SPE 38653
A Procedure to Integrate Well Test Data, Reservoir Performance History and 4-D
Seismic Information into a Reservoir Description
Jorge L. Landa, SPE, and Roland N. Horne, SPE, Stanford University
Copyright 1997, Society of Petroleum Engineers, Inc.
This paper was prepared for presentation at the 1997 SPE Annual Technical Conference and
Exhibition held in San Antonio, Texas, 5–8 October 1997.
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
This paper deals with the problem of estimating the
distributions of permeability and porosity in heterogeneous
and multiphase petroleum reservoirs by matching the dynamic
behavior. The dynamic data is in the form of field
measurements from well testing, production history,
interpreted 4-D seismic information, and other data such as
correlations between permeability and porosity, geostatistics in
the form of a variogram model and the inference of large scale
geological structure.
The issue was posed as an inverse problem and solved by
using nonlinear parameter estimation. The procedure
developed here is capable of processing all the information
simultaneously and this results in a fast and efficient method.
The procedure is also able to determine the uncertainty
associated with the estimated permeability and porosity fields.
Examples of different parameter types that may be
estimated by this approach include: (a) individual block
permeabilities and porosities; (b) geological objects such as
channels and faults; (c) pilot points that form the basis of a
kriged distribution; and (d) seismic attenuation values from 3-
D seismic images.
An important conclusion of this work is that the value of
each piece of information does not reside in its isolated use but
in the value it adds to integrated analysis of the complete set of
information. Thus data that traditionally was considered to be
of low information content for reservoir characterization
becomes useful and enhances the value of the data set as a
whole.
Introduction
Devising the optimal strategy for the development of an oil or
gas reservoir is an important and difficult task. Many
mathematical techniques for optimization can be used to deal
with problems in engineering and economics systems. These
techniques assume that we have a fairly complete
understanding of the problem and also that we can construct a
mathematical model that predicts the system’s performance
accurately in time under different scenarios; this is not a
serious concern in most engineering problems since the
parameters that define the system may not be very difficult to
obtain by direct measurement. Unfortunately this is not the
case in reservoir engineering, where the system, that is the oil-
gas reservoir, is physically inaccessible many thousands of feet
underground. Thus, any serious attempt at optimization of
reservoir development first requires the determination of the
parameters of the reservoir and the only way to obtain them is
through indirect measurement. The process of inferring the
parameters from the indirect measurements is an inverse or
parameter estimation problem. Such is the focus of this work.
Permeability and porosity are the parameters that have the
largest influence in determining the performance of the
reservoir, and thus, this work addresses the problem of
estimating permeability and porosity from a variety of
measurements that are only indirectly related to them.
Estimating permeability and porosity is difficult for the
following reasons:
• Permeability and porosity have spatial variability.
• There are very few sampling locations (wells)
compared to the areal extent of the reservoir.
• Information (data) is scarce.
• Measurements are obtained with different
technologies.
• The mathematical model of the reservoir is very
complex, usually consisting of a numerical reservoir
simulation.
2 JORGE L. LANDA AND ROLAND N. HORNE
The data set defining a reservoir is derived from
measurements that are related to the permeability and porosity
according to different laws or complex relationships. It is
common to try to solve the inverse problem by working on
each set of measurements independently of the others. In
general each individual interpretation may not be fully
consistent with the others, and as a result the final
interpretation comes after a long process of iteration to ensure
compatibility. As a final step the interpretations are validated
by history matching, since it would not be reasonable to rely
on a model that could not predict the past performance. Since
data is being collected almost continuously, the process of
updating the reservoir model never ends.
During the producing life of a reservoir, data of different
nature are always being collected. These data can be classified
as static or dynamic depending on their association with the
movement or flow of fluids in the reservoir. Data that have
originated from geology, electrical logs, core analysis, fluid
properties, seismic and geostatistics can be generally classified
as static, whereas the information originating from well
testing, pressure shut-in surveys, production history, bottom
hole pressure from permanent gauges, water-cut, and gas-oil
ratio (GOR) can be classified as dynamic.
A special consideration needs to be applied in the case of
4-D1 seismic information, which is a relatively new technology
developed in the field of geophysics. With this process it is
possible to estimate the areal distribution of change of
saturation in the reservoir due to the production or injection of
fluids. Since 4-D seismic information is related to the
movement of fluids in the reservoir it can be classified as
dynamic data. One of the outstanding features of the 4-D
seismic information is that it is areally distributed whereas the
other dynamic data are available only at the location of the
production or injection wells.
The parameter estimation problem would not only be faster
but also more reliable if it were performed with a process that
uses all or at least most of the information in the reservoir data
set simultaneously. The process of handling different data
simultaneously is known as data integration.
So far, most of the success in data integration has been
obtained with static information. Remarkably, it has not yet
become common to completely or systematically integrate
dynamic data with static data and it is currently the subject of
major research effort in several places2-30. This work addresses
this specific problem and represents a number of steps in the
direction of full integration.
Problem Statement
An example of the type of problem we want to solve has been
summarized in Fig. 1, where (a) shows a two-dimensional
SPE 38653
reservoir map in which the permeability and porosity are the
unknowns to be determined. There are four wells (marked with
triangles), one of them is injecting water and the others are
producing. The production rates and injection rates are known.
Figs. 1 (b)-(e) show the water distribution in the reservoir as a
function of time, this is not known to us but is shown here only
to illustrate the problem.
Figs. 1 (f)-(p) show graphically the field observations that
can be available to us. These observations are:
• 4-D seismic interpretation1. Fig. 1 (f) and (g) show two
changes of water saturation (∆Sw) maps. These maps are
assumed to have been made available after three
consecutive 3-D seismic surveys. The seismic surveys
were performed at the time when the reservoir was in
states shown in (b), (c) and (d). The geophysical
interpretation cannot provide us with the value of the
water saturation at each time, but it can provide the
change of saturation in the reservoir between two surveys.
• Permeability-porosity correlation. Fig. 1 (h) shows such a
correlation. This correlation may be obtained from
measurements in cores from the wells. The maximum and
minimum values possible for the permeability and
porosity are also specified, these values may come from a
priori geological information
• DST pressures. Drill stem tests (DST) have been
performed at each well and the information is available
(Figs. 1 i-l). The information includes a relatively large
number of pressure measurements over a small period of
time. Each DST consisted of a single flow followed by a
build-up. The DST’s are performed early in the
productive life of the reservoir, thus only a single phase
(oil) was produced.
• Pressure from permanent gauges. The solid lines in Figs.
1 (m)-(p) show the bottom hole pressure at each well as a
function of time. At late time the permanent gauges
recorded a simulated shut-in pressure at each well. These
events can be planned or may be due to operational
reasons. The shut-in is of shorter duration than the DST,
and fewer measurements are made
• Water cut. The dotted lines in Fig. 1 (m)-(p) show the
water cut measured at each well. This assumes that
production of each well is being measured individually
and continuously.
Other information that is available includes:
• Fluid and rock properties. Except for the permeability and
porosity, properties such as viscosity, density, relative
permeability and compressibility are known.
• Large scale geological information.
• Geostatistical information. Such as a variogram model.
Our interest is not only in finding distributions of
permeability and porosity in the reservoir that result in a match
of the data but also in other issues that are as important as the
match. Thus we look at:
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION
• What are the data we need to determine the reservoir
parameters.
• What parameters can be resolved with a given data set.
• What is the meaning of the calculated parameters.
• What is the uncertainty associated with the parameter
estimates.
Theory
The process of inversion to determine values of reservoir
parameters, such as permeability and porosity, from indirect
measurements is referred to as a parameter estimation
problem (also referred to as an inverse problem). The usual
approach to solve the parameter estimation problem is by
going through three major steps, not only for the specific case
of this work but for any general problem. These steps are as
follows:
1. To construct a mathematical model.
2. To define an objective function.
3. To apply a minimization (referred to in this work as the
parameter estimation algorithm).
Once the mathematical model has been constructed, the
objective function has been defined, and the minimization
algorithm has been chosen, the procedure for inversion works
by the following steps:
1. Assign an arbitrary, but reasonable, value to the unknown
r
set of parameters α , this is referred to as the first guess.
2. Compute the response of the system with the mathematical
model.
3. Compute the objective function, which compares the
calculated response of the system to the actual set of
measurements. If the objective function is less than a
certain predetermined value then STOP.
4. Use the minimization algorithm to compute a change in
the set of parameters. If the change in the set of
parameters is less than a certain predetermined value then
STOP
5. Return to Step 2.
Mathematical Model
The physical system under study is represented by a
mathematical model that is constructed by applying the
fundamental physical laws that are relevant to the problem.
The purpose of the mathematical model is to predict with
reasonable accuracy the behavior of the system under different
conditions. The problem of computing the response of the
mathematical model to an external perturbation is referred to
as the forward problem. The physical properties that remain
invariant for different problems are referred to as parameters
of the system. The ones that change are referred to as
variables. The opposite problem, the inverse problem, consists
of finding a set of parameters for a given model such that the
predicted behavior of the system replicates the true behavior
3
(measurements) under the same set of external conditions. In
this work the physical system under study is a reservoir. The
following fundamental laws are relevant to the dynamics of the
reservoir:
1. Mass conservation law.
2. Darcy’s law.
3. Equation of state.
4. Relative permeability and capillary pressure relationships.
The mathematical model is constructed by combining these
laws and results in a system of differential equations. In a few
cases, such as in traditional well testing theory, it is possible to
obtain an explicit solution to the differential equations, but in
the general multiphase, multiwell, heterogeneous case this is
not possible and therefore it is necessary to resort to numerical
methods to obtain a solution. In this work the mathematical
model consisted of a numerical reservoir simulation. The
r
behavior d cal of the system is represented by Eqn. 1.
r r r
d cal = d cal ( α ) (1)
r
where α ∈ R npar is the vector of the parameters of the
mathematical model. Most or all of the parameters are directly
related to the distribution of permeability and porosity in the
reservoir.
Parameterization of the mathematical model
Since in our approach we use a numerical simulator our
parameter estimation problem is to find the permeability and
porosity we should assign each cell of the simulation grid such
that the calculated data replicates the field observations. This
problem can be translated to the inverse problem formulation
from two perspectives: pixel and object modeling27,28.
r
Pixel modeling. In this approach the parameters α for the
inverse problem are the permeability and porosity at each cell
of the simulation grid. Thus the number of parameters is twice
the number of simulation cells.
Object modeling. In this approach the permeability and
porosity at each cell of the simulation grid is a function of a set
r
of parameters α , that is
r
ki = ki ( α ) (2)
r
φi = φi ( α ) (3)
Fig. 5 (a) shows an example of a channel reservoir that can
be modeled by either of the approaches. There are two main
purposes in object modeling. The first is to preserve the large
scale geological information, that is if we model the reservoir
as a channel the result of the inverse problem will always be a
channel. The second purpose is to reduce the dimension of the
inverse problem, for example the channel reservoir shown in
Fig. 5 (a) can be parameterized with only eight parameters
4 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653

