You are on page 1of 8

SPE 91820

Uncertainty Assessment Using Experimental Design: Minagish Oolite Reservoir


W.T. Peake, ChevronTexaco; M. Abadah, Kuwait Oil Co.; and L. Skander, ChevronTexaco

Copyright 2005, Society of Petroleum Engineers Inc.


This paper was prepared for presentation at the 2005 SPE Reservoir Simulation Symposium
held in Houston, Texas U.S.A., 31 January 2 February 2005.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.

Abstract
A peripheral waterflood is planned in the Minagish Oolite
Reservoir, Burgan Field. This is a large carbonate reservoir
which has been under limited primary production since 1965.
There is considerable uncertainty in the waterflood oil forecast
because of the minimal amount of data with which to
characterize the reservoir.
This study identifies key
subsurface uncertainties impacting waterflood performance
and quantifies uncertainty with P10/P50/P90 oil forecasts.
Experimental design (ED) techniques were used to
establish the minimum number of reservoir simulation runs
needed to quantify uncertainty. Both traditional and ED
techniques were used to preserve history matches for all
simulations.
Analysis of variance and multiple linear
regressions were used to identify the most significant
uncertainties and to create a proxy for the simulator. The
proxy was used in Monte Carlo simulation to develop
P10/P50/P90 oil forecasts.
Uncertainty analyses have not often been conducted on
producing reservoirs.
In this report a new technique
combining ED with history matching has been successfully
demonstrated.
The technique can be applied to other
producing reservoirs. Despite the need to restrict simulator
inputs for history matching, it was shown that considerable
uncertainty exists in the waterflood oil forecast. The results
are valuable for: a) identifying actions to reduce uncertainty
and b) input to additional development evaluations.

A key project uncertainty is the production forecast. Like


many oil field projects today, the production forecast is
usually the outcome of reservoir simulation. In the past, if
uncertainty was assessed, it was the result of personal
judgment and/or a few simulation runs. The simulations were
conducted by varying one factor (such as such as porosity or
permeability) at a time. The effects of varying each factor are
shown in a tornado chart. A better means of assessing
uncertainty is through the use of experimental design (ED).
ED is more efficient, can assess interactions and allows effects
to be estimated at several levels1.
A variety of experimental designs/methodologies have
been proposed and tested for oil and gas applications2,3. The
appropriate design for a specific application will be a function
of: number of uncertainty factors, expert knowledge, and
personal choice. In this study we used a methodology similar
to that described by Friedmann, et.al.4 and Kabir, et.al.5 Their
methodology was modified to include history matching. A
flow chart is shown in Fig. 1. More recent methodologies can
automate this process, although requiring many simulations6.
Define Objective
Build History Matched Simulation Model
Identify Uncertainty Factors and Ranges
Evaluate History Matches
Perform Experimental Design Simulations
Test for History Matches and Curvature
Generate Results by Statistical Analysis

Fig. 1 Methodology Flow Chart

Introduction
Quantification of uncertainty in reservoir development
projects has been a steadily growing industry-wide practice
since the early 1990s. By quantifying uncertainty in the early
phases of a project, it becomes possible to clearly assess risk
and plan for a range of possible project outcomes. Together,
these two abilities should yield better business decisions and
higher project values.

Background
The Minagish Oolite Reservoir, Burgan Field, Kuwait has
been on primary production since 1965. Cumulative oil
production is 8% OOIP and water production has been
minimal. Reservoir pressure has been slowly declining and is
now approaching the bubble point. The next planned stage of
development is waterflooding. Excellent response to water

www.petroman.ir

SPE 91820

injection has been observed in an analogous reservoir in a


nearby field7.

a similar nearby reservoir, and expert field knowledge. The


list was made intentionally extensive to minimize personal
bias.
Table 1. Original Uncertainty Factors and Ranges
Factor

Low

Middle

High

Top Structure
Depth

60 ft. lower on flanks

No change

60 ft. higher on
flanks

Mean Porosity

x 0.87

x 1.0

x 1.13

Mean Horizontal
Permeability

x 0.34

x 1.0

x 1.66

OOWC

6 ft. deeper

6828 ft. tvdss

6 ft. shallower

Mean Irreducible
Water Saturation

x 0.38

x 1.0

x 1.62

1 High Perm Zones

4 High Perm
Zones

3 Low Perm Zones

7 Low Perm Zones

6 High Perm
Zones
11 Low Perm
Zones

x 0.333

x 1.0

Geologic Contrast
Vertical
Permeability
Sealing Faults

Tar Zone Thickness

0 ft.

