A Comparison of Travel-Time and

Amplitude Matching for Field-Scale

Production-Data Integration: Sensitivity,
Nonlinearity, and Practical Implications
Hao Cheng, SPE, Akhil Datta-Gupta, SPE, and Zhong He, SPE, Texas A&M U.

Summary rooted in the least-squares inversion theory that attempts to mini-

The traditional approach to reconciling geologic models to pro- mize the difference between the observed production data and the
duction data involves an “amplitude matching,” that is, matching model predictions. This can be referred to as “amplitude” match-
the production history directly. These include water-cut, tracer ing. The production data can be water-cut observations, tracer
concentration, and pressure history at the wells. It is well known response, or pressure history at the wells. It is well known that
that such amplitude matching results in a highly nonlinear inverse such inverse problems are typically ill-posed and can result in
problem and difficulties in convergence, often leading to an inad- nonunique and unstable solutions. Proper incorporation of static
equate history match. The nonlinearity can also aggravate the data in the form of a prior model can partially alleviate the prob-
problem of nonuniqueness and instability of the solution. Recently, lem. However, there are additional outstanding challenges that
production data integration by “travel-time matching” has shown have deterred the routine integration of production data into res-
great promise for practical field applications. In this approach, the ervoir models. The relationship between the production response
observed data and model predictions are lined up at some reference and reservoir properties can be highly nonlinear. The nonlinearity
time such as the breakthrough or “first arrival” time. Further ex- can result in multiple local minima in the misfit function. This can
tensions have included amplitude information by a “generalized cause the solution to converge to a local minimum, leading to an
travel-time” inversion. Although the benefits of travel-time inver- inadequate history match. All these can make it difficult to obtain
sion are well documented in the context of seismic inversion, no a meaningful estimate of the parameter field, particularly if the
systematic study has been done to examine its merits for field- initial model is far from the solution.
scale history matching. Recently, streamline-based methods have shown significant po-
In this paper, we quantitatively investigate the nonlinearities in tential for incorporating dynamic data into high-resolution reser-
the inverse problems related to travel time, generalized travel time, voir models.1–14 A unique feature of the streamline-based produc-
and amplitude matching during production data integration and tion data integration has been the concept of a “travel-time match”
their impact on the solution and its convergence. In our previous that is analogous to seismic tomography. Instead of matching the
works, we speculated on the quasilinear nature of the travel-time production data directly, the observed data and model predictions
inversion without quantifying it. Our results here show, for the first are first “lined up” at the breakthrough time. This is typically
time, that the commonly used amplitude inversion can be orders of followed by a conventional amplitude match, whereby the differ-
magnitude more nonlinear compared to the travel-time inversion. ence between the observed and calculated production response is
We also examine the resulting implications in field-scale history minimized. A major part of the production data misfit reduction
matching. The travel-time inversion is shown to be more robust occurs during the travel-time inversion, and most of the large-scale
and exhibits superior convergence characteristics. The travel-time features of heterogeneity are resolved at this stage.2,4,5
sensitivities are more uniform between the wells compared to the The concept of travel-time inversion is not limited to streamline
amplitude sensitivities that tend to be localized near the wells. This models. Recently, it has been extended for application to finite-
prevents overcorrection near the wells. difference models through a “generalized travel-time” inversion.9
We have demonstrated our results using a field application The generalized travel-time inversion ensures matching of the en-
involving a multiwell, multitracer interwell tracer injection study tire production response rather than just the breakthrough times
in the McCleskey sandstone of the Ranger field, Texas. Starting and at the same time retains most of the desirable properties of the
with a prior geologic model, the traditional amplitude matching travel-time inversion. The concept follows from wave-equation
could not reproduce the field tracer response, which was charac- travel-time tomography and is very general, robust, and computa-
terized by multiple peaks. Both travel time and generalized travel tionally efficient.12,15 The generalized travel-time inversion has
time exhibited better convergence properties and could match the been utilized to extend the streamline-based production data inte-
tracer response at the wells with realistic changes to the geolog- gration methods to changing field conditions involving rate
ic model. changes and infill drilling.
The advantages of the travel-time inversion compared to am-
plitude inversion mainly stems from its quasilinear properties. The
Introduction advantages of travel-time inversion are well documented in the
Geologic models derived from static data alone often fail to re- context of seismic inversion.15 However, no systematic study has
produce the production history of a reservoir. Reconciling geologic been done to examine the benefits of travel-time inversion for
models to the dynamic response of the reservoir is critical to build- production-data integration in terms of nonlinearity and conver-
ing reliable reservoir models. In recent years, several techniques gence properties. Characterizing the degree of nonlinearity can be
have been developed for integrating production data into reservoir as important as finding the solutions to the inverse problem itself.
models.1–14 The theoretical basis of these techniques is generally However, quantitative measures of nonlinearity for the inverse
problems related to production data integration have not been ad-
equately addressed.
In this paper, we discuss the mathematical foundation for the
Copyright © 2005 Society of Petroleum Engineers
measure of nonlinearity and its implications on the production-data
This paper (SPE 84570) was first presented at the 2003 SPE Annual Technical Conference integration. We quantitatively investigate the extent of nonlinearity
and Exhibition, Denver, 5–8 October, and revised for publication. Original manuscript re-
ceived for review 28 January 2004. Revised manuscript received 22 December 2004. Paper
in travel-time inversion and amplitude inversion. In particular, we
peer approved 12 January 2005. show that the nonlinearity in travel-time inversion is orders of

