Professional Documents
Culture Documents
Williams
Williams
Williams
by
John Williams
2004
An Analysis of Monte Carlo Simulation as an Estimator of Original Oil
by
Report
Presented to the Faculty of the Graduate School of
in Partial Fulfillment
of the Requirements
Approved by
Supervising Committee:
Dedication
To my wife Shera, my parents Harry and Sarah Williams, and my children Justin and
Jenna.
Abstract
The Monte Carlo Method has been increasingly used in the petroleum industry as
range of values for hydrocarbon volumes originally in place. The distributions of the
input variables input into the Monte Carlo simulator are estimated by obtaining a
sufficient number of measurements of reservoir and fluid properties. This report will
analyze several depleted or very mature fields for which the ultimate hydrocarbon
recovery is known. For each of these fields, distributions of porosity, water saturation,
reservoir thickness, and reservoir size are obtained from cores, openhole logs, seismic,
and other data. These distributions are input into the Crystal Ball computer program to
obtain the cumulative distribution function (CDF) and probability density function (PDF)
of oil or gas volume originally in place for each of the fields. By comparing the actual
predicted by the Monte Carlo calculations, the accuracy of the Monte Carlo method will
be analyzed.
v
Table of Contents
DEDICATION....................................................................................................... iv
ABSTRACT.............................................................................................................v
TABLE OF CONTENTS....................................................................................... vi
vi
Chapter 4: Results and Discussion........................................................................44
References..............................................................................................................60
Vita ......................................................................................................................62
vii
List of Tables
Table 4-1: Summary OGIP Statistics for Case 1 Considering Correlation .......46
Table 4-2: Case 1 Values of F(x) for OGIP Considering Correlation ...............46
Table 4-4: Case 2 Values of F(x) for OOIP Considering Correlation ...............49
Table 4-5: OOIP Summary Statistics for Case 3 Considering Correlation .......51
Table 4-6: Case 3 Values of F(x) for OOIP Considering Correlation ...............51
Table 4-10: Case 5 F(x) Values for OGIP Considering Correlation ...................56
viii
List of Figures
Figure 3-15: Case 2 Probability Transform Semi-Log Plot for Water Saturation28
ix
Figure 3-18 Case 3 CDF for Porosity ...................................................................31
Figure 3-24: Case 3 Probability Transform Semi-Log Plot for Water Saturation34
x
Figure 4-3: Effect of Correlation on OGIP Calculation for Case 1 ......................47
xi
Chapter 1: Introduction
often massive structures; and usually reliable data can be obtained within only a small
understanding of any reservoir. When millions of dollars are at stake, everyone involved
strives to understand as much as they possibly can about the inherent risks in pursuing
any investment or development plan. An accurate and usable method is needed for
quantifying the uncertainty of the original volumes of oil or gas in place in immature
reservoirs. Monte Carlo simulation has been used as a tool to provide a method of
is a single answer. For example, a core analysis in a reservoir may yield values of
porosity ranging from 8% to 35% with an average value of 20%. In the deterministic
scenario approach is one in which generally three separate deterministic scenarios are
considered, one to represent the extremes and a third to represent the “most likely.” In the
porosity example above, the three scenarios might be porosity equals 8, 20, and 35 %.
However, it would be valuable to know more about the porosity values, such as how
many measurements show the porosity to be 8%, 35%, or some value in between. This is
distribution of the data. The distribution which shows the frequency of occurrence of a
1
range of values can be used to determine the likelihood of any subsequent measurement
made to be within that range, and provides a description of the data of the entire
population. This is much more descriptive than assuming a uniform average value as in
of oil and gas in place. This study uses the software package Crystal Ball to perform the
2
Chapter 2: Literature Review
where X is a continuous random variable and F(x) is the probability that a given value of
continuous random variable between 0 and 1, then F(1) = P(X ≤1) = 1. All CDF’s are
monotonically increasing and continuous from the right. The descriptive monotonically
increasing means that as x increases, F(X) must not decrease. Continuous from the right
means that small changes in x result in small changes in F(X). Additionally, for all
CDF’s, as x goes to infinity, F(X) approaches the value of 1 and as x goes to negative
infinity, F(X) approaches the value of zero. Figure 2-1 shows a sample CDF.
,
1.000
.750
.500
.250
.000
xÆ
3
When reservoir parameters are analyzed as random variables and distributions of
their values are created, it is possible to make statistical statements about the variables.
For example, it might be said of a given reservoir that 90% of the porosity values
measured will fall below 30%. This comes directly from the CDF. In this case, at the
point where F(x) = 0.9, the porosity equals 30%. Similar readings of the CDF can be
made and the corresponding values of porosity can be determined to have a certain
probability. In the example above, a porosity of 30% is considered the P(90) value. It is
common to note the P(10), P(50), and P(90) values for distributions since they give a
good representation of the entire distribution with just three numbers. By definition,
The Probability Density Function is the derivative of the CDF. The PDF displays
the same information as the CDF, but in a different format. A familiar types of PDF is the
histogram. PDFs relate the probability of occurrence of the variable f(x) for a small range
of x. The variable x should not be thought of as a fixed number, but as a small range of
values because the probability f(x) that x is any one exact number is zero. For all PDFs,
f(x) ≥ 0 for all x. Additionally, the sum of the values of f(x) over the infinite range, also
the area under the PDF curve, must be equal to one. An example of a PDF is shown in
Figure 2-2.
