SPE 89983
Probabilistic Approach Takes You Back to Reality
S. Macary, SPE, IPR Group of Companies
Copyright 2005, Society of Petroleum Engineers
This paper was prepared for presentation at the 2005 SPE Hydrocarbon Economics and
Evaluation Symposium held in Dallas, TX, U.S.A., 3 – 5 April 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
One of the most important outputs of the static model is the
volumetrics of the original hydrocarbon in place (OHIP).
According to the type of reservoir driving mechanism, the
reservoir engineer can assume the recovery factor and hence the
recoverable reserves. Most simulation studies are being
conducted on fields considering a period of production
performance to determine the remaining reserves and the best
scenario of development to produce these reserves, which is the
main function of the dynamic model.
The volumetrics could be approached using a Monte Carlo
technique, the end product of which is a range of probable
OHIP and, consequently using the proper recovery factor,
getting probability distribution of the recoverable reserves.
Therefore, using the probabilistic approach is superior in green
fields rather than brown fields because it captures the full range
of reality and where models are not yet calibrated to dynamic
data.
Sometimes, the dynamic data may take a surprise turn from
the geological understanding. Different realizations, even the
crazy ones, may help under the probability distribution
functions in defining some certainties that have not yet even
been imagined. In addition, each of the dynamic model’s
scenarios yields extra recovery, the certainty of which can be
evaluated using the probability distribution curve of the
remaining reserves. This certainty has direct influence on the
economics approval of one scenario over the others.
The current work introduces actual case histories
demonstrating how the probabilistic approach could be of real
help in selecting the production enhancement scenario and how
this tool also could be used to prioritize management interests.
Introduction
After many trials of trade off, both sides: geoscientists and
reservoir engineers could come to a mutual agreement whether
to accept the model, i.e., the static one for further use in
building the dynamic model. However, this does not mean,
under any circumstances, that the accepted model is certain. In
taking this static model to the dynamics, the domain of
uncertainties becomes wider. The main parameter that makes
the results of the dynamic model to be convergent is the
feasibility, which is, in turn, at high level of uncertainty, e.g.,
oil prices.
In almost all of the reservoir simulation studies you can
find deterministic figures counting to the second decimal of
MMBO or US$MM (i.e., ten thousand), which is sometimes as
small as daily production. The question is: how much
accuracy and certainty in our input that allows us to come
up with such deterministic numbers? Murtha (1) said “Much
of risk analysis consists of estimating something with a range
of values rather than with a single value. We report that the
wildcat well drilling will require between 56 and 87 days
instead of saying it will take exactly 65 days. Instead of the
single-point estimate of US$ 34 MM, we say the water flood
NPV is a normal distribution with a mean of US$34 MM and a
standard deviation of US$1.7 MM” (equivalent to a range of 29
to 39 MM US$).
The reliability of reserves estimates is the economic
foundation of the petroleum industry. This is the initial point
where $ sign appears. A dependable foundation for reserves
estimates helps to ensure sound choices and avoid errors in the
development of potential oil and gas properties. This, in turn,
positively affects the confidence of shareholders, corporate
staff, suppliers and contractors, government agencies, and
society as a whole (2).
In the next paragraphs will follow a backward process that
starts off by dynamic model results and ends with probabilistic
reserves. The current work tries to tie the output of the
simulator with the probability distribution of reserves in terms
of uncertainty or the chance of success.
No doubt that it will be a revolutionary approach if we
could use Monte Carlo simulation in dynamic models to
produce directly a distribution of uncertainty per each logic
scenario aiming recovery increase.
2 SPE 89983
Uncertain Reservoir Description Data
Bradley (3) and Sverdrup (4) stated that errors in reservoir
description data clearly contribute to errors in simulation model
results. Since the description data are never exactly known, one
might infer that model results are necessarily erroneous and
unreliable. A number of considerations contribute, in
contradiction of this inference, to model results being widely
used to select and to design oil-recovery processes and to
forecast oil recovery.
In any event, the pertinent question regarding reservoir-
description data is not related to correctness or uniqueness in an
absolute sense but it concerns the engineering significance of
variations in parameter values within ranges of uncertainty. The
simulation model itself might be useful in estimating this
significance.
Our case histories have a wide range of variations within
lithology, driving mechanism, volume, rock and fluid