March 2005 SPE Journal 75

magnitude smaller than that of the amplitude inversion. This leads the observed data until the cross-correlation between the two is
to better convergence properties and a robust method for produc- maximized. The approach is illustrated in Figs. 1c and 1d. It pre-
tion-data integration. We illustrate our results using both synthetic serves the robustness of a travel-time inversion and improves com-
and field applications. The field application is from the McCleskey putational efficiency by representing the production data misfit at
sandstone, the Ranger field, Texas, and involves a multiwell, a well in terms of a single travel-time shift. It can be shown to
multitracer interwell tracer injection study. The results clearly reduce to the more traditional least-squared misfit functional as we
demonstrate the benefits of travel-time inversion for field-scale approach the solution.12
production-data integration. In particular, the generalized travel- The advantages of travel-time inversion are well documented in
time inversion appears to outperform both travel-time and ampli- the geophysics literature. For example, Luo and Schuster15 pointed
tude inversion in reconciling the geologic model to the field- out that travel-time inversion is quasilinear as opposed to ampli-
tracer response. tude inversion, which can be highly nonlinear. Amplitude inver-
sion typically works well when the prior model is close to the
Background and Approach solution. This was the rationale behind our previously proposed
Travel-Time Inversion, Amplitude Inversion, and Generalized two-step approach to production-data integration: travel-time
Travel-Time Inversion. Travel-time inversion attempts to match match followed by amplitude match.2,4 In this paper, we will quan-
the observed data and model predictions at some reference time— titatively investigate the relative merits of the different methods in
for example, the breakthrough time or the peak arrival time. Thus, terms of nonlinearity and convergence properties.
we are lining up the production response along the time axis.
Fig. 1a illustrates the travel-time inversion. On the other hand, the Measures of Nonlinearity. Characterizing and assessing the non-
amplitude inversion attempts to match the production response linearity in the parameter estimation problem is critical to design-
directly. This is illustrated in Fig. 1b, wherein we match the ob- ing efficient and robust approaches to production-data integration.
served tracer concentration and model predictions at the producing There are several methods for quantifying the degree of nonlin-
well. Creatively, we can combine the travel-time inversion and earity in inverse problems. In this paper, we will use the measure
amplitude inversion into one step while retaining most of the de- proposed by Bates and Watts16 to examine the nonlinearities in
sirable features of a travel-time inversion. This is the “generalized travel-time and amplitude inversion. Grimstad and Mannseth17,18
travel-time inversion” and follows from the work of Luo and applied this measure to examine the relationship between nonlin-
Schuster15 in the context of wave-equation travel-time tomography. earity, scale, and sensitivity in parameter-estimation problems. If F
A generalized travel-time or travel-time shift is computed by represents an outcome (for example, the tracer response), then the
systematically shifting the computed production response toward nonlinearity measure is defined as ␬=㛳Fkk㛳/㛳Fk㛳2, where Fk is the

Fig. 1—Illustration of (a) travel-time inversion, (b) amplitude inversion, (c) generalized travel-time inversion, and (d) best time shift.

76 March 2005 SPE Journal

 Fig. 2—Geometric meaning of the measure of nonlinearity. Fig. 3—Synthetic permeability distribution for the nine-spot
case.

vector of the first-order derivatives with respect to the parameter earities inherent in these approaches. Fig. 4b shows the tracer
vector k, that is, the sensitivity vector, and Fkk is the vector of concentration matches after travel-time inversion. All the peak
second-order derivatives. This measure is based on the geometric times are now in agreement, although there are some discrepancies
concept of curvature, and ␬ represents the inverse of a radius of the in the details of the tracer responses. Fig. 4c shows the tracer
circle that best approximates the outcome locus F in the direction concentration matches after generalized travel-time inversion. Not
of Fk at k. A smoother and more linear outcome will have smaller only the peak arrival-times but also the amplitudes are matched
curvature (larger radius) and thus a smaller measure of nonlinear- much better compared to the travel-time inversion. Fig. 4d shows
ity, as illustrated in Fig. 2. the tracer-responses match after the amplitude inversion. Although
In our application, we evaluate ␬=㛳Fkk㛳/㛳Fk㛳2 for every iteration the matches are quite good for most wells, they are unsatisfactory
during inversion. In addition, for amplitude inversion, we compute for Wells 2 and 7. Incidentally, these are the two wells that ex-
the measure for different observations and choose the maximum. hibited maximum discrepancy based on the initial model.
The details of the computations, including the derivative calcula- Fig. 5 shows the convergence behavior for the three methods.
tions for travel-time, amplitude, and generalized travel-time will Both travel-time and generalized travel-time inversion reproduce
be discussed later. In the following section, we first illustrate the the arrival times perfectly. The generalized travel-time further re-
approach using a synthetic example. duces the tracer concentration misfit. In contrast, direct amplitude
match shows high arrival-time misfit and is unable to reproduce
Nonlinearity Measure in Production-Data Integration: A the tracer response at two wells. Fig. 6a is the estimated perme-
Simple Illustration. This example involves integration of tracer ability field after travel-time match. When comparing it to Fig. 3,
response in a heterogeneous nine-spot pattern, as shown in Fig. 3. we can identify the low-permeability areas and some of the mod-
The mesh size is 21×21. The reference permeability distribution erate-to-high-permeability areas, although the high-permeability
consists of a low-permeability trend toward the north and a high- area is not well reproduced. Fig. 6b shows the permeability field
permeability trend toward the south. The tracer responses from the derived by generalized travel-time inversion. It reproduces not
eight producers in the nine-spot pattern are shown in Fig. 4a. Also only the low-permeability area but also the high-permeability re-
superimposed in Fig. 4a are the tracer responses corresponding to gions. Fig. 6c shows the estimated permeability field after the
our initial model, a homogeneous permeability field that is condi- amplitude inversion. Clearly, the results show signs of instability
tioned at the well locations. because of the high nonlinearity as discussed in the next section.
We compare the relative performance of travel-time, ampli- Fig. 7 shows the measure of nonlinearity for the three ap-
tude, and generalized travel-time inversion and also the nonlin- proaches. We can see that both the travel time and the generalized

Fig. 4—Tracer response (a) for uniform initial permeability, (b) after peak arrival-time inversion, (c) after generalized travel-time
inversion, and (d) after direct amplitude inversion.

March 2005 SPE Journal 77

 Fig. 5—Travel-time and tracer concentration misfit for (a) travel-time, (b) generalized travel-time, and (c) amplitude inversion.

travel time exhibit the same degrees of nonlinearity. In con- initial model is far from the solution. The generalized travel-
trast, the amplitude inversion is three to four orders of magni- time inversion appears to retain most of the desirable features
tude more nonlinear than the travel-time inversion. This is partly of a travel-time inversion while obtaining an adequate ampli-
the reason for the failure of the amplitude inversion when the tude match.

Fig. 6—Estimated permeability distribution for the nine-spot case after (a) travel-time inversion, (b) generalized travel-time inver-
sion, and (c) amplitude inversion.

