Streamline-Based Time-Lapse-Seismic-Data Integration Incorporating Pressure and Saturation Effects

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/316028888
Streamline-Based Time-Lapse-Seismic-Data Integration Incorporating

Pressure and Saturation Effects
Article in SPE Journal · April 2017

DOI: 10.2118/166395-PA
CITATIONS READS
8 148
6 authors, including:
Shingo Watanabe Gill Hetz

Schlumberger Limited Texas A&M University
11 PUBLICATIONS 100 CITATIONS 14 PUBLICATIONS 66 CITATIONS
SEE PROFILE SEE PROFILE
Michael King Donald Wyman Vasco

Texas A&M University Lawrence Berkeley National Laboratory
109 PUBLICATIONS 1,663 CITATIONS 181 PUBLICATIONS 4,317 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
multiscale simulation in INTERSECT View project
Basic Energy Sciences View project
All content following this page was uploaded by Donald Wyman Vasco on 04 October 2018.
The user has requested enhancement of the downloaded file.

J166395 DOI: 10.2118/166395-PA Date: 28-March-17 Stage: Page: 1 Total Pages: 19
Streamline-Based Time-Lapse-Seismic-
Data Integration Incorporating Pressure
and Saturation Effects
Shingo Watanabe*, Jichao Han**, Gill Hetz, Akhil Datta-Gupta, and Michael J. King, Texas A&M University;
and D. W. Vasco, Lawrence Berkeley National Laboratory
Summary sequential Gaussian simulation with block Kriging, accounts for

We present an efficient history-matching technique that simulta- the volume-support differences between the seismic-interval-aver-
neously integrates 4D repeat seismic surveys with well-produc- age rock properties and the well-log point/rock properties.
tion data. This approach is particularly well-suited for the Lumley et al. (1997) and Lumley and Behrens (1998) discussed
calibration of the reservoir properties of high-resolution geologic the practical issues relevant to the successful monitoring of oil/
models because the seismic data are areally dense but sparse in water and other reservoir-fluid-flow systems. The general technical
time, whereas the production data are finely sampled in time but challenges associated with time-lapse-seismic monitoring were
spatially averaged. The joint history matching is performed by thoroughly reviewed by Lumley (2001). With these proposed
use of streamline-based sensitivities derived from either finite-dif- guidelines, the field applications of time-lapse-seismic data has
ference or streamline-based flow simulation. For the most part, matured (Fanchi 2001; Clifford et al. 2003), and time-lapse-seis-
earlier approaches have focused on the role of saturation changes, mic monitoring is now recognized as a reservoir-management tool
but the effects of pressure have largely been ignored. Here, we with a number of successful field applications (Landrø et al. 1999;
present a streamline-based semianalytic approach for computing Behrens et al. 2002; Toinet 2004; Foster 2007). The next logical
model-parameter sensitivities, accounting for both pressure and step is to use the time-lapse-seismic data to infer reservoir perme-
saturation effects. The novelty of the method lies in the semi- ability and porosity heterogeneity through a process of reservoir-
analytic sensitivity computations, making it computationally model calibration. Traditional history matching simply involves
efficient for high-resolution geologic models. The approach is production data and, because these data provide limited informa-
implemented by use of a finite-difference simulator incorporating tion regarding permeability and porosity variations away from the
the detailed physics. Its efficacy is demonstrated by use of both wells, the solution is highly nonunique. The possibility of incorpo-
synthetic and field applications. For both the synthetic and the field rating time-lapse-seismic observations as dynamic data is attrac-
cases, the advantages of incorporating the time-lapse variations are tive because they are influenced by fluid-saturation changes
clear, seen through the improved estimation of the permeability between the wells. Time-lapse-seismic images can identify
distribution, the pressure profile, the evolution of the fluid satura- bypassed oil to be targeted for infill drilling, possibly adding major
tion, and the swept volumes. reserves to extend a field’s economic life. Also, time-lapse-seismic
data can map the reservoir compartmentalization and identify the
Introduction fluid-flow properties of faults (sealing vs. leaking), which can be
3D reservoir-simulation models play an essential role in the oil and extremely useful for optimal well design in complex reservoir-flow
gas industry. They are routinely used for planning field develop- systems. Previous studies have made several attempts in quantita-
ment, calculating hydrocarbon reserves, and predicting future pro- tive time-lapse-seismic history matching. However, the reconcilia-
duction estimates. Because of sparse well distributions, the tion of reservoir heterogeneity with temporal changes in seismic
interwell volume of a reservoir model is often poorly constrained. attributes remains a particularly complex task (Gosselin et al.
A key challenge is the quantitative data integration of 3D- and 4D- 2001). Several dynamic data-integration algorithms have been pro-
seismic data to obtain a more-accurate representation of reservoir posed in the literature, which can be broadly categorized into three
properties between wells. Geostatiscal techniques have been data-integration levels. The first level is reservoir-simulation-based
widely adopted in the petroleum industry to construct reservoir integration between the pressure and saturation estimates inverted
models, and commercial and research geomodeling-software pack- from seismic-observation data and the direct simulated saturation
ages, such as GSLIB (Deutsch and Journel 1992), are now wide- and pressure responses. The second is petroelastic-based integra-
spread. There are several difficulties associated with integrating tion between the seismic-inverted-rock-elastic properties derived
seismic and well data for both 3D and 4D reservoir characteriza- from a geophysical inversion (such as acoustic impedance and
tion. First, the seismic data must be converted from time to depth, compressional velocity) and the simulated rock-elastic responses
but the velocity model is only accurately calibrated at the wells. from the simulated saturation and pressure responses by means of
Second, seismic data are band-limited, whereas well data have petroelastic models (PEMs). The third level is seismic-forward-
both high- and low-frequency components. Therefore, the seismic model-based integration between the direct seismic-trace attributes
response must be calibrated to the well data, but loses the high-fre- (e.g., reflection amplitude) and the simulated seismic responses by
quency information. The limited vertical resolution of inverted means of seismic-wave-propagation modeling. A diagram of the
seismic data has been a major obstacle to the widespread use of differing seismic-data-integration levels is shown in Fig. 1.
seismic information in modeling (Doyen et al. 1997). Behrens The forward-model-based data integration uses the seismic
et al. (1998) introduced a geostatistical method to incorporate seis- traces directly, circumventing the uncertainty associated with the
mic-attribute maps into a 3D reservoir model. The method, called seismic-inversion process. The general work flow for generating
synthetic seismic data from a reservoir-simulation model involves
generating static reservoir properties on the simulation grid; simu-
* Now with Schlumberger
lation of dynamic pressure and fluid saturations at each cell; com-
** Now with Anadarko
putation of the seismic-elastic properties; and simulation of the
Copyright V
C 2017 Society of Petroleum Engineers
seismic attribute (e.g., reflection amplitude) by applying a seis-
This paper (SPE 166395) was accepted for presentation at the SPE Annual Technical mic-wave-propagation model over the reservoir interval and the
Conference and Exhibition, New Orleans, 30 September–2 October 2013, and revised for
publication. Original manuscript received for review 30 March 2016. Revised manuscript
overburden rock (Mavko et al. 1998; Falcone et al. 2004). This is
received for review 1 December 2016. Paper peer approved 6 December 2016. a computationally demanding approach because it requires an
2017 SPE Journal 1
ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Forward Modeling Reservoir PEM Seismic

Simulation Modeling
Pressures Acoustic Simulated

Simulated
Saturations Impedance Seismogram
Data
Permeability 1 2 3
Porosity
Observation Pressures Acoustic Real

Data Saturations Impedance Seismogram
Inverse Modeling Rock-Physics Seismic Inversion

Inversion
Fig. 1—Seismic-data-integration levels: (1) reservoir-simulation level of data integration, (2) petroelastic-modeling level of data
integration, and (3) seismic-forward-modeling level of data integration.
iterative coupling between the seismic modeling and the flow sim- will also integrate this approach in a sequential fashion with other
ulation and may be prohibitively expensive for an inversion work forms of data calibration. In both cases, we will follow the seis-
flow (Gosselin et al. 2003). Despite this, there are a number of mic-data integration with a fairly conventional streamline-based
publications that use direct seismic attributes for model calibra- integration of water-cut field performance. These data are high re-
tion (Huang et al. 1997; Vasco et al. 2004; Kjelstadli et al. 2005; solution in time, but only available at the production wells.
Dadashpour et al. 2008, 2009, 2010). In this paper, we will start with the mathematical background
Unlike the direct seismic methods, the petroelastic- and the necessary to determine the streamline-based analytic sensitivities
reservoir-simulation-based approaches are more efficient because for both fluid-saturation and pressure observations. The semiana-
the seismic-volumes inversion is not a part of the model-calibra- lytic sensitivities are validated through a comparison with numeri-
tion work flow. There are several papers in the literature discus- cal estimates. We then illustrate the history-matching applications
sing the derivation of the inverted seismic responses (Tura and of time-lapse pressure and saturation changes by use of a syn-
Lumley 2000; Landrø et al. 2001). Recent efforts have focused on thetic five-spot model. Finally, we apply the proposed approach to
the petroelastic- and the reservoir-simulation-based approaches in a field example, the Norne Field, in terms of time-lapse acoustic-
various forms. For example, Landa and Horne (1997) proposed a impedance change derived from the post-stack seismic-amplitude
gradient-based method and used inverted saturation from inverted data. This set of examples demonstrates both the utility and the
time-lapse data. Gosselin et al. (2001) used synthetic acoustic-im- effectiveness of our approach.
pedance maps to calibrate their model. Arenas et al. (2001) used
the compressional velocity to update the permeability field. Dong Background and Methodology
and Oliver (2005) used a quasi-Newton method to modify the po-
Our inversion approach relies on the use of streamlines to relate
rosity and permeability dependent on changes in acoustic imped-
small changes in reservoir properties to variations in the dynamic
ance from a geophysical seismic inversion. Rey et al. (2012)
reservoir response (production, pressure, and time-lapse-seismic
applied a streamline-based sensitivity calculation to integrate the
response). Establishing these relationships, known as sensitivities,
seismic-derived water-saturation changes and the acoustic-imped-
is crucial to the data-integration process. We make extensive use
ance differences and demonstrated field-scale applications. A num-
of streamlines to describe the flow field and to calculate the fluid
ber of studies have adopted stochastic approaches. Skjervheim
time of flight along each streamline. The time of flight acts as a
et al. (2007) used the ensemble Kalman smoother to assimilate the
spatial coordinate along the streamline. However, our approach
time-lapse-seismic data of changes in acoustic impedance and
does not require the use of a streamline simulator. Rather, it is ap-
compressional velocity. Similarly, Fahimuddin et al. (2010) used
plicable to both conventional finite-difference and streamline res-
seismic-impedance estimates along with an ensemble Kalman
ervoir simulators. In fact, most of the examples presented in this
filter, including a covariance localization method. Feng and
paper use a commercial finite-difference simulator for the flow
Mannseth (2010) incorporated pseudoseismic data in the form of
simulations. For streamline simulators, the streamline trajectories
maps of saturation changes to investigate the effect of the seismic
and time of flight are readily available. However, for a finite-dif-
data on permeability estimates in the presence of noise.
ference simulator, the streamlines and the time of flight are
In this study, we examine both reservoir-simulation-based-cal-
obtained by post-processing the simulator velocity field. In this
ibration and petroelastic-based-calibration techniques. In the first
section, we discuss the mathematical details related to sensitivity
instance, we calibrate against time-lapse changes in saturation and
computations in streamline-based seismic-data integration.
pressure. In the second case, we update our model by use of time-
lapse changes in acoustic impedance with a hierarchical approach
involving global and local updates. First, we perform a global- Time of Flight and Saturation-Front-Sensitivity Calculation.
model calibration by use of geologic-model reparameterization We start with the definition of the streamline time of flight, the
and a multiobjective optimization. Next, local updates to the res- travel time of a neutral tracer along a streamline (Datta-Gupta and
ervoir model are performed by use of semianalytic, streamline- King 2007):
based-model-parameter sensitivities for saturation and pressure. ð
Although there have been attempts to estimate pore-pressure sðwÞ ¼ sðxÞdr; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ
changes from time-lapse data (Tura and Lumley 2000; Landrø w
et al. 2001, 2003) and even to use the pressure changes to estimate
permeability (Vasco 2004), the major new result of this study is where the integral is along the streamline trajectory parameterized
the use of streamline-based, semianalytic pressure sensitivities. by w, r is distance along the streamline, and sðxÞ is the slowness
As with most previous work, we will calibrate against changes in defined by the reciprocal of the interstitial velocity,
properties over the interval of the time-lapse survey, instead of 1
the properties themselves. This minimizes systematic biases intro- sðxÞ ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
j~
ðxÞj
duced by lack of calibration of the static model. In addition, we
2 2017 SPE Journal


