SPE 54310
Stochastic Modeling for the Heterogeneous Anisotropic Reservoir System by Using
Geostatistics
Dongseong Lee, Research Institute of Engineering Science, Seoul Nat'l Univ. and Joe M. Kang, SPE, Seoul Nat'l Univ.
Copyright 1999, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the 1999 SPE Asia Pacific Oil and Gas
Conference and Exhibition held in Jakarta, Indonesia, 20–22 April 1999.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
We present a method to characterize a heterogeneous
anisotropic reservoir by integrating both core data and well
testing data.
To characterize a heterogeneous anisotropic reservoir, we
introduce coordinate transformation which convert the
anisotropic reservoir into the isotropic one. Anisotropy can be
expressed by grid block permeability or correlation length or
both of those. For each case coordinate transformation
function is developed respectively. On the transformed
domain simulated annealing is employed to optimize the
permeability field. Then final heterogeneous anisotropic
permeability field is obtained by inverse transformation. To
predict the breakthrough time, the simulation of water
flooding is performed stochastically on the several types of
characterized anisotropic fields.
The results show that trend of pressure and permeability
image of the characterized field is close to that of the base
field. The breakthrough time can be predicted more accurately
within a confidence interval when water flooding is simulated
on the characterized heterogeneous anisotropic reservoirs. The
proposed method can reduce the reservoir uncertainty
significantly since both well testing data and geological data
are used.
Introduction
Fluvial reservoirs or massive fractured reservoirs appear to
exhibit anisotropy in permeability1. Reservoir characterization
and prediction of performance for those formations can play
an important role in reservoir management and secondary
production design.
Well-test permeability derived from pressure transient
analysis is important information about the permeability
distribution around the testing well. It is also available for
many active wells in a reservoir. Nevertheless, conventional
approach seldom incorporate this information into an
anisotropic reservoir model because no appropriate method is
developed to incorporate single well-test data into
characterizing an anisotropic reservoir.
In case of single well test on an anisotropic reservoir,
testing well always feels the reservoir as an isotropic one
through the pressure drawdown whether it is isotropic
reservoir or anisotropic reservoir. This reason makes it
difficult to use single well test data for an anisotropic reservoir
characterization.
In order to apply single well test data to an anisotropic
reservoir characterization, we develop transformation
equations which can convert an anisotropic reservoir into an
isotropic reservoir.
In this study a new method is proposed to characterize the
heterogeneous anisotropic reservoir by integrating well testing
data and geological data and then water flooding is simulated
stochastically to check how much reduced the uncertainty of
the reservoir performance can be on the characterized fields.
Background theory and methodology
As emphasized earlier, the goal of this work is to propose the
technique for characterizing heterogeneous anisotropic
reservoirs.
There are two main difficulties in characterizing an
heterogeneous anisotropic reservoir by incorporating single
well-test data. First, it is very difficult to define the radius of
investigation of the anisotropic reservoir in transient single
well testing. Second, the relations between well test derived
permeability and grid block permeability is not defined yet on
the anisotropic reservoir.
But these two problems above are solved on isotropic
heterogeneous reserviors by Alabert, Oliver, Feitosa, and
Sagar et al2-6. Therefore we derive the suitable coordinate
transformation method to apply the useful result achieved
form the isotropic heterogeneous reservoir to the anisotropic
reservoir characterization.
As the first simplifying assumption, the spatial variability
of reservoir properties is limited here to that of permeability.
2 DONGSEONG LEE, JOE M. KANG SPE 54310