34 ft.

108 ft.

Rock
Compressibility

5 x 10

-6

1/psi

Pb = 2571 psia

Uncertainty Factors
At the onset of the study we identified all major subsurface
uncertainties which are labeled factors and listed in Table 1.
The list was developed from: experience gained during
simulation model building and history matching, analogy with

-6

8 x 10

1/psi

Bob = 1.362
rb/STB
Rsb = 724
SCF/STB

Bob = 1.412
rb/STB
Rsb = 766
SCF/STB

Viscosity = 1.31 cp

Viscosity = 1.12 cp

Viscosity = 0.93 cp

Sorw = 0.2

Sorw = 0.28

Sorw = 0.36

krw@sorw = 0.2

krw@sorw = 0.4

krw@sorw = 0.6

Rsb = 683 SCF/STB

Water Relative
Permeability

1/psi

Pb = 2794 psia

Bob = 1.312 rb/STB

Fig. 2 Peripheral Water Injection Well Locations

-6

6.5 x 10

Pb = 2683 psia

Oil Properties

A peripheral waterflood with infill drilling has been


tentatively selected as the preferred development plan shown
in Fig. 2. The number of uncertainties remains large,
however, given no field evidence of what water injection
response will be. In addition to the uncontrollable subsurface
uncertainties, there exist additional uncertainties.
They
include production and injection well management and
facilities limitations.
The objective of this study is to quantify uncertainty of the
oil production forecast. The uncertainty assessment will be
conducted in two phases; the 1st phase to identify key
subsurface uncertainties while holding the development plan
fixed. A 2nd phase (not described here) will address
operational uncertainties and optimization for wells and
facilities.
The basis for production forecasts is a dynamic reservoir
model built in 2001. This reservoir model was history
matched by forcing oil to produce at historic rates and
adjusting model parameters to match reservoir pressure and
water cut. Within the accuracy of pressure data, all wells
exhibited a similar pressure decline. Most wells produced
little or no water. Model adjustments were: increased
horizontal permeability, addition of a few low and high
permeability layers, decreased rock compressibility and
reduced aquifer size. Wells were also calibrated using flow
tables. A very good match was obtained with minimal
changes to the original model inputs.

x 2.0

There are 12 factors.


These factors include: rock
properties, fluid properties, rock-fluid interaction, and initial
fluid contact. As seen in Table 1, oil properties consist of
several fluid properties that are expected to be highly
correlated. By combining these fluid properties into one
factor we insure a consistent set of input data. During
experimental design simulations and analysis it is assumed
that all factors are uncorrelated. This is a reasonable first
assumption for the factors considered, which later will be
removed during Monte Carlo simulation.
Due to the use of a static model with geostatistical derived
properties, there is the possibility of including geostatistical
parameters as factors. We assumed that resultant uncertainty
in oil production created by geostatistical parameters (e.g.
seed and variogram) is much smaller than that created by the
uncertainty of the mean. As a result, only a single static
model realization was utilized. Uncertainty due to porosity,
horizontal permeability and irreducible water saturation was
captured through uncertainty of their mean values.
Three levels for each factor were developed (low, middle,
high). The middle values are equal to factor values in the
original history matched model. Low and high values were
obtained by the following, in order of priority: 1. statistical
analysis of data, 2. analogy, and 3. personal judgment.
Top Structure Depth Seismic mapping and a valid time
depth relationship at the crest of the reservoir reduces the
overall (average) structure depth uncertainty between the 8
crestal wells. But this is not valid on the flanks of the
reservoir, where well data is sparse.
The first flank well was recently drilled and encountered
the reservoir top approximately 60 ft. lower than expected. In

