A Collection of Recent

Highly Cited
SPE Peer-Reviewed Papers
January 2018
A Collection of R
 ecent Highly Cited 
SPE Peer-Reviewed Papers January 2018
SPE JOURNAL
Shale Gas-in-Place Calculations Part I: New Pore-Scale Considerations
R. J. Ambrose, R. C. Hartman, M. Diaz-Campos, I. Y. Akkutlu, and C. H. Sondergeld. 2012. SPE J 17 (1): 219–229.
SPE-131772-PA. http://doi.org/10.2118/131772-PA.

Recent Advances in Surfactant EOR

G. J. Hirasaki, C. A. Miller, and M. Puerto. 2011. SPE J 16 (4): 889–907. SPE-115386-PA.
http://doi.org/10.2118/115386-PA.

Carbon Dioxide Storage Capacity of Organic-Rich Shales

S. M. Kang, E. Fathi, R. J. Ambrose, I. Y. Akkutlu, and R. F. Sigal. 2011. SPE J 16 (4): 842–855. SPE-134583-PA.
http://doi.org/10.2118/134583-PA.

Simultaneous Multifracture Treatments: Fully Coupled Fluid Flow and Fracture Mechanics
for Horizontal Wells
K. Wu and J. E. Olson. 2015. SPE J 20 (2): 337–346. SPE-167626-PA. http://doi.org/10.2118/167626-PA.

A Generalized Framework Model for the Simulation of Gas Production in Unconventional

Gas Reservoirs
Y.-S Wu, J. Li, D. -Y Ding, C. Wang, and Y. Di. 2014. SPE J 19 (5): 845–857. SPE-163609-PA.
http://doi.org/10.2118/163609-PA.

SPE RESERVOIR EVALUATION & ENGINEERING

Reservoir Modeling in Shale-Gas Reservoirs
C. L. Cipolla, et al. 2010. SPE Res Eval & Eng 13 (4): 638–653. SPE-125530-PA. http://doi.org/10.2118/125530-PA.

Comparison of Fractured-Horizontal-Well Performance in Tight Sand and Shale Reservoirs

E. Ozkan, et al. 2011. SPE Res Eval & Eng 14 (2): 248–259. SPE-121290-PA. http://doi.org/10.2118/121290-PA.

Gas Permeability of Shale

A. Sakhaee-Pour and S. L. Bryant 2012. SPE Res Eval & Eng 15 (4): 401–409. SPE-146944-PA.
http://doi.org/10.2118/146944-PA.

SPE PRODUCTION AND OPERATIONS

What is Stimulated Reservoir Volume?
M. J. Mayerhofer, E. P. Lolon, Cm Rightmire, D. Walser, C. L. Cipolla, and N. R. Warplnskl. 2010. SPE Prod & Oper
25 (1): 89–98. SPE-119890-PA. http://doi.org/10.2118/119890-PA.

Modeling of Hydraulic-Fracture-Network Propagation in a Naturally Fractured Formation

X. Weng, O. Kresse, C. Cohen, R. Wu, and H. Gu. 2011. SPE Prod & Oper 26 (4): 368–380. SPE-140253-PA.
http://doi.org/10.2118/140253-PA.

Learn more about SPE Peer-Reviewed Journals at

http://www.spe.org/publications/
 Shale Gas-in-Place Calculations Part I:
New Pore-Scale Considerations
Ray J. Ambrose, Devon Energy and University of Oklahoma; Robert C. Hartman, Weatherford Labs;
and Mery Diaz-Campos, I. Yucel Akkutlu, and Carl H. Sondergeld, University of Oklahoma
Summary
Using focused-ion-beam (FIB)/scanning-electron-microscope (SEM)
imaging technology, a series of 2D and 3D submicroscale investiga-
tions revealed a finely dispersed porous organic (kerogen) material
embedded within an inorganic matrix. The organic material has
pores and capillaries having characteristic lengths typically less than
100 nm. A significant portion of total gas in place appears to be
associated with interconnected large nanopores within the organic
material.
Thermodynamics (phase behavior) of fluids in these pores is
quite different; gas residing in a small pore or capillary is rarefied
under the influence of organic pore walls and shows a different
density profile. This raises serious questions related to gas-in-place
calculations: Under reservoir conditions, what fraction of the pore
volume of the organic material can be considered available as free
gas, and what fraction is taken up by the adsorbed phase? How
accurately is the shale-gas storage capacity estimated using the
conventional volumetric methods? And finally, do average densi-
ties exist for the free and the adsorbed phases?
We combine the Langmuir adsorption isotherm with the volu-
metrics for free gas and formulate a new gas-in-place equation
accounting for the pore space taken up by the sorbed phase. The
method yields a total-gas-in-place prediction. Molecular dynamics
simulations involving methane in small carbon slit-pores of vary-
ing size and temperature predict density profiles across the pores
and show that (a) the adsorbed methane forms a 0.38-nm mono-
layer phase and (b) the adsorbed-phase density is 1.8–2.5 times
larger than that of bulk methane. These findings could be a more
important consideration with larger hydrocarbons and suggest that
a significant adjustment is necessary in volume calculations, espe-
cially for gas shales high in total organic content. Finally, using
typical values for the parameters, calculations show a 10–25%
decrease in total gas-storage capacity compared with that using the
conventional approach. The role of sorbed gas is more important
than previously thought. The new methodology is recommended
for estimating shale gas in place.
Introduction
Production of natural gas from organic-rich shale makes up an
increasing portion of the total production in North America.
According to the US Department of Energy’s (DOE) Energy
Information Association (EIA), the gross production from shale
plays in the US nearly doubled from 2007 to 2008 from 1,180 Bscf
to more than 2,000 Bscf, and has continued to grow (EIA 2009).
The main portion of the shale-gas production is maintained by
the large US shale plays such as the Barnett and Devonian, while
growth is attributable to new production coming on line in the
Woodford, Haynesville, Marcellus, Eagle Ford, and Horn River,
to name a few. The natural gas in these reservoirs is stored by two
mechanisms, free gas and adsorbed gas. The adsorbed gas is stored
mainly on the surface of the organics. Table 1 shows typical total
organic content (TOC) by weight percent (Comer and Hinch 1987;
Jenden et al. 1993; Mancini et al. 2006; Pollastro 2007; Reynolds
and Munn 2010).
Copyright © 2012 Society of Petroleum Engineers
This paper (SPE 131772) was accepted for presentation at the SPE Unconventional
Gas Conference, Pittsburgh, Pennsylvania, USA, 23–25 February 2010, and revised for
publication. Original manuscript received for review 18 January 2010. Revised manuscript
received for review 1 December 2010. Paper peer approved 10 December 2010.
March 2012 SPE Journal
When conducting a reservoir study on a natural-gas field, one
of the primary concerns is the estimation of initial gas in place.
The estimate is the basis for disclosure of gas reserves, and it is
important for reservoir-engineering analysis, such as gas-produc-
tion forecast. Tank-type and multidimensional (simulation-based)
material-balance calculations are common industry approaches in
predicting the gas in place when sufficient field-performance data
are available. To gain confidence in the estimate, or when sufficient
data are not available to initiate the material-balance calculations,
the volumetric methods are applied. Using key reservoir param-
eters (i.e., porosity, water saturation, and formation volume factor)
associated with well logs, core data, fluid samples, and well tests,
a volumetric method allows us to predict the gas in place in terms
of a total-gas pore volume of the reservoir.
For disclosure purposes, a deterministic volumetric method, in
which a single average value is selected for each parameter in the
reserves calculations, is most commonly used in North America.
Probabilistic methods, on the other hand, are increasingly used
worldwide and give the ability to describe the full range of values
for each parameter in order to somewhat reflect spatial variability
in the parameters and structural intricacies in reservoir architecture.
In this paper, it is shown that any volumetric approach for shale-
gas reserves estimation has added complexity because in shale the
natural gas (mostly methane) exists in different thermodynamic
states, namely adsorbed, absorbed (or dissolved), and free gas, and
that an accurate estimation of the gas pore volume in shale reser-
voirs should not be considered independently of these states.
A simple model of a physical shale matrix is illustrated in Fig. 1a.
The model needs to be quantified for gas-in-place analysis, and it
is typically achieved by methods developed specifically for tight
rocks and other low-permeability formations (Luffel and Guidry
1992; Luffel et al. 1993; Mavor and Nelson 1997). However, the
effective pore volume is not directly determined in these studies;
rather, total porosity, total water volume, and total oil volume
(by weight difference and an assumed oil density of 0.8 g/cc) are
determined. Nonetheless, as shown here in Eqs. 1 and 2, the total
and effective gas porosity values are equivalent:
⎛V ⎞⎛V ⎞ V
Sg = ⎜ v ⎟ ⎜ g ⎟ = g , . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
⎝ Vb ⎠ ⎝ Vv ⎠ Vb
⎛V ⎞⎛ V ⎞ V
Sge = ⎜ ve ⎟ ⎜ g ⎟ = g . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
⎝ Vb ⎠ ⎝ Vve ⎠ Vb
The bulk volume Vb in Eqs. 1 and 2 is determined by mercury
displacement of competent cores. Grain volume, on the other hand,
is determined on crushed cores through helium porosimetry. The
difference between these two volumes yields the total void volume
Vv (associated with the total porosity ) available for all in-situ
fluids (i.e., mobile hydrocarbons, free and bound water, adsorbed
gas, solution gas, and free gas).
For total gas storage, the shale gas-in-place volumes are gener-
ally considered in terms of the following:
• A volumetric component Gf involving hydrocarbons stored in
the pore space as free gas. The free-gas volume is quantified by
modifications of standard reservoir-evaluation methods.
• A surface component Ga with the gas physically adsorbed on
the large surface area of the micro- and mesopores. The adsorbed-gas
219
 In the current industry-standard calculations, Eqs. 6 and 7 are
TABLE 1—TYPICAL TOC OF NORTH AMERICAN SHALE
GAS PLAYS not applied. The solution gas in mobile hydrocarbons and water,
and the adsorbed gas within organic matter, are combined in the
Shale or Play Average or Range of TOC (wt %) adsorption isotherm analysis; therefore, Eq. 3 is reduced to
Barnett 4%
Gst = G f + Ga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
Marcellus 1–10%
Haynesville 0-8% Note that Gf is equivalent to Eqs. 1 or 2, although, to be consistent
Horn River 3% with the total gas-sorption data Ga, it is now defined in Eq. 4 in
terms of standard cubic feet per ton (Mavor and Nelson 1997).
Woodford 5%
The current volumetric approaches for shale gas are built on
the premise that the two volumes on the right side of Eq. 8, being
associated with the inorganic pores and organic solid, respectively,
amount has generally been quantified from the sorption-isotherm mea- can be estimated independently of one another. Accordingly, the
surements by establishing an equilibrium adsorption isotherm. sorbed gas is associated with the organics, the pore volume of
• A volumetric component Gso involving gas dissolved into which is considered to be negligible, and therefore the volume does
the liquid hydrocarbon. This volume is usually combined with not need to be accounted for during the calculations of free gas;
adsorbed-gas capacity in reservoirs that contain a large fraction of all of the free gas, however, is associated with the inorganic voids
liquid hydrocarbon in the pore space. such as macropores, fissures, fractures, and so on. In this paper,
• A volumetric component Gsw involving gas dissolved into the based on new pore-scale observations, we argue that the total gas-
formation water. The amount of dissolved gas is estimated from storage capacities and the resulting gas-in-place values are being
the bulk-solubility calculations. Although it has traditionally not inadvertently inflated and overestimated by this point of view. The
been considered important, a recent study is available discussing source of the error involves the improper accounting of the pore
significant enhancement in gas solubility in formation liquids when volume occupied by the sorbed-gas phase. There has been some
confined to small pores (Diaz-Campos et al. 2009). awareness of a necessity for the sorbed-phase porosity corrections
Hence, we have Gst as the total gas in place: [see, for example, Cui et al. (2009)], although the issue has not
been addressed properly yet. This article introduces a methodologi-
Gst = G f + Ga + Gso + Gsw , . . . . . . . . . . . . . . . . . . . . . . . . . . (3) cal approach based on new pore-scale considerations.
where the components of storage on the right side are defined as
Sorbed-Phase Correction for the Void Volume
 (1 − Sw − So ) The amount of sorbed gas that is estimated to be in shale is deter-
G f = 32.0368 , . . . . . . . . . . . . . . . . . . . . . . . . (4) mined through an equilibrium adsorption-isotherm experiment. In
b Bg
this experiment, a void volume is first measured, typically using
helium. Void-volume determination is experimentally identical to
p the helium-porosimetry techniques used to determine grain density.
Ga = GsL , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)
p + pL Some authors have raised the issue of molecular size of the fluid
used (Bustin et al. 2008; Kang et al. 2010) as a source of error; this
32.0368 So Rso error will not be discussed here. After the void volume has been
Gso = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6) measured, the sorption data are collected. The mass of gas sorbed
5.6146  Bo
into the sample is measured by material balance and a given ther-
modynamic equation of state. During construction of the isotherm,
32.0368 Sw Rsw at each pressure step, the volume of the gas that is adsorbed reduces
Gsw = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
5.6146  Bw the void volume. As a result, the initially determined void volume

Non-Clay Non-Clay
Grain Volume Grain Volume

Dry Clay Dry Clay

Bulk Volume
Bulk Volume

Volume Volume

Bound (Clay) Bound (Clay)

Water Volume Water Volume

Organic Content Organic Content

Sorbed Gas
Effective Void

Volume
Connected Pore Connected Pore
Volume

Total Void
Total Void

Free Gas
Volume
Volume

Volume Containing Volume Containing

Volume
Void

Free Oil, Gas, and Water Free Oil, Gas, and Water

Isolated Pore Volume Isolated Pore Volume

(a) Old Petrophysical Model (b) New Petrophysical Model

Fig. 1—(a) Old petrophysical model and (b) new petrophysical model showing volumetric constituents of a typical gas-shale
matrix. The hatched region describes the interplay between the sorbed phase and total porosity (void volume).

220 March 2012 SPE Journal

 50
45

Storage Capacity (scf/ton)

40
35
30
25
20
15
Langmuir Isotherm Data
10
Raw Gibbs Isotherm Data
5
0
0 1000 2000 3000 4000
Pressure (psia)

Fig. 2—Methane isotherms with and without Gibbs correction.

Vv0 must be corrected at the beginning and at the end of the pressure If the changes in void volume are not taken into account, the
step, as described in Eqs. 9 and 10 (Menon 1968): isotherm will be in error and will not be usable in engineering
calculations. An example of isotherm data with and without void-
ˆ
n1 M volume corrections is shown in Fig. 2. The aforementioned void-
Vv1 = Vv 0 − , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9) volume considerations that are necessary to accurately determine
s adsorbed-gas volumes have significant implications on the “live,”
in-situ shale pore volume available for free-gas storage. There-
ˆ
n2 M fore, the effective-porosity/gas-saturation product (Sge) derived
Vv 2 = Vv 0 − . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
s from a total pore volume that is determined under static labora-
tory conditions does not reflect live-reservoir conditions. The
reservoir total pore volume is not only consumed by water and
Accordingly, over the course of the isotherm analysis, the void
oil but also by the adsorbed gas. It is for this reason we propose
volume is further reduced for each subsequent pressure step. In
that the calculated free-gas pore volume requires a correction for
practice, it is often more convenient to determine a so-called
the adsorbed-gas fraction present under reservoir-temperature and
Gibbs isotherm in terms of number of moles of adsorbed gas, as
-pressure conditions.
in Eq. 11:
Additional evidence to support the need for the correction lies
⎛ pr 1 pr 2 ⎞ ⎛ ps1 ps 2 ⎞ in visual investigation of the complex rock fabric typical of fine-
n2′ = n1′ + Vr ⎜ − + Vv 0 ⎜ − . grained, organic-rich shales. It has recently been reported that
⎝ zr 1 RTr 1 zr 2 RTr 2 ⎠⎟ ⎝ z s1 RTs1 z s 2 RTs 2 ⎠⎟ much of the gas-storage capacity within shale is predominantly
. . . . . . . . . . . . . . . . . . . . . . . (11) associated with the organic fraction of the rock matrix (Loucks
et al. 2009; Wang and Reed 2009; Sondergeld et al. 2010).
The Gibbs isotherm can then be converted to volumes using an Therefore, the free-gas volumes measured by pycnometry using
equation of state and can be adjusted for the void volume using a nonadsorbent gas such as helium must be corrected for the
the Gibbs correction factor f /s: presence of adsorbed gas within the organics. Fig. 3 shows a 2D
FIB/SEM gas-shale image supporting the preceding observations.
Ga' In the image, the matrix is represented by the gray color, with the
Ga = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)
 dark- and light-gray regions being the organic (kerogen) and the
1− f inorganic constituents (e.g., clay, silica, and feldspar), respectively,
s
whereas the pores are shown in black. Clearly, most of the pores
are significantly small, typically less than 100 nm, and are almost
exclusively found within the kerogen.
Using advanced imaging technologies, we have created 3D
shale segmentations showing kerogen and pore networks that
contain up to 600 such FIB/SEM images. A similar investigation
has previously been conducted by Tomutsa et al. (2007) for the sub-
micron petrophysical analysis of chalk. A typical kerogen network
of our study is shown in Fig. 4a. The network consists of large
interconnected pockets of kerogen, which make up an estimated
7.7% (volume) of the segment. Fig. 4b shows the superimposed
kerogen (bounded with yellow) and pore (on red) networks of the
same shale segment. As can be seen again from these images,
nearly all of the porosity within the system is associated with the
kerogen network. The image-computed porosity for this segmented
volume is 2.5%. A 3D network overlay showing the comparison
of the kerogen/pore networks and the gray-scale image is shown
in Fig. 3c. Orange denotes kerogen, whereas dark orange denotes
pores. These volumes are produced through gray-scale threshold-
Fig. 3—2D FIB/SEM image showing porosity and kerogen within ing. The setting of thresholds is subjective, and care should be
shale. Black depicts pores; dark gray is kerogen; light gray is taken, although the medium-sized pores are best resolved (Curtis
matrix (clay and silica). et al. 2010).

March 2012 SPE Journal 221

 y
z
x
(a) (b) (c)

Fig. 4—3D FIB/SEM segmentations of 300 separate 2D SEM scans. Sample size is 4 ␮m high, 5 ␮m wide, and 2.5 ␮m deep. The
kerogen network is shown in (a); yellow outlines the 3D kerogen network. In (b), yellow outlines the 3D kerogen network, while
red outlines porosity (in this sample, all porosity is found within the kerogen). (c) depicts an overlay of porosity and kerogen 3D
network; orange is kerogen, and dark orange depicts porosity within kerogen. The large orange mass is further in depth than
what is seen on the surface in (a) and (b).

Method for Shale Gas-in-Place Calculations The volume occupied by the sorbed gas must be accounted for after
In this paper, a new petrophysical model is proposed that alters the correction for water saturation. This is because of the porosity
the previous concept of effective porosity discussed earlier. Fig. 1 basis of the water-saturation measurement. Appendix A shows the
shows a comparison between the new model and the old model. derivation of the fundamental equation (Eq. 13). The porosity-
The new model emphasizes two distinct conceptual changes with reduction effects can be shown to vary. In deep reservoirs where
respect to the previous model. First, there exists a dependency of there is not much sorbed-gas content, the correction to the free gas
the connected pore space on the organics (Wang and Reed 2009; in place can be less than 5%. In other reservoirs with low porosity
Loucks et al. 2009; Sondergeld et al. 2010). Second, there is a and a high sorbed-gas content, the reduction could be up to 40%
dependency on the free pore space by the inclusion of a sorbed of the free gas in place or greater. Appendix B shows an example
phase. Fig. 5 shows a simple diagram of the current methodology calculation and the impact on the calculated gas in place.
used to determine gas in place vs. the proposed methodology. For
simplicity, the water and oil volumes are not considered in the Sorbed-Phase Density. In order to calculate the volume occupied
diagram. The simple illustration taken in context with the new by the sorbed phase, the density of adsorbed gas in the organic
information from the FIB/SEM images and segmentations visu- pores (s) must be known a priori. Measurement of the sorbed-
ally shows the errors that result by assuming that the sorbed gas phase density is not a trivial matter, however; the local density for
takes up no volume. methane is expected to vary across the pore and to be different
In order to properly account for the total and free gas in place, from its average bulk density because of the added interactions
the volume occupied by the adsorbed gas must be determined and between methane and the organic pore walls. Furthermore, in gas
subtracted from the free-gas calculation. We therefore propose that shales where the reservoir temperature is significantly greater than
the standard calculations used to calculate free-gas storage capacity the critical temperature of the natural gas, it is difficult to study
(i.e., Eq. 4) be modified as follows: phase transitions and determine if the adsorbate is in the form of
a liquid or vapor.
There have been several suggestions, which have been docu-
32.0368 ⎡  (1 − Sw ) 1.318 × 10 −6 M
ˆ ⎛ p ⎞⎤
Gf = ⎢ − ⎜ GsL ⎥ mented in physical-chemistry literature, to determine the density
Bg ⎣ b s ⎝ p + pL ⎟⎠ ⎦ of the adsorbed phase on solid surfaces. Dubinin (1960) suggested
that the adsorbate density is related to the van der Waals co-volume
. . . . . . . . . . . . . . . . . . . . . . . (13) constant b. Independently, Haydel and Kobayashi (1967) used an
experimental method and found the density values for methane
Old Methodology New Methodology and propane to be nearly equal to the van der Waals co-volume
constant. Later, it was argued that the sorbed-phase density is
Void Space Void Space equivalent to the liquid density (Menon 1968) and to the critical
Measured by Measured by density (Tsai et al. 1985) of the sorbed fluid. Ozawa et al. (1976)
Porosity Porosity considered the adsorbed phase as a superheated liquid with a den-
Measurement Measurement sity dependent upon the thermal expansion of the liquid. Recently
+ Ming et al. (2003) compared all of the previously stated methods
+ Sorbed Mass to a Langmuir-Freundlich adsorption model and found that there
Sorbed Mass Measured by exists a temperature dependence in the sorbed-phase density, but
Measured by Adsorption the value approaches those proposed by Dubinin (1960). These
Adsorption Experiment studies, although fundamentally important to our understanding of
Experiment -
gas adsorption in shale, do not show a clear and accurate path to
Free Gas prediction of the adsorbed-phase density of shale gas and, hence,
Volume Taken to estimation of shale gas in place. All the previous studies were
=Total GIP up by Sorbed
conducted at low pressure and temperature values (at most, 104oF
Gas
=Total GIP and 1,000 psi) or using an inorganic wall (silica gel, as in the
case of Haydel and Kobayashi). In this research, density values
Fig. 5—Comparison of the old and new methodologies in pre- in the first layer within the range 0.28–0.3 g/cc were found using
dicting shale gas-in-place; for simplicity, oil and water volumes molecular modeling and simulation. It is found that the density
are not shown. is sensitive to changes in temperature, pressure, and pore size.

222 March 2012 SPE Journal

 Fig. 6—Molecular simulation cell consisting of graphite walls and OPLS-UA methane.
Typically, at 3,043 psi and 176oF, the adsorbed-phase density of
methane is 0.35 g/cc for a pore width of 2.3 nm when the Langmuir
single-layer theoretical model is used.
In this study, we used a numerical molecular-modeling approach
to determine the adsorbed-phase density from the first principles
of Newtonian mechanics. Molecular modeling and simulation is a
form of computer simulation that enables us to study thermody-
namic and transport properties of many-particle systems, in which
particles (atoms and molecules that make up the natural gas and
pore walls) with initially known and instantaneously predicted posi-
tions and momenta are allowed to interact for a period of time, giv-
ing a view of the motion of the particles as trajectories in space and
time. Numerical integration of Newton’s equations of motion make
up the core of the simulation; therefore, special algorithms (e.g.,
Verlet, Leap Frog, or Beeman algorithms) have been developed and
are commonly used during a molecular simulation study. Readers
who are interested in details of molecular simulation are encour-
aged to consult textbooks by Frenkel and Smit (2002) and Allen
and Tildesley (2002) for deterministic and stochastic treatment
of the numerical-integration process. Molecular simulations have
made enormous strides in recent years and are gradually becoming
a commonly used tool in science and engineering (de Pablo and
Escobedo 2002). Today, molecular simulations are being widely
used to construct virtual experiments in cases where controlled
laboratory measurements are difficult, if not impossible, to perform.
There exists an exhaustive literature studying the equilibrium ther-
modynamics of fluids using molecular simulation involving phase
change of bulk fluids (Harris and Yung 1995), characterization of
porous materials using gas adsorption (Aukett et al. 1992; Sweat-
man and Quirke 2001), and multicomponent gas separation (Ghoufi
et al. 2009). In our case, two sets of molecular-dynamics (MD)
simulations are used to estimate and analyze the adsorbed-phase
density under typical reservoir pressure and temperature conditions:
(a) runs involving bulk-phase (i.e., in the absence of pore walls)
methane-density measurements using a fixed number of methane
molecules at fixed pressure and temperature and (b) runs involving
measurements of density of methane at the same temperature but
confined to a pore with organic (graphite) walls. We will consider
deviations in the methane-density profile along the pore width of
the second set of runs relative to the uniform density value of the
bulk as an impact of the pore-wall effects.
TABLE 2—LENNARD-JONES POTENTIAL PARAMETERS
FOR METHANE AND CARBON
Atom (nm) /kB (K)
Carbon 0.34 28
Methane 0.373 147.9
March 2012 SPE Journal
MD Simulation of Methane Adsorption in Organic Slit-Pores.
For the molecular-level investigation, methane is considered at
some supercritical conditions under thermodynamic equilibrium in
3D periodic orthorhombic pore geometry consisting of upper and
lower pore walls made of graphene (carbon) layers (see Fig. 6).
For comparison, we take into account three slit-like pores with a
pore width equal to H = 3.9, 2.3, and 1.1 nm. Pore width is an
important length scale of the study, and it is defined as the dis-
tance between the innermost graphene planes. The pores maintain
fixed dimension in the y-coordinate: Ly = 3.93 nm; however, the
dimension is changed in the x-coordinate such that both pores host
approximately the same number of methane molecules (400–450)
during the simulations. Hence, the dimensions in this direction are
equal to Lx = 4.26 nm and to Lx = 7.67 nm for the large and small
pores, respectively. While the effect of pore pressure is investi-
gated, the number of particles is increased from 400 to 600 within a
2.3-nm pore. Typically, the approximate parallel computation time
was around 1,500–3,000 seconds, and 100 to 144 cores have been
used during the simulations.
A united-atom carbon-centered Lennard-Jones potential (based
on the optimized potentials for liquid simulations, OPLS-UA
force field) is used as the model of a methane molecule; Table 2
shows the energy and distance parameters used for fluid/fluid
and solid/solid interactions (Nath et al. 1998). The methane/solid
and methane/methane interactions are of the Lennard-Jones type
(Frenkel and Smit 2002). A Lorentz-Berthelot mixing rule [ij =
(ii + jj)/2 and εij = (εiiεjj)1/2] is set up in order to describe solid/
fluid (i.e., pore-wall methane) interactions. Here, ij and εij are
the Lennard-Jones parameters accounting for interactions between
a molecular site of methane species and a carbon atom of the
organic wall. Lennard-Jones interactions were cut off at 4.1ii for
the 3.9-nm pore width, and at 3ii for the 1.1- and 2.3-nm pore
width, respectively. The van der Waals interactions were cut off at
3ii for all three systems.
The MD simulation package, DL-POLY (version 2.20), was
used to perform the simulations (Todorov and Smith 2008). Meth-
ane bulk-density computations are made considering isobaric and
isothermic conditions with a constant number of atoms, constant
pressure, and constant temperature (NPT) at three constant tem-
peratures (176, 212, and 266°F) using the Nosé- Hoover thermostat
and barostat ensemble (Frenkel and Smit 2002). The relaxation
time used for the Nosé-Hoover thermostat is optimized to 15
pico-seconds (ps). In the case of simulations involving methane in
carbon slit-pores, initially the fluid system was equilibrated at a
constant temperature, with a canonical ensemble [constant number
of atoms, constant volume, and constant temperature (i.e., NVT)].
The equilibrium is assumed to be achieved if no drift in time was
observed in the time-independent quantity, such as the total energy
of the system. The relaxation time used for the Nosé-Hoover ther-
mostat is optimized to 13, 15, and 20 ps for the pore widths of 1.1,
223

Pore Width = 3.93nm
40 4.13E+01
ρ* CH4 (1000A–3 )
32
24
16
7.24E+00
8 5.18E+00
0 0.00
0.0 3.8 7.6 11.4 15.2
Pore Half Length (Å)
Fig. 7—Number-density (left) and discrete-density (right) profile for methane at 176°F (80°C) in a 3.93 nm pore. Density values
are estimated at each 0.2-Å interval for the continuous-density profile. Discrete density corresponds to molecular-layer density
for methane across the pore. The estimated pore pressure at the center of the pores is 3,043 psi. The inset graph in the upper
right corner is the equivalent density using a Langmuir single-layer adsorption model.
2.3, and 3.9 nm, respectively. The Nosé-Hoover algorithm needs to
be used to control the system temperature while the real canonical
ensemble computations are carried out for the density computa-
tions (Frenkel and Smit 2002). The average quantities within
1 nanosecond (ns) intervals after the equilibrium is reached are
taken for the analysis. Thermodynamic equilibrium is considered
to be reached when the fluctuations in the density profile along the
time are uniform. Typically, the total run time for a simulation with
the pore was 1.5 ns and the timestep used was 0.003 ps.
At the end of each simulation run, number density Number (num-
ber of molecules per volume) for methane across the pore space
was computed at every z = 0.2 Å interval (for the continuous
density profile) and at every Lz = 3.8 Å (for the discrete density
profile) volume segment in the z-direction. The number density
for each volume segment is estimated by summing the number of
molecules counted in each interval within the volume segment and
dividing the total by the number of intervals. Then, the number
density is converted to the local mass density of methane using its
molecular weight MCH4 and Avogadro’s number as follows:
 Number M CH4
CH4 = .
6.02252 × 10 23
As explained earlier, it is expected that the local mass density
for methane across the pore will be different from its mean bulk
density, owing to changing levels of interactions between the
methane/methane and methane/carbon bodies. The purpose of our
numerical investigation was to predict a precise density profile
across the pore as an indication of the presence of these interac-
tions and to determine an average adsorbed-gas-density value, a
macroscopic quantity necessary for the gas-in-place calculations
using Eq. 13.
Fig. 7 shows the density profile for methane confined to the
3.92-nm pore at 176°F. It is clear that the predicted methane density
is not uniform across the pores; its value is significantly greater
near the wall, where adsorption takes place, and it decreases with
damped oscillations as the distance from the pore wall increases.
The oscillations are caused by the presence of adsorption in
the pores and involve structured distribution of molecules (i.e.,
molecular layers). The number of molecules is largest in the first
layer near the wall, indicating physical adsorption. The molecules
in the second layer are still under the influence of pore walls,
although intermolecular interactions among the methane molecules
begin to dominate, not allowing locally high methane densities. In
this layer, the density of methane is slightly larger than the bulk
224
0.4
0.331
0.30 0.3
0.281
0.2
0.124
0.25 0.1
0.0
0.20
ρCH4 (g/cc)
0
0.0 3.8 7.6 11.4 15.2 19.
0.163
0.15 0.133 0.126 0.124
0.10
0.05
19.0 0.0 3.8 7.6 11.4 15.2 19.0
Pore Half Length (Å)
gas density of methane 0.124 g/cc at the center of the pore. The
pore pressure is 3,043 psia, which is a quantity predicted using the
free-gas molecules and the volume at the center of the pore. The
observed density profile shows that the assumption of Langmuir
adsorption theory with monolayer is reasonable to describe the
equilibrium adsorption dynamics of methane in the organic pores.
The inset graph on the upper right shows the equivalent Langmuir
monolayer plot where the pore half-length has been divided by the
adsorbed and free gas. The adsorbed-layer density is estimated as
the total of the discrete densities associated with adsorption:
s = 0.281 + (0.163–0.124) + (0.133–0.124)
+ (0.126–0.124) = 0.33g/cm3.
Pore-Size Effects on Methane Adsorption. In essence, super-
critical methane in small organic pores is structured as a result
of pore-wall effects, showing layers of graded density across the
pore. Depending on the pore size, a bulk-fluid region may exist at
the central portion of the pore, where the influence of molecular
interactions with the pore walls is very small. In pores with sizes up
to 50 nm, a combination of molecule/molecule and molecule/wall
interactions dictates thermodynamic states of the gas and its mass
transport in the pore (Krishna 2009). On the other hand, within a
small pore (see Fig. 8), methane molecules are always under the
influence of the force field exerted by the walls; consequently, no
bulk-fluid region can be observed in the pore, and no pore pressure
can be measured. Therefore, behavior of the adsorbed molecules
should be considered rather than the motion of free-gas molecules.
Furthermore, the adsorbed-phase density at the first layer is signifi-
cantly large. The smaller the pore, the larger the adsorbed-phase
density is. In the limit, Langmuir theory is not suitable for the
description of physical interactions between the gas and solid.
Effect of Temperature on Methane Adsorption. The effect
of temperature on methane density in the large pore is shown in
Fig. 9. The estimated average adsorbed-methane density in the
first layer is around 0.3 g/cm3, although it decreases with tem-
perature (for example, to 0.259 g/cm3 at 266°F). These values
show variations caused by changing levels of kinetic energy at
the microscopic scale.
Table 3 shows that the predicted free-gas density values at the
center of the pore correspond to the bulk methane density values
predicted independently using the National Institute of Standards
and Technology’s (NIST) thermophysical properties of hydro-
carbon mixtures database, SUPERTRAPP. The comparisons also
show that our numerical simulations are accurate.
March 2012 SPE Journal
 0.4 0.350
Pore Width = 2.31nm 0.40
0.3

50 0.35 0.2
0.124
0.1
42.80 0.30 0.292
43.01 0.0
40
0.25 0.0 3.8 7.6

ρCH4 (g/cc)
ρ* CH4 (1000Å–3 )

30 0.20 0.171
0.15 0.135
20
0.10
10 7.65 7.72
0.05
4-5 (bulk)
0 0.00
0.0 3.8 7.6 11.4 15.2 19.0 22.8 0.0 3.8 7.6 11.4
Pore Length (Å) Pore Half Length (Å)
Pore Width = 1.14nm

50 4.57E+01 0.35
0.307
4.69E+01 0.30
40
ρ* CH4 (1000Å–3 )

ρCH4 (g/cc) 0.25 0.234

30 0.20

20 0.15
1.57E+01
0.10
10
0.05
4-5(bulk)
0 0.00
0.0 3.8 7.6 11.4 0.0 3.8
Pore Length (Å) Pore Half Length (Å)
Fig. 8—Number-density (left) and discrete-density (right) profiles for methane at 176°F (80°C) in pore widths of 2.31 nm (upper)
and 1.41 nm (lower). Density values are estimated at each 0.2-Å interval for the continuous-density profile. Discrete density cor-
responds to molecular-layer density for methane across the pore. Estimated pore pressure at the center of the pores is 3,043
psi. The inset graph in the upper right corner is the equivalent density using a Langmuir single-layer adsorption model.

Conclusions
In this paper, we addressed the following issues in regard to the
volume available for free gas in organic shales:
• The sorbed phase follows Langmuir theory for most of the pores
and takes up a one-molecule-thick portion of a pore, although
there is a damped oscillation density profile with an increased
density in the second layer. For a 100-nm pore, the volume is
fairly insignificant; however, for pores on the order of 1 nm, it
is quite large.
• The current industry standard disregards the volume consumed
by the sorbed phase, thus inadvertently overestimating the pore
volume available for free-gas storage. We showed that with Eq.
13, a more correct volume is calculated. Through MD simulation
and Langmuir theory, we showed that the density for adsorbed
methane typically equals 0.34 g/cm3. This value is based on
numerical work using simple flat organic pore-wall surfaces. In
small pores, the adsorbed-layer thickness will be affected by the
roughness and curvature of the solid surface, and the density will
increase because of increased surface area of the wall. Further
analysis using more-complex pore-wall structures is necessary.
• In our calculation of the gas in place, we predict a dramatic
change in the estimated gas-in-place values because of volume
adjustments. This change in gas in place persists even when the
sorbed density values reported by other researchers, such as the
liquid-methane density of 0.4223 g/cc, corresponding to 1 atm
and –161°C (Mavor et al. 2004), and of 0.374 g/cc (Haydel and
Kobayashi et al. 1967). These are clearly shown in Table 4.
In conclusion, a robust method that matches the local phys-
ics is presented to determine an estimate of the gas in place in
organic-rich gas shale. Future work includes re-evaluation of the
concepts and approaches that are presented and discussed in the
presence of multicomponent gases with varying pore sizes. It is
expected that there will be added uncertainties in the gas-in-place
estimation because of (1) multicomponent nature of gas and its
adsorbed phase and (2) volume fraction of organic nanopores that
are currently inaccessible for visual observation and measurement.
The former requires computer-simulation studies on the adsorbed-
phase thickness and density using binary, ternary, and quaternary
mixtures of gases with components such as ethane, propane, and
carbon dioxide. The latter uncertainty, on the other hand, can be
reduced by measuring the absolute area of the organic surfaces
using isotherm data. In addition, a new study on transport proper-
ties of natural gases in nanoporous organic materials is necessary
in light of the new pore-scale findings.

March 2012 SPE Journal 225

 40 0.30
176°F 176°F
0.25 212°F
32 212°F
266°F
ρ* CH4 (1000Å–3 )

266°F
0.20

ρCH4 (g/cc)
24
0.15
16
0.10
8
0.05

0 0.00
0.0 3.8 7.6 11.4 15.2 19.0 0.0 3.8 7.6 11.4 15.2 19.0
Pore Half Length (Å) Pore Half Length (Å)

Fig. 9—Number-density (left) and discrete-density (right) profiles across half-length of a 3.9-nm-width slit-pore as a function of
temperature.

TABLE 3—EFFECT OF PRESSURE ON METHANE DENSITY

Methane Bulk Density Methane Bulk Density*
3 3
Temperature (°F) Pressure (psi) (g/cm ) (g/cm )

176 2206 0.090 0.089

176 3043 0.124 0.122
176 3676 0.147 0.144
176 4141 0.160 0.159
176 4404 0.168 0.167
176 4586 0.176 0.172
176 4800 0.178 0.175
176 6272 0.214 0.211
176 7300 0.231 0.229
176 7550 0.235 0.233
176 8707 0.253 0.250
NOTE: The estimated density values are compared with the bulk density values obtained using
SUPERTRAPP® under the same pressure and temperature conditions.
* From NIST-SUPERTRAPP®

Nomenclature Ga = adsorbed-gas storage capacity, scf/ton

Bg = gas formation volume factor, reservoir volume/surface Gf = free-gas storage capacity, scf/ton
volume GsL = Langmuir storage capacity, scf/ton
Bo = oil formation volume factor, reservoir volume/surface Gso = dissolved-gas-in-oil storage capacity, scf/ton
volume Gst = total gas storage capacity, scf/ton
Bw = water formation volume factor, reservoir volume/surface Gsw = dissolved-gas-in-water storage capacity, scf/ton
volume H = pore size, Å

TABLE 4—EFFECT OF SORBED PHASE DENSITY ON FREE AND TOTAL GIP CALCULATIONS
Adsorbed-Phase Density Free Storage Capacity Total Storage Capacity
3
(g/cm ) (SCF/ton) (SCF/ton)

Reference s Shale A Shale B Shale A Shale B

Molecular dynamics 0.34 91.9 75.8 130.7 158.6

Haydel and Kobayashi (1967) 0.37 72.9 95.1 155.7 133.9
Mavor et al. (2004)
(at 1 atm, –161.2°C) 0.42 93.2 79.7 132.1 162.5

226 March 2012 SPE Journal

 kB = Boltzmann constant, 1.3806503×10–23 kJ/K–1 Bustin, R.M., Bustin, A.M., Cui, X., Ross, D.J.K., and Murthy Pathi, V.S.
Lx = length of computational cell in x-direction, Å 2008. Impact of Shale Properties on Pore Structure and Storage Charac-
Ly = length of computational cell in y-direction, Å teristics. Paper SPE 119892 presented at the SPE Shale Gas Production
Conference, Fort Worth, Texas, USA, 16–18 November. http://dx.doi.
M̂ = apparent natural-gas molecular weight, lbm/lbmole
org/10.2118/119892-MS.
MCH4 = molecular weight of methane, g/gmole Comer, J.B. and Hinch, H.H. 1987. Recognizing and quantifying expul-
n = number of moles sion of oil from the Woodford Formation and age-equivalent rocks in
n’1 = Gibbs-isotherm number of moles at the start of pressure Oklahoma and Arkansas. AAPG Bulletin 71 (7): 844–858.
step, lb-moles Cui, X., Bustin, A.M.M., and Bustin, R.M. 2009. Measurements of
n’2 = Gibbs-isotherm number of moles at the end of pressure gas permeability and diffusivity of tight reservoir rocks: different
step, lb-moles approaches and their applications. Geofluids 9 (3): 208–223. http://
p = pressure, psia dx.doi.org/10.1111/j.1468-8123.2009.00244.x.
pL = Langmuir pressure, psia Curtis, M.E., Ambrose, R.J., Sondergeld, C.S., and Rai, C.S. 2010.
pr1 = pressure of reference cell at start, psia Structural Characterization of Gas Shales on the Micro- and Nano-
pr2 = pressure of reference cell at end, psia Scales. Paper SPE 137693 presented at the Canadian Unconventional
ps1 = pressure of sample cell at start, psia Resources and International Petroleum Conference, Calgary, 19–21
October. http://dx.doi.org/10.2118/137693-MS.
ps2 = pressure of sample cell at end, psia
de Pablo, J.J. and Escobedo, F.A. 2002. Molecular Simulations in Chemical
R = universal gas constant, psia-ft3/mol-K Engineering: Present and future. AIChE Journal 48 (12): 2716–2721.
Rso = solution-gas/oil ratio, scf/STB http://dx.doi.org/10.1002/aic.690481202.
Rsw = solution-gas/water ratio, scf/STB Diaz-Campos, M., Akkutlu, I.Y., and Sigal, R.F. 2009. A Molecular
Sg = gas saturation, dimensionless Dynamics Study on Natural Gas Solubility Enhancement in Water Con-
Sge = effective gas saturation, dimensionless fined to Small Pores. Paper SPE 124491 presented at the SPE Annual
So = oil saturation, dimensionless Technical Conference and Exhibition, New Orleans, 4–7 October.
Sw = water saturation, dimensionless http://dx.doi.org/10.2118/124491-MS.
T = reservoir temperature, °F Dubinin, M.M. 1960. The Potential Theory of Adsorption of Gases and
= reference-cell temperature at start, °R Vapors for Adsorbents with Energetically Nonuniform Surfaces. Chem-
Tr1
ical Review 60 (2): 235–241. http://dx.doi.org/10.1021/cr60204a006.
Tr2 = reference-cell temperature at end, °R
Energy Information Administration (EIA). 2009. Annual Energy Out-
Ts1 = sample-cell temperature at start, °R look 2009, With Projections to 2030. Annual Report No. DOE/EIA-
Ts2 = sample-cell temperature at end, °R 0383(2009), EIA/US DOE, Washington, DC (March 2009).
Vb = bulk volume, ft3 Frenkel, D. and Smit B. 2002. Understanding Molecular Simulation: From
Vg = gas volume, ft3 Algorithms to Applications, second edition, Vol. 1. San Diego, Califor-
Vr = reference volume, ft3 nia: Computational Science Series, Academic Press.
Vv = void volume, ft3 Ghoufi, A., Gaberova, L., Rouquerol, J., Vincent, D., Llewellyn, P.L., and
Vve = effective void volume, ft3 Maurin, G. 2009. Adsorption of CO2, CH4 and their binary mixture in
Vv0 = initial void volume, ft3 Faujasite NaY: A combination of molecular simulations with gravime-
Vv1 = void volume Step 1, ft3 try–manometry and microcalorimetry measurements. Microporous and
Mesoporous Materials 119 (1–3): 117–128. http://dx.doi.org/10.1016/
Vv2 = void volume Step 2, ft3
j.micromeso.2008.10.014.
z = compressibility factor Harris, J.G. and Yung, K.H. 1995. Carbon Dioxide’s Liquid-Vapor Coex-
zr = compressibility factor of reference cell istence Curve And Critical Properties as Predicted by a Simple Molec-
z = interval across the pore space used for number density ular Model. J. Phys. Chem. 99 (31): 12021–12024. http://dx.doi.
calculations, Å org/10.1021/j100031a034.
ε = depth of the potential well, kJ Haydel, J.J. and Kobayashi, R. 1967. Adsorption Equilibria in the Methane-
b = bulk-rock density, g/cm3 Propane-Silica Gel System at High Pressures. Ind. Eng. Chem. Funda-
CH4 = mass density of methane in pore, g/cm3 men. 6 (4): 564–554. http://dx.doi.org/10.1021/i160024a010.
f = free-gas-phase density, g/cm3 Hünenberger, P.H. 2005. Thermostat Algorithms for Molecular Dynamics
Number = number density of methane, number of molecules/Å3 Simulations. Adv. Polym. Sci. 173: 105–149. http://dx.doi.org/10.1007/
s = sorbed-phase density, g/cm3 b99427.
Jenden, P.D., Drazan, D.J., and Kaplan, I.R. 1993. Mixing of Thermogenic
 = distance at which the intermolecular potential is zero, nm
Natural Gases in Northern Appalachian Basin. AAPG Bulletin 77 (6):
 = total porosity fraction, dimensionless 980–998.
a = sorbed-phase porosity fraction, dimensionless Kang, S., Fathi, E., Ambrose, R.J., Akkutlu, I.Y., and Sigal, R.F. 2010.
e = effective porosity fraction, dimensionless Carbon Dioxide Storage Capacity of Organic-rich Shales. Paper
SPE 134583 presented at the SPE Annual Technical Conference
Acknowledgments and Exhibition, Florence, Italy, 19–22 September. http://dx.doi.
We recognize the work and contributions of Mark Curtis and Gary org/10.2118/134583-MS.
Stowe in collecting many of the images in this paper. We thank the Krishna, R. 2009. Describing the Diffusion of Guest Molecules Inside
staff at the University of Oklahoma’s OSCER supercomputing cen- Porous Structures. J. Phys. Chem. C 113 (46): 19756–19781. http://
ter for the facilities to perform the MD simulations. We also thank dx.doi.org/10.1021/jp906879d.
Devon Energy—in particular, Jerry Youngblood, Bret Jameson, Loucks, R.G., Reed, R.M., Ruppel, S.C., and Jarvie, D.M. 2009. Morphol-
Jeff Hall, and Bill Coffey—for providing access to Barnett core ogy, Genesis, and Distribution of Nanometer-Scale Pores in Siliceous
and supporting much of this work. Mudstones of the Mississippian Barnett Shale. Journal of Sedimentary
Research 79 (12): 848–861. http://dx.doi.org/10.2110/jsr.2009.092.
References Luffel, D.L. and Guidry, F.K. 1992. New Core Analysis Methods for
Allen, M.P., and Tildesley, D.J. 2002. Computer Simulation of Liquids, Measuring Rock Properties of Devonian Shale. J. Pet. Tech. 44 (11):
reprint. London: Oxford University Press. 1184–1190. SPE-20571-PA. http://dx.doi.org/10.2118/20571-PA.
Aukett, P.N., Quirke, N., Riddiford, S., and Tennison, S.R. 1992. Meth- Luffel, D.L., Hopkins, C.W., and Schettler, P.D. Jr. 1993. Matrix Permeabil-
ane adsorption on microporous carbons—A comparison of experi- ity Measurement of Gas Productive Shales. Paper SPE 26633 presented
ment, theory, and simulation. Carbon 30 (6): 913–924. http://dx.doi. at the SPE Annual Technical Conference and Exhibition, Houston, 3–6
org/10.1016/0008-6223(92)90015-O. October. http://dx.doi.org/10.2118/26633-MS.

March 2012 SPE Journal 227

Mancini, E. A., Goddard D. A., Barnaby R., and Aharon P. 2006. Basin V RT
Analysis and Petroleum System Characterization and Modeling, Inte- =
n p
rior Salt Basins, Central and Eastern Gulf of Mexico. Phase 1 Final
Report, Project No. 422, Contract No. DE-FC26-03NT15395, US ft 3 psi
10.73159 519.67˚R
DOE/NETL, Alabama University, Tuscaloosa, Alabama. V ˚R lb-mol . . . . . . . . . . . . . . . . . . (A-2)
=
Mavor, M.J. and Nelson, C.R. 1997. Coalbed Reservoir Gas-In-Place n ft 3
14.696 psia = 379.48
Analysis. Report GRI-97/0263, Gas Research Institute, Chicago, lb-mol
Illinois.
Mavor, M.J., Hartman, C., and Pratt, T.J. 2004. Uncertainty in Sorption With density in g/cm3 and the desired units in scf/ton, we can use
Isotherm Measurements. Paper No. 411 presented at the International the preceding value to calculate a conversion constant.
Coalbed Methane Symposium, Tuscaloosa, Alabama, 14–16 May.
Menon, P.G. 1968. Adsorption at high pressures. Chem. Rev. 68 (3): 1 1 ton ton mol . . . . . . . . . (A-3)
⋅ = 1.318 × 10 −6
277–294. http://dx.doi.org/10.1021/cr60253a002. ft 3 2000 lb ft 3
Ming, L., Anzhong, G., Xuesheng, L., and Rongshun, W. 2003. Deter- 379.48
lb-mol
mination of the adsorbate density from supercritical gas adsorption
equilibrium data. Carbon 41 (3): 585–588. http://dx.doi.org/10.1016/ Using the conversion constant, the density of the adsorbed phase,
S0008-6223(02)00356-1. the bulk density of the rock, and the molecular weight of the
Nath, S.K, Escobedo, F.A., and de Pablo, J.J. 1998. On the simulation of adsorbed phase, we can calculate the fractional volume occupied
vapor-liquid equilibria for alkanes. J. Phys. Chem. 108 (23): 9905– by the sorbed phase:
9911. http://dx.doi.org/10.1063/1.476429.
Ozawa, S., Kusumi, S., and Ogino, Y. 1976. Physical adsorption of gases at
b ⎛ p ⎞
high pressure. IV. An improvement of the Dubinin—Astakhov adsorp- a = 1.318 × 10 −6 Mˆ G . . . . . . . . . . . . . . . . . (A-4)
tion equation. Journal of Colloid Interface Science 56 (1): 83–91. s ⎜⎝ sL p + pL ⎟⎠
http://dx.doi.org/10.1016/0021-9797(76)90149-1.
Pollastro, R.M. 2007. Total petroleum system assessment of undiscovered Assuming the oil saturation is negligible, and taking the frac-
resources in the giant Barnett Shale continuous (unconventional) tional sorbed-phase volume away from the free-gas volume, Eq.
gas accumulation, Fort Worth Basin, Texas. AAPG Bulletin 91 (4): 4 becomes
551–578. http://dx.doi.org/10.1306/06200606007.
Reynolds, M.M. and Munn, D.L. 2010. Development Update for an
Emerging Shale Gas Giant Field—Horn River Basin, British Columbia,  (1 − Sw ) − a
G f = 32.0368 . . . . . . . . . . . . . . . . . . . . . . . (A-5)
Canada. Paper SPE 130103 presented at the SPE Unconventional Gas b Bg
Conference, Pittsburg, Pennsylvania, USA, 23–25 February. http://
dx.doi.org/10.2118/130103-MS. Substituting the expression for a into this equation and simplify-
Sondergeld, C.H., Ambrose, R.J., Rai, C.S., and Moncrieff, J. 2010. Micro- ing yields Eq. 13:
Structural Studies of Gas Shales. Paper SPE 131771 presented at the
SPE Unconventional Gas Conference, Pittsburg, Pennsylvania, USA, 32.0368 ⎡  (1 − Sw ) 1.318 × 10 −6 M
ˆ ⎛ p ⎞⎤
23–25 February 2010. http://dx.doi.org/10.2118/131771-MS. Gf = ⎢ − ⎜⎝ GsL p + p ⎟⎠ ⎥
Bg ⎣ b s L ⎦
Sweatman, M.B. and Quirke, N. 2001. Characterization of Porous Materi-
als by Gas Adsorption: Comparison of Nitrogen at 77 K and Carbon
Dioxide at 298 K for Activated Carbon. Langmuir 17 (16): 5011–5020. . . . . . . . . . . . . . . . . . . . . . . (A-6)
http://dx.doi.org/10.1021/la010308j.
Todorov, I.T. and Smith, W. 2008. The DLPOLY_3 User Manual. Cheshire, Appendix B—Example Calculation
UK: Science & Technology Facilities Council (STFC). In order to quantify the pore-scale effects using the new methodol-
Tomutsa, L., Silin, D., and Radmilovic V. 2007. Analysis of Chalk Petro- ogy, we compare the results of the old method vs. the new method
physical Properties by Means of Submicron-scale Pore Imaging and on two shales. The first shale has a low sorbed-gas volume, while
Modeling. SPE Res Eval & Eng 10 (3): 285–293. SPE-99558-PA. the second has a relatively high sorbed-gas volume. A value of 0.34
http://dx.doi.org/10.2118/99558-PA. g/cm3 is used for sorbed-methane density based on the discussion
Tsai, M.C., Chen, W.N., Cen, P.L., Yang, R.T., Kornosky, R.M., Holcombe, in the “MD Simulation of Methane Adsorption in Organic Slit-
N.T., and Strakey, J.P. 1985. Adsorption of gas mixture on acti- Pores” subsection. The rest of the parameters for the two shales
vated carbon. Carbon 23 (2): 167–73. http://dx.doi.org/10.1016/0008- are shown in Table B-1.
6223(85)90008-9. Using the old calculation method for gas in place, the adsorbed,
Wang, F.P. and Reed, R.M. 2009. Pore Networks and Fluid Flow in Gas free, and total gas storage capacities for Shale A are, respectively,
Shales. Paper SPE 124253 presented at the SPE Annual Technical
Conference and Exhibition, New Orleans, 4–7 October. http://dx.doi.
 (1 − Sw )
org/10.2118/124253-MS. G f = 32.0368
Yee, D., Seidle, J.P., and Hanson, W.B. 1993. Gas sorption on coal and b Bg
measurement of gas content. In Hydrocarbons from Coal, ed. B.E. Law 0.06(1 − 0.35)
and D.D. Rice, No. 38, Chap. 9, 203–218. Tulsa, Oklahoma: AAPG = 32.0368 = 108.6 scf/ton . . . . . . . . . . . (B-1)
2.5 ⋅ 0.0046
Studies in Geology Series, AAPG.

Appendix A—Derivation of Eq. 13

p 4000
Beginning with Eq. 5, the value Ga needs to be converted into a Ga = GsL = 50 = 38.8 scf/ton . . . . . . . (B-2)
volume, and a simple unit conversion can be performed. The typi- p + pL 4000 + 1150
cal unit of the equation below is scf/ton.

Gst = G f + Ga = 108.6 scf/ton + 38.8 scf/ton = 147.7 scf/ton

p
Ga = GsL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-1)
p + pL
. . . . . . . . . . . . . . . . . . . . . . (B-3)

Given that it is in scf, we can convert scf into a mass with the Using the new calculation method for gas in place, the free, sorbed,
ideal-gas law at standard temperature and pressure: and total storage capacities for Shale A are, respectively,

228 March 2012 SPE Journal

 TABLE B-1—SHALE PROPERTIES FOR GAS-IN-PLACE EXAMPLE CALCULATION

Shale A (Low Sorption Capacity) Shale B (High Sorption Capacity)

φ 0.06 0.06
Sw 0.35 0.35
So 0.0 0.0
Bg 0.0046 0.0046
M̂ 16 lb/lb-mol 16 lb/lb-mol
GsL 50 scf/ton 120 scf/ton
p 4000 psia 4000 psia
T
o o
180 F 180 F
pL 1150 psia 1800 psia
3 3
b 2.5 g/cm 2.5 g/cm
3 3
s
0.34 g/cm 0.34 g/cm

32.0368 ⎡  (1 − Sw ) 1.318 × 10 −6 M
ˆ ⎛ p ⎞⎤ Ga = 120
4000
= 82.8 scf/ton . . . . . . . . . . . . . . . . (B-11)
Gf = ⎢ − ⎜ GsL ⎥
Bg ⎣ b s ⎝ p + pL ⎟⎠ ⎦ 4000 + 1800

⎡ 0.06(1 − 0.335) ⎤ Gst = 71.8 scf/ton + 82.8 scf/ton

32.0368 ⎢ 2.5 ⎥
= ⎢ −6 ⎥ = 155.7 scff/ton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-12)
0.0046 ⎢ 1.318 × 10 ⋅ 16 ⎛ 4000 ⎞⎥
− ⎜ 50 ⎟
⎢⎣ 0.34 ⎝ 4000 + 1150 ⎠ ⎥⎦ This represents a decrease of 32.8 and 18.6% of the free and total
= 91.9 scf/ton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-4) gas-storage capacities, respectively.

Raymond J. Ambrose is the Director of Reservoir Engineering

4000 for Reliance Holding USA, Inc. His current research interests
Ga = 50 = 38.8 scf/ton . . . . . . . . . . . . . . . . . . (B-5)
4000 + 1150 include analytical solutions for shale productivity, scanning
electron microscopy and pore structure characterization,
and storage mechanisms for shale. He holds a BS degree in
Gst = 91.9 scf/ton + 38.8 scf/ton = 130.7 scf/ton
n . . . . . . . . (B-6) chemical engineering and an MSc degree in petroleum engi-
neering, both from the University of Southern California, and a
This represents a decrease of 15.3 and 11.5% of the free and total PhD degree in petroleum engineering from the University of
Oklahoma. Ambrose currently serves SPE as a Technical Editor
gas-storage capacities, respectively. for the Journal of Canadian Petroleum Technology.
Using the old calculation method for gas in place, the adsorbed,
free, and total gas storage capacities for Shale B are, respectively, Chad Hartman is Chief Technical Advisor for Weatherford
Laboratories, based in Golden, Colorado. Hartman is acknowl-
edged as an industry expert in unconventional reservoirs and
 (1 − Sw ) integrating formation evaluation data. His credentials include
G f = 32.0368
b Bg design and development of Weatherford Laboratories’ state-
of-the-art Adsorption Isotherm Laboratory, field and laboratory
0.06(1 − 0.35)
= 32.0368 = 108.6 scf/ton . . . . . . . . . . . (B-7) protocols, and data QC/QA methods. He currently supports
2.5 ⋅ 0.0046 Weatherford Laboratories’ global business development, prod-
uct research and development, data interpretation, and client
consultation services. He holds a BS degree in chemistry from
p Fort Lewis College.
Ga = GsL
p + pL
Mery Diaz-Campos is with Schlumberger Information Solutions
4000 in Houston. She holds an MSc degree in petroleum engineering
= 120 = 82.8 scf/ton . . . . . . . . . . . . . . . . . (B-8) from the University of Oklahoma and a BSc degree in chemis-
4000 + 1800
try from National University in Lima, Peru.
I. Yucel Akkutlu is an associate professor of petroleum engi-
Gst = G f + Ga neering at the University of Oklahoma. His current academic
= 108.6 scf/ton + 82.8 scf/ton. . . . . . . . . . . . . . . . . . . (B-9) research is on thermodynamics and transport of fluids in nano-
porous materials, and on scaling-up and homogenization of
= 191.4 scf/ton coupled transport and reaction processes in low-permeability
geological formations exhibiting multiscale pore structures. He
Using the new calculation method for gas in place, the free, sorbed, holds MSc and PhD degrees in petroleum engineering from
and total storage capacities for Shale B are, respectively, the University of Southern California. Akkutlu currently serves
SPE as an Associate Editor of SPE Journal.
Carl H. Sondergeld is currently a professor and the Curtis
⎡ 0.06(1 − 0.35) ⎤ Mewbourne Chair at the Mewbourne School of Petroleum and
32.0368 ⎢ 2.5 ⎥
⎢ ⎥ Geological Engineering at the University of Oklahoma. He spent
Gf = −6
0.0046 ⎢ 1.318 × 10 ⋅ 16 ⎛ 4000 ⎞⎥ 19 years at the Tulsa Research Center of Amoco Production
− ⎜⎝ 120 ⎟ Company and holds 14 US patents. He holds BA and MA
⎢⎣ 0.34 4000 + 1800 ⎠ ⎥⎦ degrees in geology from Queens College, City University of New
= 72.9 scf/ton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-10)) York, and a PhD degree in geophysics from Cornell University.

March 2012 SPE Journal 229

 Recent Advances in Surfactant EOR
George J. Hirasaki, SPE, Clarence A. Miller, SPE, and Maura Puerto, Rice University
Summary
In this paper, recent advances in surfactant enhanced oil recovery
(EOR) are reviewed. The addition of alkali to surfactant flooding
in the 1980s reduced the amount of surfactant required, and the
process became known as alkaline/surfactant/polymer flooding
(ASP). It was recently found that the adsorption of anionic surfac-
tants on calcite and dolomite can also be significantly reduced with
sodium carbonate as the alkali, thus making the process applicable
for carbonate formations. The same chemicals are also capable
of altering the wettability of carbonate formations from strongly
oil-wet to preferentially water-wet. This wettability alteration in
combination with ultralow interfacial tension (IFT) makes it pos-
sible to displace oil from preferentially oil-wet carbonate matrix
to fractures by oil/water gravity drainage.
The alkaline/surfactant process consists of injecting alkali and
synthetic surfactant. The alkali generates soap in situ by reaction
between the alkali and naphthenic acids in the crude oil. It was
recently recognized that the local ratio of soap/surfactant deter-
mines the local optimal salinity for minimum IFT. Recognition of
this dependence makes it possible to design a strategy to maximize
oil recovery with the least amount of surfactant and to inject poly-
mer with the surfactant without phase separation. An additional
benefit of the presence of the soap component is that it generates
an oil-rich colloidal dispersion that produces ultralow IFT over a
much wider range of salinity than in its absence.
It was once thought that a cosolvent such as alcohol was
necessary to make a microemulsion without gel-like phases or
a polymer-rich phase separating from the surfactant solution.
An example of an alternative to the use of alcohol is to blend
two dissimilar surfactants: a branched alkoxylated sulfate and a
double-tailed, internal olefin sulfonate. The single-phase region
with NaCl or CaCl2 is greater for the blend than for either sur-
factant alone. It is also possible to incorporate polymer into such
aqueous surfactant solutions without phase separation under some
conditions. The injected surfactant solution has underoptimum
phase behavior with the crude oil. It becomes optimum only as
it mixes with the in-situ-generated soap, which is generally more
hydrophobic than the injected surfactant. However, some crude
oils do not have a sufficiently high acid number for this approach
to work.
Foam can be used for mobility control by alternating slugs of
gas with slugs of surfactant solution. Besides effective oil dis-
placement in a homogeneous sandpack, it demonstrated greatly
improved sweep in a layered sandpack.
Introduction
It is generally considered that only approximately one-third of the
petroleum present in known reservoirs is economically recoverable
with established technology (i.e., primary-recovery methods using
gas pressure and other natural forces in the reservoir, and second-
ary recovery by waterflooding). It has long been an objective of
the industry to develop improved processes to increase overall
recovery. However, the low oil prices that prevailed from the mid-
1980s until recently provided little incentive for research on EOR,
especially surfactant processes with substantial initial cost for
chemicals. In light of the current higher prices and accompanying
Copyright © 2011 Society of Petroleum Engineers
This paper (SPE 115386) was accepted for presentation at the SPE Annual Technical
Conference and Exhibition, Denver, 21–24 September 2008, and revised for publication.
Original manuscript received for review 19 July 2010. Revised manuscript received for
review 22 December 2010. Paper peer approved 18 January 2011.
December 2011 SPE Journal
revival of interest, it seems appropriate to review understanding of,
and prospects for, surfactant EOR.
Adding surfactant to injected water to reduce oil/water IFT
and/or alter wettability and thereby increase recovery is not a new
idea [see, for instance, Uren and Fahmy (1927)]. Indeed, a few
early field trials where small amounts of surfactant were injected
did produce small increases in oil recovery. The increases were
probably caused mainly by wettability changes, although the data
were inconclusive for assessing mechanisms. The results were not
sufficiently promising to stimulate use of surfactants on a larger
scale. A related long-held concept for improving recovery is to
generate surfactant in situ by injecting an alkaline solution (Atkin-
son 1927), which is less expensive than synthetic surfactants and
converts naphthenic acids in the crude oil to soaps. Early results
were not encouraging, and the relative importance of likely process
mechanisms was not understood (Johnson 1976). Other references
to early work on surfactants are given by Hill et al. (1973).
Two different approaches stimulated significant advances in
surfactant EOR processes in the 1960s. The surfactants were made
either by direct sulfonation of aromatic groups in refinery streams
or crude oils, or by organic synthesis of alkyl/aryl sulfonates,
which allowed for the surfactant to be tailored to the reservoir of
interest. The advantages of these surfactants are their low cost, their
wide range of properties, and the availability of raw materials in
somewhat large quantities.
Miscible flooding was an active area of research, but the sol-
vents being considered, such as enriched gas and LPG, exhibited
poor reservoir sweep because the adverse mobility ratio promoted
viscous fingering and the low solvent density led to gravity over-
ride. Seeking a solvent miscible with oil but having a higher vis-
cosity and density, Gogarty and coworkers at Marathon proposed
using a slug of an oil-continuous microemulsion made of hydro-
carbon, a petroleum sulfonate surfactant, an alcohol, and water
or brine [see review by Gogarty (1977)]. Holm and coworkers at
Union Oil advocated a similar process using a “soluble oil,” which
was also an oil-continuous microemulsion made mainly of crude
oil, some mineral oil, petroleum sulfonate, a cosolvent such as eth-
ylene glycol monobutyl ether, and water, as summarized by Holm
(1977). Slugs of these materials miscible displaced oil and with
better sweep than previous solvents. However, it was not initially
recognized that process success also depended on maintaining
ultralow IFT at the rear of the slug, where it was displaced by an
aqueous polymer solution and became a Winsor I microemulsion
(Hirasaki 1981).
The other approach involved injection of a surfactant formula-
tion made of a petroleum sulfonate and alcohol in an aqueous
electrolyte solution. Key to the success of this approach were
systematic studies of oil displacement leading to recognition that a
dimensionless capillary number N c = ␮ v /␴ controlled the amount
of residual oil remaining after flooding an oil-containing core at
interstitial velocity v with an aqueous solution having a viscosity
µ and IFT ␴ with the oil (Taber 1969; Stegemeier 1977; Melrose
and Brandner 1974; Foster 1973). In situations when gravity is
important, the Bond number must be included (Pennell et al. 1996).
This work revealed that at typical reservoir velocities, IFT had
to be reduced from crude-oil/brine values of 20 to 30 mN/m to
values in the range of 0.001 to 0.01 mN/m to achieve low values
of residual-oil saturation (<0.05).
Several research groups found that ultralow IFTs in the required
range could be achieved using petroleum-sulfonate/alcohol mix-
tures (Hill et al. 1973; Foster 1973; Cayias et al. 1977). They
also found systematic variations of IFT when changing such vari-
ables as salinity, oil composition, and temperature. An important
889
contribution was the work of Healy et al. (1976) [see also Reed
and Healy (1977)], who demonstrated a relationship between IFT
and microemulsion phase behavior. Core tests using continuous
surfactant injection at the optimal salinity also yielded the highest
recovery of waterflood residual oil. Their studies used mixtures of
an alcohol cosolvent with synthetic alkyl/aryl sulfonates, in par-
ticular C9, C12, and C15 orthoxylene sulfonates, which can be made
from oligomers of propylene with more reproducible compositions
than those belonging to petroleum sulfonates.
Conventional Phase Behavior for Ultralow IFT. The under-
standing of ultralow IFT in oil-recovery processes was advanced
when Healy et al. (1976) explained how the Winsor definition of
equilibrium microemulsion phase behavior (I, II, and III, or lower-
phase, upper-phase, and middle-phase microemulsion, respec-
tively) described the changes of phase behavior, solubilization of
oil and water, and IFT as a function of salinity for anionic surfac-
tants. The surfactant is able to solubilize an increasing amount of
oil and a decreasing amount of water as salinity is increased. The
“optimal salinity” determined from phase behavior is the salinity
at which the microemulsion solubilizes equal amounts of oil and
water. The optimal salinity at which equilibrium IFTs between
the microemulsion phase and excess-oil or excess-water phase
become equal (and thus the sum becomes a minimum) is close to
the optimal salinity from phase behavior. There are correlations
between the “solubilization parameters” (ratio of oil/surfactant
Vo /Vs or water/surfactant Vw /Vs by volume) and the IFTs of the
microemulsion with the respective excess phases (Huh 1979).
Thus, one can estimate the value of the equilibrium IFT at the
optimal salinity from the value of the solubilization parameters at
the optimal salinity (where they are equal).
Nelson and Pope (1978) recognized that the appearance of a mid-
dle-phase microemulsion (Winsor III) is dependent on the amounts
of water, oil, and surfactant present. Thus, they defined the Type III
phase environment as the range of salinity at which a middle-phase
microemulsion may exist if one were to scan the water/oil/surfactant
ternary diagram. This distinction is important at very high or very low
surfactant concentrations because the volume of the middle-phase
microemulsion is proportional to the surfactant concentration. At high
surfactant concentrations, more of the excess phases are solubilized,
and thus the excess phases have smaller volume or are not present.
If the surfactant concentration is high enough, the “middle-phase”
microemulsion phase may appear as a single phase at or near optimal
conditions. On the other hand, at low surfactant concentrations but
above the critical micelle concentration, the volume of the middle-
phase microemulsion is minute and its presence may not be visually
detected or sampled for IFT measurements.
The nanostructure of the microemulsion should be recognized
to distinguish it from macroemulsions or liquid-crystal dispersions
or phases. Macroemulsions are nonequilibrium dispersions that
change with time or may be in a metastable condition. Liquid-
crystal phases are condensed, ordered phases that usually are
birefringent (rotate polarized light), viscous, and tend to inhibit
emulsion coalescence (Healey and Reed 1974). Microemulsions
are equilibrium isotropic phases that may have a bicontinuous
structure with near-zero mean curvature at or near optimal condi-
tions (Scriven 1976). (Microemulsions are oil-swollen micelles in
water at underoptimum conditions and reversed micelles in oil at
overoptimum conditions.) It was once thought that it is necessary
to have a cosolvent (alcohol) to have a microemulsion with an
anionic surfactant. However, it is now recognized that it is possible
to have microemulsions without alcohol at room temperature by
using branched surfactants (Abe et al. 1986).
Salinity-scan tests are used routinely to screen phase behavior
of surfactant formulations before conducting more time-consuming
coreflood tests (Levitt et al. 2009; Flaaten et al. 2009; Mohammadi
et al. 2009). The minimum IFT is correlated with the solubilization
parameters at the optimal salinity. The presence of viscous, struc-
tured, or birefringent phases and/or stable macroemulsions is easily
observed. Apparent viscosities of phases present in 5-mL samples
in sealed glass pipettes can be measured by the falling-sphere
method, even for opaque phases (Lopez-Salinas et al. 2009).
890
Surfactant Requirements and Structures
In a successful displacement process, the injected surfactant slug
must first achieve ultralow IFT to mobilize residual oil and create
an oil bank where both oil and water flow as continuous phases
(Bourrel and Schechter 1988). Second, it must maintain ultralow
IFT at the moving displacement front to prevent mobilized oil from
being trapped by capillary forces. Because of the way they are pre-
pared, commercial surfactants are invariably mixtures of multiple
species, which raises questions as to whether chromatographic
separation (i.e., preferential adsorption on pore surfaces or pref-
erential partitioning into the oil phase of some species) can cause
IFT variations with possible adverse effects on oil recovery. When
alcohol is used in the formulation, it partitions among the bulk-oil
and brine phases and the surfactant films in a manner different
from the surfactant. The alcohol must then be carefully selected
and tested to ensure there is no deleterious effect of chromato-
graphic separation (Dwarakanath et al. 2008; Sahni et al. 2010).
In the surfactant films, alcohol serves as a cosolvent, making the
films less rigid and thereby increasing the rate of equilibration and
preventing formation of undesirable viscous phases and emulsions
instead of the desired low-viscosity microemulsions. Alcohol can
also serve as a cosurfactant, altering, for instance, the optimal salin-
ity required to achieve ultralow IFT. Alcohols with short chains
such as propanol increase optimal salinity for sulfonate surfactants,
while longer-chain alcohols such as pentanol and hexanol decrease
optimal salinity. For petroleum sulfonates and synthetic alkyl/aryl
sulfonates with light crude oils, it has been found that 2-butanol
acts as a cosolvent but has less effect on optimal salinity than
other alcohols.
A disadvantage of using alcohol is that it decreases solubiliza-
tion of oil and water in microemulsions, and hence increases the
minimum value of IFT achievable with a given surfactant (Salter
1977). Also, it destabilizes foam that may be desired for mobility
control with the slug and in the drive. For temperatures below
approximately 60°C, the need for alcohol’s cosolvent effect can
be reduced or eliminated by some combination of the following
strategies: (1) using surfactants with branched hydrocarbon chains,
(2) adding ethylene oxide (EO) and/or less-hydrophilic propylene
oxide (PO) groups to the surfactant, and (3) using mixtures of
surfactants with different hydrocarbon-chain lengths or structures.
Such measures counter the tendency of long, straight hydrocarbon
chains of nearly equal length to form condensed surfactant films
and the lamellar liquid-crystalline phase. At high temperatures,
increased thermal motion promotes more-flexible surfactant films
and disruption of ordered structures, but it does not always elimi-
nate viscous phases and emulsions. Further studies are needed
to investigate usefulness of these or other strategies for high
temperatures.
The use of branched hydrocarbon chains to minimize or elimi-
nate alcohol requirements was discussed by Wade, Schechter, and
coworkers (Abe et al. 1986). An isotridecyl hydrophobe was used
in Exxon’s pilot test in the Loudon field. The hydrophilic part of
the surfactant was a chain consisting of short PO and EO segments
and a sulfate group (Maerker and Gale 1992). With this combina-
tion of the first two strategies, no alcohol was required. The Neodol
67 hydrophobe developed and manufactured by Shell Chemical has
an average of 1.5 methyl groups added randomly along a straight
C15–C16 chain that provides another type of branching. A propoxyl-
ated sulfate with this hydrophobe has been blended with an internal
olefin sulfonate, which is also branched, and used to displace west
Texas crude oils in low-salinity, ambient-temperature laboratory
tests, both with (Levitt et al. 2009) and without (Liu et al. 2008)
alcohol. In this case, all three strategies were combined.
Long-term surfactant stability at reservoir conditions is another
surfactant requirement. Provided that pH is maintained at slightly
alkaline levels and calcium concentration is not too high, hydro-
lysis of sulfate surfactants is limited for temperatures up to
50–60°C (Talley 1988). Surfactants with other head groups, most
likely sulfonates or carboxylates, will be needed for reservoirs
at higher temperatures. Because the sulfonate group is added at
different points along the hydrocarbon chain during synthesis,
internal olefin sulfonate (IOS) surfactants consist of species with
December 2011 SPE Journal
a twin-tailed structure, a different type of branching. Results of
laboratory studies of IOS phase behavior at high temperatures
have recently been presented (Barnes et al. 2008, 2010; Zhao
et al. 2008, Puerto et al. 2010). Hydrocarbon-chain lengths ranged
from C15–C18 to C24–C28. However, the effect of dissolved calcium
and magnesium ions, which most likely cause surfactant precipita-
tion, was not investigated in these studies (see Fig. 3 of Liu et al.
2008). Moreover, the current availability of internal olefins of long
chain length, as would be required for low or moderate reservoir
salinities, is limited. Alpha olefin sulfonate (AOS) surfactants have
the same potential disadvantages and are not as highly branched
as IOS surfactants.
Sulfonate groups cannot be added directly to alcohols, includ-
ing those with EO and/or PO chains. One approach is to prepare
sulfonates with glycidyl chloride or epichlorohydrin, where a
three-carbon chain is added between the EO or PO chain and
the sulfonate group. Wellington and Richardson (1997) described
some results with such surfactants. Further information on their
synthesis and initial phase-behavior results for propoxy glycidyl
sulfonates with the Neodol 67 hydrophobe is given by Barnes
et al. (2008). Puerto et al. (2010) presented phase behavior showing
that suitable ethoxy or propoxy glycidyl sulfonates could produce
microemulsions with high solubilization over a wide range of opti-
mal salinities with n-octane as the oil at temperatures up to 120°C.
They also noted that because these surfactants typically exhibit
phase separation (cloud point) at high temperatures and salinities,
it may be necessary to blend them with other surfactants such as
IOS, which are also stable at high temperatures.
Until recent years, nearly all work was directed toward EOR in
sandstone reservoirs owing to concerns that in the high-divalent-
ion environment of carbonate reservoirs, petroleum or synthetic
sulfonates would adsorb excessively and/or form calcium and
magnesium salts that would either precipitate or partition into the
oil phase. The exception was the work of Adams and Schievelbein
(1987), who conducted laboratory experiments and two field tests
showing that oil could be displaced in a carbonate reservoir using
a mixture of petroleum sulfonates and ethoxylated sulfate surfac-
tants. The ethoxy groups add tolerance to divalent ions. Recent
work with carbonate reservoirs used ethoxylated or propoxylated
sulfates, as discussed in the later section on wettability.
Alcohol-Free Surfactant Slugs for Injection
The surfactant slug to be injected should be a single-phase micel-
lar solution. Especially when polymer is added to increase slug
viscosity, it is essential to prevent separation into polymer-rich
and surfactant-rich phases, which yields highly viscous phases
unsuitable for either injection or propagation through the forma-
tion (Trushenski 1977). At low temperatures, oil-free mixtures of
petroleum sulfonate/alcohol or synthetic sulfonate/alcohol mix-
tures with brine are often translucent micellar solutions at salinities
well below optimal, but contain lamellar liquid crystal and exhibit
birefringence near optimal salinity where ultralow IFT is found
upon mixing with crude oil (Miller et al. 1986). In the absence
of polymer, the lamellar phase is often dispersed in brine as
particles having maximum dimensions of at least several microm-
eters. When polymer is added to such a turbid dispersion of the
lamellar phase, it produces a polymer-rich aqueous solution and a
more concentrated surfactant dispersion (Qutubuddin et al. 1985).
This undesirable behavior can sometimes be avoided by adding
sufficient alcohol. However, use of alcohol has disadvantages, as
indicated previously.
The lamellar phase was observed in surfactant/brine mixtures in
the absence of oil even for Exxon’s Loudon formulation mentioned
previously (Ghosh 1985), where, as indicated, branching and
addition of EO/PO groups allowed low-viscosity microemulsions
to be formed and ultralow IFT to be achieved with the crude oil
without the need to add alcohol. Exxon avoided phase separation
when polymer was added to the injected slug by including a paraf-
finic white oil of high molecular weight in the formulation. That
is, they injected a white oil-in-water microemulsion (Winsor I),
which became a bicontinuous microemulsion when mixed with
December 2011 SPE Journal
substantial volumes of crude oil in the reservoir (Maerker and
Gale 1992).
As temperature increases, the lamellar liquid-crystal phase
may melt. However, the surfactant/brine mixture is still unsuitable
for injection if separation into two or more liquid phases occurs
near optimal salinity (Benton and Miller 1983). Even if bulk
phase separation does not occur, turbid solutions are sometimes
observed. These solutions usually have large, anisotropic micelles
and separate into surfactant-rich and polymer-rich phases with the
addition of polymer. These phases can separate and/or plug the
porous media into which they are injected. Adding alcohol can
reduce micelle size and prevent phase separation in some cases.
As indicated previously, addition of a paraffinic-oil which yields an
oil-in-water microemulsion with nearly spherical drops is another
approach for preventing phase separation when polymer is present.
Within limits, the higher the molecular weight of the oil added to
produce an oil-in-water microemulsion, the less oil is needed to
formulate single phases with polymer for mobility control.
Another approach to formulate single-phase injection composi-
tions would be to find surfactants or surfactant blends that neither
exhibit phase separation nor form turbid solutions or liquid crystal-
line dispersions at conditions of interest (Flaaten et al. 2009; Sahni
et al. 2010). Blends of the branched surfactant Neodol 67 propoxyl-
ated sulfate (N67-7POS), having an average of seven PO groups
with the twin-tailed surfactant IOS 15-18, an IOS (Barnes et al.
2010) made from a feedstock containing mainly C15–C18 chains, are
interesting in this respect. Fig. 1 shows phase behavior at ambient
temperature of 3 wt% aqueous solutions of such surfactant blends
containing 1 wt% Na2CO3 and varying NaCl concentration but no
alcohol or polymer. IOS 15-18 alone precipitates above 4 wt%
NaCl in such solutions. In contrast, solutions of the propoxylated
sulfate alone do not precipitate but instead become cloudy above
the same salinity, as droplets of a second liquid phase form and
scatter light. Addition of IOS 15-18 to the propoxylated sulfate
makes the mixture more hydrophilic, thereby raising the salinity at
which phase separation occurs to a value higher than for either sur-
factant alone. For instance, the 4:1 blend (hereafter NI blend) (i.e.,
80% N67-7POS and 20% IOS 15-18) exhibits phase separation
at approximately 6 wt% NaCl (plus 1% Na2CO3), although slight
cloudiness occurs above approximately 3.5 wt% NaCl. Addition
of 0.5 wt% of partially hydrolyzed polyacrylamide to a 0.5 wt%
solution of this blend in a solution containing 4 wt% NaCl and 1
wt% Na2CO3 produces phase separation, although similar addition
of polymer to a solution containing only 2 wt% NaCl does not
(Liu et al. 2008). Phase-behavior studies show that the optimal
salinity of this blend with a west Texas crude oil is approximately
5 wt% NaCl (with 1 wt% Na2CO3) when the amount of surfactant
present is much greater than soap formed from the naphthenic
acids in the crude oil. However, in alkaline/surfactant processes, it
is best to inject at lower salinities, as discussed later, because the
surfactant encounters conditions during the process with greater
ratios of soap-to-surfactant and correspondingly lower optimal
salinities. Indeed, excellent recovery of the west Texas crude oil
was observed in sandpack experiments when a single-phase mix-
ture of the 4:1 blend and polymer was injected at 2 wt% NaCl
with 1 wt% Na2CO3 (Liu et al. 2008). At 2% NaCl, the surfactant
micelles are not highly anisotropic, and polymer and surfactant can
coexist in the same phase.
A similar approach was used by Falls et al. (1994) in an
alkaline/surfactant field test. They added a small amount of the
nonionic surfactant Neodol 25-12 to the main injected surfactant, a
blend of IOSs, to make the formulation sufficiently hydrophilic to
form a single micellar solution during storage at ambient tempera-
ture. Because this surfactant becomes less hydrophilic at higher
temperatures, it did not adversely affect process performance at
the reservoir temperature of approximately 57°C.
Alkaline/Surfactant Processes: Role of Alkali
Nelson et al. (1984) proposed injection of a solution containing
both surfactant and alkali for EOR. Such processes have attracted
and continue to attract considerable interest. They have been
labeled by different names, but will be collectively described here
891
 10