Furthermore, only lateral permeability variations are 4. Repeat the above procedures for generating a permeability
considered. The reservoir is thus modeled as two-dimensional field of the minimum principal direction, y.
with constant thickness. Single phase flow is considered. Fluid
5. Assign the values above for x as x direction grid block
properties, compressibility and porosity are uniform over the
permeabilities and permeabilities for y as y direction grid
reservoir. Reservoir characteristics are summarized in table 1.
block permeabilities.
The main steps taken in this study are as followings.
The first step is to generate a base case. Anisotropic Many different multi lognormal permeability models are
permeability variations are modeled using conditional
simulation technique. The second step is a transient single generated. Their statistics are as followings. Means µ k ranges
well testing and its interpretation. The well testing is between 10 and 30 md; Dykstra-Parsons coefficients are 0.4
numerically simulated at the center of the stochastic reservoirs and 0.7. All semi-variogram models are spherical with
using a simulation scheme and its interpretation is conducted dimensionless ranges λD = a L between 0 and 1, where L is
by ISA. The third is the derivation of a coordinate
the half size of the reservoir. In anisotropic cases, anisotropic
transformation through the comparison between well test
correlation functions are considered, with the ratio of the
permeabilities and individual grid block permeabilities.
shortest to the largest correlation range valued 0.2 and 0.6.
Relations of individual block permeabilities and well testing
permeabilities are briefly discussed and the coordinate
Well testing and its analysis. Within each permeability field
transformation equations are derived for 3 types of anisotropic
generated with the techniques described above, a constant rate
reservoirs. Then we describe the technique of simulated
drawdown (pumping) test is numerically simulated at the
annealing and the procedure of heterogeneous anisotropic
central grid block. No-flow boundary conditions are imposed
reservoir characterization. The characterizations are conducted
on each side of the square domain. A fully implicit finite
for various anisotropic reservoirs. Finally, the application of
difference scheme is used to simulate the drawdowns on the
this method to water flooding is presented to check how much
square grid.
it reduces an uncertainty on the reservoir characterization.
For each of the realizations of a particular permeability
distribution, simulated drawdown data can be interpreted by
permeability field generation. First step is to generate
ISA(Inverse Solution Algorithm). ISA gives the permeability
synthetic anisotropic permeability values in stochastic manner
information according to the distance from the testing well.
which are used as input data for numerical reservoir
simulation. Stochastic modeling techniques allow building
Averaging process of a well test permeability. To
synthetic permeability fields whose spatial patterns are neither
effectively use a well-test derived permeability in a
completely erratic nor perfectly deterministic by conditioning
conditional simulation method, the well-test permeability must
the available data.
impose a deterministic constraint on the distribution of
A 2-D(x-y) anisotropic permeability field used in this study is
surrounding small scale grid block permeabilities within the
made by combining two set of isotropic permeability fields
radii of investigation. Therefore, an averaging model that
with different statistics of each other. Generally permeabilities
defines the permeability distribution within the radii of
are considered as random variables with log-normal
investigation has to be set up. The analytical solutions
distribution7. Permeability field of each other is generated by
proposed by Oliver and Feitosa et al. are investigated to
use of sequential Gaussian simulation(sGs) method provided
determine the averaging process and the region of
by GSLIB8. Desired statistics like mean, Dykstra-Parsons
investigation.
coefficient( VDP ) and correlation length are given to The radii of investigation will be the annular region
permeability values according to the principal directions of the defined by the kernel function. For a dimensionless time, t D ,
anisotropic field. For simplicity, x-direction is considered as
the maximum principal direction and y-direction is considered the radius of investigation is defined between rD = 0.12 t D
as minimum principal direction.
The steps for anisotropic lognormal permeability fields are and 2.34 t D
as followings, where,

1. Assume x direction as a maximum principal direction and k wt t

t D = 2.637 ×10 − 4 ………………………….. (1)
determine the statistics of the maximum principal direction as φµct rw2
we desire.
2. For x direction permeabilities, generate random variables
and k wt is the permeability calculated from the best-fit
with the normal distribution and the desired statistics by using
sequential Gaussian simulation method. semilog straight line slope.
3. To get the log normal permeability field, the step of An area-based power averaging scheme is considered. The
exponentiating the normal field is taken. following area-based averaging equation is used
SPE 54310STOCHASTIC MODELING FOR THE HETEROGENEOUS ANISOTROPIC RESERVOIR SYSTEM BY USING GEOSTATISTICS 3

(k ⋅ Ar )ω = ∑ (k ( s ) ⋅ Ab ( s ))ω …………...…………. (2) After transformation, the governing equation is changed into
Eq.6.
s