which is fewer than and independent of the number of cells in

r
the simulation mesh. This parameterization is illustrated in H GN ∆α = −∇E (7)
Fig. 3.
HGN is Gauss-Newton approximation to the Hessian matrix
Objective Function of E (matrix of the second derivatives of E).
The objective function is a measure of the discrepancy
between the information, that is the data, and the response When Eqn. 4 is used, ∇E and HGN are calculated as
calculated by the mathematical model using the current set of follows:
parameters. Eqns. 4 and 5 show two forms of the objective r r
function that are commonly used. ∇E = −2G TW ( d obs − d cal ) (8)
H GN = 2G TWG (9)
r r r r
( ) ( ) r
T

E = d obs − d calc W d obs − d calc (4)

where G is the matrix of the first derivatives of d cal , also
referred to as sensitivity coefficients.
r
1 r
( r
) C (dr r 
) ∂d cal
T −1
E=  d obs − d calc d obs − d calc  + G= r (10)
2  ∂α
(5)
(
1 r r
) −1 r r 
C α  α − α prior  
T
α − α prior When Eqn. 5 is used ∇E and HGN are calculated as
2  
follows:
Eqn. 4 is referred to as the weighted least square problem. r r r r
Eqn. 5 is referred to as the generalized least square problem, ∇E = −G T C d−1( d obs − d cal ) + Cα−1( α − α prior ) (11)
this form was developed from theory of probability 31, 32, 33 and −1 −1
H GN = G C G + C
T
α (12)
was used by Oliver13 to integrate well test data into reservoir d

description.
The parameter estimation algorithm will converge to
r
Parameter Estimation Algorithms α * when the following conditions are met:
Common to all parameter estimation algorithms is that they try
r
to minimize a discrepancy function (objective function). ∇E( α * ) ≤ ε1 (13)
r
One of the characteristics of reservoir parameter estimation E( α * ) ≤ ε 2 (14)
is that the objective function E is nonlinear with respect to the r
parameters; consequently all the algorithms rely on iterative where ε1 and ε2 are small positive numbers. α * is referred
procedures that minimize by a succession of changes to a to as an optimal point and should provide a set of parameters
given first set of parameters. that results in a good match of the data.

