Author’s Accepted Manuscript

Compositional Modeling of Fracture-to-Fracture

Miscible Gas Injection in an Oil-Rich Shale

Peixi Zhu, Matthew T. Balhoff, Kishore K.

Mohanty

www.elsevier.com/locate/petrol

PII: S0920-4105(17)30155-9
DOI: http://dx.doi.org/10.1016/j.petrol.2017.01.031
Reference: PETROL3840
To appear in: Journal of Petroleum Science and Engineering
Received date: 15 July 2016
Revised date: 18 November 2016
Accepted date: 17 January 2017
Cite this article as: Peixi Zhu, Matthew T. Balhoff and Kishore K. Mohanty,
Compositional Modeling of Fracture-to-Fracture Miscible Gas Injection in an
Oil-Rich Shale, Journal of Petroleum Science and Engineering,
http://dx.doi.org/10.1016/j.petrol.2017.01.031
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
Compositional Modeling of Fracture-to-Fracture Miscible Gas Injection in
an Oil-Rich Shale
a a a
Peixi Zhu *, Matthew T. Balhoff , Kishore K. Mohanty
a
Petroleum and Geosystems Engineering Department, The University of Texas at Austin, 1 University Station C 0300, Austin, TX, 78712, USA

* Corresponding author zpeixi@utexas.edu

Abstract
The ultra-low permeability of shale makes injection from a well to other wells difficult. A novel scheme
was proposed in a recent patent (Dombrowski et al., 2015) where gas is injected into a hydraulic fracture
along a horizontal well and production occurs from an adjacent fracture, intersecting the same well.
Compositional reservoir modeling was performed to investigate the effectiveness of the proposed gas
injection scheme. The computational domain consists of two hydrofrac half-stages along a horizontal
well. The results show 15.7% and 12.5% OOIP incremental recovery over 5000 days of CO 2 injection
for the base models with the matrix permeability of 10 D and 1 D, respectively, demonstrating that
the gas injection scheme has the potential to vastly improve oil recovery in oil-rich shale formations.
The effects of reservoir properties and injection conditions on oil recovery were investigated by
changing the injection pressure, reservoir heterogeneity, distribution of natural fractures, hydrofrac
spacing, size of pore space, and compositions of the injection gas. Most of them affect the oil recovery
significantly. Recovery by miscible hydrocarbon gas injection is comparable to CO2; so it should be
considered as an alternative.

Keyword
unconventional resources, compositional simulations, hydraulic fractures, natural fractures, shale oil,
miscible gas injection, phase behavior

1. Introduction
Advancement of horizontal drilling and multistage hydraulic fracturing has made oil recovery from
shale formations technically and economically viable in the last decade. As a result, shale oil is
considered as an alternative to conventional crude oil resources, which is likely to peak in the next 2-3
decades (Brandt et al., 2013). Estimated global shale resources is around 340 billion bbl of technically
recoverable oil, which represents 10% of all types of oil reserves (EIA, 2014). In the U.S., the estimated
reserve is nearly 60 billion bbl, and the production of shale oil makes up approximately 30% of total
domestic crude production (EIA, 2014).

However, the recovery factor in shale oil remains very low. A typical example is production in the
Bakken three-member formation that underlies Williston Basin and overlaps with parts of North Dakota,
Montana, Manitoba and Saskatchewan in the subsurface. The formation has a porosity of about 5% and
permeability of 0.0001-0.04 mD (Pitman et al., 2001). The estimated oil resource is 300 billion barrels
(Flannery and Kraus, 2006), but the technically recoverable oil, as reported by the U.S. Geological
Survey (Gaswirth et al., 2013), is only about 7 billion barrels. It includes the Bakken Formation in the
U.S. region and the Three Forks Formation underneath. Daily production for a new well can reach close
to 1,000 barrels, but decline in half in the first year (North Dakota Department of Mineral Resoursecs,
2012), which is a typical phenomenon of shale reservoirs due to their low permeability.

The vast amount of unrecoverable shale oil suggests a large potential for improved recovery methods.
Even a small percentage of improvement in the field is equivalent to billions of barrels of oil.
1
Traditional fluid injection such as water, efficient in most conventional reservoirs, is not viable in shale
formations due to the low permeability of shale reservoirs. Injection of miscible gas, such as CO2 or
hydrocarbon gas, however, is a promising approach for improved oil recovery in unconventional
reservoirs such as the Bakken.

Above the minimum miscibility pressure (MMP), CO2 (or a hydrocarbon gas) becomes miscible with
the oil phase and significantly reduces the capillary forces. Solubilization of gases into oil reduces oil
viscosity and swells the oil (Hawthorne et al., 2013). Although CO2 injection as an improved recovery
method has been successful in conventional reservoirs, its application in unconventional reservoirs
where permeability is low has not been widely tested at the field-scale. If successful, it may result in a
tremendous amount of oil recovery (and additionally subsurface storage of CO2). Hydrocarbon gas
injection can also be an alternative when CO2 is not readily available or there is no infrastructure to
export the associated hydrocarbon gas.