properties. Regardless of the number of scenarios (modules)
considered per each case, this study concerns only one
scenario/each reservoir that gave through simulation runs the
maximum incremental recovery above the base case, which is
“Do nothing”. It’s worth to mention that values used to
calculate reserves probabilistically (volumetrics) were taken
from the averages considered in the simulation study. Monte
Carlo simulation then was used to create probability
distributions using logic and industry experience assumptions
for each variable of the volumetrics.
Probabilistic Reserves versus Deterministic Ones
Many believe that for the first time, the official reserves
definitions allow for the use of probabilistic reserves though
they do not require this methodology. Not true-reserves
estimates have always been probabilistic, whether or not the
definition recognized that fact. Words such as “reasonably
certain”, “with high confidence”, “more likely than not” and
“possible” all lay tribute to the concept of probability.
For probable reserves, Capen (5) asked “What opportunities
do I see in this field that I can move into the proved category
over the next 2 to 3 years?” For possible reserves, consider
looking at a 4 to 6-year time horizon. How many new barrels
will begin their journey to the well bore during that time? These
possible reserves define the optimistic end of the distribution.
Treating reserves probabilistically can provide useful
information for decision makers as they plan for future
development and extension exploration.
Idrobo (6) agreed that the correct use of the combination of
deterministic and stochastic methods along with consistent
reservoir engineering criteria guarantees healthy reservoir
management decisions.
Wright (7) concluded that uncertainty is a fact of life in
reserve estimates, in which the percentage error is smaller and
less volatile than the percentage error in remaining reserves.
Additional data (in the form of production data, if nothing else)
helps narrow the range of uncertainty until late in the life of the
property. However, there is a point of irreducible uncertainty
regardless of the amount of money and effort expended.
Data Surveillance
A multireservoir field in the Western Desert-Egypt was the
subject of an integrated engineering study in 2000; the point
forward date of the prediction runs was January 2001, while the
forecast termination is June 2006. The simulation study
included 55 modules (scenarios) out of which only 37 were
reported. Among these scenarios one can find well optimization
(different workover jobs: add perforation, new completion,
cleaning, etc), infill drilling (vertical or horizontal) and
secondary projects (water injection).
Table 1 introduces a comparison of descriptive statistics of
Monte Carlo probabilistic simulation (Mean, Min, Max, P10,
P50 and P90) versus the simulation study output (both OOIP
and reserves) as calculated either volumetrically or by material
balance (MBC) for all reservoirs, in addition to the total field
(aggregated probabilistically). The simulation study did not
provide complete picture of reserves through MBC due to the
absence of pressure data. Anyhow, this is not the concern of the
current work.
Table 2 illustrates the production history of our concerned
case; December 2000 stands for the end of production history
match period used in the simulation study; December 2004 is
the last oil cumulative used in the analysis (current work). This
table also shows the incremental production for the last four
years and the certainty of the cumulative production with
respect to the entire range of ultimate reserves. Table 3 shows
the incremental recovery from the best development scenarios,
both individually and as a combined total field development
scenario.
Discussion
At that stage of the field age regarding the best scenarios
concerned, the author believes as SPE (8) does that the
incremental recovery gained through simulation study scenarios
belongs to the category of proved reserves (developed). In our
case (i.e., the best scenarios not all examined modules) there is
no step drilling, infill seismic,
or pressure maintenance projects; the best scenarios are either
recompletion or ESP installation.
First of all, engineers have to understand that any value within
the range of probabilistic reserves frequency distribution is
probably correct as any other one; the matter is how to narrow
the range. Again, the older the reservoir is (more data) the more
realistic are the ranges of properties used in the volumetrics
(porosity, Bo, RF and gross volumes, Sw). One interesting
approach to narrow reserves range was introduced by
Verbruggen (9) in which the ultimate recovery (UR) range is
SPE 89983
narrowed by setting the current cumulative production value at
the P100 value, i.e. 100% confidence level of achieving this
recovery.
Analyzing the tabulated data we can make the following
comments:
a) OOIP values from the simulation study do match with
both the Mean and P50 output of Monte Carlo (Table 1). This is
expected, as the avareges (most likely values of volumetrics
parameters) are the same in both studies. However;
b) Simulation study reserves show the following
inequalities with respect to the Monte Carlo ones: AD >P50,
