energies

Article
Practical Aspects of Upscaling Geocellular Geological
Models for Reservoir Fluid Flow Simulations:
A Case Study in Integrating Geology, Geophysics,
and Petroleum Engineering Multiscale Data from the
Hunton Group
Benmadi Milad 1, * , Sayantan Ghosh 1 , Roger Slatt 1 , Kurt Marfurt 1 and Mashhad Fahes 2
1 School of Geosciences, University of Oklahoma, Norman, OK 73019, USA; Sayantan.Ghosh-1@ou.edu (S.G.);
rslatt@ou.edu (R.S.); kmarfurt@ou.edu (K.M.)
2 Mewbourne School of Petroleum & Geological Engineering, University of Oklahoma, Norman, OK 73019,
USA; mashhad.fahes@ou.edu
* Correspondence: Benmadi.Milad@gmail.com




Received: 22 February 2020; Accepted: 20 March 2020; Published: 1 April 2020


Abstract: Optimal upscaling of a high-resolution static geologic model that reflects the flow performance
of the reservoir is important for reasons such as time and calculation efficiency. In this study, we
demonstrate that honoring reservoir heterogeneity is critical in predicting accurate production and
reducing the time and cost of running reservoir flow simulations for the Hunton Group carbonate.
We integrated three-dimensional (3D) seismic data, well logs, thin sections, outcrops, multiscale
fracture studies, discrete fracture networks, and geostatistical methods to create a 100 × 150 × 1 ft
gridded representative geologic model. We calibrated our gridded porosity and permeability
parameters, including the evaluation of fractures, by history matching the simulated production rate
and cumulative production volumes from a baseline fine-scale model generated from petrophysical
and production data obtained from five wells. We subsequently reperformed the simulations using a
suite of coarser grids to validate our property upscaling workflow. Compared to our baseline history
matching, increasing the horizontal grid cell sizes (i.e., horizontal upscaling) by factors of 2, 4, 8, and 16
results in cumulative production errors ranging from +0.5% for two time (2×) coarser to +74.22% for
16× coarser. The errors associated with vertical upscaling were significantly less, i.e., ranging from
+0.5% for 2× coarser to +10.8% for 16× coarser. We observed higher production history matching
errors associated with natural fracture size. Results indicate that greater connectivity provided
by natural fracture length has a larger effect on production compared to the higher permeability
provided by larger apertures. We also estimated the trade-off between accuracy and run times using
two methods: (a) using progressively larger grid cell sizes; (b) applying 1, 5, 10, and 20 parallel
processes. Computation time reduction in both scenarios may be described by simple power law
equations. Observations made from our case study and upscaling workflow may be applicable to
other carbonate reservoirs.

Keywords: Upscaling; Error Modeling; Reservoir Modeling; Geological Modeling; Data Integration;
Flow Simulation; Seismic for flow simulation; Outcrops; Carbonate; DFN; Fracture Modeling

1. Introduction
Time is money. Therefore, despite the availability of unlimited computer resources, a reservoir
analyst may have a larger impact on field performance by quickly history matching dozens of wells

Energies 2020, 13, 1604; doi:10.3390/en13071604 www.mdpi.com/journal/energies

 Energies 2020, 13, 1604 2 of 27

using a more economical upscaled model (coarser model) than by slowly history matching a single
well using a higher accuracy fine-scale model. Upscaling is the process of reducing a large number of
the fine-scale grid blocks of a geological model to a smaller number of coarse-scale grid blocks while
retaining the underlying geological properties [1–4]. There are copious upscaling methods that include
arithmetic, harmonic, geometric, power law, pressure solver, and weighted averages algorithms, where
the choice depends on the rock property [4].
The necessity of the upscaling process arises from the limitation on computing capacity and
amount of available time required to process large geologic models. Although high cell resolution is
preferred, such bulky models containing millions of cells are prohibitive for iterative production history
matching, sensitivity analyses, production forecasting, reservoir optimization, uncertainty assessment,
proposing new or infill drilling locations, and risk analysis using a probabilistic approach [5–8]. When
a large number of discrete fractures is modeled, the model becomes even more cumbersome [9–11].
Therefore, upscaling is necessary in most cases. However, upscaling a fine-scale reservoir model must
be performed with care. Excessive upscaling leads to loss of the detailed heterogeneities in the geologic
model that directly affects the result accuracy [4].
The flow from thin bed conduits and thin bed baffles is particularly sensitive to inaccurate
upscaling. Santacruz et al. [12] simulated the flow of a leveed-channel system, assigning different
permeabilities to the thin slump deposits. They concluded that slump deposits acted as barriers to
waterflood penetration into the oil zones, whereby the lower permeability of the slumps significantly
reduced the production rate. In many reservoir simulation studies, upscaling is performed only in
the vertical direction, as the areal grid cell size of the reservoir model typically approximates the
heterogeneity provided by three-dimensional (3D) seismic data [13]. Furthermore, Ma et al. [13] found
a 50% error in the simulated production volumes after upscaling. Figure 1 illustrates the change in
production as a function of upscaling permeability.

A A’
A Permeability (mD)
1
0.01
0.0001 Production Rate
2009 2010 2011 2016
(STB/d)
Q_Upscaled 2 60 37 29 20
Upscaled 2 Q_Upscaled 1 60 23 19 14
Grid cell size 400x600x1
A’ Total number of cells 87,318 60 60
100x150x1ft 400x600x1ft actual production rate
Oil productionrate (STB/d)

50 50

40 40

30 30 % increase after upscaling 30

Upscaled 1 20 20
Grid cell size 100x150x1
Total number of cells 2,530,134 10 10

0 0
2009 2011 2012 2014 2015 2016
Date (Year)
Fine-Scale
Grid cell size 20x30x1
Total number of cells 13,292,228

Figure 1. Illustrations of permeability upscaling processes that lead to loss of the detailed heterogeneities
in the
ustrations of permeability geologic processes
upscaling model thatthat
directly
leadaffects theof
to loss accuracy of production
the detailed profile. The
heterogeneities in fine
the scale permeability
geologic model that directly affects the
profile. The fine scale permeability
model, which model, which
consists consists
of more thanof13
more thancells,
million 13 million
has cellcells,
sizeshas cellftsizes
of 20 of 20 ft in X-direction,
in X-direction, 30 ft in 30 ft in Y-direction
direction. This model was upscaled by enlarging the grid size (reducing the number of cells to 2,530,134)
Y-direction, and 1 ft in vertical (Z) direction. This model was upscaled by enlarging the grid size and averaging the permeabilities. Sim
d for the excessively upscaled model resulting in only 87, 000 cells. Note the permeability heterogeneities
(reducing the number of cells to 2,530,134) and averaging the permeabilities. Similar process was along A-A’ section are not the same, e
val shape. The right figure shows the change in production rate over time for the two upscaled models.
repeated for the excessively upscaled model resulting in only 87,000 cells. Note the permeability The fine scale model could not be
s handling 13 million cells is beyond their capacity. Notice the 30% increase in production rate after upscaling (Upscaled 1 to Upscaled 2).
heterogeneities along A-A’ section are not the same, especially in the black oval shape. The right figure
shows the change in production rate over time for the two upscaled models. The fine scale model
could not be run on our computers as handling 13 million cells is beyond their capacity. Notice the 30%
increase in production rate after upscaling (Upscaled 1 to Upscaled 2).
Energies 2020, 13, 1604 3 of 27

Meddaugh [14] examined the impact of vertical and areal upscaling on fluid flow, using multiple
geostatically generated porosity model realizations. He concluded that the changes in the cumulative
water injection are more sensitive to the number of realizations than to the level of vertical upscaling,
and that the recovery factor decreases when the areal upscaling increases. A challenge in upscaling
geologic reservoir model parameters is to determine a sufficiently coarse reservoir grid cell size [2,3,8]
without sacrificing the detail of the original structures, rock properties, layered nature of the reservoir,
and reservoir heterogeneity. Regardless of the upscaling process used, a proper production history
match using fluid flow simulation validates the upscaling process. The purpose of this study is,
therefore, to find the optimal level of upscaling for the Hunton carbonate oil wells, beyond the
production history match grid cell size, at which the most significant geological and petrophysical
properties and the simulation accuracy is retained (Figure 1). Moreover, this study demonstrates the
relative error magnitudes arising from the numerical uncertainty versus reservoir heterogeneity that it
is essential for modelers to understand the sources of error.
We begin our study with a review of the geology of the Hunton carbonate of Oklahoma as it
appears in the three study areas. We then describe our methodology, based on the well- established
workflow of defining and then history matching a finely gridded, high resolution model (Figure 1).
We construct a fine-grid model as our reference and use of it for simulating production from available
Hunton wells. We then use a series of homogenous models to simulate production; thus, removing the
effect of geologic heterogeneity while divulging the numerical errors associated with each subsequent
increase in grid cell sizes. Next, we show the effect of upscaling both the horizontal and vertical grid
cell properties on the computation run times and accuracy, constructing power law relations that can
be used to predict the run times for future models. We conclude with a summary of our results and
recommendations that the reservoir analysis requires a balance between the need for accuracy of a
given well simulation to the value of having less accurate results for multiple wells in the oil field.
The significance of this work lies in:
1. demonstrating seamless integration of lithologic, natural fracture, petrophysical, 3D seismic,
reservoir, and production data to obtain a comprehensive reservoir description;
2. dividing reservoir property upscaling errors (during reservoir simulation and history matching)
into that associated with numerical instability and that associated with geologic heterogeneity in a
quantifiable manner (i.e., through empirical equations) for different grid cell sizes;
3. quantifying the reservoir simulation time required for different number of parallel processes in
a predictable manner through empirical equations;
4. quantifying the simulation time vs. number of grid cells (or size) through empirical equations;
5. describing the effect of upscaling in different directions on reservoir simulation accuracy;
6. describing the effect of natural fracture size and aperture on production profile.

