CANADIAN HEAVY

OIL ASSOCIATION

SPE/PS-CIM/CHOA 97719
PS2005-320
New Developments in Streamflood Modeling
M. Kumar, SPE, C. Satik, SPE, and V. Hoang, SPE, Chevron Energy Technology Co.

Copyright 2005, SPE/PS-CIM/CHOA International Thermal Operations and Heavy Oil Symposium
This paper was prepared for presentation at the 2005 SPE International Thermal Operations
and Heavy Oil Symposium held in Calgary, Alberta, Canada, 13 November 2005.
This paper was selected for presentation by an SPE/PS-CIM/CHOA 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,
Petroleum SocietyCanadian Institute of Mining, Metallurgy & Petroleum, or the Canadian
Heavy Oil Association and are subject to correction by the author(s). The material, as
presented, does not necessarily reflect any position of the SPE/PS-CIM/CHOA, its officers, or
members. Papers presented at SPE and PS-CIM/CHOA meetings are subject to publication
review by Editorial Committees of the SPE and PS-CIM/CHOA. Electronic reproduction,
distribution, or storage of any part of this paper for commercial purposes without the written
consent of the SPE or PS-CIM/CHOA 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
Using recent results from fine-scale, multi-pattern,
geostatistical models of the Kern River field, California, this
paper reviews key issues related to steamflood modeling and
shows that fine-grid models depict the near-vertical steam
override, and corroborates that heavy oil steamflooding is not
a displacement process; rather the oil drains by gravity.
Further, models with unconfined boundaries result in steam
zone pressures similar to those observed in field. Including
the common operating practice of cyclic steaming of
producers at early time reduces pressures and accelerates
steam breakthrough time and recovery.
Furthermore, pattern element and single sand models
used in many previous studies are not sufficient to explain
observed field performance, and that larger, heterogeneous
model give more realistic recovery predictions. Discontinuous
shales allow significant drainage to occur from the upper to
the lower sands. Consequently, the upper zones may contain
less reserve than expected and the lower zones can give
apparent high recovery. Use of parallel models showed
significant speed up over serial models allowing significantly
larger models to be run in a reasonable time. Apparent higher
speed up is gained for larger models.
The paper demonstrates that the current
improvements make larger-scale modeling of steamflood
projects viable compared with what was possible earlier and
that a realistic forecast of steamflood performance is attained
when the necessary details are included in the model.

Introduction
Steam injection is the most widely applied enhanced oil
recovery (EOR) method.1-6 Current oil production by steam
injection is estimated to be over 1.1 million BOPD.1 Most
conventional heavy oil steamflooding projects in California,
Canada, Indonesia and Venezuela employ vertical wells;
although, the use of horizontal producers is growing.5,6 On the
other hand, extra heavy oils may require both horizontal
injectors and producers, such as, steam assisted gravity
drainage (SAGD).5 Oil recovery can exceed 20% of the
original oil in place (OOIP) for cyclic steaming and over 50%
OOIP by continuous steam injection (for small well
spacing).2,4-6
Early steamflood performance prediction methods
used analytical and semi-analytical models,7-8 that did not
account for gravity override of steam. Later, Neuman9
developed an analytical gravity override model for steamdrive.
The analytical models are heat flow and energy balance
models with an assumed shape of the steam zone and uniform
properties; therefore, they can not account for the effects of
variation in geology, fluid property, and operating conditions.
Scaled-physical models were used as improvement over the
early analytical models.2,10,11 Although, the scaled models
portray most mechanisms accurately, they are time consuming
and cumbersome. Further, they may be limited by availability
of materials and fluids to achieve proper scaling of a particular
reservoir and oil.2. Consequently, they are seldom used now
as a forecasting tool.
Steamflood modeling using numerical simulation has
been done since the 1970s to interpret performance, evaluate
recovery methods and optimize field development, and it is
the method of choice for performance prediction.12-19
Numerical models have been able to duplicate the results of
physical models quite well.20,21 Compared with black oil
models, steamflood modeling requires smaller grid sizes (to
capture process physics) and large computational times,
because the energy equation has to be solved simultaneously
with the material balance equation, transport equation and the
equation of state.12-14 As a result, we can practically model
only a segment of a reservoir or project. Further, early
thermal simulation models15-18 were small pattern-element
models because of computing hardware and computation time
limitations. Furthermore, to describe observed field behavior,
such models typically required history matching