4
xÆ
Figure 2-2: Example of PDF.
There are many different shapes that the CDF and PDF can take depending on the
data they are describing. Mathematicians have developed models of the CDF and PDF for
certain distribution types that occur frequently in applications. We will use these models
to describe the reservoir data in the case studies.
where µ is the mean and σ is the standard deviation of x. The expected value of the
normally distributed random variable is equal to the mean. The normal distribution is also
called the Gaussian distribution and the PDF and CDF are symmetrical in shape.
5
Figure 2-3: PDF(left) and CDF(right) of a normal distribution.
is normally distributed.
frequently associated with processes that divide large quantities into smaller ones. This is
exactly the type of process that occurs during sediment deposition. Therefore, log-normal
distributions are common in the petroleum industry.8 The PDF for the log-normal
distribution is
xσ lnx 2π
6
where µlnx is the mean of ln(x) and σ lnx is the standard deviation of ln(x). The
Figure 2-4 shows the shape of the PDF and CDF curves for a log-normal
seen by the large value of f(x) for small values of x and a long tail extending to the right.
The triangular distribution is useful when limited data are available. In this study,
the triangular distribution is used to describe the reservoir thickness in each case studied,
7
where the number of wells drilled limits the number of reliable data points. This
distribution gets its name from the fact that values are distributed between a minimum, a
maximum, and a most likely value. The equations for the PDF are
0 x<a
f(x) = 2 (x-a)/(b-a) a ≤ x < b (Eq. 2-8)
(c-a) (c-x)/(c-b) b ≤ x < c
0 c≤ x
0 x<a
(x-a)2/[(c-a)(b-a)] a≤x<b (Eq. 2-9)
F(x) = 1 – (c-x)2/[(c-a)(c-b)] b ≤ x < c
1 c≤x
where a is the minimum, b is the most likely, and c is the maximum possible value of x.
The triangular distribution has an expected value given by the following equation:
Figure 2-5 shows the shape of the PDF and CDF for the triangular distribution.
8
Figure 2-5: PDF(left) and CDF(right) of a triangular distribution.
The uniform distribution is a two parameter distribution and is the simplest of all
distributions discussed here. It is useful when little is known about the distribution of the
subject parameter other than a maximum and a minimum. In this study, the uniform
distribution is used as an estimate for the aerial extent of the studied reservoirs. The PDF
0 x<a
f(x) = 1/(b-a) a ≤ x < b (Eq. 2-12)
0 b≤x
0 x<a
F(x) = (x-a)/(b-a) a ≤ x < b (Eq. 2-13)
1 b≤x
9
where a is the minimum possible value and b is the maximum possible value.
Figure 2-6 below shows the shape of the PDF and CDF for the uniform distribution.
There are many other distribution types found in the literature. However, for the
cases studied, it was found that the distribution types described here are sufficient to
describe the data. For a description of other distribution types, please refer to reference
Ball for the simulations, empirical CDF’s and PDF’s are created for porosity and water
saturation. In all cases, the reservoir thickness is assumed to have a triangular distribution
and the area is assumed to have a uniform distribution. Porosity and water saturation are
calculated from core data and from wireline logs. The number of data points available
from each well depends on the number of core measurements made. For each core
porosity measurement, a water saturation value is calculated by reading the resistivity and
gamma ray measurements from the log at the depth at which the core measurement was
10
made. The water saturation is calculated using either Archie’s equation or the Simandoux
equation.11
calculating the thickness of the reservoir penetrated by each well using log analysis. The
minimum, maximum, and most likely values of the triangular distribution used to
describe reservoir thickness come directly from this analysis. The analysis of the
reservoir thickness and seismic data are used to create reservoir isopach maps. The
uniform distributions used to describe the reservoir area are based on the most pessimistic
and optimistic isopach map that can be created with the data. For example, in some cases
there are faults within the reservoir. The area of the reservoir in communication with the
wells depends on the sealing nature of the fault. Since it is not known for sure if these
faults are sealing, the actual area in communication with the wells is not precisely known.
Similarly, the original downdip limit of the reservoirs is not known if there is no water
To create an empirical CDF, the following steps are taken. First, the data is sorted
from smallest to largest, Xi=1 to Xi=N. Then, the corresponding value of F(x)i is calculated
where Ni is the rank of the data point and N is the total number of data measurements. A
11
To calculate the empirical PDF, representative ranges of values for the
data, or bins, are created. The size of the bins are equal, and are in general determined by
the formula
Then the data is examined to determine the number of occurrences of values that
fall into each bin. These occurrences are tabulated, and then the value of f(x) is calculated
by dividing the number of occurrences in each bin by the total number of data points. A
Two tools are used to evaluate the empirical CDF’s and PDF’s. The first is visual
inspection and comparison to the idealized distribution CDF and PDF shapes. The
second, and more reliable tool, is to plot the data on probability paper. For a normal
distribution, the data forms a straight line on probability paper. For a log-normal
distribution, the logarithm of the data forms a straight line on probability paper.
