Numerical and Experimental Modelling of the

Steam Assisted Gravity Drainage (SAGD) Process

K. SASAKI, S. AKIBAYASHI
Akita University

N. YAZAWA
Technology Research Centre, Japan National Oil Company

Q. DOAN, S.M. FAROUQ ALI

University of Alberta

Abstract
For complex petroleum recovery processes, an experimental
investigation is usually performed with a numerical simulation to
study the recovery mechanism(s). In this paper, both physical and
numerical simulations of the steam assisted gravity drainage
(SAGD) process were performed. One of the objectives of the
numerical investigation was to determine the match between
numerical results with data generated from scaled model experiments. The Computer Modelling Groups (CMG) STARSTM
thermal simulator was used. Results from the numerical simulation were found to be in reasonable agreement with those
obtained from the experiments for oil production rates, and
cumulative oil production. In addition, the steam chamber volume and temperature distribution were also examined.
Effects of different parameters, such as steam injection pressure, vertical separation between injection and production wells,
and reservoir thickness, on the performance of the SAGD process
were investigated. They were observed to have the same effects
on both experimental and numerical results. The numerical simulator was also used to study the influence of rock and fluid properties, such as oil viscosity, permeability, porosity, and the
amount of heat loss from the reservoir to the surroundings.

FIGURE 1: Schematic model of SAGD process [after Butler(1)].

Introduction

Description of the Experimental Model

The steam assisted gravity drainage (SAGD) process was developed by Butler(1), and is illustrated in Figure 1. It has been applied
in several projects, including the Underground Test Facility (UTF)
and has shown promise of achieving high recovery (more than 50%
of OOIP in some cases).
Many experimental and numerical studies of the SAGD process
have been carried out over the last ten years, on different aspects of
the process. One of the recent numerical studies was presented by
Chow and Butler(2). They focused on history matching the oil
recovery and the steam temperature interface position with those
observed in the SAGD experiments by Chung and Butler(3).
In the present study, numerical history matching of the experimental data, such as the oil production and the steam chamber temperature contours [Sasaki et al.(4)] is the main focus. Furthermore,
time to establish initial communication between the two steam
injection and production wells (steam breakthrough time) was
investigated. The physical and operational conditions in the experimental study were different, compared to those used in Chung and
Butlers experiments(3). They included a pressure drop of Pi = 20
kPa, permeability of k =142 D; no pre-heating was employed. The
experiments were configured to examine phenomena associated
with the rising chamber. More details are provided in the following,
for both experimental investigation and numerical simulation.

Several 2D visual, scaled physical models were used in the

experiments. They were designed to represent a vertical section of
a heavy oil reservoir. The models had sidewalls of acrylic resin (of
20 mm in thickness). For one of the models, both the width and
height of the model were 380 mm; its thickness was 44.5 or 49.5
mm. Another model had width and height of 300 mm, and thickness of 4.5 and 9.5 mm. Simulation results were history matched
with experimental data obtained from the second model. For each
experiment, the model was first packed uniformly with glass beads
of 0.22 mm average diameter and saturated with motor oil
(COSMO#1000TM) through the two 10-mm fittings located along
the upper and lower edges of the model. The transparent acrylic
resin walls enabled physical and thermal visualization of the
expanding steam chamber. The copper pins installed on the front
acrylic plate allowed monitoring of temperature distributions in the
model reservoir. Two horizontal wells were used, one for steam
injection and one for fluid production. The vertical separation
between them was 100 mm. The wells were covered with stainless
steel screen (100 mesh) to filter the glass beads. They were
designed to allow efficient steam injection and liquid production.
Figure 2 shows the experimental model. It was estimated that
experimental errors caused by incomplete breaking of emulsion
was + 3% and -3% by weight for water and oil, respectively.