There are many methods30, 34, 35 to minimize the objective
function. These methods are usually classified depending on
whether they use the gradient of the objective function or not.
The gradient of E is defined as:
 ∂E 
T
Computation of the Sensitivity Coefficients
As shown in Eqns. 8, 9, 11, and 12. The computation of G
(Eqn. 10) is crucial to the Gauss-Newton method. The
efficient evaluation of this matrix has been the subject of
intensive research2, 3, 4, 5, 14, 21, 23, 25, 27, 28.

∇E =  r  (6) Because of the different types of field observations it is

 ∂α 
necessary to compute the sensitivity of the calculated data with
respect to the parameters of the problem, that is:
In this work we used the Gauss-Newton method, which is
classified as a gradient method. The Gauss-Newton algorithm
r ∂p ( t )
was combined with the Marquardt stabilization method to si p ( t ) = i r (15)
compute the direction for the change in the parameters, and ∂α
r wc ∂wci ( t )
with a linear search procedure to calculate the size of the si ( t ) = r (16)
change of parameters at each iteration. The constraints on the ∂α
parameters were included by using penalty functions. r ∂∆S w j ( t )
s j∆S ( t ) =
w
r (17)
∂α
At each iteration of the Gauss-Newton algorithm a linear
systems of equations is solved (Eqn. 7): where pi is the pressure measured in the i well, wci is the
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 5

water cut measured in the i well, and ∆Sw j is the change of

water saturation in the j cell of the simulation grid After this approximation it is possible to apply the theory
All the field observations are directly related to the of resolution and variance developed for linear models33,36.
pressure and saturation calculated by the numerical reservoir This theory is based on the singular value decomposition of G.
simulator. Thus we first compute the sensitivity of the
calculated pressure and saturation with respect to the G = USV T = U p S pV pT (25)
parameters of the problem, and then convert to the data
dimension. where U and V are orthogonal matrices and S is a diagonal
In this work the sensitivity coefficients are calculated with matrix of the singular values of G. p is the number of nonzero
a multiphase extension of the Jacobian method shown in singular values. Up, Vp and Sp are the matrices constructed
Appendices A and B of Ref. 27 and detailed in Ref. 28. The with the columns of U, V and S corresponding to the nonzero
method is a variant of the “gradient simulator” approach singular values. The generalized inverse G-g is calculated as:
described by Anterion et al.21 The method is relatively
straightforward to implement and can be applied to both pixel
G − g = V p S p−1U Tp (26)
and object modeling.
r
Thus the vector of sensitivities Z with respect to the
Thus the resolution matrix R is calculated as:
parameter αi at the time step k+1 of the time discretization in
the simulator is obtained from Eqn. 1828.
R = V pV pT (27)
r r r
JZαk +1 = DZαk + Wαk +1
i i i
(18)
and the covariance matrix of the parameter estimates
with r
r r Cov{ α * } as:
Zα0 = 0 (19)
r
i