78 March 2005 SPE Journal

Mathematical Formulation: Sensitivity Sensitivity and Nonlinearity of Travel Time. Streamline meth-
Computations and Measures of Nonlinearity ods decouple flow and transport by a coordinate transformation
from the physical space to the time of flight along streamlines.22
We now discuss the mathematical details related to sensitivity
The time of flight is defined as
computation and measure of nonlinearity for travel-time, general-
ized travel-time, and amplitude inversion. Although the approach
is generally applicable, we will use a streamline simulator here
␶= 兰s共x兲dr, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
␺
because of the advantages in sensitivity computations. The sensi-
tivities quantify change in production response because of a small where the integral is along the streamline trajectory, ⌿, and s is the
change in reservoir properties. They are an integral part of most slowness defined as the reciprocal of the interstitial velocity,
inverse modeling methods. We also need the sensitivities to quan- 1 ␾
tify nonlinearities in the various inverse methods examined in this s= = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
|v| k␭rt|ⵜP|
study. Several approaches can be used to compute sensitivity co-
efficients of model parameters. Most of these methods fall into one The first-order derivative of slowness with respect to permeabili-
of the three categories (perturbation method, direct method, and ty is
adjoint state method19–21) and can be computationally demanding, ⭸s s
particularly for large-scale field applications. However, for stream- = − . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)
⭸k k
line models, it is possible to analytically derive a relationship
between perturbations in reservoir properties, such as permeability and the second-order derivative of slowness is
or porosity, and changes in observations such as watercut and ⭸2s s
tracer response. Streamline-based sensitivity computation is very =−. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4)
fast and involves quantities computed by a single streamline ⭸k 2
k2
simulation. Hence, we will limit our discussion to streamline mod- If we assume that the streamlines do not shift because of small
els only. perturbations in reservoir properties, we can then relate the change
We use the theory of Bates and Watts16 to measure the non- in travel time ␦␶ to the change in slowness by
linearity in production-data integration. Bates and Watts16 separate
the nonlinearity measures into parameter-effect curvature and in- ␦␶ = 兰␦s共x兲dr
␺
trinsic curvature; thus, they decompose the second-order derivative

兰冋− k共共x兲 ␦k共x兲 + ␾共共x兲 ␦␾共x兲册dr.

Fkk into one component parallel to the tangent plane defined by Fk s x兲 s x兲
for all directions and another component normal to that plane. = . . . . . . . . . . . . . . . . . . (5)
Here, we do not separate the intrinsic curvature and parameter ␺

effect curvature; neither do we consider the direction in the pa- The travel-time sensitivity along a single streamline at a producer
rameter space, because it is not practical to do so for our problem. with respect to permeability for a gridblock at location x is given
However, the theory we applied is essentially the same as that of by integrating Eq. 3 from the inlet to the outlet of the streamline ⌿
Bates and Watts.16 within the gridblock:

Fig. 7—Measure of nonlinearity for (a) travel-time inversion, (b) generalized travel-time inversion, and (c) amplitude inversion.

March 2005 SPE Journal 79

 ⭸␶共␺兲
⭸k共x兲
= 兰 冋− k共共x兲册dr共␺兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)
outlet

inlet
s x兲
C(t) = C0 t − 冉 兰s(x)dr
␺
冊 , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (21)

The overall travel-time sensitivity is then obtained by summing the where we have used the definition of time of flight from Eq. 1.
sensitivities over all streamlines contributing to the arrival time of Now, consider a small perturbation in reservoir properties, such
a particular concentration (for example, the peak concentration): as permeability. The resulting changes in slowness and concentra-
tions can be written as
⭸␶ ⭸␶共␺兲
⭸k共x兲
= 兺 ⭸k共x兲 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
all ␺
s共x兲 = s0共x兲 + ␦s共x兲; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22)

The second-order derivative of travel time along a single C共t兲 = C0共t兲 + ␦C共t兲, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23)
streamline is obtained by integrating Eq. 4, where s0 and C0 are initial slowness distribution in the reservoir
outlet and the associated tracer response, respectively. Applying Eqs. 21
⭸2␶共␺兲 s x兲
⭸k 共x兲
2
= 兰 − k 共共x兲dr共␺兲, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
2
and 22, the change in concentration response can be expressed as
inlet ␦C共t兲 = C共t兲 − C0共t兲
and then integrating over all streamlines contributing to a producer,
⭸2␶
= 兺 ⭸k 共x兲
⭸2␶共␺兲
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
= C0 t −冉 兰[s (x) + ␦s(x)]dr
␺
0
冊 冉 − C0 t − 兰s (x)dr)
␺0
0
冊 .

⭸k2共x兲 ␺
2 . . . . . . . . . . . . . . . . . . . . . . . . . . (24)
The components of the tangent vector Fk and acceleration vec- Using a Taylor series expansion and assuming ⌿ ⳱ ⌿0 (stationary
tor Fkk can now be obtained from Eqs. 7 and 9: streamlines), we get

Fk = 冉 ⭸␶ ⭸␶
,
⭸k1 ⭸k2
, ··· ,
⭸␶
⭸knb 冊 T
; . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10) 冉
C0 t − 兰[s (x) + ␦s(x)]dr 冊
␺
0

Fkk = 冉 ⭸␶ ⭸␶
2
2
⭸␶
, 2, ··· , 2
⭸k1 ⭸k2
2

⭸knb
2

冊 T

. . . . . . . . . . . . . . . . . . . . . . . . . . . . (11) 冉
= C0 t − 兰s (x)dr − 兰␦s(x)dr
␺0
0

␺0
冊
The 2-norms are used to calculate the vector norms,
冉
≈ C0 t − 兰s (x)dr 冊 + 冋 冉 t − 兰s (x)dr − 兰␦s(x)dr 冊
冉 兺冉 冊 冊
0 0
1Ⲑ2
⭸␶
nb 2
␺0 ␺0 ␺0

冉 兰s (x)dr 冊册C⬘冉 t − 兰s (x)dr 冊

||Fk|| = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)
⭸kj
j=1
− t− 0 0

冉 兺冉 冊 冊
0
1Ⲑ2 ␺0 ␺0
⭸␶
nb 2 2

冉 兰s (x)dr 冊 − 兰␦s(x)dr ⭈ C⬘冉 t − 兰s (x)dr 冊 . . . . (25)

||Fkk|| = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)
j=1 ⭸kj2 = C0 t − 0
0
0

␺0 ␺0 ␺0
Now we can calculate the nonlinearity measure of travel-time in-
version ␬tt according to the theory of Bates and Watts16 by Therefore the perturbation in C(t) and s(x) are related by
␬tt = ||Fkk|| Ⲑ ||Fk||2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (14)

Sensitivity and Nonlinearity of Amplitude. Tracer transport can

␦C(t) = −C⬘0 t −
冉 兰s (x)dr 冊兰␦s(x)dr.
␺0
0

␺0
. . . . . . . . . . . . . . . . . (26)

be described by the following convection-diffusion equation, The tracer-concentration sensitivity along a single streamline ⌿
is then
⭸C共x,t兲

