Você está na página 1de 4

A Theory for SeismicInterpretation

CH3.8

J A. C. Jacobs,Instut Franqaisdu P&role, PSI ResearchConsortium, France

SUMMARY

the reasoningprocessfor interpretationprocess.The paperconcludes with a comprehensiveexample of reasoningduring the


determinationof a velocity model.

Integrationof geophysics
and geology becomesmore and
more important.On the one handthe number and the variety
of datathatare availableis increasing,On the other hand there
existsnowadaysa greatnumberof techniques
that contributes
to the imagingof (complex)subsurfaces:
III geophysicse.g.
travel time inversion,prestacktime- or depthmigration,forward
modellingtechniques
and in the geologicaldomain geological
modellingmethodsand reconstruction
techniques. A problem
thatis notyet solvedis theintegrated
useof all thesepossibilities
and the integrationof knowledgefrom one domain in the other
domain. It can be foreseen that this is needed in the near
future otherwise many of the sophisticatedtools will be too
restrictedin their application. Integrationcan only be reached
when its methodologyis well defined. To theseends,this paper
provides the resultsof epistemologicalstudy of the processof
seismicinterpretation.A generallyapplicablereasoningscheme
is presented,which allows integrationof all types of available
information. The schemewas testedusingthe resultsof a case
studyon real data (Jacobsand Jardin 1991). The exampleshows
the feasibility of the approach. This methodologymakes the
integrationof geology and geophysicspracticable.

INTERPRETATION AS A PROCESS OF
THEORY BUILDING
The seismicmethodcan be representedin a (too) simplified
way as in figure 1. There a certain number of processesare
appliedon the seismicdatain orderto obtaina certainsubsurface
representation.Of courseeach processis selectedsuchthat we
obtain (after a certain numberof processes)a consistentidea of
the subsurface.This is the classicalway to representthe method.
Interpretationactson the outputof certainblack boxes,reduces
or summarizesthe output.
In a more broadersenseinterpretationcan be viewed as a
processof envisioning,that is to say it is the processof forming
an idea about the subsurfacegeology. In order to gain more
insight in the generalprinciplesof seismicinterpretationit will
be comparedwith settingup a theory about the subsurface.A

Seismic
data

kd

Processing ~hterpretatio~

INTRODUCTION
Prestackstructuralinterpretationplays an importantrole in

Figure 1. The seismic merhod

seismicstructuralimaging over complex structures.It prepares


the input models(structuraland velocity) for the prestackinversion processes(traveltime inversion and prestackdepth migration) and its techniquesare used to interpret the resultsof the
inversion processes>This is a complicatedprocess,becauseof
the many different processesand data types that are involved.
Thereto a more structuredapproachto prestackinterpretation
was developed (Jacobsand Jardin 19!90),in which three main
types of interpretationare distinguished:Direct interpretation,
the interpretationof differences and the connectiveinterpretation. Interpretationmethodsfor all of the three typesexcept the
latter one exist or are developableasisolatedcases.However interpretationin generaland connectiveinterpretationin particular
are lacking a firm theoreticalfoundation,which is neededbefore
any techniqueor methodcan be developed,which integratesall
different aspectsof the problem. Techniquesand algorithmsfor
direct interpretationand the interpretationof differencescan be
basedmainly on parametricmethods.The connective
interpretation shouldbe basedon symbolicreasoning.In orderto develop

a theoryof interpretation
thatallowsan integrated
approach,
two
main subjectswill be discussedfirstly: The interpretation
process
from an epistemologicalviewpoint and a schemethat represents
231

goodtheoryis a setof logicalcoherent,non-conflictingassertions


concerningsomereality, suchthat it is possibleto derive testable
conjecturesor hypotheses.It is assertedherethatfor eachspecific
set of data we have to build a theory concerningthat specific
subsurface.The referential scopefor the theory are the seismic
data, the a priori information and other additional subsurface
information. The completesubsurfacetheoryis used afterwards
to calculate additional subsurfaceparameters(such as depth,
velocity, etc.). Generally we will use it for severalpurposes:
to explain certaindata, to predictfacts aboutthe subsurfaceand
to testcertainpremises.For eachpartof the subsurface
theorywe
will use someformal means,applycertainprocessesandgenerate
conclusions.Thus each stepis a buildingblock. One sucha step
in theory building (to be followed by explaining,predictingor
testing) can be generalizedas follows:
1. Observation (collecting and establishinghypotheses). In
this phaseall available informationaboutthe subsurfaceis
gatheredand the first conjecturesare made.
2. Induction (formulationof hypotheses).Departingfrom the
observeddata,one canstarthypothesizing
aboutsubsurface
parameters.This will result in certaingeneralitiesaboutthe

