Implementing Convection

in a Reservoir Simulator: A Key Feature

in Adequately Modeling the Exploitation
of the Cantarell Complex
E. Manceau, E. Delamaide, SPE, J.C. Sabathier, SPE, S. Jullian, and F. Kalaydjian, SPE, IFP, and
J.E. Ladron De Guevara, SPE, J.L. Sanchez Bujanos, and F.D. Rodriguez, Pemex

Summary
As with some thick and highly fractured Iranian fields, the
Cantarell complex located offshore Mexico presents features
[decreases in the production gas/oil ratio (GOR) and bubblepoint
pressure with time] that reveal the effect of convection.
This effect on the past homogenization of fluid properties is discussed and supported by a thorough characterization of the thermodynamic properties of actual reservoir fluids.
To model convection, the reservoir simulator used for this
study was purpose adapted. Sensitivity runs were performed to
demonstrate the necessity of accounting for convection when
matching the past history of the Akal field, which is part of the
Cantarell complex.
Introduction
Presentation of the Cantarell Complex. The Cantarell complex
is the most important oil field in Mexico, and the sixth-largest in
the world. To economically optimize its value, it has been decided
to initiate a major recovery process by injecting nitrogen for pressure-maintenance purposes. Cantarell field is a thick, highly fractured reservoir; therefore, it is the kind of reservoir where convection phenomena may occur. Convection is a complex process that
is characterized by a vertical homogenization of fluid properties in
the fractures. This may have an essential impact on production and
injection profiles, in particular on the quantity of nitrogen in the
effluents as well as nitrogen breakthrough times, and therefore on
the overall nitrogen-injection efficiency.
The Cantarell complex is located offshore approximately 85 km
from Ciudad del Carmen. It includes four adjacent oil fields known
as Akal, Chac, Kutz, and Nohoch. Akal is the largest oil accumulation, with more than 90% of the 35 billion barrels of oil in place.
The reservoir is an anticline producing from the fractured carbonates of the Cretaceous and upper Jurassic formations, which also
contain many vugs and caves. The Upper Cretaceous is the most
fractured and brecciated. Fracturing decreases with depth in the
Middle and Lower Cretaceous. The average thickness of the whole
reservoir is about 775 m, and the depth of the top Cretaceous
ranges between 1100 and 3600 m true vertical depth subsea (SS).
Below the Cretaceous sequence, the Upper Jurassic (Oxfordian,
Kimmeridjian, Tithonian) is a stratigraphic reservoir with poor
reservoir characteristics.
Field production started in June 1979, reaching a peak of 1.157
MMBOPD in April 1981, with 40 producing wells. A total of 184
wells were drilled in Cantarell, among them 173 wells in Akal
alone. Cantarell crude is a 19 to 22API Maya type, with an initial
bubblepoint pressure close to 150 bar. Initially, the reservoir pressure was above the bubblepoint pressure and was equal to 266 bar
at 2300 mSS; therefore, there was no initial gas cap. The reservoir
pressure rapidly reached the bubblepoint pressure, and a secondary
gas cap appeared in 1981. The gas/oil contact (GOC) was located
Copyright 2001 Society of Petroleum Engineers
This paper (SPE 71303) was revised for publication from paper SPE 59044, first presented
at the 2000 SPE International Petroleum Conference and Exhibition in Mexico, Villahermosa,
Mexico, 13 February. Original manuscript received for review 16 March 2000. Revised manuscript received 14 February 2001. Paper peer approved 16 February 2001.