Pi dtðSw ; s; wÞ dsðwÞ dFw
Pw Producer ¼ : . . . . . . . . . . . . . . . . . . ð11Þ
d/ðxÞ d/ðxÞ dSw
i
w After water breakthrough on a streamline, these arrival times are
evaluated at the total time of flight, sðwÞ, for the streamline.
Appendix A provides the derivation of the semianalytic
Injector expressions for the sensitivity of water saturation with respect
i
ΔP to variations in permeability. As noted in Appendix A, the water-
Streamline = 1D Grids w
Pi Pw saturation sensitivity at location s at a given time t can be calcu-
lated by
i w
dSw ðs; tÞ t Sw ðs; tÞ Sw ðs; t DtÞ ds
¼ ; . . . . . . . . . ð12Þ
Fig. 2—A streamline between well pairs connecting gridblocks. dk s Dt dk
where the last partial derivative of travel time with respect to per-
The parameter w might represent the takeoff angle of the meability can be obtained from Eq. 8. In the derivation, it was
streamline from a well. Following Vasco et al. (2004) and by use assumed that the streamline trajectories do not change over time.
of Darcy’s law, we write the slowness in terms of the porosity / The large time intervals typically between seismic surveys may
and permeability k: require that we break up the period into several increments during
which the streamlines do not shift significantly. Also, most often
/ we are calibrating against the changes in saturation, rather than
sðxÞ ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ the saturation itself. The generalizations of the sensitivity calcula-
krt kjrPj
tion to account for these effects are also discussed in Appendix A.
where krt is the total relative mobility (krt ¼ kro þ krw þ krg ) and
jrPj is the pressure gradient along the streamline. Because slow- Pressure-Data Integration. Pressure-data integration is per-
ness is a composite quantity involving reservoir properties, its first- formed by mapping the spatial distribution of pressure in the res-
order variation, assuming a fixed pressure gradient, will be given by ervoir model into a spatial distribution of pressure along
@s @s streamlines. Specifically, for a particular gridblock i in the reser-
ds ¼ dk þ d/: . . . . . . . . . . . . . . . . . . . . . . . . ð4Þ voir model with total fluid pressure Pi , the pressure drop DPjiw
@k @/ along the streamline passing through the gridblock to well w, with
From Eq. 3, the partial derivatives are bottomhole pressure (BHP) Pwi (Fig. 2), is given by
@s / s Pi ¼ Pw þ DPjiw : . . . . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
¼ ; . . . . . . . . . . . . . . . . . . . . . ð5Þ
@k krt ðkÞ2 jrPj k
This uses the (known) bottomhole flowing pressure at the time
@s 1 s at which the spatial distribution of pressure was obtained from the
¼ : . . . . . . . . . . . . . . . . . . . . . . . . ð6Þ
@/ krt kjrPj / time-lapse-seismic observations. If distributed time-lapse pressure
data and BHP are available, we can compute the pressure drop
The approximation in Eqs. 5 and 6 is that the local perturba- from Eq. 13 and use it as our observed data:
tions in permeability or porosity generate negligible pressure
changes. This approximation implies that, to leading order, the DPjiw;obs ¼ Pi;obs Pw;obs : . . . . . . . . . . . . . . . . . . . . .ð14Þ
streamlines do not shift because of the small perturbations (Vasco
and Datta-Gupta 1999, 2016). We can relate a change in time of Now, the misfit between the simulation response and an obser-
flight ds to the change in slowness by integrating along each vation can be written as
streamline trajectory as
ð ð ddi ¼ DPjiw;obs DPjiw;cal
@s @s
dsðwÞ ¼ dsdr ¼ dk þ d/ dr: . . . . . . . . . .ð7Þ ¼ ðPi;obs Pw;obs Þ ðPi;cal Pw;cal Þ
w w @k @/
¼ ðPi;obs Pi;cal Þ ðPw;obs Pw;cal Þ; ð15Þ
Thus, the tracer-travel-time sensitivity along a single stream-
line, w, with respect to permeability and porosity at location x, where the first term is the pressure difference at location i and the
follows Eq. 7 by integrating from the inlet to the outlet of the second term is the BHP difference at well w.
streamline within the gridblock:
ð Pressure-Sensitivity Calculation. The total pressure drop along
ds s Ds
¼ dr ¼ ; . . . . . . . . . . . . . . . . . . . . . ð8Þ a streamline can be expressed as the sum
dk w k k
ð X
ds s Ds DP ¼ w
DPðxÞ: . . . . . . . . . . . . . . . . . . . . . . . . . .ð16Þ
¼ dr ¼ : . . . . . . . . . . . . . . . . . . . . . . . ð9Þ
d/ w / /
This can be computed by simply adding up the pressure drops
Here Ds is the time of flight across the gridblock at location x. across the gridblocks intersected by the streamline, as shown in
To relate the time of flight sensitivity to the travel-time sensitiv- Fig. 2. Furthermore, by use of Darcy’s law, we can express the
ity of the water saturation, we use the Buckley-Leverett equation local pressure drop along a streamline as
(Buckley and Leverett 1942) written by use of the streamline time
qðxÞDL
of flight as the spatial coordinate (Datta-Gupta and King 2007). As DPðxÞ ¼ ; . . . . . . . . . . . . . . . . . . . . . ð17Þ
shown in Appendix A, the derivation of the travel-time sensitivity AðxÞkrt ðxÞkðxÞ
relates the travel time of a particular fluid saturation, tðSw ; s; wÞ, to
where AðxÞ is the cross-sectional area, qðxÞ is the flow rate along
the time of flight s: We can now compute the sensitivity of the satu-
a streamline, krt ðxÞ is the total relative mobility, and DL is the arc
ration arrival time by use of that of the streamline time of flight:
length of the streamline increment within the gridblock. As is evi-
dent in Eq. 17, the pressure drop is a composite quantity involving
dtðSw ; s; wÞ dsðwÞ dFw
¼ ; . . . . . . . . . . . . . . . . . . . ð10Þ reservoir and fluid properties. We assume that the streamline tra-
dkðxÞ dkðxÞ dSw jectories do not change because of small perturbations in
2017 SPE Journal 3

Table 1—Rock and fluid properties. Pressure/volume/temperature

values are at the reference pressure of 5,864 psi. FVF 5 formation
volume factor. Table 2—Simulation specifications of five-spot synthetic case.
permeability. We can now relate the change in local pressure drop

In Eq. 22, dd is the misfit between the observed data and the val-
to a small change in permeability as
ues estimated from the reservoir model, the matrix S contains the
@DP sensitivities of model responses with respect to reservoir parame-
dDP ¼ dk; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð18Þ ters, dm is the vector of reservoir-model-parameter updates, and L is
@k
a second spatial-difference-operator matrix. The first term ensures
where the partial derivative is the differences between the observed data and calculated model
predictions are minimized. The second term, called a norm penalty,
@DP qDL DP penalizes deviations from the prior model. This prevents exces-
¼ : . . . . . . . . . . . . . . . . . . . . . ð19Þ
@k Akrt k2 k sively large changes during any single iteration, and helps preserve
geologic realism by preserving major features of the prior model.
The pressure drop to a location i, DPjiw , will be given by Finally, the third term, referred to as a roughness penalty, constrains
integration along the streamline trajectory passing through i to the model changes to be spatially smooth, in line with the resolution
well w: provided by the data. The weights b1 and b2 determine the relative
ð ð strengths of the norm and smoothness constraints; guidelines exist
@DP
dDPjiw ¼ dDPdr ¼ dkdr; . . . . . . . . . . . . . ð20Þ in the literature for their selection. An iterative sparse matrix solver,
w w @k the least squares with QR factorization algorithm (Paige and Saun-
and the pressure-drop sensitivity for a particular gridblock at loca- ders 1982), is used for solving the augmented linear system effi-
tion x follows from Eq. 20, ciently. We perform the data integration in a sequential manner
starting with pressure data first followed by the saturation changes,
ð outlet and then the well-production information. This sequential frame-
@DPjiw DP DP
¼ dr ¼ : . . . . . . . . . . . . . ð21Þ work is analogous to the structured approach used in the industry
@k location k k
(Williams et al. 1998; Cheng et al. 2008) and accounts for the differ-
The sensitivity calculations for both the pressure and the satu- ent length scales associated with different data types. The approach
ration have been validated in Appendix B. also facilitates convergence of the inversion algorithm. The norm
constraint in Eq. 22 maintains consistency between different steps
in the sequential process by limiting the changes, thus preserving
Objective-Function Minimization. After computing the sensi-
the calibration results from the previous step.
tivities of the data with respect to the gridblock permeabilities,
we perform the data integration through the minimization of a
penalized misfit function (Vasco and Datta-Gupta 1999; He et al. History-Matching Applications
2002; Vasco 2004): Five-Spot Synthetic Case. Our first example of seismic history
matching is a synthetic 2D model. The model is composed of a
oðdmÞ ¼ dd Sdm þ b1 dm þ b2 Ldm: . . . . . . . . . . . . ð22Þ 50 50 grid of cells and a five-spot-well configuration, with one
central injector and four producers at the corners. The wells are
1.0
constrained by the historical (constant) reservoir-flow rates. The
rock and fluid properties and the simulation specifications are
0.9 Krw summarized in Tables 1 and 2, respectively. The relative perme-
Kro ability curves are shown in Fig. 3. The reference permeability
0.8
model is generated by use of sequential Gaussian simulation with
Relative Permeability
0.7 well-permeability values as conditioning data (Fig. 4a). The prior

permeability model, shown in Fig. 4b, is also generated by se-
0.6 quential Gaussian simulation but with different geostatistical pa-
0.5 rameters from the reference model. The reference data are
generated from the model by use of a commercial reservoir simu-
0.4 lator. These reference data include well-production values and
0.3
maps of the estimated saturation and pressure changes that
occurred between 260 and 1,560 days from the start of production
0.2 (Figs. 5a and 5b). The production data are available as water cut
and BHP at each of the wells. Our work flow involves an initial
0.1
time-lapse-pressure match, followed by a match of the saturation
0.0 changes, by use of the pressure and saturation-sensitivity calcula-
0.0 0.2 0.4 0.6 0.8 1.0 tions discussed previously. The water cut is matched by use of a
Normalized Water Saturation generalized travel-time-inversion (GTTI) method (He et al. 2002).
The reasoning behind our sequential approach is that pressure
Fig. 3—Relative permeability curves. data are well-suited to capture the large-scale spatial variations in
4 2017 SPE Journal