9 Multiphase
Region
8 Phase boundary
7 Clear solution
6 Two clear phases
% NaCl
5 Precipitation
4 One-Phase
Region Cloudy solution
3
Cloudy after nine
2 months

0
1:1 4:1 9:1
IOS N67
N67:IOS (w/w)

Fig. 1—Effect of added NaCl on phase behavior of 3 wt% solutions of N67/IOS mixtures containing 1 wt% Na2CO3 (Liu et al.
2008).

as alkaline/surfactant processes (Nelson et al. 1984; Peru and
Lorenz 1990; Surkalo 1990; Baviere et al. 1995).
The primary role of the alkali in an alkaline/surfactant process
is to reduce adsorption of the surfactant during displacement
through the formation and sequestering divalent ions. An addi-
tional benefit of alkali is that the soap is formed in situ from the
naphthenic acid in the crude oil (Johnson 1976). As indicated pre-
viously, the generation of soap allows the surfactant to be injected
at lower salinities than if used alone, which further reduces adsorp-
tion and facilitates incorporation of polymer in the surfactant slug.
Also, alkali can alter formation wettability to reach either more
water-wet or more oil-wet states. In fractured oil-wet reservoirs,
the combined effect of alkali and surfactant in making the matrix
preferentially water-wet is essential for an effective process. These
benefits of alkali will occur only where alkali is present. Thus, it
is important to determine “alkali consumption,” which controls the
rate of propagation of alkali through the formation.
Reduced Surfactant Adsorption. The discussion here will be
limited to anionic surfactants (Wessen and Harwell 2000). The
primary mechanism for the adsorption of anionic surfactants on
sandstone- and carbonate-formation material is the ionic attrac-
tion between positively charged mineral sites and the negative
surfactant anion (Tabatabal et al. 1993; Zhang and Somasundaran
2006). Thus, the role of the alkali is to be a “potential-determining
ion” to reverse the charge on positively charged mineral sites. The
potential-determining ions for oxide minerals are the hydronium
and hydroxide ions. The pH at which the charge reverses is the
“isoelectric point” if measured by electrophoresis (zeta potential)
and is the “point-of-zero-charge” if determined by titration. The
values are tabulated for most common minerals (Lyklema 1995).
Silica is negatively charged at reservoir conditions and exhibits
negligible adsorption of anionic surfactants. Clays (at neutral pH)
have negative charge at the faces and positive charge at the edges.
The clay edges are alumina-like and thus are expected to reverse
their charge at a pH of approximately 9. Carbonate formations and
sandstone-cementing material can be calcite or dolomite. These
latter minerals also have an isoelectric point of approximately pH
9, but carbonate ions, as well as the calcium and magnesium ions,
are more significant potential-determining ions. The zeta potential
of calcite is negative even at neutral pH in the presence of 0.1 N
carbonate/bicarbonate ions (Hirasaki and Zhang 2004). If a for-
mation contains iron minerals, the oxidation/reduction conditions
influence whether the surface iron sites are Fe3+ or Fe2+. Adsorption
of anionic surfactant for one sandstone was found to be lower by
more than a factor of two for reducing rather than for oxidizing
conditions (Wang 1993). Surfactant adsorption is only one com-
ponent of surfactant retention. Phase trapping of surfactant can be
more significant and will be discussed later.
Alkaline preflush had been advocated for both sequestering
divalent ions and reducing sulfonate adsorption (Holm and Rob-
ertson 1981). In subsequent work, alkali has been injected with the
surfactant. Adsorption of anionic surfactants on Berea sandstone
was reduced several-fold with addition of sodium carbonate for
petroleum sulfonate (Bae and Petrick 1977) or with addition of
sodium silicate or hydroxide for alcohol ethoxysulfate (Nelson
et al. 1984). The reduction of adsorption on Berea sandstone with
sodium bicarbonate was 68% in a dynamic experiment (Peru and
Lorenz 1990). Static and dynamic adsorption of anionic surfactants
on calcite and dolomite was decreased by an order of magnitude
with addition of sodium carbonate, but insignificantly with sodium
hydroxide (see Figs. 2 through 4) (Hirasaki and Zhang 2004;
Seethepalli et al. 2004; Zhang et al. 2006; Liu et al. 2008; Tabata-
bal et al. 1993). (The TC Blend of Figs. 2 through 4 is an earlier
blend of isotridecyl 4PO sulfate and C12 3EO sulfate. Research
on it was discontinued because its optimal salinity was too high
for the application of interest.)
Divalent-Ion Sequestration. The phase behavior of anionic sur-
factant systems is much more sensitive to a change in divalent ions
(e.g., Ca2+ and Mg2+) compared to monovalent ions (e.g., Na+),
especially at low surfactant concentrations (Nelson 1981). This is
problematic in sandstones because of ion exchange between the
clay, brine, and surfactant micelles (Hill et al. 1977; Pope et al.
1978; Hirasaki 1982). This exchange can result in the phase behav-
ior becoming overoptimum, with resulting large surfactant reten-
tion (Glover et al. 1979, Gupta 1982). Alkali anions (e.g., carbon-
ate, silicate, and phosphate) that have low solubility product with
divalent cations will sequester divalent cations to low concentra-
tions (Holm and Robertson 1981). Hydroxide is not as effective for
sequestration of calcium because the solubility product of calcium
hydroxide is not very low. Sodium metaborate has recently been
introduced as an alkali that may sequester divalent ions (Flaaten
et al. 2009; Zhang et al. 2008). A common problem with alkali
injection is that softened water is needed to avoid scaling.
Generation of Soap. The original concept of alkali flooding was
the reduction of oil/water IFT by in-situ generation of soap, which
is an anionic surfactant, sodium naphthenate (Jennings 1975).
Ultralow IFT usually required injection of relatively fresh water
with a low concentration of alkali because optimal salinity (total
electrolyte concentration) of the in-situ-generated soap is usually
low (e.g., <1% electrolyte). If the alkali concentration is too low,
alkali consumption reactions may result in a large retardation of

892 December 2011 SPE Journal

 2.5

2.0

Adsorption Density,
10–3 mmol/m2
1.5

1.0

0.5

0.0
0.0 0.5 1.0 1.5 2.0 2.5

Residual Surf. Conc., mmol/l

Original 0.05%TC blend 0.05%TC blend 0.05%TC blend/3% Na2CO3

0.05%TC blend/3.5% Na2CO3 0.05%TC blend/4% Na2CO3 0.05%TC blend/4.5% Na2CO3
Original 0.1%TC blend 0.1%TC blend 0.1%TC blend//3% Na2CO3

Fig. 2—Static adsorption of TC blend surfactant on dolomite sand. BET surface area of the calcite: 17.8 m2/g (Zhang et al.
2006).

the alkali displacement front. The concept of alkaline/surfactant
flooding is to inject a surfactant with the alkaline solution such that
mixture of the in-situ-generated soap and injected surfactant has
an optimal salinity that is tailored to the reservoir fluids (Nelson
et al. 1984; Surkalo 1990).
The common method used to determine the amount of naph-
thenic acid in crude oil is the total acid number (TAN), determined
by nonaqueous titration with a base (Fan and Buckley 2007). If
sodium naphthenate is to act as a surfactant, it should partition
into the aqueous phase at low electrolyte concentrations and be
measurable by hyamine titration for anionic surfactants. It was
found that the sodium naphthenate determined by extraction into
the aqueous phase and measured by hyamine titration is less than
the TAN value (Liu et al. 2010). It is hypothesized that the TAN
includes components that are too lipophilic to be extracted to the
aqueous phase and/or too hydrophilic to be detected by hyamine
titration.
Alkali Consumption. The ASP process should be designed such
that displacement fronts of the alkali, surfactant, and polymer
travel together. The mechanisms responsible for the retardation of
the alkali front include silica dissolution, clay dissolution with zeo-
lite precipitation, anhydrite or gypsum dissolution with calcite (or
calcium hydroxide or silicate) precipitation, dolomite dissolution
with calcium and magnesium silicate precipitation, hydrogen-ion
exchange, divalent-ion exchange with precipitation, and mixing
with divalent ions in formation water with precipitation (Ehrlich
and Wygal 1977; Holm and Robertson 1981; Southwick 1985;
Cheng 1986; Novosad and Novosad 1984; Jensen and Radke 1988;
Mohammadi et al. 2009). Naphthenic acids in crude oil also react
with alkali and thus contribute to consumption, but the amount
is usually small compared to the mentioned inorganic mineral
reactions. Silica dissolution can be controlled by using a buffered
system such as sodium carbonate or silicate rather than hydroxide
(Southwick 1985). Clay dissolution is strongly dependent on the

1.0
Expr. 1
0.9 v=1.2 feet/day
beta=0.34±0.03 Expr. 2
Cl−
0.8 Φ=0.335
Dimensionless Concentration

0.7 Expr. 2
Expr. 1
v=12 feet/day
0.6 beta=0.22±0.03
Φ=0.338
0.5
Calculated
0.4 from dolomite
isotherm
0.3
beta=0.04
0.2

0.1

0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Injected Volume, PV

Fig. 3—Dynamic adsorption of 0.2% TC blend surfactant without Na2CO3 on dolomite sand (Zhang et al. 2006).

December 2011 SPE Journal 893

 1.0
0.1%CS330+0.1%TDA-4PO
0.9 +0.3MNa 2 CO 3
0.8 beta=0.07±0.04
Dimensionless Concentration
v=1.2 feet/day
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.0 0.5
Fig. 4—Dynamic adsorption of 0.2% TC blend/0.3-M Na2CO3 on dolomite sand (Zhang et al. 2006).
pH and type of clay, and is kinetically limited (Sydansk 1982).
Thus, acidic clay such as kaolinite, as well as high temperature,
will increase the importance of this mechanism.
A limitation of the application of sodium carbonate in carbonate
formations is that if anhydrite or gypsum is present, it will dissolve
and precipitate as calcite (Hirasaki et al. 2005; Liu 2007). This is
detrimental for dolomite formations because they may have origi-
nated from evaporite deposits where gypsum is usually present. An
alternative alkali is sodium metaborate (Zhang et al. 2008; Flaaten
et al. 2009). However, longer-term experiments and equilibrium
calculations indicate that this metaborate will also precipitate.
Alkaline/Surfactant Processes: Wettability
Alteration
Wettability is the next most important factor in waterflood recov-
ery after geology (Morrow 1990). The recovery efficiency of a
flooding process is a function of the displacement efficiency and
sweep efficiency. These efficiencies are a function of the residual-
oil saturation (waterflood and chemical flood) and mobility ratio,
respectively. The residual-oil saturation to waterflooding is a
function of wettability, with the lowest value at intermediate wet-
tability (Jadhunandan and Morrow 1995). The mobility ratio is a
function of the ratio of water relative permeability to oil relative
permeability at their respective endpoints or at a specific satura-
tion. The mobility ratio or relative permeability ratio becomes
progressively larger as the wettability changes from water-wet to
oil-wet (Anderson 1987b). When a formation is strongly oil-wet,
it can have both a high waterflood residual-oil saturation and
unfavorable mobility ratio. In addition, an oil-wet formation will
have capillary resistance to imbibition of water (Anderson 1987a).
Formation wettability can be altered by pH (Wagner and Leach
1959; Ehrlich et al. 1974; Takamura and Chow 1985; Buckley
et al. 1989, Dubey and Doe 1993), surfactants that adsorb on the
minerals (Somasundaran and Zhang 2006) or remove adsorbed
naphthenic acids (Standnes and Austad 2000), and acids or bases
(Cuiec 1977). These processes are now incorporated into chemi-
cal-flood simulators (Anderson et al. 2006; Delshad et al. 2006;
Adibhatla and Mohanty 2007, 2008).
Sandstone Formations. Wettability alteration to more water-wet
or more oil-wet conditions was proposed as one of the mechanisms
of caustic flooding (Wagner and Leach 1959; Ehrlich et al. 1974;
Johnson 1976). Our current understanding of microemulsion phase
behavior and wettability is that the system wettability is likely to
894
Experimental Data for NaCl
Experimental Data for Surfactant
Simulation Curve for Surfactant
Simulation Curve for NaCl
1.0 1.5 2.0 2.5 3.0
Injected Volume, PV
be preferentially water-wet when the salinity is below the optimal
salinity (Winsor I). When the system is overoptimum (Winsor
II), macroemulsions tend to be oil-external. An oil-external mac-
roemulsion will trap water and have a low oil and water relative
permeability, similar to what one expects with oil-wet porous
media. The optimal salinity for a conventional alkali flooding
system is dependent on the in-situ-generated sodium naphthenate
soap, and is usually below approximately 1% electrolyte strength.
Because salinity of reservoir brine typically exceeds this value, a
conventional alkali flood often generates overoptimum and oil-wet
conditions. We show later in this review that this behavior can
be avoided by injecting the alkali and surfactant in the Winsor I
region. After mixing with the fluids in the reservoir of interest, it
will pass through the Winsor III, low-IFT region. Even a high-
salinity sandstone formation that is initially oil-wet may be altered
to preferentially water-wet by injecting alkali with a hydrophilic
surfactant in the Winsor I region.
Carbonate Formations. Wettability alteration has received more
attention recently for carbonate formations compared to sand-
stones because carbonate formations are much more likely to be
preferentially oil-wet (Treiber and Owens 1972). Also, carbonate
formations are more likely to be fractured and will depend on
spontaneous imbibition or buoyancy for displacement of oil from
the matrix to the fracture.
Wettability-alteration tests on plates of calcite, marble, lime-
stone, and dolomite with different surfactants and sodium carbon-
ate have been used to identify many systems that are altered to
preferentially water-wet with low anionic-surfactant concentra-
tions (Hirasaki and Zhang 2004; Seethepalli et al. 2004; Zhang
et al. 2006; Adibhatla and Mohanty 2008; Gupta et al. 2009).
Sodium carbonate has an important role because the carbonate ion
is a potential-determining ion for calcite and dolomite (Hirasaki
and Zhang 2004).
Spontaneous Imbibition. Spontaneous imbibition is the process
by which a wetting fluid is drawn into a porous medium by capil-
lary action (Morrow and Mason 2001). The presence of surfactant
in some cases lowers the IFT, and thus the capillary pressure, to
negligible values. Spontaneous displacement can still occur in this
case by buoyancy or gravity drainage (Schechter et al. 1994).
The research group of Austad has investigated spontane-
ous imbibition into chalk-formation material with enhancement
by cationic and nonionic surfactants and/or sulfate ions present in
December 2011 SPE Journal
seawater (Austad et al. 1998; Standnes and Austad 2000; Milter
and Austad 1996a, 1996b; Høgnesen et al. 2004, 2006). Sponta-
neous imbibition was most effective with dodecyl trimethylam-
monium or amine surfactants. The mechanism is thought to be
removal of adsorbed naphthenic acids through ion pairing with the
cationic surfactant. Capillarity may dominate during earlier time,
with gravity dominating later.
Laboratory and field testing of surfactant-enhanced imbibition
was investigated for the dolomite formation of the Yates field
(Chen et al. 2000; Yang and Wadleigh 2000). Both mass-balance
and CT scans showed increased oil recovery with 0.35% nonionic
and ethoxylated sulfate surfactants compared to reservoir brine.
A single-well injection, soak, and production test showed increased
oil recovery with decrease in water/oil ratio (WOR). The CT scans
showed that with formation brine, the oil recovery involved coun-
tercurrent imbibition, but with surfactant, the displacement was
dominated by gravity after early time.
An investigation compared the spontaneous recovery of the
Yates system with 0.35% ethoxylated alcohol and dodecyl tri-
methylammonium bromide [C12TAB, Standnes et al. (2002)].
The superior recovery of C12TAB was interpreted to be caused by
advancing contact angle of 32° for C12TAB, compared with 107°
for the ethoxylated alcohol and 133° for brine.
Laboratory measurements of spontaneous oil recovery were
made on the Yates system or outcrop limestone with alkaline/
surfactant solutions with 0.05% anionic surfactant and sodium
carbonate (Hirasaki and Zhang 2004; Seethepalli 2004; Zhang
et al. 2006; Adibhatla and Mohanty 2008). There was no recovery
(Yates cores) or only little recovery (outcrop cores) with brine
as the imbibing fluid. However, the alkaline/surfactant solutions
recovered as much as 60% of the oil. For a given system, tem-
perature is an important factor for wettability alteration and rate
of imbibition oil recovery (Gupta and Mohanty 2010).
Surfactant-aided wettability alteration and spontaneous oil
recovery may not have a significant contribution from capillary
pressure if the IFT is reduced to ultralow values. However, in these
conditions, gravity becomes an important contribution (Babadagli
2001; Hirasaki and Zhang 2004, Zhang et al. 2006; Adibhatla and
Mohanty 2007; Gupta and Mohanty 2010, 2008). Fig. 5 illustrates
the simulated velocity field during spontaneous displacement for
a cylindrical core immersed in a 0.05% alkaline/surfactant solu-
tion. The aqueous phase enters from the sides, and oil flows out
from the top of the core. The significance of gravity-dominated
displacement is that time scales proportionally with the character-
istic length of the matrix, whereas it would scale with the square
of the length if the process were capillary-dominated (Hirasaki
and Zhang 2004).
Najafabadi et al. (2008) and Delshad et al. (2009) simulated
cases of fractured oil-wet systems where negative capillary pres-
sure inhibited oil displacement from the matrix. Wettability altera-
tion and low IFT reduced the unfavorable capillary pressure so that
the pressure gradient from flow in the fractures could displace oil
from the matrix by viscous forces.
Emulsification of Heavy Oil. Alkali and surfactant can be used to
recover viscous, heavy oil by emulsification and wettability altera-
tion along with displacing the oil as a lower-viscosity, oil-in-water
emulsion (Liu et al. 2006; Bryan et al. 2008; Li et al. 2010; Kumar
and Mohanty 2010; Kumar et al. 2010).
Alkaline/Surfactant Processes: Wide Region
of Ultralow IFT
Surfactant processes for enhancing oil recovery are based on
achieving ultralow IFT (e.g., <10–2 mN/m) to either raise the capil-
lary number in forced displacement or to raise the Bond number
for gravity-driven displacement (Pennell et al. 1996) from a matrix
that is surrounded by fractures or vugs bearing surfactant solution.
For most anionic surfactants, especially those sensitive to divalent
ions, ultralow IFT occurs only over a limited range of conditions
[Healy et al. (1976); IFT vs. salinity plot shows a narrow region
of ultralow IFT]. It has always been a process-design challenge
to maintain or pass through these narrow conditions of Winsor
December 2011 SPE Journal
III or middle-phase microemulsion to have ultralow IFT during
the oil-recovery displacement process. Recently, it was found that
blending alkyloxylated sulfate and IOS surfactants in an alkaline
environment produced ultralow IFT for a wide range of conditions
even when the system has Winsor I phase behavior. This is an
important discovery because surfactant retention is much less for
Winsor I compared with Winsor III phase behavior. These results
are reviewed here.
Middle Layer of Lower-Phase Microemulsion. When a salin-
ity scan test is conducted at low surfactant concentrations (e.g.,
0.05%), the equilibrium phase behavior appears to go from a
lower-phase microemulsion to an upper-phase microemulsion over
a narrow salinity range (Fig. 6) (Zhang et al. 2006: Liu et al.
2008). Middle-phase microemulsions are rarely seen at low sur-
factant concentrations; thus, IFT measurements for salinity scans
for low surfactant concentration are usually between the upper
and lower phases observed in the sample tubes. The phases may
be (1) lower-phase microemulsion and excess oil (Winsor I), (2)
excess brine and upper-phase microemulsion (Winsor II), or (3)
excess brine and excess oil (Winsor III). The value of the IFTs
for these systems, measured between the upper and lower phases,
was not reproducible until a protocol was developed to include a
small volume of intermediate-density material in conducting spin-
ning-drop measurement. In Winsor III, this material would include
the middle-phase microemulsion. In Winsor I (lower-phase micro-
emulsion or underoptimum salinity), this intermediate-density
material appeared to be a colloidal dispersion in the lower-phase
microemulsion (Fig. 7). This colloidal dispersion formed a middle
layer between the excess oil and lower-phase microemulsion of
alkaline/surfactant systems with crude oil with a TAN of 0.3 mg
KOH/g or greater. This middle layer was more opaque than the
rest of the lower-phase microemulsion, and can be interpreted to
be more oil-rich from its intermediate density. When the oil/water
ratio is increased, the volume of this layer also increased. Because
this layer is not observed in the absence of alkali, it is hypothesized
that the dispersed colloidal material is an oil-and-sodium naphthe-
nate-rich microemulsion that is in equilibrium with the remainder
of the lower-phase microemulsion. As shown later, the soap is
more lipophilic than the added surfactant. This layer is not just a
macroemulsion dispersion that has not yet coalesced because the
presence of this material affects the value of the IFT.
IFT With and Without Alkali. The IFT of the system with and
without alkali was measured to test the hypothesis that the soap
generated by the alkali is responsible for the ultralow IFT in the
presence of the colloidal-dispersion material. Fig. 8 compares
the measured IFT with and without alkali. In the absence of
alkali, the lower-phase microemulsion was homogeneous, and the
ultralow IFT occurred only near the optimal salinity, as expected
for conventional surfactant EOR systems. In the presence of
alkali and using the protocol to ensure that a small volume of
colloidal dispersion was present, a wider, ultralow IFT region was
observed, especially for underoptimum conditions. If the colloidal
dispersion is not present as a result of creaming or centrifugation,
IFT behavior is similar to that in the absence of alkali [Liu et al.
(2008), Fig. 14]. Apparently, this colloidal dispersion contains
surface-active species responsible for lowering IFT between the
lower-phase microemulsion and the excess-oil phase in a manner
similar to behavior of the middle-phase microemulsion between
the excess-brine and excess-oil phases.
Consistent With Solubilization Parameter. The wide range of
ultralow IFT was verified by comparison with the Huh correlation
(Huh 1979). The solubilization ratios of the alkaline NI blend with
Yates crude oil are shown in Fig. 9. The IFTs calculated from
the Huh correlation and measured by spinning drop are shown
in Fig. 10. Thus, the measured IFT is consistent with the volume
of oil that is solubilized into the microemulsion phase(s)/volume
of surfactant. The generality of the wide ultralow IFT was tested
with another crude oil with a TAN of 4.79 mg KOH/g. The lower-
phase microemulsion of this crude oil was too dark to observe the
895
 Aqueous Phase Velocity 10.8852

0.09

0.08

Distance From Bottom, m ---> 0.07

0.06

0.05

0.04

0.03

0.02

0.01

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
(a) Distance From Center, m --->

Oil Phase Velocity 10.8852

0.09

0.08

0.07
Distance From Bottom, m --->

0.06

0.05

0.04

0.03

0.02

0.01

0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
(b) Distance From Center, m --->

Fig. 5—Velocity field in a cylindrical core immersed in surfactant solution at 10.89 days: (a) aqueous phase, (b) oil phase (Gupta
and Mohanty 2010).

spinning drop for IFT measurements. The IFT estimated from the
Huh correlation shows this crude oil to also have a wider, ultralow
IFT region of salinities, especially in the underoptimum region
(Liu et al. 2010).
Alkaline/Surfactant Processes: Phase
Behavior of Soap/Surfactant
Alkali saponifies the naphthenic acid in crude oil in situ to generate
sodium naphthenate, a soap that helps to generate low IFT during
the displacement process. Thus, an alkaline/surfactant system
should be considered as a pseudo-two-surfactant system featuring
the injected surfactant and the soap. The two surfactants will likely
have different optimal salinities. Thus, a mixing rule is needed to
model how the optimal salinity changes with surfactant and soap
concentrations.
WOR and Surfactant Concentration. Optimal salinity was
observed to be a function of surfactant concentration and WOR

896 December 2011 SPE Journal

x= 0.2 0.8 1.4 2.0 2.6 3.2 3.6 4.0 4.5 5.0

Excess
Oil

Colloidal
Dispersion

Lower-Phase
Microemulsion

Fig. 6—Salinity scan for 0.2% NI blend, 1% Na2CO3 with MY4

crude oil for WOR = 3 after settling time of 28 days at 25°C.
x = wt% NaCl (from Liu et al. 2008).

Fig. 7—View of colloidal dispersion region near interface for

for an alkaline/surfactant system (Fig. 11). However, all of these
curves can be reduced to a single curve if plotted as a function of
the soap/surfactant ratio (Fig. 12). The latter figure compares the
curves of optimal salinity for the TC blend and NI blend surfactant
formulations and the same crude oil.
Mixing Rule. The modeling of alkaline/surfactant flooding will
benefit from a mixing rule for the optimal salinity. When the TAN
was used for the soap content, the experimental data deviated
significantly from the mixing rule of Salager et al. (1979). An
alternative approach to determine the soap content of crude oil is
to extract soap from the crude oil into alkaline, alcoholic water
and titrate for anionic-surfactant content by hyamine titration.
2% NaCl sample from salinity scan, after 23 days settling (from
Liu et al. 2008).
(An alternative to analytical determination of the extractable
soap content is to estimate a value that will result in the best fit
to Eq. 1.)
The Salager et al. (1979) mixing rule was found to be followed
reasonably well when the aqueous-titration method was used to
quantify the soap content of the crude oil (Fig. 13, right panel).
The expression for the mixing rule is
( ) ( )
log (Optmix ) = X soap log Optsoap + 1 − X soap log (Optsurfactant )

1.E+01
Without Na2CO3

With 1% Na2CO3
1.E+00

1.E-01
IFT, mN/m

1.E-02

1.E-03

1.E-04
0 1 2 3 4 5 6
Salinity, % NaCl

Fig. 8—Measured IFT of system with and without Na2CO3 (from Liu et al. 2008).

December 2011 SPE Journal 897

 1000 1.E-01
Chun-Huh Correlation

Vw /Vs Spinning Drop Measurement

Solubilization Ratio

100 1.E-02

IFT, mN/m
10 1.E-03
Vo/Vs

1 1.E-04
2 2.5 3 3.5 4 2.0 2.5 3.0 3.5 4.0
NaCl, % NaCl, %

Fig. 9—Measured solubilization ratios of salinity scan (from Fig. 10—Comparison of the IFT from the solubilization param-
Liu et al. 2008). eter and spinning-drop measurements (from Liu et al. 2008).

14 14
WOR=1 (TC blend)

Optimal NaCl Conc., %

12 12
Optimal NaCl Conc., %

WOR=3 (TC blend)

10 10 WOR=10 (TC blend)
8 8 NI blend
NI blend
6
6 TC blend
4
4 WOR=10
2
2 WOR=3
WOR=1 0
0 1.E-02 1.E-01 1.E+00 1.E+01
0.01 0.1 1 10
Soap/Synthetic Surfactant Mole Ratio
Surfactant Concentration, %
Fig. 12—Optimal salinity as a function of soap/surfactant ratio
Fig. 11—Optimal sodium chloride concentration of TC blend for NI and TC surfactant blends with MY4 crude oil (Liu et al.
as a function of WOR and surfactant concentration (settled for 2008).
more than 6 months) (from Zhang et al. 2006).

where ␴ *M = X soap ␴ soap

*
+ X surfactant ␴ surfactant
*
. . . . . . . . . . . . . . . . . . . . . (2)

Soap where ␴ *M , ␴ soap

*
, and ␴ surfactant
*
X soap = , mole fraction . . . . . . . . . . . . . . (1) are the optimal solubilization ratio of
Soap + Surfactant the mixture, soap, and surfactant, respectively.

The optimal-salinity mixing rule is used in UTCHEM for simu- IFT Measurements. The dependence of the optimal salinity on
lation of alkaline/surfactant processes (Mohammadi et al. 2009). the soap/surfactant ratio can be used to explain the difference of
In addition, the researchers found that the optimum solubilization minimum-equilibrium IFT that may be observed with different
ratio follows a linear mixing rule: surfactant concentrations and WOR. This also explains why the

Opt vs Soap Fraction (theory) Opt vs Soap Fraction (exp) Opt vs Soap Fraction (theory) Opt vs Soap Fraction (exp)
10 10
Optiman Salinity

Optimal Salinity

1
1

0.1
0.1
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
Xsoap Xsoap
((a)
a) (b)

Fig. 13—Relationship of optimal salinity and soap mole fraction by different acid-number methods for NI Blend and Yates oil
(Liu et al. 2010). (a) Nonaqueous phase titration; acid number = 0.75 mg KOH/g; (b) soap extraction by NaOH; acid number =
0.44 mg KOH/g.

898 December 2011 SPE Journal

 0.4
Brine Oil Bank
(Two Phases)
0.3
Residual Oil Saturation, PV
0.2
Chemical Slug
0.1
Polymer Drive
0 with the surfactant for the Winsor I conditions behind the active
0 0.4
Chemical Slug/Drive Injected, PV
Fig. 14—Oil saturation and surfactant production during ex-
periment with finite, overoptimum surfactant slug and salinity
gradient (Hirasaki et al. 1983).
minimum IFT of small oil drops on a calcite plate occurred at the
optimal salinity of the TC blend surfactant with zero soap fraction
(i.e., 10–12% NaCl) [Zhang et al. (2006); Figs. 4 and 13]. The
WOR of the small drops was very high. Equilibrium IFT measure-
ments had lower optimal salinity because the soap/surfactant ratios
were larger owing to lower WORs.
The dependence of optimal salinity on the soap/surfactant ratio
also explains the transient minimum IFT observations in spinning-
drop measurement of a fresh oil drop in fresh surfactant solution
(Liu 2007). The soap concentration of the oil drop is changing
as soap is being extracted from the small oil drop into the much
larger volume of surfactant solution. Thus, the soap/surfactant
concentration ratio of the oil drop changes from a large value to
near zero with time, and the minimum IFT (in time) occurs when
the soap/surfactant ratio of the oil drop corresponds to the ratio
that is optimal for the salinity of the surfactant solution.
Composition Gradients
Displacement of residual oil by surfactant flooding requires reduc-
ing the IFT to ultralow values such that disconnected oil droplets
can be mobilized. The ultralow IFT generally exists only in a nar-
row salinity range near the optimal salinity. During the 1970s and
1980s, two schools of thought developed about how ultralow IFT
could be achieved in the displacement process (Gupta and Trush-
enski 1979). One approach is to either preflush the formation to
reduce the formation salinity to a value near optimal, or to design
the surfactant formulation such that the optimal salinity is equal to
the formation salinity, with the surfactant slug and drive injected
at the formation salinity (Maerker and Gale 1992). In the former
case, success was limited because the more viscous surfactant slug
contacted portions of the reservoir that the preflush bypassed. In
the latter case, this problem is avoided because there is no change
in salinity because of dispersive mixing or crossflow. The other
approach is to have a salinity gradient such that the system has
overoptimum salinity ahead of, and underoptimum salinity behind,
the active region. In this case, the salinity profile is certain to pass
through the optimal salinity somewhere in the displacement-front
region (Nelson 1981).
Whether the salinity is constant or a salinity gradient is used,
the electrolyte composition is further challenged by divalent ions in
the formation brine and ion exchange from the clays to the flowing
December 2011 SPE Journal
phases (Hill et al. 1977; Pope et al. 1978; Glover et al. 1979; Gupta
Transition Zone 1982). It was discovered that the surfactant micelles or microemul-
sion droplets have an affinity for divalent ions similar to that of the
clays, and thus act as a flowing ion-exchange medium (Hirasaki
Two-Phases - II (+)
Two-Phases - II (−)
Three-Phases - III
1982; Hirasaki and Lawson 1986). The problem of divalent ions
is avoided by use of an alkali such as sodium carbonate or sodium
silicate (Holm and Robertson 1981).
Salinity Gradient. It was demonstrated that with a salinity
gradient:
1. Ahead of the active region, the system is overoptimum; sur-
factant is retarded by partitioning into the oil-phase.
2. The system passes through the active region of ultralow IFT
(Winsor III) where residual-oil displacement takes place.
3. Behind the active region, the system is underoptimum, with
lower-phase microemulsion, and the surfactant propagates with
the water velocity (Glover et al. 1979; Pope et al. 1979; Hirasaki
et al. 1983). Thus, the salinity gradient tended to focus the sur-
factant near the advancing displacement front where salinity is
optimal and the phase behavior is Winsor III (Fig. 14). Also, the
salinity gradient helps to maintain polymer flow in the same phase
0.8 1.2 1.6 region. The polymer is in the excess-brine phase in the Winsor II
and III phase environments (Gupta 1982; Tham et al. 1983). The
example in Fig. 15 was injected overoptimum only for illustra-
tion of surfactant transport with respect to salinity environment.
Overoptimum salinity environments (Winsor II) can have viscous,
high-internal-phase, water-in-oil emulsions [Hirasaki et al. (1983),
Fig. 14] that may be bypassed by the subsequent lower-salinity
fluids. In practice, the surfactant slug is injected in a near-opti-
mal to underoptimum salinity environment. Therefore, the gradi-
ent basically provides assurance that if overoptimum conditions
are unexpectedly reached during the process, the lower salinity
injected later will allow optimal conditions to be achieved and will
release surfactant trapped in the oil.
Soap/Surfactant Gradient. It was mentioned earlier that the
optimal salinity changes as the soap/surfactant ratio changes.
Thus, an alkaline/surfactant flood will have a gradient of optimal
salinity because of a gradient in the soap/surfactant ratio unless the
soap content is negligible or the surfactant and soap have identical
optimal salinity (although surfactant would likely not be used if the
soap had a suitable optimal salinity). A gradient in the soap/sur-
factant ratio exists because soap is generated in situ by interaction
between the alkali and the naphthenic acids in the crude oil, while
the synthetic surfactant is introduced with the injected fluid.
The role of the soap/surfactant gradient in the ASP process was
evaluated with a 1D finite-difference simulator (Liu et al. 2008,
2010). Example composition and IFT profiles (Fig. 15) show the
IFT dropping to ultralow values in a narrow region of the profile
as the optimal salinity passes across the system salinity, which
was constant in this example. There is only a short distance for
the oil saturation to be reduced to a low value before the IFT
again increases and traps any oil that has not been displaced. The
oil saturation that will be trapped is approximately the saturation
where the slope of the ultralow-IFT oil/water fractional-flow curve
becomes less than the dimensionless velocity of the displacement
front (Pope 1980; Hirasaki 1981; Ramakrishnan and Wasan 1988,
1989). Thus mobility control is important for displacement effi-
ciency in addition to sweep efficiency for ASP flooding. Finite-
difference simulation showed that recovery decreased from 95 to
86% as the aqueous viscosity decreased from 40 to 24 cp for oil
with viscosity of 19 cp (Liu et al. 2008). This is consistent with a
pair of experiments that differed only in polymer concentration.
The effects of salinity, surfactant concentration, acid number,
slug size, and dispersion on oil recovery are illustrated for a 0.2-
PV slug and laboratory-scale dispersion (Pe = 500) in Fig. 16 (Liu
et al. 2010). The system is the one discussed in the IFT and phase-
behavior sections, and the black dot in Fig. 17 represents condi-
tions of the successful sandpack experiment mentioned previously.
The effects of surfactant concentration and acid number (soap
content of the crude oil) are combined in a single parameter—the
899
 Fig. 15—Profiles for large slug (0.5 PV) with low dispersion near optimal salinity (2% NaCl) (from Liu et al. 2010).
soap fraction at the waterflood residual-oil saturation. The range
of salinity for greater-than-90% oil recovery is a function of the
soap fraction. The salinity for maximum oil recovery decreases
from optimal salinity of the surfactant to that of soap as the soap
fraction increases (straight line in the figure). The range of salini-
ties for potentially high oil recovery is substantial, especially in
the underoptimum region below the optimal line.
If the dispersion is increased to a representative field-scale
value [Pe = 50, Lake (1989)] with constant salinity and a 0.2-PV
slug, the region of greater-than-90% recovery all but disappears.
Dilution by mixing at the front and back of the surfactant slug low-
ers the surfactant concentration more than the soap concentration,
and the propagation velocity of the soap/surfactant ratio for optimal
salinity is greatly retarded. However, if the system is operated with
a salinity gradient, high oil recovery is again possible (Liu et al.
900
2010). The lower salinity of the drive compensates for the lower
surfactant concentration such that the region of optimal salinity
again propagates with a near-unit velocity. In addition, injection
of the surfactant slug and polymer drive with a salinity that is less
that the optimal salinity of the surfactant alone makes it possible
to inject the surfactant slug with polymer without separation of the
surfactant and polymer into separate phases (Gupta 1982; Liu et al.
2008). Also, the salinity gradient avoids the large surfactant reten-
tion from microemulsion trapping by the polymer drive (Glover
et al. 1979; Hirasaki et al. 1983).
Foam Mobility Control
Foam is usually considered as a means of mobility control for
gas-injection processes such as steam foam or CO2 foam. Foam
mobility control for surfactant flooding is a natural progression
December 2011 SPE Journal
 5.0

4.0
70% 50%
3.0 30%

Injected Salinity, %NaCl

2.0
90%

90%

1.0
Optimum
70%
Curve
50%
0.5

30%

0 0.2 0.4 0.6 0.8 1

Sor
X Soap

Acid No.=0.2mg/g, surfactant concentration=0.14%, salinity=4.0% NaCl (over-optimum)

Acid No.=0.2mg/g, surfactant concentration=0.14%, salinity=2.0% NaCl (near-optimum)

Acid No.=0.2mg/g, surfactant concentration=0.14%, salinity=1.0% NaCl (under -optimum)

Fig. 16—Recovery factor with small slug (0.2 PV) and low dispersion (Pe = 500) (Liu et al. 2010).

because the system already has surfactant present (Lawson and
Reisberg 1980). Moreover, at high temperatures, foam may be
favored because polymer degradation is a concern (Srivastava and
Nguyen 2010). In fact, foam was used for mobility control for
alkaline/surfactant flooding in China (Zhang et al. 2000; Wang
et al. 2001). Recently, it has been used to improve sweep in
surfactant/polymer flooding (Kang et al. 2010). It has also been
used as mobility control for surfactant aquifer remediation (Hira-
saki et al. 1997, 2000). Nonionic surfactants have been evaluated
for mobility control of CO2 EOR (Adkins et al. 2010a, b). They
can be injected dissolved in the CO2 phase and have less adsorp-
tion on carbonate formations compared with anionic surfactants.

Fig. 17—Displacement profiles for the displacement of MY residual crude oil by ASPF in 40-darcy sandpack (Li et al. 2010).

December 2011 SPE Journal 901

 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 2.0
Total PV
0 0.1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
Liquid PV

NIP Air NIP Air NIP Air IOS Air IOS Air IOS Air IOS Air IOS Air IOS

1 ft/day 20 ft/day

Fig. 18—Profiles of the displacement of 266-cp Crude B with ASP and ASPF (Li et al. 2010).

If, in addition, the oil/water IFT can be reduced to ultralow values,
a low-tension CO2 EOR process may be applicable for reservoirs
with pressures below the minimum miscibility pressure.
ASP Foam. The reduction of surfactant adsorption with alkali
may result in the polymer being the most expensive chemical in
the ASP process. Experiments in 1D sandpacks have shown that
an ASP process with the polymer drive replaced by a foam drive
is equally efficient. Fig. 17 is an experiment in which the ASP
slug is alternated with equal-sized slugs of gas. The foam drive
consists of slugs of the better-foaming surfactant component
(without polymer) alternated with equal-sized slugs of gas (Li
et al. 2010). Practically all of the 19-cp oil was recovered after 1.2
TPV injected, but with only 0.6 PV of liquid injected. Experiments
with different sands indicated that foam reduced mobility more
in higher-permeability media, making it particularly attractive in
layered systems.
ASP foam was used to recover a 266-cp, 4.8-mg KOH/g TAN
crude oil (Fig. 18). What was remarkable is that the apparent
viscosity of the displacement process was only 80 cp or less.
Apparently, the viscous oil was being transported as an oil-in-water
emulsion with much less resistance than that of the crude oil.
Sweep of Layered Sands. Fig. 19 compares sweep in two cases
of a 19:1 permeability contrast layered sandpack initially filled
with water dyed green. The sandpack is nearly completely swept
with 1.0 TPV of surfactant alternating with gas (SAG) while the
low-permeability layer is only one-quarter swept with water only
(Li et al. 2010). The sweep efficiency is compared in Fig. 20 as
a function of the PV of liquid injected for SAG, water alternating
gas, and waterflooding.
Potential for Fractured Formations. The improvement of sweep
in layered sands suggests that foam may be helpful in the sweep
of a system of fractures. Yan et al. (2006) showed that pregener-
ated foam does improve the sweep of parallel plates with different
apertures to simulate heterogeneous, parallel fractures. Also, the
increased pressure gradient caused by foam flow in the fractures
increases the driving force for displacement of oil from the matrix
(Haugen et al. 2010; Abbasi-Asl et al. 2010). Farajzadeh et al.

SAG, 6 psi, fg=1/3 Water only, 4 psi

0.0 TPV

0.2 TPV

0.4 TPV

0.6 TPV

0.8 TPV

1.0 TPV

Fig. 19—Comparison of SAG with waterflood in 19:1 permeability ratio sandpack (Li et al. 2010).

902 December 2011 SPE Journal


SAG
1.0
Sweep Efficiency
0.8
0.6
0.4
0.2
0.0
0 0.5 1 1.5
PVs of Liquid Injected
Fig. 20—Sweep in 19:1 permeability contrast sandpack with SAG, WAG, and waterflood (Li et al. 2010).
(2010) describe how the viscous pressure gradient caused by foam
flow in fractures can accelerate the production of oil that would
otherwise be produced from the matrix by gravity drainage.
Field Pilots
A pilot of the alkaline/surfactant process is described by Falls et al.
(1994). This pilot was tested without polymer, with the intention
of a subsequent test with polymer. Nevertheless, the interpretation
of induction logs suggested 100% displacement efficiency in the
region swept by the injected fluids.
A refinement of the alkaline/surfactant process with recent
understanding about the IOS is described by Buijse et al. (2010).
The paper also discusses the importance of the crude-oil composi-
tion. The process was tested in the field with a single well chemi-
cal-tracer test. Stoll et al. (2010) describe pilot tests in Oman.
Gao and Gao (2010) summarized the pilots in Daqing oil
field.
Conclusions
The technology of surfactant flooding has advanced to overcome
many of the past causes of failures and to reduce the amount of sur-
factant required. These developments are summarized as follows:
1. Surfactant adsorption can be significantly reduced in sandstone
and carbonate formations by injection of an alkali such as
sodium carbonate. The alkali also sequesters divalent ions. The
reduced adsorption permits lower surfactant concentrations.
2. A wide selection of surfactant structures is now available to meet
requirements for specific applications.
a. Branched alcohol alkoxylate sulfates and sulfonates are toler-
ant of divalent ions. Ethoxylation increases optimal salinity;
propoxylation decreases optimal salinity. In both cases,
EO or PO, the optimal salinity decreases with increasing
temperature.
b. Alkyloxylated glycidyl ether sulfonate is more expensive
than sulfate but is stable at elevated temperatures.
c. IOSs are low-cost, double-tailed surfactants.
3. Aqueous solutions of a blend of N67-7PO sulfate and IOS1518
with alkali have a larger single-phase region extending to higher
salinities and calcium-ion concentrations than either alone.
This blend, without alcohol, can form a single phase for injec-
tion with polymer but can form microemulsions with crude oil
without forming a gel.
4. Soap generated in situ by the alkali is a cosurfactant that can
change the phase behavior of the injected surfactant solution
from lower- to middle- to upper-phase microemulsions. It is
lower phase when injected, middle phase at the displacement
front, and upper phase ahead of the displacement front.
5. Injection of the surfactant and polymer at salinity that is under-
optimum with respect to the injected surfactant avoids surfac-
tant/polymer phase separation and microemulsion trapping.
December 2011 SPE Journal
SAG fg=2/3, 8psi
SAG fg=2/3, 6psi
SAG fg=4/5, 4psi
SAG fg=2/3, 4psi
WAG
SAG fg=3/4, 4psi
SAG fg=2/3, 2psi
Waterflood
SAG fg=1/3, 6psi
SAG fg=1/2, 4psi
WAG fg=4/5, 4psi
WAG fg=3/4, 4psi
2 2.5 3 WAG fg=2/3, 4psi
WAG fg=1/2, 4psi
Water fg=0, 4psi
6. The soap generated in situ by the alkali causes a middle layer
to form and coexist with the lower-phase microemulsion, which
results in ultralow IFT over a wide range of salinity.
7. Anionic surfactants and sodium carbonate can alter wettability
for either sandstone or carbonate formations. Spontaneous oil
displacement can occur by gravity drainage.
8. Foam can be used as the drive of the alkaline/surfactant process
in place of the polymer drive.
9. Foam can efficiently sweep layered and fractured systems.
Acknowledgments
The authors acknowledge the financial support by DOE grant
DE-FC26-03NT15406 and the Rice University Consortium on
Processes in Porous Media. The information and insight we gained
from our long collaboration with Gary Pope and Kishore Mohanty
are also acknowledged.
References
Abbasi-Asl, Y., Pope, G.A., and Delshad, M. 2010. Mechanistic Modeling
of Chemical Transport in Naturally Fractured Oil Reservoirs. Paper
SPE 129661 presented at the SPE Improved Oil Recovery Symposium,
Tulsa, 24–28 April. doi: 10.2118/129661-MS.
Abe, M., Schechter, D., Schechter, R.S., Wade, W.H., Weerasooriya, U.,
and Yiv, S. 1986. Microemulsion Formation with Branched Tail Poly-
oxyethelene Sulfonate Surfactants. Journal of Colloid and Interface
Science 114 (2): 342–356. doi: 10.1016/0021-9797(86)90420-0.
Adams, W.T. and Schievelbein, V.H. 1987. Surfactant Flooding Carbon-
ate Reservoirs. SPE Res Eng 2 (4): 619–626. SPE-12686-PA. doi:
10.2118/12686-PA.
Adibhatia, B. and Mohanty, K.K. 2007. Simulation of Surfactant-Aided
Gravity Drainage in Fractured Carbonates. Paper SPE 106161 pre-
sented at the SPE Reservoir Simulation Symposium, Houston, 26–28
February. doi: 10.2118/106161-MS.
Adibhatla, B. and Mohanty, K.K. 2008. Oil Recovery From Fractured
Carbonates by Surfactant-Aided Gravity Drainage: Laboratory Experi-
ments and Mechanistic Simulations. SPE Res Eval & Eng 11 (1):
119–130. SPE-99773-PA. doi: 10.2118/99773-PA.
Adkins, S.S., Chen, X., Chan, I., Torino, E., Nguyen, Q.P., Sanders, A.,
and Johnston, K.P. 2010a. Morphology and Stability of CO2-in-Water
Foams with Nonionic Hydrocarbon Surfactants. Langmuir 26 (8):
5335–5348. doi: 10.1021/la903663v.
Adkins, S.S., Chen, X., Nguyen, Q.P., Sanders, A.W., and Johnston, K.P.
2010b. Effect of branching on the interfacial properties of nonionic
hydrocarbon surfactants at the air-water and carbon dioxide-water
interfaces. Journal of Colloid and Interface Science 346 (2): 455–463.
doi: 10.1016/j.jcis.2009.12.059.
Anderson, G.A., Delshad, M., King, C.B., Mohammadi, H., and Pope, G.A.
2006. Optimization of Chemical Flooding in a Mixed-Wet Dolomite
Reservoir. Paper SPE 100082 presented at the SPE/DOE Symposium on
Improved Oil Recovery, Tulsa, 22–26 April. doi: 10.2118/100082-MS.
903
Anderson, W.G. 1987a. Wettability Literature Survey— Part 4: Effects of
Wettability on Capillary Pressure. J Pet Technol 39 (10): 1283–1300.
SPE-15271-PA. doi: 10.2118/15271-PA.
Anderson, W.G. 1987b. Wettability Literature Survey –Part 5: The Effects
of Wettability on Relative Permeability. J Pet Technol 39 (11): 1453–
1468. SPE-16323-PA. doi: 10.2118/16323-PA.
Atkinson, H. 1927. Recovery of Petroleum from Oil Bearing Sands. US
Patent No. 1,651,311.
Austad, T., Matre, B., Milter, J., Sævareid, A., and Øyno, L. 1998. Chemi-
cal flooding of oil reservoirs 8. Spontaneous oil expulsion from oil- and
water-wet low permeable chalk material by imbibition of aqueous sur-
factant solutions. Colloids and Surfaces A: Physico. Eng. Aspects 137
(1–3): 117–129. doi: 10.1016/S0927-7757(97)00378-6.
Babadagli, T. 2001. Scaling of Co-current and Counter-Current Capillary
Imbibition for Surfactant and Polymer Injection in Naturally Fractured Res-
ervoirs. SPE J. 6 (4): 465–478. SPE-74702-PA. doi: 10.2118/74702-PA.
Bae, J.H. and Petrick, C.B. 1977. Adsorption/Retention of Petroleum
Sulfonate in Berea Cores. SPE J. 17 (5): 353–357. SPE-5819-PA. doi:
10.2118/5819-PA.
Barnes, J.R., Dirkzwager, H., Smit, J.R., Smit, J.P., On, A., Navarrete, R.C.,
Ellison, B.H., and Buijse, M.A. 2010. Application of Internal Olefin
Sulfonates and Other Surfactants to EOR. Part 1: Structure—Perfor-
mance Relationships for Selection at Different Reservoir Conditions.
Paper SPE 129766 presented at the SPE Improved Oil Recovery Sym-
posium, Tulsa, 24–28 April. doi: 10.2118/129766-MS.
Barnes, J.R., Smit, J.P., Smit, J.R., Shpakoff, P.G., Raney, K.H., and
Puerto, M.C. 2008. Development of Surfactants for Chemical Flooding
at Difficult Reservoir Conditions. Paper SPE 113313 presented at the
SPE/DOE Symposium on Improved Oil Recovery, Tulsa, 20–23 April.
doi: 10.2118/113313-MS.
Baviere, M., Glenat, P., Plazanet, V., and Labrid, J. 1995. Improved EOR
by Use of Chemicals in Combination. SPE Res Eng 10 (3): 187–193.
SPE-27821-PA. doi: 10.2118/27821-PA.
Benton, W.J. and Miller, C.A. 1983. Lyotropic liquid crystalline phases
and dispersions in dilute anionic surfactant-alcohol-brine systems. 1.
Patterns of phase behavior. J. Phys. Chem. 87 (24): 4981–4991. doi:
10.1021/j150642a042.
Bourrel, M. and Schechter, R.S. 1988. Microemulsions and Related Sys-
tems: Formulation, Solvency, and Physical Properties, Vol. 30. New
York: Surfactant Science Series, Marcel Dekker.
Bryan, J, Mai, A., and Kantzas, A. 2008. Investigation into the Process
Responsible for Heavy Oil Recovery by Alkali-Surfactant Flooding.
Paper SPE 113993 presented at the SPE/DOE Symposium on Improved
Oil Recovery, Tulsa, 20–23 April. doi: 10.2118/113993-MS.
Buckley, J.S., Takamura, K., and Morrow, N.R. 1989. Influence of Electric
Surface Charges on the Wetting Properties of Crude Oils. SPE Res Eng
4 (3): 332–340. SPE-16964-PA. doi: 10.2118/16964-PA.
Buijse, M.A., Prelicz, R.M., Barnes, J.R., and Cosmo, C. 2010. Applica-
tion of Internal Olefin Sulfonates and Other Surfactants to EOR. Part
2: The Design and Execution of an ASP Field Test. Paper SPE 129769
presented at the SPE Improved Oil Recovery Symposium, Tulsa, 24–28
April. doi: 10.2118/129769-MS.
Cayias, J.L., Schechter, R.S., and Wade, W.H. 1977. The utilization of
petroleum sulfonates for producing low interfacial tensions between
hydrocarbons and water. J. Colloid Interface Sci. 59 (1): 31–38. doi:
10.1016/0021-9797(77)90335-6.
Chen, H.L., Lucas, L.R., Nogaret, L.A.D, Yang, H.D., and Kenyon, D.E.
2000. Laboratory Monitoring of Surfactant Imbibition Using Com-
puterized Tomography. Paper SPE 59006 presented at the SPE/DOE
International Petroleum Conference and Exhibition in Mexico, Villa-
hermosa, Mexico, 1–3 February. doi: 10.2118/59006-MS.
Cheng, K.H. 1986. Chemical Consumption During Alkaline Flood-
ing: A Comparative Evaluation. Paper SPE 14944 presented at the
SPE Enhanced Oil Recovery Symposium, Tulsa, 20–23 April. doi:
10.2118/14944-MS.
Cuiec, L. 1977. Study of Problems Related to the Restoration of the Natural
State of Core Samples. J Can Pet Technol 16 (4): 68–80. JCPT Paper
No. 77-04-09. doi: 10.2118/77-04-09.
Delshad, M., Najafabadi, N.F., and Sepehnoori, K. 2009. Scale Up Method-
ology for Wettability Modification in Fractured Carbonates. Paper SPE
118915 presented at the SPE Reservoir Simulation Symposium, The
Woodlands, Texas, 2–4 February. doi: 10.2118/118915-MS.
904
Delshad, M., Najafabadi, N.F., Anderson, G.A., Pope, G.A., and Sepehr-
noori, K. 2006. Modeling Wettability Alteration in Naturally Fractured
Reservoirs. Paper SPE 100081 presented at the SPE/DOE Symposium
on Improved Oil Recovery, Tulsa, 22–26 April. doi: 10.2118/100081-
MS.
Dubey, S.T. and Doe, P.H. 1993. Base Number and Wetting Properties
of Crude Oil. SPE Res Eng 8 (3): 195–200. SPE-22598-PA. doi:
10.2118/22598-PA.
Dwarakanath, V., Chaturvedi, T., Jackson, A.C., Malik, T., Siregar, A., and
Zhao, P. 2008. Using Co-Solvents to Provide Gradients and Improve
Oil Recovery During Chemical Flooding in a Light Oil Reservoir. Paper
SPE 113965 presented at the SPE/DOE Symposium on Improved Oil
Recovery, Tulsa, 19–23 April. doi: 10.2118/113965-MS.
Ehrlich, R. and Wygal, R.J. 1977. Interrelation of Crude Oil and Rock
Properties with the Recovery of Oil by Caustic Waterflooding. SPE J.
17 (4): 263–270. SPE-5830-PA. doi: 10.2118/5830-PA.
Ehrlich, R., Hasiba, H.H., and Raimondi, P. 1974. Alkaline Waterflooding
for Wettability Alteration-Evaluating a Potential Field Application.
J Pet Technol 26 (12): 1335–1342. SPE-4905-PA. doi: 10.2118/
4905-PA.
Falls, A.H., Thigpen, D.R., Nelson, R.C., Ciaston, J.W., Lawson, J.B.,
Good, P.A., Ueber, R.C., and Shahin, G.T.1994. Field Test of Cosur-
factant-Enhanced Alkaline Flooding. SPE Res Eng 9 (3): 217–223.
SPE-24117-PA. doi: 10.2118/24117-PA.
Fan, T. and Buckley, J.S. 2007. Acid Number Measurements Revisited. SPE
J. 12 (4): 496–500. SPE-99884-PA. doi: 10.2118/99884-PA.
Farajzadeh, R., Wassing, B., and Boerrigter, P. 2010. Foam Assisted Gas Oil
Gravity Drainage in Naturally-Fractured Reservoirs. Paper SPE 134203
presented at the SPE Annual Technical Conference and Exhibition,
Florence, Italy, 19–27 September. doi: 10.2118/134203-MS.
Flaaten, A.K., Nguyen, Q.P., Pope, G.A., and Zhang, J. 2009. A System-
atic laboratory Approach to Low-Cost, High-Performance Chemical
Flooding. SPE Res Eval & Eng 12 (5): 713–723. SPE-113469-PA. doi:
10.2118/113469-PA.
Foster, W.R. 1973. A Low Tension Waterflooding Process. J Pet Technol
25 (2): 205–210; Trans., AIME, 255. SPE-3803-PA. doi: 10.2118/3803-
PA.
Gao, S. and Gao, Q. 2010. Recent Progress and Evaluation of ASP Flood-
ing for EOR in Daqing Oil Field. Paper SPE 127714 presented at the
SPE EOR Conference at Oil & Gas West Asia, Muscat, Oman, 11–13
April. doi: 10.2118/127714-MS.
Ghosh, O. 1985. Liquid Crystal to Microemulsion Transitions in Anionic
Surfactant-Oil-Brine Systems. PhD thesis, Rice University, Ann Arbor,
Michigan (October 1985).
Glover, C.J., Puerto, M.C., Maerker, J.M., and Sandvik, E.L. 1979. Sur-
factant Phase Behavior and Retention in Porous Media. SPE J. 19 (3):
183–193. SPE-7053-PA. doi: 10.2118/7053-PA.
Gogarty, W.B. 1977. Oil recovery with surfactants: History and a current
appraisal. In Improved Oil Recovery by Surfactant and Polymer Flood-
ing, ed. D.O. Shah and R.S. Schechter, 27–54. New York: Academic
Press.
Gupta, R., Mohan, K., and Mohanty, K.K. 2009. Surfactant Screening for
Wettability Alteration in Oil-Wet Fractured Carbonates. Paper SPE
124822 presented at the SPE Annual Technical Conference and Exhibi-
tion, New Orleans, 19–27 September. doi: 10.2118/124822-MS.
Gupta, R. and Mohanty, K. 2010. Temperature Effects on Surfactant-Aided
Imbibition Into Fractured Carbonates. SPE J. 15 (3): 588-597. SPE-
110204-PA. doi: 10.2118/110204-PA.
Gupta, R. and Mohanty, K.K. 2008. Wettability Alteration of Fractured
Carbonate Reservoirs. Paper SPE 113407 presented at SPE/DOE
Symposium on Improved Oil Recovery, Tulsa, 20–23 April. doi:
10.2118/113407-MS.
Gupta, S.P. 1982. Dispersive Mixing Effects on the Sloss Field Micellar
System. SPE J. 22 (4): 481–492. SPE-9782-PA. doi: 10.2118/9782-
PA.
Gupta, S.P. and Trushenski, S.P. 1979. Micellar Flooding—Compositional
Effects on Oil Displacement. SPE J. 19 (2): 116–128; Trans., AIME,
267. SPE-7063-PA. doi: 10.2118/7063-PA.
Haugen, A., Ferno, M.A., Graue, A., and Bertin, H.J. 2010. Experimental
Study of Foam Flow in Fractured Oil-Wet Limestone for Enhanced
Oil Recovery. Paper SPE 129763 presented at the SPE Improved Oil
Recovery Symposium, Tulsa, 24–28 April. doi: 10.2118/129763-MS.
December 2011 SPE Journal
Healey, R.N. and Reed, R.L. 1974. Physicochemical Aspects of Micro-
emulsion Flooding. SPE J. 14 (5): 491–501; Trans., AIME, 265. SPE-
4583-PA. doi: 10.2118/4583-PA.
Healy, R.N., Reed, R.L., and Stenmark, D.K. 1976. Multiphase Microemul-
sion Systems. SPE J. 16 (3): 147–160; Trans., AIME, 261. SPE-5565-
PA. doi: 10.2118/5565-PA.
Hill, H.J., Helfferich, F.G., Lake, L.W., and Reisberg, J. 1977. Cation
Exchange and Chemical Flooding. J Pet Technol 29 (10): 1336–1338.
SPE-6642-PA. doi: 10.2118/6642-PA.
Hill, H.J., Reisberg, J., and Stegemeier, G.L. 1973. Aqueous Surfactant
Systems For Oil Recovery. J Pet Techol 25 (2): 186–194. SPE-3798-
PA. doi: 10.2118/3798-PA.
Hirasaki, G. and Zhang, D.L. 2004. Surface Chemistry of Oil Recovery
From Fractured, Oil-Wet, Carbonate Formation. SPE J. 9 (2): 151–162.
SPE-88365-PA. doi: 10.2118/88365-PA.
Hirasaki, G., Miller C.A., Pope, G.A., and Jackson, R.E. 2005. Surfactant
Based Enhanced Oil Recovery and Foam Mobility Control. 2nd Annual
Technical Report (July 2004–June 2005), Contract No. DE-FC26-
03NT15406, US DOE, Washington, DC (July 2005).
Hirasaki, G.J. 1981. Application of the Theory of Multicomponent, Multi-
phase Displacement to Three-Component, Two-Phase Surfactant Flood-
ing. SPE J. 21 (2): 191–204. SPE-8373-PA. doi: 10.2118/8373-PA.
Hirasaki, G.J. 1982. Ion Exchange With Clays in the Presence of Surfactant.
SPE J. 22 (2): 181–192. SPE-9279-PA. doi: 10.2118/9279-PA.
Hirasaki, G.J. and Lawson, J.B. 1986. An Electrostatic Approach to the
Association of Sodium and Calcium with Surfactant Micelles. SPE
Res Eng 1 (2): 119–130; Trans., AIME, 281. SPE-10921-PA. doi:
10.2118/10921-PA.
Hirasaki, G. J. et al. 1997. Field Demonstration of the Surfactant/Foam
Process for Aquifer Remediation. SPE 39292 presented at the SPE
Annual Technical Conference and Exhibition, San Antonio, 5–8 Octo-
ber. doi: 10.2118/39292-MS.
Hirasaki, G.J., Jackson, R.E., Jin, M., Lawson, J.B., Londergan, J.,
Meinardus, H., Miller, C.A., Pope, G.A., Szafranski, R., and Tanzil,
D. 2000. Field Demonstration of the Surfactant/Foam Process for
Remediation of a Heterogeneous Aquifer Contaminated with DNAPL.
In NAPL Removal: Surfactants, Foams, and Microemulsions, ed. S.
Fiorenza, C.A. Miller, C.L. Oubre, and C.H. Ward, Part 1, 3–163.
AATDF monograph series, Boca Raton, Florida, USA: Lewis Publish-
ers, CRC Press.
Hirasaki, G.J., van Domselaar, H.R., and Nelson, R.C. 1983. Evaluation of
the Salinity Gradient Concept in Surfactant Flooding. SPE J. 23 (3):
486–500. SPE-8825-PA. doi: 10.2118/8825-PA.
Høgnesen, E.J., Olsen, M., and Austad, T. 2006. Capillary and Gravity
Dominated Flow Regimes in Displacement of Oil from an Oil-Wet
Chalk Using Cationic Surfactant. Energy Fuels 20 (3): 1118–1122.
doi: 10.1021/ef050297s.
Høgnesen, E.J., Standnes, D.C., and Austad, T. 2004. Scaling Sponta-
neous Imbibition of Aqueous Surfactant Solution Into Preferential
Oil-Wet Carbonates. Energy Fuels 18 (6): 1665–1675. doi: 10.1021/
ef040035a.
Holm, L.W. 1977. Soluble oils for improved oil recovery. In Improved Oil
Recovery by Surfactant and Polymer Flooding, ed. D.O. Shah and R.S.
Schechter, 453–485. New York: Academic Press.
Holm, L.W. and Robertson, S.D. 1981. Improved Micellar/Polymer Flood-
ing With High-pH Chemicals. J Pet Technol 33 (1): 161–171. SPE-
7583-PA. doi: 10.2118/7583-PA.
Huh, C. 1979. Interfacial tensions and solubilizing ability of a micro-
emulsion phase that coexists with oil and brine. Journal of Col-
loid and Interface Science 71 (2): 408–426. doi: 10.1016/0021-
9797(79)90249-2.
Jadhunandan, P.P. and Morrow, N.R. 1995. Effect of Wettability on Water-
flood Recovery for Crude-Oil/Brine/Rock Systems. SPE Form Eval 10
(1): 40–46. SPE-22597-PA. doi: 10.2118/22597-PA.
Jennings, H.Y. Jr. 1975. A Study of Caustic Solution-Crude Oil Interfacial
Tensions. SPE J. 15 (3): 197–202. SPE-5049-PA. doi: 10.2118/5049-
PA.
Jensen, J.A. and Radke, C.J. 1988. Chromatographic Transport of Alkaline
Buffers Through Reservoir Rock. SPE Res Eng 3 (3): 849–856. SPE-
14295-PA. doi: 10.2118/14295-PA.
Johnson, C.E. Jr. 1976. Status of Caustic and Emulsion Methods. J Pet
Technol 28 (1): 85–92. SPE-5561-PA. doi: 10.2118/5561-PA.
December 2011 SPE Journal
Kang, W., Liu, S., Meng, L., Cao, D., and Fan, H. 2010. A Novel Ultra-low
Interfacial Tension Foam Flooding Agent to Enhance Heavy Oil Recov-
ery. Paper SPE 129175 presented at the SPE Improved Oil Recovery
Symposium, Tulsa, 24–28 April. doi: 10.2118/129175-MS.
Kumar, R. and Mohanty, K.K. 2010. ASP Flooding of Viscous Oil. Paper
SPE 135265 presented at the SPE Annual Technical Conference and
Exhibition, Florence, Italy, 19–27 September. doi: 10.2118/135265-
MS.
Kumar, R., Dao, E., and Mohanty, K.K. 2010. Emulsion Flooding of Heavy
Oil. Paper SPE 129914 presented at the SPE/DOE Enhanced Oil Recov-
ery Symposium, Tulsa, 20–23 April. doi: 10.2118/129914-MS.
Lake, L.W. 1989. Enhanced Oil Recovery, 166. Englewood Cliffs, New
Jersey: Prentice Hall.
Lawson, J.B. and Reisberg, J. 1980. Alternate Slugs of Gas and Dilute
Surfactant for Mobility Control During Chemical Flooding. Paper SPE
8839 presented at the SPE/DOE Enhanced Oil Recovery Symposium,
Tulsa, 20–23 April. doi: 10.2118/8839-MS.
Levitt, D.B., Jackson, A.C., Heinson, C., Britton, L.N., Malik, T., Dwara-
kanath, V., and Pope, G.A. 2009. The Identification and Evaluation
of High-Performance EOR Surfactants. SPE Res Eval & Eng 12 (2):
243–253. SPE-100089-PA. doi: 10.2118/100089-PA.
Li, R.F., Yan, W., Liu, S., Hirasaki, G., and Miller, C.A. 2010. Foam
Mobility Control for Surfactant Enhanced Oil Recovery. SPE J. 15 (4):
928-942. SPE-113910-PA. doi: 10.2118/113910-PA.
Liu, Q., Dong, M., and Ma, S. 2006. Alkaline/Surfactant Flood Potential in
Western Canadian Heavy Oil Reservoirs. Paper SPE 99791 presented
at the SPE/DOE Symposium on Improved Oil Recovery, Tulsa, 22–26
April. doi: 10.2118/99791-MS.
Liu, S. 2007. Alkaline Surfactant Polymer Enhanced Oil Recovery Pro-
cesses. Ph.D. thesis, Rice University, Ann Arbor, Michigan (Decem-
ber 2007). http://www.owlnet.rice.edu/~gjh/Consortium/Manuscripts/
Liu_Thesis.pdf.
Liu, S., Feng Li, R., Miller, C.A., and Hirasaki, G.J. 2010. Alkaline/Surfac-
tant/Polymer Processes: Wide Range of Conditions for Good Recovery.
SPE J. 15 (2): 282-293. SPE-113936-PA. doi: 10.2118/113936-PA.
Liu, S., Zhang, D.L., Yan, W., Puerto, M., Hirasaki, G.J., and Miller, C.A.
2008. Favorable Attributes of Alkali-Surfactant-Polymer Flooding. SPE
J. 13 (1): 5–16. SPE-99744-PA. doi: 10.2118/99744-PA.
Lopez-Salinas, J.L., Miller, C.A., Koo, K.H.K., Puerto, M., and Hirasaki,
G.J. 2009 Viscometer for opaque, sealed microemulsion samples.
Paper SPE 121575 presented at the SPE International Symposium on
Oilfield Chemistry, The Woodlands, Texas, USA, 20–22 April. doi:
10.2118/121575-MS.
Lyklema, J. 1995. Fundamentals of Interface and Colloid Science. Volume
II: Solid-Liquid Interfaces, Appendix 3, A3.1–A3.6. San Diego, Cali-
fornia: Academic Press.
Maerker, J.M. and Gale, W.W. 1992. Surfactant Flood Process Design for
Loudon. SPE Res Eng 7 (1): 36–44; Trans., AIME, 293. SPE-20218-
PA. doi: 10.2118/20218-PA.
Melrose, J.C. and Brandner, C.F. 1974. Role of Capillary Forces In
Detennining Microscopic Displacement Efficiency For Oil Recovery
By Waterflooding. J Can Pet Technol 13 (4): 54–62. JCPT Paper No.
74-04-05. doi: 10.2118/74-04-05.
Miller, C.A., Ghosh, O., and Benton, W.J. 1986. Behavior of dilute lamellar
liquid-crystalline phases. Colloids and Surfaces 19 (2–3): 197–223. doi:
10.1016/0166-6622(86)80336-5.
Milter, J. and Austad, T. 1996a. Chemical flooding of oil reservoirs
6. Evaluation of the mechanism for oil expulsion by spontaneous
imbibition of brine with and without surfactant in water-wet, low-
permeable, chalk material. Colloids and Surfaces A: Physicochemi-
cal and Engineering Aspects 113 (3): 269–278. doi: 10.1016/0927-
7757(96)03631-X.
Milter, J. and Austad, T. 1996b. Chemical flooding of oil reservoirs 7.
Oil expulsion by spontaneous imbibition of brine with and without
surfactant in mixed-wet, low permeability chalk material. Colloids
and Surfaces A: Physicochemical and Engineering Aspects 117 (1–2):
109–115. doi: 10.1016/0927-7757(96)03693-X.
Mohammadi, H., Delshad, M., and Pope, G.A. 2009. Mechanistic Model-
ing of Alkaline/Surfactant/Polymer Floods. SPE Res Eval & Eng 12
(4): 518–527. SPE-110212-PA. doi: 10.2118/110212-PA.
Morrow, N.R. 1990. Wettability and Its Effect on Oil Recovery. J Pet Tech-
nol 24 (12): 1476–1484. SPE-21621-PA. doi: 10.2118/21621-PA.
905
Morrow, N.R. and Mason, G. 2001. Recovery of Oil by Spontaneous Imbi-
bition. Current Opinion in Colloid & Interface Science 6 (4): 321–337.
doi: 10.1016/S1359-0294(01)00100-5.
Najafabadi, N.F., Delshad, M., Sepehrnoori, K., Nguyen, Q.P., and Zhang,
J. 2008. Chemical Flooding of Fractured Carbonates Using Wettability
Modifiers. Paper SPE 113369 presented at the SPE/DOE Symposium
on Improved Oil Recovery, Tulsa, 20–23 April. doi: 10.2118/113369-
MS.
Nelson, R.C. 1981. Further Studies on Phase Relations in Chemical Flood-
ing. In Surface Phenomena in Enhanced Oil Recovery, ed. D.O. Shah,
73–104. New York: Plenum Publishing.
Nelson, R.C. and Pope, G.A. 1978. Phase Relationships in Chemical
Flooding. SPE J. 18 (5): 325–338; Trans., AIME, 265. SPE-6773-PA.
doi: 10.2118/6773-PA.
Nelson, R.C., Lawson, J.B., Thigpen, D.R., and Stegemeier, G.L. 1984.
Cosurfactant-Enhanced Alkaline Flooding. Paper SPE 12672 presented
at the SPE Enhanced Oil Recovery Symposium, Tulsa, 15–18 April.
doi: 10.2118/12672-MS.
Novosad, Z. and Novosad, J. 1984. Determination of Alkalinity Losses
Resulting From Hydrogen Ion Exchange in Alkaline Flooding. SPE J.
24 (1): 49–52. SPE-10605-PA. doi: 10.2118/10605-PA.
Pennell, K.D., Pope, G.A., and Abriola, L.M. 1996. Influence of Viscous
and Buoyancy Forces on the Mobilization of Residual Tetrachloro-
ethylene during Surfactant Flushing. Environ. Sci. Technol. 30 (4):
1328–1335. doi: 10.1021/es9505311.
Peru, D.A. and Lorenz, P.B. 1990. Surfactant-Enhanced Low-pH Alkaline
Flooding. SPE Res Eng 5 (3): 327–332. doi: 10.2118/17117-PA.
Pope, G.A. 1980. The Application of Fractional Flow Theory to
Enhanced Oil Recovery. SPE J. 20 (3): 191–205. SPE-7660-PA. doi:
10.2118/7660-PA.
Pope, G.A., Lake, L.W., and Helfferich, F.G. 1978. Cation Exchange in
Chemical Flooding: Part 1—Basic Theory Without Dispersion. SPE J.
18 (6): 418–434. SPE-6771-PA. doi: 10.2118/6771-PA.
Pope, G.A., Wang, B., and Tsaur, K. 1979. A Sensitivity Study of Micel-
lar/Polymer Flooding. SPE J. 19 (6): 357–368. SPE-7079-PA. doi:
10.2118/7079-PA.
Puerto, M., Hirasaki, G.J., and Miller, C.A. 2010. Surfactant Systems for
EOR in High-Temperature, High-Salinity Environments. Paper SPE
129675 presented at SPE Improved Oil Recovery Symposium, Tulsa,
24–28 April. doi: 10.2118/129675-MS.
Qutubuddin, S., Miller, C.A., Benton, W.J., and Fort, T. Jr. 1985. Effects
of polymers, electrolytes, and pH on microemulsion phase behavior. In
Macro- and Microemulsions: Theory and Applications, ed. D.O. Shah,
No. 272, 223–251. Washington, DC: Symposium Series, ACS.
Ramakrishnan, T.S. and Wasan, D.T. 1988. The Role of Adsorption in
High pH Flooding. Paper SPE 17408 presented at the SPE California
Regional Meeting, Long Beach, California, USA, 23–25 March. doi:
10.2118/17408-MS.
Ramakrishnan, T.S. and Wasan, D.T. 1989. Fractional-Flow Model for
High-pH Flooding. SPE Res Eng 4 (1): 59–68. SPE-14950-PA. doi:
10.2118/14950-PA.
Reed, R.L. and Healy, R.N. 1977. Some physico-chemical aspects of
microemulsion flooding: a review. In Improved Oil Recovery by
Surfactant and Polymer Flooding, ed. D.O. Shah and R.S. Schechter,
383–437. New York: Academic Press.
Sahni, V., Dean, R.M., Britton, C., Weerasooriya, U., and Pope, G.A. 2010.
The Role of Co-Solvents and Co-Surfactants in Making Chemical
Floods Robust. Paper SPE 130007 presented at the SPE Improved Oil
Recovery Symposium, Tulsa, 24–28 April. doi: 10.2118/130007-MS.
Salager, J.L., Bourrel, M., Schechter, R.S., and Wade, W.H. 1979. Mix-
ing Rules for Optimum Phase-Behavior Formulations of Surfactant/
Oil/Water Systems. SPE J. 19 (5): 271–278. SPE-7584-PA. doi:
10.2118/7584-PA.
Salter, S.J. 1977. The Influence of Type and Amount of Alcohol on
Surfactant-Oil-Brine Phase Behavior and Properties. Paper SPE
6843 presented at SPE Annual Meeting, Denver, 9–12 October. doi:
10.2118/6843-MS.
Schechter, D.S., Zhou, D., and Orr, F.M. Jr. 1994. Low IFT drainage
and imbibition. J. Pet. Sci. Eng. 11 (4): 283–300. doi: 10.1016/0920-
4105(94)90047-7.
Scriven, L.E. 1976. Equilibrium bicontinuous structure. Nature 263 (09
September 1976): 123–125. doi: 10.1038/263123a0.
906
Seethepalli, A. , Adibhatla, B., and Mohanty, K.K. 2004. Physicochemical
Interactions During Surfactant Flooding of Fractured Carbonate Res-
ervoirs. SPE J. 9 (4): 411–418. SPE-89423-PA. doi: 10.2118/89423-
PA.
Somasundaran, P. and Zhang, L. 2006. Adsorption of surfactant on minerals
for wettability control in improved oil recovery processes. J Pet Sci Eng
52 (1–4): 198–212. doi: 10.1016/j.petrol.2006.03.022.
Southwick, J.G. 1985. Solubility of Silica in Alkaline Solutions: Implica-
tions for Alkaline Flooding, SPE J. 25 (6): 857–864. SPE-12771-PA.
doi: 10.2118/12771-PA.
Srivastava, M. and Nguyen, Q.P. 2010. Application of Gas for Mobility
Control in Chemical EOR in Problematic Carbonate Reservoirs. Paper
SPE 129840 presented at the SPE Improved Oil Recovery Symposium,
Tulsa, 24–28 April. doi: 10.2118/129840-MS.
Standnes, D.C. and Austad, T. 2000. Wettability alteration in chalk 2.
Mechanism for wettability alteration from oil-wet to water-wet using
surfactants. J. Pet. Sci. Eng. 28 (3): 123–143. doi: 10.1016/S0920-
4105(00)00084-X.
Standnes, D.C., Nogaret, L.A.D., Chen, H.-C., and Austad, T. 2002. An
Evaluation of Spontaneous Imbibition of Water into Oil-Wet Carbonate
Reservoir Cores Using a Nonionic and a Cationic Surfactant. Energy
Fuels 16 (6): 1557–1564. doi: 10.1021/ef0201127.
Stegemeier, G.L. 1977. Mechanisms of entrapment and mobilization of
oil in porous media. In Improved Oil Recovery by Surfactant and
Polymer Flooding, ed. D.O. Shah and R.S. Schechter, 55–91. New
York: Academic Press.
Stoll, W.M., al Shureqi, H., Finol, J., Al-Harthy, A.A., Oyemade, S., de
Kruijf, A., van Wunnik, J., Arkesteijn, F., Bouwmeester, R., and Faber,
M.J. 2010. Alkaline-Surfactant-Polymer Flood: From the Laboratory
to the Field. Paper SPE 129164 presented at the SPE EOR Confer-
ence at Oil & Gas West Asia, Muscat, Oman, 11–13 April. doi:
10.2118/129164-MS.
Surkalo, H. 1990. Enhanced Alkaline Flooding. J Pet Technol 42 (1): 6–7.
SPE-19896-PA. doi: 10.2118/19896-PA.
Sydansk, R.D. 1982. Elevated-Temperature Caustic/Sandstone Interaction:
Implications for Improving Oil Recovery. SPE J. 22 (4): 453–462. SPE-
9810-PA. doi: 10.2118/9810-PA.
Tabatabal, A., Gonzalez, M.V., Harwell, J.H., and Scamehorn, J.F. 1993.
Reducing Surfactant Adsorption in Carbonate Reservoirs. SPE Res Eng
8 (2): 117–122. SPE-24105-PA. doi: 10.2118/24105-PA.
Taber, J.J. 1969. Dynamic and Static Forces Required To Remove a Discon-
tinuous Oil Phase from Porous Media Containing Both Oil and Water.
SPE J. 9 (1): 3–12. SPE-2098-PA. doi: 10.2118/2098-PA.
Takamura, K. and Chow, R.S. 1985. The electric properties of the bitu-
men/water interface Part II. Application of the ionizable surface-
group model. Colloids and Surfaces 15: 35–48. doi: 10.1016/0166-
6622(85)80053-6.
Talley, L.D. 1988. Hydrolytic Stability of Alkylethoxy Sulfates. SPE Res
Eng 3 (1): 235–242. SPE-14912-PA. doi: 10.2118/14912-PA.
Tham, M.J., Nelson, R.C., and Hirasaki, G.J. 1983. Study of the Oil Wedge
Phenomenon Through the Use of a Chemical Flood Simulator. SPE J.
23 (5): 746–758. SPE-10729-PA. doi: 10.2118/10729-PA.
Treiber, L.E. and Owens, W.W. 1972. A Laboratory Evaluation of the
Wettability of Fifty Oil-Producing Reservoirs. SPE J. 12 (6): 531–540.
SPE-3526-PA. doi: 10.2118/3526-PA.
Trushenski, S.P. 1977. Micellar flooding: sulfonate-polymer interaction. In
Improved Oil Recovery by Surfactant and Polymer Flooding, ed. D.O.
Shah and R.S. Schechter, 555–575. New York: Academic Press.
Uren, L.C. and Fahmy, E.H. 1927. Factors Influencing the Recovery of
Petroleum from Unconsolidated Sands by Waterflooding. SPE-927318-
G. Petroleum Transactions, AIME, 77: 318–335.
Wagner, O.R. and Leach, R.O. 1959. Improving Oil Displacement Effi-
ciency by Wettability Adjustment. SPE-1101-G. Trans., AIME, 216:
65–72.
Wang, D., Cheng, J., Yang, Z., Li, Q., Wu, W., and Yu, H. 2001.
Successful Field Test of the First Ultra-Low Interfacial Tension
Foam Flood. Paper SPE 72147 presented at the SPE Asia Pacific
Improved Oil Recovery Conference, Kuala Lumpur, 6–9 October.
doi: 10.2118/72147-MS.
Wang, F.H.L. 1993. Effects of Reservoir Anaerobic, Reducing Conditions
on Surfactant Retention in Chemical Flooding. SPE Res Eng 8 (2):
108–116. SPE-22648-PA. doi: 10.2118/22648-PA.
December 2011 SPE Journal
Wellington, S.L. and Richardson, E.A. 1997. Low Surfactant Concentration
Enhanced Waterflooding. SPE J. 2 (4): 389–405. SPE-30748-PA. doi:
10.2118/30748-PA.
Wessen, L.L. and Harwell, J.H. 2000. Surfactant Adsorption in Porous
Media. In Surfactants: Fundamentals and Applications in the Petroleum
Industry, ed. L.L. Schramm. New York: Cambridge University Press.
Yan, W., Miller, C.A., and Hirasaki, G.J. 2006. Foam sweep in frac-
tures for enhanced oil recovery. Colloids and Surfaces A: Physi-
cochem. Eng. Aspects 282–283: 348–359. doi: 10.1016/j.colsurfa.
2006.02.067.
Yang, H.D. and Wadleigh, E.E. 2000. Dilute Surfactant IOR—Design
Improvement for Massive, Fractured Carbonate Applications. Paper
SPE 59009 presented at the SPE International Petroleum Conference
and Exhibition in Mexico, Villahermosa, Mexico, 1–3 February. doi:
10.2118/59009-MS.
Zhang, D.L., Liu, S., Puerto, M., Miller, C.A., and Hirasaki, G.J. 2006.
Wettability alteration and spontaneous imbibition in oil-wet carbonate
formations. Journal of Petroleum Science and Engineering 52 (1–4):
213–226. doi: 10.1016/j.petrol.2006.03.009.
Zhang, J., Nguyen, Q.P., Flaaten, A.K., and Pope, G.A. 2008. Mechanisms
of Enhanced Natural Imbibition with Novel Chemicals. Paper SPE
113453 to be presented at the SPE/DOE Symposium on Improved Oil
Recovery, Tulsa, 20–23 April. doi: 10.2118/113453-MS.
Zhang, R. and Somasundaran, P. 2006. Advances in adsorption of sur-
factants and their mixtures at solid/solution interfaces. Adv. Colloid
Interface Sci. 123–126 (Special Issue, 16 November 2006): 213–229.
doi: 10.1016/j.cis.2006.07.004.
Zhang, Y., Yue, X., Dong, J., and Liu, Y. 2000. New and Effective Foam
Flooding To Recover Oil in Heterogeneous Reservoir. Paper SPE 59367
presented at the SPE/DOE Improved Oil Recovery Symposium, Tulsa,
3–5 April. doi: 10.2118/59367-MS.
Zhao, P., Jackson, A.C., Britton, C., Kim, D.H., Britton, L.N., Levitt, D.B.,
and Pope, G.A. 2008. Development of High Performance Surfactants
for Difficult Oils. Paper SPE 113432 presented at the SPE/DOE
Symposium on Improved Oil Recovery, Tulsa, 20–23 April. doi:
10.2118/113432-MS.
December 2011 SPE Journal
George J. Hirasaki had a 26-year career with Shell Development
and Shell Oil Companies before joining the chemical engineer-
ing faculty at Rice University in 1993. At Shell, his research areas
were reservoir simulation, enhanced oil recovery, and forma-
tion evaluation. At Rice, his research interests are in NMR well
logging, reservoir wettability, surfactant enhanced oil recovery,
foam-mobility control, gas-hydrate recovery, asphaltene depo-
sition, and emulsion separation. He holds a BS degree from
Lamar University and a PhD degree from Rice University, both in
chemical engineering. He received the SPE Lester Uren Award in
1989 and was named an Improved Oil Recovery Pioneer at the
1998 SPE/DOE IOR Symposium. Clarence A. Miller is Louis Calder
Professor Emeritus of Chemical and Biomolecular Engineering
at Rice University and a former chairman of the department.
Before coming to Rice, he taught at Carnegie-Mellon University.
He has been a Visiting Scholar at Cambridge University, the
University of Bayreuth (Germany), and the Delft University of
Technology (Netherlands). His research interests center on emul-
sions, microemulsions, and foams and their applications in deter-
gency, enhanced oil recovery, and aquifer remediation. He is
coauthor of the book Interfacial Phenomena, now in its second
edition. He holds BS and PhD degrees in chemical engineering
from Rice and the University of Minnesota. Maura C. Puerto is
a research scientist at Rice. She retired from ExxonMobil as a
Research Associate after more than 25 years of service in the
Reservoir Division. She is known as an expert in application of
surfactants to processes involving flow through porous media,
and is skilled in the area of designing laboratory coreflood test-
ing and history matching the results by simulation. Before joining
Exxon in 1974, she was employed in quality control and prod-
uct development for Almay Hypoallergenic Cosmetics and was
technical director of the cosmetic branch at the Cuban Institute
of Soap and Cosmetics. Since her retirement from Exxon, she has
participated on a part-time basis in research on enhanced oil
recovery and aquifer remediation at Rice. She was the recipi-
ent in 1980 of the SPE Cedric K. Ferguson Award, and in 1996
she received, based on her work on CO2 solubility in LVP solvents
for aerosols, the Quality Award from Exxon Chemicals for turning
environmental regulations into opportunities. She holds a chemi-
cal engineering degree from Oriente University in Cuba.
907
 Carbon Dioxide Storage Capacity
of Organic-Rich Shales
S.M. Kang, E. Fathi, R.J. Ambrose, I.Y. Akkutlu, and R.F. Sigal, The University of Oklahoma
Summary
This paper presents an experimental study on the ability of
organic-rich-shale core samples to store carbon dioxide (CO2). An
apparatus has been built for precise measurements of gas pressure
and volumes at constant temperature. A new analytical methodol-
ogy is developed allowing interpretation of the pressure/volume
data in terms of measurements of total porosity and Langmuir
parameters of core plugs. The method considers pore-volume
compressibility and sorption effects and allows small gas-leakage
adjustments at high pressures. Total gas-storage capacity for pure
CO2 is measured at supercritical conditions as a function of pore
pressure under constant reservoir-confining pressure. It is shown
that, although widely known as an impermeable sedimentary rock
with low porosity, organic shale has the ability to store significant
amount of gas permanently because of trapping of the gas in an
adsorbed state within its finely dispersed organic matter (i.e., kero-
gen). The latter is a nanoporous material with mainly micropores
(< 2 nm) and mesopores (2–50 nm). Storage in organic-rich shale
has added advantages because the organic matter acts as a molecu-
lar sieve, allowing CO2—with linear molecular geometry—to
reside in small pores that the other naturally occurring gases cannot
access. In addition, the molecular-interaction energy between the
organics and CO2 molecules is different, which leads to enhanced
adsorption of CO2. Hence, affinity of shale to CO2 is partly because
of steric and thermodynamic effects similar to those of coals that
are being considered for enhanced coalbed-methane recovery.
Mass-transport paths and the mechanisms of gas uptake
are unlike those of coals, however. Once at the fracture/matrix
interface, the injected gas faces a geomechanically strong porous
medium with a dual (organic/inorganic) pore system and, there-
fore, has choices of path for its flow and transport into the matrix:
the gas molecules (1) dissolve into the organic material and diffuse
through a nanopore network and (2) enter the inorganic material
and flow through a network of irregularly shaped voids. Although
gas could reach the organic pores deep in the shale formation fol-
lowing both paths, the application of the continua approximation
requires that the gas-flow system be near or beyond the percola-
tion threshold for a consistent theoretical framework. Here, using
gas permeation experiments and history matching pressure-pulse
decay, we show that a large portion of the injected gas reaches the
organic pores through the inorganic matrix. This is consistent with
scanning-electron-microscope (SEM) images that do not show
connectivity of the organic material on scales larger than tens of
microns. It indicates an in-series coupling of the dual continua in
shale. The inorganic matrix permeability, therefore, is predicted to
be less, typically on the order of 10 nd. More importantly, although
transport in the inorganic matrix is viscous (Darcy) flow, transport
in the organic pores is not due to flow but mainly to molecular
transport mechanisms: pore and surface diffusion.
Introduction
One of the primary considerations in subsurface sequestration of
anthropogenic CO2 is the knowledge of the gas storability of the
Copyright © 2011 Society of Petroleum Engineers
This paper (SPE 134583) was accepted for presentation at the SPE Annual Technical
Conference and Exhibition, Florence, Italy, 20–22 September 2010, and revised for
publication. Original manuscript received for review 28 June 2010. Revised manuscript
received for review 28 September 2010. Paper peer approved 5 October 2010.
842
geological formation. At a glance, a formation with a high pore vol-
ume would appear to be a good candidate for the purpose. However,
this initial impression needs to be tempered because not all high-
porosity formations are suitable for permanent storage of the gas.
Some of them lack a suitable storage environment that will foster
physical mechanisms of gas trapping. In the absence of a trapping
mechanism, a free-gas cap is artificially created in the formation,
which may not warrant long-term storage of the injected gas. The
trapping is associated with fluid/fluid or fluid/solid interactions in
porous media, such as dissolution, physical adsorption, or some
homogeneous and heterogeneous reactions. Depleted oil reservoirs,
for example, could be considered among the acceptable target loca-
tions. The trapping mechanism is mainly driven by absorption of the
injected gas by the immobile residual oil. This is often accomplished
through multiple-contact (dynamic) miscibility, which involves
simultaneous phase-change and mass-transport phenomena. These
reservoirs can also trap CO2 that is not dissolved in the reservoir
fluids to the extent that the reservoir seals for hydrocarbons also
form a capillary seal for CO2. Saline aquifers, on the other hand,
allow aqueous-phase precipitation reactions as well as absorption by
the formation water. However, the dissolved gas promotes density-
driven natural convection of water and the related hydrodynamic
instabilities; consequently, the injected gas could be transported
and dispersed over large distances, leading to uncertainties in its
fate. Coalbeds and naturally occurring gas-hydrate reservoirs have
been proposed as economically feasible choices of location because
the injected CO2 not only could be sequestered in sorbed states
(adsorbed on microporous coal material surfaces and absorbed into
organic macromolecular openings in coal and in water) but also
could participate in enhanced recovery of natural gas by an in-situ
molecular swapping mechanism that promotes release and displace-
ment of methane as a free gas. Introduction of greenhouse gases into
these formations, however, is a difficult field operation because of
a significant loss in well injectivity or no injectivity.
The present work is a fundamental-level investigation on
gas shales as another location for subsurface CO2 sequestration.
Thermally mature organic-rich shales have proved to be prolific
reservoirs for natural-gas production. In the United States, there is
large production from numerous wells and an extensive pipeline
structure associated with organic-shale-gas reservoirs located near
major population centers, such as the Barnett, adjacent to Dallas/
Fort Worth, and the Marcellus, in New York state. These shales
store methane over geologic time spans both in sorbed states and
as free gas. Also, much like the other shales playing the roles of a
barrier or seal in a petroleum-reservoir system, they have ultralow
permeability. Therefore, as the gas wells are depleted, these shales
will become a natural candidate to sequester CO2.
Shale sediments with potential for natural-gas production are
generally rich in organic matter, also known as kerogen. Table 1
shows that the total organic content (TOC) in shale could be up
to 10% of the total shale weight for the North American shale-gas
plays. Although this may not be perceived as a large value, the
significance of the organic matter in the storage of gases in shale
becomes more obvious when one considers its contribution to the
total shale pore volume. In the unconventional-gas-research com-
munity, it is increasingly accepted that the organic matter is the
main constituent of total pore volume, which is associated with
in-situ generation and storage of natural gas. In this study, we
experimentally predict an organic-pore-volume/total-pore-volume
ratio up to 70% for Barnett-shale samples. The organic matter in
shale could be a suitable place for CO2 sequestration.
December 2011 SPE Journal

