SPE-169471-MS

Original Gas in Place Probabilistic Assessment Using Design of

Experiments on Material Balance Models in the Dolphin Field
A. Guerrero, G. Gomez, A. Lerza, and J. Dolan, Chevron

Copyright 2014, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Latin American and Caribbean Petroleum Engineering Conference held in Maracaibo, Venezuela, 21–23 May
2014.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Dolphin field is a mature gas field located in the East Coast Marine Area, offshore Trinidad. One of the
methods to estimate gas initially in place (GIIP) of the field is material balance; however, this approach
presents some particular challenges which include: sands are not well consolidated and exhibit very high
rock compressibilities, communicating reservoir compartmentalization and wells are produced at very
high rates. These three factors don’t allow for a straight-forward understanding of the aquifer size,
normally assessed with a deterministic Cole Plot analysis. As a result of these complexities, analysis
demands a more in-depth understanding of the uncertainties of the reservoirs and their implications on
fluid in place estimation to avoid bias and anchoring. This paper presents a case study of the workflow
used in determining probabilistic material balance models for one sand in the Dolphin Field.
Initially, an uncertainty analysis was conducted to determine the appropriate ranges of petrophysical
data and aquifer properties that were used as input information for the material balance models.
Experimental Design was then successfully applied to generate probabilistic material balance models:
Placket-Burman design of experiment was used to identify the key uncertainties in a linear model. Monte
Carlo simulation was used with the resultant response surface equation to determine probabilistic volumes
of the field.
This project successfully achieved a probabilistic GIIP range for a single method testing sand of the
field, ultimately yielding a better understanding of the significant uncertainties and their implications. The
resulting probabilistic material balance models also enable the use of an integrated asset model to generate
fit for purpose forecasts and a provides a more rigorous quality checking method on reservoir simulation
results.
This paper will discuss a simple workflow used in probabilistically determining GIIP for one producing
horizon in Dolphin field. Given the high confidence results encountered on the single sand, this approach
will be applied to all Dolphin sand horizons to provide higher confidence GIIP estimates for the entire
portfolio.
2 SPE-169471-MS

Figure 1—Methodology

Background
Dolphin is an offshore dry gas field located within
the East Coast Marine Area (ECMA) of Trinidad and
Tobago. Found approximately 65km off Galeota
Point, Dolphin field was discovered in 1976 by Ten-
neco and started production in 1996 and currently is
operated by BG Trinidad and Tobago in a joint
venture with Chevron. Dolphin sands were deposited
in Plio-Pleistocene shelf depositional environment,
with a gradual deepening from a delta front area,
strongly progradational sequences influenced by
proto-Orinoco. These sands have low rock strength
with high rock compressibility. The methodology
used to assess probabilistic Gas Initially in Place
(GIIP) was based on the Lower E Sand found in Figure 2—Uncertainty Analysis
Dolphin field; this sand has one well that produces
the sand and 5 other wellbores throughout the field
that provide reservoir characterization information for the sand. During the production history of the well,
there have been several planned and unplanned shutdowns which provide a very good record of static
reservoir pressure of the sand. Material balance [1] is the technique of choice used to estimate GIIP for
mature gas fields; it is simple and reliable, however, the results could be biased depending upon the
analyst and the choice of deterministic properties used when performing the analysis. In this mature gas
field the uncertainties that most impact the GIIP estimates are rock compressibility and aquifer properties
[2]
. Cole plot [3][4] analysis is a great way to diagnos aquifer influence[5], however, this could lead to a bias
by neglecting small aquifer influence when a depletion drive is diagnosed. The method used to model the
aquifer influence is the Carter-Tracy[6][7]analytical model, which requires some degree of knowledge of
the properties of the rock in the aquifer zone and its extension.

Methodolgy

Initially, an uncertainty analysis is performed to develop appropriate ranges for each uncertainty.
Plackett-Burman experimental design was then used to generate the experiments and this input data was
SPE-169471-MS 3

Table 1—Experimental Design Table

Exp # Sw Cr Aquifer Thickness Aquifer Ro/Ri Aquifer Encroachment Angle Aquifer Permeability Normalized GIIP

1 ⫺1 1 1 ⫺1 1 ⫺1 0.92
2 1 ⫺1 1 ⫺1 ⫺1 ⫺1 0.98
3 ⫺1 ⫺1 ⫺1 1 1 1 0.84
4 1 ⫺1 ⫺1 ⫺1 1 1 0.98
5 1 ⫺1 1 1 ⫺1 1 0.86
6 ⫺1 1 ⫺1 ⫺1 ⫺1 1 0.93
7 1 1 1 ⫺1 1 1 0.90
8 ⫺1 ⫺1 ⫺1 ⫺1 ⫺1 ⫺1 1.00
9 0 0 0 0 0 0 0.89
10 ⫺1 1 1 1 ⫺1 1 0.69
11 ⫺1 ⫺1 1 1 1 ⫺1 0.83
12 1 1 ⫺1 1 ⫺1 ⫺1 0.76
13 1 1 ⫺1 1 1 ⫺1 0.69

Figure 3—Regression Quality: MBAL® vs PROXY GIIP

loaded into MBAL®* to create material balance models. For every model the GIIP was regressed to
history match the pressure history of the sand. A PROXY equation was created by regression of the GIIP
results for every experiment. Monte Carlo simulation was performed to generate the probabilistic GIIP
results. Figure 1 below reflects a graphical representation of the process.
Uncertainty Analysis
A comprehensive analysis was undertaken to capture appropriate ranges of the uncertainties and ulti-
mately to generate the various material balance models. Water saturation, rock compressibility, aquifer
thickness and aquifer permeability were modeled with cumulative distribution functions (cdf). These cdf’s
were generated from log information taken from the wells that intersect the sand within the field. It is also