r r r
yk Cov{α * } = G − g Cov{d obs }G − gT (28)
Zαk = (20)
i
∂α i
r k +1 Eqns. 27 and 28 can be modified to take into account data
∂f
J = r k +1 (21) weights. The weight matrix W is usually:
∂y
r r −1
∂f k +1 W = Cov{d obs } 2 (29)
D = rk (22)
∂y
rk
[ ]
T Applications
y = p1k , S wk _1 , p2k , S wk _ 2 ,K , pnbloc
k
, S wk _ nblock (23) Some practical examples of the application of the procedure
are shown next.
Variance and Resolution of the Parameter Estimates
Since the reservoir parameter estimation problem we are Channel Model
dealing with is nonlinear (Eqn. 1), it is not possible to have a The data (field observations) are shown in Fig. 2. The
simple way to calculate the covariance matrix of the reservoir was discretized with a 40 × 30 Cartesian mesh. The a
parameters estimates. The importance of the covariance matrix priori large scale information consisted of the knowledge of a
is in that it provides information about the uncertainty channel, thus we used the dynamic object approach. The
associated with each parameter and also about the correlation object channel can be parameterized with eight parameters as
between the parameters. One way to overcome the difficulty shown in Fig. 3.
associated with the nonlinearity is by making a linear In the object model approach the parameters are no longer
approximation to the problem. This type of analysis will the permeability and porosity at each cell of the simulation
provide very valuable qualitative information not only about grid (pixel model) but the parameters that define the object.
the level of uncertainty in the parameter estimates but also in Thus in this approach the object is allowed to “float” in the
the information content in each type of data. The linear reservoir, that is the channel can change shape, translate, and
analysis is performed by assuming that the nonlinear system rotate. Also, the permeability inside and outside the channel
r
can be modeled in the neighborhood of an optimal point α * are considered as parameters. Fig. 4 shows a sequence of
with a first order approximation as: frames that depicts the evolution of the parameters that define
the channel, which reveals two main features of the method.
r r First, is the capability to handle objects in the space. Second,
d cal = Gα + constant (24)
is the speed of the method, in this case we were able to match
r r the data and converge to the “true” reservoir in only 41
where the sensitivity matrix G is computed at α = α * iterations. Fig. 2 shows the match of the data.
6 JORGE L. LANDA AND ROLAND N. HORNE
The same data were used for another inverse problem, this
time without using the information about the existence of a
channel. We kept the 40 × 30 simulation grid but now we
parameterized with 100 parameters, each parameter
representing the permeability of 12 adjacent cells. We refer to
this method of parameterization as the “large pixel” model
because of the resemblance to the pixel approach. The initial
guess was a homogeneous reservoirs. Fig. 5 (b) shows the
calculated permeability distribution. Fig. 5 (c) shows also the
boundaries of the channel that was used originally to generate
the data. This problem was much harder to solve because of
the larger number of parameters. It took approximately 400
iterations to obtain a good match of the data. Thus a first
impression is that the object modeling approach was 10 times
faster, but it must be remembered that each iteration of the 100
parameters requires much longer CPU time. Hence the object
model approach was actually 1000 times faster than the large
pixel approach.
Black and White 4-D Seismic Data
One observation in the previous example is that we were using
“exact” data. This can be considered unrealistic, especially in
the case of the 4-D seismic data since with the current
technology it is not possible for the geophysicists to prepare an
“exact” map of change of saturation. Most likely they will be
able to prepare a map where they can assert the areas of the
reservoir where there were changes in the saturation but the
magnitude of the changes will be unknown. Thus we used the
data shown in Fig. 1 but instead of using the exact change of
saturation maps in Fig. 1 (f) and (g) we used the coarse maps
shown in Fig. 6 (b) and (d). Fig. 7 (a) shows the true
permeability field, (b) the calculated permeability when
“exact” data is used, and (c) the calculated permeability when
the 4-D seismic data is used in a coarse “black and white”
format. We see from (c) that the “black and white” data can be
used in the approach we developed here and still provide a
reasonable description of the reservoir.
Fault Model
We can use the object model approach to find the location of
faults in a reservoir by using a diversified data set similar to
the one depicted in Fig. 1. As a first approximation, we can
model a sealing-fault as a rectangle, where the permeability
inside is very low (10-5 md) and the width is small compared to
the dimensions of the reservoir. We set free all the other
parameters that define the rectangle object. Thus the rectangle
can change its length, rotate and translate in space. Fig. 8
shows an example of finding the location of a single fault. This
figure shows the location of the fault and the water saturation
distribution in the reservoir at the end of the simulated time.
We do not use the saturation information in the match. The
first frame shows the first guess for the fault. The first guess is
substantially different than the “true case” (last frame) from
the point of view of water saturation (see the well closest to the
right bottom corner). The procedure developed in this work
SPE 38653
was able to recover the true location of the fault after 51
iterations.
Kriging Model
We can use the object model approach to solve another type of
problem by using a technique presented by Fasanino et al.37.
With this technique we can introduce geostatistical
information into our inverse problem without the need to use
the generalized least square formulation (Eqn. 5). The
technique works as follows.
If there were a relatively large number of permeability and
porosity samples in the reservoir, then it would be possible to
estimate the permeability in rest of the reservoir by using a
linear estimation technique, such as kriging, that takes into
consideration a predefined spatial correlation of the parameters
given in the form of a covariance matrix. In our case we do not
have the sample points, therefore we make the sampled
permeability values the unknown parameters in the inverse
problem. The covariance matrix used by the kriging estimator
allows us to introduce the a priori information about the
spatial correlation.
The permeability k at simulation cell j is calculated as:
npar
kj = ∑λ k
i,j i (30)
i =1
where λi,j is the kriging weight calculated from a
predefined variogram model at the location of the npar
hypothetical sample points.
Thus the problem now can be stated as to find a set of
permeability values that we should assign to the pilot sample
points that result in a kriged reservoir that match the actual
dynamic data.
Figs. 9 (a) and (b) show an example of the use of this
approach. In this case we used 48 hypothetical sample points
in the reservoir. (a) shows the true reservoir and (b) shows the
calculated reservoir when exact data are used. Since in real
reservoirs the appropriate locations or number of pilot points
and the variogram model are unknown, we can use this
procedure to generate several realizations of the reservoir by
using different pilot points and variogram models that are
considered plausible for that particular reservoir. This idea was
tested by solving the same inverse problem with three different
variogram models that differed from the true variogram in the
azimuth angle. Figs. 9 (c) and (d) show the results when the
variogram is rotated 90° and 45° respectively. Figs. 9 (b), (c),
and (d) all provided a good match of the dynamic data
generated with the true reservoir shown in (a) even though
they look substantially different from each other. The first
guess in all cases was a homogeneous reservoir. This could be
considered as the most unfavorable condition because it
presumes an unrealistic ignorance of the major features in the
reservoir.
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION
Another procedure closely related to the kriging-pilot
points procedure is the one referred to as “the self-calibrated
method” by Wen, Gomez-Hernandez, Capilla and
Sahuquillo38. This method was developed in the ground water
modeling field but can be adapted for use in the multiphase
problem of petroleum reservoir modeling. The essence of the
method is, given a first guess of the reservoir or realization
obtained with a geostatistical tool, to calculate a perturbation
to the field such that its response matches the dynamic data
from the real reservoir. The perturbation is calculated as a
kriging estimate from perturbations at certain predefined
control or pilot points. Mathematically it is:
npar
k j = k 0j + ∑ λi , j ∆ki
(31)
i =1
3-D Seismic Model
The object model approach can also be used to integrate
information from 3-D seismic interpretations. For example
Fig. 10 (a) shows a 3-D interpretation (adapted from Ref. 39) .
This image was discretized as a 16 color map and then we
assigned a permeability/porosity value to each color. Thus the
colors themselves become the “objects” in the problem, and
we use the procedure to determine the permeability/porosity
values we should assign to each color in order to match the
data. Fig. 10 shows in (a) the true reservoir and in (b) the
calculated reservoir. In this case we simulated the field
considering 10 wells. The data used in the inverse problem
consisted of field observations from one DST and permanent
gauges at each well over a 200 day period. The simulation grid
consisted of a 30 × 30 mesh.
Variance of Parameter Estimates and Value of the
Information
The previous section showed some examples of solutions to
reservoir inverse problems by the integrated use of data of
different nature. The inverse problem is not completely solved
by providing sets of permeability and porosity values that
result in a good match of the data. There are other issues that
come along with the inverse problem such as the uncertainty
associated with each parameter estimate and the value of the
information. These issues can be explored with the theory used
to develop Eqns. 24 to 29. A simple example will illustrate the
use of this analysis.
For the sake of simplicity let us assume that we have a
homogeneous (but unknown) reservoir, and a data set similar
to the one illustrated in Fig. 1. The reservoir was discretized
with 40 × 30 mesh and parameterized with 100 large pixel
r reaches the wells28. This is confirmed in the color map,
parameters. After a permeability distribution ( α * ) that
matches the data is obtained, the first question we would like
to answer what is the uncertainty associated to each of the 100
parameters we have calculated.
7
We start the analysis by first computing the matrix G of the
r r
sensitivity coefficients at α = α * , then we compute the
singular value decomposition of G (or GTG)28, next we
compute the covariance matrix of the parameter estimates
r
(Eqn. 28). The diagonal elements of Cov{ α * }, that is the
variances of the parameter estimates ( σ α2i ) provide
information about the uncertainty. We can visualize this
information by translating σ α−1i to the reservoir map. This is
shown in Fig. 11. The red colors show which section of the
reservoir are best resolved by the data, next are the areas in
green. The areas in blue or black are the areas not well defined
by the data, that is the permeability in such areas can take any
arbitrary value within a large range without affecting the
quality of the match. This is a basic analysis, next we calculate
the variance for each parameter and for each data set, both
individually and in combination. The results are summarized
in Fig. 13, which maps the inverse of the normalized variance
σ αi
in the reservoir.
αi
The following combinations of data were analyzed:
1. DST pressure only.
2. Long term pressure only.
3. Pressure only (DST pressure and long term pressure).
4. Water cut only.
5. Change of saturation only (two 4-D seismic
interpretations).
6. Pressure and water cut.
7. Pressure and change of saturation.
8. Water cut and change of saturation.
9. All (pressure and water cut and change of saturation).
The following remarks and observations can be drawn
from Fig. 13:
• Combinations of data always reduce the uncertainty. That
is, adding more uncertain information to a data set does
not downgrade the solution, on the contrary, it always
produces a reduction in the level of uncertainty.
• The addition of 4-D seismic information produced a
remarkable improvement despite this information having a
relative high uncertainty. This is because the 4-D seismic
data provides information that is spatially distributed.
• The water cut information seems to help little in this
specific problem.
• The long term pressure history seems to provide strong
information. The highest sensitivity in the long term
pressure is concentrated at the time when the water front
here the best area determined by the long term pressure is
the area behind the water front.
• The DST information alone will not help much in the
inverse problem in the way it was posed, that is the DST
data alone cannot be used to compute 100 spatially
8 JORGE L. LANDA AND ROLAND N. HORNE
distributed parameters. DST’s resolve simple models only.
On the other hand the combination of the apparently poor
information from DST with the long term pressure data
(permanent gauges) produced a remarkable improvement
in the quality of the data.
• The DST plot reveals how this data should be used in
reservoir inverse problems. If no other data are available
then the number of parameters should be small enough to
ensure a low variance. Later in the life of the reservoir,
when other data are available, such as production history
and 4-D seismic, then the DST information can be used
again but without the need to constrain the reservoir to a
small number of parameters.
• The appropriate number of parameters, or level of
parameterization (size of the model) can be adjusted with
the amount of data, that is when data are scarce the model
should be kept small. When the data are abundant then
certain areas of the reservoir can be parameterized with a
more complex model.
From these analyses it was observed that the pressure data
play an important role in the reservoir parameter problem.
Therefore a more detailed analysis was performed in the case
of the pressure data. The analysis was performed by breaking
down the pressure into major components, which are: DST,
shut-in, and data from permanent gauges (referred to here as
long term pressure only). The results are shown in Fig. 14.
The focus here is on the shut-in pressure information. The
apparently poor shut-in pressure information enhances the
quality of the data set. The implication is important. If
permanent gauges are installed, then any sporadic or
unpredicted shut-off of a well will produce a short transient,
and indirectly will enhance the quality of the rest of the data.
Analysis of the Results from a Reservoir Inverse Problem
The previous sections presented a systematic procedure to
analyze the issue of how each data type contributes to the
reduction of the uncertainty attached to each parameters. To
perform such analysis we assumed that we knew the true
values of the parameters. Here we will use the variance
approach to analyze the results of an inverse problem, that is
after the data have been matched to a calculated set of
parameters. Then a new question arises, namely how good are
such calculated parameters. This issue was investigated by
using the variance analysis.
This case used the same heterogeneous reservoir and data
shown in Fig. 1. The first guess for the inverse problem was a
homogeneous distribution of permeability. The result of the
match and the variance calculation are shown along with the
true case in Fig. 12. Both true and calculated sets of
parameters provide a good match to the data, the “true case”
provides the exact match because the data was generated with
this distribution of permeability. The true and the calculated
permeability fields seen to be different, but when both are
SPE 38653
k
evaluated from the point of view, they are seem to have
σ
essentially the same uncertainty. Both maps reveal that the
permeability at the middle of the South and West boundaries
will be ill-determined, consequently any value there cannot
taken seriously, since any arbitrary but reasonable value
assigned to those regions would provide an acceptable match
of the data. That is the same as saying that the available data
are not sufficient to resolve the parameters in these regions of
the reservoir.
Both variance maps are similar from the qualitative point
of view, this implies that it is not necessary to know the true
reservoir values to perform the variance analysis, a reservoir
description that provides a good match of the field
observations is sufficient for the analysis.
Application of Variance Analysis in Reservoir Monitoring
The variance analysis shown in the previous sections can be
used to organize and optimize the data collection policy in a
reservoir. In general it is possible to obtain a first
approximation to the reservoir model before production and/or
dynamic data are collected in the field. Then it is possible to
use this approximated reservoir to perform the variance
analysis28 and to prepare plots similar to those shown Figs. 13
and 14 before any data is actually collected in the field. Then
the best strategy for data collection can be devised considering
that each piece of information will be used later in conjunction
with the others. The analysis can also be used to find the best
level of parameterization at a given time considering the data
available. The level of parameterization will change as more
field observations are made available. Both processes,
reservoir parameter estimation and future data collection
policy, have to be updated continually since every time new
field data arrives the reservoir model is updated via the inverse
problem solution, then the variance analysis is updated and the
cycle starts over again.
Since the variance analysis uses a covariance model for the
data (field observations) which is generally assumed to be
diagonal and which is related to the accuracy or reliability of
the field measurements, then this method can be used to
investigate how the accuracy of the instruments used to collect
the information has an effect on the calculated permeability
and porosity.
Conclusions
We showed procedures; (1) to estimate the distribution of
permeability and porosity in the reservoir by using dynamic
data; (2) to assess the uncertainty associated with the
calculated permeability and porosity and; (3) to determine the
value of each data type from the point of view of the inverse
problem.
The procedures allow us to integrate information from
several sources to compute distributions of permeability and
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION
porosity. The information that can be integrated includes: well
test data, shut-in pressure, long term pressure (from permanent
gauges), production history (water cut, rates), interpreted 4-D
seismic maps, permeability-porosity correlations, variogram
models and large scale geological information (object
modeling). This seems to cover most of the information that
can be realistically obtained from a program for reservoir
monitoring with current technology, but it is not necessary to
have all of them to apply the method described in this paper.
In this work the interpreted 4-D seismic information was
assumed to be in the form of maps of change of saturation in
the reservoir. The procedure can be adapted to the case when
the 4-D seismic interpretation is not related to a simple change
of saturation but to a more complex relationship that could be
a function of not only saturation but pressure, density, rock
type, etc.
The method developed in this work relies on the
computation of the sensitivity coefficients for the field
observations with respect to the parameters of the inverse
problem. This computation might be considered an
unnecessary burden since there are other methods to solve the
inverse problem without the sensitivity coefficients, but we
found such an effort worthwhile for the following reasons:
• The sensitivity coefficients are necessary to perform the
variance analysis, and this is key to understanding the
meaning of what it is being calculated and to plan
reservoir monitoring. The sensitivity coefficients are also
needed to find the most adequate parameterization of the
reservoir for a given amount of information.
• The method to compute the sensitivity coefficients, as
developed in this work, is relatively simple to implement
when it is possible to have access to the numerical
reservoir simulator code. Thus the complexity of
modeling the reservoir is left to the simulation part and
not to the inverse problem as it would be the case for the
optimal control approach.
• The Gauss-Newton approach is very efficient as a
parameter estimation algorithm, compared to direct search
methods that do not require sensitivity coefficients.
• The computation of the sensitivity can be performed with
high efficiency, specially in the case of object or large
pixel modeling. The efficiency can also be enhanced by
parallel computing.
From the results of our numerical experiments it is possible
to conclude that the method can be utilized as a tool for
reservoir characterization. The method seems to be more
useful when the dynamic data set consists of a large number of
measurements that are difficult to honor with the existing
geostatistical approaches; this may be the case in mature fields
that have already gone through a secondary recovery process
and in which data have been gathered over many years.
9
The more interesting conclusions obtained from this work
are about the roles of well testing, 4-D seismic data and
integration.
Well testing has often been considered as an isolated tool
to solve relatively simple reservoir models in the
neighborhood of the wells and to improve the productivity.
These are very important considerations from the engineering
and economical point of view, but since the well tests are
performed generally during the earlier part of the life of the
reservoir then this information is not used later mainly because
of the assumption that the area investigated is small and also
because of the lack of a mathematical tool to be used in a more
complex environment. The analysis performed here shows that
the well test information can be used to resolve much more
complex reservoirs models later in the life of the reservoir, not
as a standalone approach but as a piece of information in a
larger data set. The addition of well test information enhances
the resolution value of the other information, thus well tests
early in the life of the reservoir will also be useful later for
reservoir characterization.
The value of 4-D seismic data is due to the fact that it
provides information that is spatially distributed in the
reservoir in contrast to the other data that are localized at the
wells. Even so, it was found that the 4-D information does not
assist much in reservoir characterization when it is considered
alone. The value of this type of information appears when it is
combined with other more traditional data such as production
history. Also it was found that this kind of information does
not need to be very accurate since data in coarse “black and
white” format can still be useful. In all the examples shown in
this work it was assumed that two consecutive 3-D seismic
surveys were performed after production had begun in the
reservoir. If the first survey were to be performed before any
production then it is very likely that the 4-D information will
provide even more value to the reservoir inverse problem.
The value of data integration can be evaluated with the
variance analysis. It was shown that the addition of data
always enhances the overall resolution of the parameters. The
addition of data that seems to provide little information value
when considered in isolation results in a large increase in the
resolution of the other data. A dramatic example is in the case
of the shut-in pressure data, which includes only a very small
number of pressure measurements obtained when the wells are
closed at the surface for a short period. This information when
considered by itself cannot be used to solve even the simplest
models, but its value is considerable when it is used in an
integrated procedure such as the one developed in this work.
Acknowledgments
The authors gratefully acknowledge the financial support from
the Stanford University Research Consortium for Innovation in
Well Test Analysis (SUPRI-D).
10 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653