adjustments to relative permeability, transmissivity, and rock

compressibility.22
To optimize capital requirements and facility usage,
most steamflood projects are developed in phases, resulting in
unconfined boundaries. This is not captured in confined,
pattern-element models. Additionally, most of the previous
models considered a single sand zone. Multiple zones were
handled as layered models, with uniform (average) properties
in a layer.22. In reality, however, intervening shales maybe
discontinuous, resulting in some vertical flow between zones.
Because of the above limitations, both real and perceived, the
use of simulation in steamflood reservoir management is less
prevalent.
Recent steamflood flow simulations have shown the
importance of capturing detailed heterogeneity23,24, and
modeling of multi-pattern areas.25 New advancements in
computer hardware and reservoir simulation technologies have
enabled the use of larger and more representative
heterogeneous reservoir models. Further, the use of parallel
computing26 and dynamic gridding27 have enabled significant
improvement in thermal simulation capabilities.

SPE/PS-CIM/CHOA 97719

The Kern River reservoir is composed of a sequence

of several alternating sand and shale members. Figure 2
shows a schematic of the reservoir zones and compares the
measured core permeability and transform used in the model
at a well. The individual sand bodies, which are typically 50100 ft thick and are separated by competent and correlatable
shale layers, are steamflooded one at a time. However, some
of the shales may not be continuous over the entire project
area. In addition, shale continuity varies both areally and
vertically.31. As a result, significant fluid migration can occur
between sands, making reservoir management and analysis
more challenging. The reservoir dips gently at about 3-4
from northeast to southwest. Typical reservoir depth is 300 to
850 ft.
The Kern River field properties of low reservoir
pressure, high permeability and high oil saturations are all
favorable for steamflooding. The reservoir is generally
processed from the bottom up. That is, steam injection is
initiated in the lowest zone and the remaining upper reservoir
zones are sequentially processed.

Input Parameters and Flow Simulation Model

Using recent results from fine-scale, multi-pattern,
geostatistical models of the Kern River field,25,28 California,
this paper reviews key issues related to steamflood modeling
including: (1) the level of detail needed to capture process
physics, (2) methods to properly represent the influence of
surrounding patterns, (3) effect of including cyclic steaming of
producers, (4) impact of discontinuous shales in multi-sand
reservoirs on recovery from individual zones, and (5) methods
to speed up computation.
The paper demonstrates that the current
improvements make larger-scale modeling of steamflood
projects viable compared with what was possible earlier and
that a realistic forecast of steamflood performance is attained
when the necessary details are included in the model.

Reservoir Geology and Geologic Models

The geologic model used in this study is from portions of the
Kern River field (see Fig. 1), which is a shallow heavy oil
field located five miles northeast of Bakersfield, California. It
is one of the largest oil fields in the United States based on oilin-place and reserves. The field has been on steam injection
since the mid 1960s.29-30 The Kern River field comprises of
several different leases, and steam injection projects were
started at different times in various leases.
The Kern River field produces from several distinct
zones. These sands were deposited in a braided river
environment, and have high permeabilities and porosities.
The sandstones are typically medium to very coarse-grained,
poorly to very poorly sorted, and have little to no detriatal
clay. The high-quality reservoir sandstones are interbedded
with poor quality sand-stones, siltstones, and mudstones,
which may be barriers to fluid flow.

Input Parameters. Table 1 lists the important reservoir and

fluid properties used in the model. An average sand porosity
of 32% was used in this study. Core data showed that average
sand porosity varied in a narrow range.
Horizontal
permeability distributions were obtained using geostatistical
techniques. The average permeability was about 2500 md.
The vertical permeability was considered to be one-half of the
horizontal permeability.
The initial reservoir temperature was 90F. The
initial reservoir pressure was 260 psia at a subsea depth of 565
ft. However, because of depletion, the top of many of the
zones currently have a gas cap at close to the atmospheric
pressure. The initial oil saturation was about 50% and the
initial gas saturation was 0% in the oil zone. Reservoir (pore
volume) compressibility was 5010-6 psi-1; this is within the
range of our recent measurements on unconsolidated cores.
The crude gravity was 14API and its molecular
weight was 400. This heavy oil was represented by two
pseudo-components: methane and a nondistillable component.
Table 2 lists dead crude oil viscosities as a function of
temperature. The steam quality at the sandface was 70%.
Measured relative permeability values were used.25
Endpoint saturations and relative permeabilities were
considered to be independent of temperature.32 Three-phase
oil relative permeabilities were calculated using the linear
(straight line) interpolation model33 because it predicts the
steamflood residual oil saturation more accurately.21

Flow Simulation Model.