Permeability (md)
1000
100
(a) (b) (c)
10
(d) (e)
Fig. 4—Permeability fields for the inverted five-spot synthetic test. (a) Reference model, (b) initial model, (c) after-pressure-
change-data integration, (d) after-water-saturation-change-data integration, and (e) final model after generalized travel-time-
production-data integration.
Pressure (psi) Water Saturation

30 0.9
20 0.8
10
0 0.6
–10
0.4
–20
–30 0.2
–40
0.0
(a) (b)
Fig. 5—Time-lapse reference data between 260 and 1,560 days: (a) pressure change and (b) water-saturation change.
reservoir properties, whereas the saturation and production data The misfit reduction caused by the integration of the pressure
capture small-scale or local variations. This sequence is analogous changes, shown in Fig. 6a, display a notable reduction after only a
to the structured approach to history matching widely practiced in few iterations. The updated permeability field, shown in Fig. 4c,
the industry (Williams et al. 1998; Cheng et al. 2008). appears to identify a low-permeability barrier in the right lower
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
Normalized Objective
Function (saturation)
Function (pressure)
0.7 0.7 0.7

Function (GTT)
0.6 0.6 0.6

0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
0 5 10 15 20 25 30 0 5 10 15 20 0 5 10 15 20
Iteration No. Iteration No. Iteration No.
(a) (b) (c)
Fig. 6—Objective function of inverted five-spot for synthetic 2D model. (a) Pressure-change-data integration, (b) saturation-
change-data integration, and (c) generalized travel-time-production-data integration.
2017 SPE Journal 5

Pressure (psi)
30
20
10
0
(a) (b) (c)
–10
–20
–30
–40
(d) (e)
Fig. 7—Time-lapse pressure changes. (a) Reference model, (b) initial model, (c) after-pressure-change-data integration, (d) after-
water-saturation-change-data integration, and (e) final model after generalized travel-time-production-data integration.
corner of the model. The fit to the time-lapse-pressure changes is model, as shown in Figs. 7 and 8. The permeability-model updates
improved significantly, as shown in Fig. 7. Next, saturation during the inversion steps are compared in Fig. 4. These results
changes are integrated into the model. The reduction in the misfit confirm that the consistent integration of both time-lapse-pressure
to the water-saturation changes is shown in Fig. 6b. The time- and -saturation changes, together with production data, can con-
lapse-saturation change from this stage is shown in Fig. 8d and strain the history-matched solution.
reveals further improvement over the pressure-data integration as
shown in Fig. 8c. Finally, the GTTI is applied to match the well by The Norne Field Case. The Norne Field was discovered in De-
well-water-cut data. The objective function reduction is shown in cember 1991, development drilling began in August 1996, and oil
Fig. 6c and the production-data history-matching result is shown production started in November 1997. The field has high-quality
in Fig. 9. The predictions derived from the final model match the time-lapse-seismic data, production data, and well logs in addition
time-lapse-pressure and -saturation changes in the reference to the reservoir model. The geological model consists of five
Water Saturation
0.9
0.8
(a) (b) (c) 0.6
0.4
0.2
0.0
(d) (e)
Fig. 8—Time-lapse water-saturation-change comparisons. (a) Reference model, (b) initial model, (c) after-pressure-change-data
integration, (d) after-water-saturation-change-data integration, and (e) final model after generalized travel-time-production-data
integration.
6 2017 SPE Journal

1 P1 1 P2
0.9 0.9
0.8 0.8
Well Water Cut
Well Water Cut

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 500 1,000 1,500 2,000 2,500 0 500 1,000 1,500 2,000 2,500
Time (days) Time (days)
P3 P4
1 1
0.9 0.9
0.8 0.8
Well Water Cut
0.7
Well Water Cut

0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 500 1,000 1,500 2,000 2,500 0 500 1,000 1,500 2,000 2,500
Initial model Reference model After-pressure-change-data integration
After-water-saturation-change-data integration Final model
Fig. 9—Water-cut history-matching results. Reference model responses are plotted as dot points, initial model responses are blue
lines, model responses after accounting for the pressure changes are purple lines, model responses after accounting for the satu-
ration changes are green lines, and final updated model responses are red lines.
reservoir zones: Garn, Not, Ile, Tofte, and Tilje. Oil is mainly 2003–2004, 2004–2006). The production data were in the form of
found in the Ile and Tofte formations, and gas in the Garn forma- water, oil, and gas rates and BHPs at the producers. The seismic
tion. The sandstones are buried at the depth of 2500–2700 m. The data, which were processed externally, are available for the model
porosity is in the range of 25–30%, whereas permeability varies calibration as near, mid, far, and full offset stacked 3D volumes of
from 20 to 2500 md (Steffensen and Karstadt 1996; Osdal et al. reflection amplitudes. The time-lapse differences were calculated
2006). We demonstrate the feasibility of our approach by per- from these volumes. The interpreted horizons from the top to the
forming a full-field history matching of the Norne Field. The res- base of the reservoir were used for identification of the water/oil
ervoir model consists of 44,431 active cells and it contains 36 contact. Additional details on the entire data set may be found in
wells (nine injectors and 27 producers), as shown in Fig. 10. A Rwechungura et al. (2012).
detailed reservoir model was provided as an initial model for the PEM. Unlike the synthetic test, where interpreted saturation
inversion. For the history match, we consider all available data and pressure changes are considered in the inversion, this is a
from 1997 to 2006. These data include production and injection much-more-realistic case in which seismic responses are pro-
information from 1997 to the end of 2006, and multiple sets vided. Therefore, we need to consider an additional step and intro-
of time-lapse-seismic data for the same period (2001–2003, duce PEM to completely specify sensitivities to changes in
reservoir properties. A PEM is a set of equations relating reservoir
properties (pore volume, pore-fluid saturations, reservoir pres-
sures, and rock composition) to seismic-rock-elastic parameters
Injectors (P-wave and S-wave velocities, Vp and Vs, respectively). Elastic-
rock properties are functions of temperature, compaction, fluid
Producers
saturation, and reservoir pressure, although we may neglect the
effects of temperature in the Norne Field.
Permeability (md) The Gassmann equation (Gassmann 1951) and the Hertz-
Mindlin contact theory (Mindlin 1949) are the basis for estimates
1000 of the changes in elastic parameter caused by fluid-saturation and
reservoir-pressure variations, respectively. The Hertz-Mindlin
100 model (Mindlin 1949) is used to compute changes in seismic-
rock-elastic parameters caused by pressure changes (Mavko et al.
10 1998). The effective bulk modulus of a pack of randomly distrib-
uted identical spheres in the absence of fluid is given by
1
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
KHM ¼ Kma n Peff =ðPext Pint Þ; . . . . . . . . . . . . . . . . ð23Þ
Fig. 10—Structure of the Norne Field with a gridblock configu- where KHM is the bulk modulus at critical porosity (Dadashpour
ration of 46 3 112322 and average grid-cell dimensions of et al. 2009, 2010). Here, Peff is the effective pressure, the differ-
121 3 12939.5 m, showing the reservoir permeability. ence between the lithostatic pressure Pext and the hydrostatic
2017 SPE Journal 7

Table 3—Input parameters for the PEM in Norne Field.
pressure P (Christensen and Wang 1985), Kma is the bulk modulus listed in Table 3. The density of the saturated rock is given by the
of the matrix, and n is the coordination number. For the Norne weighted average of the component densities:
Field application, the initial pressure Pint is set to 270 bar and the
lithostatic pressure depends on the true vertical depth (TVD) as qsat ¼ ð1 /Þqma þ /ðSo qo þ Sw qw þ Sg qg Þ; . . . . . . . ð27Þ
Pext ¼ 0:0981ð9 105 TVD þ 1:7252Þ TVD: . . . ð24Þ where qo ; qw ; qg , and qma are the densities of oil, water, gas, and
the rock matrix, respectively. With the saturated-rock bulk modu-
In the Hertz-Mindlin (Mindlin 1949) theory, the
ffi velocity
pffiffiffiffiffiffiffi lus and shear modulus and density, we can compute the compres-
varies as Peff raised to the one-sixth power (i.e., 6 Peff ), although sional (P-wave) velocity for an isotropic, layered, elastic medium
measurements on some laboratory samples suggest other values. (Kennett 1983) as
For this field application, we use n ¼ 5, a value taken from the lit- vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
erature (Dadashpour et al. 2009). The Gassmann equation (Gass- uK þ 4 G
mann 1951) expresses the bulk modulus of a fluid-saturated rock t sat 3 fr
Vp ¼ : . . . . . . . . . . . . . . . . . . . . . . . . ð28Þ
from three terms: the bulk modulus of the mineral matrix KHM , qsat
the bulk modulus of the porous-rock frame Kfr , and the bulk mod-
ulus of the pore-filling fluids Kf ; as given by the following for- In Eq. 28, Gfr is the frame shear modulus that is not affected
mula (Dadashpour et al. 2009): by fluid saturations. The acoustic (P-wave) impedance can be
computed as
ðKHM Kfr Þ2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Ksat ¼ Kfr þ ; . . . . . . . ð25Þ 4
KHM Kfr Zp ¼ qsat Vp ¼ qsat Ksat þ Gfr : . . . . . . . . . . . . . ð29Þ
KHM 1/þ/ 3
Kf KHM
where / is the effective porosity of the medium and the bulk mod- With the previously discussed PEM for the Norne Field, we
ulus of the pore fluid, a mixture of oil, water, and gas, and is esti- examine the sensitivity of the acoustic impedance with pressure
mated by Wood’s law given as (Reuss 1929) and saturation changes. For a simple two-phase oil/watersystem,
Fig. 11a shows the increase of acoustic impedance with increas-
1 So Sw Sg ing water saturation for a fixed pressure. Fig. 11b shows the
¼ þ þ ; . . . . . . . . . . . . . . . . . . . . . . . ð26Þ decrease of acoustic impedance with increasing pressure at a
Kf Ko Kw Kg
fixed saturation.
where So , Sw , and Sg are the oil, water, and gas saturations, Seismic-Data Processing. The first step in our data-calibra-
respectively, and Ko , Kw , and Kg are their respective bulk moduli. tion procedure is to invert the seismic volumes of reflection
For the Norne Field application, the rock-elastic properties are amplitudes to changes in acoustic (P-wave) impedance. By use of
8 2017 SPE Journal