Since common computer software does not easily allow creation of a probability
plot, an alternative, but equivalent method of performing this analysis was done. The
NORMSINV function in Excel is used to create a plot that mimics the behavior of the
data if it were plotted on probability paper. The resulting parameter is referred to as the
probability transform. The Excel function NORMSINV returns the inverse of the
standard normal cumulative distribution. The distribution has a mean of zero and a
standard deviation of one. In order to use this function to create the probability plot, first
the data is ordered and F(x) is calculated as was done in creating the empirical CDF’s.
12
Then two additional Excel columns are created. One is for frequency, which is the
number of times a certain value of the data occurs, and the other is for cumulative
frequency, which is the cumuluative sum of the frequencies. This is done to account for
data values that have more than one occurrence in the data set. Once the cumulative
frequency has been calculated, then the NORMSINV function is used to calculate the
inverse of the standard normal distribution with the cumulative frequency as the
argument. Plotting this on the x-axis with the parameter value on the y-axis gives the
The Monte Carlo method was originally developed by Nicholas Metropolis and
Stanislaw Ulam while working on the Manhattan Project in Los Alamos around the time
of World War II as a way to model the random behavior of sub atomic particles.7 Since
then, the theory has been used by numerous industries and for countless purposes to solve
deterministic problems through the use of random numbers.6 By the early 1970’s
petroleum engineers were beginning to use the technique to model reserves estimates.5
The Monte Carlo method has two requirements. The first is a mathematical model. The
second is knowledge of the CDF’s of the variables to be fed into the mathematical
model.4 When the CDF’s are known each variable needed in the model is randomly
sampled and the model is used to calculate the unknown quantity. This process, known as
a trial, is repeated many times until a sufficient number of trials have been made to create
13
2.5.1 Volumetric Model
For the problem of interest, the model is the volumetric equations for oil and gas
Boi = Initial formation oil volume factor [reservoir barrels per surface
barrels]
Bgi = Initial formation gas volume factor [reservoir ft3 per surface ft3]
The input variables in the above equations are area, reservoir thickness, porosity,
water saturation, Bgi and Boi. In this study, area, thickness, porosity, and water saturation
are treated as random variables with known distributions based on the available data. Boi
Crystal Ball can sample the distributions of the input variables and perform the
random number between 0 and 1, which represents a value of F(x) on the CDF. Once the
random number has been generated, the program reads the corresponding data value from
14
the CDF plot and inserts it into the model. For each trial, a random number is generated,
and the corresponding parameter value is determined, for each of the four input variables,
area, thickness, porosity, and water saturation, required in the volumetric model. The
process is repeated for each trial. A total of 5,000 trials were performed for each
many petroleum reservoirs, the permeability is related to and dependent upon the
porosity. Similar dependencies are known to exist in some reservoirs between aerial
extent and reservoir thickness and water saturation and porosity. When parameters such
as these exhibit a dependence on one another, they are said to be correlated. In this study,
only correlation between porosity and water saturation are considered. Correlations
between these variables are accounted for to assure the model accurately samples the
dependent distributions.
defined as follows:
where
value of zero indicates the variables are completely independent. In general, stronger
correlations between input variables will result in a larger mean value of the resulting
increase.
The degree of correlation between variables affects the way Crystal Ball generates
the random numbers needed to sample the input variable distributions. When no
correlation is present, Crystal Ball generates independent random numbers for each input
variable for each trial. However, when two variables are perfectly correlated, for
example, the same random number should be used for both variables for every trial of the
simulation.14 For variables that are correlated, but not perfectly correlated, there should
be a certain number of trials for which the random number used to sample the CDF’s is
the same for each variable to appropriately model the correlation. It has been shown that
the correlation coefficient can be used to determine the exact number of trials for which
this should be done. In the case of two variables being correlated with a correlation
coefficient of 0.5, exactly one-half of the trials should draw from the CDF’s using the
same random number and exactly one-half of the trials should use independent random
variables. Correlation between porosity and water saturation is accounted for in this
16
study. In general, cases where -0.4 < r < 0.4 can be attributed to statistical noise and in
Crystal Ball also uses a procedure for accounting for correlated variables called
rank correlation.14 This procedure, first proposed by Spearman in 1904, uses the ranks of
the data according to magnitude instead of the data values in calculating the correlation
coefficient.15 This is done to avoid the circumstance in which one or a few data points can
artificially skew the calculation for the correlation coefficient. The rank correlation is
where d is the difference in ranks between the correlated variables, and N is the total
The central limit theorem states that a random variable X, created by the addition
saturation distributions, we would expect the resulting distributions of OOIP and OGIP to
distributions together. Still, the central limit theorem can be used to predict the shape of
17
our resulting OOIP and OGIP distributions. To illustrate this, consider the random
variable Y generated by multiplying together distributions X1, X2, X3, and X4 as shown
Y = X1 X2 X3 X4 (Eq. 2-21)
The logarithm of each of these variables can be taken to arrive at the equivalent
expression
The central limit theorem predicts that Log(Y) will have a normal distribution. But a log-
normal distribution is one in which the logarithm of the variable is normally distributed.
