ResSimCh7 PDF

Permeability Upscaling
7
CONTENTS
1 SINGLE-PHASE FLOW 3.3.3 Geopseudo Example

1.1 INTRODUCTION 3.3.4 When to use the Geopseudo Method
1.2 Upscaling Porosity and Water Saturation 3.4 Uncertainty and Upscaling
1.3 Averaging Permeability 3.5 Upscaling Summary
1.3.1 Flow Parallel to Uniform Layers
1.3.2 Flow Across Uniform Layers 4 REFERENCES
1.3.3 Flow through Correlated Random Fields
1.3.4 Additional Averaging Methods
1.3.5 Summary of Permeability Averaging
1.4 Numerical Methods
1.4.1 Recap on Flow Simulation
1.4.2 Boundary Conditions
1.5 Upscaling Errors
1.5.1 Correlated Random Fields
1.5.2 Evaluating the Accuracy of Upscaling
1.5.3 Upscaling of a Sand/Shale Model
1.6 Summary of Single-Phase Upscaling
2 TWO-PHASE FLOW
2.1 Introduction
2.2 Applying Single-Phase Upscaling to a
Two-Phase Problem
2.3 Improving Single-Phase Upscaling
2.3.1 Non-Uniform Upscaling
2.3.2 Well Drive Upscaling
2.4 Introduction to Two-Phase Upscaling
2.5 Steady-State Methods
2.5.1 Capillary-Equilibrium
2.6 Dynamic Methods
2.6.1 Introduction
2.6.2 The Kyte and Berry Method
2.6.3 Discussion on Numerical Dispersion
2.6.4 Disadvantages of the Kyte and Berry
Method
2.6.5 Alternative Methods
2.6.6 Example of the PVW Method
2.7 Summary of Two-Phase Flow
3 ADDITIONAL TOPICS
3.1 Upscaling at Wells
3.2 Permeability Tensors
3.2.1 Flow Through Tilted Layers
3.2.2 Simulation with Full Permeability
Tensors
3.3 Small-Scale Heterogeneity
3.3.1 The Geopseudo Method
3.3.2 Capillary-Dominated Flow
Learning Objectives
After reading through this Chapter, the Student should be able to do

the following:
• Appreciate why upscaling is necessary.
• Know how to calculate effective permeability in simple models by

averaging.
• Understand how to perform numerical upscaling of single-phase flow.
• Be aware of the effects of heterogeneity on two-phase flow.
• Realise the limitations of applying single-phase upscaling to a two-phase

problem.
• Know how to carry out steady-state, capillary-equilibrium upscaling for two-

phase flow.
• Become familiar with two-phase dynamic upscaling (the Kyte and Berry
Method), and understand the advantages and disadvantages of applying dynamic
upscaling.
• Understand how to upscale around a well.
• Appreciate that permeability is a full tensor property.
• Know how to upscale from the core-scale to the scale of a geological model,
taking account of fine-scale structure and capillary effects.
2
7
1 SINGLE-PHASE FLOW
1.1 Introduction
Reservoir modelling often involves generating multi-million cell models, which
are too large for carrying out flow simulations using conventional techniques. The
number of cells must therefore be reduced by “upscaling” (Figure 1). Some quanti-
ties, such as porosity and water saturation, are easy to upscale, because they may
be averaged arithmetically. However, other quantities – notably permeability – are
much more difficult to upscale.
Full-field model
Geological model
Figure 1
Upscaling Example
We usually refer to the upscaled permeability as the effective permeability. The

effective permeability is defined as the permeability of a single homogeneous cell
which gives rise to the same flow as the fine-scale distribution when the same pressure
gradient is applied. We assume that Darcy’s Law holds at the coarse scale:
Ak eff ∆P
Q=−
µ ∆x (1)
where Q = total flow, A = area, keff = effective permeability, μ = viscosity, and ΔP/Δx
is the pressure gradient.
True effective permeability is an intrinsic property of the model and ought to be

independent of the applied boundary conditions (Section 1.4). However, in practice,
the effective permeability often does depend on the boundary conditions, and on the
method used for calculation. Upscaling must always be carried out with care in order
to obtain “sensible” results.
In Figure 1, the geological model on the left is a fine-scale model with 20 million cells,
and the coarse-scale model on the right consists of about 300,000 cells. Each of the
coarse-scale cells contains an effective permeability. An example of fine-scale and
coarse-scale grids is shown in the 2D model in Figure 2. An effective permeability
is calculated for each coarse-scale cell, either by averaging the fine-grid values, or
by performing a numerical simulation.
Institute of Petroleum Engineering, Heriot-Watt University 3

fine grid coarse grid
one coarse cell
Figure 2
The upscaling procedure
In this section of the upscaling course, we assume that there is only one phase present
– water or oil, and that we have steady-state linear flow. We show how simple aver-
aging may sometimes be used to estimate upscaled parameters, and then move on to
methods which involve numerical simulation. This is followed by a set of examples
which demonstrate how errors may arise, and how to avoid them.
1.2 Upscaling Porosity and Water Saturation

We start by averaging porosity and water saturation, using a simple model (Figure 3).
(Note that the water saturation is not required for a single-phase problem. However,
we include it here because it is simple to upscale.) There are 10 grid blocks of size
1 m3, 4 of which have a porosity of 0.15, and 6 of which have a porosity of 0.20.
φ = 0.15 φ = 0.20 Figure 3

Example for averaging
porosity and water
saturation
Sw = 0.50 Sw = 0.40
The average porosity, φ , is given by:
total pore volume

φ=
total volume (2)
Therefore, in this case:
4 × 0.15 + 6 × 0.20
φ= = 0.18.
10
When averaging the water saturation, we need to take the porosity into account. In the
previous example, suppose the water saturation was 0.5 in the blocks with porosity of
0.15, and 0.4 in the blocks with porosity 0.2, then the average water saturation is:
total amount of water

Sw =
total pore volume (3)
Here, the average water saturation is:
4
7
4 × 0.15 × 0.5 + 6 × 0.20 × 0.4
Sw =
4 × 0.15 + 6 × 0.20
0.3 + 0.48 0.78
= = = 0.433.
1.8 1.8
1.3 Averaging Permeability

In some simple models, such as parallel layers or a random distribution, the effective
permeability may be calculated by averaging.
1.3.1 Flow Parallel to Uniform Layers
P1 P2
Qi ki, ti
Figure 4
Along-layer flow ∆x
Consider a set of (infinite) parallel layers of thickness, ti and permeability ki, where i
= 1, 2, .. n (the number of layers). The effective permeability of these layers is given
by the arithmetic average, ka.
n
∑t k i i
k eff = k a = i =1
n
∑t i
i =1 (4)
(Equation (4) may be proved by applying a fixed pressure gradient along the layers.)
Example 1
x 1 t1 = 3 mm, k1 = 10 mD
Figure 5 2 t2 = 5 m
mm, k2 = 100 mD
z
A simple, two-layer model
Suppose we have two layers as shown in Figure 5. The effective permeability for
flow in the x-direction is given by Equation (4), and is:
3 × 10 + 5 × 100 530
ka = = = 66.25 m
mD
3+5 8

1.3.2 Flow Across Uniform Layers
∆Pi ki, ti
Figure 6
∆x Across-layer flow
For flow perpendicular to the layers, the effective permeability is given by the har-
monic average, kh:
n
∑t i
k eff = k h = i =1
n .
ti
∑
i =1 k i (5)
(Equation (5) may be proved by assuming a constant flow rate through each layer.)
Example 2
Equation (5) may be used to calculate the effective permeability for flow across the
two layers in the model shown in Figure 5, i.e. flow in the z-direction.
3+5 8
kh = = = 22.86 mD
3 10 + 5 100 0.35
From Examples 1 and 2, we see that the permeability is different in different direc-
tions. In reservoirs with approximately horizontal layers, the arithmetic average
may be used for calculating the effective permeability in the horizontal direction,
and the harmonic average may be used for calculating the effective permeability in
the vertical direction.
1.3.3 Flow through Correlated Random Fields

Figure 7 shows an example of a correlated random permeability distribution. Corre-
lated random fields are described in Section 1.5.1. Basically, “correlated” means that
areas of high or low permeability tend to be clustered, so that the spatial distribution
is smoother than a totally random one. The “correlation length” is approximately the
size of patches of high or low permeability. The longer the correlation length, the
longer will be the range of the semi-variogram for the permeability distribution.
Assuming that we are averaging over many correlation lengths, permeability should
be isotropic (same in the x-, y- and z-directions). The effective permeability for a
random permeability distribution is proportional to the geometric average, which is
given by:
6
7
 n 
ln( k i )
∑ 
k g = exp i =1 
 n 
  (6)
where i = 1, 2, .. n is the number of cells in the distribution.
Correlation Length
Figure 7
A correlated, random
permeability distribution
(white = high permeability,
dark = low permeability)
The results given below have been derived theoretically for log-normal distributions,
with a standard deviation of σY, where Y = ln(k). The results depend on the number
of dimensions:
k eff = k g (1 − σ 2Y 2)
k eff = k g
k eff = k g (1 + σ 2Y 6)
in 1D
in 2 D
in 3D
} (7)
These formulae are approximate, and assume σY is small (< 0.5). (You are not required
to know the proof.) The 1D result is an approximation of the harmonic average.
Note that the results do not depend on the correlation length of the field, provided it
is much smaller than the system size.
Also note that ka > kg > kh, and the effective permeability always lies between the
two extremes: ka and kh.
Example 3
Figure 8
A random arrangement
of the permeabilties in the
simple example 75 cells of 10 mD 125 cells of 100mD

Suppose that the permeability values in the simple example are jumbled up, so that
there are 75 small cells of 10 mD, and 125 small cells of 100 mD. See Figure 8. The
effective permeability of this model is:
 75 × log(10 ) +125 × log(100 ) 

 
k eff = k g = 10 200
 75 + 250 
 200 
= 10
= 101.625 = 42.17 mD.
1.3.4 Additional Averaging Methods

Since averaging is very quick (compared with numerical simulation), many engi-
neers use this technique in more complex models. Sometimes engineers increase the
accuracy by using power averaging. The power average is defined as:
1/ α
 n α
 ∑ ki 
k p =  i =1  ,
 n 
 
