Seismic Inversion 1: Analysis

Monday Afternoon, November 11th

Comparison of PoststackSeismicInversion Methods

Downloaded 08/20/15 to 155.69.4.4. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

s11.1
Brian Russell* and Dun Hampson, Hampson-RussellSoftware ServicesLtd., Canada

equation,whichcanbe written

SUMMARY: st = [rt i wt t nt] - at, (1)

Single-tracepost-stackseismicinversionis a processing where st = the seismictrace,
rechniquewhoseaim is to extract acousticimpedance
rt = the earth’sreflectivity,
informationfrom stackedseismicdata. In principle,the wt = the seismicwavelet,
processis straightforward, and involvesthe assumptions nt = additivenoise,
that the seismic trace can be modelled usin the
convolutionaland verticalincidencereflectioncoefBlcient at = amplitudescaling,
equations.In practice,thereare manyseparatemethods * denotesconvolution,
usedto performpost-stack inversion.Thesemethodscan and . denotesmultiplication.
be convenientlygroupedinto three basiccategories: 1)
classicalRecursivemethods;2) sparse-spikemethods; Equation (1) suggeststhat if we are able to remove(or
and 3) model-basedmethods. reduce) the noise component,deconvolvethe wavelet,
and restore the original amplitudes(i.e. remove the
The three types of inversion will be compared and amplitude scalar), we would be left with the earth’s
contrastedusingtwo modeldatasets, and two real seismic normal incidencereflectivity,whichis relatedto acoustic
datasets.The modelswill includea detailedsyntheticin impedanceby the equation
which the result is know, as well as the claisic wedge
model. The real datasetsincludea gassandexampleand
a carbonatereef example,both from Alberta. Basedon
theseexamples,a number of observations will be made
aboutthe variousinversionmethods.

INTRODUCTION: where Zt = p,V, = acousticimpedanceof layert,

Inversion of post-stackseismic data to reduce an

estimateof the earth’s acousticimpedanceKas been an and
objectiveof geophysicists for a numberof years,and the
publishedliterature containsdescriptionsof numerous In practice,we can never exactlyrecoverthe reflection
approaches(Lindseth, 1979; Oldenburg et al, 1983; coefficientsfrom the seismictrace,and therewill always
Cookeand Schneider,1983). Figure 1 is a summaryof be amplitude,noise and residualwaveletproblems. All
the currentlyavailableseismicinversionmethods. The post-stackinversion methods listed in Figure 1 are
mostcompleteapproachto inversioninvolvesthe use of therefore approximationsof the ideal situation. The
pre-stackdata, and two formsof pre-stackinversionare eneral conceptof post-stackinversionis shownin the
recognized: travel-time inversion (often called a owchartin Figure 2. The two inputs to inversionare
tomography)and amplitudeinversion(a simplificationof stackedseismicdata and a setof geologicalconstraintsin
which is AVO analysis). Post-stackinversion,however, the form of a model. Theseare combinedin somewayto
remains the most popular current method, probably produce the final inversion. The way in which the
because of its greater robustness and simpler informationis combineddependson the algorithmused.
assumptions.As seen in Figure 1, there are two major In straightforward terms,bandlimitedinversioninvolves
approaches to post-stackinversion,broadbandinversion integratmg the seismic data directly to produce a
and bandlimitedinversion. The broadbandmethodhas bandlimitedinvertedtraceand then derivingthe missing
been furthersubdividedinto the sparse-spikemethodand low frequencytrend from the <geological ,model. The
the model-basedmethod. Thesethreetypesof post-stack sparse-spike inversionmethodmvolvesestimatinga set
inversion, bandlimited, sparse-sike and model-based, of sparsereflection coefficientsfrom the seismicdata,
are discussedin detail by the t ree apers referred to constrainingthesereflectioncoefficientswith the model
above,and will be briefly summarized
R Rere. and then inverting the coefficients produce the
impedance. In model-basedinversion,we startwith an
First, we will summarizethe basictheorybehindall post- initial model of the earth’s geologyand perturb this
stackmethods.All are basedon the assumptionthat the modeluntil the derivedsyntheticseismicsectionbestfits
seismictrace can be modeled with the convolutional the observedseismicdata. It shouldbe pointedout that
the results of both the sparse-spikeand model-based
methods are similar, and have the appearanceof a
“blocky” looking impedance. The differenceswill be
discussed in the nexttwo sections.