×106 ×106
8 7.7
Acoustic Impedance (Pa·s/m)
Acoustic Impedance (Pa·s/m)

7.9 7.6
7.8
7.5
7.7
7.4
7.6
7.5 7.3
7.4 7.2
0 0.2 0.4 0.6 0.8 1 200 250 300 350 400 450 500
Sw Pressure (bar)
(a) (b)
Fig. 11—Acoustic-impedance-calculation sensitivity by PEM model in oil/water two-phase system: (a) with respect to water-saturation
changes under a fixed pressure (270 bar) and (b) with respect to pressure changes under a fixed saturation value (Sw 5 0.5).
commercial software, we conduct seismic-data processing that responses: Positive change reflects the aquifer encroachment,
consists of time-to-depth-data conversion; well-log-quality check whereas negative changes below the water/oil contact correspond
and acoustic-impedance-log calculation; genetic inversion for to pressure increases in Fig. 14. The cross validation of the pre-
generating an acoustic-impedance map from the seismic-ampli- dicted acoustic-impedance values from the genetic inversion and
tude data; and resample the acoustic-impedance map into the sim- the training-acoustic-impedance logs show overall agreement
ulation grid. As for the post-stack seismic sections, we decided (Fig. 13).
to use the near-offset stacking data set because the acoustic Global-to-Local Hierarchical-History-Matching Work Flow.
(P-wave)-impedance changes are more evident in the small-angle The reservoir model provided by the operator was already cali-
reflection waves in amplitude variation with offset analysis (Aki brated to match the reservoir energy (regional pressure and pore
and Richards 1980). volume). They adjusted fault transmissibilities, regional relative
The time-to-depth conversion is achieved by use of a layered permeability parameters, large-scale absolute permeability, and
velocity model. The layering of the model is consistent with the porosity heterogeneity by use of regional and constant multipliers.
depth-converted seismic-amplitude data, as shown in Fig. 12. Our objective was to update the permeability model only at those
Well-log data, especially for the density and the sonic logs, are locations and scales required to improve large-scale transport
used to compute acoustic impedances at the well locations, as within the reservoir induced by the production, water injection,
shown in Fig. 13. We adapt a genetic algorithm for the inversion and aquifer support, but to otherwise leave the prior unchanged.
of the seismic amplitudes, as proposed by Veeken et al. (2009). We apply a hierarchical-history-matching work flow that consists
The method only requires the post-stack seismic cube as input. of two stages (Yin et al. 2011): a global update and a local update.
The acoustic-impedance logs at the wells are used as training data For the global update, the geological model is first parameterized
for a neural network to construct the nonlinear operator that trans- by use of a grid-connectivity transform (GCT) (Bhark et al.
forms the seismic traces into the equivalent acoustic-impedance 2011). It is a linear transformation where the heterogeneity is
response. The operator weights are adjusted to minimize the dif- updated in a transform domain that is characterized by the spectral
ference between the predicted acoustic-impedance response and modes of the reservoir-model grid. This change of basis from the
the acoustic-impedance logs at wells. One of the challenges in spatial to spectral domain is performed by multiplication of the
calibrating the reservoir model by use of seismic data is the differ- heterogeneity field with the transformation-basis functions, which
ence in the vertical-grid-cell resolution. In this case, the vertical- are constructed from the eigenvectors of a grid Laplacian. The
cell resolution for the simulation model is approximately 12.5 m, discrete, grid-based, spatial field is mapped into the transform do-
which is approximately the same as the seismic vertical resolu- main by use of an orthogonal transformation:
tion. The neural-network operator is applied to the seismic cube
to generate acoustic-impedance-map estimates in a simulation- v ¼ UT u () u ¼ Uv; . . . . . . . . . . . . . . . . . . . . . . ð30Þ
grid resolution, as shown in Fig. 14. The acoustic-impedance
changes are thought to correspond to the motion of the water/oil where u represents the spatial field and has dimension N, where N
contact between surveys. They are consistent with the PEM is the discretization of the property field, such as the number of
Permeability (md) 2,000 4,000 6,000 8,000 10,000 12,000 –1,600 –800 0 800 1,600 2,400
–2,880 –2,800 –2,720 –2,640
1000
100
10
(a) (b) (c)
Fig. 12—Time-to-depth data conversion. (a) Reservoir model intersected by the depth-domain seismic-amplitude inline and cross-
line slices, (b) the inline slice with reservoir-model-layer horizons, and (c) the crossline slice with reservoir-model-layer horizons.
2017 SPE Journal 9

6608/10-B-2-H (SSTVD) 6608/10-C-4-AH (SSTVD) 6608/10-D-4-H (SSTVD) 6608/10-E-3-H (SSTVD) 6608/10-F-1-H (SSTVD)
2200
2000 2200
2200
2300
2300
2100 2300
2300 2400
2400
2400
2200 2400 2500
2500
2500 2600
2500
2300
2600
2600 2700
2600
2400
2700
2700 2800
2700
2500
2800
2800 2900
2800
2600
2900
2900 3000
Fig. 13—The acoustic-impedance-log comparisons. The calculated acoustic-impedance log (black) and the response extracted
from the acoustic-impedance cube as a result of genetic inversion (red).
gridblocks. The column vector v is composed of transform coeffi- The GCT parameterization of a multiplier field is shown in
cients of length M, and U is an (N M) matrix containing M col- Fig. 16. In this case, we used a total of 420 coefficients (20 basis
umns, where each column defines a discrete-basis function of vectors per layer and 21 active layers) to represent the geologic
length N. For model calibration, a spatial-multiplier field has been model consisting of 44,431 active cells. For the Pareto-based mul-
posed in the multiplicative formulation as follows: tiobjective minimization, we define three objective functions:
gridblock acoustic-impedance-change (AIDIFF) misfit ðAI ¼ Zp Þ;
u ¼ u0 Uv; . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð31Þ cumulative field-water-production (FWPT) misfit; and cumulative
field-gas-production (FGPT) misfit expressed as
where u0 is the prior property field, also called the initial model; rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Uv defines the multiplier field in the spatial domain; and signi- X3 XNcell
fies elementwise multiplication (Schur product). This honors the Objective 1 ¼ time¼1 i¼1
obs
ðdAIi;time dAIi;time cal
Þ2 ;
prior permeability heterogeneity in the model updates.
In addition, we use a Pareto-based multiobjective history- ð32Þ
matching work flow proposed by Park et al. (2013) to update the rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
XNwell XNtime
cal 2
GCT coefficients. This approach is well-suited for minimizing the Objective 2 ¼ j¼1 i¼1
ðFWPTi;obs j FWPTi; j Þ ;
multiple, and potentially conflicting, objectives involved in
matching both seismic data and production data. For the local ð33Þ
update, the gridblock-permeability changes are introduced by rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
XNwell XNtime
means of the streamline-based inversion algorithm, introduced in Objective 3 ¼ ðFGPTi;obs cal 2
j¼1 i¼1 j FGPTi; j Þ ;
the current study. The time-lapse acoustic-impedance changes
and well-by-well water-cut-production data are integrated and the ð34Þ
fine-scale permeability variations between well locations are
refined. The diagram of the work flow is shown in Fig. 15. where “time” is the time-lapse period, Ncell is the total number of
For the global update, we first parameterize the permeability gridblocks, Nwell is the total number of history-matching wells,
field of the individual layers to preserve the vertical stratification. and Ntime is the total number of timesteps. Fig. 17 shows the
Acoustic Impedance Change Pressure Change Sw Change

(Pa·s/m) 4,000 5,000 6,000 7,000 8,000 9,000 (bar) 4,000 5,000 6,000 7,000 8,000 9,000 4,000 5,000 6,000 7,000 8,000 9,000
–2,500 –2,550 –2,600 –2,650 –2,700 –2,750 –2,800
–2,500 –2,550 –2,600 –2,650 –2,700 –2,750 –2,800
–2,500 –2,550 –2,600 –2,650 –2,700 –2,750 –2,800

–2,800 –2,750 –2,700 –2,650 –2,600 –2.550 –2,500
–2,800 –2,750 –2,700 –2,650 –2,600 –2.550 –2,500
–2,800 –2,750 –2,700 –2,650 –2,600 –2.550 –2,500
0.4
150000 60
125000 50 0.35
100000 40
0.3
75000 30
50000 20 0.25
25000 10 0.2
0 0
–25000 0.15
–10
–50000 0.1
–20
–75000
–30 0.05
–100000
–40
–125000 0
–150000 –50
4,000 5,000 6,000 7,000 8,000 9,000 4,000 5,000 6,000 7,000 8,000 9,000 4,000 5,000 6,000 7,000 8,000 9,000
Fig. 14—Acoustic-impedance changes from an inline slice between 2001 and 2003 surveys. The interpreted water/oil contacts are
superimposed (red line is at 2001 survey and black line is at 2003 survey). The water-saturation and pressure changes from the ini-
tial model are compared.
10 2017 SPE Journal

Global Update: Basis coefficients (v)
Initialize populations
(v)
Generate models Local Update: Grid permeability (k)

GCT parameterization
Streamline-based sensitivity
u = u0 . (Φv)
(Acoustic impedance change =
Water-cut GTTI)
Model = Prior x Multiplier Flow
simulation
Generation
Genetic operator
Pareto-based
crossover
mutation model ranking
Optimal solutions
Fig. 15—Global-to-local hierarchical-history-matching work flow.
Reservoir-permeability- Leading GCT-basis vector

multiplier field
Multiplier
9
8
7
6
5 = + ... + ... +
4
3 v1 v2 v10 v20
2
1
Fig. 16—Parameterization of the permeability-multiplier field as the weighted linear combination of leading GCT-basis vectors.
Initial Models
7
x 10
6.2 8 7 7
6 x 10 6.2 x 10 6.2 x 10
6
5 6 6
5.8
AIDIFF
5.8 5.8
AIDIFF
AIDIFF
5.6
FGPT
4
5.6 5.6
5.4
3
5.2 5.4 5.4
2 5.2 5.2
5
6
FG 4 2 1 5 5
1 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2 1 2 3 4 5 6
PT 0
FWPT x 10
6 FWPT x 106 FWPT x 106 FGPT x 108
Final Update Models

Prior model Population
7
x 10
6.2
x 108 x 107 x 107
6 6.2 6.2
6
5.8 5 6 6
AIDIFF
5.6 5.8 5.8