44

Journal of Canadian Petroleum Technology

were similar to those used in the experiments, with most of the

experimental values being measured physically(5).
Relative Permeability
Different opinions exist in the literature on the nature of relative
permeability function. For instance, Dake(6) showed that the apparent relative permeabilities are, assuming segregated flow, linear
functions of saturation. On the other hand, relative permeabilities
could be defined as non-linear functions of saturation(7).
In this numerical simulation study, both linear and non-linear
two-phase relative permeability functions, with zero and non-zero
end-point saturations, were tested separately to determine the function(s) matching the experimental results most closely. Figure 4
shows the two-phase relative permeability specifications employed
in the numerical study.

Representation of Heat Loss

FIGURE 2: Reservoir physical model (300 300 mm, 4.5 or

9.5 mm in thickness).

A convective heat transfer model in the simulator was used to

estimate the experimental heat loss from the model. For the acrylic
resin sidewalls, a heat conductivity value of = 0.13 J/cmmin. C
was used, and the heat transfer was from the reservoir to the surroundings (air). The overall convective heat transfer coefficient
was calculated as = 0.041 J/cm2.min. C, and was input for each
of the grid blocks of the resin plates.

Representation of Wells

Numerical Modelling
Selection of Grid System

Two horizontal wells were assigned and the Discretized

Wellbore (DW) model was used to simulate effects of the transient
flow behaviour in the steam injector wellbore.

For the numerical investigation, a Cartesian grid system (x, y, z)

was employed, due to the rectangular shape of the experimental
reservoir model. The 3D configuration was selected to evaluate
experimental convective heat loss more accurately, with appropriate definition of the thermal characteristics of the acrylic resin
walls. Grid-size sensitivity study was conducted to determine the
appropriate vertical and horizontal lengths of grid blocks, in order
to model the vertical rise and sideways expansion of the steam
chamber more closely. This study led to a selection of 15 3 15
configuration. The net reservoir contained in the numerical model
extended along the range of (2 to 14, 2, 2 to 14), while the remaining blocks represented the acrylic resin walls. Figure 3 shows the
configuration of the grid blocks with their numbers and respective
sizes, and location of the wells.

Production Well
A horizontal producer was assigned into the grid blocks (8,1) to
(2,2), i.e., being horizontal in the z direction through two grid
blocks and within one block in each of x and y directions.
A starting operating bottomhole pressure (BHP) of 101.3 kPa
(atmospheric) and a minimum BHP of the same value were set.
This was to prevent escape of excessive steam at the time of breakthrough, as was done in the experiments.
Injection Well
A horizontal injector was assigned into the grid blocks (8,1) to
(2,6) in the same manner, except that it was positioned 100 mm
(i.e., L = 100 mm above the production well.

Reservoir Parameters
Rock and heavy oil properties, initial reservoir conditions, and
saturation endpoints are listed in Tables 1 4. These properties
TABLE 1: Properties of experimental reservoir model.
Porosity, fraction
Average Permeability, k
Thermal Conductivity,
Volumetric Heat Capacity
Resin Thermal Conductivity
Resin Volumetric Heat Capacity

0.38
1.42
0.70
1.99
0.13
1.67

105 mD
J/cmmin. C
J/cm3 C
J/cmmin. C
J/cm3 C

TABLE 3: Heavy oil viscosity [see Sasaki et al.(4)].

Temp.
C

Viscosity
cP

Temp.
C

Viscosity
cP

15
20
25
35
45
55

15,000
9,200
6,913
4,015
2,412
1,495

65
75
85
95
105
120

996
624
418
286
200
60

TABLE 4: Initial conditions and saturation endpoints.