2. Geology of the Hunton Reservoir

The areas of study are in Pottawatomie, Pontotoc, and Murray Counties, Oklahoma (Figure 2).
We conducted field measurements of the Hunton Anticline exposure as well as the Ada and Fittstown
carbonate outcrops of the Chimneyhill subgroup to build a realistic fracture model for our upscaling
and simulation studies. The subsurface data (logs, core, and seismic) comes from a well located in
Pottawatomie County.
In the studied area (Figure 2), the Hunton Group Limestone is highly heterogeneous as inferred
from the Dykstra Parsons coefficient [15]. This quantity is used to quantify the level of heterogeneity
in a reservoir using permeability values from core data. Twenty-five Klinkenberg corrected core
permeability values (Figure 3) available in our study reveal a Dykstra Parson’s coefficient of 0.902
(calculated using Equation (1)) which represents a very high reservoir heterogeneity. Lake and Jensen
(1991) describe various other methods for the estimation of reservoir heterogeneity.
 surface with less than one degree dip [18, 19]. In the area of study (Figure 2), the stratigraphic thickness
ranges from 50 to 120 ft. Three primary lithologies of the Chimneyhill in the study area include
wackestone, mudstone, and mud-dominated wackestone [20]. The Hunton Group facies comprise a
series of progradational and aggradational parasequences (shallowing-upward cycles) [21]. Fractures
in the Hunton Group are considered as one of the key components to enhance the hydrocarbon
Energiesproduction [22]. Ghosh et al. [23] and Ghosh et al. [24] provide further details on the outcrop and
2020, 13, 1604 4 of 27
subsurface fractures, current tectonic stress regime, and paleo tectonic stress regimes in the study area
and surrounding areas in Oklahoma.

FigureFigure 2. Locations
2. Locations of of
thethe studiedareas.
studied areas. (a)
(a)Oklahoma
Oklahoma mapmapwith adjacent
with states.states.
adjacent (b) Data
(b)locations in
Data locations in
Pottawatomie, Pontotoc, and Murray Counties. Location of three-dimensional (3D) seismic,
Pottawatomie, Pontotoc, and Murray Counties. Location of three-dimensional (3D) seismic, subsurface subsurface
log and cores data in Pottawatomie County is indicated by the star. Yellow bars indicate outcrops in
log and cores data in Pottawatomie County is indicated by the star. Yellow bars indicate outcrops in
Pontotoc and Murray Counties. (c) Limits of the Pottawatomie County survey showing subsurface well
Pontotoc and Murray Counties. (c) Limits of the Pottawatomie County survey showing subsurface
locations.
well locations.

Dykstra Parsons Coefficient

100
y = 4.2804e-0.068x
R² = 0.9288
Permeability (mD)

10

1
K50=0.14 mD
0.1
K84.1=0.014 mD
0.01

0.001
0 20 40 60 80 100
Percent of samples having permeability greater than

Figure 3. Dykstra Parsons

Figure analysis for core
3. Dykstra permeability
Parsons analysismeasured
for core in this study.
permeability measured in this study.
The Hunton Group is a prolific oil and gas-producing reservoir in the Midcontinent [16].
The cumulative oil production is 290 Million Barrels of Oil (MMBO) and the cumulative gas production
is 5 Trillion Cubic Feet of Gas (TCFG) [17], with undiscovered 9 MMBO and 38 Billion Cubic Feet
(BCF) of oil and gas, respectively [16]. The Hunton Group carbonates were deposited during Late
Ordovician-Silurian-Early Devonian time in Oklahoma and is bounded by the underlying conformable
Sylvan Shale and the overlying unconformable Woodford Shale (Figure 4). The Hunton Group, including
the Chimneyhill subgroup, is associated with a shallow-warm water carbonate ramp of a gently sloping
surface with less than one degree dip [18,19]. In the area of study (Figure 2), the stratigraphic thickness
ranges from 50 to 120 ft. Three primary lithologies of the Chimneyhill in the study area include
wackestone, mudstone, and mud-dominated wackestone [20]. The Hunton Group facies comprise a
series of progradational and aggradational parasequences (shallowing-upward cycles) [21]. Fractures
in the Hunton Group are considered as one of the key components to enhance the hydrocarbon
production [22]. Ghosh et al. [23] and Ghosh et al. [24] provide further details on the outcrop and
subsurface fractures, current tectonic stress regime, and paleo tectonic stress regimes in the study area
and surrounding areas in Oklahoma.
K50 − K84.1 0.1429 − 0.0141
Dykstra Parsons coe f f icient = = = 0.902 (1)
K50 0.1429

where,
K50 = 50th percentile of permeability
K84.1 = 84.1th percentile of permeability
 Energies 2020, 13, x FOR PEER REVIEW 6 of 29

Energies 2020, 13, 1604 5 of 27

2
3 Figure 4. Stratigraphic column of the Chimneyhill subgroup of the Hunton Group carbonate in the study area (Pottawatomie County, Oklahoma) from a
Figure 4. Stratigraphic column of the Chimneyhill subgroup of the Hunton Group carbonate in the study area (Pottawatomie County, Oklahoma) from a subsurface
4 subsurface well. Tracks from left to right are: geologic time, stratigraphic units (Viola carbonate, Sylvan shale, Hunton Group carbonate, and Woodford
well. Tracks from left to right are: geologic time, stratigraphic units (Viola carbonate, Sylvan shale, Hunton Group carbonate, and Woodford shale), true vertical depth
5 shale), true vertical depth (ft), gamma ray, deep resistivity (RT90), photoelectric (PE), and sonic (DT) logs, neutron porosity (NPHI) and bulk density
(ft), gamma
6 ray,(RHOB),
deep resistivity (RT90),
electrofacies, photoelectric
core porosity, (PE), and sonic
core permeability, (DT) logs,borehole
core mineralogy, neutronimage,
porosity
and (NPHI)
lithology.and bulk density (RHOB), electrofacies, core porosity, core
permeability, core mineralogy, borehole image, and lithology.
Energies 2020, 13, 1604 6 of 27

3. Methodology

3.1. Heterogeneous Model

The general workflow includes the following steps:
1. creation of the geologic model including the discrete fracture network (DFN),
2. upscaling the geologic and DFN model for subsequent flow simulation,
3. production history matching of the fine grid baseline model (Upscaled 1 in Figure 1), and
repeated forward modeling of the production rate and volume at increasingly coarser grid sizes and
fracture dimensions.
Figures 5 and 6 describe the workflow used to populate the 3D volume, where we:
(1) construct the relationship between the acoustic impedance logs, lithology (using thin sections),
and porosity logs at well locations,
(2) derive core porosity and vertical/horizontal permeability relationships (derived using
laboratory experiments,
(3) calculate fluid saturation from well logs,
(4) invert the seismic data to estimate a compressional wave (P-wave) impedance volume,
(5) correlate the P-wave impedance to the well log electrofacies and porosity measurements to
construct a horizontal variogram to geostatistically construct 3D estimates of lithology and porosity,
(6) build DFN models based on the outcrop field measurements, and finally,
(7) upscale the fracture and matrix properties together for production simulations.
The original geologic model using a 20 × 30 × 1 ft geologic grid resulted in 13,292,228 cells
which exceeded the capabilities of our computer resources. We therefore upscaled the initial grid to
100 × 150 × 1 ft to form our baseline model, which was modified to match the production history. This
baseline model was then further upscaled to coarser 200 × 300 × 1 ft, 400 × 600 × 1 ft, 800 × 1200 × 1 ft,
and 1600 × 2400 × 1 ft grid cell sizes with a constant bed thickness of 1 ft for the horizontal upscaling
to compare difference in production with the 100 × 150 × 1 model. Doubling the grid cell size in
the X and Y directions (constant 1 ft bed thickness in Z direction) reduces the lateral heterogeneity
and modifies the production rate and cumulative production. To understand the effect of vertical
upscaling on production we coarsened the cell size to 100 × 150 × 2 ft, 100 × 150 × 4 ft, 100 × 150 × 8 ft,
and 100 × 150 × 16 ft. Finally, to understand the effect of natural fractures on production, we increased
the fracture lengths and widths by factors of two and four for the baseline 100 × 150 × 1 ft grid used in
history matching.
 Energies 2020, 13, x FOR PEER REVIEW 8 of 29

Energies 2020, 13, 1604 7 of 27

26

27
28 Figure 5. Methodology depicting
Figure the integration
5. Methodology of various
depicting thesources of data
integration offor building
various the static
sources reservoir
of data model. the static reservoir model.
for building
 and 1600 x 2400 x 1 ft grid cell sizes with a constant bed thickness of 1 ft for the horizontal upscaling
to compare difference in production with the 100 x 150 x 1 model. Doubling the grid cell size in the
X and Y directions (constant 1 ft bed thickness in Z direction) reduces the lateral heterogeneity and
modifies the production rate and cumulative production. To understand the effect of vertical
upscaling on production we coarsened the cell size to 100 x 150 x 2 ft, 100 x 150 x 4 ft, 100 x 150 x 8 ft,
and 100 x 150 x 16 ft. Finally, to understand the effect of natural fractures on production, we increased
Energies 2020, 13, 1604 8 of 27
the fracture lengths and widths by factors of two and four for the baseline 100 x 150 x 1 ft grid used
in history matching.