Therefore we can expect that our computed distributions of OOIP and OGIP will be log-
normally distributed.1 The PDF’s for OOIP and OGIP for the reservoirs modeled will
allow us to see this.
18
Chapter 3: Case Studies
Data from five depleted or nearly depleted reservoirs were collected and analyzed.
In each case, distributions of porosity, water saturation, reservoir thickness, and aerial
extent of the reservoir were developed with the help of cores, wireline logs, and seismic
were used to estimate the parameters needed in the volumetric equations. The derived
distributions were loaded into the Crystal Ball Monte Carlo simulation software and
The first case study is a gas reservoir located in the Gulf of Mexico. The reservoir
drive mechanism was water drive. The reservoir is at an average depth of 11,000 feet and
is a clean, unconsolidated sandstone. Seven wells produced this reservoir from 1962 until
1998 with an ultimate recovery of 117,798 MMcf. The production profile for the
and PDF of reservoir porosity was developed. Because the wells were drilled with a
water based mud, reliable estimates of water saturation could not be made from the core
data. As a result, the resistivity and gamma ray logs were analyzed and water saturation
was calculated using the Simandoux equation.11 Based on the known water-gas contact,
the downdip limit was established. Some uncertainty of the size of the reservoir exists
due to complex faulting. The net sand thickness was estimated from the sand thicknesses
as determined by the logs of the seven wells that penetrated and produced the reservoir.
Figures 3-2 and 3-3 show the CDF and PDF generated for porosity from the actual data.
1.00
0.90
0.80
0.70
0.60
F (x )
0.50
0.40
0.30
0.20
0.10
0.00
18 20 22 24 26 28 30 32
POROSITY %
20
0.45
0.4
0.35
0.3
0.25
f(x)
0.2
0.15
0.1
0.05
0
18 20 22 24 26 28 30 32 34
Porosity - %
The CDF and PDF plots for porosity are characteristic of a normally distributed
variable. To check whether or not the distribution is truly normal, a plot of porosity
versus the probability transform is made. This plot is equivalent to a probability plot and
can be used to check for normality. This plot is shown in Figure 3-4. For a normally
distributed variable, a straight line is expected. Figure 3-4 shows that the data does lie in
35
30
P O R O S IT Y %
25
20
15
-3 -2 -1 0 1 2 3
PROBABILITY TRANSFORM
21
Based on the result of Figure 3-4, the porosity is estimated to be reasonably
normally distributed. Using Cyrstal Ball, a normal distribution is assumed with a mean of
26.41 % and a standard deviation of 2.55 %. Figures 3-5 and 3-6 show the CDF and PDF
1.00
0.90
0.80
0.70
0.60
F(x)
0.50
0.40
0.30
0.20
0.10
0.00
10 15 20 25 30 35 40 45 50 55 60
Sw %
0.2
0.18
0.16
0.14
0.12
f(x)
0.1
0.08
0.06
0.04
0.02
0
10 15 20 25 30 35 40 45 50
Sw %
The CDF and PDF plots for water saturation appear similar in character to the
log-normal distribution, with the exception of the increased frequency of measured values
in the range of 35% to 47%. These values may be due to error in measurements. To
22
check if the distribution is truly log-normal, a plot of water saturation values versus the
probability transform on a semi-log plot is made. For a log-normal distribution, this plot
should be a straight line. The plot is shown in Figure 3-7. There is some deviation from a
true straight line on the right side of this curve due to the perceived measurement errors
described above. Otherwise, the data indicate that the distribution is indeed
approximately log-normal.
100
Sw %
10
1
-3 -2 -1 0 1 2 3
PROBABILITY TRANSFORM
distribution with mean of 23.9% and standard deviation of 9.08%. This distribution is
input into the Crystal Ball software for use in the Monte Carlo simulation.
value of 20 feet, most likely value of 49 feet, and maximum value of 98 feet. The area is
approximated with a uniform distribution varying from 1286 acres to 1402 acres. Initial
reservoir pressure was measured to be 5911 psi and the compressibility factor is 0.93.
The porosity and water saturation are checked for correlation by computing the
23
correlation coefficient. Figure 3-8 shows a crossplot of porosity and water saturation. The
33
31
29
27
Porosity %
25
23
21
19
17
15
0 10 20 30 40 50 60
Water Saturation %
To check for the importance of the correlation between porosity and water
coefficient of -0.42 with respect to these variables and again assuming no correlation
between any variables. The required variable distributions are input into the simulator
and 5,000 trial calculations are made to create a distribution of the OGIP based on the
The second case study is an oil reservoir in the Gulf of Mexico. The drive
mechanism was water drive. This reservoir began producing in 1997 and has an ultimate
recovery of 1,218,000 barrels of oil. The reservoir is currently not producing. The
24
Figure 3-9: Case 2 Oil Production History.