128

at 1800 mSS in 1997. The corresponding cumulative oil production

was around 5.5 billion STB.
Accounting for Convection in Cantarell Complex. Cantarell
field appears to have all the characteristics of a reservoir where
convection may occur. As observed, for instance, in a major Iranian
field,1 convection is a complex phenomenon that occurs in thick
and highly fractured reservoirs. As explained in detail by Saidi,2 it
results from a combination of thermal gradients, gas liberation at
the GOC, and gravity segregation, and it is made possible by high
vertical permeabilities. When the oil initially reaches the bubblepoint pressure, it liberates gas in solution, thus becoming heavier.
Because of the high vertical permeability, this heavier oil can move
downward while lighter oil heated from below expands and rises.
A convection flux is then established, finally leading to a fast
homogenization of the oil properties along the vertical depth. This
leads to a reduction of the bubblepoint pressure in a vertical oil column. Indications that convection is taking place include more
homogeneous oil properties and temperature on the vertical,
change of the oil composition with time, and decline of bubblepoint pressure and production GOR with time.
For Akal, producing GORs were plotted vs. time for each well.
The initial mean GOR value is approximately 90 vol/vol. The wells
were organized into four classes: wells with decreasing then
increasing GOR, wells with increasing GOR, wells with a constant
GOR, and wells with a decreasing GOR. Fig. 1 shows a typical
well with a decreasing behavior. Such a well generally begins producing with an initial GOR value of 90 vol/vol, then its GOR slowly decreases down to around 60 vol/vol. This means that the oil
produced becomes heavier with time. Typically, this can be
explained by convection. Fig. 2 shows the location of all the wells
on the Cantarell field with their classification as of 1993. The corresponding GOC is also drawn. One can observe that the major
part of the wells areally allocated close to the top of the structure,
despite the vertical position of their completion, shows a GOR
behavior other than constant as the reservoir pressure goes down,
while the wells allocated through the flanks of the structure show
a constant GOR behavior. This means that a complex phenomenon
affecting the original fluid properties is taking place. Even though
there is no evidence of convection, it is assumed that convection
also takes place in the gas cap, leading to a faster homogenization
in this area. However, this is not the main focus of this paper.
To confirm this statement, three oil samples were taken in 1997
from three different zones of the reservoir: sample 1.07 was found
close to the GOC, sample 1.11 was at an intermediate location, and
sample 1.16 was in a deep zone, close to the water/oil contact
(WOC). The bubblepoint pressures, as well as the flash GOR
measured for each oil sample, are presented in Table 1. It can be
observed that, for each crude sample, the oil is heavier than the initial oil in place and that the deeper the oil, the lighter it is.
All this contributes to the declaration that a convection phenomenon takes place in the thick and highly fractured Cantarell complex. Therefore, this process must be taken into account in a
Cantarell reservoir model to accurately match the production profiles and to properly predict the nitrogen-concentration evolution in
the gas effluents, thereby optimizing the nitrogen-injection process.
April 2001 SPE Reservoir Evaluation & Engineering

Producing GOR, vol/vol

120

D = L/t,

100
80
60
40
20
0
06/1979

06/1982

06/1985

06/1988

06/1991

06/1994

06/1997

Date

Fig. 1Typical trend of a producing GOR under a convection

phenomenon.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)

where L=the mean distance made by a concentration front (cm) in

a time t (sec). Thanks to convection, the concentration front can go
as quickly as 100 m per month, sometimes even more. This gives
the order of magnitude for diffusion values accounting for convection at approximately 1 to 100 cm2s-1. When convection does not
take place, the diffusion value is close to 10-5 cm2s-1. Therefore, the
convection induces a multiplying factor of which the order of magnitude is between 105 and 107. For Cantarell, a COEFF of 106 was
finally used in the fracture network. As explained later in this paper,
this value gives quite an accurate match to the decline in GORs.
Moreover, when convection takes place, the lighter components
of the oil are liberated, making the oil heavier. This is why the pressure/volume/temperature (PVT) should be as precise as possible.
As a consequence, the use of a compositional model is strongly
advised when considering convection phenomena.
Convection implies the use of a dual-porosity, single-permeability, detailed compositional model, as well as some specific diffusion coefficients for the study of Cantarell. In spite of these
time-consuming requirements, ATHOS enables the model to
run very efficiently.
Reservoir Model Calibration
The purpose of the proposed reservoir model was to qualify the
impact of convection on the history match and on the production
forecasts for the nitrogen-injection phase. Therefore, the match
was not on a well-per-well basis. As a consequence, a phenomenological model was used with a limited number of cells.

Fig. 2Convection diagnostic through GOR evolution in

Cantarell field.

TABLE 1MEASURED OIL PROPERTIES AT DIFFERENT

LOCATIONS IN AKAL RESERVOIR
Sample
Number
Sample 1.07
Sample 1.11
Sample 1.16

Depth
(mSS)

Bubblepoint
Pressure (bar)

Flash GOR
(vol/vol)

1965
2267
2630

84
112
124

52
65
72

Convection Modeling in a Reservoir Simulator

Conventional numerical simulators are not equipped with a convection option because it requires very specific flow conditions
and reservoir structure, particularly large vertical transmissivities.
These conditions mainly account for thick and highly fractured
reservoirs. Therefore, an adapted dual-porosity model must be
used. ATHOS software, which was used for this study, has a
well-developed dual porosity option,3 based on the Warren and
Root theory.4 In Akal, the greater part of the oil in place is assumed
to be in the matrix. On the other hand, as the fracture permeability
is much higher than the matrix permeability, the flow occurs mainly within the fracture network. Therefore, a dual-porosity, singlepermeability model was chosen.
The convection phenomenon was modeled by amplifying the
natural vertical oil-diffusion process in the fractures. Increasing
vertical diffusion results in a rapid change of oil properties over
the whole thickness of the reservoir. As gas is liberated at the
GOC, the changes of the oil composition are transferred to the oil
directly below, which is the exact effect of convection. Therefore,
convection has been taken into account in this study by implementing in the reservoir simulator a very large vertical multiplying oil-diffusion coefficient in the fractures (COEFF). The final
vertical oil-diffusion coefficient used for the simulations is therefore the result of the multiplication of the natural oil-diffusion
coefficient by the COEFF. The diffusion coefficient D can be
defined as follows.
April 2001 SPE Reservoir Evaluation & Engineering