(8)
where α is the power. The value of the power depends on the type of model, and
must be calibrated against numerical simulation (Section 1.4).
Also, sometimes, engineers use a combination of the arithmetic and harmonic aver-
ages, e.g. they take the arithmetic average of the permeabilities in each column and
then calculate the harmonic average of the columns.
1.3.5 Summary of Permeability Averaging

To summarise, there are two types of simple model in which we can calculate the
effective permeability by averaging:
• Parallel layers
• Correlated random fields
Since averaging is very quick, it is frequently used as an approximation for the effec-
tive permeability in more complex models.
1.4 Numerical Methods

In general, the permeability distribution will not be simple enough for us to be able
to calculate the effective permeability analytically (i.e. by averaging), and we will
have to perform a numerical simulation. We can use a finite difference method to
calculate the pressures.
1.4.1 Recap on Flow Simulation

The continuity equation tells us that there is no net accumulation or loss of fluid
within a grid block:
8
7
q xin + q zin = q xout + q zout . (9)
(We are assuming incompressible rock and fluids, here.)
x qzin
i,j-1
z
qxin qxout i-1,j i,j i+1,j

Figure 9
i,j+1
Recap on numerical flow
simulation qzout
Darcy’s law is used to express the flows in terms of the pressures and permeabilities.
For example, if the grid blocks in Figure 9 are of length Δx and height Δz (and unit
width in the y-direction), then:
q xin = −
(
k x, i −1/ 2, j ∆z Pi , j − Pi −1, j )
µ ∆x (10)
where kx,i-1/2,j is the harmonic average of the permeabilities in the x-direction in blocks
(i-1,j) and (i,j). (You now should know now why the harmonic average is used here.)
The other flows are calculated in a similar manner.
It is useful to use the transmissibilities, Tx = kxΔz/Δx and Tz = kzΔx/Δz. (Assume

the width, Δy = 1.) We can therefore derive the pressure equation:
(T x , i −1 / 2 , j )
+ Tx, i +1/ 2, j + Tz, i , j −1/ 2 + Tz, i , j +1/ 2 Pi , j
− Tx, i −1/ 2, j Pi −1, j − Tx, i +1/ 2, j Pi +1, j
− Tz, i , j −1/ 2 Pi , j −1 − Tz, i , j +1/ 2 Pi , j +1
=0 (11)
An equation is set up for each Pij, i = 1, 2, .. nx and j = 1, 2, .. nz. The transmissibilities

are known, and using the appropriate boundary conditions, we can solve this set of
linear equations to obtain the pressure in each grid block. The effective permeability is
than calculated from the total flow and the total pressure drop, as described below.
Note that the boundary conditions are applied to each coarse grid cell in turn, and
they may not be a good approximation to the pressures which would arise in a fine-
grid simluation. This leads to errors in the results. Upscaling errors are discussed
in Section 1.5.
1.4.2 Boundary Conditions

Boundary conditions are required to specify what happens at the edges of the
model.

a) No-Flow Boundaries
no flow through the sides
P1 P2
Figure 10
Fixed pressure, or no-flow,
boundary conditions
no flow through the sides
The pressure is fixed on two sides of the model, and no flow is allowed through the
others sides of the model. This type of boundary condition is suitable for models
where there is little cross-flow: for example, models with approximately horizontal
layers, or a random distribution. These are the most commonly applied boundary
conditions. Figure 11 illustrates how an effective permeability may be calculated
in the x-direction.
Pressure= P1 Pressure= P2
on left face y
on right face
x
Area, A
Flow Rate, Q z
Figure 11
The calculation of effective
permeability using no-flow
boundary conditions
L
1. Solve the steady-state equation to give the pressures, Pij, in each grid block.
2. Calculate the inter-block flows in the x-direction using Darcy’s Law. (See
Equation 10.)
3. Calculate the total flow, Q, by adding the individual flows between two y-z
planes. (Any two planes will do, because the total flow is constant.)
4. Calculate the effective permeability for flow in the x-direction, using the
equation:
k eff , x A( P1 − P 2)
Q=
µL (12)
Repeat the calculation for flow in the y- and z-directions, to obtain keff,y and keff,z.
(b) Periodic Boundary Conditions

Periodic boundary conditions are useful for calculating the effective permeabilities
in models where there are infinitely repeated geological structures in each direction.
(See Section 3.2.) The use of periodic boundary conditions ensures that we also
10
7
have an infinitely repeated pattern of flows and pressure gradients. In the example
shown in Figure 12, there is a net pressure gradient in the x-direction. The blocks
are numbered i = 1, 2, ..nx in the x-direction, and j = 1, 2, .. nz in the z-direction.
x
P(i,0) = P(i,nz)
Figure 12
P(0,j) = P(nx,j)+∆P P(nx+1,j) = P(1,j)-∆P
Periodic boundary
conditions P(i,nz+1) = P(i,1)
One advantage of using periodic boundary conditions, is that fluid can flow through
the sides of the model. This method can be used to calculate a full tensor effective
permeability (Section 3.2).
(c) Linear Pressure Boundary Conditions

In linear pressure boundary conditions (Figure 13), the pressure is fixed at each end,
as in the fixed-pressure boundary conditions. Then, the pressure at the edges of the
model is interpolated linearly from one side to the other. Like the periodic boundary
conditions, the linear pressure boundary conditions allow flow through the edges.
P1 P2
P1 P2
Figure 13
Linear pressure boundary
P1 P2
conditions
(d) Flow Jacket, or Skin

To reduce the effect of boundary conditions when calculating the effective permeability,
some engineers perform the simulations on a larger grid than necessary. The extra
grid blocks round the edges are referred to as a “jacket” or “skin”. See Figure 14.

boundary conditions applied
to outer edges of model
keff calculated for

this block Figure 14
Example of a flow jacket
round a model. In this case
the jacket is 4 cells thick
1.5 Upscaling Errors

In the process of upscaling, information about the fine-scale structure is lost, and
upscaling usually gives rise to errors. However, in some cases, the errors will be
larger than in others. In this section, we examine a series of models which show
examples of where upscaling is successful and where it is not.
1.5.1 Correlated Random Fields

We introduce the concept of a correlated random field. Although the permeability
distribution in real rocks may not follow this type of model, it is a useful way to
parameterise heterogeneity. We assume that the probability density function (pdf)
of the model is normal or log normal, as shown in Figure 15. In an isotropic model
(i.e. same in all directions), the field is then characterised by three parameters: the
mean, μ, the standard deviation, σ, and correlation length, λ. The standard deviation
determines the width of the pdf (i.e. the permeability contrast), and the correlation
length determines approximately the distance over which the permeability values are
similar. Figure 16 shows examples of the 4 models with varying σ and λ.
a) b)
0.06 0.06
0.05 0.05
Frequency
Frequency
0.04 0.04 Figure 15

0.03 0.03 Normal and log-normal
0.02 0.02 permeability distributions.
0.01 0.01 a) Normal distribution
0.00
40 60 80 100 120 140 160
0.00 with mean = 100 mD, and
0.5 1.0 1.5 2.0 2.5 3.0 3.5
Perm eabil ity (m D) log (p erm eabil ity) standard deviation = 20
mD.
c)
0.06 b) Log-normal distribution
0.05 with mean = 2.0, and
Frequency
0.04 standard deviation = 0.5.

0.03 c) Log-normal distribution
0.02 as above, but with
0.01 permeability plotted on
0.00
the x-axis, rather than
0 500 1000 1500 2000
Per m eab ility (m D) log(permeability)
12
7
a) small σ, small λ b) large σ, small λ
a) small σ, large λ b) large σ, large λ
Figure 16
Models with different
Permeability (mD)
standard deviations and
correlations lengths 0 50 100 150 200
1.5.2 Evaluating the Accuracy of Upscaling

One way to evaluate the accuracy of upscaling is to compare upscaling in two stages,
k 2eff , with upscaling in a single stage, k1eff , as shown in Figure 17. If upscaling is
2 1
accurate, then k eff = k eff . This is the case for upscaling along and across parallel
layers in the idealised models of Sections 1.3.1 and 1.3.2.
fine-scale model
single cell
k1eff
k2eff
Figure 17
Comparison of one-stage
and two-stage upscaling
The accuracy of scale-up is affected by the correlation length and standard deviation
of the distribution, and we use the method shown in Figures 10 and 11 to demonstrate
this effect. Instead of generating many fine-scale models with different correlation
lengths, we create 1 fine-scale model, but upscale by different factors – so that the

coarse block size varies relative to the correlation length. Figure 18a shows a fine-
scale model with a correlated random permeability distribution. The model has 400
x 400 grid cells, each of size 1 m3. The permeability distribution is ln-normal, with
a mean of 4.6 (corresponding to 100 mD), and a correlation length of 10 m. Three
different versions of the model were created with different standard deviations: 0.5,
0.75 and 1.0. The following scale-up factors were tested:
4 × 4, 5 × 5, 8 × 8, 10 × 10, 16 × 16, 20 × 20, 40 × 40, 80 × 80

In terms of the correlation length, this gives coarse-scale cells of size:
0.4, 0.5, 0.8, 1.0, 1.6, 2.0, 4.0, 8.0.
Figure 18b and c show examples of coarse-scale models with scale-up factors of 5
and 50. In each case the ratio of two-stage upscaling to single-stage upscaling was
calculated. The results are plotted in Figure 19. The results are least accurate when
the scale-up factor is 10 – 50, i.e. when the coarse block size is 1 – 5 times the cor-
relation length. Also, the error increases with the standard deviation of the model,
as one might expect.
a) b) c)
Figure 18
a) Fine-scale model with
400 x 400 cells; b) coarse-
scale model with 80 x 80
Permeability (mD) cells; c) coarse-scale model
0.1 1 10 100 1000 10000 with 8 x 8 cells
1.03
1.02
keff2/keff1
sigma = 1.00
sigma = 0.75
sigma = 0.50
1.01
1.01
Figure 19
2 1
1.00 The ratio of k eff k eff for
0 20 40 60 80
Scale-up Factor different scale-up factors
The conclusions from these examples are:
• Upscaling will be least accurate when the coarse cell size is comparable to, or
slightly larger than the correlation length.
• Upscaling errors increase as the standard deviation of the model increases.
14
7
1.5.3 Upscaling of a Sand/Shale Model
Upscaling errors are largest in models where there are high permeability contrasts.
Unfortunately, high permeability contrasts frequently occur in reservoir rocks. For
example, we often require to model the following:
• Low permeability shales in a high permeability sandstone