Figure 6. Workflow depicting history match. The reference fine grid has cells that measure ∆x = 20 ft,
Figure 6. Workflow depicting history match. The reference fine grid has cells that measure Δx = 20 ft,
∆yΔy
= =3030ft,ft,and
and ∆z
Δz ==11ft.ft.Flow
Flow
waswas simulated
simulated for
for the the coarser
coarser horizontally
horizontally andupscaled
and vertically vertically upscaled
models,
models,asaswell
well as forthe
as for themodels
models with
with upscaled
upscaled fracture
fracture lengthslengths and apertures.
and apertures.

3.2.3.2.
Homogenous
Homogenous Model
Model

WeWe constructed homogenous

constructed homogenous models
modelsin order to quantify
in order the production
to quantify errors, resulting
the production errors,from
resulting from
thethe upscalingprocesses,
upscaling processes, which
whichareare
associated withwith
associated reservoir rock heterogeneity.
reservoir The homogenous
rock heterogeneity. The homogenous
models were derived from the already-existing heterogeneous models using average values for the
models were derived from the already-existing heterogeneous models using average values for the
matrix and fracture parameters that were obtained from the fine grid 20 x 30 x 1 model (Table 1).
matrix The
andpartial
fracture parameters
differential that used
equations wereforobtained from
the reservoir the fine include
simulators × 30
grid 20the × 1 model
material balance(Table 1).
equation, flow equations for three-phase flow of oil, water, and gas, and formation and fluid
Table 1. Reservoir
compressibility simulation
equations. parameters
The finite differencefor is
thea homogenous models
numerical method used
that in allto
is used vertical
solve and horizontal
partial
upscaling and production simulation cases. Where: Φ = porosity; Kv = matrix
differential equations. However, the numerical method only gives an approximate solution to the permeability in vertical
discretized
direction;grid systemsaturation;
Sw: water of partial Kdifferential equations
I , KJ , KK : fracture resulting ininax,truncation
permeability error [3,25,26].
y, and z directions. Sigma factor:
Therefore, the production
link between the matrixerrors can be associated
and fracture properties with
thatboth reservoir
describes heterogeneity
fluid flow between andthe
numerical
matrix and the
errors.
dual-porosity/permeability model in a porous medium. If Sigma factor = 0, no communication between
We quantified the effect of truncation error (numerical error) in the different simulations, first
the matrix and natural fractures occurs.
by making a series of homogenous vertical and homogenous horizontal upscaled models with
different grid cell sizes Parameters
Matrix (horizontal upscaling: 100 x 150 x 1 ft, 200 x 300 x 1 ft,Fracture
Natural 400 x 600Parameters
x 1 ft, 800 x
1200 x 1 ft, and 1600 x 2400 x 1 ft; vertical upscaling: 100 x 150 x 2 ft, 100 x 150 x 4 ft, 100 x 150 x 8 ft,
Parameters
and 100 x 150 x 16Average Values
ft grid cell sizes). We then Units
performed the Parameters
calculations asAverage Values
shown below. We used Units
Φ
the 150 x 100 x 4 grid cell size
0.07 as an example:Fraction Φ 0.00009 Fraction
K Energies
in X direction 0.131REVIEW
2020, 13, x; doi: FOR PEER mD KI 1
www.mdpi.com/journal/energies mD
K in Y direction 0.131 mD KJ 1 mD
KV 0.0001 mD KK 1 mD
Sw 5 % Sigma factor 7.6 dimensionless

The partial differential equations used for the reservoir simulators include the material balance
equation, flow equations for three-phase flow of oil, water, and gas, and formation and fluid compressibility
equations. The finite difference is a numerical method that is used to solve partial differential equations.
However, the numerical method only gives an approximate solution to the discretized grid system of
partial differential equations resulting in a truncation error [3,25,26]. Therefore, the production errors
can be associated with both reservoir heterogeneity and numerical errors.
We quantified the effect of truncation error (numerical error) in the different simulations, first by
making a series of homogenous vertical and homogenous horizontal upscaled models with different
grid cell sizes (horizontal upscaling: 100 × 150 × 1 ft, 200 × 300 × 1 ft, 400 × 600 × 1 ft, 800 × 1200 × 1 ft,
and 1600 × 2400 × 1 ft; vertical upscaling: 100 × 150 × 2 ft, 100 × 150 × 4 ft, 100 × 150 × 8 ft, and
Energies 2020, 13, 1604 9 of 27

100 × 150 × 16 ft grid cell sizes). We then performed the calculations as shown below. We used the
150 × 100 × 4 grid cell size as an example:

heterog150×100×4 − heterog150×100×1
Horiz_upsc_prod_ err_ heterog150×100×4 = (2)
heterog150×100×1

homog150×100×4 − homog150×100×1
Horiz_upsc_prod_ err_ homog150×100×4 = (3)
homog150×100×1

Horizontal upscaling production error due to geologic heterogeneity =

(4)
Horiz_upsc_prod_ err_ heterog150×100×4 −horiz_upsc_prod_ err_ homog150×100×4

Horizontal upscaling production error due to numerical uncertainty =

(5)
Horiz_upsc_prod_ err_ homog150×100×4 , i.e., same as (3)
Percentage errors for all grid cell sizes are calculated with respect to the history match grid cell
size (i.e., 150 × 100 × 1 ft). Similar calculations may be performed for the error in vertical upscaling.
However, this limitation is overcome using percentage errors instead of actual errors as shown in the
equations above.

3.3. Estimation of Reservoir Parameters

Electrofacies form one of the key components of the models described in Figure 5. We begin with
Principal Component Analysis (PCA) of the electric logs. These values serve as input to subsequent
Self-Organizing Mapping (SOM) to construct electrofacies logs. We corroborate the geologic interpretation
of the electrofacies predictions using thin sections and borehole images. We constructed the relationship
between the porosity and vertical and horizontal permeabilities from cores. Porosity logs were
correlated with core porosity measurements which in turn are linked to the facies distribution
(Figure 7e). We estimated fluid saturation by applying Archie’s law (Equation (6)) on the well logs.
We measured porosities as well as vertical and horizontal permeabilities from core plugs facilitating
the construction of porosity-permeability relationships. Following previous work by Milad et al. [11],
Milad and Slatt [20], Milad et al. [27], we constructed vertical variograms from the well logs to populate
the petrophysical data spatially.
a Rw
Swn = m (6)
∅ Rt
where:
n: saturation exponent (n = 2.5 for this study)
m: cementation exponent (m = 1.75)
a: tortuosity factor (a = 1)
Rw: brine resistivity (Rw = 0.035 ohm.m)
Rt: true deep resistivity; ohm.m

3.4. Seismic Data

We used seismic data to delineate formation boundaries, to define major faults, and to provide a
volumetric estimation of compressional (P-wave)- and shear (S-wave) impedances. These impedances
were crucial in differentiating fossiliferous wackestone from dolomitized mudstone rocks. The low
frequency background model used in impedance inversion used five wells (Figures 2c and 7a) with
P-wave velocity, S-wave velocity, and density logs and ranged from the top of the Woodford surface to
the top of the Viola surface (Figure 4). Six common angle gathers ranging from 0o to 30o at 5o increments
were used for prestack inversion using © Hampson-Russell Software. Detailed descriptions about
our inversion methodology and its results are available in Milad et al. [28] and Milad [29]. The main
purpose of the inversion is to provide horizontal variograms for porosity distribution. Another purpose
Energies 2020, 13, 1604 10 of 27

is to find the major and minor directions of lithologic variation (Figure 7b), impedance (Figure 7c),
and the facies distribution (Figure 7d).

3.5. DFN Modeling

Figure 7f shows a DFN model for one of the four fracture sets. We considered four main parameters
in the creation of the fracture network: intensity, geometry, orientation, and aperture. Parameters
defining the fracture intensity and dimension can be found in Milad et al. [27]. The mean dip and mean
dip azimuths were assigned for each fracture set based on the borehole measurements. We assigned
the statistical distributions of the borehole-measured fracture apertures (calibrated to the maximum
aperture value observed in the core) to the model. We used the stochastic technique described by
Schlumberger [30] in © Petrel Software manual to create a DFN volume for the entire seismic cube with
a distribution guided by the fracture intensity map (grid cell intensity values) for each set, where the

Fracture porosity = fracture aperture/fracture spacing (7)

Natural fracture permeabilities were calculated using a cubic aperture (y) law

y3
Permeabiltiy = (8)
12

3.6. Upscaling
We follow Oda [31] to upscale the fracture porosity, the fracture sigma length in the x, y, and z
directions (spanned by J, K, and I grid indices), and the fracture permeability. Assuming Darcy’s Law
for laminar flow through an isotropic porous medium, Oda [31] shows how to upscale the geometry
and properties of the fractures within a given cell to construct the permeabilities of an equivalent grid
cell. Projecting the fracture isotropic permeability onto the fracture plane and then scaling it to the
ratio between the fracture volume and grid cell volume provides a net permeability tensor for each
grid cell. Eigenvectors of the net permeability tensor provide the principal permeabilities.
The simulation software requires two simulation cells to model a dual porosity/dual permeability
system where adjacent grid cells communicate through both the fractures and the matrix. This construct
allows one to define porosity and permeability where the transmissibilities are connected or coupled
to each other using a quantity called the sigma factor [30].