冉 兰s (x)dr 冊兰 ⭸k(x) dr
␾ = ⵜ ⭈ 关D共x兲 ⭈ ⵜC共x,t兲兴 − u ⭈ ⵜC共x,t兲. . . . . . . . . . . . (15) ⭸C(t) ⭸s(x)
⭸t = −C⬘0 t − 0
⭸k(x)
Ignoring the dispersion term, Eq. 15 can be rewritten as ␺ ␺

⭸C共x,t兲
冉 兰s (x)dr 冊兰冋− k(x)册dr
s(x)
␾ + u ⭈ ⵜC共x,t兲 = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . (16) = −C⬘0 t − 0

⭸t ␺ ␺

冉 兰s (x)dr 冊
⭸␶(⌿)
Applying a transformation to the time-of-flight coordinate, the
tracer transport equation along a streamline can be expressed as22 = −C⬘0 t − 0
. . . . . . . . . . . . . . . . . . . . . (27)
⭸k(x)
␺
⭸C共␶,t兲 ⭸C共␶,t兲
+ = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17) The second-order derivative of the tracer concentration with
⭸t ⭸␶
respect to permeability is
For a unit-impulse concentration at (␶, t) ⳱ (0,0), the solution is22
C共x,t兲 = ␦关t − ␶共x兲兴, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
where ␦ is the Dirac-delta function. If the input is C0, then
⭸2C(t)
⭸k2(x)
= −C⬘0 t −
冉 兰s (x)dr 冊兰 ⭸k (x) dr
␺
0

␺
⭸2s(x)
2

C共x,t兲 = C0共t − ␶兲. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (19)

Summing the contributions of all streamlines reaching a producer,
= −C⬘0 t −
冉 兰s (x)dr 冊兰 k (x)dr
␺
0

␺
s(x)
2

冉 兰s (x)dr 冊 ⭸k (x) . . . . . . . . . . . . . . . . . . . . (28)

we get the tracer response at a producer as ⭸2␶(⌿)
= −C⬘0 t − 0

C共t兲 = 兰 C 共t − ␶兲d␺. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)

all ␺
0 ␺
2

As before, we need to sum over all streamlines reaching a producer

From Eq. 19, tracer response at the producer along a single stream- to get the final first-order and second-order derivatives of the
line is concentration response at the producer.

80 March 2005 SPE Journal

 Now, we need to evaluate the tangent vector Fk, the accelera-
tion vector Fkk, and measure of nonlinearity ␬ at different obser-
vation times. The vectors and norms are expressed as follows:
.
f⌬␶ = 冋
⭸f共x, ␶兲
⭸␶ r=⌬␶
册
⭸C共t + ⌬␶兲o ⭸t
兰
冉 冊
1
⭸C共ti兲 ⭸C共ti兲 ⭸C共ti兲 T = dt C共t兲c . . . . . . . . . . . . . . . . . . . . . . . . (37)
Fk共ti兲 = , , ···, , . . . . . . . . . . . . . . . . . . . . (29) A ⭸t ⭸␶
⭸k1 ⭸k2 ⭸knb ⭸C共t + ⌬␶兲o
兰
1

冉 冊
= dt C共t兲c = 0
⭸2C共ti兲 ⭸2C共ti兲 ⭸2C共ti兲 T
A ⭸t
Fkk共ti兲 = , , ···, ; . . . . . . . . . . . . . . . . (30)
⭸k21 ⭸k22 ⭸k2nb Note that ⭸/⭸␶ ⳱ 1 in this derivation. Eq. 37 is the function that is

冉 兺冉 冊 冊
used to compute the sensitivity of the generalized travel time.
1Ⲑ2
nb
⭸C共ti兲 2 Using Eq. 37 and the rule for the derivative of an implicit
||Fk共ti兲|| = , . . . . . . . . . . . . . . . . . . . . . . . . (31) function, we get
j=1 ⭸kj
.

冉 兺冉 冊 冊
⭸共f⌬␶兲
1Ⲑ2
nb
⭸2C共ti兲 2
⭸⌬␶ ⭸k共x兲
||Fkk共ti兲|| = . . . . . . . . . . . . . . . . . . . . . . . (32) = − . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (38)
j=1 ⭸kj2 ⭸k共x兲 ⭸共f⌬␶兲
⭸⌬␶
By definition, the measure of nonlinearity at observation time ti is
␬共ti兲 = ||Fkk共ti兲|| Ⲑ ||Fk共ti兲||2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (33) Taking the derivatives of ḟ⌬␶ with respect to k(x) and ⌬␶, we have
The final measure of nonlinearity for amplitude inversion ␬am is ⭸共 ḟ⌬␶兲 1 ⭸C共t + ⌬␶兲o ⭸C共t兲c
given by the maximum over all observed data, ⭸k共x兲 A
= 兰dt ⭸t ⭸k共x兲
␬am = max关␬共ti兲, ␬共t2兲, ···, ␬共tno兲兴. . . . . . . . . . . . . . . . . . . . . . . . (34) ⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸t ⭸␶
兰dt
1
=
A ⭸t ⭸t ⭸␶ ⭸k共x兲
Sensitivity and Nonlinearity of Generalized Travel Time. In
⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸␶
兰dt
generalized travel-time inversion, we define the misfit between the 1
= . . . . . . . . . . . . . . . . (39)
calculated and observed tracer concentrations in terms of the fol- A ⭸t ⭸t ⭸k共x兲
lowing correlation function12,15:
and
C共x, t + ␶兲o
f共x, ␶兲 = 兰 dt
Ao
C共x, t兲c , . . . . . . . . . . . . . . . . . . . . . . (35) ⭸ ḟ⌬␶ 兰dt ⭈ E, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (40)
=
where A is the maximum amplitude of tracer concentration and ␶ ⭸⌬␶ A
is the shift time between calculated and observed tracer concen- where
trations. We seek a ␶ that shifts the calculated tracer response so
that it best matches the observed tracer response.
The criterion for the “best” match is defined as the travel-time E=
⭸C共t + ⌬␶兲o ⭸C共t兲c
+ C共t兲c
⭸ 冋 ⭸C共t + ⌬␶兲o
⭸t 册
residual ⌬␶ that maximizes the previous correlation function, that is, ⭸t ⭸⌬␶ ⭸⌬␶
f共x, ⌬␶兲 = max兵 f共x, ␶兲|␶ ∈关−T, T兴其, . . . . . . . . . . . . . . . . . . . . . (36) ⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸t ⭸2C共t + ⌬␶兲o ⭸t
= + C共t兲c
where T is the estimated maximum travel-time difference between ⭸t ⭸t ⭸⌬␶ ⭸t2 ⭸⌬␶
the observed and calculated tracer responses. Therefore, the de-
rivative of f(x, ␶) with respect to ␶ should be zero at ⌬␶ unless the ⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸2C共t + ⌬␶兲o
= + C共t兲c . . . . . . . . . . . . . . (41)
maximum is at an endpoint T or –T, ⭸t ⭸t ⭸t2

Fig. 8—Tracer response for a quarter-five-spot heterogeneous case.

