Journal of Petroleum Science and Engineering 131 (2015) 34–50

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

Integration of multiscale carbonate reservoir heterogeneities

in reservoir simulation
Manuel Gomes Correia n, Célio Maschio, Denis José Schiozer
Petroleum Engineering Department, Faculdade de Engenharia Mecânica - UNICAMP Caixa Postal 6122, 13.083-970 Campinas, São Paulo, Brazil

art ic l e i nf o a b s t r a c t

Article history: The construction of robust reservoir models considering geological carbonate heterogeneities, such as
Received 15 September 2013 fractures and vugs, upscaling and numerical flow simulation remains a challenge. This work performs a
Accepted 10 April 2015 methodology aiming a suitable representation of flow simulation and upscaling techniques according to
Available online 25 April 2015
small and large scale carbonate reservoir heterogeneities. The methodology is applied to one specific
Keywords: flow unit type considering diffuse fractures, sub-seismic fractures and isolated vugs.
carbonate reservoir The methodology follows four elementary steps: division of reservoir into flow units, geostatistical
fracture upscaling modeling, upscaling procedures and flow simulation modeling of carbonate reservoirs. Given the
dual porosity presence of multiscale geological features and computational limitations, the upscaling procedure is
dual permeability
separated as a function of heterogeneities scale.
The process is shown for one flow unit of a specific case where flow progress in matrix and fracture
occur at different time steps which make the single porosity approach difficult. For regions of fractures
not fully connected, the flow takes place in the matrix. A dual permeability flow model is the best
approach since it can take into account both flow progresses in connected fracture networks and in
matrix medium.
The methodology has shown several advantages: sequential control over static properties and
pseudo-functions integrated with upscaling procedure; definition of the accurate simulation flow
model; and improvement of the integration of multiscale heterogeneities in reservoir simulation.
This work presents a new methodology and analysis that can be useful for multidisciplinary areas.
& 2015 Elsevier B.V. All rights reserved.

1. Introduction rock. Upscaling is therefore a pre-requisite for flow simulation

Carbonate oil reservoirs hold significant quantities of the world
oil and gas reserves. About half of the world proven oil reserves are
trapped in fractured carbonate reservoirs. The high number of
uncertainty involved in the development of carbonate reservoirs
leads to the need of understanding the flow behavior in all scales
(Ahr, 2008).
For flow simulation purposes, the development of a naturally
fractured carbonate reservoir keeping its heterogeneous behavior
and in a reasonable simulation consumption time, can be a
challenge. Fracture scales could range from small scale diffuse
fractures, intermediate scale sub-seismic faults to large scale seismic
faults.
The discrete fracture network (DFN) model cannot be included
into field scale models because of the computational limitation to
fix possibly billions of fractures in each cubic kilometer of reservoir
n
Corresponding author.
E-mail address: manuel@dep.fem.unicamp.br (M.G. Correia).
(Dershowitz et al., 1998).
The translation of DFN models and/or implicit fracture models
into an equivalent medium is set by upscaling procedures. Either
analytical or numerical methods can be used to upscale fracture
permeability. Oda (1985) proposed an analytical equation to
calculate fracture-permeability tensor and Lough et al. (1997)
presented an approach using the boundary-element method. Oda's
solution is faster since it is an analytical solution. The equation can
be described as:
X NBFracs
NBsets X
1 C f ;s
Kf ¼ V f ;s N
V cell 12ef ;s f ;s
s¼1 f ¼1
where in V cell is respect to block volume, C f ;s and ef ;s are related to
fracture conductivity and fracture aperture, respectively. NBsets is
the number of fracture sets, NBFracs is the number of fracture
per set, V f ;s is the fracture volume, N f ;s is respective to matrix
projection relatively to pressure block gradient in fracture plan.
However, this method is only valid for well-connected and high
density DFN. The second method, numerical solution, consists in

http://dx.doi.org/10.1016/j.petrol.2015.04.018
0920-4105/& 2015 Elsevier B.V. All rights reserved.
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50 35

solving a steady-state flow problem, under boundary conditions, define guidelines for selecting the proper flow modeling approach
on the discrete fractured network with application of Poiseuille's of a fractured reservoir. This methodology identifies three drive
formula for fracture flows. This method takes into account the full criteria for the selection of a proper flow simulation model: the
geometry of the system but requires a larger CPU simulation time. fracture scale compared to simulator grid cells; the connectivity of
Because of the high computational effort, this method is normally the fracture network and continuity of matrix medium; and the
applied only to small density of fractures.
Another issue in upscaling procedures is related to the multi-
scale fractures length relative to simulation grid block size. Lee
et al. (2001) demonstrated that long fractures, whose scale is much
larger than grid block, have more influence in flow than short or
medium scale fractures. Any homogenization method will under-
estimate the real effect of these geological features. Small scale
fractures can be modeled as part of an equivalent medium using
dual porosity simulators or implicitly in matrix component. Some
works, for instance Li and Lee (2006), consider a numerical method
to upscale medium discrete fractures and an analytical method to
upscale small scale fractures. In addition to fracture length scale,
flow properties of fractures and interaction with matrix medium
can be important to determine a representative model (Single
Porosity system (SP), Dual Porosity system (DP), Dual Permeability
(DK) or explicit fault modeling).
Setting up the flow model for a naturally fractured reservoir is
still a challenge. Bourbiaux (2010) developed a methodology to Fig. 2. Main drives of methodology.

