You are on page 1of 13

SPE 124625-PP

A Robust Measure of Heterogeneity for Ranking Earth Models: The F-PHI


Curve and Dynamic Lorenz Coefficient
G. Michael Shook, SPE, Chevron Energy Technology Company; and Kameron M. Mitchell, Chevron International
Exploration and Production

Copyright 2009, Society of Petroleum Engineers

This paper was prepared for presentation at the 2009 SPE Annual Technical Conference and Exhibition held in New Orleans, Louisiana, USA, 4–7 October 2009.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of th e paper have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohi bited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
A new method for estimating heterogeneity in Earth models is presented. The new approach requires running a streamline
simulator only a few timesteps, so multi-million cell models are evaluated in minutes. Using streamline time of flight and
volumetric flow rate information from the simulation, a flow capacity diagram and sweep efficiency history can be
determined. Five different measures of heterogeneity can be obtained from these two diagnostic plots.

We then evaluated each of these measures of heterogeneity for 450 models that were constructed using a wide range of
Dykstra-Parsons coefficient and correlation lengths, and 2 different well patterns. The measures of heterogeneity were
plotted against discounted oil recovery to determine the best measure of heterogeneity. From this study we conclude the
Lorenz coefficient – as determined from dynamic data – is the single best measure of heterogeneity.

Introduction
Uncertainties in reservoir performance under secondary or tertiary displacement arise from a variety of sources. The
complex interaction between depositional environment and subsequent diagenesis is poorly understood, but nevertheless
gives rise to spatial variability in reservoir properties (e.g., porosity, permeability, rock type, etc.). The methods available to
estimate reservoir properties all suffer from concerns of spatial resolution. Some, for example seismic surveys and pressure
transient tests, estimate bulk properties at an intermediate scale but cannot provide “local” values. Others such as well logs
and core analysis may provide accurate estimates of properties at a single point, but cannot provide information other than at
those (sparse) locations. To acknowledge the inherent uncertainty in any given Earth model, a stochastic approach is most
frequently taken. A number of equally probable models are constructed to evaluate the effect of parameter uncertainty on
reservoir performance. The single biggest drawback to this approach is the investment in time required to scale these multi-
million cell models up for simulation, history-matching field performance, and making predictive forward model runs. For
this reason, a method for ranking the behavior of alternative Earth models is attractive, so that a minimum number of models
are used in full field reservoir simulation.

The key to ranking alternative models is to identify a robust indicator of that model‟s performance under displacement
conditions. We submit that an appropriate measure of dynamic heterogeneity is a useful means of ranking models. Primary
recovery is less sensitive to heterogeneity; in fact, Walsh and Lake (2007) show primary recovery in heterogeneous media
with crossflow can be approximated with a suitably averaged, homogeneous permeability. It is during secondary or tertiary
recovery that thief zones or other short circuiting of injected fluids become costly.

Static measures of heterogeneity describe the distribution in permeability and porosity of a given model; dynamic measures
implicitly include the distribution in flowpath length and the connected structure of the flowpaths. Large dynamic
heterogeneity is manifest in premature breakthrough of injected fluids. This early breakthrough may be caused by low
volume, high permeability flowpaths (i.e., thief zones), or merely by a distribution in flowpath lengths where some paths are
much shorter than the mean. The appropriate heterogeneity measure – and therefore ranking tool – should be sensitive to
either of these conditions.

Static methods to estimate heterogeneity and predict recovery were developed in the 1940s and 1950s (Miller and Lents,
2 SPE 124625-PP

1947; Dykstra and Parsons, 1950; Schmalz and Rahme, 1950). Dykstra and Parsons determine the “permeability variation”
(VDP) from the median and variance of an ordered set of permeability data. Their method for estimating recovery as a
function of heterogeneity is based on non-communicating layers with constant permeability and 1-D flow. Lake (1989)
shows the method can be generalized to variable layer thickness and porosity; however, there is no intra-layer heterogeneity,
and all flow path lengths are equal.