TABLE 1—TYPICAL TOC
OF NORTH AMERICAN SHALE-GAS PLAYS*
Shale or Play Average TOC (wt%)
Barnett 4
M arcellus 1–10
Haynesville 0-8
Horn River 3 Gas Shale
W o o d f o rd 5 Fig. 1a shows a typical 2D focused-ion-beam/SEM (FIB/SEM)
* Ambrose et al. 2010
The paper is organized as follows. At micron scale, the
organic-rich gas shales have been visually investigated recently
by several groups (Loucks et al. 2009; Wang and Reed 2009;
Sondergeld et al. 2010). Salient results relevant to sequestration
will be summarized from these petrophysical investigations. Fol-
lowing that, we introduce an approach that uses typical equilibrium
adsorption data, assumes Langmuir monolayer adsorption, and
yields effective organic pore size as an indication of organic pore
space available for the storage. It is shown that the shale organics
are nanoporous materials. Although they may not be suitable for
storing large volumes of CO2 as free gas, they are characterized
by a large internal surface area suitable for trapping significant
amounts of gas at an adsorbed state. Having small organic
pores introduces new complications into the storage-capacity
and transport measurements, however. The organic pore volume
is a dynamic stress-dependent quantity. Next, in the light of this
observation, we introduce a new methodology for the measure-
ment of total shale porosity and gas-storage capacity where the
pore volume is allowed to change as a function of pore pressure
because of both adsorption/absorption and because of pore-volume
compressibility. The method is based on a five-parameter nonlin-
ear model of storability under equilibrium conditions, where the
parameters will require a minimum five-stage laboratory-measure-
ment process using the hydrocarbon gas in the reservoir or CO2.
The method will be applied to the estimation of storage with pure
methane and pure CO2. Finally, shale-gas-permeation experiments
and pressure-pulse-decay measurements are performed, investi-
gating mass-transport mechanisms in the presence of adsorption
using various gases. Although pulse-decay analysis is common in
the industry for the laboratory measurement of low-permeability
samples, our approach is new and unconventional. It involves a
simulation-based nonlinear history-matching algorithm where the
new pore-scale considerations of gas shales have been incorporated
(a)
Fig. 1—(a) 2D backscatter FIB/SEM image of a gas-shale sample. (b) 3D FIB/SEM shale segmentation. The shale segmentation
size is 4 µm high, 5 µm wide, and 4 µm deep.
December 2011 SPE Journal
into the pulse-decay analysis through forward simulation of gas
flow and transport in the core plug. We predict shale-gas quanti-
ties of practical interest, such as organic pore volume, porosity,
permeability, and molecular-diffusion coefficients, as a function
of pore pressure. These measurements will allow us to investigate
the nature of gas transport in organic-rich shale.
Pore-Scale Considerations in Organic-Rich
image of gas shale at micron scale. The image shows organic
matter in dark gray as a finely dispersed porous material imbed-
ded within light-gray inorganic clays. Pores are shown in black,
and note that most of them are within organic islands, or kerogen
pockets, that have characteristic size between 200 and 500 nm.
The average size of the organic pores is typically much smaller
than those we observe elsewhere in the inorganic matrix. In the
next section, we present an analytical approach showing that the
organic material mainly consists of micropores (pore lengths less
than 2.0 nm) and mesopores (pore lengths between 2 and 50 nm),
with an average pore size below 4–5 nm. Unfortunately, this is a
length-scale that is too close to the molecular realm and beyond the
maximum resolution of the current SEM technology. This suggests
that, if one were allowed to look carefully at the fine details of
the pore structure in the organics in Fig. 1, uniformly distributed
micropores would have appeared, surrounding the currently visible
ones, giving a sponge-like appearance to the kerogen pockets.
Ambrose et al. (2010) recently reported the results of an
analysis involving hundreds of 2D images of organic shale samples,
sequentially obtained using ion milling, and their recombined 3D
shale segmentations. The digital segment analysis reveals kerogen
pore networks. A typical network consisting of large interconnected
pockets of kerogen is shown in Fig. 1b. Although the samples repre-
sent an extremely small portion of the reservoir and additional stud-
ies are currently necessary for statistical inference, the following
fundamental conclusions will be made for the consideration of CO2
storage and transport: The organic-rich-shale matrices consist of
organic and inorganic materials that could be dispersed within each
other or bicontinuous and, at times, intertwined; a major fraction of
the total porosity within the system is associated with the kerogen
network, and therefore much of the gas-storage capacity within
shale is predominantly associated with the organic fraction of the
rock matrix. In the following pages, we will take advantage of these
observations and develop a new conceptual shale-matrix model that
involves a porous medium with dual (organic/inorganic) -porosity
continua. The approach will allow us to quantify and measure total
gas storage and transport in organic-rich shale samples.
(b)
843
Effective Organic Pore Size. An effective-pore-size-estimation
approach considers that gas adsorption takes place only on the
organic pore walls in the kerogen pockets. The approach is based
on the premise that an estimate of total surface area of the organic
material in shale can be realized if a method is developed whereby
the amount of adsorbed gas corresponding to monolayer coverage
can be determined. Physically, the monolayer approach may not
mean much when pores are so small, although a simple surface-
science concept could help us illustrate the potential features of
organic shale. For simplicity, we consider that all of the pores in
the organics will have the same shape: either spherical or cylindri-
cal because both shapes will be investigated. We used methane for
the analysis, and the shale has the physical characteristics shown
in Table 2. First, the total volume available for gas sorption must
be calculated. It is assumed that the pores containing water will
not have any surfaces where gas adsorption can occur and that
the volume of gas that can dissolve in the clay-bound water is
negligible. Hence, total shale gas pore volume per bulk volume
is equal to  (1−Sw). Consider that only k0 fraction of this total
gas pore volume is associated with the organics; hence, the gas
pore volume in the organics is k0  (1−Sw). The expression can be
written in terms of volumes (cm3) as follows: [mass/density]rock k0
 (1−Sw). Thus, using the values in Table 2, the gas-filled organic
pore space for 1.0 ton of shale rock is equal to
⎛ 1 ton ⎞ ⎛ 907,184.75 g ⎞
⎜ 2.5g cm 3 ⎟ × ⎜⎝ ⎟⎠
⎝ ⎠ ton
×0.55 × 0.05 × (1 − 0.35) = 5, 896.7 cm 3.
The number of lbm mol in 1.0 scf of gas is
n P
= std
V RTstd
14.69 psi
=
⎛ scf psi ⎞
⎜⎝ 10.7316 ⎟ × ( 60 + 459.67 ) °R
°R lbm mol ⎠
lbm mol
= 2.6340 × 10 −3 . typically observe phase-transition layers where the molecules are
scf
Therefore, the number of methane molecules in 50 scf/ton of gas
(Langmuir volume) would be
(50.0 scf ) × ⎛⎜⎝ 2.6340 × 10 −3
lbm mol ⎞ ⎛ g ⎞
⎟⎠ × ⎜⎝ 453.5924 ⎟
scf lbm ⎠
⎛ molecules ⎞
× ⎜ 6.0221 × 10 23 = 3.5976 × 10 25 molecules.
⎝ g mol ⎟⎠
If one adsorbed methane molecule covers 0.16 nm2 (Anderson and
Pratt 1985), assuming monolayer coverage, we can estimate the
total area covered by this number of methane molecules as
0.16 nm 2
3.5976 × 10 25 molecules × = 5.7562 × 10 24 nm 2.
molecule
This is a significant value, indicating that the organic pores have
relatively large internal surface area. We can use the ratio of vol-
ume to surface area to calculate the radius the molecules needed
to cover for a uniform size sphere and cylinder. In the case of a
cylindrical pore, we will assume that it is a cylinder without any
flat ends so only the curved surface area will be calculated. For
a sphere, the volume is V = 4r3/3 and surface area is As = 4r2.
Hence, for the spherical pore, the volume/surface-area ratio would
be V/As = (4r3/3)/(4r2) = r/3. Using similar arguments, it could
be shown that the ratio for a cylindrical pore is equal to r/2. Next,
we calculate the effective pore sizes corresponding to spherical and
cylindrical geometries using the gas-filled organic pore volume of
5.8967 × 1024 nm3 (or 5,896.7 cm3):
⎛V⎞ ⎛ 5.8967 × 10 24 nm 3 ⎞
rsphere = 3 ⎜ ⎟ = 3 ⎜ ≅ 3.0 nm .
⎝ As ⎠ ⎝ 5.7562 × 10 24 nm 2 ⎟⎠
⎛V⎞ ⎛ 5.8967 × 10 24 nm 3 ⎞
rcylinder = 2 ⎜ ⎟ = 2 ⎜ ≅ 2.0 nm .
⎝ As ⎠ ⎝ 5.7562 × 10 24 nm 2 ⎟⎠
Table 3 shows the estimated values of spherical and cylindri-
cal pore radii in the organics as a function of k0. Assuming slit-
type pore geometry, Schettler et al. (1989) estimated an average
pore diameter of 5.5 nm for Devonian shales using the Brunauer-
Emmett-Teller multilayer-adsorption isotherm method. This is a
same-order-of-magnitude value. On the basis of the calculations,
it can be argued that a significant portion of the organic pores are
small (less than 4–6 nm) and, therefore, may not be accessible
with the resolution of the SEM image shown in Fig. 1. These small
organic pores, with their large surface area, are an ideal place for
trapping gas in an adsorbed state and for its long-term storage.
However, they make the measurements and analyses of gas storage
and transport somewhat complicated because, at that length scale
• The thickness of the adsorption layer becomes comparable
to the organic pore size. Under thermodynamic equilibrium (no
flux of mass or energy), some of the gas molecules inside the
organic pore are closer and, therefore, under stronger influence
of the pore walls. These molecules have limited mobility and
relatively low kinetic energy. They participate in the formation of
a dense (liquid-like) adsorption layer that covers internal surfaces
of the pore walls; see Fig. 2a. Next to the adsorption layer, we
constantly being adsorbed and desorbed under equilibrium. The
molecules in phase transition are relatively less dense and more
mobile with some kinetic energy, although they are under some-
what reduced influence of the walls. The rest of the gas molecules
are at the central portion of the pore (if any space is left), and they
are not under the influence of pore walls. They are the free-gas
molecules, with a density equal to the fluid bulk density in the
absence of pore walls. They mainly interact among themselves and
with the gas molecules in phase transition. In the case of methane
adsorption in a 3.6-nm-thick organic slit pore shown in Fig. 2, in
total 2 × 0.37 = 0.74 nm of space next to the walls will be occupied
by the adsorption layer, thus leaving only 3.6−0.74 = 2.86-nm-
thick pore space available for the phase-transition molecules and
TABLE 3—ESTIMATED ORGANIC PORE RADII
FOR VARYING k0 VALUES
Fraction of Porosity Spherical Pore Cylindrical Pore
in Organic ( k0) Radius, rsphere (nm) Radius, rcylinder (nm)

0.1 0.61 0.41

0.2 1.23 0.82
TABLE 2—SHALE PARAMETERS
FOR EFFECTIVE-PORE-SIZE CALCULATIONS 0.3 1.84 1.23
0.4 2.46 1.64
Parameter Symbol Value
0.5 3.07 2.05
Total porosity φ 5% 0.6 3.69 2.46
Organic/total pore volume ratio k 0.5 0.7 4.30 2.87
Water saturation Sw 35% 0.8 4.92 3.28
3
Bulk density 2.5 g/cm 0.9 5.53 3.69
Langmuir volume (methane) VL 50 scf/ton 1 6.15 4.10

844 December 2011 SPE Journal


0.35 Adsorption layer
0.30
Phase
0.25
transition layers
ρCH4, g/cm3
0.20
Free gas
0.15
0.10
0.05
0.00
0.0 3.8 7.6 11.4 15.2
Pore Half Length, Å
(a)
(c)
Fig. 2—(a) Molecular layer density for methane at 176°F (80°C) across the half-length of a 3.74-nm organic slit pore. Results
obtained using equilibrium molecular dynamics simulation carried out in the canonical ensemble. The estimated pore pressure
is approximately 2,000 psi. (b) Schematic representation of gas slippage and hopping mechanisms of molecular transport. (c)
Schematic of gas mass transport in organic nanopores. The arrows represent diffusive mass fluxes of the free-gas phase (thick
arrow) and the adsorbed phase (thin arrow) in the x direction. JK,x and JS,x are the contributions of free-gas pore diffusion and
adsorbed-phase surface diffusion, respectively.
for the free-gas molecules. Density structure of gas molecules in
small organic pores and its effect on shale-gas in-place estima-
tions have been studied recently by our group; see Ambrose et al.
(2010).
• The mean free path of gas molecules (i.e., average distance
gas molecules travel between two successive collisions) becomes
comparable to the available pore space. Consequently, the free-gas
molecules interact vigorously with the molecules in phase transi-
tion and with those that make up the adsorption layer and pore wall,
compared with the level of interaction the free-gas molecules have
among themselves at the central portion of the pore. When thermo-
dynamic equilibrium is perturbed because of an externally applied
molecular flux, for example, these amplified interactions may lead
to kinetic effect that causes streaming of the gas molecules—in
particular, those in phase transition. This is the occurrence of the
so-called slippage effect (or molecular streaming) in the organic
pore. It can cause the collapse of laminar-flow (parabolic) -veloc-
ity profile with zero velocity at the wall. Consequently, the free-
gas mass transport would show deviations from Darcy’s law and
would be represented better with molecular (pore) diffusion. A
useful dimensionless parameter in the literature to determine the
changes in flow regime of a gas system is the Knudsen number
(NKn). This number is the ratio of the molecular mean free path to
the pore length scale. When NKn > 0.1, the kinetic effect becomes
dominant by the wall in the phase-transition region. Further dis-
cussion on the flow regimes can be found in Roy et al. (2003).
In Fig. 3, we consider methane existence in organic pore volume
with characteristic pore length varying between 2 and 10 nm and
estimate the NKn for varying pore pressures. The values inside the
ellipse (typical initial shale-gas-reservoir conditions) indicate that
December 2011 SPE Journal
Nonequilibrium
Slippage
Hopping
Direction of external molecular flux
(b)
J K,x
J S,x
the gas mass transport in the organics should be in the transitional
slippage-flow regime. This means that the assumption of Darcy
flow is invalid in the organic pores in shale. Therefore, a new
approach is necessary for modeling the gas mass transport, which
will incorporate non-Darcy transport of the free-gas molecules.
• Adsorbed-phase transport could be another important mass-
transfer mechanism. Fig. 2c shows the free-gas and adsorbed-phase
components of total mass flux in the organic micro- and mesopore.
Under the influence of an external flux, some of the molecules in
the adsorbed-phase and those in phase transition can overcome the
local interactions with the wall and develop a hopping mechanism.
(Much like drifting of a tumbleweed or tree leaves on the ground
under the influence of a mild wind.) This transport mechanism, also
known as surface diffusion, is not important if it takes place in a
large pore with an almost infinite number of free-gas molecules;
but, if the discussion is on nanopores with a large surface area, it
may easily contribute to the total mass flux and, hence, enhance
the kinetic effect by the pore walls. Let us reconsider the layers of
molecules indicating nonuniform structure (damped oscillation) in
methane density in a 3.6-nm organic pore in Fig. 2. It is not dif-
ficult to imagine that the molecular layers corresponding to free gas
would not develop and all of the existing molecules would be either
adsorbed or under the influence of pore walls if the pore size were
less then 2×1.08 = 2.16 nm. Then, the only transport mechanism
during the gas release from or uptake by the pore would be because
of surface diffusion. We note that, although they both take place in
the phase-transition region, surface diffusion is a different transport
mechanism than the slippage; at a given pressure and temperature,
surface diffusion is driven by gradients in adsorbed-phase density,
whereas slippage is driven by the pore size.
845
 100.00
2-nm pore radius
10-nm pore radius
10.00

N Kn
1.00

0.10

0.01
10 100 1,000 10,000
Pressure, psia

Fig. 3—Knudsen number vs. pore pressure for pore-throat radii calculated from the Langmuir isotherm at reservoir temperature
of 188°F.

• The assumption of continuum-fluid mechanics may −2r 8 RT ⎛ dn ⎞

become questionable—in particular, in the micropores. A 2.86- J K ,x = ⎜ ⎟ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
3  M ⎝ dx ⎠
nm pore can, at most, include only seven methane molecules in a
row across the pore space. Similarly, approximately 30 methane Equivalently, this expression can be reformulated in terms of
molecules would fit in a 10-nm organic pore. Hence, the number moles of concentration C rather than molecular concentration:
of molecules is limited inside the organic pores, which may not
allow the assumption of a representative elementary volume −2r 8 RT ⎛ dC ⎞
J K ,x = ⎜ ⎟ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
for the fluid (i.e., the classical fluid-particle concept of fluid 3  M ⎝ dx ⎠
mechanics) with average intensive properties, such as density
and viscosity. By analogy to Fickian gas diffusion, we now define a molecu-
In this work, these pore-scale considerations will be part of lar-diffusion coefficient for transport in the capillary as
the gas-storage and -transport measurements. Gas-storage mea-
surements do not necessarily require the assumption of continuum, 2r 8 RT
DK ,capillary = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3)
although the transport calculations do. Our group has other investi- 3 M
gations based on noncontinuum approaches (e.g., molecular mod-
eling and molecular/continuum multiscale modeling), although Thus, the diffusion coefficient is proportional to the effective
they are discussed elsewhere; see Diaz-Campos et al. (2009) and (adsorption layer-corrected) pore radius and the mean molecular
Ambrose et al. (2010). During the measurements at varying pore velocity. One can relatively easily obtain typical values to compare
pressures, the organic pore volume available for the free gas will with the experimentally obtained organic pore diffusion coeffi-
be adjusted for the isothermal pore compressibility effect (effective cient. For the purpose, we consider a 1.5-nm cylindrical pore and
stress) and also for the presence of an adsorption layer. During perform the calculations at our laboratory conditions:
transport calculations, we will differentiate free-gas and adsorbed-
phase diffusive transport in the organics from gas flow in inorganic DK ,capillary =
−9
2r 8 RT 2 × 1.5 × 10 m
=
( )
matrix and in fractures using a separate transport mechanism for 3 M 3
each. In the organics, diffusion of the adsorbed phase accounts for
⎛ kg m 2 ⎞
surface diffusion effects of molecules in phase transition, whereas 8 × ⎜ 8.3145 2 × ( 298.15 K )
diffusion of free molecules represents transport in the pore space ⎝ s K mol ⎟⎠
×
and slippage in the phase-transition region. ⎛ kg ⎞
3.1416 × ⎜ 16.042 × 10 −3 ⎟
⎝ mol ⎠
Interpretation of Organic Pore Diffusion. Consider flux of
free-gas molecules across the available organic pore space in the m2 cm 2
= 6.3 × 10 −7 = 6.3 × 10 −3 .
direction parallel to its inlet and outlet ends. The net free-gas mass s s
flux is proportional to the difference in gas number densities n1
One would expect this estimate to be somewhat larger than the
and n2 at the two ends: JK = wc(n2− n1), where w is a dimension-
experimentally obtained molecular diffusivity because the mea-
less probability factor and c is the mean molecular speed. The
surement is the outcome of an expression derived for a single
mean molecular speed could be estimated from the kinetic theory
capillary pore. Typically, the two can be roughly related through
of gases using c = [8 RT ( M )] , where M is the gram molecular the porosity/tortuosity ratio as
weight, T is the absolute temperature, and R is the universal gas
constant (Roy et al. 2003). The probability factor is dependent on ⎛ ε ⎞⎛  ⎞
DK = ⎜ k ⎟ ⎜ 2 ⎟ DK ,capillary, . . . . . . . . . . . . . . . . . . . . . . . . . (4)
the geometry of the pore and requires knowledge of the appropri- ⎝ ε TOC ⎠ ⎝  ⎠
ate scattering law. Considering diffusive scattering and simple
geometry, such as a capillary with a length L and radius r, the where TOC is the kerogen/total-volume ratio and  is tortuosity.
value of w for the condition L >> r is equal to 2/3(r/L). Substitut- Because it cannot be measured directly, there exists uncertainty in
ing these expressions into the flux equation so that it gives mass the values of tortuosity of nanoporous materials. Values between
flux along the axis of a relatively long capillary pore and writing 1.5 and 8 have been reported by Wei et al. (2007) and Busch et al.
in differential form using the convention that the molecules move (2008). If the kerogen porosity (the first parenthesis term on the
from high to low concentration of molecules, we obtain right-hand side of Eq. 4 multiplied with ) is 20%, and a kerogen

846 December 2011 SPE Journal

tortuosity of 3.0–6.0 is assumed, one would expect that the organic
pore-diffusion coefficient measured using pressure-pulse decay DK
should be on the order of 10−5 cm2/s (i.e., two orders of magnitude
smaller). In addition, the adsorbed-phase transport can be captured
similarly using Fick’s law:
⎛ dC ⎞
J S ,x = − DS ⎜  ⎟ , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5)
⎝ dx ⎠
where Cµ is the adsorbed-phase amount in terms of moles per solid
organic volume and DS is the porosity- and tortuosity-corrected
surface diffusivity. Typically, for Cµ, we would expect one-order-
of-magnitude-smaller values then the free-gas amount C, although
its gradient could vary and be relatively large during the production
and sequestration operations.
When evaluating the capacity of a shale reservoir to store
CO2, an essential first step is to evaluate the total porosity and
how much gas per unit weight (scf/ton) the porous material of the
reservoir can store. In general, this storage capacity is a function
of the porous-material properties, the reservoir pore pressure, and
the temperature. When multiple storage mechanisms are present, it
is difficult to characterize the total porosity and storage. First, we
explain a multistage gas-uptake process that allows determination
of the total storage capacity and total porosity as a function of
pressure and temperature and the parsing of the storage into the
various storage mechanisms. This measurement can also be taken
under in-situ subsurface conditions.
Measurement of Shale-Gas-Storage
Capacity
For a shale sample, total porosity is defined as the volume of flu-
ids (i.e., liquid and free gas, not including adsorbed or absorbed
gas) contained in the sample. Hence, the storage capacity of shale
should be divided up into the sorbed component and the pore-vol-
ume component. In this work, an isothermal multistep gas-uptake
process measures the storage capacity. The minimum number of
steps required is discussed later.
Fig. 4 shows the basic components needed for the mea-
surement. For temperature control, or to make measurements at
reservoir conditions, the apparatus can be enclosed in an oven to
provide a constant high-temperature environment. Currently, we
have two different storability-measurement implementations in the
laboratory, one in an environmental chamber and one that operates
at ambient temperature in the laboratory. The apparatus consists
of a pressure vessel that contains the core sample to be tested
along with a dead volume, Volume 2; a gas reservoir, Volume 1;
a valve, Valve 2, to separate the reservoir volume from the sample
container; a gas source, typically a pressure tank; as valves, Valve
1, to isolate the rest of the system from the gas tank; and pressure
gauges. Not shown are pumps to raise the gas from the bottle
pressure to the needed reservoir pore pressure and the pump and
plumbing to provide confining pressure on the sample. The pres-
Valve 1
Reservoir
Volume 1
Valve 2
Gas
Source
Volume 3 = Volume 2
+ Sample Volume
Fig. 4—Diagram of gas-storage-capacity measurement method.
December 2011 SPE Journal
sure vessel is plumbed so that the sample can be kept at a confining
pressure that corresponds to the confining stress that the sample is
exposed to in the reservoir. During the measurements, the confin-
ing pressure is always kept greater than the pore pressure.
The storability measurement is a multistage measurement.
Each stage starts with Valve 2 closed. To start, Volume 1 is raised
to a pressure P1, and then Valve 1 is closed. The sample and dead
Volume 2 are at an initial starting pressure P2. The test consists of
opening Valve 2 and recording the pressure as a function of time
until it has stabilized at a new pressure Pf . The pressure-decay
curve can be used to obtain the transport coefficients. The test can
be run with P1 less than P2 (for the gas-release measurements) or P1
greater than P2 (for the gas-uptake measurements). In the first case,
the decrease in storability with decrease in pressure is measured;
and, in the second case, the increase in storability is measured.
For the second case, the analysis of a given stage proceeds
as follows. The moles of gas n1 in Volume 1, V1, at pressure P1
is given by
1 1 ( z1 RT ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)
n1 = PV
where R is the universal gas constant, T is the absolute temperature,
and z1 is the correction factor that accounts for deviation from the
ideal-gas law. The correction factor depends on temperature, pres-
sure, and the type of gas. Numerous tables and programs exist that
provide accurate values of z1. Similarly, the moles of gas contained
in V2 initially are given by
n2 = P2V2 ( z 2 RT ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
After Valve 2 is opened and the pressure has stabilized at Pf ,
the moles in V1 and V2 are given by
n1 f = Pf V1 ( z RT )
f
and
n2 f = Pf V2 ( z RT ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
f
Mass balance requires that the decrease of moles of gas in V1,
n1, is equal to the increase in moles of gas in V2, n2, plus the
increase in moles of gas stored in the organic pores of the shale
sample ns. From the previous equations,
V1 ⎛ P1 Pf ⎞
n 1 = − ,
RT ⎜⎝ z1 z f ⎟⎠
V2 ⎛ Pf P2 ⎞
n2 = − ,
RT ⎜⎝ z f z2 ⎟⎠
and
ns = n1 − n2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
The additional moles of gas ns stored in the sample provide the
increase in storability as a function of the change in pressure.
When this process is iterated over multiple steps, it provides the
storability as a function of pore pressure. This is the basic informa-
tion needed to characterize the reservoir and is an essential input
into simulations that model the way gas will be produced from the
reservoir or be sequestered in it. It should be noted that, because
there can be hysteresis in the storability curves, both the increasing
and decreasing storability cases may need to be measured.
To separate the storage in the rock system into the adsorbed/
absorbed component and the component stored as free gas in the
pore space, a model of the storage mechanisms needs to be used.
The moles of gas nsp stored in the sample pore space Vp in terms
of the sample pore pressure Ps (Ps equals P2 or Pf) satisfies the
gas law. Therefore,
nsp = PsVp ( zs RT )
847
and
⎛V P V P ⎞
nsp = ⎜ pf f − p 2 2 ⎟ RT . . . . . . . . . . . . . . . . . . . . . . . . (10)
⎝ zf z2 ⎠
In general, Vp is a function of the pore pressure and the gas
type. As we will see, it may not be a single-valued function. The
dependence of Vp on pressure needs to be included in the model
and will be discussed after the discussion of the adsorption/absorp-
tion term. Vp2 and Vpf are the pore volumes at pressures P2 and Pf ,
respectively.
For both the shale-gas and the coalbed-methane cases, the
amount of gas adsorbed/absorbed by the rock matrix is generally
assumed to be described by a Langmuir isotherm. It is known that
this is only an approximation and other more-exact models could
be used to model the adsorption/absorption component of the
storage. For the Langmuir model, all the nonfree gas is assumed
to be adsorbed. The Langmuir isotherm is parameterized by two
quantities, the total moles of adsorbed gas at infinite pore pressure
Samax and the Langmuir pressure PL.
The Langmuir equation gives the adsorbed gas storage Sa as
a function of the pore pressure P as
Sa =
(
Sa max P PL ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
1 + P PL
so that nsa is given by
⎡ Pf PL P2 PL ⎤
nsa = Sa max ⎢ − would not provide any test of the model assumptions. Combining
( P2 PL ) ⎥⎥⎦
. . . . . . . . . . . . . . . . (12)
⎢⎣ 1 + ( Pf PL ) 1 +
and
ns = na + nsp.
The pore volume can change as a function of pressure because
of both adsorption/absorption and because of pore-volume com-
pressibility. Adsorption can affect pore volume by the adsorbed
molecules occupying pore space that otherwise would be occupied
by the free gas or by the adsorbed gas closing off pore throats so
gas cannot get through to the larger pore. Absorption can cause the
matrix to swell or, by swelling, to close off access to pores. The
effects would look very similar for both adsorption and absorption.
Using the Langmuir model associates all the pore-volume changes
with adsorption. The Langmuir model can be used to calculate the
pore-volume loss because of adsorbed gas. This calculation does
not account explicitly for pore-blockage effects, but, to the extent
that a Langmuir isotherm successfully fits the gas adsorption, it is
implicitly including these effects. In the Langmuir equation, the
term (P/PL)/(1+P/PL) represents the fraction of the moles adsorbed
at any pressure. If Vamax represents the maximum adsorbed-gas
volume and Va the adsorbed-gas volume at any pressure, then,
assuming Langmuir adsorption,
Va =
(
Va max P PL ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)
1 + P PL
If Vp0 is the pore volume at zero pore pressure, then the
decrease in Vp at pressure P because of adsorption is given by
Vp0−Vamax (P/PL)/(1+P/PL). The decrease in Vp in each stage is
⎡ Pf PL P2 PL ⎤
Vpa = −Va max ⎢ −
( P2 PL ) ⎥⎥⎦
. . . . . . . . . . . . . . (14)
⎢⎣ 1 + ( Pf PL ) 1 +
If the molar density of the adsorbed gas ads max is known,
then
Vamax = Samax/ads max. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (15)
The last consideration is accounting for pore-volume com-
pressibility. Over the pressure ranges applicable, this should be a
small effect with values typically on the order of 10−6. It can be
determined in two ways, as part of the multistage measurement
848
process using the adsorbed gas or as a separate process using
helium, which is not adsorbed (so only a simple two-stage process
needs to be performed). In either case, the pore-volume compress-
ibility is defined by Cp so that the volume increase because of
increasing pore pressure is given by
Vpp = CpVp0 (Pf –P2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (16)
The total change in pore volume going from pore pressure
P2 to Pf is given by
Vp = Vpp+Vpa, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
so that
Vpf = Vp2+Vp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
Here, the second-order effect of Cp being dependent on pore
pressure is ignored. This model of storability has five parameters
to determine, Samax, Vamax, PL, Vp0, and Cp. Simultaneous estima-
tion of these quantities will require, at a minimum, a five-stage
measurement process if they are all determined by measurements
using the hydrocarbon gas in the reservoir or CO2. Measurements
with helium could be used to determine Cp and an approximate
value for Vp0, assuming the density of the adsorbed-state gas is
constant, and then the measurements could be reduced to two-stage
measurements. The second option is experimentally easier but
the second option with a few detailed multistage measurements
to obtain model-independent storability to test against the model
might prove to be a cost-effective option.
Multiscale Characterization and
Measurement of Shale-Gas-Transport
Mechanisms
Our arguments based on the indirect laboratory observations
(SEM images and storage-capacity measurements) suggest that a
flow model for shale gas should include dispersed organic porous
material within the inorganic matrix. A conceptual model showing
the organic- and inorganic-material distribution in shale is given
in Fig. 5. Note that, compared to coal, the main difference in
shale-gas production is that, before reaching the fractures, a large
portion of shale gas in place should be transported first inside
the kerogen network imbedded in the matrix. It should be noted
that a large contrast in length scales exists between gas flowing
in the organic pores, outside of them in the matrix, and in the
fractures. This requires a multicontinuum modeling approach
involving separate but coupled hydraulic components of the porous
medium, where each component (organic, inorganic, and fracture)
Organic material
Inorganic
Organic porous material Inorganic porous material
with uniform fracture network with nonuniform fracture network
(a) (b)
Fig. 5—Distributions of the organic and inorganic materials in
coal and shale.
December 2011 SPE Journal
 Transport in parallel
Adsorption+diffusion
Kerogen
Fracture
Inorganic
matrix
Viscous
flow
(a)
Fig. 6—Conceptual multicontinuum models for gas flow and transport in organic-rich shale during sequestration.
is distributed continuously in space and holds the fundamental
porous-medium conditions specified by Bear and Bachmat (1991).
Multicontinuum models can be identified by the number of
components characterizing the hydraulic behavior of the system
and the type of coupling and mass-exchange terms between the
components. The interaction between the components, defined by
finite mass-exchange terms, affects the overall transport behavior
of the system significantly, while a single-continuum approach
neglects this important effect. There are three different types of
coupling between the components: parallel, series, and selective
coupling. In parallel coupling, all specified continua are coupled
with one another directly, while, in the series coupling, they are
connected in the order of hydraulic conductivity (Lee and Tan
1987). In organic-rich-shale-gas reservoirs, multicontinuum could
be defined, at the least, by two matrix continua (i.e., organic and
inorganic, along with a fracture continuum).
In the Effective Organic Pore Size subsection, we argued that
the gas mass transport in the organics is transitional slippage with
adsorbed-phase surface-diffusion effects (i.e., non-Darcy). Model-
ing of such transitional flow regimes in the presence of adsorption
is not straightforward, however. Here, following the work by Fathi
and Akkutlu (2009, in press), we investigate gas-transport behavior
in the kerogen by experimentally quantifying the pore and surface
diffusion coefficients DK and DS , respectively. Pore diffusion is the
mechanism of transport for the free-gas molecules occupying the
available pore space, whereas surface diffusion is the mechanism
of transport for the adsorbed-phase near the internal surfaces of the
kerogen. During the production, gas in the kerogen pore network
is eventually released to the relatively larger pores, voids, micro-
fractures, and cracks in the inorganic matrix or directly into the
nonuniform fracture network shown in Fig. 5b. In its new environ-
ment, the mechanism of its transport is primarily viscous (Darcy)
flow and, because of limited surface area, the adsorbed amount is
considered negligible. In the case of CO2 sequestration, the same
arguments are valid, although in the opposite transport direction
(i.e., from the fractures deep into the inorganic shale matrix and
into the kerogen). Fig. 6 shows two conceptual models of shale-gas
flow and transport that are considered in this study for methane
production and CO2 sequestration. In Fig. 6a, parallel coupling is
shown where organic and inorganic continua are coupled directly
with fractures. In Fig. 6b, all three continua are coupled in series
(i.e., fracture → inorganic → organic). We used these conceptual
models during the development of a mathematical formulation
(a group of coupled partial-differential equations) describing the
mechanisms of gas storage and transport. The formulation is used to
simulate gas-permeation experiments using organic-rich shale core
plugs. Later, the developed simulators are used in the experimental
investigation of multiscale characterization of transport phenomena
in organic-rich shales. Further discussion on the modeling and
simulation of shale-gas flow can be found in Fathi (2010).
Measurements of Transport Coefficients—Unsteady Gas-Per-
meation Experiments. Steady-state methods for the measure-
December 2011 SPE Journal
Transport in series
K
e Inorganic
Fracture
r matrix
o
g Adsorption Viscous
e + flow
n diffusion
(b)
ments of permeability and diffusivity of shale are difficult to
implement because of very low mass fluxes and the extremely
long time needed to reach the steady-state condition. Therefore,
transient methods such as pressure-pulse-decay experiments have
been used for the measurements. This method is much faster in
comparison to the conventional methods and can be used to mea-
sure permeability valuse as low as 10−9 md (Brace et al. 1968; Ning
1992; Finsterle and Persoff 1997; Jones 1997). In this method,
a small perturbation is introduced to a shale/gas system under
thermodynamic equilibrium by rapidly changing the downstream
(or upstream) pressure, typically by an approximate 10% of the
equilibrium pressure. Consequently, small pressure gradients are
generated across the sample, leading to transport of gas across the
core plug. During the gas permeation, decline in the upstream or
downstream pressure is measured to extract transport properties of
the shale using an analytical or numerical inversion method.
Results
Table 4 shows results from the models that fit the measured stor-
age data for helium, CO2, and methane with two organic-rich shale
samples, Sample 21 and Sample 23. These samples are selected
because TOC and mineralogy were known. Because, in the initial
measurement stage, the CO2 passes through a pressure where it
can form a liquid phase, there is considerably more uncertainty in
the model fits to the measured storage data. Note the estimated Vp0
values. The values are quite different for Sample 23, giving 5.38,
3.95, and 2.89% porosity for the gases, respectively. The model
fits to the experimental data are not unique. Assuming that all
reasonable fits to the data have the properties of this fit, the varia-
tions indicate that methane molecules with spherical molecular
geometry cannot have access to some of the micropores that CO2
molecules with linear molecular geometry have access to, and the
CO2 molecules cannot have access to some of the micropores that
the small helium molecules can access. In essence, the organic
material of Sample 23 acts as a molecular sieve for fluids used
during measurements; see Fig. 7. Although this effect creates
uncertainties during the storage-capacity measurements for Sample
23, in general, it points to an important CO2-sequestration quality
for the shale-gas formations. It suggests that a large fraction of
the injected CO2 can be stored in small pores in which the other
naturally occurring gases, such as methane and nitrogen, cannot
be stored. Compared to methane, for example, CO2 can access an
additional 100 × (0.4771−0.345)/0.345 = 38.3% of the organic pore
volume when injected into Sample 23 under the same conditions.
The molecular sieving becomes a secondary effect in the case of
Sample 21, however. In this case, methane and CO2 have access
to the same pore space, whereas helium can penetrate into an
additional 7.7% organic-micropore volume.
As the consequence of changes in the estimated values of the
pore volume Vp0, the predicted gas-sorption parameters in Table 4
are quite different. Effect on the Langmuir adsorption isotherm is
shown in Fig. 8. Note that, because the pore-volume estimation for
Sample 21 is consistent, the predicted methane and CO2 isotherms
849
 TABLE 4—MEASURED GAS-STORAGE CAPACITY FOR THE OR GANIC-RICH SHALE SAMPLES USING VARIOUS GASES

Sample 21 S a m p le 2 3

He CO2 CH4 He CO2 CH4

Vp0 (cc) 0.42 0.39 0.39 0.65 0.48 0.35

Cp (psi )
–1 6 6 6 6 6 6
5.416 10 5.416 10 5.416 10 7.05 10 7.05 10 7.05 10
PL, CH4 (psi) using Vpo,He 3,232 1,000
using Vpo,CH4 1,800 2,502
Vamax, CH4 (cm )
3
using Vpo,He 0.4193 0.4056
using Vpo,CH4 0.1962 0.1423
Samax, CH4 (mol) using Vpo,He 0.0072 0.0021
using Vpo,CH4 0.0049 0.0041
3
ads, CH4 (mol/cm ) using Vpo,He 0.017 0.005
using Vpo,CH4 0.025 0.029
PL,CO2 (psi) using Vpo,He 500 655
using Vpo,CH4 500 2610
using Vpo,CO2 500 1410
Vamax,CO2 (cm )
3
using Vpo,He 0.4118 0.4300
using Vpo,CH4 0.3850 0.3200
using Vpo,CO2 0.3850 0.3400
Samax, CO2 (mol) using Vpo,He 0.0240 0.0280
using Vpo,CH4 0.0237 0.0211
using Vpo,CO2 0.062 0.0220
3
ads, CO2 (mol/cm ) 0.065

appear distinctly in Fig. 8a. The isotherms for the other sample
are shown in Fig. 8b. Indeed, methane isotherms show that the
gas stored as an adsorbed phase is significantly larger when the
pore volume obtained using helium is included in the analysis. The
contrast is even larger in the case of CO2. These results indicate
the importance of fluid type used for the adsorption analysis. In the
rest of the paper, we show only the storage and transport results
for a particular fluid using a sample pore volume measured with
that fluid.
Modeled total gas storage is given in Figs. 9 and 10. The
adsorbed-gas and free-gas amounts stored are shown in Fig. 9a
for Sample 21 and in Fig. 9b for Sample 23. Clearly, most of the
injected CO2 is stored in an adsorbed state. For example, at 2,500
psia and for Sample 21, the ratio of adsorbed to free CO2 is approx-
imately 13:1. For Sample 23, that ratio drops to 13:5, although the
total amount stored is nearly doubled. Hence, the measurements
reveal that the samples have larger adsorption capacity for CO2
at any given pore pressure. For both samples, the total CO2-gas-
storage capacity is 4 times greater than the total methane-gas-stor-
age capacity at 2,500 psia. Such a large contrast in methane and
CO2 storage is important for sequestration, although it cannot be
explained by the molecular-sieving arguments related to the organ-
ics of certain types of samples. It can be explained only by changes
in the fluid’s internal energy with the changes in the fluid type.
Indeed, the molecular-interaction energies (both among the gas
molecules and between the organic pore wall and gas molecules)
are different, which leads to significantly enhanced adsorption in
the case of CO2. Molecular-level investigation further comparing
adsorption of fluids and their mixtures in kerogen in the small
organic pores under reservoir conditions is currently necessary.
Results indicate that adsorption is the dominant mechanism
during the CO2 storage in organic-rich shale. This may also be the
case for methane, although the dominance is not as amplified and
distinct as in the case of CO2; see Fig. 10. At 2,500 psia, the ratio

Pure Binary mixture Ternary mixture

Fig. 7—Schematic of molecular sieving of gases in kerogen.

850 December 2011 SPE Journal

 CH4, Vp0=0.39 cm3 by CO4 CH4, Vp0=0.35 cm3 by CO4
350 0.0012 800
CH4, Vp0=0.42 cm3 by He CH4, Vp0=0.65 cm3 by He
CO2, Vp0=0.39 cm3 by CO4
0.0025

Adsorbed Gas Storage, SCF/ton

Adsorbed Gas Storage, mol/cm3

CO2, Vp0=0.35 cm3 by CO4

Adsorbed Gas Storage, SCF/ton

CO2, Vp0=0.39 cm3 by CO2 700

Adsorbed Gas Storage, mol/cm3

300 0.0010 CO2, Vp0=0.48 cm3 by CO2
CO2, Vp0=0.42 cm3 by He CO2, Vp0=0.65 cm3 by He
600 0.0020
250
0.0008
500
200 0.0015
0.0006 400
150
300 0.0010
0.0004
100
200
0.0002 0.0005
50 100

0 0.0000 0 0.0000
0 1,000 2,000 3,000 4,000 5,000 0 1,000 2,000 3,000 4,000 5,000
Pressure, psia
(a) (b) Pressure, psia

Fig. 8—Estimated Langmuir isotherms for Sample 21 (a) and Sample 23 (b). The parameters used are given in Table 4. The mark-
ers show the five experimental pressure data points used during the prediction of the gas-storage parameters.

500 500
Adsorbed gas, CO2 Adsorbed gas, CO2 0.0016
450 0.0016 450 Adsorbed gas, CO4
Adsorbed gas, CO4 0.0014
0.0014 Free gas, CO2
400 Free gas, CO2 400

Gas Storage, mol/cm3

Free gas, CO4
Gas Storage, SCF/ton

Gas Storage, mol/cm3

Gas Storage, SCF/ton
0.0012
350 Free gas, CO4 0.0012 350
300 300 0.0010
0.0010
250 250 0.0008
0.0008
200 200
0.0006 0.0006
150 150
0.0004 0.0004
100 100
50 0.0002 0.0002
50
0 0.0000 0 0.0000
0 1,000 2,000 3,000 4,000 5,000 0 1,000 2,000 3,000 4,000 5,000
Pressure, psia Pressure, psia

Fig. 9—Estimated adsorbed- and free-gas amounts for Sample 21 (a) and Sample 23 (b). Solid lines are with CO2, and the dashed
lines are with methane.
Adsorbed Gas Storage/Total Gas Storage

Adsorbed Gas Storage/Total Gas Storage

1.5 1.5

CO2 CO2
1 1

CH 4
CH 4
0.5 0.5

0 0
25 1,025 2,025 3,025 4,025 5,025 25 1,025 2,025 3,025 4,025 5,025
Pressure, psia Pressure, psia

(a) (b)

Fig. 10—Estimated ratio of adsorbed gas to total gas stored in Sample 21 (a) and in Sample 23 (b).

December 2011 SPE Journal 851

 1,630
1,620
1,610
Upstream Pressure, psi
1,600
Confining pressure 5,000 psi
1,590
Dark gray curve: experimental data
1,580 Light gray curve: mean of realizations
1,570 Black curve: 112 realizations
1,560
1,550
1,540
1,530
0 0.5 1 1.5 2 2.5
Time, seconds
(a)
Fig. 11—(a) Example history-matching results for Sample 21 using upstream pressure decline. 112 realizations are used to obtain
the optimum values of the transport coefficients shown in the first row in Table 5. (b) Permeability for Sample 21 is obtained
using the pulse decay shown on the left and using Jones’ straight-line method (Jones 1997).
of adsorbed CO2 to total CO2 gas storage is approximately 0.93
for Sample 21; on the other hand, the ratio for methane is approxi-
mately 0.54 for the same sample. In the case of Sample 23, at the
same pressure, the ratios are 0.72 and 0.46, respectively.
The rapid change in free-gas-CO2 measurements near 950 psia
in Figs. 9 and 10 is caused by dramatic changes in the z factor of
the gas near the critical fluid state.
The estimated coefficient of isothermal pore-volume-com-
pressibility values in Table 4 are nearly the same for the two sam-
ples, taking values on the order of from 5.0×10−6 to 7.0×10−6 psi−1.
During our gas-storage measurements and calculations, we found
that the results would have been less than 5% different when the
pore-compressibility effect had been ignored. Thus, the primary
correction necessary for the organic pore volume is the adsorption
effect. This behavior is quite different than in coals where most of
the matrix is organic and the pore-volume-compressibility values
could be one or two orders of magnitude larger; hence, its effect
on the pore volume measurements in coal becomes comparable to
the sorption effects (Koleowo 2010).
Finally, the estimated adsorbed-phase density values in Table
4 for methane and CO2 are in agreement with those reported in the
literature (Kurniawan et al. 2006; Ambrose et al. 2010), indicating
that the gas-storage models are reasonable and that the density for
the adsorbed-phase is close to the liquid density of the chemical
species used during the experiment.
A typical upstream pressure-decline curve of the pulse-decay
gas-permeation experiments is shown in Fig. 11. In this case,
methane initially in equilibrium with shale Sample 21 is allowed
to permeate. The curve typically yields two distinct behaviors
having completely different characteristic times. At early times,
the curve shows a rapid decline in pressure (fast transport), and,
later, it slows down, gradually decreasing the upstream pressure
(slow transport). These distinct portions of the decline curve
include information related to Darcy flow in the inorganic matrix
consisting of large voids, and pore/surface diffusion in the organic
pore network. In addition, the total time required for the shale/gas
system to reach the new thermodynamic equilibrium is related
to gas storability for the shale and can be used to obtain basic
information such as organic-/inorganic-pore-volume ratio. Having
the pressure-decline curve, we used a simulation-based history-
matching algorithm based on the randomized maximum likelihood
method and estimated the optimized transport properties of the
shale samples. The optimization requires different realizations of
the parameter space, typically on the order of hundreds. Details
of the automated nonlinear history-matching process are described
elsewhere (Oliver et al. 2008). Here, salient results related to the
852
Time, seconds
0 10,000 20,000 30,000 40,000 50,000
0
–0.5
–1
–1.5
–2
ln (ΔPD)
–2.5
–3
–3.5
–4 μgLcg
kg=|slope|
–4.5 6.805×10–5f1A(1/V1+1/V2)
3 3.5 4 4.5
×104 –5
(b)
nature of multiscale gas transport in shale are discussed for the
sequestration. Most importantly, we found that it is not possible
to obtain converging (optimum) results using unipore shale-matrix
models for the forward simulation; the matrix model used for the
history matching should be a dual-porosity (organic/inorganic)
system. History-matching results at various confining-stress condi-
tions and using a multiscale transport model are shown in Table 5
for Samples 21 and 23. Note that the time required to reach the new
equilibrium is significantly less as the pore pressure is increased.
This indicates that the transport paths are sensitive to effective
stress. The observation is supported further by the increase in the
estimated inorganic permeability and organic diffusivity values.
We note that permeability estimates are significantly less than
those obtained using conventional measurements, such as Jones’
method (see also Fig. 11b for details). This is possibly because of
the explicit introduction of ultralow-permeability organics into the
description of shale-gas transport in this work.
It was argued in the Interpretation of Organic Pore Diffusion
subsection that the pore-diffusion coefficient for a single capil-
lary should typically be two orders of magnitude larger than the
experimentally obtained apparent free-gas transport coefficient
DK. Here, using Eq. 4, we experimentally predict tortuosity val-
ues within the range of 4.0 to 6.0 for the kerogen. The estimated
values are consistent with the literature, showing high tortuosity
values for microporous materials and suggesting that the mea-
sured diffusion coefficient indeed represents the coefficient for
pore diffusion of free gas. More interestingly, the coefficient for
adsorbed-phase transport DS is predicted to be typically three
orders of magnitude larger than that for the free gas. This is a
somewhat curious behavior, which suggests that the surface diffu-
sion could be the dominant mechanism of transport in the organ-
ics. Typically, the adsorbed-phase concentration and its gradient
in the organics are smaller than those values associated with the
free gas; however, when multiplied with an extremely large sur-
face diffusivity, the product (i.e., mass flux of the adsorbed phase)
becomes a quantity that significantly contributes to the total mass
flux in the kerogen. The importance of surface diffusion to the
gas transport in heterogeneous coals and shales has been discussed
recently by Fathi (2010) and Fathi and Akkutlu (in press) on the
basis of theoretical arguments in the spectral (Fourier-Laplace)
domain. Here, the experimental results support the observation,
suggesting that the sequestered gases will also be transported as
part of the adsorbed phase in the organics. Fig. 12 shows that the
estimated mass flux from surface diffusion of the adsorbed gas
molecules is comparable to and could be even significantly larger
than the mass flux from pore diffusion of the free-gas molecules.
December 2011 SPE Journal
 TABLE 5—ESTIMATED METHANE TRANSPORT PARAMETERS* FOR ORGANIC-RICH SHALE SAMPLES
USING ISOTHERMAL PULSE-DECAY MEASUREMENTS AT VARYING PORE PRESSURES

Estimated Matrix Parameters for Shale Sample 21

Effective Stress Conditions Pulse
Total
Equilibrium Time To
Confining Pore Reach Jones’ TOC φ Inorganic Kerogen
Pressure Pressure p Equilibrium Permeability (mass/mass) (vol/vol)
(psi) (psi) (psi) (seconds) (nd) % % k (nd) k0 DK (cm2/s) DS (cm2/s)

0 1.73
5 2
5,000 1,623 83 44,485 70.4 3.9 1.33 15.8 53.87 4.45 3.50 10 1.10 10
5 2
5,000 2,588 84 18,063 102.2 3.9 1.24 20.4 64.16 4.39 4.00 10 4.20 10
5 2
5,000 2,974 70 12,060 122.9 3.9 1.22 28.7 67.56 4.04 4.90 10 8.80 10

Estimated Matrix Parameters for Shale Sample 23

Effective Stress Conditions Pulse
Total
Equilibrium Time To
Confining Pore Reach Jones’ TOC φ Inorganic Kerogen
Pressure Pressure p Equilibrium Permeability (mass/mass) (vol/vol)
(psi) (psi) (psi) (seconds) (nd) % % k (nd) k0 DK (cm2/s) DS (cm2/s)

0 2.85
5 3
5,000 1,772 78 72,060 12.65 3 2.4 2.6 43.58 6.14 3.50 10 1.55 10
5 3
5,000 25,008 74 54,480 16.31 3 2.32 3.52 47.09 5.10 5.30 10 3.50 10
5 3
5,000 3,066 73 32,520 17.7 3 2.27 5.24 48.23 4.10 8.20 10 6.35 10
* In the tables, porosity represents the void volume available for free gas only; hence, porosity is corrected for the pore com pressibility and methane-adsorption effects;
TOC is measured independently; k0 represents the ratio of kerogen pore volume to the total pore volume; Dk is the molecular diffusivity of free gas in kerogen.

Further investigation using different shale samples is currently
necessary.
Conclusions
In this paper, we have developed methodologies that allow us to
investigate in detail CO2 sequestration in organic-rich gas-shale
samples from Forth Worth basin. It is found that the pore-volume
estimation is a crucial step for the sequestration considerations
in gas shale. This volume may not be important for the storage
of free gas, however, because up to 97% of the uptaken gas is
stored in adsorbed state inside the organic pores, depending on the
shale-gas-reservoir pressure and temperature. It is also found that
the gas transport takes place in the presence of dynamic porosity
and permeability, and it could be dominated by the adsorbed-
phase transport taking place in the organics. It is a difficult task
to experimentally differentiate slippage of free-gas molecules
and surface-diffusion effects taking place in the phase-transition
region near the organic nanopore walls, although strong depen-
dence on the adsorbed-phase gradient indicates the significance of
surface diffusion in the organics. Although the study is important
for our understanding of gas storage and transport in organic-rich
shale, gas shales in North America may differ in terms of organic
content, pore structure of the organic material, and temperature.
Further investigation involving samples from different shale
plays is necessary for statistical inference and generalization of
the results.

Sample 21 Sample 23
2.5×10–9 2.5×10–9
Free gas diffusive mass flux Free gas diffusive mass flux
JK,x and JS,x(mol-cm/cm3/s)

JK,x and JS,x(mol-cm/cm3/s)

Adsorbed gas diffusive mass flux Adsorbed gas diffusive mass flux
2×10–9 2×10–9
Darcy flux
Darcy flux

1.5×10–9 1.5×10–9
Direction of
gas permeation
1×10–9 1×10–9

5×10–10 5×10–10

0 0
0 1 2 3 4 5 0 1 2 3
Distance, cm Distance, cm
(a) (b)

Fig. 12—Predicted mass fluxes across the shale Sample 21 at initial equilibrium pore pressure of 2,974 psia and Sample 23 at
3,066 psia during the pressure-pulse-decay measurements. The mass fluxes are computed using forward simulation with the
optimized transport parameters shown in Table 5. Only the flux values at the halftime to reach the new equilibrium pressure
during the pulse-decay measurement are shown.

December 2011 SPE Journal 853

Nomenclature Gas Conference, Pittsburgh, Pennsylvania, 23–25 February. doi:
As = surface area, nm2 10.2118/131772-MS.
C = moles of concentration Anderson, J.R. and Pratt, K.C. 1985. Introduction to Characterization and
Cp = the pore-volume compressibility, psi−1 Testing of Catalysts. Sydney, Australia: Academic Press.
Cµ = adsorbed-phase amount in moles per solid organic Bear, J. and Bachmat, Y. 1991. Introduction to Modeling of Transport
volume, mol/cm3 Phenomena in Porous Media, paperback edition. Dordrecht, The
DK = pore-diffusion coefficient, cm2/s Netherlands: Theory and Applications of Transport in Porous Media,
DK, capillary = molecular-diffusion coefficient for gas transport in the Kluwer Academic Publishers.
capillary, cm2/s Brace, W.F., Walsh, J.B., and Frangos, W.T. 1968. Permeability of Gran-
DS = porosity- and tortuosity-corrected surface diffusivity, ite Under High Pressure. J. Geophys. Res. 73 (6): 2225–2236. doi:
cm2/s 10.1029/JB073i006p02225.
JK,x = pore-diffusion mass flux, mol·cm/cm3/s Busch, A., Alles, S., Gensterblum, Y., Prinz, D., Dewhurst, D.N., Raven,
JS,x = surface-diffusion mass flux, mol·cm/cm3/s M.D., Stanjek, H., and Krooss, B.M. 2008. Carbon Dioxide Storage
n = number of moles, mol Potential of Shales. International Journal of Greenhouse Gas Control
n1 = moles of gas in V1, mol 2 (3): 297–308. doi: 10.1016/j.ijggc.2008.03.003.
n1f = moles of gas in V1 at Pf , mol Fathi, E. 2010. Fluid Flow, Transport and Reaction in Heterogeneous
n2 = moles of gas in V2, mol Porous Media. PhD thesis, University of Oklahoma, Norman, Okla-
n2f = moles of gas in V2 at Pf , mol homa.
nsp = moles of gas stored in the sample, mol Fathi, E. and Akkutlu, I.Y. 2009. Matrix Heterogeneity Effects on Gas
NKn = Knudsen number Transport and Adsorption in Coalbed and Shale Gas Reservoirs.
P = pore pressure Transport in Porous Media 80 (2): 281–304. doi: 10.1007/s11242-
P1 = equilibrium pressure in V1, psia 009-9359-4.
P2 = initial pressure in V3, psia Fathi, E. and Akkutlu, I.Y. In press. Mass Transport of Adsorbed-phase
Pf = new equilibrium pressure in dead volume and sample in Stochastic Porous Medium with Fluctuating Porosity Field and
volume, psia Nonlinear Gas Adsorption Kinetics. Transport in Porous Media (sub-
PL = Langmuir pressure, psia mitted 2010).
Ps = pore pressure in the sample, psia Finsterle, S. and Persoff, P. 1997. Determining Permeability of Tight
Sa = adsorbed-gas storage, mol Rock Samples Using Inverse Modeling. Water Resour. Res. 33 (8):
Samax = total moles of adsorbed gas at infinite pore pressure, 1803–1811.
mol Jones, S.C. 1997. A Technique for Faster Pulse-Decay Permeability Mea-
Sw = water saturation, % surements in Tight Rocks. SPE Form Eval 12 (1): 19–26. SPE- 28450-
V1 = reservoir volume, cm3 PA. doi: 10.2118/28450-PA.
V2 = dead volume, cm3 Koleowo, O.B. 2010. Gas Storage Capacity and Transport in Coals. MS
Va = adsorbed-gas volume, cm3 thesis, University of Oklahoma, Norman, Oklahoma.
Vamax = maximum adsorbed gas, cm3 Kurniawan, Y., Bhatia, S.K., and Rudolph, V. 2006. Simulation of Binary
VL = Langmuir volume, scf/ton Mixture Adsorption of Methane and CO2 at Supercritical Conditions in
Vp = sample pore volume, cm3 Carbons. AIChE Journal 52 (3): 957–967. doi: 10.1002/aic.10687.
Vp0 = pore volume at zero pore pressure, cm3 Lee, B. and Tan, T. 1987. Application of Multiple Porosity/Permeability
Vp2 = pore volume at P2, cm3 Simulator in Fractured Reservoir Simulation. Paper SPE 16009 pre-
Vpf = pore volume at Pf, cm3 sented at the SPE Symposium on Reservoir Simulation, San Antonio,
z1 = correction factor at P1 Texas, USA, 1–4 February. doi: 10.2118/16009-MS.
z2 = correction factor at P2 Loucks, R.G., Reed, R.M., Ruppel, S.C., and Jarvie, D.M. 2009. Morphol-
zf = correction factor at Pf ogy, Genesis, and Distribution of Nanometer-Scale Pores in Siliceous
zs = correction factor at Ps Mudstones of the Mississippian Barnett Shale. Journal of Sedimentary
n1 = change of moles of gas in V1, mol Research 79 (12): 848–861. doi: 10.2110/jsr.2009.092.
n2 = change of moles of gas in V2, mol Ning, X. 1992. The Measurement of Matrix and Fracture Properties
ns = change of moles of gas in sample, mol in Naturally Fractured Low Permeability Cores using a Pressure
nsa = change of moles of adsorbed gas in sample, mol Pulse Method. PhD thesis, Texas A&M University, College Station,
nsp = change of moles of gas stored in sample, mol Texas.
Vpa = change of volume of adsorbed gas in sample, cm3 Ozdemir, E., Morsi, B.I., and Schroeder, K. 2004. CO2 Adsorption Capac-
Vpp = change of pore volume at the pore pressure in ity of Argonne Premium Coals. Fuel 83 (7–8): 1085–1094.
sample, cm3 Oliver, D.S., Reynolds, A.C., and Liu, N. 2008. Inverse Theory for Petro-
k0 = organic pore volume to total pore volume, ratio leum Reservoir Characterization and History Matching. Cambridge
TOC = kerogen to total volume ratio University Press.
 = bulk density of gas, g/cm3 Roy, S., Raju, R., Chuang, H.F., Cruden, B.A., and Meyyappan, M. 2003.
ads = molar density of adsorbed gas, mole/cm3 Modeling Gas Flow Through Microchannels and Nanopores. J. Appl.
 = tortuosity Phys. 93 (8): 4870–4879. doi: 10.1063/1.1559936.
 = porosity Schettler, P.D., Parmely, C.R., Juniata, C., and Lee, W.J. 1989. Gas Stor-
age and Transport in Devonian Shales. SPE Form Eval 4 (3): 371–376;
Trans., AIME, 287. SPE-17070-PA. doi: 10.2118/17070-PA.
Acknowledgments
Sondergeld, C.H., Ambrose, R.J., Rai, C.S., and Moncrieff, J. 2010. Micro-
We recognize the work and contributions of Gary D. Stowe in the Structural Studies of Gas Shales. Paper SPE 131771 presented at the
IC3 laboratory at The University of Oklahoma. We thank Mery SPE Unconventional Gas Conference, Pittsburg, Pennsylvania, USA,
Diaz-Campos for her contribution to the generation of Fig. 2. 23–25 February 2010. doi: 10.2118/131771-MS.
For insightful discussions, we thank Chandra S. Rai and Carl H. Wang, F.P. and Reed, R.M. 2009. Pore Networks and Fluid Flow in
Sondergeld in shale petrophysics and Dean S. Oliver in nonlinear Gas Shales. Paper SPE 124253 presented at the SPE Annual Tech-
optimization. nical Conference and Exhibition, New Orleans, 4–7 October. doi:
10.2118/124253-MS.
References Wei, X.R., Weng, G.X., Massarotto, P., Rudolph, V., and Golding, S.D.
Ambrose, R.J., Hartman, R.C., Diaz-Campos, M., Akkutlu, I.Y., and Son- 2007. Modeling Gas Displacement Kinetics in Coal With Maxwell-
dergeld, C.H. 2010. New Pore-scale Considerations in Shale Gas in Place Stefan Diffusion Theory. AIChE Journal 53 (12): 3241–3252. doi:
Calculations. Paper SPE 131772 presented at the SPE Unconventional 10.1002/aic.11314.

