effective processing of nonrepeatable 4-d seismic data to...

The concept of time-lapse (or 4-D)seismic monitoring is straightfor-ward: A baseline 3-D survey describ-ing initial reservoir conditions and asubsequent monitor survey (or sur-veys) recorded after conditions havechanged are calibrated, compared,and the intersurvey differences ana-lyzed and interpreted in terms of netvariations in pore fluid saturation,pressure, or temperature inside ahydrocarbon reservoir.

Seismic monitoring thus dependslargely upon whether “repeatability”of surveys is achievable. Ideally, thedifference between surveys shouldcontain nothing but the changes inreservoir properties. However, it isalmost impossible to obtain perfectlyrepeatable (or identical) time-lapseseismic surveys in the real world.Ross and Altan (1997) report thateven a “zero-time repeatability test”(in which two offshore 3-D surveyswere shot only one day apart by thesame crew using identical equipmentand survey geometry in an area of noproduction-related dynamic influ-ences) could yield visually coherentenergy on difference sections, if theprocessing sequence was not opti-mal.

Therefore, uniform processing oftime-lapse surveys is crucial for suc-cessful 4-D reservoir interpretation,especially when there are large dif-ferences in acquisition or in ambientrecording conditions. An importantcomponent of uniform 4-D data pro-cessing is cross-equalization toremove artificial effects caused bynonreservoir factors, such as changesin acquisition and/or processingparameters, so that time-lapse 3-Ddata will be identical everywhereexcept within the reservoir zone(where dynamic changes should beexpected).

This article presents a case studyof effective 4-D seismic data pro-cessing. Time-lapse seismic surveysmonitoring a heavy oil thermal EORsite in west Saskatchewan, Canada,were uniformly processed by asequence including: spatial realign-

ment, amplitude balancing, andcross-equalization. This caseinvolved three 3-D seismic surveyswith significantly different parame-ters—different types of source(Vibroseis, weight-drop, and dyna-mite), varying orientations in geom-etry, and varied spatial parameterssuch as line spacing and bin size. Wefound that carefully designed pro-cessing based on cross-equalization iscrucial for legacy data and wouldhelp minimize major differences andnonrepeatability in data acquisition,thus making it possible to extract reli-able information about a reservoir’sdynamic changes from time-lapseseismic data.

Data acquisition. East Senlac is closeto the Alberta-Saskatchewan border(Figure 1). The producing reservoir isunconsolidated fluvial channel sandof the Lower Cretaceous DinaFormation, which directly overliesthe erosional surface of Paleozoic car-bonate rocks. The 15-m thick sandhas average porosity of 33% and per-

meability of 5-10 Darcy. At about 730m, this shallow reservoir is highlysaturated with a viscous heavy oil. Tomake this oil mobile and producible,steam-assisted gravity drainage(SAGD) technology has been used,and high-temperature steam is con-tinuously injected into the reservoir.Three pairs of horizontal wells, A1-A3, were first drilled (Figure 1). Theupper well injects steam, and thelower well, a few meters below andparallel to the injector, produces oiland condensate from the heated zone(referred to as the steam chamber). Afourth infill horizontal well pair, A4,was drilled later to boost production.Three time-lapse 3-D surveys werecollected at various times to monitorsteam chamber growth, determinesteam sweep efficiency, and examinethe effects of reservoir heterogeneityon the distribution of the heatedreservoir zones.

The surveys were acquired withquite different field parameters. Thefirst (baseline) survey was shot in thewinter of 1990 for static reservoir

0000 THE LEADING EDGE JANUARY 2001 JANUARY 2001 THE LEADING EDGE 0000

Effective processing of nonrepeatable 4-D seismic data to monitor heavy oil SAGD steamflood at East Senlac, Saskatchewan, Canada

GUOPING LI and GREG PURDUE, PanCanadian Petroleum Limited, Calgary, Alberta, CanadaSTEVEN WEBER and RODNEY COUZENS, Sensor Geophysical Limited, Calgary, Alberta, Canada

Figure 1. East Senlac study area and cover of the time-lapse seismic surveys.

characterization. Vibrators were thesource. An I/O System I did therecording. A standard orthogonal 3-D acquisition geometry, oriented E-W, was used. Source and receiverlines were separated by 180 m; 60-msource intervals and 30-m receiverintervals yielded a natural bin size of15 � 30 m.

