Professional Documents
Culture Documents
By DAN IfAMPSON
Hampson-RussellSoftwareServices
Calgac, Cam&
b
Figure 8. Model and real data dlsplaysbeforeand a&r GLI.
Figure 7. Common offset stack producedby real data aud
modelgeneratedusingsonicand densitylogs.
concept,we can now say that the model so far perfectlymatches
Colonysand example.The real data to which theseconcepts the well-log informationbut doesnot acceptablymatch the real
will he appliedinvolvethe Cretaceous Colony sandof southeastern seismicdata.
Alberta. This sand, which has high porosity and often displays GLI will changethings.
good stratigraphictrappingconditions,is an excellenthydrocar- The top of Figure 8 showsthe initial guess(in whichthe sand
bon reservoir. The seismicdata also frequently contain classic layer is modeledwith a Poisson’sratio of 0.25, equalto the snr-
bright-spotanomalieswhich are indicativeof gassands. roundinglayers).The real datashowsignificantamplitudeincrease
Sonic and densitylogs from this area are shownin Figure 5, at both the top and bottomof the layer but the syntheticmodel
with an expandedview of the zone of interestin Figure 6. We doesnot.
havesimplifiedthe problemby approximatingit as threeuniform After GLI (shownat the bottomof Figure 8), the Poisson’s
layers:a sandencasedby two shales.The simplifiedlogsarecalled ratio within the sandlayer has been reducedto a value of 0.16
“blocked” to differentiatethem from the originals. andthe Poisson’sratioon the surroundinglayershasaceally been
Note thatthe majorP-wave responseis expectedto comefrom i&eased slightly to 0.3. At the sametime the thicknessof the
the changein density,not the c@ge in P-wave velocity. The drop layer hasincreasedfrom 7.3 m to about 10 m.
in densityis indicativeof the presenceof gas. As expected,the new syntheticmodel is a much &tter match
The objectiveof the inversionis to seewhetherwe can deduce to the real data. Therefore we have achievedthe model-based
the presenceof gasby analyzingthe conventionalseismicdata. inversionobjectiveof mat$ing the real data while not deviatini
The real datahavebeenprocessed to producethe commonoff- too far from the initial guess.However, a majorquestionremaius:
set stackshownin the right panel of Figure 7. This type of dis- How unique is this answer?
play is sometimescalledan Ostrandergatherandresultsfrom sum-
ming traces,fromseveralCDP gathers,which havethe sameoff-
set. of course,the summingcausessomesmearingof amplitude M onte Carlo analysis.Sincethe AVO probleminvolvessuch
information,but the benefit is increasedsignal-to-noiseratio. a small dataset,we might be temptedto try all possiblemodels-
Figure 7 alsoshowsa modelgeneratedfrom thesonicandden- an approachwe couldcall trial-and-errorinversion.But we quick-
sity logs. The model assumesthat the sand layer has the same ly see that this is impossibleeven if the phrase “all possible
Poisson’sratio as the surroundingmaterial-in other words,that models” is alteredto “all modelswithin somereasonablelimits.”
the sandlayer doesnot containg&s. For example,if we were to plot the thickuessof ihe gass+md
Figure 7 showsthat the real data contain a very significant layer (in terms of traveltime)againstPoisson’sratio, the accept-
AVO anomaly at the Colony sandlevel but the model, basedon able zone might contain solutionsin which the layer thickness
the initial guess,doesnot. In termsof the model-basedinversion rangesfrom 2-20 ms and Poisson’sratio from 0.2-0.4. This is
Poisson’s Ratio
Poisson’s Ratio
illustrated in Figure 9. Each dot in the upper right quadrant C onclusions. It is obviousthat full, multilayer AVO inversion
representsa solution.Supposewe decidedthatevaluating10 values is feasible usingcurrent technology.
(or dots) per parameteris sufficient to give us a good idea of the However, the magnitudeof the nonuniquenessproblem has to
entire zone. As previously explained, the AVO modeling tcchni- make US stop and ask: What do we really mean by inversion?In-
que that we are using has four parametersper layer. So, if we steadof thinking about inversion as a processwhich gives us the
modeled a six-layer caSe and the computer could generate one answer. we have to think of it as a process which gives US
completemodel per second, the time required to try all combina- reasonableanswers,i.e., answersthat fit the data.
tions of our 10 dots within the acceptablezone is 3 x 10lh years. Monte Carlo analysisoffers a tool for economic~y analyzing
This is quite a long time abouta million times longer than the es- that set of possible answers, but it does not remove the non-
timated age of the earth. uniquenessproblem. It’s the job of the interpreterto determineif
Thus our trial-and-error technique is not possible even for the most reasonableanswer that emerges from our inversion is
models with reasonablelimits. reasonableenough g