Grid Building. The scope of the study required that a simple

model be built. It was decided to represent the whole Akal block
with only two cells in the x-direction. The y-axis was oriented perpendicularly to the slope (southwest/northeast). The width of the
cells in the x-direction was 5000 m. A total of 25 cells was selected in the y-direction. The width of the cells in the y-direction was
500 m, except at the extremities, where it went up to 1500 m. The
grid used for the simulations is therefore far from a square grid.
This may have a strong impact on the accuracy of the production
results. However, the goal of this phenomenological study was neither to get an accurate history match, nor to compute accurate production forecasts, but to quantify the impact of convection on the
nitrogen-injection efficiency. The horizontal gridding used was
therefore considered satisfactory because it allowed for correct
representation of the physics of convection.
The most important phenomenon in Akal was the formation of
a secondary gas cap, in addition to convection. Moreover, as nitrogen will be injected into the gas cap, it is important to have a fine
gridding in the vertical direction to model the gas cap expansion as
precisely as possible. Another reason to use a fine gridding is that
gravity plays an important role in the convection equilibration. As
a consequence, 42 layers are defined in the z-direction, with a
mean thickness of about 20 m. This appears to be sufficiently discretized in the vertical direction to correctly represent the physical
phenomena. However, for accurate production forecasts, the use of
more cells in the vertical direction is recommended.
The final grid size was 2 by 25 by 42; thus, 4,200 cells had the
dual-porosity option.
PVT Characterization. As already explained, convection modeling requires a thorough characterization of the thermodynamic
properties of reservoir fluids. Two types of data were used. First,
some initial data from 1979 samples were directly used. Each sample included a compositional analysis as well as density, viscosity,
flash, and differential liberation. The different samples were consistent with each other and indicated that the oil composition was
initially identical throughout the reservoir.
Because convection occurs in the Akal reservoir, the oil composition continuously changes with time. It was therefore primordial to find additional data from another date to fit the evolution of
the oil composition in the model. This is why it was decided in
1997 to retrieve three additional samples of crude oil at different
129

PVT Model With Four Pseudocomponents. In a first attempt, it

was decided to use four pseudocomponents to calibrate the PVT
model: N2 alone; C1, H2S, and CO2 together; all components from
C2 to C6; and C7+. With this model, it has been found that a single
pseudocomponent could not be used for the C7+ fraction to get a
correct representation of bubblepoint evolution. As the bubblepoint
pressure evolution was one of the most important parameters for
controlling convection phenomenon, it was decided to compute a
new PVT model with five pseudocomponents to optimize the calibration of the heavy fraction.
PVT Model With Five Pseudocomponents. Instead of considering
only one component, C7+, for the heavy fraction, as was done previously, the heavy fraction was modeled by two pseudocomponents: C7 through C14 together, and C15+. The tuning was performed automatically by the ATHOS PVT package by constraining the model with the initial PVT data (1979) and with the
lab measurements performed on the three 1997 samples. Table 2
presents the comparison of the PVT results computed with the EOS
with the initial PVT data (1979). Figs. 3 and 4 show the results in
TABLE 2COMPARISON OF PVT MODEL
WITH INITIAL PVT DATA (1979)
Thermodynamic Properties
Bubblepoint Pressure (bar)
Liquid Density at pb, Tres (g/cm3)
Liquid Viscosity at pb, Tres (cp)

130

Initial Sample
(1979)

PVT
Model

149.1
0.7874
2.320

149.5
0.7844
2.324

Formation Volume Factor, vol/vol

1.35

1.30

1.25

1.20

Initial PVT (1979)