*
MBAL® is a trademark of Petroleum Experts Ltd.
4 SPE-169471-MS

Figure 4 —Pareto Chart: Statistical Significance

Figure 5—Linear Response Regression Form

very important to represent the full uncertainty range if the range is deemed to behave as a normal
distribution; if the sample size is less than 30, a t-student distribution was a good option to increase the
uncertainity range. Deterministic ranges of aquifer size and encroachment angle were used based upon
geological structural uncertainties that led to low, mid or high scenarios. Figure 2 below represents
acquifer size and cdf’s associated with select uncertainties.

Experimental Design
Experimental Design is a technique used to minimize the number of trials or experiments performed
ultimately yielding significant information in an efficient, concise way. In this study a linear model with
7 terms was created with Plackett-Burman experimental design, which included 6 factors and one center
point. The resulting 13 experiments were used to populate deterministic MBAL® models. The GIIP of
every MBAL® model was regressed to history match the pressure behavior over the 15 years of
production of the sand. Table 1 shows variables used and ranges, and the resultant GIIP values.
Uncertainties and GIIP results were then regressed to calculate the PROXY linear function.
The quality of the regression can be analyzed using a crossplot of the MBAL® GIIP regression and the
PROXY results. Ideally, the crossplot data should exhibit a slope of 1. In this case, the linear correlation
is very reasonable with a correlation coefficient R2 value of 0.94. Finally, another quality check was to
create a blind test by assuming values of the different variables and estimating the GIIP in the MBAL®
SPE-169471-MS 5

Figure 6 —Normalized GIIP S-Curve

Figure 7—Tornado Plot

model and the PROXY; it was quite evident that the test follows the linear trend of the crossplot. Figure
3 below shows regression quality of the proxy values versus MBAL® GIIP.
Another important output of the regression was the Pareto Chart, which allows for identifying
statistically significant factors. In this case the t-critical for 6 degrees of freedom is deemed a significant
limit. Figure 4: Pareto shows that rock compressibility and Aquifer Size were statistically significant. In
consequence, if GIIP uncertainty plays a key role in a decision, then these two uncertainties should be
narrowed down with additional studies or data acquisition.
Monte Carlo Simulation
Cdf of every variable is used in conjunction with the linear response regression model. Monte Carlo is a
simple stochastic simulation, where values are assigned randomly to every factor honoring the cdf. Figure
5 shows the form of the linear PROXY equation in conjunction with the shape of the cdf for every factor.
6 SPE-169471-MS

50,000 realizations were simulated in this study to obtain the probabilistic GIIP results. The resulting
S-Curve shows a 38% variation in GIIP which is deemed to be reasonable. Recognizing the complete
range of uncertainty is important to prepare robust production forecasts and create an appropriate
development plan. Figure 6 shows cdf of normalized GIIP results.
The last task was to select 3 realizations to represent the entire range of uncertainties in GIIP. In this
case a Tornado chart is very useful to determine the combination of variables that should yield an
approximation of the P10, P50 and P90 GIIP models with special focus on the more significant
uncertainties. For instance, a realization with a GIIP very similar to the P10 volume, with high rock
compressibility and a large aquifer was selected. Figure 7 shows a Tornado chart of the variables.
These models were then incorporated into a fit for purpose Integrated Asset Model, which is used to
review development alternatives for the field.
Conclusions
● Experimental Design applied to material balance models is a robust and successful
technique to generate reliable probabilistic GIIP outcomes using Monte Carlo
simulation, minimizing the number of trails and ultimately reducing bias asso-
cianted with deterministic methodologies.
● A linear PROXY appropriately models the GIIP dependence on the uncertain
factors for this sand.
● Uncertainty quantification must incorporate appropriate ranges otherwise results
will likely not be reliable.
● Recognizing the significant uncertainties is very valuable in minimizing effort and
could ultimately result in reduced investments.

Acknowledgements
The authors of would like to thank BG Trinidad and Tobago, BG Group and Chevron for permission to
publish this paper.

References
1 Dake, L.P.: Fundamentals of Reservoir Engineering, Elsevier Scientific Publishing Co., Amster-
dam (1978).
2 Bruns, J.R., Fetkovitch, M.J., and Meitzen, V.C: “The Effect of Water Influx on p/z-Cumulative
Gas Production Curves”, JPT (March 1965).
3 Dake, L.P: The Practice of Reservoir Engineering, Elsevier, Amsterdam (1994).
4 Cole, F.W: Reservoir Engineering Manual, Gulf Publishing Co., Houston (1969).
5 J.L. Pletcher, Marathon Oil Co. “Improvements to Reservoir Material-Balance Methods”. SPE
75354-PA.
6 Carter R.D. and Tracey G.W.: “An Improved Method for Calculating Water Influx”, JPT Sep.
1960, SPE 2072.
7 MBAL®, version 10.5. User Guide. Edinburgh, United Kingdom.
8 Mohan, K. 2008. Natural Gas. Production Engineering. First Edition. Tulsa, Oklahoma: Pen-
nWell Corp.
9 Charnes, John. 2007. Financial Modeling with Crystal Ball and Excel 2007. p. 89. Hoboken, New
Jersey: John Wiley & Sons, Inc.