References
Nomenclature 1. Seymour, R.H and Barr, F.J.: “Seabed-Seismic 4D-Collection
r
α = parameter model vector Method for Reservoir Monitoring,” JPT (January 1997) Vol. 49,
r No. 1, 40-41.
α * = optimal parameter model vector
r 2. Jacquard, P., and Jain, C.: “Permeability Distribution from Field
α prior = a priori parameter model vector Pressure Data,” Soc. Pet. Eng. J. (Dec. 1965) 281; Trans.,
k= permeability AIME, Vol. 234.
k0 = permeability from realization 3. Carter, R.D., Kemp, L.F., Pierce, A.C., and Williams, D.L:
“Performance Matching With Constraints,” Soc. Pet. Eng. J.
λ= kriging weight
(Apr. 1974) 187; Trans., AIME, Vol. 257.
φ= porosity 4. Dogru, A.H., and Seinfeld, J.H.: “Comparison of Sensitivity
σ2 = variance Coefficient Calculation Methods in Automatic History
∇E = gradient of E Matching,” Soc. Pet. Eng. J. (Oct. 1981) 551.
Cd= data covariance matrix 5. Carter, R.D., Kemp, L.F., and Pierce, A.C.: “Discussion of
Cα = a priori parameter covariance matrix Comparison of Sensitivity Coefficient Calculation Methods in
Cov{ } = covariance matrix operator Automatic History Matching,” Soc. Pet. Eng. J. (Apr. 1982)
205.
DST = drill stem test
r 6. Chen, W.H., Gavalas, G.R., Seinfeld, J.H., and Wasserman,
d = data vector M.L: “A New Algorithm for Automatic History Matching,” Soc.
E= objective function Pet. Eng. J. (Dec. 1974) 593; Trans., AIME, Vol. 257.
r
f = material balance equations vector 7. Chavent, G., Dupuy, M., and Lemonnier, P.: “History Matching
by Use of Optimal Theory,” paper SPE 4627 presented at the
G= matrix of sensitivity coefficients 1973 SPE Annual Technical Conference and Exhibition, Las
H= Gauss-Newton approximation to the Hessian matrix Vegas, NV, Sep. 30 - Oct. 3.
of objective function 8. Watson, A.T., Seinfeld, J.H., Gavalas, G.R., and Woo, P.T.:
J= Jacobian of material balance equations “History Matching in Two-Phase Petroleum Reservoirs,” paper
R= resolution matrix SPE 8250 presented at the 1979 SPE Annual Technical
Sw = water saturation Conference and Exhibition, Las Vegas, NV, Sep. 23-26.
9. Yang, P.H, and Watson, A.T.: “Automatic History Matching
S= singular value matrix
With Variable-Metric Methods,” paper SPE 16977 presented at
t= time the 1987 SPE Annual Technical Conference and Exhibition,
U= factor from singular value decomposition Dallas, TX, Sep. 27-30.
V= factor from singular value decomposition 10. Kamal, M.M.: “Use of Pressure Transients to Describe
W = weighting matrix Reservoir Heterogeneity,” JPT, (Aug. 1979) 1061, Trans.,
r
y = vector of pressure and saturation in the simulation AIME, Vol. 257, SPE Reprint Series No. 14.
11. Tang, Y.N., and Chen, Y.M.: “Application of GPST Algorithm
mesh
r to History Matching of Single-Phase Simulator Models,” paper
Z = sensitivity vector SPE 13410, 1985.
12. Tang, Y.N., and Chen, Y.M.: “Generalized Pulse-Spectrum
Technique for 2-D and 2-Phase History Matching,” Applied
Subscripts Numerical Mathematics (1989) Vol. 5, 529-539.
obs = observed value 13. Oliver, D.S.: “Incorporation of Transient Pressure Data Into
calc = calculated value Reservoir Characterization,” In Situ (1994) Vol. 18, 243-275.
npar = number of parameters 14. Chu, L., Reynolds, A.C., and Oliver, D.S.: “ Computation of
Sensitivity Coefficients for Conditioning The Permeability Field
nblocks = number of blocks in simulation grid
to Well-Test Pressure Data,” In Situ (1995) Vol. 19, 179-223.
p = number of nonzero singular values 15. Oliver, D.S.: “Multiple Realizations of the Permeability Field
T = transpose of matrix From Well Test Data,” paper SPE 27970 presented at the 1994
University of Tulsa Centennial Petroleum Engineering
Symposium, Tulsa, OK, Aug. 29-31.
Superscripts 16. Chu, L., Reynolds, A.C., and Oliver, D.S.: “Reservoir
r Description From Static and Well-Test Data Using Efficient
= vector
-g = generalized inverse of a matrix Gradient Methods,” paper SPE 29999 presented at the 1995
SPE International Meeting on Petroleum Engineering, Beijing,
k = iteration index
PR China, Nov. 1995.
p = pressure 17 Reynolds, A.C., He, N., Chu, L., and Oliver, D.S.:
wc = water cut “Reparameterization Techniques for Generating Reservoir
∆Sw= change of water saturation Descriptions Conditioned to Variograms and Well-Test Pressure
Data,” paper SPE 30588 presented at the 1995 SPE Annual
Technical Conference and Exhibition, Dallas, TX, Oct. 22-25.
18. Oliver, D.S.: “A Comparison of the Value of Interference and
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 11