The extraction of oil by CO2 from oil shales was demonstrated by Hawthorne et al. (2013). Bakken rock
samples were exposed to CO2 flow at the reservoir conditions (5000 psi, 230 °F) in a laboratory
extractor, which is well above supercritical conditions and MMP. CO2 successfully extracted 90% of oil
from the middle Bakken rods (9mm x 9mm x 30mm) within 4 hours, compared to 2 hours from a
conventional reservoir rock. Only 20% of the oil was extracted within 4 hours from the lower
permeability lower Bakken square rods. Although the contact surface area between CO2 and rock was
unrealistic (for field), and translating their results directly to field injection is difficult, the results
demonstrate the potential of CO2 enhanced oil recovery in low permeability formations. They also
pointed out the importance of diffusion for CO2 recovery, especially in tight formations.

Limited number of core-flood experiments have been conducted to evaluate CO2 or gas injection in tight
formations. Kovscek et al. (2008) and Vega et al. (2010) tested the performance of both immiscible CO2
and miscible CO2 in a fractured shale core with permeability of 0.02-1.3 mD. Two different injection
modes, countercurrent and cocurrent flow, were investigated. Their results show that miscible injection
significantly outperforms immiscible or near miscible injection. Virtually all the oil was recovered
when countercurrent mode was followed by cocurrent mode. They concluded that both diffusion and
convective-dispersion are important to the recovery mechanism. Gamadi et al. (2013) injected N2 in
different shale core samples with a cyclic or Huff-n-Puff injection mode. Their results showed 20-70%
improved recovery above miscible pressure, depending on the core type and number of cycles. Song
and Yang (2013) compared the performance of cyclic CO2 injection versus water injection and showed
that miscible CO2 is more advantageous over immiscible or water injection.

Reservoir simulation studies have been performed to explore gas injection on fractured shale reservoirs.
Unlike in conventional reservoirs, where the presence of a fracture network is often undesirable, both
hydraulic and natural fractures serve as a flow path that delivers the injected gas into the formation.
Therefore, hydraulic fractures must be taken into account when modeling gas injection in shales. Sheng
and Chen (2014) simulated gas injection in a fractured 0.1 D reservoir. While primary depletion
resulted in nearly 6% original oil in place (OOIP) recovery, they obtained 15% of total recovery from
continuous well-to-well flooding over 70 years and 14% from cyclic injection. Wan et al. (2013)
demonstrated the positive effect of fracture spacing on recovery through a fractured-network model with
cyclic gas injection. They concluded that for gas injection, more effort should be put on maximizing the
fracture network (ideally by utilization of the natural fractures) than increasing the fracture conductivity.
In a similar model, Wan et al. (2014) later showed the significance of gas diffusion on oil recovery.

2
Chen et al. (2014) used compositional modeling to investigate the effect of heterogeneity on cyclic CO2
injection. They found that introducing heterogeneity into the reservoir resulted in less production.
Sanchez-Rivera et al. (2015) optimized the injection conditions for CO2 Huff-n-Puff in the Bakken
formation. They pointed out the positive role of natural fractures that improves CO2 transport and oil
recovery. Yu et al. (2014) compared oil recovery for different completion schemes along a single
horizontal well for cyclic CO2 injection. Both Yu et al. (2014) and Sanchez-Rivera et al. (2015)
demonstrated the importance of injection rate (or injection pressure), injection time and number of
cycles on recovery over other parameters such as soaking time. In addition, Yu et al. (2014) observed
higher recovery when diffusion effect considered. Shoaib and Hoffman (2009) performed CO2 flooding
simulation at the field scale for 0.01-1 md type formations and compared different horizontal well
patterns on recovery. In a follow-up study, Hoffman (2012) showed that injection of hydrocarbon gas
and CO2 are similarly effective. Therefore, in the case when CO2 supply is scarce, hydrocarbon gas or a
mixture of both can be considered as an alternative (Wang et al., 2010; Sanchez-Rivera et al., 2015).

The proposed injection modes in most of the literature are well-to-well flooding and cyclic injection.
Dombrowski et al. (2015) have proposed a novel injection/production scheme, i.e., injecting miscible
gas in one hydraulic fracture along a horizontal well and producing from an adjacent fracture in the
same horizontal well (Dombrowski et al., 2015), as shown in Figure 1a or in an adjacent well
(Figure 1b). New completion schemes have to be developed where alternate fractures can be used to
inject the gas and produce the oil in the same well (Figure 1a). If the well spacing is small enough, the
fractures in two wells can be staggered in a zipper pattern (Figure 1b). The two schemes are
conceptually similar, but very unorthodox. Due to their similarity, lots of general conclusions can be
transferred from one to another. The goal of this work is to understand the first injection scheme. The
effects of key parameters (e.g., injection pressure, reservoir heterogeneity, natural fractures, fracture
spacing, gas composition and matrix pore size) on oil recovery in the fracture-to-fracture injection mode
are investigated by compositional modeling. Geomechanics and completion difficulty are not considered
in this work, although we recognize that they can be important due to the large pressure gradient
between adjacent fractures.