The second 3-D, a monitoring sur-vey, was recorded in September 1997after 18 months of continuous steaminjection. A truck-mounted weight-drop source was used to avoid dam-age to surface facilities surroundingthe steam plant and wells. The sur-vey used an orthogonal geometry ina NE-SW orientation. Finer line spac-ing and station intervals resulted ina smaller CDP bin size (10 � 20 m).The recording system was I/OSystem II.

Our third 3-D survey wasacquired in the spring of 1998, whensteam had been continuously injectedfor almost two years and the maxi-mum reservoir temperature hadachieved 260°C. Its purpose was tomonitor movement of the steam frontand prepare for new SAGD locationsin the eastern portion of the area. Weapplied a new proprietary 3-D acqui-sition technique, MegaBin, to recordwider-band seismic signals contain-ing higher S/N ratio, and better off-

set and azimuth distributions.Dynamite charges, shot from 6-mdrill holes, did not damage the sur-rounding surface facilities and wells.An I/O System II was again used.Parallel source lines and receiver lineswere spaced at 80 m. Source locationswere staggered relative to receiverlocations. Station intervals were 80m for sources and 40 m for receivers,producing CDP bins of 20 � 40 mwith a nominal fold of over 80 at off-sets equivalent to reservoir depth.

This was much higher fold than in thetwo previous surveys. Table 1 detailsacquisition information for the sur-veys.

Obviously, these 3-D surveyswere not optimally designed for time-lapse monitoring and certainly didnot meet generally accepted require-ments for repeatability. Therefore itis not surprising to observe big dif-ferences in the data. Figure 2 showstypical shot records from the 1997and the 1998 surveys. As can be seen


Table 1. Detailed acquisition information for all surveysField parameters

DateSource type

Geophones

Recording equipment

Record lengthSource line spacingReceiver line spacingSource intervalReceiver intervalTypical patch

Natural bin size

Survey 3

Spring 1998Dynamite1 Kg, single hole6-m burial depth

OYO 10 Hz6 over 20 m inlineI/O system II532 channels maxSEG-D format, FP gain3(12)-411(275) Hz (dB/oct)3/4 Nyquist linear phasenotch out3.0 s @ 1 ms80 m NW-SE80 m NW-SE80 m40 m19 lines � 28 stations1440 m � 1080 m20 m � 40 m

Survey 2

Fall 1997Weight-dropTruckmount16 pops per shot

SM-24 10 Hz6 over 20 m inlineI/O system II728 channels maxSEG-D format, FP gain3(12)-207(298) Hz (dB/oct)3/4 Nyquist min phasenotch out3.0 s @ 2 ms160 m NE-SW120 m NW-SE40 m20 m13 lines � 56 stations1440 m � 1100 m10 m � 20 m

Survey 1

Winter 1990Vibroseis1 vibrator center on flag8 sweeps, 6 seconds18-144 Hz linear6 dB/oct boostGeospace 14 Hz12 over 2 m inlineI/O system I360 channels maxSEG-D format, FP gain3(12)-180(75) Hz (dB/oct)notch out9.0 s listen3.0 s @ 2ms180 m EW180 m NS60 m30m8 lines � 45 stations1260 m � 1320 m15 m � 30 m

Figure 2. Raw shot records from 1997 (left) and 1998 monitoring surveys.

clearly, data quality varies remark-ably. The weight-drop source tends toyield lower-frequency signals (left)than the dynamite source. The latterapparently has, as planned, recordedan improved S/N ratio within awider frequency band. It is extremelyimportant to remove these acquisi-tion-related differences from the datasets during processing beforeattempting to infer reliable informa-tion about reservoir changes fromthem.