www.petroman.ir

SPE 91820

the absence of other data, we assumed +/- 60 ft. as the range of


uncertainty on the flanks. The low value for structure depth
was obtained by adding 60 ft. to all mapped depths on the
flanks. The high value was obtained by subtracting 60 ft. on
the flanks.
Mean Porosity The most significant components of
uncertainty in mean porosity were assumed to be: (1)
measurement error and (2) non-representative sampling8.
Non-representative sampling is the difference between sample
and reservoir statistics.
Reservoir model porosities were derived from log
porosities calibrated to core measurements.
Systematic
measurement error, including core measurement error and log
to core calibration error was assumed to be +/- 0.015.
Reviews of all possible sources of measurement error can be
found in the literature9.
Uncertainty resulting from non-representative sampling
was obtained by considering an areal distribution of porosities.
The porosities considered are vertical averages over reservoir
thickness. The vertical mean (average) was obtained from
well logs. The variance of the mean of mean porosities from
well logs provides our estimate of uncertainty due to
sampling. If: (1) wells are uniformly distributed and (2)
internal reservoir stratigraphic layering is uniform, then the
result should not be biased. Both conditions are met in the
Minagish Oolite Reservoir.
Mean porosities were calculated in each of 7 wells. The
mean (of means) was 0.24 with standard deviation of 0.02.
The Central Limit Theorem states that the distribution of
sample means is approximately normal, regardless of the
distribution of the parent population. This justifies an
assumed normal distribution for mean porosity.
Both components of uncertainty are independent. As a
result, variances are additive and the combined standard
deviation of 0.025 was obtained using a root-mean-square
calculation. It is seen that sampling error dominates. Mean
porosities at 10% and 90% probability were calculated and
converted into multipliers shown in Table 1. This uncertainty
estimate was only applied to the oil column and did not impact
the aquifer pore volume.
Mean Horizontal Permeability Uncertainty in mean
horizontal permeability was calculated in a similar manner as
porosity. Core Labs determined that the reproducibility of
core air permeabilities to be +/- 10% for permeabilities
between 500 and 5000 md. Using this value, the measurement
uncertainty is +/- 61 md.
Lacking a permeability well log, we substituted core
measurements from 5 wells for an estimate of sampling
uncertainty. The sample mean and standard deviation of the
mean horizontal permeability was 609 and 307 md,
respectively. The combined standard deviation is 313 md.
Sampling error is seen to dominate. Mean porosities at 10%
and 90% probability were calculated and converted into
multipliers shown in Table 1.
Original Oil-Water Contact Three credible original OWC
depths were obtained from wells drilled early in the life of the
reservoir. The range of these measurements is 8 ft., however,
a water-up-to depth from a well drilled at the same time
indicates the possibility of a slightly larger range. The mean

value from measurements is 6828 TVD ftss. Low and high


values were selected to be +/- 6 ft. from the mean.
Mean Irreducible Water Saturation Uncertainty in the
mean irreducible water saturation was calculated in a similar
manner as porosity.
A +/- 0.05 measurement error was
assumed. From well logs, the sample mean and standard
deviation is 0.108 and 0.014, respectively. The combined
standard deviation is 0.052. Measurement error is seen to
dominate. Mean irreducible water saturation at 10% and 90%
probability were calculated and converted into multipliers
shown in Table 1.
Geologic Contrast With few modern logs and vintage
1970s logs and core data, it has not been possible to construct
a facies based static model. Thus, the presence and lateral
extent of very high and very low permeability layers, that may
have a profound effect on fluid movement, can not be obtained
from existing data. The best estimate of lateral extent and
location of the high/low permeability streaks was derived from
an analogous reservoir, where both types of permeability
layers were observed.
Using the analogy as a guide, it was assumed a most likely
case (geologic contrast middle) as having a similar number
of layers, at the same stratigraphic depths and with the same
approximate areas. These layers were assigned 0.01 and 5000
md permeabilities. These layers and their permeabilities
provide internal reservoir geologic contrast that would
otherwise not exist in the static model derived strictly from
well logs and geostatistical modeling of permeability. A range
of values for geologic contrast was obtained by adding and
removing permeability layers, as shown in Table 1.
Vertical Permeability Static model vertical permeability
was calculated using a vertical/horizontal permeability ratio of
0.6, an average value observed in cores. A majority of core
vertical/horizontal permeability ratios lie within the range of
1.2 and 0.2. This range was converted to multipliers shown in
Table 1. It is noted that vertical/horizontal permeability ratios
were input to the static model, however the multipliers are
applied post scale-up. This apparent inconsistency is minimal
however, as post scale-up model layers are 1-2 ft. thick.
Sealing Faults There are 5 mapped faults in this
reservoir, all of which are nearly vertical. None juxtapose pay
and non-pay.
Shut-in bottom hole pressure (SIBHP)
measurements from all wells follow the same general trend as
shown in Fig. 3. However, the accuracy of SIBHP data is
such that it is not possible to completely rule out reservoir
compartmentalization due to sealing faults. The most likely
scenario is that, based on several pressure buildup analyses
and the regional stress field, 3 faults are sealing. The range of
possibilities is that 0 to 5 faults are sealing.
Tar Zone Thickness Core descriptions from the 1970s
include observations of heavy oil near the oil-water contact.
To date there has been no investigation of the nature and
extent of this observation. However, the properties and
location of an immobile tar zone have been extensively
studied in the analogous reservoir10. There the tar zone is
located directly above the oil-water contact and has a
maximum thickness of 108 ft. It is located only where
permeability is greater than 100 md. By analogy, we assumed
a range of 0 108 ft. thickness with a most likely value of 54
ft. It was also assumed that the tar zone exists only where