There are three well penetrations, all with core data. The core data is somewhat
limited although given the small size of the reservoir it is determined that the data is
geology due to faulting caused by an underlying salt dome. As a result, the reservoir has a
very steep dip. Based on the core data, CDF’s and PDF’s for porosity were created. These
plots are shown in Figures 3-9 through 3-10 and the probability transform is plotted in
Figure 3-11. The PDF for the porosity has an unusual shape in that f(x) increases as
porosity increases to approximately 30% and then abruptly decreases. The peaks and
troughs likely represent holes in the data which would be filled if more cores had been
available. The probability transform for porosity indicates that an assumption of a normal
25
1
0.8
0.6
F(x)
0.4
0.2
0
10 15 20 25 30 35 40
Porosity %
0.2
0.15
f(x )
0.1
0.05
0
10 15 20 25 30 35 40
POROSITY %
40
35
30
Porosity %
25
20
15
10
5
0
-3 -2 -1 0 1 2 3
Probability Transform
standard deviation of 5.36 %. This distribution is input into the Crystal Ball simulator for
Values of water saturation are obtained by a combination of core and log analysis.
The water saturation data are used to create CDF and PDF plots. These plots indicate that
Figures 3-12 and 3-13. Note that the PDF for water saturation appears skewed to the left.
Based on this, a log-normal distribution with mean of 26.45 % and standard deviation of
0.8
0.6
F(x)
0.4
0.2
0
0 10 20 30 40 50
Sw %
27
0.14
0.12
0.1
0.08
f(x)
0.06
0.04
0.02
0
0 10 20 30 40 50 60
Sw %
100
Sw
10
1
-3 -2 -1 0 1 2 3
PROBABILITY TRANSFORM
Figure 3-15: Case 2 Probability Transform Semi Log Plot for Water Saturation.
The distribution for the reservoir thickness is derived based on the three wellbore
penetrations. A triangular distribution is assumed with a minimum thickness of 46 feet, a
maximum thickness of 142 feet and a most likely thickness of 52 feet. The area is
modeled with a uniform distribution assuming a minimum area of 50 acres and a
28
maximum area of 120 acres. Possible correlation between porosity and water saturation is
checked by calculating the correlation coefficient. Figure 3-X shows a plot of porosity vs
water saturation. The correlation coefficient is -0.63. The correlation between porosity
and water saturation may be important. To check this, the simulation is run assuming a
correlation between porosity and water saturation and again assuming no correlation.
40
35
30
Porosity %
25
20
15
10
5
0
0 20 40 60 80 100
Water Saturation %
Based on the parameter distributions, and a known value for the formation
volume factor of 1.27 reservoir barrels per stock tank barrel, the Monte Carlo simulation
The third data set comes from an oil field that was produced as a
waterflood unit. At the time of writing, all water injection has ceased and the unit will
have a very limited economic remaining life. Figure 3-15 shows the production profile.
29
The current cumulative oil production is 14,641,000 barrels and the expected ultimate
The reservoir is at approximately 12,000 feet. There are eleven total penetrations
and core data is available on seven of those wells. From the core data, the CDF and PDF
plots are constructed. The porosity has a mean of 16.27 % and a standard deviation of
6.27 %. The PDF character indicates a bimodal distribution, meaning that there are two
peaks in the plot. This is indicative of two distinct geologic depositional environments
within the reservoir. Each depositional environment is referred to as a parent. Plotting the
porosity against the probability transform gives further evidence that there are two
Further inspection of the data does not indicate that there are certain areas of the
reservoir for which one parent dominates. For example, it might be expected that the
deeper sections of the reservoir underwent one type of deposition and the shallower
sections underwent another. If this were the case, it would be appropriate to separate the
reservoir accordingly and then perform two separate Monte Carlo simulations with
separate porosity, water saturation, thickness, and area distributions for each segment.
30
Because of the complexity of the geology of this reservoir, it is not possible to separate
the parents. Plotting the porosity values against the probability transform show that the
These plots are shown in Figures 3-18 and 3-19. Nevertheless, in order to carry out the
Monte Carlo simulation, a normal distribution for porosity with a mean of 16.7 % and a
standard deviation of 6.27 % is assumed. It is noted that this input distribution is not
completely representative of the actual distribution in the reservoir, and the results from
this simulation will be analyzed to determine if this assumption affects the accuracy of
the calculations. Figures 3-16 and 3-17 show the CDF and PDF for porosity.
1.00
0.90
0.80
0.70
0.60
F(x)
0.50
0.40
0.30
0.20
0.10
0.00
0 5 10 15 20 25 30 35
Porosity %
0.140
0.120
0.100
0.080
f(x)
0.060
0.040
0.020
0.000
0 5 10 15 20 25 30 35
Poros ity %
31
30
25
20
Porosity
15
10
0
-3 -2 -1 0 1 2 3
Transform
100
P oros ity
10
1
-3 -2 -1 0 1 2 3
Transform
The water saturation CDF and PDF plots are created by analyzing the core and
wireline log data; these plots are shown in Figures 3-20 and 3-21. The CDF and PDF are
positive skewedness and a characteristic tail to the right. Plotting the water saturation
against the probability transform on a semi-log plot yields a relatively straight line. This
plot is shown in Figure 3-22. Based on this analysis, a log-normal distribution with a
32
mean of 20.57 % and a standard deviation of 7.4 % is used in the Monte Carlo simulation
( )
1.00
0.90
0.80
0.70
0.60
F(x) 0.50
0.40
0.30
0.20
0.10
0.00
0 10 20 30 40 50 60
Sw %
0.250
0.200
0.150
f(x)
0.100
0.050
0.000
0 10 20 30 40 50 60 70
Sw %
33
100.0
Sw %
10.0
1.0
-3 -2 -1 0 1 2 3
PROBABILITY TRANSFORM
Figure 3-24: Case 3 Probability Transform Semi-Log Plot for Water Saturation.
reservoir.