• Low permeability faults in a high permeability sandstone
• High permeability channels in a low net/gross reservoir
• High permeability fractures in a low permeability reservoir
All these cases are difficult to model. As an example, we consider a sand/shale model,
where the shale has zero permeability. Figure 20 shows the fine-scale model, which
has to be upscaled to 3 coarse blocks, as shown. Since there is a shale lying across
each coarse block, each coarse block will have zero permeability in the z-direction
(vertical). However, fluid can flow through the model vertically, as shown. This
error arises because the coarse block size is similar to the characteristic length of the
shales. Upscaling would be more accurate, if the coarse block size was much larger,
or much smaller than the shales. Alternatively, using a “skin” or “flow jacket” will
increase the accuracy of upscaling (Section 1.4.2, Figure 14).
Figure 20
Sand/shale model
1.6 Summary of Single-Phase Upscaling

The main points to remember from this section are:
• Fine-scale geological models usually require upscaling for full-field

simulation.
• Upscaled permeability is generally referred to as effective permeability.
• Some quantities, such are porosity and water saturation are easy to upscale,
because they may be averaged arithmetically.
• In some simple models, permeability may also be upscaled using averaging, as
follows:
− the arithmetic average for along-layer flow;
− the harmonic average for across-layer flow;
− the geometric average for a random distribution.
• In more complex models, the effective permeability is calculated using a
numerical simulation.
• Different boundary conditions may be used when calculating the effective
permeability numerically: constant pressure (or no-flow), periodic and linear.
• Upscaling is least accurate when the coarse cell size is comparable to, or slightly
larger that the correlation of the permeability distribution.
• Upscaling errors increase as the standard deviation of the model increases.

2 TWO-PHASE FLOW
2.1 Introduction
Often we need to simulate two-phase systems, e.g. a water flood or a gas flood of an
oil reservoir, or an oil reservoir with a gas cap or an aquifer. The aim of upscaling
in this case is to calculate a coarse-scale model which can reproduce the flow rates
of the different fluids. The coarse model should also provide a good approximation
to the saturation distribution in the reservoir with time.
The paths which the injected fluid takes through the reservoir depends on the forces
present:
• Viscous – due to injection of a fluid

• Capillary
• Gravity
Therefore, the balance of forces should be taken into account during upscaling.
Before learning how to upscale two-phase flow, we show the effects which geologi-
cal heterogeneity may have on hydrocarbon recovery.
Consider the following simple model (Figure 21), with alternating horizontal layers of
100 mD and 10 mD (referred to as facies 1 and facies 2). We assume that the model
is filled with oil and connate water initially, and simulate a water flood, by injecting
at uniform rate at the left side, and producing from the right side (at constant bottom-
hole pressure). The density of the two fluids is the same for this example, so that
there are no gravity effects. Figure 22 shows the relative permeabilities and capillary
pressures, and Table 1 lists the properties of the 3 cases simulated with this model.
100 mD
10 mD
Figure 21
Simple layered model for
demonstrating viscous and
capillary effects
12 0.9
krw 1 Pc 1
0.8
10 kro 1 Pc 2
krw 2 0.7
Cap Pressure
8 kro 2 0.6
Rel Perm
0.5
6
0.4
Figure 22
4 0.3
0.2
The relative permeability

2
0.1
0 0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8 and capillary pressure
Water Saturation Water Saturation
curves
Case Porosity Rel Perm/Pc Curve No. Flow Regime

facies 1 facies 2 facies 1 facies 2
1 0.2 0.2 1 1 viscous Table 1
2 0.2 0.05 1 1 viscous Properties of the first set of
3 0.2 0.1 1 2 visc+capillary
examples
16
7
Figure 23 shows the distribution of oil saturation after the injection of 0.2 PV, for
Cases 1 and 2. Both cases use only a single relative permeability curve and the flow
regime is viscous-dominated. Water flows faster along the high permeability layers,
as one would expect. However, notice that this effect is reduced when the porosity
of facies 2 is reduced.
Figure 23
The oil saturation for Cases
1 and 2, after the injection Oil Saturation
of 0.2 PV
0.3 0.4 0.5 0.6 0.7
In Case 3, both relative permeability and capillary pressure tables are used. These
curves are typical of a water-wet rock: the capillary pressure is much higher in the
low permeability facies, and the connate water saturation is higher. In this case, water
is imbibed along the low permeability layers, and also there is cross-flow from the
high permeability layers to the low permeability ones (Figure 24). Due to the effects
of capillary pressure, the front is nearly level in the two facies.
Figure 24
The oil saturation for Case
3, after the injection of 0.2 Oil Saturation
PV
0.3 0.4 0.5 0.6 0.7
Figure 25 shows the recovery as a function of pore-volumes injected, and demon-

strates that models may have the same effective absolute permeability, but different
recoveries.

0.5
0.4
0.3 Case 1
Case 2
0.2 Case 3
0.1
Figure 25
0 Cumulative recovery and
0 0.1 0.2 0.3 0.4 0.5 0.6
Pore Volumes Injected
watercut for Cases 1 - 3
In the next example, graded models are used, as shown in Figure 26. There are two
versions: Case 4 with permeability increasing upwards (referred to as coarsening-up)
and Case 5 with permeability decreasing upwards (referred to as fining-up). The
permeabilities range from 200 mD to 1000 mD, the porosity was kept constant at
0.2, and the first relative permeability curve was used. In this model, however, the
densities of the fluids were different: the density of water was set to 1000 kg/m3 and
that of oil was set to 200 kg/m3.
a) Coarsening-up b) Fining-up
1000 mD 200 mD
Figure 26
The graded layer models for
200 mD 1000 mD
Cases 4 and 5
Again a waterflood was performed, and the results are shown in Figure 27. Since
water is more dense than oil, water has a tendency to slump down. In Case 4 (coars-
ening-up), this tendency is reduced by the fact that the viscous forces tend to move
the fluid faster in the upper layers. However, in Case 5 (fining-up), the slumping
effect is reinforced by the viscous force moving fluid faster along the lower layers.
This means that the breakthrough time is earlier in Case 5 than in Case 4, as shown
in Figure 27. The effective absolute permeability in these two models is the same,
but the two-phase flow effects are different, due to the effect of gravity. (If the den-
sity of water equalled the density of oil, the recovery and watercut curves would be
identical.)
a) Coarsening-up a) Fining-up
Figure 27
Oil Saturation The oil saturation after the
injection of 0.2 PV in the
0.3 0.4 0.5 0.6 0.7 graded layer models
18
7
0.5 1.0
Fractional Recovery
0.4 0.8
Water Cut
0.3 0.6
Case 4
0.2 0.4 Case 5
Figure 28 0.1 0.2
Cumulative recovery and 0.0 0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
watercut for the graded Pore Volumes Injected Pore Volumes Injected
layer models
In summary, in a viscous-dominated flood, permeability heterogeneity disperses the
flood front (Cases 1 and 2), so that breakthrough occurs earlier, and the water cut
curve rises less steeply. However, the effect of heterogeneity also depends on the
balance of fluid forces. In a water-wet system, capillary pressure can help the front
to advance more evenly along the layers (Case 3). (More information on capillary
effects is given in Section 3.3.) Gravity effects may increase or reduce the viscous
effects, depending on permeabilities in the model (Cases 4 and 5).
2.2 Applying Single-Phase Upscaling to a Two-Phase Problem

Most engineers only perform single-phase upscaling although, as shown above, het-
erogeneities give rise to a variety of effects in two-phase flow. The reason for this
is that two-phase upscaling is time-consuming and the results are not always robust
(i.e. they may contain large errors). We deal with two-phase upscaling in Sections
2.4 – 2.6.
Figure 29 (left side) shows a very heterogeneous 2D model, with correlated random
permeabilities. The permeability distribution was ln-normal (i.e. natural logs), with
a standard deviation of 2.0 The model is assumed to be in the horizontal plain. The
details of the model are given in Table 2. The model was upscaled using the pres-
sure solution method with no-flow boundaries. Three different scale-up factors were
used, and the coarse-scale models 1 and 3 are also shown in Figure 29 (middle and
right side).
Figure 29
Fine- and coarse-
scale models used for
demonstrating the effects
of applying single-phase
upscaling to a two-phase log10(k)
problem -2 -1 0 1 2 3 4 5 6
Model Number of Cells Cell dimension (m) Scale-up Factor

Fine 105 x 105 5 -
Coarse 1 21 x 21 25 5x5
Coarse 2 15 x 15 35 7x7
Table 2 Coarse 3 7x7 75 15 x 15

A waterflood with a quarter 5-spot well pattern was performed (i.e. 2 wells in
diagonally opposite corners). The same relative permeability curve was used for
the whole model (Figure 30), and the capillary pressure was set to zero. The flood
was therefore viscous-dominated. The viscosity of water was 0.3 and that of oil was
3.0. The resulting recovery curves are shown in Figure 31. As one might expect, the
error increases with the scale-up factor.
1
0.8
Relative Permeability
0.6
krw
kro
0.4 Figure 30
The relative permeability
0.2
curve used for the random
0 model. (Capillary pressure
0 0.2 0.4 0.6 0.8 1
Water Saturation was set to zero)
0.6
0.5
Fractional Recovery
0.4
fine
ups 5x5
0.3
ups 7x7
ups 15x15
0.2
0.1
Figure 31
0
Comparison of recovery for
0 0.2 0.4 0.6 0.8 1
Pore Volumes Injected different scale-up factors
The model was modified by reducing the standard deviation to 0.2 (lowering the
permeability contrast), and the simulations were repeated with the low heterogeneity
model. The recovery is shown in Figure 32 for the scale-up factor of 15 x 15. As
expected, the errors are smaller for the low heterogeneity model.
0.6
0.5
Fractional Recovery
0.4
lo het fine
lo het coarse
0.3
hi het fine
hi het coarse
0.2
0.1
Figure 32
Comparison of recovery for
0
0 0.2 0.4 0.6 0.8 1 models with different levels
Pore Volumes Injected of heterogeneity
In addition, the errors caused by using only single-phase upscaling, are larger when
the coarse block size similar to the correlation length. Also, the upscaling error tends
20
7
to be larger in unstable floods (injected fluid is of lower viscosity than the in situ
fluid) than in stable floods.
In summary, single-phase upscaling may be adequate for upscaling two-phase sys-

tems, provided that:
• The scale-up factor is small

• The permeability contrasts are small
• The correlation length is very large, or very small compared to the coarse cell
size
• The flood is stable (favourable mobility ratio)
• The flood is not capillary-dominated (See Section 3.3)
• The flood is not gravity-dominated
2.3 Improving Single-Phase Upscaling