www.petroman.ir

SPE 91820

Evaluate History Matches


It was anticipated that some factor values (low/high) would
cause a mismatch during experimental design simulations,
regardless of the values of other factors. It is industry practice
that a valid simulation model is obtained only through
reasonableness of input data and history matching. It is our
premise that ED results from simulation models that do not
match history are highly questionable. Therefore, the
methodology that follows is designed to preserve history
matches for all simulation models.
Two different approaches were used in this study to insure
that all simulation models were adequate history matches.
The approaches are:
1. Traditional (vary one-factor-at-a-time)

2. Experimental Design
Both approaches rely on simulation results to identify
factors that have the largest impact on the history match.
Once these factors were identified, their range was modified to
achieve better history matches for the next set of simulations.
The traditional approach was used initially because it was
fast. Partial history matches (15 out of 38 years) without
forecasts were run in 2 hours. In hindsight, we could have
used ED with the same partial history matches to obtain
similar results (described later). From the traditional approach
a tornado chart was used to identify the factors listed in Table
2 that caused the greatest mismatches. Fortunately only water
cut match was problematic; all simulations produced
reasonable pressure matches. Adjusted factor values, which
were carried forward to the first ED set of simulations, are
listed in Table 2.
Table 2. Pre-ED History Match Adjustments
Factor

Low

High

Mean Horizontal
Permeability

x 0.67

3 High Perm Zones

Geologic Contrast
4 Low Perm Zones

12 Low Perm Zones

Tar Zone Thickness

20 ft.

54 ft.

Oil Properties

Viscosity = 1.215 cp

Water Relative
Permeability

Sorw = 0.32

krw@sorw = 0.5

First Experimental Design Simulations


A Plackett-Burman 2 level design was chosen for its
usefulness in screening a relatively large number of factors
with a minimum number of simulations. Its disadvantages are
the inability to thoroughly investigate interactions and higher
order effects. With 12 factors the design has 16 runs. A 17th
center point run was added to evaluate curvature (caused by
nonlinear and interaction effects).
All simulations began in 1965, the year production began.
The history match concluded near the end of 2003. As shown
in Fig. 3 a reasonable pressure match was obtained for all
runs. The center point run (containing only middle values)
lies in the middle of all the results. As shown in Fig. 4 the
water cut match was excellent for the center point run and
many other runs. However, two runs had unacceptably high
water cuts (pink and grey curves).
3300