Two model segments were
extracted from the detailed geologic model. The first model

SPE/PS-CIM/CHOA 97719

contained a single sand zone, isolated by continuous shales

above and below the zone of interest.28. The sand was 60-ft
thick and contained a 6-ft thick gas zone at the top, which
represents current conditions and is caused by depletion from
the zone before initiation of steam injection. The second
model comprised of all sand zones and was used to determine
the effect of discontinuous shales on oil recovery factor.
The average pattern size in Kern River is about 2.6
acres. Several different grid sizes were used in the study. The
fine-scale geologic model was scaled up to the desired level.
The typical scaled up models had areal grid size of 20-40 ft
and vertical size of about five feet. The injection wells were
open to the bottom third of each injection zone or sand
interval. The production wells were open to the entire sand
interval.
Chevron's scalable, parallel, multi-purpose reservoir
simulator, CHEARS,26 was used in this study. The thermal
option of CHEARS is a three-phase, three-dimensional, fully
implicit, compositional simulator. The simulator accounts for
and rigorously models important physical processes taking
place during steamflooding.

Results and Discussion

Process Physics and Steam Override. As noted earlier, a
single zone model was used to evaluate the level of detail
needed to capture process physics. Figure 3 shows the
permeability distribution for the single zone model segment
consisting of 25 inverted five-spot patterns. The distance
between the injector and the producer was 233 ft. Steam
injection rate was 298 BSPD, cold water equivalent, and the
steam quality was 70%. This model was used to evaluate the
level of detail or grid sizes (vertical and areal) needed to
capture process physics properly. Grid sizes for this multipattern model were varied from 10-85 ft. in the areal and from
2-6 ft in the vertical direction. The base case model shown in
Fig. 3 had areal and vertical dimensions of 20 ft and 3 ft,
respectively.

after the steam breakthrough to the producer, the subsequent

steam zone growth is predominantly vertically downward with
a near horizontal steam/oil interface as the heated oil drains to
the producer by gravity.
These results clearly show that the steamflooding for
heavy oils is primarily a gravity drainage process and not a
displacement process. Therefore, the grid sizes in the
numerical models should be sufficiently small to accurately
capture process physics. Figure 6 shows that for California
heavy oils, the 3-D saturation distributions start to deteriorate
for areal grid sizes grater than ~40 ft (and vertical grid size
greater than ~ 5 ft). Clearly, grid-size sensitivity should be
conducted for any new thermal simulation studies and a
careful consideration of study objectives is needed for proper
grid size selection.

Effect of Surrounding (Undeveloped) Patterns. As noted

earlier, most steamflood projects are developed in phases,
resulting in unconfined boundaries. The effects of
unconfinement are not captured in the confined, patternelement models. Here unconfinement is defined as the
additional undeveloped reservoir volume, around the actual
project or model area, which is not under steam injection.
Figure 7 compares injector pressure at center pattern of the
model (see Fig. 3), for various level of unconfinement. The
undeveloped area surrounding the pattern has a pore volume
equal to 20% of the project pore volume or higher.
Figure 7 clearly shows that a confined model with
no-flow boundaries or modeling a small pattern element done
in the past, assuming that the entire field is developed
simultaneously, results in unusually high injection pressure.
On the other hand, lower peak pressures are observed when
models are unconfined. These results suggest that higher
pressures observed in some of the prior pattern-element
models were caused by the unconfined models or the inability
to properly include the effects of boundary conditions. Also,
note that unconfined models better represented the typical
phased development of projects.

Figure 4 shows steam (mole fraction) distribution in a

vertical cross-section through the middle of the model at
various times until the steam breakthrough to the producer.
Note that the vertical scale in Fig. 4 is exaggerated by about
four times to enhance the vertical detail. It shows a near
vertical rise of the injected steam until it reaches the top of the
formation. For the conditions of this model, the injected
steam reached the top of the formation in about 12 days.
Subsequently, the stem zone moves laterally toward the
producer. The steam breakthrough to the producers occurs in
about one year. Note that the near vertical steam override and
the shape of the steam zone is typical of all steam injection
processes.