Fig. 1. Overview of challenges in fractured carbonate reservoirs.

36 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50

Fig. 3. DFN upscaling issue for vugs, given limitation in grid cell size.

Fig. 4. DFN upscaling issue for vugs, given limitation in grid cell size.

Fig. 5. DFN upscaling issue for vugs, given limitation in grid cell size.

time scale of flow interaction between fracture and matrix medium
compared to the time scale of fluid transport within fracture
medium. This information can be useful for selecting the proper
flow model approach, but with uncertainties. Another challenge
issue is the possible presence of multi-scale fractures and vugs in
the same cell volume, which is not well addressed in Bourbiaux
(2010) methodology.
Following Aguilera (2001), natural fractures can be classified, by
source, as tectonic fractures, regional fractures, contractional frac-
tures or surface related fractures. Tectonic and regional fractures
are normally associated to large scale faults distribution and
regional patterns, respectively. Surface related fractures do not
appear important from the point of view of petroleum production.
Contractional fractures can be a result of desiccation (probably of
minimal importance to direct hydrocarbon production), syneresis
(closely, regularly spaced, isotropically distributed), thermal con-
tractional (probably rare at depth in the subsurface) and mineral
phase (often irregular geometry). So, fractures distribution is
normally related to a regional pattern.
Well-test information, mainly pressure behavior, can be useful
to identify the presence of a double porosity behavior. This kind of
data can reveal the large scale behavior of reservoir (double
porosity or homogeneous behavior) and gross properties, which
can be useful to reduce uncertainty at flow model simulator
approach.
Under transient conditions, Warren and Root (1963) develop a
procedure for fractured reservoir evaluation, which is based on a
similar processing approach as used in a conventional reservoir.
For single porosity reservoirs, or homogeneous behavior, the well
pressure vs. log time is represented by a straight-line. For double
porosity reservoirs, the pressure plotted vs. log time exhibits a
double straight-line over three stages of different behavior. This
behavior should be obtained before the effect of the reservoir
boundary at the well, which could occur during the later stages of
pressure build-up. So, caution needs to be taken for the existence
of additional conditions (injection fluid or finite reservoirs, for
example). Pressure drawdown behavior in the presence of inter-
ference tests like fluid injection, has been studied by some authors
(for instance, Prat, 1990).
The terminology used for fracture classification is many times
confuse and varies significantly, especially considering the subject
area, reservoir simulation or geology characterization. In this work,
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50 37

we consider diffuse fractures as the minor fractures and sub- represents an overview of the complexity of fractured carbonate
seismic fractures the intermediate length scale. reservoirs development. The first two images of the overview illustrate
Some of the challenges in carbonate reservoirs development can be an isolate vugular system, diffuse and sub-seismic fractures as discrete
exemplified by flow simulation of a carbonate geologic scenario. Fig. 1 properties and their representation in refined cells (50  50  50 cm3)

Fig. 6. Methodology approach.

Fig. 7. Scale hierarchy approach.