Miller and Lents (1947) appear to be the first to use layer flow capacity (kh) and storage (h) to estimate recovery from gas
cycling operations. They assume parallel flow in each layer, which is equivalent to the Dykstra-Parsons assumption of 1-D
flow. Stiles (1949) and Schmalz and Rahme (1950) constructed a flow capacity diagram, which is a plot of cumulative flow
capacity vs. cumulative thickness. Variations in porosity were neglected in these studies. Stiles (1949) used his “capacity
distribution” curve to estimate oil recovery from waterflooding, and Schmalze and Rahme (1950) propose a Lorenz
Coefficient as a measure of heterogeneity. Both the flow capacity curve and Lorenz coefficient are discussed in more detail
below. In the present context, these static estimators also fail to account for permeability connectivity which would affect
recovery.

More recently, a number of dynamic methods have been proposed to measure heterogeneity and rank geologic models. In
broad terms the methods can be grouped as: fast simulation, permeability-connectivity estimates, and streamline simulation.

Ballin et al. (1992) proposed using a Fast Simulator (FS) to rank various models. They then use a comprehensive flow
simulator (CS) on a select few of the models to generate approximate probability distribution functions for a variety of
simulation output. The method relies on identifying an appropriate set of FS response parameters that are rank-preserving
approximations of the CS model. Deutsch and Srinivasan (1996) also used FS models to rank alternative Earth models. In
that study, the FS were simply upscaled models of the originals. Both of these methods showed some success in ranking
models; however, one disadvantage is that model scaleup is required.

Hird and Dubrule (1998) and Ballin et al. (2002) developed methods for ranking models by estimating reservoir connectivity.
The Resistivity Index (RI) method measures the least resistive path between an arbitrary grid block and a producer. The
method appears best suited for depletion studies (Ballin et al., 2002), as there is no provision to consider flowpath
distributions arising from injection.

Streamline modeling has an advantage over the other classes of model ranking in that no scaleup is required, the models in
general are fast and have minimal numerical dispersion. Saad et al. (1996) proposed using a simulated tracer mean swept
volume as a means of ranking multiple model realizations. Idrobo et al. (2000) show that total oil recovered correlates well
with sweep efficiency at water breakthrough for 50 model realizations they generated. Both of these measures uses recovery
at a fixed time (or volume injected), which fails to consider the effect of recovery as a function of time. The rate of oil
recovered is missing in identifying an appropriate means of ranking models.

The purpose of this paper is two-fold. First, we introduce a method for estimating the flow geometry of an Earth model using
streamline simulation. The F- plot is a generalization of Flow Capacity-Storage Capacity, or Lorenz plots (Stiles, 1949;
Schmalz and Rahme, 1950; Lake, 1989; Gunter et al., 1997) that have historically been constructed from static data. Our
method differs in that it determines flow geometry from residence time data, and thus accounts not only for static
heterogeneity but also streamline length distribution and well pattern geometry. The F- plot can be thought of as a
generalized miscible displacement model, and we also present a graphical method for estimating sweep efficiency directly
from the F- curve. The method is analogous to the Welge (1952) graphical interpretation of the Buckley-Leverett equation
(1942).

We then construct a suite of 224 synthetic Earth models that encompass a wide range of stochastic model properties.
Streamline simulation of each Earth model (and two well patterns) was performed, and a set of “heterogeneity indicators”
were calculated from the simulation output. The models were then ranked according to each of the indicators, and the most
robust measure of dynamic heterogeneity was identified. The ability to rank Earth models using streamline simulation
provides the advantage that the select few models can be upscaled and used in uncertainty analysis.

Streamline Simulation
Streamline simulation offers a rapid means of assessing heterogeneity in Earth models. Briefly, streamline models solve for
fluid pressures on a regular grid, then construct streamlines to describe flow between sources and sinks. Streamlines are
constructed such that they are everywhere tangential to the velocity field, but otherwise may take arbitrary shape; i.e., they
are not constructed along the finite difference grid. A more rigorous description of streamline simulation is found in Idrobo
et al. (2000) and Datta-Gupta and King (2007).
SPE 124625-PP 3

The output data that is used in this analysis from streamline models is simply the “time of flight” (TOF) of the streamlines, i,
and their volumetric flow rate, qi. This information is sufficient to estimate flow geometry and sweep efficiency of a given
model. The flow geometry plots generated are also referred to as Lorenz Plots, or F- curves, as discussed below. The F-
curve can be used directly in calculating a Sweep Efficiency history as well.