TABLE 2: Heavy oil properties.
Density,
Molecular
Compressibility
Thermal Expansion Coefficient
Heat Capacity
Critical Pressure
Critical Temperature
Capillary Pressure
Phase Equilibrium Constant
January 2001, Volume 40, No. 1

g/cm3

0.998
490 g/gmole
7.0 10-7 kPa-1
6.0 10-4 C-1
411.7 J/gmole C
1,115 kPa
494 C
0.0 kPa
0.0

Temperature
Pressure
Oil Saturation fraction
Water Saturation fraction
Gas Saturation fraction
Reference Pressure
Reference Temperature
Sor fraction
Swc fraction
Sgc fraction

20.0 C
101.3 kPa
1.0
0.0
0.0
101.3 kPa
20 C
0.05
0.10
0.05
45

FIGURE 3: Grid block

configuration of the numerical
reservoir model.

In the experiments, the BHP was observed to be almost constant

at 121.6 kPa. But, due to the history matching, the operating constraint was set to be the steam injection rate (in terms of cold water
equivalent, CWE) which was then altered, based on the experimental data, at the beginning of each of the 10-minute periods.

Representation of Steam Injection and

Production Control
In the experimental study, steam was injected at a constant
gauge pressure of 20 kPa, corresponding to an absolute pressure of
121.3 kPa (with a quality of 100% at 105 C). To history match, the
injected steam quantity based on the experimental data was input
at each data point for the duration of the experiments.
In the experiments, steam was trapped at the production well, to
prevent its escape at the time of breakthrough, by a controlling
valve which was set to be automatically closed at a rate of about
90% when temperature of the production fluid reached 95 C. In
the numerical model, steam trapping was provided by an operating
constraint set by a differential temperature of 5 C (between the
steam saturation temperature corresponding to the well bottom
hole pressure and the temperature of the produced water). For this
condition, the flowing bottom hole pressure of the production well
was kept high enough that live steam does not appear in the well
block; this was very similar to that in the experiments.

Simulation Results and Discussion

Sensitivity Study
The effects of several parameters on the simulation results were

investigated, prior to history matching the experiments. Those

parameters included relative permeability functions and the
amount of heat loss.
Several numerical runs were conducted with different relative
permeability functions and endpoint saturations. It was noted that
both the shape of the rising steam chamber and cumulative oil production were affected by the relative permeability functions
employed. Figure 5 shows the simulation-generated steam chamber, where the cumulative oil productions were 44 and 35 cm3 for
linear relative permeability functions with zero and non-zero endpoint saturations [see Figure 4(a)], respectively. The steam chamber for a cumulative oil recovery of 30 cm3 was, for the relative
permeabilities shown in Figure 4(b), is shown in Figure 6. These
results disagree with the conclusion drawn by Chow and Butler(2).
This disagreement could possibly be attributed to the large difference in the average permeability values of the two studies (see
Introduction).
Another factor having a significant effect on the results was the
amount of heat loss from the reservoir model (outer grid blocks) to
the surrounding (air). Several numerical runs confirmed that
decreasing heat losses (both convective and conductive) from the
reservoir led to earlier initial communication between the wells
(i.e., shorter breakthrough time), and resulted in less steam injection for more oil production. The cumulative oil productions and
the steam-oil ratios for various amounts of heat loss from the reservoir to the surroundings at various times are compared in Figures
7 and 8, respectively.
As was carried out in the experimental studies, several other
numerical runs were conducted to see if varying several physical
conditions would show similar effects on the performance of the
SAGD process.

FIGURE 4: Twophase relative

permeability
functions, (a) linear
relative permeability
functions, (b) nonlinear relative
permeability
functions.
46

Journal of Canadian Petroleum Technology

FIGURE 5: Numerical simulation

temperature contours (at = 550 min.)
with linear relative permeability
functions as shown in Figure 4(a); (a)
with zero-end point saturations, (b)
with non-zero end point saturations.