38 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
Fig. 8. Reference model definition considering M3 constant.
Table 1
DFN characterization.
Type of fracture Frac. Set Density
(m2/m3)
Sub-seismic Frac. 1 0.015
Sub-seismic Frac. 2 0.015
Sub-seismic Frac. 3 0.02
Sub-seismic Frac. 4 0.02
Diffuse Frac. 5 0.2
Diffuse Frac. 6 0.2
Vugs 7 0.2
within a model with dimensions close to a simulator grid block
(100  100  5 m3). Water front shows a highly heterogeneous beha-
vior, and average pressure drop indicates a possible dual porosity
behavior. This example shows the reservoir behavior complexity that
could occur within a small space volume (nearly a block simulator
dimension) of a possible carbonate geologic scenario. Besides flow
complexity, the presence of multiscale heterogeneities can lead to
different upscaling approaches in the same flow unit.
This example shows several challenges: how to represent
properly this complexity in a simulation block? Which upscaling
procedure is the best approach? How to apply different upscaling
procedures in the same flow unit for different heterogeneities?
Which flow simulation model represents properly this behavior?
How to assign this complexity in a full field in a reasonable
development time?
A detailed methodology needs to be applied in order to consider
all these challenges in a reasonable development time.
Despite the carbonate reservoir recovery factors being usually
smaller than siliciclastic reservoirs, they show geological and
petrophysical characteristics that induce into characterization,
production and management challenges. The presence of multi-
scale heterogeneities and the higher number of uncertainties and
parameters induce in several challenges: geostatistical modeling of
small scale heterogeneities like vugular porosity; preserve the
multiscale behavior of reservoir by upscaling techniques and
pseudo-functions; and, uncertainty in the proper numerical flow
model that represents the dynamic behavior. In the flow simula-
tion studies of production prediction in this type of reservoirs, an
unsuitable execution of the previous steps can induce into devel-
opment risks for carbonate reservoirs.
Length h/z Azimuth Aperture
(m) (m)
100/10 0 0.0005
100/10 90 0.0005
30/5 0 0.001
30/5 90 0.001
3/1 0 0.0004
3/1 90 0.0004
0.3/0.3 – 0.4
Over the last decades there has been an increase in research
development for accurate geostatistical modeling approaches
(Strijker et al., 2012; Asadi-Eskandar et al., 2013), upscaling proce-
dures (Oda method) (Oda, 1985; Lough et al., 1997; Ding et al., 2006;
Li and Lee, 2006) and flow models (Lee et al., 2001; Bourbiaux,
2010), regarding naturally fractured carbonate reservoir. However,
there is a lack of studies regarding a methodology that: (1) focus in
integrating all these stages considering the multiscale complexity of
carbonate reservoirs; (2) allows the use of a reference solution
within a reasonable CPU time consumption; (3) focus in an
hierarchical upscaling method by using accessible commercial soft-
ware's; (4) is independent of the heterogeneities scale and hetero-
geneities type present in the reservoir.
An example is the recent methodology by Bourbiaux (2010) to
define guidelines for selecting the proper flow modeling for fractured
reservoirs. It is only applied considering fractures and not validated
under the presence of vugs or matrix. Furthermore, it does not
address the presence of multiscale heterogeneities in the same flow
unit. Oda method, (Oda, 1985), is also the conventional upscaling
procedure for discrete fracture networks used by industry, but is only
valid under some restrictions. The use of a reference solution could be
useful for Oda method validation or chose a proper upscaling method
based on a numerical approach (Lough et al., 1997).
However, Bourbiaux methodology (Bourbiaux, 2010) and Oda's
solution (Oda, 1985) continues to be some of the most important
studies applied for nowadays carbonate reservoirs development. This
work does not aim to improve those methods, but to integrate and
complement their applicability for a fully reservoir development.
Additionally, most of works that addresses upscaling proce-
dures are generally developed considering only static properties
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50 39

Fig. 9. Studied geological scenario.

Fig. 10. Vugs and diffuse fractures as discrete property. Fig. 11. Sub-seismic fracture as discrete property.

(porosity, net to gross, permeability). Even numerical upscaling
procedures for static properties, which are flow-based approaches
based on finite differences, do not address the dynamic behavior
which is always dependent from rock/fluid data (capillary pres-
sure and relative permeability). The rock wettability could result
in different upscaling and match procedures as the presence
or absence of imbibition forces in water-wet rocks. So, for a
proper upscaling approach the use of a reference model is always
necessary especially for water-wet cases as imbibition proce-
sses are not fully addressed by dual porosity formulations
(Rangel-German et al., 2010). In order to integrate static and
dynamic properties in upscaling procedures, a hierarchical ups-
caling approach is necessary in order to define a reference solu-
tion (refined grid), with a reasonable numerical simulation time
consumption, and get sequential control over static properties
and pseudo functions (pseudo-relative permeability or capillary
pressure curves for numerical match between the fine and coarse
grid) along the upscaling procedure.
The main purpose of this study is to develop a methodology to
select a suitable representative simulation flow model and upscaling
40 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50

Fig. 12. Permeability for model M1. Fig. 14. Permeability for model F1.

Fig. 15. Relative permeability curve for M1.

Fig. 13. Porosity for model M1.

procedure, according to small and large scale carbonate reservoir

heterogeneities, in a reasonable development time.

2. Methodology

The development of this work follows three elementary steps:

(1) The division of reservoir into flow units, (2) geostatistical
modeling, (3) upscaling procedures and (4) flow simulation mod-
eling of carbonate reservoirs. Fig. 2 shows a simplified illustration
of methodology. Geostatistical modeling is related to discrete
fracture network modeling. In the upscaling procedures we define
the average porosity and permeability for fracture and matrix
system, separately, and the reference model for upscaling matches.
The flow simulation modeling is the final step in which is defined
the numerical flow model that matches the reference model. A
detailed description of the methodology is provided below:
1. The first step follows the division of reservoir into flow units. A
flow unit is a portion of reservoir within which geological and
petrophysical properties that affect the flow of fluids is distinct
from the other reservoir units. The flow units separation is not
addressed in this work because this step should follow a
complex geologic framework (identify the major lithofacies,
regional fracture behavior, etc.), that is not the goal of our
study. The next steps must be applied for the integration of each
flow unit in reservoir simulation.
2. The geostatistical modeling follows mainly the stochastic discrete
fracture network (DFN) modeling technique. Besides diffuse and
sub-seismic fractures, vugs are also treated in discrete fracture
network. The common software used for geostatistical modeling
has limitation at block grid size (40 cm) for the upscaling of
discrete fractures to grid properties. In DFN upscaling procedures,
Fig. 16. Relative permeability curve for F1.
a small porous volume of fractures is enough for high perme-
ability. This means that if a small porous volume of a vug
intercepts a grid cell size, it is enough for an output high
permeability after applying DFN upscaling for the chosen grid
block size. Given the small size of vugs, a careful treatment
associated to a cut off procedure is necessary in order to preserve
the isolated behavior of vugs in the reference grid. Fig. 3 exem-
plifies this issue. A small portion in the blue vug is crossing some
grid blocks with 40 cm size. The output permeability after DFN
upscaling shows a residual permeability that could be enough for
some connectivity to the other vug. Fig. 4 shows the applied
correction. The small portion of vugs that intercepts other grid
blocks is removed, and consequently, the isolated behavior of vugs
is preserved. However, it is necessary to take into account that real
vugs length or intensity is always an impossible task to define as
deterministic. The uncertainty in this kind of small scale data is
always present. Thus, the loss of information in this stage due to
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
Table 2
Scale approach for models and grid block.
Model dimension
M1 100  100  5 m3
M2
M3
F1 500  500  5 m3
F2
R 500  500  5 m3
FU 500  500  5 m3
Fig. 17. Oil recovery for M2.
the cutoff procedure is irrelevant comparing to unknown informa-
tion. Fig. 5 shows the results if a refined grid, smaller than vugs
size, could be applied. In that case, the output permeability should
preserve the full isolate behavior of vugs, without requiring any
data treatment. Flow based upscaling techniques can also auto-
matically solve this issue, but they are totally unreliable given
computational restrictions for refined grids. Some studies
(Khvoenkova and Delorme, 2009; Gosselin et al., 2010) compare
the time consumption for analytical and numerical upscaling
methods for discrete fracture networks. The software guide used
in this work also warns to CPU restrictions for numerical upscaling
methods.
3. Given the presence of multiscale geological heterogeneities
and computational limitations, the upscaling procedure is
separated in function of heterogeneities scale (Fig. 6). The
scale hierarchy (Fig. 7) is subjective because it depends on the
block simulation size that reservoir engineers are prepared to
use because of hardware and software restrictions. This work
suggests a hierarchical upscaling method based on heteroge-
neities scale. Besides the heterogeneity scale, the minimum
block size is always dependent on the model dimension under
study as the number of blocks also impacts in CPU perfor-
mance. For small scale features, the reference model and
respective block dimension has to be smaller than larger scale
features. Following Bourbiaux (2010) concepts, the term “het-
erogeneities scale” is a function of simulation grid block
stablish by the user. The reason for this concept is related to
the fact that upscaling numerical formulations are linked to
heterogeneities scale in function of block size. Another pre-
caution is relative to grid orientation. If fracture system follows
an azimuth pattern the grid axis should be aligned with the
same azimuth direction. This condition prevents numerical
corrections relative to the computation of effective perme-
ability tensor for x and y axis.
41
Grid-cell-size
40  40  50 cm3
10  10  5 m3
100  100  5 m3
1  1  1 m3
100  100  5 m3
1  1  1 m3
100  100  5 m3
Fig. 18. Water cut for M2.
Fig. 19. Pseudo relative permeability curve for M2.
The upscaling procedure is divided into three steps: Upscaling
process M, Upscaling process F and the definition of a global
reference model R.
3.1. M represents the upscaling process for matrix heterogene-
ities with dimensions much smaller than simulation block
scale (depositional porosity, vugs, diffuse fractures, etc.) in a
region relative to a simulation block (up to 100 m). For flow
representation purposes, two virtual wells (injector and
producer) are used. M1 is the reference model and repre-
sents the discrete fracture network and/or depositional
porosity as a fine grid property. M2 is an intermediate model
used to match the flow behavior since M3 represents only
one block. In this case two virtual wells are used for flow
validation. M3 needs to be matched with M2 by histograms
of porosity and permeability since it is only one block and
consequently unreliable for flow match purposes. M3 repre-
sents the equivalent porosity and permeability of small scale
heterogeneities in a reservoir simulation block (up to
100 m). M2 is only important for cases with minimum
permeability, for flow match purposes. Otherwise, the direct
42 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50

Fig. 20. Overview of upscaling process M for vertical and horizontal permeability.

Fig. 21. Overview of upscaling process M for porosity.

Fig. 22. Horizontal permeability (mD) for F2 by Analytical method. Fig. 23. Permeability (mD) for F2 by Numerical method.