854 December 2011 SPE Journal

Seung Mo Kang is an MS student in petroleum engineer-
ing at The University of Oklahoma. He holds a BS degree in
mechanical engineering from Hanyang University in South
Korea. His current research interest is quantification of gas-
storage mechanisms in shale gas reservoirs. Ebrahim Fathi is a
research associate at The University of Oklahoma. Fathi holds
BS and MS degrees in exploration of mining and petroleum
engineering from Tehran University. He holds a PhD degree in
petroleum engineering from The University of Oklahoma. His
research interests are up-scaling of the fluid flow, transport
and reaction processes in heterogeneous porous media, gas
transport and storage in shale gas and coalbed methane res-
ervoirs, and simulation of flow dynamics in naturally occurring
nanoporous materials. Raymond J. Ambrose is a PhD student
in petroleum engineering at The University of Oklahoma and
director of reservoir engineering for Reliance Holding USA. He
holds a BS degree in chemical engineering and an MS degree
in petroleum engineering, both from the University of Southern
California. His current research interests are analytical solu-
tions for gas shale productivity, SEM imaging and pore struc-
ture characterization for organic-rich gas shale, identification
December 2011 SPE Journal
of gas-storage mechanisms in unconventional gas resources,
and estimation of gas in place. I. Yucel Akkutlu is professor
of petroleum engineering at The University of Oklahoma. He
holds MS and PhD degrees from the University of Southern
California. His work finds applications in reservoir engineering,
particularly in the areas of unconventional gas recovery and
improved and enhanced oil recovery. His current research
deals with the thermodynamics of fluids in nanoporous materi-
als and with the scaling up of coupled transport and reaction
processes in low-permeability geological formations exhibiting
multiscale pore structures. Richard F. Sigal joined The University
of Oklahoma in 2004 as Unocal Centennial Professor with a
joint appointment in petroleum engineering and geology and
geophysics. His primary research activity since then has been
on tight gas, shale gas, and other unconventional gas reser-
voirs. Before joining The University of Oklahoma, he worked in
the oil and gas industry for 26 years on reservoir characteriza-
tion, core measurements, and petrophysics. Among his areas
of special expertise are nuclear magnetic resonance, mercury
capillary pressure measurements, and the stress dependence
of permeability.
855
 J167626 DOI: 10.2118/167626-PA Date: 10-April-15
Simultaneous Multifracture Treatments:
Fully Coupled Fluid Flow and Fracture
Mechanics for Horizontal Wells
Kan Wu, SPE, and Jon E. Olson, SPE, University of Texas at Austin
Summary
Successfully creating multiple hydraulic fractures in horizontal
wells is critical for unconventional gas production economically.
Optimizing the stimulation of these wells will require models that
can account for the simultaneous propagation of multiple, poten-
tially nonplanar, fractures.
In this paper, a novel fracture-propagation model (FPM) is
described that can simulate multiple-hydraulic-fracture propagation
from a horizontal wellbore. The model couples fracture deformation
with fluid flow in the fractures and the horizontal wellbore. The dis-
placement discontinuity method (DDM) is used to represent the
mechanics of the fractures and their opening, including interaction
effects between closely spaced fractures. Fluid flow in the fractures
is determined by the lubrication theory. Frictional pressure drop in
the wellbore and perforation zones is taken into account by applying
Kirchoff’s first and second laws. The fluid-flow rates and pressure
compatibility are maintained between the wellbore and the multiple
fractures with Newton’s numerical method. The model generates
physically realistic multiple-fracture geometries and nonplanar-frac-
ture trajectories that are consistent with physical-laboratory results
and inferences drawn from microseismic diagnostic interpretations.
One can use the simulation results of the FPM for sensitivity
analysis of in-situ and fracture treatment parameters for shale-gas
stimulation design. They provide a physics-based complex frac-
ture network that one can import into reservoir-simulation models
for production analysis. Furthermore, the results from the model
can highlight conditions under which restricted width occurs that
could lead to proppant screenout.
Introduction
Multiple hydraulic fracturing in combination with horizontal dril-
ling significantly improves well productivity in shale reservoirs.
Limited-entry fracturing techniques are widely used to create
simultaneously propagating multiple fractures (Daneshy 2011). In a
single fracturing stage, there can be three to six perforation clusters.
In comparison with modeling single fracturing propagation, multi-
ple-fracture-propagation modeling needs to consider mechanical
fracture interaction, which is a dynamic process of subsurface geo-
mechanical stress changes induced by fracture growth. The signifi-
cant consequences of multifracture treatments are stress-shadow
effects that lead to complex fracture geometry and fracture-width
restriction. Fracture geometry can be further complicated by pre-
existing natural fractures in shale-gas reservoirs (Gale et al. 2007;
Olson 2008; Chong et al. 2010), which is not the focus of this work
but which will be discussed in a future publication. To optimize
multiple-fracture treatments, characterization of the fracture geom-
etry is very necessary. However, the complex-fracture geometry
cannot be measured directly and completely by any fracture diag-
nostic techniques. Therefore, modeling remains an important tool
for engineers to predict the fracture geometry.
Recent efforts to model multifracture treatments have taken var-
ious forms. Multicrack models were described by Siriwardane and
Copyright V
C 2015 Society of Petroleum Engineers
This paper (SPE 167626) was accepted for presentation at the SPE Annual Technical
Conference and Exhibition, New Orleans, 30 September–2 October 2013, and revised for
publication. Original manuscript received for review 1 November 2013. Revised manuscript
received for review 8 March 2014. Paper peer approved 17 April 2014.
April 2015 SPE Journal
ID: jaganm Time: 11:58 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037
Stage: Page: 337 Total Pages: 10
Layne (1991), Mack et al. (1992), and Elbel et al. (1992), in which
fracture propagation was modeled according to the Geerstma and
DeKlerk geometry, the Perkins and Kern geometry, and a pseudo-
3D geometry, respectively, with fluid flow in multiple fractures and
the wellbore, but with no mechanical interaction between multiple
fractures nor independent pressure drop in each fracture. Olson
(2008) used a pseudo-3D approach for vertical but laterally nonpla-
nar fractures, assuming zero-viscosity injection fluid to reduce
computation time for multifracture problems. Roussel and Sharma
(2011) used a 3D finite-difference numerical model to calculate
stress distributions around uniform-pressure fractures, accounting
for propagation by assuming fractures would follow principal-stress
trajectories. Weng et al. (2011) and Wu et al. (2012) describe a
model for simulating multiple hydraulic fractures in naturally frac-
tured reservoirs combining standard pseudo-3D equations for frac-
ture-width calculations coupled with the displacement discontinuity
approach of Olson (2004) to account for stress-interaction effects
between fractures. Meyer and Bazan (2011) presented a discrete-
fracture-network model to simulate planar-fracture growth in the
three principal planes, with the Olson (2004) analytical solution to
describe the interaction between induced fractures. Wong and Xu
(2013) modeled multiple-fracture propagation in horizontal wells
with mechanical interaction through the boundary integral formula-
tion. Castonguay et al. (2013) implemented the weakly singular,
symmetric Galerkin boundary-element method to calculate fracture
geometry, in which flow in the fracture is modeled as power-law
fluid flow in the arbitrary curved channels. Nagel et al. (2011) pro-
posed a discrete-element model, in which the discrete-fracture net-
work consists of planar, finite fractures in three dimensions that
bound intervening rigid or deformable reservoir blocks. Dahi Tale-
ghani and Olson (2011) presented a coupled fluid-flow and rock-
mechanics model that was based on the extended finite-element
method to simulate the fracture geometry of plane-strain problems.
Keshavarzi and Mohammadi (2012) also applied the extended fi-
nite-element method to analyze the interaction between hydraulic
and natural fractures under plane-strain and quasistatic conditions.
The model presented in this paper follows the single nonplanar-
fracture model of Olson and Wu (2012). The model incorporates
fracture deformation and non-Newtonian fluid flow in the wellbore
and fractures. Fracture widths and stress-intensity factors are calcu-
lated with the modified, pseudo-3D DDM approach of Olson
(2004), accounting for all mechanical interactions between every
fracture element at every timestep. The flow rate in each fracture
satisfies mass balance and pressure equilibrium between the hori-
zontal wellbore and hydraulic fractures. The Newton-Raphson nu-
merical method is applied to solve nonlinear fluid-flow equations.
The Picard iterative method is used to fully couple fracture
mechanics and fluid flow. We validate our model against two ana-
lytical solutions and published numerical results. A field case study
is performed to predict multiple-fracture geometry and to calculate
stress distribution with the purpose of demonstrating the feasibility
of the model.
Fracture Mechanics
Fracture mechanics determines fracture displacements and propa-
gation direction. The 2D DDM (Crouch 1976) with a 3D correc-
tion factor (Olson 2004) is used to calculate normal and shear
displacements considering mechanical interaction. The fracture-
337
 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 338 Total Pages: 10

y y

Dx Dy

x z
σzz
2a

σxx y

h
+
S x
–
S

x
Fig. 2—Sketch of a uniformly loaded, isolated, vertical fracture
Fig. 1—Illustration of a 2D boundary with crack-like geometry. with a finite height and an infinite length.

propagation direction is determined by the maximum-circumfer- Because hydraulic fractures are best modeled accounting for
ential-stress theory (Erdogan and Sih 1963), a criterion based on their 3D geometry (finite height and length), we use a 3D correc-
the stress-intensity factors at the crack tips. tion factor Gij derived by Olson (2004),

DDM. The 2D DDM, a special boundary-element method, was dijb

Gij ¼ 1  h ib=2 ; . . . . . . . . . . . . . . . . . . . ð3Þ
developed by Crouch (1976) and designed for handling problems
with crack-like geometries. Stresses at a point n within an elastic dij2 þ ðh=aÞ2
region are integrals of displacements Dui over the boundary S6 of
the region (Fig. 1), where dij is the distance between the centers of fracture elements i
ð and j, h is the fracture height, and a and b are empirically deter-
mined constants found by comparing displacements and stresses
rjk ðnÞ ¼ Eijk ðn; gÞDui ðgÞdSðgÞ
S ; . . . . . . . . . . . . . ð1Þ with a true 3D fracture solution (Shou 1993). This correction fac-
þ
Dui ¼ ui  u i
tor was modeled after the analytical plane-strain equation for the
normal stress, rxx, acting perpendicular to a uniformly loaded, iso-
where the quantity Eijk(n,g) is the tensor field that represents the lated, vertical crack of finite height and infinite length (Pollard
influences of a concentrated force at point n on displacements at and Segall 1987). The use of a ¼ 2, b ¼ 3, and dij ¼ x in Eq. 3
point g. Because analytical solutions are very hard to obtain for returns the dimensional factor from the following equation for
Eq. 1, a numerical solution was applied to deal with this problem. fracture-induced normal stress rxx along the x-axis (Fig. 2),
The boundary S6 is divided into N planar elements. An analytical 8 9
solution of a constant displacement-discontinuity element in an >
< 3 >
=
infinite elastic solid is obtained with the Green’s-function ap- jxj
rxx ðxÞ ¼ DrI 1  h i : . . . . . . . . . . . . ð4Þ
proach that is based on the solution of Kelvin’s problem. On the >
:
3=2 >
;
x2 þ ðh=2Þ2
basis of the analytical solution, one can calculate stresses at arbi-
trary points with the principle of superposition by linearly com-
bining the contributions of the displacement discontinuities of all Maximum-Circumferential-Stress Criterion. The opening
elements. mode and sliding-shear mode stress-intensity factors, KI and KII,
For fracture problems, displacement or stress-boundary condi- are functions of normal- and shear-displacement discontinuities at
tions are generally defined on the crack surfaces along with any the crack tip element, Dn and Ds, respectively, because (Olson
remote-boundary conditions. In addition, the simultaneous solu- 1990, 2007)
tion of the system of equations accounts for stresses caused by the pffiffiffi
mechanical interaction with other nearby elements (Crouch and 0:806E p
Starfield 1983). Thus, because of the normal- (rin ) and shear- (ris ) KI ¼ pffiffiffiffiffi Dn
4ð1   2 Þ 2a
stress boundary conditions on each element i determined with the and pffiffiffi . . . . . . . . . . . . . . . . . . . . . ð5Þ
fluid-flow solution, we can determine the normal- and shear-dis- 0:806E p
placement discontinuities, Dnj and Dsj , respectively, at each ele- KII ¼ pffiffiffiffiffi Ds :
4ð1   2 Þ 2a
ment j by solving a linear system of equations given by
These expressions were shown to be accurate for stress-inten-
X
N X
N
rin ¼ Gij Cijns Dsj þ Gij Cijnn Dnj sity-factor determination in fully 3D displacement-discontinuity
j1 j1
models as well (Sheibani and Olson 2013). Knowing the stress-in-
: . . . . . . . . . . . . . . . ð2Þ tensity factors, one can determine the fracture-propagation direc-
X
N X
N
ris ¼ Gij Cijss Dsj þ Gij Cijsn Dnj tion h by the maximum-circumferential-stress theory (Erdogan
j1 j1 and Sih 1963) as
ij ij
C and G are factors that account for the elastic mechanical KI sinh þ KII ð3cosh  1Þ ¼ 0; . . . . . . . . . . . . . . . . . . . ð6Þ
interaction between elements. Cijns is the plane-strain, elastic-influ-
ence coefficient matrix representing the normal stress at element i where h is the counterclockwise positive angle from the in-plane
induced by a shear-displacement discontinuity at element j, and propagation of the fracture ahead of the tip. For pure opening
Cijnn gives the normal stress at element i because of an opening dis- mode (KI > 0 and KII ¼ 0), we would predict planar-fracture prop-
placement discontinuity at element j. We can attribute analogous agation along h ¼ 0 . For pure sliding mode (KI ¼ 0 and KII = 0),
meanings to Cijss and Cijsn . The detailed derivation of C is given by the strongest possible kink in fracture direction occurs along
Crouch and Starfield (1983). h ¼ 670.5 for positive or negative KII, respectively.

338 April 2015 SPE Journal

ID: jaganm Time: 11:58 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 339 Total Pages: 10

Fluid Flow pcf,i is pressure loss in the horizontal wellbore given by (Valko
Multiple-fracture propagation is driven by an incompressible and and Economides 1995)
non-Newtonian fluid. Fluid flow consists of two parts—fluid flow X
i
0
in the fracture and fluid flow in the horizontal wellbore between pcf ;i ¼ Ccf ðxj  xj1 ÞQwj n ;
injection points. j¼1
2ðX
j1Þ
Fluid Flow in a Single Fracture. Fluid flow through a rock frac- Qwj ¼ QT  Qk ;
ture is modeled by the Navier-Stokes equations of fluid mechanics k¼1 : . . . . . . . . . .ð11Þ
(Schlichting 1968). When assuming a uniform pressure gradient Qwj ¼ QT at j ¼ 1;
in the channel between two parallel and smooth fracture surfaces,
one can simplify the Navier-Stokes equations (Valko and Econo- and  n0
mides 1995): 0 0 1 þ 3n0 0
Ccf ¼ 23n þ2 pn k0 Dð3n þ1Þ :
 0 n0
@p 0 1 þ 2n0 n n0 ð2n0 þ1Þ n0
¼ 2n þ1 k0 h w Q : . . . . . . . . . . ð7Þ
@s n0 The pressure drop between the wellbore intervals is a function of
distance between two neighboring fractures and flow rate Qwj in the
From Eq. 7, we can obtain the relationship between pressure drop wellbore intervals. The longer the distance between two neighboring
and fluid properties, fracture geometry, and flow rate. Pressure fractures, the larger the pressure drop. This is a rough approach to esti-
drop increases with increasing flow rate Q and decreases with mate frictional pressure drop in the wellbore without considering
increasing fracture width w. In our model, to enhance conver- flowing conditions. Siriwardane and Layne (1991) gave a more-accu-
gence stability with little cost in accuracy, the flow-rate gradient rate approach to calculate the pressure loss in the wellbore by taking
along the fracture is ignored. Spence and Sharp (1985) and Zim- into account friction factors under turbulent and laminar conditions.
merman et al. (1991) discussed the validation of the lubrication
theory for modeling fluid flow inside the fractures. According to Material Balance. One must satisfy material balance in the
their analysis, fracture aperture cannot change too abruptly and whole fluid-flow system. The total volume of fluid pumped during
should be small in comparison to the other dimensions of the frac- injection time is equal to the fluid volume preserved in the multi-
tures, and fluid flow should be laminar. ple fractures and the volume leaking into the surrounding porous
medium. The global volume balance is
Fluid Flow in Horizontal Wellbore. A numerical model for
multilayer-fracturing treatments in vertical wells was presented ðt i ðtÞ
Lð i ðtÞ ð
Lð t
X
N X
N
by Siriwardane and Layne (1991), Mack et al. (1992), and Elbel QT ðtÞdt ¼ hwds þ qL ðs; tÞdtds: . . . . ð12Þ
et al. (1992). The model described the conservation of volume 1 1
0 0 0 0
and the continuity of pressure by use of a set of constraints, such
as hydrostatic-pressure change, perforation frictional pressure The leakoff volume is described by the Carter leakoff model
drop, and frictional pressure loss in each casing interval. Fluid (Carter 1957), which is expressed as
flow in the horizontal wellbore is similar to fluid flow in the verti-
cal wellbore with multiple layers. The only difference is that we 2hCL
qL ðs; tÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
assume no hydrostatic-pressure change in the horizontal wellbore. t  sðsÞ
Fluid flow in the wellbore is analogous to the flow of electric
current through an electrical circuit network, by applying Kirch- The model assumes that leakoff is 1D in the direction perpen-
off’s first and second laws to obtain the flow-rate distribution of dicular to the fracture plane.
multiple fractures (Siriwardane and Layne1991; Mack et al. 1992;
Elbel et al. 1992). Our model assumes that multiple-transverse Iterative Algorithm
fractures are started perpendicular to the horizontal wellbore at
The fracture mechanics and fluid flow are coupled numerically
perforation-cluster locations. The total injection rate QT is divided
through an iterative algorithm. The elastic DDM solution deter-
into injection rates of each fracture Qi. The flow-rate distribution
is a dynamic process, and the diversion of the fluid into various mines the displacement discontinuities (opening and shearing) for
perforation clusters is a function of wellbore friction, perforation each fracture element because of its internal pressure and resolved
friction, and the fluid pressure within the fractures (Fig. 3). external stresses, whereas the internal-pressure distribution is
By ignoring wellbore-storage effects, total injection rate related to fracture opening by the fluid flow. The mechanical solu-
tion is a linear set of equations, whereas the fluid flow is nonlin-
should be the sum of the injection rates of all fractures,
ear. Fluid flow is discretized by the finite-difference method on a
X
N fixed grid of boundary elements. Newton’s numerical method is
QT ¼ Qi : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð8Þ used to solve the system of nonlinear equations. Picard coupling
i¼1 (Li 1991; Adachi et al. 2007) is used to couple the fracture
mechanics and fluid flow at every updated timestep. If the conver-
Kirchoff’s second law describes the continuity of pressure, in gence criterion is not satisfied, the net pressure is modified at each
which the pressure in the wellbore heel is equal to the sum of iteration step as
wellbore frictional pressure drop, perforation frictional pressure
drop, and pressure in the first element of the fracture as (Elbel pkþ1 ¼ ð1  ar Þpk1 þ ar pk : . . . . . . . . . . . . . . . . . . . ð14Þ
et al. 1992)
X
i Displacement discontinuities are recalculated with the modi-
P0 ¼ pw;i þ ppf ;i þ pcf ; j : . . . . . . . . . . . . . . . . . . . ð9Þ fied pressure from the previous iteration step. The convergence
j¼1 criterion is checked again after computing the new net pressure
on the basis of the displacements. ar is a retardation coefficient
pw,i is pressure in the first element of fracture, which is calculated ranging from 0 to 0.5 that dampens variations to avoid stress
by Eq. 7; ppf,i is the perforation frictional pressure loss that is pro- jumps and extreme changes in pressure. If the convergence crite-
portional to the square of flow rate, estimated by (Elbel et al. rion cannot be satisfied, decreasing ar is an approach to improve
1992) convergence. Smaller ar can cause convergence to be too slow,
qs whereas a large ar can result in diverging oscillations. We found
ppf ;i ¼ Qi 2 : . . . . . . . . . . . . . . . . . . . ð10Þ that ar ¼ 0.2 provides a reasonably stable convergence. The flow
0:323qw n2p;i dp;i
4
chart in Fig. 4 describes the structure of the program.

April 2015 SPE Journal 339

ID: jaganm Time: 11:58 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 340 Total Pages: 10

Frac. 1 Frac. 2 Frac. n

Q1 Pw,1 Q3 Pw,3 Q2n-1 Pw,2n-1

QT,P0 Ppf,1 QT-Q1-Q2 Ppf,3 QT – (Q1+...+ Q2n-2) Ppf,2n-1

Ppf,2 ΔPcf,2 Ppf,4 ΔPcf,n Ppf,2n

ΔPcf,1

Q2 Pw,2 Q4 Pw,4 Q2n Pw,2n

Fig. 3—Illustration of flow-rate partition and pressure drop.

Model Validation
Comparison With Analytical Solutions. Our FPM extends from
a nonplanar single-fracture model to a multiple-fracture model.
The single-fracture, nonplanar model developed by Olson and Wu
(2012) was validated against planar pseudo-3D models from the
Gas Research Institute Staged Field Experiment No. 3 case (War-
pinski et al. 1994). In this paper, we compare our model to the
end-member analytical solutions of Khristianovich-Geertsma-
Deklerk (KGD) and Perkins-Kern-Nordgren (PKN) (Table 1).
The KGD model presumes plane strain in the horizontal plane and
a rectangular vertical cross section (no width variation over the
height). The PKN model assumes plane strain in a vertical plane
normal to the fracture-propagation direction, and the vertical cross
section of the fracture is elliptical in shape. To compare with the
KGD and PKN solutions, we simulate a case with a fracture height
of 200 ft. The fracture geometry calculated by KGD is much fatter
and shorter with very low net pressure that climbs with time, com-
pared with the PKN model. At the earlier time of propagation,
KGD gives more-accurate results than PKN because fracture length
is shorter than fracture height. As fracture length grows, KGD grad-
ually loses its accuracy, and PKN can give more-accurate results.
The other input parameters are listed in Table 2.
A comparison of results made for fracture length (Fig. 5), frac-
ture width at the wellbore (Fig. 6), and net injection pressure
(Fig. 7) is presented. As shown, our model can give a very good
agreement with the KGD in the earlier time and tends to match
PKN analytical solutions as the fracture grows longer, which
lends credence to our model results and analysis.
Comparison With Unconventional Fracture Model. For multi-
ple-fracture propagation, fracture interaction plays an important
role in fracture geometry. For nonplanar propagation validation,
we refer for comparison to the nonplanar model of Weng et al.
(2011) and Wu et al. (2012) for the configurations of a single

Initialization

A new timestep

Element length Add a new element Propagation direction

Displacement
Fracture mechanics
Stress shadow
Pressure
Fluid flow
Flow rate

No Converge

Yes

Mass balance

Check treatment time Continue

Finished

Postprocessing

Fig. 4—Flow chart of the numerical scheme in the model.

340 April 2015 SPE Journal

ID: jaganm Time: 11:58 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 341 Total Pages: 10

TABLE 1—ANALYTICAL SOLUTION OF KGD AND PKN (VALKO AND ECONOMIDES 1995)

Fracture Wellbore Injection

Half-Length Width Pressure
!1=6 !1=6
KGD
i3 E0 li3 Shmin þ 1:09ðE0 2 lÞ1=3 t1=3
0:539 t2=3 2:36 t1=3
lh3f E0 h3f
!1=5  2 1=5 !1=5
i3 E0 4=5 li E0 4 li2
PKN 0:524 t 3:04 0 t1=5 Shmin þ 1:52 t1=5
lh4f E hf h6f

1200
TABLE 2—INPUT PARAMETERS
FPM KGD PKN
1000

Fracture half-length, ft
l 100 cp
Injection Rate 20 bbl/min 800
Young’s Modulus 6,530,000 psi
Poisson’s Ratio 0.2 600
Height 200 ft
400
Minimum Horizontal Stress 4,450 psi
200

0.4 0
Average width at the wellbore, in.

FPM KGD PKN 0 10 20 30 40 50 60 70

0.35
Injection time, min
0.3

0.25 Fig. 5—Comparison of fracture half-length vs. injection time

between the KGD and PKN models and our model.
0.2

0.15
1200
0.1 FPM KGD PKN
Net injection pressure, psi

0.05 900
0
0 10 20 30 40 50 60 70
Injection time, min 600

Fig. 6—Comparison of fracture width at the wellbore between

the KGD and PKN models and our model. 300

horizontal wellbore with two initial fractures (Fig. 8a) and two
horizontal wellbores with one fracture each (Fig. 8b). We simu- 0
0 10 20 30 40 50 60 70
lated these two configurations under isotropic and anisotropic in-
situ stress conditions (see Table 3 for input parameters). The dif- Injection time, min
ferential stress for the anisotropic-stress reservoir is 130 psi, with
the maximum horizontal stress oriented in the y-direction. Fig. 7—Comparison of fracture pressure at the wellbore
between the KGD and PKN models and our model.
Two fractures, started parallel with each other in a horizontal
wellbore for isotropic and anisotropic far-field stresses, are illus- trated in Fig. 9. The two fractures propagate away from each other
because of the mechanical fracture interaction (i.e., stress-shadow
effects). An opening fracture exerts additional stresses on sur-
rounding rock and adjacent fractures, which can result in a local
change in the direction of maximum horizontal stress and a devia-
tion of the fracture path from a planar geometry. In addition, the
fractures have larger curvature under the isotropic-stress condition
than that under the anisotropic condition. This is because the ani-
sotropic far-field stresses tend to suppress fracture curving (Olson
and Pollard 1989). The stress shadow and the anisotropic far-field
stresses compete with each other. To quantify the competition,
Olson and Dahi Taleghani (2009) defined a parameter called the
relative net pressure Rn,
ðPfrac  Shmin Þ
Rn ¼ ; . . . . . . . . . . . . . . . . . . . . . . . ð15Þ
ðShmax  Shmin Þ

(a)
Fig. 8—Two initiation fractures in a horizontal wellbore (a) and
two horizontal wellbores (b); blue lines represent horizontal
wellbores, and red lines represent initial fractures.
(b) where the numerator is the net pressure and the denominator is hori-
zontal differential stress. Stress-shadow effects strongly depend on
the net pressure within the fractures (Pollard and Segall 1987; War-
pinski and Branagan 1989). Larger net pressure will lead to stronger
stress-shadow effects. When Rn is much greater than unity, the net

April 2015 SPE Journal 341

ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 342 Total Pages: 10

in a horizontal wellbore. We explore the details of simultaneous

TABLE 3—INPUT PARAMETERS (AFTER WU ET AL. 2012)
Injection Rate 40 bbl/min
Stress Anisotropy 130 psi
Poisson’s Ratio 0.35
Young’s Modulus 4.35106 psi
Fluid Viscosity 1 cp
Fracture Height 394 ft
Minimum Horizontal Stress 6,773 psi
Maximum Horizontal Stress 6,903 psi
Distance Between Initiation Points 33 ft
pressure is much greater than differential stress, which suggests that
stress-shadow effects will dominate the fracture geometry. However,
if the Rn is much less than unity, the fractures will propagate along a
straight line under the domination of differential stress. However,
this parameter only indicates whether a nonplanar fracture-propaga-
tion path is possible, which means that strong stress-shadow effects
still affect fracture width even if the fractures are perfectly straight in
a strongly anisotropic stress field. As seen from Fig. 9, we found that
these two models give a very similar fracture-propagation path with
side-by-side configuration.
Fig. 10 shows two fractures in an en-echelon arrangement
from two adjacent horizontal wellbores under isotropic and aniso-
tropic far-field stress conditions. The two fractures grow toward
each other, because of induced shear stress at the fracture tips
caused by mechanical interaction. In the anisotropic far-field case,
there is a reduced curvature of fracture paths compared with the
isotropic case. In summary, our model compares favorably with
the nonplanar results generated by Wu et al. (2012).
A Case Study for Six Fractures Propagating
Simultaneously
Generally, there are three to six perforation clusters for each frac-
turing stage to create simultaneous multiple-fracture propagation
fracture propagation from multiple clusters by studying a case of
six initial fractures investigating propagation geometry, pressure
distribution, and stress-shadow effects. The input parameters
(Table 4) are based on the published data for the Barnett shale.
Fig. 11 displays the fracture trajectory and fracture width at the
end of 12 minutes of injection time. This fracture geometry is
dominated by two factors—distance from injection point and me-
chanical interaction between fractures. To clearly illustrate fracture
aperture distribution for each fracture, we enlarge the aperture 500
times in Fig. 11b. The injection point is at the heel of the horizontal
wellbore (–200, 0). Fractures close to the heel obtain more fluid
because of viscous dissipative flow through the wellbore from ear-
lier fractures. Fractures obtaining more fluid have greater length
and aperture. Fractures 1 and 6 are under the same stress-shadow
position because they are on the outside of the array, but Fracture 1
is closer to the injection point. Therefore, Fracture 1 has higher net
pressure and grows longer with larger width compared with Frac-
ture 6. All of the interior fractures (Fractures 2, 3, 4, and 5) have
similar trends. The space of 50 ft puts adjacent fractures within the
region of stress-shadow influence of their neighbor, resulting in a
strong mechanical interaction that affects propagation path and
width. Fractures 2 and 5 are close to the exterior Fractures 1 and 6,
respectively, and are strongly affected by induced stresses of the
fractures, slowing the growth rate of Fractures 2 and 5 and causing
them to deviate from their original propagation path (Fig. 12). Frac-
tures 3 and 4 are less affected by the two exterior fractures. In gen-
eral, the interior fractures are much narrower and shorter compared
with the exterior fractures (Fig. 12); this implies a greater flow re-
sistance for these fractures that diverts fluid elsewhere. The net
pressures for the interior fractures are almost the same as that of the
exterior fractures (they all have to have roughly the same net pres-
sure at the injection point because of the low pressure drop in the
wellbore) (Fig. 13). For that to be true, because the widths are less,
the flow rate has to be less into those fractures. Consequently, inte-
rior fractures have less ability to accept proppant during fracturing
and a greater risk of proppant screenout. In summary, the key fea-
tures for the multiple-fracture geometry are that the interior

–850 –850
FPM FPM
Wu et al. 2012 Wu et al. 2012

–900 –900

–950 –950

–1000 –1000
y, ft
y, ft

–1050 –1050

–1100 –1100

–1150 –1150

–1200 –1200
–60 –10 40 90 –60 –10 40 90
x, ft x, ft

Fig. 9—Comparison of propagation paths for two initially parallel fractures in isotropic and anisotropic stress fields: left figure
refers to isotropic stress field; right figure refers to anisotropic stress field.

342 April 2015 SPE Journal

ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 343 Total Pages: 10

–850
–850 FPM
FPM
Wu et al. 2012
Wu et al. 2012

–900 –900

–950 –950

–1000 –1000

y, ft
y, ft

–1050 –1050

–1100 –1100

–1150 –1150

–1200 –1200
–60 –10 40 90 –60 –10 40 90
x, ft x, ft

Fig. 10—Comparison of propagation paths for two initially offset fractures in isotropic and anisotropic stress fields: left figure
refers to isotropic stress field; right figure refers to anisotropic stress field.

fractures are shorter, narrower, and more nonplanar than the exte- rxx is much larger than that for ryy, because fractures open against
rior fractures. the rock and directly squeeze the rock in the x-direction, whereas
One can further understand the resultant fracture paths and y-direction effects are reduced by Poisson’s ratio effects. Thus, at
widths by examining the stress distributions induced by the frac- some specific points the change in rxx is greater than the change
tures. Before fracturing, the minimum horizontal stress of rxx is ori- in ryy, such that rxx might become larger than ryy. This flip in the
ented parallel to the x-direction, and the maximum horizontal stress orientation of the compressive stress explains why interior frac-
is the ryy. Fractures grow along the y-direction because that path pro- tures change their propagation directions to seek the path with
vides the least resistance. As the fractures grow, geomechanical least resistance.
stresses around fractures are dynamically changed. In the Figs. 14 One also can use the induced stress field to predict the poten-
and 15, red colors indicate an increase of compressional stress, and tial for microseismic activity. Microseismic events are induced by
blue colors represent a decrease of compression. Nearby and ahead changes in stresses and pore pressure caused by hydraulic fractur-
of the fracture tips, compressional stresses of rxx and ryy decrease ing (Warpinski et al. 2001; Vulgamore et al. 2007), and are often
because of induced tensile stresses in the near-tip field. In contrast, used to estimate fracture geometry. The stress changes generated
rock behind the tips is compressed because of the fracture opening. by opening fractures might lead to the shear failure of pre-existing
In comparing the change in rxx and ryy at the end of injection fractures. Fig. 16 shows the change of the maximum shear stress
(Figs. 14 and 15), one can note that the area of stress change for at the end of fracturing. As shown in the figure, areas representing
0 psi of the shear stress change indicate no change from the initial
state. Red colors mean an increase in the maximum shear stress,
TABLE 4—INPUT PARAMETERS FOR CASE STUDY whereas blue colors indicate a decrease. The initial maximum
shear stress is 50 psi because the differential stress (SHmax–Shmin)
Fracture Spacing 50 ft is 100 psi. A larger maximum shear stress suggests the greater
Layer Height 100 ft likelihood to induce shear failure. Fig. 16 illustrates that the large
Minimum Horizontal Stress 4,450 psi increase of the shear-stress primarily occurs around fracture tips
Maximum Horizontal Stress 4,550 psi and the region between two fractures. Accordingly, in these areas,
Fluid Leakoff coefficient 0.00001 ft/min0.5
there is a possibility to induce shear failure of pre-existing fac-
tures and generate microseismic activity. Warpinski et al. (2001)
Injection Rate 60 bbl/min
reported that considerable shear stress near fracture tips developed
n0 0.7 by hydraulic fractures provides a primary mechanism for microse-
k0 0.02 lbf-secn/ft2 isms to accurately determine the length and height of fractures,
Young’s Modulus 6.53106 psi and pore-pressure change because of fluid leakoff is an important
Poisson’s Ratio 0.2 factor in microseismic development behind the tip. Our analysis,
Number of Perforations 60 however, did not take into account fluid-leakoff effects.
Density of Slurry 1.2 g/cm3
Density of Water 1.0 g/cm3
Conclusions
Diameter of Perforations 5 mm
The FPM presented in this paper is able to simulate simultaneous
Diameter of Wellbore 0.1 m
multiple-fracture propagation. The model integrates fracture

April 2015 SPE Journal 343

ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 344 Total Pages: 10

Frac. 1 Frac. 2 Frac. 3

Frac. 4 Frac. 5 Frac. 6

500

400

300

200

100
y, ft

–100

–200

–300

–400

–500
–200 –100 0 100

x, ft
(a) Fracture trajectory (b) Fracture width distribution

Fig. 11—Fracture trajectory and aperture distribution after fracturing (map view), twofold exaggeration of x-axis scale.

interaction between multiple fractures and fluid flow in the single • Shear stress is generated during fracturing and might cause shear
fracture and the horizontal wellbore. Mass balance and pressure slippage of pre-existing fractures and develop microseismic ac-
equilibrium between multiple fractures and the horizontal well- tivity to help determine the geometry of multiple fractures.
bore are satisfied. From our model, we can conclude: • A map of the maximum shear stress change can assess the
• Simultaneous propagating fractures extend, attracting or repel- expected area of the shear failure of pre-existing fractures
ling each other as a result of stress-shadow effects, resulting in induced by hydraulic fracturing.
the direction and magnitude change of maximum and minimum
horizontal stresses.
• Under the same situation, fractures closer to the injection point Nomenclature
can obtain more fluid and propagate longer and wider. a ¼ half-length of the element, ft
• Nonplanar-geometry reduced propagation occurs for interior CL ¼ fluid-loss coefficient, ft/min0.5
fractures, causing a significant restriction of fracture aperture D ¼ diameter of wellbore, in.
and reduction of length. Ds ¼ shear displacement of the element, in.
• Compression beside the fractures increases, but is decreased at Dn ¼ normal displacement of the element, in.
fracture tips. At some points around fractures, the increment of dij ¼ distance between the center of elements i and j, ft
minimum horizontal stress is much larger than that of maxi- dp,i ¼ diameter of perforation, in.
mum horizontal stress, resulting in direction change of maxi- E ¼ Young’s modulus, psi
mum horizontal stress and nonplanar fracture geometry. E0 ¼ plane-strain modulus, psi

800
0.25 Frac. 1 Frac. 2 Frac. 3
Frac. 1 Frac. 2 700 Frac. 4 Frac. 5 Frac. 6
Frac. 3 Frac. 4
0.02 Frac. 5 Frac. 6 600
Net pressure, psi
Fracture width, in.

500
0.15
400
300
0.01
200
0.05 100
0
0 0 100 200 300 400 500
0 100 200 300 400 500
Fracture length, ft
Fracture length, ft
Fig. 13—Net pressure in each fracture changing along fracture
Fig. 12—Fracture width vs. fracture length. length.

344 April 2015 SPE Journal

ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15 Stage: Page: 345 Total Pages: 10

Change of normal stress in x-direction/psi Change of normal stress in y-direction/psi

500 300 500 300

400 400
200 200
300 300
200 200
100 100
100 100
y, ft

y, ft
0 0 0 0
–100 –100
–100 –100
–200 –200
–300 –200 –300 –200
–400 –400
–500 –300 –500 –300

–200 100 0 200 200 –200 100 0 200 200

x, ft x, ft

Fig. 14—Stress change of normal stress rxx (map view); twofold Fig. 15—Stress change of normal stress ryy (map view); twofold
exaggeration of x-axis scale. exaggeration of x-axis scale.

h ¼ fracture height, ft t ¼ injection time, minutes

i ¼ injection rate, bbl/min u
i ¼ displacement over the boundary S of the region, in.
KI ¼ stress-intensity factor of opening mode, psim0.5 uþ
i ¼ displacement over the boundary Sþ of the region, in.
KII ¼ stress intensity factor of shearing mode, psim0.5  ¼ Poisson’s ratio
k0 ¼ fluid-consistency index, lbf-secn/ft2 w ¼ fracture width, in.
Li(t) ¼ total fracture length of ith fracture at time t, ft a ¼ empirically determined constants, 2.0
N ¼ the number of fractures ar ¼ retardation coefficient
n0 ¼ fluid power-law index b ¼ empirically determined constants, 2.0
np,i ¼ the number of perforations DrI ¼ the mode I driving stress
p ¼ internal pressure, psi l ¼ fluid viscosity, cp
Pfrac ¼ pressure in the fractures, psi qs ¼ density of slurry, g/cm3
pk ¼ net pressure at the kth iteration step, psi qw ¼ density of water, g/cm3
q ¼ fluid-flow rate in the fracture, bbl/min sðsÞ ¼ time at which fracture reaches position s, minutes
Qi ¼ injection volume of each fracture, bbl/min h ¼ counterclockwise positive angle from the direction of
qL(s,t) ¼ volume rate of fluid loss to the formation per unit length in-plane propagation
of fracture, bbl/min
QT ¼ total injection rate, bbl/min References
s¼ distance along the fracture, ft
Shmax ¼ maximum horizontal stress, psi Adachi, J., Siebrits, E., Peirce, A. et al. 2007. Computer Simulation of Hy-
Shmin ¼ minimum horizontal stress, psi draulic Fractures. International J. Rock Mechanics & Mining Sci. 44:
739–757.
Carter, R.D. 1957. Appendix to Optimum Fluid Characteristics for Frac-
ture Extension. In Drill. and Prod. Prac., ed. G.C. Howard and C.R.
Change of maximum shear stress/psi Fast. API 267.
500 Castonguay, Stephen T., Mear, Mark E., Dean, Rick H. et al. 2013. Predic-
tions of the Growth of Multiple Interaction Hydraulic Fractures in
400 Three Dimensions. Presented at the SPE Annual Technical Conference
100
and Exhibition, New Orleans, Louisiana, 30 September–2 October.
300
SPE-166259-MS. http://dx.doi.org/10.2118/166259-MS.
200 50 Chong, K.K., Grieser, W.V., Passman, A. et al. 2010. A Completions
Guide Book to Shale-Play Development: A Review of Successful
100
Approaches Towards Shale-Play Stimulation in the Last Two Decades.
y, ft

0 0 Presented at the Canadian Unconventional Resources and International

Petroleum Conference, Calgary, Alberta, Canada, 19–21 October.
–100 CSUG/SPE-133874-MS. http://dx.doi.org/10.2118/133874-MS.
–200 –50 Crouch, S.L. 1976. Solution of Plane Elasticity Problems by the Displace-
ment Discontinuity Method. International J. Numerical Methods in
–300 Eng. 10: 301–343.
–400 –100 Crouch, S.L. and Starfield, A.M. 1983. Boundary Element Methods in
Solid Mechanics. London: George Allen & Unwin.
–500 Dahi_Taleghani, A. and Olson, J.E. 2011. Numerical Modeling of Multi-
–200 100 0 200 200 stranded-Hydraulic-Fracture Propagation: Accounting for the Interac-
x, ft tion Between Induced and Natural Fractures. SPE J. 16 (3): 575–581.
SPE-124884-PA. http://dx.doi.org/10.2118/124884-PA.
Fig. 16—The change of maximum shear stress (map view); two- Daneshy, A. 2011. Multistage Fracturing Using Plug-and-Perf Systems.
fold exaggeration of x-axis scale. Shale Energy/Fracturing 232 (10).

April 2015 SPE Journal 345

ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037

 J167626 DOI: 10.2118/167626-PA Date: 10-April-15
Elbel, J.L., Piggott, A.R., and Mack, M.G. 1992. Numerical Modeling of
Multilayer Fracture Treatments. Presented at the SPE Permian Basin
Oil and Gas Recovery Conference, Midland, Texas, 18–20 March.
SPE-23982-MS. http://dx.doi.org/10.2118/23982-MS.
Erdogan, F. and Sih, G.C. 1963. On the Crack Extension in Plates Under
Plane Loading and Transverse Shear. J. Basic Eng. 519–527.
Gale, Julia F.W., Reed, Robert M., and Holder, Jon. 2007. Natural Frac-
tures in the Barnett Shale and Their Importance for Hydraulic Fracture
Treatments. AAPG Bull. 91 (4): 603–622. http://dx.doi.org/10.1306/
11010606061.
Keshavarzi, R. and Mohammadi, S. 2012. A New Approach for Nu-
merical Modeling of Hydraulic Fracture Propagation in Naturally
Fractured Reservoirs. Presented at the SPE/EAGE European Un-
conventional Resources Conference and Exhibition, Vienna, Aus-
tria, 20–22 March. SPE-152509-MS. http://dx.doi.org/10.2118/
152509-MS.
Li, Yi. 1991. On Initiation and Propagation of Fractures From Deviated
Wellbores. Dissertation, The University of Texas at Austin (May
1991).
Mack, Mark G., Elbel, Jacques L., and Piggott, Andrew R. 1992. Numeri-
cal Representation of Multilayer Fracturing. Rock Mechanics, ed. Till-
erson and Wawersik. Balkema, Rotterdam. ISBN 90 5410 0451.
Meyer, B.R. and Bazan, L.W. 2011. A Discrete Fracture Network Model
for Hydraulically Induced Fractures: Theory, Parametric and Case
Studies. Presented at the SPE Hydraulic Fracturing Technology Con-
ference and Exhibition, The Woodlands, Texas, 24–26 January. SPE-
140514-MS. http://dx.doi.org/10.2118/140514-MS.
Nagel, N., Gil, I., Sanchez-Nagel, M. et al. 2011. Simulating Hydraulic
Fracturing in Real Fractured Rock—Overcoming the Limits of Pseudo
3D Models. Presented at the SPE Hydraulic Fracturing Technology
Conference and Exhibition, The Woodlands, Texas, 24–26 January.
SPE-140480-MS. http://dx.doi.org/10.2118/140480-MS.
Olson, J.E. 1990. Fracture Mechanics Analysis of Joints and Veins. Dis-
sertation, Stanford University, Stanford, California (December 1990).
Olson, J.E. 2004. Predicting Fracture Swarms—The Influence of Subcriti-
cal Crack Growth and the Crack-Tip Process Zone on Joint Spacing in
Rock. In The Initiation, Propagation, and Arrest of Joints and Other
Fractures, ed. J.W. Cosgrove and T. Engelder, 73–87. Geological So-
ciety of London Special Publication 231.
Olson, J.E. 2007. Fracture Aperture, Length and Pattern Geometry Devel-
opment Under Biaxial Loading: A Numerical Study With Applications
to Natural, Cross-Jointed Systems. In Fracture-Like Damage and
Localisation, ed. G. Couples and H. Lewis, Vol. 289, 123-142. Geo-
logical Society of London, Special Publications.
Olson, J.E. 2008. Multi-Fracture Propagation Modeling: Applications to
Hydraulic Fracturing in Shales and Tight Gas Sands. Presented at the
42nd U.S. Symposium on Rock Mechanics, San Francisco, California,
29 June–2 July.
Olson, J.E. and Wu, K. 2012. Sequential Versus Simultaneous Multi-zone
Fracturing in Horizontal Wells: Insights From a Non-planar, Multi-
frac Numerical Model. Presented at the SPE Hydraulic Fracturing
Technology Conference, The Woodlands, Texas, 6–8 February. SPE-
152602-MS. http://dx.doi.org/10.2118/152602-MS.
Olson, J.E. and Dahi Taleghani, A. 2009. Modeling Simultaneous Growth
of Multiple Hydraulic Fractures and Their Interaction With Natural
Fractures. Presented at the SPE Hydraulic Fracturing Technology Con-
ference, The Woodlands, Texas, 19–21 January. SPE-119739-MS.
http://dx.doi.org/10.2118/119739-MS.
Olson, J.E. and Pollard, D.D. 1989. Inferring Paleostresses From Natural
Fracture Patterns: A New Method. Geology 17 (4): 345–348. http://
dx.doi.org/10.1130/0091-7613.
Pollard, D.D. and Segall, P. 1987. Theoretical Displacements and Stresses
Near Fractures in Rock: With Applications to Faults, Joints, Veins,
Dikes and Solution Surfaces. In Fracture Mechanics of Rock, ed. B.K.
Atkinson, 277-350. London: Academic Press.
Roussel, Nicolas P. and Sharma, Mukul M. 2011. Strategies to Minimize
Frac Spacing and Stimulate Natural Fractures in Horizontal Comple-
tions. Presented at the SPE Annual Technical Conference and Exhibi-
tion. Denver, Colorado, 30 October–2 November. SPE-146104-MS.
http://dx.doi.org/10.2118/146104-MS.
346
ID: jaganm Time: 11:59 I Path: S:/3B2/J###/Vol00000/140037/APPFile/SA-J###140037
Stage: Page: 346 Total Pages: 10
Schlichting, H. 1968. Boundary-Layer Theory, sixth edition. New York:
McGraw-Hill.
Sheibani, Farrokh and Olson, Jon. 2013. Stress Intensity Factor Determi-
nation for Three-Dimensional Crack Using the Displacement Disconti-
nuity Method With Applications to Hydraulic Fracture Height Growth
and Non-Planar Propagation Paths. In Effective and Sustainable Hy-
draulic Fracturing, ed. A.P. Bunger, J. McLennan, and R. Jeffrey,
Chap. 37. Rijeka, Croatia: InTech. http://dx.doi.org/10.5772/56308.
Shou, K.J. 1993. A High Order Three-Dimensional Displacement Discon-
tinuity Method With Application to Bonded Half-Space Problems. PhD
thesis, University of Minnesota, Minnesota (August 1993).
Siriwardane, H.J. and Layne, A.W. 1991. Improved Model for Predicting
Multiple Hydraulic Fracture Propagation From a Horizontal Well. Pre-
sented at the SPE Eastern Regional Meeting, Lexington, Kentucky,
22–25 October. SPE-23448-MS. http://dx.doi.org/10.2118/23448-MS.
Spence, D.A. and Sharp, P.W. 1985. Self-Similar Solution for Elastohy-
drodynamic Cavity Flow. Proc. of the Royal Society of London, Series
A (400), pp. 289–313.
Valko, P. and Economides, M.J. 1995. Hydraulic Fracture Mechanics.
New York: John Wiley & Sons.
Vulgamore, T., Clawson, T., Pope, C. et al. 2007. Applying Hydraulic
Fracture Diagnostics to Optimize Stimulations in the Woodford Shale.
Presented at the SPE Annual Technical Conference and Exhibition,
Anaheim, California, 11–14 November. SPE-110029-MS. http://
dx.doi.org/10.2118/110029-MS.
Warpinski, N.R. and Branagan, P.T. 1989. Altered-Stress Fracturing. J Pet
Technol 41 (9): 990–997. SPE-17533-PA. http://dx.doi.org/10.2118/
17533-PA.
Warpinski, N.R., Moschovidis, Z.A., Parker, C.D. et al. 1994. Comparison
Study of Hydraulic Fracturing Models—Test Case: GRI Staged Field
Experiment No. 3. SPE Prod & Fac 9 (1): 7–16. SPE-28158-PA.
Warpinski, N.R., Wolhart, S.L., and Wright, C.A. 2001. Analysis and Pre-
diction of Microseismicity Induced by Hydraulic Fracturing. Presented
at the SPE Annual Technical Conference and Exhibition, New Orle-
ans, Louisiana, 30 September–3 October. SPE J. 9 (1): 24–33. SPE-
87673-PA. http://dx.doi.org/10.2118/87673-PA.
Weng, X., Kresse, O., Cohen, C. et al. 2011. Modeling of Hydraulic Frac-
ture Network Propagation in a Naturally Fractured Formation. Pre-
sented at the SPE Hydraulic Fracturing Technical Conference and
Exhibition, The Woodlands, Texas, 24–26 January. SPE-140253-MS.
http://dx.doi.org/10.2118/140253-MS.
Wong, Sau-Wai and Xu, Guanshui. 2013. Interaction of Multiple Hydraulic
Fractures in Horizontal Wells. Presented at the SPE Middle East Uncon-
ventional Gas Conference and Exhibition, Muscat, Oman, 28–30 Janu-
ary. SPE-163982-MS. http://dx.doi.org/10.2118/163982-MS.
Wu, R., Kresse, O., Weng, X. et al. 2012. Modeling of Interaction of Hy-
draulic Fractures in Complex Fracture Networks. Presented at the SPE
Hydraulic Fracture Technology Conference, The Woodlands, Texas,
6–8 February. SPE-152052-MS. http://dx.doi.org/10.2118/152052-MS.
Zimmerman, R.W., Kumar, S., and Bodvarsson, G.S. 1991. Lubrication
Theory Analysis of the Permeability of Rough-Walled Fractures. Inter-
national J. Rock Mechanics Mine Sci. & Geomechanics. Abstr. 28 (4):
325–331.
Kan Wu is a PhD candidate in the Department of Petroleum and
Geosystems Engineering at the University of Texas at Austin (UT).
Her research is focused on developing a fracture-propagation
model to simulate multiple fracture propagation in naturally
fractured reservoirs. Wu holds MS and BS degrees in petroleum
engineering from the China University of Petroleum. She is a
member of SPE. Wu won the First Place in the PhD Division at the
2013 SPE Gulf Coast Region Paper Contest and Second Place in
the PhD Division at the 2013 SPE Local Paper Contest.
Jon E. Olson is an associate professor in petroleum and geosys-
tems engineering at UT. He is co-principal investigator of the
Fracture Research and Application Consortium at UT, and spe-
cializes in rock-mechanics applications to structural geology,
hydraulic fracturing, and reservoir characterization. Before join-
ing UT in 1995, Olson spent 6 years as a research engineer at
Mobil Research and Development Corporation. He holds a
PhD degree in geomechanics from Stanford University.
April 2015 SPE Journal
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
A Generalized Framework Model for the
Simulation of Gas Production in
Unconventional Gas Reservoirs
Yu-Shu Wu, SPE, Colorado School of Mines; Jianfang Li, SPE, Research Institute of Petroleum Exploration and
Development; Didier-Yu Ding, SPE, IFP Energies Nouvelles; Cong Wang, SPE, Colorado School of Mines; and
Yuan Di, Peking University
Summary
Unconventional gas resources from tight-sand and shale gas reser-
voirs have received great attention in the past decade around the
world because of their large reserves and technical advances in
developing these resources. As a result of improved horizontal-
drilling and hydraulic-fracturing technologies, progress is being
made toward commercial gas production from such reservoirs, as
demonstrated in the US. However, understandings and technolo-
gies needed for the effective development of unconventional res-
ervoirs are far behind the industry needs (e.g., gas-recovery rates
from those unconventional resources remain very low). There are
some efforts in the literature on how to model gas flow in shale
gas reservoirs by use of various approaches—from modified com-
mercial simulators to simplified analytical solutions—leading to
limited success. Compared with conventional reservoirs, gas flow
in ultralow-permeability unconventional reservoirs is subject to
more nonlinear, coupled processes, including nonlinear adsorption/
desorption, non-Darcy flow (at both high flow rate and low flow
rate), strong rock/fluid interaction, and rock deformation within
nanopores or microfractures, coexisting with complex flow geome-
try and multiscaled heterogeneity. Therefore, quantifying flow in
unconventional gas reservoirs has been a significant challenge, and the
traditional representative-elementary-volume- (REV) based Darcy’s
law, for example, may not be generally applicable.
In this paper, we discuss a generalized mathematical framework
model and numerical approach for unconventional-gas-reservoir
simulation. We present a unified framework model able to incorpo-
rate known mechanisms and processes for two-phase gas flow and
transport in shale gas or tight gas formations. The model and numer-
ical scheme are based on generalized flow models with unstructured
grids. We discuss the numerical implementation of the mathematical
model and show results of our model-verification effort. Specifically,
we discuss a multidomain, multicontinuum concept for handling
multiscaled heterogeneity and fractures [i.e., the use of hybrid mod-
eling approaches to describe different types and scales of fractures
or heterogeneous pores—from the explicit modeling of hydraulic
fractures and the fracture network in stimulated reservoir volume
(SRV) to distributed natural fractures, microfractures, and tight ma-
trix]. We demonstrate model application to quantify hydraulic frac-
tures and transient flow behavior in shale gas reservoirs.
Introduction
Even with the significant progress made in producing natural gas
from unconventional, low-permeability shale gas and tight gas
reservoirs in the past decade, gas recovery remains very low (esti-
mated at 10 to 30% of gas in place). Gas production or flow in
such extremely low-permeability formations is complicated fur-
ther by many coexisting processes, such as severe heterogeneity,
a large Klinkenberg effect (Klinkenberg 1941), nonlinear or non-
Copyright V
C 2014 Society of Petroleum Engineers
This paper (SPE 163609) was accepted for presentation at the SPE Reservoir Simulation
Symposium, The Woodlands, Texas, USA, 18–20 February 2013, and revised for
publication. Original manuscript received for review 4 March 2013. Revised manuscript
received for review 13 January 2014. Paper peer approved 5 February 2014.
October 2014 SPE Journal
ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 845 Total Pages: 13
Darcy flow behavior, adsorption/desorption, strong interactions
between fluid (gas and water) molecules, and solid materials
within tiny pores, as well as micro- and macrofractures of shale
and tight formations. Currently, there is little in basic understand-
ing of how these complicated flow behaviors impact gas flow and
the ultimate gas recovery in such reservoirs. In particular, only a
few effective reservoir simulators and few modeling studies cur-
rently are available (e.g., Kelkar and Atiq 2010) in the industry
for assisting reservoir engineers to model and develop the uncon-
ventional natural-gas resources.
Shale formations are characterized by extremely low perme-
ability from subnanodarcies to microdarcies, and it is different for
different shale types, even under similar porosity, stress, or pore
pressure. As summarized by Wang et al. (2009), the permeability
of deep organic-lean mudrocks ranges from smaller than to tens
of nanodarcies, whereas permeability values in organic-rich gas
shales range from subnanodarcies to tens of microdarcies. The
Klinkenberg effect (Klinkenberg 1941), or gas-slippage effect,
has been practically ignored in conventional gas reservoir studies,
except when analyzing pressure responses or flow near gas-pro-
duction wells at a very low pressure. This is because of a larger
pore size and relatively high pressure existing in those conven-
tional gas reservoirs. In shale gas reservoirs, however, the Klin-
kenberg or slippage effect is expected to be significant because of
the nanosized pores of such rock, even under a high-pressure con-
dition. Wang et al. (2009) show that gas permeability in the Mar-
cellus shale increases from 19.6 md at 1,000 psi to 54 md at 80 psi
because of the strong slippage effect.
Unconventional reservoir dynamics are characterized by the
highly nonlinear behavior of multiphase flow in extremely low-
permeability rock, coupled by many coexisting physical processes
(e.g., non-Darcy flow). Because of complicated flow behavior, a
strong interaction between fluid and rock, and multiscaled hetero-
geneity, the traditional Darcy’s-law/REV-based model may not
be generally applicable for describing flow phenomena in uncon-
ventional gas reservoirs. Blasingame (2008) and Moridis et al.
(2010) provide very comprehensive reviews of flow mechanisms
in unconventional shale gas reservoirs. Both studies point out that
the nonlaminar/non-Darcy flow concept of high velocity may turn
out to be important in shale gas production. The nonlaminar/non-
Darcy flow concept of high-velocity flow in shale gas reservoirs
may not be represented by Darcy’s law, and the Forchheimer
equation is probably sufficient for many applications.
Natural gas in shale gas formations is present both as a free-gas
phase and as gas adsorbed onto solids in pores. In these reservoirs,
gas or methane molecules are adsorbed mainly to the carbon-rich
components (i.e., kerogen) (Silin and Kneafsey 2011; Mengal and
Wattenbarger 2011; EIA 2011). The adsorbed gas represents a sig-
nificant percentage of total gas reserves (20 to 80%) as well as a sig-
nificant factor in recovery rates, which cannot be ignored in any
model or modeling analysis. In shale gas formations, past studies
found that methane molecules are adsorbed mainly to the carbon-
rich components (i.e., kerogen), correlated with total organic con-
tent (TOC) in shales, as a function of reservoir pressure.
In conventional oil or gas reservoirs, the effect of geome-
chanics on rock deformation or permeability is generally small
845
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
and has been mostly ignored in practice. However, in unconven-
tional shale formations with nanosized pores or nanosized micro-
fractures, such geomechanics effects can be relatively large and
may have a significant impact on both fracture and matrix perme-
ability, which has to be considered, in general. Wang et al. (2009)
show that permeability in the Marcellus shale is pressure-depend-
ent and decreases with an increase in the confining pressure (or
total stress). The effect of confining pressure on permeability is
caused by a reduction of porosity. Bustin et al. (2008) report the
effect of stress (confining pressure) in Barnett, Muskwa, Ohio,
and Woodford shales and show that the degree of permeability
reduction with confining pressure is significantly higher in shales
than in consolidated sandstone or carbonate.
This paper presents a generalized mathematical model and nu-
merical approach for unconventional-gas-reservoir simulation.
We present a unified framework model that is able to incorporate
many known mechanisms and processes for two-phase gas flow in
shale gas or tight gas formations with a continuum modeling
approach. The numerical scheme is based on generalized flow
models with unstructured grids. We discuss the numerical imple-
mentation of Klinkenberg effects, non-Darcy flow, gas adsorption,
and geomechanics effects into the mathematical model. In the nu-
merical-modeling examples, we apply geomechanics-coupled per-
meability to fractures, and apply the Klinkenberg effect to matrix
media. This is because the Klinkenberg effect is apparent with
extremely low permeability or tiny pores, whereas fractures are
very sensitive to the change in effective stress. Results of our
model-verification effort are also presented. We demonstrate
model application to quantify hydraulic fractures and transient
flow behavior in shale gas reservoirs.
One of the critical issues in shale-gas-reservoir simulation is
how to handle fracture flow and fracture/matrix interaction. This
is because the gas flow and production rely on fractures in these
reservoirs. Cipolla et al. (2009) built a methodology on modeling
complex fracture geometry and heterogeneity from the microseis-
mic data. In this paper, we present a hybrid fracture-modeling
approach, defined as a combination of explicit fracture modeling
and multicontinuum, multiple-interacting-continua (MINC) (Pruess
and Narasimhan 1985), and single-porosity modeling approaches,
which seems the best option for modeling a shale gas reservoir
with both hydraulic fractures and natural fractures. This is because
hydraulic fractures, which have to be dealt with for shale gas pro-
duction, are better handled by the explicit fracture method but can-
not be modeled, in general, by a dual-continuum model. On the
other hand, naturally fractured reservoirs are better modeled by a
dual-continuum approach, such MINC, for extremely low-perme-
ability matrix in shale gas formations, which cannot be modeled by
an explicit fracture model. Specifically, we demonstrate how to use
the hybrid modeling approach to describe different types and scales
of fractures from the explicit modeling of hydraulic fractures and
fracture network in the SRV to distributed natural fractures, micro-
fractures, and tight matrix.
Flow-Governing Equations
In most cases of gas production from shale gas formations, a two-
phase (gas/liquid) -flow model or a multiphase-flow model is con-
sidered to be sufficient for simulation studies. This is because
what we are most concerned with in shale-gas-reservoir simula-
tion is the modeling of gas flow from reservoir to well. However,
in addition to the gas phase, liquid-phase flow is often occurring
simultaneously with gas flow; it needs to be considered when two
cases exist—mobile in-situ connate water and an abundance of
aqueous hydraulic-fracturing fluids, which are sucked into the for-
mations surrounding the wells. Therefore, in this paper, we pri-
marily discuss the two-phase (gas and liquid) -flow model and
formulation and treat single-phase gas flow as a special case of
the two-phase flow for the simulation studies of unconventional
gas reservoirs.
A multiphase system of gas and water (or liquid) in a porous
or fractured unconventional reservoir is assumed to be similar to
what is described in a black-oil model, composed of two phases:
846
ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 846 Total Pages: 13
gaseous and aqueous. For simplicity, the gas and water compo-
nents are assumed to be present only in their associated phases
and adsorbed gas is within the solid phase of rock. Each fluid
phase flows in response to pressure and gravitational and capillary
forces according to the multiphase extension of Darcy’s law or
several extended non-Darcy-flow laws, discussed next. In an iso-
thermal system containing two mass components, subject to mul-
tiphase flow and adsorption, two mass-balance equations are
needed to fully describe the system, as described in an arbitrary
flow region of a porous or fractured domain for flow of phase b
(b ¼ g for gas and b ¼ w for water),
@
ð/Sb qb þ mb Þ ¼ r  ðqb vb Þ þ qb : . . . . . . . . . . . . ð1Þ
@t
where / is the effective porosity of the porous or fractured media;
Sb is the saturation of fluid b; qb is the density of fluid b; vb is the
volumetric velocity vector of fluid b, determined by Darcy’s law
or non-Darcy-flow models, discussed next; t is time; mb is the
adsorption or desorption mass term for the gas component per
unit volume of formation; and qb is the sink/source term of phase
(component) b per unit volume of formation.
Incorporation of Gas Adsorption and Desorption. The amount
of adsorbed gas in a given shale gas formation is generally
described with the Langmuir’s isotherm (e.g., Moridis et al. 2010;
Mengal and Wattenbarger 2011; Silin and Kneafsey 2011; EIA
2011; Wu et al. 2012; Wu and Wang 2012) (i.e., it is correlated to
reservoir gas pressure). To incorporate the gas adsorption or de-
sorption mass term in the mass-conservation equation, the amount
of adsorbed gas is determined according to the Langmuir’s iso-
therm as a function of reservoir pressure. As the pressure
decreases with continuous gas production through production
wells in reservoirs, more adsorbed gas is released from the solid
to the free-gas phase in the pressure-lowering region, contributing
to the total gas flow or production. In our model, the mass of
adsorbed gas in unit formation volume is described (Leahy-Dios
et al. 2011; Silin and Kneafsey 2011; Wu et al. 2012) as
mg ¼ qR qg VE ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
where mg is adsorbed gas mass in unit formation volume; qR is
rock bulk density; qg is gas density at standard condition; and VE
is the adsorption isotherm function or gas content in scf/ton
(or standard gas volume adsorbed per unit rock mass). If the
adsorbed-gas terms can be represented by the Langmuir isotherm
(Langmuir 1916), the dependency of adsorbed-gas volume on
pressure at constant temperature is given as
P
VE ¼ VL ; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ
P þ PL
where VL is the Langmuir’s volume in scf/ton; P is reservoir gas
pressure; and PL is Langmuir’s pressure, the pressure at which
50% of the gas is desorbed. In general, Langmuir’s volume VL is
a function of the organic richness (or TOC) and thermal maturity
of the shale.
Note that Eq. 3 is valid only for the case when the Langmuir
model is applicable. In general, VE in Eq. 2 can be determined
from any correlation of gas adsorption as a function of reservoir
gas pressure, which may be defined by a table lookup from labora-
tory studies for a given unconventional reservoir.
In the literature, the most commonly used empirical model
describing sorption onto organic carbon in shales is analogous to
that used in coalbed methane and follows the Langmuir isotherm
(Gao et al. 1994; Moridis et al. 2010), such as Eq. 2. This adsorp-
tion-modeling approach is based on the assumption that an instan-
taneous equilibrium exists between the sorbed and the free gas
(i.e., there is no transient-time lag between pressure changes and
the corresponding sorption/desorption responses; i.e., the equilib-
rium model of the Langmuir sorption is assumed to be valid,
which provides a good approximation in shale gas modeling).
October 2014 SPE Journal
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
1E–01
Gas Permeability, md
1E–02
1 microdarcy
1E–03
1E–04 Barnett Muskwa
Ohio Marcellus
Woodford-1 Woodford-2
Huron
1E–05
0 1600 3200 4800
Effective Stress, psi
Fig. 1—Effect of confining pressure on gas permeability in gas
shales (Wang et al. 2009).
Several kinetic sorption models exist in the literature that use dif-
fusion approaches; however, the subject has not been fully inves-
tigated or fully understood (Moridis et al. 2010).
Coupled Flow and Geomechanics Effect. In this section, we
will propose a simple-to-implement modeling approach, easy to
incorporate into an existing reservoir simulator, to couple geome-
chanics with two-phase flow in unconventional reservoirs. The
following discussion is based on our previous work (e.g., Wu
et al. 2008; Winterfeld and Wu 2011). The effective porosity, per-
meability, and capillary pressure of rock are assumed to correlate
with the mean effective stress (r0m ), defined as
r0m ¼ rðx; y; z; PÞ  aP; . . . . . . . . . . . . . . . . . . . . . . ð4Þ
where a is the Biot constant and
rm ðx; y; z; PÞ ¼ ½rx ðx; y; z; PÞ þ ry ðx; y; z; PÞ þ rz ðx; y; z; PÞ=3;
                   ð5Þ
where rx, ry, and rz are total stress in the x-, y-, and z-direction,
respectively. With the definition of the mean effective stress in
Eq. 5, the effective porosity of the formation (fractures or porous
media) is defined as a function of mean effective stress only,
/ ¼ /ðr0m Þ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð6Þ
Similarly, the intrinsic permeability is related to the effective
stress; that is,
k ¼ kðr0m Þ: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð7Þ
For capillary pressure functions, the impact of rock deforma-
tion or pore change is accounted for with the Leverett function
(Leverett 1941),
qffiffiffiffiffiffiffiffiffiffiffiffi
k0 =/0
Pc ¼ Cp P0c ðSw Þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; . . . . . . . . . . . . . . . . ð8Þ
kðr0m Þ=/ðr0m Þ
where Pc is the capillary pressure between gas and water as a
function of water or gas saturation; Cp is a constant; and the
superscript 0 denotes reference or zero-stress condition.
Several correlations have been used for porosity as a function
of effective stress and permeability as a function of porosity (Da-
vies and Davies 1999; Rutqvist et al. 2002; Winterfeld and Wu
2011, 2012). In our numerical implementation, the function for
porosity and permeability presented by Rutqvist et al. (2002) is
adopted, which is obtained from laboratory experiments on sedi-
mentary rock (Davies and Davies 1999),
0
/ ¼ /r þ ð/0  /r Þear ; . . . . . . . . . . . . . . . . . . . . . ð9Þ
October 2014 SPE Journal
ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 847 Total Pages: 13
where /0 is zero-effective-stress porosity; /r is high-effective-
stress porosity; and the exponent a is a parameter. Rutqvist et al.
(2002), Davies and Davies (1999), and Winterfeld and Wu (2011,
2012) also present an associated function for permeability in
terms of porosity,
 
/
c 1
k ¼ ko e /0 ; . . . . . . . . . . . . . . . . . . . . . . . . . . ð10Þ
where c is a parameter. Fig. 1 shows the effect of confining pres-
sure on gas permeability in gas shales.
One can use an alternative, table-lookup approach for the cor-
relation of reservoir porosity and permeability as a function of
effective mean stress, from laboratory studies, for a given uncon-
6400 ventional reservoir.
One must note that Fig. 1 from Wang et al. (2009) presents the
permeability measurement from core plugs in which potential natu-
ral microfractures in core plugs play an important role for the con-
nectivity. If one uses crushed samples to measure the matrix
permeability only by eliminating natural and drilling induced
microfactures, the permeability value is one or two orders lower.
The geomechanics has a much stronger impact on the fracture than
on the matrix. So, when using a dual-porosity approach in the mod-
eling, if microfractures are considered as a part of the matrix media,
one can directly apply the relations in Eqs. 4 through 10. However,
if microfractures are considered as a part of the fractured media,
the geomechanics effect is more complex because fracture conduc-
tivities are subjected to different laws according to microfractures,
partially propped fractures, or propped fractures (Cipolla 2009).
The applicability of these mechanics-coupling models in mul-
tiphase-flow simulations for a rock-deformation effect requires
that the initial distribution of effective stress or total stress field be
predetermined as a function of spatial coordinates and pressure
fields, as in Eq. 5. In practice, the stress distribution may be esti-
mated analytically, numerically, or from field measurements
because changes in effective stress are primarily caused by changes
in reservoir pressure during production. These models can be sig-
nificantly simplified for coupling multiphase gas flow with rock de-
formation in stress-sensitive formations in numerical simulation, if
the in-situ total stress in reservoirs is nearly constant or a function
of spatial coordinates as well as fluid pressure only during the
production. The constant-total-stress requirement may be approxi-
mately satisfied for deep reservoirs.
Incorporation of Klinkenberg or Gas-Slippage Effect. In low-
permeability shale gas formations with nanosized pores or under a
low-reservoir-pressure condition, the Klinkenberg effect (Klin-
kenberg 1941) may be significant and should be accounted for
when modeling gas flow in such reservoirs (Wu et al. 1998; Wang
et al. 2009). As discussed previously, the Klinkenberg effect is
expected to be larger or stronger in unconventional reservoirs
because of small pore size and low permeability in comparison
with those in conventional reservoirs. The Klinkenberg effect, if
existing, will enhance gas permeability or productivity in a low-
pressure zone, such as the region near a well or matrix portions
near fractures, of low-permeability unconventional formations,
and, therefore, it should be included as an additional beneficial
factor of gas-flow enhancement.
The Klinkenberg effect is incorporated in gas-flow models by
modifying absolute permeability for the gas phase as a function of
gas pressure (e.g., Wu et al. 1998),
 
b
kg ¼ k1 1 þ ; . . . . . . . . . . . . . . . . . . . . . . . . ð11Þ
Pg
where k1 is constant, absolute gas-phase permeability under very
large gas-phase pressure (in which the Klinkenberg effect is mini-
mized); and b is the Klinkenberg beta factor and could be pressure-
or temperature-dependent, accounting for the gas-slippage effect.
In a conventional-gas-reservoir simulation, the beta factor is
commonly treated as constant and depends on the pore structure
of the medium and formation temperature for a particular reser-
voir. Several recent studies on dynamic gas slippage with
847
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 848 Total Pages: 13

60 60

Permeability, µD

Permeability, µD
40 40

20 20

0 0
0 1000 2000 3000 0 0.005 0.01 0.015
Pore Pressure, psia Reciprocal of Pore Pressure, 1/psia

Fig. 2—Contribution of the Klinkenberg effect to the apparent matrix permeability (Ozkan et al. 2010).

microscale or pore-scale models have considered the beta factor
as a function of gas pressure or the Knudsen number. In applica-
tion, the Klinkenberg effect should be modeled with the labora-
tory-determined beta factor either as a constant or as a pressure-
dependent function or simply treating the apparent gas permeabil-
ity as a function of pressure from a table lookup to include the
Klinkenberg effect or the Knudsen diffusion. An example relation
between permeability and pressure, as shown in Fig. 2, can be
directly used for the reservoirs concerned, if a site-specific study
provides such correlations or plots.
After a comparison of Fig. 1 and Fig. 3, the Klinkenberg effect
seems to have less impact than that of geomechanics, and they are
going in the opposite directions. Geomechanics has an effect
mainly on the microfractures and stimulation fractures, whereas
the Klinkenberg effect is primarily on the matrix media with
nanosized pores or fractures.
A dynamic gas-slippage theory was proposed on the basis of
the assumption that gas travels under the influence of a concentra-
tion field (random molecular flow) and a pressure field (macro-
scopic flow) (Ertekin et al. 1986). According to this theory, the
Klinkenberg factor is not a constant anymore, but a pressure-de-
pendent value. They gave the expression of the Klinkenberg fac-
tor as
We analyze the Klinkenberg effect with three different matrix
permeabilities—1:0  103 md, 1:0  105 md; and 1:0  107
md; as shown in Fig. 2 and Fig. 4. We can see that the contribu-
tion of the Klinkenberg effect is more significant at low pressures
and for lower values of permeability. This estimation also pro-
vides reliable values of the beta factor for analyzing the Klinken-
berg effect.
The Incorporation of NonDarcy Gas Flow. In addition to mul-
tiphase Darcy flow, non-Darcy flow may also occur between and
among the continua, such as along fractures, in unconventional
gas reservoirs. The flow velocity, vß, for the non-Darcy flow of
each fluid may be described with the multiphase extension of the
Forchheimer equation (e.g., Wu 2002),
lb
ðrUb Þ ¼ vb þ bb qb vb jvb j; . . . . . . . . . . . . . ð14Þ
k  krb
where ßß is presented the effective non-Darcy-flow coefficient
with a unit m–1 for fluid b under multiphase-flow conditions. The
correlation proposed by Evans and Civan is used to determine the
non-Darcy-flow beta factor in the Forchheimer equation (Evans
and Civan 1994) in our simulation examples, such as

Pcg Dg l 1:485  109

b¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð12Þ b¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . ð15Þ
kg k1:021

where cg is gas compressibility and D is diffusivity coefficient. where the unit of k is md and the unit of b is ft1 . This correlation
In Eq. 12, the correlation to compute the diffusivity constant is for b matched with more than 180 data points, including those for
given by Ertekin et al. (1986): propped fractures (correlation coefficient ¼ 0.974).

31:57 Numerical Model

Dg ¼ pffiffiffiffiffiffi k0:67 ; . . . . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
Mg As discussed previously, the partial-differential equation that gov-
erns gas and liquid flow in shale gas reservoirs is nonlinear. In
where Dg is in ft2 =D. addition, gas flow in unconventional reservoirs is subject to many

1.0E–2
3200
1.0E–3 km = 1E–7 md
km = 1E–3 md
1600
Permeability, md

1.0E–4
b factore, psi

1.0E–5 800
km = 1E–5 md km = 1E–5 md
1.0E–6 400

1.0E–7 200
km = 1E–7 md
1.0E–8 km = 1E–3 md
100
300.0 600.0 1200.0 2400.0 4800.0
300.0 600.0 1200.0 2400.0 4800.0
Pressure, psi
Pressure, psi
Fig. 3—Effect of pore pressure on gas permeability in the Mar-
cellus shale, with a confining pressure of 3,000 psi (Soeder Fig. 4—Estimations of Klinkenberg beta factor for three perme-
1988; Wang et al. 2009). ability values.