2. Modeling method
2.1. Fluid model
Initial fluid composition for each pseudo component is given in Table 1. Fluid properties at initial
reservoir pressure, 6840 psi, and temperature, 241 °F, were calculated using the cell-to-cell method from
CMG WINPRO and given in Table 2. MMP of all the injected gas were also calculated. The Peng-
Robinson-Penelous equation of state (Pedersen et al., 2004), which includes a volume-shift parameter
for accurate fluid density, was used for all computations.

In nano pores, such as the shale reservoirs studied here, the fluid critical properties are changed from
their bulk values (Teklu et al., 2014). In this work, we assume that the pore size is uniform throughout
the reservoir domain. For a single (or pseudo) component, the relative shifts in critical temperature and
pressure are given by
2

Tcb  Tcp  LJ  
T   0.9409  0.2415  LJ  (1)
c
Tcb rp  rp
 
2

Pcb  Pcp  LJ  
P   0.9409  0.2415  LJ  (2)
c
Pcb rp  rp
 
3
where rp is the pore radius, T and P are temperature and pressure, and the subscripts cb and cp represent
the bulk properties and those in confined pore space, respectively. LJ is the Lennard-Jones collision
diameter, which can be obtained from Bird et. al. (2007)
1
 T 3
 LJ  0.244  cb  (3)
 Pcb 
In exploring the effect of nano pore size on the fluid phase behavior (and therefore oil recovery), Tcp and
Pcp for individual components were calculated with given rp and were used in place of Tcb and Pcb in
compositional modeling.

In compositional modeling of miscible injection, the importance of diffusion on oil recovery in

unconventional reservoirs was shown by Wan et al. (2014). Diffusion was included in this work by
using the empirical correlation proposed by Sigmund (1976). The diffusion coefficient Dik of a
component i in phase k can be calculated by
1  yik
Dik  (4),
 yik Dij1
j i

where yik is the mole fraction of component i in phase k. Dij is the binary diffusion coefficient between
component i and j. It can be calculated by
k0 Dij0   k  
2 3
k  k 
Dij   0.99589  0.096016 0  0.22035  0   0.032874  0   (5),
k  k  k    k  

where k is the density of phase k and the superscript 0 represents the quantities at zero pressure limit.
Further details of the diffusion model can be found in Sigmund (1976). Adsorption/desorption of
kerogen is not considered because the mechanism is important at low pressure, while the proposed
scheme uses high injection pressure.

2.2. Reservoir description and computational domain

A schematic of the fractured reservoir model is given in Figure 2a. It consists of two horizontal wells
parallel to each other with multistage hydraulic fractures. The fracture half-length is 200 ft and the well
spacing is 2000 ft. Hydraulic fracture spacing, S, is 160 ft. The computational domain is represented
by the red shaded area in the figure and consists of two half-stages. Therefore, the hydraulic fractures
are located right at the domain boundaries. The dimension of the domain is 160 ft (length) × 1000 ft
(width) × 40 ft (height).

A single-porosity model was used for computation. The matrix was assigned a permeability, kmatrix, of
either 10 D or 1 D and a porosity of 8%. The two hydraulic fractures had a permeability of 10 D,
43% porosity and 0.0025 ft fracture half-width. As a result, the fracture conductivity is 10 D× 0.0025 ft
× 2 = 0.05 ftD. The small size of the fracture requires very fine grid blocks in its neighborhood for
domain discretization. Together with the strong permeability contrast between matrix and fractures,
they can lead to numerical instability and require long computation times. To avoid numerical
difficulties, equivalent fractures with effective permeability of 25 mD and effective fracture half-width
of 1 ft were used (Rubin, 2010; Sanchez-Rivera et al., 2015). Since the fracture conductivity remains
0.05 ftD, the flux in the fractures were not altered and the simulation retained its accuracy. An effective
porosity of 0.11% was also employed, so that the fracture pore volume remains unchanged.

4
Natural fractures, especially those intersecting with hydraulic fractures, render the delivery of injected
fluid (Sanchez-Rivera et al., 2015) and are incorporated in the model. The natural fractures are assumed
to have a permeability of 100 mD, porosity of 100% and fracture width of 0.003 ft. With the same
numerical treatment as was done to the hydraulic fractures, equivalent natural fractures were modelled
using discrete grid blocks with permeability 0.15 md, porosity 15% and fracture width 2 ft. Dual
porosity/permeability models were not used in this work.

The domain was discretized into 35 × 64 × 1 Cartesian grid blocks. Further discretization did not
improve the results but required longer computation time. Due to geometrical symmetry of the domain,
no-flow conditions were imposed at all boundaries. Figure 2b shows the upper half of the domain with
finer grid blocks that is represented by the red shaded area in Figure 2a. 2D discretization was chosen
because gravitational effects are assumed minimal. This is because any phase separation would tend not
to distribute vertically due to low permeability of the formation. The Corey-Brooks model was used to
create the oil-water and gas-liquid relative permeability curves as a function of water saturation and
liquid saturation, respectively. The corresponding parameters are directly taken from Sanchez-Rivera
(2014) and shown in Table 3. The shales are considered mixed-wet. The wettability is reflected in the
relative permeabilities used in the simulations. All simulations were performed using CMG GEM. The
reservoir and operational conditions of the two base cases are given in Table 4.