Seismic interpretation theory

subsurface.
3. Deduction (derivationof predictionsusing the hypotheses).
The hypothesesare acceptedfor the momentas true, which
allows to predict certain phenomena.
4. Testingof the hypotheses.This is the phaseof verificationof
thehypotheses.It will resultin therefutationor confirmation
of certain hypotheses.
5. Evaluation.This is in fact the sameas phasea. observation,
which closesthe loop.
The processof theory building is here presentedas an empirical cycle in which the feedbackbetweenthe variousphases
is essential. A remarkableconclusionis the existenceof both
inductionand deductionin the samecycle. Although the notion
of deductionis heavily criticized (a.o. Popper 1970) it seems
indispensablein our approach. In the seismic method we are
not only inducingusing the data in order to create a subsurface
model, but we are also applying certaingeneralgeophysicalor
geologicalprinciples(the theory) from which by deductioncertain postulatescan be formulatedconcerningthe data. This cycle
can be illustratedby the following example: Supposea typical
phenomenaas a velocity pull-up is observed
(phase1.) in seis-

mic data. It can be conjectured


(phase2.) thatprobablysaltis
presentabovethatpull-up. So it can be postulated
thatin a velocity scanat that specificpositiona relatively high velocity will
be present(phase3.). The next step is to verify this hypothesis,
for instanceby actuallymaking that velocity scanandto check
the existenceof a velocity anomaly(phase4.). In the last phase
the effect of this hypothesis(eitherrefutedor confirmed)is taken
as a new observationand the cycle startsagain.
T

/t\
H

or induction.However at the samemomentH, is verified by deriving 01~. One level up it can be seenthat theoryT is derived
by regressive
reductionfrom hypotheses
HI and HP, while H3
can be derivedfrom T with 03 as the actualderived observation. Finally we need to reach the ideal situation,where from
the theory all hypothesesand observationcan be derived. That
is to say all arrowsshouldpointdown or the theoryexplainsall
observations.
From this first approachit can be inferred that epistemological researchin seismicinterpretationremovesits seemingly
inexact nature. This should be sufficientif the processes(that
are at the basisof the observations)shouldlead to observations
that were independentof the processes.However the observation carry the characteristicsof the underlyingprocessin them,
Thereforeif we cannotapply this schemastraightforwardon our
interpretationproblem as a general schemefor reasoning. To
solve this problem the notionsof attributesand influenceshave
to be introduced
A MODEL FOR REASONING DURING
INTERPRETATION

Sofarit hasbeendemonstratedthat interpretationis a process


of observation,followed by inductionanddeductionof facts. The
underlyingprocessesare invisible. In thatview the underlying
processes
werelessimportantandconsidered
to be independent
of thegeneralline of reasoning.
In orderto createa usefulinterpretationscheme,the behaviorof a processand the parameters
that influencethat processare essentialand have to be included.
Interpretationthat reasononly using the outcomeof a process
without taking the behavioralpropertiesof the processinto account will fail in reachinga consistentresult. It shouldbe emphasizedthat the underlyingalgorithmicprocess(as long as the
processgeneratesthe requiredinformation)is of lessimportance.
In orderto includein a reasoningschemetheprocess
behavior

we introducetwo additionalfactors:the attributeof and influence on an observation. Both can either processdependentor
observationdependent.The observationdependent(od) attribute
indicatesthe presenceor absenceof a specificpropertyof that
observation.Theseare the size or extendof an observationor its
associateda priori knowledge. The processdependent(pd) attribute relatesinformationaboutthe underlyingprocess
directly

Figure 2. Theorybuilding

In reality explanation(induction)and verification(deduction)


are usedsimultaneous.One tries to explainan observation,which
is often directly followed by a verification. The collected hypothesesare usedto establish
the subsurface
theory.Following
Bochenski(1962) the developmentof a theory can be depicted
as in figure 2. On top we find the theory (T), at the base the
observationsare presented(01,...,Oa) andin betweenthe hypotheses(i-i, ,..., Ha), The up-arrowsstandfor explanation,while
the downarrowsstandfor verification. E.g. observations01
and 0j2 lead to the hypothesisHA trough regressivereduction