848 October 2014 SPE Journal

ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
other nonlinear flow processes, such as adsorption and non-Darcy
flow. In general, the flow model needs to be solved with a numeri-
cal approach. This work follows the methodology for reservoir
simulation (i.e., the use of numerical approaches to simulate gas
and water flow), following three steps: (1) spatial discretization of
mass-conservation equations; (2) time discretization; and (3) iter-
ative approaches to solve the resulting nonlinear, discrete alge-
braic equations.
Discrete Equations. The component mass-balance equations
(Eq. 1) are discretized in space with a control-volume or inte-
grated finite-difference concept (Pruess et at. 1999). The control-
volume approach provides a general spatial discretization scheme
that can represent a 1D, 2D, or 3D domain with a set of discrete
meshes. Each mesh has a certain control volume for a proper
averaging or interpolation of flow and transport properties or ther-
modynamic variables. Time discretization is carried out with a
backward, first-order, fully implicit finite-difference scheme. The
discrete nonlinear equations for components of gas and water at
gridblock or node i can be written in a general form:
h i
b;n Vi
ð/qSÞb;nþ1
i þ mb;nþ1
i  ð/qSÞb;n
i  mi
X Dt
¼ flowb;nþ1
ij þ Qb;nþ1
i 2ðkbb Þijþ1=2 >
j 2 gi
ðb ¼ gas and liquidÞ and ði¼ 1; 2; 3;…;NÞ;
                   ð16Þ
where superscript b serves also as an equation index for gas and
water components, with b ¼ 1 (gas) and b ¼ 2 (water); superscript
n denotes the previous time level, with n þ 1 as the current time
level to be solved; subscript i refers to the index of gridblock or
node i, with N as the total number of nodes in the grid; Dt is time-
step size; Vi is the volume of node i; gi contains the set of direct
neighboring nodes ( j) of node i; and mki , flowkij , and Qki are the
adsorption or desorption, the component mass “flow” term
between nodes i and j, and sink/source term at node i for compo-
nent k, respectively.
Eq. 16 presents a precise form of the balance equation for each
mass component of gas and water in a discrete form. It states that
the rate of change in mass accumulation (plus adsorption or desorp-
tion, if existing) at a node over a timestep is exactly balanced by an
inflow/outflow of mass and also by sink/source terms, when exist-
ing for the node. As long as all flow terms have the flow from node
i to node j equal to and opposite to that of node j to node i for fluids,
no mass will be lost or created in the formulation during the solu-
tion. Therefore, the discretization in Eq. 16 is conservative.
The “flow” terms in Eq. 16 are mass fluxes by advective proc-
esses and are described, when Darcy’s law is applicable, by a dis-
crete version of Darcy’s law; that is, the mass flux of fluid phase b
along the connection is given by
flowbij ¼ kb;ijþ1=2 cij ðUbj  Ubi Þ; . . . . . . . . . . . . . . . . ð17Þ
where kb,i jþ1/2 is the mobility term to phase b, defined as
! commercial reservoir simulators, developed for conventional-
qb krb
kb;ijþ1=2 ¼ : . . . . . . . . . . . . . . . . . . . . ð18Þ
lb
ijþ1=2
In Eq. 17, cij is transmissivity and is defined, for a Voronoi
grid, as (Pruess et al. 1999)
Aij kijþ1=2
cij ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð19Þ
Di þ D j
where Aij is the common interface area between the connected
blocks or nodes i and j; Di is the distance from the center of block
i to the common interface of blocks i and j; and kijþ1/2 is an aver-
aged (such as harmonic-weighted) absolute permeability along
the connection between elements i and j.
October 2014 SPE Journal
ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 849 Total Pages: 13
In this numerical approach, we apply the upstream weighting
method to the mobility term and the harmonic mean method to
the transmissivity term to guarantee the convergence and accu-
racy of the calculation. The flow-potential term in Eq. 17 is
defined as
Ub j ¼ Pbi  qb;ijþ1=2 gZi ; . . . . . . . . . . . . . . . . . . . . . ð20Þ
where Zi is the depth to the center of block i from a reference
datum.
Handling the Klinkenberg Effect. To include the Klinkenberg
effect on gas flow, the absolute permeability to gas phase in Eq.
19 should be evaluated with Eq. 11 as a function of gas-phase
pressure.
Handling the Non-Darcy Flow. Under the non-Darcy-flow con-
dition of Eq. 14, the flow term (flowbij ) in Eq. 17 along the con-
nection (i, j), between elements i and j, is numerically defined as
(Wu 2002)
8
2 !2 31=2 9
>
< >
=
Aij 1 1
flowb;ij ¼  þ4  c ij ðUb j  Ub i Þ5 ;
: kb kb >
;
                   ð21Þ
in which the non-Darcy-flow transmissivity is defined as
4ðk2 qb bb Þijþ1=2
c ij ¼ : . . . . . . . . . . . . . . . . . . . . . . . ð22Þ
Di þ D j
In evaluating the “flow” terms in Eqs.17 through 22, subscript
ij þ 1/2 is used to denote a proper averaging or weighting of
fluid-flow properties at the interface or along the connection
between two blocks or nodes i and j. For example, we use
upstream weighting for relative permeability, density, and non-
Darcy coefficient. The convention for the signs of flow terms is
that flow from node j into node i is defined as “þ” (positive) in
calculating the flow terms.
Handling Fractured Media. Handling flow through fractured
media is critical in shale-gas-reservoir simulation, because gas
production from such low-permeability formations relies on frac-
tures, from hydraulic-fracture networks to various-scaled natural
fractures, to provide flow channels for gas flow into producing
wells. Therefore, any unconventional reservoir simulator must
have the capability of handling fractured media. The published
modeling exercises in the literature have paid much attention to
modeling fractures in shale gas formations (e.g., Cipolla 2009;
Freeman et al. 2009a,2009b; 2010; Moridis et al. 2010; Cipolla
et al. 2010; Rubin 2010; Li et al. 2011; Wu et al. 2012). However,
note that very few studies have been carried out to address the
critical issues as how to accurately simulate fractured unconven-
tional gas reservoirs or to select the best approach for modeling a
given shale gas formation. Most of the modeling exercises use
fractured-reservoir simulation, which have very limited capabil-
ities for modeling multiscaled or complicated fractured reservoirs.
On the other hand, to simulate fractured unconventional gas reser-
voirs, more effort on model development is needed—from new
conceptual models to in-depth modeling studies of laboratory to
field-scale application.
In our opinion, the hybrid fracture-modeling approach—
defined as a combination of explicit fracture modeling (discrete-
fracture model) and MINC (Pruess and Narasimham 1985; Pruess
1983) and single-porosity modeling approaches—provides the
best option for modeling a shale gas reservoir with both hydraulic
fractures and natural fractures. This is because hydraulic fractures,
which have to be dealt with for shale gas production, are better
handled by the explicit fracture method, and they cannot be
849
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
Fractures
Matrix Blocks
Fig. 5—Schematic of MINC concept (Pruess and Narasimham
1985).
modeled, in general, by a dual-continuum model. On the other
hand, naturally fractured reservoirs are better modeled by a dual-
continuum approach, such as MINC, for extremely low-perme-
ability matrix in shale gas formations, which cannot be modeled
by an explicit-fracture model.
An explicit-fracture-modeling, or discrete-fracture, concept is
to include every fracture explicitly in the modeled system by the
use of refined grids to discretize fractures and the matrix surround-
ing fractures. This approach is a good option for simulating hy-
draulic fractures for gas production from hydraulic-fractured wells
in a nonfractured/shale gas reservoir. The advantage of this
approach is that it can model hydraulic fractures accurately when
the fractures are known for their spatial distributions, determined
from other fracture-characterization studies. The disadvantage is
that it cannot be used for simulating natural fractures or microfrac-
tures, in general, because the number of natural fractures or micro-
fractures in a shale gas reservoir is too large for the model to
handle and their actual distributions in formations are unknown.
For the low matrix permeability or large matrix-block size, the
traditional double-porosity model may not be applicable for mod-
eling natural fractures in unconventional reservoirs. This is
because it takes years to reach the pseudosteady state under which
the double-porosity model applies. The MINC concept (Pruess
and Narasimham 1985) is able to describe gradients of pressures,
temperatures, or concentrations near the matrix surface and inside
the matrix—by further subdividing individual matrix blocks with
1D or multidimensional strings of nested meshes, as shown in
Hydraulic fracture geometry
Microseismic
cloud
Hydraulic fracture and SRV
Fig. 6—Hybrid fracture model built methodology from microseismic cloud.
850
ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 850 Total Pages: 13
Fig. 5. Therefore, the MINC method treats interporosity flow in a
fully transient manner by computing the gradients that drive inter-
porosity flow at the matrix/fracture interface. In comparison with
the double-porosity or dual-permeability model, MINC does not
rely on the pseudosteady-state assumption to calculate fracture/
matrix flow and is able to simulate fully transient fracture/matrix
interaction by subdividing nested-cell gridding inside matrix
blocks. The MINC concept should be generally applicable for
handling fracture/matrix flow in fractured-shale gas reservoirs, no
matter how large the matrix-block size is or how low the matrix
permeability is, and it is more suitable for handling fractured-
shale gas reservoirs. However, the MINC approach may not be
applicable to systems in which fracturing is so sparse that the frac-
tures cannot be approximated as a continuum.
As Fig. 6 shows, in our hybrid fracture model, both the hy-
draulic fractures and SRV are evaluated from the microseismic
cloud. Recent advances in microseismic-fracture mapping tech-
nology have provided previously unavailable information to char-
acterize hydraulic-fracture growth and SRV, and have documented
surprising complexities in many geological environments. We will
have a primary hydraulic-fracture system and an associated stimu-
lated volume in each hydraulic-fracture stage. First, we define a pri-
mary fracture on the basis of the orientation and region of the
microseismic cloud. The hydraulic fractures are modeled by the
discrete-fracture method. We assume the SRV near the hydraulic
fractures is the region with natural fractures, and we apply MINC
in this region. Single-porosity is applied in the region outside the
SRV, in which there are no natural fractures. Local grid refinement
(LGR) is used to improve simulation accuracy because pressure
gradients change substantially over short distances in the regions
near hydraulic fractures. LGR is performed near the hydraulic-frac-
ture region.
Numerical Solution. In this work, we use the fully implicit
scheme to solve the discrete nonlinear Eq. 16 with a Newton itera-
tion method. Let us write the discrete nonlinear equation, Eq. 16,
in a residual form as
h i
b;n Vi
Rb;nþ1
i ¼ ð/qSÞb;nþ1
i þ mb;nþ1
i  ð/qSÞb;n
i  mi
X Dt
 flowb;nþ1
ij  Q b;nþ1
i ¼ 0
j 2 gi
ðb ¼ 1; 2; i¼ 1; 2; 3;…; NÞ:           ð23Þ
Eq. 23 defines a set of 2  N coupled nonlinear equations that
need to be solved for every balance equation of mass components,
respectively. In general, two primary variables per node are
needed to use the Newton iteration for the associated two equa-
tions per node. The primary variables selected are gas pressure
LGR mesh
Dual-porosity, or dual-permeability, or
Well
MINC
Discrete fracture Single porosity
Slide view of this model
October 2014 SPE Journal
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 851 Total Pages: 13

1.2E+7 8.E+05
Analytical c_β = 0
Analytical b = 7.6E6 Analytical c_β = 1.0E–3
Numerical b = 7.6E6 7.E+05 Analytical c_β = 1.0E–4
1.0E+7 Numerical b = 0 Numerical c_β = 0
Numerical c_β =1.0E–3
Analytical b = 0 6.E+05 Numerical c_β =1.0E–4

Pressure, Pa
8.0E+6

Pressure, Pa
5.E+05
6.0E+6 4.E+05

4.0E+6 3.E+05
2.E+05
2.0E+6
1.E+05
0.0E+0 0.E+00
0 2 4 6 8 10 0 2 4 6 8 10
Horizontal Distance, m Horizontal Distance, m

Fig. 7—Analytical and numerical results for linear flow with the Fig. 8—Analytical and numerical results for linear non-Darcy
Klinkenberg effect. flow.

and gas saturation. The rest of the dependent variables—such as
relative permeability, capillary pressures, viscosity and densities,
adsorption term, and nonselected pressure and saturation—are
treated as secondary variables, which are calculated from selected
primary variables.
In terms of the primary variables, the residual equation, Eq.
23, at a node i is regarded as a function of the primary variables at
not only node i, but also at all its direct neighboring nodes j. The
Newton iteration scheme gives rise to
X @Rb;nþ1 ðxm;p Þ
i
ðdxm;pþ1 Þ ¼ Rb;nþ1
i ðxm;p Þ; . . . . . .
m
@xm
where xm is the primary variable m with m ¼ 1 and 2, respectively,
at node i and all its direct neighbors; p is the iteration level; and
i ¼ 1, 2, 3, …, N. The primary variables in Eq. 23 need to be
updated after each iteration,
xm;pþ1 ¼ xm;p þ dxm;pþ1 : . . . . . . . . . . . . . . . . . . . . . ð25Þ
The Newton iteration process continues until the residuals
Rb;nþ1
i or changes in the primary variables dxm;pþ1 over iteration
are reduced below preset convergence tolerances.
Numerical methods are generally used to construct the Jaco-
bian matrix for Eq. 24, as outlined in Forsyth et al. (1995). At
each Newton iteration, Eq. 24 represents a system of (2  N) line-
arized algebraic equations with sparse matrices, which are solved
by a linear equation solver.
Numerical-Model Verification
To examine the accuracy of our simulator formulation in simulat-
ing porous-medium gas flow with the Klinkenberg, non-Darcy-
flow, gas-adsorption, and geomechanics effects, several relevant
steady and transient analytical solutions are derived or used for
considering these flow mechanisms. The problem concerns steady-
state and transient gas flow across a 1D reservoir. The system con-
tains steady-/transient-state gas flow at an isothermal condition, and
a constant gas mass injection/production rate is imposed at one side
of the rock or well. The other boundary of the rock/reservoir is kept
at constant pressure. Eventually, the system will reach steady state,
if the production is maintained for a long period of time. A compar-
ison of the pressure profiles along the rock block from the simula-
tion and the analytical solution is shown in Figs. 7 and 8,
indicating that our simulated pressure distribution is in excellent
ð24Þ agreement with the analytical solutions for all the problems of 1D
linear flow with the Klinkenberg or non-Darcy-flow effect.
Details about the analytical solution derivation considering the
Klinkenberg and non-Darcy-flow effect are included in our previ-
ous work (Wu et al. 2012), and we will show their verification
results only in this section for the 1D linear-flow steady-flow situa-
tion. Comparisons between the analytical and numerical solutions
for the radial-flow and transient-flow cases are also presented in
our former work. Constant coefficients for the Klinkenberg effect
and correlation (Eq. 13) for the non-Darcy-flow coefficient are used
with comparison results shown in Figs. 7 and 8.
Verification for Flow With Adsorption. For the gas flow with
adsorption, the approximate analytical solution is given in Appen-
dix A. The parameters used for this comparison study are porosity
U ¼ 0.15; permeability k ¼ 100 md; formation temperature
T ¼ 25 C; gas viscosity l ¼ 1:64  102 cp; initial pressure
Pi ¼ 105 Pa; and thickness of the radial system is 1 m. The
well-boundary condition is a constant gas/mass-injection rate:
Q ¼ 1:0  104 kg=s.
Fig. 9 presents the comparisons of the pressure profile at 1.67
days from the numerical and analytical solutions. Two situations,

200000 200000
P_analytical
P_analytical
180000 180000 P_numerical
P_numerical
Pressure, Pa
Pressure, Pa

160000 160000

140000 140000

120000 120000
VL = 0 VL = 50
t = 1.67 days t = 1.67 days
100000 100000
0 0.5 1 1.5 2 2.5 0 1 2 3
Radius, m Radius, m

Fig. 9—Comparison of gas-pressure profiles considering gas adsorption in a radial system at 1.67 days, calculated with the numer-
ical and analytical solutions.

October 2014 SPE Journal 851

ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 852 Total Pages: 13

3.0E+7
TABLE 1—PARAMETERS FOR CHECKING INTEGRAL
SOLUTION FOR FLOW WITH GEOMECHANICS EFFECT 2.5E+7

Parameter Value Unit

Pressure, Pa
2.0E+7
7
Initial pressure Pi ¼ 10 Pa
Initial porosity Ui ¼ 0:20 1.5E+7
3 Constant K Pressure (N)
Initial fluid density qw ¼ 975:9 kg=m
1.0E+7 Constant K Pressure (A)
Cross-sectional area a ¼ 1.0 m2 unconstant K Pressure (A)
Formation thickness h ¼ 1.0 m 5.0E+6 unconstant K Pressure (N)
Fluid viscosity l ¼ 0:35132  103 Pa  s
Fluid compressibility Cf ¼ 4:556  1010 Pa1 0.0E+0
Rock compressibility Cr ¼ 5:0  109 Pa1 0 5000 10000
Initial permeability k0 ¼ 9:860  1013 m2 Time, seconds
Water-injection rate qm ¼ 0:01 kg=s
Fig. 10—Comparison of injection pressures calculated from in-
Hydraulic radius rw ¼ 0:1 m
tegral and numerical solutions for linear flow in a permeability-
Exponential index c ¼ 2.22 dependent medium with constant and nonconstant permeabil-
ity function.

Langmuir volume VL ¼ 0 and VL ¼ 50m3 =kg, are considered.

The analytical solutions give an excellent match with the numeri-
cal solution.
Verification for Linear Flow With Geomechanics. Wu and
Pruess (2000) presented an analytical method for analyzing the
nonlinear coupled rock-permeability-variation/fluid-flow problem.
Approximate analytical solutions for 1D linear and radial flow are
obtained by an integral method, which is widely used in the study
of steady and unsteady heat-conduction problems. The accuracy
of integral solutions is generally acceptable for engineering appli-
cations. When applied to fluid-flow problems in porous media, the
integral method consists of assuming a pressure profile in the
pressure-disturbance zone and determining the coefficients of the
profile by making use of the integral mass-balance equation.
The parameters, as shown in Table 1, are used to evaluate
both the numerical solution and the integral solution. A compari-
son of injection pressures from integral and numerical solutions is
shown in Fig. 10. The agreement between the two solutions is
excellent for the entire transient period.
Model Application
In the following model-application examples, we are concerned
with gas flow toward one horizontal well and a 10-stage hydrau-
lic-fracture system in an extremely tight, uniformly porous and/or
fractured reservoir (Fig. 11). The reservoir formation is at liquid/
gas, two-phase condition; however, the liquid saturation is set at
residual values as an immobile phase. This is a single-phase gas-
flow problem and is modeled by the two-phase-flow reservoir
simulator. The immobile liquid flow is controlled by liquid rela-
tive permeability curves.
We demonstrate the application of the proposed mathematical
model for modeling gas production from a producer with 10-stage
hydraulic fracturing in a shale gas reservoir. The stress alteration
induced by hydraulic fracturing may activate existing natural frac-
tures, and therefore opens microflow channels in the drainage area
of the stimulated well. Here, we present the simulation of a hy-
draulic-fracturing problem as an example case to illustrate the
capability of our hybrid fracture model to capture such a complex
fracture network in these reservoirs. Three different fracture mod-
els (as shown in Fig. 12) are built, and their flow behavior is com-
pared. The first one considers that there is no natural-fracture-
active area, and the whole formation is single-porosity shales with
low permeability. In the second model, we assume that only the
natural fractures within the SRV near the hydraulic fractures are
active and the rest of the natural fractures outside the SRV remain
inactive. An increase in pore pressure around the hydraulic frac-
ture causes a significant reduction in the effective stresses, poten-
tially reopening the existing healed natural fractures or creating
new fractures. As a result, the permeability near the well of the
reservoir is significantly improved. This effect would help
increase the well productivity in the initial production. The third
fracture model is that all the formation is naturally fractured.
To simulate the performance of this system with our model,
hydraulic fractures are represented by the discrete-fracture model
and an active, naturally fractured reservoir area is described by
the multicontinuum-fracture model, whereas a nonactive-natural-
fracture reservoir area is represented by the single-porosity model.
The basic parameter set for the simulation and discussion is sum-
marized in Table 2, which are chosen field data.
We first compare the gas-production behavior for these three
fracture models. Then, on the basis of the second fracture model
(i.e., reactivated natural fractures only in SRV), we analyze the
cumulative-gas-production curves with the Klinkenberg, geome-
chanics, and adsorption/desorption effects.

Fig. 12—Three different fracture models: From left to right are

no-natural-fracture model, SRV model, and all-formation-natu-
Fig. 11—Horizontal and multistaged hydraulic-fracture model. rally-fractured model.

852 October 2014 SPE Journal

ID: jaganm Time: 16:29 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 853 Total Pages: 13

TABLE 2—DATA USED FOR THE CASE STUDIES

Reservoir length, Dx, ft 5,500 Hydraulic-fracture permeability, khf , md 1105

Reservoir width, Dy, ft 2,000 Natural-fracture porosity, Unf 0.001
Formation thickness, Dz, ft 250 Natural-fracture total compressibility, cnf , psi1 2.5104
Reservoir depth, h, ft 5,800 Natural-fracture permeability, knf , md 1,600
Reservoir temperature, T, oF 200 Matrix total compressibility, ctm , psi1 2.5104
Initial reservoir pressure, Pi , psi 3,800 Matrix permeability, km , md 3.2105
Horizontal well length, Lh , ft 4,800 Matrix porosity, Um 0.05
Constant flowing bottomhole pressure, Pwf , psi 1,000 Viscosity, l, cp 0.0184
Hydraulic-fracture number 10 Langmuir’s volume, VL , scf/ton 77.56
Distance between hydraulic fractures, 2ye , ft 500 Langmuir’s pressure, PL , psi 2,285.7
Hydraulic-fracture porosity, Uhf 0.5 Non-Darcy-flow constant, b, ft1 1.29106
Hydraulic-fracture total compressibility, chf , psi1 2.5104
Hydraulic-fracture half-length, Xf , ft 250

8000 105 md, and the initial reservoir pressure is 3,800 psi. With this
Cumulative Production, MMscf

permeability value and under higher pressure, the Klinkenberg

7000
6000
5000
4000
3000
2000
1000
0
0 20 40 60
Time, years
Fig. 13—Simulated gas-production performance for the three
fracture models.
Fig. 13 compares the performance of the fractured horizontal
well for the three fracture models. The comparison indicates that
the fracture model makes a difference in well performance. The
contribution from active natural fractures is evident and helps to
yield higher production rates for a long period. A larger SRV
leads to a higher gas-production rate.
For the second fracture model, pressure distributions at 1 year
and 20 years are presented in Fig. 14.
Fig. 15 shows the cumulative-production comparison between
cases with and without the Klinkenberg effect. Here, our simula-
tor handles the Klinkenberg beta factor not as a constant value,
but a changing value with matrix permeability and pressure. As
shown in Table 2, the input data of matrix permeability are 3:2 
effect will not have an obvious influence on gas-flow permeability
on the basis of the estimation in Fig. 3. However, the constant bot-
tomhole production pressure is set as 1,000 psi, which is much
smaller than the reservoir initial pressure. When the pressure of
the region near the wellbore and hydraulic fracture decreases
quickly, the Klinkenberg effect becomes important for the flow in
this region. On the basis of Eqs. 12 and 13, the effective perme-
No Fracture ability considering the Klinkenberg effect at initial pressure
SRV (3,800 psi) is 3:69  105 md, whereas that at bottomhole pres-
Total Fracture
sure (1,000 psi) is 5:0  105 md. Our simulation result in Fig. 15
also shows the influence of the Klinkenberg effect. It leads to
approximately a 4% increase to the total gas production.
80 100 We studied the non-Darcy flow in the preceding scenario of a
horizontal well with multistage hydraulic fractures and natural
fractures to see its influence on gas production. The simulation
result is shown in Fig. 16 and 17. The difference is observed on
the gas cumulative production between the case considering the
non-Darcy flow and the case not considering the non-Darcy flow
in the first 6 years. Not considering the non-Darcy flow inside hy-
draulic fractures could lead to an overestimate of approximately
5% of cumulative gas production. After that, the difference
between cases diminishes until these two curves coincide at
approximately 40 years.
This simulation result is reasonable with the following analy-
sis. In Fig. 18, we compare the calculated gas-flow velocities
from Darcy’s law and the Forchheimer equation for different pres-
sure gradients. The parameters of permeability, viscosity, and the
non-Darcy-flow factor in this calculation are the same as those in
Table 2. When the pressure gradient is less than 1:0  103 psi/ft
or velocity is less than10 ft/D, there is almost no difference
between these two calculations. However, if the pressure gradient

Pg Pg
2.4E+07 2.4E+07
2.2E+07 2.2E+07
2E+07 2E+07
1.8E+07 1.8E+07
1.6E+07 1.6E+07
1.4E+07 1.4E+07
1.2E+07 1.2E+07
1E+07 1E+07
8E+06 8E+06
6E+06 6E+06
4E+06 4E+06

Fig. 14—Pressure distribution at 1 year (left) and 20 years (right) of Fracture Model #2 (unit: Pa).

October 2014 SPE Journal 853

ID: jaganm Time: 16:30 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 854 Total Pages: 13

6000 3000

Cumulative Prodcution, MMscf

Total Prodcution, MMscf 5000 2500

4000 2000

3000
1500
case without Darcy flow
2000
Klinkengberg 1000
case with non-Darcy flow
1000 Klinkenberg
500
0
0 20 40 60 80 100 0
Time, years 0 2 4 6 8
Time, years
Fig. 15—Gas-cumulative-production behavior with the Klinken-
berg effect. Fig. 16—Gas-cumulative-production behavior with non-Darcy
flow in the first 6 years.

6000 1.0E+5
Cumulative Production, MMscf

5000 1.0E+4

1.0E+3
4000

Velocity, ft/D
1.0E+2
3000
1.0E+1
Darcy flow
2000
non-Darcy flow Darcy Velocity
1.0E+0
1000 non-Darcy Velocity
1.0E–1
0
1.0E–2
0 20 40 60 80 100
1.0E–6 1.0E–4 1.0E–2 1.0E+0
Time, years
Pressure Gradient, psi/ft
Fig. 17—Gas cumulative-production behavior with non-Darcy Fig. 18—Darcy and non-Darcy velocities with pressure
flow in 100 years. gradient.

keeps increasing from 1:0  103 psi/ft, the difference will input of the figure data. As shown in Fig. 20, geomechanics/flow
become larger. Fig. 19 shows the calculated average gas flow rate coupling has a large impact on formation permeability, especially
inside hydraulic fractures with production time in the case that for the natural-fracture system. Consider the Muska formation, for
does not consider the non-Darcy flow. For the first 6 years, flow example, when the effective stress increases from 1,600 to 4,800
velocities locate in the range in which the difference between the psia and permeability decreases to 1/20 of its original value. With
Darcy flow and the non-Darcy flow is obvious. After that, flow the gas production, reservoir effective stress increases as pore
velocities move to the area in which the difference is negligible. pressure decreases, leading to the large reduction of cumulative
Fig. 20 shows the simulated-well cumulative production vs. gas production.
time with and without geomechanics effect. The relationship used Figs. 21 and 22 present the results for adsorption analysis with
for describing the effective stress and permeability of the uncon- the numerical model. On the basis of the data in Table 2, we cal-
ventional reservoir is shown in Fig. 1, by use of a table-lookup culate the total gas mass as free gas in the micropores and

1.0E+5 6000
Cumulative Production, MMscf

1.0E+4 Gas Flow Velocity in HF 5000

Velocity in HF, ft/D

1.0E+3
4000
1.0E+2
3000
1.0E+1
2000 case without
1.0E+0 geomechanics
1000 case with
1.0E–1 geomechanics

1.0E–2 0
0 10 20 30 0 10 20 30 40 50 60 70 80 90 100
Time, years Time, years

Fig. 19—Calculated gas-flow velocity with time in hydraulic Fig. 20—Gas-cumulative-production behaviors with geo-
fractures. mechanics.

854 October 2014 SPE Journal

ID: jaganm Time: 16:30 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14
6000
Cumulative Production, MMscf
5000
4000
3000
2000
case without adsorption
1000
case with adsorption
0 0
0 20 40 60 80 100
Time, years
Fig. 21—Gas-cumulative-production behaviors with adsorption.
adsorbed gas at initial condition. The proportion of gas stored in
the pore space is approximately 77%, whereas that stored as
adsorption is 23%. Then, we compare the cumulative gas produc-
tion with and without considering adsorption. Simulation results
(Fig. 21) show that the estimated gas production will increase
with considering adsorption. This difference will become more
and more evident. For the situation considering gas adsorption/de-
sorption, gas production from the desorption is approximately
13%, and the produced portion of the free gas consists of 87%, as
shown in Fig. 22.
Summary and Conclusions
This paper discusses a generalized-framework mathematical
model for modeling gas production from unconventional gas res-
ervoirs. The model formulation incorporates known nonlinear
flow processes, associated with gas production from low-perme-
ability unconventional reservoirs, including the Klinkenberg,
non-Darcy-flow, and nonlinear-adsorption effects. The model for-
mulation and numerical scheme are based on a generalized two-
phase (gas/liquid) -flow model with unstructured grids. Specifi-
cally, a hybrid modeling approach is presented by combining dis-
crete fracture, multidomain, and multicontinuum concepts for
handling hydraulic fractures and a fracture network in SRV, dis-
tributed natural fractures, microfractures as well as porous matrix.
We have verified the numerical models against analytical solu-
tions for the Klinkenberg, non-Darcy-flow, and nonlinear-adsorp-
tion effects.
As application examples, we present modeling studies with
three fracture models for gas production from a 10-stage hydrau-
lic-fractured horizontal well, incorporating the Klinkenberg, non-
Darcy-flow, and nonlinear-adsorption effects. The model results
show that there is a large impact of various fracture models on
gas-production rates as well as cumulative production.
Acknowledgments
This work was supported in part by EMG Research Center and
UNGI of the Petroleum Engineering Department at Colorado
School of Mines; by Foundation CMG; by RIPED of PetroChina
Company; and by IFPEN.
References
Blasingame, T.A. 2008. The Characteristic Flow Behavior of Low-Perme-
ability Reservoir Systems. Paper SPE 114168 presented at the SPE
Unconventional Reservoirs Conference, Keystone, Colorado, 10–12
February. http://dx.doi.org/10.2118-MS.
Bustin, R.M., Bustin, A.M.M., Cui, X. et al. 2008. Impact of Shale Proper-
ties on Pore Structure and Storage Characteristics. Paper SPE 119892
October 2014 SPE Journal
ID: jaganm Time: 16:30 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016
Stage: Page: 855 Total Pages: 13
6000
Cumulative Production, MMscf
5000
4000 gas production
from free gas
3000 gas production
from adsorption
total production
2000
1000
0 20 40 60 80 100
Time, years
Fig. 22—Gas-production analysis from free gas and adsorbed
gas.
presented at the SPE Gas Production Conference, Fort Worth, Texas,
16–18 November. http://dx.doi.org/10.2118/119892-MS.
Cipolla, C.L. 2009. Modeling Production and Evaluating Fracture Per-
formance in Unconventional Gas Reservoirs. J. Pet Tech 61 (9): 84–90
(Distinguished Author Series). http://dx.doi.org/10.2118/118536-PA.
Cipolla, C.L., Lolon, E.P., Erdle, J.C. et al. 2010. Reservoir Modeling in
Shale-Gas Reservoirs. Res Eval & Eng. 13 (4): 638–653. http://
dx.doi.org/10.2118/125530-PA.
Cipolla, C.L., Warpinski, N.R., Mayerhofer, M.J. et al. 2009. The Rela-
tionship Between Fracture Complexity, Reservoir Properties, and
Fracture Treatment Design. Paper SPE 115769 presented at the SPE
ATC&E, Denver, Colorado, 21–24 September. http://dx.doi.org/
10.2118/115769-MS.
Davies, J.P. and Davies, D.K. 1999. Stress-Dependent Permeability Char-
acterization and Modeling. Paper SPE 56813 presented at the SPE
ATCE, Houston, Texas, 3–6 October. http://dx.doi.org/10.2118/
56813-MS.
EIA (US Energy Information Administration). 2011. World Shale Gas
Resources: An Initial Assessment of 14 Regions Outside the United
States (April).
Ertekin, T., King, G.R., and Schwerer, F.C. 1986. Dynamic Gas Slippage:
A Unique Dual-Mechanism Approach to the Flow of Gas in Tight For-
mations. SPE Form Eval 1 (1): 43–52. http://dx.doi.org/10.2118/
12045-PA.
Evans, R.D., and Civan, F. 1994. Characterization of Non-Darcy Multi-
phase Flow in Petroleum Bearing Formation. Final Report for US
Department of Energy Assistant Secretary for Fossil Energy.
Forsyth, P.A., Wu, Y.S., and Pruess, K. 1995. Robust Numerical Methods
for Saturated-Unsaturated Flow With Dry Initial Conditions in Hetero-
geneous Media. Advance in Water Resources 18: 25–38.
Freeman, C.M., Moridis, G.J., Ilk, D. et al. 2010. A Numerical Study of
Transport and Storage Effects for Tight Gas and Shale Gas Reservoir
Systems. Paper SPE 131583 presented at the SPE International Oil and
Gas Conference and Exhibition, Beijing, China, 8–10 June. http://
dx.doi.org/10.2118/131583-MS.
Freeman, C.M., Moridis, G.J., and Blasingame, T.A. 2009a. A Numerical
Study of Microscale Flow Behavior in Tight Gas and Shale Gas Reser-
voir Systems. Proceedings of the 2010 TOUGH Symposium, Berkeley,
California, 14–16 September.
Freeman, C.M., Moridis, G.J., Ilk, D. et al. 2009b. A Numerical Study of
Tight Gas and Shale Gas Reservoir Systems. Paper SPE 124961 pre-
sented at the SPE Annual Technical Conference and Exhibition, New
Orleans, Louisiana, 4–9 October. http://dx.doi.org/10.2118/124961-
MS.
Gao, C., Lee, J.W., Spivey, J.P. et al. 1994. Modeling Multilayer Gas Res-
ervoirs Including Sorption Effects. Paper SPE 29173 presented at the
SPE Eastern Regional Conference and Exhibition, Charleston, West
Virginia, 8–10 November. http://dx.doi.org/10.2118/29173-MS.
Kelkar M. and Atiq, M. 2010. Upgrading Method for Tight Gas Reser-
voirs. Paper SPE 133301 presented at the SPE Annual Technical
855
 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 856 Total Pages: 13

Conference and Exhibition, Florence, Italy, 19–22. http://dx.doi.org/ Wu, Y.S. 2002. Numerical Simulation of Single-Phase and Multiphase
10.2118/133301-MS. Non-Darcy Flow in Porous and Fractured Reservoirs. Transport in Po-
Klinkenberg, L.J. 1941. The Permeability of Porous Media to Liquids and rous Media 49: 209–240.
Gases. In API Drilling and Production Practice, 200–213. Wu, Y.S. and Pruess, K. 2000. Integral Solutions for Transient Fluid Flow
Langmuir, I. 1916. The Constitution and Fundamental Properties of Solids Through a Porous Medium With Pressure-Dependent Permeability.
and Liquids. J. Am. Chem. Soc. 38 (11): 2221–2295. International J. Rock Mechanics and Mining Sci. 37: 51–61.
Wu, Y.S., Pruess, K., and Persoff, P. 1998. Gas Flow in Porous Media
Leahy-Dios, A., Das, M., Agarwal, A. et al. 2011. Modeling of Transport
With Klinkenberg Effects. Transport in Porous Media 32: 117–137.
Phenomena and Multicomponent Sorption for Shale Gas and Coalbed
Wu, Y.S., Rutqvist, J., Karasaki, K. et al. 2008. A Mathematic Model for
Methane in an Unstructured Grid Simulator. Paper SPE 147352 pre-
Rock Deformation Effect of Flow in Porous and Fractured Reservoirs.
sented at the SPE Annual Technical Conference, Denver, Colorado, 30
Paper ARMA-08-142 presented at the 42nd US Rock Mechanics Sym-
Octomber–2 November. http://dx.doi.org/10.2118/147352-MS.
posium and 2nd US–Canada Rock Mechanics Symposium, San Fran-
Leverett, M.C. 1941. Capillary Behavior in Porous Media. Trans AIME cisco, California, 29 June–2 July.
142: 341–358. Wu, Y.S. and Wang, C. 2012. Transient Pressure Testing Analysis of Gas
Li, J., Du, C.M., and Zhang, X. 2011. Critical Evaluation of Shale Gas Wells in Unconventional Reservoirs. Paper SPE-SAS-312 presented at
Reservoir Simulation Approaches: Single-Porosity and Dual-Porosity the SPE Annual Technical Symposium and Exhibition (ATS&E), Kho-
Modeling. Paper presented at the SPE Middle East Unconventional bar, Saudi Arabia, 8–11 April.
Gas Conference and Exhibition, Muscal, Oman, 31 January–2 Wu, Y.S., Wang, C., Li, J. et al. 2012. Transient Gas Flow in Unconventional
February. Gas Reservoirs. Paper SPE 154448 presented at the EAGE Annual Con-
Mengal, S.A. and Wattenbarger, R.A. 2011. Accounting for Adsorbed Gas ference and & Exhibition Incorporating SPE Europec, Copenhagen, Den-
in Shale Gas Reservoirs. Paper SPE 141085 presented at the SPE Mid- mark, 4–7 June. http://dx.doi.org/10.2118/154448-MS.
dle East Oil and Gas Show and Conference, Manama, Bahrain, 25–28
September. http://dx.doi.org/10.2118/141085-MS. Appendix A
Moridis, G.J., Blasingame, T.A., and Freeman, C.M. 2010. Analysis of Here, we derive the analytical solution for gas flow with adsorp-
Mechanisms of Flow in Fractured Tight-Gas and Shale-Gas Reser- tion/desorption. If the system is isothermal, the ideal-gas law
voirs. Paper SPE 139250 presented at the SPE Latin American and applies, and the gravity effect is negligible, then gas flow in porous
Caribbean Petroleum Engineering Conference, Lima, Peru, 1–3 media with adsorption is described by the following equations:
December. http://dx.doi.org/10.2118/139250-MS.
Ozkan, E., Raghavan, R., and Apaydin, O.G. 2010. Modeling of Fluid
@ð/q þ mg Þ
r  ðqvÞ ¼  ; . . . . . . . . . . . . . . . . . . ðA-1Þ
Transfer From Shale Matrix to Fracture Network. Paper SPE 134830 @t
presented at the SPE ATC&E, Florence, Italy, 19–22 September. where q is the gas density; v is the gas-flow velocity; / is the po-
http://dx.doi.org/10.2118/134830-MS. rous-media porosity; mg is the adsorbed gas mass in a unit forma-
Pruess, K. 1983. GMINC—A Mesh Generator for Flow Simulations in tion volume at a given pressure; and t is the time.
Fractured Reservoirs, Report LBL-15227, Berkeley, California: Law- According to the ideal-gas law,
rence Berkeley National Laboratory.
Pruess, K. and Narasimhan, T.N. 1985. A Practical Method for Modeling PV ¼ nRT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-2Þ
Fluid and Heat Flow in Fractured Porous Media. Soc. Pet. Eng. J. 25:
14–26. and
Pruess, K., Oldenburg, C., and Moridis, G. 1999. TOUGH2 User’s Guide, M
q¼ P ¼ bP; . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-3Þ
Version 2.0, Report LBNL-43134, Berkeley, California: Lawrence RT
Berkeley National Laboratory.
where M is gas molecular weight; R is the universal gas constant;
Rubin, B. 2010. Accurate Simulation of Non-Darcy Flow in Stimulated M
Fractured Shale Reservoirs. Paper SPE 132093 presented at the SPE b is a coefficient, for simplicity, defined as b ¼ ; and T is the
Western Regional Meeting, Anaheim, California, 27–29 May. http://
RT
system temperature.
dx.doi.org/10.2118/132093-MS. From Darcy’s law and the Langmuir isotherm (Eqs. 2 and 3),
Rutqvist, J.Y.,Wu, S., Tsang, C.F. et al. 2002. A Modeling Approach for
Analysis of Coupled Multiphase Fluid Flow, Heat Transfer, and Defor- k
v ¼  rP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-4Þ
mation in Fractured Porous Rock. International J. Rock Mechanics l
and Mining Sci. 39: 429–442.
and
Silin, D. and Kneafsey, T. 2011. Gas Shale: From Nanometer-Scale Obser-
vations to Well Modeling. Paper CSUG/SPE 149489 presented at the P P
mg ¼ qR qg VE ¼ qR qg VL ¼ Va ; . . . . . ðA-5Þ
Canadian Unconventional Resources Conference, Calgary, Alberta, P þ PL P þ PL
Canada, 15–17 November. http://dx.doi.org/10.2118/149489-MS.
where qR is rock bulk density; qg is gas density at standard condi-
Soeder, D.J. 1988. Porosity and Permeability of Eastern Devonian Gas
tion; VE is the adsorption isotherm function for gas content; VL is
Shale. SPE Form Eval 3 (1): 116–124. http://dx.doi.org/10.2118/
the Langmuir’s volume in scf/ton; and PL is Langmuir’s pressure.
15213-PA.
a is a coefficient, for simplicity, defined as a ¼ qR qg VL .
Wang, F.P., Reed, R.M., Jackson, J.A. et al. 2009. Pore Networks and By substituting Eqs. A-4 and A-5 into Eq. A-1, we obtain
Fluid Flow in Gas Shales. Paper SPE 124253 presented at the SPE An-  
nual Technical Conference and Exhibition, New Orleans, Louisiana, P
  @
4–7 October. http://dx.doi.org/10.2118/124253-MS. k @P P þ PL
r  b PrP ¼ /b þa : . . . . . . . .ðA-6Þ
Winterfeld, P.H. and Wu, Y.S. 2011. Parallel Simulation of CO2 Seques- l @t @t
tration With Rock Deformation in Saline Aquifers. Paper SPE 141514
prepared for presentation at the SPE Reservoir Simulation Symposium, In radial coordinates,
 
The Woodlands, Texas, 21–23 February. http://dx.doi.org/10.2118/ P
 2
 @
141514-MS. 1@ @P 2/l @P 2al P þ PL @P
r ¼ þ ; . . . . ðA-7Þ
Winterfeld, P.H. and Wu, Y.S. 2012. A Novel Fully Coupled Geomechan- r @r @r k @t bk @P @t
ical Model for CO2 Sequestration in Fractured and Porous Brine Aqui-
fers. Paper presented at the XIX International Conference on Water   " #
1@ @P2 2/l 2al PL @P
Resources CMWR 2012, University of Illinois at Urbana-Champaign, r ¼ þ ; . . . . . . ðA-8Þ
r @r @r k bk ðP þ PL Þ2 @t
Illinois, 17–22 June.

856 October 2014 SPE Journal

ID: jaganm Time: 16:30 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

 J163609 DOI: 10.2118/163609-PA Date: 9-October-14 Stage: Page: 857 Total Pages: 13

and earned BS (Eqv.) and MS degrees in petroleum engineering in

China, and MS and PhD degrees in reservoir engineering from
  " #
1@ @P2 /l al PL @P2 the University of California at Berkeley. At CSM, he is teaching
r ¼ þ 2
: . . . . . . ðA-9Þ petroleum reservoir engineering courses, supervising graduate
r @r @r Pk Pbk ðP þ PL Þ @t students, and conducting research in the areas of multiphase
fluid and heat flow in porous media, enhanced oil recovery
Eq. 9 becomes (EOR), CO2 geosequestration and CO2 EOR, reservoir simula-
tion, enhanced geothermal systems, geomechanics coupling
1 @ @P2 1 @P2 and rock-deformation effects, and unconventional hydrocar-
ðr Þ¼ ; . . . . . . . . . . . . . . . . . . . . . ðA-10Þ
r @r @r A @t bon reservoirs. Previously, Wu was a staff scientist with LBNL for
14 years (1995 through 2008).
where we define the coefficient
Jianfang Li is a senior research engineer at Research Institute
1 /l al PL of Petroleum Exploration and Development (RIPED), Petro-
¼ þ : . . . . . . . . . . . . . . . . . . ðA-11Þ China. She earned BS and MS degrees in applied mathemat-
A Pk Pbk ðP þ PL Þ2 ics from Beijing University of Aeronautics and Astronautics and
a PhD degree in reservoir engineering from the graduate
We propose to use a history-dependent, constant, averaged school of RIPED. Li’s research interests include optimization,
pressure within the pressure-changed domain (Wu et al. 1998), algorithms for solving linear algebraic equations, reservoir simula-
X tion (carbonate-reservoir simulation, chemical-flooding reservoir
Vj Pj simulation, natural-gas-reservoir simulation, and unconventional-
P X ; . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-12Þ hydrocarbon-reservoir simulation).
Vj
Didier-Yu Ding is a senior research engineer at IFP Energies
where Vj is a controlled volume at the geometric center of which Nouvelles in France. His research interests include numerical
the pressure was Pj at the immediately preceding modeling and optimization, reservoir simulation and charac-
X time when the terization, complex wells, and near-well flow. Ding holds a BS
solution was calculated. The summation, Vj , is performed degree in mathematics from Peking University in China, and
over all Vj in which pressure increases (or decreases) occurred at MS and PhD degrees in applied mathematics from University
the preceding time value. Pj is always evaluated analytically at de Paris, France.
point j, on the basis of the previous estimated, constant Cong Wang is a PhD candidate in petroleum engineering at
diffusivity. CSM. He earned a BS degree from Peking University and an
The well boundary proposed as a line source/sink well is MS degree from the Petroleum Engineering Department, CSM.
Wang’s Master’s degree study was on the development of an
pkhrb @P2 unconventional reservoir simulator and its application in tran-
lim ¼ Qm : . . . . . . . . . . . . . . . . . . . . . ðA-13Þ sient-pressure analysis and fracture characterization. He has
x!0 l @r
made several conference presentations at SPE events, and his
Then, we could get a transient-pressure solution for gas flow current research involves fully coupling geomechanics and
with adsorption/desorption, fluid flow in unconventional reservoirs.
  Yuan Di is an associate professor in the Department of Energy
lQm r2 and Resources Engineering, College of Engineering, Peking
P2 ðr; tÞ ¼P2i  Ei  :             ðA-14Þ University. He earned a BS degree from Xi’an Jiaotong Univer-
2pkhrb 4At
sity, an MS degree from Xi’an University of Architecture and
Technology, and a PhD degree from Tongji University, Shang-
Yu-Shu Wu is a professor and Foundation CMG Research Chair hai—all in civil engineering. Di’s research interests include the
in Petroleum Engineering at the Colorado School of Mines numerical simulation of multiphase flow in porous media,
(CSM). He is also a guest scientist at the Earth Sciences Division wave propagation in porous media, and the analysis of bore-
of the Lawrence Berkeley National Laboratory (LBNL). Wu hole stability.

October 2014 SPE Journal 857

ID: jaganm Time: 16:30 I Path: S:/3B2/J###/Vol00000/140016/APPFile/SA-J###140016

Reservoir Modeling in Shale-Gas Reservoirs
C.L. Cipolla and E.P. Lolon, StrataGen Engineering; and J.C. Erdle and B. Rubin, Computer Modelling Group
Summary
The exploitation of unconventional gas reservoirs has become an
ever increasing component of the North American gas supply.
The economic viability of many unconventional gas developments
hinges on effective stimulation of extremely low-permeability rock
by creating very complex fracture networks that connect huge
reservoir surface area to the wellbore. In addition, gas desorption
may be a significant component of overall gas recovery in many
shale-gas reservoirs. The widespread application of microseismic
(MS) mapping has significantly improved our understanding of
hydraulic fracture growth in unconventional gas reservoirs (pri-
marily shale) and has led to better stimulation designs. However,
the overall effectiveness of stimulation treatments is difficult to
determine from MS mapping because the location of proppant
and the distribution of conductivity in the fracture network cannot
be measured (and are critical parameters that control well perfor-
mance). Therefore, it is important to develop reservoir-modeling
approaches that properly characterize fluid flow in and the prop-
erties of a complex fracture network, tight matrix, and primary
hydraulic fracture (if present) to evaluate well performance and
understand critical parameters that affect gas recovery.
This paper illustrates the impact of gas desorption on produc-
tion profile and ultimate gas recovery in shale reservoirs, showing
that in some shale-gas reservoirs desorption may be a minor com-
ponent of gas recovery. In addition, the paper details the impact of
changing closure stress distribution in the fracture network on well
productivity and gas recovery. In shale-gas reservoirs with lower
Young’s modulus rock, stress-dependent network-fracture conduc-
tivity may reduce ultimate gas recovery significantly. The paper
includes an example that contrasts the application of numerical
reservoir simulation and advanced decline-curve analyses to illus-
trate issues associated with conventional production-data-analysis
techniques when applied to unconventional reservoirs.
Selected examples from the Barnett shale are included that
incorporate MS fracture mapping and production data to illustrate
the application of production modeling to evaluate well perfor-
mance in unconventional gas reservoirs. This paper highlights
production modeling and analysis techniques that aid in evaluat-
ing stimulation and completion strategies in unconventional gas
reservoirs.
Introduction
Gas shales are organic-rich shale formations and are apparently
both the source rock and the reservoir. The gas is stored in the
limited pore space of these rocks and a sizable fraction of the gas
in place may be adsorbed on the organic material. The natural-
gas resource potential for gas shales in the USA is estimated to
be from 500 to 1,000 Tcf (Arthur et al. 2008). Typical shale-gas
reservoirs exhibit a net thickness of 50 to 600 ft, porosity of 2 to
8%, and total organic carbon of 1 to 14% and are found at depths
ranging from 1,000 to 13,000 ft. The success of the Barnett shale
has illustrated that gas can be economically produced from rock
that was previously thought to be source rock and/or caprock, not
reservoir rock. This revelation has led to the development of many
other shale-gas reservoirs, including the Woodford, Fayetteville,
Marcellus, and the Haynesville (Fig. 1). In addition to increasing
natural-gas prices (until recently), the economic development of
Copyright © 2010 Society of Petroleum Engineers
This paper (SPE 125530) was accepted for presentation at the SPE Eastern Regional
Meeting, Charleston, West Virginia, USA, 23–25 September 2009 and revised for publication.
Original manuscript received for review 21 July 2009. Revised manuscript received
for review 29 December 2009. Paper peer approved 1 March 2010.
638
many shale reservoirs was made possible through improved stimu-
lation techniques and horizontal drilling.
The economic viability of many unconventional gas develop-
ments hinges on effective stimulation of extremely low-permeability
rock, typically 10 to 100 nanodarcies (10−6 md). In most cases,
economic production is possible only if a very complex, highly
nonlinear fracture network can be created that effectively connects
a huge reservoir surface area to the wellbore. Many conventional
hydraulic-fracture treatments use high-viscosity fluids to reduce
fracture complexity and promote planar fractures, allowing the
placement of high concentrations of large proppant. However,
stimulation treatments in shale reservoirs use large volumes of
low-viscosity fluid (water) to promote fracture complexity and
place very low concentrations of small proppant—an approach
that is completely different from conventional hydraulic fracturing.
Hydraulic-fracture treatment stages in horizontal wells can require
over a million gallons of water and up to 1 million pounds of prop-
pant and are pumped at rates of 75–150 bbl/min. Because of the
complexity of fracture growth in many shale-gas reservoirs, it is not
possible to accurately predict the distribution of proppant within
the network and the conductivity of the network fractures is dif-
ficult to predict. Previous work has characterized proppant distribu-
tion within complex fracture networks using two limiting cases:
either the proppant is evenly distributed throughout the complex
network or the proppant is confined to a primary or main fracture
(Cipolla et al. 2008a). This previous work showed that for prop-
pant volumes that are typical for many shale-gas hydraulic-fracture
treatments, if the proppant is evenly distributed throughout the
complex fracture network, proppant concentrations (lbm/ft2) are
likely to be too low to materially affect network-fracture conductiv-
ity and the network fractures will essentially behave as unpropped
fractures. If the proppant is confined to a primary or main hydraulic
fracture and is not transported into the network fractures, then the
relative conductivity of the primary fracture should be high, but the
fracture network will be unpropped. Therefore, unpropped-fracture
conductivity may play an important role in shale-gas stimulation
performance. Network-fracture conductivity in many shale-gas
reservoirs is estimated by modeling well performance. Combining
detailed reservoir modeling with MS-mapping data that define the
size of the fracture network and laboratory cores tests that provide
estimates of matrix permeability and propped- and unpropped-
fracture conductivity can result in a much better understanding of
stimulation performance.
Gas flow from ultralow-permeability rock through the complex
fracture network must be modeled to evaluate stimulation designs
and completion strategies properly. Therefore, the complex frac-
ture network and primary hydraulic fracture (if present) must be
discretely characterized in our reservoir-simulation models. MS
fracture mapping provides a measurement of the overall reservoir
volume that is stimulated [referred to as stimulated reservoir vol-
ume (SRV)], and special core analysis can provide measurements
of matrix permeability. With these two parameters, reservoir-simu-
lation models can be used to estimate the spacing of the network
fractures (complexity) and their conductivity. Simulated produc-
tion profiles can then be compared to actual well performance to
evaluate the effectiveness of stimulation designs and completion
strategies and diagnose the relative conductivity of the primary
fracture (if present). This paper builds upon and extends previous
shale-gas reservoir-modeling work (Mayerhofer et al. 2006, 2010;
Warpinski et al. 2008; Cipolla et al. 2008a, 2009b), focusing on
the impact of gas desorption and stress-sensitive network-fracture
conductivity on well performance. In addition, this work discusses
various shale-gas reservoir-simulation approaches, detailing an
August 2010 SPE Reservoir Evaluation & Engineering
 Fig. 1—Lower 48 states shale-gas plays.

approach that appears best suited for most shale-gas reservoirs.
The highlights of the previous work are summarized to provide
background to support the work presented in this paper.
Background
The initial MS-mapping work in the Barnett shale revealed that
fracture growth can be surprisingly complex in some environments.
Fig. 2 shows the MS-mapping data for a typical vertical well in
the Barnett shale (Fisher et al. 2002). Stimulation treatments in
vertical wells consist of approximately 750,000 gal of slickwater
and 150,000 lbm of proppant pumped at concentrations of 0.1 to
0.25 ppa and rates of 50–70 bbl/min. Early work showed that very
large fracture networks were created in the Barnett shale, typically
in excess of 2,500 ft long and 1,000 ft wide, covering the Barnett
reservoir thickness (≈300 ft). In Fig. 2, the fracture network is
more than 3,000 ft long and approximately 1,000 ft wide, extend-
ing more than 70 acres and contacting 900 million ft3 of reservoir
rock. Fig. 2 also shows that five offset wells were “killed” by
stimulation fluid from the treatment well, confirming that large
volumes of water are transversing the complex fracture network.
The volume of reservoir rock that is contacted by the stimulation
treatment (SRV), as evidenced by the MS activity, has a direct
impact on well performance—with larger SRVs resulting in better
wells (Mayerhofer et al. 2010). This early MS-mapping work led
to a much better understanding of fracture growth in the Barnett
shale and was instrumental in many improvements in stimulation
and completion strategies, including horizontal drilling.
Fig. 3 shows the MS-mapping data for a horizontal Barnett
shale well completed with an uncemented liner (five sets of per-
forations) and stimulated using a single-stage hydraulic-fracturing
treatment consisting of 4.2 million gal of slickwater and 853,000
lbm of proppant pumped at 130 bbl/min [Fisher et al. 2004]. In this
case, the single-stage hydraulic-fracture treatment covered most of
the 2,400 ft long lateral, but there are areas with little MS activity
that may not be stimulated effectively. The stimulation treatment
appears to have effectively covered an area of approximately

1500 2000.00

Wells killed by stimulation fluid 1600.00

1000 1200.00

800.00
500 Fairway “Length”
“Length”
400.00
Observation Well Fairway“Width”
Fairway “Width”
0.00
South-North (ft)

0
South-North (ft)

–400.00

–500 –800.00
Treatment Well
–1200.00
–1000
–1600.00

–2000.00
–1500
–2400.00

–2000 –2800.00

Wells killed by stimulation fluid –3200.00

–2500
–3600.00

–4000.00
–3000
–2800

–2400

–2000

–1600

–1200

–800

–400

400

800

1200

1600

2000

2400

2800

–1000 –500 0 500 1000 1500 2000 2500 3000

West-East (ft) West-East (ft)

Fig. 2—MS-mapping data from a vertical Barnett shale well Fig. 3—MS-mapping data for a horizontal Barnett shale well,
(Fisher et al. 2002). uncemented liner, single-stage treatment (Fisher et al. 2004).

August 2010 SPE Reservoir Evaluation & Engineering 639

2,200×2,000 ft (square), or 100 acres, not that much larger than
the vertical-well treatment shown in Fig. 2. This led operators to
cased-and-cemented horizontal completions in which multistage
stimulation treatments can be pumped, resulting in larger SRVs
and possibly more-complex fracture networks—with horizontal
completions yielding more than three times the estimated ultimate
recovery (EUR) of vertical wells (Frantz et al. 2005). Although
MS mapping provided invaluable insights into fracture growth and
the impact of SRV on well performance, there was still a need to
better understand the mechanisms that controlled production in the
Barnett shale (and other shale reservoirs) to improve completion
strategies and stimulation designs, which required detailed numeri-
cal reservoir modeling.
In addition to confirming the relationship between SRV and
well performance, reservoir-modeling work has shown that the
complexity and conductivity of the fracture network are key
components that control well productivity in shale-gas reservoirs
(Mayerhofer et al. 2006; Cipolla et al. 2008a). In addition, the
location of the proppant within the complex fracture network and
the conductivity of the primary or main fracture (if present) play
a significant role in well performance (Cipolla et al. 2009a). This
previous work showed that more-complex fracture networks (i.e.,
smaller blocks or more-dense fracture spacing) result in higher
production rates. Reservoir-simulation history matching has indi-
cated that network-fracture conductivity may be relatively low
in the Barnett shale, ranging from 0.5 to 5.0 md-ft, limiting gas
production in many cases (Mayerhofer et al. 2006; Cipolla et al.
2008a). In addition, the presence of a high-relative-conductivity
primary fracture will improve production significantly when net-
work-fracture conductivity is low; however, work to date suggests
that high-conductivity primary fractures may not be present in the
Barnett shale (Cipolla et al. 2009c). With a better understanding
of the mechanisms that control well productivity, operators have
focused on completion techniques and stimulation designs that
maximize network complexity, increase SRV, and improve fracture
conductivity such as more fracture-treatment stages, larger volume
and higher injection rate treatments, diversion to increase fracture
complexity, larger proppant volumes and higher-strength and/or
lower-density proppants to increase fracture conductivity. In addi-
tion, operators are experimenting with simultaneous or alternating
stimulations in offsetting wells to increase fracture complexity and
improve gas recovery.
Shale-Gas Reservoir Simulation
Given the complex nature of hydraulic-fracture growth and the very
low permeability of the matrix rock in many shale-gas reservoirs—
in combination with the predominance of horizontal completions—
reservoir simulation is commonly the preferred method to predict
and evaluate well performance. However, semianalytical solutions
for hydraulically fractured horizontal wells in fractured reservoirs
have been published (Medeiros et al. 2008). Analytical solutions
for fluid flow in naturally fractured reservoirs were published by
Warren and Root (1963) and Kazemi (1969); however, analytical
solutions to fluid flow in naturally fractured reservoirs typically
cannot capture the very long transient behavior in the matrix blocks
exhibited by shale-gas reservoirs (caused by the very low perme-
ability of the shale matrix). These “pseudosteady-state” solutions
to fluid flow in naturally fractured reservoirs have been adapted
to numerical reservoir-simulation models to improve run time, but
again, they lack the ability to model transient flow in the matrix
blocks. Techniques have been developed to model the transient
behavior of matrix blocks in reservoir simulators, but many of
these techniques can still rely on analytical approximations that
are used within the numerical model to reduce run time. The most
rigorous method to model shale-gas reservoirs is to grid the entire
reservoir discretely, including the network fractures, hydraulic
fracture, matrix blocks, and unstimulated areas—but this increases
computational time. However, with the continual advances in
computing power, much-more-complex numerical models can be
used efficiently. The reservoir simulations in this paper were per-
formed using a detailed numerical grid that rigorously represents
the complex fracture network, primary fracture, and tight shale
640
matrix. A more detailed discussion of shale-gas reservoir modeling
is provided in Appendix A.
Gas Desorption and Stress-Dependent
Fracture Permeability
Previous reservoir-modeling work has largely neglected the effect
of gas desorption and stress-dependent fracture permeability on
well productivity and gas recovery. A series of reservoir simula-
tions was performed using reservoir properties typical for the
Barnett and Marcellus shales to evaluate the likely impact of
gas desorption and stress-dependent fracture permeability in two
important shale developments in the USA (Fig. 1). The simulations
focus on horizontal completions because the majority of shale-gas
reservoirs are being developed using cased-and-cemented hori-
zontal wells with multiple fracture treatment stages. The length of
the horizontal section is 3,000 ft. The reservoir model consists of
main, or primary, fractures and network, or secondary, fractures.
The main (or primary) fracture representing the propped fracture
is connected to the wellbore and is oriented perpendicular to the
direction of the horizontal wellbore. The primary-fracture spacing
is controlled by the number of fracture-treatment stages along the
lateral, with more stages resulting in closer spacing of the primary
fractures. The secondary (or network) fractures are assumed to be
unpropped and have no direct communication with the wellbore
and are placed in the perpendicular and parallel directions to the
primary fracture (assumed square in dimension). The spacing
of the secondary fractures (size of the network blocks) is varied
within a range that appears consistent with actual well performance
and estimates of matrix permeability and fracture conductivity
(Fredd et al. 2001; Mayerhofer et al. 2006; Cipolla et al. 2008a,
2009c). It is assumed that gas flow occurs into the wellbore
through the main fractures only. A wide range of reservoir and
fracture parameters was evaluated, and only a limited number of
simulations are presented to illustrate the results of this work. All
simulations are single phase (dry gas).
Barnett Shale: Impact of Gas Desorption. Table 1 shows the
main reservoir parameters used for Barnett shale reservoir simula-
tions. Pore pressure is 3,800 psi or 0.54 psi/ft. The reservoir size
is 3,400×2,400 ft, with a fracture-network size of 2,000×3,400 ft
(SRV=2,040×106 ft3). Fig. 4 shows a graph of gas content (scf/ton)
vs. pressure for the gas porosity and adsorbed gas used for the
Barnett example (Montgomery 2004). The gas in the matrix poros-
ity is shown as a nearly straight line (circles) while the desorption
isotherm is represented by a curved line (squares). Both data sets
add together to form the total gas content. In this case, 41% of
the total gas is adsorbed. At higher pressures, more gas is stored
in the matrix porosity and the gas porosity will contribute more
to production than desorption because of the difference in curve
shapes (straight line vs. curved line). Although there appears to
be a considerable amount of adsorbed gas, it may be difficult to
capture this gas when reservoir permeability is low.
Figs. 5 and 6 show the impact of gas desorption on cumulative
gas production for a typical Barnett horizontal completion. The
TABLE 1—BARNETT SHALE RESERVOIR PROPERTIES
Depth (ft) 7,000
Reservoir pressure (psi) 3,800
Pore pressure gradient (psi/ft) 0.54
Closure pressure (psi) 5,000
Closure pressure gradient (psi/ft) 0.71
Net pay (ft) 300
Porosity (% ) 3
Matrix permeability (md) 1 E 5; 1 E 4
Water saturation (%) 30
Reservoir temperature (°F) 180
FBHP (psi) 1,000
August 2010 SPE Reservoir Evaluation & Engineering
 Fig. 4—Example of Barnett gas content, adsorbed and free.
simulations assume a uniform network-fracture conductivity of
2 md-ft. The production for two cased-and-cemented Barnett hori-
zontal wells is shown for comparison, Well A and Well B. These
wells were selected because MS-mapping data were available to
determine the SRV. The SRV, lateral length, and primary-fracture
spacing were similar to many of the simulation cases. Well A has
a lateral section of 2,600 ft. The fracture treatments in this well
consisted of four stages (one perforation cluster per stage, with
cluster spacing of 700 ft). Therefore, the primary- or main-frac-
ture spacing is 700 ft. The total amount of proppant pumped was
670,000 lbm (40/70 sand) with 120,000 bbl of fluid. The MS map-
ping of this well showed an SRV of 1,880×106 ft3, which is close
to the simulated SRV of 2,000×106 ft3. Well B has a lateral length
similar to that of Well A (2,600 ft).The fracture treatment for Well
B consisted of 830,000 lbm of 40/70 sand and 117,000 bbl of fluid
pumped in two stages with six perforation clusters, three per stage
(spaced 500 ft apart). The primary-fracture spacing for Well B is
500 ft. The MS mapping of Well B indicated an SRV of 2,017×106
ft3, which is also very close to the simulation model.
In Fig. 5, (k = 10−4 md) and Fig. 6 (k = 10−5 md), the primary-
fracture spacing is 600 ft for the simulated data (solid and dashed
lines). The secondary- or network-fracture spacing was varied from
50 to 600 ft. For the larger network-spacing case (600 ft apart),
the desorbed gas contributed approximately 8.5% of the total gas
production after 30 years when the reservoir permeability was 10−4
md, and only 6.9% after 30 years when the reservoir permeability
was 10−5 md. Figs. 5 and 6 show that the desorbed-gas contribution
is relatively small within the first 5 years of production and, thus,
not likely to have a material impact on production economics. If
Fig. 6—Impact of desorbed gas, Barnett shale (k = 10−5 md,
uniform conductivity network = 2 md-ft).
August 2010 SPE Reservoir Evaluation & Engineering
Fig. 5—Impact of desorbed gas, Barnett shale (k = 10−4 md,
uniform conductivity network = 2 md-ft).
the network-fracture spacing is smaller (50-ft square blocks), the
desorbed gas results in roughly 15% more production after 30
years (for both reservoir permeability cases). Thus, the impact of
desorption increases when the network-fracture spacing is smaller,
with most of the additional gas produced later in the well life. Other
studies have also reported that desorption does not play a major
role in the production profile or gas recovery in the Barnett shale
(Frantz et al. 2005).
The production profile for Well B follows the simulation
results for the larger network block size (Dx = 600 ft) and matrix
permeability of 10−4 md relatively closely, indicating that fracture
complexity in Well B may be low. The production profile for Well
A follows a similar but somewhat higher trend, possibly indicating
a more-complex fracture network. It should be emphasized that
the production comparisons for Wells A and B are qualitative only
and are not intended to be detailed “history matches.” However,
they provide actual production data from Barnett horizontal wells
with SRV and primary-fracture spacing similar to that used in
the reservoir simulations and could be considered “approximate”
history matches.
Fig. 7 shows pressure distribution after 1, 5, and 15 years for
the two different reservoir-permeability cases. Because of the sym-
metry of the reservoir area investigated, only one-half of the total
reservoir size is simulated. The primary fractures extend 1,000
ft from the wellbore (half-length). The fracture conductivity is
2 md-ft for both the primary and network fractures (uniform con-
ductivity). The pressure distribution includes the effect of desorp-
tion. Dark red indicates initial reservoir pressure, while dark blue
indicates flowing bottomhole pressure (FBHP) (1,000 psi). The
left side of the graph shows the pressure distribution for a matrix
permeability of 10−5 md, primary fracture spacing of 300 ft, and
a network block size (Dx) of 300 ft. When matrix permeability is
very low and network-block size is large, pressures in the matrix
blocks remain relatively high even after 15 years of production.
The right side of the graph shows the pressure distribution for a
matrix permeability of 10−4 md, primary fracture spacing of 100 ft,
and network-block size of 50 ft, showing that when network-block
size is small (i.e., more-complex fracture network), the tight matrix
can be effectively drained after 15 years. However, because of the
relatively high FBHP and the Barnett desorption isotherm (Fig. 4),
very little gas is desorbed from the shale even with efficient drain-
age of the tight matrix.
Impact of FBHP. The impact of FBHP is illustrated in Figs. 8
and 9. Figs. 8 and 9 compare the 30-year gas recovery with and
without gas desorption for a FBHP of 500 and 1,000 psi, a typi-
cal range for many Barnett shale wells. The simulations in Fig. 8
assume a matrix permeability of 10−4 md and a network-fracture
spacing of 600 ft, while the simulations in Fig. 9 assume a matrix
permeability of 10−5 md and a network-fracture spacing of 100 ft.
The primary-fracture spacing is 600 ft, and a uniform network
641
 k = 1e-5 mD, primary spacing = 300 ft, Dx = 300 ft k = 1e-4 mD, primary spacing = 100 ft, Dx = 50 ft

1 year

5 years

15 years

Fig. 7—Pressure distribution after 1, 5, and 15 years for the Barnett shale example (with desorbed gas).

conductivity of 2 md-ft is used in the simulations in Figs. 8 and 9.
Fig. 9 shows that reducing the FBHP from 1,000 to 500 psi results
in increases in cumulative gas production of up to 10% after 30
years for the case with 10−4 md matrix permeability (Fig. 8) and
12.5% after 30 years of production for the case with 10−5 md matrix
permeability (Fig. 9). Comparing the results with desorbed-gas
production (solid lines) to those without desorbed-gas production
(dashed lines) in Figs. 8 and 9 shows that reducing the FBHP from
1,000 to 500 psi does not significantly increase the production of
desorbed gas. In addition, reviewing the production profiles for
the first 5 years shows that reducing the FBHP from 1,000 to 500
psi has very little impact on the early-time production profile,
especially in the first 3 years. However, the impact of reducing
the FBHP on gas production is greater for more-complex fracture
networks (i.e., smaller blocks).
Comparing the production profile from Well A to the simulation
results in Fig. 9 shows that Well A follows the production profile
for a matrix permeability of 10−5 md and a network-fracture spacing
of 100 ft. This is not a unique solution, because other combina-
tions of matrix permeability and network-fracture spacing may
result in similar production profiles, but the modeling may indicate
that Well A has a more complex fracture network compared with
Well B. Non-unique solutions to production history matching
are common in shale-gas reservoirs because matrix permeability,
network-fracture spacing, and the conductivity of the network
and primary fractures are not known exactly. However, laboratory
core tests can be performed to estimate matrix permeability and
unpropped-fracture conductivity, constraining the solution and
allowing significant insights into well performance.
Barnett Shale: Impact of Stress-Dependent Network-Fracture
Permeability. The impact of increasing closure stress on propped-
fracture conductivity has been studied for decades, but little atten-
tion has been given to the conductivity of unpropped hydraulically
created fractures until recently. There is a likelihood that large
portions of the fracture networks created during shale-gas stimula-
tion treatments may be unpropped or ineffectively propped (Fredd
et al. 2001; Cipolla et al. 2008a). Therefore, unpropped- or partially-
propped-fracture conductivity may play a significant role in shale-
gas production. The impact of primary- and network-fracture

Fig. 8—Impact of BHFP for the desorbed-gas case (k = 10−4 Fig. 9—Impact of BHFP for the desorbed-gas case (k = 10−5
md, Dx = 600 ft). md, Dx = 100 ft).

642 August 2010 SPE Reservoir Evaluation & Engineering

Fig. 10—Effect of closure stress and Young’s modulus on
unpropped-fracture conductivity.
conductivity on well performance was presented in previous work
(Cipolla et al. 2008a, 2009a), showing that production in many
shale-gas reservoirs may be limited by insufficient fracture con-
ductivity. Laboratory data showing the effect of closure stress on
unpropped- and partially-propped-fracture conductivity for rock
with relatively high Young’s modulus was published by Fredd
et al. (2001). Fredd’s laboratory data were extrapolated to lower
Young’s modulus rock for unpropped fractures with shear offsets
to illustrate the impact of modulus on unpropped-fracture conduc-
tivity (Cipolla et al. 2008a). These data and this approach were
used to evaluate stress-sensitive network-fracture conductivity in
the Barnett shale. Fig. 10 shows an estimation of the impact of
closure stress and Young’s modulus on unpropped-fracture con-
ductivity, illustrating the dramatic decrease in fracture conductivity
as closure stress increases and Young’s modulus decreases. In this
example, a reference conductivity of 10 md-ft at 1,200-psi closure
stress was used for all cases. The reference conductivity of 10 md-
ft was found to be a reasonable estimate for the Barnett shale (E ≈
5 10+6 psi) on the basis of comparisons to actual well performance,
but would actually be a history-matching parameter and could
vary considerably depending on stimulation designs, completion
strategies, and geologic variability. The same reference conductiv-
ity was used for lower-modulus rock to allow easy comparison of
the simulation results for lower-modulus rock and/or the impact of
more-severely stress-dependent fracture conductivity.
Closure stress on the fracture is defined as the horizontal stress
perpendicular to the fracture (h) minus the pressure inside the
fracture (pf). For simplicity, the two horizontal stresses are assumed
to be equal. In the Barnett shale, a horizontal stress of 5,000 psi
Fig. 12—Impact of stress-dependent fracture conductivity for
the desorbed-gas case (k = 10−5 md).
August 2010 SPE Reservoir Evaluation & Engineering
Fig. 11—Impact of stress-dependent fracture conductivity for
the desorbed-gas case (k = 10−4 md).
was used to calculate the closure stress on the network fractures.
The initial closure stress in the Barnett example is 1,200 psi before
production (i.e., no drawdown), with maximum closures stress of
4,000 psi when the pressure in the network fractures reaches the
FBHP of 1,000 psi. In reference to Fig. 10, the conductivity of the
network fractures is approximately 0.55 md-ft when closure stress
is 4,000 psi, almost a 20-fold decrease from the initial conductiv-
ity of 10 md-ft.
Figs. 11 and 12 show the impact of stress-dependent fracture
conductivity on well performance (including desorbed gas). In
these cases, the primary-fracture spacing is 600 ft and network-
fracture spacing is varied from 50 to 200 ft, while matrix perme-
ability is varied from 10−5 md to 10−4 md. The solid lines represent
the cases without stress effect and with 2 md-ft constant fracture
conductivity, while the dashed lines represent the cases with stress
effect (assuming E = 5 10+6 psi) and 10-md-ft initial conductivity
(Fig. 10). In this case, the impact of variable closure stress in the
network fractures reduces the ultimate gas production by approxi-
mately 8 to 10% and may alter the production profile.
The production profiles for different fracture spacings show that
stress-dependent fracture conductivity results in materially higher
initial production rates when block sizes are small, while there
is very little impact on the initial production profile when block
sizes are larger. In all cases, stress-dependent fracture conductiv-
ity reduces ultimate gas production, as drawdown in the fracture
network continues to increase throughout the well life and network
fractures “close.” However, the impact of changing closure stress
on well production may not be critical in the Barnett shale owing
to the relatively high Young’s modulus. However, additional res-
ervoir-modeling work is required to better understand the impact
of stress-dependent fracture conductivity.
Comparing the production profile for Well B to the simulation
results in Figs. 11 and 12 shows that Well B follows the produc-
tion trend for the case with stress-dependent fracture conductivity,
10−4 md matrix permeability, and 600-ft network-fracture spacing
slightly better than the same case without stress-dependent fracture
conductivity (Fig. 11, lower set of curves). More-detailed history
matching should provide a better, but still not unique, match for
both wells.
Impact of Young’s Modulus and Severity of Stress-Dependent
Fracture Conductivity. As Young’s modulus decreases (i.e., softer
rock), the impact of increasing closure stress on unpropped- or par-
tially-propped-fracture conductivity will likely be more severe. An
approximation of the impact of Young’s modulus on unpropped-
fracture conductivity is shown in Fig. 10, illustrating the dramatic
decrease in unpropped-fracture conductivity that is likely for softer
rocks. A series of reservoir simulations was performed using the
Barnett shale reservoir properties and the fracture-conductivity
profiles in Fig. 10. Figs. 13 and 14 show the simulation results
643
Fig. 13—Impact of Young’s modulus on stress-dependent frac-
ture conductivity for the desorbed-gas case (k = 10−4 md).
for matrix permeability of 10−4 and 10−5 md, respectively. The
simulations assume a primary- or main- fracture spacing of 600 ft
and a network-fracture spacing of 100 ft (k=10−4 md) and 50 ft
(k=10−5 md). The cumulative production for a constant and uni-
form 2-md-ft network-fracture conductivity and is shown for
comparison. The results show that more-severe stress-dependent
fracture conductivity can reduce gas recovery significantly. When
Young’s modulus is 4×106 psi, stress-dependent fracture conduc-
tivity reduces gas recovery by 11 to 14% in comparison with the
constant-conductivity case. For lower-modulus rock (2×106 psi),
gas recovery decreases by 40 to 43% compared to the E=5×106 psi
case. At lower modulus, the fracture conductivity is reduced sig-
nificantly as drawdown in the fracture network increases with time,
resulting in poor drainage of the tight matrix rock and significantly
lower ultimate gas recovery. However, initial well performance is
not markedly different for the various cases, and the severity of
stress-dependent fracture conductivity may not be evident during
the first 1–2 years of production, requiring 3–5 years of produc-
tion history to better define the possible effects of stress-dependent
network-fracture conductivity.
The production profile for Well A follows the trend for a matrix
permeability of 10−5 md and a fracture-network spacing of 50 ft
(lowest curve in Fig. 14) very closely, while also following the
trend for a matrix permeability of 10−4 and network-fracture spac-
ing of 100 ft (lowest curve in Fig. 13). This illustrates some of the
issues with potentially non-unique solutions when history matching
shale-gas production profiles—because fracture conductivity, matrix
permeability, and network-fracture spacing are many times not well
known. However, reservoir modeling can provide important insights
into production mechanisms and stimulation performance when
values of matrix permeability and unpropped-fracture conductivity
can be bounded using laboratory data. It may be possible to estimate
network-fracture spacing and network structure using MS-mapping
data, further constraining the history match.
Summary of Barnett Shale-Gas Desorption and Stress-Sensi-
tive-Permeability Simulations. The simulation study using typical
Barnett shale reservoir and rock properties indicated that gas
desorption and stress-dependent fracture conductivity probably
have a minor, but not necessarily insignificant, impact on well
performance. The impact of gas desorption is primarily in the
later life of the well when pressures in the tight matrix blocks may
become low enough to produce meaningful amounts of gas, result-
ing in an increase in 30-year gas recovery of 5–15%. However, as
network-block size increases and fracture conductivity and matrix
permeability decrease, the ability to produce adsorbed gas becomes
increasingly difficult. While decreasing FBHP from 1,000 to 500
psi does not significantly increase the production of adsorbed
gas, it will improve 30-year gas recovery by approximately 10%
(mostly because of the additional production of free gas). Because
of the relatively high modulus of the Barnett shale, the impact of
644
Fig. 14—Impact of Young’s modulus on stress-dependent frac-
ture conductivity for the desorbed-gas case (k = 10−5 md).
stress-dependent network-fracture conductivity does not appear
to be significant, mostly affecting the mid- to late-life production
behavior, and could reduce gas recovery by approximately 10%.
However, if the relationship between closure stress and unpropped-
fracture conductivity is more severe than that shown in Fig. 10
(E = 5×106 psi curve), then ultimate gas recovery could be affected
materially.
Marcellus Shale: Impact of Stress-Dependent Network-Fracture
Permeability. The impact of gas desorption and stress-dependent
fracture conductivity was also evaluated for typical Marcellus shale
reservoir properties, although the Marcellus is more variable in
depth and thickness and much earlier in its development than the
core area of Barnett—making it more difficult to define average
or generic properties. In addition, actual production data were not
available to “calibrate” the Marcellus simulations. Table 2 lists the
generic reservoir properties used for the Marcellus shale. The area
of the reservoir model is the same as that used in the Barnett shale,
3,400×2,400 ft (≈190 acres), and a cased-and-cemented horizontal
completion with multiple fracture-treatment stages is assumed. The
SRV for the Marcellus simulations is 680×106 ft3 (100 ft of pay).
Fig. 15 shows a graph of gas content (scf/ton) vs. pressure
for the gas in the matrix porosity and the adsorbed gas used for
the Marcellus shale (Faraj and Duggan 2008). The free gas in the
matrix porosity is shown as a nearly straight line (circles), while
the adsorbed gas is represented by a curved line (squares). In this
case, 46.5% of the total gas is adsorbed, a percentage similar to
that in the Barnett shale. At higher pressures, more gas is stored
in the matrix porosity and the gas porosity will contribute more
to production than gas desorption because of the difference in the
curve shapes (straight line vs. curved line).
The Marcellus simulations include the effect of stress depend-
ent network-fracture conductivity. Assuming uniform and constant
TABLE 2—GENERIC MARCELLUS SHALE PROPERTIES
Depth (ft) 5,000
Reservoir pressure (psi) 2,750
Pore pressure gradient (psi/ft) 0.55
Closure pressure (psi) 3,500
Closure pressure gradient (psi/ft) 0.70
Net pay (ft) 100
Porosity (% ) 8
Matrix permeability (md) 1 E 4; 5 E 4
Water saturation (%) 25
Reservoir temperature (°F) 105
FBHP (psi) 500
August 2010 SPE Reservoir Evaluation & Engineering

Fig. 15—Example Marcellus shale-gas content, adsorbed and
free gas.
network-fracture conductivity may be optimistic for lower-modu-
lus shale, such as the Marcellus (E≈2×106 psi). Fig. 16 shows an
estimate of the effect of closure stress on unpropped-fracture con-
ductivity for a Young’s modulus of 2 MMpsi. It should be empha-
sized that this is only an estimate based on previously published
work (Fredd et al. 2001; Cipolla et al. 2008a) and revised on the
basis of Barnett shale history-matching results, which indicated
that the previously published estimates may be optimistic. The
initial network-fracture conductivity is 2 md-ft before production,
but declines to 0.02 md-ft when the pressure in the network frac-
tures decreases to the FBHP of 500 psi.
The Marcellus simulations assume that a higher-conductivity
primary fracture is present with an initial conductivity of 50 md-ft.
The conductivity of the primary fracture is also stress dependent,
decreasing to approximately 0.25 md-ft at 3,000-psi closure stress
(when the pressure in the primary fracture is 500 psi). The impact
of a high-conductivity primary fracture will be discussed later in
this section.
Figs. 17 and 18 show the impact of gas desorption on cumula-
tive gas production. In Fig. 17 (k = 10−4 md) and Fig. 18 (k = 5×10−5
md), the secondary or network-fracture spacing is fixed (100 ft)
and network fracture conductivity is 2 md-ft. The primary-fracture
spacing (Dm) is 100 and 300 ft. For the smaller primary-spacing
case (100 ft apart), the desorbed gas contributes approximately
10% of the total production after 30 years (for both matrix-perme-
ability cases). As was seen in the Barnett example, these graphs
show that the desorbed gas contribution is relatively small within
the first 5 years of production. If the primary-fracture spacing is
larger (300 ft apart), the desorbed gas results in only 8% more
production after 30 years (for both matrix-permeability cases).
Fig. 17—Impact of desorbed-gas production in the Marcellus
shale (k = 10−4 md).
August 2010 SPE Reservoir Evaluation & Engineering
10
E=2E+6 psi
Conductivity (mD-ft)
1
2 mD-ft
0.02 mD-ft
0.1
0.01
Initial closure stress Closure stress at 500 psi FBHP
0.001
0 500 1000 1500 2000 2500 3000 3500 4000
Stress (psi)
Fig. 16—Effect of closure stress on unpropped-fracture con-
ductivity, Marcellus shale example.
Thus, the impact of desorption decreases when the primary-frac-
ture spacing is larger or farther apart (i.e., fewer fracture-treatment
stages along the lateral). Free gas (porosity) is likely the dominant
contributor to production in the Marcellus shale.
Fig. 19 shows pressure distribution after 1, 5, and 15 years for
the two different matrix-permeability cases. Because of the sym-
metry of the reservoir area investigated, only one-half of the total
reservoir size is simulated. The primary fractures extend 1,000 ft
from the wellbore (half-length). The fracture conductivity is 2 md-
ft for both the primary and network fractures (uniform conductiv-
ity). The pressure distribution includes the effect of desorption and
stress-dependent fracture conductivity. Dark red indicates initial
reservoir pressure, while dark blue indicates FBHP (500 psi). The
left side of the graph shows the pressure distribution for a matrix
permeability of 5×10−4 md, primary-fracture spacing of 300 ft, and
a network block size (Dx) of 300 ft. When matrix permeability is
very low and network block size is large, pressures in the matrix
blocks remain relatively high even after 15 years of production.
The right side of the graph shows the pressure distribution for a
matrix permeability of 10−4 md, primary-fracture spacing of 100 ft,
and network-block size of 50 ft, showing that when network-block
size is small (i.e., more-complex fracture network), the tight matrix
can be effectively drained after 15 years. In this comparison, matrix
permeability is five times lower for the case with 50-ft network
blocks in comparison to the 300-ft case and gas recovery is still
significantly higher (as evidenced by the lower pressures). In many
shale-gas reservoirs, gas recovery may be more dependent on the
size of the network blocks (degree of fracture complexity) than
on the permeability of the matrix rock. However, because of the
relatively high FBHP (500 psi) and the behavior of the Marcellus
desorption isotherm (Fig. 15), very little gas is desorbed from the
shale even with efficient drainage of the tight matrix.
Fig. 18—Impact of desorbed-gas production in the Marcellus
shale (k = 5×10−4 md).
645
 k = 5e-4 mD, primary spacing = 300 ft, Dx = 300 ft k = 1e-4 mD, primary spacing = 100 ft, Dx = 50 ft

1 year

5 years

15 years

Fig. 19—Pressure distribution after 1, 5, and 15 years for the generic Marcellus shale example (with desorbed gas).