2.3. Generation of heterogeneous permeability field

To investigate the effect of heterogeneity on miscible gas injection, the reservoir was assumed to have
log-normally distributed permeabilities, given by Y = ln(k), where k is the permeability. The mean and
standard deviation of Y are given by  and , which relate to the mean and variance of k as
   
E  k   exp    2 / 2 , and Var  k   exp  2   1 exp  2   2  (6)
Variance of heterogeneity is usually represented by Dykstra-Parsons coefficient VDP, which can be
calculated as
VDP  1  exp( ) (7)
Stanford Geostatistical Modeling Software (SGeMS) was used to create various heterogeneous reservoir
domains. 0.25 ft × 0.25 ft was used as the grid block size. With given correlation length L which
indicates the spatial continuity of the permeability, SGeMS generates a log-normally distributed domain
that has zero mean and standard deviation 1. The distribution was then converted to mean of  and
standard deviation . Finally, the permeabilities were upscaled, subject to the discretization scheme of
the stimulated domain. When investigating heterogeneity effect in this work, L was fixed at 40 ft, and
VDP varied from 0 to 0.8.

2.4. Generation of the reservoir domain with natural fractures

The effect of random natural fractures on oil recovery was also considered. The natural fractures were
included in the stimulated region of the heterogeneous reservoir domain that has a correlation length of
L = 40 ft and VDP = 0.7. Fractures were modelled discretely in a single porosity model as described in
section 2.2. While the main area of the reservoir domain was discretized in the same way as described
before, the stimulated region of the reservoir was discretized into 17 × 21 coarse blocks. Each of these
357 blocks were further refined into 5 × 5 grids using local grid refinement. The discrete natural
fractures are assumed aligned with the Cartesian coordinate and placed randomly in 4 possible directions
in each of the coarse blocks. The coarse blocks that are located at the boundaries of the reservoir domain
were given special treatment so that placement of natural fractures makes geological sense. The
approach leads to N = 1287 possible locations where discrete natural fractures can be included. The

5
probability of natural fractures is defined as p = Nf/N, where Nf is the number of fractures. p was
manually varied from 0 to 0.68 to investigate the effect of natural fractures on recovery.

3. Results
3.1. Base cases
In the two base cases chosen (Figure 3), oil recovery was modeled for 500 days of primary depletion
followed by 5000 days of gas injection. No attempt was made to optimize the primary depletion
duration in this work. Extending the depletion days will definitely improve the ultimate recovery. At
day 500, the production rate dropped to less than 3% of the peak value, which is considered low enough
to start CO2 injection. Following primary depletion, CO2 was injected from the left fracture in Figure 2b
at a fixed injection pressure Pin = 7000 psi, while oil was produced from the right fracture at
Pout = 1000 psi. For 10 μD matrix, CO2 breaks through at approximately 1000 days (after 500 days of
CO2 injection) at an oil recovery of 24.3% OOIP. The oil rate decreases considerably beyond the
breakthrough reaching a cumulative recovery of 32.8% in 5500 days. Primary depletion produces 11.7%
at 1000 days and 17.1% at 5500 days. Thus the incremental recovery is 12.6% at 1000 days and 15.7%
at 5500 days. The primary recovery factors here are high compared to realistic cases because the chosen
permeability is at the high end of shale permeability. The high injection pressure injects the CO2 into
the reservoir and displaces oil. Since the injection pressure is much higher than the MMP, oil is
displaced miscibly by CO2.

As the matrix permeability decreases to 1 μD, the CO2 injection rate decreases (compared to the 10 μD
matrix). The recovery mechanisms are the same as those in the last case, but the rate is lower. CO2
breaks through at approximately 3600 days at an oil recovery of 17.5%. The oil rate decreases after
breakthrough, but not as much as in the last case. The cumulative oil recovery is 21.9% at 5500 days.
Primary depletion produces only 8.2% at 3600 days and 9.4% at 5500 days. Thus the incremental
recovery is 9.3% at 3600 days and 12.5% at 5500 days. This matrix permeability is close to those of
Bakken shale. Recovery numbers are presented in Table 5.

In the rest of this work, the effect of several operational conditions and reservoir properties on the oil
recovery was investigated by the proposed injection scheme. In addition to the parameters shown
below, the influence of mechanical dispersion was explored by changing the dispersitivity in Table 4.
However, the results are not presented since it was found that due to the slow fluid velocity in shale,
dispersion only has a small to negligible effect on the oil recovery. Simulations were performed with
and without the consideration of molecular diffusion. It has been shown that diffusion may have a large
contribution to oil recovery in conventional reservoirs (Yanze and Clemens, 2012). In our simulations,
however, no difference in ultimate recovery was found in the case of 1m reservoir, while only 2.9%
improvement in oil recovery for 10m reservoir was observed when molecular diffusion was included in
the model. Therefore, diffusion can have a small effect on the oil recovery.