Steam Injection Pressure

Figure 9 compares the numerical oil production rates for different injection pressures. It is seen that operating at higher injection
pressures raised the temperature of the steam chamber. This
allowed oil to drain more rapidly, and the communication between
the wells was established earlier. Hence, the spreading rate of the
steam chamber was increased, which consequently resulted in a
higher amount of cumulative oil production.
Injector Location
As the vertical spacing between the two horizontal wells
decreased, communication between the two wells was established
earlier. Figure 10 compares the effects of well spacing on the
breakthrough time in the experimental and numerical simulations.
There is good agreement between these results.
Model Thickness
Figures 11 and 12 show the experimental and numerical results
of effect of reservoir thickness on the process, respectively. In both
cases, oil production increased with increasing reservoir model
thickness (in the z direction). The amount of steam needed to produce a unit volume of oil was less for the thicker reservoir. Varying
the thickness had no effect on breakthrough time.
In light of the sensitivity studies mentioned above, and by
selecting appropriate values as input data for the simulator, the
experimental steam injection quantities are history matched at each
data point for the entire period of the experimental time. The following section presents the production performance results of the
numerical simulation runs.

FIGURE 7: Effect of heat loss on cumulative oil production

(STARS).
January 2001, Volume 40, No. 1

FIGURE 6: Numerical simulation temperature contours (at =

550 min.) with non-linear relative permeability functions as shown
in Figure 4(b).

History Matching Results

Matching Steam Injection
In order to history match, the steam injection quantity based on
the experimental data was input at each data point (at the beginning
of each 10-min. interval) for the entire period of the experiment.
Figure 13 shows the experimental cumulative steam injection at

FIGURE 8: Effect of heat loss on steam-oil ratio (STARSTM).

47

FIGURE 9: Effect of injection pressure on oil production rate

(STARSTM).

FIGURE 11: Effect of model thickness on experimental results

(dual-well SAGD).

FIGURE 10: Effect of well spacing on breakthrough time.

FIGURE 12: Effect of model thickness on numerical results (dualwell SAGD).

various times. The overall cumulative amount is 548 cm3 at the end
of the simulation time.
Oil Production
In Figure 14, the experimental cumulative oil production is
compared with that from the numerical simulation. At the end of
550 min., the simulation reported a cumulative oil recovery of 49
cm3. This value was less than the experimental volume of 64 cm3.
The slopes of the cumulative oil production curves, however,
match satisfactorily. As seen in Figure 14, the numerical cumulative oil production curve extends ahead of the experimental one,
between 40 and 180 min. This is believed to have been caused by
somewhat incomplete representation of transient flow behaviour in
the numerical simulator, which resulted in earlier production. On
the other hand, adjustments in relative permeabilities could be
needed. After 180 min., the numerical production curve trails
behind the experimental curve. It was during this period that the
maximum degree of emulsification was found and the oil was
mostly produced as an emulsion (water in oil) in the experimental
studies.
Figure 15 compares the numerical oil production rate with the
experimental rate. There is good agreement between them, except
for more noticeable fluctuations in the experimental production
rate curve. These fluctuations were due to the fact that the experimental production rates were plotted at the end of each 10 min.
interval. The numerical oil production rate increases with time and
attains a maximum rate at 160 min., which is the time just before
steam breakthrough. Steam breakthrough was noted at 170 min. in
the experiment. Then, a steady decrease in the oil production rate
is observed because of reservoir depletion.
48

Water Production
The experimental and numerical cumulative water productions
are compared in Figure 16. In the first 30 min. there were no cumulative water productions. This was due to the transient flow behaviour in the injection well. In addition, heating the reservoir, including the rock and fluid, meant that oil production is delayed until the
oil is sufficiently mobilized. Hence, water production is also
delayed. The larger water production value by the numerical simulation at 550 min. is believed to be caused by the relative permeability functions employed in the simulation runs.
Steam Chamber and Temperature Contours
Figure 17 compares the numerical and experimental steam
chambers, with temperature contours at the end time, = 550 min.
Clearly, satisfactory agreement was obtained between the shapes
of the experimental and the numerical steam chambers, for the relative permeability functions used in the numerical simulation. It
was noted that the numerical simulator can model the vertical rise
and sideways and upward expansion of the steam chamber. Both
experimental and numerical steam chambers were observed to be
growing at similar rates from the beginning to the end of the recovery process.