FGPT
AIDIFF
4
AIDIFF
5.4 5.6 5.6

3
5.2 5.4 5.4
5 2 5.2 5.2
6
FG 4 2 1
0 0.5 1 1.5 2
5
0 0.5 1 1.5 2
5
PT 2 0 1 1 2 3 4 5 6
FWPT x 10
6
FWPT x 106 FWPT x 106 FGPT x 108
Fig. 17—Multiobjective function comparisons between initial models and the final models in the global-step model calibration. The
left-most panel shows the full 3D view, whereas the three right panels are projections on pairs of axes.
2017 SPE Journal 11

1 analysis in the objective space, as shown in Fig. 17. For the local
update, we need the sensitivity of the acoustic impedance with
respect to gridblock permeability. This can be obtained by means
Normalized AIDIFF Misfit
0.95 of a chain rule:

ddZp @Zp dSw @Zp dSg @Zp dP
SZp ¼ ¼ þ þ ; . . . . . . ð35Þ
0.9 dk @Sw dk @Sg dk @P dk
where the partial derivatives of acoustic impedance,

0.85
@Zp =@Sw ; @Zp =@Sg ; and @Zp =@P, are computed by numerical per-
turbation from the current saturation and pressure gridblock val-
ues by use of Eq. 29, whereas water-saturation sensitivity @Sw =@k
is computed by use of Eq. 12. Notice that we have ignored poros-
0.8
ity dependence in Eq. 35 and assumed that the changes in porosity
caused by compaction are not significant. For the gas-saturation
sensitivity, @Sg =@k, we follow the derivation used for the water-
0.75 saturation sensitivity in Eq. 12. This assumption applies mainly
Prior Global Local
for gas/oil two-phase system near the top layers of the Norne
Field where the free-gas cap and oil rim are located. For the
Fig. 18—The objective-function misfit for AIDIFF data integra-
tion comparisons among the prior model, global-step-cali- pressure sensitivity, we use the pressure-drop sensitivity given
brated model, and the final updated model from the local step by Eq. 21.
calibration. As discussed previously and illustrated in the previous test
case, our history matching follows a sequential approach. The
approach begins with pressure effects on acoustic-impedance
results of the multiobjective function minimization in the global changes, which are are used to calibrate the model. Next, the satu-
step of the model calibration. The Pareto-based evolutionary algo- ration effects on the acoustic-impedance changes are integrated to
rithm is solved by use of a genetic algorithm that allows us to cap- update the model. Water-saturation sensitivity and gas-saturation
ture the uncertainty of the model parameters. It produces a suite sensitivity are inverted separately in the third step. The inversion
of optimal solutions from the diverse initial population around the performance for acoustic-impedance change is shown in Fig. 18
prior model in a multidimensional objective space. In the 2D pro- for one of the models. The majority of the reduction of the
jection spaces in Fig. 17, the Pareto fronts are clearly shown to AIDIFF data misfit was achieved in the global step of the model
depict the tradeoff between objectives. calibration. In the final step, the GTTI is applied to integrate the
Local Step Time-Lapse Acoustic-Impedance Change and water-cut data. Samples of six well-by-well water-cut-response
Production-Data Integration. After the global calibration step, plots out of 27 producers are shown in Fig. 19. The final updated
we select a few candidate models for the local updates by a cluster model for some selected layers is shown in Fig. 20. The
B-1BH B-2H
0.8 0.8
Water Cut
Water Cut
0.6 0.6 OBS

0.4 0.4 Initial
0.2 0.2 Final
0 0
0 500 1,000 1,500 2,000 2,500 3,000 3,500 0 500 1,000 1,500 2,000 2,500 3,000 3,500
D-1CH D-4AH
0.7 0.3
0.6 0.25 OBS
Water Cut
Water Cut
0.5 0.2
0.4 Initial
0.15
0.3
0.2 0.1 Final
0.1 0.05
0 0
0 500 1,000 1,500 2,000 2,500 3,000 3,500 0 500 1,000 1,500 2,000 2,500 3,000 3,500
E-3CH E-3AH
1 1
Water Cut
0.8 0.8 OBS

Water Cut
0.6 0.6
Initial
0.4 0.4
0.2 0.2 Final
0 0
0 500 1,000 1,500 2,000 2,500 3,000 3,500 0 500 1,000 1,500 2,000 2,500 3,000 3,500
Fig. 19—Water-cut-production-data history-matching comparisons between the initial model and the final updated model.
OBS 5 observation data.
12 2017 SPE Journal

L1 (Garn) L5 (Ile) L11 (Ile) L17 (Tofte) L21 (Tilje)

Prior Model Permeability (md)
1000
100
10
1
Global Updated Model
Permeability (md)
1000
100
10
Final Updated Model

Permeability (md)
1000
100
10
Model Change (Final-Prior) Permeability (md)

600
500
400
300
200
100
0
–100
–200
–300
–400
–500
–600
Fig. 20—Permeability-model-update-layer comparisons: the prior model (top), the global updated model, the final updated model,
and the model changes between the prior and the final models.
improvements in the water-cut responses can be related to the per- 1. We have proposed a methodology for streamline-based ana-
meability updates. For example, positive permeability changes at lytic approaches to compute parameter sensitivities that relate
the top of Layer 5 are related to Well D-4AH and permeability the time-lapse seismic data to reservoir properties while
changes at the bottom of Layers 5 and 11 are related to Wells B- accounting for both pressure and saturation effects by means of
2H, D-1CH, and B-1BH. appropriate rock-physics models.
The final model responses in terms of acoustic-impedance 2. Our numerical experiments validate the proposed sensitivity
changes are compared in Fig. 21. For the second time-lapse pe- calculations for the saturation and pressure drop by comparison
riod (2003–2004), a large negative time-lapse AIDIFF misfit (red with a numerical-perturbation method. However, unlike nu-
color) in the prior model in the top right part of reservoir (G-seg- merical perturbation, our proposed approach to sensitivity
ment) was corrected by the inversion. This resulted from an computations requires only a single forward simulation.
improved matching of the time-lapse-pressure change in the final 3. A synthetic example shows the importance of accounting for
updated model. Overall, the misfit associated with the time-lapse both saturation and pressure changes in the reservoir responses
acoustic-impedance change and the well-production-breakthrough to constrain the history-matching solutions.
times is notably reduced from the initial misfit calculated by use 4. The proposed approach accounts for the model-parameter
of the prior model. uncertainty by updating the ensemble of models, as demon-
strated in the Norne Field case.
Summary and Conclusions 5. The Norne Field application demonstrates the feasibility of the
structured work flow of time-lapse-seismic data and produc-
In this paper, we have presented an efficient history-matching
tion-data integration by use of seismic-data processing, petro-
approach that integrates 4D repeated seismic surveys with well-
elastic modeling, and the efficient streamline-based data
production data while accounting for both pressure and saturation
integration. The calibrated model shows the targeted global-
effects on the seismic response. Although our approach relies on
and fine-scale-model updates and improvement of the matches
streamline-based sensitivity calculation to relate seismic response
to the acoustic-impedance changes between time-lapse sur-
to the reservoir parameters, it can be applied with either stream-
veys. The well-by-well match to the water-cut-breakthrough
line simulators or conventional finite-difference simulators. For
times is also substantially improved.
finite-difference simulators, the streamline and time of flight can
be computed by means of post-processing of the velocity field
(Cheng et al. 2005). We have demonstrated the effectiveness of Nomenclature
our proposed approach through synthetic and field applications. Fw ¼ fractional flow of water
Some of the conclusions from this paper are summarized as Gfr ¼ shear modulus of the porous-rock frame
the following. k ¼ permeability
2017 SPE Journal 13

Layer 14: Time-Lapse 1 Layer 3: Time-Lapse 2 Layer 16: Time-Lapse 3
(Pa·s/m) (Pa·s/m) (Pa·s/m)

Acoustic-Impedance- 300000 300000 200000
Change-Observation 200000 200000 150000
Data 100000 100000 100000
0 0 50000
0
–100000 –100000
–50000
–200000 –200000
–100000
–300000 –300000
–150000

300000 300000 200000
Prior Model Acoustic- 200000 200000 150000
Impedance Change 100000 100000
100000
0 50000
0
0
–100000 –100000
–50000
–200000 –200000
–100000
–300000 –300000
–150000

300000 300000 200000
Global Model Acoustic- 200000 200000 150000
Impedance Change 100000 100000 100000
50000
0 0
0
–100000 –100000
–50000
–200000 –200000
–100000
–300000 –300000
–150000

300000 300000 200000
200000 200000 150000
Final Model Acoustic- 100000 100000 100000
Impedance Change 50000

0 0
0
–100000 –100000
–50000
–200000 –200000
–100000
–300000 –300000
–150000
Fig. 21—Time-lapse acoustic-impedance-change comparisons in selected layers among the observation data, the prior model
responses, the global updated, and the final updated model responses.
kHM ¼ bulk modulus of the Hertz-Mindlin (Mindlin 1949) / ¼ porosity

formula w ¼ streamline trajectory
Kf ¼ bulk modulus of the pore-filling fluids
Kfr ¼ bulk modulus of the porous-rock frame
Kma ¼ bulk modulus of the matrix Acknowledgments
Ksat ¼ bulk modulus of the fluid-saturated rock The authors acknowledge the sponsors of Model Calibration and
P ¼ hydrostatic pressure Efficient Reservoir Imaging at Texas A&M University. We would
Peff ¼ effective pressure also like to thank Statoil (operator of the Norne Field) and its
Pext ¼ lithostatic pressure license partners Eni and Petoro for the release of the Norne data.
s ¼ slowness Further, we acknowledge the Center for Integrated Operations at
Sw ¼ water saturation Norwegian University of Science and Technology for cooperation
¼ interstitial velocity
~ and coordination of the Norne cases. Finally, we would like to
Vp ¼ compressional (P-wave) velocity thank the US Department of Energy’s Basic Energy Sciences pro-
Zp ¼ acoustic (P-wave) impedance gram for the ongoing support of this research.
dm ¼ model-parameter change
DP ¼ pressure drop along streamlines
k ¼ mobility References
qsat ¼ density of the fluid-saturated rock Aki, K. and Richards, P. G. 1980. Quantitative Seismology. San Francisco:
s ¼ time of flight along streamlines Freeman.
14 2017 SPE Journal