There are two approaches which may make single-phase more accurate when applying
it to two-phase problems. The first is to use non-uniform upscaling, and the second
is to perform a global single-phase simulation (i.e. over the whole fine-scale model)
using the correct boundary conditions, including wells. We refer to this second
method as Well Drive Upscaling (WDU).
2.3.1 Non-Uniform Upscaling

Consider a model with horizontal layers, as shown in Figure 33. There is a high
permeability streak running across the model. The model details are given in Table 3.
In both coarse-scale models there are 3 coarse cells in the vertical direction. In model
Coarse 1, the cells are each 5 m thick. However, in model Coarse 2, the thicknesses
are: 7 m, 1 m, 7 m, so that the high permeability streak is still resolved.
b) C
Coa
oars
oarse
rs e1 c) Coar
c) Coar
oarse
se 2
Figure 33
Model with a high
permeability streak
Model No. of cells Cell size (m)

Fine 100 x 15 1x1
Coarse 1 20 x 3 5x5
Table 3 Coarse 2 20 x 3 variable
Figure 34 shows the recovery for these models. It can be seen that the model with
uniform coarse cells (Coarse 1) gives very inaccurate results, and the model which
maintains the high permeability streak (Coarse 2) is much more accurate.

0.25
0.20
Fractional Recovery
0.15 fine
coarse 1
0.10 coarse 2
0.05
Figure 34
0.00
Recovery and watercut
0.0 0.1 0.2 0.3 0.4 for fine- and coarse-scale
Pore Volumes Injected
models
In a practical upscaling application, much attention is paid to upgridding the model

– i.e. deciding how to amalgamate the layers, so that upscaling is as accurate as pos-
sible. The coarse grid may also be non-uniform in the x- and y-directions.
Several methods for performing non-uniform upscaling have been developed. For
example, Durlofsky et al. (1996, 1997) first carry out a single-phase simulation. Then
they use the inter-block flows to determine the coarse block boundaries. Smaller
coarse blocks are assigned to regions where there are high flow rates.
2.3.2 Well Drive Upscaling

When upscaling using numerical simulation, boundary conditions are applied to
each coarse-scale cell in turn (Figure 2.) These are called local boundary condi-
tions. However, the boundary conditions described in Section 1.4.2 may be quite
different from the pressures which actually occur in a fine-scale simulation, lead-
ing to inaccuracies in the upscaled model. To overcome this problem, we may use
global boundary conditions. Figure 35 demonstrates the difference between local
and global boundary conditions.
Boundary conditions Injector

applied to coarse cell
P1 P2
Figure 35
Producer
Local and global boundary
Local Global
conditions
In the Well-Drive Upscaling method (WDU), a single-phase simulation is performed

on the whole fine grid (Zhang et al, 2005). (It is feasible to perform a single pressure
solve on grids with several million grid cells.) Then the effective transmissibility
between coarse-scale cells is calculated, rather than the effective permeability. The
upscaled transmissibility is give by:
22
7
T=
∑q
PΙ − PΙΙ (13)
Where q denotes the fine-scale flows (Figure 36) and PI and PII are the (pore-volume
weighted) average pressures in coarse cells I and II.
Figure 36
Upscaling transmissibility Ι ΙΙ
Upscaling is also performed at the wells, using the method described in Section
3.1. This method produces very accurate single-phase upscaling, which leads to an
increase in the accuracy of two-phase flow at the coarse scale.
Many tests on upscaling methods have been carried out using the model generated
for the 10th SPE comparative solution project, which was on upscaling (Christie and
Blunt, 2001). This model is referred to as the SPE 10 model (Figure 37a). We use
layer 59 (Figure 37b) as an example of well drive upscaling, because this layer is
particularly heterogeneous. There is an injection well at the centre and production
wells in each corner.
a) b) P1 P2
P4 P3
Figure 37
The SPE 10 model, and log10(k)
layer 59 -3 -2 -1 0 1 2 3 4 5
Layer 59 was upscaled using the WDU method and also the conventional method
with local boundary conditions. Figure 38 shows the oil saturation distribution. It
can clearly be seen that the WDU method reproduces the results of fine-scale model
much better than the conventional approach. This is because appropriate boundary
conditions have been applied to the model.

fine local WDU
Figure 38
Comparison of oil
saturation distribution
in fine- and coarse-scale
Soil
simulations of Layer 59 of
0.20 0.35 0.50 0.65 0.80 the SPE 10 model
2.4 Introduction to Two-Phase Upscaling

So far, when performing upscaling, we have assumed that there is only one phase
present, and that the flow is in a steady state. We only need to upscale the absolute
permeability. However, when there are two phases flowing, such as water displac-
ing oil, the system is not, in general, in a steady state. We need to simulate fine-
scale floods in order to upscale relative permeability and capillary pressure. This
is referred to as dynamic upscaling, and the upscaled relative permeabilities are
known as pseudo relative permeabilities, or pseudos. Pseudos can be calculated to
take account of physical dispersion, and also to compensate for numerical dispersion
(Section 2.6.3).
When upscaling, we should use the phase permeabilities:
k f = k abs k rf (14)
Where “f” stands for fluid – oil, gas or water. Generally, we assume that both the
absolute and the relative permeabilities are homogeneous and isotropic at the smallest
scale ( k x = k z ). As we upscale, the absolute and relative permeabilities may become
anisotropic ( k rx ≠ k rz ). To obtain effective (or pseudo) relative permeabilities, the
absolute permeability must be scaled-up separately. Then the pseudo relative perme-
ability is calculated as follows:
k rf , x = k f , x k abs, x (15)
Similar equations are used for flow in the y- and z- directions.
2.5 Steady-State Methods

If fluids are in a steady state, the saturation does not change with time and the fractional
flow (flow of water/total flow) is constant. Although floods are dynamic processes,
sometimes a flood may approach a steady state. For example, over small scales (20
cm, or less), oil and water may come into capillary equilibrium.
24
7
In a steady-state upscaling method, we assume that within a short interval of time the
zone of interest is in a steady-state, but we allow the fluid saturation to change gradu-
ally, so that a full range of saturation is obtained. At steady-state, the water saturation
does not change with time, i.e. ∂Sw/∂t = 0, so the continuity equation becomes:
∇ ⋅ u f = 0, (16)
where u is Darcy velocity, and f is fluid. From Darcy’s law:
∇ ⋅ ( k f ⋅ ∇Pf ) = 0. (17)
There are several steady-state methods, depending on the balance of forces:
• Capillary equilibrium,
• Vertical equilibrium (gravity-dominated flood)
• Viscous-dominated steady-state
We concentrate here on the capillary equilibrium method.
The advantage of steady-state methods is that they turn two-phase upscaling into a
series of single-phase upscaling calculations. This means that steady-state methods
are feasible for models with large numbers of grid cells. (See, for example, Pickup
and Stephen, 2000; and Pickup et al, 2000.)
2.5.1 Capillary-Equilibrium
Assume that the injection rate is very low, gravity forces are negligible, and that the
fluids have come into capillary equilibrium with a coarse-scale cell. This means that
the saturation distribution is determined by the capillary pressure curves.
The method is as follows:
1. Choose a Pc level.
2. Determine the water saturations, and then the relative permeabilities.
3. Calculate the pore volume-weighted average water saturation.
4. Calculate the phase permeabilities: ko = kabskro, kw = kabskrw.
5. Calculate the effective water phase permeability, kw
6. Calculate the effective oil phase permeability, ko
7. Calculate the relative permeabilities, krw = kw/kabs, etc.
8. Repeat the process with another value of Pc.
Steps 5 and 6 may be carried out analytically or numerically, depending on the

distribution.

Example 4
Consider a model with two layers of equal thickness, as shown in Figure 39. The
absolute permeabilities are 100 mD and 20 mD. Assume that the porosity in each
layer is equal to 0.2. The relative permeability and Pc curves for each layer are
shown in Figure 40.
kabs (mD)
100 Figure 39
20 Model with horizontal
layers
12
0.8
0
lo
hi lo
Cap Pressure
8 0.6
Rel Perm
6
hi 0.4
4
0.2 hi Figure 40
2
lo
0 0 Relative permeability and
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Water Saturation Water Saturation capillary pressure curves
Using the arithmetic and harmonic averages (Section 1.3), the effective permeability
is:
k x = 60.00 k z = 33.33
Suppose we choose a capillary pressure of Pc = 0.45.
In the high perm layer: Sw = 0.34,
krw = 0.0013, kw = 0.13, kro = 0.5, ko = 50.
In the low perm layer: Sw = 0.44,
krw = 0.0016, kw = 0.032, kro = 0.48, ko = 9.6.
Figure 41 shows the phase permeabilities.
kw (mD) ko (mD)
0.13 50.0
0.032 9.6 Figure 41

Phase permeabilities
Since the layers are of equal width, the average saturation is Sw = 0.39. The effective
phase permeabilities are then calculated using the arithmetic and harmonic averages.
Then the relative permeabilities are calculated using Equation 15.
26
7
k wx = 0.081 k rwx = 0.081 / 60.00 = 0.00135
k wz = 0.051 k rwz = 0.051 / 33.33 = 0.00153
k ox = 29.8 k rox = 29.8 / 60.00 = 0.50
k oz = 16.1 k roz = 16.1 / 33.33 = 0.48
Note that the kv/kh ratio ( = k z k x ) is different for oil and water:
k w, z k w, x = 0.63 k o, z k o, x = 0.54
Effective relative permeability curves may be derived by repeating this calculation for
a range of capillary pressure values (Figure 42). The capillary-equilibrium method
is useful as a quick method for upscaling small-scale models (Section 3.3). However,
it is only valid in cases where the flow rate is very low.
0.9
0.8
0.7
0.6 krox
Rel Perm
0.5
0.4
kroz
0.3 krwz
0.2 krwx
Figure 42 0.1
0
Effective relative 0.2 0.3 0.4 0.5 0.6 0.7 0.8
permeability curves Water Saturation
2.6 DYNAMIC METHODS
2.6.1 Introduction
For dynamic (or non steady-state methods), we need to perform a two-phase flow
simulation on a fine grid. There are basically two types of dynamic method:
a) Weighted Pressure Methods