Effective permeability = fracture area/cell volume (9)

Fracture porosity = total fracture area *fracture aperture/cell volume (10)

3.7. Geocellular Gridding and History Matching

The grid cells are aligned with an orthogonal X, Y, Z Cartesian coordinate system. We used small
vertical cell sizes to accommodate the heterogeneities observed in the well logs. The relative grid size is
a function of the heterogeneity, where coarser spacing can be used if the reservoir properties are more
homogeneous in one direction (Figure 7b). We used proportional slicing between the top and base
correlation surfaces to divide the reservoir into appropriate layers. Based on P-impedance from seismic
inversion, we rotated the computational grid to align with the direction of the depositional trend (30 ft
in the major direction of North 15 East (N15E) [x] and 20 ft in the minor [y] direction) (Figure 7b).
The history matching process was undertaken to adjust the uncertain model parameters of fluid
reservoir properties, including well production rates, reservoir pressure and reservoir temperatures
(Table 2) without changing the properties of static geological models. Once the fluid model parameters
and well data constraints were refined and individual well production histories were matched in the
100 × 150 × 1 ft model, we used the same reservoir simulation parameters for the subsequent upscaled
Energies 2020, 13, 1604 11 of 27

models. Flow simulations of structured gridding models were performed on the discretized models
using the commercial Eclipse reservoir simulator.

a) b)

W3 W1

W4 Elevation depth (ft)

W5 W2

c) d)
P impedance
(g/cm3-ft/s) Wackestone
60000
Mudstone
30000 Mud-Wack.
10000

Figure 7. Results
trajectories. (b) M
seismic inversion
range: 15800 ft).
impedance map.
1 mile
e) NE-SW fracture s
f)
Porosity (ft3/ft3)
0.2

0.1

Figure 7. Results of fine scale model. (a) Wells showing four vertical and one horizontal well trajectories.
(b) Major and minor variogram directions derived from impedance maps of seismic inversion (major
direction: N15E, major range: 18,840 ft, minor direction: S75E, minor range: 15,800 ft). (c) Impedance
map from inversion. (d) Lithology derived from seismic impedance map. (e) Porosity maps from the
inversion-derived variogram. (f) Discrete fracture network (DFN) map of the North East-South West
(NE-SW) fracture set. All other sets have similar spatial relative density variation.
Energies 2020, 13, 1604 12 of 27

4. Results

4.1. Static Modeling

4.1.1. Fine Scale Geologic Models

Figure 7a shows the trajectories of five wells used in this study. Geologic models in Figure 7b–f
were obtained using the method described in Figure 5, i.e., by utilization of core, well log, outcrop,
and seismic data.

4.1.2. Coarse Scale Reservoir Simulation Models

Figure 8a shows a 3-dimensional porosity model that has been upscaled to 150 × 100 × 1 ft grid
from a 20 × 30 × 1 ft (Figure 7e) grid. Figure 8b shows the water saturation calculated from the well
logs, upscaled onto the 20 × 30 × 1 ft grid, and then subsequently upscaled onto the 150 × 100 × 1 ft
grid. Figure 8c,d show the upscaled combined fracture and matrix permeabilities.

a) b)
Water saturation (fraction)
Porosity (ft3/ft3) 1
0.15
0.8
0.12
0.6
0.09
0.06 0.4
0.03 0.2
0 0

1 mile

c) d)
Horizontal permeability (mD) Vertical permeability (mD)
Figure
1 0.1 100x150
0.001
0.01 (c) hori
0.00001
0.0001 0.0000001
Vertical

Figure 8. Static heterogeneous models upscaled to 100 × 150 × 1 ft grid cell size: (a) porosity, (b) water
saturation, (c) horizontal permeability, and (d) vertical permeability. Vertical exaggeration is 20 times.
Energies 2020, 13, 1604 13 of 27

4.2. Dynamic Simulation

Table 2 provides detailed information and properties of the reservoir.

Table 2. Reservoir simulation parameters and well data constraints. These data are kept the same for
all upscaling simulation cases. STB/d: Stock Tank Barrels per day; API: American Petroleum Institute.

Well Rate and Pressure Constraints Fluid Model Parameters

Well No Oil rate (STB/d ) Phases Gas, Oil, Water
Well 1 50 Min pressure 300 psi
Well 2 130 Max pressure 2000 psi
Well 3 20 Reference pressure 1860 psi
Well 4 40 Temp. 170.33 o F
Well 5 125 Gas Sp.gr 0.815
Oil gravity 45 API
Simulation period: 1976-January to 2016-December Bubble point pressure 1860 psi
Water salinity 30,000 ppm

4.2.1. Effect of Horizontal Upscaling

Table 3 shows total errors resulting from the truncation computation and geologic heterogeneity.
Table 4 shows numerical errors due to computation truncation from homogenous models. Table 5 shows
the effect of geologic heterogeneity on the effect of production using Equations (1–3) as mentioned
earlier. The effect of horizontal upscaling on production is highly variable.

Table 3. Cumulative production comparisons in horizontal upscaling cases for the heterogeneous
models. Well by well actual cumulative production, history-matched (100 × 150 × 1 grid) cumulative
production, and comparisons of larger grid cell sizes with history-matched cumulative production are
shown. Errors in the history match column are with respect to the actual cumulative production and
those in remaining columns are with respect to the history-matched production.

100 × 150 × 1
Actual 200 × 300 × 1 400 × 600 × 1 800 × 1200 × 1 1600 × 2400 ×
(History Match)
Cumulative Cumulative Cumulative Cumulative 1 Cumulative
Cumulative
Production Production Production Production Production
Production
(STB) Error Error Error Error
Error
Well 1 10,170 9626 (–5%) +131% +155% +345% +537%
Well 2 10,217 7542 (–26%) +62.6% +99% +261% +285%
Well 3 37,398 39,634 (6%) –4% +71% +72% +189%
Well 4 27,348 @ 1995 29,333 (7%) +22% +29% +100% +153%
Well 5 37,956 @ 2014 34,254 (–10%) +36% +44% +80% +106%
All wells
175,064 170,015 (0.5%) +28% +54% +120% +166%
combined
Energies 2020, 13, 1604 14 of 27

Table 4. Cumulative production comparisons in horizontal upscaling cases for the homogenous models.
Well by well actual cumulative production and comparisons of larger grid cell sizes with 100 × 150 × 1
ft history matched grid cell size are shown.

100 × 150 × 1 200 × 300 × 1 400 × 600 × 1 800 × 1200 × 1 1600 × 2400 × 1
Cumulative Cumulative Cumulative Cumulative Cumulative
Production Production Production Production Production
Error Error Error Error Error
Well 1 107.99% 16.57% 50.48% 85.95% 123.84%
Well 2 69.55% 19.12% 54.95% 91.25% 121.43%
Well 3 54.01% 11.91% 32.00% 67.21% 92.54%
Well 4 131.78% 12.78% 35.50% 66.05% 72.97%
Well 5 –36.87% 16.56% 50.30% 79.46% 108.21%
All wells
4.77% 14.02% 39.90% 75.22% 91.78%
combined

Table 5. Cumulative production comparisons in horizontal upscaling cases after subtracting the
numerical errors in Table 4 from total errors in Table 3 for 200 × 300 × 1, 400 × 600 × 1, 800 × 1200 × 1,
and 1600 × 2400 × 1 cell sizes.

100 × 150 × 1
Actual 200 × 300 × 1 400 × 600 × 1 800 × 1200 × 1 1600 × 2400 ×
(History Match)
Cumulative Cumulative Cumulative Cumulative 1 Cumulative
Cumulative
Production Production Production Production Production
Production
(STB) Error Error Error Error
Error
Well 1 10,170 9626 (–5%) 114.43% 104.52% 259.05% 413.16%
Well 2 10,217 7542 (–26%) 43.48% 44.05% 169.75% 163.57%
Well 3 37,398 39,634 (6%) –15.91% 39.00% 4.79% 96.46%
Well 4 27,348 @1995 29,333 (7%) 9.22% –6.50% 33.95% 80.03%
Well 5 37,956 @ 2014 34,254 (–10%) 19.44% –6.30% 0.54% –2.21%
All wells
175,064 170,015 (0.5%) 13.98% 14.10% 44.78% 74.22%
combined

For the total (numerical plus heterogeneity) error cases, Well 3 (Figure 9c) shows little variation
(-4%) in cumulative production (Table 3) between the 100 × 150 × 1 (history match) green line and the
next larger grid (200 × 300 × 1 ft). In other wells (Figure 9a,b,d,e), the next larger grid (200 × 300 × 1 ft)
shows a significant change (total error) in production rate (+22 to + 131%) (Table 3). The remaining
larger grid cell sizes (400 × 800 × 1 ft, 800 × 1600 × 1 ft, and 1600 × 2400 × 1 ft) induced significant
differences in cumulative production and production rates when compared to the history-matched
grid cell size (i.e., +22% to +537%) [Figure 9a–f, Table 3]. Note that progressively larger grid cell sizes
do not necessarily result in a progressively larger departure from the history match (green curve) for
individual wells. However, such a correlation can be observed for the cumulative production in all
wells combined (Table 3, Table 4, and Table 5). For example, the orange (800 × 1200 × 1 ft) and black
(1600 × 2400 × 1 ft) curves swap position from one well to the other (Figure 9a–e).
Energies 2020, 13, x FOR PEER REVIEW 7 of 29

grid cell sizes do not necessarily result in a progressively larger departure from the history match
(green curve) for individual wells. However, such a correlation can be observed for the cumulative
production
Energies in1604
2020, 13, all wells combined (Tables 3, 4, and 5). For example, the orange (800 x 1200 x 1 15
ft)ofand
27
black (1600 x 2400 x 1 ft) curves swap position from one well to the other (Figures 9a-9e).
35 35

Oil productionrate (STB/d)