Data processing. Because of severenonrepeatability with the time-lapsesurveys, a special data processingsequence, aiming at preserving rela-tive amplitudes and intersurvey bal-ancing, had to be designed, tested,and applied. The sequence includedspatial realignment (regridding), pre-processing, surface-consistent decon-

volution, single-trace zero-phasedeconvolution, surface consistentamplitude recovery, various staticcorrections, detailed velocity analy-sis, NMO and CDP stack, poststackf-xy deconvolution, trace interpola-tion, migration, and cross-equaliza-tion. During each step, effort wasmade to ensure consistency from onesurvey to another in the selection ofprocessing parameters. This articlefocuses only on regridding, ampli-tude balancing, and cross-equaliza-tion.

1) Spatial realignment. The consider-able variations in acquisition geom-etry for the 4-D surveys have twoimportant consequences: varying binsizes and different orientationaldirections. Figure 3, a close up of theinitial CDP bin grids of the surveys,reveals the nonrepeatability problem.

Regridding was done at the verybeginning of prestack processing torealign the 4-D surveys onto a com-mon grid. This new grid aligned thesurveys in the NE-SW direction,which is nearly the azimuthal direc-tion of the 1997 and 1998 surveys.The new bin size is 20 � 40 m. Everytrace from prestack gathers of eachsurvey was reassigned according tothe new bin locations and dimen-sions. Traces were reorganizedaccording to their new bin positionsto form new CDP gathers. Afterregridding, processing continuedfrom prestack through poststack.Once the data were stacked, traceinterpolation produced smaller bins(20 � 20 m).

2) Amplitude balancing. To facilitatetime-lapse analysis and the possibil-ity of subsequent AVO analysis, dataprocessing was required to preserverelative amplitudes. Bad and exces-sively noisy traces were eliminatedearly in the processing sequence, and60-Hz power line noise was sub-tracted from contaminated traces.Offset-dependent spherical diver-gence gain corrections and an inelas-tic attenuation correction wereapplied to the cleaned-up data. Apass of surface-consistent deconvo-lution followed (source, receiver, andoffset waveform components wereresolved, but only source andreceiver components were removed).Surface-consistent amplitude scalarsadjusted the relative amplitude con-tributions of sources and receiversbefore traces were CDP stacked. Notethat these scalars were derived fromband-limited data, using only fre-quencies with the best signal to noisecharacteristics, but were applied tounfiltered traces.

Two complicating factors causedus to modify the basic amplitude-preserving processing. One was theobservation that data output fromsurface-consistent deconvolutionlacked the temporal resolutiondesired. This led us to insert a single-trace zero-phase spiking deconvolu-tion step immediately after the initialsurface-consistent deconvolutionpass. This did increase the effectivebandwidth of the data, but unfortu-nately single-trace deconvolutiondoes not maintain amplitude rela-tionships, especially in the presenceof noise. Consequently, after thisadditional deconvolution step, wesaw, for example, that amplitudes onnear-offset traces that were overlainby ground roll increased relative to


Figure 4. Prestack amplitude balancing. Shot record (a) after surface-con-sistent deconvolution and single-trace zero-phase spiking deconvolution;(b) after adding amplitude rescaling to (a); and (c) after adding a globalscale correction to (c).

a) b) c)

Figure 3. Differences in CDP bin geometries.

amplitudes on longer-offset tracesfree of ground roll (Figure 4a). Toaddress this problem, a rescaling step(after the single-trace deconvolution)restored the rms amplitude values ofindividual traces to what they hadbeen before the zero-phase deconvo-lution. Figure 4b shows a shot recordafter rescaling.

The other complication was theobservation that the front ends ofsome traces output from the decon-volutions were abnormally high inamplitude. The temporal amplitudebalancing first employed to addressthis problem naively failed to accountfor the fact that surface-consistentscalars derived from bandlimited

traces were subsequently applied towideband traces and therefore failedto produce the desired results. Thiswas easily corrected by applying aglobal scale correction that compen-sated for applying scalars derivedfrom filtered traces to unfilteredtraces (Figure 4c).

Figure 5 shows traces from amigrated stack section (a) before and(b) after properly normalizing ampli-tudes in the manner discussed above.The green box highlights differencesin waveform around the reservoirzone.

3) Cross-equalization. All data wereprocessed in a consistent mannerfrom prestack to poststack, but00severe differences in phase, reflec-tion strength and time shifts stillexisted. This is apparent in Figure 6,which shows three migrated stacksections along the same NW-SE line(in-line 95) before cross-equalization.The reservoir zone is highlighted forcomparison. No doubt, it would bevery difficult to reliably infer changesin fluid-flow patterns directly fromsuch data without cross-equalization.