Conclusions
In this study, CMGs STARSTM simulator was used to simulate
experiments of the SAGD process. The results from the numerical
simulation were found to be mostly in good agreement with those
from the experiments, including oil production rate, cumulative oil
Journal of Canadian Petroleum Technology

FIGURE 13: Experimental cumulative steam injection vs. time

(dual-well SAGD).

FIGURE 15: Comparison of oil production rate vs. time (dual-well

SAGD).

FIGURE 14: Comparison of cumulative oil production vs. time

(dual-well SAGD).

FIGURE 16: Comparison of cumulative water production vs. time

(dual-well SAGD).

production, steam chamber size and shape, and temperature contours (i.e., temperature distribution in the reservoir). Varying physical conditions (steam injection pressure, vertical separation
between injection and production wells, and reservoir thickness)
were found to have similar effects on the performance of the
SAGD process both in the experimental and numerical studies.
Specifically, the following conclusions were drawn:
1. The numerical simulation study using CMGs STARSTM provided relatively good results for the history matching of
experimental data for the SAGD process obtained from the
scaled reservoir model.
2. Phenomena associated with the rise of the steam chamber,

together with operational constraints such as the steamtrapping mechanism at breakthrough time and heat losses
could be modelled reasonably well.
3. The numerical simulations using linear relative permeability
functions with non-zero end-point saturations provided good
agreement between the steam chamber shape and experimental observations, while the linear functions as well as
non-linear functions did not. This sensitivity of the steam
chamber shape to the shape of the relative permeability functions employed for this particular model appears to be in disagreement with the conclusion drawn by Chow and Butler(2).

FIGURE 17: Comparison of steam chambers at = 550 min.; (a) numerical result, (b) experiment.
January 2001, Volume 40, No. 1

49

Acknowledgements
This study has been supported by the Technology Research
Centre (TRC) of Japan National Oil Company (JNOC). We would
like to thank Mr. S. Demir and Mr. Yamazaki for their enthusiastic
assistance, Mr. K. Ohno and Dr. H.K. Sarma of JNOC for their
helpful advice; and Mr. W.L. Buchanan and Ms. V. Oballa of CMG
for their kind support.

NOMENCLATURE
H
k
Kr
L
t
T

Pi

=
=
=
=
=
=
=
=
=
=

reservoir height, mm
permeability, D (Darcy)
relative permeability
vertical spacing between two wells, mm
reservoir model thickness, mm
temperature
convective heat transfer coefficient, J/min. C.cm2
pressure difference between two wells, kPa
heat conductivity, J/cm.min. C
elapsed time from start of steam injection, min.