30 30
Well 1
25 25
20 20
15 15
10 10
5 5
0 0
2006 2008 2011 2014 2016
Date (Year)
(a)

15 15
Oil productionrate (STB/d)

Well 2

10 10

5 5

0 0
2009 2010 2012 2013 2014 2016
Date (Year)

(b)
15 15
Oil productionrate (STB/d)

Well 3
10 10

5 5

0 0
1976 1984 1992 2000 2008 2017
Date (Year)

(c)
30 30
Oil productionrate (STB/d)

Well 4
20 20

10 10

0 0
1984 1989 1995 2000 2005 2011
Date (Year)
(d)
Figure 9. Cont.
 Energies 2020, 13, 1604 16 of 27
Energies 2020, 13, x FOR PEER REVIEW 8 of 29

80 80

Oil productionrate (STB/d)

70 70
60 Well 5 60
50 50
40 40
30 30
20 20
10 10
0 0
2009 2010 2012 2013 2014 2016 2017
Date (Year)
(e)
120 120
Oil productionrate (STB/d)

110
100 All wells 100
90
80 80
70
60 60
50
40 40
30
20 20
10
0 0
1975 1983 1991 2000 2008 2016
Date (Year)

(f)
100x150x1ft 200x400x1ft 400x600x1ft
800x1200x1ft 1600x2400x1ft actual production rate

Figure 9. Effect
Figure 9. Effectofof horizontal
horizontal upscaling
upscaling on production.
on production. (a–e) Rate(a)-(e) Rate for
production production for wells 1–5
wells 1–5 respectively.
respectively. (f) Production
(f) Production rates for allrates
wells.for all actual
Dots: wells.production
Dots: actual
rates;production rates;forProduction
Production rates different gridrates
cell for
sizes—green
different grid cellcurves: 100 × 150
sizes—green × 1 (history
curves: 100 xmatch);
150 x blue 200 × 300blue
curves:match);
1 (history × 1; curves:
red curve:
200400x ×300
600x×1;1; red
curve:orange curves:
400 x 600 x 1;800 × 1200
orange × 1; black
curves: 800curves:
x 1200 1600 × 2400curves:
x 1; black × 1. 1600 x 2400 x 1.
4.2.2. Effect of Vertical Upscaling
Table 3. Cumulative production comparisons in horizontal upscaling cases for the heterogeneous
Table 6Well
models. shows by total errors,cumulative
well actual Table 7 shows numerical
production, errors, and Table
history-matched (100 x8 150
shows
x 1 errors due to geologic
grid) cumulative
production, and comparisons of larger grid cell sizes with history-matched cumulative
heterogeneity for the vertical upscaling cases. For the total (numerical plus heterogeneity) error cases, production
are 10a–c,e–f
Figure shown. Errors
show in the history
little matchbetween
difference column are thewith respect
history to the(green
match actualcurves)
cumulative
andproduction
larger grid cell
and In
curves. those in remaining
addition, columns
essentially all are with respect
vertically to thematches
upscaled history-matched production.as reasonable history
may be considered
matches to the observed data (Figure 10a–c,e–f). Larger grid cell size always leads to a jump in the
100 x 150 x 1 1600 x
800 x 1200 x grid cell size
cumulative production (Table 6) and production rate compared to the base history-match
Actual (history 200 x 300 x 1 400 x 600 x 1 2400 x 1
of 100 × 150 × 1 ft. Vertical upscaling makes a noticeable difference in only well 4 1(Figure 10d) which
cumulative match) cumulative cumulative cumulativ
cumulative
exhibits a noticeable 17% deviation in production rate and cumulative production (Table 6) with the
production cumulative production production e
largest vertical grid cell size (16 ft). production
(STB) production error error productio
error
error n error
Well 1 10,170 9626 (–5%) +131% +155% +345% +537%
Well 2 10,217 7542 (–26%) +62.6% +99% +261% +285%
Well 3 37,398 39,634 (6%) –4% +71% +72% +189%
27,348 @
Well 4 29,333 (7%) +22% +29% +100% +153%
1995
37,956 @
Well 5 34,254 (–10%) +36% +44% +80% +106%
2014
All wells 170,015
175,064 +28% +54% +120% +166%
combined (0.5%)
Energies 2020,
Energies 13, 1604
2020, 13, x FOR PEER REVIEW 1017
of of
2927

10 10

Oil productionrate (STB/d)

8 Well 1 8

6 6

4 4

2 2

0 0
2006 2008 2011 2014 2016
Date (Year)
(a)
14 14
Oil productionrate (STB/d)

12 12
Well 2
10 10
8 8
6 6
4 4
2 2
0 0
2009 2010 2012 2013 2014 2016
Date (Year)

(b)
10 10
Oil productionrate (STB/d)

8 Well 3 8

6 6

4 4

2 2

0 0
1976 1984 1992 2000 2008 2017
Date (Year)
(c)
20 20
Oil productionrate (STB/d)

18 18
16 Well 4 16
14 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
1984 1989 1995 2000 2005 2011
Date (Year)
(d)
Figure 10. Cont.
Energies 2020, 13, 1604 18 of 27
Energies 2020, 13, x FOR PEER REVIEW 11 of 29

60 60

Oil productionrate (STB/d)

50 50
Well 5
40 40
30 30
20 20
10 10
0 0
2009 2010 2012 2013 2014 2016 2017
Date (Year)
(e)
60 60
Oil productionrate (STB/d)

50 50
All wells
40 40
30 30
20 20
10 10
0 0
1975 1983 1991 2000 2008 2016
Date (Year)

(f)
100x150x1ft 100x150x2ft 100x150x4ft
100x150x8ft 100x150x16ft actual production rate

Figure 10. Effect of vertical upscaling on production rate. (a–e) Production rate for individual wells.
Figure 10. Effect of vertical upscaling on production rate. (a)-(e) Production rate for individual wells.
(f)(f)Sum of production rates for all wells (field production). The difference in the field production dates
Sum of production rates for all wells (field production). The difference in the field production dates
is is
duedue different
to startstart
to different timetime
of production for these
of production for five wells.
these five Dots:
wells.actual
Dots:production rates; production
actual production rates;
rates for different
production ratesgrid cell sizes—green
for different grid cell sizes—green × 150 ×100
curves: 100 curves: 1 (approximate history match);
x 150 x 1 (approximate historyblue curves:
match);
100blue× 150 × 2; 100
curves: red xcurve: 100
150 x 2; × curve:
red 150 × 4; orange
100 x 150 xcurves: × 150 ×100
100curves:
4; orange 8; xblack
150 xcurves:
8; black100 × 150100
curves: × 16
x of
cumulative production for individual wells.
150 x 16 of cumulative production for individual wells.

Table 6. Cumulative production comparisons in vertical upscaling cases for the heterogeneous models
Table 6. Cumulative production comparisons in vertical upscaling cases for the heterogeneous
showing total (numerical uncertainty plus geologic heterogeneity). Well by well actual cumulative
models showing total (numerical uncertainty plus geologic heterogeneity). Well by well actual
production, history match (100 × 150 × 1) cumulative production, and comparisons of larger grid cell
cumulative production, history match (100 x 150 x 1) cumulative production, and comparisons of
size with history match cumulative production are shown. Errors in the history match column are with
larger grid cell size with history match cumulative production are shown. Errors in the history match
respect to the actual cumulative production and those in remaining columns are with respect to the
column are with respect to the actual cumulative production and those in remaining columns are with
history
respectmatch
to the(i.e., notmatch
history actual(i.e.,
production)
not actualcase (100 × 150
production) × 1).
case (100 x 150 x 1).

100 100 × 150

x 150 x 1× 1
Actual 100 × 150 × 2 100 × 150 × 4 100 × 150 × 8 100 × 150 × 16
(History Match)100 x 150 x 100 x 150 x 100 x 150 x
Cumulative
Actual (history Cumulative Cumulative 100 Cumulative
x 150 x 8 Cumulative
Cumulative 2 4 16
Production Production Production Production Production
cumulative match)
Production cumulative
(STB) Error
cumulative Error
cumulative Error Error
cumulative
production Error
cumulative production
production production production
Well 1 10,170
(STB) 9626 (–5%)
production +2.3% +3.8% +4.9%
error +7.6%
error error error
Well 2 10,217 error
7542 (–26%) +6.7% +8.7% +9.4% +10.9%
Well
Well 31 10,170
37,398 962639,634
(–5%)(6%) +4.4%
+2.3% +5.7%
+3.8% +5.7%
+4.9% +8.3%
+7.6%
Well
Well 42 10,217
27,348 @ 1995 754229,333
(–26%)(7%) +4.3%
+6.7% +8.5%
+8.7% +14.6%
+9.4% +17.1%
+10.9%
Well
Well 53 37,398
37,956 @ 2014 39,634 (6%)
34,254 (–10%) +0.5%
+4.4% +2.6%
+5.7% +0.18%
+5.7% +3.6%
+8.3%
All wells 27,348 @
Well 4 175,064 170,015
29,333 (0.5%)
(7%) +3.4%
+4.3% +6.0%
+8.5% +8.0%
+14.6% +10.8%
+17.1%
combined 1995
37,956 @ 34,254 (–
Well 5 +0.5% +2.6% +0.18% +3.6%
2014 10%)
Energies 2020, 13, 1604 19 of 27

Table 7. Cumulative production comparisons in vertical upscaling cases for the homogenous models
showing errors due to numerical uncertainty. Well by well actual cumulative production and
comparisons of larger grid cell sizes with (100 × 150 × 1 grid) cumulative production are shown.