As in single-phase numerical upscaling, a common approach in two-phase upscaling
is to sum the flow, average the pressure gradient and use Darcyʼs Law to obtain the
pseudo phase permeability. However the pressure may be averaged in different ways.
Here, we shall concentrate on the Kyte and Berry (1975) method.
b) Total Mobility Methods

In total mobility methods, we avoid averaging the pressure, and scale-up the total
mobility. Then the average fractional flow is used to calculate the pseudo relative
permeabilities.
The total mobility is:
k ro k rw
λt = λo + λw = + .
µo µw (18)

The fractional flow is the flow of water divided by the total flow:
qw q
fw = = w
qo + qw qt (19)
Again there are a number of variations of this method, the most commonly used
being that of Stone (1991).
2.6.2 The Kyte and Berry Method

A simple version of the Kyte and Berry (1975) method is presented here, using the
grid shown in Figure 43.
i=1 2 3 4 5 6 7 8 9 10
j=1
2
3
4
5 ∆z
∆x
Figure 43
Model used for describing
Pseudo calculated
for this coarse block DX
∆X DZ
∆Z the Kyte and Berry Method.
The thickness of the model
is Δy
Δ
The diagram shows two coarse grid blocks, each of which is made up of 5 x 5 fine
blocks. The equations below show how to calculate the pseudo relative permeabilities
and capillary pressure for the left coarse block.
The first step is to perform a fine-scale, two-phase simulation (e.g. in ECLIPSE),

saving the pressures and inter-block flows at specified intervals of time (in the re-start
files). The method proceeds as follows:
1. Calculate the effective absolute permeability in the area shown in Figure 44,
i.e. half way between the two coarse blocks.
i=3 i=7
j=1
Figure 44
The area used for
calculating the effective
j=5
absolute permeability
Kyte and Berry approximate the effective permeability using the arithmetic average
in each column, and then taking the harmonic average of the columns. The area
between the two coarse blocks is used, for reasons explained below.
28
7
5
∑ ∆z k
j =1
j ij
ki =
∆Z (20)
where Δzzj and ΔZ are the thicknesses of the fine and coarse blocks, respectively. (In
this case, all the blocks are of equal size.)
∆X
kI = 7
∑ ∆x i ki
i=3 (21)
where Δxi and ΔX are the lengths of the fine and coarse blocks, and k I is the required
effective absolute permeability.
The pseudos are then calculated, at certain times during the simulation. (These are
the times at which the restart files are written in the Eclipse simulation.)
2. Calculate the average water saturation:

5 5
∑∑S
j =1 i =1
φ ∆x i ∆z j
w , ij ij
Sw = 5 5
∑ ∑ φ ∆x ∆z ij i j
j =1 i =1 (22)
where φij is the porosity.
3. Calculate the total flow of oil and water out of the left coarse block (Figure 45).
5
q f = ∑ q f 5, j ,
j =1 (23)
where qf5,j is the flow of fluid “f” from fine block number (5,j).
i=5
j=1
Figure 45
Calculation of the total flow j=5
4. Calculate the average phase pressures in the central column of each coarse
block. In this example, we use the fine blocks in columns 3 and 8, the shaded
areas in Figure 46.

i=3 i=7
j=1
Figure 46
j=5 The cells used for averaging
I II the phase pressures
In the Kyte and Berry method, the pressures are weighted by the phase permeabilities
times the height of the cells (which in this case are all the same size). This is so that
more weight is given to regions where there is greater flow. However, there is no
scientific justification for using this weighting. In the first coarse block (numbered,
I), the average pressure is:
5
∑k
j =1
3j (
k rf 3 ∆z 3 j Pf 3 j − gρf (D3 j − D) )
P fI = 5
∑k 3j k rf 3 ∆z 3 j
j =1 (24)
where D3j is the depth of cell (3,j) and D is the average depth of coarse cell I. The
term gρf(D3j - D ) is to normalise the pressure to the grid block centre. The average
pressure for coarse block II is calculated in the same manner, but using column 8
instead of column 3. The pressure difference is then calculated as:
∆P f = P fI − P fII . (25)
5. The pseudo rel perms are then calculated using Darcy’s law. Firstly, calculate
the pseudo potential difference. (Potential is defined as Φ = P-ρgz, so that the
flow rate is proportional to ∇Φ .)
∆Φ f = ∆P f − gρf ∆
∆D, (26)
where ΔD is the depth difference between the two coarse grid centres. Then:
−µ f q f ∆X
k rf =
∆Zk I ∆Φ I (27)
6. Calculate the pseudo capillary pressure using:
P c = P oI − P wI (28)
The Eclipse PSEUDO package can be used for calculating Kyte and Berry
pseudos.
30
7
2.6.3 Discussion on Numerical Dispersion
One advantage of pseudo-isation methods, such as that of Kyte and Berry is that they
can take account of numerical dispersion. When a simulation is carried out using a
larger grid, the front between the oil and water becomes more spread out. However,
the Kyte and Berry method counteracts this effect by calculating the flows on the
down-stream side of the coarse block, instead of the middle. This is illustrated by a
simple example. Figure 47 shows an example of input relative permeability curves
(“rock” curves).
0.9
0.8
0.7
0.6
Perm Rel
0.5
0.4
0.3
0.2
0.1
Figure 47 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Water Saturation
Example of “rock” curves
If the water saturation is Sw = 0.5, the rock curves show that there is a small amount
of oil and water flowing. However, when the average saturation, Sw , is 0.5 in the
coarse block, the distribution could be as shown in Figure 48.
oil
coarse
water block
Figure 48
Example of the water
saturation in a coarse block
Since the water has reached only half way across the coarse block, there should be no
water flowing out of the right side. The Kyte and Berry method calculates the pseudo
relative permeabilities using the flow on the downstream side of the coarse block, to
prevent water breaking through too soon. The pseudo water relative permeability
curve is moved to the right, relative to the rock curves, as shown in Figure 49.

1
pseudos - solid lines
0.9
ave. rock curves - dashed lines
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8
Figure 49
Water Saturation Example of pseudo relative
permeability curves
2.6.4 Disadvantages of the Kyte and Berry Method

There are certain problems with the Kyte and Berry method.
• Negative rel perms are produced, if ∆Φ f has the same sign as q f .

• Infinite rel perms occur if ∆Φ f is zero.
• The method of averaging the pressures, using relative permeability as a weighting
function, may cause errors when the fluids are separated due to gravity. For
floods which are gravity-dominated, the TW method works better (Section
2.6.5).
• Non-zero pseudo capillary pressure may be produced, even if there is no capillary
pressure in the fine-scale simulation. This is because a different weighting is
used for calculating the average pressure in each phase.
• The capillary pressure may be different in different directions, because only the
central column is used for averaging the pressures.
Because of the first two disadvantages, i.e. negative, or infinite rel perms, pseudos
obtained from packages like the PSEUDO must be vetted before using at the coarse
scale. Often “odd” values of relative permeability are set to zero.
Good reviews of various methods for calculating pseudos are presented in Barker
and Dupouy (1999) and Barker and Thibeau (1997).
Note that dynamic upscaling methods, such as that of Kyte and Berry are difficult
to apply in practice. Ideally, a fine-scale two-phase flow simulation is required for
each coarse-scale cell (plus a “flow jacket”), and this is time consuming. Also, it is
difficult to determine the correct boundary conditions to use, so the results may not
be accurate.
If a pseudo is calculated for each coarse cell, in each direction, there may be 10,000s
of pseudos in the coarse-scale model. The number of pseudos must be reduced, by
grouping similar pseudos together.
32
7
Pseudo relative permeability curves depend on a number of factors, including:
(a) The balance of forces

The shape and end points of a pseudo depend on the ratio of viscous/capillary and
viscous/gravity forces. These ratios may be different in different parts of the res-
ervoir.
(b) The well locations

The wells determine the flow rate and direction. If a new well is drilled, the pseudos
ought to be re-calculated.
Because of these problems, two-phase upscaling is rarely used for upscaling from a
geological model to a full-field simulation.
2.6.5 Alternative Methods

There are a number of similar methods to the Kyte and Berry Method.
(a) The Pore-Volume Weighted Method

The problems of non-zero capillary pressure and directional capillary pressure,
mentioned in Section 2.6.4, may be overcome by using a pore volume weighted
average of the pressures over the entire coarse block. ECLIPSE uses this method
for calculating the average capillary pressure in the Kyte and Berry method. Also,
pore volume weighting may be used for averaging the pressures when calculating
the pseudo relative permeabilities. In this case, the method is called the Pore Volume
Weighted Method. It is available in the Eclipse PSEUDO package.
(b) The TW Method

This method was developed by Nasir Darman at Heriot-Watt University (Darman et al,
1999). It is similar to the Kyte and Berry method, except transmissibility weighting is
used when calculating the average pressure. The method works better than the Kyte
and Berry method in cases where gravity effects are significant (e.g. a gas flood).
Both these methods share the same problems discussed in Section 2.6.4, namely,
they are difficult to apply in practice.
2.6.6 Example of the PVW Method