232

to the observation
(e.g. the pd-attribute
of a velocitypull up
is a strong lateral velocity variation). In the reasoningscheme
thesetwo attributesplay the role of backgroundinformationand
justification of the observation. In generalinfluencesare the

conditionalfactorsthatplay a centralrole duringthe reasoning


process.They are decisivein the sensethatdepending
on their
informationa certainconclusion
is rebutted.An exampleof an
observationdependentinjluenceis: The velocityasderivedfrom
well data,which representsa greatcertaintyaboutthecorrectness
of thevelocity value. Process dependentinfluencesrepresent
directly the limitationsof a processingstep. E.g. irrespective
of

Seismic interpretationtheory

therealvelocitydistribution,
constant
velocityanalyses
arecommonly madein orderto obtaina stackedseismicsection.Our
observation
howeverconsists
of a setof moreor lesshorizontal layerswith constantvelocities.The behaviorof the process
hasan essentialinfluenceon our observation
and may prevent
a right conclusion.
Takingtheseall togetherresultsin the followinglist of elementswhich shouldbe includedin a reasoningscheme:observations,
attributes
(asbackground
andwarrant),influences(as
rebuttals)and a conclusion.In figure3. the generalmodelof
reasoningis depicted(Toulmin1960). The schemeconsistsof
the followingelements:

Qualifier

Backing

Althoughall elementsin thisschemearegivenas singleelements,it is nevertheless


possiblethatan elementconsistof a
numberof itemsas long as theyare not mutuallyexclusive.In
thisscheme
twodirections
of reasoning
canbe found:Thebacking andthewarrantcontributes
in a deductivesense.They try to
verify theobservation.
Goingfromobservation
to conclusion
is
typicallyinductive. This schemerepresentsa small part of the
scheme
aspresented
in figure2. Thedifference
withthatscheme
is the additionof influencesand attributes.
A schemeaspresented
hereabovecan be filled up in many
differentways. The examplegiven startstypicallywith a fact
derivedfrom seismicdata. As will be shownin the real data
exampletheobservation
can alsoconsistof theresultof a processingstep(whichis of coursealsosometypeof interpretation).
while therebuttalcomesfrom anotherobservation.
There are a few additionalremarksto be made. Firstly
observations
can belongto a so calledmacro-or microworld.
This impliesthat differentattributesand/orinfluencesneedto
be considered
for the sametypeof conclusion.We call thisthe
regionalizationof thereasoning
process.Secondlyit is possible
thatthereareno warrants,
backingor rebuttals
available.In that
casesomedefaultsneedto be selected.This doesnt change
anythingin our reasoningscheme,but affectsin an important
way therelationsbetweenthedifferentcompleted
schemes.

Conclusion

AN ILLUSTRATIVE EXAMPLE
Thisexamplecomesfroma testusingrealdata.The scheme
for thisexampleis presented
in figure4.

Figure 3. Reasoning
scheme

0: Observation
(or seriesof observations).
This is always
the startof the line of reasoning.A velocitypull up.
W: The warrantcontainsusuallythe generalinformation
aboutthe whole possibleclassof observations.It is the
attributeof our observation.
Velocitypull upsarecausedby
stronglateralvelocityvariationswith somelateralextend.
B: The backing contains additional information that
strengthen
thewarrant.Stronglateralvelocityvariations
with
somelateralextendareoftenfoundnearsaltbodies.
Q: The qualifierindicatesthe belief one can have in the
conclusion.This is often expressed
in termsof: For all
casesor highly probableetc.
RB: Therebuttalcandisprove
theresultto beobtained.Here
the essentialinfluenceon the observation
comesin. It is
expressed
in termsof: Unlessa velocityanalysisdoesnt
showany stronglateralvelocitychange.
C: Theclaimorconclusion,
whichis in facta newhypothesis