Well-Test Data for Mapping Permeability and Porosity,” In Situ Theory, Academic Press, Inc., San Diego, CA, 1989.
(1996) Vol. 20, 41-59. 34. Gill, P.E., Murray, W., and Wright, M.H.: Practical
19. Oliver, D.S., He, N. and, Reynolds, A.C.: Conditioning Optimization, Academic Press, San Diego, CA, 1981.
Permeability Fields to Pressure Data,” paper presented at the 5th 35. Bard, Y.: Nonlinear Parameter Estimation, Academic Press,
European Conference on the Mathematics of Oil Recovery, New York, NY, 1970.
Leoben, Austria, (1996), Sep. 3-6. 36. Jackson, D.: “Interpretation of Inaccurate, Insufficient and
20. He, N., Reynolds, A.C., and Oliver, D.S.: “Three-Dimensional Inconsistent Data,” Geophysical Journal of the Royal
Reservoir Description from Multiwell Pressure Data,” paper Astronomical Society,(1972), Vol. 28, 97-109.
SPE 36509 presented at the 1996 SPE Annual Technical 37. Fasanino, G., Molinard, J., and de Marsily, G.: “Inverse
Conference and Exhibition, Denver, CO, Oct. 6-9. Modeling in Gas Reservoirs,” paper SPE 15592 presented at the
21. Anterion, F., Eymard, R., and Karcher, B.: “Use of Parameter 1986 SPE Annual Technical Conference and Exhibition, New
Gradients for Reservoir History Matching,” paper SPE 18433 Orleans, LA, Oct. 5-8.
presented at the 1989 SPE Symposium on Reservoir Simulation, 38. Wen, X., Gomez, J., Capilla, J. and Sahuquillo, A:
Houston, TX, Feb. 6-8. “Significance of Conditioning to Piezometric Head Data for
22. Bissell, R., Sharma, Y., and Killough, J.E.: “History Matching Predictions of Mass Transport in Groundwater Modeling,”
Using the Method of Gradients: Two Case Studies,” paper SPE Mathematical Geology, (1996), Vol. 28, No. 7, 951-968.
28590 presented at the 1994 SPE Annual Technical Conference 39. Sheriff, R.E, and Geldart, L.P.: Exploration Seismology,
and Exhibition, New Orleans, LA, Sep. 25-28. Cambridge University Press, New York, NY, 1995.
23. Tan, T.B., and Kalogerakis, N.: “A Fully Implicit, Three-
Dimensional, Three-Phase Simulator with Automatic History-
Matching Capability,” paper SPE 21205 presented at the 1991
SPE Symposium on Reservoir Simulation, Anaheim, CA, Feb.
17-20.
24. Tan, T.B.: “A Computational Efficient Gauss-Newton Method
for Automatic History Matching,” paper SPE 29100 presented
at the 1995 SPE Symposium on Reservoir Simulation, San
Antonio, TX, Feb. 12-15.
25. Killough, J.E., Sharma, Y., Dupuy, A., and Bissell, R.: “A
Multiple Right Hand Side Solver for History Matching,” paper
SPE 29119 presented at the 1995 SPE Symposium on Reservoir
Simulation, San Antonio, TX, Feb. 12-15.
26. Bissell, R.: “History Matching A Reservoir Model by the
Positioning of Geological Objects,” paper presented at the 5th
European Conference on the Mathematics of Oil Recovery,
Leoben, Austria, (1996), Sep. 3-6.
27. Landa, J.L., Kamal, M.M., Jenkins, C.D., and Horne, R.N.:
“Reservoir Characterization Constrained to Well Test Data: A
Field Example,” paper SPE 36511 presented at the 1996 SPE
Annual Technical Conference and Exhibition, Dallas, TX, Oct.
6-9.
28. Landa, J.L.: “Reservoir Parameter Estimation Constrained to
Pressure Transients, Performance History and Distributed
Saturation Data,” Ph.D. dissertation, Stanford University,
Stanford, CA, 1997.
29. Sultan, A.J., Ounes, A. and, Weiss, W.W.: “ Automatic History
Matching for an Integrated Reservoir Description and
Improving Oil Recovery,” paper SPE 27712 presented at the
1994 SPE Permian Basin Oil and Gas recovery Conference,
Midland, TX, Mar. 16-18
30. Ounes, A, Bhagavan, S., Bunge, P.H., and Travis, B.J.:
“Application of Simulated Annealing and Other Global
Optimization Methods to Reservoir description: Myths and
Realities,” paper SPE 28415 presented at the 1994 SPE Annual
Technical Conference and Exhibition, New Orleans, Sep. 25-28.
31. Tarantola, A., and Valette, B.: “Generalized Nonlinear Inverse
Problems Solved Using The Least Squares Criterion,” Reviews
of Geophysics and Space Physics (May 1982) Vol. 20, No. 2,
219-232
32. Tarantola, A.: Inverse Problem Theory - Methods for Data
Fitting and Model Parameter Estimation, Elsevier Science
Publishers, Amsterdam 1987.
33. Menke, W.: Geophysical Data Analysis: Discrete Inverse
12 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653