876
 2 Poststack seismic inversion methods

MODEL RESULTS: the objective.Again, the geologicalconstraintusedwasa

vertically and laterally stretchedmodel based on the
Two models were used to test the three inversion
Downloaded 08/20/15 to 155.69.4.4. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

acousticimpedancederivedfrom a well logthat hadbeen

algorithms.The first wasa 6-tracesyntheticseismogram drilled into the reef. The resultsare shownin Figures3,
derivedfrom the sonicand densitylogsfrom a well which 4 and 5 for a few tracesclosethe well. All threemethods
encountereda carbonatereef. Smce the geological reduced good estimatesof the reef porosity. The
answer is known, the outputsof the various inversion gandlimited method, shownin Figure 3 producedthe
algorithmscould be judged basedon the known result. most laterally contmuousresults,whereasthe results
Two separategeologicalconstraints wereused,one which from the model-basedand sparse-spikemethodswere
wascloseto the known answer,and a secondwhichwasa more discontinuousin a lateral sense. However, the
smoothedversionof the knownanswer.As expected,the sharpness in the geolo ical interfacescamethroughmuch
inversionresultsdegradedwhen the poorerimtial model moreclearlyin both oBthe “blocky”methods,and a better
wasused. Specificresultsfor the threeinversionmethods matchto the originallog wasobtained. The sparse-spike
were as follows. The bandlimitedinversion reduceda resultsshownin Figure 4 were less detailed than the
.K smoothed
frequencylimited versionof the soniclog wit modelbasedresults.The bestoverallfit wasfoundusing
interface boundaries. The sparse-spikeinversion was the model-based method,asseenin Figure5.
sensitiveto the detectionthresholdusedfor detectin the
spikesand thereforemissedsomeof the loweramplitude
reflection coefficients. However, the interfaceswere CONCLUSION:
much better definedthan in the bandlimitedinversion.
Finally, the model-basedinversion produceda similar Three different post-stackseismicinversionalgorithms,
result to the sparse-spikemethod, but defined more were used to invert four datasets,two of which were
blocksin the final result. modelsand two of which were real examples. It was
foundthat the mostrobustof the threemethodswas the
The secondmodel datasetused was the classicwedge classicalbandlimited a preach. However,this method
model. In thiscase,we alsousedtwo differentgeological produceda smoothed,Pfrequency limited estimateof the
modelsas constraints,one whichwascloseto the actual Impedancewhichwasnot as sharpin detail as the other
resultand the otherwhichwasa constantthicknessblock. two methods.The bandlimitedmethodalsofailed in the
It was observedthat bandlimited inversionand model- case of a very “sparse”model. The sparse-spikeand
basedinversionwereboth sensitiveto the startingmodel, model-basedmethodsproducedvisually similar results.
but sparse-sike inversion was able to overcomethe The sparse-spike approachproducedsuperiorresultson
limitation oP an incorrectgeologicalconstraint. Of the a totally “sparse”model, but producedless detailed
three methods,bandlimited inversion performed the resultsihan &model-basedinversionwhen appliedto real
worst and sparse-spikeinversion performed the best. data. Althouehthe model-basedinversionmethodis the
Model-basedinversiondid not performaswell assparse- most intuititely appealing, it has to be carefully
spikeinversionbecauseof the difficultiesin converging constrainedto avoidthe problemof non-uniqueness.
to a sparsemodel, Bandlimitedinversion reducespoor
resultsbecauseof the noticeablewaveletePfects. This studyshowedthat thereis no absolutelyrightwayto
do post-stackinversion. However,it suggested typesof
data in which one inversion may be preferred over
REAL DATA RESULTS: another.

Two real datasetswere used to test the inversion

algorithms,a gas sand and a carbonatereef. The gas REFERENCES:
sandexamplewas a classic“bright spot”,and was taken
from an areawhereAVO inversronworksverywell. The Lindseth,R.O., 1979: Syntheticsoniclogs- a processfor
objectiveof doing post-stackinversionwas thereforeto s&%tigraphlcinterpretation:Geophysics, v. 44, p.
estimate the lateral changesin thicknessand acoustic
impedanceof the sand. In all cases,the modelusedwas
a verticallyand laterallystretchedversionof the acoustic Oldenburg,D.W., Scheuer,T., and Levy,S., 1983.
impedance derived from the producing well. As Recoveryof the acousticimpedancefrom
expected,bandlimited inversion produceda smoothed reflectionseismograms: Geophysics, v. 48,p.
estimateof the sand. The sparse-sike and model-based 1318-1337
methodsboth reducedsimilarlool?mg blockyinversions,
but the lateraP continuityon the model-basedinversion Cooke,D.A., and Schneider,W.A., 1983. Generalized
wasbetter. linear inversionof reflectionseismicdata:
Geophysics, v. 48, p. 665-676
The seconddatasetwas a carbonatereef in which the
lateral extentof a highporosityzone within the reef was

877
 Poststeck seismic inversion methods 3
Downloaded 08/20/15 to 155.69.4.4. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Bandlimitedinversionappliedto
carbonatereef example. Acoustic
impedancefromwell is shownas
heavytrace.
Figure 7 A summaryof currentseismic
inversionmethods.

Seismic Geological
Data Constreints
t t
Reflectivity Figure4 Sparse-spikeinversionof carbonate
Estimate %b”d”ei” reef example. Acousticimpedance
fromwell is shownas heavytrace.
L : 1

Figure2 Thegeneralmethodof post-stack

seismicinversion.

Figure5 Model-basedinversionof carbonate

reef example. Acousticimpedance
fromwell is shownas heavytrace.