Streamline simulation allows for compressible fluids by solving the pressure equation at various times during the simulation.
Multiple pressure solutions are also required if displacement forces are not balanced (e.g., the mobility ratio is not unity, or
buoyancy forces are significant). In these cases, the distribution in streamlines is not at steady state; they vary in time. This
causes ambiguity in describing „heterogeneity,‟ since intuitively heterogeneity should be a property of the reservoir model,
not the displacement mechanism. For that reason, our application of streamline simulation requires the following:
 small, constant compressibility
 straight line relative permeability curves and negligible capillary pressure
 unit mobility ratio and no density differences

Constant (and small) compressibility is easier to solve numerically than incompressible flow, and yet attenuates transients
associated with compressible fluids rapidly. With the requirements of no capillary pressure or viscous or buoyancy
imbalances, the method allows a very rapid assessment of the flow geometry. We run a streamline simulator (in our case,
FrontSim) a few timesteps to attenuate pressure transients, and then output time of flight and volumetric flow of the
streamlines. This approach has an advantage of other streamline-based methods in that we take only a few time steps with
the streamline simulator, and are thus able to assess heterogeneity in multi-million cell models in minutes.

Flow Capacity – Storage Capacity Diagrams and F- Curves


Flow capacity – storage capacity diagrams have appeared in reservoir engineering literature for decades (Stiles, 1949;
Schmalz and Rahme, 1950; Lake, 1989; Gunter et al., 1997). Also known as F-C curves, they were originally derived for 2-
D, vertical cross section, non-communicating, layered reservoirs. Under those conditions, Flow Capacity, F, and Storage
Capacity, C are readily calculated.

Imagine a collection of N permeable medium layers, each having a different permeability, k, porosity, , and thickness, h. If
there is no communication between these layers, we can uniquely describe the flow capacity and storage capacity of the
individual layers.

First, the layers are ordered according to decreasing fluid velocity. A variable R i = ki/i is defined, and the layers are ordered
according to decreasing R. Note that since the layers do not communicate the reordering doesn‟t affect the layers‟ behavior.
Individual layer flow capacity is defined as the volumetric flow of that layer, divided by the total volumetric flow. Writing
Darcy‟s law for each layer and cancelling like terms (recall the cross sectional area and length is equal for all N layers) gives

qi (kh ) i
fi  N
 N
(1)
 qi  (kh ) i
i 1 i 1

The storage capacity of layer “i” is simply the layer pore volume divided by the total pore volume:

VP i (h) i
ci  N
 N
(2)
 VP i  (h) i
i 1 i 1

The F-C diagram is constructed as a cumulate distribution function of the layer f and c. A point F i represents the volumetric
flow of all layers with velocity ≥ the “ith” and the corresponding Ci is the pore volume associated with those layers:

i i
qj  (kh ) j
j1 j1
Fi  N
 N
(3)
qj  (kh ) j
j1 j1
4 SPE 124625-PP

i i
 Vp j  (h) j
j 1 j 1
Ci  N
 N
(4)
 Vp j  (h) j
j 1 j 1

Calculations for a simple, 5-layer model are given in Table 1 below, and the resulting F-C curve is given in Figure 1.

Table 1. Example calculation of F- data for a 5 layer model.


h (ft) k (md)  kh h F C
0 0
5 500 0.25 2500 1.25 0.7564 0.333
5 100 0.2 500 1 0.9077 0.6
5 50 0.15 250 0.75 0.9834 0.8
5 10 0.1 50 0.5 0.9985 0.933
5 1 0.05 5 0.25 1 1
3305 3.75

0.75
Flow Capacity, F

0.5

0.25

0
0 0.25 0.5 0.75 1
Storage Capacity, 

Figure 1. F-C diagram for the 5-layer case described in Table 1.

The concepts behind the F-C construction can be generalized to 3-D, heterogeneous media in a simple fashion using
streamlines. To emphasize the differences in the assumptions in our development (e.g., streamlines in a heterogeneous 3-D
medium vs. parallel flow in layered media with constant intralayer properties; flowpaths arising from displacement vs. static
layer properties), we refer to the 3-D, dynamic Flow Capacity plot as an “F- plot.”