In two-phase dynamic upscaling methods, pseudo relative permeabilities are cal-
culated, so that (hopefully) the results of a coarse-scale simulation provide a good
approximation to the fine-scale results. Layer 59 of the SPE 10 upscaling study
(Figure 37b) is used here as an example. A global simulation was performed on the
fine grid, i.e. the whole of the fine grid was included in the fine-scale flow simulation.
(Note that this would not be done, in practice, because there is no point in upscaling,
if you can simulation the whole fine grid.) The fine-scale model had 60 x 220 cells
and the coarse-scale model had 10 x 22, which corresponded to a scale-up factor of
6 x 10. Pseudo relative permeabilities were calculated for each coarse cell, in each
direction.
Figure 50 shows the oil saturation for the fine-scale simulation, a coarse-scale simulation
using the WDU method (Section 2.3.2), and a coarse-scale simulation using pseudos

from the PVW method. Both the WDU and the PWV methods give reasonable oil
saturation distributions. The oil recovery rate, for well P4, for these models is shown
in Figure 51, along with the oil rate for a coarse-scale simulation using single-phase
upscaling with local boundary conditions (Sections 2.2. and 2.3).
fine WDU PVW
Figure 50
The oil saturation
distribution for the fine-
scale model of layer 59 and
the coarse-scale models
Soil obtained using the WDU
and the PVW methods
0.20 0.35 0.50 0.65 0.80
60
50
Well Rate (m3/day)
40
Fine
30
Local Figure 51
WDU
PVW
The oil recovery rate for
20
well P4 for the fine-scale
10 model and coarse-scale
models obtained using
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
the WDU, PVW and local
Pore Volumes Injected upscaling methods
It can be seen in Figure 51 that both the WDU and the PVW methods agree well with
the fine-scale simulation. However, the results from the single-phase upscaling with
local boundary conditions are poor. This example shows that two-phase upscaling
is not necessarily always more accurate than single-phase upscaling.
2.7 Summary of Two-Phase Flow

The main points on two-phase flow are:
• Two-phase upscaling is time consuming and not always robust, so is rarely used
by engineers.
• Usually, only single-phase upscaling is performed.
• But, heterogeneity interacts with two-phase flow, and tends to produce dispersion of
the flood front, which is not taken into account using single-phase upscaling.
• So, single-phase upscaling may give rise to errors, especially when there is
a large scale-up factor, and the reservoir model is very heterogeneous (large
standard deviation).
34
7
• Errors in single-phase upscaling may be reduced by using non-uniform upscaling,
or well-drive upscaling.
• Ideally, two-phase upscaling should be performed to take account of two-phase
flow.
• Steady-state upscaling is relatively quick to apply, and is feasible for large
models. However, it is only valid in limited cases, e.g. when the fluids are
approximately in capillary equilibrium.
• Dynamic methods are potentially more accurate.
• The Kyte and Berry (1975) Method was described as an example.
• Dynamic methods can compensate for the effects of numerical dispersion.
• Dynamic methods are difficult to apply in practice.
3 ADDITIONAL TOPICS
This course has, so far, focussed mainly on common methods for upscaling a geo-
logical model for full-field simulation. Most of the single-phase upscaling methods
presented may be found in geological packages, such as IRAP/RMS and Petrel. (The
WDU and TW methods which were developed at Heriot-Watt are not available in
commercial packages.) However, there are a number of other important issues which
should be taken into account when upscaling. In this section, we cover these issues
in a variety of additional topics:
• Upscaling as Wells
• Permeability Tensors
• The Geopseudo Method
• Uncertainty and Upscaling
3.1 Upscaling at Wells

In the single-phase upscaling methods described in Chapter 1, we assumed that the
flow was linear. This means that the upscaling methods were not appropriate for
regions containing wells, where there is radial flow. We start with a brief overview
of simulation in blocks containing a well.
A grid block in the simulator is much larger than the diameter of a well, and the
pressure calculated for a block containing a well is different from the actual bottom
hole pressure. These are related by:
Iw
q=
µ
(Pw − Pb )
(29)
where Pw is the well-bore pressure and Pb is the pressure of the block. Iw is the well
index, given by:
2πk∆z
Iw =
ln( ro rw ) (30)

where rw is the well-bore radius and ro is the equivalent radius, given by Peacemanʼs
equation (Peaceman, 1978; Peaceman, 1983). Iw is also referred to as the well con-
nection factor, or the connection transmissibility factor.
Durlofsky et al. (2000) put forward a method for upscaling in the near-well region.
Others have put forward similar methods. The method is only approximate, but
improves the accuracy of coarse-scale simulations. The first step is to calculate
effective single-phase permeabilities, using one of the conventional methods (e.g.
periodic boundary condition applied to each block in turn). Then, a fine-scale single-
phase simulation of the well block and surrounding blocks is carried out (Figure 52).
From the results, the total flows out of the coarse-scale well block, and the average
pressures in the coarse blocks are calculated. These are used to calculate upscaled
transmissibilities between the coarse-scale well block and the surrounding blocks,
and a coarse-scale well index.
q
T4
T3 T1
T2
Figure 52
Near-well upscaling (after
Durlofsky et al., 2000)
This method improves the accuracy of upscaling at well, and it is also incorporated
into the well drive upscaling method (WDU), described in Section 2.3.2.
3.2 Permeability Tensors

Suppose that we have layers which are tilted at an angle to the horizontal, as in
Figure 53.
net flow in z-dir
x Figure 53
Cross-flow due to tilted
net flow in x-dir z layers. Light-coloured
layers represent high
permeability and dark-
coloured layers represent
low permeability
∆P
A pressure gradient has been applied in the x-direction. This will obviously give
rise to a flow in the x-direction. The fluid takes a path through the medium, so that
it expends a minimum amount of energy. There will be a component of flow up the
high permeability, and only a small amount of flow across the low permeable layer,
as shown. This gives rise to a net flow in the z-direction, or cross-flow. Here, the
term cross-flow is used to describe flow perpendicular to the applied pressure gradi-
ent. When calculating the effective permeability of this model, we need to take this
cross-flow into account. This may be done using a tensor effective permeability, k ,
where:
36
7
 k xx k xy k xz 
 
k =  k yx k yy k yz 
 k zx k zy k zz 
 (31)
The first index applies to the flow direction, and the second to the direction of the
pressure gradient. For example kxy is the flow in the x-direction caused by a pres-
sure gradient in the y-direction. The terms kxx, kyy, kzz are known as the diagonal
terms. These are the terms which are usually considered – the horizontal and vertical
permeabilities, kh and kv, respectively. The other terms, which describe the cross-
flow, are the off-diagonal terms.
With tensor permeabilities, Darcyʼs Law becomes:
k
u=− ⋅ ∇P,
µ (32)
where u is the Darcy velocity (vector) and P is pressure (scalar).
1 ∂P ∂P ∂P 
u x = −  k xx + k xy + k xz 
µ ∂x ∂y ∂z 
1 ∂P ∂P ∂P 
u y = −  k yx + k yy + k yz 
µ ∂x ∂y ∂z 
1 ∂P ∂P ∂P 
u z = −  k zx + k zy + k zz 
µ ∂x ∂y ∂z  (33)
In Sections 1.3.1 and 1.3.2, we studied flow along and across horizontal layers. The
model in Figure 5 is repeated here (Figure 54), showing the arithmetic average for
the effective permeability for along-layer flow and the harmonic average for across
layer flow.
fine-
fine-sca
ne-scale
sca le coar
coarse
ar se--sca
se scale
le
Figure 54 x t1 = 3 mm, k1 = 10 m
mD
D 66.2
66 .25
.2 5 mD
The simple two-layer model z t2 = 5 mm

mm,, k 2 = 100 mD
mD 22..86 mD
22 mD
In this model, there is no cross-flow, so we may write the effective permeability in

tensor form with zero off-diagonal terms, as follows:
66.25 0 
k=
 0 222.886
or in 3D:

66.25 0 0 
k=  0 66.25 0 
 
 0 0 22.86
3.2.1 Flow Through Tilted Layers
Figure 55
x' Layers tilted at an angle of
z' θ to the horizontal
This model is essentially the same as the one in Figures 4 and 6, although the layers
have are repeated, and they have been tilted. In the frame of reference defined by the
x ′ and z′ axes, the effective permeability may be calculated using the arithmetic and
harmonic averages as before. However, in the x-z co-ordinate system, the effective
permeability should be represented by a full tensor. The terms of the tensor may be
calculated from the arithmetic and harmonic averages, as follows:
 k a cos2 θ + k h sin 2 θ ( k a − k h ) sin θ cos θ 

k= 2 
 ( k a − k h ) sin θ cos θ k a sin θ + k h cos θ
2
(34)
This formula is obtained by rotating the co-ordinate axes through an angle θ. (You
are not required to know the proof.) This example is in 2D, so only the kxx, kxz, kzx
and kzz are shown. Further rotations may be carried out around the x ′ or z ′ axes to
obtain a full 3D tensor.
Note that:
• The tensor is symmetric (kxz = kzx).

• Depending on the sign of θ, the off-diagonal terms may be positive or negative.
Example 5
Suppose the example in Figure 54 is rotated by 30º (Figure 56), and calculate the
effective permeability tensor.
38
7
'z
x
Figure 56
Layered model tilted by 30º z 30o
cos230 = 0.75, sin230 = 0.25, sin30.cos30 = 0.433.
From before, ka = 66.25 mD, and kh = 22.86 mD.
k xx = 66.25 × 0.75 + 22.86 × 0.25 = 55.40 mD,
k xz = (66.25 − 22.86) × 0.433 = 18.79 mD,
k zz = 66.25 × 0.25 + 22.86 × 0.75 = 33.71 mD.
55.40 18.79
k= .
18.79 33.71
Full tensor permeabilities may also be calculated from numerical simulations. It is

useful to use periodic boundaries, as described in Section 1.4.2. When a pressure
gradient is applied in the x-direction, there will be flow in the x-direction, and also
flow in the z-direction due to internal heterogeneity. These flows can be used to
calculate the kxx and kzx tensor terms. Then a pressure gradient is applied in the z-
direction to obtain the kzz and kxz terms. (In 3D, a pressure gradient should also be
applied in the y-direction.)
3.2.2 Simulation with Full Permeability Tensors

Having calculated full effective permeability tensors, we need special software to
handle them at the larger scale. Conventional finite difference simulators use a 5-point
scheme in 2D and a 7-point scheme in 3D, and only take diagonal tensors – e.g. when
running ECLIPSE, you usually specify PERMX, PERMY and PERMZ. Simulation
with full tensors is more complicated and more time-consuming, but some packages
allow the user to input full tensors. In Eclipse, there is a full tensor option which
allows you to specify terms such as PERMXY.
In 2D, a 9-point scheme is required to take account of cross-flow. This means that
there are 9 terms in each of the pressure equations, as illustrated in Equation (35).
a1Pi , j − a 2 Pi −1, j − a 3 Pi +1, j − a 4 Pi , j −1 − a 5 Pi , j +1