transition from M1 to M3 is necessary by using histograms
to match.
3.2. F represents the upscaling process for fracture scale close to
simulation block (up to 150 m) in a region relative to a flow
unit - FU (greater than 500 m). F1 is the reference grid for
upscaling process F and represents the permeability ten-
sors of discrete fracture network in a refined grid. This data
is only applied for definition of reference solution. The
upscaling of discrete fracture network to F2 by conventional
procedures (analytical or numerical methods) computes
the tensor of fracture permeability for the coarser grid (F2),
porosity, and a sigma factor property that is related
to connectivity between fractures. For high connected
fracture systems it is possible to directly infer that Oda's
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
Fig. 24. Sigma factor.
Fig. 25. Porosity for reference model “R”.
Fig. 26. Horizontal permeability (mD) for reference model “R”.
solution is always the best method, given computational
restrictions of flow-based techniques (numerical method).
This connectivity can be deduced by well test analysis in
the previous discrete fracture network (not the purpose of
this work) or analysis of sigma factor property. In that case,
the validation of upscaling procedure from DFN to F2 can be
done at this stage. For lower connected systems or uncer-
tainty cases it is required the comparison between flow-
based techniques and Oda's method.
43
Fig. 27. Average reservoir pressure for reference model “R”.
Fig. 28. Water front for “R”.
Fig. 29. Water front for “R”.
The computation of porosity and permeability for both M3
and F2, separately, is the first stage of upscaling process and
it will be very important to define later the proper simula-
tion model approach.
3.3. The next stage in upscaling process is to define a reference
model (R) with both models from upscaling process M and F. R
is a single porosity fine grid model and is the combination of
44 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50

Fig. 30. Close up in water front showing the dual permeability flow behavior and respective rock type.

Fig. 31. Oil recovery for FU.

Fig. 33. Reservoir pressure for FU.

Fig. 32. Water cut for FU.

fractures from F1. The pore volume occupied by F1 needs to be

M3 and F1. The average porosity and permeability calculated in
block M3 is assigned among void between fractures from F1
(Fig. 8). As M3 is defined for a smaller space volume than F1, it
can be assumed as constant in its spatially distribution over
model F1, which is geologically accepted given the tendency of
horizontal connectivity or can be defined as necessary in order
to fill the empty space between fractures from F1. This could be
an advantage for uncertainty or highly heterogeneous cases. It
is important to refer that a pore volume correction needs to
be done in the reference model because the average poro-
sity from M3 has not been defined considering the crossing
subtracted from M3 when defining the reference model.
4. The transition from R to FU is the final step and it is related
to flow simulation modeling stage. The flow simulation
modeling stage has the purpose of specifying if FU can be
treated as dual porosity, dual permeability system or as a
single porosity medium, and determine the dynamic and
static properties linked to model FU. The flow simulation
modeling stage is divided into two steps: validation and
numerical adjustment.
4.1. The first step of the validation process consists in analyz-
ing the output reservoir parameters (water cut, water front,
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50 45

Fig. 34. Oil recovery for FU.

Fig. 37. Water front for FU in a single porosity version after 6 months.

Fig. 35. Water cut for FU.

Fig. 38. Water front for FU in a single porosity version after 1 year.

Fig. 36. Average reservoir pressure for FU.

reservoir pressure) of reference solution (R). These output

reservoir parameters gives the indication of a possible dual
porosity, dual permeability or single porosity behavior. A
careful investigation needs to be considered in well pres-
sure analysis since the possible three stages in pressure
drop that characterize a dual porosity behavior is only valid
for infinite boundary conditions and in the absence of fluid
injection. The guidelines defined by Bourbiaux (2010) for
selecting the proper flow model for fractured reservoirs
have great applicability at this stage. The simulation time
Fig. 39. Water front for FU in a single porosity version after 2 years.
step ðΔtÞ required to describe the flow progress in the
fracture medium is compared to the time (te) required to
occur a matrix-fracture fluid transfer. If Δt o t e then
matrix-fracture transfers are delayed with respect to the
fluids transfer taking place in the fracture network. The use
46 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50

Table 3
Simulation time consumption.

Simulation time consumption

(seconds)

Reference model (R) 1,271,545

Dual permeability (FU) 1.92
Dual porosity (FU) 1.33
Single porosity (FU) 0.61

Fig. 40. Capillary pressure transformation.

Fig. 43. Water cut for FU.

Fig. 41. Pseudo relative permeability curve for FU in matrix component.

Fig. 44. Average reservoir pressure for FU.

network all over reservoir, the dual permeability flow

Fig. 42. Oil recovery for FU.
of a dual porosity flow model is recommended in this case.
If Δt 4 t e then matrix blocks exchange fluids very rapidly
with the fracture network. This quasi-static equilibrium is
very often satisfied by a single porosity flow model. For
cases where fractures do not constitute a connected
model offers the possibility to consider the flow in both
non-fractured regions and well-connected fractured net-
works (Bourbiaux, 2010). The time can be estimated by
analytical solutions. Otherwise, the use of fine-grid models
can be a numerical solution. In this work we propose the
use of refined grids to evaluate the proper flow model.
4.2. The second step of the validation process consists in
applying the expected flow model, from the previous
analysis, for FU case and compare if is the best approxima-
tion to reference solution. If the expected flow model is a
dual porosity or a dual permeability system, FU results in F2
(fracture component) and M3 (matrix component). If the
expected flow model is a single porosity system, some
average between F2 and M3 should be an approximate
solution.
4.3. The final step is to adjust the flow model selected with the
reference solution.
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50 47

Fig. 45. Water front for FU in fracture component after 6 months. Fig. 48. Water front for FU in matrix component after 1 year.

Fig. 46. Water front for FU in matrix component after 6 months.

Fig. 49. Water front for FU in fracture component after 2 years.