3.2. Injection pressure Pin

The role of CO2 injection pressure was investigated by varying Pin from 3500 (near the MMP of 3750
psi) to 7000 psi at the injection fracture. Figure 4a shows the percentage of oil recovered as a function
of time for primary depletion and CO2 injection at different Pin when kmatrix = 10 μD. For all investigated
pressures, the oil recovery drops below that of primary depletion for a short period of time during early
injection. During depletion, both fractures are used for oil production. As soon as CO2 injection is
started, oil is produced from only one fracture. This startup leads to a low production in the short-term.

6
Nonetheless, the oil recovery by CO2 injection quickly surpasses the base depletion case. For all
injection pressures, the ultimate recovery substantially outperforms that of primary depletion. At
Pin = 3500 psi, injection of CO2 until day 5500 gives 26.9% OOIP of ultimate recovery, resulting in
9.9% of incremental recovery above primary depletion that recovers 17.0% OOIP. This is already a
rather substantial recovery improvement at a relatively low injection pressure. Raising the injection
pressure to 5000 psi recovers 30.4% oil. This increase is due to the increase in sweep as shown in
Figure 5. The increase of pressure drives more CO2 into the unfractured zone between the wells.

The slopes of the recovery curves in Figure 4a indicate a higher production rate during early injection,
especially at Pin = 7000 and 5000 psi. The improvement of production over primary depletion is
followed by an abrupt decline at about day 1000 for injection pressure 7000 psi, and day 1400 for
5000 psi, indicating the onset of CO2 breakthrough. At Pin = 7000 psi, estimated oil rate before
breakthrough is approximately 5 bbl/day. Since the computational domain consists of two half-stage,
single-wing and 10 ft tall fractures, extrapolating the result to a full horizontal well with 30 stages and
40 ft fractures results in an oil rate of 1200 bbl/day (multiplication factor is 240). Production declines
to 0.12 bbl/day towards the end of injection for the computational domain, equivalent to 28.8 bbl/day for
a full well, which is probably no longer economical. The average oil rate before the breakthrough was
estimated and plotted against the injection pressure (Figure 4b). The linear relationship indicates that
the main recovery mechanism is pressure maintenance.

Figure 5 shows the distribution of CO2 at the end (day 5500) of injection. Compared to the areas away
from the fractures, the stimulated region (area between the two fractures) is mostly saturated with CO2.
The oil located in the stimulated area is the easiest to be recovered, due to CO2 contact and high pressure
gradient. Further recovery requires migration of CO2 into the unfractured region which is larger at
higher injection pressures. At 5500 days for Pin = 3500 psi, there is still some oil near the production
fracture. Raising the injection BHP to 5000 psi helps CO2 sweep out all the oil in the fractured region.
A higher injection pressure also drives CO2 into the unfractured region and improves the recovery.
Further increase of the injection pressure to 7000 psi improves the sweep in the unfractured region
slightly (Figure 4a).

Recovery is shown in Figure 6 as a function of pore volumes of CO2 injected, Vpi. The early flat part of
the curve is due to pressurization after which oil is pushed by CO2 miscibly. The recovery at
Pin = 3500 psi injection is higher than the recovery for Pin = 7000 or 5000 psi injection because less CO2
is injected into the unfractured region in the first case. After the CO2 breakthrough, most of the oil
production is from the unfractured region. Therefore, recovery after breakthrough is higher for
Pin = 7000 and 5000 psi than Pin = 3500 psi.

The sensitivity to matrix permeability was investigated by developing a reservoir model with 1 μD. Oil
recovery versus time for the reservoir is given in Figure 7a. Due to the low permeability and
insufficient pressure gradient, the benefit of injecting CO2 at 3500 psi is very limited. Raising the
injection pressure to 5000 psi substantially increases ultimate recovery. Even at Pin = 5000 psi the
stimulated area is still only partially saturated with CO2. Since there is still a lot of unrecovered oil in
this region, raising the pressure to 7000 psi can substantially recover more oil. A near-linear correlation
between daily oil rate and Pin suggest that pressure gradient is still the controlling factor for the
production (Figure 7b).

7
3.3. Heterogeneity
Reservoir heterogeneity affects the displacement efficiency of gas injection and henceforth the ultimate
recovery. In this work, 1 D was chosen as the mean permeability, E(k). The effect of Dykstra-Parsons
coefficient VDP on recovery subjected to CO2 injection was investigated by fixing the correlation length
at 40 ft. It was assumed that correlation lengths are equal in both directions. Heterogeneity was only
imposed in the upper half of the computational domain where grid blocks are finer (Figure 2b). Beyond
this region the random fields cannot be accurately captured due to block size being larger than L.
Therefore, a homogeneous permeability of 1 D was implemented.