−a 6 Pi −1, j −1 − a 7 Pi −1, j +1 − a 8 Pi +1, j −1 − a 9 Pi +1, j +1 = 0.
(35)
The coefficients, ai, in Equation (35) depend on the transmissibilities between the
blocks. There are several different methods of discretisation which give slightly dif-

ferent results. To extend this to 3D, we need either a 19-point scheme or a 27-point
scheme. See Figure 57. (The 19-point scheme leaves out the 8 corners of the cube.)
Obviously, it takes longer to solve equations with a larger number of terms.
a) b)
x
i-1,j-1 i,j-1 i+1,j-1
z
i-1,j i,j i+1,j
Figure 57
i-1,j+1 i,j+1 i+1,j+1 a) 9-point scheme for 2D.
b) 27-point scheme for 3D
Often the off-diagonal elements of the permeability tensor (kxy, etc) are negligible,
so the limitations of using a 5-point (2D) or a 7-point (3D) scheme are not serious.
In layered systems, the size of the off-diagonal term may be gauged from Equation
(34) in Section 3.2.1:
k xz = ( k a − k h ) sin θ cos θ. (36)
This is a maximum for θ = 45º, and increases as (ka – kh) increases. Therefore, full
permeability tensors become more important as the angle of the lamination or bed-
ding increases, and as the permeability contrast increases.
3.3 Small-Scale Heterogeneity

Most reservoirs are modelled using, what is commonly termed a “fine-scale geological
model”. This is a stochastic model with grid cells of size approximately 50 m in the
horizontal directions, and about 0.5 m in the vertical. There are typically about 107
such cells in a full field model. These cells must be reduced in number to about 104 for
full-field simulation. However, each of the grid cells in the geological model is likely
to be heterogeneous, containing, for example, sedimentary structures. Petrophysical
data (permeabilities, relative permeabilities, and capillary pressures) are acquired from
core plugs, which are only a few cm long. When small-scale structure is present,
petrophysical data should be upscaled before being applied to the grid blocks of the
geological model. Figure 58 shows the ranges of scales of sedimentary structures,
along with the scales of measurements and typical sizes of models.
100
10
0 Seismic data Para-
Vertical thickness (m)
sequences
Log
1 Flow model
0.1 Core Geological model
Beds
0.01 Probe
Laminae
0.001
0.001 0.01 0.1 1.0 1 10 100 1000 10000 Figure 58
Horizontal length (m)
Length scales
40
7
For convenience, we consider upscaling as two separate stages (Figure 59). Stage
1 is upscaling from the smallest scale at which we may treat the rock as a porous
medium (rather than a network of pores), up to the scale of the stochastic geologi-
cal model, i.e. from the mm – cm scale to the m – Dm scale. Stage 2 is upscaling
from the stochastic geological model to the full-field simulation model, which has
already been described.
Core plug Sedimentary

Structure
Stage 1
Upscaling
Geological Model
~ 107 blocks
Figure 59
Two separate stages of
upscaling. (Geological Stage 2
model taken from “Tenth Upscaling
SPE Comparative Solution
Project: A Comparison of
Techniques”, by Christie Simulation Model
and Blunt, 2001.) ~ 104 blocks
3.3.1 The Geopseudo Method

Upscaling from the core-scale to the scale of the geological model (Stage 1 in Figure
59) is frequently ignored by engineers, who apply core plug permeabilities and “rock”
relative permeability curves directly to the geological model. However, work carried
out at Heriot-Watt University has demonstrated that small-scale structures, such as
sedimentary lamination may have a significant effect on oil recovery (Corbett et al.,
1992; Ringrose et al., 1993; Huang et al., 1995). For example, in a waterflood of
a water-wet rock, water is imbibed into the low permeability laminae, and oil may
become trapped in the high permeability laminae.
The Geopseudo Method is an approach, where upscaling is carried out in stages,

using geologically significant length-scales (Figure 60). Models of typical sedi-
mentary structures are created and permeability values are assigned to the laminae
(from probe permeameter measurements, or by analysing core plug data). Relative
permeabilities and capillary pressure curves are also assigned to each lamina-type (by
history matching SCAL experiments on core plugs). Flow simulations are carried
out to calculate the effective single-phase permeability and the two-phase pseudo
parameters. Additional stages of modelling and upscaling may be required – e.g.
upscaling from beds to bed-sets.
In the finest-scale model, the grid cells may be a mm cube, or less. If we upscale to
blocks of 50 m x 50 m x 0.5 m, we are upscaling by a factor of at least 5 x 104 in the
horizontal directions and 500 in the vertical.

Low Perm High Perm
Individual Rel. Perm Curves
Figure 60
Illustration of the
Effective Perm Pseudo Rel. Perm Curves
Geopseudo Method
3.3.2 Capillary-Dominated Flow

At small scales the flow is often capillary-dominated. Figure 24 showed a moderate
capillary effect: the imbibition of water along the low permeability layer made the
flood front approximately level in the two layers (instead being ahead in the high
permeability layer, in the case of a viscous-dominated flood). In that model (Figure
21), the layers were 1 m thick, and the grid cells were 10 cm square. If the size of
the model is reduced by a factor 100, so that the layers are 1 cm thick, and represent
sedimentary laminae, the effects of capillary pressure are much stronger, as shown in
Figure 61. In this case, strong capillary imbibition draws water into the low perme-
ability layers (black) so that the front advances faster in this layer. Notice that, in this
figure, there is little lateral variation in the shading, showing that the water saturation
is almost constant in each layer. This is because the front has been spread out by the
effects of capillary pressure, and the model is almost in capillary equilibrium.
Figure 61
Example of capillary-
Oil Saturation dominated flood in a
0.3 0.4 0.5 0.6 0.7 layered model
When the flow is across the layers, as in Figure 62, the effects of capillary pressure
are even more striking. This figure shows the same small-scale model of sedimen-
tary lamination. When the injection rate is low (average frontal advance rate of
0.3 m/day), the flood is capillary-dominated (Figure 62a), water (black) has been
imbibed into the low permeability layers leaving oil trapped in the high permeability
laminae (grey). As the injection rate is increased, the oil has more viscous force and
can overcome the capillary forces leading to less trapping of oil (Figures 61b). In
the case of Figure 62c, the flood is viscous dominated and all the movable oil has
been displaced by water.
42
7
a) b) c)
Figure 62
Examples of across-
layer flow. a) capillary
Oil Saturation
dominated, b) intermediate,
0.3 0.4 0.5 0.6 0.7
c) viscous-dominated
The examples shown in Figures 61 and 62 demonstrate the significance of capillary

effects at the small-scale. When upscaling from the lamina-scale, these effects should
not be ignored, and two-phase upscaling should be performed.
3.3.3 Geopseudo Example

All the upscaling methods described in the previous sections may be used in the
Geopseudo approach, depending on the type of heterogeneities and the fluids flowing
– averaging, single-phase numerical methods, two-phase steady-state methods, or
two-phase dynamic methods. Since a flood is often capillary-dominated, as shown
above, steady-state upscaling using the capillary equilibrium method is often appro-
priate. Two-phase dynamic upscaling may also be used, and we show an example
of the Kyte and Berry method below.
Figure 63 shows a model of sedimentary ripples. Kyte and Berry pseudos were cal-
culated for the model using four different flow rates. There is a factor of 10 between
each flow rate, with rate 1 being the fastest. Figure 64 shows the resulting pseudos
(from Pickup and Stephen, 2000).
Figure 63
3 cm, 54 cells 10 mD
A model of ripples (based
on the Ardross Outcrop, 1 cm, 200 mD
18 cells
near St. Monance in Fife,
Scotland)
0.9 0.9
0.8 rate 1 0.8 rate 1
rate 2 rate 2
0.7 rate 3 0.7 rate 3

rate 4 rate 4
0.6 0.6
0.5 0.5
0.4 0.4
Figure 64 0.3 0.3
Pseudo relative 0.2 0.2

0.1 0.1
permeabilities for different 0.0
0.0
flow rates, for oil (left) and 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Water Saturation Water Saturation
water (right)

Note the following:
1. At high rates, the pseudos are shifted to the right. This is to compensate for
numerical dispersion.
2. At very low flow rates (rate 4), the flood is capillary-dominated, and the oil is
trapped. The pseudo oil relative permeability goes to zero around Sw = 0.46.
3.3.4 When to use the Geopseudo Method

Geopseudo upscaling may be time-consuming, and there is no point in upscaling from
the smallest scales, unless cores are available for the field. Cores must be studied to
identify the sedimentary structures present, and probe permeability measurements
should be taken to populate the small-scale models. Additionally, reliable SCAL
data is also required.
Ringrose et al. (1999) give a list of guidelines for when Geopseudo upscaling may
be necessary:
1) Are immiscible fluids flowing?

2) Are significant small-scale heterogeneities present? Specifically:
• Is the permeability contrast greater than 5:1?
• Is the layer thickness less than 20 cm?
• Is the mean permeability less than 500 mD?
3) What is the large-scale structure of the reservoir? In many cases, large-scale
connectivity may be the dominant issue, in which case, small-scale structure
may have to be ignored. The Weber and van Geuns (1990) classification may
be used to describe the large-scale structures:
• Layer cake reservoirs – small-scale structure will usually have primary
importance.
• Jigsaw puzzle reservoirs – small-scale structure may be important.
• Labyrinth reservoirs – small-scale structure will usually be of secondary
importance.
3.4 Uncertainty and Upscaling

During the 1990s, reservoir modelling developed (along with computing power) so
that geologists could create models containing millions of grid cells. Such models are
often time-consuming to generate, and only a few are created for each reservoir. These
detailed models are too large for full-field flow simulation, and must be upscaled to
reduce the number of cells to, about 104 or 105. Research into upscaling has focussed
on trying to develop methods to accurately upscale these types of models.
However, it is now recognised that there are many uncertainties in the reservoir
modelling, and instead of concentrating on a few detailed models, geologists and
engineers are starting to generate thousands of models in order to characterise the
effects of uncertainty. These models must be coarse so that the simulations can run
very quickly.
These changes mean that, in future, people are less likely to follow the “traditional”
upscaling approach. However, if the effects of fine-scale structure are ignored, this
44
7
will lead to errors in the predicted recovery. It is therefore very important to understand
the effects of possible sub-grid heterogeneity on absolute and relative permeability,
and to include these effects, when necessary. This is an area of active research at
Heriot-Watt University.
3.5 Upscaling Summary

Several reviews have been published on upscaling. These give an overview of some
of the methods described in this chapter: e.g. Christie (1996), Renard and Marsily
(1997) and Christie (2001).
Here is a summary of the main points:
• The effective permeability of simple permeability models (layered or random)

may be calculated using averaging.
• In general, effective permeability should be calculated using a numerical

simulation, along with suitable boundary conditions.
• Permeability upscaling is often inaccurate, particularly when the coarse cell size
is comparable, or slightly larger than the correlation length of the permeability
distribution, and when the standard deviation is large.
• Usually only single-phase upscaling is used in two-phase systems. However,

this can give rise to errors, especially when the scale-up factor is large and when
the standard deviation of the permeability distribution is large.
• Single-phase upscaling for two-phase systems may be made more accurate by

using a non-uniform coarse grid, or by using the Well Drive Upscaling method,
which increases the accuracy of single-phase upscaling by using the “correct”
boundary conditions.
• The capillary equilibrium method is useful, particularly for small-scale models.