8000

c c 6000

Permeability
? 4000

log(k)
c c 2000
φ - Porosity
0
(a) (h)

c c Pressure
Water Cut Production History
DST Well # 1
Well # 1
c c
c c
(b) (i) (m)

c c c c
(f) Production History
Well # 2
c c DST
c c Well # 2

(c) (j) (n)

c c c c
DST Production History
(g) Well # 3 Well # 3

c c
(d) (k) (o)

c c 0.6
DST
Saturation

Well # 4
0.4
Injector Injection History
c c 0.2
Well # 4 - Injector

(e) (l) (p)

Fig. 1: Problem statement.

(a): Reservoir geometry. (b)-(e): Water saturation distribution as a function of time. (f): Change of water saturation between (b)
and (c). (g): Change of water saturation between (c) and (d). (h): Permeability-Porosity relationship. (i)-(l): Pressure vs. time for
DST’s. (m)-(p): Long term pressure and watercut vs. time.
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 13

Iter #1 E= 197,097 Iter #2 E= 113,596 Iter #3 E= 41,512

a a a a a a
True Reservoir Calculated Reservoir 15000

a a a a a a a a a a

Permeability
10000

Permeability
Iter #7 E= 15,617 Iter #11 E= 14,506 Iter #16 E= 14,340
5000

a a Field
a a a a a a a a

a a a a
a a a a a a
Change of

Match a a Saturation #1
a a
0.30
Iter #19 E= 14,360 Iter #22 E= 13,680 Iter #25 E= 13,510
0.25
of the Data

∆Sw
a a a a a a
a a a a
0.20

0.15
Change of a a a a a a
a a Saturation #2
a a Iter #28 E= 13,252 Iter #31 E= 12,756 Iter #34 E= 10,530

a aa aaaaa aaaaaa aa
a
a aa a a a a a a
a
True Pressure
a DST
True Water Cut Pressure & Water-Cut History
aaaa Well # 1 a
a Calculated Values Well # 1 a
aa
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a aaaaaa a a a a a a
a aaa
a a a aaa
aaaaa a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a
aaaa
a Iter #36 E= 2,778 Iter #38 E= 71 Iter #41 E= 0.01
aa
15000

a
a
a a a a aa a
aaaa
aaa a a a a a a
a a aa a a a a
a a a a
10000
aaaaaaaaaa
aa
Pressure & Water-CutaHistory
aa aa a a aa a
aaaaa a

md
Well # a2 a a
a a a a a
DST a a a 5000
a a a a a a
Well # 2 a
a a
a a a a a a a a a a a a a a a a
aaaaa a a a a a a a a a a a a a a a a a a a a a
0
aaaa
aa
a aa aaaa aaaaaaa
a a
DST aaaaaaa
Pressure & Water-Cut History a a a a a a a a Fig. 4 Channel model. Evolution of parameters.
a
a Well # 3 a a a
a a aa
Well # 3 a a
aaa a a a a a a a a a a a a a a a a a a a a a a
a a a a a aaa
a a a a a a a a
a a a a
aaaaa a a a a a a a a a a a a a a a a a a a a a a a a a a True Permeability
aa
a
a DST
aa a a a a
a a a a a a a
a a
Well # 4 a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
a Injector Pressure History
a aaaa aaaa
a Well # 4 - Injector
a a aa a a aaa aaa
aa a

Fig. 2 Channel model - Data and match of the data.

a a

(a)
(c)

Calculated Permeability
15000

a a
Curve Analysis - Parametric Channel Model 10000

md
α5
5000

α2 = kmatrix a a
α3
0
(b)
-1
π α4
Calculated Permeability
α1 =
kch
anne
l
a a

α8
α7

a a

α6 (c)

Fig. 5 Channel reservoir. (a): True reservoir. (b):

Fig. 3 Parameterization of a channel. Calculated reservoir using large pixel model. (c):
Comparison of pixel and true reservoir.
14 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653

Iter #1 Iter #2 Iter #3

c c c c c c

True ∆Sw #1 Filtered ∆Sw #1 0.4

0.3 c c c c c c

∆Sw
0.2
Iter #7 Iter #11 Iter #15

0.1 c c c c c c

0.0
(a) (b)
c c c c c c

True ∆Sw #2 Filtered ∆Sw #2 0.4

Iter #20 Iter #25 Iter #30

0.3
c c c c c c

∆Sw
0.2

0.1 c c c c c c

0.0
(c) (d) Iter #35 Iter #42 Iter #45

c c c c c c
Fig. 6 4-D seismic data in “black and white” format.

c c c c c c

Linear Search Iter #48 Iter #51

True Permeability
0.6
c c c c c c

a a

∆Sw
0.4

c c c c c c 0.2

Fig. 8 Fault model. Evolution of parameters.

a a

(a)
True Permeability - α=30° Calculated Permeability - α=30°
Calculated Permeability - Exact ∆Sw 6000

a a a a 103

a a 5000

md
4000

10
md

3000

2000 a a a a

a a 1000
(a) (b)
0
(b)
Calculated Permeability - α=120° Calculated Permeability - α=75°

Calculated Permeability - Filtered ∆Sw

a a a a

a a

a a a a

a a (c) (d)

(c) Fig. 9 Kriging model. (a): True reservoir. (b): Calculated

Fig. 7 Comparison of results. (a): True reservoir. (b):
Calculated reservoir using exact data. (c): Calculated
reservoir using 4-D seismic data in “black and white”
format.
reservoir using exact data. (c): Calculated reservoir -
variogram rotated 90°. (d): Calculated reservoir -
variogram rotated 45°.
SPE 38653 INTEGRATING WELL TEST DATA, PERFORMANCE HISTORY AND 4-D SEISMIC INFORMATION 15

True k Calculated k
True Permeability
104
10000
c c c c
a
c 103

k md
5000
c
a c

k - md
102 c c c c
c c c (a) (c)

c 10 k/σk - True k k/σk - Calculated k

c 10

c c c c
1

k/σk md/md
(a)
1

Calculated Permeability
c c c c
a 10-1
c (b) (d)

c Fig. 12 Variance maps: (a): True reservoir . (b): k/σ map for true
a c
reservoir. (c): Calculated reservoir. (d): k/σ map for calculated reservoir
c c c

c
c

(b)

Fig. 10 3-D seismic model. (a): True reservoir. (b):

Calculated reservoir.

Map of σ-1
1

c c
10-1
1/σ md-1

10-2

c c

c well location

Fig. 11 Map of σ . Homogeneous reservoir.

-1
16 JORGE L. LANDA AND ROLAND N. HORNE SPE 38653

Data: DST Pressure Data: Long Term Pressure

c c c c

c c c c

k/σ scale
Data: Pressure
10
c c

k/σk md/md
1

c c
10-1

Data: Pressure + Water Cut Data: Pressure + Water Cut + ∆Sw Data: Pressure + ∆Sw

c c c c c c

c c c c c c

Data: Water Cut Data: Water Cut + ∆Sw Data: ∆Sw

c c c c c c

c c c c c c

Fig. 13 Data analysis.

k/σ scale
Data: DST Pressure
10

c c

k/σk md/md
1

10-1

10-2
c c
10-3

Data: DST + Shut-in Pressure Data: DST + Shut-in + Long Term Pressure Data: DST + Long Term Pressure

c c c c c c

c c c c c c

Data: Shut-in Pressure Data: Shut-in + Long Term Pressure Data: Long Term Pressure

c c c c c c

c c c c c c

Fig. 14 Analysis of the pressure information.