1.15

PVT Model
1.10

1.05

1.00
0

20

40

60

80

100

120

140

160

180

200

Pressure, bar

Fig. 3Final match of the FVF with the initial PVT data (1979).
100
90

Solution GOR, vol/vol

locations in the reservoir, namely samples 1.07, 1.11, and 1.16. In

the lab, these oil samples were analyzed with great accuracy. The
first results were obtained without nitrogen. For the three downhole oil samples investigated, a regular and steady evolution of the
analytical and PVT characteristics as a function of depth was
observed. GOR was found to increase from 52 to 72 m3/m3 when
measured at 15C and atmospheric pressure. The bubblepoint pressure, the compressibility (at both bubblepoint pressure and reservoir pressure), the swelling factor (at bubblepoint pressure), and
the density (at bubblepoint pressure, reservoir pressure, and atmospheric conditions) were also measured. Finally, the oil viscosity as
a function of pressure was measured for the three samples. Again,
a steady evolution of these characteristics was observed for the
three downhole oil samples. This was the case for every pressure
except those close to the bubblepoint pressure, where some asphaltene aggregation could occur and thus increase the apparent viscosity. Indeed, using an n-heptane titration test, an asphaltene content of 8.8% weight was measured.
To complete this characterization task, and to complete the PVT
data set, a series of tests was performed in the presence and
absence of nitrogen. A constant mass/composition expansion test
of the three downhole oils was carried out for which the compressibility factor and the oil density were measured at each pressure step. A differential expansion of the deepest crude oil with and
without nitrogen was performed as well. At each pressure, the
released gas and the residual liquid were analyzed. The main conclusions derived from the comparison of this series of tests were
that there was an important increase in the bubblepoint pressure
when relatively small amounts of nitrogen were dissolved in the
crude, and that the effect of nitrogen on asphaltene aggregation
was negligible.
Once all these lab measurements were performed, they had to
be implemented into the reservoir model. However, in reservoir
simulations, it is not possible to take into account all the oil components because doing so would lead to prohibitive simulation
times. The current practice is therefore to use pseudocomponents,
obtained by lumping together the actual oil components. Then an
equation of state (EOS) is used to match the available oil properties as successfully as possible. In this study, the Peng-Robinson5
equation was used. N2 usually can be lumped with other components; however, for the present study, it was decided to leave N2 as
a single component because nitrogen was the focus of the study.

80
70
60
50
40

Initial PVT(1979)

30

PVT Model

20
10
0
0

50

100

150

200

Pressure, bar

Fig. 4Final match of the solution GOR with the initial PVT data
(1979).

terms of formation volume factor and solution GOR vs. pressure.

For all those properties, the match is rather accurate except at low
pressure, where the FVF error reaches a maximum of 2%.
However, this is not critical because the static pressure in the reservoir never reaches such a low level.
The match of the five-pseudocomponent PVT model with the
1997 lab experiments in terms of bubblepoint pressure and flash
GOR is presented in Table 3. Except for the sample 1.07, the
match is rather accurate, with a maximum error of 4% in terms of
bubblepoint pressure. The match is also rather accurate for the
GOR values. However, when the oil becomes heavier, the match
does not seem as good. This is especially the case for sample 1.07:
the computed bubblepoint pressure using the PVT model is 95 bar,
compared to an experimental value of 84 bar. Note, however, that
the value obtained with only four pseudocomponents was much
higher (equal to 103 bar). This clearly indicates that the heavy fraction is a critical parameter and that a better match would require
some additional pseudocomponents in the heavy range.
Unfortunately, this would considerably increase the CPU time and
was therefore discarded in this study. In any case, the PVT model
with five pseudocomponents clearly leads to a better match than
did the previous PVT model. For instance, for sample 1.16, the calculated bubblepoint pressure is 124.8 bar for five pseudocomponents (compared to 124.0 measured), whereas the value computed
by using four pseudocomponents was 127.9 bar. This clearly
demonstrates that using five pseudocomponents instead of four
enables a more efficient modeling of the heavy fraction, which is a
primordial step in characterizing convection phenomenon.
A last comparison is presented in Table 4 regarding the differential liberation results of mixture downhole crudes with and without nitrogen. Once again, the match is rather accurate. The GOR
match is good; the main discrepancy concerns the FVF, especially
at low pressure. As explained earlier, this difference was already
observed when comparing the PVT model with the initial sample
data (1979). However, the reservoir pressure still remains above 80
bar, thanks to nitrogen injection. Therefore, this should not have a
significant impact on the production forecasts.
April 2001 SPE Reservoir Evaluation & Engineering

TABLE 3COMPARISON OF LAB RESULTS AND PVT MODEL WITH THE PENG-ROBINSON
EOS FOR THREE 1997 OIL SAMPLES

Sample Number
Sample 1.07
Sample 1.11
Sample 1.16
Sample 1.07 + N2 min
Sample 1.07 + N2 max
Sample 1.11 + N2 min
Sample 1.11 + N2 max
Sample 1.16 + N2 min
Sample 1.16 + N2 max

Saturation
Lab Results
(bar)

Pressure
PVT Model
(bar)

Flash
Lab Results
(vol/vol)

GOR
PVT Model
(vol/vol)

84.0
112.0
124.0
184.0
270.0
206.0
333.0
210.0
238.0

95.0
115.9
124.8
191.2
267.6
205.7
323.2
207.3
243.8

52
65
72
63
69
73
96
80
86

54
65
74
58
63
71
93
78
82

TABLE 4COMPARISON OF THE PVT MODEL WITH EXPERIMENTS OF DIFFERENT

LIBERATIONS OF MIXTURE DOWNHOLE CRUDES (WITH AND WITHOUT NITROGEN)
Mixture

Thermodynamic Properties

Lab Results

PVT Model

Crude Oil
Without Nitrogen

Bubblepoint Pressure (bar)

Liquid Density at pb, Tres (g/cm3)
Bo Flash
GOR Flash (vol/vol)