X Field Data
Pressure (psia)

permeability exceeds 100 md. The permeability criterion left


gaps which permit communication between the oil column and
the aquifer, although at a reduced amount than without the tar
zone.
Rock Compressibility As a result of history matching,
reservoir model rock compressibility is near the lower limit of
core measurements. Unlike the other factors, we recognized
that obtaining a range for rock compressibility from core
measurements would lead to values far outside the range
needed for a pressure match. Therefore a smaller range (5 - 8
x 10-6 1/psi) was assumed that would likely permit an
acceptable match.
Uncertainty in rock compressibility was also used as a
surrogate for uncertainty in aquifer size. The history matched
reservoir model has an aquifer to oil column pore volume ratio
of 23. The effect of aquifer size was indistinguishable from
that of rock compressibility.
Oil Properties There are 4 credible oil PVT analyses
conducted on bottom hole fluid samples obtained early in the
life of the reservoir. After review of the PVT and sampling
depth data, it was decided that although an oil property
gradient with depth might exist, it could not be determined
from this data set. As a result, the model contains a single set
of black oil fluid properties. The mean, P10 and P90 values
of key oil properties at bubble point pressure and reservoir
temperature were obtained using t-statistics to account for the
small number of data points. These key oil properties are:
bubble point pressure, gas-oil ratio, oil formation volume
factor and oil viscosity. Their values are shown in Table 1.
Measurement error was assumed small in comparison to the
variance in PVT measurements. Published correlations were
used to calculate consistent sets of low, middle, high oil
properties for all pressures.
Water Relative Permeability There exists no relative
permeability measurements for this reservoir. However,
relative permeability measurements were made in the
analogous reservoir. From this data, a set of most likely
relative permeabilities were derived and input to the reservoir
model as a power-law function. History matching was used to
determine the middle values: a 0.28 residual oil saturation and
a 0.4 water relative permeability. Based on experience, water
relative permeability endpoints were selected that could span
the range of very favorable to very unfavorable (to oil
recovery) as shown in Table 1. These values might also be
considered as spanning the range of possible wettabilities.

3100

2900

Center Point Run


2700
1965

www.petroman.ir

1975

1985

1995

Fig. 3 First ED Pressure Match

2005

SPE 91820

50
Water Cut (%)

X Field Data
40
30
Center Point Run
20
10
0
1965

1975

1985

1995

2005

Fig. 4 First ED Water Cut Match

The history matches were further evaluated to identify


those factors with the greatest impact on water production.
Analysis of variance (ANOVA) and calculation of main
effects were performed on the cumulative water production at
the end of the history match period. Those factors impacting
cumulative water production, in order of significance, are:
water relative permeability, mean porosity, OOWC, vertical
permeability, geologic contrast, mean horizontal permeability,
and rock compressibility. This is shown in a normalized
Pareto chart of main effects in Fig. 5. As a result of this
evaluation, those factor ranges were adjusted to improve the
history match during the next (second) ED design runs.
Relperm
Porosity
OOWC
Vperm
Geolcon
Hperm
Comp
Oil Prop
Curvature
Str top
Swirr
Fault Seal
Tar Zone

injection rates are controlled by voidage to maintain a constant


reservoir pressure above bubble point. Wells produce against
a constant bottom hole pressure and plug backs occur at 95%
water cut, starting with the lowest completion. The only
facility limitation is a maximum field oil rate.
A Pareto chart of main effects for cumulative oil recovery
at 2040 is shown in Fig. 6. In this chart the absolute value of
effects are ranked and shown against a 95% significance limit.
Significance is obtained from an F-statistic. From this result it
is seen that at least 4 factors are significant, from porosity
through horizontal permeability. The least significant factors
are: structure top, fault seal, curvature, and geologic contrast.
This observation generally holds for other years, although the
ranking may differ slightly and the position of the 95%
significance limit varies.
Porosity
Tar Zone
Relperm
Hperm
Oil Prop
OOWC
Swir
Comp
Vperm
Geolcon
Curvature
Fault Seal
Str top

Fig. 6 Pareto Chart of First ED Cumulative Oil at 2040

Fig. 5 Pareto Chart of First ED Cumulative Water at 2004

Despite the 2 water cut mismatches; it was decided to


proceed ahead with predictions. It was considered that these
predictions would provide a first-order estimate of the relative
importance of each factor. Cumulative oil recovery and
recovery efficiency (%OOIP) were selected as response
variables. Once the prediction runs were completed, the
factors were prioritized through ANOVA and calculation of
main effects.
From 2003 to 2007 two wells remain on primary
production. The water flood prediction begins in 2007 when
new injectors and producers are drilled, completing a ring of
peripheral injectors and crestal producers on 250 acre spacing.
Injectors are completed above and below the OOWC and