35
30
25
Porosity %
20
15
10
0
0 20 40 60 80 100
Water Saturation %
34
To determine the effect of correlation, the simulation is run twice, once using the
correlation coefficient of -0.80 for the porosity-water saturation correlation, and again
assuming no correlation between variables. The overall reservoir distributions are input
into the Crystal Ball software and 5,000 trials are made for each case.
The data for the next case is taken from an oil reservoir in Australia. The drive
mechanism was depletion drive. The reservoir has 10 penetrations and core data is
available in three wells. The reservoir is approximately 6,500 feet deep and the ultimate
production is 29,100,000 barrels of oil. The porosity has a mean of 22.42% and a
standard deviation of 2.28%. The CDF, PDF, and probability transform plots show that
the porosity is normally distributed. The PDF shows that the majority of the porosity
values measured fall within a fairly narrow range and the shape is similar to the idealized
normal distribution. The plot of porosity against the probability transform yields a nearly
perfectly straight line. Based on this, a normal distribution with the mean and standard
deviation given above are input into the Crystal Ball simulator. Figures 3-23 through 3-25
0.8
0.6
F(x)
0.4
0.2
0
10 15 20 25 30
Poros ity %
35
0.3
0.25
f(x) 0.2
0.15
0.1
0.05
0
10 15 20 25 30 35
Porosity-%
35
30
25
Po ro sity
20
15
10
5
0
-3 -2 -1 0 1 2 3
Probability Transform
The water saturation CDF and PDF are also characteristic of a normally
distributed variable as shown in Figures 3-26 and 3-27. The probability transform plot
confirms that this is a good assumption as the data plots are nice straight lines with the
exception of a few outlying data points. The plot of the probability transform is shown in
36
Figure 3-28. A normal distribution is input into the Monte Carlo simulator with a mean of
1
0.9
0.8
0.7
0.6
F(x)
0.5
0.4
0.3
0.2
0.1
0
10 20 30 40 50 60 70
Sw (%)
C S S ( C U )
0.14
0.12
0.1
0.08
f(x)
0.06
0.04
0.02
0
20 25 30 35 40 45 50 55
Sw-%
37
80
70
60
50
Sw %
40
30
20
10
0
-3 -2 -1 0 1 2 3
TRANSFORM
seismic amplitude with a minimum value of 1800 acres and a maximum value of 2200
acres. The correlation coefficient porosity and water saturation is -0.14. A plot of porosity
against water saturation is shown in Table 3-X.
Case 4: Porosity vs Water Saturation
30
25
Porosity %
20
15
10
10 20 30 40 50 60 70 80
Water Saturation %
38
The correlation between porosity and water saturation is very weak. Therefore,
correlation is ignored in this case. The distributions are input into the Monte Carlo
simulator and 5,000 trial are made to create the resulting distribution of OOIP.
Case 5 is a gas reservoir located in the Gulf coast of Louisiana. The drive
mechanism was depletion drive. The reservoir is at approximately 10,000 feet and has
good core data from the three penetrations, and the available data is somewhat limited.
The reservoir produced 22,476 MMcf from 1943 to 1953. Figure 3-29 shows the
100,000
10,000
J-43 J-44 J-45 J-46 J-47 J-48 J-49 J-50 J-51 J-52 J-53 J-54
TIME
The core data are used to construct the PDF, CDF, and probability transform plots
for porosity. These plots are shown in Figures 3-30 through 3-32. There is a cluster of
porosity values around 30%, as can be seen on the PDF plot. The probability transform
plots as a reasonably straight line, and it is determined that the porosity is approximately
normal with a mean of 27.2 % and a standard deviation of 5.4%.
39
100%
80%
% Less Than
60%
40%
20%
0%
10 15 20 25 30 35
Porosity (%)
0.3
0.25
0.2
f(x)
0.15
0.1
0.05
0
15 20 25 30 35
POROSITY %
35
30
25
POROSITY
20
15
10
5
0
-2 -1 0 1 2
PROBABILITY TRANSFORM
40
The water saturation data is obtained from wireline log measurements in
conjunction with the core porosity data. The CDF, PDF, and probability plots are shown
in Figures 3-33 through 3-35. Based on the positively skewed PDF plot and the
reasonably straight line seen on the semi-log plot of the probability transform, the water
100%
80%
% Less Than
60%
40%
20%
0%
0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0%
Porosity (%)
0.6
0.5
0.4
f(x)
0.3
0.2
0.1
0
0.0 10.0 20.0 30.0 40.0 50.0
Sw %
41
100.00
POROSITY
10.00
1.00
-2 -1.5 -1 -0.5 0 0.5 1
PROBABILITY TRANSFORM
value of 41 feet, and a maximum value of 71 feet. The aerial extent of the reservoir is
estimated with uniform distribution between 375 and 450 acres. Figure 3-X shows a plot
of porosity against water saturation for this reservoir. The correlation coefficient between
these variables is -0.69.
