Professional Documents
Culture Documents
based method we call Case Base Reasoning (CBR) was 5. Outer Ramp: Formed below the storm wave base, this
developed concurrently with the cloud transform method. zone is comprised of open marine bioturbated
CBR compares a facies guided library of known cases (cored wackestones and mud dominated packestones. Moderate
wells) to the current case (uncored well) for each one of the to low permeability is favoured.
newly defined forty-seven geological layers. When a statistical
best match is found, the permeability from the cored well is In addition to the above, two additional sedimentary packages
transferred to the uncored well. with permeability tendencies were identified:
These three field wide permeability estimations allowed us to 1. Dolomitized packstones / float grainstones in two
test and compare results for each geological layer and select geological layers favoured moderate to high permeability.
the better for that particular layer. 2. Float grainstones in one geological layer favoured high
permeability.
Traditional well-test analysis provides good insight into the
average properties of the reservoir in the vicinity of a well2. Dolomitization: Each layer’s permeability is influenced by its
The use of well test data to descripe reservoir heterogeneity3, degree of dolomitization though not always in the same
and to constrain and condition reservoir permeability models2,4 manner. In the shoal and lagoonal facies, moderate amounts of
has been reported in the literature. dolomitization degrade permeability, while near complete
dolomitization enhances it. In the peritidal environment of the
In our large model, the proper permeability estimator for each SE part of the field, dolomitization enhances permeability.
layer was also conditioned to test K, to produce some other
permeability realizations. Before conditioning, however, the Data Quality
test K data was scaled to core K which gave more vertical The validity of a reservoir model is a reflection of the quality
refinement. of the data used to construct it. Log interpretation, core
analyses and well test data suffer from data quality issues.
Geological Model
Eleven thousand feet of core material from thirty-six wells Log Interpretation: Log interpretation of early wells
were studied in detail, including thin section description to overestimates dolomite content ten to twenty percent.
establish a sequence stratigraphy driven layering scheme. Pervasive overestimation of reservoir porosity (13% to 17%)
is present in certain geologic layers. To correct the above, over
Lithology: The reservoir is composed of dolomite and half the wells in the field require reinterpretation.
limestone with intervening anhydrite layers. Core Data: Low porosity plugs often display elevated
Pore System Geometry: Wide variations in pore size to pore permeabilities. This is the effect of cracks or fractures6. To
throat ratio occur. Over 600 high pressure mercury injection address this, plugs with porosities of 6% or less with unusually
pore throat size distribution measurements with accompanying high permeability were removed. Cracked plugs above 6%
thin section descriptions enabled our geologists to identify porosity remain hidden in the database.
sixteen rock types which were later narrowed to nine.
Depositional Model: The reservoir is a distally steepened Permeability Estimation Methods
ramp characterized by thick regressive events or high stand ESTIME: This approach assumes for any rock physical
systems tracts, and thin trangressive systems tracts5. It covers parameter such as permeability, there is a unique log response.
a broad platform to outer ramp profile with a set of facies Core air permeability data is employed to create a cellular
characterizing each tract. Broad statements are possible about database. Each cell is built by assigning it the logarithm of
the permeability tendencies of each depositional setting permeability for a finite range. Cells containing multiple
despite the complicating influence of digenetic overprinting. values are averaged to produce a single permeability. The
number of permeability values used to compute the average is
1. Peritidal (Inner Most Ramp): Comprised of anhydrites, recorded. Specific ranges of NPHI, RHOB, and MSFL derived
stromatolites and vermicular or patterned dolomites. porosity are equated with a specific permeability value within
Permeability is generally low. each cell. Conflicts in the database are resolved. If the three
2. Lagoon (Inner Ramp): Here evaporatic flats and lagoonal log values equate with three different cells, each with a
deposits occupy the platform. The lagoonl facies are different permeability value, then a weighted average
bioturbated algal wackestones and packstones. These permeability is predicted. A cell whose permeability was
textures favour moderate to low permeabilities. derived from the greater number of core values is weighted
3. Shoal (Ramp Crest): These are high-energy deposits more than other cells. For example, suppose one cell contains
dominated by shoal facies comprised of foraminiferal a permeability average of log1020 md from three original
packstones and grainstones. Rock textural qualities values, the second cell contains log1050 md obtained from two
display a broad range of permeabilities from low to high, original values, and our third cell contains log10200 md, a