Cross-equalization requires awavelet operator, estimated in a timewindow, to shape and match datafrom the monitor survey, A2, to thebaseline survey.

Theoretically, the differencebetween A1 and A2 after cross-equal-ization should equal zero everywhere(inside and outside the design win-dow) except in the reservoir zonewhere changes are expected. Ross etal. (1996) said the filter can containup to four basic elements: time cor-rections, rms energy balancing, fre-quency bandwidth normalization,and phase matching. They suggestedthat the filter operator be designedover a range of static reflectors thatexcludes the reservoir zone wheremeaningful changes in pore fluidstate or other reservoir propertiesmay occur. We found that this is notnecessarily the best approach.

We computed an operator from afairly long time window (250-1350ms) and used the baseline 1990 sur-vey as the reference.

Figure 7 shows data after cross-equalization. By comparing dataabove and below the reservoir, onecan see that time shifts, phase varia-tions, and even amplitude differences(all evident in Figure 6), have beenminimized, and that the sections in(b) and (c) now look almost identicalto the baseline section in (a). Somesubtle changes in terms of amplitude


Figure 5. Effect of prestack amplitude balancing. Migrated stack sections(a) without and (b) with amplitude normalizations applied subsequent toprestack single-trace zero-phase spiking deconvolution following surface-consistent deconvolution.

a)

b)

a) b) c)

Figure 6. 4-D migrated stack sections before cross-equalization: (a) baselinein-line 95 from 1990 survey; (b) monitor in-line 95 from 1997 survey; and (c)monitoring in-line 95 from 1998 survey. Note phase variations, amplitudedifferences, and time shifts.

and waveform within the reservoirzone are evident in all panels inFigure 7. These subtle changes areindeed the expected consequence ofthe changing reservoir conditions(steam chamber effect), as explainedin the next section.

Now let us consider the effect of

design windows on cross-equaliza-tion. Ideally, one would prefer toexclude the reservoir zone (wheredynamic change is anticipated) fromthe design window. Marine time-lapse studies use very large time win-dows (800 ms) that exclude the zoneof interest. In another EOR monitor-

ing study, Sun and Harrison (1998)also normalized time-lapse seismicdata in a 600-ms time window thatexcluded the reservoir. However,when we tested the sensitivity ofcross-equalization design windowson our data sets, we found that a longanalysis window containing thereservoir zone tends to outperformshorter windows excluding the tar-get.

Figures 8a and 8b show migratedsections from the monitoring sur-veys. The input to cross-equalizationwas from 1998 with 1997 as the ref-erence. The cross-equalization filtercontained frequency content balanc-ing, rms energy equalization, timestatic corrections, and phase match-ing. The design window was 250-1350 ms. Note that our reservoir isaround 710 ms. The output is givenin panel (c) and the difference inpanel (d). The cross-equalizationoperator worked very well.

We also tried a shorter cross-equalization operator on the sameinput section. It was designed froma window of 200-600 ms, with thereservoir zone excluded. Figures 9aand 9b show the new cross-equalizedsection and the difference.

By comparing Figure 9a (the out-put) with Figure 8a (the reference),one can see that continuity of reflec-tions in the time zone immediatelyabove the highlighted zone on theoutput is severely disrupted becauseof this short operator. The loss ofreflection continuity is amplified onthe difference section (Figure 9b)where much coherent energy hasleaked into the nonreservoir zone.

Ross et al. (1997) concluded thatphase correction is a major compo-nent of the total cross-equalizationoperation. It was also true in our case.Figure 9c shows the result of apply-ing an operator designed from thesame short window (200-600 ms) butexcluding the phase-matching com-ponent. One can clearly observe thefollowing consequences:

1) The cross-equalized data in a shal-low window (300-600 ms) inFigure 9c look very similar to thebaseline data in Figure 8a.

2) However, data below the designwindow (i.e., below 600 m) havedifferent phase than the baselinedata.

3) A large amount of reflectionenergy has leaked into the differ-ence section (Figure 9d).