Effect of Stress-Dependent Fracture Conductivity. The impact 10% additional gas production in 30 years, while having little effect
of stress-dependent conductivity is shown in Fig. 20 for a matrix on the initial production profile. Therefore, desorbed-gas produc-
permeability of 5×10−4 md, primary-fracture spacing of 100 ft, tion is probably a minor component in the economic development
and a network-fracture spacing of 100 ft. All simulations include of the Marcellus shale. In contrast to the Barnett shale, the lower
gas desorption. The cumulative production for a constant 2-md-ft Young’s modulus of the Marcellus could mean that stress-depen-
uniform-conductivity network is shown by the top curve, indicating dent network-fracture conductivity will play a significant role in
a 30-year gas recovery of 8,342 MMscf. The bottom curve (dashed well performance and could substantially reduce initial production
line) shows the production for a uniform-conductivity network rates and ultimate gas recovery. However, the detrimental effects
with stress-dependent fracture conductivity (initial conductivity of stress-dependent network-fracture conductivity may be offset
of 2 md-ft, declining to 0.02 md-ft when the pressure in a net- by stimulation designs that result in high-relative-conductivity
work fracture reaches 500 psi, Fig. 16). The 30-year gas recovery primary fractures.
is reduced significantly when stress-dependent network-fracture
conductivity is modeled, decreasing by almost 60%. Vertical-Well Case Study
The detrimental effects of stress-dependent network-fracture The example well is a vertical well located in the core area of the
permeability can be offset, to some degree, by designing stimulation Barnett shale; it was fracture treated and mapped in 2001 (May-
treatments that create a high-relative-conductivity primary fracture. erhofer et al. 2006). MS mapping clearly illustrates that very large
The presence of a high-relative-conductivity primary fracture fracture networks are generated, with the predominant hydraulic-
significantly reduces network-fracture conductivity requirements fracture azimuth in the northeasterly direction and a secondary
(Cipolla et al. 2008a, 2009a). The impact of a primary fracture
with an initial conductivity of 50 md-ft is also shown in Fig. 20,
middle curve with higher initial production and 7200×106 ft3 cumu-
lative gas production. It should be emphasized that the conductiv-
ity of the primary fracture is also stress dependent, declining to
0.25 md-ft when the pressure in the primary fracture reaches
500 psi. The cumulative gas production after 30 years is roughly
doubled for the case with a high relative conductivity primary
fracture in comparison with the 2-md-ft uniform-fracture-conduc-
tivity case (with stress-dependent network-fracture conductivity)—
essentially offsetting the detrimental affects of stress-dependent
network-fracture conductivity. It is interesting to note that the case
with a high-relative-conductivity primary fracture (Fig. 20, middle
curve) exhibits much higher initial gas-production rates compared
to the 2-md-ft constant- and uniform-conductivity case (Fig. 20,
top curve), yet 30-year cumulative gas production is approximately
15% lower. These simulations emphasize the importance of under-
standing the stress dependence of network-fracture conductivity in
fracture design and production forecasting.
Summary of Marcellus Shale Gas-Desorption and Stress-
Sensitive-Permeability Simulations. The impact of desorbed-gas Fig. 20—Impact of stress-dependent fracture conductivity, with
production in the generic Marcellus reservoir-simulation cases was and without a high-relative-conductivity primary fracture, Mar-
similar to that for the Barnett shale, possibly adding approximately cellus shale example.

646 August 2010 SPE Reservoir Evaluation & Engineering


South/North (ft)
Fracture-Network Model
MS mapping
West/East (ft)
Fig. 21—Fracture network as determined from MS mapping
is approximated with numerical network model (Mayerhofer
et al. 2006).
component orthogonal to that. One of the remaining questions
concerns how much of the generated network is actually effective
during production. Fig. 21 shows the MS-mapping results. The
total extent of the fracture network is at least 2,700 ft, with apparent
asymmetric growth to the southwest. It appears that the northeast
wing of the network structure did not develop as effectively, on the
basis of created MS events (Note: The observation well was in a
position that should have detected growth on the northeast side).
The network width is fairly small in this case (approximately 350
ft), and network height shows adequate coverage of the Lower
Barnett interval. The network structure (bold lines) in Fig. 21
represents the effective network modeled in the reservoir simulator
as an approximation of the MS results. This is the final network
structure used to history match well production and is a result of
several iterations that explored alternative answers. In short, the
conclusion is that once matrix permeability is fixed, the resulting
total effective network length (sum of all open segments) and
conductivity are fairly unique.
Table 3 shows the reservoir input parameters used in the simu-
lation. Three parameters were varied to obtain the history match:
fracture length, fracture density, and fracture conductivity. Actual
Wellhead Pressure (psi)
Time (days)
Fig. 22—History match of flowing tubing and casing pressures,
open-ended tubing (Mayerhofer et al. 2006).
August 2010 SPE Reservoir Evaluation & Engineering
TABLE 3—EXAMPLE-WELL RESERVOIR PARAMETERS
Parameters V a lu e
Depth (ft) 7,000
Net thickness (ft) 415
Porosity (% ) 6
Water saturation (%) 30
Initial pressure (psi) 3,800
Temperature (°F) 180
Gas gravity 0.6
gas rate was used as a constraint (Fig. 22). The total well history
includes 4 years of production. Fig. 23 shows the history match of
flowing tubing and casing pressures vs. time. The model predicts
higher wellhead flowing pressures in late time, which tracks the
surface annulus pressure (open-ended tubing) and represents liquid
loading. The loading condition was confirmed by higher flow rates
after a foam cleanout. Although this match is not entirely unique,
the effective total-fracture-network length (sum of all fracture seg-
ments open to flow in the two principal frac directions) is on the
order of 6,000 to 7,800 ft, with an average conductivity of 4 md-ft
(matrix permeability was fixed at 0.0001 md or 100 nanodarcies,
which was estimated from core tests). The effective network width
was modeled to be 180 ft (90-ft fracture spacing in the northeast
direction and four orthogonal stringers) with a network extent of
approximately 2,000 ft in the southwest direction. While the total
effective network length (sum of all open segments) must fall
within the 6,000- to 7,800-ft range, the aspect ratio of effective
network extent to width could vary to some degree. An alternative
match, which would also be within the MS-mapped network area,
was achieved with a wider 360-ft network (180 ft fracture spacing)
and shorter 1,300-ft extent in the southwest direction. Decreasing
the matrix permeability by an order of magnitude (to 10 nanodar-
cies) would result in a similar pressure history match; however, the
effective fracture-network length would increase to approximately
16,500 ft and result in closer fracture spacing of 60 ft with 360-ft
network width and 2,700-ft total extent. The exact position, aspect
ratio of effective network width to length, and fracture spacing
within the network can be uniquely determined only with the help
of multiwell production data.
Fig. 24 shows the pressure distribution after 1 year of produc-
tion. Red represents initial reservoir pressure, and dark blue is
approximately the FBHP. Fig. 24 illustrates that nanodarcy matrix
permeability will not drain very far beyond the created fracture
network. Additionally, low fracture conductivities play an impor-
tant role because they create higher pressure drops in the network
resulting in less-efficient drainage farther from the well.
Fig. 23—Gas-flow rate was used as a constraint for production
modeling (Mayerhofer et al. 2006).
647
 Conductivity=
4 md⋅ft

Fig. 24—Pressure distribution after 1 year of production (May-

erhofer et al. 2006).

Fig. 25—Blasingame analysis, Barnett shale (xf = 7,000 ft).

Advanced Decline-Curve Analysis. The production data for this
well were evaluated in Cipolla et al. (2008b). The Blasingame
production type-curve analyses of the example well are shown in
Figs. 25 and 26, illustrating the non-uniqueness that is possible in
unconventional reservoirs when reservoir permeability is uncertain
and fracture growth is not measured. Figs. 25 and 26 show two
matches to the production from the referenced Barnett shale well: one
indicating an effective fracture half-length of 7,000 ft, matrix perme-
ability of 5×10−6 md, and a drainage area of 69 acres and a second
match indicating an effective fracture half-length of only 130 ft, matrix
permeability of 4×10−3 md, and a drainage area of 8.5 acres.
Although constraining the matrix permeability would result in
less uncertainty in the type-curve match, the examples illustrate
how the assumption of a single planar fracture and the inability
to describe the network-fracture system limit the application of
type-curve analyses in unconventional gas reservoirs. It can be very
important when evaluating production data to determine, even qual-
itatively, the effects of matrix permeability and primary-fracture
conductivity (plus the impact of network size, fracture complexity,
and network-fracture conductivity) on well performance.
The Barnett shale is an example in which selecting the appro-
priate diagnostic technologies was the key to success. In this case,
fracture growth is too complex to be modeled using commercial
fracture models; therefore, treatment designs are based on the inte-
gration of MS data and well-performance evaluations. To evaluate
well performance properly requires numerical reservoir modeling
that captures the details of the complex fracture network, which
limits the application of type-curve production-data-analysis tech-
niques. The combination of MS mapping and numerical reservoir
modeling resulted in significant insights into stimulation designs
and completions strategies, leading to the application of large-
volume high-rate water-frac treatments with low concentrations
of small proppants and horizontal wells to maximize the fracture-
network size and complexity.
economics in many moderate-to-deep shale-gas plays may be
insignificant.
3. The impact of stress-dependent network-fracture conductivity
in higher-Young’s-modulus shale reservoirs, such as the Barnett
shale, may be small. In these hard-rock shale-gas reservoirs,
reductions in network-fracture conductivity as closure stress
increases could reduce ultimate gas recovery by 10%, but may
not have a material impact on initial well productivity.
4. Production from shale-gas reservoirs with lower Young’s modulus,
such as the Marcellus and Haynesville shales, could be affected
significantly by stress-dependent network-fracture conductivity,
resulting in significantly lower gas recovery than anticipated on the
basis of initial well productivity. Although the impact of increasing
closure stress on network fractures can be severe for lower-modulus
rock, the effects may not be evident during the initial 1–2 years of
production, complicating production forecasting.
5. Fracture-treatment designs that create a high-relative conduc-
tivity primary fracture may offset, at least to some degree, the
detrimental effects of stress-dependent or low network-fracture
conductivity, resulting in significant improvements in produc-
tion rates and gas recovery.
6. Recovery of adsorbed gas may be improved if more-complex
fracture networks can be created (i.e., smaller network blocks).
7. Numerical-reservoir-simulation history matching can be non-
unique in complex shale-gas reservoirs. However, the modeling
can be reasonably constrained when estimates of matrix perme-
ability are available from special core analyses and MS-map-
ping data are gathered to determine SRV and provide insights
into fracture-network development. Combining these data with
laboratory measurements of unpropped- and partially-propped-
fracture conductivity can significantly improve the insights
gained from reservoir simulation.

Conclusions
The reservoir-simulation work focused on evaluating the effects of
gas desorption and stress-dependent network permeability in hori-
zontal completions using reservoir properties typical of the Barnett
and Marcellus shales, but the results and modeling approach should
be applicable to many other shale-gas reservoirs. The following
conclusions can be drawn from this work.
1. Gas desorption may not be a significant component of produc-
tion in many moderate-to-deep shale-gas reservoirs, such as the
Barnett, Marcellus, and Haynesville shales. Although adsorbed
gas may constitute 40–50% of the total gas in place, the ability
to produce the adsorbed gas is limited because of the ultratight
matrix rock, relatively high FBHP, and the desorption profile,
which requires relatively low pressures to produce significant
amounts of adsorbed gas.
2. Desorbed gas may account for 5–15% of ultimate gas produc-
tion, but the desorbed gas is produced primarily during the later
life of the well, having little impact on the initial well produc-
tivity. Therefore, the impact of gas desorption on development Fig. 26—Blasingame analysis, Barnett shale (xf = 130 ft).

648 August 2010 SPE Reservoir Evaluation & Engineering

8. Production-data analyses, such as advanced type-curve solutions,
cannot adequately describe the complex fluid flow in shale-
gas reservoirs. However, these techniques may have important
qualitative applications.
9. Discrete numerical modeling of the fracture network and tight
matrix rock is necessary to accurately simulate production
from many shale-gas reservoirs. Numerical-reservoir-simulation
approaches that use dual-porosity/dual-permeability approxima-
tions may not adequately capture the very long transient behav-
ior in the ultratight shale matrix blocks.
Nomenclature
Dm = primary-fracture spacing, L
Dx = network-fracture spacing, square blocks, L
E = Young’s modulus
Fc = fracture conductivity, L3
FCD = dimensionless fracture conductivity kf /kxf
h = net pay, L
k = matrix permeability, L2
kf = fracture permeability, L2
kwf = fracture conductivity, L3
pf = pressure in fracture
xf = hydraulic-fracture wing or half-length, L
xn = hydraulic-fracture-network width (from MS event pattern), L
h = horizontal stress
Acknowledgments
The authors would like to thank Carbo Ceramics and StrataGen
Engineering for supporting the publication of this work and Com-
puter Modelling Group for supporting the shale-gas reservoir-mod-
eling studies. The authors would also like to acknowledge Michael
Mayerhofer for his many contributions to shale-gas reservoir
modeling that were instrumental in this work.
References
Arthur, J.D., Bohm, B., and Layne, M. 2008. Hydraulic Fracturing Con-
siderations for Natural Gas Wells of the Marcellus Shale. Presented at
The Ground Water Protection Council 2008 Annual Forum, Cincinnati,
Ohio, USA, 21–24 September.
Cipolla, C.L., Warpinski, N.R., Mayerhofer, M.J., Lolon, E.P., and Vincent,
M.C. 2008a. The Relationship Between Fracture Complexity, Reser-
voir Properties, and Fracture Treatment Design. Paper SPE 115769
presented at the SPE Annual Technical Conference and Exhibition,
Denver, 21–24 September. doi: 10.2118/115769-MS.
Cipolla, C.L., Lolon, E.P., and Mayerhofer, M.J. 2008b. Resolving Created,
Propped, and Effective Hydraulic Fracture Length. Paper IPTC 12147
presented at the International Petroleum Technology Conference, Kuala
Lumpur, 3–5 December. doi: 10.2523/12147-MS.
Cipolla, C.L., Lolon, E.P., Mayerhofer, M.J., and Warpinski, N.R. 2009a. The
Effect of Proppant Distribution and Un-Propped Fracture Conductivity on
Well Performance in Unconventional Gas Reservoirs. Paper SPE 119368
presented at the SPE Hydraulic Fracturing Technology Conference, The
Woodlands, Texas, USA, 19–21 January. doi: 10.2118/119368-MS.
Cipolla, C.L., Lolon, E.P., and Dzubin, B. 2009b. Evaluating Stimulation
Effectiveness in Unconventional Gas Reservoirs. Paper SPE 124843
presented at the SPE Annual Technical Conference and Exhibition,
New Orleans, 4–7 October. doi: 10.2118/124843-MS.
Cipolla, C.L., Lolon, E.P., Mayerhofer, M.J., and Warpinski, N.R. 2009c.
Fracture Design Considerations in Horizontal Wells Drilled in Uncon-
ventional Gas Reservoirs. Paper SPE 119366 presented at the SPE
Hydraulic Fracturing Technology Conference, The Woodlands, Texas,
USA, 19–21 January. doi: 10.2118/119366-MS.
Faraj, B. and Duggan, J. 2008. Marcellus Shale Gas Potential in the South-
ern Tier of New York. Presented at the Annual Canadian Society for
Unconventional Gas (CSUG) Conference, Calgary, 19–21 November.
Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A.,
and Dunn, K.P. 2004. Optimizing Horizontal Completion Techniques
in the Barnett Shale Using Microseismic Fracture Mapping. Paper SPE
90051 presented at the SPE Annual Technical Conference and Exhibi-
tion, Houston, 26–29 September. doi: 10.2118/90051-MS.
August 2010 SPE Reservoir Evaluation & Engineering
Fisher, M.K., Wright, C.A., Davidson, B.M., Goodwin, A.K., Fielder,
E.O., Buckler, W.S., and Steinberger, N.P. 2002. Integrating Fracture
Mapping Technologies to Optimize Stimulations in the Barnett Shale.
Paper SPE 77411 presented at the SPE Annual Technical Conference
and Exhibition, San Antonio, Texas, USA, 29 September–2 October.
doi: 10.2118/77441-MS.
Frantz, J.H. Jr., Williamson, J.R., Sawyer, W.K., Johnston, D., Waters, G.,
Moore, L.P., Macdonald, R.J., Pearcy, M., Ganpule, S.V., and March,
K.S. 2005. Evaluating Barnett Shale Production Performance Using an
Integrated Approach. Paper SPE 96917 presented at the SPE Annual
Technical Conference and Exhibition, Dallas, 9–12 October. doi:
10.2118/96917-MS.
Fredd, C.N., McConnell, S.B., Boney, C.L., and England, K.W. 2001.
Experimental Study of Fracture Conductivity for Water-Fracturing and
Conventional Fracturing Applications. SPE J. 6 (3): 288–298. SPE-
74138-PA. doi: 10.2118/74138-PA.
Kazemi, H. 1969. Pressure Transient Analysis of Naturally Fractured
Reservoir With Uniform Fracture Distribution. SPE J. 9 (4): 451–462;
Trans., AIME, 246. SPE-2156-A. doi: 10.2118/2156-A.
Mayerhofer, M.J., Lolon, E.P., Youngblood, J.E., and Heinze, J.R. 2006.
Integration of Microseismic Fracture Mapping Results With Numeri-
cal Fracture Network Production Modeling in the Barnett Shale.
Paper SPE 102103 presented at the SPE Annual Technical Conference
and Exhibition, San Antonio, Texas, USA, 24–27 September. doi:
10.2118/102103-MS.
Mayerhofer, M.J., Lolon, E.P., Warpinski, N.R., Cipolla, C.L., Walser, D., and
Rightmire, C.M. 2010. What Is Stimulated Reservoir Volume? SPE Prod
& Oper 25 (1): 89–98. SPE-119890-PA. doi: 10.2118/119890-PA.
Medeiros, F., Ozkan, E., and Kazemi, H. 2008. Productivity and Drainage
Area of Fracture Horizontal Wells in Tight Gas Reservoirs. SPE Res Eval
& Eng 11 (5): 902–911. SPE-108110-PA. doi: 10.2118/108110-PA.
Montgomery, S.L. 2004. Barnett Shale: A New Gas Play in the Fort Worth
Basin. IHS Energy Petroleum Frontiers 20 (1).
Warpinski, N.R., Mayerhofer, M.J., Vincent, M.C., Cipolla, C.L., and Lolon,
E.P. 2008. Stimulating Unconventional Reservoirs: Maximizing Network
Growth While Optimizing Fracture Conductivity. Paper SPE 114173
presented at the SPE Unconventional Reservoirs Conference, Keystone,
Colorado, USA, 10–12 February. doi: 10.2118/114173-MS.
Warren, J.E. and Root, P.J. 1963. The Behavior of Naturally Fractured
Reservoirs. SPE J. 3 (3): 245–255; Trans., AIME, 228. SPE-426-PA.
doi: 10.2118/426-PA.
Appendix A—Reservoir Simulation, Shale-Gas
Gridding, and Modeling Issues
Numerical modeling of gas production from hydraulically frac-
tured shale-gas wells has been studied to determine how best to
represent network and/or natural fractures (both inside and outside
the stimulated reservoir volumes) accurately and efficiently.
Modeling Single-Plane Propped Fractures in Single-Porosity
Reservoirs. This work extends previous work for modeling single-
plane propped fractures in single-porosity reservoirs (e.g., tight
gas sands) in which it was determined that transient, non-Darcy
flow could be accurately (in comparison to very finely gridded,
single-porosity reference models and analytical solutions) and
efficiently (run times of a couple of minutes vs. hours) modeled
with a finite-difference simulator using a locally refined grid: (a)
whose innermost row of cells representing the propped fracture
can be orders of magnitude larger in width than the actual propped
fracture width (i.e., ≈2 ft) as long as both the fracture permeability
and the non-Darcy-flow coefficient are scaled appropriately, and
(b) whose spacing perpendicular to the length of the propped
fracture is logarithmic. A graph of the grid design is shown in
Fig. A-1. A comparison of FBHP behavior with this grid design
(Fig. A-1) vs. a very finely gridded single-porosity reference
model and Gringarten’s infinite-conductivity type-curve solutions
is presented in Fig. A-2.
Modeling Fracture Networks. This section presents an extension
of previous work on single-plane propped vertical fractures in
vertical and horizontal wells to the modeling of networks of frac-
tures in shale-gas wells that have been completed with multistage
649

Fig. A-1—Grid design for single-plane vertical-propped-fracture simulation.
fracture treatments. Several options were evaluated to determine
which “grid design” and “porosity model” approach resulted
in the best agreement with very finely gridded, single-porosity
reference models, including the use of the multiple interacting
continuum (MINC) and the dual-permeability “dual-continuum”
models, while minimizing model run time.
Our work has determined that the best method for both accu-
rately and efficiently modeling transient, non-Darcy gas produc-
tion from hydraulically fractured horizontal shale-gas wells is
to (a) use the dual-permeability method to represent all network
fractures in both the unstimulated and stimulated volumes and
to (b) locally refine the grid in the stimulated volumes using the
same logarithmically spaced grid design as was developed for the
single-plane vertical-fracture work mentioned previously. This new
approach is henceforth referred to in this paper as the “DK-LS-
LGR method” (i.e., for dual permeability, logarithmically spaced,
locally refined grid) for modeling hydraulically fractured shale-gas
wells/reservoirs.
Fig. A-3 compares the grid designs within the stimulated vol-
umes for both the very-fine-grid reference model (a), which uses
43×43 LGR with 0.001-ft-wide grids representing each fracture
and the simplified model (b), which uses 9×9 LGR with 2-ft-wide
grids representing each fracture. Both the reference grid and the
simplified grid use logarithmic spacing to capture transient flow
behavior to and within the fractures. The gas-production rates from
these two grids are shown in Fig. A-4 and confirm that the simpli-
fied approach gives an excellent match to the reference model.
100.00
10.00
PD
1.00
0.10
0.01
0.0001
(a)
Fig. A-2—FBHP comparison between Fig. A-1 grid design and (a) a very-finely gridded single-porosity reference model and (b)
Gringarten’s infinite-conductivity type-curve analytical solution.
650
Detailed Gridding of Hydraulic Fracture (xf = 500 ft)
Hydraulic Fracture
The MINC approach for modeling the fracture network was
also evaluated but proved inadequate at early time, as can be seen
in Fig. A-5, which compares logarithmically and equally spaced
gridding for both MINC (with 4 levels of refinement) and the new
simplified 9×9 LGR grid with 2-ft-wide grids representing the
fractures. The MINC approach provided worse results when flow
was allowed from the shale surrounding the stimulated volumes
(not shown).
Modeling Variations in Fracture Intensity in the Stimulated
Volume. We extended the new DK-LS-LGR approach to account
for variable fracture intensity in the stimulated volume by (a)
using a base gridblock size throughout the model whose length is
set equal to the largest fracture spacing observed in a given situ-
ation, (b) applying logarithmic-spaced LGR within the simulated
volumes, and (c) varying the dual-permeability parameters within
the stimulated volumes to account for variable fracture intensity
(i.e., spacing) and effective permeability. Fig. A-6a depicts the
stimulated-volume region surrounding a horizontal shale-gas well
as is typically interpreted from MS data gathered during hydraulic-
fracturing treatments. The base grid length in this example is 200 ft
on a side and represents the coarsest fracture spacing observed in
this particular shale formation. Fig. A-6b shows how we apply log-
arithmically spaced, local grid refinement to the stimulated-volume
region, and also how we modify effective fracture permeability (the
green areas) and the fracture spacing (not shown) to represent vari-
able-intensity fracturing within the stimulated volumes. Fig. A-7
xe/xf=3
xe/xf=7
xe/xf=15
Infinite
IMEX Infinite
IMEX xe/xf-15
IMEX xe/xf-7
IMEX xe/xf=3
0.001 0.01 0.1 1 10 100 1000 10000
tDxf
(b)
August 2010 SPE Reservoir Evaluation & Engineering
 Reference Grid Simplified Grid
4.0 md-ft fractured shale LGR model 4.0 md-ft fractured shale LGR model
Permeability J (md) 2000-01-01 K layer: 1 Effective Permeability J (md) 2010-01-01 K layer: 1
10,099.80 10,099.90 10,100.00 10,100.10
9,800 9,900 10,000 10,100 10,200 10,300 10,400

-9,900
-10,099.90

-10,099.90

-10,000

-10,000
Network Fractures
Network Fractures
-10,100.00

-10,100.00
Producer1

-10,100

-10,100
Producer1

-10,200
-10,200
-10,100.10

-10,100.10

-10,300
-10,300
0.00 75.00 150.00 feet
0.00 25.00 50.00 meter

10,099.80 10,099.90 10,100.00 10,100.10 9,800 9,900 10,000 10,100 10,200 10,300 10,400

(a) ~0.45 ft (b) ~700 ft

Fig. A-3—Grid designs for (a) reference grid (43×43 LGR with 0.001-ft-wide grid for each fracture) and (b) simplified grid (9×9
LGR with 2-ft-wide grid for each fracture) representing network of propped fractures in stimulated volume of a hydraulically
fractured shale-gas well.

compares the production from a very-fine-grid single-porosity • Fractures of greater or variable intensity inside the stimulated
reference model and the DK-LS-LGR model (Fig. A-6) for shale volumes are modeled using scaled dual permeability properties
matrix permeability of (a) 0.0001 md and (b) 0.01 md, showing that (matrix and fracture permeability and fracture spacing).
the simplified gridding approach provides acceptable accuracy. • Outside the stimulated volume the grid is not refined.
The new approach to modeling hydraulically fractured shale- The advantages of the approach described in this paper
gas wells can be summarized as follows: for modeling hydraulically fractured shale-gas wells are as
• A dual-permeability approach has been validated for mod- follows:
eling the matrix and natural fractures both inside and outside of the • Dual-permeability behavior is accounted for both inside and
stimulated volumes in hydraulically fractured shale-gas wells. outside the stimulated volumes (important in some shale plays in
• The base fracture network in the stimulated volumes is mod- which unstimulated fracture permeability may be significant).
eled with 9×9 locally refined, logarithmically spaced grids in the • Variations in fracture intensity can be accommodated without
stimulated volumes. having to vary the base gridblock size.

Noflow Outside Stimulated Region

Pseudoization Test for Fine Grid
2.87e+6

2.37e+6 Fracture Pseudoized using 2.000 ft thick blocks

Fracture Modelled using 0.001 ft thick blocks
Gas Rate SC (ft3/day)

1.87e+6

1.37e+6

8.71e+5

3.71e+5
0 10 20 30 40 50 60
Time (day)

Fig. A-4—Gas-production-rate comparison between Fig. A-3 grid designs for the case of 0.0001-md matrix permeability.

August 2010 SPE Reservoir Evaluation & Engineering 651

 Noflow Outside Stimulated Region Noflow Outside Stimulated Region
First 30 Days Equally and Log Spaced MINC vs. Equally and Log Spaced LGR
Equally and Log Spaced MINC vs. Equally and Log Spaced LGR 3.00e+6
3.00e+6

2.50e+6
2.50e+6

Gas Rate SC (ft3/day)

Gas Rate SC (ft3/day)

2.00e+6
2.00e+6 9x9 LGR 2 ft Bf Other Cells Equally Spaced
9x9 LGR 2 ft Bf Log Scaled
Minc 4 Equally Spaced
1.50e+6 1.50e+6 Minc 4 Log Scaled

1.00e+6 1.00e+6
9x9 LGR 2 ft Bf Other Cells Equally Spaced
9x9 LGR 2 ft Bf Log Scaled
5.00e+5 Minc 4 Equally Spaced 5.00e+5
Minc 4 Log Scaled

0.00e+0 0.00e+0
0 10 20 30 0 1,000 2,000 3,000 4,000
Time (day)
Time (day)
(a) (b)

Fig. A-5—Comparison of MINC dual-porosity approach to 9×9 LGR with 2-ft-wide grid representing each fracture.

Fig. A-6—(a) Typical stimulated volume resulting from a hydraulic-fracture treatment in a shale-gas well, and (b) the DK-LS-LGR
grid design created to model gas-production rate accurately in a numerical reservoir simulator.

New Dual Perm Model (0.0001 mD Shale) New Dual Perm Model (0.01 mD Shale)
Using Variable Fracture Spacing to Represent Using Variable Fracture Spacing to Represent
Different Fracture Intensity Different Fracture Intensity
Representing 66 ft Fracture Spacing in 200 ft DK blocks Representing 66 ft Fracture Spacing in 200 ft DK blocks
4.00e+6 4.00e+6
Single Porosity Run (Explicit Modelling Single Porosity Run (Explicit Modelling
or Variable Fracture Spacing or Variable Fracture Spacing
DK Run (Using Fracture Spacing DK Run (Using Fracture Spacing
3.00e+6 Scaling or 200 ft Blocks) 3.00e+6
Gas Rate SC (ft3/day)
Gas Rate SC (ft3/day)

Scaling or 200 ft Blocks)

2.00e+6 2.00e+6

1.00e+6 1.00e+6

0.00e+0 0.00e+0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Time (day) Time (day)
(a) (b)

Fig. A-7—Gas-production-rate comparison between Fig. A-6 grid design and a very finely gridded single-porosity reference model
for matrix-permeability cases of 0.0001 and 0.01 md.

652 August 2010 SPE Reservoir Evaluation & Engineering

 and chemistry from the U. of Nevada-Las Vegas and a Masters
SI Metric Conversion Factors degree in petroleum engineering from the U. of Houston.
Elyezer Lolon is a reservoir engineer consultant for the Reservoir
acre × 4.046 873 E+03 = m2 Technology Group at Anadarko Petroleum Corporation, focus-
bbl × 1.589 873 E–01 = m3 ing on reservoir modeling studies for shale plays. Previously, he
ft × 3.048* E–01 = m worked for StrataGen Engineering and Pinnacle Technologies
ft2 × 9.290 304* E–02 = m2 in Houston and Chevron Pacific Indonesia, where he designed
ft3 × 2.831 685 E–02 = m3 and evaluated hydraulic fracturing treatments, analyzed diag-
nostic fracture injection tests, performed pressure transient and
gal × 3.785 412 E–03 = m3 production data analyses, and conducted integrated reservoir
lbm × 4.535 924 E–01 = kg studies. Lolon holds a BS degree in petroleum engineering from
psi × 6.894 757 E+00 = kPa Bandung Institute of Technology in Indonesia and MS and Ph.D.
degrees in petroleum engineering from Texas A&M U. J.C. Erdle
*Conversion factor is exact. is Computer Modelling Group’s vice president for software sales
and support for the USA and Latin America. He has 35 years
of industry experience, primarily in reservoir and production
Craig L. Cipolla is an engineering advisor for Schlumberger in engineering-related positions within the services and software
Houston, currently focusing on the application of MS-fracture segments of the E&P industry. Erdle holds BS (1971) and PhD
mapping to improve stimulation designs and field development (1974) degrees from Penn State in petroleum engineering. Barry
strategies. Before joining Schlumberger in 2009, his most recent Rubin is a senior staff engineer at Computer Modelling Group in
positions were VP of stimulation technology for Carbo Ceramics Calgary. He has worked in the area of simulator development
(2008–2009) and VP of engineering for Pinnacle Technologies at the Computer Modelling Group in Calgary for 20 years and
(1996–2008). Cipolla’s 29 years of worldwide experience at BP Canada, BP Exploration, and BP Research International for
includes the application of MS and tilt-meter fracture mapping a total of 12 years. Rubin’s research interests include the mod-
technologies, the design and evaluation of hydraulic fractur- eling of naturally fractured reservoirs and, more recently, the
ing treatments, tight gas reservoir engineering, integrated field modeling of non-Darcy gas flow in shales. He holds a Masters
studies, training, and supervising stimulation treatments. He was degree in chemical engineering from the U. of Calgary and
an SPE Distinguished Lecturer on hydraulic fracturing in 2005– a Bachelors degree in chemical engineering from McGill U. in
2006. Cipolla holds undergraduate degrees in engineering Montreal, Canada.

August 2010 SPE Reservoir Evaluation & Engineering 653

 Comparison of Fractured-Horizontal-Well
Performance in Tight Sand and Shale
Reservoirs
E. Ozkan, SPE, and M. Brown, SPE, Colorado School of Mines; R. Raghavan, SPE, ConocoPhillips (Retired);
and H. Kazemi, SPE, Colorado School of Mines
Summary
This paper presents a discussion of fractured-horizontal-well per-
formance in millidarcy permeability (conventional) and micro- to
nanodarcy permeability (unconventional) reservoirs. It provides
interpretations of the reasons to fracture horizontal wells in both
types of formations. The objective of the paper is to highlight the
special productivity features of unconventional shale reservoirs.
By using a trilinear-flow model, it is shown that the drainage
volume of a multiple-fractured horizontal well in a shale reservoir
is limited to the inner reservoir between the fractures. Unlike con-
ventional reservoirs, high reservoir permeability and high hydrau-
lic-fracture conductivity may not warrant favorable productivity
in shale reservoirs. An efficient way to improve the productivity
of ultratight shale formations is to increase the density of natural
fractures. High natural-fracture conductivities may not necessar-
ily contribute to productivity either. Decreasing hydraulic-fracture
spacing increases the productivity of the well, but the incremental
production gain for each additional hydraulic fracture decreases.
The trilinear-flow model presented in this work and the informa-
tion derived from it should help the design and performance predic-
tion of multiple-fractured horizontal wells in shale reservoirs.
Introduction
The objective of hydraulic fracturing in conventional tight reser-
voirs (as in tight sands with permeabilities in the milli- to micro-
darcy range) has been to create a high-conductivity flow path to
improve flow convergence in the reservoir. Accordingly, a practical
interpretation of the objective of fracturing a horizontal well is to
create a system whose long-term performance is identical to that
of a single effective (total) fracture of length equal to the spacing
between the outermost fractures (Raghavan et al. 1997; Chen and
Raghavan 1997). Fig. 1 provides a sketch of this interpretation.
With this interpretation, performances of fractured horizontal wells
can be correlated in terms of an effective fracture conductivity
(CF,eff) and effective fracture half-length (xF,eff). The conductivity of
this effective fracture depends on the permeability of the reservoir
and the number of, distance between, and conductivities of the
individual hydraulic fractures.
Raghavan et al. (1997) have presented practical results to evalu-
ate the long-term performances of multiple-fractured horizontal
wells in low-permeability, conventional formations. Assuming
equally spaced fractures of identical properties, they correlated
the long-term performances (after the onset of pseudoradial flow)
of multiple-fractured horizontal wells in terms of the effective
wellbore radius of individual fractures (normalized by the dis-
tance between the outermost fractures, rwj/D) vs. effective (total)
system conductivity (CFD,eff). The Raghavan et al. (1997) correla-
tion, shown in Fig. 2, is similar to the correlation presented by
Prats (1961) and Cinco-Ley and Samaniego-V. (1981) to obtain
the effective wellbore radius of a single fracture as a function of
fracture conductivity and half-length.
Copyright © 2011 Society of Petroleum Engineers
This paper (SPE 121290) was accepted for presentation at the Western North American
Regional Meeting, Anaheim, California, USA, 26–30 May 2010, and revised for publication.
Original manuscript received for review 26 January 2009. Revised manuscript received for
review 11 October 2010. Paper peer approved 11 October 2010.
248
As shown in Fig. 2, the Raghavan et al. (1997) correlation
includes multiple curves for different numbers of fractures along
the well. The top curve in Fig. 2 represents the correlation for a
single fracture (Prats 1961; Cinco-Ley and Samaniego-V. 1981).
Fig. 2 may be used in two ways. It can be used to determine the
effective (total) system conductivity (CFD,eff) for given individual
fracture properties (conductivity CFD and half-length xF), distance
between the outermost fractures (D), and the number of fractures
along the well (nF). Then, using the effective system conductivity
(CFD,eff) and half-length (xF,eff=D/2), effective wellbore radius (rw,eff)
of the (total) system is obtained from the single-fracture correla-
tion (the top curve in Fig. 2). Alternatively, we may use Fig. 2 to
estimate the conductivity of the individual fractures (CFD) required
for a desired effective wellbore radius (rw,eff) for a given D and n.
As noted by Chen and Raghavan (1997), the total effective
wellbore radius can be used in the following well-known deliver-
ability equation:
qsc =
7.08 × 10 −3 kh (
p − pwf ) . . . . . . . . . . . . . . . . . . . (1)
B ⎛1 4A ⎞
ln
⎜ 2 e C r 2 ⎟
⎝ A w ,eff ⎠
Then, using Fig. 2 and Eq. 1, it is possible to design a multiple-
fractured horizontal well to accomplish the desired productivity.
For example, for given hydraulic-fracture properties, it is possible
to optimize the number of fractures along a given horizontal-well
length. Similarly, the optimum number of fractures, the length of the
horizontal well, and the properties of the hydraulic fractures required
to drain a given reservoir area of A (with the shape determined by
the shape factor CA) can be found from Fig. 2 and Eq. 1.
It must be emphasized, however, that this approach is appropri-
ate for the long-term performances of multiple-fractured horizontal
wells dominated by flow beyond the tips of the fractures and the
horizontal well. In other words, pseudoradial-flow convergence
must have developed around the multiple-fractured horizontal well
for a constant effective wellbore radius to be defined. Pseudoradial
flow develops around a single infinite-conductivity vertical fracture
for times (Gringarten et al. 1974)
1.14 × 10 4 ct  x F2
t≥ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
k
For long fractures (large xF) and tight formations (small k), time
for the start of pseudoradial flow may be very long. For a mul-
tiple-fractured horizontal well, because xF in Eq. 2 corresponds to
xF,eff=D/2, the start of pseudoradial flow may be delayed even for
moderate-to-large permeabilities. Therefore, most of the produc-
tive life of a fractured horizontal well is dominated by transient
production from the stimulated volume between the hydraulic
fractures (Raghavan et al. 1997). In fact, pseudoradial flow may
not be applicable in unconventional tight reservoirs with micro- to
nanodarcy permeability (as in shale) in which the contribution of
the reservoir beyond the stimulated volume is usually negligible
(Fisher et al. 2004; Mayerhofer et al. 2005; Maxwell et al. 2009;
Medeiros et al. 2008). Under these conditions, flow convergence
is mainly in the linear direction perpendicular to the surfaces
April 2011 SPE Reservoir Evaluation & Engineering

Fig. 1—Multiple-fractured horizontal well in a conventional tight
formation and the effective (total) fracture concept.
of the hydraulic fractures and the long-term performance of a
fractured-horizontal-well/reservoir system can be represented by
that of a single hydraulic fracture. The length of the equivalent
hydraulic fracture is equal to the aggregate length of the hydraulic
fractures, and the conductivity of the equivalent fracture is equal
to the average of the conductivities of the individual fractures, as
shown in Fig. 3. The drainage volume around the equivalent frac-
ture is a parallelepiped with the length of the equivalent fracture
(2xF,eff=2nFxF) and the width equal to the average hydraulic-fracture
spacing (dF).
In the following sections, we will introduce a trilinear-flow
model based on the premise that the drainage volume is limited
to the reservoir volume between hydraulic fractures. In extremely
tight formations, such as shale, an invasive (natural or induced)
fracture network is required for the efficient drainage of the matrix.
This introduces additional flow characteristics without changing
the predominantly linear-convergence of flow from reservoir to
hydraulic fractures. Because the linear flow performances of
multiple-fractured horizontal wells in conventional and unconven-
tional tight formations can be evaluated along the same lines, we
will discuss them together. First, we will introduce the analytical
trilinear-flow model used in this work to simulate the performances
of multiple-fractured horizontal wells. Then we will use it for the
numerical computations and demonstration examples.
Trilinear-Flow Model
The basis of our trilinear-flow model is the premise that the
productive life of a multiple-fractured horizontal well in a tight
Fig. 3—Effective-fracture concept for a multiple-fractured hori-
zontal well in an unconventional tight (shale) reservoir.
April 2011 SPE Reservoir Evaluation & Engineering
r w /xf vsCCFFD
D
r wj /D
CFD,
FD,eff
,eff
Fig. 2—Effective wellbore radius vs. effective system conduc-
tivity for a multiple-fractured horizontal well in a conventional
tight reservoir [after Raghavan et al. (1997)].
formation is dominated by linear-flow regimes. Our interpretation
is similar to the ideas used for fractured vertical wells in commin-
gled reservoirs by Camacho-V. (1984), Bennett et al. (1985), and
Camacho-V. et al. (1987). As shown by the schematic in Fig. 4,
the trilinear-flow model couples linear flows in three contiguous
flow regions: the outer reservoir (denoted by the subscript O), the
inner reservoir between fractures (denoted by the subscript I), and
the hydraulic fracture (denoted by the subscript F). In this work,
we assume uniform distribution of identical hydraulic fractures
along the length of the horizontal well. (If fractures are not equally
spaced, then the trilinear model is applicable before the interaction
of the hydraulic fractures.) Although it is possible to incorporate
the individual hydraulic-fracture properties in the model by an
approach similar to that of Raghavan et al. (1997), the common
field practice of creating equally spaced hydraulic fractures of
similar properties justifies our assumption. Furthermore, unless
there is significant contrast in individual-fracture properties, our
model should closely approximate the performance of the system
when average fracture properties are used.
To simulate the performances of multiple-fractured horizontal
wells in shale, we allow the inner reservoir to be naturally frac-
tured. We use the dual-porosity idealization to represent the natu-
rally fractured medium. In this paper, we consider transient fluid
transfer from the matrix (m) to natural fractures (f) as idealized by
Kazemi (1969), de Swaan O (1976), and Serra et al. (1983). The
model, however, is general and can be used with the pseudosteady-
fluid-transfer formulation of Warren and Root (1963). Our choice
of the transient dual-porosity model in this paper is based on the
expectation of elongated transient flow periods in tight matrices.
This expectation is supported by the quarter-slope behavior of the
log-log pressure response of pressure-transient tests observed in
the field, which is a result of the intermediate-time behavior of the
transient-fluid-transfer model. We discuss the flow-regime charac-
teristics of transient and pseudosteady dual-porosity models and
their implications for pressure-transient analysis in a companion
paper (Brown et al. 2009). We note only that the pseudosteady-
fluid-transfer model may be applicable in formations with matrix
permeabilities much higher than those observed in common appli-
cations of multiple-fractured horizontal wells.
The derivation and verification of the trilinear model are given
by Brown (2009) and Brown et al. (2009). Final analytical expres-
sions are presented in Appendix A. For our discussions, here we
note some essential definitions. We let qF,sc denote the flow rate
for each hydraulic fracture such that qF,sc = qsc/nF, where qsc is the
total flow rate from the horizontal well and nF is the number of
hydraulic fractures along the well length. For a naturally fractured
inner zone, we use the intrinsic properties of the matrix and the
249
 Line of symmetry
No-Flow Line
y e =d F /2
No-Flow Boundary
OUTER RESERVOIR
HYDRAULIC kOO, φO,, cctO
tO
FRACTURE
k F, w FF, φF, c tF
xe
INNER RESERVOIR
NATURALLY
xF FRACTURED
k f, φf, c tf ,
k m, φm , c tm
Line of symmetry
No-Flow Line
HORIZONTAL WELL
Conceptual Representation of the Trilinear Model
1/nF of the Well’s Drainage Area
Producing q/nF stb/d (oil) or Mscf/d (gas)
Fig. 4—Schematic of the trilinear-flow model used for the analytical solution of multiple-fractured-horizontal-well performance.
fractures. Specifically, kIh = kf hf , where kf is the intrinsic perme-
ability of the natural fractures, and hft is the total thickness of the
natural fractures intersecting the height of the hydraulic fracture
hF (in this work, hF is equal to the pay thickness h). If nf is the
count of natural fractures intersecting the height of the hydraulic
fracture and hf is the aperture of the natural fractures, we have hft =
nf hf . We also introduce the density of natural fractures, f = nf /h,
in units of the number of natural fractures per foot of pay zone
(f/ft). An estimate of f can be obtained from cores, logs, rate of
penetration, and other sources.
For the clarity of the discussions presented later, it is useful
to comment on the no-flow boundaries and the lines of symmetry
considered in the trilinear model. As shown in Fig. 4, for the
assumptions of the trilinear model, the lines bisecting the width
of each fracture and the lines bisecting the distance between two
identical fractures (dF) are no-flow lines of symmetry. Similarly,
the axis of the horizontal well forms a no-flow line of symmetry.
Therefore, it is sufficient to model a symmetry element, which is
one-quarter of a fracture’s drainage area with a production rate of
qF,sc/4=qsc/(4nF). In this model, the effect of the no-flow bound-
ary at ye = dF/2 corresponds to the effect of interference between
hydraulic fractures.
We also consider another no-flow boundary at a distance xe ≥
xF from the axis of the horizontal well. The boundary at xe may
correspond to a no-flow boundary between two wells or to a physi-
cal barrier to flow. Assuming xe = xF limits the drainage area to the
stimulated inner reservoir between hydraulic fractures. If xe > xF
and the outer reservoir has considerably lower permeability com-
pared with the inner reservoir, then the flow across the boundary
between the inner and the outer reservoirs is negligible initially;
that is, x = xF is a no-flow boundary at early times. The outer res-
ervoir may contribute to flow when pressure drops significantly to
start depletion in the inner reservoir. The magnitude and the time
of contribution of the outer reservoir depend on the permeability
and volume contrast between the inner and the outer reservoirs.
Interplay among the effects of the boundaries just discussed is
responsible for the changes in flow and production characteristics
of multiple-fractured horizontal wells in shale reservoirs.
We should also explain our assumption of linear flow in hydrau-
lic fractures, which is similar to fractured-vertical-well models.
This assumption is applicable when the wellbore intercepts the
hydraulic fracture through its entire thickness. In the case of a
fractured horizontal well, the wellbore communicates with the
fracture through a small cylindrical surface piercing through the
fracture surface. As a result of this restricted communication, flow
250
y e =d F /2
OUTER RESERVOIR
HYDRAULIC
kOO, φO,, cctO
tO
FRACTURE
k F, w F φ
F, F, ctF
tF
xe
INNER RESERVOIR
NATURALLY
xF FRACTURED
k f, φf, c tf ,
km, φm,, cctm tm
HORIZONTAL WELL
Symmetry Element Considered in the Mathematical Model
1/4 of the Fracture’s Drainage Area
Producing qF /4 =q/(4nF) stb/d (oil) or Mscf/d (gas)
convergence in fractures is radial around the wellbore intersection.
However, unless the fracture length (2xF) is comparable to fracture
height, flow within the fracture is predominantly linear, except
with the radial-flow convergence around the wellbore that causes
a choking effect (Soliman et al. 1990; Larsen and Hegre 1994;
Al-Kobaisi et al. 2006). We account for the effect of flow choking
around the intersection of the horizontal well and hydraulic frac-
ture by incorporating a choking-skin (sc) expression (see Appendix
A, Eq. A-23) given by Mukherjee and Economides (1991). The
choking-skin concept should be valid after the end of radial flow
in the fracture. Early-time radial flow within hydraulic fractures
does not last long and is masked by the effect of wellbore stor-
age. For the purposes of this work, we do not consider the effect
of wellbore storage in our discussions. However, wellbore-storage
effect can be incorporated easily into the trilinear model by using
Eq. A-24 in Appendix A. For the discussion of wellbore-storage
effect on trilinear-model responses, we refer the reader to Brown
et al. (2009).
Effect of the Outer Reservoir
In Figs. 5 and 6, we present some results to support the premise
and field observations that the drainage of fractured horizontal
wells is confined to the inner reservoir between the hydraulic frac-
tures (also referred to as the stimulated volume). Before discussing
the results, we must comment on the value of the matrix perme-
ability (km=1,000 md) used in Figs. 5 and 6. When we used matrix
permeabilities in the millidarcy range in our simulations, we did
not observe any influence of the outer reservoir because the outer
reservoir is made of the same matrix (without natural fractures).
A tighter matrix causes faster pressure drop in the inner zone,
but the support of the outer zone is also less efficient and takes a
longer period to be felt. Therefore, the cases presented in Figs. 5
and 6 may not correspond to tight, unconventional reservoirs but
they are useful to make the pertinent observations on the general
characteristics of the support of the outer reservoir.
In Figs. 5 and 6, we consider a dry-gas reservoir and for the
identification of the conventional flow regimes, we present the
results in terms of pseudopressure (Al-Hussainy et al. 1966;
Al-Hussainy and Ramey 1966) defined by Eq. A-3 in Appendix A.
Pseudopressure change is defined as the difference from the initial
pseudopressure (Δmwf = mi−mwf) and is presented as a function of
time for three different distances to the outer reservoir boundary
(xe). In Figs. 5 and 6, the case for xe = xF = 250 ft represents the
conditions in which the drainage volume is limited to the inner
reservoir and the hydraulic-fracture spacing (dF=2ye) is 500 ft. For
April 2011 SPE Reservoir Evaluation & Engineering
 (a)
A – k f = 2,000 md
Fig. 5—Effect of the outer reservoir on multiple-fractured-horizontal-well performance—influence of natural-fracture permeability.
this fracture spacing, the effect of fracture interference (that is, the
effect of the no-flow boundary at ye=dF/2) starts at approximately
100 hours. The unit-slope behavior of the responses after 1,000
hours is the same as pseudosteady-state flow behavior. This behav-
ior is a result of the fact that the outer region does not respond
significantly until the inner zone is depleted to a low pressure. The
deviation of the pseudopressures from the xe = xF case (deviation
from unit-slope behavior) indicates the start of the support of the
outer reservoir. If, for example, the outer region in Figs. 5 and 6
had permeability comparable to that of the inner region, the sup-
port of the outer reservoir could cause a well-defined half-slope
behavior (unit slope for pseudosteady state in the inner reservoir
and half slope for infinite-acting liner flow in the outer reservoir).
For the permeability contrast used in Figs. 5 and 6, the value of
the late-time slope tends toward one-half before becoming unity at
late times when the outer reservoir goes into depletion under the
effect of its no-flow boundary at xe.
Fig. 5 shows the effect of natural-fracture permeability. In
Fig. 5a, for a realistic value of natural-fracture permeability (kf =
2,000 md), the support of the outer reservoir becomes evi-
dent after approximately 100 years for a well spacing of 107 ft
(xe = 5×106 ft). For smaller well spacing, it takes even longer to feel
the effect of the outer boundary (the relative volume of production
(a)
A – ρ f = 0.8 f / ft
Fig. 6—Effect of the outer reservoir on multiple-fractured-horizontal-well performance—influence of natural-fracture density.
April 2011 SPE Reservoir Evaluation & Engineering
(b)
B – k f = 20,000 md
from the outer reservoir is even more insignificant compared with
the volume produced from the inner zone). It is important to note
that mwf = 8.69×108 psi2/cp corresponds to pwf = pi−pwf ≈ 4,000
psi for these examples. Because the initial pressure is pi=4,000
psi, the absolute limit of production is reached at approximately
10.5 years. Therefore, subject to the constraints of the model con-
sidered in our work, for the case presented in Fig. 5a, the effect of
the outer boundary will not be felt during the life of the well.
In Fig. 5b, the natural-fracture permeability is increased to
kf = 20,000 md, which may not be realistic for most natural
fractures. This case, however, may be used to simulate a hydrau-
lic-fracture network (resulting from dense, complex hydraulic-
fracture growth) with a fracture spacing of 1.25 ft (f = 0.8 f/ft).
As indicated by the comparison of the results in Figs. 5a and 5b,
higher natural-fracture permeability causes the support of the outer
reservoir to become evident earlier; however, the effect of the outer
reservoir is still outside the practical life span of the well. These
observations support the common practice of small well spacing
to efficiently drain tight, unconventional reservoirs with fractured
horizontal wells.
Fig. 6 demonstrates the effect of natural-fracture density on the
contribution of the outer reservoir. Fig. 6a is the same as Fig. 5b
having a natural-fracture density of f = 0.8 f/ft. In Fig. 6b, the
(b)
B – ρ f = 2.0 f / ft
251
 PSEUDOPRESSURE DERIVATIVE, dmwf /dln t , psi2/cp
1E+9 1E+9
Hydraulic Fractures Natural Fractures Matrix Hydraulic Fractures Natural Fractures Matrix
mwf, psi2/cp ---------------------------- ------------------------- -------------- ---------------------------- ------------------------- --------------
k = 105 md kf = 2000 md φm = 0.05 k = 105 md kf = 2000 md φ = 0.05
F F m
wF = 10-2 ft hf = 10-3 ft cm = 10-5 psi-1 1E+8 wF = 10-2 ft hf = 10-3 ft cm = 10-5 psi-1
1E+8
φ = 0.38 φ = 0.45 φ F = 0.38 φ f = 0.45
F f

cF = 10-5 psi-1 cf = 10-5 psi-1 cF = 10-5 psi-1 cf = 10-5 psi-1

ρ = 0.8 f / ft Matrix km (md) 1E+7 ρ = 0.8 f / ft
PSEUDOPRESSURE CHANGE,

xF = 250 ft f xF = 250 ft f
1E+7 ---------------------
dF = 500 ft 1E-8 dF = 500 ft
1E-6 Matrix km (md)
1E-4 1E+6
--------------------- qF,sc = 100 Mscf/d
1E-2
1E-8
1E+6 1E-1
1E-6 Reservoir
Reservoir
1E+5 1E-4 -------------------
-------------------
1E-2 h = 250 ft
h = 250 ft
1E-1 T = 150oF
T = 150oF γ = 0.55
1E+5 γ = 0.55
qF,sc = 100 Mscf/d 1E+4 pi = 4000 psi
pi = 4000 psi
μ = 0.0184 cp
WITH CHOKING SKIN μ = 0.0184 cp i
i
xe = 500 ft
WITHOUT CHOKING SKIN xe = 500 ft
1E+4 1E+3
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6
TIME, t, hr TIME, t, hr
(a) (b)

Fig. 7—Effect of matrix permeability on multiple fractured horizontal-well performance—naturally fractured inner reservoir.

fracture density is increased to f = 2 f/ft. Comparing the results cannot be used. For the particular example in Figs. 7 and 8, the
in Figs. 6a and 6b, higher fracture density does not cause a notice- effective, or bulk, permeability of the natural-fracture system is
able change in the time required to feel the effect of the outer k f =kf hft/h=1.6 md. Flow capacity of the natural fractures will be
reservoir; and although at higher fracture densities the contribution discussed in the next section. Here, we note that the validity of
of the outer reservoir is more significant at late times, the effect the preceding discussion requires a large permeability contrast
becomes evident only at impractically late times. The discussion between the matrix and fractures; that is, the production of the
in this section, therefore, supports the common interpretation that matrix must be by virtue of the natural fractures as in the dual-
fractured horizontal wells do not drain the region beyond the tips porosity idealization (Warren and Root 1963; Kazemi 1969; de
of the hydraulic fractures for practical matrix permeabilities. Swaan O. 1976). Most tight gas reservoirs and shale-gas reservoirs
satisfy this requirement and the preceding conclusions can be
Effect of Matrix Permeability generalized for practical applications.
When the matrix permeability decreases from milli- to micro- To demonstrate the difference between the performances of
and nanodarcy range, the reservoir description is changed from multiple-fractured horizontal wells in conventional tight (e.g.,
conventional to unconventional. For unconventional reservoirs, a tight gas) and unconventional tight (e.g., shale-gas) formations,
dense fracture network is required for a sufficient effective perme- we present two cases in Fig. 9. The unconventional tight reservoir
ability in the system. Here, we will first present a discussion of the has a natural-fracture network with matrix and natural-fracture
effect of matrix permeability when the inner reservoir is naturally permeabilities of km = 10−6 md and kf = 2,000 md, respectively (the
fractured, as in unconventional tight formations. Then, we will corresponding effective fracture-network permeability is 1.6 md).
present a comparison with the case in which there are no natural The conventional reservoir is homogeneous and has a permeability
fractures, as in conventional tight-sand applications. of k=10−1 md. Despite the large contrast between the matrix perme-
Fig. 7 shows the pseudopressure change vs. time (Fig. 7a) abilities of the two systems, Fig. 9a shows similar pseudopressure
and log-derivative of pseudopressure vs. time (Fig. 7b) for matrix responses. In fact, Fig. 9b shows that the productivity of the higher-
permeabilities in the range of 10−8 md ≤ km ≤ 10−1 md. In Fig.
7a, pseudopressure responses are also shown for the idealized
case of no choking-skin effect (sc) within hydraulic fractures. 1E-3
Fig. 7a indicates that the choking skin masks the characteristics
PRODUCTIVITY INDEX, J / nF , Mscf - cp / d - psi2

of the early-time responses. The derivative responses in Fig. 7b,

however, are not influenced by the choking skin and display all
early-time flow characteristics. This observation emphasizes the 1E-4
Natural Fractures
importance of derivative analysis for multiple-fractured horizontal -------------------------
kf = 2000 md
wells (especially for the identification of flow characteristics from
hf = 10-3 ft
production data in which derivative analysis may not be possible φ = 0.45
Matrix km (md)
f ---------------------
or may be overlooked). -5
cf = 10 psi -1
1E-5 1E-8
Fig. 7 shows that higher matrix permeabilities reduce the pseu- ρ = 0.8 f / ft 1E-6
f Hydraulic Fractures
dopressure drop at intermediate times. At late times, all responses ----------------------------
1E-4
Reservoir
merge, but this is not caused by the lack of the effect of matrix ------------------- kF = 105 md
permeability. Fig. 8 shows the productivity indices for the same h = 250 ft wF = 10-2 ft
cases considered in Fig. 7 but only with the effect of choking 1E-6 T = 150oF
γ = 0.55 φ = 0.38
F Matrix
skin. Productivity indices are computed by using the procedure pi = 4000 psi cF = 10-5 psi-1 --------------
φ = 0.05
outlined in Appendix B with the drainage area equal to the area μ = 0.0184 cp
i
xF = 250 ft m

of the inner reservoir. Note that the productivity index stabilizes xe = 500 ft dF = 500 ft cm = 10-5 psi-1
when boundary-dominated flow is established at late times. Figs. 1E-7
7 and 8 indicate that the well performance improves as the matrix 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
permeability increases but that matrix permeabilities higher than or TIME, t, hr
equal to 10−4 md do not contribute to the well performance. This
indicates that when the maximum flow capacity of the natural- Fig. 8—Effect of matrix permeability on the productivity of a mul-
fracture system is reached, additional productivity of the matrix tiple-fractured horizontal well—naturally fractured reservoir.

252 April 2011 SPE Reservoir Evaluation & Engineering

 1E+9 1E-3
Hydraulic Fractures Natural Fractures Matrix Natural Fractures Matrix Hydraulic Fractures

PRODUCTIVITY INDEX, J / nF , Mscf - cp / d - psi2

---------------------------- ------------------------- -------------- ------------------------- ------------------- ----------------------------

mwf, psi2/cp
kF = 105 md kf = 2000 md km = 10-6 md kf = 2000 md km = 10-6 md kF = 105 md
wF = 10-2 ft hf = 10-3 ft φm = 0.05 hf = 10-3 ft φm = 0.05 wF = 10-2 ft
φ = 0.38 φ = 0.45 c = 10-5 psi-1 φ = 0.45 c = 10-5 psi-1 φ F = 0.38
1E+8 F f m f m
cF = 10-5 psi-1 cf = 10-5 psi-1 cf = 10-5 psi-1 cF = 10-5 psi-1
ρ = 0.8 f / ft ρ = 0.8 f / ft
PSEUDOPRESSURE CHANGE,

xF = 250 ft f f xF = 250 ft
dF = 500 ft dF = 500 ft

1E+7 1E-4
qF,sc = 100 Mscf/d
Reservoir
Reservoir
-------------------
-------------------
h = 250 ft
h = 250 ft
1E+6 T = 150oF
T = 150oF γ = 0.55
γ = 0.55 qF,sc = 100 Mscf/d
pi = 4000 psi
pi = 4000 psi
μ = 0.0184 cp NATURALLY FRACTURED
NATURALLY FRACTURED μ = 0.0184 cp i
i
xe = 500 ft HOMOGENOUS, k = 10-1 md
HOMOGENOUS, k = 10-1 md xe = 500 ft
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)

Fig. 9—Effect of matrix permeability on the change in pseudopressure (a) and productivity (b) of a multiple-fractured horizontal
well—homogeneous and naturally fractured inner reservoirs.

permeability conventional, homogeneous reservoir is less than that A higher density of natural fractures, however, increases the flow
of the unconventional tight reservoir with natural fractures. This capacities of both the natural fractures and the matrix because the
example underlines the importance of natural fractures in uncon- increase in surface area for fluid transfer leads to a more significant
ventional tight formations, such as shale. increase in productivity. From a practical viewpoint, this is a desir-
able feature because increasing the permeability of natural fractures
Effect of Natural-Fracture Permeability and may not be a viable option while increasing the density of natural
Density fractures may be accomplished by designing complex hydraulic
In the preceding section, we emphasized the importance of the fractures in brittle formations (Cramer 2008; Cipolla 2009). It can
flow capacity of natural fractures for production from tight for- also be deduced that creating a complex hydraulic-fracture network
mations. Figs. 10 and 11 indicate that a higher natural-fracture of high permeability (but relatively low density) is not a good option
permeability may not significantly improve the well performance if hydraulic fracturing does not also increase the density of natural
while a higher density of natural fractures may have a significant fractures. This indicates that the experience in brittle shale forma-
effect on productivity. As discussed in the preceding section, this tions, such as the Barnett, may not be extrapolated directly to soft-
is because of the relative flow capacities of the matrix and frac- shale formations such as the Marcellus and the Haynesville.
ture systems. A higher permeability increases the flow capacity
of natural fractures but does not change the flow capacity of the Effect of Hydraulic-Fracture Permeability
matrix because the flow capacity of the matrix is proportional to and Spacing
the surface area. If there is not enough fluid transfer from matrix Determining the optimum number of hydraulic fractures to be
to fracture, a high permeability in the natural fractures does not created along the length of the well is an important practical con-
create additional productivity. cern. A relevant, but more conventional, concern is determining

1E+7 1E-3
Natural Fractures Natural Fracture Matrix Natural Fractures Natural Fracture Matrix
------------------------- Permeability
PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2

--------------- ------------------------- Permeability ---------------

mwf, psi2/cp

hf = 10-3 ft kf (md) km = 10-6 md hf = 10-3 ft kf (md) km = 10-6 md

φ f = 0.45 --------------------- φ m = 0.05 φ f = 0.45 --------------------- φm = 0.05
2000 2000
cf = 10-5 psi-1 50000 cm = 10-5 psi-1 cf = 10-5 psi-1 50000 cm = 10-5 psi-1
ρ = 0.8 f / ft ρ = 0.8 f / ft
f f
PSEUDOPRESSURE CHANGE,

Reservoir
-------------------
h = 250 ft
1E+6 T = 150oF 1E-4
γ = 0.55 Hydraulic Fractures Hydraulic Fractures Reservoir
pi = 4000 psi ---------------------------- ---------------------------- -------------------
μ = 0.0184 cp kF = 105 md kF = 105 md h = 250 ft
i
T = 150oF
xe = 500 ft wF = 10-2 ft wF = 10-2 ft γ = 0.55
φ F = 0.38 φ F = 0.38 pi = 4000 psi
cF = 10-5 psi-1 c = 10-5 psi-1 μ = 0.0184 cp
F i
xF = 250 ft xF = 250 ft xe = 500 ft
qF, sc = 100 Mscf/d
dF = 500 ft dF = 500 ft
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4

(a) TIME, t, hr (b) TIME, t, hr

A – PSEUDOPRESSURE VS. TIME B – PRODUCTIVITY INDEX VS. TIM E

Fig. 10—Effect of natural-fracture permeability on the change in pseudopressure (a) and productivity (b) of a multiple-fractured
horizontal well.

April 2011 SPE Reservoir Evaluation & Engineering 253

 1E+9 1E-3
Fracture Density Natural Fractures Matrix Matrix Hydraulic Fractures
ρ (f / ft) ------------------------- --------------- --------------- ----------------------------

PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2

mwf, psi2/cp
f
--------------------- kf = 2000 md km = 10-6 md km = 10-6 md kF = 105 md xF = 250 ft
1.2E-0 hf = 10-3 ft φm = 0.05 φm = 0.05 wF = 10-2 ft dF = 500 ft
8.0E-1
1E+8 φ = 0.45 cm = 10-5 psi-1 cm = 10-5 psi-1 φ F = 0.38
4.0E-1 f
2.0E-1 cf = 10-5 psi-1 cF = 10-5 psi-1
1.2E-1 1E-4
PSEUDOPRESSURE CHANGE,

Fracture Density
Hydraulic Fractures ρ (f / ft)
f
---------------------------- ---------------------
1E+7 kF = 105 md xF = 250 ft 1.2E-0
8.0E-1
wF = 10-2 ft dF = 500 ft
Reservoir 4.0E-1
φ F = 0.38 2.0E-1 Reservoir
-------------------
cF = 10-5 psi-1 h = 250 ft 1E-5 1.2E-1 -------------------
Natural Fractures h = 250 ft
T = 150oF
1E+6 γ = 0.55 ------------------------- T = 150oF
kf = 2000 md γ = 0.55
pi = 4000 psi
μ = 0.0184 cp hf = 10-3 ft pi = 4000 psi
i
φ = 0.45 μ = 0.0184 cp
qF, sc = 100 Mscf/d xe = 500 ft f i
cf = 10-5 psi-1 xe = 500 ft
1E+5 1E-6
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – PRODUCTIVITY INDEX VS. TIME

Fig. 11—Effect of natural-fracture density on the change in pseudopressure (a) and productivity (b) of a multiple-fractured hori-
zontal well.

the optimum conductivity of hydraulic fractures. Fig. 12 is a density. Therefore, it is desirable to obtain higher dimensionless
demonstration of the conventional concept that increasing hydrau- hydraulic-fracture conductivities by creating larger natural-fracture
lic-fracture conductivity, kFwF, leads to an increase in productivity. densities. It must be noted, however, that higher natural-fracture
The incremental benefit, however, decreases as the conductivity densities also increase the effective natural-fracture permeability
increases. From a fundamental viewpoint, higher productivity and, therefore, decrease the dimensionless hydraulic-fracture con-
is achieved by making the dimensionless conductivity, CFD = ductivity. This leads to a complex design problem for hydraulic
kFwF/(kxF), larger. This concept is relatively straightforward for fractures in tight, naturally fractured formations.
the design of hydraulic fractures in conventional tight formations As shown in Fig. 13a, reducing the hydraulic-fracture spacing
(such as tight gas sands) in which higher conductivities can be significantly influences the magnitudes of the intermediate- and
achieved for low reservoir permeability k. A large dimensionless late-time pseudopressures. On the other hand, Fig. 13b shows
conductivity indicates that the hydraulic fracture can transmit all that hydraulic-fracture spacing does not significantly change flow
the fluid provided by the reservoir. characteristics (it does, however, affect the duration of the flow
For unconventional tight formations, dimensionless hydraulic- regimes). It must be noted that the flow rate of each fracture has
fracture conductivity is defined on the basis of the effective fracture been kept constant while changing the hydraulic-fracture spacing
permeability (k = k f ) and may appear to be relatively large; yet, in Fig. 13. If the total flow rate was kept constant, then the indi-
as discussed in the preceding section, flow capacity of the matrix vidual-fracture rates would have to be reduced when the fracture
may not be sufficient to supply the fracture network to its full spacing decreased. This would yield a separation of the pseudo-
flow capacity because of low matrix permeability or low surface pressure curves at early times (with the smallest pseudopressure
area available for fluid transfer because of low natural-fracture change corresponding with the smaller fracture spacing).

1E+9 1E-3
Hydraulic Fractures Natural Fractures Hydraulic Fractures Reservoir
---------------------------- Natural Fractures ------------------------- ---------------------------- -------------------
PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2
mwf, psi2/cp

wF = 10-2 ft ------------------------- kf = 2000 md wF = 10-2 ft h = 250 ft

kf = 2000 md
φ F = 0.38 hf = 10-3 ft φ F = 0.38
T = 150oF
hf = 10-3 ft γ = 0.55
cF = 10-5 psi-1 φ = 0.45 c = 10-5 psi-1
1E+8 φ = 0.45 f F pi = 4000 psi
xF = 250 ft Fracture Permeability f
cf = 10-5 psi-1 xF = 250 ft μ = 0.0184 cp
kF (md) cf = 10-5 psi-1 i
dF = 500 ft Matrix ρ = 1.2 f / ft dF = 500 ft xe = 500 ft
PSEUDOPRESSURE CHANGE,

------------------------------ ρ = 1.2 f / ft f
f ---------------
1 x 105 km = 10-6 md
8 x 104
1E+7 φm = 0.05 1E-4
5 x 104
3 x 104 cm = 10-5 psi-1
Reservoir
------------------- Fracture Permeability
h = 250 ft kF (md)
1E+6 T = 150oF Matrix
γ = 0.55 ------------------------------ ---------------
pi = 4000 psi 1 x 105 km = 10-6 md
μ = 0.0184 cp 8 x 104 φm = 0.05
qF, sc = 100 Mscf/d
i
5 x 104
xe = 500 ft cm = 10-5 psi-1
3 x 104
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – PRODUCTIVITY INDEX VS. TIME

Fig. 12—Effect of hydraulic-fracture conductivity on the change in pseudopressure (a) and productivity (b) of a multiple-fractured
horizontal well.

254 April 2011 SPE Reservoir Evaluation & Engineering

 1E+9 1E+9

PSEUDOPRESSURE DERIVATIVE, dmwf / dlnt, psi2/cp

Fracture Spacing Natural Fractures Matrix Fracture Spacing Natural Fractures Matrix
dF ( ft) ------------------------- --------------- dF ( ft) ------------------------- ---------------

mwf, psi2/cp
--------------------- kf = 2000 md km = 10-6 md --------------------- kf = 2000 md km = 10-6 md
1000 hf = 10-3 ft φm = 0.05 1000 hf = 10-3 ft φ = 0.05
750 1E+8 750 m
1E+8 500 φ = 0.45 cm = 10-5 psi-1 φ f = 0.45 cm = 10-5 psi-1
f 500
250 cf = 10-5 psi-1 250 -5
cf = 10 psi -1

ρ = 0.4 f / ft ρ = 0.4 f / ft
PSEUDOPRESSURE CHANGE,

f f
1E+7
Hydraulic Fractures
Hydraulic Fractures
----------------------------
1E+7 ----------------------------
kF = 105 md
kF = 105 md
wF = 10-2 ft Reservoir 1E+6 wF = 10-2 ft Reservoir
φ = 0.38 ------------------- -------------------
F φ F = 0.38
-5
cF = 10 psi -1 h = 250 ft h = 250 ft
T = 150oF cF = 10-5 psi-1 T = 150oF
1E+6 xF = 250 ft γ = 0.55 γ = 0.55
xF = 250 ft
pi = 4000 psi 1E+5 pi = 4000 psi
μ = 0.0184 cp μ = 0.0184 cp
i i
qf, sc = 100 Mscf/d xe = 500 ft qF, sc = 100 Mscf/d xe = 500 ft

1E+5 1E+4
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – LOG DERIVATIVE VS. TIME

Fig. 13—Effect of hydraulic-fracture spacing on the pseudopressure responses of a multiple-fractured horizontal well.

The productivity of each fracture is expected to increase when increase. This conclusion supports the field practice of creating
the fracture spacing is decreased. As shown in Fig. 14a, however, complex fractures to increase natural-fracture density and decrease
the incremental gain in productivity with each additional fracture hydraulic-fracture spacing.
decreases. To find the total productivity of the system, the produc-
tivity indices in Fig. 14 must be multiplied by the corresponding Conclusions
number of hydraulic fractures. For example, for a 4,000-ft horizon- In this paper, we investigated the performance characteristics of
tal well, 1,000-ft and 250-ft hydraulic-fracture spacing corresponds fractured horizontal wells and discussed the effects of hydraulic
to two and eight hydraulic fractures, respectively. The total produc- fractures and reservoir properties. The trilinear-flow model pre-
tivity of the well with 1,000-ft fracture spacing is then obtained sented in this work has provided an effective tool for understanding
by multiplying the corresponding individual-fracture productivity and predicting the productivities of fractured horizontal wells.
given in Fig. 14 by two. Similarly, the well productivity for 250- The following conclusions are warranted on the basis of this
ft fracture spacing is obtained by multiplying the corresponding work:
productivity index by eight. This would yield a larger productivity 1. Differences in the performance of fractured horizontal wells in
for the well with eight fractures (250-ft spacing) than for the well conventional and unconventional tight formations are mostly
with two fractures (1,000-ft spacing). determined by the existence of a natural- or induced-fracture
To complete our discussions, in Fig. 14b we present individual- network in the latter.
fracture productivities for two values of natural-fracture density 2. Unlike conventional tight gas reservoirs, relatively high matrix
and two values of hydraulic-fracture spacing. As expected, Fig. 14b permeability and increased hydraulic-fracture conductivity
indicates that increasing natural-fracture density while decreas- are not sufficient to have good productivity in unconventional,
ing hydraulic-fracture spacing provides the greatest productivity shale-gas reservoirs. Existence of a dense natural- or induced-

1E-3 1E-3
Fracture Spacing Natural Fractures Matrix Natural Fractures Hydraulic Fractures Matrix Reservoir
PRODUCTIVITY INDEX, J / nF , Mscf - cp / d - psi2

------------------------- ---------------------------- --------------- -------------------

PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2

dF ( ft) ------------------------- ---------------

kf = 2000 md km = 10-6 md kf = 2000 md kF = 105 md km = 10-6 md h = 250 ft
---------------------
hf = 10-3 ft T = 150oF
1000 hf = 10-3 ft φm = 0.05 wF = 10-2 ft φm = 0.05 γ = 0.55
750 φ = 0.45 φ = 0.45 φ = 0.38 c = 10-5 psi-1
500 f cm = 10-5 psi-1 f F m pi = 4000 psi
250 cf = 10-5 psi-1 cf = 10-5 psi-1 cF = 10-5 psi-1 μ = 0.0184 cp
Reservoir i
ρ = 0.4 f / ft ------------------- xF = 250 ft xe = 500 ft
f
h = 250 ft
T = 150oF
1E-4 γ = 0.55 1E-4
pi = 4000 psi
Hydraulic Fractures μ = 0.0184 cp
i
----------------------------
xe = 500 ft
kF = 105 md
wF = 10-2 ft
φ = 0.38 Natural Fracture Fracture Spacing
F
Density ρf (f / ft) dF ( ft)
cF = 10-5 psi-1
------------------------ ---------------------
xF = 250 ft 0.4 1000
1.2 250
1E-5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5

(a) TIME, t, hr (b) TIME, t, hr

A – EFFECT OF HYDRAULIC FRACTURE SPACING FOR B – COMBINED EFFECT OF HYDRAULIC FRACTURE

FIXED NATURAL FRACTURE DENSITY SPACING AND NATURAL FRACTURE DENSITY

Fig. 14—Effect of hydraulic-fracture spacing and natural-fracture density on the productivity of multiple-fractured horizontal
wells.