Figure 8 shows the permeability in the upper half of the computational domain for the Dykstra-Parsons
coefficient VDP = 0.5. Plots of recovery versus time for different VDP are provided in Figure 9a. The
mean permeability is 1 μD and L is 40 ft, which is a moderate value in geological formations. Since
reservoir heterogeneity leads to uneven flow, it is expected that production decline with the increase of
VDP in primary depletion (Chen et al., 2014). As Figure 9a shows, the effect is more pronounced for gas
injection. Ultimate incremental recovery versus VDP is also provided in Figure 9b. Error bars that were
estimated from the results of five realizations for each Dykstra-Parsons coefficient are also provided.
With the consideration of the uncertainties, incremental recovery is reduced by about a half when the
reservoir model is changed from homogeneous (VDP = 0) to very heterogeneous (VDP = 0.8). The effect
of correlation length was also investigated but no significant change in the oil recovery was found.

3.4. Natural fractures

The presence of natural fractures can facilitate the transport of injection fluid into the shale matrix and
improve oil recovery. Their effect was investigated by including the natural fractures in the reservoir
model that has an average permeability of 1D, VDP = 0.5 and L = 40 ft (Figure 10). Recovery results
are shown in Figure 11a. While the ultimate recovery due to primary depletion are similar for all the
cases, an increase in incremental recovery was observed as the natural fracture probability increases
from p ≈ 0 (base case) to p = 0.2. The improvement due to higher p is even more significant when p was
changed from 0.2 to 0.5 for a long period of injection time since more natural fractures can facilitate gas
injectivity. However, the production rate for p = 0.5 (pink curve in Figure 11a) starts to drop at around
day 2000, and its ultimate recovery at day 5500 is only close to the case of p = 0.2.

The production drops in the case of p = 0.5 can be explained by early breakthrough of the injection fluid
due to the large natural fracture density, even though there is no percolation path for the natural fractures
from one hydraulic fracture to another. As indicated by the slope of the curves, an even earlier
breakthrough was seen for the case of p = 0.68, a probability under which a percolation path exists. The
ultimate recovery falls below the case of p = 0.1. A plot of the oil recovery versus cumulative gas
injected for different natural fracture probability is also provided in Figure 11b. While the curves for p
= 0, 0.1 and 0.2 almost overlap with each other, those for p = 0.5 and 0.68 show a lower recovery at the
same amount of gas injected. The results suggest that a reservoir with higher natural fracture density is
beneficial for improved recovery at miscible gas injection, but too many natural fractures can cause a
loss of injected fluid and early breakthrough.

While the previous cases are mostly parametric by assuming that natural fractures distribute evenly
across the reservoir domain, realistic natural fractures tend to be more complicated than that. It is quite
likely that some natural fractures, particularly those near the wellbores, can be activated during
fracturing and get connected to the hydraulic fractures. That can lead to shattered formation near the
wellbores with lots of intersecting fractures and unaffected fractures away from the wellbores. A

8
configuration of such with p = 0.63 was created and shown in Figure 10e. The corresponding oil
recovery result is shown by the blue curve in Figure 11c. While the probability of natural fractures (p =
0.63) represented by the blue curve is less than that by the black curve (p = 0.68), the primary recovery
is actually higher. Furthermore, the ultimate recovery from both cases are nearly the same. The results
suggest that the recovery is not only affected by the number of natural fractures, but also by their
distribution, especially those natural fractures that intersecting with hydraulic fractures.

3.5. Hydraulic fracture spacing S

For a given Pin, S affects the pressure gradient from fracture to fracture. Computational domains with
different S were created to investigate the role of spacing on oil recovery.

Comparisons of oil recovery between S = 80 and 160 ft for both kmatrix = 10 μD and 1 μD are given in
Figure 12. Since the pressure gradient was doubled when S was reduced to 80 ft, a higher production
rate was achieved before the CO2 breakthrough. When the matrix permeability is 1 μD and a full
horizontal well is considered, the estimated production before the breakthrough increases from
136 bbl/day for S = 160 ft to 257 bbl/day for S = 80 ft and lasts for nearly one year, which is an 11%
increase in incremental recovery before CO2 breakthrough (Figure 12b). Oil rate at surface condition
versus. S-1 is shown in Figure 13. The proportional relationship indicates that the pressure gradient is
the main driving force for the recovery. When kmatrix = 10 μD, a fracture spacing of 80 ft increased the
production rate, but led to an early breakthrough at about day 600.

After CO2 breakthrough, the production declined sharply (for both kmatrix = 10 μD and 1 μD) and
dropped even below the case with S = 80 ft. In the former case (S = 80 ft), the decline led to a lower
ultimate recovery than for S = 160 ft. This is due to gas channeling which causes lower sweep
efficiency for shorter fracture spacing. Even though in the reservoir model with kmatrix = 1 μD, the
ultimate recovery for S = 80 ft is higher than 160 ft, the two curves are expected to crossover if the
simulation time is extended.