Similar analysis was conducted for recovery factor at


2040. This analysis showed that structure top ranks high in
significance and porosity, tar zone and OOWC rank much
lower.
This first set of ED forecasts, however imperfect the
history match, yielded valuable information. It was observed
that fault seal and geologic contrast were consistently the least
significant factors for both cumulative oil and recovery factor.
We also determined that due to the lack of significant
curvature, both factor interactions and quadratic effects are
unimportant.
Second Experimental Design Simulations
Iteration is a normal component of ED methodology. A
second set of ED simulations were conducted to improve the
water cut matches. Factor values for the second ED
simulations are shown in Table 3. Adjustment was made to
those values having the highest impact on water production, as
previously identified. It is not known a priori, whether these
adjustments will be sufficient to provide matches for all ED
simulations. The middle water relative permeability was also
adjusted to remain centered between low and high values.

www.petroman.ir

SPE 91820

Table 3. Second ED Uncertainty Factors and Ranges


Factor

Low

Middle

High

Top Structure
Depth

60 ft. lower on
flanks

as is

60 ft. higher on
flanks

Mean Porosity

x 0.94

x 1.0

x 1.13

Mean Horizontal
Permeability

x 0.8

x 1.0

x 1.66

OOWC

6 ft. deeper

6828 ft. tvdss

3 ft. shallower

Mean Irreducible
Water Saturation

x 0.382

x 1.0

x 1.618

Vertical
Permeability

x 0.333

x 1.0

x 1.4

Tar Zone Thickness

20 ft.

34 ft.

54 ft.

Rock
Compressibility

5 x 10-6 1/psi

6.5 x 10-6 1/psi

7.25 x 10-6 1/psi

Pb = 2571 psia

Pb = 2683 psia

Pb = 2794 psia

Bob = 1.312
rb/STB
Rsb = 683
SCF/STB

Bob = 1.362
rb/STB
Rsb = 724
SCF/STB
Oil Viscosity = 1.12
cp

Bob = 1.412
rb/STB
Rsb = 766
SCF/STB
Oil Viscosity =
0.93 cp

Oil Properties

Viscosity = 1.22 cp
Water Relative
Permeability

Sorw = 0.2

Sorw = 0.25

Sorw = 0.3

krw@sorw = 0.2

krw@sorw = 0.3

krw@sorw = 0.4

From the first ED simulations it was seen that two factors


can be neglected, reducing the number of factors from 12 to
10. This had the desirable consequence of reducing the
number of required simulation runs. As before, we used a
Plackett-Burman ED for the second set of simulations but now
the number of required runs is 12. Adding a center point run
gives a total of 13 runs.
Acceptable history matches were obtained for all runs as
shown in Fig. 7. The pressure match was qualitatively the
same as shown in Fig. 3.

Water Cut (%)

50
40

Cumulative oil recovery versus time is shown in Fig. 8. It


is observed that despite of the reduction of factor ranges due to
history matching there remains a considerable uncertainty in
oil recovery.
ANOVA and effects for cumulative oil is shown in Fig. 9
with a significance limit of 80%. Comparing Figs. 3 & 6, it is
observed that the exact position of factors has changed;
however, the relative importance of each factor remains
unchanged. As before, interactions and higher order terms are
not important.
Tar Zone
Relperm
Porosity
Hperm
Swir
Vperm
OOWC
Oil Prop
Comp
Str top
Curvature

Fig. 9 Pareto Chart of Second ED Cumulative Oil at 2040

Generate Simulator Proxies


Without the need to include interactions and higher order
terms, we can continue using 2-level ED simulation results for
building multiple linear regression models. These models will
become proxies for the reservoir simulator in Monte Carlo
simulations.
Table 4 lists factor rankings for cumulative oil at selected
years. Factor rankings were obtained from Pareto charts like
Fig. 9. Only the top 6 factors (shaded in green) were used as
independent variables while performing regressions.
Table 4. Cumulative Oil Factor Rankings by Year