∂ 2 P ∂ 2 P φµCt ∂P
+ = ………………………….. (6)
Where s is the location vector of the grid block, Ar is the ∂x ' 2 ∂y '2 k x k y ∂t
radial investigation area at a given time by ISA, and Ab is the
Just like the homogeneous case, the governing equation for the
area of the grid block that falls in the radial investigation area.
heterogeneous anisotropic reservoir can be expressed by Eq.7.
Sagar showed that on an isotropic heterogeneous reservoir, the
equivalent radial permeability, k determined in geometric ∂  ∂P  ∂  ∂P 
 k x ( x , y )  +  k y ( x, y ) 
averaging( ω = 0 ) suitably approximates the instantaneous ∂x  ∂x  ∂y  ∂y 
well test permeability from the ISA method. ………...…. (7)
∂P
= φµC t
ln k ⋅ Ar ≈ ∑ ln k ( s ) ⋅ Ab ( s ) ……………………….. (3) ∂t
s
We suggest transformation equation as the followings,
Derivation of coordinate transformation equation. It is
known that existing technique can characterize the
permeability distribution by integrating the statistics data from ky kx
core analysis and permeability from well-testing. But in that
x' = 4 x, y' = 4 y ……………….……. (8)
kx ky
case well-testing data can only be used to characterize the
reservoir if the reservoir is isotropic. Where,
To characterize the anisotropic reservoir, we derive the
1
∑k
coordinate transformation equation which can convert an
anisotropic reservoir into an isotropic reservoir. kx = x (i, j ) ………………….……………. (9)
nx n y i, j
In this study, we categorize the permeability anisotropic
fields into 3 followings, anisotropic field of grid block
1
permeability, anisotropic field of permeability correlation
length, and composite scale anisotropic field of permeability.
ky =
nx n y
∑k
i, j
y (i, j ) ………….…………………. (10)
Anisotropic field of grid block permeability is the field in
which the mean of permeabilities is different on each principal Eq.7 can be converted into Eq.11.
direction but correlation length of permeability field is the
same irrespective of the direction. Anisotropic field of ∂  k y ∂P  ∂ 
+
k ∂P 
permeability correlation is the field in which the mean of k x ( x, y )  k y ( x, y ) x 
∂x'  k x ∂x'  ∂y '  k y ∂y ' 
permeabilities is the same in all directions but correlation  
length is different according to the direction. Composite scale
∂P
anisotropic field of permeability is the field in which both = φµC t ............................................................. (11)
mean and correlation length of permeability are different ∂t
according to the direction. For each case, the coordinate
transformation equation is developed. If the ratio of k x ( x, y ) / k y ( x, y ) is equal to the ratio of
1) Transformation equation for anisotropic field of grid
k x / k y , the Eq.<11> is changed into Eq.12 which is the
block permeability
Governing equation of the homogeneous anisotropic reservoir governing equation of heterogeneous isotropic reservoir.
can be expressed by Eq.4.
∂  ∂P  ∂  ∂P 
+  ∂y ' 
 k x ( x, y ) k y ( x, y )  k x ( x, y ) k y ( x, y )
∂ P
2
∂ P 2
∂P ∂x'  ∂x'  ∂y ' 
kx + k y 2 = φµCt ……...………………. (4)
∂x 2
∂y ∂t ∂P
= φµC t ................................................................. (12)
And that can be replaced by an equivalent homogeneous ∂t
isotropic reservoir through the shrinking or expansion of the Anisotropic field of grid block permeability can be changed
coordinate system for which the transformed coordinates from into isotropic field.
(x, y) to (x', y') are defined by
2) Transformation equation for anisotropic field of
ky permeability correlation length
kx
x' = 4 x, y' = 4 y ……...…………….... (5) In this case, the permeability correlation length is different
kx ky according to the direction though mean of permeability is the
same regardless of the direction. In our study, we consider the
4 DONGSEONG LEE, JOE M. KANG
variogram of geometric anisotropy type in which the
correlation length of the x-direction is different from that of
the y-direction but the seal values of each direction is the
same.
Although on a local scale(individual grid block) we have
the isotropic medium, on a larger scale, due to anisotropy in
correlation length, we have an apparent anisotropic
permeability medium.
The elliptical shape of pressure propagation on this field is
similar with that on the anisotropic field of grid block
permeability even the mean of grid block permeability of each
principle direction is isotropic. Our example in which
a x > a y results in the elliptical drainage profile shown in
Fig.1. The statistics of the reservoir is as followings;
k x = k y = 20md , a x = 2,000 ft , a y = 400 ft , and
VDP = 0.7 . This effect will also be observed if k x > k y .
It is known that on the isotropic heterogeneous field, area
based geometric average of grid block permeabilities within
the radius of investigation properly approaches well test
permeability from ISA. This fact must be satisfied by the new
transformation. Just like the anisotropic field of grid block
permeability, we hypothesize that the transformation equation
is the ratio of correlation length. In order to find a suitable
coordinate transformation equation, we change ∆x , ∆y
according to the following Eq.13 until the difference between
the area-based geometric average of the grid block
permeabilities and well test permeability is minimized.
ay ax
∆x' = 2n ∆x , ∆y ' = 2 n ∆y ……………… (13)
ax ay
Fig.2 shows the suitable n value for the above example which
minimize the difference at a given field. In this case, n is about
3. Although we can’t find the universal n value in this study,
but it is not difficult to find the suitable n for the
transformation at the given field which you want to
characterize.
3) Transformation equation for composite scale
anisotropic field of permeability
A composite scale anisotropic field of permeability is the
anisotropic field in which both mean value of permeability and
correlation length are different according to the direction.
In this case, the transformation can be made by applying
two former cases sequentially.
Characterization by the annealing. This step is to solve the
inverse problem; characterizing a permeability field with the
known data. We assume that the univariate and spatial
statistics of the distribution and the transient pressure response
of the permeability field are known a priori.
Incorporating the conditioning data, the permeability
distribution function and the variogram constraints follows the
conventional simulated annealing approach. In that case, the
energy function for the variogram is as
SPE 54310
Nd
 γ k (hi ) − γ 0 (hi ) 
E1k = ∑ 
i =1  γ 0 (hi )  ….…………………. (14)