but favour moderate to high values. single original value, the average is
4. Rudstones (Ramp Margin): Coarse grain rudstones [(log1020*3)+(log1050*2)+(log10200*1)]/[3+2+1] = 39.8 md.
forming discontinuous prograding clinoforms occur at the Empty cells containing no permeability value for the log
margin of the shoal facies. Permeability is high. values are handle by linear interpolation between filled cells.
Since each point used to create the prediction model is include
SPE 84599 3
in the prediction of any cored well, the bias is towards Case Base Reasoning (CBR): Geological mapping revealed
acceptable prediction for that particular well. reservoir permeability magnitude and distribution is
influenced by: (1) geologic layer, (2) facies within a geologic
Advantages cycle, and (3) depth (base to top). These concepts were
• Links permeability variations to three log responses. developed into a statistical software program we call “Case
• Applicable to hydrocarbon and water filled porosity. Base Reasoning” or CBR. Well log-derived attributes from
uncored wells are statistically compared to those of a library
Limitations of uncored wells within a single geologic layer and facie. The
• Misses high permeability streaks. presumption is; similar log patterns and values from different
• Smoothes the data. Poor dynamic range. wells (same layer and facie as judged by the reservoir
geologist) equate with similar permeability. Similarity is
Cloud Transforms (CLTD): Each reservoir depositional judged based on a weighted comparison of the standard
setting is dominated by grainstones, packstones or deviation and mean of porosity and lithology. (Lithology was
wackestones. These rocks displayed different permeability awarded a low weight due to the previously mentioned data
distribution patterns (Figure 1). This suggested a statistical quality issues.) In other words, “if they look the same, than
permeability estimation guided by depositional setting and its their other attributes must be alike”. Once the most similar
dominant rock fabric is appropriate. Cloud transforms are one match of an uncored well to an uncored well is found, a
such statistical method7. Cloud transforms stochastically transfer of permeability from the cored well to the uncored
estimate one parameter (permeability) from another, i.e. well occurs. This is accomplished through recognition curves
porosity. Each geological layer and facie defined by the (RC). Recognition curves are the weighted product of porosity
geologist was cross-plotted, core porosity versus core and lithology calculated along the depth track of both cored
permeability. This resulted in eighty-three geologic layer and and uncored wells. A moving window of a user selected fixed
facie related individual plots. Some geological layers were length (one or more depth levels) is progressively migrated
combined where properties were similar. Bins of 2.5 porosity over the recognition curves. At each depth level, the window
units width were created along the X-axis of each plot. One of covering the RC of the cored well is shifted up and down one
three (Log10) permeability distributions (log normal, normal, or more depth levels. At each station, the average RC is
triangular) was assigned to each of these porosity bins based calculated. When the minimum absolute difference in RCs’ is
on permeability distribution analyses with the software encountered between the uncored well and the cored well, the
package Best Fit (manufacturer Palisade). Log10 core data sets two RC’s are considered equivalent. Permeability from the
too few in number (a frequent occurrence) to quantify the current window position is transferred from the cored well to
distribution type, were assigned a normal distribution. A the uncored well. Both windows are then depth shifted one or
random number generator duplicates each permeability more depth levels and the process repeated (Figure 3).
distribution or any single data point belonging to that
distribution. For each well, using log porosity and the Advantages
Reservoir Geologist’s assigned geological layer and facie, one • Maintains the permeability heterogeneity of the
permeability probability value was generated at one foot formation. Averaging of the data is not performed.
spaced depth levels using the appropriate cloud transform and • Permeability depth profiles are maintained.
porosity bin. Figure 2 illustrates the fundamental elements of • Applicable to hydrocarbon and water filled porosity.
this technique.