Arenas, E., Kruijsdijk, C. V., and Oldenziel, T. 2001. Semi-Automatic Fanchi, J. R. 2001. Time-Lapse Seismic Monitoring in Reservoir Manage-
History Matching Using the Pilot Point Method Including Time-Lapse ment. The Leading Edge 20 (10): 1140–1147. https://doi.org/10.1190/
Seismic Data. Presented at the SPE Annual Technical Conference and 1.1487246.
Exhibition, New Orleans, 30 September–3 October. SPE-71634-MS. Feng, T. and Mannseth, T. 2010. Impact of Time-Lapse Seismic Data for
https://doi.org/10.2118/71634-MS. Permeability Estimation. Computat. Geosci. 14 (4): 705–719. https://
Batzle, M. and Wang, Z. 1992. Seismic Properties of Pore Fluids. Geophy- doi.org/10.1007/s10596-010-9182-6.
sics 57 (11): 1396–1408. https://doi.org/10.1190/1.1443207. Foster, D. 2007. The BP 4-D Story: Experience Over The Last 10 Years
Behrens, R., Condon, P., Haworth, W. et al. 2002. 4D Seismic Monitoring And Current Trends. Presented at International Petroleum Technology
of Water Influx at Bay Marchand: The Practical Use of 4D in an Conference, Dubai, 4–6 December. IPTC-11757-MS. https://doi.org/
Imperfect World. SPE Res Eval & Eng 5 (5): 410–420. SPE-79961- 10.2523/IPTC-11757-MS.
PA. https://doi.org/10.2118/79961-PA. Gassmann, F. 1951. Elastic Waves Through a Packing of Spheres. Geo-
Behrens, R., MacLeod, M., Tran, T. et al. 1998. Incorporating Seismic At- physics 16 (4): 673–685. https://doi.org/10.1190/1.1437718.
tribute Maps in 3D Reservoir Models. SPE Res Eval & Eng 1 (2): Gosselin, O., Aanonsen, S., Aavatsmark, I. et al. 2003. History Matching
122–126. SPE-36499-PA. https://doi.org/10.2118/36499-PA. Using Time-lapse Seismic (HUTS). Presented at the SPE Annual
Bhark, E. W., Jafarpour, B., and Datta-Gupta, A. 2011. A Generalized Technical Conference and Exhibition, Denver, 5–8 October. SPE-
Grid Connectivity–Based Parameterization for Subsurface Flow Model 84464-MS. https://doi.org/10.2118/84464-MS.
Calibration. Water Resour. Res. 47 (6): W06517. https://doi.org/ Gosselin, O., van den Berg, S., and Cominelli, A. 2001. Integrated His-
10.1029/2010WR009982. tory-Matching of Production and 4D Seismic Data. Presented at the
Buckley, S. E. and Leverett, M. C. 1942. Mechanism of Fluid Displace- SPE Annual Technical Conference and Exhibition, New Orleans, 30
ment in Sands. In Transactions of the Society of Petroleum Engineers, September–3 October. SPE-71599-MS. https://doi.org/10.2118/71599-
Vol. 146, Part 1, 107–116. SPE-942107-G. Richardson, Texas: Society MS.
of Petroleum Engineers. https://doi.org/10.2118/942107-G. He, Z., Yoon, S., and Datta-Gupta, A. 2002. Streamline-Based Production
Cheng, H., Dehghani, K., and Billiter, T. 2008. A Structured Approach for Data Integration With Gravity and Changing Field Conditions. SPE J.
Probabilistic-Assisted History Matching Using Evolutionary Algo- 7 (4): 423–436. SPE-81208-PA. https://doi.org/10.2118/81208-PA.
rithms: Tengiz Field Applications. Presented at the SPE Annual Tech- Huang, X., Meister, L., and Workman, R. 1997. Reservoir Characteriza-
nical Conference and Exhibition, Denver, 21–24 September. SPE- tion by Integration of Time-lapse Seismic and Production Data. Pre-
116212-MS. https://doi.org/10.2118/116212-MS. sented at the SPE Annual Technical Conference and Exhibition, San
Cheng, H., Kharghoria, A., He, Z. et al. 2005. Fast History Matching of Fi- Antonio, Texas, 5–8 October. SPE-38695-MS. https://doi.org/10.2118/
nite-Difference Models Using Streamline-Based Sensitivities. SPE Res 38695-MS.
Eval & Eng 8 (5): 426–436. SPE-89447-PA. https://doi.org/10.2118/ Kennett, B. 1983. Seismic Wave Propagation in Stratified Media. Cam-
89447-PA. bridge, UK: Cambridge University Press.
Christensen, N. and Wang, H. 1985. The Influence of Pore Pressure and Kjelstadli, R. M., Lane, H. S., Johnson, D. T. et al. 2005. Quantitative His-
Confining Pressure on Dynamic Elastic Properties of Berea Sandstone. tory Match of 4D Seismic Response and Production Data. Presented at
Geophysics 50 (2): 207–213. https://doi.org/10.1190/1.1441910. Offshore Europe, Aberdeen, 6–9 September. SPE-96317-MS. https://
Clifford, P., Robert, T., Parr, R. et al. 2003. Integration of 4D Seismic doi.org/10.2118/96317-MS.
Data into the Management of Oil Reservoirs with Horizontal Wells Landa, J. and Horne, R. 1997. A Procedure to Integrate Well Test Data,
between Fluid Contacts. Presented at Offshore Europe, Aberdeen, 2–5 Reservoir Performance History and 4-D Seismic Information into a
September. SPE-83956-MS. https://doi.org/10.2118/83956-MS. Reservoir Description. Presented at the SPE Annual Technical Confer-
Dadashpour, M., Ciaurri, D. E., Mukerji, T. et al. 2010. A Derivative-Free ence and Exhibition, San Antonio, Texas, 5–8 October. SPE-38653-
Approach for the Estimation of Porosity and Permeability Using MS. https://doi.org/10.2118/38653-MS.
Time-Lapse Seismic and Production Data. J. Geophys. Eng. 7 (4): Landrø, M., Digranes, P., and Strønen, L. 2001. Mapping Reservoir Pres-
351–368. https://doi.org/10.1088/1742-2132/7/4/002. sure and Saturation Changes Using Seismic Methods: Possibilities and
Dadashpour, M., Echeverrı́a-Ciaurri, D., Kleppe, J. et al. 2009. Porosity Limitations. First Break 19 (12): 671–684. https://doi.org/10.1046/
and Permeability Estimation by Integration of Production and Time- j.1365-2397.2001.00226.x.
Lapse Near and Far Offset Seismic Data. J. Geophys. Eng. 6 (4): Landrø, M., Solheim, O.A., Hilde, E. et al. 1999. The Gullfaks 4D Seismic
325–344. https://doi.org/10.1088/1742-2132/6/4/001. Study. Petrol. Geosci. 5 (3): 213–226. https://doi.org/10.1144/petgeo.
Dadashpour, M., Landrø, M., and Kleppe, J. 2008. Nonlinear Inversion for 5.3.213.
Estimating Reservoir Parameters from Time-Lapse Seismic Data. J. Landrø, M., Veire, H. H., Duffaut, K. et al. 2003. Discrimination Between
Geophys. Eng. 5 (1): 54–66. https://doi.org/10.1088/1742-2132/5/1/ Pressure and Fluid Saturation Changes from Marine Multicomponent
006. Time-Lapse Seismic Data. Geophysics 68 (5): 1592–1599. https://
Datta-Gupta, A. and King, M. J. 2007. Streamline Simulation: Theory and doi.org/10.1190/1.1620633.
Practice, Vol. 11. Richardson, Texas: Society of Petroleum Engineers. Lumley, D. E. 2001. Time-lapse Seismic Reservoir Monitoring. Geophy-
Deutsch, C. V. and Journel, A. G. 1992. GSLIB: Geostatistical Software sics 66 (1): 50–53. https://doi.org/10.1190/1.1444921.
Library and User’s Guide. New York City: Oxford University Press. Lumley, D. and Behrens, R. 1998. Practical Issues of 4D Seismic Reser-
Dong, Y. and Oliver, D. 2005. Quantitative Use of 4D Seismic Data for voir Monitoring: What an Engineer Needs to Know. SPE Res Eval &
Reservoir Description. SPE J. 10 (1): 91–99. SPE-84571-PA. https:// Eng 1 (6): 528–538. SPE-53004-PA. https://doi.org/10.2118/53004-
doi.org/10.2118/84571-PA. PA.
Doyen, P. M., Psaila, D. E., and Jans, D. 1997. Reconciling Data at Seis- Lumley, D. E., Behrens, R. A., and Wang, Z. 1997. Assessing the Techni-
mic and Well Log Scales in 3-D Earth Modelling. Presented at the cal Risk of a 4-D Seismic Project. The Leading Edge 16 (9):
SPE Annual Technical Conference and Exhibition, San Antonio, 1287–1292. https://doi.org/10.1190/1.1437784.
Texas, 5–8 October. SPE-38698-MS. https://doi.org/10.2118/38698- Mavko, G., Mukerji, T., and Dvorkin, J. 1998. The Rock Physics Hand-
MS. book: Tools for Seismic Analysis in Porous Media. Cambridge, UK:
Fahimuddin, A., Aanonsen, S., and Skjervheim, J.-A. 2010. 4D Seismic Cambridge University Press.
History Matching of a Real Field Case With EnKF: Use of Local Anal- Mindlin, R. D. 1949. Compliance of Elastic Bodies in Contact. J. Appl.
ysis for Model Updating. Presented at the SPE Annual Technical Con- Mech. 16: 259–268.
ference and Exhibition, Florence, Italy, 19–22 September. SPE- Osdal, B., Husby, O., Aronsen, H. A. et al. 2006. Mapping the Fluid Front
134894-MS. https://doi.org/10.2118/134894-MS. and Pressure Buildup Using 4D Data on Norne Field. The Leading
Falcone, G., Gosselin, O., Maire, F. et al. 2004. Petroelastic Modelling as Edge 25 (9): 1134–1141. https://doi.org/10.1190/1.2349818.
Key Element of 4D History Matching: A Field Example. Presented at Paige, C. C. and Saunders, M. A. 1982. LSQR: An Algorithm for Sparse
the SPE Annual Technical Conference and Exhibition, Houston, 26–29 Linear Equations and Sparse Least Squares. ACM Trans. Math. Soft-
September. SPE-90466-MS. https://doi.org/10.2118/90466-MS. ware 8 (1): 43–71. https://doi.org/10.1145/355984.355989.355989.
2017 SPE Journal 15