We simulate the annealing procedure until kik ≈ k ISAi for
each t D . For N d different time step, we define N d different
radii of investigation. The energy function incorporating the
well-testing information can be written as
Nd
 k ik − k ISAi 
E 2k = ∑ 
i =1  k ISAi 
 …………………………... (15)
And the overall energy function is given by
E1k Ek
E k = Ψ1 + Ψ2 2 ……………………………. (16)
E01 E02
The annealing algorithm must simultaneously satisfy the
variogram and conditioning data constraints and the overall
energy function within some tolerance.
The procedure of characterization is briefly here. At a
given field we want to characterize, we obtain static statistics
from the core data and dynamic statistics from the well test.
On the base of permeability mean values, first transformation
is taken and on the base of correlation length anisotropy the
second transformation is taken considering a suitable n value.
After first and/or second transformation, we conducted
simulated annealing on the transformed isotropic domain by
incorporating static data and dynamic data. Then we can get
the final field by back transformation in the reverse manner of
the transformation. The whole procedure is shown at Fig.3.
Results and discussion
In this section, the results of characterization are presented and
discussed. For space constraints, we present the results of
characterizations on the type three anisotropic reservoir:
composite scale anisotropic field of permeability.
By use of the simulated annealing and the transformation
technique outlined above, we reproduced permeability fields
such that the spatial and univariate statistics and the pressure
transient data of the synthetic base case were honored. The
statistics of the base case is V DP =0.7, a x =3,200 ft, a y =550
ft, k x =30md, k y =10md, and the other geometric data is the
same as table 1. Fig.4 shows the fluid simulation results of the
pressure responses compared to the base case pressure
response. The match between the pressure and pressure
derivatives of the base case and reproduced permeability fields
shows a good agreement. The base case permeability field and
the reproduced permeability distributions are shown in Fig.5.
We can see the image of the characterized permeability field is
similar to that of base case. For application of this method,
water flooding is simulated stochastically to check how much
it reduces the uncertainty of the reservoir performance. The
SPE 54310STOCHASTIC MODELING FOR THE HETEROGENEOUS ANISOTROPIC RESERVOIR SYSTEM BY USING GEOSTATISTICS 5