Datta-Gupta and King (1995) show the pore volume of the ith streamline is given as

Vpi = qi i (5)

Where Vpi is its pore volume, qi is the volumetric flow rate assigned to the streamline, and i is the time of flight (TOF).
Using this simple concept, the F- relationships can be derived from streamline simulation rapidly.
 Run a streamline simulator a few time steps, so pressure transients are attenuated and the simulation is at
steady state.
 Output streamline volumetric flow rate and time of flight from the simulation
 Calculate the individual streamlines‟ pore volume using Eqn 5 above.
 Order the streamlines according to increasing residence time (increasing Vp/q). Note that by definition
steady state streamlines do not interact or communicate.
 Calculate F and  and plot.

i i
qj  Vp j
j 1 j 1
Fi  N
and i  N
(6)
 qi  Vp j
j 1 j 1

Figure 2 below shows a comparison in the streamline-derived F- curve and the analytical calculation from a 2-D model
with input values of permeability, porosity, and layer thickness. This is one of the few cases in which the F- curve derived
SPE 124625-PP 5

from streamline simulation (i.e., a dynamic estimate) agrees with the static F-C calculation. This is because for this case the
streamlines are parallel, all flowpath lengths are equal, and streamline TOF is proportional only to k/. For the more general
case, streamlines have arbitrary (nonuniform) length, the streamline TOF is then proportional to k/ and streamline length,
and so the dynamic F- curve cannot be inferred a priori from static data.

Figure 2. Comparison in F- data for a Dykstra-Parsons model (log-normal permeability distribution, 2-D cross sectional model).
Symbols are the analytic calculation of F-C; solid line is the F- curve inferred from streamline behavior. The static and dynamic F-
 curve in this case are identical, since streamlines have equal length, cross-sectional area is unity. This represents one of the few
cases in which static and dynamic heterogeneity are equivalent.

Two measures of dynamic heterogeneity can be determined from the F- plot. The first is the Lorenz coefficient (Schmalz
and Rahme, 1950; Lake, 1989). Lc is defined as

1
LC  2(  Fd  0.5) (7)
0

A Lorenz coefficient of zero falls along the 45º line on the F- curve, which is a homogeneous displacement (equal
volumetric flow from every incremental pore volume). A value of 1 is referred to as “infinitely heterogeneous”; it can be
interpreted as “all of the flow comes from a small (approaching 0) portion of the pore volume. These are shown
schematically in Figure 3.

F- curve for V DP = 0.98.


Note ~99% of the flow
comes from 25% of the F- curve for V DP = 0.7.
pore volume
0.75
F-PHI curve for V DP = 0
FHI for VDP = 0.7.
(homogeneous
F=0.72;  = 0.28, so
displacement). Each
FHI=2.57
Flow Capacity, F

layer has same pore


volume, same
0.5 volumetric flow.

0.25

0
0 0.25 0.5 0.75 1
Storage Capacity, PHI

Figure 3. Example F- plots showing limits of homogeneous displacements, very heterogeneous displacements, and an intermediate case. FHI is
calculated from the slope of the F- curve as shown.

Zheng et al. (2000) refer to point on the F- curve with unit slope as the “effective flow interval.” It can be shown (Wu et
al., 2007) that the derivative of the F- curve is

dF t *
 (8)
d  i

where t* is the mean residence time of all streamlines and  is the TOF of the ith streamline, so a unit tangent can be
6 SPE 124625-PP

considered as representative of the domain (i = t*). This point has also been used to calculate the Flow Heterogeneity Index
(FHI; personal communication, C. Harrison, Chevron). FHI is the value of F/ on the diagram where the tangent to the
curve has unit slope.

F
FHI  (9)
 m 1

For homogeneous media, FHI=1; it has no upper limit. FHI is also shown schematically in Figure 3.

A third measure of heterogeneity is the coefficient of variation of the streamline TOF. The coefficient of variation is defined
as (Jensen et al., 2000)

Var ( )
CV  (10)
t*

where Var() is the variance of the residence time distribution (i.e., the 2 nd temporal moment of the TOF distribution), and t*
is defined as above.