DS > P30 and others are even less the P10 (Table 1). This
complete dismatch could be a result of incorrectness of either
way: Monte Carlo overestimated recovery factor or simulation
study pessimism! Anyhow, this should be interpreted in the
light of OOIP match;
c) All reservoirs to date cumulative is higher than the
simulation study UR (Table 2); however, this cumulative still
far less than P10 with the exception of AD;
d) Almost all reservoirs for the last four years have
produced more than two folds the predicted values through the
best scenario calculated to June 2006 (Table 3). One may
attribute this positive situation to good reservoir management
but it is better to evaluate what was the uncertainty in achieving
incremental production by applying one or another scenario.
Now, let us go to definite situations to judge the added
value driven from certainly levels. Both reservoirs AS because
of its small size and DS because of being shut in are not
included in the next analysis. By any means, approaching
higher production with respect to predicted certain one by
Monte Carlo could be a result of better reservoir understanding
and management
We start off by AD reservoir. Cumulative production to
December 2000 was 62.93 MMBO covering 72% from the
reserves range produced by Monte Carlo. Other words, this
means having just 28%
chance of success of getting more (see Fig. 1). In subtracting
the cumulative oil produced from the total range of
probabilistic reserves we got what we can call “Remaining
Reserves”. Figure 2 shows the frequency distribution of these
remaining reserves. The best scenario gave an incremental
recovery of 0.446 MMBO, which has just 24% certainty
(chance of success). In real life, the incremental recovery from
this reservoir (Jan. 2001-Dec. 2004) was only 0.15 MMBO-just
one third of the predicted incremental and produced within
about 80% of the prediction interval (Jan. 2001-June 2006).
The other two reservoirs: AR/G and RS have a
completely contradictory status in comparison with AD
reservoir. RS may have more superiority, as it’s three times
bigger by volume as AR/G as illustrated in Figures 3 and 4.
Regardless this fact, they’re twins with respect to the analysis
tying simulation results with the probabilistic approach. For RS
3
reservoir, the incremental production for 80% prediction period
is almost two folds the predicted recoverable reserves by
simulation. In addition, even December 2004 cumulative value
covers less than 1.5% from the probabilistic range. Hence, there
is up to 98% certainty to get bigger values of recoverable
reserves, i.e., there is an extremely high chance of success to
feasibly develop this reservoir.
Very interesting point is to compare between the
uncertainties of remaining recoverable reserves in AD and RS.
It’s clear that the chance of success for RS reservoir to produce
reserves up to P50 (about 5.4 MMBO incremental) is much
higher than the chance of AD to go to P90 reserves value (4
MMBO incremental). The other reservoir AR/G has the same
property, very high chance of success to go to P50 with
incremental recovery of about 1.8 MMBO. Then, both
reservoirs can produce with high certainty two folds of what
AD may produce with higher uncertainty.
The purpose of the current work was just to compare
numbers gained from both simulation study and probabilistic
Monte Carlo simulation using probability distribution functions
(PDF), we did not want to go into details like reservoir driving
mechanism, reservoir characterization, stratigraphy or any other
parameters either favorable or not. However, the simple
example given above about certainty of producing some
incremental oil lets us to have a look at the economics of the
development by each scenario. It was found that the sum of
investments to implement best scenario for both RS and AR/G
is just one sixth of the needed investment to develop AD. This
is another justification to highly prioitize developing these
reservoirs.
This combination of oil volumes, uncertainty analysis and
economics could be effectively used in prioritizing
development schedule and making successful decisions.
Decision makers should take into account that the simulation
study is an ongoing process
that needs to be updated on regular basis as long as there is a
volume of data added. This will produce dynamic behavior of
reserves, which regardless of being a growth or a decline means
better understanding of the reservoirs attidude and capabilities.
There’re another two points we need to highlight. First, any
new evaluation should start from two positions: certainty of
produced volumes with respect to the total ultimate recovery
and probability distribution of the remaining reserves. Figure 5
is the “Overlay Plot” illustrating the share of the three reservoir
in the total remaining reserves, while Fig. 6 is a trend plot
showing a comparison between the remaining reserves in the
three reservoirs expressed in terms of P10, P50 and P90. Both
plots confirm the above mentioned analysis about the
superiority of some reservoirs above the others. The second
point is to use this trade off between the two concepts to tune
our decisions.
Conclusions
1. Simulation engineer should use a pragmatic approach
to quantify the uncertainty ranges around the base case.
4 SPE 89983