It is feasible, even for models with a relatively large number of grid cells.
• Two-phase dynamic upscaling methods should be able to reproduce two-phase

flow on a coarse scale. The Kyte and Berry (1975) method is an example of a
pressure averaging approach.
• In general, two-phase upscaling is difficult to apply. It is more time consuming

than single-phase upscaling, and the results are not robust (negative or infinite
values may be obtained).
• The regions around wells should be treated as a special case, because the flow
is radial. The well index and the transmissibilities around the well block should
eb upscaled.
• Permeability is actually a tensor quantity (4 terms in 2D, 9 terms in 3D). Full

tensors may be used to take account of cross-flow within a grid cell. However,
in general, only the diagonal terms are used (kxx, kyy and kzz, often referred to
as kx, ky and kz).

• It is important to take account of small-scale (mm – m) heterogeneity in some
reservoirs. This may be done using the Geopseudo Method, in which models
of sedimentary structures are generated and upscaled.
• Capillary effects are often significant at small-scales, and it is important to take

these into account using two-phase upscaling (steady-state or dynamic).
46
7
4 REFERENCES
Barker, J. W. and Thibeau, S., 1997. “A Critical Review of the Use of Pseudo Relative
Permeabilities for Upscaling”, SPE Reservoir Engineering, May, 1997, 138-143.
Barker, J. W. and Dupouy, P., 1999. “An Analysis of Dynamic Pseudo-Relative Per-
meability Methods for Oil-Water Flows”, Petroleum Geoscience, 5 (4), 385 - 394.
Christie, M. A., 1996. “Upscaling for Reservoir Simulation”, J. Pet. Tech., November
1996, 48, 1004-1008.
Christie, M. A., 2001. “Flow in Porous Media – Scale Up of Multiphase Flow”,

Current Opinion in Colloid and Interface Science”, 6, 23 – 241.
Christie, M. A. and Blunt, M. J., 2001. “Tenth SPE Comparative Solution Project: A
Comparison of Upscaling Techniques”, presented at the SPE Reservoir Simulation
Symposium, Houston Texas, 11 – 14 February, 2001.
Corbett, P. W. M., Ringrose, P. S., Jensen, J. L. and Sorbie, K. S., 1992. “Laminated
Clastic Reservoirs - The Interplay of Capillary Pressure and Sedimentary Architec-
ture”, SPE 24699, presented at the 67th Annual Technical Conference of the SPE,
Washington, DC, 4 - 7 October, 1992.
Darman, N. H., 2000. “Upscaling of Two-Phase Flow in Oil-Gas Systems”, Ph.D.

Thesis, Heriot-Watt University.
Darman, N. H., Pickup, G. E. and Sorbie, K. S., 2002. “A Comparison of Two-Phase

Dynamic Upscaling Methods Based on Fluid Potentials”, Computational Geosciences,
6, 5 – 27.
Durlofsky, L. J., Behrens, R. A., Jones, R. C. and Bernath, A., 1996. “Scale Up of
Heterogeneous Three Dimensional Reservoir Descriptions”, SPEJ, 1, 313-326.
Durlofsky, L. J., Jones, R. C. and Milliken, W. J., 1997. “A Nonuniform Coarsening

Approach for the Scale Up of Displacement Processes in Heterogeneous Porous
Media”, Advances in Water Resources, 20, 335 – 347.
Durlofsky, L. J., Milliken, W. J. and Bernath, A., 2000. “Scaleup in the Near-Well
Region”, SPEJ, 5 (1), 110 – 117.
Huang, Y., Ringrose, P. R. and Sorbie, K. S., 1995. “Capillary Trapping Mechanisms
in Water-Wet Laminated Rocks”, SPE RE, 10 (4), 287 – 292.
Kyte, J. R. and Berry, D. W., 1975. “New Pseudo Functions to Control Numerical
Dispersion”, SPEJ, August 1975, 269-276.
Peaceman, D. W., 1978. “Interpretation of Well-Block Pressures in Numerical Res-

ervoir Simulation”, SPEJ, June 1978, 183-194.

Peaceman, D. W., 1983. “Interpretation of Well-Block Pressures in Numerical
Reservoir Simulation with Nonsquare Grid Blocks and Anisotropic Permeability”,
SPEJ, June 1983, 531-543.
Pickup, G. E. and Stephen, K. D., 2000. “An Assessment of Steady-State Scale-Up

for Small-Scale Geological Models”, Petroleum Geoscience, 6 (3), 203 – 210.
Pickup, G. E., Ringrose, P. S. and Sharif, A., 2000.”Steady-State Upscaling: From

Lamina-Scale to Full-Field Model”, SPEJ, 5 (2), 208 – 217.
Renard, P. and de Marsily, G., “Calculating Equivalent Permeability: A Review”,

Advances in Water Resources, 20 (5/6), 253 – 278.
Ringrose, P. S., Sorbie, K. S., Corbett, P. W. M. and Jensen, J. L., 1993. “Immiscible
Flow Behaviour in Laminated and Cross-bedded Sandstones”, J. Petroleum Science
and Engineering, 9(2), 103-124.
Ringrose, P. S., Pickup, G. E., Jensen, J. L. and Forrester, M. M., 1999. “The Ardross
Reservoir Gridblock Analog: Sedimentology, Statistical Representivity, and Flow
Upscaling”, in Reservoir Characterization – Recent Advances, eds R. Schatzinger
and J. Jordan, AAPG Memoir 71, p 256 – 276.
Stone, H. L. 1991. “Rigorous Black Oil Pseudo Functions”, SPE 21207, presented
at the 11th SPE Symposium on Reservoir Simulation, Anaheim, CA, February, 17-
20, 1991.
Weber, K. J. and van Geuns, L. C., 1990. “Framework for Constructing Clastic Res-
ervoir Simulation models”. JPT, October 1990, p 1248 – 1297.
Zhang, P., Pickup, G. E. and Christie, M. A., 2005. “A New Upscaling Approach
for Highly Heterogeneous Reservoirs”, presented at the SPE Reservoir Simulation
Symposium , February 2005.
48

ResSimCh7 PDF

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

ResSimCh7 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ResSimCh7 PDF

Uploaded by

Copyright:

Available Formats

Permeability Upscaling

1 SINGLE-PHASE FLOW 3.3.3 Geopseudo Example

After reading through this Chapter, the Student should be able to do

• Appreciate why upscaling is necessary.

• Know how to calculate effective permeability in simple models by

• Understand how to perform numerical upscaling of single-phase flow.

• Be aware of the effects of heterogeneity on two-phase flow.

• Realise the limitations of applying single-phase upscaling to a two-phase

• Know how to carry out steady-state, capillary-equilibrium upscaling for two-

• Understand how to upscale around a well.

• Appreciate that permeability is a full tensor property.

We usually refer to the upscaled permeability as the effective permeability. The

True effective permeability is an intrinsic property of the model and ought to be

Institute of Petroleum Engineering, Heriot-Watt University 3

one coarse cell

1.2 Upscaling Porosity and Water Saturation

φ = 0.15 φ = 0.20 Figure 3

The average porosity, φ , is given by:

total pore volume

Therefore, in this case:

total amount of water

Here, the average water saturation is:

1.3 Averaging Permeability

1.3.1 Flow Parallel to Uniform Layers

Institute of Petroleum Engineering, Heriot-Watt University 5

1.3.3 Flow through Correlated Random Fields

where i = 1, 2, .. n is the number of cells in the distribution.

Institute of Petroleum Engineering, Heriot-Watt University 7

 75 × log(10 ) +125 × log(100 ) 

1.3.4 Additional Averaging Methods

1.3.5 Summary of Permeability Averaging

1.4 Numerical Methods

1.4.1 Recap on Flow Simulation

(We are assuming incompressible rock and fluids, here.)

qxin qxout i-1,j i,j i+1,j

It is useful to use the transmissibilities, Tx = kxΔz/Δx and Tz = kzΔx/Δz. (Assume

An equation is set up for each Pij, i = 1, 2, .. nx and j = 1, 2, .. nz. The transmissibilities

1.4.2 Boundary Conditions

Institute of Petroleum Engineering, Heriot-Watt University 9

no flow through the sides

(b) Periodic Boundary Conditions

(c) Linear Pressure Boundary Conditions

(d) Flow Jacket, or Skin

Institute of Petroleum Engineering, Heriot-Watt University 11

keff calculated for

1.5 Upscaling Errors

1.5.1 Correlated Random Fields

0.04 0.04 Figure 15

0.04 standard deviation = 0.5.

a) small σ, large λ b) large σ, large λ

1.5.2 Evaluating the Accuracy of Upscaling

Institute of Petroleum Engineering, Heriot-Watt University 13

4 × 4, 5 × 5, 8 × 8, 10 × 10, 16 × 16, 20 × 20, 40 × 40, 80 × 80

0.4, 0.5, 0.8, 1.0, 1.6, 2.0, 4.0, 8.0.

The conclusions from these examples are:

• Low permeability shales in a high permeability sandstone

1.6 Summary of Single-Phase Upscaling

• Fine-scale geological models usually require upscaling for full-field