April 2011 SPE Reservoir Evaluation & Engineering 255

 fracture network is a more important requirement for favorable mwf = flowing pseudopressure, psi2/cp
productivities in shale-gas reservoirs. However, increasing the nf = number of natural fractures
permeability of the fracture network has a relatively less sig- nF = number of hydraulic fractures
nificant effect on productivity than increasing the density of the p = pressure, psi
fracture network. pb = base pressure, psi
3. The limitations of the interaction between the tight matrix and
pi = initial reservoir pressure, psi
natural-fracture network may place restrictions on productivity
improvements. pwD = dimensionless flowing wellbore pressure
4. Even for relatively large matrix permeabilities, the drainage pwf = flowing wellbore pressure, psi
volume of fractured horizontal wells is limited to the naturally p = average reservoir pressure, psi
or artificially fractured inner reservoir between the hydraulic p D = dimensionless average reservoir pressure
fractures. Under these conditions, flow convergence is linear pwD = dimensionless flowing wellbore pressure in the Laplace
for most of the productive life of the well. space
5. Hydraulic-fracture conductivities should be optimized on the pwD,storage = dimensionless wellbore-storage pressure in the Laplace
basis of the flow capacities of the matrix and natural-fracture space
network. Decreasing fracture spacing increases the productivity q = production rate, B/D (oil), Mcf/D (gas)
of the well, but the incremental gain for each additional fracture qF = individual-hydraulic-fracture production rate, B/D (oil),
decreases. scf/D (gas)
qF,sc = individual-hydraulic-fracture production rate at surface
Nomenclature conditions, STB/D (oil), Mscf/D (gas)
A = drainage area, ft2 qsc = production rate at surface conditions, STB/D (oil),
AD = dimensionless drainage area Mscf/D (gas)
B = formation volume factor, bbl/STB rw,eff = effective wellbore radius, ft
ct = total compressibility, psi−1 rwj = fracture effective wellbore radius, ft
ctf = total natural-fracture compressibility, psi−1 s = Laplace-transform parameter
ctF = total hydraulic-fracture compressibility, psi−1 sc = hydraulic-fracture flow-convergence skin factor (Eq.
ctI = total compressibility of the inner reservoir, psi−1 A-23)
ctIi = initial total compressibility of the inner reservoir, t = time, hours
psi−1 tD = dimensionless time
ctm = total matrix compressibility, psi−1 T = reservoir temperature, °R
ctO = total compressibility of the outer reservoir, psi−1 u = parameter defined as the product of the dual-porosity
CA = shape factor function and the Laplace-transform parameter (Eq.
CD = dimensionless wellbore-storage coefficient A-18)
CF = hydraulic-fracture conductivity, md-ft wF = hydraulic-fracture width, ft
CF,eff = effective hydraulic-fracture conductivity, md-ft wF,eff = effective hydraulic-fracture width, ft
CFD = dimensionless hydraulic-fracture conductivity wFD = dimensionless hydraulic-fracture width
CFD,eff = effective dimensionless hydraulic-fracture conductivity xF = hydraulic fracture half-length, ft
CRD = dimensionless reservoir conductivity xF,eff = effective hydraulic-fracture half-length, ft
dF = distance between two hydraulic fractures, ft xe = reservoir size in x-direction, ft
D = distance between outermost hydraulic fractures, ft xeD = dimensionless reservoir size in x-direction
f = dual-porosity function defining fluid transfer from ye = reservoir size in y-direction, ft
matrix to fracture (Eq. A-20) ye,eff = effective reservoir size in y-direction, ft
h = reservoir thickness, ft yeD = dimensionless reservoir size in y-direction
hf = thickness of natural fractures, ft z = real-gas compressibility factor
hft = total thickness of natural fractures, ft F = inner-reservoir parameter used in trilinear-flow model
hF = height of hydraulic fractures, ft (Eq. A-10)
hm = thickness of matrix slabs, ft R = outer-reservoir parameter used in trilinear-flow model
J = transient productivity index, STB/D/psi (oil), Mscf/D/ (Eq. A-13)
psi2/cp (gas) F = inner-reservoir parameter used in trilinear-flow model
k = homogeneous reservoir permeability, md (Eq. A-12)
kf = natural-fracture permeability, md R = outer-reservoir parameter used in trilinear-flow model
kF = hydraulic-fracture permeability, md (Eq. A-15)
kI = permeability of the inner reservoir, md  = Euler’s constant (= 0.5772…)
km = matrix permeability (dual-porosity model), md I = inner-reservoir diffusivity, ft2/hr
kO = permeability of the outer reservoir, md f = natural-fracture diffusivity, ft2/hr
k f = effective (bulk) natural-fracture permeability, md F = hydraulic-fracture diffusivity, ft2/hr
kI = effective (bulk) permeability of the inner reservoir, FD = hydraulic-fracture diffusivity ratio
md O = outer-reservoir diffusivity, ft2/hr
m = pseudopressure (Eq. A-2), psi2/cp RD = outer-reservoir diffusivity ratio
mi = initial pseudopressure, psi2/cp
= dual-porosity transmissivity ratio (Eq. A-22)
mwD = dimensionless flowing pseudopressure  = fluid viscosity, cp
m = average reservoir pseudopressure, psi2/cp i = initial fluid viscosity, cp
m D = dimensionless average pseudopressure f = density of natural fractures, fractures/ft
mwD = dimensionless flowing wellbore pseudopressure in the  = reservoir porosity, fraction
Laplace space f = natural-fracture porosity, fraction
mwD,storage = dimensionless wellbore-storage pseudopressure in the F = hydraulic-fracture porosity, fraction
Laplace space I = inner-reservoir porosity, fraction

256 April 2011 SPE Reservoir Evaluation & Engineering

 Ii = initial inner reservoir porosity, fraction
m = matrix porosity, fraction
O = outer-reservoir porosity, fraction
 = dual-porosity storativity ratio (Eq. A-21)
Acknowledgments
This research has been conducted under the Marathon Center of
Excellence for Reservoir Studies (MCERS). Parts of this work
have been completed to fulfill the MS degree requirements of
Margaret Brown at the Colorado School of Mines. The financial
support for Margaret Brown’s graduate studies has been provided
by the MCERS and the Petroleum Engineering Department at the
Colorado School of Mines (CSM).
References
Al-Hussainy, R. and Ramey, H.J. Jr. 1966. Application of Real Gas Flow
Theory to Well Testing and Deliverability Forecasting. J Pet Technol
18 (5): 637–642; Trans., AIME, 237. SPE-1243-B-PA. doi:
10.2118/1243-B-PA.
Al-Hussainy, R., Ramey, H.J. Jr., and Crawford, P.B. 1966. The Flow of
Real Gases Through Porous Media. J Pet Technol 18 (5): 624–636;
Trans., AIME, 237. SPE-1243-A-PA. doi: 10.2118/1243-A-PA.
Al-Kobaisi, M., Ozkan, E., and Kazemi, H. 2006. A Hybrid Numerical/
Analytical Model of a Finite-Conductivity Vertical Fracture Intercepted
by a Horizontal Well. SPE Res Eval & Eng 24 (10): 345–355. SPE-
92040-PA. doi: 10.2118/92040-PA.
Bennett, C.O., Camacho-V., R.G., Reynolds, A.C., and Raghavan, R. 1985.
Approximate Solutions for Fractured Wells Producing Layered Reser-
voirs. SPE J. 25 (5): 729–742. SPE-11599-PA. doi: 10.2118/11599-PA.
Brown, M. 2009. Analytical Trilinear Pressure Transient Model for Mul-
tiply Fractured Horizontal Wells in Tight Shale Reservoirs. MS thesis,
Colorado School of Mines, Golden, Colorado, USA.
Brown, M., Ozkan, E., Ragahavan, R., and Kazemi, H. 2009. Practical
Solutions for Pressure Transient Responses of Fractured Horizontal
Wells in Unconventional Reservoirs. Paper SPE 125043 presented at
SPE Annual Technical Conference and Exhibition, New Orleans, 4–7
October. doi: 10.2118/125043-MS.
Camacho-V., R.G. 1984. Response of Wells Producing Commingled Res-
ervoirs: Unequal Fracture Length. MS thesis, University of Tulsa,
Tulsa.
Camacho-V., R.G., Raghavan, R. and Reynolds, A.C. 1987. Response of
Wells Producing Layered Reservoirs: Unequal Fracture Length. SPE
Form Eval 2 (1): 9–28. SPE-12844-PA. doi: 10.2118/12844-PA.
Chen, C.-C. and Raghavan, R. 1997. A Multiply-Fractured Horizontal Well
in a Rectangular Drainage Region. SPE J. 2 (4): 455–465. SPE-37072-
PA. doi: 10.2118/37072-PA.
Cinco-Ley, H. and Samaniego-V., F. 1981. Transient Pressure Analysis for
Fractured Wells. J Pet Technol 33 (9): 1749-1766. SPE-7490-PA. doi:
10.2118/7490-PA.
Cipolla, C.L. 2009. Modeling Production and Evaluating Fracture Perfor-
mance in Unconventional Gas Reservoirs. J Pet Technol 61 (9): 84–90.
SPE-118536-MS. doi: 10.2118/118536-MS.
Cramer, D.D. 2008. Stimulating Unconventional Reservoirs: Lessons
Learned, Successful Practices, Areas for Improvement. Paper SPE
114172 presented at the SPE Unconventional Reservoirs Conference,
Keystone, Colorado, USA, 10–12 February. doi: 10.2118/114172-MS.
de Swaan O., A. 1976. Analytical Solutions for Determining Naturally
Fractured Reservoir Properties by Well Testing. SPE J. 16 (3): 117-122;
Trans., AIME, 261. SPE-5346-PA. doi: 10.2118/5346-PA.
Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A.,
and Dunn, K.P. 2004. Optimizing Horizontal Completion Techniques
in the Barnett Shale Using Microseismic Fracture Mapping. Paper SPE
90051 presented at the SPE Annual Technical Conference and Exhibi-
tion, Houston, 26–29 September. doi: 10.2118/90051-MS.
Gringarten, A.C., Ramey, H.J. Jr., and Raghavan, R. 1974. Unsteady-State
Pressure Distributions Created by a Well With a Single Infinite-Con-
ductivity Vertical Fracture. SPE J. 14 (4): 347–360; Trans., AIME, 257.
SPE-4051-PA. doi: 10.2118/4051-PA.
Kazemi, H. 1969. Pressure Transient Analysis of Naturally Fractured
Reservoir With Uniform Fracture Distribution. SPE J. 9 (4): 451–462;
Trans., AIME, 246. SPE-2156-A. doi: 10.2118/2156-PA.
April 2011 SPE Reservoir Evaluation & Engineering
Larsen, L. and Hegre, T.M. 1994. Pressure Transient Analysis of Multifrac-
tured Horizontal Wells. Paper SPE 28389 presented at the SPE Annual
Technical Conference and Exhibition, New Orleans, 25–28 September.
doi: 10.2118/28389-MS.
Maxwell, S.C., Waltman, C.K., Warpinski, N.R., Mayerhofer, M.J., and
Boroumand, N. 2009. Imaging Seismic Deformation Induced by
Hydraulic Fracture Complexity. SPE Res Eval & Eng 12 (1): 48–52.
SPE-102801-PA. doi: 10.2118/102801-PA.
Mayerhofer, M.J., Bolander, J.L., Williams, L.I., Pavy, A., and Wolhart,
S.L. 2005. Integration of Microseismic-Fracture-Mapping Fracture and
Production Analysis With Well Interference Data to Optimize Fracture
Treatments in the Overton Field, East Texas. Paper SPE 95508 pre-
sented at the SPE Annual Technical Conference and Exhibition, Dallas,
9–12 October. doi: 10.2118/95508-MS.
Medeiros, F., Ozkan, E., and Kazemi, H. 2008. Productivity and Drainage
Area of Fracture Horizontal Wells in Tight Gas Reservoirs. SPE Res
Eval & Eng 11 (5): 902–911. SPE-108110-PA. doi: 10.2118/108110-
PA.
Mukherjee, H. and Economides, M.J. 1991. A Parametric Comparison
of Horizontal and Vertical Well Performance. SPE Form Eval 6 (2):
209–216. SPE-18303-PA. doi: 10.2118/18303-PA.
Prats, M. 1961. Effect of Vertical Fractures on Reservoir Behavior—Incom-
pressible Fluid Case. SPE J. 1 (2): 105–118; Trans., AIME, 222. SPE-
1575-G. doi: 10.2118/1575-G.
Raghavan, R.S., Chen, C.C., and Agarwal, B. 1997. An Analysis of Hori-
zontal Wells Intercepted by Multiple Fractures. SPE J. 2 (3): 235–245.
SPE-27652-PA. doi: 10.2118/27652-PA.
Serra, K.V., Reynolds, A.C., and Raghavan, R. 1983. New Pressure Tran-
sient Analysis Method for Naturally Fractured Reservoirs. J Pet Technol
35 (12): 2271−2283. SPE-10780-PA. doi: 10.2118/10780-PA.
Soliman, M.Y., Hunt, J.L., and El Rabaa, A.M. 1990. Fracturing Aspects
of Horizontal Wells. J Pet Technol 42 (8): 966–973. SPE-18542-PA.
doi: 10.2118/18542-PA.
Warren, J.E. and Root, P.J. 1963. The Behavior of Naturally Fractured
Reservoirs. SPE J. 3 (3): 245–255; Trans., AIME, 228. SPE-426-PA.
doi: 10.2118/426-PA.
Appendix A—Trilinear-Flow Model:
Analytical Expressions and Definitions
Here, we present the analytical expressions for the trilinear model
used in this paper. For a detailed derivation of these expressions,
we refer the reader to Brown et al. (2009) and Brown (2009). For
convenience, we introduce the trilinear-flow model in terms of
dimensionless variables. We define a dimensionless pressure by
pwD =
kI h
141.2qF ,sc B
( )
pi − pwf , . . . . . . . . . . . . . . . . . . . . . (A-1)
for liquid flow, and
mwD =
kI h
1422qF ,scT ⎣ ⎦ ( )
⎡ m ( pi ) − m pwf ⎤, . . . . . . . . . . . . . . . . (A-2)
for gas flow, with m(p) denoting real-gas pseudopressure (Al-Hus-
sainy et al. 1966; Al-Hussainy and Ramey 1966),
p
p′
m ( p) = 2 ∫ dp′ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-3)
pb
z
We note that kI h = kf hft for a naturally fractured inner zone
Dimensionless time is defined by
2.637 × 10 −4 kI t
tD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-4)
⎡⎣(ct )I ⎤⎦ i x F2
i
where t is in hours and (ct)I = (ct)f for a naturally fractured inner
zone. The dimensionless variables of the reservoir and fracture
geometry are defined by
xe
xeD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-5)
xF
257
 ye for a naturally fractured inner reservoir with the transient dual-
yeD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-6) porosity idealization (Serra et al. 1983). In Eq. A-20, the dual-
xF
porosity parameters are defined by
and
(ct h )m
= . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-21)
wFD
w
= F. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-7) (ct h ) f
xF
and
The dimensionless trilinear-flow solution for a multiple-fractured
horizontal well is given, in the Laplace-transform domain, by 12 x F2 km

= . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-22)
hm h f k f

pwD = mwD = + sc, . . . . . . . . . . . . . . (A-8)
CFD s  F tanh ( F ) All properties appearing in Eqs. A-21 and A-22 are intrinsic.
In Eq. A-8, sc denotes the contribution of flow choking around
where s is the Laplace-transform parameter with respect to the the intersection of the horizontal well and hydraulic fracture.
dimensionless time, tD, defined in Eq. A-4, Mukherjee and Economides (1991) have provided the following
expression to estimate sc:
k F wF
CFD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-9) kI h ⎡ ⎛ h ⎞  ⎤
kI x F sc = ⎢ ln − ⎥ . . . . . . . . . . . . . . . . . . . . . . . . (A-23)
kF wF ⎣ ⎜⎝ 2rw ⎟⎠ 2 ⎦

2F s The expression given by Eq. A-23 should be valid after the end of
F = + , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-10) the radial flow within the hydraulic fracture, which corresponds to
CFD FD
a very short period of time at the beginning of flow. For practical
cases, the early-time radial flow within the hydraulic fracture is
F completely masked by the effect of wellbore storage. The effect
FD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-11) of wellbore storage can be incorporated easily by using the fol-
I
lowing relationship:

pwD
F =  R tanh ⎡⎣  R ( yeD − wFD 2 ) ⎤⎦ , . . . . . . . . . . . . . . . (A-12) pwD ,storage = mwD ,storage = , . . . . . . . . . . . . . . . . (A-24)
1 + C D s 2 pwD

where pwD is the trilinear-model solution given by Eq. A-8 and CD

R is the dimensionless wellbore-storage coefficient given by
R = + u, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-13)
C RD yeD
 cC
CD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-25)
2 (ct h )I x F2
kI x F
C RD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-14) where c = 5.615 or 1 for oil and gas flow, respectively.
kO ye

R = s RD tanh ⎡⎣ s RD ( xeD − 1) ⎤⎦ , . . . . . . . . . . . . . . (A-15) Appendix B—Computation of the

Productivity Index
The productivity index of a multiple-fractured horizontal well is
and defined by
O
RD = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-16) J=
qsc q
= nF F ,sc , . . . . . . . . . . . . . . . . . . . . . . . . . (B-1)
I p − pwf p − pwf
In Eqs. A-11 and A-16, ( = F , I , or O ) is defined by
for liquid flow, where p denotes the average pressure in the reser-
k voir. We rearrange Eq. B-1 to write
= . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-17)
(ct )  qF ,sc
J = nF , . . . . . . . . . . . . . . . . . . . (B-2)
141.2qF ,sc B
For a naturally fractured inner reservoir, I = f in Eqs. A-9, A-11,
A-14, A-16, and A-17. Also, for a naturally fractured inner zone, kI h
( pwD − p D )
all definitions are based on intrinsic properties, except for kI = k f where pwD is computed from Eq. A-8 and
in Eqs. A-9 and A-14 where k f is the bulk (effective) permeability
(k f = k f h ft h ). In Eq. A-13, tD
p D = 2 , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-3)
(1 +  ) AD
u = sf ( s ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-18)
with
where
A
AD = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-4)
x F2
f ( s ) = 1, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-19)
For gas flow, we define
for a homogeneous inner reservoir and
qsc qF ,sc
J= = nF , . . . . . . . . . . . . (B-5)

 ⎛ 3s ⎞ m − mwf 1422qF ,scT
f (s) = 1 + tanh ⎜ ⎟ , . . . . . . . . . . . . . . . . . . . . (A-20)
( mwD − m D )
3s ⎝
⎠ kI h

258 April 2011 SPE Reservoir Evaluation & Engineering

where m denotes the average pseudopressure in the reservoir and 2007 SPE Formation Evaluation Award and a Distinguished
m D can be approximated by Member of SPE. Margaret Brown is a reservoir engineer work-
ing for Marathon Oil in Houston. E-mail: mbrown6@marathonoil.
tD com. She holds an MSc degree in petroleum engineering from
m D ≈ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (B-6)
(1 +  ) AD
Although we use constant-rate-production equations in this paper,
the productivity index can also be used for variable-rate condi-
tions. A more detailed discussion around computation of the gas
productivity index (including variable-rate production) is given by
Medeiros et al. (2008).
Erdal Ozkan is a professor of petroleum engineering and co-
director of MCERS at CSM. E-mail: eozkan@mines.edu. He holds
BS and MS degrees from Istanbul Technical University and a
PhD degree from the University of Tulsa, all in petroleum engi-
neering. Ozkan’s main research interests are pressure-tran-
sient analysis, modeling fluid flow in porous media, horizontal
and multilateral well technology, and unconventional reser-
voirs. He has served as the Executive Editor of SPE Res Eval &
Eng and chaired several SPE forums, workshops, and confer-
ences. Ozkan is a member of the SPE Reservoir Description
and Dynamics Advisory Committee. He is the recipient of the
April 2011 SPE Reservoir Evaluation & Engineering
CSM. Rajagopal Raghavan retired from Phillips Petroleum as its
formation evaluation manager in Bartlesville, Oklahoma, USA.
His interests are in well performance and reservoir description
methods. Raghavan has served SPE in various capacities over
the years, including serving as Senior Technical Editor, and is
a Distinguished and an Honorary Member of SPE. Hossein
Kazemi is the Chesebro’ Distinguished Professor of Petroleum
Engineering at CSM and Codirector of MCERS. E-mail: hka-
zemi@mines.edu. He holds BS and PhD degrees in petroleum
engineering from the University of Texas at Austin (1961 and
1963, respectively). Kazemi is a Distinguished and an Honorary
Member of SPE and has served as Distinguished Author and
Speaker for SPE. He retired from Marathon in 2001 after serving
as senior technical consultant, director of production research,
manager of reservoir technology, and executive technical
fellow at Marathon Petroleum Technology Center in Littleton,
Colorado, USA. At CSM, Kazemi has taught graduate courses
in petroleum engineering and supervised graduate research
in reservoir modeling, well testing, and improved oil and gas
recovery.
259
 Gas Permeability of Shale
A. Sakhaee-Pour and Steven L. Bryant, The University of Texas at Austin
Summary
Permeability is one of the most fundamental properties of any res-
ervoir rock required for modeling hydrocarbon production. How-
ever, shale permeability has not yet been understood fully because
of the complexities involved in modeling flow through nanoscale
throats. In this paper, we analyze the effects of adsorbed layers of
methane (CH4) and of gas slippage at pore walls on the flow
behavior in individual conduits of simple geometry and in net-
works of such conduits. The network is based on scanning elecron
microscope (SEM) images and a drainage experiment in shale. To
represent the effect of adsorbed gas, the effective size of each
throat in the network depends on the pressure. The hydraulic con-
ductance of each throat is determined on the basis of the Knudsen-
number (Kn) criterion (Knudsen 1909). The combined effects of
adsorption and slip depend strongly on pressure and on conduit di-
ameter. The results indicate that laboratory measurements made
with N2 at ambient temperature and 5-MPa pressure, which is typi-
cal for a transient pulse-decay (TPD) method, overestimate the gas
permeability at early life of production by a factor of four. This ra-
tio increases if the measurement is run at ambient condition
because the low pressure enhances the slippage and reduces the
thickness of the adsorbed layer. Moreover, the permeability
increases nonlinearly as the in-situ pressure decreases during pro-
duction. This effect contributes to mitigating the decline in the pro-
duction rate of shale-gas wells. Laboratory data available in the
literature for CH4 permeability at pressures below 7 MPa agree
with model predictions of the effect of pressure.
Introduction
Shale has provoked a great deal of research recently because of
the considerable volume of natural gas stored in reservoirs. The
natural gas produced from these reservoirs is called shale gas and
is envisaged to provide a substantial fraction of US gas produc-
tion, with some reports of as much as half by 2020 (Polczer
2009). However, debate continues about how much gas can be
produced from these types of reservoirs (Urbina 2011).
An accurate measurement of shale permeability is challenging
because it is so small, on the order of 10–21 m2 (nanodarcies). The
challenge appears in both experimental and theoretical investiga-
tions. For instance, constant-pressure steady-state-flow measure-
ment is not used for shale because it requires a considerable time
(Mallon and Swarbrick 2008). Consequently, other approaches
such as TPD (Billiotte et al. 2008; Cui et al. 2009) and crushed
rock (Luffel et al. 1993) are commonly employed. These methods
reduce the time required for measurement by changing the bound-
ary condition and sample size. The test is altered from steady state
to transient in TPD by changing the constant-pressure boundary
conditions to pressures that vary with time. Therefore, it is con-
ducted more quickly because sustained flow at steady state is not
attempted. In the crushed-rock method, the rock is broken into
small pieces so the transient flow of gas out of the pieces is more
rapid.
Studies of gas flow through shale have invoked a different set
of parameters. For example, the gas evacuated from the core was
modeled in terms of conventional permeability and desorption
(Javadpour et al. 2007). Because many of the voids in shale are
Copyright V
C 2012 Society of Petroleum Engineers
This paper (SPE 146944) was accepted for presentation at the SPE Annual Technical
Conference and Exhibition, Denver, 30 October–2 November 2011, and revised for
publication. Original manuscript received for review 28 June 2011. Revised manuscript
received for review 17 January 2012. Paper peer approved 29 March 2012.
August 2012 SPE Reservoir Evaluation & Engineering
within organic material, methane molecules can adsorb to an
appreciable extent. For instance, the influence of the adsorbed
layer on the void space was considered and the adsorbed porosity
was introduced (Cui et al. 2009). In addition, the volume of
adsorbed gas was estimated using reservoir simulation (Cipolla
et al. 2010). The volume and producibility of adsorbed methane
have direct implications for reserves and, thus, have been widely
discussed. However, the impact of an adsorbed layer on perme-
ability has been less widely explored. Yet because the voids are
so small, the adsorbed layer can also affect the transport proper-
ties of the shale.
This paper analyzes the dependency of the in-situ shale-matrix
permeability on reservoir pressure. In this regard, we implement
the effect of an adsorbed layer in simple conduit geometries as a
function of gas pressure. Subsequently, we adopt noncontinuum
flow models to account for nonzero slip velocity in the conduits.
The presence of an adsorbed layer of gas molecules has a geomet-
ric implication (it reduces the cross section available for transport)
and an influence on microscale boundary condition (slippage of
gas being transported through the pore). Here, we examine a mac-
roscopic consequence of these phenomena—namely, the effect on
the permeability of the rock to gas—by implementing these phe-
nomena in a network model of the void space. The effect of
adsorption and slippage depends on the size of the conduit, and
though a network is not a literal model of shale pore space, it does
serve as a convenient vehicle for examining the effect of a distri-
bution of pore sizes. The results have implications for interpreting
measurements of shale permeability to gas at laboratory condi-
tions and for interpreting flow rates from production wells over
time. The flow behavior from a rock mass into a production well
is controlled by resistance of flow paths that are combinations of
matrices and fractures. Thus, for operational decisions, the effect
of pressure on matrix permeability should be incorporated with
natural- and induced-fracture dependency on the pressure (Philip
et al. 2005).
We refer to laboratory and field conditions extensively in the
following discussion, and for the sake of clarity, we define them
here. The temperature is equal to 300 and 360 K at the laboratory
and field conditions, respectively. The pressure at the laboratory
condition is presumed to be 5 MPa, which is a common pressure
for permeability measurement using TPD (Billiote et al. 2008).
The gas permeability of shale at this condition is on the order of
10 nanodarcies. We assume the thickness of the adsorbed layer is
negligible at laboratory condition. This is a plausible assumption
because the ratio of no-slip hydraulic conductance without an
adsorbed layer to no-slip hydraulic conductance with an adsorbed
layer at 5 MPa is close to unity, regardless of throat size, as we
will show later in the paper. For estimating the gas conductance
and permeability, we further suppose that gases at the laboratory
and in-situ conditions are N2 and CH4, respectively. For conven-
ience, we use the terms “liquid conductance” and “liquid perme-
ability” to refer to hydraulic conductance computed with a
continuum model and no-slip boundary condition at pore walls.
We use the terms “gas conductance” and “gas permeability” to
refer to the conductance when slippage is taken into account.
Both “gas” and “liquid” conductance can be computed when an
adsorbed layer is present. We denote the gas and liquid perme-
abilities at the laboratory condition by kg,lab and kl,lab, respec-
tively. The pressure varies at the field condition, controlling the
thickness of the adsorbed layer. The pressure also affects the Kn
and, hence, the slippage. For the field condition, the gas and liq-
uid permeabilities are represented by kg,in-situ and kl,in-situ,
respectively.
401

TABLE 1—FLUID-FLOW REGIMES DEFINED BY
RANGES OF Kn
Kn 0–10–3 10–3–10–1 10–1–101
Flow regime Continuum Slip Transition Free-molecular
Gas-Flow Regimes
Kn (Knudsen 1909) is used to differentiate flow regimes in con-
duits at micro- and nanoscale. This nondimensional parameter is
defined as
k assumption, the gas permeability relative to the no-slip permeabil-
Kn ¼ ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ
K
where k denotes the mean free path of gas molecules and K the
characteristic length of the channel. The flow regime changes
from continuum model to discrete particles as the Kn increases.
Substituting the mean free path yields the following relation (Roy
et al. 2003):
rffiffiffiffiffiffiffiffiffi
p l
Kn ¼  ; . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
2RT qK
where l is the gas viscosity, q the density, T the prevailing tem-
perature, and R the specific gas constant. The specific gas constant
is the ratio of universal gas constant R to the molar mass m. On
the other hand, the length scale K is equal to the diameter and
aperture height in the circular tube and slit, respectively. For N2
and CH4 at the range of temperatures (300 to 360 K) and pres-
sures (5 to 28 MPa) of interest here, the range of values of k is 0.3
to 2.2 nm. For pore sizes K in the range of 3 to 30 nm, we thus en-
counter values for Kn in the range of 10–2 to 100.
Continuum Regime. The flow regime is categorized on the basis
of Kn, as summarized in Table 1 (Roy et al. 2003). This table indi-
cates that the Navier-Stokes equations with no-slip boundary con-
dition are appropriate only for small Kn (Kn < 10–3). In other
words, the classical continuum-model assumptions (momentum
transfer by means of bulk phase viscosity, fluid velocity matches
solid velocity at walls) cannot always provide accurate results.
Slip Flow Regime. The ratio of molecule/wall to molecule/mole-
cule collisions increases as Kn increases. However, the molecule/
molecule interaction is still dominant in the slip flow regime.
Therefore, Navier-Stokes equations were suggested (Karniadakis
et al. 2005) to remain valid for this domain (103< Kn < 101).
A first-order slip-boundary condition was applied to include the
effect of molecule/wall collisions (Roy et al. 2003). After imple-
menting the tangential momentum coefficient as shown in Appen-
dix A, the gas permeability can be estimated as
kg ¼ kl ð1 þ a1 Kn Þ; . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ
where kl is the permeability of the conduit to liquid (when the no-
slip boundary condition applies), kg the gas permeability of the
conduit, and a1 the permeability enhancement. The liquid perme-
ability is assumed constant (i.e., independent of pressure) and esti-
mated on the basis of the characteristic length size of the throat
from conventional theory. The coefficient a1 depends on the ge-
ometry of the conduit. It is approximately 5 for circular cross sec-
tions (see Appendix A).
The slip flow regime can also be simulated using the dusty gas
model (DGM) (Mason and Malinauskas 1983). This model adopts
a linear combination of gas-transport mechanisms to predict the
overall flow rate. It was mainly based on empirical observation
(Graham 1833, 1846, 1849, 1863), and the theoretical explanation
was proposed later (Mason and Malinauskas 1983). The total
mass flux of a single-size conduit is calculated as
402
kl
Jtotal ¼ Jvisc þ JKn ¼  nrp  DKn rn
l
 
kl rp
¼ p þ DKn ;
>101 l RT                   ð4Þ
where rp is the pressure gradient and DKn is the Knudsen diffu-
sivity coefficient. This relation can also be written in terms of gas
permeability, as
 
kl rp kg p rp
Jtotal ¼  p þ DKn ¼ ; . . . . . . . . . . . ð5Þ
l RT l RT
where kg is the single-phase gas permeability. By substituting the
Knudsen diffusivity coefficient and implementing the ideal-gas
ity in terms of Kn is yielded
 
kg DK l
¼ 1þ n ¼ 1 þ k1 Kn ; . . . . . . . . . . . . . . . . . ð6Þ
kl kl p
where k1 is the permeability enhancement equal to 13.58 for the
circular cross section, as clarified in Appendix A. This relation
indicates that the first-order models (Eq. 3) and the DGM (Eq. 6)
suggest a similar trend for the gas permeability (i.e., a linear func-
tion of Kn).
Eq. 6 is the analog of Klinkenberg’s correction (Klinkenberg
1941) for a single conduit. In the Klinkenberg correction, the perme-
ability to gas is evaluated at several pressures for a core sample
instead of a single conduit. For steady-flow experiments, the pres-
sure is replaced by the average of the inlet and outlet pressures across
the core. The term DKknl l is determined empirically as the slope of the
measured gas permeabilities plotted against the reciprocal pressure.
While the first-order slip model and the DGM indicate a quali-
tatively similar trend for the gas permeability, the rate of change
with Kn is different. The high Kn is attained by decreasing the
throat size in the first-order slip model. In this case, molecule/mol-
ecule collisions are not necessarily infrequent; rather, they are less
frequent than molecule/wall collisions. On the other hand, the low
density of the gas is the reason for the slippage in the DGM. Note
that reductions in both the throat size and the density increase Kn
(see Eq. 2). Consequently, the permeability enhancements are
obtained by measuring the flow rates. The in-situ condition for gas
shale corresponds to the large gas density (approximately 100 kg/
m3). Kn in this situation is large because shale conduits are very
small. Thus, in this study, we use the first-order slip model in the
range of 103< Kn < 101.
Transition Flow Regime. The physics of the transition flow re-
gime is complicated, and most models are used to predict the
computational results of the Monte Carlo simulations (Karniada-
kis et al. 2005). The proposed models (Karniadakis et al. 2005)
implement different shear stress laws in the Navier-Stokes equa-
tions. The higher-order gradients of the velocity are implemented
in the proposed models, and the velocity profile is subject to
higher-order boundary conditions. To capture the flow rate, the
corresponding mass-flow rate for a circular tube was proposed as
(Karniadakis et al. 2005)
 
  4Kn pR4
M_ ¼ 1 þ 1:358tan1 ð4Kn0:4 Þ  Kn 1 þ qrp;
1 þ Kn 8l
                   ð7Þ
where M_ is the mass flux and q is the average density. The coeffi-
cients are obtained by fitting the equation to the Monte Carlo sim-
ulation results. We can express the apparent permeability by
dividing the mass-flow rate to the no-slip condition. Therefore,
the ratio of gas permeability to no-slip permeability can be
expressed as
 
kg  1 0:4
 4Kn
¼ 1 þ 1:358tan ð4Kn Þ  Kn 1 þ : . . . . ð8Þ
kl 1 þ Kn
August 2012 SPE Reservoir Evaluation & Engineering
 14
12
10
Fraction (%)
8 phase is flowing. The purpose of this calculation is simply to iso-
6
4 sidering the SEM images (Ambrose et al. 2010; Wang and Reed
2 the conductance of the circular tube depends on the fourth power
0
0 10 20 30 40 50
d (nm)
Fig. 1—Pore-size distribution of a Barnett shale sample
obtained from mercury-intrusion capillary pressure.
This indicates that the nonlinearity of permeability increases with
Kn. For the sake of convenience, we develop a polynomial form
for the permeability enhancement for 0.1 < Kn < 0.8, which lies
within the range of interest for shale-gas reservoirs. A second-
order polynomial works well:
kg
¼ 0:8453 þ 5:4576Kn þ 0:1633Kn2 ; . . . . . . . . . . . . . ð9Þ
kl
where the coefficients are calculated using the nonlinear regres-
sion model (Mendenhall et al. 1989), and the coefficient of regres-
sion is found to be 0.99. Note that Eq. 9 is valid only for
transition flow regime. Also, the gas permeability is no longer a
linear function of Kn. This reveals that the Klinkenberg correction
cannot be employed for higher-Kn flow regimes.
Free Molecular Regime. For completeness, we include the free
molecular regime, which is relevant for gas-phase transport at am-
bient conditions. The mass-flow rate of the free molecular regime
was modeled by Knudsen (1909). This model implements flux
because of a density gradient of the molecules as follows:
JKn ¼ DKn rni ; . . . . . . . . . . . . . . . . . . . . . . . . . . ð10Þ
where JKn is the mass flux of component i, DKn the Knudsen diffu-
sivity coefficient, and rni the density gradient. Unlike ordinary
diffusion, only one component is considered to predict the flow
rate in this mechanism. This model requires Knudsen diffusivity,
which is usually measured experimentally (Reinecke and Sleep
2002). It was derived analytically only for a long tube with a con-
stant circular cross-sectional area (Roy et al. 2003), as follows:
rffiffiffiffiffiffiffiffiffi
d 8RT d
DKn ¼ ¼ v; . . . . . . . . . . . . . . . . . . . . . . ð11Þ
3 pm 3 layer and slippage, if the network elements are slits rather than
where d is the tube diameter, m the molar mass, T the ambient
temperature, R the universal gas constant that is equal to 8.314 in
SI units, and v the average velocity of the molecules.
Adsorbed Layer
The effect of adsorbed gas is neglected in modeling flow through
conventional rock. This is reasonable because the occupied vol-
ume is negligible compared to the total void space in conventional
rocks. However, the adsorbed volume of CH4 is crucial in shale
because the throats are often smaller than 10 nm and much of the
void space is in the organic material, for which CH4 has a large
affinity. While molecular adsorption can occur on any surface, the
August 2012 SPE Reservoir Evaluation & Engineering
adsorbed layer of interest here is on the organic material. This
layer diminishes in thickness as the pressure decreases.
We analyze the effect of the adsorbed layer on the single-
throat conductance. The ratio of the throat conductance without
the adsorbed layer to that of the in-situ condition (gas adsorbed at
in-situ temperature and pressure) is evaluated. The conductance
without the adsorbed layer is computed by assuming a liquid
late the geometric effect of the adsorbed layer. We assume the
thickness of the adsorbed layer is 0.7 nm at 28 MPa (Ambrose
et al. 2010) and decreases linearly with reducing pressure. We fur-
ther suppose the cross sectional area of the throat is circular, con-
2009). For the calculation of liquid (i.e. no-slip) flow, we assume
of characteristic size, in agreement with Stokes flow of a Newto-
nian fluid (Sakhaee-Pour 2012). The influence of the adsorbed
layer is important for throat sizes smaller than 50 nm, which con-
stitute a significant fraction of shale pores. For a typical shale
sample, the characteristic throat size of 6 nm corresponds to the
largest fraction of conduits, as shown in Fig. 1.
We build our pore-space model on the basis of the SEM
images of shale and the drainage experiment. Some images of
Barnett shale available in the literature are shown in Fig. 2. The
high-resolution images indicate many throats exist in organic
materials. The affinity of gas molecules for organic materials
means that these throats constitute the void space that is altered
by desorption. It has been argued that the fluid flow in organic
materials is mainly single gas phase (Wang and Reed 2009).
Therefore, we propose a network model for gas permeability that
includes only voids within the organic matter.
Results
We analyze the conductance of a single-sized throat to study the
effects of the adsorbed layer and slippage. Then, we build a net-
work model to examine these effects when the connected throats
exhibit a distribution of sizes. Although the network does not ex-
plicitly represent the arrangement of voids in a shale, it does yield
a first-order estimate of difference in the gas-flow behavior
through the shale at different conditions—for example, high pres-
sure (early in the life of a well) vs. low pressure (after substantial
gas production) or field condition vs. laboratory condition. We
take into account the effects of the adsorbed layer and slippage on
the conductance of each throat in the network. Consequently, we
develop a characteristic plot, mapping the measurements taken at
typical laboratory conditions to the in-situ conductance of each
throat.
Analysis of a Single Cylindrical Conduit. Before considering
the network, we investigate the effects of the adsorbed layer and
flow slippage on the conductance of a single throat. In this regard,
we consider pore-size distribution (inferred from mercury poros-
imetry) and SEM of the Barnett shale sample shown in Figs. 1
and 2, respectively. We assume the void space is a network of cy-
lindrical throats. Our analysis (Sakhaee-Pour 2012) shows qualita-
tively similar flow behavior, in terms of response to the adsorbed
cylinders.
We begin by analyzing the effect of the adsorbed layer on the
liquid conductance of a single-sized throat. The goal is simply to
quantify the geometric effect of reducing the throat size as the
thickness of the adsorbed layer increases. Thus, slip is ignored;
the calculations assume no slip on the adsorbed layer. The role of
pressure in this analysis is merely to change the thickness of the
layer. The results are presented in terms of the ratio of the throat
conductance without the adsorbed layer to the throat conductance
with the adsorbed layer (see Fig. 3). We observe that the adsorbed
layer is crucial in estimating the conductance (with a no-slip
boundary condition), especially at larger pressures and in smaller
pore throats in the ranges of interest. The ratio of liquid conduc-
tances is larger at smaller throat sizes because the layer thickness,
403

Fig. 2—SEM of Barnett shale (Ambrose et al. 2010; Wang and
Reed 2009).
which is constant at a given pressure, occupies a greater fraction
of the cross section in smaller throats.
Next, we explore the importance of slippage with respect to
the gas conductance. To this aim, we calculate the ratio of gas to
liquid conductances. The conductances are computed assuming
that no adsorbed layer exists. The only effect of pressure is thus to
change Kn and, hence, the degree of slip. The temperature is fixed
at 300 K. As before, the liquid conductance is computed with a
no-slip boundary condition to serve as a reference value. The ratio
404
8 P=28 MPa
P=20 MPa
7
P=10 MPa
P=5 MPa
6
k(Kn=0)/kads
5
1
101
d (nm)
Fig. 3—Effect of adsorbed CH4 layer(s) on hydraulic conduct-
ance of a single cylindrical conduit. The ratio of “liquid” con-
ductance without adsorbed layer [k(Kn 5 0)] to “liquid”
conductance with adsorbed layer (kads) increases as pressure
increases and as conduit diameter decreases. Liquid conduct-
ance refers to single-phase laminar flow with no-slip boundary.
Thickness of adsorbed layer is proportional to pressure.
of gas to liquid conductances is shown in Fig. 4. The calculation
reveals that slippage plays an important role in smaller throat
sizes and at lower pressures. For instance, the gas conductance
of a 6-nm throat, which has the largest fraction of throat sizes in
Fig. 1, is larger than the liquid conductance by a factor of 1.5 at
P ¼ 20 MPa. This ratio increases nonlinearly with the decrease of
pressure, revealing that the slippage becomes more important at
late production.
Now, we consider the combination of the competing effects of
the adsorbed layer and slippage on the gas conductance. The
adsorbed layer reduces the conductance by reducing the cross-sec-
tional area. Slippage enhances the conductance by facilitating the
molecule movements at the throat surface. Both effects depend on
pressure. To analyze this competition, we calculate the ratio of
gas conductance with the adsorbed layer to liquid conductance
without the adsorbed layer. As in the previous analysis, the liquid
conductance is calculated with a no-slip boundary condition to
serve as a reference value. The gas conductance is obtained by
3 P=28 MPa
P=20 MPa
P=10 MPa
k[Kn(P,T=300 K)]/k(Kn=0)
P=5 MPa
2.5
1.5
1
101
d (nm)
Fig. 4—Effect of slip at pore walls on hydraulic conductance of
a single cylindrical conduit [ratio of gas conductance with slip-
page (Kn evaluated at 300 K and at pressure as per legend) to
liquid conductance without slippage k(Kn 5 0)]. Both conduc-
tances assume no adsorbed layer is present. The value of the
Kn changes with pressure.
August 2012 SPE Reservoir Evaluation & Engineering
 3.5 P=28 MPa
P=20 MPa
kin-situ[Kn(P,T=360 K)]/k(Kn=0)
3 P=10 MPa
P=5 MPa
2.5
2 6
1.5
1 2
0.5
0 d (nm)
101
d (nm)
Fig. 5—Combined effect of adsorbed layer and slip on hydraulic
conductance of a single cylindrical conduit [ratio of gas con-
ductance kin-situ with adsorbed layer and slippage (in-situ con-
dition of 360 K and P as per legend) to liquid conductance
k(Kn 5 0) without adsorbed layer and without slippage]. This
behavior is relevant to evolution of permeability during
production.
computing slip flow within the open cross section of the throat
remaining after implementing the adsorbed layer at the specified
pressure. The temperature is held constant at 360 K. Fig. 5 shows
qualitatively different trends, depending on pressure. At large
pressures, the ratio of gas conductance at in-situ conditions to the
corresponding no-slip conductance increases as the pore size
increases. At small pressures, the ratio decreases as the pore size
increases.
This calculation helps us understand which effect governs the
flow behavior at different stages of production. The purpose of
the comparison between gas and liquid conductances is to analyze
the gas-conductance variation vs. a reference value. It compares
the gas conductance (360 K, various P) that is representative of
reservoir condition to liquid conductance, the reference value
evaluated in the absence of an adsorbed layer and ignoring slip.
The analysis shows the adsorbed layer dominates the flow behav-
ior at high pressure (P ¼ 28 MPa in Fig. 5) regardless of the throat
size. That is, the gas conductance is smaller than the reference
conductance for all throat sizes because the adsorbed layer signifi-
cantly reduces the cross section for gas flow. The influence is
greater for smaller throats. However, the slippage has the primary
effect at small pressures (P ¼ 10 MPa or P ¼ 5 MPa). That is, the
gas conductance is greater than the reference conductance for all
throats, with the effect being greater in narrower throats. The
implication of Fig. 5 is that the adsorbed layer should be taken into
account for modeling gas transport during early production, while
the influence of slippage is dominant during later production.
Finally, for the single-tube study, we compare the gas conduc-
tances at laboratory and in-situ conditions. This is useful for
estimating gas permeability in the field from a laboratory measure-
ment. The adsorbed layer is presumed negligible at the laboratory
condition because the pressure (5 MPa) is low; the curve for the
corresponding pressure in Fig. 3 shows that the correction is less
than 25% for all throat sizes. The thickness of the adsorbed layer
varies at the in-situ condition, depending on the pressure. Note that
the slippage occurs in both conditions. We calculate the gas con-
ductance at the laboratory condition (T ¼ 300 K) from the liquid
conductance, accounting for slippage at laboratory pressure in the
relevant flow regime (Kn criterion evaluated at 300 K and 5 MPa)
by means of Eq. 2, Eq. 3, or Eq. 9. For the in-situ condition, we
compute the cross-sectional area open to flow, depending on the
thickness of the adsorbed layer. The gas conductance is computed
from the liquid conductance at the field pressure after implement-
ing the slippage. Fig. 6 shows that the gas conductance at labora-
August 2012 SPE Reservoir Evaluation & Engineering
12 P=28 MPa
P=20 MPa
klab[Kn(P=5 MPa, T=300 K)]
10 P=10 MPa
/kin-situ[Kn(P,T=360 K)]
8
4
0
101
Fig. 6—The ratio of gas conductance klab without adsorbed
layer and with slippage (laboratory condition of 300 K and 5
MPa) to gas conductance kin-situ with adsorbed layer and slip-
page (field condition of 360 K and P as per legend) increases as
pressure increases and as cylindrical throat size decreases.
These curves are relevant to estimating field permeability from
laboratory measurements.
tory condition is notably higher than at the in-situ condition at
larger pressures. This is because of the lack of an adsorbed layer
and the enhancement of slippage at laboratory condition vs. in-situ
condition. Clearly, the laboratory measurements should be cor-
rected for any flow modeling at the field condition. Considering the
importance of this issue, we will analyze the laboratory measure-
ments using network modeling and propose a correction in the fol-
lowing paragraphs.
Liquid permeability is also of interest in shale, either for the
problem of water production or for the challenge of “tight oil”
production. To estimate liquid permeability at the field condition
from gas permeability measured at the laboratory condition, we
calculate the ratio of gas to liquid conductances of a single cylin-
drical conduit. As in the preceding, the adsorbed layer at the labo-
ratory condition is negligible because of low pressure. Therefore,
the gas conductance at the laboratory condition is computed from
the liquid conductance by implementing the slippage at the labo-
ratory pressure and temperature (5 MPa, 300 K). For the liquid
conductance at the in-situ condition, we calculate the throat area
not obstructed by the thickness of the adsorbed layer. The thick-
ness of the adsorbed layer depends on pressure. We ignore the
swelling effect for the in-situ condition, which may reduce the hy-
draulic conductance. This may occur because of the presence of
water in the formation, which is not available at the same amount
at the laboratory condition. The ratio of gas to liquid conductan-
ces at laboratory and field conditions is shown in Fig. 7. The
results reveal the gas permeability obtained at the laboratory con-
dition is significantly larger than the liquid permeability at the in-
situ condition. The ratio of permeabilities estimated here is larger
than in the preceding analysis (Fig. 6) because the liquid does not
slip at laboratory condition. In reality, the ratio of gas to liquid
conductances may be even greater than in Fig. 7 because the hy-
draulic conductance of the throat decreases because of possible
swelling clays available in the formation.
Analysis of Network of Cylindrical Conduits of Distributed
Sizes. We employ a regular square lattice network (Bennett and
Myers 1962) to model the gas flow through organic materials in
shale. We adopt the pore-size distribution from the drainage
experiment shown in Fig. 1. We also assume the width of the net-
work (transverse to the direction of flow) and the throat length are
17 and 50 nm, respectively. The model width is chosen so that the
network model yields a desired porosity. Here, 17 nm is the value
that results in the desired porosity of 10%. The small width means
that the network corresponds to a thin 2D slice through the or-
ganic material oriented along the axes of the circular holes. That
is, we consider flow perpendicular to the plane of the images in
405
 25 P=28 MPa
klab[Kn(P=5 MPa, T=300 K)]/kads P=20 MPa
P=10 MPa
20
P=5 MPa
15
10
5 1.5
0
1
10
d (nm)
Fig. 7—For a single cylindrical tube, a laboratory measurement
of gas permeability greatly overestimates the permeability to
liquid in the reservoir, as shown by the ratio of gas conduct-
ance klab with slippage and without adsorbed layer (laboratory
condition of 300 K and 5 MPa) to liquid conductance kads with-
out slippage and with adsorbed layer (field condition of 360 K
and P as in legend). Pressure determines the thickness of the
adsorbed layer.
Fig. 2. The objective is not to represent the actual connectivity of
the voids, but to capture the influence of the adsorbed layers and
slip in a connected collection of throat sizes. When the throat
sizes are distributed randomly on the network, the gas permeabil-
ity of the network is 93 nanodarcies at laboratory condition [i.e.,
without the adsorbed layer; the calculation accounts only for slip
in each throat, using Eqs. 9 and 3 (high Knudsen flow models) to
relate flow in each throat to the pressure gradient along that
throat]. The network permeability is a plausible overestimate of
the measured value because the network represents only the or-
ganic portion of the shale. The organic matter accounts for 10%
of the core volume, so prorating the network contribution accord-
ingly would yield an estimate of 9 nanodarcies, which is typical
for shale (Billiotte et al. 2008).
Effect of Laboratory Conditions vs. Field Conditions on Gas
Permeability. First, we evaluate the gas permeability at in-situ
conditions by considering the influence of the adsorbed layer on
flow through each conduit in the network. To compute the gas
conductance at the in-situ condition, we calculate the liquid con-
ductance with no-slip boundary condition based on the cross-sec-
tional area, depending on the thickness of the adsorbed layer. The
gas conductance of each throat is then computed from the liquid
conductance.
The ratio of network permeabilities at laboratory conditions to
those at in-situ conditions is presented in Fig. 8. The results show
the measurements made at the laboratory condition would overes-
timate the permeability by a factor of four at P ¼ 28 MPa (i.e., at
the start of production). This factor is almost equal to the ratio of
laboratory to in-situ permeabilities for a single tube of 6 nm (see
Fig. 6), which has the largest fraction of pore-throat sizes as
deduced from the mercury-intrusion data shown in Fig 1. This
means that the throats, of which characteristic sizes constitute the
largest population, dominate the flow behavior. The overestima-
tion would be even larger if the prediction were based on labora-
tory measurements at ambient pressure and temperature. This is
because of enhancement in slippage occurring at low pressures.
The ratio reaches unity at a slightly larger pressure than 5 MPa
because the field temperature (360 K) is greater than the labora-
tory temperature (300 K) and because the gas for the laboratory
measurement is N2 and not CH4. The difference in temperatures
makes Kn slightly larger at the laboratory condition, but for a
given pressure, temperature, and throat size, the properties of N2
yield a smaller Kn than for CH4. Thus, smaller slippage occurs for
406
4.5
4
3.5
kg, lab/kg, in-situ
3
2.5
2
1
0.5
5 10 15 20 25 30
P (MPa)
Fig. 8—The ratio of gas permeability with slippage and without
adsorbed layer (laboratory condition of 300 K and 5 MPa) to gas
permeability with adsorbed layer and slippage (field condition
of 360 K and P from the x-axis) obtained from a network of cy-
lindrical throats having the size distribution shown in Fig. 1.
N2 than for CH4 because of a smaller Kn, resulting in a smaller
permeability. Thus, the gas permeability of the network model
exposed to CH4 at a slightly larger pressure than 5 MPa at field
condition is equal to the N2 permeability at the laboratory
condition.
The ratio of laboratory to in-situ permeabilities decreases as
the pressure decreases. This is because of the opposing effects of
the adsorbed layer and slippage at different pressures. To provide
a quick tool for estimating the in-situ permeability from the labo-
ratory measurement, we fit a curve to the results in Fig. 8. The
nonlinear regression model indicates the coefficient of regression
is 0.99. Therefore, the in-situ gas permeability can be estimated
from the laboratory measurement using the following equation:
kg;lab
¼ 0:001P2 þ 0:0898P þ 0:783; . . . . . . . . . . . ð12Þ
kg;in-situ
where kg,in-situ is the gas permeability at the in-situ condition, kg,lab
is the gas permeability at the laboratory condition (300 K, 5
MPa), and P is the gas pressure in MPa.
Effects of Field Conditions on Gas Permeability During
Production. Finally, we investigate the effect of declining field
pressure on the gas permeability using the network model. In this
regard, we calculate the ratio of the field permeability at low pres-
sures (kg2,in-situ) to the gas permeability value at the start of pro-
duction (kg1,in-situ). The initial field pressure is presumed to be 28
MPa. We take into account the effect of the adsorbed layer,
depending on the pressure. In addition, we account for the effect
of pressure on slippage. The ratio of the gas permeabilities is pre-
sented in Fig. 9. The network modeling suggests that the shale
permeability at the start of production is significantly smaller than
during late production. The adsorbed layer is the reason for this
effect. Furthermore, the dependency of the permeability on the
pressure is nonlinear. Hence, the gas permeability of shale is
highly dependent on pressure, unlike the conventional reservoir.
This has major implications in reservoir simulations and ultimate-
recovery-estimation models.
Validation Against Laboratory Data. Figs. 8 and 9 summa-
rize the main implications of the theory. Fig. 8 estimates gas
permeability at the in-situ condition from the laboratory measure-
ment, and Fig. 9 predicts the enhancement of gas permeability
during production. Two sets of experiments are required to vali-
date these implications. The first set requires comparison of the
measurements with N2 at the laboratory condition and CH4 at the
reservoir condition. For the second set, we must analyze the
August 2012 SPE Reservoir Evaluation & Engineering
 4.5
4
3.5
kg 2, in-situ /kg1, in-situ
3
2.5
2
1.5
1 1
5 10 15 20 25 30
P (MPa)
Fig. 9—The ratio of gas permeability at pressures below initial
reservoir pressure (kg2,in-situ) to gas permeability at initial pres-
sure (P 5 28MPa) (kg1,in-situ) increases as production continues
and pressure declines accordingly. The ratio is calculated from
a network of cylindrical throats having the size distribution
shown in Fig. 1.
dependency of rock resistance against gas flow on the pore pres-
sure when the moving fluid is methane. The test with methane
should be run at reservoir temperature and pressures to represent
the in-situ conditions. Our model predicts that the effect of the
adsorbed layer becomes more important as the organic content of
shale increases. The organic-rich region has more tendancy to
adsorb methane to its pore wall, which must be considered when
choosing the core samples.
Data across the range of conditions of Figs. 8 and 9 for rocks
with different organic content are not currently available. How-
ever, measurements of CH4 transport reported on a particular type
of shale at pressures below 7 MPa (Letham 2011) let us partially
test the effect of slippage. The experimental data were measured
with methane for a sample with a permeability of 390 nanodarcies
at 5 MPa and 300 K. The effective stress was kept constant in the
laboratory measurements (Letham 2011); that is, the confining
stress was lowered to yield the same effective stress at a lower
pore pressure. The porosity of the laboratory sample was reported
to be 13%. The porosity of our original sample was 10%, which is
slightly different from the laboratory sample. To compensate for
these differences, we increased the pore-throat sizes of our origi-
nal network model, of which permeability
qwas
ffiffiffiffiffiffi 9 nanodarcies
at 5
390
MPa and 300 K, by a factor of 6.58 9 ¼ 6:58 . We also
increased the width of the network by a factor of 43.3 (¼ 6.582) to
keep the porosity of the network model unaffected by the change
in the pore-size distribution. Using the modified model, we com-
pute gas permeabilities for a range of small pressures (between 1
and 6 MPa) and at 300 K. Since the effective stress was kept con-
stant during laboratory measurements, we do not change the
modified pore-throat-size distribution with pore pressure. (In the
field, effective stress increases with reservoir depletion, and the
effect on pore-throat-size distribution, which is not included in
the model presented here, would diminish the slippage effect in
Fig. 10.) Fig. 10 plots these gas permeabilities, normalized by the
nominal liquid permeability at laboratory condition. Similar to ev-
ery calculation performed for the laboratory condition through
this study, the gas flow here is with slippage and without the
adsorbed layer, and the liquid flow is without slippage and with-
out the adsorbed layer. However, unlike the laboratory condition
we defined in which pressure was 5 MPa, we change the pressure
to investigate its influence on gas permeability because of slip-
page. The liquid permeability for the laboratory data is 390 nano-
darcies, estimated by extrapolating the measured gas permeability
to large pressure. Fig. 10 shows that the model slightly underesti-
August 2012 SPE Reservoir Evaluation & Engineering
1.9
1.8 Laboratory data (Letham 2011)
1.7 Corrected throats model
klab(P, T=300 K)/k(Kn=0)
1.6
1.5
1.4
1.3
1.2
1.1
0 1 2 3 4 5 6 7
P (MPa)
Fig. 10—The ratio of gas permeability without an adsorbed
layer and with slippage [klab(P, T5300 K)] to the liquid perme-
ability without an adsorbed layer and without slippage
[k(Kn 5 0)] is obtained from a network of conduits and com-
pared to the laboratory measurements (Letham 2011). The
effective stress is constant, and hence, the pore-pressure
change does not affect pore-throat-size distribution. Gas per-
meability increases with lowering pressure because of slip-
page, unlike the liquid permeability, which is constant. The
flowing gas is methane.
mates the normalized gas permeability, the maximum difference
with the laboratory measurement being only 6%. This indicates
that the slippage model provides a good estimate for the effect of
pressure on gas permeability at moderate pressures.
Conclusions
Pore throats of shale are mostly narrower than 10 nm and are
inside organic material on which CH4 adsorbs. As a result, the gas
permeability of these rocks is affected significantly by the
adsorbed gas and by the slip of flowing gas on the pore walls (or
on layers of molecules adsorbed on the walls). The adsorbed layer
does not play an important role in conventional rocks because the
pore throats are much wider, nor is the gas volume desorbed from
the organic material significant in conventional reservoirs. To bet-
ter understand gas-flow behavior through shale, we evaluated
these effects in individual conduits of a cylindrical cross section
and in simple networks of such conduits. The effect of the
adsorbed layer was treated as purely geometric: The cross section
open to flow was reduced by the thickness of the adsorbed layer,
which was assumed to vary linearly with pressure. The effect of
slip was accounted for by applying the model appropriate to the
flow regime, according to the Knudsen number (Kn). The latter
depends on pressure, temperature, and conduit diameter. For the
slip flow regime, the first-order slip model was judged more suita-
ble than the dusty gas model (DGM) for the shale-gas application.
At large pressures such as typical initial shale-gas-reservoir
pressures, the effect of the adsorbed layer dominates the effect of
slip on gas-phase permeability. Slip dominates at smaller pres-
sures typical of those after longer periods of production. Conse-
quently, the reservoir matrix permeability is predicted to increase
significantly throughout the life of a well, by a factor of 4.5, as
production continues and pressure declines. The models predict
that the typical conditions for laboratory measurements of perme-
ability cause those values to overestimate field permeability by as
much as a factor of four. The model results are captured in simple
analytical expressions that allow convenient estimation of these
effects.
For complete validation of the proposed model, laboratory
measurements at elevated pressures (on the order of 28 MPa) and
reservoir temperature made with CH4 are needed. Experimental
data at such conditions, for which the adsorbed layer is expected
407
to have the dominant effect, are not available. Comparing model burgh, Pennsylvania, USA, 23–25 February. http://dx.doi.org/10.2118/
predictions with the laboratory data at lower pressures (<7 MPa) 131772-MS.
and constant effective stress permits evaluating the importance of Bennett, C.O. and Myers, J.E. 1962. Momentum, Heat, and Mass Transfer.
slippage. When normalized by the nominal absolute permeability New York: Series in Chemical Engineering, McGraw-Hill.
of the network (i.e., the value at which both slip and adsorbed Billiotte, J., Yang, D., and Su, K. 2008. Experimental study on gas permeabil-
layers are negligible), the predicted trend agrees well with the ity of mudstones. Physics and Chemistry of the Earth, Parts A/B/C 33
measurements. The maximum difference between the predicted (Supplement 1): S231–S236. http://dx.doi.org/10.1016/j.pce.2008.10.040.
normalized gas permeability and the measured value is 6%. This Cipolla, C.L., Lolon, E.P., Erdle, J.C. et al. 2010. Reservoir Modeling in
means that our adapted network model provides a reasonable ba- Shale-Gas Reservoirs. SPE Res Eval & Eng 13 (4): 638–653. SPE-
sis for understanding the effect of gas pressure on the matrix per- 125530-PA. http://dx.doi.org/10.2118/125530-PA.
meability of a shale-gas reservoir. Cui, X., Bustin, A.M.M., and Bustin, R.M. 2009. Measurements of gas
permeability and diffusivity of tight reservoir rocks: different
approaches and their applications. Geofluids 9 (3): 208–223. http://
Nomenclature
dx.doi.org/10.1111/j.1468-8123.2009.00244.x.
d ¼ tube diameter Graham, T. 1833. On the law of the diffusion of gases. J. Membr. Sci. 100
DKn ¼ Knudsen diffusivity coefficient (1): 17–21. http://dx.doi.org/10.1016/0376-7388(94)00228-Q.
JKn ¼ Knudsen flux Graham, T. 1846. On the Motion of Gases. Phil. Trans. R. Soc. Lond. A
Jtotal ¼ total flux 136 (1846): 573–631.
Jvisc ¼ viscous flux Graham, T. 1849. On the Motion of Gases. Part II. Phil. Trans. R. Soc.
k(Kn ¼ 0) ¼ no-slip conductance for a single cylin- Lond. A 139 (1849): 349–391.
drical tube with no adsorbed layer Graham, T. 1863. On the Molecular Mobility of Gases. Phil. Trans. R.
k[Kn(P,T ¼ 300 K)] ¼ gas conductance for a single cylindri- Soc. Lond. A 153 (1863): 385–405.
cal tube with slippage and without an Javadpour, F., Fisher, D., and Unsworth, M. 2007. Nanoscale Gas Flow in
adsorbed layer
Shale Gas Sediments. J Can Pet Technol 46 (10): 55–61. JCPT Paper
kads ¼ no-slip conductance for a single cylin-
No. 07-10-06. http://dx.doi.org/10.2118/07-10-06.
drical tube in the presence of an
Karniadakis, G., Bes kök, A., and Aluru, N. 2005. Microflows and Nano-
adsorbed layer
flows: Fundamentals and Simulation, Vol. 29. New York: Interdisci-
kg ¼ single-phase gas permeability
plinary Applied Mathematics, Springer. ISBN 0387221972.
kg,in-situ ¼ gas permeability of the network
Klinkenberg, L.J. 1941. The permeability of porous media to liquids and
model at the in-situ condition
gases. API Drilling & Production Practice (1941): 200–213.
kg,lab ¼ gas permeability of the network
Knudsen, M. 1909. Die Gesetze der Molekularströmung und der inneren
model at the laboratory condition
Reibungsströmung der Gase durch Röhren (The laws of molecular and
kg1,in-situ ¼ gas permeability of the network
viscous flow of gases through tubes). Ann. Phys. 333 (1): 75–130.
model at the start of production repre-
senting the in-situ condition http://dx.doi.org/10.1002/andp.19093330106.
kg2,in-situ ¼ gas permeability of the network Letham, E.A. 2011. Matrix permeability measurements of gas shales: gas
model at a different reservoir pressure slippage and adsorption as sources of systematic errors. BSc thesis,
representing the in-situ condition University of British Columbia, Vancouver, Canada.
kin-situ[Kn(P,T ¼ 360 K)] ¼ gas conductance for a single cylindri- Luffel, D.L., Hopkins, C.W., and Schettler Jr., P.D. 1993. Matrix Perme-
cal tube at the in-situ condition with ability Measurement of Gas Productive Shales. Paper SPE 26633 pre-
slippage and an adsorbed layer sented at the SPE Annual Technical Conference and Exhibition,
kl ¼ permeability of a conduit to liquid Houston, 3–6 October. http://dx.doi.org/10.2118/26633-MS.
with no-slip boundary condition Mallon, A.J. and Swarbrick, R.E. 2008. How should permeability be meas-
klab[Kn(5 MPa,300 K)] ¼ gas conductance for a single cylindri- ured in fine-grained lithologies? Evidence from the chalk. Geofluids 8
cal tube at the laboratory condition (1): 35–45. http://dx.doi.org/10.1111/j.1468-8123.2007.00203.x.
with slippage and without an adsorbed Mason, E.A. and Malinauskas, A.P. 1983. Gas Transport in Porous
layer Media: The Dusty-Gas Model. Amsterdam, The Netherlands: Chemi-
Kn ¼ Knudsen number cal Engineering Monographs, Elsevier Science.
m ¼ molar mass Mendenhall, W., Reinmuth, J.E., and Beaver, R. 1989. Statistics for Man-
M_ ¼ mass flux agement and Economics. Boston, Massachusetts: PWS-Kent Publish-
p ¼ gas pressure ing Company.
P ¼ gas pressure in MPa Philip, Z.G., Jennings Jr., J.W., Olson, J.E. et al. 2005. Modeling Coupled
R ¼ universal gas constant Fracture-Matrix Fluid Flow in Geomechanically Simulated Fracture
R ¼ specific gas constant Networks. SPE Res Eval & Eng 8 (4): 300–309. SPE-77340-PA.
T ¼ temperature http://dx.doi.org/10.2118/77340-PA.
a1 ¼ permeability enhancement Polczer, S. 2009. Shale expected to supply half of North America’s gas.
k ¼ mean free path of gas molecule Calgary Herald, 9 April 2009.
K ¼ characteristic length of the channel Reinecke, S.A. and Sleep, B.E. 2002. Knudsen diffusion, gas permeability,
l ¼ gas viscosity and water content in an unconsolidated porous medium. Water Resour.
q ¼ gas density Res. 38 (12): 1280. http://dx.doi.org/10.1029/2002wr001278.
Roy, S., Raju, R., Chuang, H.F. et al. 2003. Modeling gas flow through
microchannels and nanopores. J. Appl. Phys. 93 (8): 4870–4879.
http://dx.doi.org/10.1063/1.1559936.
Acknowledgments
Sakhaee-Pour, A. 2012. Gas flow through shale. PhD dissertation (in pro-
The authors acknowledge financial contributions from Exxon- gress), The University of Texas at Austin, Austin, Texas.
Mobil and Jackson School of Geosciences at The University of Urbina, I. 2011. Insiders Sound an Alarm Amid a Natural Gas Rush. N.Y.
Texas at Austin. Times, 25 June 2011, http://www.nytimes.com/2011/06/26/us/26gas.
html?_r¼1&pagewanted¼all.
References Wang, F.P. and Reed, R.M. 2009. Pore Networks and Fluid Flow in Gas
Ambrose, R.J., Hartman, R.C., Diaz-Campos, M. et al. 2010. New Pore- Shales. Paper SPE 124253 presented at the SPE Annual Technical
scale Considerations in Shale Gas in Place Calculations. Paper SPE Conference and Exhibition, New Orleans, 4–7 October. http://
131772 presented at the SPE Unconventional Gas Conference, Pitts- dx.doi.org/10.2118/124253-MS.

408 August 2012 SPE Reservoir Evaluation & Engineering

Appendix A ð ð1    
2  rv
First-Order Slip Model. The velocity profile in the slip flow re- Q ¼ us ðrÞ2prdr ¼ a 1  x2 þ 2 Kn xdx
gime for a cylindrical conduit is obtained by adding a correction rv
0
term to the solution of the Navier-Stokes equation after applying
the Newtonian-fluid assumption with no-slip boundary condition ð1      
Q 2  rv 1 2  rv
(Roy et al. 2003): Q¼ 1  x2 þ 2 Kn xdx ¼ þ Kn ;
a rv 4 rv
0
us ¼ uno-slip þ ucorrection ; . . . . . . . . . . . . . . . . . . . . . ðA-1Þ
                   ðA-8Þ
where us is the velocity profile in the slip flow regime, uno-slip is where Q is the flow rate, a is a characteristic flow rate, and Q is
the no-slip velocity of the Newtonian fluid, and ucorrection is the the nondimensional flow rate. The first term on the right-hand
correction term to implement the effect of slippage. The correc- side of Eq. A-8 represents the no-slip flow rate, and the second
tion term is constant, independent of location. The slip velocity term is the enhancement because of slippage. Therefore, the effect
profile can also be expressed in a nondimensional form as of slip on permeability can be readily measured by dividing
QðKn Þ by QðKn ¼ 0Þ :
Us ¼ Uno-slip þ Ucorrection ; . . . . . . . . . . . . . . . . . . . . ðA-2Þ

1 2rv  
where the nondimensional no-slip velocity is defined as follows: kg 4 þ rv Kn 2  rv
¼ 1
¼ 1 þ 4 Kn : . . . . . . . . . ðA-9Þ
r
2 kl 4
rv
Uno-slip ¼ 1  ¼ 1  x2 ; . . . . . . . . . . . . . . . . ðA-3Þ
R Here kg is the permeability of the conduit to gas phase, and kl is
where R is the radius of the tube, r is the distance from the center the permeability of the same conduit when the no-slip boundary
of the tube, and x is the normalized radial distance from the center condition applies. The latter condition usually holds when a liquid
of the tube, which is equal to unity at the wall. To include the fills the conduit, hence the choice of subscript l. After implement-
effect of molecule/wall collisions, the first-order slip boundary ing rv ¼ 0.9, we obtain the enhancement of apparent permeability,
condition is imposed as (Klinkenberg 1941) called a1 in the text (refer to Eq. 3), to be equal to 5.

2  rv @Us  DGM Expression in Terms of Kn. The apparent permeability of
Us j w ¼ Kn ; . . . . . . . . . . . . . . . . . . . . ðA-4Þ
rv @n w the conduit based on the DGM is calculated as
 
where n is the outward (unit) vector, which is normal to the tube kg DK l b
¼ 1þ n ¼ 1 þ : . . . . . . . . . . . . . . . . . ðA-10Þ
wall, and rv is the tangential momentum-accommodation coeffi- kl kl p p
cient. The tangential momentum accommodation is calculated by
measuring the gas-flow rate at the slip flow regime and is close to The second term on the right-hand side of Eq. A-10 can be simpli-
0.9 (Roy et al. 2003). fied after substituting the Knudsen diffusivity coefficient in terms
The first-order slip boundary condition is applied to derive the of a pertinent parameter, using Eq. 11. We also use the Hagen-
nondimensional velocity profile (see Eq. A-2) (Klinkenberg Poiseuille model for the permeability term, as follows:
1941): rffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi
    d 8RT 8RT
l
@Us  @Uno-slip @Ucorrection  @Uno-slip  b DKn l 3 pm  l 32 1 pm
¼ þ ¼ ¼ ¼ ¼
@n w @n @n 
w @n w p kl p d2 3 d p
p
¼ ð2xÞjx¼1 ¼ 2:                  ðA-5Þ 32
rffiffiffiffiffiffiffiffiffi
2RT
The minus sign is because of the direction of the normal vector. l
64 1 pm
The correction term is constant, and, thus, its gradient vanishes. ¼ :                    ðA-11Þ
Therefore, Eq. A-4 simplifies to (Klinkenberg 1941) 3 d p
   
2  rv @Us  2  rv Here, we implement the ideal-gas assumption p ¼ q mR T ¼ qRT
K ¼ 2 Kn : . . . . . . . . . . . . . . ðA-6Þ to express the pressure in terms of density, which yields the
rv @n w
n
rv results in terms of Kn:
Thus, we can express the nondimensional velocity profile based rffiffiffiffiffiffiffiffiffi
2RT rffiffiffiffiffiffiffiffiffi
on Eqs. A-2 and A-3 as follows (Klinkenberg 1941): l
b 64 1 pm 64 l 2m
  ¼ ¼
2  rv p 3 d R 3 qd pRT
U s ¼ 1  x2 þ 2 Kn : . . . . . . . . . . . . . . . . ðA-7Þ q T
rv r mffiffiffiffiffiffiffiffiffi
128 l p
To cast this profile in terms of permeability, we obtain the flow ¼ ¼ 13:58Kn :              ðA-12Þ
3p qd 2RT
rate as

August 2012 SPE Reservoir Evaluation & Engineering 409

 What Is Stimulated Reservoir Volume?
M.J. Mayerhofer, Pinnacle; E.P. Lolon, CarboCeramics; N.R. Warpinski, Pinnacle; C.L. Cipolla, CarboCeramics;
D. Walser, Pinnacle; and C.M. Rightmire, Forrest A. Garb and Associates
Summary
Ultralow-permeability shale reservoirs require a large fracture
network to maximize well performance. Microseismic fracture
mapping has shown that large fracture networks can be gener-
ated in many shale reservoirs. In conventional reservoirs and tight
gas sands, single-plane-fracture half-length and conductivity are
the key drivers for stimulation performance. In shale reservoirs,
where complex network structures in multiple planes are created,
the concepts of single-fracture half-length and conductivity are
insufficient to describe stimulation performance. This is the rea-
son for the concept of using stimulated reservoir volume (SRV)
as a correlation parameter for well performance. The size of the
created fracture network can be approximated as the 3D volume
(stimulated reservoir volume) of the microseismic-event cloud.
This paper briefly illustrates how the SRV can be estimated from
microseismic-mapping data and is then related to total injected-
fluid volume and well performance. While the effectively produc-
ing network could be smaller by some proportion, it is assumed
that the created and the effective network are directly related.
However, SRV is not the only driver of well performance. Fracture
spacing and conductivity within a given SRV are just as important,
and this paper illustrates how both SRV and fracture spacing for a
given conductivity can affect production acceleration and ultimate
recovery. The effect of fracture conductivity is discussed separately
in a series of companion papers. Simulated-production data are
then compared with actual field results to demonstrate variability
in well performance and how this concept can be used to improve
completion design, well spacing, and placement strategies.
Introduction
Fisher et al. (2002), Maxwell et al. (2002), and Fisher et al. (2004)
were the first papers to discuss the creation of large fracture networks
in the Barnett shale and show initial relationships between treatment
size, network size and shape, and production response. Microseis-
mic-fracture-mapping results indicated that the fracture-network size
was related to the stimulation-treatment volume. Fig. 1 shows the
relationship between treatment volume and fracture-network size for
five vertical Barnett wells, showing that large treatment sizes resulted
in larger fracture networks. It was observed that as fracture-network
size and complexity increase, the volume of reservoir stimulated
also increases. Fisher et al. (2004) detailed microseismic-fracture-
mapping results for horizontal wells in the Barnett shale. This work
illustrated that production is related directly to the reservoir volume
stimulated during the fracture treatments. In vertical wells, larger
treatments are the primary way to increase fracture-network size and
complexity. Horizontal-well geometry provides other optimization
opportunities. Longer laterals and more stimulation stages can also
be used to increase fracture-network size and SRV. Mayerhofer et al.
(2006) performed numerical reservoir simulations to understand
the impact of fracture-network properties such as SRV on well
performance. That paper also showed that well performance can be
related to very long effective fractures, forming a network inside a
very tight shale matrix of 100 nanodarcies or less.
Fig. 2 illustrates the various types of fracture growth, rang-
ing from simple fractures to very complex fracture networks.
Complex-fracture networks are desirable in “supertight” shale
Copyright © 2010 Society of Petroleum Engineers
This paper (SPE 119890) was accepted for presentation at the SPE Shale Gas Production
Conference, Fort Worth, Texas, 16–18 November 2008, and revised for publication. Original
manuscript received for review 10 September 2008. Paper peer approved 27 March 2009.
February 2010 SPE Production & Operations
reservoirs because they maximize fracture-surface contact area
with the shale through both size and fracture density (spacing).
Chances for creating large tensile-fracture networks are increased
by pre-existing healed or open natural fractures and favorable
stress-field conditions such as a small difference in principal hori-
zontal stresses. Fig. 3 shows an example of a complex-fracture
network (in a vertical well) measured with microseismic mapping.
The figure illustrates the development of a large-scale network with
two distinct, orthogonal fracture orientations. This paper aims to
expand on the papers cited by discussing in more detail the concept
of SRV and its relationship with shale-well performance.
Estimating SRV
It has been well documented by Albright and Pearson (1982),
Warpinski et al. (2004), and Rutledge and Phillips (2003) that
microseismic events are created mainly as a result of shear slip-
pages around the hydraulic fractures. The mechanisms include
shear slippages induced by altered stresses near the tip of the
fractures, and shear slippages related to leakoff-induced pore-
pressure changes. In conventional reservoirs with higher reservoir
permeabilities and/or oil/water reservoirs, the microseismic-event
cloud can be fairly wide but may not be related to the generation of
complex-fracture networks. In this case, most of the events may be
related to pore-pressure increase as a result of rapidly moving pres-
sure transients in high-permeability formations and/or reservoirs
with fairly incompressible fluids. In supertight shale reservoirs,
diffusivity-related pore-pressure changes cannot move very far
from the actual fracture planes, unless natural fractures in other
directions are opened and hydraulically enhanced as a network
structure, thus serving as a conduit for fluid movement. This means
that a large event cloud structure must be approximately equivalent
to the actual fracture-network size. Thus, the microseismic-event
cloud structure observed by microseismic fracture mapping pro-
vides a means to estimate the SRV in very tight reservoirs.
Fig. 4 shows an automated method using discrete bins to esti-
mate stimulated reservoir area (SRA) from microseismic-mapping
data in a horizontal well. Constant-width bins (i.e., 100-ft wide) are
drawn in the principal fracture direction from the wellbore to the
farthest event in the specific bin on both sides of the wellbore. The
individual-bin areas are then summed up to approximate the total
SRA. The calculation of SRV (a 3D structure) requires an estimate
of the stimulated-fracture-network height in each discrete bin within
the contacted shale section, in addition to knowing the approximate
SRA. This calculation is also performed within the selected bins
and is performed by calculating the network height as the differ-
ence between the shallowest and deepest event within the specific
bin and top and bottom of the shale section (Fig. 5). While this
method is not an analytically exact calculation, it does provide a
fast automated method to approximate a very complex 3D structure,
while honoring the contact with the actual shale section. The SRV
in this paper is specified in millions of ft3 or in acre-ft.
An important aspect of the SRV measurement is the proper
setup of observation well or multiple observation wells to guar-
antee that the entire SRV can be observed. The proper design
of a microseismic-mapping setup takes into account maximum
observation distance to ensure that the entire SRV can be imaged.
Microseismic-moment magnitude vs. distance plots (Zimmer et al.
2007) can be used to ascertain if all events were within observable
range or if the SRV could in fact be larger than imaged. Other
issues that affect correlations between SRV and well performance
include associated formation-water production and condensate
yield, which becomes relevant in less-mature shales close to the
89
 45000

40000

35000

Fracture Network Length (ft)

30000

25000

20000

15000

10000

5000

0
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Fluid Volume (bbl)

Fig. 1—Relationship of total fracture-network length as a function of total job fluid volume pumped (Fisher et al. 2004).

oil window. Although general guidelines are applicable for SRV dimensions are given by the total fracture-network length (2xf) and
measurements, it is important to evaluate any given SRV measure- width (xn). However, the key property for well production is really
ment in conjunction with the particular reservoir setting. the total sum of all fracture-network segments (linear feet) within
the SRA, which is a strong function of fracture spacing (density).
SRV and Fracture Spacing in Shales Eq. 1 shows the calculation of total fracture length, L ftotal , for
It is important to note that the SRV is just the reservoir volume the entire SRV as a function of fracture-network half-length, xf ,
affected by the stimulation. It does not provide any details of the width, xn, and fracture spacing, Δxs:
effectively producing fracture structure or spacing. Maxwell et al.
(2006) introduced a concept that could eventually be used to charac- 4 x f xn
L ftotal = + 2 x f + x n . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
terize fracture density. In this approach, “additional seismic-signal ⌬x s
characteristics allow investigation of the source of the mechanical
deformation resulting in the microseisms. In particular, the seismic 1500
moment, a robust measure of the strength of an earthquake or All microseisms
microearthquake, can be used to quantify the seismic deformation.”
Besides this potential geophysical approach, reservoir modeling
1000 ~2,700 ft
also provides an avenue to evaluate the effectively producing net-
work better. The details of this approach will be described in the
following section. As an introduction, Fig. 6 shows an identical
SRV with different fracture spacings (densities) and the effect of 500
fracture spacing on the gas-recovery factor. The SRV in this graph is
approximately 2,000 × 106 ft3 (for h = 300 ft). In contrast to conven-
tional single-fracture modeling using fracture half-length, the SRA Frac
well
Northing (ft)

Simple fracture
Fracture Complex
Complexfracture
Fracture –500 ~1,200 ft

Possible
–1000
aligned
features
–1500

Observation well
–2000
–1000 –500 0 500 1000 1500
Complex fracture
Fracture Complex fracture
Fracture Easting (ft)
With fissure
with Fissureopening
Opening Network
network
Fig. 3—Microseismic fracture mapping shows complex-net-
Fig. 2—Types of fracture growth (Warpinski et al. 2008). work growth in shales (Warpinski et al. 2008).

90 February 2010 SPE Production & Operations

 Fig. 4—Estimating SRA from microseismic-mapping data.

Fig. 7 shows a plot of Eq. 1 on a log-log plot for the given SRV
in Fig. 6. The resulting curve forms an approximate straight line
on a log-log plot. As an example, reducing fracture spacing from
Free spacing: 200 ft Free spacing: 100 ft
300 to 50 ft would result in a more than five-fold increase in total
fracture length (from 48,600 to 264,600 ft) and would accelerate
the 3-year cumulative gas recovery by about the same multiplier.
This illustrates the importance of viewing SRV in context with the

Xf
Treatment horizontal well
Xn After 1 year production
Gas Recovery Factor (%)

Events outside of shale section

Time (day)

Fig. 6—Effect of fracture spacing (⌬xs) on well performance

and recovery factor for SRV of approximately 2,000 × 106 ft3
Fig. 5—Conceptual sketch of SRV-height calculation. (Mayerhofer et al. 2008; Warpinski et al. 2008).