To summarize, cross-equalization


a) b) c)

Figure 7. 4-D migrated stack sections after cross-equalization. The filteroperator was designed from a time window of 250-2350 ms. Input data tofilter are in Figure 6. (a) Baseline 1990 section as reference; (b) monitoring1997 section; and (c) monitoring 1998 section. Filter contains bandwidthnormalization, rms energy balancing, phase matching, and time correction.

Figure 8. Cross-equalization results using a long design window (250-1350ms). (a) In-line 95 of 1997 survey as reference; (b) in-line 95 of 1998 surveyas input; (c) output of cross-equalization applied to (b); and (d) differenceof (a) and (c). Filter contains bandwidth normalization, rms energy, phasematching, and time correction.

a) b) c) d)

Figure 9. Cross-equalization results using a short design window (200-600ms). (a) Filtered in-line 95 section; (b) Difference between (a) and Figure 8a;(c) Filtered section using an operator excluding the phase-matching compo-nent; and (d) difference between (c) and Figure 8a.

a) b) c) d)

is necessary to remove from time-lapse seismic data the differences notcaused by the reservoir changes, anoperator from a long design window(even including the dynamicallychanging reservoir zone) works bet-ter on our data sets, and phase cor-rection is important for effectivecross-equalization.

4-D seismic interpretation. Our 4-Dseismic data have strong seismicamplitude anomalies around all thehorizontal well pairs, A1-A4. Theseanomalous seismic signatures arecharacterized by a trough amplitudeincrease on the reflection at the topof the porous reservoir zone (Figure10).

This 15-m thick reservoir is welldefined by (at its top) a trougharound 705 ms and (at its base) a peakabout 715 ms. The base of the reser-voir, a boundary between soft sandsand an older hard carbonate rock,results in a significant impedancecontrast in this area.

Within the lower horizontal pro-ducer of the infill SAGD well pairA4, in-situ temperature/pressurelogs were recorded almost concur-rently with acquisition of the 1998monitoring survey. The in-situ mea-

surements show that the pressurealong the well bore within the reser-voir more or less remains at the samelevel as the injection pressure whilethe reservoir temperature hasincreased beyond 240°C over thewhole length of the horizontal leg.Engineering measurements and pro-duction history consistently confirmthat a sizable steam chamber has beengenerated around the well.

This steam chamber effect is wellreflected on the time-lapse seismicdata. Figure 10 shows traces of fouradjacent CDP bins, separated by 20m, extracted around horizontal wellpair A4. We have put the same CDPtraces, from different surveys, sideby side. One can clearly see thatreflections on all three traces in eachCDP group have the same character-istic patterns above the reservoir zone(705 ms), indicating that the over-burden has not been affected by thesteam flood. On the other hand, at thetop of the reservoir (705 ms), one canidentify increasingly stronger troughamplitudes, as indicated by the colorchanges. We interpret this anomalousamplitude variation as being causedlargely by a significantly lower veloc-ity and, to a lesser extent, by adecreasing bulk density as a directresult of steam flooding. The intro-duction of steam into the reservoirmay have caused partial gas satura-tions that slow down wave propa-gation in the reservoir and decreasedoil viscosity that softens the reservoirformation. Both weaken acousticimpedance. We also believe that,although started initially at the baseof the reservoir, the steam chamberhas actually grown upward andreached the top of the reservoir, atleast in the vicinity of well A4. Theseobservations and interpretation arestrongly supported and validated bythe in-situ downhole tempera-ture/pressure logs collected in theA4 producer.

Figure 11 shows time-lapse pro-files from the 1990 survey through the1998 survey. These enlarged sectionsare cut along the same line across thetoes of the four SAGD horizontal wells.The time-lapse effect of steam cham-ber migration on the seismic is evi-denced again by an increasingly strongamplitude event at the reservoir top.After 24 months of steam injection, thereflection from the reservoir top isstrong and thickening, compared tothose on the 1990 survey (beforesteaming) and the 1997 survey (after18 months of steaming).

Figure 12 displays time-lapse seis-


Figure 10. Time lapse trace-to-trace comparison, illustrating the effect ofsteam heating on seismic signatures.