March 2005 SPE Journal 81

 Fig. 9—Sensitivity for (a) travel-time and (b) generalized travel-time inversion.

In the previous derivation, we have applied the relationship Finally, to calculate measures of nonlinearity, the components
of the tangent vector Fk and acceleration vector Fkk are obtained
⭸t ⭸t from Eqs. 42 and 43 as follows:

冉 冊
= = 1 at ␶ = ⌬␶. ⭸⌬␶ ⭸⌬␶ ⭸⌬␶
⭸⌬␶ ⭸␶ T
Fk = , , ···, ; . . . . . . . . . . . . . . . . . . . . . . . . . . (44)
⭸k1 ⭸k2 ⭸knb
Substitution of Eqs. 39 through 41 into Eq. 38 gives
Fkk = 冉 ⭸2⌬␶ ⭸2⌬␶
,
⭸2⌬␶
, ···, 2 冊 T
. . . . . . . . . . . . . . . . . . . . . . . . (45)

冋 册
⭸k1 ⭸k2
2 2
⭸knb
⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸␶
⭸⌬␶ 兰 dt
⭸t ⭸t ⭸k共x兲 The 2-norms of the vectors are calculated by

兰dt冋 册 冉兺 冉 冊 冊
= . . . (42)
⭸k共x兲 ⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸2C共t + ⌬␶兲o nb
⭸⌬␶
2 1Ⲑ2
+ C共t兲c ||Fk|| = ; . . . . . . . . . . . . . . . . . . . . . . . . . . . (46)
⭸t ⭸t ⭸t2 j=1 ⭸kj

The second-order derivative of generalized travel-time with

respect to permeability is then
||Fkk|| = 冉兺 冉 冊 冊
nb

j=1
⭸2⌬␶
⭸kj2
2 1Ⲑ2

. . . . . . . . . . . . . . . . . . . . . . . . . . . (47)

The measure of nonlinearity for the generalized travel-time inver-

兰dt冋 册
⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸ ␶
2 sion is evaluated using Eqs. 46 and 47:
⭸2⌬␶ ⭸t ⭸t ⭸k2共x兲 ␬gt = ||Fkk|| Ⲑ ||Fkk||2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (48)

兰dt冋 册
= . . . (43)
⭸k2共x兲 ⭸C共t + ⌬␶兲o ⭸C共t兲c ⭸2C共t + ⌬␶兲o Sensitivity Computations: A Quarter-Five-Spot Example. We
+ C共t兲c
⭸t ⭸t ⭸t2 illustrate sensitivity computations for the three methods using the
tracer response in a heterogeneous quarter five-spot pattern (Fig. 8).
Fig. 9a is the sensitivity distribution for the peak travel-time, and
⭸2␶
where is calculated by Eq. 8. Fig. 9b is the sensitivity distribution for the generalized travel-
⭸k2 time. Figs. 10a through 10c show the sensitivity distribution for

Fig. 10—Sensitivity distribution for amplitude inversion (a) before peak time, (b) at peak time, and (c) after peak time.

82 March 2005 SPE Journal

Fig. 11—A two-phase example with infill drilling: (a) reference permeability model, triangle for infill wells in the midterm of
production; (b) estimated permeability by travel-time inversion; (c) estimated permeability by generalized travel-time inversion;
and (d) estimated permeability by amplitude inversion.

the amplitude before, at, and after peak time, respectively. From available geologic and static information related to the reservoir.
Figs. 9 and 10, we can see that the sensitivity distribution between Finally, the third term, a roughness penalty, simply recognizes the
the wells for travel-time inversion is more uniform than that for fact that production data are an integrated response and are thus
amplitude inversion. Also, the magnitude of the amplitude sensi- best suited to resolve large-scale structures rather than small-scale
tivity is much smaller than that of the travel-time sensitivity. This property variations.
smaller sensitivity contributes to the high nonlinearity of ampli- The minimum in Eq. 49 can be obtained by an iterative least-
tude inversion, because the nonlinearity is evaluated by 㛳Fkk㛳/ squares solution to the augmented linear system
㛳Fk㛳2, where Fk is the sensitivity vector. Such a relationship be-

冉 冊 冉冊
tween nonlinearity and sensitivity for inverse modeling has also S ␦d
been observed by Grimstad and Mannseth.17,18
␤1I ␦R = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (50)
Data Inversion ␤2L 0
Our goal is to reconcile high-resolution geologic models to field-
production history—for example tracer response. This typically The weights ␤1 and ␤2 determine the relative strengths of the prior
involves the solution of an underdetermined inverse problem. The model and the roughness term. The selection of these weights can
mathematical formulation behind such streamline-based inverse be somewhat subjective, although there are guidelines in the lit-
problems has been discussed elsewhere.2,4,5 Briefly, in our ap- erature.23 In general, the inversion results will be sensitive to the
proach we start with a prior static model that already incorporates choice of these weights.
geologic, well-log, and seismic data. We then minimize a penal- In Eq. 50, ␦d is replaced by ␦␶ for travel-time inversion, ␦C for
ized misfit function consisting of the following three terms: amplitude inversion, and ␦⌬␶ for generalized travel-time inver-
||␦d − S␦R||+␤1||␦R||+␤2||L␦R||. . . . . . . . . . . . . . . . . . . . . . . . . . (49) sion. The sensitivity matrix S is also replaced by the correspond-
ing expression.
In Eq. 49, ␦d is the vector of data residuals at the wells, while S is Note that one of the major advantages of travel time and the
the sensitivity matrix containing the sensitivities of the observed generalized travel-time approach is that the size of the sensitivity
data with respect to the reservoir parameters. Also, ␦R corresponds matrix S is dependent only on the number of wells regardless of
to the change in the reservoir property, and L is a second-spatial- the number of data points. This leads to considerable savings
difference operator. The first term ensures that the difference be- in computation time. We use an iterative sparse matrix solver,
tween the observed and calculated production response is mini- LSQR, for solving this augmented linear system efficiently.24 The
mized. The second term, called a norm constraint, penalizes de- LSQR algorithm is well suited for highly ill-conditioned systems
viations from the initial model. This helps preserve geologic and has been widely used for large-scale tomographic problems
realism because our initial or prior model already incorporates in seismology.25

Fig. 12—Water-cut response (a) for uniform initial permeability, (b) after peak arrival-time inversion, (c) after generalized travel-time
inversion, and (d) after direct amplitude inversion.