Limitations
Advantages • Boundaries between facies within the database are not
• Reproduces the permeability frequency distribution gradational. Sharp changes in permeability range can
within each reservoir layer and facie. occur if the geologic layer is classed as lagoon, outer
• Maintains the dynamic range (no smoothing). platform or shoal.
• Applicable to hydrocarbon and water filled porosity. • The assumption, if recognizable physical features of the
layer are similar, than the permeability must be similar,
Limitations may not valid for all geologic layers. For example, one
• Permeability depth trends are not maintained. layer displays little in the way of porosity or lithology
• Log porosity is a poor proxy for core porosity in the case character to differentiate one well from the next, yet large
of the Peritidal layers. Log porosity smoothes the porosity changes in permeability do occur. Since the changes in
profile displaying much less heterogeneity than is present permeability are rhythmic cycles, the depth processing of
in core. the CRB reproduces these rhythms but not their exact
• Cloud transforms from layers and facies with sparse core placement.
data sets are not statistically representative of the
reservoir. Validation
• Presents a more heterogeneous picture of the reservoir Building confidence in the predictive models requires data
than is actually present. from different sources providing supptive answers. Used
collectively they comprise:
4 SPE 84599
Blind Tests: Estimate the core permeability of a cored well KH Cross Plots
(which also matches well test) without the benefit of the core. The field is subdivided into six main sub-reservoirs. Test KH
3D Visualization: 3D visualization ensures the plausibility of data was cross-plotted against the KH of the three
a model. It allows us to step back from an interpretation and permeability estimators for the wells of any of these sub-
“look”, either the model makes sense based on what the reservoirs, as shown in Figure 5. On these plots, the half-
geologist and engineer knows about the geologic and flow decade lines are plotted on both sides of the ideal 1:1, 45° line.
characteristics of the reservoir, or it does not. They represent an acceptable level of uncertainty in the data.
Bootstrapping: Select only uncored wells to build a new The following conclusions were made about the six main
model. With 270 wells minus the cored wells, the new model reservoir layers.
should look like the cored wells at the cored well locations.
Water Flood Dynamics: Areas of rapid and slow • In sub-reservoir 1, all three predictors underestimate well
encroachment are compared to the high and low permeability test permeability. However, more CBR KH data than
paths and bubble maps (amount of production). ESTIME and CLDT fall in between the half-decade lines.
• In sub-reservoir 2, the cross plot is inconclusive which
Integration of Core K and Well Test K Data estimator is better than the others.
Core permeability is preferred to well test permeability for its • CLDT K performs better than the ESTIME and CBR K
high vertical resolution. Well test permeability however is estimators in the sub-reservoir 3, as shown in Figure 5.
believed to present more representative reservoir permeability • CLDT K does better than the ESTIME and CBR K
due to the large rock volume investigated. Integrating core predictors in matching the test KH in sub-reservoir 4.
and well test permeability takes advantage of the strengths of • In sub-reservoir 5, CLDT and CBR K predictors perform
both. A script written using the internal programming better than ESTIME. However, it is not clear enough
language of IRAP-RMS++, scales core K to well test K. The which is better when they are compared with each other.
steps followed to achieve this are explained below (Figure 4). • In sub-reservoir 6, CBR K predictor is performing better
than CLDT and ESTIME K predictors. The number of
• Before loading into the geological model, well test KH data points however is sparse.
was divided by the reservoir layer thickness to determine
test K, which was distributed on foot-by-foot over the test
interval. Cumulative KH Plots
Well cumulative test KH is plotted versus estimator KH for
• Overlap between the well test intervals and core K data
the six main layers of the reservoir, Figure 6. Ideal is for
are recorded. Over any overlap, the script computes the
cumulative estimated KH curve to overlap with the cumulative
well test KH (Test K * H), and the core KH (C_Kc-ave*
well test KH curve. If not, the profiles should be similar.
H), and the ratio between them. By multiplying the core K
(C_Kc) by this ratio, well test scaled core permeability
• In sub-reservoirs 1 and 2 of the field, all three K
(C_Kc-c), is generated.
estimators are found to underestimate permeability
• Well test K remains the same in all wells except in the
confirming the conclusion of the KH cross plots
overlap interval where it is replaced by C_Kc-c.
technique. Figure 6 illustrates this for sub-reservoir 2.