X Field Data

30

Center Point Run

20
10
0
1965

1975

1985

1995

2005

Cumulative Oil

Fig. 7 Second ED Water Cut Match

Center Point Run


0
2005

2010

2015

2020

2025

2030

Fig. 8 Second ED Cumulative Oil Recovery

2035

2040

2010
2

2015

2020

2025

2030

2035

2040

tarzone

relperm

porosity

hperm

swirr

vperm

oowc

11

11

oilprop

comp

11

11

11

top

10

11

10

10

curvature

10

10

10

10

11

Multiple linear regressions were calculated for cumulative


oil production every 5 years from 2010 to 2040. Coded values
for the independent variables were used because certain
variables are qualitative functions. An example result is given
below for cumulative oil at 2040. The variable subscripts
correspond to the rank given in Table 4.
Y = 561 -43.2 X1 41.7 X2 + 39.2 X3+ 23.4 X4 -21.0 X5 +11.6 X6

www.petroman.ir

SPE 91820

For all years the coefficient of determination is between 0.92


and 0.98, indicating an excellent fit to simulator results.
A further demonstration of the ability of proxies to serve
as a substitute for the simulator is given by calculation of
residuals. Residuals for cumulative oil at 2040 are shown in
Fig. 10 for 13 ED runs and 5 check runs (#14 #18).
Residuals for other years are similar.

Residual (%)

3
1
-1

9 10 11 12 13 14 15 16 17 18

cumulative probabilities were assigned at -1 and +1,


respectively.
As described previously, mean porosity, mean horizontal
permeability, and mean irreducible water saturation are
characterized by a normal pdf. The assigned pdfs have 90%
and 10% cumulative probabilities at -1 and +1, respectively.
From fundamental principles, we expect: (1) a negative
correlation between porosity and irreducible water saturation
and (2) a positive correlation between porosity and horizontal
permeability. A negative 0.50 correlation between porosity
and irreducible water saturation was obtained from analysis of
well log data. A 0.34 correlation between porosity and
horizontal permeability was obtained from analysis of core
plug data. Lacking log or core data for both horizontal
permeability and water saturation, we approximated the
correlation as a negative 0.2.

Fig. 10 Residuals for Cumulative Oil at 2040

The five check runs were used to provide further validation


of the regression models. Factor values in the check runs were
2-level and chosen at random. The residuals from the check
runs are similar to that obtained from the ED runs providing
confidence in the results.
The above calculations were repeated for oil recovery
factor at 2040. The formula for recovery factor is:
Y = 33.9 2.62 X1 1.92 X2 + 1.29 X3+ 1.12 X4 + 0.91 X5 0.74 X6

where X1 is water relative permeability; X2 is structure top;


X3 is oil properties; X4 is horizontal permeability; X5
irreducible water saturation; and X6 is tar zone. The
coefficient of determination is 0.92 and the maximum residual
is 5.7%.
Generate Probablility Distributions
Monte Carlo Simulation (MCS) was used to obtain probability
distributions for both response variables11.
Input data
requirements for MCS include the simulator proxies,
probability density functions (pdf) for each independent
variable and correlation coefficients where applicable.
Probability density functions and correlations can be obtained
from fundamental principles, expert opinion, and historical
data12. Following the convention used to develop the proxies,
each pdf is in coded variables.
The factors which have discrete values (tar zone, oil
properties, top structure, and water relative permeability) were
assigned pdfs with probabilities corresponding to each value.
Values -1, 0, and 1 were given 20%, 60% and 20%
probabilities, respectively.
Rock compressibility was assigned a uniform pdf between
-1.25 and +1.25. As a result, 90% and 10% cumulative
probabilities (equal to or less than) are at -1 and +1,
respectively.
Vertical permeability was assigned a triangular
distribution, with zero as the most likely value. 90% and 10%

P10
P50
P90

0
2005 2010 2015 2020 2025 2030 2035 2040
Fig. 11 Cumulative Oil Uncertainty

MCS was conducted with 25,000 trials and the results are
shown in Fig. 11 & 12. In Fig. 11 cumulative oil production is
shown as three curves depicting 90%, 50% and 10%
probabilities (equal to or less than). Values were calculated
every 5 years, with graphical interpolation in between. In
comparison to Fig. 8, it is seen that the P90 and P10 curves
fall within the range of results obtained from the ED
simulations. This result is expected, as ED simulations
contain only low and high values and should include more
extreme outcomes.
The cumulative probability for recovery factor at year
2040 is shown in Fig. 12. 90% and 10% (equal to or less
than) recovery factors are 30.3 and 37.4 % respectively.
Cumulative Probability
(%)