Fig. 47. Water front for FU in fracture component after 1 year. Fig. 50. Water front for FU in matrix component after 2 years.

Well index calibration is also necessary due to the well information

It is unlikely that the previous combination F2 and M3 gives a lost in upscaling procedures. In that way, the use of pseudo
perfect match with the reference solution because there are functions (capillary pressure or relative permeability curves) and
physical phenomena (imbibition, gravitational forces, etc.), not well index calibration should be applied in order to get a good
only related to static properties, that could lead to a poor match. match with the reference solution. If the expected flow model is a
48 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
single porosity system it is unreliable to use more than one relative
permeability curve for the same block. Some studies (ex:. Lingen et
al., 2001) presents a methodology to obtain a pseudo relative
permeability curve for reservoir simulation model that represents
the flow behavior of a reference solution. This detailed study is not
the focus of this work, but some simplified concepts could also be
applied. The relative permeability for fractures is normally
assumed as two straight-lines functions with endpoints at zero
and 100% saturation. It is predictable that the fracture relative
permeability curve is not suitable if the terminal points from the
matrix relative permeability curve are different, as it leads to
different initial conditions from the reference solution, in terms
of fluids volume. So, the terminal points for the single flow model
approach have to follow matrix relative permeability curve in order
to have the same initial fluids volume distribution. A practical
approach to modify the original matrix relative permeability curve
is by using the corey exponents. It is also predictable that the final
relative permeability curve should be a transition between the
fracture and the matrix relative permeability curve. After all these
procedures, FU represents the proper representation of a fractured
carbonate reservoir flow unit in the reservoir simulation scale.
The model S is the full simulation model, it represents a set of
flow units but this work includes the application only to one flow
unit in order to exemplify the procedure.
Fig. 51. Overview of main results.
3. Application
The geological scenario considered to exemplify the methodol-
ogy approach assumes that vugs provide the essential porosity of
the reservoir and permeability is provided by diffuse and sub-
seismic fractures. Vugs and diffuse fractures are defined in upscal-
ing process M, since its size is much smaller than simulation grid
block approach in this case (100 m). Table 1 resumes the geological
scenario characterization. Fig. 9 represents an overview of the
geological scenario and respective scale. Fig. 10 represents vugs
and diffuse fractures as discrete properties. Fig. 11 represents sub-
seismic fractures in discrete form. Fig. 12 and Fig. 13 represent both
vugs and diffuse fractures, in a representative grid (M1). Fig. 14
represents sub-seismic fractures in a representative grid (F1).
Fig. 15 and Fig. 16 represent the relative permeability curve for
M1 and F1, respectively.
Table 2 resumes the ranking of scales for model dimension and
respective grid-cell-size.
4. Results and discussion
From the transition from M1 and M2 the output reservoir para-
meters (oil recovery factor and water cut) show that the numerical
 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
upscaling approach has irrelevant differences with analytical mean
(geometric mean) for permeability upscaling approach (Fig. 17 and
Fig. 18), so both methods could be applied, despite the higher
computational time for the numerical upscaling procedure. Using a
pseudo relative permeability curve it is possible to obtain a perfect
match. Fig. 19 represents the pseudo relative permeability curve for M2.
The difference between the initial relative permeability curve and
pseudo relative curve is small due to the small adjustment required to
match both models. In the transition from M2 to M3, the arithmetic
mean is applied for porosity, geometric mean for horizontal perme-
ability and harmonic mean for vertical permeability. M3 is defined by
an average horizontal and vertical permeability value of 187 mD and
60 mD, respectively, and, an average porosity value of 7%. These
petrophysical values correspond to the permeability and porosity
provided from heterogeneities much smaller than the simulation cell
size. Fig. 20 and Fig. 21 represent an overview of upscaling process M
for permeability and porosity, respectively.
Fig. 22 and Fig. 23 represent the upscaled model F2 by analytical
and numerical approach, respectively. The boundary condition
applied for numerical approach is the flow restriction through
the cell sides. It is possible to induce by a previous analysis that
permeability from analytical upscaling approach is overestimated,
given the major differences with numerical approach. Fig. 24
represents the sigma factor which is related to connectivity
between fractures. This data is used only if in the final model a
dual porosity or a dual permeability system is selected. Fig. 25 and
Fig. 26 represent the reference model R with a spatial distribution
of homogenized petrophysical values from M3 (for matrix) and
petrophysical values from the reference model for sub-seismic
fractures, F1 (for fractures).
Fig. 27 represents the average reservoir pressure drop for the
reference model R. Reservoir pressure drop has typical behavior of
dual porosity medium, since three stages could be distinguished in
reservoir behavior. This analysis is not deterministic as the model