March 2005 SPE Journal 83

Fig. 13—Measure of nonlinearity for the two-phase, infill example: (a) travel-time inversion, (b) generalized travel-time inversion,
and (c) amplitude inversion.

Applications homogeneous or smooth starting models, we can obtain a reason-

A Two-Phase Example With Infill Drilling. So far, we have able solution by careful choice of inversion parameters. But for
focused on single-phase tracer flow. We now consider a two-phase models with significant heterogeneity, especially for field appli-
waterflood example with changing streamlines.12 The flood pat- cations, direct amplitude inversion often fails.
tern is a nine-spot. We start with one central injector and four side
producers. Four corner producers are introduced at 300 days. Pres-
sure and streamlines are updated every 100 days. Fig. 11a shows
the reference permeability and the well pattern. The reference
permeability is the same as the one used for the tracer example.
The water-cut responses from the eight producers are shown in
Fig. 12a. Also superimposed in Fig. 12a are the water-cut re-
sponses from the initial model, a homogeneous permeability field
conditioned at the well locations.
Figs. 12b through 12d show the water-cut match by travel-time
inversion, generalized travel-time inversion, and amplitude inver-
sion, respectively. Clearly, the match by generalized travel-time
inversion is the best, followed by travel-time inversion. Amplitude
match did not work for Well 8.
Fig. 11b is the estimated permeability field after travel-time
match. On comparing with Fig. 11a, we can see that the low-
permeability areas are reproduced well; however, the high-
permeability contrast to the south is not detected properly. Fig. 11c
shows the permeability field derived by generalized travel-time
inversion. It reproduces not only the low-permeability areas but
also the high-permeability regions. Fig. 11d shows the estimated
permeability field after the amplitude inversion. Clearly, the re-
sults show signs of instability, as discussed before.
Fig. 13 shows the measure of nonlinearity for the three ap-
proaches. We can see that both the travel time and the generalized
travel time have a similar magnitude of nonlinearity. In contrast,
the amplitude inversion is three to four orders of magnitude more
nonlinear than the travel-time inversion. This is partly the reason
for the failure of the amplitude inversion. Our experience with
amplitude inversion indicates that the results tend to be more sen-
sitive to the choice of inversion parameters (␤1, ␤2 in Eq. 49). For Fig. 14—Tracer injection pattern: the Ranger field case.

84 March 2005 SPE Journal

 Fig. 15—NaSCN tracer response for the initial permeability field at (a) Well 40, (b) Well 37, and (c) Well 39.

Field Application: The Ranger Field, Texas. A multiwell, multi- able for analysis. We did not have well- and depth-specific data,
tracer, interwell tracer injection study was carried out in the but rather a summary of the core data for all wells. The core data
McCleskey sandstone of the Ranger field, Texas. The first descrip- indicated a fair degree of permeability heterogeneity in the reser-
tion of this data set was published by Lichtenberger.26 The dataset voir, but only slight variation in porosity. For the initial model, we
was also described later by Allison et al.27 The 320-acre area of used a uniform value of porosity and a heterogeneous permeability
interest includes 13 producers and four injectors, injecting seven field generated using Sequential Gaussian Simulation29 based on
different tracers. The seven tracers injected included five conser- well data (Fig. 16). We assume that kx ⳱ ky , kz ⳱ 0.1kx , and only
vative tracers consisting of four decaying (Tritium, Cobalt-57, Co- kx is altered during inversion.
balt-58, and Cobalt-60), one chemical (sodium thiocyanate, Estimating Permeability. We matched the NaSCN data to ob-
NaSCN), and two partitioning tracers (tertiary butyl alcohol, TBA, tain the permeability distribution (kx) in the study area using the
and isopropyl alcohol, IPA). three different approaches: travel-time inversion, generalized
All tracers were injected in small slugs on the same day except
for TBA, which was injected in a small slug 20 days later. Tracer
sampling continued for 826 days after injection of the first set of
tracers. The tracer injection pattern is shown in Fig. 14. Detailed
information for injection locations and the amounts of each tracer
injected can be found elsewhere.27,28
We can use the conservative tracers (Tritium and NaSCN) to
obtain permeability distribution in the study area. However, the
Tritium response may be affected by a chromatographic delay
because of tritium exchange with immobile hydrogen.26 We se-
lected NaSCN as the conservative tracer for permeability inver-
sion. Totally, 5,655 lbs of NaSCN was injected into Well 38, and
four wells (Wells 19, 37, 39, and 40) showed tracer response as
indicated in Fig. 14. The observed tracer responses in Wells 37, 39,
and 40 are shown in Fig. 15, along with the calculated response
from the initial permeability model. The data from Well 19 were
not used because of its low production rate (<20 B/D).
Choice of an Initial Model. During inverse modeling, a proper
selection of the initial model can be critical to ensure a plausible
solution. Such an initial model should incorporate all available
prior information. For our simulation studies, we use a 31×45×6
grid which corresponds to 100×100-ft gridblocks areally, and 2- to Fig. 16—Initial permeability distribution for the Ranger field
4-ft gridblocks vertically. A total of 141 core samples were avail- case.