35
30
25
Porosity %
20
15
10
5
0
0 10 20 30 40 50 60
Water Saturation %
42
Based on the value of the correlation coefficient, porosity and water saturation are
correlated strongly enough to make two separate simulations to determine the effect of
the correlation on the OGIP distribution. The parameter distributions are input into the
Monte Carlo simulator and 5,000 trial calculations are made for the case of correlated
porosity and water saturation and also for the case assuming no correlation of variables.
43
Chapter 4: Results and Discussion
In four out of the five cases, porosity and water saturation were correlated to such
a degree that it was considered necessary to consider this factor when running the Monte
Carlo simulations. These correlations were always between porosity and water saturation.
correlation and the simulations not including correlation, it is concluded that correlation
has a very small effect on the calculations. However, the effect of correlation between the
variables manifests itself in the form of distributions with larger standard deviations and
The CDF and PDF of the OGIP distribution in Case 1 when correlation is taken
44
Figure 4-2: Case 1 PDF for OGIP Considering Correlation.
Table 4-1 shows some summary statistics of the OGIP distribution. Statistics such
as the mean, median, mode, variance, and standard deviation are well known. Others such
as skewness, kurtosis, coefficient of variability, and mean standard error may not be.
Therefore a brief description is given here. The kurtosis is a measure of the peakedness of
a distribution. The ideal normal distribution has a kurtosis of zero. Distributions with a
positive kurtosis, i.e. flatter than the standard normal distribution are said to be
leptokurtic and those with a negative kurtosis are said to be platykurtic. Skewness is a
measure of the asymmetry of the distribution. A negative skewness indicates the
distribution is skewed to the right. The normal distribution has zero skewness. The
coefficient of variation is a measure of the dispersion of a distribution. It is defined as the
standard deviation divided by the mean. Lastly, the mean standard error is simply the
standard deviation divided by the square root of the sample size.18
Table 4-2 shows certain probabilities of the magnitude of OGIP. The range is
bounded by P(0) and P(100). As mentioned previously, P(50) is the median of the
distribution.
45
Table 4-1: Summary OGIP Statistics for Case 1 Considering Correlation.
The results of the simulation when no correlation between water saturation and
porosity is assumed are very similar to the calculations of OGIP when correlation is
considered. A plot of F(x) vs OGIP for both simulations, shown as Figure 4-3, illustrates
this fact.
46
100%
80%
60%
F(x)
NO CORR
CORR
40%
20%
0%
0 100,000 200,000 300,000 400,000 500,000 600,000
OGIP (MMCF)
The resulting CDF and PDF of original oil in place for Case 2 accounting for the
correlation between porosity and water saturation, along with the relevant statistics of the
distribution is shown below in Figures 4-4 and 4-5 and in Tables 4-3 and 4-4.
47
Figure 4-5: Case 2 PDF for OOIP Considering Correlation.
48
Table 4-4: Case 2 Values of F(x) for OOIP Considering Correlation.
To check the effect of correlation, F(x) vs OOIP is plotted for the correlated and
the uncorrelated cases. Figure 4-6 shows this plot. Correlation has a small effect on the
calculations and is not an important consideration in this case. However the correlated
100%
90%
80%
70%
60%
F(x)
NO CORR
50%
CORR
40%
30%
20%
10%
0%
0 5,000 10,000 15,000 20,000 25,000 30,000
OOIP (MBBLS)
The resulting CDF and PDF for the third case are generated for each case. For the
case considering correlation, these plots are displayed in Figures 4-7 and 4-8.
49
Figure 4-7: Case 3 CDF for OOIP Considering Correlation.
50
Table 4-5: OOIP Summary Statistics for Case 3 Considering Correlation.
For this case some of the values of OOIP calculated in the Monte Carlo runs are
negative. It is obviously physically impossible to have a negative OOIP. To explain this,
the input distributions were reviewed. The porosity was assumed to be normally
distributed even though it was pointed out that the reservoir distribution was inadequately
Crystal Ball using the normal distribution for porosity are negative. Therefore, the
51
unsettling to have this problem, the overall distribution of OOIP is considered relevant
for OOIP values larger than 25 million barrels. The values of F(x) vs calculated values of
OOIP for the correlated and uncorrelated simulations are plotted to check for the
100%
90%
80%
70%
60%
F(x)
NO CORR
50%
CORR
40%
30%
20%
10%
0%
-25,000 25,000 75,000 125,000 175,000 225,000
OOIP (MBBLS)
The resulting CDF and PDF plots for the fourth case study are shown in Figures
52
Figure 4-11: Case 4 PDF for OOIP.