75
0.817
1.181
43

83
0.819
1.189
44

Crude Oil
+
5 mol% N2
(at 75 bar)

Bubblepoint Pressure (bar)

Liquid Density at pb, Tres (g/cm3)
Bo Flash
GOR Flash (vol/vol)

75
0.839
1.145
27

80
0.843
1.130
28

Crude Oil
+
5 mol% N2
(at 50 bar)

Bubblepoint Pressure (bar)

Liquid Density at pb, Tres (g/cm3)
Bo Flash
GOR Flash (vol/vol)

50
0.840
1.13
22

56
0.850
1.10
19

History Match
The main concern during the history match was enabling oil from
the matrix to be sufficiently produced. Indeed, it appeared that,
using the initial values of porosity, permeabilities, relative permeability and capillary-pressure curves, block size, and matrix/fracture porosity distribution, the major part of the oil produced came
from the fracture. In a first approach, the matrix/fracture porosity
distribution was assumed to be a fixed parameter; 17% of the original oil in place (OOIP) belonged to the fracture, and 83%
belonged to the matrix. Therefore, other parameters were modified
to get a proper match in terms of mean reservoir pressure, GOC,
WOC, and bubblepoint pressure distribution vs. time. However, it
has not been possible to properly match all these properties without drastically changing the permeabilities and the Kr-Pc properties, which would have no physical meanings.
In a second attempt, based on vuggy porosity connected
through the fractures system, it was admitted that the matrix/fracture porosity distribution can be a matching parameter, as long as
the overall OOIP of Akal reservoir remains constant and equal to
32 billion barrels. Finally, a proper match was obtained with 35%
of the OOIP in the secondary porosity. This value seems reasonable
because of the presence of many vugs in the reservoir. Moreover,
this value of fracture volume allows a smooth match while keeping
all the other parameters in ranges compatible with field observations. Some slight changes were performed, including the matrix
value of waterflood residual oil saturation and the water capillarypressure curve. These changes were made as a consequence of
some experimental values measured in IFP laboratories. The water
capillary curve was modified to account for the fact that, in the lab,
a water-pressure limit of around -50 mbar had been observed
before any oil could move out.
Figs. 5 through 7 show the history match obtained respectively in terms of mean reservoir pressure, GOC, and WOC. The presApril 2001 SPE Reservoir Evaluation & Engineering

sure and the GOC matches are accurate; however, the mean reservoir pressure is a little too high at the beginning of the history. This
is a consequence of the coarse gridding and may be caused by a
lack of definition at the top of the structure. The WOC is accurate,
even though it seems to increase too quickly at the beginning of the
history match. The field data available for matching the WOC
allow uncertainty in this matching parameter and, because the main
focus of the study was the injection process in the gas cap, the
WOC match was considered satisfactory.
The bubblepoint pressure was also an important parameter to
match. Its distribution is mainly affected by the convection phenomenon; therefore, the tuning parameter for matching the bubblepoint pressure distribution was the COEFF. The simulations performed showed that the bubblepoint pressure evolution is strongly
dependent on the COEFF value. The values of bubblepoint pressures to match were the ones measured in the lab on the three crude
samples retrieved in 1997. The sample locations were identified in
300

Field Data

250

Pressure, bar

Finally, this PVT model with five pseudocomponents was used

for matching the model and for nitrogen injection runs.

Reservoir Model
200
150
100
50
0
06/1979

06/1983

06/1987

06 /1991

06/1995

Date
Fig. 5Final match of the mean reservoir pressure for Akal field.
131

1120

2,200

1220

2,400

Depth, m SS

Depth, m SS

Field Data

1320

Reservoir Model

1420
1520
1620
1720

Field Data

2,600

Reservoir Model

2,800
3,000
3,200

1820
3,400

1920
06/1975

06/1979

06/1983

06/1987

06/1991

06/1995

05/1975

05/1979

05/1983

05/1987

05/1991

05/1995

Date

Date

Fig. 6Final match of the GOC for Akal field.

Fig. 7Final match of the WOC for Akal field.

the model grid and the experimental values of bubblepoint pressures were compared to the computed values obtained with the
model at that location in the grid in 1997. A few runs with different COEFF values were performed. The final match is presented in
Table 5, corresponding to a COEFF value of 106.
There is a significant difference for the heavier sample, sample
1.07. However, the oil composition obtained by the reservoir-simulation model at that location at that time is very similar to the real
oil composition measured in the lab. This means that the distribution of the oil composition in the reservoir is well-matched; therefore, the convection phenomenon has been modeled accurately.
The discrepancy observed on the bubblepoint pressure is actually
the consequence of the too-coarse lumping of the heavy fraction in
PVT modeling. To be more consistent in terms of bubblepoint
pressure values, it is necessary to recompute the PVT model with
more pseudocomponents, as explained earlier, even though this
means more CPU time. Finally, with this match, the GOR evolution obtained for each well was compared to the real field GOR
evolution with time. Once again, the match is accurate and the tendency to have a decreasing GOR from 90 to around 60 vol/vol was
confirmed for many of the wells located below the gas cap.
As explained earlier, a final COEFF value of 106 is acceptable; it
remains in the order of magnitude of a classical convection process.

breakthroughs observed on some production wells at the center of

the field.
The convection phenomenon therefore plays an essential role in
Cantarell, especially on fluid thermodynamics allowing excess gas
in the gas cap, which affects reservoir mean pressure as well as
GOC and WOC evolution.