8. www.spe.org “ Petroleum Reserves and Resources

2. It’s an open invitation to simulator designers to build
Definitions” .
in some uncertainty analysis for further improvement of
probabilistic reserve estimates.
9. Rini Verbruggen et al: “Understanding Reserves Uncertainties
in a Mature Field by Reservoir Modeling”, SPE 77896 presented
3. Bigger reservoirs give better reasonable results in at the SPE Asia Pacific Oil & Gas Conference and Exhibition
comparison as they’re subjected to higher level of exploitation (APOGCE) held in Melbourne, Australia, Oct. 8-10, 2002.
activities.

4. Probability distribution functions could be used as a

quality check tool to evaluate the level of reservoir
management.

5. Such approach in evaluating reserves reveals the

priorities in developing multiple choices.

6. combination of oil volumes, uncertainty analysis and

economics could be effectively used in prioritizing
development schedule and making successful decisions.

Acknowledgments

The author extends his thanks to IPR Group of Companies for

permission upon using the data of the simulation study in
addition to the access to the current production performance.
He also does thank IPR Group of Company for supporting the
publication of this paper. Many other thanks are to be addressed
to Dr. Jim Pollin upon his valuable comments and Mr. Dale
Hinshaw upon his instructive directions.

References

1. James A. Murtha: “Monte Carlo Simulation: Its Status and

Future”, SPE 37932 presented at the SPE ATCE, San Antonio,
TX, Oct. 5-8, 1997.
2. W. Gary McGilvray: “Independent Evaluation for Reliable
Reserves Estimates”, JPT, Dec. 2004.
3. Howard Bradley: “Petroleum Engineering Handbook”, Third
Edition, Richardson, TX, USA, 1992.

4. Einar Sverdrup: “Modeling Streamlines Workflows,” E&P,

Aug. 2004, pp. 69-72.

5. E.C Capen: “Probabilistic Reserves! Here at Last?”

SPE Reservoir Evaluation & Engineering, Oct. 2001, pp. 387-94.

6. E. A. Idrobo et al: “A New Tool to Make Quick Estimates of

Probabilistic Reserves from Production Trends”, SPE 68597
presented at SPE Hydrocarbon Economics and Evaluation
Symposium, Dallas, TX, Apr. 2001.

7. John Wright: “Irreducible Uncertainty: A Fact of Life in

Reserve Estimates”, SPE 84146 presented at SPE ATCE, Denver,
CO, Oct. 5-8, 2003.
SPE 89983 5