Effect of Cyclic Steaming of Producers. The cyclic steam

stimulation (CSS) of production wells is a common field
practice in most California and other steamflood operations.
The CSS is typically performed during the early phase of a
steamflood operation, prior to steam breakthrough to the
producers, to heat the reservoir around the producers and to
establish quicker thermal communication between injectors
and producers. In the results presented here CSS was
considered only during the first year of steamflooding. A total
of three cycles were included, with each having one week of
injection, one week of soak and 3.5 months of production
periods.

Figure 5 shows the growth of the steam zone (or gas

saturation distribution) at different times after steam
breakthrough to the producer. For illustration purposes, only a
quarter of the central pattern is shown in Fig. 5. It shows that

Figure 8 shows that the peak pressures reduced even

further, when the effect of CSS is included in the model. Note
that the peak pressures are similar to those in the field, when
the effects of cyclic steaming and unconfinement both are

included in the model. Further, previous spongy rock or

high rock-compressibility requirements in some patternelement simulations were modeling artifacts.
The models results show that CSS also causes earlier
steam breakthrough to the producers, as desired. The CSS
resulted in acceleration of the oil recovery process but it did
not affect the ultimate recovery (see Ref. 28). Cyclic steaming
is often not included in steamflood modeling because it
increases computation time substantially. These results show
that if cumulative oil recovery is the primary parameter of
interest, then not including the details of early CSS is quite
acceptable.

Effect of Shale Breaks. As noted earlier, the Kern River

reservoir is typically processed from bottom up. However,
because the shales may not be continuous over a project area,
significant fluid drainage can occur between sands. Therefore,
proper modeling of these shales is necessary to capture fluid
movement accurately.
Figure 9 shows the shale regions in the second
model25, which is comprised of all sand zones and consists of
18 steamflood patterns. A lateral log cutoff was used to define
shale; this was consistent with core data, and resulted in 22%
volumetric population of shale in the entire interval. Shales
were assigned a permeability of 1 md. It is evident that many
of the shale layers are not continuous and that they have a
Swiss-cheese appearance. Several holes in the shale layers
can be seen clearly seen. Note further, that the discontinuity
can occur at a different areal location for different shales.
Also, some of the shales are more continuous over the model
area. These shale discontinuities form conduits for fluid flow
between zones and allow steam to flow up and heated oil to
drain down to the lower zones.
Figure 10 shows that the field measured historical oil
production compares well with the calculated results for this
model comprising all zones. It shows excellent agreement
between the calculated recovery and field data from 1984
onward; however, the model underpredicts oil recovery before
1984. The lower recovery in the model in the early years was
attributed to not including the effect of cyclic steaming of the
producers. As discussed in the previous section, cyclic
steaming results in acceleration of oil production, but has little
impact on total recovery. Figure 11 shows that calculated and
measured temperature profiles at observation well 610-T1
compare well also.
Pattern Element vs. Large Heterogeneous Models.
To quantify the differences between a small, pattern-element,
layered model vs. the multi-pattern, detailed, heterogeneous
model, a smaller, quarter of a five-spot model was
constructed. The properties of the layered model were
obtained by averaging the values for each layer of the detailed
heterogeneous model, and the shale layers were assumed to be
continuous, as had been done in earlier studies.