Sweep Efficiency Histories


A second diagnostic plot that is readily obtained from F- data is a sweep efficiency history plot. Sweep is defined as:

Ev (t)= Volume of reservoir contacted by displacing agent at time t (11)


Volume contacted at t=

Idrobo et al. (2000) show swept volume as a function of time can be determined from the streamline TOF distribution.
Shook and Forsmann (2005) show E v can be determined directly from F- data as

q t
EV   [1  F ( )]d (12)
VP 0

In Appendix A we show Sweep Efficiency can also be determined graphically from the F- curve as:

1  Fi (t i )
E v (t i )   i (t i )  (13)
dF / d i

The procedure is analogous to the Welge (1952) graphical solution to the Buckley-Leverett (1942) problem; thus, the F-
curve can be interpreted as a generalized flow curve, one that describes displacements in 3-D. The two methods for
estimating Ev from streamline data (Eqns. 12 and 13) for a homogeneous 5-spot are shown in Figure 4. The two curves are
indistinguishable, and agree well with the analytic solution to the problem (Morel-Seytoux, 1966).

Several more measures of dynamic heterogeneity are obtained from the sweep efficiency history. Idrobo et al. (2000)
propose sweep efficiency at breakthrough as a good means of ranking heterogeneity. This measure is the inverse of the
Koval factor (Koval, 1963), originally proposed as a means of predicting performance in unstable miscible displacements.
Saad et al. (1996) use tracer data to estimate the mean swept volume (volume swept at 1 pore volume injected) as a measure
of heterogeneity. We replace their metric with sweep efficiency at one pore volume injected.

Therefore, five measures of dynamic heterogeneity have been identified. These are measures of dynamic heterogeneity
because they are developed from the F- curve based on streamline simulation (i.e., dynamic data), and all are readily
measured for a given simulation. The candidates for describing dynamic heterogeneity are summarized in Table 2 below.
We now must construct Earth models, calculate the 5 candidate values for each model, rank the Earth models according to a
“response” to heterogeneity (discussed below), and identify the most robust among the candidates. This is described in the
following section.
SPE 124625-PP 7

0.9

0.8

Sweep Efficiency 0.7

0.6

0.5

0.4

0.3
Direct calculation
0.2
Welge Integration
0.1
Analytic Solution
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Dimensionless time, tD

Figure 4. Comparison in sweep efficiency histories for a homogeneous ¼ 5-spot. The direct calculation is Eqn. 12 above, the Welge integration
method is Eqn. 13, the analytic solution is given by Morel-Seytoux (1966). The numerical and graphical solutions are indistinguishable.

Table 2. Summary of candidates for reporting dynamic heterogeneity


Name Formula Description
Cv Var ( ) Coefficient of variation, recognized as ‘dimensionless variance’
CV 
t*
Ev at BT Sweep efficiency at breakthrough
Ev at tD=1 Sweep efficiency at 1 pore volume injected
FHI F The ratio of Flow-to-Storage where the F- curve has unit slope
FHI  (which is representative of mean bulk flow)
 m 1
Lc 1 Standard statistical measure of CDFs; a measure of deviation from
LC  2(  Fd  0.5) a homogeneous model
0

Earth Models
In order to identify the most robust measure of dynamic heterogeneity, a series of Earth models was constructed. These
synthetic models were created in Gocad and exported for FrontSim simulations. All of the models are 20 acre squares, with a
total thickness of 25 ft, and all have constant porosity (so that all pore volumes are equal), and log-normal distribution in
permeability. The numerical model was built with an areal grid of 101101 and 10 layers, and we used Sequential Gaussian
Simulation (SGS) to generate our permeability fields. SGS in Gocad requires the following input:

 Mean of the field. We used a constant value of 100 md for all cases.
 Standard deviation. For log-normal permeability fields, there is a 1:1 correspondence between the Dykstra-
Parsons coefficient, VDP, and standard deviation, so this is equivalent to fixing VDP for our models. We built
models with VDP varying between 0.6 and 0.9 by increments of 0.05, so 7 different static heterogeneity
measures.
 Horizontal correlation lengths. Horizontal correlation lengths used in the study were 66 ft., 660 ft., 2640 ft.,
and 33000 ft. For a quarter 5-spot, these lengths represent 0.05, 0.5, 2.0, and 25 well spacings. We further
assumed horizontal correlation structure is isotropic (x = y).
 Vertical correlation lengths. Vertical correlation lengths were 2.5 ft. (0.1 reservoir thicknesses), and 12.5 ft.
(0.5 thicknesses).

These combinations constitute 56 different Earth models. We ran each of the cases with 4 ratios of vertical-to-horizontal
permeability (10-3, 10-2, 0.1, 0.5), and so have 224 heterogeneous models, and include a homogeneous model as well (225
total). Finally, we considered a quarter 5-spot pattern and a line-drive pattern; thus 450 sets of model runs are reported.