Oil recovery in reservoirs with different fracture spacing is compared at a constant injection rate subject
to a maximum injection pressure of 7000 psi. The injection rate increases proportionally with the
fracture spacing (Table 6). The results are provided in Figure 14. It is obvious that primary recovery is
lower at a larger fracture spacing. However, for CO2 injection (Figure 14a), a significant improvement
is seen in both the recovery at breakthrough and ultimate recovery with the increase of S, at least up to
S = 160 ft (in the case of kmatrix = 10 μD). The sweep efficiency increases with the increased spacing
between fractures. The improvement for oil recovery is less substantial when S is increased from 160
ft to 200 ft because the set maximum injection pressure (7000 psi) is reached and the injection rate is
decreased. The observation indicates that for a given cost on injection fluid, it is more beneficial to
increase the spacing between adjacent hydraulic fractures, provided that the injection pressure does not
limit the injection rate.

In the case of kmatrix = 1 μD, recovery decreases with the increase of fracture spacing before day 4500.
This is a result of the lower injection rate for higher fracture spacing due to the low matrix permeability.
Reducing the fracture spacing by creating more hydraulic fractures is crucial for the miscible recovery.
After day 4500, the curves start to crossover and the same trend for the recovery as the case of
kmatrix = 10 μD is observed.

9
3.6. Effect of pore size on phase behavior
The two phase envelops of the reservoir fluid at different pore sizes are shown in Figure 15a. The
figure clearly shows a dramatic decrease of Tc and Pc when the pore radius is smaller than 30 nm,
indicating that in the reservoir when miscible gas injection is applied, the minimal miscibility pressure
of the fluid can possibly be reduced in smaller pores, potentially leading to easier recovery. On the other
hand, the matrix permeability becomes smaller with pore size rp. For the investigation of pore size on
oil recovery, it was assumed that kmatrix = 10 D in a 50 nm pore. The phase behavior at these large pore
sizes should be very close to that of the bulk (green line in Figure 15a), according to Eq.1 and 2. It was
also assumed that kmatrix decreases proportionally with rp2. Figure 15b shows recovery by gas injection
in nano pores of different sizes, assuming the reservoir model has uniform rp. A decrease in ultimate
incremental recovery with pore size can be observed, suggesting that the effect of pore size on
permeability dominates over its effect on phase behavior.

3.7. Hydrocarbon gas injection

The proposed CO2 injection scheme may not always be feasible, particularly if CO2 is not available or
the cost is high. However, many shale oil reservoirs produce natural gas as a byproduct, which is often
flared because transportation pipelines are scarce. It has been suggested by other authors (Hoffman,
2012; Sanchez-Rivera et al., 2015) that natural gas or hydrocarbon gas can be injected to recover shale
oil with a similar recovery performance as CO2. The use of hydrocarbon gas can therefore be an
alternative.

The cumulative hydrocarbon gas that was produced after 500 days of primary depletion was reinjected
into the reservoir in this study. The recycled gas compositions for both kmatrix = 10 D and 1 D are
provided in Table 1. Oil recovery due to the recycled gas injection is compared with CO2, as shown in
Figure 16a for kmatrix = 10 μD and Figure 16b for kmatrix = 1 μD. For both matrix permeabilities,
hydrocarbon gas injection improved the production over CO2. Ultimate recovery after 5500 days was
4.2% and 2.0% above that with CO2 injection for kmatrix = 10 μD and 1 μD, respectively, even though the
MMP of both the recycled gases are no less than that of CO2 (Table 2). It was observed that the
viscosity of the recycled gas is about 0.044 cp at reservoir conditions, lower than the CO2 viscosity
(0.064 cp). These values are about a third of the viscosity of the reservoir oil (0.142 cp). Therefore, the
miscible fingering effect is negligible for both hydrocarbon gas and CO2. Since the hydrocarbon gas has
a lower viscosity than CO2, it can lead to a higher injectivity, resulting in a faster oil recovery. While it
should be noted that the results may be applicable for the chosen reservoir and fluid properties, the
results at least show that recycled hydrocarbon can be considered an alternative when CO2 is not
available.

4. Conclusions
A novel gas injection scheme was implemented to improve oil recovery in shale by injecting gas from
one fracture and producing oil from another fracture. A computational domain was created with two
half-stages along a well to represent shale reservoirs with horizontal wells and multistage hydraulic
fractures. Main conclusions drawn from the computational results are:

1. Base cases show a substantial improvement of oil recovery over the primary production by using the
fracture-to-fracture gas-flooding scheme. At injection pressure Pin = 7000 psi, the incremental recovery
for 5000 days of CO2 injection following 500 days of depressurization was 15.7% and 12.5% OOIP for
a matrix permeability of 10 μD and 1 μD, respectively, suggesting that this injection scheme has the
potential to improve recovery in shale oil.
10
2. Injection pressure Pin is an important operating parameter for the ultimate recovery. Increasing Pin
can lead to a significant improvement if the matrix permeability is low. For the reservoir domain with
kmatrix = 10 μD, injection even at near miscible condition (Pin = 3500 psi) can lead to a substantial
incremental recovery over primary depletion. When kmatrix = 1 μD, Pin needs to be well above the MMP
for the injection scheme to be beneficial due to insufficient pressure gradient.