On a real seismicdata set four differenttechniques


were
appliedin orderto definethevelocitydistribution:A classical
constantvelocityanalysis,a shotorientedanalyses,
modelingby
raytracingand prestack
traveltimeinversion.
1. The analysisstartedwith a constantvelocitystackfrom
which a stackwas produced. The observation
is a constant
velocitydistribution
withnearlyhorizontal
layerboundaries.
The
dependentattribute). As
stackis reasonable
well (observation
influenceswe note the inherentlimitationsof fhe process(it
assumes
on forehandhorizontallayering)andtheclearlyvisible
dipsin the zerooffsetsection.Thusthe conclusion
is thatthis
velocitydistributionis unlikely.
2. Throughprestack
interpretation
we obtaina certainnumber of traveltimesfrom eachshotrecord.It is testedwhether
thesetraveltimessatisfiesa simplelinear velocitydistribution.
The processinfluenceis theaccuracyof theresultandobservationinfluenceis thelackof differentialcompaction.
Observarion
dependentatrriburesare theinformationaboutthetypeof velocity distributionand the basicassumptionthatvelocityboundaries
usually coincide with structuralboundaries. In conclusionthis
resultsin a likely dipping linear velocity distribution.

or observationto be tested. Dependingon the rebuttal the


conclusionwill be there is salt or there is no salt.

3. Modelling by raytracing.Using the resultsfrom interpre-

tationof zerooffset andstackedsections(processnndnhsewafion

233

Seismic interpretationtheory

atfributes)andtheconclusion
fromstep2. (observurion
attribute)
the kinematicresponsefrom a deepevent was calculatedand
compared
with thereal seismicdata. Undertheassumption
that
the deepeventis a normalreflector(observation
influence)it is
inferredthata fan shaped
velocitydistribution
is highlyprobable.

Other type o
velocity
distribution

4. The fourth observationis the velocitydistributionas


calculatedby prestacktraveltimeinversionin which a priori
informationaboutdeepreflectors
is takeninto account(process
afrribufe).In the sedimentary
sequence
the velocityboundaries
are almostperpendicular
to thepresumed
geologicalboundaries.
Otherattributeinformationcomesin from step3. The residues,
whichare a measurefor the qualityof the inversion,are low
(processinfluence).Howeverthe backingof the observation
consistof two contradictory
attributes.Hencewe end up with
two really differentvelocitymodels.

Stack shows Limitationof


reasonable
process
and
image
i
Interpretation
zero offset
shows dips
c

Followingour reasoningschemewe haveto verify thetwo


contradictory
hypothesis,
a fan shaped
velocitydistribution
anda
distribution
fromprestacktraveltimeinversion,by derivingfrom
themotherpostulates.
If oneof thesepostulates
fits into the set
of observations,
we can accepttheunderlyinghypothesis.

1velocity boundaries

1e

CONCLUSIONS
It hasbeenshownthatseismicinterpretation
canbe consideredasa logicalprocess
of deduction
andinduction.By addingto Raytrace
a generalreasoning
scheme
thenotionsof attributes
of andinflu- tests
enceson observations,
a scheme
hasbeendeveloped
in whichail
externaleffectson thereasoning
process
aretakeninto account.
The exampleshowsthe feasibilityof thisapproach.Integration
of geophysics
andgeologyis the key for successful
imagingof
the subsurface.
This methodology
offersa goodpossibilityfor
this.
ACKNOWLEDGMENTS

Highly prob
t

Velocity

The datafor thisresearch


werekindlyprovidedby TOTALCFP. The research
wascarriedout aspartof thePrestack
StrutturalInterpretation
consortium
project(PSI).The authorshereby
acknowledge
thesupport
providedby thesponsors
of thisproject,
REFERENCES
Bochenski
I.M. 1962: Wijsgerigemethodenin modemeWeten
schap,Aula, Amsterdam/Antwerpen.
Jacobs,J.A.C. and Jardin,A., 1990: Prestackinterpretation
01
seismicdataovercomplexsubsurface
structures.
60 * Am
SEGMeering,Expanded
Abstracts,
215-218.
Jacobs,
J.A.C.andJardin,A., 1991:Prestack
structural
interpre
tationof seismicdata;a casestudy61 AnnualSEGMeeting
ExpandedAbstracts.
ToulminS. 1962: The philosophy
of Science;an IntrodUCtiOn
HarperTorchBooks513, New York.

234

Figure4.

Fan shaped
vel. distribution
/

Você também pode gostar