An important aspect in identifying the most robust measure of dynamic heterogeneity is a meaningful metric with which to
rank models models. In this study, Discounted Oil Production is used as that metric. The injection/production rate is fixed,
8 SPE 124625-PP

such that mean interstitial velocity for the 5-spot case is 0.3 ft/day (Parsons, 1974). Incremental sweep efficiency –
calculated each quarter year for the duration of the recovery history – is converted to oil recovered, and discounted to present
day using a discount rate of 10%. This metric allows the full sweep efficiency history to be included as a means of assessing
heterogeneity. Previous studies considered only a single point on that curve.

We include a single limiting case in this analysis: that of perfect displacement. That is, complete sweep and 100% oil
recovery at 1.0 pore volumes injected. For our injection and discount rates, this results in Discounted Oil Recovery of
0.6321.

Study Results
Figures 5-9 show results of Discounted Oil Recovered vs. each of the 5 candidates heterogeneity measures, and for both well
patterns (blue is 5-spot, magenta is line drive data). Several observations are apparent from these figures:
 The coefficient of variation, Cv, has too much variation to be a useful measure of heterogeneity. Further
analysis of these results indicates that the curve in Figure 5 actually consists of multiple individual curves
corresponding to individual correlation lengths. Thus, Cv is not a robust measure of heterogeneity, as it fails
to account for differing correlation lengths uniquely.

0.7

0.6
Discounted Oil Recovery

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25

Coefficient of Variation, Cv

Figure 5. Coefficient of Variation vs. Discounted Oil Recovery.

 Sweep efficiency at breakthrough (or the inverse of the Koval factor, 1/H K) is likewise too variable to be
useful (see Figure 6). This curve also comprises multiple individual curves for different correlation lengths.

0.7

0.6
Discounted Oil Recovery

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1

Sweep Efficiency at Breakthrough (1/HK)

Figure 6. Sweep efficiency at breakthrough vs. Discounted Oil Recovery.


SPE 124625-PP 9

 Sweep efficiency at 1 pore volume injected (Figure 7) is an excellent indicator of dynamic heterogeneity.

Figure 7. Sweep at 1 PVI vs. Discounted Oil Recovery

 FHI is a good measure of heterogeneity; however, it does not perform as well as some of the others.

0.7

0.6
Discounted Oil Recovery

0.5

0.4

0.3

0.2

0.1

0
1 1.5 2 2.5 3 3.5 4 4.5 5

Flow Heterogeneity Index, FHI

Figure 8. FHI vs. Discounted Oil Recovery

 Lc is another excellent indicator of heterogeneity, and slightly outperforms Ev at 1 PVI as discussed below.
10 SPE 124625-PP

0.7

0.6

Discounted Oil Recovery 0.5

0.4

0.3

0.2

y = -0.4541x2 - 0.0437x + 0.6212


R2 = 0.9961
0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Dynamic Lorentz Coefficient, Lc

Figure 9. Lc vs. Discounted Oil Recovery

Of five of the heterogeneity measures discussed here, two appear excellent means of ranking Earth models; sweep efficiency
at 1 PVI and Lc both show amazingly small variation over the complete set of Earth models considered. There are no overly
compelling reasons to select Lc as the better tool, however, we do so for the following reasons:

 Using the regression equations in Figures 7 and 9 we calculated the root mean square error for Lc and Ev. Lc
had a lower RMS error by 50%. Both have RMS error of order 10-2.
 Lc is slightly better in preserving rank of the models. That is, the change in discounted oil recovery is in the
correct direction for changes in Lc 3% more frequently than for Ev.
 Sweep efficiency could be considered a “secondary variable” in that it is calculated directly from the F-
curve (from which Lc is determined).

Thus, we have identified a measure of dynamic heterogeneity that is the most robust of five candidates for the 225 Earth
models and 2 well patterns considered here. We therefore conclude the “dynamic Lorenz coefficient” is the preferred means
of ranking models according to dynamic heterogeneity.