Statistics from this simulation are shown in Tables 4-7 and 4-8.
53
Table 4-8: Case 4 F(x) Values for OOIP.
The resulting CDF and PDF of OGIP in the fifth case study when correlation is
considered are shown in Figures 4-12 and 4-13 and some of the critical results of the
54
Figure 4-13: Case 5 PDF for OGIP Considering Correlation.
55
Table 4-10: Case 5 F(x) Values for OGIP Considering Correlation.
The values of F(x) vs calculated values of OGIP for the correlated and
uncorrelated simulations are plotted to check for the significance of correlation on the
100%
90%
80%
70%
60%
F(x)
NO CORRELATION
50%
CORRELATION
40%
30%
20%
10%
0%
0 20,000 40,000 60,000 80,000 100,000 120,000
PREDICTED OGIP (MMcf)
The distributions for OOIP and OGIP obtained from the Monte Carlo simulation
for each case can now be compared to the actual production histories for each of these
56
reservoirs. The recovery factor is the percentage of oil or gas in the ground actually
produced. Table 4-11 shows the recovery factors of each case based on the actual
volumes produced corresponding to the calculated P(10), P(50), and P(90) values of oil
or gas in place as calculated in the Monte Carlo simulations. When correlation was
considered, those values of OOIP or OGIP were used when calculating the recovery
factors.
Real recovery factors vary from 5% to 90% depending on the reservoir fluid
properties and the reservoir drive mechanism. The recovery factors shown in Table 4-11
are considered reasonable for the drive mechanism of the reservoirs and the associated
fluid types. None of the Monte Carlo simulation results shown in Table 4-11 predict
greater than 100% recovery, which would be impossible. In this respect, the results
indicate that the Monte Carlo simulations provide reliable estimates of hydrocarbons
originally in-place. There have been attempts made to estimate reasonable ranges of
Table 4-12. The values given in Table 4-12 are estimates only and there may be cases
where the actual recovery factor of a reservoir is outside of the ranges given.
57
Table 4-12: Typical Recovery Factors for Certain Reservoirs.
definition the P(50) value of the distribution, and calculating the resulting recovery, the
predictions of oil and gas in place are in fact quite good. The recovery factors calculated
against the P(50) values of OOIP or OGIP are considered to be typical.
58
Chapter 5: Summary and Conclusions
determine its effectiveness as an estimator of oil and gas in place. The resulting
distributions from the simulations allow us to conclude that the procedure worked well in
predicting the OOIP and OGIP for the reservoirs studied. The back calculated recovery
factors in general fall within accepted ranges of what would be expected from the
reservoirs given their drive mechanism and hydrocarbon properties. By inspection of the
distributed as predicted by the Central Limit Theorem. The effect of correlation between
variables does not appear to have a large effect on the distributions, though in cases
where correlation was considered, the mean value of the distribution tended to be larger
correlation built in to the volumetric model tended to be more dispersed than those
A review of the literature reveals that an analysis of this type has not been done in
any great detail. Only one case study was found3 in which a probabilistic study was
analyzed for its accuracy. I believe that since the Monte Carlo method is so widely used
in the petroleum industry today that further verification of its accuracy by performing
59
References
1. James Murtha. 2002. “Sums and Products of Distributions: Rules of Thumb and
Applications.” SPE paper 77422
2. Jensen, Lake, et al. 1997. Statistics for Petroleum Engineers and Geoscientists.
4. James Murtha. 1994. “Incorporating Historical Data Into Monte Carlo Simulation.”
SPE paper 26245
5. Evers and Jennings. 1973. “How to Use Monte Carlo Simulation in Profitability
Analysis.” SPE paper 4401
6. Gentle, J.E. 1998. “Random Number Generation and Monte Carlo Methods.”
Springer- Verlag New York, Inc., 1998, 247 pp.
7. Metropolis, N. and Ulam, S. 1949. “The Monte Carlo Method.” J. Amer. Stat. Assoc.
44, 335-341.
10. James Murtha. 1997. “Monte Carlo Simulation: Its Status and Future.” SPE paper
37932
11. Bassiouni. 1994. Theory, Measurement, and Interpretation of Well Logs. pp. 16-17
12. Craft and Hawkins. 1959. Applied Petroleum Reservoir Engineering
13. James Murtha. 2004. “When Does Correlation Matter?”
14. David Morgan. 2004. “What To Do When You Need 240 Correlations.” Proceedings
of the 2004 Crystal Ball User Conference
15. J.J. Arps and T.G. Roberts. 1955. Journal of Petroleum Technology p. 124
16. J.J. Arps. 1956. “Estimation of Primary Oil Reserves.” AIME 4331
17. D. Vose. 1996. Quantitative Risk Analysis. p 53-54
18. Decioneering. 1996. Crystal Ball Manual, Version 3.
60
19. Merran Evans. 2000. Statistical Distributions, 3rd Edition
61
Vita
John Williams was born on June 30, 1972 in West Allis, Wisconsin, son of Harry
Engineering from The University of Texas in 1995. He has worked in the petroleum
62
63