March 2005 SPE Journal 85

travel-time inversion, and amplitude inversion. Fig. 15 shows the Fig. 21 shows the permeability fields derived by travel-time
NaSCN responses from a streamline simulator using the initial inversion and generalized travel-time inversion. Fig. 22 shows the
permeability field. Also superimposed are the observed NaSCN permeability change after travel-time inversion and generalized
concentrations. Clearly, there is a large difference between the travel-time inversion. In Fig. 23, we show that there is a general
calculated and observed NaSCN response. Fig. 17 shows the agreement between our final model and the permeability distribu-
NaSCN concentration match after travel-time inversion. The peak tion reported by Allison et al.27 by a manual history matching of
arrival times are now in agreement with the observed data. The the tracer data. The most significant change by Allison et al. was
tracer concentration amplitudes show improvement but the overall introduction of a high-permeability streak in the original perme-
match is still not satisfactory. Fig. 18 is the NaSCN concentration ability model between Wells 38 and 40. Our results from gener-
match after the generalized travel-time inversion. From Fig. 18, we alized travel-time inversion also indicate the presence of higher
can see that not only are the peak arrival times well matched, but permeability between Wells 38 and 40 in the corresponding layer
the calculated concentration amplitudes are also in close agree- (Fig. 23). However, our results did not require the additional
ment with the observed data. This shows that generalized travel- changes near the boundary obtained by Allison et al.
time inversion is an effective one-step inversion process. Fig. 19
displays the NaSCN concentration match after direct amplitude
inversion. Clearly, the calculated responses have changed very Summary and Conclusions
little from the initial responses. The results indicate that amplitude We have presented three approaches to production-data integration
inversion may not be as effective as the travel-time inversion, and examined their relative merits using quantitative measures of
particularly when the initial model is far from the solution. Gen- nonlinearity. These are travel-time, generalized travel-time, and
eralized travel-time inversion stands out as the best among the the commonly used amplitude inversion. The travel-time inversion
three inversion methods. of production data is robust and computationally efficient. Unlike
Fig. 20 summarizes nonlinearity for the three inversion meth- conventional amplitude matching that can be highly nonlinear, the
ods. The measure of nonlinearity for the field example is given by travel-time inversion has quasilinear properties. This makes the
the maximum among the three producers. Amplitude inversion method particularly attractive for field-scale applications in which
displays the highest measure of nonlinearity, approximately 200 to the prior geologic model might be far from the solution. The
250, while travel-time inversion is quasilinear, with a nonlinearity generalized travel-time inversion appears to retain most of the
of approximately 0.2 to 0.4. The generalized travel-time inversion desirable features of the travel-time inversion and also accom-
is between these two cases in terms of nonlinearity measure. How- plishes the amplitude match. Some specific findings from this
ever, it is one order of magnitude larger than the travel-time in- study can be summarized as follows:
version, while two orders of magnitude smaller than that of the 1. We have quantitatively investigated the nonlinearities associ-
amplitude inversion. Generalized travel-time inversion keeps most ated with travel-time and amplitude inversion for production
of the favorable features of travel-time inversion and has a much data integration. The nonlinearity is expressed in terms of a
better tracer-concentration amplitude match than travel-time inver- simple and intuitive geometric measure of curvature as pro-
sion. The severe nonlinearity of the amplitude inversion is partly posed by Bates and Watts16 and later used by Grimstad
responsible for its poor performance for the field case. and Mannseth.17

Fig. 17—NaSCN tracer response after travel-time inversion at (a) Well 40, (b) Well 37, and (c) Well 39.

86 March 2005 SPE Journal

 Fig. 18—NaSCN tracer response after generalized travel-time inversion at (a) Well 40, (b) Well 37, and (c) Well 39.

Fig. 19—NaSCN tracer response after direct amplitude inversion at (a) Well 40, (b) Well 37, and (c) Well 39.

March 2005 SPE Journal 87

Fig. 20—Measure of nonlinearity for (a) travel-time inversion, (b) generalized travel-time inversion, and (c) amplitude inversion.

2. The nonlinearity in travel-time inversion is found to be orders of desirable properties of the travel-time inversion. For the field
magnitude smaller than the conventional amplitude inversion. example studied here, the generalized travel-time inversion out-
As a result, the travel-time inversion has better convergence performed both travel-time and amplitude inversion.
properties and is less likely to be trapped in local minimum.
3. Travel-time sensitivity is more uniform between the wells. In Nomenclature
contrast, the amplitude sensitivity can be localized near the
wells. The higher magnitude of the travel-time sensitivity also con- d ⳱ data vector
tributes to its quasilinearity and improved convergence properties. Cc ⳱ calculated tracer concentration
4. The generalized travel-time inversion effectively combines Co ⳱ observed tracer concentration
travel-time and amplitude inversion while retaining most of the D ⳱ dispersion coefficient

Fig. 21—Derived permeability field after NaSCN concentration match by (a) generalized travel-time inversion and (b) travel-time
inversion.

88 March 2005 SPE Journal

 Fig. 22—Permeability change after (a) generalized travel-time match and (b) travel-time match.

Fk ⳱
tangent vector Acknowledgments
Fkk ⳱
acceleration vector The authors would like to acknowledge financial support from
I ⳱
identity matrix members of the Joint Industry Project on reservoir data integration
k ⳱
permeability and also from the U.S. DOE.
L ⳱
spatial difference operator
nb ⳱
number of gridblocks
no ⳱
number of dynamic data observations References
s ⳱
slowness 1. Bissel, R.C.: “Calculating Optimal Parameter for History Matching,”
S ⳱
sensitivity matrix paper presented at the 1994 European Conference on the Mathematics
t ⳱
time of Oil Recovery, Topic E: History Match and Recovery Optimization,
u ⳱
Darcy velocity Røros, Norway, 7–10 June.
v ⳱
interstitial velocity 2. Vasco, D.W., Yoon, S., and Datta-Gupta, A.: “Integrating Dynamic
␤ ⳱
weighting factor Data Into High-Resolution Reservoir Models Using Streamline-Based
␭rt ⳱
total relative mobility Analytic Sensitivity Coefficients,” SPEJ (December 1999) 389.
3. Oliver, D.S. et al.: “Integration of Production Data into Reservoir
⌬␶ ⳱
generalized travel time or travel-time shift
Models,” Petroleum Geoscience (1994) 18, No. 3, 243.
␬am ⳱
measure of nonlinearity for amplitude inversion 4. Datta-Gupta, A. et al.: “Streamlines, Ray Tracing and Production To-
␬gt ⳱
measure of nonlinearity for generalized travel-time mography: Generalization to Compressible Flow,” Petroleum Geo-
inversion science (May 2001) 75.
␬tt ⳱ measure of nonlinearity for travel-time inversion 5. Yoon, S. et al.: “A Multiscale Approach to Production-Data Integration
␶ ⳱ time of flight Using Streamline Models,” SPEJ (June 2001) 182.

Fig. 23—(a) Permeability multipliers from the manual history match in Layer 3, by Allison et al.,27 and (b) permeability change from
generalized travel-time inversion in the corresponding layer.