SPE/PS-CIM/CHOA 97719

Figure 10 also shows that the calculated oil recovery

using a quarter-pattern, layered model, is higher than the
multi-pattern, heterogeneous model or the field data. In
addition, the oil recovery is delayed and the oil production
peak is sharper. The multi-pattern heterogeneous model
allows breakthrough to occur more quickly and at different
times for different producers. Furthermore, the volumetric
sweep or recovery at the end of the project is lower for the
heterogeneous model. These results suggest that layered,
small pattern-element models are inadequate to explain
observed field behavior in this type of reservoir.
Oil Recovery from Individual Zones. Figure 12
shows the oil recovery from seven individual zones of the
model. It shows that, when the shales are discontinuous, the
lower sands can have significantly higher recovery than the
average, whereas the upper sands have significantly lower
recovery. Note that, oil recovery decreases sequentially from
71% to 52% as one moves up from the lowest R1 sand to the
K1 sand. As steam injection is expanded vertically upwards,
discontinuous shales allow heated oil from upper sand to drain
into the lower sand that was already processed by steam. This
oil migration continues until project termination, long after
steam injection into the lower sand has stopped.
On the contrary (see Fig. 12), oil recovery from the K
sand is higher than K1, even though K is above the K1. This
is because the shale above the K1 sand is quite continuous
over the model area, and it prevents oil drainage from sands
above. However, shales between C1, G, and K sands are not
continuous; therefore, we observe the same decreasing oil
recovery in sands above the K sand.
Figure 13 more clearly illustrates the effect of
continuous vs. discontinuous shale above a sand zone. It
compares the variation in average oil saturation in the R1 and
K1 sands during the entire project duration. Note that the
shale above the R1 sand has holes in it, where as shale above
the K1 sand is continuous over the model area. The steam
injection period for each sand is also marked on the figure.
A rapid initial decrease in R1 sand oil saturation is
caused by oil production during steam injection into that zone.
After completion of steam injection, the heated sand continues
to produce oil; therefore, oil saturation decreases further.
However, as the successive upper sands are steamed, some of
the heated oil continues to drain to the lowest, R1 sand. Some
of the oil draining into the R1 zone at late times is not
produced (because of the low gravity head of the zone); as a
result, the average saturation in the R1 zone increases. At
project termination, the saturation in the R1 zone had
increased by 8% from its lowest point.
On the other hand, average oil saturation in the K1
zone continually decreases during the entire project duration.
A decrease in the oil saturation prior to the initiation of steam
injection into the zone is caused by hot-plate heating by steam
injection into the K2 sand immediately below.
The
conductive and convective heating through the leaky shale
causes the oil in K1 zone to become mobile, drain into the

SPE/PS-CIM/CHOA 97719

zone below, and get produced. The decrease in oil saturation

during the steam injection into the K1 sand (Fig. 13), is similar
to R1 sand. Because the shale above the K1 is continuous,
K1 sand only experiences a net oil outflow, and the oil
saturation continues to decrease.
Table 3 compares field-observed oil recovery at the
end of steam injection into a zone vs. oil recovery at project
termination for each zone. Note that oil recovery at the end of
steam injection includes any production from upper or lower
zones (because producers are open to all zones) and is
attributed to the zone being steamed during that period. Table
3 shows that the actual production allocation can be quite
complex and that sand underlain by continuous shale tend to
produce more over the project life. Table 3 also shows that
the variation in oil recovery from various zones at the end of
project termination is a lot larger than that at the end of steam
injection into a zone.
These results clearly demonstrate that when the
shales are discontinuous, the oil recovery from lower zones is
higher. Consequently, recovery factors based on lower sands
alone will result in optimistic reserves estimate. Conversely,
zones above discontinuous shales have lower than anticipated
recovery and reserves (because of oil drainage to the lower
zones).
Methods to Speed up Computation. As noted earlier,
various techniques, including more powerful hardware,
numerical solver improvement, parallel computing and
dynamic gridding have been employed to enhance thermal
simulations. In the past few years, parallel computing has
become an efficient tool to reduce computation time and to
run thermal problems that were earlier considered too large
and impractical to run.
We evaluated the performance improvement of
thermal simulations using Chevrons CHEARS parallel code
running on a Linux cluster. We used a half million-cell, single
sand model, with 25 inverted five-spot patterns as a base
simulation model in the evaluation. Results show that parallel
computing can duplicate the solutions that of a single CPU for
both field wide and individual wells and reduce the run time
considerably. Figure 14 compares relative speed using a
single CPU and 16 CPUs for a 560,000-cell model. The
scalability of the parallel process was fairly efficient. The 16CPU run increased the speed of the simulation 8 folds (Fig.
14). In general, we found that parallel computing provided
greater improvement for larger models as shown in Fig. 15.
We expect increased utilizations of parallel
computing for field scale thermal simulations with further
simulator advancement. Darche, et al.27 have recently shown
that using dynamic gridding and parallel computing they were
able to achieve speed up by a factor of 10 for a thermal
problem.