Park, H.-Y., Datta-Gupta, A., and King, M. 2013. Handling Conflicting Appendix A—Water-Saturation-Change-
Multiple Objectives Using Pareto-Based Evolutionary Algorithm for Sensitivity Calculations
History Matching of Reservoir Performance. Presented at the SPE Res- Saturation-Front-Arrival-Time Sensitivity. We have already
ervoir Simulation Symposium, The Woodlands, Texas, 18–20 Febru- developed expressions for sensitivity of the streamline time of
ary. SPE-163623-MS. https://doi.org/10.2118/163623-MS. flight to reservoir porosity and permeability. We can now relate
Reuss, A. 1929. Berechnung der Fließgrenze von Mischkristallen auf the time-of-flight sensitivity to the travel-time sensitivity of the
Grund der Plastizitätsbedingung für Einkristalle. ZAMM-Journal of water saturation. For two-phase flow, this sensitivity is used to
Applied Mathematics and Mechanics 9 (1): 49–58. https://doi.org/ calibrate the reservoir properties using water-cut-breakthrough
10.1002/zamm.19290090104. times, and can be generalized to three-phase flow. More generally,
Rey, A., Bhark, E., Gao, K. et al. 2012. Streamline-Based Integration of the arrival-time sensitivities may be used to relate the onset time
Time-Lapse Seismic and Production Data into Petroleum Reservoir of a change in a geophysical observable to reservoir-flow proper-
Models. Geophysics 77 (6): M73–M87. https://doi.org/10.1190/ ties (Vasco et al. 2014, 2015). Consider the incompressible flow
geo2011-0346.1. of oil and water described by the Buckley-Leverett equation
Rwechungura, R., Bhark, E., Miljeteig, O. et al. 2012. Results of the First (Buckley and Leverett 1942) written by use of the streamline time
Norne Field Case on History Matching and Recovery Optimization of flight as the spatial coordinate (Datta-Gupta and King 2007):
Using Production and 4D Seismic Data. Presented at the SPE Annual @Sw @Fw
Technical Conference and Exhibition, San Antonio, Texas, 8–10 Octo- þ ¼ 0: . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-1Þ
@t @s
ber. SPE-157112-MS. https://doi.org/10.2118/157112-MS.
Skjervheim, J.-A., Evensen, G., Aanonsen, S. et al. 2007. Incorporating The velocity of a given fluid saturation Sw along a streamline
4D Seismic Data in Reservoir Simulation Models Using Ensemble is given by the slope of the fractional-flow curve:

Kalman Filter. SPE J. 12 (3): 282–292. SPE-95789-PA. https:// @s dFw
doi.org/10.2118/95789-PA. ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-2Þ
@t Sw dSw
Steffensen, I. and Karstadt, P. 1996. Norne Field Development–Fast Track
From Discovery to Production. J Pet Technol 48 (4): 296–339. SPE- Eq. A-2 relates the travel time of a particular fluid saturation,
30148-JPT. https://doi.org/10.2118/30148-JPT. s dFw
tðSw ; s; wÞ to the time of flight s, ¼ . We can now compute
Toinet, S. 2004. 4D Feasibility and Calibration Using 3D Seismic Model- t dSw
ing of Reservoir Models. Presented at Abu Dhabi International Confer- the sensitivity of the saturation arrival time by use of that of the
ence and Exhibition, Abu Dhabi, 10–13 October. SPE-88783-MS. streamline time of flight, as follows:
https://doi.org/10.2118/88783-MS.
dtðSw ; s; wÞ dsðwÞ dFw
Tura, A. and Lumley, D. 2000. Estimating Pressure and Saturation ¼ ; . . . . . . . . . . . . . . . . . ðA-3Þ
dkðxÞ dkðxÞ dSw
Changes From Time-Lapse AVO Data. Presented at Offshore Technol-
ogy Conference, Houston, 1–4 May. OTC-12130-MS. https://doi.org/ dtðSw ; s; wÞ dsðwÞ dFw
¼ : . . . . . . . . . . . . . . . . . ðA-4Þ
10.4043/12130-MS. d/ðxÞ d/ðxÞ dSw
Vasco, D. W. 2004. Seismic Imaging of Reservoir Flow Properties: Time-
After water breakthrough on a streamline, these arrival times
Lapse Pressure Changes. Geophysics 69 (2): 511–521. https://doi.org/
are evaluated at the total time of flight, sðwÞ, for the streamline.
10.1190/1.1707071.
Vasco, D. W. and Datta-Gupta, A. 1999. Asymptotic Solutions for Solute Water-Saturation-Sensitivity Calculation. We now derive
Transport: A Formalism for Tracer Tomography. Water Resour. Res. semianalytic expressions for the sensitivity of water saturation
35 (1): 1–16. https://doi.org/10.1029/98WR02742. with respect to variations in permeability, as in Vasco et al.
Vasco, D. W. and Datta-Gupta, A. 2016. Subsurface Fluid Flow and Imag- (2004). For two-phase flow, water saturation is a function of the
ing: With Applications for Hydrology, Reservoir Engineering, and streamline time of flight s and time t. First consider self-similar
Geophysics. Cambridge, UK: Cambridge University Press. solutions to Eq. A-1, for which the saturation is a function of the
Vasco, D. W., Bakulin, A., Baek, H. et al. 2015. Reservoir Characteriza- dimensionless ratio s=t. This solution allows us to relate the deriv-
tion Based Upon the Onset of Time-Lapse Amplitude Changes. Geo- ative of saturation with respect to time to the derivative of the sat-
physics 80 (1): M1–M14. https://doi.org/10.1190/geo2014-0076.1. uration with respect to s, as follows:
Vasco, D. W., Daley, T. M., and Bakulin, A. 2014. Utilizing the Onset of @Sw s dSw ðs=tÞ
Time-Lapse Changes: A Robust Basis for Reservoir Monitoring and
¼ 2 ; . . . . . . . . . . . . . . . . . . . . . . ðA-5Þ
@t t dðs=tÞ
Characterization. Geophys. J. Int. 197 (1): 542–556. https://doi.org/
10.1093/gji/ggt526.
@Sw 1 dSw ðs=tÞ
¼ ; . . . . . . . . . . . . . . . . . . . . . . . . ðA-6Þ
Vasco, D. W., Datta-Gupta, A., Behrens, R. et al. 2004. Seismic Imaging
@s t dðs=tÞ
of Reservoir Flow Properties: Time-Lapse Amplitude Changes. Geo- Hence,
physics 69 (6): 1425–1442. https://doi.org/10.1190/1.1836817.
@Sw t @Sw
Vasco, D. W., Yoon, S., and Datta-Gupta, A. 1999. Integrating Dynamic ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-7Þ
Data Into High-Resolution Reservoir Models Using Streamline-Based @s s @t
Analytic Sensitivity Coefficients. SPE J. 4 (4): 389–399. SPE-59253- giving us the water-saturation sensitivity,
PA. https://doi.org/10.2118/59253-PA.
dSw @Sw ds t @Sw ds
Veeken, P. C., Priezzhev, I. I., Shmaryan, L. E. et al. 2009. Nonlinear ¼ ¼ : . . . . . . . . . . . . . . . . . ðA-8Þ
Multitrace Genetic Inversion Applied on Seismic Data Across the dk @s dk s @t dk
Shtokman Field, Offshore Northern Russia. Geophysics 74 (6): The partial derivative of water saturation with respect to time
WCD49–WCD59. https://doi.org/10.1190/1.3223314. in Eq. A-8 can be calculated numerically by a backward time dif-
Williams, M. A., Keating, J. F., and Barghouty, M. F. 1998. The Strati- ference as
graphic Method: A Structured Approach to History Matching Complex
@Sw ðs; tÞ Sw ðs; tÞ Sw ðs; t DtÞ
Simulation Models. SPE Res Eval & Eng 1 (2): 169–176. SPE-38014- ; . . . . . . . . . . . . . ðA-9Þ
PA. https://doi.org/10.2118/38014-PA. @t Dt
Yin, J., Park, H.-Y., Datta-Gupta, A. et al. 2011. A Hierarchical Stream- where Dt is the timestep size. Use of this expression requires us to
line-Assisted History Matching Approach With Global and Local Pa- save the saturation information for the timestep immediately
rameter Updates. J. Pet. Sci. Eng. 80 (1): 116–130. https://doi.org/ before the time-lapse-survey time. Therefore, the saturation sensi-
10.1016/j.petrol.2011.10.014. tivity at location s at a given time t can be calculated by
16 2017 SPE Journal

dSw ðs; tÞ t Sw ðs; tÞ Sw ðs; t DtÞ ds

¼ ; ...... ðA-10Þ SWAT 0.00 0.25 0.50 0.75 1.00
dk s Dt dk
where the last partial derivative of travel time with respect to perme- I1 P1
ability can be obtained from Eq. 8. In the derivation, it was assumed
that the streamline trajectories do not change over time. The large
time intervals typically between seismic surveys may require that we
break up the period into several increments during which the stream-
lines do not shift significantly. Also, most often we are calibrating
against the changes in saturation, rather than the saturation itself.
Fig. B-1—A 1D simulation model.
Water-Saturation-Change-Sensitivity Calculations. In the water-
saturation-sensitivity derivation, we assumed that the streamline
trajectory is fixed from the initial condition to the particular time Water-Saturation-Sensitivity Validation. We set up a simple
of interest for calculating the sensitivity. Also, we used the self- 1D homogeneous model with a single injector at one end and a
similar solution to estimate partial derivatives in Eqs. A-5 and A- producer at the other (Fig. B-1), similar in nature to a coreflood
6, which assumes uniform initial conditions. To generalize the experiment. The two immiscible fluids in the system are oil and
formulation to account for changing field conditions, we separate water. The rock and fluid properties are the same as in Table 1,
the total time t into small timestep sizes as except for permeability of 200 md and porosity of 0.15. The rela-
XN tive permeability curves are as shown in Fig. 3. The wells are con-
t¼ Dti ; . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-11Þ trolled by BHP constraints and we conduct the simulation for 50
i¼1 days, in 10 timesteps of 5 days each. The details of the simulation,
and the total change in the arrival time as the weighted average of such as the grid geometry, are summarized in Table B-1.
arrival-time change for each timestep as To validate the semianalytic sensitivity estimates, we compare
XN Dti them with a numerical-perturbation approach that computes the
dt ¼ dti : . . . . . . . . . . . . . . . . . . . . . . . ðA-12Þ sensitivities directly as
i¼1 t
@d gðm þ dmÞ gðmÞ
Then, the arrival-time sensitivity becomes ; . . . . . . . . . . . . . . . . . . . ðB-1Þ
" # " # @m dm
dt XN Dti dti XN Dti @ti ds
¼ ¼ where d is the observed quantity, g(m) is a nonlinear function rep-
dk s0 i¼1 t dk s0 ;ti i¼1 t @s s0 ;ti dk s0 ;ti resenting the reservoir simulation, and m is the reservoir property
" # " #
XN Dti t ds XN Dti ds of interest. The computation of all the sensitivities requires Ngþ1
¼ ¼ : simulations, where Ng is the number of gridblocks, whereas the
i¼1 t s0;i dk s0 ;ti i¼1 s dk
0;i s0 ;ti proposed analytic sensitivity calculation requires only a single
simulation. For the following experiment, the perturbation is 10%
ðA-13Þ
of the original parameter value for numerical-sensitivity calcula-
Substituting Eq. A-13 into Eq. A-10, we obtain the generalized tions. Water-saturation sensitivities, associated with perturbations
form of saturation sensitivity at gridblock location s0 as in reservoir permeabilities, are shown in Fig. B-2. The stream-

line-based sensitivities show very good agreement with the nu-
dSw ðs0 ; tÞ @Sw ðs0 ; tÞ XN Dti ds merical-perturbation estimates. The peak value of the sensitivity
¼ j
dk @t i¼1 s0;i dk s0 ;ti profile corresponds to the water-saturation-front location where a