Sensitivity Tests Related to Convection

Phenomenon
Once the history match was validated, some sensitivity tests were
performed to characterize the real impact of convection modeling
on the model itself.
Sensitivity Tests on the Matched General Reservoir Properties.
For the three general properties matched during this study (mean
reservoir pressure, GOC, and WOC), the impact of convection cannot be visualized very easily. Convection has limited direct influence on the mean reservoir pressure decline and on the WOC.
However, it does have a more significant impact on the GOC evolution; when the reservoir pressure declines, the excess gas in the
matrix blocks diffuses and is released to the gas cap. This was confirmed by the study; the gas cap was as much as 50 m lower when
considering convection than without convection. This gas-cap
extension has a strong impact on the gas breakthrough times, and
therefore on the overall production profile of the field. Note that
this convection phenomenon has been confirmed by the early gas
TABLE 5BUBBLEPOINT PRESSURE
MATCHES BETWEEN LAB MEASUREMENTS AND
RESERVOIR SIMULATION RESULTS IN 1997
Sample Number

Lab Measurement

Reservoir Simulation

Sample 1.07

84.0

95.8

Sample 1.11

112.0

107.1

Sample 1.16

124.0

129.3

132

Impact of Convection on the Bubblepoint Pressure Distribution

in the Matrix. The convection phenomenon takes place in the
fracture, leading to a significant decline of the bubblepoint pressure in the fracture. The purpose of this sensitivity test was to
observe the impact of the decline on the matrix oil properties.
Three runs were computed for 5,800 days and the bubblepoint
pressures in the matrix were compared at a specific location of the
grid. The first run consisted of a reference case without lateral diffusion and without convection. In that case, the bubblepoint pressure in the matrix and in the fracture remained constant, equal to
150 bar. In a second run, convection was considered, but without
lateral diffusion. In that case, in the oil zone below the gas cap, the
bubblepoint pressure in the fracture dropped down to 113 bar
because of the convection phenomenon. In the matrix, the bubblepoint pressure also decreased, but only up to 145 bar. This decline
can be explained by gravity, which enables the heavier oil of the
fracture to enter the matrix just below. Finally, the third run considered both convection and lateral diffusion. In that case, the bubblepoint pressure in the matrix was even lower (equal to 139 bar);
in fact, lateral diffusion enabled additional exchanges between the
matrix and the fracture. This exchange is strongly dependent on the
block sizes. In fact, the block size is a geometric representation of
the matrix/fracture exchanges. It is assumed that the matrix is constrained in cubic blocks, surrounded by fractures. The smaller the
block size, the larger the fracture density, and the larger the
exchanges between the matrix and the fracture.
The Cantarell field is highly fractured and the matrix is poorly
permeable, which explains why most of the flow goes through
matrix/fracture exchanges. Therefore, the bubblepoint pressure in
the matrix decreases with time through matrix/fracture exchanges
caused by two main phenomena: lateral diffusion and gravity.
Impact of Convection on the Nitrogen Molar Fraction in the
Gas Effluents. Nitrogen injection is planned as a major secondary
recovery process for pressure-maintenance purposes. One of the
key points when designing such an installation is the design of surface facilities for effluent separation and treatment, which strongly
depend on the quantity of nitrogen in the gas effluents. The reservoir model built in this study, though of phenomenological nature
and therefore not dedicated to deliver such accurate values on a
well-per-well basis, was unique as far as convection is concerned.
It was, therefore, a real benefit to analyze the relative impact of
convection on the results obtained in terms of nitrogen-molar fraction in gas effluents per platform.
The nitrogen injection began around April 2000. Before this
date, there was a very small quantity of nitrogen in the gas effluent,
owing to the presence of nitrogen in the original Cantarells hydrocarbon mixture. Nitrogen is injected at the top of the structure and
April 2001 SPE Reservoir Evaluation & Engineering