Conclusions
Current improvements in thermal simulation make larger-scale
modeling of steamflood projects viable compared with what
was possible earlier and a realistic forecast of steamflood
performance is attained when the necessary details are
included in the model. The following conclusions are based
on detailed, multi-pattern, multi-zone steamflood simulation
studies using actual reservoir properties.
1.

Finely-gridded steamflood models capture near vertical

steam override and oil drainage by gravity with a near
horizontal steam/oil interface accurately.

2.

Models with unconfined boundaries result in steam zone

pressures similar to those observed in the field.

3.

Cyclic steaming of producers at early time reduces

pressures and accelerates steam breakthrough time and
recovery.

4.

Detailed, heterogeneous, models provide a more realistic

description of reservoir geology and project performance.

5.

Calculated oil recovery and temperature profiles using a

multi-zone, multi-pattern model compare well with field
data.

6.

Discontinuous shales allow significant oil to drain from

upper to lower sands and fluid migration across zones.
Consequently, the upper zones may contain less reserve
than expected and the lower zones may give apparent
high recovery.

7.

Small pattern-element or single-sand models are

inadequate to explain observed field behavior in this type
of reservoir.

8.

Parallel thermal models significantly reduce computation

times, making larger-scale steamflood models possible.

Acknowledgements
The authors would like to thank Anil Ambashta, Mike
Basham, Cedric Cease, Wen Chen, Sam DeFrancisco, Bill
Fong, Lee Williams, and other colleagues for their valuable
contributions to this study.

References
1.

Worldwide EOR Survey, Oil & Gas Journal, April 12,

2004, pp. 53-65

2.

Matthews, C.S: Steamflooding, SPE Distinguished

Author Series, JPT (March 1983) 465-471.

SPE/PS-CIM/CHOA 97719

3.

Blevins, T. R.: "Steamflooding in the U.S.: A Status

Report," JPT (May 1990) 548-54.

4.

Boberg, T. C.: Thermal Methods of Oil Recovery, John

Wiley and Sons, New York (1988).

19. Kumar, M. and Ziegler, V.M.: Injection Schedule and

Production Strategies for Optimizing Steamflood
Performance, SPERE (May 1993) 101-107.
20. Myhill, N.A.: A Check on Numerical Thermal
Simulation. Paper SPE 8822 presented at the 1980
Enhanced Oil Recovery Symposium, Tulsa, April 20-23.

5.

Butler, R.: Thermal Recovery of Oil and Bitumen,

Prentice Hall, Englewood Cliffs, New Jersey (1991).

6.

Hong, K.C.: Steamflood Reservoir Management,

PennWell Books, Tulsa, Oklahoma (1994)

7.

Marx, J.W. and Langeheim, K.H.: Reservoir Heating by

Hot Fluid Injection, Trans., AIME (1959) 216, 312-15.

21. Kumar, M. and Do, T.N.: Numerical Simulation of

Scaled-Model Steamflooding Experiments in Tar Sands
With Bottomwater Zone, paper CIM/SPE 90-96
presented at the Petroleum Society of CIM/SPE
International Technical Meeting, Calgary, June 10-13,
1990.

8.

Myhill, N.A. and Stegemeier, G.L.: Steam Drive

Correlation and Prediction, J. Pet. Tech. (Feb. 1978)
173-82.

22. Johnson, R.S., Chu, C., Mims, D.S., and Haney, K.L.:
History Matching of High and Low Quality Steamfloods
in Kern River Field, CA, SPERE (Feb. 1992) 75-82.

9.

Neuman, C.H.: A Gravity Override Model of Steam

drive, Journal of Petroleum Technology (Jan 1985) pp.
163-169.

23. Kumar, M.: Interwell Heterogeneity Representation and

Its Effect on Steamflood Performance, paper SPE 24932
presented at the 67th ATCE of SPE held in Washington,
D.C., October 4-7, 1992