does not follow the boundary conditions for well test analysis
given the presence of fluid injection and reservoir finite bound-
aries. Water front (Fig. 28 and Fig. 29) shows a delayed water
advance in fracture comparing to water advance in matrix. Fig. 30
shows a close up in water front and the respective rock type. The
good capillary continuity between matrix blocks in contrast to poor
connectivity between fractures leads to this flow progress differ-
ence. Furthermore, the medium matrix permeability in arrange-
ment to a water-wet rock leads to high imbibition rates and
delayed flow progress in the fractures. This means that a small
water saturation in a fracture is enough for high imbibition rates in
matrix medium (M3). The fractures are partially connected and
water imbibition in matrix medium follows fracture water satura-
tion only for some regions in reservoir. For regions with fractures
not fully connected, water front takes place essentially in matrix
medium. In this first stage of validation, it is possible to presume a
dual permeability behavior. This analysis is also supported by
Bourbiaux (2010) guidelines to choose a proper flow model.
Figs. 31–33 show the reservoir match results for a dual perme-
ability and dual porosity approach using the upscaled model F2 by
analytical and numerical approach. The dual permeability model
using the numerical upscaling procedure is the best approach for
this case. Oda´s method overestimates the fracture permeability
tensor value given the poor connectivity of fracture network. Given
that, the oil recovery factor is lower for numerical upscaling
approach since the time step related to flow progress in fracture
medium is much smaller than the simulation time step required
for full matrix-fracture fluid transfers, which justifies the oil that is
left behind. The numerical approach is a proper approach due to
the poor connected fracture network. The dual permeability flow
model is also a better approach since dual porosity system can-
not take into account the flow between matrix blocks, which is
49
important for this case because of the higher matrix permeability
(km ¼187 mD). Given the lack of connectivity in fracture compo-
nent, the dual permeability flow model enables the flow in matrix
blocks where fractures are not well connected and the flow in
blocks where fractures behaves as flow channels. Although the
dual permeability approach provides a good match with reference
model, the single porosity approach should always be verified. The
reference model shows that connected fractures provide trends for
water front, and in regions with non-connected fractures the flow
takes place in matrix. A possible single porosity approach should
follow this idea. For regions with non-connected fractures the
permeability is equal to M3 and for connected fractures should be
equal to F2. Figs. 34–36 show a good match between a single
porosity approach and reference model for water cut, oil recovery
and reservoir pressure drop. The single porosity approach results
from an average between fracture and matrix permeability and a
pseudo relative permeability curve. Despite the good match in
output reservoir parameters, the concern is related to water front
(Figs. 37–39) which cannot reflect the heterogeneity provided by
the difference in flow progress between fracture and matrix system
in reference model. Connected fractured regions provide trends for
flow progress in fracture and similarly in matrix component
because high matrix permeability values provide high imbibition
recovery from matrix and also continuity between matrix blocks. A
single porosity flow model can give an approximation of flow
reservoir complexity by average techniques or pseudo-functions
but a dual permeability flow model can directly give a perfect
match, despite of computational time (Table 3). It can be possible
to improve the water front in the single porosity model by other
kind of pseudo-functions or by block to block adjustments, but it
could be numerically expensive. The second stage of validation
shows that the dual permeability flow model is the best approach.
Despite the good approximation between the dual perme-
ability flow model and the reference, it is possible to improve the
match by a pseudo relative capillary pressure, pseudo relative
permeability curve (Fig. 40 and Fig. 41) and well index calibra-
tion. Figs. 42–44 show the match improvement of the chosen
dual permeability flow model. Figs. 45–50 show the water front
for the match case, in fracture and matrix component. In a global
point of view it is possible to observe a good match with water
front in the reference model, despite the larger simulation block
dimension (100 m) that mask the resolution. For some specific
points the water front does not follow the reference model due
to wrong approximations in effective permeability derived from
the flow based upscaling approach. This event shows that flow
based upscaling techniques are also not fully self-assured and
different boundary conditions can give different effective
permeabilities.
Resuming, the flow unit (FU) is represented by a dual perme-
ability flow model with heterogeneities smaller than simulation
block represented implicitly in matrix component by analytical
upscaling approach and with sub-seismic fractures near simulation
block scale represented in fracture component by numerical
upscaling method. Fig. 51 represents an overview of the proce-
dures, averaging techniques and pseudo-function applied over the
upscaling procedure. How much information is lost if we ignore all
these steps?
5. Conclusions
This work performs a methodology aiming a suitable representa-
tion of simulation flow model and upscaling procedures approach
according to multiscale carbonate reservoir heterogeneities. A complex