propagation of water front is shown in Fig.6. On the k ISA = well test permeability from ISA method, L2 , md
characterized fields with variogram constraints and well test
data, the propagation shape is much similar to that of base ND = number of radius of investigation
case. Ψ= weighting factor
We also examine the breakthrogh time on the
characterized fields. The comparison of breakthrough time
E0 = initial base energy state
between the fields characterized with only variogram data and
the fields characterized with both variogram data and well test References
imformation clearly shows a reduction of uncertainty on the 1. Ramey, H.J.Jr.: "Interference Analysis for Anisotropic
anisotropic heterogeneous reservoir descriptions(Fig.7). Y- Formations-A Case History," JPT (Oct. 1975).
2. Alabert, F.G.: "Constraining Description of Randomly
axis means breakthrough time (day).
Heterogeneous Reservoirs to Pressure Test Data: A Monte
Carlo Study," paper SPE 19600 presented at the 1989 Soc. of
Conclusions Pet. Eng. Ann. Tech. Conf. and Exh., San Antonio, TX, Oct.
We found that the method proposed in this study can be used 8-11.
to characterize the anisotropic heterogeneous reservoir 3. Oliver, D.S.: "The Averaging Process in Permeability
successfully by the transformation technique. After a suitable Estimation From Well-Test Data," SPEFE (Sept. 1990) 319.
coordinate transformation, anisotropic reservoirs can be 4. Oliver, D.S. :"Estimation of Radial Permeability Distribution
treated as a isotropic reservoirs. On the transformed domain, From Well-Test Data," SPEFE (Dec. 1992) 290.
we solve the inverse problem by the simulated annealing. 5. Feitosa, G., L. Chu, L.G. Thompson, and A.C. Reynolds:
"Determination of Radial Permeability Distribution from
Then the final permeability field can be obtained by the back
Well-Test Pressure Data," JPT (July 1994) 607-615.
transformation. The important conclusion of this work are as 6. Sagar, R.K., B.G. Kelkar, and L.G. Thompson: "Reservoir
followings. Description by Integrating Well-Test Data and Spatial
1. The transformation equation for the anisotropic field of Statistics," SPEFE(1995) 267-274.
n 7. Journel, A. and C.J. Huijbregts: Mining Geostatistics,
correlation length can be expressed by the 2 root of the
ratio of principal correlation length. Academic Press, New York, NY (1978)
8. Deutsch, C.V. and A.G. Journel: GSLIB Geostatistical
2. In case of the composite scale ansiotropic field of
Software Library and User's Guide, Oxford Univ. Press, New
permeability, we can convert the anisotropic reservoir into York, NY, 1992
isotropic reservoir by taking two transformation steps
sequentially; the transformation for permeability anisotropy
and the transformation for correlation length anisotropy.
3. The pressure responses of the simulated permeability
distribution were consistent with the pressure responses of TABLE 1—Summary of input data used for
the base case distribution within 10 % error. simulation study
4. We observe that the image of reproduced permeability fields Reservoir Length 4,025 ft
is good agreement with that of base case. configuration Width 4,025 ft
5. By the stochastic approach, we show that the uncertainty of Thickness 10 ft
the performance prediction can be reduced significantly by Rock properties Porosity 30.0%
Viscosity 0.4 cp
4.9 × 10 1/psi
the proposed method on anisotropic heterogeneous Fluid properties Compressibility -7

reservoirs. FVF 1.2 bbl/STB

Well condition Wellbore radius 0.5 ft
Nomenclature Well location 2,012.5ft × 2,012.5 ft
a = correlation length, L, ft Flow rate 106.3 STB/D
Initial condition Initial pressure 2,000 psi
λD = dimensionless length Grid block No. of grid blocks 115 × 115 × 1
VDP = Dykstra-Parsons coefficient configuration Block dimension 35 ft × 35 ft × 35 ft

∆x = grid block length for x, L, ft

∆y = grid block length for y, L, ft
L = half lengh of the reservoir, L, ft
x = major principal direction, L, ft
y = minor principal direction, L, ft
k = average of grid block permeabilites, L2 , md
ki = well test permeability of i-th region, L2 , md
k wt = well test permeability from honor plot, L2 , md
6 DONGSEONG LEE, JOE M. KANG SPE 54310

Fig. 2 Contour map of absolute difference between area-base

averaged grid block permeabilities and ISA well test permeability.

Fig. 1 Pressure distribution on the global scale anisotropic

reservoir.

Fig. 3 Schematic representation of coordinate transformation process

SPE 54310STOCHASTIC MODELING FOR THE HETEROGENEOUS ANISOTROPIC RESERVOIR SYSTEM BY USING GEOSTATISTICS 7

Fig. 4 Comparison of pressures and pressure derivatives of base

case and simulated realization

Fig. 5 Base case and simulated permeability distribution honoring

semi-variogram and ISA constraint (a) base case (b) simulated
case 1 (c) simulated case 2.
8 DONGSEONG LEE, JOE M. KANG SPE 54310

Fig. 6 Distributions of pressure and water saturation

Fig. 7 Uncertainty interval of breakthrough time