions: Geophysics. v. 43, p. 49-76.
Stolt, R. H., 1978, Migration by Fourier transform: Geophysics. Y. 43. p.
23-48.
Weglein, A. B., 1982. Near-field inverse scattering formalism for the
three-dimensional wave equation: The inclusion of a priori Velocity information: 3. Acoust. Sot. Am. (to appear May 1982)

Downloaded 19 Jan 2010 to 129.7.53.87. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Potrebbero piacerti anche