100 × 150 × 1 100 × 150 × 2 100 × 150 × 4 100 × 150 × 8

Cumulative Cumulative Cumulative Cumulative
Production Error Production Error Production Error Production Error
Well 1 108.02% 0.00% –0.07% 0.18%
Well 2 69.55% 0.09% 0.25% 0.59%
Well 3 54.01% –0.05% –0.09% 0.12%
Well 4 131.78% 0.01% 0.27% 0.86%
Well 5 –36.87% –0.58% –1.96% –1.59%
All wells combined 4.77% –0.07% –0.17% 0.21%

Table 8. Cumulative production comparisons in vertical upscaling cases after subtracting the numerical
errors in Table 7 from total errors in Table 6 for 100 × 150 × 2, 100 × 150 × 4, 100 × 150 × 8, and 100 ×
150 × 16 cell sizes.

100 × 150 × 1
Actual 100 × 150 × 2 100 × 150 × 4 100 × 150 × 8 100 × 150 × 16
(History Match)
Cumulative Cumulative Cumulative Cumulative Cumulative
Cumulative
Production Production Production Production Production
Production
(STB) Error Error Error Error
Error
Well 1 10,170 9626 (–5%) 2.30% 3.87% 4.72% 7.60%
Well 2 10,217 7542 (–26%) 6.61% 8.45% 8.81% 10.90%
Well 3 37,398 39,634 (6%) 4.45% 5.79% 5.58% 8.30%
Well 4 27,348 @ 1995 29,333 (7%) 4.29% 8.23% 13.74% 17.10%
Well 5 37,956 @ 2014 34,254 (–10%) 1.08% 4.56% 1.77% 3.60%
All wells
175,064 170,015 (0.5%) 3.47% 6.17% 7.79% 10.80%
combined

4.2.3. Comparison
The difference in production rates from vertical upscaling is not nearly as dramatic as that observed
Energies 2020, 13, x FOR PEER REVIEW 13 of 29
from horizontal upscaling (Figure 11a,b), though both follow a logarithmic relationship.

3402 22176 87318 361746 2530134

(1600x2400x1) (800x1200x1) (400x600x1) (200x300x1) (100x150x1)

(a)

Figure 11. Cont.

296322
(100x150x16)
547056
 3402 22176 87318 361746 2530134
Energies 2020, 13, 1604 (1600x2400x1) (800x1200x1) (400x600x1) (200x300x1) (100x150x1) 20 of 27

(a)

296322
(100x150x16)
547056
(100x150x8)
1048524
(100x150x4)
2028666
2530134
(100x150x2)
(100x150x1)

(b)
Figure 11. Comparison between vertical and horizontal upscaling production error. Total errors that
Figure 11. Comparison between vertical and horizontal upscaling production error. Total errors that
are associated with numerical errors and rock heterogeneity due to upscaling (orange), numerical
are associated with numerical errors
(uncertainty) error resulting from and rock heterogeneity
the truncation due(black),
computational errors to upscaling (orange),
and upscaling errors numerical
(uncertainty) error resulting from the truncation computational errors (black), and upscaling errors
(rock heterogeneity) resulting from changes in grid cell sizes (green). Some orange points are
overlapping black points in the upper plot (a) and black points are overlapping each other on the
(rock heterogeneity) resulting from changes in grid cell sizes (green). Some orange points are
overlapping black points in the upper plot (a) and black points are overlapping each other on the lower
plot (b). a) Semilog plot depicting logarithmic best-fit equations (inversely proportional) for the three
horizontal upscaling curves (orange, black, and green). Note numerical production errors due to the
computational truncation errors (black) are higher than the production errors due to the upscaling of
geologic heterogeneity (green) for grid cell sizes. The exact error values for the horizontal upscaling
are found in the last columns (“all wells combined”) in Table 3, Table 4, and Table 5. b) Semilog plot
depicting logarithmic best-fit equations (inversely proportional) for the total errors and numerical errors
vertical upscaling. Note numerical errors with the vertical upscaling is very low (black). Note much
lower numerical errors with vertical upscaling (black) and higher errors with horizontal upscaling
(black). The exact error values for the vertical upscaling are found in the last columns (“all wells
combined”) in Table 6, Table 7, and Table 8.

4.2.4. Effect of Fracture Size

Table 9 and Figure 12a–f show that increase in aperture has a lesser effect on production than
the length. For example, Well 1 shows a difference of 11% in cumulative production compared to the
history match using 2x aperture, and 301% using 2x length (Table 9, also see Figure 12a solid green,
solid blue, and solid red curves). Well 1 also shows 190% increase compared to the history match using
4x aperture and 673% increase using 4x length (Table 9, also see Figure 12a solid green, broken blue,
broken red curves). For Well 2, 2x fracture aperture results in a 63% increase, while 2x fracture length
results in 861% increase in production. Similarly, for 4x increase in aperture and length, the increase is
169% and 1595% in production respectively. With 4x aperture and 4x length, the jump in production
rate with respect to the respective 2x cases is much higher than the difference between the respective 2x
and history match cases. The remainder of the wells show a similar increase for 2x cases. The 4x values
indicated by the arrows in Figure 12c–e (constant values) should be ignored as the model boundaries
were most likely reached. In addition, the cumulative production of the whole field for 4x aperture
and length (arrows in Figure 12f) should be ignored as that includes all five wells.
 upscaling of geologic heterogeneity (green) for grid cell sizes. The exact error values for the horizontal
upscaling are found in the last columns ("all wells combined") in Tables 3, 4, and 5. b) Semilog plot
depicting logarithmic best-fit equations (inversely proportional) for the total errors and numerical
errors vertical upscaling. Note numerical errors with the vertical upscaling is very low (black). Note
much lower numerical errors with vertical upscaling (black) and higher errors with horizontal
Energies 2020, 13, 1604
upscaling (black). The exact error values for the vertical upscaling are found in the last columns ("all 21 of 27
wells combined") in Tables 6, 7, and 8.

Table 9. Cumulative production comparisons for history match, 2x, and 4x fracture aperture and
4.2.4. Effect of Fracture Size
fracture lengths. Errors in the history match column are with respect to the actual cumulative production
andTable
those9inand
theFigure 12a-fcolumns
remaining show that
are increase in aperture
with respect has a match
to the history lesser effect onactual)
(i.e., not production than the
production
length. For×example,
case (100 150 × 1). Well 1 shows a difference of 11% in cumulative production compared to the
history match using 2x aperture, and 301% using 2x length (Table 9, also see Figure 12a solid green,
solid blue, and solid 100 × 150 × 1 2X Fracture 4X Fracture 2X Fracture 4X Fracture
Actualred curves). Well 1 also shows 190% increase compared to the history match
(History Match) Aperture Aperture Length Length
using 4x apertureCumulative
and 673% increase using 4x length (Table Cumulative
9, also see Figure 12a solid green, broken
Cumulative Cumulative Cumulative Cumulative
Production
blue, broken red curves). For Well 2, 2x fracture
Production aperture results
Production in a 63% increase,
Production Productionwhile Production
2x fracture
(STB)
length results in 861% increase inError Error
production. Similarly, for 4xError Error
increase in aperture Error the
and length,
Well 1 is 169% 10,170
increase +11% With 4x
9626 (–5%) respectively.
and 1595% in production +190% +301%
aperture and 4x length, the+673%
jump in
production
Well 2 rate with
10,217respect to7542 respective 2x+63%
the(–26%) cases is much+169%
higher than+861%
the difference+1596%
between
the respective
Well 3 2x37,398
and history match
39,634cases.
(6%) The remainder
+224% of the wells show +642%
642% a similar increase for 2x
642%
cases. The
Well 4 4x values indicated
27,348 @ 1995 by the arrows
29,333 (7%) in Figure
+86% 12c-12e (constant
470% values)
+470%should be ignored
470% as
the model boundaries were most likely reached. In addition, the cumulative production of the whole
Well 5 37,956 @ 2014 34,254 (–10%) +37% 480% +539% 555%
field for 4x aperture and length (arrows in Figure 12f) should be ignored as that includes all five wells.
40 40
Oil productionrate (STB/d)

35 35
30 Well 1 30
25 25
20 20
15 15
10 10
5 5
0 0
2006 2008 2011 2014 2016
Date (Year)

(a)
120 120
Oil productionrate (STB/d)

100 100
Well 2
80 80
60 60
40 40
20 20
0 0
2009 2011 2012 2013 2015 2016
Energies 2020, 13, x FOR PEER REVIEW Date (Year) 15 of 29

(b)
Oil productionrate (STB/d)

20 20
16 Well 3 16
12 12
8 8
4 4
0 0
1976 1984 1992 2000 2008 2017
Date (Year)

(c)
45 Figure 12. Cont. 45
Oil productionrate (STB/d)

40 40
35 Well 4 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
1984 1989 1995 2000 2005 2011
Date (Year)
(d)
140 140
STB/d)

120 120
 12 12

Oil productionr
8 8
4 4
0 0
1976 1984 1992 2000 2008 2017
Energies 2020, 13, 1604 22 of 27
Date (Year)

(c)
45 45

Oil productionrate (STB/d)

40 40
35 Well 4 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
1984 1989 1995 2000 2005 2011
Date (Year)
(d)
140 140
Oil productionrate (STB/d)

120 120
100 100
80 80
60 Well 5 60
40 40
20 20
0 0
2009 2010 2012 2013 2014 2016 2017
Date (Year)
(e)
300 300
Oil productionrate (STB/d)

250 All wells 250

200 200
150 150
100 100
50 50
0 0
1975 1983 1991 2000 2008 2016
Date (Year)
(f)
history match (100x150x1ft grid size) 2x fracture aperture (100x150x1ft grid size)
2x fracture length (100x150x1ft grid size) 4x fracture aperture (100x150x1ft grid size)
4x fracture length (100x150x1ft grid size) actual production rate

Figure 12. Effect of fracture aperture and length on production rate showing history match curves, 2x,
Figure 12. Effect of fracture aperture and length on production rate showing history match curves, 2x,
and 4x fracture length and aperture. Arrows indicate curves, which may be considered as incorrect
and 4x fracture length and aperture. Arrows indicate curves, which may be considered as incorrect
due to constant production rates over long periods because model boundaries were probably reached
due to constant production rates over long periods because model boundaries were probably reached
due to high connectivity. (a) and (b) show all curves correctly. (c), (d), and (e) show history match, 2x
due to high connectivity. (a) and (b) show all curves correctly. (c), (d), and (e) show history match, 2x
length, and 2x apertures correctly but not 4x. Dots: actual production rates, Solid green: history match,
length, and 2x apertures correctly but not 4x. Dots: actual production rates, Solid green: history match,
Solid red: 2x fracture length, Broken red: 4x fracture length; Solid blue: 2x fracture aperture, Broken
Solid red: 2x fracture length, Broken red: 4x fracture length; Solid blue: 2x fracture aperture, Broken
blue: 4x4x
blue: fracture aperture.
fracture aperture.