March 2005 SPE Journal 89

 6. Landa, J.L. et al.: “Reservoir Characterization Constrained to Well 21. Sun, N.-Z. and Yeh, W.W.-G.: “Coupled Inverse Problems in Ground-
Test Data: A Field Example,” paper SPE 36511 presented at the 1996 water Modeling—I, Sensitivity Analysis and Parameter Identification,”
SPE Annual Technical Conference and Exhibition, Denver, 6–9 Octo- Water Resources Research (1990) 26, 2507.
ber. 22. Datta-Gupta, A. and King, M.J.: “A Semianalytic Approach to Tracer
7. Milliken, W.J., Emanuel, A.S., and Chakravarty, A.: “Application of Flow Modeling in Heterogeneous Permeable Media,” Advances in Wa-
3D Streamline Simulation to Assist History Matching,” paper SPE ter Resources (1995) 18, No. 1, 9.
63155 presented at the 2000 SPE Annual Technical Conference and 23. Parker, R.L.: Geophysical Inverse Theory, Princeton U. Press, Prince-
Exhibition, Dallas, 1–4 October. ton, New Jersey (1994).
8. Wang, Y. and Kovscek, A.R.: “A Streamline Approach to History- 24. Paige, C.C. and Saunders, M.A.: “LSQR: An Algorithm for Sparse
Matching Production Data,” paper SPE 59370 presented at the 2000 Linear Equations and Sparse Least Squares,” ACM Trans. Math. Soft-
SPE/DOE Improved Oil Recovery Symposium, Tulsa, 3–5 April. ware (1982) 8, No. 1, 43.
9. Wu, Z. and Datta-Gupta, A.: “Rapid History Matching Using a Gen- 25. Nolet, G.: “Seismic Wave Propagation and Seismic Tomography,”
eralized Travel-Time Inversion Method,” SPEJ (June 2002) 113. Seismic Tomography, G. Nolet (ed.), D. Reidel, Dordrecht, The Neth-
10. Reynolds, A.C., He, N., and Oliver, D.S.: “Reducing Uncertainty in erlands (1987) 1–23.
Geostatistical Description With Well Testing Pressure Data,” paper 26. Lichtenberger, G.J.: “Field Application of Interwell Tracers for Reser-
presented at the 1997 Intl. Reservoir Characterization Conference, voir Characterization of Enhanced Oil Recovery Pilot Areas,” paper
Houston, 2–4 March. SPE 21652 presented at the 1991 SPE Production Operations Sympo-
11. Landa, J.L. and Horne, R.N.: “A Procedure to Integrate Well Test Data, sium, Oklahoma City, Oklahoma, 7–9 April.
Reservoir Performance History, and 4D Seismic Information Into a 27. Allison, S.B., Pope, G.A., and Sepehrnoori, K.: “Analysis of Field
Reservoir Description,” paper SPE 38653 presented at the 1997 SPE Tracers for Reservoir Description,” J. Pet. Sci. Eng. (1991) 5, 173.
Annual Technical Conference and Exhibition, San Antonio, Texas, 5–8 28. Illiassov, P.A. and Datta-Gupta, A.: “Field-Scale Characterization of
October. Permeability and Saturation Distribution Using Partitioning Tracer
12. He, Z., Yoon, S., and Datta-Gupta, A.: “Streamline-Based Production Tests: The Ranger Field, Texas,” SPEJ (December 2002) 409.
Data Integration With Gravity and Changing Field Conditions,” SPEJ 29. Deutsch, C.V. and Journel, A.G.: GSLIB: Geostatistical Software Li-
(December 2002) 423. brary and User’s Guide, Oxford U., Oxford, U.K. (1998).
13. King, M.J. and Datta-Gupta, A.: “Streamline Simulation: A Current
Perspective,” In Situ (1998) 22, No. 1, 91.
Hao Cheng is a PhD candidate in petroleum engineering at
14. Wen, X.-H., Deutsch, C.V., and Cullick, A.S.: “High-Resolution Res-
Texas A&M U. e-mail: [email protected]. His research in-
ervoir Models Integrating Multiple-Well Production Data,” paper SPE terests are reservoir engineering, reservoir characterization,
38728 presented at the 1997 SPE Annual Technical Conference and and dynamic data integration. Cheng holds a PhD degree
Exhibition, San Antonio, Texas, 5–8 October. from the U. of Petroleum, Beijing, and a BS degree from Xi’an
15. Luo, Y. and Schuster, G.T.: “Wave-Equation Traveltime Inversion,” Petroleum Inst., both in petroleum engineering. Akhil Datta-
Geophysics (1991) 56, No. 5, 645. Gupta is the Rob L. Adams Endowed Professor in Petroleum
16. Bates, D.M. and Watts, D.G.: “Relative Curvature Measures of Non- Engineering at Texas A&M U. e-mail: datta-gupta@spindletop.
linearity,” J. R. Statist. Soc. B (1980) 42, No. 1, 1. tamu.edu. He has worked for BP plc Exploration/Research and
17. Grimstad, A.-A. and Mannseth, T.: “Nonlinearity, Scale, and Sensitiv- the Lawrence Berkeley Natl. Laboratory. Datta-Gupta holds a
PhD degree in petroleum engineering from the U. of Texas at
ity for Parameter Estimation Problems,” SIAM J. Sci. Comput. (2000)
Austin. He is the recipient of the 2003 SPE Lester C. Uren Award
21, No. 6, 2096. for significant technical contributions in petroleum reservoir
18. Mannseth, T. et al.: “Functional Representation and Model Nonlinear- characterization and streamline-based flow simulation. He is
ity in Estimation of Porous Media Properties From Laboratory Experi- an SPE distinguished member (2001), distinguished lecturer
ments,” paper presented at the 1997 SIAM Conference on Mathemati- (1999–2000), and distinguished author (2000), and he was se-
cal and Computational Issues in the Geosciences, Albuquerque, New lected as an outstanding technical editor (1996). He also re-
Mexico, 16–19 June. ceived the SPE Cedric K. Ferguson Certificate (2000) and the
19. Yeh, W.W.-G.: “Review of Parameter Identification Procedures in AIME Rossitter W. Raymond Award (1992). Zhong He is current-
Groundwater Hydrology: The Inverse Problem,” Water Resources Re- ly employed by Schlumberger Data and Consulting services
in Denver. His research interests include reservoir engineer-
search (1986) 22, No. 2, 95.
ing, reservoir simulation, and characterization. He holds a PhD
20. Bissel, R.C., Killough, J.E., and Sharma, Y.: “Reservoir History degree from Texas A&M U., a BS degree from Jianghan Pe-
Matching Using the Method of Gradients on a Workstation,” paper SPE troleum Inst., and an MS degree from the Research Inst. of
24265 presented at the 1992 SPE European Petroleum Computer Con- Petroleum Exploration and Development in China, all in petro-
ference, Stavanger, 25–27 May. leum engineering.

90 March 2005 SPE Journal