Other Applications
In this study we focused on a single well pair in ranking models. There are a number of modifications that can be made to
enhance the applicability of the method. These are summarized briefly below:
 In a full field study, separate streamline data (TOF, q) by injector, and conduct the study as outlined about.
This provides the data to examine the spatial distribution of heterogeneity. Further separation into specific
well pairs provides information on anisotropy.
 Calculate the mean residence time of the streamlines, and then neglect the contribution of those whose
residence time is greater than some predetermined cutoff (e.g., greater than twice the mean, or t i > 2t*).
This provides the total pore volume drained in some appropriate time frame (e.g., two pore volumes
injected), and provides information similar to connected geobodies (Balin et al., 2002). This could also be
used to estimate permeability/porosity cutoffs.

It is further clear the method could be applied to an Earth model and the subsequent scaled up flow model, to ensure
heterogeneity was preserved during scaleup.

Summary and Conclusions


Several specific conclusions are drawn from this work, namely
 A simple method for calculating flow geometry from streamline simulation of Earth models is presented.
We term this variation of flow capacity – storage capacity diagrams “F-” curves, to emphasize these are
calculated from dynamic – not static – data.
SPE 124625-PP 11

 The advantage of our approach is that we require only a few time steps of a given streamline model to
achieve steady state conditions. This allows us to assess heterogeneity in multi-million cell models in a
matter of minutes.
 Through a simulation study encompassing a wide range of static properties that describe heterogeneity, we
have identified the Dynamic Lorenz Coefficient, Lc, as the most robust indicator of heterogeneity, and
therefore the best metric to rank Earth models.
 The method is easily generalized to assess the spatial distribution of heterogeneity or field anisotropy, and
can be used to demonstrate heterogeneity was preserved during scaleup.

Acknowledgements
The authors gratefully thank Tengizchevroil for their encouragement and support of this study, Dennis Dull and Ricardo
Combellas for help in testing the methods, Janet Wilton for help preparing the manuscript, and Chevron management for
permission to publish the paper.

References
Ballin, P.R., A.G. Journel and K. Aziz, 1992, “Prediction of uncertainty in reservoir performance forecast,” Journal of Cadadian Petroleum
Technology, 31(4), April 1992, pp 52-62.

Balin, P.R., R. Solano, K.B. Hird and R.F. Volz, 2002, “New Reservoir Dynamic Connectivity Measurement for Efficient Well Placement
Strategy Analysis Under Depletion,” 2002, SPE paper 77375, presented at 2002 SPEAnnual Techniical Conference, San Antonio,
Tx., 29 Sept – 2 Oct, 2002.

Buckley, S.E. and M.C. Leverett, 1942, “Mechanisms of Fluid Displacement in Sands,” Trans. AIME 146 (1942) 107-116.

Datta-Gupta, A. and M.J. King, 1995, “A semianalytic approach to tracer flow modeling in heterogeneous permeable media,” Advances in
Water Resources, 18 (1995) 9-24.

Datta-Gupta, A. and M.J. King, 2007, Streamline Simulation: Theory and Practice, Society of Petroleum Engineers.

Deutsch, C.V. and S. Srinivasan, 1996, “Improved Reservoir Management Through Ranking Stochastic Reservoir Models,” SPE paper
35411, presented at the 1996 SPE/DOE Tenth Symposium on Improved Oil Recovery, Tulsa OK, 21-24 April, 1996.

Dykstra, H. and R.L. Parsons, 1950, “The Prediction of Oil Recovery by Water Flood,” Secondary Recovery of Oil in the United States,
Principles and Practice, 2d ed., American Petroleum Institute (1950, pp 160-174.

Gunter, G.W., J.M. Finneran, D.J. Hartmann, and J.D. Miller, 1997, “Early Determination of Reservoir Flor Units Using an Integrated
Petrophysical Method,” SPE Annual Technical Conference and Publication, SPE paper 38679, pp. 373-380.

Hird, K.B. and O. Dubrule, 1998, “Quantification of Reservoir Connectivity for Reservoir Description Applications,” SPE Reservoir
Evaluation and Engineering, February, 1998, pp 12-17.

Idrobo, E.A., M.K. Choudhary, and A. Datta-Gupta, 2000, “Swept Volume Calculations and Ranking of Geostatistical Reservoir Models
Using Streamline Simulation,” SPE paper 62557, presented at the 2000 SPE/AAPG Western Regional Meeting, Long Beach, Ca.

J.L. Jensen, L.W. Lake, P.W.M. Corbett, and D.J. Goggin, 2000, Statistics for Petroleum Engineers and Geoscientists, Elsevier.