takes approximately 3 years to reach the producer wells. However,

the results differ strongly from one platform to another. Fig. 8
shows the nitrogen molar fraction evolution for two platforms vs.
time through the year 2023. For the first platform, located close to
the nitrogen injection site in the center of the field, the nitrogen
injected is rapidly present in an important quantity in the gas effluent. Indeed, thanks to thermodynamic diffusion and convection
phenomena, nitrogen moves downward quite quickly. On the other
hand, when considering a platform producing far from the nitrogeninjection location (on the edges of the field, close to the WOC), the
quantity of nitrogen in the gas effluent remains low. As no direct
compound exchange occurs in this area of the field between the gas
cap and the oil zone, nitrogen moves much more slowly. This clearly indicates that convection is one of the main mechanisms affecting the quantity of nitrogen in the gas effluents. Of course, the
blending of the production of the different platforms is highly recommended because it could alleviate the problem of high nitrogen
concentration levels in some platforms in the center of the structure.
To validate this approach, an additional run was performed
without activating the convection option in the model. The comparison of the nitrogen molar fraction in the gas effluent obtained
with this model, and the value obtained with the reference model
(with convection), is presented in Fig. 9. Note that these results
concern a platform in the center of the structure, under convection
impact. When convection is activated, nitrogen moves much more
quickly in the oil column below the gas cap. Consequently, the
nitrogen breakthrough comes earlier3 years earlier, in the case of
the platform shown in Fig. 9. For both configurations, the platform
is shut down when the gas cap reaches the perforations, and
approximately the same molar fraction of nitrogen is found at that
time, corresponding to the gas-cap nitrogen concentration.

Molar fraction of nitrogen in gas effluent

Conclusions
A thorough analysis of thermodynamic properties and production
data clearly indicates that convection phenomena occur in

0.40

Platform in the center of the structure, below the gas cap

0.30

Platform on the edges, far from the gas cap

0.20

0.10

0.00
06/1979

06/1987

06/1995

06/2003

06/2011

06/2019

Date

N2 molar fraction in gas effluent, % mol

Fig. 8Nitrogen molar fraction in gas effluent for a platform as

a function of the platform location.
30
25

Nomenclature
Bo = oil formation volume factor, vol/vol
D = diffusion coefficient, cm2s-1
L = mean distance made by a concentration front
in a time t, cm
pb = oil bubblepoint pressure, bar
t = time, seconds
Tres = reservoir temperature, C
Acknowledgments
We would like to thank Pemex for their technical contribution and
permission to publish this paper, as well as the IFP personnel
involved in this integrated study, from core laboratories to fluid
analysis and reservoir simulations.

With Convection
20

Cantarell field. As this may have a significant impact on the overall efficiency of nitrogen injection, a phenomenological reservoir
model was built to quantify the impact.
The ATHOS software was adapted to deal with such a phenomenon, in a dual-porosity, single-permeability, and compositional configuration. Convection was modeled by a large vertical
diffusion multiplying coefficient in the fractures.
Convection implies some strong thermodynamic exchanges, so
a thorough characterization of the PVT samples was also performed. This included the detailed analysis of three samples taken
from the reservoir in 1997 at three different locations, as well as the
optimization of the PVT lumping. It was demonstrated that the
modeling of heavy fractions in the Peng-Robinson equation is a
critical parameter for convection. Finally, five pseudocomponents
were used for the PVT model.
A history match was then performed. The general reservoir
properties, such as mean reservoir pressure, GOC, and WOC, were
properly matched. The bubblepoint distribution and the GOR
declines were also well-matched, with a COEFF of 106.
Finally, some sensitivity tests were performed and the following observations were made:
1. Convection has limited direct influence on the mean reservoir pressure and on the WOC. However, it does have a considerable impact on the GOC evolution.
2. Convection has a strong influence on the production profiles
of the wells. Even though the purpose of this study was not
to match the history on a well-per-well basis, it was clearly
observed, both in real field data and in the model, that the
wells located below the gas cap had a GOR that significantly declined with time while, on the edges, the GORs
remained constant.
3. Convection has a direct, strong impact on the bubblepoint
distribution in the fractures. Moreover, because of gravity
and lateral diffusion, thermodynamic exchanges also occur in
the matrix, and the bubblepoint pressure distribution in the
matrix is significantly altered.
4. Convection has a primordial impact on the nitrogen molar
fraction in the gas effluents. This by itself justifies the need
to account for convection when looking at nitrogen injection forecasts.
As a general conclusion, in any field such as Cantarell where
convection clearly occurs, it is highly recommended to perform
detailed three-dimensional (3D) reservoir simulations with an
improved reservoir and PVT characterization, taking into account
convection as well as gravity and lateral diffusion. This is crucial
to properly predict production and injection profiles, and thereby to
optimize the design of future surface facilities.

Without Convection

15
10

References
5
0
1/6/79

1/6/87

1/6/95

1/6/03

1/6/11

1/6/19

Date

Fig. 9Impact of convection on the nitrogen molar fraction in

the gas effluent of a platform.
April 2001 SPE Reservoir Evaluation & Engineering

1. Saidi, A.M.: Twenty Years of Gas Injection History into WellFractured Haft Kel Field (Iran), paper SPE 35309 presented at the
1996 SPE International Petroleum Conference and Exhibition in
Mexico, Villahermosa, Mexico, 57 March.
2. Saidi, A.M.: Reservoir Engineering of Fractured Reservoirs, TOTAL
Edition Presse, Paris (1987) 1, 307395.
3. Quandalle, P. and Sabathier, J.C.: Typical Features of a Multipurpose
133

Reservoir Simulator, SPERE (November 1989) 475.