geological scenario was used in order to exemplify the methodology
approach.
50 M.G. Correia et al. / Journal of Petroleum Science and Engineering 131 (2015) 34–50
The developed methodology presents several advantages:
 Get sequential control over static properties and pseudo func-
tions along upscaling procedure as the upscaling is ranked by
heterogeneities scale;
 Simplification of the final match procedure as the match is done
sequentially by heterogeneity scale, over the upscaling;
 Allows the use of a reference model within a reasonable simu-
lation time consumption as the upscaling is ranked by repre-
sentative units (flow units);
 Reduction of the uncertainty in the accurate simulation flow model;
 Improvement of the integration of multiscale heterogeneities in
reservoir simulation;
 Extends Bourbiaux (2010) approach to vugular and matrix
systems;
 Integrate the upscaling of dynamic and static properties.
For this case study, the dual permeability flow model is the best
approach because of several observations:
 Both flow progress in matrix and fracture occur at different time
steps which make difficult the single porosity approach;
 Reservoir pressure drop has the typical three stages of a dual
porosity reservoir;
 Dual porosity approach underestimates oil recovery because of
high permeable matrix medium;
 Dual permeability flow model can take into account both flow
progresses in connected fracture networks and in matrix medium;
For a proper representation of a carbonate reservoir in flow
simulation, important steps should be followed:
 Upscaling procedure must be done differently by flow units and
consequently by type of heterogeneity scale;
 Flow progress and reservoir pressure drop analysis of a refer-
ence solution is helpful to define the flow model selection;
 Pseudo-functions are normally necessary to match procedures
in fractured carbonate reservoirs because of dynamic phenom-
ena (imbibition, gravitational forces, etc.) that cannot be
upscaled by the common averages used for static properties.
This work presents a methodology and an analysis that can be
useful for multidisciplinary areas of expertise since it integrates
geostatistical modeling of carbonate reservoir heterogeneities with
reservoir simulation.
Acknowledgments
The authors are grateful to the Center of Petroleum Studies
(Cepetro-Unicamp/Brazil), PETROBRAS S/A (Grant Agreement
No.0050.0022715.06.4), UNISIM and the Petroleum Engineering
Department (DEP-FEM-Unicamp/Brazil) for their support of this
work. The authors are also grateful to Schlumberger Information
Solution for the use of Petrels.
References
Aguilera, N., 2001. Geologic Analysis of Naturally Fractured Reservoirs. Gulf Publish-
ing Company.
Asadi-Eskandar, A., Rahimpour-Bonab, H., Hejri, S., Afsari, K., Mardani, A., 2013.
Consistent geological-simulation modeling in carbonate reservoirs, a case study
from the Khuff Formation, Persian Gulf. J. Pet. Sci. Eng. 109, 260–279.
Ahr, Wayne M., 2008. Geology of Carbonate Reservoirs: The Identification, Descrip-
tion, and Characterization of Hydrocarbon Reservoirs in Carbonate Rocks. John
Wiley and Sons, New Jersey.
Bourbiaux, B., 2010. Fractured reservoir simulation: a challenging and rewarding
issue. Oil Gas Sci. Technol. – Rev. IFP 65 (No. 2), 227–238.
Dershowitz, Bill, LaPointe, P., Eiben, T., Wei, L., 1998. Integration of discrete feature
network methods with conventional approaches. SPE 49069. In: Proceedings of the
Annual Technical Conference and Exhibition, New Orleans, September 27–30.
Ding, Y., Basquet, R., Bourbiaux, B., 2006. Upscaling fracture networks for simulation
of horizontal wells using a dual-porosity reservoir simulation. SPE 92774. In:
Proceedings of the Reservoir Simulation Symposium, January 31.
Gosselin, O., Cottereau, N., Garcia, M., Vigier, L., 2010. Effective Fracture Network
Permeability–Comparative Study of Calculation Methods. SPE Europec. SPE-
131126. Barcelona.
Khvoenkova, N., Delorme, M., 2009. Performance analysis of the hybrid fracture
media upscaling approach on a realistic case of naturally fractured reservoir. In:
Proceedings of the International Petroleum Technology Conference, Doha, Qatar,
December 7–9.
Lee, S.H., Lough, M.F., Jensen, C.L., 2001. Hierarchical modeling of flow in naturally
fractured formations with multiple length scales. Water Resour. Res. 37,
443–455.
Li, L., Lee, Seong H. 2006. Efficient field-scale simulation of black oil in a naturally
fractured reservoir via discrete fracture networks and homogenized media. In:
Proceedings of the SPE 103901, Internacional Oil and Gas Conference, Beijing,
China, 5–7 December.
Lingen, P.v., Sengul, M., Daniel, J.M., Consentino, L., 2001. Single medium simulation
of reservoirs with conductive faults and fractures. SPE Paper 68165. In: SPE
Middle East Oil Show, Bahrain, March 2001.
Lough, M.F., Lee, S.H., Kamath, J., 1997. A new method to calculate effective
permeability of gridblocks used in the simulation of naturally fractured
reservoirs. SPE Reserv. Eng. (Soc. Pet. Eng.) 12 (3), 219–224.
Oda, M., 1985. Permeability tensor for discontinuous rock mass. Geotechnique
35 (4), 483–495.
Prat, G.D.A., 1990. Well Test Analysis for Fractured Reservoir Evaluation. Elsevier
Science Publishers, New York.
Rangel-German, E.R., Kovscek, A.R., Akin, S., 2010. Time-dependent shape factors for
uniform and non-uniform pressure boundary conditions. Transp. Porous Media
83 (3), 591–601.
Strijker, G., Bertotti, G., Luthi, S., 2012. Multi-scale fracture network analysis from an
outcrop analogue: a case study from the Cambro-Ordocivian clastic succession
in Petra, Jordan. Mar. Pet. Geol. 38, 104–116.
Warren, J.E., Root, P.J., 1963. The behavior of naturally fractured reservoirs. SPE J.
September, 245–255.