CBR K in both sub-reservoirs is closer to well test K than
These new well test permeability profiles are blocked to the
ESTIME or CLDT K.
resolution of geological model fine layers. They are then
modeled by Kriging between wells, and upscaled by a flow • In sub-reservoir 3, CLDT and CBR estimators follow
based averaging technique8. closely the well test KH. CLDT KH is closer than CBR,
confirming the finding of the KH cross plot. ETIME
Which Permeability Estimator is More Applicable? cumulative KH falls below all three curves.
Despite the above validation tests, it was not clear which of • In sub-reservoir 4, the cumulative KH plot shows both
the new permeability estimates was the most appropriate for CLDT and CBR closely follows the cumulative well test
the reservoir simulation model. To explore the quality of these KH. CLDT estimator seems to perform better for the
permeability estimates, each was compared in turn against the latest wells, confirming the finding of the KH cross plot.
wells’ test KH (conditioned to core KH) as the reference. For • For sub-reservoirs 5 and 6, KH cross plots were not
reference the older ESTIME prediction was also included. The definitive. CLDT estimator is closest to the trend of well
analysis methodology involves four techniques: test cumulative KH than the other two predictors for sub-
reservoir 5. As for sub-reservoir 6, both the ESTIME and
• KH cross-plots CBR predictors follow the curve of well test K closely.
• Cumulative KH plots CBR, however, seems to be a little better than ESTIME.
• KH difference histograms
By using the KH cross plots and cumulative KH plots, it was
• Visualization of simple permeability maps.
possible to conclude that CBR K provides a better K estimator
in sub-reservoirs 1, 2 and 6, while CLDT K is better in sub-
reservoirs 3, 4 and 5.
++
Mark of Roxar
SPE 84599 5
realization, which did not involve any conditioning to well test Acknowledgements
K (just a combination of the permeability estimators). Model The authors would like to thank ADNOC and ADMA OPCO
performance is better with the second permeability realization senior management for their continuous support and for
for five out of the six sub-reservoirs of the field, Figure 10. permission to publish this work. Samir Azir, ADMA-OPCO,
The third permeability realization gave better performance for interpreted the geology and built the 3D geological model.
onlt one sub-reservoir.
References
Since none of the permeability realizations performed best in
all the field’s sub-reservoirs, we decided to apply the third K 1. Plomb, J.P.: “Umm Shaif Arab Petrophysical Study”,
realization in sub-reservoir 1, and the second K realization in ADNOC/Schlumberger Joint Internal Report, Oct. 1991.
all other sub-reservoirs. A combination of the second and the 2. Kamal, M. M.: “Use of Pressure Transients To Describe
third realization better describes the heterogeneity of the Reservoir Heterogeneity,” JPT (August 1979) 1060.
reservoir. 3. He, N., Oliver, D. S., and Reynolds, A. C.: “Conditioning
Stochastic Resevoir Models to Well-Test Data,” SPE
Reservoir Eval. & Eng. 3 (1), February 2000.
Summary and Conclusions 4. Landa, J. L., Horne, R. N., Kamal, M. M., and Jenkins, C.
D.: ”Reservoir Characterization Constrained to Well-Data:
• Permeability modeling of a large carbonate reservoir is A Field Example,” SPE Resevoir Eval. & Eng., Vol. 3, No.
improved when more than one estimation is available. 4, August 2000.
• Four different techniques were utilized to aid the selection 5. Azer, S., Al-Aruri, A., Peebles, R., Boyd, D., and Lyon, T.:
of the best permeability estimator for each sub reservoir. “Umm Shaif Field Arab C and D Integrated
Multidisciplinary Reservoir Characterization Study”, Ref.
• The simulation model first K realization is built by PDD/752, 3 Volumes, Internal, Report, Nov. 1999.
combining the K of the most appropriate estimators to the 6. Boyd, D.: “Trustworthy Core Data”, Internal Document,
different sub reservoirs. File Note Ref. No. 28/33330, Feb. 2002.