ONE-SIXTH VER-TICAL 54 X 119 mm

1/6 V

mic sections along the horizontal tra-jectory of well pair A4. Obvious againis the significant influence of long-duration steam flooding (18-24months). Note the remarkable varia-tions in amplitudes before and aftersteaming. The steam chamber clearlyhas expanded around the well.

The difference sections providemore quantitative information aboutthe net changes in a reservoir withtime. In Figure 13, two difference sec-tions, along horizontal well A4, effec-tively demonstrate the length of thesteam chamber around the wellboreand its vicinity. Production logging inthis well confirms our seismic inter-pretation that, during the 24 monthsof continuous steam injection, thereservoir rock in the neighborhood ofthe whole length of this horizontal

well has been heated effectively andevenly.

Our 4-D interpretation also indi-cates that the steam chambers are notequally developed along all well-bores. At some locations, the steamflood patterns seem more spatiallyheterogeneous and unheated spots(namely, reflections without time-lapse changes) have been identified.

Discussion and conclusions. Thiscase study shows that, when pro-cessing time-lapse seismic data, pre-serving relative amplitudes (in amanner similar to AVO processing)and cross-equalization are key com-ponents of the sequence. Even whenmajor differences in acquisition para-meters exist (resulting in significantnonrepeatability of the data), cross-


Figure 11. Steam chamber developing across the pattern of four SAGDhorizontal wells.

ONE-THIRD VER-TICAL

54X249 MM

1/3 V

equalization can satisfactorily removemost of the nonreservoir differences(or artifacts) and permit reliableextraction of reservoir information.

Our study also shows that thecross-equalization design window isimportant. For time-lapse seismicdata such as those described here, alonger window that includes the zoneof interest is preferred. One possibleexplanation for this surprising situa-tion is that a longer design windowwill have better statistical balancingcharacteristics on seismic signals thana short window. It holds true espe-cially when the shallow parts of theseismic sections above the reservoirzone are inconsistent due to rapidgeologic and near-surface changes.

In particular, it is likely true for shal-low reservoir targets.

Although repeatable acquisitionin time-lapse seismic monitoring isalways very desirable, strictly repeat-able seismic data may not be obtain-able. Actually, most often one has todeal with very different 4-D data setsfor monitoring. Nevertheless, oneshould always pay attention to care-ful planning and implementation oftime-lapse 4-D seismic data acquisi-tion to minimize unnecessary nonre-peatability. When legacy data have tobe used or when it is impossible toacquire repeatable time-lapse data,uniform processing becomes veryimportant and sometimes essential.


Figure 12. Steam chamber developing along horizontal well pair A4.

ONE-THIRD VER-TICAL

54X249 MM

1/3 V

Suggested reading. “MegaBin land 3-D seismic: Toward a cost effective ‘sym-metric patch geometry’ via regularspatial sampling in acquisition designand cooperative processing for signif-icantly improved S/N and resolution”by Goodway and Ragan (CSEGConvention, 1996). “Time-lapse seismicmonitoring: Some shortcomings innonuniform processing” by Ross andAltan (TLE, 1997). “Inside the cross-equalization black box” by Ross et al.(TLE, 1996). “Time-lapse seismic datanormalization and calibration” by Sunand Harrison (Joint CSEG/CSPG/CWLS Convention, 1998).

Acknowledgments: We acknowledge the sup-port of Sung Youn and Chayan Chakrabarty,and the contributions of Brent Ragan, AlGoodfellow, and Rick Filafilo during theMegaBin 3-D survey design and acquisition.Trevor Hamill was involved in the design ofthe 1997 survey. Many thanks to GaryLaskoski, Debby Guy, and Bev Bridger fordata management assistance. SensorGeophysical processed all 4-D data. Finally,but not least, we thank the management atPanCanadian Petroleum Limited for supportand permission to publish this paper.

Corresponding author: [email protected]


Figure 13. 4-D seismic difference section. Highlighted zone shows that thesteam chamber is expanded along horizontal well pair A4. (a) Differencebetween 1997 and 1990 surveys and (b) difference between 1998 and 1990surveys.

ODD-SIZE HOUSE AD

a)

b)

effective processing of nonrepeatable 4-d seismic data to...

Documents