Table 1 - Comparison: Probabilistic versus Study Deterministic Output

Monte Carlo Results Simulation Study

NAME MEAN MIN MAX P10 P50 P90 Determ. MBC
MMSTB
OOIP 17.13 5.00 27.50 12.86 16.90 21.79 17.48 N.A.
1 AR/G
Reserves 4.27 1.20 7.20 2.93 4.21 5.68 1.788
OOIP 39.89 15.00 65.00 28.85 39.61 51.39 45.071 N.A.
2 RS
Reserves 12.95 5.90 22.50 9.17 11.99 17.15 5.83
OOIP 11.83 2.50 20.00 8.17 11.65 15.82 12.737 12.51
3 DS
Reserves 4.16 1.00 8.00 2.72 4.17 5.64 3.501
OOIP 189.55 125.00 250.00 164.48 189.16 214.72 192.966 209.84
4 AD
Reserves 59.58 40.00 80.00 50.98 59.53 67.16 62.881
OOIP 6.99 3.00 12.00 4.98 6.86 9.28 7.193 N.A.
5 AS
Reserves 1.06 0.38 2.50 0.73 1.06 1.42 0.09
OOIP 265.80 200.00 350.00 238.62 271.17 292.55 270.11
Total Field
Reserves 82.35 60.00 105.00 72.64 81.98 92.38 74.09

Table2 - Production History of the Concerned Case

Cum Des. 2000 Cum Dec. 2004 Incremental
Reservoirs Date of Qi Certainty
MSTB MSTB MSTB
AR/G 1988 1,788 2,440 < P10 652
RS 1981 5,829 6,610 < P10 781
DS 1974 3,501 3,501 P30 < 3.5 < P50 0
AD 1968 62,930 63,079 P70 < 63.1 < P80 149
AS 1996 94 134 < P10 40
Total Field 74,142 75,764 P20 < 76.3 < P30 1,622

Table 3 - Simulation Study Results

Reservoir Incremental Recovery, 01.2001-06.2006, MSTB

Name Do Nothing Best Scenario

AR/G 238 446

RS 264 404

DS 0 50

AD 76 446

AS 87 161

TOTAL FIELD * 665 1459

* Not a sum of the above (separate combined scenario)

6 SPE 89983

Forecast: Reserves, Dolomite

996 Trials Frequency Chart

.036 36

.027 27

.018 18

.009 9

Mean = 59.27
.000 0

40.00 50.00 60.00 70.00 80.00

Certainty is 28.31% from 62.93 to 80.00 MMSTB

Fig. 1 – Reservoir AD Probabilistic Reserves and Certainty of Cumulative Oil Produced

Forecast: REmaining Reserves

962 Trials Frequency Chart

.746 718

.560 538.5

.373 359

.187 179.5

.000 0

0.00 2.25 4.50 6.75 9.00

Certainty is 24.22% from 0.42 to 9.00 MMBO

Fig. 2– Reservoir AD Probabilistic Remaining Reserves

Forecast: Reserves, AR/G

995 Trials Frequency Chart

.029 29

.022 21.75

.015 14.5

.007 7.25

Mean = 4.25
.000 0

1.50 3.00 4.50 6.00 7.50

Certainty is 97.99% from 2.40 to 7.20 MMSTB

Fig. 3– Reservoir AR/G Probabilistic Reserves and Certainty of Cumulative Oil Produced
SPE 89983 7

Forecast: Reserves, Razzak Sand

996 Trials Frequency Chart

.032 32

.024 24

.016 16

.008 8

Mean = 12.94
.000 0

5.00 9.38 13.75 18.13 22.50

Certainty is 98.69% from 6.93 to 22.50 MMSTB

Fig. 4– Reservoir RS probabilistic Reserves and Certainty of Cumulative Oil Produced

Overlay Chart

Frequency Comparison
.212
3-Reservoirs Remaining Reserves

.159

AR/G Remaining Reserves

.106

RS Remaining Reserves
.053

Dolomite Remaining Reserves

.000

0.00 6.88 13.75 20.63 27.50

Fig. 5 – Overlay Plot Showing the Share of each Reservoir in the Remaining Reserves

Trend Chart

Trend Chart: P10, P50 and P90 for the Three Reservoirs
15.00

90%

10.00

50%

5.00

10%

0.00
Dolomite Remaining Reserves

RS Remaining Reserves

AR/G Remaining Reserves

Certainties Centered on Means

Fig. 6 – Trend Chart Showing How Much Certainty of Remaining Reserves within Each Reservoir
8 SPE 89983