Subscripts
g
l
o
S
w

=
=
=
=
=

gas (steam)
liquid
oil
saturation
water

SI Metric Conversion Factor

cP 1.0 E-03 = Pas
D 1.01 E+12 = m2

REFERENCES
1. BUTLER, R.M., Thermal Recovery of Oil and Bitumen; PrenticeHall Inc., New Jersey, xiii, 7-10, pp. 287-290, 1991.
2. CHOW, L. and BUTLER, R.M., Numerical Simulation of the Steam
Assisted Gravity Drainage Process; Journal of Canadian Petroleum
Technology, Volume 35, No. 6, p. 55, June 1996.
3. CHUNG, K.H. and BUTLER, R.M., Geometrical Effect of Steam
Injection on the Formation of Emulsions in the Steam Assisted
Gravity Drainage Process; Journal of Canadian Petroleum
Technology, Vol. 27, No. 1, January February 1988.
4. SASAKI, K., AKIBAYASHI, S., KOSUKEGAWA, H., KATO, M.,
and ONO, K., Experimental Study on Initial Stage of SAGD Process
Using Two-Dimensional Scaled Model for Heavy Oil Recovery; SPE
37089, Proceedings of the Petroleum Society/SPE (2nd Three Day
International Conference on Horizontal Well Technology, Calgary,
AB, November 18 20, 1996.
5. SASAKI, K., AKIBAYASHI, S., KATO, M., and ONO, K., A New
Concept of Enhanced SAGD Process by Adding Intermittent SteamStimulation on Lower Horizontal Production-Well (SAGD-ISSLW);
Proceedings of 15th World Petroleum Congress, Beijing, China,
Vol. 2, John Wiley & Sons, pp. 511-513, 1998.
6. DAKE, L.P., Fundamentals of Reservoir Engineering; Elsevier
Scientific Publishing Co., New York, 378, 1978.
7. HONARPOUR, M., KOEDERITZ, L., and HARVEY, A., Relative
Permeability of Petroleum Reservoirs; CRC Press Inc., Boca Raton,
FL, pp. 45-48, 1986.

ProvenanceOriginal Petroleum Society manuscript, Numerical

and Experimental Modelling of the Steam Assisted Gravity
Drainage (SAGD) Process (99-21), first presented at the 50th
Annual Technical Meeting, June 14 18, 1999, in Calgary,
Alberta. Abstract submitted for review December 16, 1998; editorial comments sent to the authors June 5, 2000; revised manuscript
received June 29, 2000; paper approved for pre-press November 3,
2000; final approval January 4, 2001.

50

Authors Biographies
Kyuro Sasaki is an associate professor of
Earth Science and Technology at Akita
University. His research interests are EOR,
fluids mechanics, and heat and mass transfer phenomena in mineral engineering. He
holds B.S., M.S. and Ph.D. degrees from
Hokkaido University, Japan, and is a member of the Petroleum Society.

Satoshi Akibayashi is a professor of Earth

Science and Technology at Akita
University. His principal interests are EOR
and reservoir simulation. He holds a B.S.
degree in mining engineering from Akita
University and Ph.D. degree in geothermal
engineering from Kyushu University.
Satoshi Akibayashi has been a vice president of the Geothermal Research Society of
Japan since 1998 and served on the 1994
1997 Board of Directors of the Japanese
Associations for Petroleum Technology.
Nintoku Yazawa is the director of the
Reservoir and Recovery Laboratory at the
Technology Research Centre of Japan
National Oil Corporation. His research
interests are gas injection process, MEOR,
and heavy oil recovery. Mr. Yazawa holds
B.E. and M.S. degrees from Waseda
University, all in petroleum engineering.

Quang Doan is presently an assistant professor of petroleum engineering in the

School of Mining and Petroleum
Engineering at the University of Alberta.
His research activities are focused in the
areas of heavy oil recovery, multiphase flow
transport, and enhanced oil recovery. He
obtained B.Sc., M.Sc., and Ph.D. degrees in
petroleum engineering from the University
of Alberta. He is a member of the Petroleum
Society, Society of Petroleum Engineers,
and is a registered professional engineer with APEGGA.
S.M. Farouq Ali is general manager of
PERL: Petroleum Engineering and
Research Laboratories Canada Ltd. and
H.O.R. Heavy Oil Recovery Technologies
Ltd. He holds B.Sc. (Hons.), M.S. and
Ph.D. degrees in petroleum engineering. He
was professor of petroleum engineering for
38 years, in Alberta and at the Pennsylvania
State University, USA. He specializes in
reservoir engineering, oil recovery and simulation. He has authored over 500 papers,
supervised 200 graduate theses, and has done more than 200 petroleum reservoir studies. He has received several awards from the
Society of Petroleum Engineers, including the Lester C. Uren
award. Recently, he was awarded honorary doctorates by two leading universities in Russia. This past June, Dr. Farouq Ali was presented with the Petroleum Societys Distinguished Service Award.

Journal of Canadian Petroleum Technology