10. Baker, P.E.: Effect of Pressure and Rate on Steam Zone

Development in Steamflooding, Society Of Petroleum
Engineers Journal, (Oct., 1973) pp. 274-284.
11. Stegemeier, G.L., Laumbach, D.D., and Volek, C. W.:
Representing Steam Processes with Vacuum Models,
Soc. Pet. Eng. J. (June 1980) 151-74.
12. Coats, K.H.: A Highly Implicit Steamflood Model,
Soc. Pet. Eng. J. (Oct. 1978) 369-83.
13. Grabowski, J.W., et al: A Fully Implicit General Purpose
Finite-Difference Thermal Model for In-situ Combustion
and Steam, SPE 8396 presented at the 1975 ATCE Las
Vegas, Nevada, Sept. 23-25.
14. Abou-Kassem, J.H. and Aziz, K.: Sensitivity of
Steamflood Model Results to Grid and Timestep Sizes,
Soc. Pet. Eng. J. (Feb. 1984) 65-74.
15. Chu, C. and Trimble, A.E.: Numerical Simulation of
Steam Displacement - Field Performance Applications,
J. Pet. Tech. (June 1975) 765-776.
16. Gomaa, E.E., Duerksen, J.H. and Woo, P.T.: Designing a
Steamflood Pilot in the Thick Monarch Sand of the
Midway-Sunset Field, SPE 5853, Journal of Petroleum
Technology (Dec. 1977) pp. 1559-1569.
17. Hong, K.C. and Ault, J.W.: Effects of Noncondensable
Gas Injection on Oil Recovery by Steamflooding, J. Pet.
Tech. (Dec. 1984), pp. 2160-2170.
18. Ziegler, V.M.: A Comparison of Steamflood Strategies:
Five-Spot Pattern vs. Inverted Nine-Spot Pattern, SPERE
(Nov. 1987) 549-58.

24. Dehghani, K., Basham, W.M., Durlofsky, L.J., Tucker,

K.E.: Modeling and Scaleup of Steamflood in a
Heterogeneous Reservoir, SPERE (Nov. 1995) 237-246.
25. Williams, L.L., Fong, W.S. and Kumar, M.: Effects of
Discontinuous Shales on Multizone Steamflood
Performance in the Kern River Field, SPE Reservoir
Evaluation and Engineering (Oct. 2001) pp. 350-357.
26. Chien, Mark C.H., Tchelepi, Hamdi A., Yardumian,
Hrant E., and Chen, Wen H.: A Scalable Parallel MultiPurpose Reservoir Simulator, paper SPE 37976
presented at the 1997 Reservoir Simulation Symposium,
Dallas, Texas, June 8-11.
27. Darche, J.E., Grabenstetter, P.H., and Sammon, G.: The
Use of Parallel Processing with Dynamic Gridding,
Paper SPE 93023 presented at the 2005 Reservoir
Simulation Symposium, The Woodlands, Texas, Jan. 31Feb 2.
28. Satik, C., Kumar, M., DeFrancisco, S., Hoang, V., and
Basham, M.: Important Modeling Parameters for
Predicting Steamflood Performance, paper SPE 90713
presented at the 2004 ATCE, Houston, Texas, Sept. 2629.
29. Bursell, C. G. and Pittman, G. M.: Performance of
Steam Displacement in the Kern River Field, JPT (Aug.
1975) 997-1005.
30. Blevins, T.R. and Billingley, R.H.: The Ten-Pattern
Steamflood, Kern River Field, California, JPT (Dec.
1975) 1505-14.

SPE/PS-CIM/CHOA 97719

31. Vogt, J.P.: Comparison of Geostatistical Methods for

the Kern River Field, paper SPE 22338, presented at the
1992 SPE International Meeting on Petroleum
Engineering, Beijing, March 24-27.
32. Kumar, M. and Inouye, T.A.: Low-Temperature Analogs
of High-Temperature Water/Oil Relative Permeabilities,
Paper SPE 28616 presented at the SPE 69th ATCE of
SPE held in New Orleans, LA, USA, 25-28 September
1994.
33. Baker, L. E.: Three-Phase Relative Permeability
Correlations, paper SPE 17369 presented at the Sixth
SPE/DOE Symposium on EOR, Tulsa, OK, April 1988.