• Conditioning the model first K realization to the test KH 7. Pfeffer, D., Prezbindowski, D., Sherman, B. and Waite, M.:
derived two other K realizations. “Comparison of Deterministic and Stochastic Methods of
• The model K maps is either multiplied by the ratio at he Distributing Reservoir Properties in a Complex Carbonate
well level then mapped or by mapping the multipliers Reservoir: an Example from the Cretaceous Ratawi Oolite
in the Kuwait-Saudi Arabia Partitioned Neutral Zone”, Geo
then multiplying by the permeability predictors. 2000 abstracts, p. 162.
• The permeability arrays that are conditioned to the test K 8. Ghedan, S. G., Cunningham, T. W., Ehmaid, B. H, and
(second and third realizations) performed much better Azer, S. R., “Upscaling of Multimillion Cell Geological
than the first K array in matching the model production Model into a Practical Simulation Model of a Major
and pressure history. Carbonate Oil Reservoir in Offshore Abu Dhabi”, Paper
• By inspecting the quality of the simulation model history SPE 78578 presented at the 10th Abu Dhabi International
match, neither the second nor the third K realization did Petroleum Exhibition and Conference, Abu Dhabi, UAE,
well in all sub reservoirs. 13-16 October, 2002.
• A combination of the second and the third realization
better describes the heterogeneity of the reservoir to
produce a quality history matching of the simulation
model.
SPE 84599 7
Wackestone Facies
120 Packstone Facies
180
100
160
80 140
Frequency
Frequency
120
60
100
40 80
60
20
40
0 20
20
Fig. 2- Geologically driven cloud transforms duplicate mean, maximum, minimum and frequency of
permeability for each reservoir layer and facie.
8 SPE 84599
Processing Flow
Terminate Predictive Process for Layer When End of Data Column Reached
1.E+05
1.E+04
CLDT KH
1.E+05
1.E+03
1.E+02 1.E+04
ESTIME KH
1.E+05 1.E+03
1.E+04
1.E+02
1.E+02 1.E+03 1.E+04 1.E+05
CBR KH
Test KH
1.E+03
2.E+05
2.E+05
Cum Test KH
Cum CLDT KH
1.E+05
Cum ESTIME KH
Cum CBR KH
1.E+05
Cum. KH, md.ft
1.E+05
8.E+04
6.E+04
4.E+04
2.E+04
0.E+00
005
025
031
035
041
043
047
051
072
076
091
095
098
101
104
107
110
112
113
115
118
119
124
125
140
145
157
161
169
174
179
183
208
Well Number
8
CLDT KH Diff.
7 ESTIME KH Diff.
CBR KH Diff.
6
5
Frequency
0
-2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2
Log (Test KH) - Log (Est KH)
Fig. 8- Test, RDS, CLDT, and CBR Permeability Distribution in a Sample layer in Sub-Reservoir 4. CLDT K is the Closest to Test K.
70 70
CBR K Multipliers Frequency CLDT K Multipliers Frequency
60 60
50 50
Frequency
40
Frequency
40
30 30
20 20
10 10
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
CBR K Multiplie r CLDT K Multiplier
30
CLDT K Multiplier Frequency
Fig. 9- CBR K Multiplier Frequency of CBR K in Sub-
25 Reservoir 2, CLDT K in Sub-Reservoirs 3 and 4.
20
Frequency
15
10
0
0 2 4 6 8 10 12 14
CLDT K Multiplier
SPE 84599 11
4500
4400
4300
4200
Sector Pressure, Psia
4100
4000
3900
3800
3700
3600
3500
Jan-62 Jan-66 Jan-70 Jan-74 Jan-78 Jan-82 Jan-86 Jan-90 Jan-94 Jan-98 Jan-02
Time, Ye ars
Fig. 10- Observed BHCIPs Data of Certain Sector, Matched by the Model Average Wells WBP9
in that Sector. It Seems that the Model with Second K Realization is better matched than with
the third K Realization.