-5

Cumulative Oil

-3

100
75
50
25
0
25

30

35

40

45

Recovery Factor (%OOIP)

Fig. 12 Uncertainty in Recovery Factor at 2040

Conclusions
A valid oil production uncertainty assessment has been
obtained for a potential waterflood in the Minagish Oolite
Reservoir. Validity was established by history matching and

www.petroman.ir

SPE 91820

statistical analysis. A methodology combining: reservoir


simulation, Experimental Design, ANOVA, multiple linear
regression analysis and Monte Carlo simulation was used.
The methodology was a modification to that previously
described in the literature. The modification included steps to
evaluate and preserve history matching.
Despite constraints imposed by the history match,
significant uncertainty was calculated. The major subsurface
factors creating uncertainty in cumulative oil production were:
tar zone thickness, water-oil relative permeability, mean
porosity and mean horizontal permeability. The major
subsurface factors creating uncertainty in recovery factor
were: water-oil relative permeability, top structure depth, oil
properties, and mean horizontal permeability.
Both
cumulative oil and recovery factor were linear functions of the
uncertainty factors. Cumulative oil production forecasts
developed at 10%, 50% and 90% probabilities will be useful
for: (a) an estimate of reservoir performance uncertainty and
(b) input to obtaining the optimal development plan.

10. Al-Ajmi, et.al.: The Minagish Field Tar Mat, Kuwait: Its
Formation, Distribution and Impact on Water Flood, GeoArabia,
Vol. 6, No. 1, 2001, pp. 7-24.
11. Murtha, J.A.: Monte Carlo Simulation: Its Status and Future, J.
Petroleum technology, April 1997, pp.361-373.
12. Murtha, J.A.: Incorporating Historical Data in Monte Carlo
Simulation, paper SPE 26245 presented at the 1993 SPE
Petroleum Computer Conference, New Orleans, Louisiana, July
11-14.

Acknowledgment
We are grateful to the management of the Kuwait Oil
Company and to the Energy Ministry of the State of Kuwait
for permission to publish this work. We thank Adwait
Chawathe of ChevronTexaco for his advice and
encouragement on ED.
References
1. Montgomery, D.: Design and Analysis of Experiments, John
Wiley and Sons, New York City, 2001.
2. Peng, C. and Gupta, R.: Experimental Design and Analysis
Methods in Multiple Deterministic Modeling for Quantifying
Hydrocarbon In-Place Probability Distribution Curve, paper
SPE 87002 presented at the 2004 SPE Asia Pacific Conference
on Integrated Modeling for Asset Management, Kuala Lumpur,
Malaysia, March 29-30.
3. White, C. and Royer, S.: Experimental Design as a Framework
for Reservoir Studies, paper SPE 79676 presented at the 2003
SPE Reservoir Simulation Symposium, Houston, Texas,
February 3-5.
4. Friedmann, R., Chawathe, A., and Larue, D.: Assessing
Uncertainty in Channelized Reservoirs Using Experimental
Designs, SPE Reservoir Evaluation & Engineering, August
2003, pp. 264-274.
5. Kabir, C. et.al.: Developing New Fields Using Probabilistic
Reservoir Forecasting, paper SPE 77564, presented at the 2002
SPE Annual Technical Conference and Exhibition, San Antonio,
Texas, September 29 October 2.
6. Castellini, A. et.al.: Practical Methods for Uncertainty
Assessment of Flow Predictions for Reservoirs with Significant
History A Field Case Study, paper A-33, presented at the 9th
European Conference on the Mathematics of Oil Recovery,
Cannes, France, August 10 September 2, 2004.
7. Matar, S. et.al.: A Full Field Simulation Model of a Giant
Carbonate Reservoir in Kuwait, paper SPE 81498 presented at
the 2003 SPE 13th Middle East Oil Show & Conference, Bahrain,
April 5-8.
8. Bu, T. and E. Damsleth, Errors and Uncertainties in Reservoir
Performance Predictions, SPE Formation Evaluation,
September 1996, pp. 194-200.
9. American Petroleum Institute: Recommended Practices for Core
Analysis, Recommended Practise No. 40, February 1998.

www.petroman.ir