TABLE 3 Oil Recovery Including Drainage across

Stacked Zones

Sand

Oil Recovery at
End of Injection,
% OOIP

C1
G
K
K1
K2
R
R1

56.9
55.4
59.4
55.8
60.3
62.1
60.6

Oil Recovery
at Project
Termination,
%OOIP
51.6
55.9
60.6
51.6
54.1
58.9
70.5

Difference
,
%OOIP
-5.3
+0.4
+1.2
-4.2
-6.2
-3.2
+9.9

Tables
TABLE 1 Input Parameters
Average Pattern Size, acres

2.6
234

710

710

400
400

Crude Gravity, API

Crude Molecular Weight

14
400

Average Sand Porosity

Average Sand Permeability, md

0.32
2,500

220

307

711

263

401
401

240

Location of SingleSand Model

403
403

236

186

402

711

275

404

402

238

203

Initial Oil Saturation

Initial Gas Saturation in Oil Zone

0.45 .55
0.0

253

271

258

414
414
130

276

272

416
416

225

410
410

295
295

Oil Compressibility, psi-1

Rock Compressibility, psi-1

90

302

273

510-6
5010-6

265

294

318
318

310

250
0.70

303
153

750
750

229

301

212

298

210

270
264

502
502

500
289

291

292

503A
503A

260

503C
503C

308

286

282

408
305

313

504B
504B

501
501

504A

284

283

503B
503B

299

504A
269

280

281

511
511

311
304

288

287

Location of Multi223
Sand Model

277

194

309

297

407

274

290

156

407

195

296

239

206

TABLE 2 Crude Oil Viscosity

406
406

261

217

278

268

Steam Injection Pressure, psia

Injected Steam Quality

404

279

242

300
211

Initial Reservoir Temperature, F

405
405

311

509
306

205

702 702

411A
411A

509

320

314

411B

510510

411B

506

T, F
90
100
150
200
250
300
350
400

Viscosity, cp
4,740
3,000
410
93
30
13
6.3
3.6

506

505

413
413

412
412

505

Figure 1 Kern River field (and model segment

locations)

409
409

SPE/PS-CIM/CHOA 97719

L o g P e r m e a b ility
0

825
850

Depth, ft

875
900
925
950

x*1 ; y*1 ; z*3.8

975
Figure 4 Steam distribution in a vertical crosssection at various times before steam breakthrough

1000
Figure 2 Kern River reservoir zones: comparison of
measured core permeability and transform used in
the model at a well

x*1 ; y*1 ; z*5

dx=dy=20ft
dz=3ft

Figure 3 Permeability distributions for the single

zone model segment ( 25 inverted five-spot patterns)

Figure 5 3-D gas saturation distribution at different

times after steam breakthrough

SPE/PS-CIM/CHOA 97719

Grid blocks with Sg < 2% are not displayed

Figure 6 Effects on areal grid size on steam

saturation distributions (vertical lines represent well
locations)

Figure 8 Effect of cyclic steaming of producers on

reservoir pressures (near injection well)

Figure 7 Effect of model confinement on reservoir

pressures (near injection well)

Figure 9 Schematic of discontinuous shales in the

model area.

10

SPE/PS-CIM/CHOA 97719

6,000

C1

Cum Oil Production, MBBLS

Field Data

5,000

Detailed, Multi-Pattern Model

Average, Quarter-Pattern Model

4,000

3,000

K1

2,000

K2

1,000

R
R1

0
76

78

80

82

84

86

88

90

92

94

96

40

50

60

70

80

Time, Calendar Year

Figure 10 Comparison of measured and calculated

oil recovery using detailed, multi-pattern and
average, pattern-element models

Figure 12 Oil recovery from individual zones

0.50
K1 Injection

100

200

300

500
TOW 610-T1

600
Simulation

Average Oil Saturation

in Each Sand

Temperature, F

0.40
K1

0.30
R1
0.20

R1 Injection

0.10
0.00

Depth, ft

Time
700

Figure 13 Oil saturation changes with time in R1 and

K1 zones

800

900

Figure 11 Comparison of measured and calculated

temperature profiles at Well 610-T1

SPE/PS-CIM/CHOA 97719

11

Parallel Processing Comparison

560,000 cell Thermal Model
12.00

Relative Speed

10.00
8.00
6.00
4.00
2.00
0.00
1-CPU

2-CPU

4-CPU

9-CPU

16-CPU

24-CPU

Model

Figure 14 Parallel computing can improve the speed

of thermal simulations significantly

Parallel model Performance

Relative Speed vs. Model Size
9
8
Relative Speed

60,000cells
145,000cells
151,000cells
560,000cells

25-pattern
homogeneous

5
4
3

25-pattern
homogeneous

2
1

25-pattern
heterogeneous

9-pattern
heterogeneous

0
60,000cells

145,000cells

151,000cells

Model Size

Figure 15 Parallel computing provided greater

improvement for larger model

560,000cells