4.2.5. Run Time and Accuracy

There is a fourth order negative relationship between solution time vs. grid cell size (Figure 13a),
and solution time vs. number of processes (Figure 13b). In other words, using five parallel processes
does not reduce runtime to one-fifth of the time taken by one processor. For example, for the history
match case, five processors consume nearly 13 hours which is one-third, rather than one-fifth, the time
taken by a single processor. For the 200 × 300 × 1 grid cell size, this ratio is different as for the remaining
grid cell sizes. The time-gap closes with an increase in the number of processes (Figure 13b).
Energies 2020, 13, 1604 23 of 27
Energies 2020, 13, x FOR PEER REVIEW 17 of 29

Elapsed time for flow simulation (hours) 100.000

y = 1E-07x1.3706
1 Process R² = 0.9812
y = 1E-07x1.2512
5 Processes
10.000 R² = 0.9972
10 Processes y = 9E-07x1.0482
R² = 0.9944
y = 1E-06x1.0334
1.000 20 Processes
R² = 0.9956

0.100

0.010

0.001
1000 10000 100000 1000000 10000000
Number of grid cells
(a)

40
Elapsed time for flow simulation

35
y = 35.836x-0.746
2530134 (100x150x1) R² = 0.9429
30
y = 6.4194x-0.873
25 361746 (200x300x1) R² = 0.9533
(hours)

y = 0.7496x-0.664
20 87318 (400x600x1) R² = 0.8739

15 y = 0.0607x-0.368
22176 (800x1200x1) R² = 0.8565
10
5
0
0 5 10 15 20 25
Number of grid cells
(b)
Figure
Figure 13.13.Time
Timerequired
requiredfor forflow-simulation
flow-simulation completion.
completion. (a)(a)Log-log
Log-logplot
plotdepicting
depicting simulation
simulation time time
required using a different number of grid cells (or grid cell size) while applying different
required using a different number of grid cells (or grid cell size) while applying different numbers numbers of of
processes:
processes: a single(blue),
a single (blue),five
five(orange),
(orange),ten
ten(red),
(red),twenty
twenty (black)
(black) processes.
processes. All
Allcurves
curveswerewerebest
bestfitfitbyby a
a power
power law.law.
(b) (b) Plot
Plot (both
(both x andy-axes
x and y-axesare
arelinear)
linear) depicting
depicting simulation
simulationtime
timeusing
using different number
different number
of of processes
processes (curvesdepicting
(curves depictingsingle
singlegrid
grid cell
cell sizes)
sizes) while
while using
usingvarious
variousgrid
gridcell
cellsizes:
sizes:100 x 150
100 × 150 x 1× 1
(green), 200 x 300 x 1 (blue), 400 x 600 x 1 (red), and 800 x 1200 x 1 (orange). All curves
(green), 200 × 300 × 1 (blue), 400 × 600 × 1 (red), and 800 × 1200 × 1 (orange). All curves were best were best fit fit
by negative power law equations.
by negative power law equations.

5. Discussion
The aforementioned data integration at different scales from core to seismic volume were used to
5. Discussion
develop a high-resolution stratigraphic framework and complex structures to build detailed fine grid
Energies 2020, 13, 1604 24 of 27

cell size static geological models consisting of lithofacies, natural fractures, porosities, permeabilities,
and saturation. The resulting detailed (20 × 30 × 1 ft grid cell size) reservoir model was upscaled to
100 × 150 × 1 ft grid cell size for flow (reservoir) simulation. We used a dual porosity/dual permeability
model for higher accuracy even though the calculations require higher time for upscaling and
flow simulation.
A non-linear relationship between the grid cell size (and consequently the number of grids)
and run duration was observed. The runtime gap closes with larger grid cell size as the number of
grid cells decreases. Therefore, if the grid cell size is large enough, parallel processes may not be
required. The increase in grid cell size, however, results in unacceptably large errors in the simulation
results. Therefore, the optimal grid cell size and the run time taken based on the number of available
processes requires optimization. Comparison between vertical and horizontal upscaling shows greater
error with horizontal upscaling. In this study, the vertical grid cell can be reasonably upscaled to
100 × 150 × 2 ft or 100 × 150 × 4 ft without incurring more than 6% error compared to the history match
(100 × 150 × 1 ft). Larger vertical grid cell sizes, i.e., up to 10 ft may be used without incurring more
than 11% error. In other words, a reasonable history match and forward modeling can be obtained
with 1–10 ft vertical grid cell size showing that thin beds do not affect production in the current study.
Increase in lateral grid cell size, however, results in a substantial total error (0.5% to 166%) that includes
upscaling processes and numerical errors. The magnitude of numerical truncation (i.e., numerical
uncertainty or instability) errors ranges from 4.77% to 91.78% and is inversely proportional to the
number of grid cells, or directly proportional to the grid cell size, in the model. Thus, the effect of
numerical errors can be eliminated by decreasing the grid cell sizes in the model. These results agree
with Ezekwe [3] and Lantz [25] where the numerical errors increase with increases in the areal grid cell
sizes. Increase in lateral grid cell size due to upscaling results in errors in the range of 0.5%–74.2%
compared to the vertical grid cell sizes errors which are in the range of 0.5%–10.8%. This observation is
counterintuitive, given more frequent changes in rock properties in the vertical direction than in the
lateral directions. However, it is also noteworthy that the vertical cell size is much smaller compared
to horizontal cell size. Moreover, each step in vertical upscaling increases the grid cell volume by a
factor of two, while each step in horizontal upscaling increases grid cell volume by a factor of four
(i.e., twice in the x-direction and twice in the y-direction). Based on the overall production, a lateral
grid cell size of more than the minimum required for successful simulation is not recommended.
The equation in Figure 11, i.e., the one defining a monotonous (logarithmic) increase in error with
grid cell size (or decrease in the number of grid cells), is presented only for reference and likely to vary
in different areas based on geology. The error can also decrease with an increase in grid cell size as
observed with individual wells. However, the equation in Figure 13 (grid cell size vs. time), i.e., power
law relationship between grid cell size and solution time, is likely to hold true at other places in other
geological settings.
Regarding the effect of natural fracture geometry on production, natural fracture length has a
higher impact on the production compared to the aperture. Therefore, assuming a constant natural
fracture intensity or surface area, greater production may be expected with longer natural fractures.
This observation likely results from a higher reservoir connectivity through longer natural fractures.
Since fracture length can affect reservoir production significantly [32], its accurate estimation is
paramount, even though it is very difficult to achieve [10]. An increase in natural fracture aperture,
on the other hand, does not result in higher natural fracture connectivity but still increases production.
This observation indicates that natural fracture conductivity is not infinite. Moreover, constant high
production rates with 4x apertures imply a shift from finite conductivity (1x and 2x aperture cases) to
infinite conductivity leading to model boundary dominated flow.

6. Conclusions
This study combines several geologic, geophysical, and engineering data and parameters
methodically to establish the feasibility of such effort in understanding simulation accuracy and
Energies 2020, 13, 1604 25 of 27

practicability. We demonstrated that it is essential for modelers to understand the sources of error
(i.e., numerical uncertainty vs. reservoir heterogeneity) and acquire ability to estimate the relative
error magnitudes arising from these sources. Additionally, we found that vertical upscaling is a better
option than horizontal upscaling for preservation of computational accuracy. For horizontal upscaling,
we showed that the lateral grid cell size should be kept at the original bin size (~110 × 110 ft) of the
seismic data used in its construction in order to balance runtime and accuracy. Moreover, reservoir
connectivity attributable to natural fracture length was found to have a significantly greater impact
on the production response compared to natural fracture aperture. A less intuitive phenomenon
(i.e., overhead of inter process communication) causing a non-linear relationship between the number
of parallel processes and the total simulation time was demonstrated. Our workflow of using a fine
grid and comparing the history match to those of a coarser grid exhibits a means of estimating the
accuracy of a given simulation.