February 2010 SPE Production & Operations 91

 1.E+06

Total Fracture Length (ft)

1.E+05

1.E+04
10 100 1000

Fracture Spacing (ft)

Fig. 7—Fracture spacing vs. total fracture length based on Eq. 1 for SRV of 2,000 × 106 ft3.

potential fracture spacing. A dual-porosity approach is not used at
this point because the fracture spacing is still relatively sparse and
the numerical approach better illustrates the linear flow patterns in
a low-permeability shale system.
SRV and Well Performance
Previous publications have briefly discussed the relationship
between actual SRV and well performance. Here we will expand
on the subject with more details and comparisons with longer-term
production data from the Barnett shale. Fig. 8 shows a plot of
SRV measured from microseismic fracture mapping vs. 6-month
cumulative gas production from a group of horizontal wells in one
specific county in the Barnett shale. The plot shows that SRVs in
this county range from as low as 120 × 106 ft3 (2,755 acre-ft) to
approximately 1,900 × 106 ft3 (43,618 acre-ft). Small SRVs were
generated partly as a result of shorter laterals and smaller treatment
sizes (fluid volumes). While the data show some scatter, it clearly
indicates a trend that large SRV’s will result in better well perfor-
mance. A similar trend is still valid when plotting SRV vs. longer
term 3-year cumulative gas production (Fig. 9). The scatter in the
data is to be expected as a result of variations in fracture spacing
within the SRV and potential differences in effective network size
and conductivity. While the presented trends in Figs. 8 and 9 show
a general relationship between SRV generated from microseismic
mapping and production data, they do not provide any deterministic
quantification of how the actual effective network (portion of the
network contributing to gas production) is structured to provide the
specific gas production. The quantification of the effectively pro-
ducing fracture network can be performed only with flow simula-
tions, such as those provided by a numerical reservoir simulator.
Previous work has provided extensive numerical reservoir mod-
eling (Mayerhofer et al. 2006, Warpinski et al. 2008, Cipolla et al.

1000

900

800
6-month Cum Gas (MMscf)

700

600

500

400

300

200

100

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
SRV (Mio ft3)

Fig. 8—SRV trend vs. 6-month cumulative horizontal-well production for one Barnett-shale county.

92 February 2010 SPE Production & Operations

 2500
2000
3-year Cum Prod (MMscf)
1500
1000
500
0
0 400
Fig. 9—SRV trend vs. 3-year cumulative horizontal-well production for one Barnett-shale county.
2008) to show how specific fracture-network properties can affect
well performance. The work demonstrated that numerical model-
ing using a black-oil reservoir simulator can history match well
performance adequately. The numerical modeling describes the
fracture network as a discrete set of high-permeability fractures in
two orthogonal directions. One direction represents the principal
hydraulic-fracture direction as a result of newly created fractures
or the opening of natural fractures. The orthogonal set represents
the opening of conjugate natural-fracture sets. Besides geochemical
shale properties and pore pressure, the key fracture-network proper-
ties that affect well performance include SRV, shale-matrix perme-
ability, fracture spacing, and fracture conductivity. It was discussed
in these papers that once the value for shale permeability is fixed,
the solutions for total cumulative fracture length (the sum of all
fracture segments in the network) and conductivity become fairly
unique. Fig. 10 shows the results from a vertical-well history match
(Mayerhofer et al. 2006), where the network was approximated with
an effectively producing network having almost 8,000 ft of total
Fig. 10—Reservoir-simulation results for vertical well (Mayerhofer et al. 2008).
February 2010 SPE Production & Operations
800 1200 1600 2000
SRV (Mio ft3)
fracture length within a microseismically mapped SRV of 219 ×
106 ft3 (gray dots in graph). Fracture spacing was modeled to be
approximately 180 ft in the principal fracture direction, with a uni-
form fracture conductivity of 4 md-ft and shale-matrix permeability
of 100 nanodarcies (0.0001 md). The fracture network is larger on
the southwest side of the wellbore, with only a few microseismic
events to the northeast.
Fig. 11 shows the simulation setup for a 3,000-ft-long horizon-
tal well, from which an SRV of 5,000 × 106 ft3 with equidistant
fracture spacing of 400 ft was generated. The plot clearly shows
that the ultimate drainage area is constrained to the created SRV in
supertight shale reservoirs. Using this approach, a series of simula-
tions was generated with different SRV sizes and fracture spacings
to evaluate the effect of these properties on well performance. The
SRV was varied from approximately 800 × 106 ft3 to 6,000 × 106
ft3 and equidistant fracture spacing from 50 to 800 ft.
To cross check the generic simulations with actual well-production
data and a more-detailed production history match, the vertical-well
93
 After 1 year After 15 years

k (matrix) = 0.0001 md

SRV = 5000 x 106 ft^3

2,450 ft

3,000 ft
Treatment Horizontal Well Treatment Horizontal Well

Fig. 11—Simulation of horizontal-well fracture network (SRV = 5,000 × 106 ft3, frac spacing = 400 ft, lateral length approximately
3,000 ft).

example in Fig. 10 was plotted against the more-generic simulation
results. Although the well configuration is different and flowing
pressures are not exactly the same, the correlation between SRV,
well performance, and fracture spacing is approximately consistent,
showing that the 6-month cumulative production of this vertical
well would fall on the 200-ft fracture-spacing curve (Fig. 12).
Fig. 13 shows a plot of the same generic simulations with actual
horizontal-well Barnett-shale data throughout the producing areas
in north Texas. The figure shows a wide scatter of well results,
with cumulative gas production ranging from less than 100 to 500
MMscf for SRVs ranging from 100 × 106 ft3 to 3,000 × 106 ft3. The
data indicate that most of the wells plot at a fracture spacing of 200
ft or larger, with many wells at fracture spacings larger than 800 ft,
indicating a sparse fracture network (smaller total fracture length)
within the measured SRV (Fig. 7) and therefore a less effective
fracture network. The plot also shows the significant acceleration
component of smaller fracture spacings on short-term production.
For example, at an SRV of 1,000 × 106 ft3, reducing fracture spacing
from 400 to 100 ft would result in a 3.5-fold increase in production.
The larger the SRV, the more significant the fold of increase in
production with smaller fracture spacing (because of much longer
total fracture length with larger SRV). On the other hand, generat-
ing a larger SRV with large 800-ft fracture spacing will not provide
increased production, which can be explained by the low fracture
conductivity (and associated increasing pressure drop with distance
from the well) in a sparsely fractured network.
After comparing short-term production, the discussion will now
focus on longer-term production (3 to 15 years). Fig. 14 shows a
plot of the SRV from the same generic horizontal-well simulations
vs. 3-year cumulative gas production, fracture spacing, and actual
horizontal-well Barnett-shale data. The data indicate a slight shift
of the real data to smaller fracture spacings, which may indicate
improved cleanup for the longer-term production. The actual cumu-
lative production ranges from as low as 100 to 2,000 MMscf. The
simulations also show that the folds of increase in production as a
function of smaller fracture spacing become less for longer-term

Vertical Well from Mayerhofer et al. 2006 falls on 200 ft frac spacing curve

Network spacing=400 ft Network spacing=200 ft Network spacing=100 ft

Network spacing=50 ft Vertical well Network spacing=800 ft

Fig. 12—Simulated SRV and actual 6-month vertical well production response (Mayerhofer et al. 2008).

94 February 2010 SPE Production & Operations

 Network spacing=400 ft Network spacing=200 ft Network spacing=100 ft
Network spacing=50 ft Vertical well Network spacing=800 ft

Fig. 13—Simulated SRV and simulated 6-month cumulative production vs. actual Barnett-shale data (entire north Texas area).

recovery, as previously illustrated in Fig. 6 (early-time accelera-
tion). The plot also clearly shows that significant improvements in
well performance by generating larger SRVs is more significant at
low to moderate SRVs (less than 1,000 × 106 ft3). For larger SRVs,
the benefits can be achieved only with a simultaneous decrease of
fracture spacing. For example, increasing SRV from 1,500 × 106
ft3 to 3,000 × 106 ft3 with 800-ft fracture spacing will create no
meaningful increase in gas production.
Fig. 15 shows the results for 15-year cumulative produc-
tion. In this case, no actual production data is available, because

3-Year Cumulative Gas vs. SRV

2000

1600
3-Year Cumulative Gas (MMscf)

1200

800

400

0
0 500 1000 1500 2000 2500 3000
SRV (Million cuft)
Network spacing=800 ft Network spacing=400 ft Network spacing=200 ft
Network spacing=100 ft Network spacing=50 ft Actual data

Fig. 14—Simulated SRV and simulated 3-year cumulative production vs. actual Barnett-shale data (entire north Texas area).

February 2010 SPE Production & Operations 95

 15-Year Cumulative Gas vs. SRV
16000

Inputs:
Lateral length, ft: 3000
14000
Reservoir permeability, md: 0.0001 Largest observed SRV per well
Pore pressure, psi 3000
Net pay thickness, ft: 300
15-Year Cumulative Gas (MMscf)

12000 Frac conductivity, md-ft: 4

Minimum BHP, psi: 1000

10000

8000

6000

4000

2000

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000
SRV (Million cuft)
Network spacing=800 ft Network spacing=400 ft Network spacing=200 ft
Network spacing=100 ft Network spacing=50 ft

Fig. 15—Simulated SRV vs. 15-year cumulative production.

horizontal-well development started in 2002 in the Barnett shale.
The plot shows that for longer-term production, the impact of SRV
becomes more significant even at larger fracture spacing because
SRV provides the ultimate drainage area and diffusion has had
time to access the entire network. In this case, increasing SRV
from 1,500 × 106 ft3 to 3,000 x106 ft3 with 800-ft fracture spacing
will result in a 50% cumulative-production increase from 2,000 to
3,000 MMscf (compared to no meaningful increase after 3 years
of production).
Applying SRV and Network Azimuth to
Well-Placement and -Spacing Strategies
Figs. 16 and 17 show conceptual diagrams illustrating how SRV
and fracture-network azimuth can be used for well-placement and
-spacing considerations, respectively. Fig. 16 shows the importance
of lateral orientation with respect to fracture-network azimuth and
SRV. The graph assumes that the lateral will be drilled in a north-
west/southwest orientation. If fracture direction follows the same
southwest-to-northwest trend, the scenario on the left side of Fig. 16

Fig. 16—Importance of SRV and network azimuth for lateral orientation and well placement.

96 February 2010 SPE Production & Operations


Adjacent SRV = Optimizing areal SRV 3
coverage but not recovery factor?
SRV 2 Well 3
SRV 1
SRV 1 Well 2
Well 1
Fig. 17—Horizontal-well-placement strategy considerations in conjunction with SRV and fracture spacing in shale reservoirs.
will promote longitudinal fractures with a very small SRV. If wells
are to be drilled in this orientation, more wells with closer spacing
will be required to drain a given section of reservoir. The fracture
orientation provided in the center of Fig. 16 provides an oblique ori-
entation of the fracture direction with respect to the lateral and also
results in less-efficient reservoir coverage, but SRV is now larger
than in the longitudinal case. The best scenario with largest SRV
is illustrated on the right side of Fig. 16, with transverse network
orientation providing the largest SRV per lateral.
Keeping these considerations in mind, one could postulate that
well spacing (distance between laterals) should roughly coincide
with the extent of the SRV perpendicular to the wellbore, as illus-
trated in Fig. 17 on the left-hand side (SRV from each lateral is
adjacent to the offset well). This approach may appear to be the
optimum configuration. While the areal coverage of SRV is defi-
nitely optimized with this approach, gas-recovery factors within
the SRV can be optimum only with sufficiently close fracture spac-
ing, as illustrated in Fig. 6. If fracture spacing cannot be optimized
with large SRVs, it may be more beneficial to drill laterals at closer
spacing, with the chance of overlapping SRVs (right-hand side in
Fig. 17). This is the reason for the so-called “simul-fracs” and “zip-
per fracs” in shale reservoirs, where multiple parallel wellbores are
either stimulated simultaneously or in short alternating sequence
from one lateral to the next. The hypothesis in these strategies
is to maximize fracture density between laterals as a result of
stress diversion from simultaneously growing opposing fractures
(simul-fracs) and/or creating new branches as a result of increased
stresses around newly opened fractures. Real data from Barnett-
shale completions may suggest that this approach has merits, but
no specific study has been published on this topic. The challenge
lies in understanding the practical and physical limitations of
what is possible in terms of network size and fracture spacing at
reasonable completion costs (i.e., lateral length, number of stages
and perforations, and treatment size).
Conclusions
1. Low-permeability shale reservoirs require a large fracture
network with small fracture spacing and adequate conductivity
to maximize well performance. The size of the SRV, fracture
spacing, and azimuth are key components for well-placement
and -spacing strategies.
2. Natural features controlling SRV and fracture spacing in shales
include shale thickness, stress field and magnitude, presence of
pre-existing open or healed natural fractures, fracture barriers,
rock brittleness, and geologic features such as faults and karsts.
While these parameters are not under direct human control, it is
February 2010 SPE Production & Operations
SRV 3
Overlapping SRV = Higher recovery factor
(smaller frac spacing)
SRV 2
Well 3
Well 2
Well 1
critical to understand how they will impact well results, comple-
tions and well-development strategies.
3. Engineering measures to increase SRV and fracture spacing
include lateral length and orientation, treatment sizes, number
of stages, perforation clusters, and diversion techniques and/or
openhole packer completion systems. More perforation clusters
and stages in cemented-and-cased completions increase the
likelihood of dense fracturing.
4. The key completion strategy is to balance the creation of a large
SRV with maximum possible fracture density. Large SRVs with
small fracture spacing provide maximum well performance
and gas-recovery factors. However, the practical and physical
limitations of achieving this goal will require an economic
optimization process.
Nomenclature
hf = fracture height, L
xf = hydraulic-fracture wing or half-length, L
xn = hydraulic-fracture network width (from microseismic-
event pattern), L
⌬xs = orthogonal fracture spacing, L
L ftotal = total composite fracture length
Acknowledgments
The authors would like to thank Pinnacle (a Halliburton Service)
and Carbo Ceramics for supporting the publication of this work.
Some of this work is also based on previous publications of Devon
Energy Corporation data, and we thank them for their support of
those efforts.
References
Albright, J.N. and Pearson, C.F. 1982. Acoustic Emissions as a Tool for
Hydraulic Fracture Location: Experience at the Fenton Hill Hot Dry Rock
Site. SPE J. 22 (4): 523–530. SPE-9509-PA. doi: 10.2118/9509-PA.
Cipolla, C.L., Warpinski, N.R., Mayerhofer, M.J., Lolon, E.P., and Vincent,
M.C. 2008. The Relationship between Fracture Complexity, Reservoir
Properties, and Fracture Treatment Design. Paper SPE 115769 pre-
sented at the SPE Annual Technical Conference and Exhibition, Denver,
21–24 September. doi: 10.2118/115769-MS.
Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A.,
and Dunn, K.P. 2004. Optimizing Horizontal Completion Techniques
in the Barnett Shale Using Microseismic Fracture Mapping. Paper SPE
90051 presented at the SPE Annual Technical Conference and Exhibi-
tion, Houston, 26–29 September. doi: 10.2118/90051-MS.
Fisher, M.K., Wright, C.A., Davidson, B.M., Goodwin, A.K., Fielder,
E.O., Buckler, W.S., and Steinsberger, N.P. 2002. Integrating Fracture
97
 Mapping Technologies to Optimize Stimulations in the Barnett Shale.
Paper SPE 77441 presented at the SPE Annual Technical Conference
and Exhibition, San Antonio, Texas, USA, 29 September–2 October.
doi: 10.2118/77441-MS.
Maxwell, S.C., Urbancik, T.I., Steinsberger, N.P., and Zinno, R. 2002.
Microseismic Imaging of Hydraulic Fracture Complexity in the Bar-
nett Shale. Paper SPE 77440 presented at the SPE Annual Technology
Conference and Exhibition, San Antonio, Texas, USA, 29 September–2
October. doi: 10.2118/77440-MS.
Maxwell, S.C., Waltman, C.K., Warpinski, N.R., Mayerhofer, M.J., and
Boroumand, N. 2006. Imaging Seismic Deformation Induced by
Hydraulic Fracture Complexity. Paper SPE 102801 presented at the
SPE Annual Technology Conference and Exhibition, San Antonio,
Texas, USA, 24–27 September. doi: 10.2118/102801-MS.
Mayerhofer, M.J., Lolon, E.P., Youngblood, J.E., and Heinze, J.R. 2006.
Integration of Microseismic Fracture Mapping Results With Numeri-
cal Fracture Network Production Modeling in the Barnett Shale.
Paper SPE 102103 presented at the SPE Annual Technical Conference
and Exhibition, San Antonio, Texas, USA, 24–27 September. doi:
10.2118/102103-MS.
Rutledge, J.T. and Phillips, W.S. 2003. Hydraulic stimulation of natural
fractures as revealed by induced microearthquakes, Carthage Cotton
Valley gas field, East Texas. Geophysics 68 (2): 441. doi: 10.1190/
1.1567212.
Warpinski, N.R., Mayerhofer, M.J., Vincent, M.C., Cipolla, C.L., and
Lolon, E.P. 2008. Stimulating Unconventional Reservoirs: Maximizing
Network Growth While Optimizing Fracture Conductivity. Paper SPE
114173 presented at the SPE Unconventional Reservoirs Conference,
Keystone, Colorado, USA, 10–12 February. doi: 10.2118/114173-MS.
Warpinski, N.R., Wolhart, S.L., and Wright, C.A. 2004. Analysis and
Prediction of Microseismicity Induced by Hydraulic Fracturing. SPE J.
9 (1): 24–33. SPE-87673-PA. doi: 10.2118/87673-PA.
Zimmer, U., Maxwell, S., Waltman, C., and Warpinski, N. 2007. Microseis-
mic Monitoring Quality Control (QC) Reports as an Interpretive Tool
for Nonspecialists. Paper SPE 110517 presented at the SPE Annual
Technical Conference and Exhibition, Anaheim, California, USA,
11–14 November. doi: 10.2118/110517-MS.
SI Metric Conversion Factors
acre × 4.046 873 E+03 = m2
bbl × 1.589 874 E–01 = m3
cp × 1.0* E–03 = Pa·s
ft × 3.048* E–01 = m
*Conversion factor is exact.
Michael Mayerhofer is the technology manager for fracture
diagnostics at Pinnacle, a Halliburton Service in Houston. His
responsibilities include the application of tiltmeter and micro-
seismic hydraulic-fracture-mapping results for optimizing frac-
ture completion, well placement and infill drilling strategies,
the design and evaluation of hydraulic-fracturing treatments,
reservoir engineering, and integrated field studies. His 18-year
involvement with hydraulic fracturing and reservoir engineer-
ing includes fundamental research and real-field applications
in various global producing areas and has resulted in more
than 40 technical papers and journal articles. Prior to joining
Pinnacle Technologies in 1997, he worked for Union Pacific
Resources in Fort Worth, Texas, USA. He holds a PhD degree in
petroleum engineering from Mining University Leoben in Austria.
He was a member of the SPE Well Completions committee from
1998 to 2001 and has served on the JPT editorial committee.
Elyezer Lolon is a senior staff engineer at StrataGen Engineering
(CARBO Ceramics) in Houston. e-mail: elyezer.lolon@strata-
genengineering.com. His responsibilities include designing and
98
evaluating hydraulic-fracturing treatments and reservoir simu-
lation studies, as well as interpreting diagnostic fracture injec-
tion test, pressure transient, and production data. Lolon has 8
years of reservoir engineering experience in the petroleum
industry. He previously worked for Pinnacle Technologies as a
staff petroleum engineer and at Chevron Pacific Indonesia as
a petroleum engineer. He holds a BS degree in petroleum engi-
neering from Bandung Institute of Technology in Indonesia and
MS and Ph.D. degrees in petroleum engineering from Texas
A&M University. Norm Warpinski is the director of technology
development for Pinnacle, a Halliburton Service, in Houston. He
is in charge of developing new tools and analyses for hydrau-
lic-fracture mapping, reservoir monitoring, hydraulic-fracture
design and analysis, and integrated solutions for reservoir devel-
opment. He received his MS and PhD degrees in mechanical
engineering from the University of Illinois, Champaign/Urbana
after receiving a BS degree in mechanical engineering from
Illinois Institute of Technology. He worked for 28 years for Sandia
National Laboratories performing oil, gas, geothermal, waste
repository, and carbon sequestration research before join-
ing Pinnacle in 2005. He served as executive editor of SPE
Production & Operations from 2006–08, as review chair of JPT
from 1990–92, on the SPE R&D committee from 2002–05, as an
SPE distinguished lecturer in 1998–99, and on various confer-
ence and forum committees. Craig Cipolla is chief engineer-
ing advisor for Schlumberger, Hydraulic Fracture Monitoring,
focusing on the application of microseismic fracture mapping
measurements to improve stimulation designs and field devel-
opment. Prior to joining Schlumberger, he was vice president of
stimulation technology for Carbo Ceramics. Prior to the sale of
Pinnacle Technologies’ fracture mapping and reservoir moni-
toring businesses to Halliburton in 2008, Cipolla led Pinnacle’s
engineering consulting and fracture mapping applications
groups for 12 years. Cipolla remained with Carbo Ceramics-
StrataGen Engineering when the consulting arm of Pinnacle
was split off and renamed in 2008. Craig’s 28 years of experi-
ence includes the design and evaluation of hydraulic-fractur-
ing treatments, training engineers in the use of real-time data
analysis, tight gas reservoir engineering, integrated field stud-
ies, and supervising stimulation treatments. In addition, He is an
expert in the application of microseismic and tiltmeter-fracture
mapping technologies. Prior to joining Pinnacle in 1996, Cipolla
held positions with Union Pacific Resources, CER Corporation,
and Dresser Titan. Doug Wasler has worked for Pinnacle for the
past five years. He has 30 years of experience in the Permian
Basin, mid-continent, Appalachia, Rockies, and South Texas with
Dowell Schlumberger, The Western Company of North America,
BJ Services, and Pinnacle. He specializes in the calibration of
3D fracture modeling through a number of methods, includ-
ing historical production transient analysis and calibration by
various fracture mapping processes. He has taught numerous
seminars and short courses on subjects related to his fields of
interest, including Mini-frac Evaluation, 3-D Pressure Matching,
On-Location Stimulation QC, Fracturing Situations, Horizontal
Devonian Completion Best Practices, Fracture Mapping
Techniques, and others. Most recently, he has specialized in
the examination and comparison of the various emerging
resource plays in North America. He has authored 14 papers
and holds three patents in his areas of interest. He holds a BS
degree in natural gas engineering from Texas A&M University.
Mike Rightmire is a senior vice president with Forrest A. Garb
and Associates where he is responsible for reservoir engineer-
ing projects and reserves appraisals. During his 25-year career,
Rightmire also worked for Arco and Pinnacle Technologies. His
tenure at Arco included experience in reservoir and produc-
tion engineering, software development, numerical simula-
tions, chemical transport modeling, and reservoir engineering
training. Responsibilities with Pinnacle Technologies included
reservoir surveillance applications and project management.
Rightmire holds a BS degree in petroleum engineering from
Texas A&M University and a BS degree in biology from the
University of Alaska.
February 2010 SPE Production & Operations
 Modeling of Hydraulic-Fracture-Network
Propagation in a Naturally Fractured
Formation
X. Weng, O. Kresse, C. Cohen, R. Wu, and H. Gu, Schlumberger
Summary
Hydraulic fracturing in shale-gas reservoirs has often resulted in
complex-fracture-network growth, as evidenced by microseismic
monitoring. The nature and degree of fracture complexity must be
understood clearly to optimize stimulation design and completion
strategy. Unfortunately, the existing single-planar-fracture models
used in the industry today are not able to simulate complex frac-
ture networks.
A new hydraulic-fracture model is developed to simulate complex-
fracture-network propagation in a formation with pre-existing
natural fractures. The model solves a system of equations govern-
ing fracture deformation, height growth, fluid flow, and proppant
transport in a complex fracture network with multiple propagat-
ing fracture tips. The interaction between a hydraulic fracture and
pre-existing natural fractures is taken into account by using an
analytical crossing model and is validated against experimental
data. The model is able to predict whether a hydraulic-fracture
front crosses or is arrested by a natural fracture it encounters, which
leads to complexity. It also considers the mechanical interaction
among the adjacent fractures (i.e., the “stress shadow” effect). An
efficient numerical scheme is used in the model so it can simulate
the complex problem in a relatively short computation time to
allow for day-to-day engineering design use.
Simulation results from the new complex-fracture model show
that stress anisotropy, natural fractures, and interfacial friction play
critical roles in creating fracture-network complexity. Decreasing
stress anisotropy or interfacial friction can change the induced-
fracture geometry from a biwing fracture to a complex fracture
network for the same initial natural fractures. The results presented
illustrate the importance of rock fabrics and stresses on fracture
complexity in unconventional reservoirs. These results have major
implications for matching microseismic observations and improv-
ing fracture stimulation design.
Introduction
Hydraulic fracturing in shale gas reservoirs has often resulted in
complex-fracture-network growth, as evidenced by microseismic
monitoring (Fisher et al. 2002; Maxwell et al. 2002; Daniels et al.
2007; Le Calvez et al. 2007). This renders the hydraulic-fracture
models used in the industry today for simulating a biwing planar
fracture in a conventional reservoir unsuitable for designing frac-
ture treatments in a shale-gas formation where complex fracturing
is evident. To overcome the limitation of the conventional fracture
models, hydraulic-fracture models for simulating propagation of
fracture networks consisting of two orthogonal sets of parallel and
uniformly spaced fractures (referred to as a “wiremesh” model in
this paper) have been developed (Xu et al. 2009; Xu et al. 2010;
Meyer and Bazan 2011). Although the assumed fracture geometry
may not accurately represent the actual complex fracture network,
which may consist of irregularly oriented and distributed fractures,
it is a reasonable approximation because it properly accounts for
the increased storage and surface area and mechanical interaction
Copyright © 2011 Society of Petroleum Engineers
This paper (SPE 140253) was accepted for presentation at the SPE Hydraulic Fracturing
Technology Conference, The Woodlands, Texas, USA, 24–26 January 2011, and revised for
publication. Original manuscript received 10 February 2010. Revised manuscript received
11 July 2011. Paper peer approved 9 September 2011.
368
of the fracture network, as compared to biwing fractures, and is
thus able to predict a reasonable network footprint. This is the
case provided that the assumed fracture spacings and stress anisot-
ropy (defined as the difference between maximum and minimum
horizontal stresses) have been calibrated properly against micro-
seismic observations. Even though the microseismic-event cloud
does not provide the precision to delineate the exact boundary of
the stimulated volume and there are inherent uncertainties in the
estimated dimensions, it provides engineers with an independent
measurement to validate the assumed fracture spacings and to
make adjustments if needed because these parameters are highly
uncertain.
While the wiremesh model is a significant improvement over
the conventional biwing-fracture model in terms of being able to
simulate complex fracturing in shale gas, and it provides an esti-
mate of the fracture-network dimensions and proppant placement
in the network, it has some limitations. One of the limitations
is that the fracture-network pattern (i.e., the fracture spacings)
cannot be linked directly to the pre-existing natural fractures.
Calibration against microseismic data is required to obtain frac-
ture spacings consistent with the stimulated volume indicated by
the microseismic data. However, the calibrated fracture spacings
cannot be used easily in other wells, or even other stages in the
same well, if the reservoir characteristics and pumping param-
eters change significantly. When used for parametric studies to
optimize a treatment, the outcome may be skewed because the
model assumes fixed spacings, and any effect of treating param-
eters on the resulting fracture-network pattern is not taken into
account. Another limitation is that network geometry is assumed
symmetric with respect to the injection point, and the shape is
elliptical, which limits the ability to match microseismic data that
show significant asymmetry or irregular shape in the stimulated
region.
A more rigorous general hydraulic-fracture model incorporat-
ing a full solution of coupled elasticity and fluid equations has been
presented by Zhang et al. (2007), but it is limited to the 2D plane-
strain conditions, which is suitable for situations where fracture
length is much greater than height but is not suitable for general
field applications where fractures are often contained in height.
Additionally, the model is computationally intensive and not well
suited for engineering design applications. Olson (2008) presented
a complex-fracture-network model that is capable of predicting
hydraulic-fracture propagation and interaction with pre-existing
natural fractures. However, the model is based on only fracture
mechanics and does not include fluid flow and proppant transport.
A constant fracturing pressure is assumed in the simulation. This
limits its ability to accurately simulate actual hydraulic-fracture
treatments.
In this paper, a new hydraulic-fracture model is presented that
simulates propagation of complex fractures in a formation with
pre-existing natural fractures. The model solves a system of equa-
tions governing fracture deformation, height growth, fluid flow,
and proppant transport in a complex fracture network with mul-
tiple propagating fracture tips. The interaction between hydraulic
fractures and pre-existing natural fractures is taken into account by
use of an analytical crossing model. Numerical-simulation results
and the impact of natural-fracture distribution, interfacial friction,
and stress anisotropy on fracture pattern are illustrated through a
simple example.
November 2011 SPE Production & Operations
 illustrates the hydraulic-fracture network created by the fracturing
fluid in a formation with pre-existing natural fractures.
The equation for flow of a power-law fluid in any given fracture
branch can be expressed using the Poiseuille law (Nolte 1991):
Hydraulic frac 2 n ′+1
n′
2 K ′ ⎛ 4 n′ + 2 ⎞ 1 ⎛ w( z ) ⎞ n′
␣0 = n' ⎜ ⎟ ; ␾ ( n′ ) = ∫ ⎜⎝ ⎟ dz
␾ ( n′ ) ⎝ n′ ⎠ H fl H fl
w ⎠

n ′−1
∂p 1 q q
= −␣0 2 n′+1 , . . . . . . . . . . . . . . . . . . . . . . . . (1)
∂s w H fl H fl

where p is fluid pressure; q is the local flow rate in the fracture;

Hfl is the height of the fluid in the fracture; w(z) is fracture width
Natural frac σh
as a function of depth z; w is the average width; n' and K' are fluid
power-law index and consistency index, respectively; and s is the
σH distance along the fracture.
The local condition for mass balance is given by the continu-
ity equation:
Fig. 1—Diagram of hydraulic-fracture network and pre-existing
natural fractures. ∂q ∂( H fl w) 2hLC L
+ + qL = 0, qL = , t > ␶ 0 (s) , . . . . . . . (2)
∂s ∂t t − ␶ 0 (s)
Unconventional Fracture Model where CL is the total leakoff coefficient, hL is the height of leakoff
The complex-fracture-network model, referred to as an uncon- zone, and ␶ 0 (s) is the time when each element of a fracture is
ventional fracture model (UFM) in this paper, simulates the first exposed to fracturing fluid. Also, the global volume balance
propagation, deformation, and fluid flow in a complex network must be satisfied:
of fractures. The model solves the fully coupled problem of fluid t L (t ) L (t ) t

flow in the fracture network and the elastic deformation of the ∫ Q (t ) d t = ∫ Hw d s + ∫ ∫q L d t d s , . . . . . . . . . . . . . . . . . (3)
fractures, which has similar assumptions and governing equations 0 0 0 0

to those of conventional pseudo-3D fracture models. But instead
of solving the problem for a single planar fracture, UFM solves
the equations for the complex fracture network. Fracture-height
growth is modeled in the same manner as in conventional pseudo-
3D models. A three-layer proppant-transport model, consisting of
a proppant bank at the bottom, a slurry layer in the middle, and
clean fluid at the top, is adopted for simulating proppant transport
in the fracture network. Transport equations are solved for each
component of the fluids and proppants pumped. A key difference
where Q(t) is pump rate, L(t) is the total length of all fracture
branches at time t, and H(s, t) is fracture height. Additionally, at
any junction of fracture branches, the sum of flow rates into the
junction from all branches must be equal to zero. At the wellbore,
the sum of flow rates into all open perforation intervals must be
equal to the prescribed injection rate Q(t); i.e.,
∑ q (t ) = Q(t ),
i i = 1,.. N perf , . . . . . . . . . . . . . . . . . . . . . . . . (4)
i

between UFM and the conventional planar-fracture model is being
able to simulate the interaction of hydraulic fractures with pre-
existing natural fractures (i.e., determining whether a hydraulic
fracture propagates through or is arrested by a natural fracture
when they intersect and whether it subsequently propagates along
the natural fracture). The branching of the hydraulic fracture at the
intersection with the natural fracture gives rise to the development
of a nonplanar, complex fracture pattern. A crossing model that is
extended from the Renshaw and Pollard (1995) interface crossing
criterion, applicable to any intersection angle, has been developed
(Gu and Weng 2010) and validated against the experimental data
(Gu et al. 2011), and it is integrated in the UFM. In addition
to the hydraulic-fracture/natural-fracture interactions, the UFM
also takes into account the interaction among adjacent hydraulic-
fracture branches by computing the stress-shadow effect on each
fracture by the adjacent fractures.
Governing Equations. To simulate the propagation of a complex
fracture network that consists of many intersecting fractures, the
equations express the underlying physics of the fracturing process
must be satisfied. These include the equation expressing the fluid
flow in the fracture network, the equation expressing the fracture
deformation, and the fracture-propagation criterion. In the most
general sense, hydraulic fractures are 3D in nature, and the fracture
planes may not be vertical. This is especially the case when the
initial natural fractures are not vertical, potentially leading to non-
vertical hydraulic fractures when the fracturing fluid opens up the
natural fractures. However, solving a fully 3D nonplanar fracture
problem is extremely computation intensive and not suitable for
engineering design models. In the model presented in this paper,
we make the assumption that all natural and hydraulic fractures
are vertical. Therefore, the model is applicable only to a formation
with vertical or near-vertical natural fractures. Fig. 1 schematically
where qi(t) is the injection rate into each perforation cluster (i =
1,..Nperf) and is determined as a part of the solution.
Eqs. 1 and 2 provide the expressions for solving pressure and
flow distribution in the fracture network for a given fracture width
distribution. On the other hand, fracture-width is related to fluid
pressure through the elasticity equation. To have a computationally
efficient model for general engineering design applications, the
same assumption of 2D plane-strain solution for fracture width
as in conventional pseudo-3D models is adopted. For a vertical
fracture in a layered medium with variable minimum horizontal
stress σh(x, y, z) and fluid pressure p, the width profile can be
determined from an analytical solution given as (refer to Eq. 9
below for detailed form of solution)
w( x , y, z ) = w( p( x , y), H , z ). . . . . . . . . . . . . . . . . . . . . . . . . . (5)
In the limiting case of a fracture in a zone of uniform in-situ
stress and constant height, the width/pressure relation in Eq. 5
reduces to that of a 2D PKN model (here E' is the plane-strain
Young’s modulus):
␲ H( p − ␴n )
w= . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)
2E '
At fracture tips, the following boundary conditions are
satisfied:
p = ␴ n , W = 0, q = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
Eqs. 1 through 7 represent the complete set of governing
equations for the problem and must be solved at each timestep to
find fracture opening, fluid pressure, and flow rate in the entire
fracture network.

November 2011 SPE Production & Operations 369

 At the first timestep, it is assumed that biwing fractures are initi-
ated from all perforation intervals in the lowest-stress layer. A 2D
analytical solution for power-law fluid is used to calculate initial fluid
pressure for a prescribed initial fracture length L0 at a known injection
rate Q0. The initial time t0 can be found from volume balance.
Height Growth. In a multilayered formation, the fracture height
and width profile of a fracture cross section depends on fluid
pressure, the in-situ stress, fracture toughness, layer thickness,
and elastic modulus of each layer covered by the fracture height.
Calculation of height growth takes the same approach as that used
for cell-based pseudo-3D models (Mack and Warpinski 2000).
Each element (equivalent to a “cell” in the cell-based pseudo-3D
models) has its own height. The height and width profile of the
fracture cross section is determined only on the basis of local
pressure and vertical stress profile. Interaction between adjacent
cells is neglected.
Similar to a pseudo-3D model, the fluid flow and the associated
pressure gradient in the vertical direction are negligible except near
the upper- and lower-fracture-tip region. That is,
p = pcp + ␳ f g(hcp − z ), . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
where pcp is the pressure in the fracture at a reference depth hcp
(chosen as the perforation depth) measured from the bottom tip, z
is the position in the fracture, and ␳f is the fluid density.
The stress-intensity factors at the fracture top and bottom tips
are calculated from the pressure inside the fracture, the fracture
geometry (height, in this case), and the layer stresses. A stable frac-
ture height is determined by matching the stress-intensity factors
at the tips to the fracture toughness in the layers that contain the
fracture tips. The relations for the stress-intensity factors and the
width, as functions of pressure and tip positions (fracture height),
are given in Mack and Warpinski (2000):
␲h ⎡ ⎛ 3 ⎞⎤
K Iu = pcp − ␴ n + ␳ f g ⎜ hcp − h ⎟ ⎥
2 ⎢⎣ ⎝ 4 ⎠⎦
2 n −1 ⎡h ⎛ h − 2h ⎞ ⎤
+ ∑ (␴ − ␴ i ) ⎢⎣ 2 arccos ⎜⎝ h i ⎟⎠ − hi (h − hi ) ⎥⎦ ,
␲ h i =1 i +1
␲h ⎡ ⎛ h⎞ ⎤
K Il = pcp − ␴ n + ␳ f g ⎜ hcp − ⎟ ⎥
2 ⎢⎣ ⎝ 4⎠ ⎦
2 n −1
⎡h ⎛ h − 2h ⎞ ⎤
+ ∑ (␴ − ␴ i ) ⎢⎣ 2 arccos ⎜⎝ h i ⎟⎠ + hi (h − hi ) ⎥⎦ ,
␲ h i =1 i +1
and w( z ) =
4
( )
⎡ pcp − ␴ n + ␳ f g hcp − z ⎤ z (h − z )
E′ ⎣ ⎦ T0 − ␴ yy
⎡ ⎛ h − 2hi ⎞
⎢ z⎜
− ⎝
4 n −1 ⎢(hi − z ) cosh 1
+ ∑
␲ E ′ i =1
(␴ i +1 − ␴ i ) ⎢
⎢ −
⎢ + z (hh − z ) arccos ⎜⎛ h
⎝ h ⎠ ⎥⎦
⎢⎣
. . . . . . . . . . . . . . . . . . . . . . . . (9)
where KIu and KIl are the stress intensity-factors at the upper and
lower tips, respectively; w(z) is the width at depth z; ␴n and ␴i are
the in-situ stresses at the top tip and the ith layer, respectively; h
is the fracture height; and hi represents the height from the lower
tip to the top of the ith layer.
This model, called the equilibrium height model, is extended
further to a nonequilibrium height model, in which the pressure
gradient because of the fluid flow in the tip regions in the verti-
cal direction is taken into account. This adds an apparent fracture
toughness proportional to the fracture’s top and bottom velocities
in the equations above. The resulting equations are solved numeri-
cally to determine the tip velocities.
370
Fracture Interaction. Fracture interaction is one of the most
important factors that must be taken into account to model hydrau-
lic-fracture propagation in naturally fractured reservoirs. This
includes the interaction between hydraulic fractures and natural
fractures, as well as between hydraulic fractures.
The interaction between hydraulic and natural fractures is
influenced by the natural fractures’ orientation and geometry, in-
situ stresses, rock properties (e.g., rock modulus, tensile strength,
toughness, permeability, and porosity), fluid properties (e.g., pres-
sure and rheology), and interface properties (e.g., friction coef-
ficient, cohesion, and permeability), and it represents a coupled
rock-deformation/frictional-sliding/fluid-flow problem. It is in
itself a very complex process to model.
The behavior of a hydraulic fracture when it intersects natural
fractures (crossing, arresting, branching, or reinitiating with an off-
set) is one of the root causes of the creation of complex fractures.
If the hydraulic fracture crosses a natural fracture, the hydraulic
fracture remains planar, and there is no complexity generated as
long as the fluid pressure does not subsequently exceed the open-
ing pressure of the natural fracture. Conversely, if the hydraulic
fracture does not cross the natural fracture, but instead opens the
natural fracture and propagates it along, a complex fracture net-
work may form. It is important to determine whether a hydraulic
fracture crosses natural fractures under particular field conditions
of in-situ stress and rock and natural-fracture properties.
Theoretical, numerical, and experimental works have been
conducted by various researchers on fracture propagation across
interfaces (Renshaw and Pollard 1995; Blanton 1982; Warpinski
and Teufel 1987; Thiercelin and Makkhyu 2007; Zhang and Jeffrey
2006). Warpinski and Teufel (1987) observed in their mineback
experiments that the hydraulic fracture was often arrested at the
rock joints, leading to branching of multiple fractures and a com-
plex-fracture pattern. Some of the mechanisms have been studied,
and component models for the mechanisms were proposed. Ren-
shaw and Pollard (1995) developed a simple criterion for predict-
ing whether a fracture will propagate across a frictional interface
orthogonal to the approaching fracture. The criterion is based on
the linear elastic fracture mechanics solution for the stresses near
the fracture tip to determine the stresses required to prevent slip
along the interface at the moment when the stress on the opposite
side of the interface is sufficient to reinitiate a fracture. Compres-
sional crossing will occur if the magnitude of the compression act-
ing perpendicular to the frictional interface is sufficient to prevent
slip along the interface at the moment when the stress ahead of the
fracture tip is sufficient to initiate a fracture on the opposite side
of the interface. The criterion is given as
0.35
0.35 +
−␴ xx kfric
>
1.06
, . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
⎤
⎟⎠ + hi ⎥ where ␴xx and ␴yy are far-field effective stresses (tensile is positive)
h
⎥ acting parallel with and perpendicular to a propagating hydraulic
z − hi ⎥, fracture correspondingly, T0 is rock tensile strength, and kfric is a
⎥ coefficient of friction at the interface. This criterion was validated
2 hi ⎞ ⎥
⎟ by laboratory experiments.
Natural fractures are often not aligned with the contemporary
principal in-situ-stress directions, in which case the intersection
angle of a hydraulic fracture approaching a natural fracture is
between 0 and 90° (angle ␤ in Fig. 2). Therefore, it is important
to extend the Renshaw and Pollard criterion to nonorthogonal
intersections.
The Renshaw and Pollard model recently has been extended
to nonorthogonal intersections by Gu and Weng (2010). Because
the model is based on an analytical solution, it is efficient in com-
putation and suitable for incorporation in the hydraulic-fracture-
network model described in this paper.
For hydraulic-fracture applications, the ratio of the remote
in-situ stresses (␴H to ␴h) is larger than unity. The coefficient of
friction for many rocks is typically between 0.1 and 0.9. Fig. 3
shows the results from the extended criterion for a number of
November 2011 SPE Production & Operations

h
βx
y 75°
interface
x
y
βy
fracture
Fig. 2—Schematic of fracture approaching an interface.
intersection angles. The region to the right of each curve represents
the crossing condition. It can be seen that as the angle decreases
from 90°, crossing becomes increasingly difficult. In other words,
a hydraulic fracture is more likely to be arrested and propagate
along the natural fracture it intersects. The extended criterion also
has been compared to laboratory experimental results and showed
good agreement (Gu et al. 2011).
Stress Shadow. Fracture-network-growth pattern is affected by
the mechanical interaction among the adjacent fractures. Generally
known as the stress-shadow effect, the stress field of one fracture
is perturbed by the opening and shearing displacements of other
nearby fractures. In a 2D, plane strain, displacement discontinuity
solution, Crouch and Starfield (1983) described the normal and
shear stresses (␴n and ␴s) acting on one fracture element induced
by the opening and shearing displacement discontinuities (Dn and
Ds) of all fracture elements as the following:
␴ ni = ∑ j =1 Cnsij Dsj + ∑ j =1 Cnn
N N ij
Dnj,
and ␴ si = ∑ j =1 Cnsij Dsj + ∑ j =1 Cnn
N N ij
Dnj . . . . . . . . . . . . . . . . . (11)
where Cij are the 2D, plane-strain elastic influence coefficients,
and their expressions can be found in Crouch and Starfield (1983).
This method is used in our model to compute the additional stress
induced on each fracture element from the displacements of adja-
cent elements (Fig. 4). In addition, a 3D correction factor suggested
by Olson (2004) was further introduced to the influence coefficients
(Cij in Eq. 11) to account for the 3D effect caused by finite fracture
height that leads to decaying of interaction between any two fracture
elements when the distance increases. This additional normal stress
Dsj
Elem j
σsi Elem i
σni
Fig. 4—2D displacement discontinuity method (DDM) approach:
influence of displacement discontinuities from jth element on
stress of ith element.
November 2011 SPE Production & Operations
10
90°
Stress ratio (hmax / hmin)
60°
45°
30°
15°
H
1
0 0.2 0.4 0.6 0.8 1
x Coefficient of friction
Fig. 3—Plot of crossing criterion for stress ratio > 1 at several
intersection angles (for T0 = 0).
because of the stress-shadow effects is computed at each timestep in
the UFM model and is then added to the initial in-situ-stress field
on each fracture element during pressure and width iteration. At the
time of writing of this paper, only the contribution of the normal dis-
placement (i.e., fracture-opening width) is considered. The effect of
stress shadow, especially the shear stress, on the directional change
of propagating fracture tips is not considered at present.
Proppant Transport. Proppant transport is an important feature
of the modeling work involved in the development of a hydraulic-
fracturing simulator because it determines the final conductivity
of the created fracture network. The model presented in this
paper considers the transport of multiple fluid and proppant spe-
cies inside the fracture network. It accounts for several physical
mechanisms such as bridging, packing, and the settling of proppant
to form a bank, as well as the erosion of the existing bank. To
achieve computational efficiency, a 1D proppant-transport model
in the horizontal direction has been adopted, while the vertical
flow of proppant as a consequence of both settling and bank ero-
sion is computed to track the height of the bank, the slurry, and
the clean fluid locally.
The fluid and proppant species are described in the model
through volumetric concentrations. They are calculated in the trans-
port algorithm as averaged values over the volume inside the frac-
ture element, from the top of the bank to the top of the fracture:
w ′ ⌬x ′
xc +
H 2 2
1
X k ( x ′, y′, z ) d x d y′ d z ,
⌬x ′w ( H − H bank ) H∫bank ∫ ∫
ck =
w ′ ⌬x ′
− xc −
2 2
. . . . . . . . . . . . . . . . . . . . . (12)
where Xk is the volume fraction of a species identified by the index
k, Δx' is the length of the element, H is the height of the fracture,
Hbank is the height of the bank, and ck is the concentration of species
k (fluid or proppant). The 1D transport model inside the fracture is
similar to the one in Adachi et al. (2007) except that it considers
only advective transport in the horizontal direction. For example,
Dnj Eq. 13 describes the variation of fluid concentration cfl,k as governed
by the flow inside the fracture and the leakoff into the matrix:
∂ ( H − H bank ) wc fl ,k
+
(
∂ q fl c fl ,k )=−f c fl ,k , . . . . . . . . . . (13)
∂t ∂x
leakoff
where qfl is the flow rate (m3/s) and fleakoff is the flux of fluid through
the fracture walls (m2/s) divided by the fracture height above the
bank. The equation is similar in the case of proppant species,
except that there is no proppant leakoff into the matrix. For the
calculation of slurry height, a similar transport equation is used to
calculate the volumetric concentration of the slurry. The numerical
implementation of these transport equations uses an explicit
scheme in time and a total variation diminishing method for the
advection term (LeVeque 1992).
371
 ( )
1/ n ′
New guessed ⎡ 1 ␳ prop,k − ␳ fl ⎤
time step vset,k = ⎢ n′−1 gDkn′+1 ⎥ , . . . . . . . . . . . . . . . . (14)
⎢⎣ 3 18 K′ ⎥⎦

Extend all tips by
ΔLi based on velocity
Apply junction/
crossing rules
Solve coupled flow
and elasticity
Eqs. for p, q, w
Iterate
Update Δt based
on mass balance
Apply proppant transport
Update fracture height
where vset,k is the settling velocity for proppant type k, ␳prop,k is its
density, and Dk is its diameter. K' and n' are the averaged flow con-
sistency index and averaged flow behavior index, respectively, both
weighted by the fluid concentration. A more accurate calculation
of the settling velocity has also been implemented, on the basis of
the correlation from Shiller and Naumann (1933), to calculate the
drag-force coefficient on a spherical particle.
The modeling of erosion of the bank is based on the correlation
from Wang et al. (2003), which describes the minimal height of
clean fluid and slurry at given flow conditions, under which the
proppant does not settle and the bank is eroded.
Numerical Solution. At any given timestep, the created hydraulic-
fracture network is represented by connected fracture elements.
New elements are added as fracture tips advance. Many new
elements can be added at all propagating fracture tips in a single
timestep, and new element numbers are assigned to these new
elements. As a result, the element numbers are not continuous
from one element to the next. Therefore, a special code structure
is established to track the neighbors, and special junction elements
at intersection of multibranches are also introduced.
The combined Poiseuille and continuity equations (Eqs. 1
and 2) are discretized in terms of pressure pi and average width
wi defined at the center of each element and flow rate qij defined
at the boundary between Element i and Element j. The resulting
mass-balance equation is written as
d Volfluid (i, t ) = d Volfrac (i, t ) + d Volleakoff (i, t ), . . . . . . . . . . . . (15)

or, in the form of pressure, width, and flow rate at each element,
Compute stress shadow

− ∑ qijt d t = (h tfl ,i wit lit − h tfl−,idt wit − dt lit − dt )

Fig. 5—Structure of propagation-solution loop. j

h ( k , i ) ⎡ hocc ( k , i )li t − ␶ 0 ( k, i ) ⎤
t t
N zones
⎧
+ 2 ∑ ⎨2C L ( k , i ) L ⎢ ⎥
k =1 ⎩ h( k , i ) ⎢ − hocc ( k , i )li
t − dt t − d t
t − ␶ 0 ( k , i ) ⎥⎦
old
⎣
In the vertical direction, the model for proppant considers two hL ( k , i ) t ⎫
+ S p (k, i) ⎡ h ( k , i )lit − h t − d t ( k , i )lit − d t ⎤⎦ ⎬ . . . . . . . . . . . (16)
h( k , i ) ⎣
competing mechanisms: the proppant settling into the bank and the
erosion of the existing bank. The settling is calculated by first cal- ⎭
culating the settling velocity for each proppant and then calculating
the increase of the bank height on the basis of both the proppant Eq. 16 includes the leakoff coefficient CL (k, i) for each Zone k
concentration and the height of the slurry. The settling velocity for covered by Fracture Element i, as well as the height of each zone
each proppant was calculated initially on the basis of the extension h (k, i), leakoff height of each zone hL (k, i), the height of the
t
of the Stokes’s law to power fluids given by Daneshy (1978): occupied part of the zone hocc ( k , i ) at position of Element i, and

Injection Pressure vs. Time Fracture Half-Length vs. Time

1500
33600
1300
33500
Fracture Length (m)
Inj. Pressure (KPa)

33400 1100
33300 900
33200
700
33100
500
33000
UFM
32900 Conventional bi-wing 300 UFM
32800 Conventional bi-wing
100
32700
0 2000 4000 6000 8000 –100
0 1000 2000 3000 4000 5000 6000 7000 8000
Time (sec)
Pumping Time (sec)

Fig. 6—Comparison of injection pressure and fracture half-length between UFM and a conventional model for a biwing fracture.

372 November 2011 SPE Production & Operations

 terms of pressure by using a damped Newton-Raphson method.
TABLE 1—INPUT PARAMETERS FOR THE BIWING CASE
The Newton-Raphson method (or Newton method) is a popular and
3
Injection rate 0.5 m /s powerful method that is applicable to solution of multidimensional
Minimum confining stress 4.8e+07 Pa nonlinear systems of equations. It is based on the simple first-
Young’s modulus 3.5e+10 Pa
order Taylor-series expansion. For any given approximate solution
p, f(p) ≠ 0; otherwise, p is already a solution, an improved solution
Fracture height 10 m p + δp is sought after such that f(p + δp) ≈ f(p) + Jf(p)δp = 0. Here
Viscosity 0.001 kg/m.s Jf(p) is the Jacobian matrix for f, defined as

the spurt Sp (k, i). Fluid-flow rate between adjacent elements is ⎡⎣ J f ( p ) ⎤⎦ =

( ).
∂fi p kj
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
calculated by integrating the Poiseuille equation (Eq. 1) and the ij ∂p j
elasticity equation (Eq. 5). For example, in the limiting case of
constant fracture height, it has the form { }
If ␦ p satisfies the linear system,

qij = ␥ ij ⋅ ␤ij 1/ n '

, ␥ ij =
h fl { } { ( )}
⎡⎣ J f ( p ) ⎤⎦ ␦ p j = − fi p kj . . . . . . . . . . . . . . . . . . . . . . . . (19)
ij

(␣ ⋅ (l + l ) ⋅ ( n′ + 1))
1/ n ′
i j Then (p + δp) is taken as an approximate zero of f, or solution
of f(p) = 0. In this sense, Newton’s method replaces a system of
␤ij = ( pi − ␴ i ) ( ) ( ) ( )
2 n ′+ 2 2 n ′+ 2 2 n ′+ 2 2 n ′+ 2
− p* − ␴ i + p* − ␴ j − pj − ␴ j , nonlinear equations with a system of linear equations that must be
solved repeatedly to converge on an adequate solution. In this way,
pi l j + p j li as it iterates, the solution is updated as
p* = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
li + l j
p kj +1 = p kj + ␤␦ p kj +1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)
where p* is an averaged value of the pressure at the interface
between Elements i and j. Convergence is achieved when some norm associated with
Different forms of flow-rate equation are used and implemented the right-hand side of Eq. 19 (i.e., solution) reaches the specified
in the model for the cases of laminar, turbulent, and Darcy fluid minimum value.
flow. A combination of the above equations leads to a nonlinear After the solution converges, a new timestep starts, and an
system of equations, simplified as f(p) = 0, which is solved in explicit method is used to extend the fracture tips. The algorithm

X-axis
–3200 –2900 –2400 –2000 –1000 –1200 –800 –400 0 400

1400 1400

1200 1200

1000 1000
NF 2
800 800
NF 1
600 600
H perfs
400 400
NF 3 h
200 200

0 Treatment Well 0

–2900 –2400 –2000 –1000 –1200 –800 –400 0 400

X-axis

Fig. 7—Top view of wellbore, perforation, and natural fractures (NF) (x- and y-coordinate in feet).

TABLE 2—INPUT PARAMETERS FOR EXAMPLE NO. 1

Zone Top Min Horiz Young’s Poisson’s
Depth (ft) Thickness (ft) Stress (psi) Modulus (psi) Ratio Permeability (md)

6700 200 5338 4.0e6 0.24 0.0001

6900 300 4538 4.0e6 0.24 0.0001
7300 200 5537 4.0e6 0.24 0.0001
Injection rate 50 bpm
Fluid viscosity 5 cp
Rock tensile strength 500 psi

November 2011 SPE Production & Operations 373

 X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0

–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600

X-axis

(a) t=1 minute

X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0

–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600

X-axis

(b) t=15 minutes 19 seconds

Figs. 8a and b—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).

first finds all the hydraulic-fracture tips that satisfy the propagation
criterion (stress-intensity factor is greater than rock toughness). For
each propagating tip element, the flow rate into the tip is calculated
on the basis of the Poiseuille equation (Eq. 1) to find the corre-
sponding tip velocity (equal to fluid velocity near the tip):
vtip = qtip / (h fl w). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (21)
The tip with the highest velocity is extended by a prescribed
maximum length lfix. The other tips are extended proportionally to
their velocities:
ltip = lfix vtip / vtip
max
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22)
A special tip element is implemented, and the pressure at the
propagating tip element is prescribed by using asymptotic PKN tip
solution as a function of the element’s length and velocity.
Fig. 5 shows the general numerical scheme at each timestep.
Model Validation and Examples
Comparison With Conventional Model for a Biwing Fracture.
To check the UFM validity, it is compared with the limiting case
of biwing fractures. Fig. 6 shows the comparison of UFM results
shown as a red curve with a conventional hydraulic-fracture model
shown as a blue curve (Gulrajani et al. 1997), for the input param-
eters as given in Table 1. The predicted injection pressure and
fracture length from the two models are identical.

374 November 2011 SPE Production & Operations

 X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0

–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600

X-axis

(c) t=18 minutes 6 seconds

X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0

–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600

X-axis

(d) t=24 minutes 59 seconds

Figs. 8c and d—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).

Example No. 1: Simple Natural Fractures. A major advantage
of the UFM is to be able to predict the impact of natural fractures
and in-situ stresses on the created hydraulic-fracture network. The
factors that influence fracture complexity include natural-frac-
ture size, orientation, and distribution; natural-fracture mechanical
properties (interfacial friction coefficient, cohesion, and fracture
toughness); stress anisotropy (defined as difference between maxi-
mum and minimum horizontal stresses); and rock tensile strength.
As shown in Fig. 4, whether a hydraulic fracture crosses a natural
fracture when they intersect depends largely on the interfacial
friction coefficient and stress anisotropy. Therefore, these two
parameters play a critical role in fracture complexity.
To illustrate the importance of the natural friction coefficient
and stress anisotropy, a simple example involving a rock containing
only a few natural fractures is given for different values of these
parameters and different intersection angles. Fig. 7 shows the top
view of the wellbore and a single perforation cluster where the
fracture is initiated and the prescribed natural fractures. The input
parameters for the example are given in Table 2.
Case 1: Friction Coefficient = 0.5, Stress Anisotropy = 200
psi, Intersection Angle = 90°. In this example case, the property of
NF 3 is chosen so that a hydraulic fracture cannot propagate along
NF 3 easily (a large toughness is used). This biases the fracture
propagation to the right side of the wellbore, so the fracture geom-
etry does not become very complex. Fig. 8 shows the simulation
fracture geometry at different times, containing top views of both
the hydraulic fracture and natural fractures and 3D views with
color contour of width for the hydraulic fracture only.

November 2011 SPE Production & Operations 375

 X-axis
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800
X-axis
Fig. 8e—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).
Even though this is a simple case, it illustrates some impor-
tant behaviors of complex fracture. As shown in Fig. 8a, a single
fracture is initially created at the perforation cluster. Because
of the low fluid viscosity, the fracture width is relatively small,
approximately 0.05 in. When the fracture tips reach NF 1 and NF
3, the friction coefficient and stress anisotropy are such that the
hydraulic fracture is not able to cross the natural fractures. In the
ensuing injection, the fracture length is not growing, but its width
balloons up to 0.4 in. (Fig. 8b), while the net pressure increases
from 40 to approximately 230 psi.
When the pressure is sufficiently large to exceed the maximum
horizontal stress acting on the orthogonal NF 1, NF 1 is hydrauli-
cally opened. The fracture propagates quickly along NF 1 and
reaches the ends of the fracture, where the fracture branches off and
propagates in the preferred fracture direction again (Fig. 8c). The
orthogonal fracture NF 1 acts as a choke for the fluid in the main
fracture connected to the perforation and maintains a large width
1000
1 2
Net Pressure Fracture 1 (psi)
100
10
0.1 1 10
Custom Pumping Time (min)
Fig. 9—Net pressure for Example No. 1, Case 1.
376
Width
–600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0
–600
(e) t=68 minutes 15 seconds
and a net pressure of approximately 230 psi, while the width in NF
1 remains small because of a larger confining stress. The fractures
that branch off the ends of NF 1 also have a small width because
of low fluid viscosity and net pressure. The fracture branches that
are directly opposing the main fracture have a smaller width com-
pared with their opposite wings as a result of fracture-interference
(stress-shadow) effect (Fig. 8d). These fractures then intersect NF 2
and NF 3 and are stopped by these natural fractures. As before, the
pressure in these fracture branches starts to build up, and the width
balloons until the net pressure inside these fractures exceeds the
maximum horizontal stress to allow NF 2 to open up (Fig. 8e).
This example shows that when a hydraulic-fracture branch
intersects a natural fracture and cannot cross, the propagating
branch is stopped and some other branches can continue to grow.
When all propagating tips are stopped by natural fractures, the net
pressure inside the fracture network will have to rise to exceed the
stress anisotropy. That is, the net pressure in the fracture system
is regulated by stress anisotropy. This can lead to much greater
fracture width than predicted by the conventional biwing-fracture
model, or even the wiremesh model, in which straight fracture
planes is assumed. The net pressure response typically shows
a rapid rise in early time and then flattens after it exceeds the
maximum horizontal stress (Fig. 9). The high net pressure resulting
from width restriction and large pressure drop through the natural
fractures has also been observed by Jeffrey et al. (2009), though
the net pressure may not have to exceed the anisotropy to open up
the natural fracture in their 2D-plane-strain model.
Case 2: Friction Coefficient = 0.5, Stress Anisotropy = 500
psi, Intersection Angle = 90°. In this case, the stress anisotropy
is increased to 500 psi. At this condition, the crossing criterion
predicts that the hydraulic fracture will cross a natural fracture
when they intersect. Consequently, the UFM predicts a biwing
fracture as shown in Fig. 10. Similar geometry is obtained if the
stress anisotropy is kept at 200 psi, while the friction coefficient is
increased to greater than 0.7, which also corresponds to the condi-
tion of a hydraulic fracture crossing natural fractures.
Case 3: Friction Coefficient = 0.5, Stress Anisotropy = 500 psi,
Intersection Angle = 60°. In this case, the stress anisotropy is 500
100 psi, and the hydraulic fracture intersects natural fracture at an angle
of 60°. Fig. 11 shows the fracture geometry at t = 75 minutes, and
Fig. 12 shows the predicted net pressure. The figures show that with
nonorthogonal intersection, the natural fractures stop the hydraulic
November 2011 SPE Production & Operations
 X-axis
–2400 –2000 –1600 –1200 –800 –400

0.12
0.11
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0

–2400 –2000 –1600 –1200 –800 –400

X-axis

Fig. 10—Fracture geometry for Example No. 1, Case 2 (large anisotropy, t = 9 minutes, x- and y-coordinate in feet).

fractures from crossing, resulting in complex fracture rather than
biwing fracture as in Case 2 for an orthogonal intersection. Because
of the greater normal stress acting on the natural fractures (375 psi),
the fluid net pressure has to rise above 380 psi before the fluid pres-
sure can open up NF 1. As a result, the fracture width in the main
fracture is approximately 0.6 in., much wider than in the previous
cases because of the greater net pressure.
Example No. 2: More-Complex Natural Fractures. In this exam-
ple, more-complex natural fractures are generated on the basis of
statistical parameters given in Table 3. Other input parameters are
the same as those given in Table 2. Stress anisotropy is assumed to
be 200 psi. The pumping schedule is given in Table 4.
Fig. 13 shows the top view of the statistically generated natural
fractures and the predicted hydraulic-fracture network. Fig. 14
shows the fracture width contour and area proppant concentration
in 3D view. It shows that while a large hydraulic-fracture network
is created, proppant occupies only a small fraction of the fracture
surface area created. This is mainly because of fast proppant
settling as a result of the low-viscosity fluid used, but it is also

X-axis Width
–2000 –1600 –1200 –800
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35 –2400
0.3 –2000
0.25 X-axis
0.2 –1600
0.15 –1200
0.1
0.05 –800
0

–2000 –1600 –1200 –800

X-axis

Fig. 11—Fracture geometry for Example No. 1, Case 3 (large anisotropy, 60° angle, x- and y-coordinate in feet).

November 2011 SPE Production & Operations 377

 Stage 1_Pumping schedule_1_Design1 More-detailed discussion of the impact of various formation
Net pressure (Fracture 1) parameters on the resulting fracture geometry and a field case
1000 study using the UFM are beyond the scope of this paper. Readers
1 2 are referred to the paper by Cipolla et al. (2011).

Conclusions
The new complex-hydraulic-fracture model presented in this paper
Net Pressure Fracture 1 (psi)

is capable of simulating propagation of hydraulic fractures in a

formation consisting of pre-existing natural fractures. Simulation
results from the model show that stress anisotropy, natural fractures,
100 and interfacial friction play critical roles in creating fracture net-
work complexity. Decreasing stress anisotropy or interfacial friction
can change the induced-fracture geometry from a biwing fracture
to a complex-fracture network for the same initial natural fractures.
The results illustrate the importance of rock fabrics and stresses on
fracture complexity in unconventional reservoirs. A complex-frac-
ture model provides a critical tool for engineers working on shale-
gas stimulation to be able to simulate the fracture geometry, prop-
10
pant distribution, and fracture conductivity; assess the influence
0.1 1 10 100 of natural fractures and various parameters on fracture geometry;
Custom Pumping Time (min) match the fracture geometry against the microseismic observations;
and optimize job design to maximize the production.
Fig. 12—Net pressure for Example No. 1, Case 3 (large anisot-
ropy, 60° angle).
Nomenclature
partially because of proppant bridging in the side fractures that Cnsij , Cnn
ij
, Cssij , Csnij = plane-strain elastic influence coefficients in 2D
have very small width. DD method
E' = plane-strain elastic modulus of the rock
H = fracture height
General Applications
Hfl = fluid height
As illustrated in the preceding simple examples, the pre-existing K' = fluid consistency index
natural fractures and the stress state in a formation (both minimum
KIc = rock toughness
and maximum horizontal stresses) have a major impact on the
hydraulic-fracture network created during a fracture treatment. L = fracture length
To predict the fracturing behavior, proper characterization of the n' = fluid power-law index
natural fractures is very important. However, detailed understand- P = absolute fluid pressure
ing of the natural fractures is often unavailable or incomplete. In pnet = net fluid pressure inside the fracture,
these circumstances, the UFM can be used to do “what-if” studies pnet (s, t) = p – σn
to assess the possible stimulation outcome for a range of uncer- q = flow rate of fluid
tain reservoir parameters. The predicted fracture geometry can be qL = leakoff rate per unit height through the fracture
compared to the effective stimulated volume as determined from surface into the rock
microseismic measurement, and these measurements can be used Q = pumping rate
to help constrain the uncertain parameters, leading to a more-pre- t = time
dictive design model. Furthermore, the predicted fracture geometry
T0 = tensile strength of rock
needs to be integrated with a production simulator to assess the
potential production from the fracture stimulation, which allows Pf = unconventional-fracture-model fluid density
the production engineer to match the production performance σn = horizontal stress acting perpendicular to fracture
further and to evaluate design changes to optimize the production. element
All these techniques require close integration of a fracture model σs = shear stress acting on fracture element
such as the UFM with reservoir characterization, geomechanical w = average fracture opening at vertical cross section
modeling, microseismic interpretation, and production-simulation
tools. The UFM has been integrated into a software platform that Acknowledgment
enables these integrations. The general workflow can be found in The authors wish to thank Schlumberger for permission to publish
Cipolla et al. (2010). this paper.

TABLE 3—NATURAL-FRACTURE PROPERTIES FOR EXAMPLE NO. 2

Length (ft) Orientation Spacing (ft) Friction Coeff. Cohesion (psi)

Average 200 120° 100 0.5 0

Std. deviation 50 10° 50 0 0

TABLE 4—PUMP SCHEDULE

Stage Name Pump Rate (bpm) Fluid Fluid Vol (gal) Proppant Prop Conc. (ppga) Proppant Mass (lbs)

Pad 50 Slickwater 100,000 40/70 mesh sand 0 0

0.5 ppga 50 Slickwater 100,000 40/70 mesh sand 0.5 50,000
0.75 ppga 50 Slickwater 100,000 40/70 mesh sand 0.75 75,000
1 ppga 50 Slickwater 100,000 40/70 mesh sand 1.0 100,000

378 November 2011 SPE Production & Operations

 X-axis
–3200 –2800 –2400 –2000 –1600 –1200 –800 –400 0 400

2000

1800

1600

1400

1200

1000

800

600

400

200

–200

–400

–600

–800

–3200 –2800 –2400 –2000 –1600 –1200 –800 –400 0 400

X-axis

Fig. 13—Natural fractures and created-hydraulic-fracture geometry for Example No. 2 (x- and y-coordinate in feet).

References
Adachi, J., Seibrits, E., Peirce, A., and Desroches, J. 2007. Computer simulation
of hydraulic fractures. Int. J. Rock Mech. Min. Sci. & Geomech. Abstracts
44 (5): 739–757. http://dx.doi.org/10.1016/j.ijrmms.2006.11.006.
Blanton, T.L. 1982. An Experimental Study of Interaction Between Hydrau-
lically Induced and Pre-Existing Fractures. Paper SPE 10847 presented
at the SPE Unconventional Gas Recovery Symposium, Pittsburgh,
Pennsylvania, USA, 16–18 May. http://dx.doi.org/10.2118/10847-MS.
Cipolla, C.L., Weng, X., Mack, M.G., et al. 2011. Integrating Microseismic
Mapping and Complex Fracture Modeling to Characterize Hydraulic
Fracture Complexity. Paper SPE 140185 presented at the SPE Hydrau-
lic Fracturing Technology Conference, The Woodlands, Texas, USA,
24–26 January. http://dx.doi.org/10.2118/140185-MS.
Cipolla, C.L., Williams, M.J., Weng, X., Mack, M.G., and Maxwell,
S.C. 2010. Hydraulic Fracture Monitoring to Reservoir Simulation:
Maximizing Value. Paper SPE 133877 presented at the SPE Annual

Width Proppant–Areal_Distribution
0.5 5.5
0.45 5
0.4 4.5
0.35 400 ft 4
3.5
0.3
0.25 3
0.2 2.5
0.15 2
1.5
0.1 1
0.05 0.5
0 0

Fig. 14—Fracture width and proppant area concentration for Example No. 2.

November 2011 SPE Production & Operations 379

 Technical Conference and Exhibition, Florence, Italy, 19–22 Septem-
ber. http://dx.doi.org/10.2118/133877-MS.
Crouch, S.L. and Starfield, A.M. 1983. Boundary Element Methods in
Solid Mechanics: With Applications in Rock Mechanics and Geological
Engineering. London: George Allen & Unwin.
Daneshy, A.A. 1978. Numerical Solution of Sand Transport in Hydraulic
Fracturing. J Pet Technol 30 (1): 132–140. SPE-5636-PA. http://dx.doi.
org/10.2118/5636-PA.
Daniels, J., Waters, G., Le Calvez, J., Lassek, J., and Bentley, D. 2007.
Contacting More of the Barnett Shale Through an Integration of Real-
Time Microseismic Monitoring, Petrophysics, and Hydraulic Fracture
Design. Paper SPE 110562 presented at the SPE Annual Technical Con-
ference and Exhibition, Anaheim, California, USA, 11–14 November.
http://dx.doi.org/10.2118/110562-MS.
Fisher, M.K., Wright, C.A., Davidson, B.M., et al. 2002. Integrating
Fracture Mapping Technologies to Optimize Stimulations in the Bar-
nett Shale. Paper SPE 77441 presented at the SPE Annual Technical
Conference and Exhibition, San Antonio, Texas, USA, 29 September–
2 October. http://dx.doi.org/10.2118/77441-MS.
Gu, H. and Weng, X. 2010. Criterion For Fractures Crossing Frictional
Interfaces At Non-orthogonal Angles. Paper ARMA 10-198 presented
at the 44th U.S. Rock Mechanics Symposium and 5th U.S.-Canada
Rock Mechanics Symposium, Salt Lake City, Utah, USA, 27–30
June.
Gu, H., Weng, X., Lund, J.B., Mack, M.G., Ganguly, U., and Suarez-
Rivera, R. 2011. Hydraulic Fracture Crossing Natural Fracture at Non-
Orthogonal Angles, A Criterion, Its Validation and Applications. Paper
SPE 139984 presented at the SPE Hydraulic Fracturing Technology
Conference, The Woodlands, Texas, USA, 24–26 January. http://dx.doi.
org/10.2118/139984-MS.
Gulrajani, S.N., Nolte, K.G., and Romero, J. 1997. Evaluation of the M-Site
B-Sand Fracture Experiments: The Evolution of a Pressure Analysis
Methodology. Paper SPE 38575 presented at the SPE Annual Technical
Conference and Exhibition, San Antonio, Texas, USA, 5–8 October.
http://dx.doi.org/10.2118/38575-MS.
Jeffrey, R.G., Zhang, X., and Thiercelin, M. 2009. Hydraulic Fracture
Offsetting in Naturally Fractured Reservoirs: Quantifying a Long-
Recognized Process. Paper SPE 119351 presented at the SPE Hydrau-
lic Fracturing Technology Conference, The Woodlands, Texas, USA,
19–21 January. http://dx.doi.org/10.2118/119351-MS.
Le Calvez, J.H., Klem, R.C., Bennett, L., Erwemi, A., Craven, M., and Pala-
cio, J.C. 2007. Real-Time Monitoring of Hydraulic Fracture Treatment:
A Tool to Improve Completion and Reservoir Management. Paper
SPE 106159 presented at the SPE Hydraulic Fracturing Technology
Conference, College Station, Texas, USA, 29–31 January. http://dx.doi.
org/10.2118/106159-MS.
LeVeque, R.J. 1992. Numerical Methods for Conservation Laws, second
edition, 165. Basel, Switzerland: Lectures in Mathematics, Birkhaüser
Verlag.
Mack, M.G. and Warpinski, N.R. 2000. Mechanics of Hydraulic Fracturing.
In Reservoir Stimulation, third edition, ed. M.J. Economides and K.G.
Nolte, Chap. 6. West Sussex, UK: John Wiley & Sons.
Maxwell, S.C., Urbancik, T.I., Steinsberger, N.P., and Zinno, R. 2002.
Microseismic Imaging of Hydraulic Fracture Complexity in the Bar-
nett Shale. Paper SPE 77440 presented at the SPE Annual Technology
Conference and Exhibition, San Antonio, Texas, USA, 29 September–
2 October. http://dx.doi.org/10.2118/77440-MS.
Meyer, B.R. and Bazan, L.W. 2011. A Discrete Fracture Network Model
for Hydraulically Induced Fractures - Theory, Parametric and Case
Studies. Paper SPE 140514 presented at the SPE Hydraulic Fracturing
Technology Conference, The Woodlands, Texas, USA, 24–26 January.
http://dx.doi.org/10.2118/140514-MS.
Nolte, K.G. 1991. Fracturing-Pressure Analysis for Nonideal Behav-
ior. J Pet Technol 43 (2): 210–218. SPE-20704-PA. http://dx.doi.org/
10.2118/20704-PA.
Olson, J.E. 2004. Predicting fracture swarms—the influence of subcritical
crack growth and the crack-tip process zone on joint spacing in rock. In
The Initiation, Propagation, and Arrest of Joints and Other Fractures,
ed. J.W. Cosgrove and T. Engelder, No. 231, 73–87. Bath, UK: Special
Publication, Geological Society Publishing House.
Olson, J.E. 2008. Multi-fracture Propagation Modeling: Application to
hydraulic fracturing in shales and tight gas sands. Paper ARMA 08-327
380
presented at the 42nd US Rock Mechanics Symposium, San Francisco,
29 June–2 July.
Renshaw, C.E. and Pollard, D.D. 1995. An experimentally verified criterion
for propagation across unbounded frictional interfaces in brittle, linear
elastic materials. Int. J. Rock Mech. Min. Sci. & Geomech. Abstracts
32 (3): 237–249. http://dx.doi.org/10.1016/0148-9062(94)00037-4.
Schiller, L. and Naumann, A. 1933. A drag coefficient correlation. VDI
Zeitschrift 77: 318–320.
Thiercelin, M. and Makkhyu, E. 2007. Stress field in the vicinity of a natu-
ral fault activated by the propagation of an induced hydraulic fracture.
Proc., 1st Canada-US Rock Mechanics Symposium (Rock Mechanics:
Meeting Society’s Challenges and Demands), Vancouver, Canada,
27–31 May, 1617–1624.
Wang, J., Joseph, D.D., Patankar, N.A., Conway, M., and Barree, R.D.
2003. Bi-power law correlations for sediment transport in pressure
driven channel flows. Int. J. Multiphase Flow 29 (3): 475–494. http://
dx.doi.org/10.1016/s0301-9322(02)00152-0.
Warpinski, N.R. and Teufel, L.W. 1987. Influence of Geologic Discontinui-
ties on Hydraulic Fracture Propagation. J Pet Technol 39 (2): 209–220.
SPE-13224-PA. http://dx.doi.org/10.2118/13224-PA.
Xu, W., Calvez, J.H.L., and Thiercelin, M.J. 2009. Characterization of
Hydraulically-Induced Fracture Network Using Treatment and Micro-
seismic Data in a Tight-Gas Sand Formation: A Geomechanical
Approach. Paper SPE 125237 presented at the SPE Tight Gas Comple-
tions Conference, San Antonio, Texas, USA, 15–17 June. http://dx.doi.
org/10.2118/125237-MS.
Xu, W., Thiercelin, M.J., Ganguly, U., et al. 2010. Wiremesh: A Novel
Shale Fracturing Simulator. Paper SPE 132218 presented at the Inter-
national Oil and Gas Conference and Exhibition in China, Beijing, 8–10
June. http://dx.doi.org/10.2118/132218-MS.
Zhang, X. and Jeffrey, R.G. 2006. The role of friction and secondary flaws
on deflection and reinitiation of hydraulic fractures at orthogonal pre-
existing fractures. Geophys. J. Int. 166 (3): 1454–1465. http://dx.doi.
org/10.1111/j.1365-246X.2006.03062.x.
Zhang, X., Jeffrey, R.G., and Thiercelin, M. 2007. Effects of Frictional
Geological Discontinuities on Hydraulic Fracture Propagation. Paper
SPE 106111 presented at the SPE Hydraulic Fracturing Technology
Conference, College Station, Texas, USA, 29–31 January. http://dx.doi.
org/10.2118/106111-MS.
Xiaowei Weng is a principal engineer and project manager
of modeling and mechanics with Schlumberger in Sugar
Land, Texas, USA. His research interests include hydraulic-frac-
ture modeling, acid fracturing, multifractured horizontal well
completion and production, wellbore hydraulics, and coiled-
tubing cleanout. He received holds MS and PhD degrees in
engineering mechanics from The University of Texas at Austin.
He is a member of SPE. Olga Kresse is a senior engineer with
Schlumberger in Sugar Land, Texas, USA. Her primary research
interests are fracture and fluid mechanics, acoustics, and
applied math. Her most recent work is concentrated on
numerical modeling of hydraulic fracturing in unconventional
reservoirs. She holds two PhD degrees – one in solid mechanics
from T. Shevchenko National University and the other in fracture
mechanics from the University of Minnesota. She is a member
of SPE, ASME, ARMA, AGU, and ASA. Charles-Edouard Cohen is
a modeling engineer with Schlumberger in Sugar Land, Texas,
USA. His research interests include modeling and simulation,
fluid mechanics, matrix acidizing in carbonate reservoir, frac-
turing, and production. His most recent works are on numeri-
cal modeling of hydraulic fracturing in unconventional reser-
voirs. He holds a PhD degree in reservoir engineering from the
French Institute for Petroleum and the National Polytechnique
Institute of Toulouse, an MSc degree in hydraulics and fluid
mechanics, and an MSc degree in fluid dynamics from the
ENSEEIHT in Toulouse, France. He is a member of SPE and ARMA.
Ruiting Wu is a geomechanics engineer with Schlumberger.
Her research interests include fracture mechanics and sand
management. She holds a PhD degree from the Georgia
Institute of Technology. She is a member of SPE. Hongren Gu
is a principal engineer with Schlumberger in Sugar Land, Texas,
USA. His research interests include reservoir stimulation, fracture
mechanics, and wellbore hydraulics. He holds an MS degree
from Xian Jiaotong University in China and a PhD degree from
University of Texas at Austin, both in engineering mechanics.
November 2011 SPE Production & Operations