Sw ðs0 ; tÞ Sw ðs0 ; t DtN Þ XN Dti ds sharp saturation change occurs. Behind the water front, the sensi-
j : tivity profile decays because the saturation changes diminish.
DtN i¼1 s0;i dk s0 ;ti
ðA-14Þ
Pressure-Sensitivity Validation. The pressure sensitivities are
To account for the changing field conditions, the streamlines verified by use of the same model as in the preceding subsection.
are traced for every time interval Dti , and the saturation sensitivity However, the boundary condition at the wells is now changed to a
in Eq. A-14 is computed by integrating the sensitivity values for rate constraint of 50 RB/D. Pressure-drop sensitivities with
the entire time interval. Also, if the streamlines do not change, respect to changes in reservoir permeability are shown in Fig. B-3.
Eq. A-14 reverts to Eq. 12. Now the saturation-change sensitivity The streamline-based sensitivities show good agreement with esti-
can be computed by calculating the saturation sensitivities at two mates derived by use of a numerical approach. The sensitivity pro-
different times ðt ¼ t1 ; t ¼ t2 ; t1 < t2 Þ, as follows: file shows that the sensitivities contain both positive and negative

dDSw ðs0 Þ t2 dSw ðs0 ; t2 Þ dSw ðs0 ; t1 Þ
dk ¼ dk

dk
t1
" #
@Sw ðs0 ; t2 Þ X N2 Dti ds
¼
@t i¼1 s0;i dk s0 ;ti
" #
@Sw ðs0 ; t1 Þ XN1 Dti ds
;
@t i¼1 s0;i dk s0 ;ti
ðA-15Þ
where the saturation-time partial derivatives can be approximated
by a backward differencing from the simulation responses, as in
Eq. A-9.
Appendix B—Sensitivity-Calculation Validations Table B-1—Simulation specifications of saturation-sensitivity-

In Appendix B, numerical tests are conducted to verify the pro- validation case. Pressure/volume/temperature values are at the
posed water-saturation- and pressure-drop-sensitivity calculations. reference pressure of 5,000 psi.
2017 SPE Journal 17

4.E–04 1
0.9
4.E–04
Water-Saturation Sensitivity
Streamline 0.8
3.E–04
Water Saturation
Numerical 0.7
3.E–04 0.6
2.E–04 0.5
2.E–04 0.4
0.3
1.E–04
0.2
5.E–05 0.1
1.E–19 0
0 10 20 30 40 50 0 10 20 30 40 50
–5.E–05
Grid No. Grid No.
(a) (b)
4.E–04 1
4.E–04 Streamline 0.9
Water-Saturation Sensitivity
0.8
3.E–04 Numerical
Water Saturation
0.7
3.E–04
0.6
2.E–04 0.5
2.E–04 0.4
1.E–04 0.3
0.2
5.E–05
0.1
1.E–19 0
0 10 20 30 40 50 0 10 20 30 40 50
–5.E–05
Grid No. Grid No.
(c) (d)
Fig. B-2—Water-saturation sensitivities with respect to perturbations in permeability. (a) Sensitivity values with respect to the per-
meability of Gridblock No. 15 at a time of 25 days. The red line is the semianalytic streamline-based calculation. The green line is
the sensitivity estimated by a numerical-perturbation approach. (b) Water-saturation profile at a time of 25 days. (c) Sensitivity val-
ues with respect to permeability of Gridblock No. 15 after 50 days of production. (d) Water-saturation profile at 50 days.
4.E–02 5,150
Streamline
3.E–02 Numerical 5,100
Pressure–Drop Sensitivity
2.E–02
Pressure (psi)
1.E–02 5050
0.E+00 5,000
0 10 20 30 40 50
–1.E–02
4,950
–2.E–02
–3.E–02 4,900
0.00 10.00 20.00 30.00 40.00 50.00
–4.E–02
Grid No. Grid No.
(a) (b)
5,150
4.E–02
3.E–02 5,100
Pressure–Drop Sensitivity
Pressure (psi)
2.E–02
5,050
1.E–02
0.E+00 5,000
0 10 20 30 40 50
–1.E–02
4,950
–2.E–02 Streamline
4,900
–3.E–02 Numerical
0 10 20 30 40 50
–4.E–02 Grid No.
Grid No.
(c) (d)
Fig. B-3—Pressure sensitivities with respect to changes in reservoir permeability. (a) Sensitivity with respect to a perturbation of
the permeability of Gridblock No. 15 at a time of 25 days. The red line is the analytic streamline-based calculation, and the green
line is the numerical perturbation. (b) Pressure profile at a time of 25 days. (c) Sensitivity values with respect to permeability of
Gridblock No. 15 at a time of 50 days. (d) Pressure profile at a time of 50 days.
18 2017 SPE Journal

values symmetrically with respect to the perturbed gridblock (No. neering. Datta-Gupta holds a PhD degree from the University
15). A comparison of sensitivity profiles at different times shows of Texas at Austin.
that the shape of the sensitivity profile does not change signifi- Michael J. King is a professor and assistant department head
cantly over time. However, the magnitude of the sensitivity does in the Department of Petroleum Engineering at Texas A&M Uni-
change because the local pressure gradient changes when water versity. He is the holder of the Foundation CMG Chair in Robust
displaces oil, providing a change in the total mobility. Reduced Complexity Modeling and the LeSuer Chair in Reser-
voir Management. King joined Texas A&M University in 2009 af-
ter retiring from BP America as a senior adviser in reservoir
Shingo Watanabe is an Intersect research and prototype modeling and simulation. He was with BP for 27 years, both in
developer at Schlumberger. His research interests include mul- the US and overseas, in a variety of technology and opera-
tiscale reservoir-simulation methods, fast nonlinear solvers, tional roles. King’s research interests include reservoir charac-
automatic history-matching methods, and reservoir characterization, reservoir management, fundamentals of flow in
terization. Watanabe is a member of SPE. He holds master’s porous media, upscaling of geologic models for flow simula-
and PhD degrees in petroleum engineering, both from Texas tion, and, more recently, pressure and rate transient and
A&M University. unconventional reservoir analysis. He has served as an SPE Dis-
Jichao Han is a reservoir engineer at Anadarko Petroleum Cor- tinguished Lecturer and is the coauthor of the SPE textbook
poration. His research interests include unconventional reser- Streamline Simulation: Theory and Practice. King became an
voir simulation, history matching, optimization, geomodel SPE Distinguished Member in 2013, and is recipient of the 2011
rankings, and chemical flooding. Han is a member of SPE. He SPE Reservoir Description and Dynamics Award. He holds a
holds a PhD degree in petroleum engineering from Texas A&M PhD degree in theoretical physics from Syracuse University and
University. bachelor’s degrees in physics and mathematics from The
Cooper Union for the Advancement of Science and Art.
Gill Hetz is a PhD degree candidate in petroleum engineering
at Texas A&M University. His current research interests include D. W. Vasco is currently a research scientist at the Lawrence
inverse modeling and multiscale data integration and history Berkeley National Laboratory, developing and implementing
matching of production and 4D-seismic data to improve reser- algorithms for geophysical and hydrologic data analysis and
voir characterization. Hetz is a member of SPE. He holds a mas- inversion. In 1987, he worked at the same institution on the inver-
ter’s degree in oil and gas engineering from the Technion- sion of surface deformation to image-volume change in the sub-
Israel Institute of Technology. surface. From 1987 to 1989, Vasco was a post-doctoral-degree
scholar at the Phillips Geophysics Laboratory, working on the
Akhil Datta-Gupta is Regents Professor and holder of the L. F. inversion of airborne-gravity-gradiometry observations.
Peterson ’36 Endowed Chair in Petroleum Engineering at Texas From1991 to 1992, he was a visiting fellow at the Australian
A&M University. Previously, he worked for BP Exploration and National University. Vasco’s research interests are inverse theory,
Research and Lawrence Berkeley National Laboratory. Datta- asymptotic techniques for modeling flow and transport, the
Gupta is the recipient of the 2009 SPE John Franklin Carll analysis of surface deformation, and techniques for imaging res-
Award, the 2003 SPE Lester C. Uren Award, the 2000 and 2006 ervoir heterogeneity. He is a member of SPE and a recipient of
SPE Cedric K. Ferguson Certificate, and the 1992 AIME Rossiter the 2000 SPE Cedric K. Ferguson Award. Vasco holds a bache-
W. Raymond Memorial Award. He is an SPE Distinguished lor’s degree in geophysics from the University of Texas at Austin
Member and a member of the US National Academy of Engi- and a PhD degree from the University of California at Berkeley.
2017 SPE Journal 19
View publication stats


Streamline-Based Time-Lapse-Seismic-Data Integration Incorporating Pressure and Saturation Effects

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Streamline-Based Time-Lapse-Seismic-Data Integration Incorporating Pressure and Saturation Effects

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Streamline-Based Time-Lapse-Seismic-Data Integration Incorporating

Article in SPE Journal · April 2017

Shingo Watanabe Gill Hetz

SEE PROFILE SEE PROFILE

Michael King Donald Wyman Vasco

SEE PROFILE SEE PROFILE

multiscale simulation in INTERSECT View project

Basic Energy Sciences View project

The user has requested enhancement of the downloaded file.

Summary sequential Gaussian simulation with block Kriging, accounts for

2017 SPE Journal 1

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Forward Modeling Reservoir PEM Seismic

Pressures Acoustic Simulated

Observation Pressures Acoustic Real

Inverse Modeling Rock-Physics Seismic Inversion

2 2017 SPE Journal

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

2017 SPE Journal 3

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Table 1—Rock and fluid properties. Pressure/volume/temperature

permeability. We can now relate the change in local pressure drop

0.7 well-permeability values as conditioning data (Fig. 4a). The prior

4 2017 SPE Journal

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Pressure (psi) Water Saturation

0.7 0.7 0.7

0.6 0.6 0.6

2017 SPE Journal 5

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

(a) (b) (c) 0.6

6 2017 SPE Journal

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Well Water Cut

Well Water Cut

Well Water Cut

Initial model Reference model After-pressure-change-data integration

After-water-saturation-change-data integration Final model

2017 SPE Journal 7

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Table 3—Input parameters for the PEM in Norne Field.

8 2017 SPE Journal

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Acoustic Impedance (Pa·s/m)

Acoustic Impedance (Pa·s/m)

(a) (b) (c)

2017 SPE Journal 9

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Acoustic Impedance Change Pressure Change Sw Change

–2,500 –2,550 –2,600 –2,650 –2,700 –2,750 –2,800

–2,500 –2,550 –2,600 –2,650 –2,700 –2,750 –2,800

–2,800 –2,750 –2,700 –2,650 –2,600 –2.550 –2,500

–2,800 –2,750 –2,700 –2,650 –2,600 –2.550 –2,500

10 2017 SPE Journal

ID: jaganm Time: 20:22 I Path: S:/J###/Vol00000/170006/Comp/APPFile/SA-J###170006

Global Update: Basis coefficients (v)

Generate models Local Update: Grid permeability (k)

Fig. 15—Global-to-local hierarchical-history-matching work flow.

Reservoir-permeability- Leading GCT-basis vector

Final Update Models

5.6 5.8 5.8

5.4 5.6 5.6