Koval, E.J., 1963, “A Method for Predicting the Performance of Unstable Miscible Displacement in Heterogeneous Media, Society of
Petroleum Engineers Journal, 3 (June 1963), 145-154.

Lake, Larry W., 1989, Enhanced Oil Recovery, Prentice Hall.

Miller, M.G. and M.R. Lents, “Performance of Bodcaw Reservoir, Cotton Valley Field Cycling Project: New Methods of Predicting Gas-
Condensate Reservoir Performance Under Cycling Operations Compared to Field Data,” Drill. And Prod. Prac., API (1947) 128-
149.

Morel-Seytoux, H.J., “Unit Mobility Ratio Displacement Calculations for Pattern Floods in Homogeneous Medium,” Society of Petroleum
Engeers Journal, 6 (September 1966), 217-227.

Parsons, R.W., 1974, “Velocities in Developed 5-Spot Patterns,” JPT, May 1974, 550.
12 SPE 124625-PP

Saad, N., V. Maroongroge, and C.T. Kalkorney, 1996, “Ranking Geostatistical Models Using Tracer Production Data,” SPE 35494 presented at
the European 3-D Reservoir Modeling Conference, Stavanger, Norway, 16-17 April 1996.

Shook, G.M. and J.H. Forsmann, 2005, “Tracer Interpretation using Temporal Moments on a Spreadsheet,” INL report 05-00400, June 2005
(Rev. 1 Sept 2005).

Schmalz, J.P. and H.D. Rahme, “The Variation of Waterflood Performance with Variation in Permeability Profile,” Prod. Monthly (1950) 15,
No. 9, 9-12.

Stiles, W.E., “Use of Permeability Distribution in Waterflood Calculations,” Trans. AIME (1949) 186, 9-13.

Walsh, M.P. and L.W. Lake, 2003, A Generalized Approach to Primary Hydrocarbon Recovery, Elsevier.

Welge, H.J., 1952, “A Simplified Method for Computing Oil Recovery by Gas or Water Drive,‟ Trans. AIME 195 (1952), 91-98.

Wu, X., G.A. Pope, G.M. Shook, and S. Srinivasan, 2007, “Prediction of enthalpy production from fractured geothermal reservoirs using
partitioning tracers,” International Journal of Heat and Mass Transfer 51 (2007), 1453-1466.

Zheng, S-Y, P.W.M. Corbett, A. Ryseth, and G. Stewart, “Uncertainty in well test and core permeability analysis: a case study in fluvial
channel reservoirs, northern North Sea, Norway,” AAPG Bull, 84, No. 12 (December 2000), 1929-1954.

Appendix A. Calculating Sweep Efficiency Graphically from F- Curves


Shook and Forsmann (2005) show Sweep Efficiency can be calculated from F- data as:

q t
EV (t )   [1  F ( )]d (A1)
VP 0
Where the injection rate, q is assumed constant. Strictly speaking, F and  can only be determined at individual streamlines‟
breakthrough times, i. Therefore, in what follows we will use the following naming convention:

(ti) = i
F(ti) = Fi
ti = the ith streamline TOF

Equation A1 can be integrated as

qt i q ti
EV (t i )    F ( )d (A2)
VP VP 0

Integrating by parts gives:

qt i q  ti 
EV (t i )    Fi t i   t i dF  (A3)
VP VP  0 

or
qt i q  i dF 
EV (t i )    Fi t i   t i d  (A4)
V P V P  0 d 

which can be integrated to

qt i qt i qt dF
EV (t i )   Fi  i i (A5)
VP VP V P d

Wu et al. (2007) show the derivative of the F- curve is:


SPE 124625-PP 13

dF t*
 (A6)
d i t i

Where ti is the TOF of the ith streamline. Solving for ti gives

t* t*
ti   (A7)
dF F'
d

Inserting this into Equation A5 and using the identify qt*/Vp = 1 gives

1  Fi
EV (t i )   i (A8)
F'

Thus, at any point on the F- curve (and therefore, implicitly, at any given time), sweep efficiency can be directly estimated
from F and . The similarity between Eqn. A8 and the graphical solution to the Buckley Leverett equation (Welge, 1952)
suggests the F- curve can be interpreted as a generalized, 3-D flow curve, albeit without saturation effects (relative
permeability and residual saturations).

You might also like