4. Warren, J.E. and Root, P.J.: The Behavior of Naturally Fractured
Reservoirs, SPEJ (September 1963) 245; Trans., AIME, 228.
5. Peng, D.Y. and Robinson, D.R.: A rigorous method for predicting the
critical properties of multicomponent systems from an equation of
state, AICHE J. (March 1977) 23, No. 2, 137.

SI Metric
API
bar
bbl
F
ft
in.
in.2

Conversion Factors
141.5/(131.5 + API)
=
1.0*
E + 05 =
1.589 873
E - 01 =
(F - 32)/1.8
=
3.048*
E - 01 =
2.54*
E + 00 =
6.451 6*
E + 00 =

*Conversion factor is exact.

g/cm3
Pa
m3
C
m
cm
cm2
SPEREE

Emmanuel Manceau is a research engineer in the Reservoir

Engineering Division of IFP. Currently, he is in charge of a
research project for uncertainty quantification and risk analysis.
He has performed many extended reservoir-engineering studies, especially for fractured reservoirs in Latin America and the
Middle East. Manceau holds a BS degree in aeronautical engineering from ENSICA (France) and an MS degree in mechanical engineering from Louisiana State U. Eric Delamaide is the
technical manager of IFP Technologies (Canada) Inc. of
Calgary, an IFP subsidiary. He joined IFP in 1991 as a reservoir
engineer and has been responsible for reservoir engineering
studies in 20 countries in North and Central America, Europe,
and China, among others. He has authored or coauthored 15
papers and is an associate professor at the IFP School.
Delamaide holds a BS degree in mining engineering from Ecole
Nationale des Mines de St Etienne and an MS degree in petroleum engineering from Ecole Nationale Suprieure du Ptrole
et des Moteurs (IFP School), France. Jean-Claude Sabathier
recently retired from IFP, where he was Associate Director of
Research for reservoir engineering research programs. Before
joining IFP in 1993, he was a project leader for full-field studies

134

at Franlab, an IFP subsidiary, and a member of the French

Conservation Board. Sabathier holds MS degrees in aeronautical engineering from ENSAE, Paris, and in petroleum engineering from ENSPM, Rueil (France). He has served on the Editorial
Review Committee and the 1994 Forum Series Europe Steering
Committee. Sophie Jullian is currently Head Manager of the
Development Dept. at IFP. She joined IFP in 1983 and has
served as Project Manager of Gas Treatment and Head
Manager of the Thermodynamic Dept. Jullian holds an engineering degree from the chemical engineering school in Lyon,
France, and a PhD degree in chemical engineering. Franois
Kalaydjian is Director of the IFP Reservoir Engineering research
division. His research interests include the modeling and simulation of various recovery processes such as imbibition in fractured reservoirs, gas injection, WAG injection, and production
of gas condensates. He has authored or coauthored more
than 50 articles related to these topics. Kalaydjian holds an
engineering degree from the School of Mines in France and a
PhD degree in physics from the U. of Bordeaux, France. He has
served on various SPE program committees and was Chairman
of the 2000 SPE EUROPEC Program Committee for Integrated
Reservoir Management. Jos L. Snchez is a reservoir engineer
at Pemex Exploration and Production in Mexico City, where he
coordinates the Naturally Fractured Reservoirs Technologies.
Previously, he was a geophysicist and reservoir engineer in drygas fields at IMP and Pemex. Snchez holds a BS degree in
geophysics from UNAM, an MS degree in petroleum engineering from UNAM, and a PhD degree from the U. of Texas. Juan E.
Ladrn de Guevara is a reservoir engineer at Cantarell Asset in
Ciudad del Carmen, Campeche. Previously, he was a reservoir
engineer in Naturally Fractured Reservoirs at IMP and at the
South and Marine Regions of Pemex E&P. He holds a BS degree
in petroleum engineering from UNAM and is a graduate student in petroleum engineering at the School of UNAM.
Fernando Rodrguez is a reservoir engineer, head of the
Reservoir Management area of Pemex E&Ps Technology
Division, and a part-time professor at UNAM. Previously, he
worked for the Mexican Petroleum Inst. and was a professor at
UNICAMP in Brazil. His background is in reservoir simulation and
well-testing; recently, he has been involved in fractured reservoir management issues. Rodrguez holds BS and MS degrees
from IPN and UNAM, respectively, and a PhD degree from
Stanford U., all in petroleum engineering.

April 2001 SPE Reservoir Evaluation & Engineering