Author Contributions: For research articles with several authors, a short paragraph specifying their individual
contributions must be provided. The following statements should be used “Conceptualization, B.M., S.G. and
R.S.; methodology, B.M. and S.G.; software, B.M.; validation, B.M., S.G., K.M. and M.F.; formal analysis, B.M. and
S.G.; investigation, B.M.; resources, B.M.; data curation, B.M.; writing—original draft preparation, B.M. and S.G.;
writing—review and editing, S.G., R.S., K.M. and M.F.; visualization, B.M. and M.F.; supervision, R.S.; project
administration, B.M. and R.S.; funding acquisition, R.S. All authors have read and agreed to the published version
of the manuscript.”, please turn to the CRediT taxonomy for the term explanation. Authorship must be limited to
those who have contributed substantially to the work reported.
Funding: This research was funded by STACK-MERGE-SCOOP- consortium members at the Institute of Reservoir
Characterization at the University of Oklahoma. Check carefully that the details given are accurate and use the
standard spelling of funding agency names at https://search.crossref.org/funding, any errors may affect your
future funding.
Acknowledgments: Thanks to the STACK-MERGE-SCOOP- consortium members at the Institute of Reservoir
Characterization at the University of Oklahoma for their generous funding and support. The authors are greatly
thankful to Abeed Awotunde for his critical reviews and comments. I would like to thank the late Roger Slatt for
his extraordinary guidance and support during my Ph.D. study and SLATT_GRAD. We thank Schlumberger for
providing Petrel, Techlog, and Eclipse and CGG Geosolutions for providing Hampson-Russell software for use in
research and education. We also used software generated by the OU AASPI consortium. We would like to thank
the reviewers for their great comments and feedback.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. King, M.J.; MacDonald, D.G.; Todd, S.P.; Leung, H. Application of Novel Upscaling Approaches to the
Magnus and Andrew Reservoirs. In Proceedings of the European Petroleum Conference, The Hague,
The Netherlands, 20–22 October 1998. [CrossRef]
2. Stern, D. Practical aspects of scaleup of simulation models. J. Pet. Technol. 2005, 57, 74–81. [CrossRef]
3. Ezekwe, N. Petroleum Reservoir Engineering Practice; Pearson Education, Inc.: New York, NY, USA, 2010.
4. Mehmood, S.; Awotunde, A.A. Sensitivity-based upscaling for history matching of reservoir models. Pet. Sci.
2016, 13, 517–531. [CrossRef]
5. Durlofsky, L.J. Upscaling and Gridding of Fine-Scale Geological Models for Flow Simulation. In Proceedings
of the 8th International Forum on Reservoir Simulation Iles Borrmees, Stresa, Italy, 20–24 June 2005;
Volume 2024, pp. 1–59.
6. Adamson, G.; Crick, M.; Gane, B.; Gurpinar, O.; Hardiman, J.; Ponting, D. Simulation throughout the life of a
reservoir. Oilfield Rev. 1996, 2, 16–27.
7. Olalotiti-Lawal, F.; Onishi, T.; Datta-Gupta, A.; Fujita, Y.; Hagiwara, K. Post-combustion CO2 EOR
development in a mature oil field: Model calibration using a hierarchical approach. In Proceedings
of the SPE Annual Technical Conference and Exhibition, Denver, CO, USA, 5–7 October 2020.
8. Avansi, G.; Rios, V.; Schiozer, D.J. Numerical tuning in reservoir simulation: It is worth the effort in practical
petroleum applications. J. Braz. Soc. Mech. Sci. Eng. 2019, 41, 1–21. [CrossRef]
9. Hui, M.H.; Karimi-Fard, M.; Mallison, B.; Durlofsky, L.J. A general modeling framework for simulating
complex recovery processes in fractured reservoirs at different resolutions. SPE J. 2018, 23, 598–613. [CrossRef]
Energies 2020, 13, 1604 26 of 27

10. Ghosh, S. Integrated Studies on Woodford Shale Natural Fracture Attributes, Origin, and Their Relation to
Hydraulic Fracturing. Ph.D. Thesis, The University of Oklahoma, Norman, OK, USA, 2017.
11. Milad, B.; Ghosh, S.; Slatt, R. Comparison of rock and natural fracture attributes in karsted and non-karsted
Hunton Group Limestone: Ada and Fittstown area, Oklahoma: The Shale Shaker. J. Okla. City Geol. Soc.
2018, 69, 70–86.
12. Santacruz, C.; Correa-Correa, H.; Slatt, R.M. 3D geological modeling and performance simulation of
a leveed-channel outcrop with application to deepwater leveed-channel reservoirs. AAPG Search and
Discovery Article 2012, 50728. Available online: http://www.searchanddiscovery.com/pdfz/documents/2012/
50728santacruz/ndx_santacruz.pdf.html (accessed on 22 February 2020).
13. Ma, E.; Ryzhov, S.; Wang, Y.; Al-houti, R.; Dashti, L.; Ali, F.; Ibrahim, M. SPE 164187 Answering the Challenge
of Up-scaling a 900 Million-Cells Static Model to a Dynamic Model—Greater Burgan Field. In Proceedings
of the Kuwait, Paper SPE 164187 presented at the MEOS conference, Manama, Bahrain, 10–13 March 2013;
pp. 1–11. [CrossRef]
14. Meddaugh, W.S. Reservoir Modeling for Mature Fields—Impact of Work Flow and Upscaling on Fluid-Flow
Response. In Proceedings of the SPE Europec/EAGE Annual Conference and Exhibition, Vienna, Austria,
12–15 June 2006.
15. Dykstra, H.; Parsons, R.L. The Prediction of Oil Recovery by Water Flood. Secondary Recovery of Oil in the United
States: Principles and Practice; American Petroleum Institute: New York, NY, USA, 1950.
16. Gaswirth, S.B.; Higley, D.K. Petroleum Systems and Assessment of Undiscovered Oil and Gas in the Anadarko
Basin Province, Colorado, Kansas, Oklahoma, and Texas: USGS Province 58; U.S. Geological Survey: Reston, WV,
USA, 2014.
17. IHS Energy Group. Petroleum information/Dwights petroROM Rocky Mountain Region production data; IHS
Energy Group: Englewood, CO, USA, 2011; CD-ROM.
18. Amsden, T.W. Hunton Group, Late Ordovician, Silurian, and Early Devonian in the Anadarko Basin of Oklahoma;
Oklahoma Geological Survey, University of Oklahoma: Norman, OK, USA, 1975.
19. Stanley, T.M.; Rottmann, K. Stratigraphy and Facies Relationships of the Hunton Group, Northern Arbuckle
Mountains and Lawrence Uplift; University of Oklahoma: Norman, Oklahoma, 2001.
20. Milad, B.; Slatt, R. Impact of lithofacies variations and structural changes on natural fracture distributions.
Interpretation 2018, 6, T873–T887. [CrossRef]
21. Fritz, R.D.; Medlock, P.L. Sequence stratigraphy of the Hunton Group as defined by core, outcrop, and log
data. Houst. Geol. Soc. Bull. 1994, 36, 29–58.
22. Al-Shaieb, Z.; Beardall, G.; Medlock, P.; Lippert, K.; Matthews, F.; Manni, F. Overview of Hunton Facies
and Reservoirs in the Anadarko Basin: Presented at the Hunton Group Core Workshop and Field Trip; Oklahoma
Geological Survey Special Publication: Norman, Oklahoma, 1993.
23. Ghosh, S.; Bontempi, C.P.; Hooker, J.N.; Slatt, R.M. High-Resolution Stratigraphic Characterization of Natural
Fracture Attributes in the Woodford Shale, Arbuckle Wilderness and US-77D Outcrops, Murray County,
Oklahoma. Interpretation 2018, 6, SC29–SC41. [CrossRef]
24. Ghosh, S.; Galvis-Portilla, H.A.; Klockow, C.M.; Slatt, R.M. An application of outcrop analogues to
understanding the origin and abundance of natural fractures in the Woodford Shale. J. Pet. Sci. Eng. 2018,
164, 623–639. [CrossRef]
25. Lantz, R.B. Quantitative evaluation of numerical diffusion (truncation error). Soc. Pet. Eng. J. 1971, 11,
315–320. [CrossRef]
26. Fanchi, J.R. Multidimensional numerical dispersion. Soc. Pet. Eng. J. 1983, 23, 143–151. [CrossRef]
27. Milad, B.; Ghosh, S.; Suliman, M.; Slatt, R. Upscaled DFN Models to Understand the Effects of Natural Fracture
Properties on Fluid Flow in the Hunton Group tight Limestone. In Proceedings of the Unconventional
Resource Technology Conference, Houston, Texas, 23–25 July 2018; pp. 1–16. [CrossRef]
28. Milad, B.; Alali, A.; Slatt, R.; Marfurt, K. Geostatistical Integration of Multiscale Data to Construct the Hunton
Group Geocellular Model: Upscaling Logs and Downscaling Seismic Impedance Volumes. In Proceedings of
the AAPG 2018 Annual Convention & Exhibition, Salt Lake City, UT, USA, 20–23 May 2018.
29. Milad, B. Integrated Reservoir Characterization and Geological Upscaling for Reservoir Flow Simulations of
the Sycamore/Meramec and Hunton Plays in Oklahoma. Ph.D. Thesis, University of Oklahoma, Norman,
OK, USA, 2019.
Energies 2020, 13, 1604 27 of 27

30. Schlumberger. Petrel Fracture Modeling Manual; Schlumberger: Houston, TX, USA, 2014.
31. Oda, M. Permeability tensor for discontinuous rock masses. Geotechnique 1985, 35, 483–495. [CrossRef]
32. Ghosh, S.; Busetti, S.; Slatt, R.M. Analysis and prediction of stimulated reservoir volumes through hydraulic
fracturing: Examples from western Arkoma Basin. J. Pet. Sci. Eng. 2019, 182, 106338. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).