3. For a fixed correlation length, larger Dykstra-Parsons coefficients significantly reduce the ultimate
incremental recovery by gas injection.

4. The presence of natural fractures can facilitate miscible gas injection. However, too many natural
fractures result in early breakthrough and low ultimate incremental recovery.

5. The natural fractures not only affect the oil recovery by their population, but also by how they
intersect with hydraulic fractures.

6. Reducing the hydraulic fracture spacing, S, improves the production proportionally before
breakthrough, especially for the lower matrix permeability. However, shorter S also leads to
significant production decline after breakthrough and lower sweep efficiency.

7. As the pore size in the matrix becomes smaller, recovery decreases because of permeability reduction.
The effect of phase behavior change due to pore size is less important for recovery.

8. Injection of recycled hydrocarbon gas improves ultimate recovery over the injection of CO2 because
the recycled gas has a lower viscosity (and both are multi-contact miscible at typical pressures and
temperatures).

Acknowledgement
We would like to thank the Gas EOR IAP under CPGE at the University of Texas at Austin for partial
financial support.

References
Brandt, A.R., Millard-Ball, A., Ganser, M., Gorelick, S.M., 2013. Peak Oil Demand: The Role of Fuel
Efficiency and Alternative Fuels in a Global Oil Production Decline. Environ. Sci. Technol. 47,
8031–8041.
Chen, C., Balhoff, M.T., Mohanty, K.K., 2014. Effect of Reservoir Heterogeneity on Primary Recovery
and CO2 Huff “n” Puff Recovery in Shale-Oil Reservoirs. SPEREE 17, 404–413.
Dombrowski, R.J., Fonseca, E.R., Karanikas, J.M., Moya, M.J.F., and Lamantia, B.C., 2015. Fluid
Injection in Light Tight Oil Reservoirs. US Patent 9127544.
EIA, 2014. Shale Oil and Shale Gas Resources Are Globally Abundant. Today in Energy.
http://www.eia.gov/todayinenergy/detail.cfm?id=14431.
Flannery, J., Kraus, J., 2006. Integrated Analysis of the Bakken Petroleum System, U.S. Williston Basin,
presented at the AAPG Annual Convention, Houston.
Gamadi, T.D., Sheng, J.J., Soliman, M.Y., 2013. An Experimental Study of Cyclic Gas Injection to
Improve Shale Oil Recovery. SPE-166334-MS.
Gaswirth, S.B., Marra, K.R., Cook, T.A., Charpentier, R.R., Gautier, D.L., Higley, D.K., Klett, T.R.,
11
 Lewan, M.D., Lillis, P.G., Schenk, C.J., Tennyson, M.E., Whidden, K.J., 2013. Assessment of
Undiscovered Oil Resources in the Bakken and Three Forks Formations, Williston Basin Province,
Montana, North Dakota, and South Dakota, USGS.
Hawthorne, S.B., Gorecki, C.D., Sorensen, J.A., Steadman, E.N., Harju, J.A., Melzer, S., 2013.
Hydrocarbon Mobilization Mechanisms from Upper, Middle, and Lower Bakken Reservoir Rocks
Exposed to CO2. SPE-167200-MS.
Hoffman, B.T., 2012. Comparison of Various Gases for Enhanced Recovery from Shale Oil Reservoirs.
SPE-154329-MS.
Kovscek, A.R., Tang, G.Q., Vega, B., 2008. Experimental Investigation of Oil Recovery From Siliceous
Shale by CO2 Injection. SPE-115679-MS.
North Dakota Drilling and Production Statistics. https://www.dmr.nd.gov/oilgas/stats/statisticsvw.asp
(accessed 2015/06/10)
North Dakota Department of Mineral Resoursecs Presentation,
https://www.dmr.nd.gov/oilgas/presentations/TribalLeaderSummit09052012_50.pdf
Pedersen, K.S., Milter, J., Sørensen, H., 2004. Cubic Equations of State Applied to HT/HP and Highly
Aromatic Fluids. SPE Journal, 186-192.
Pitman, J.K., Price, L.C., LeFever, J.A., 2001. Diagenesis and Fracture Development in the Bakken
Formation, Williston Basin: Implications for Reservoir Quality in the Middle Member, USGS.
Rubin, B., 2010. Accurate Simulation of Non Darcy Flow in Stimulated Fractured Shale Reservoirs.
SPE-132093-MS.
Sigmund, P.M., 1976. Prediction of Molecular Diffusion at Reservoir Conditions. Part II - Estimating
the Effects of Molecular Diffusion and Convective Mixing in Multicomponent Systems. Journal of
Canadian Petroleum Technology, 15, 3-62.
Sanchez-Rivera, D., 2014. Reservoir Simulation and Optimization of CO2 Huff-and-Puff Operations in
the Bakken Shale. Master Thesis
Sanchez-Rivera, D., Mohanty, K., Balhoff, M., 2015. Reservoir Simulation and Optimization of Huff-
and-Puff Operations in the Bakken Shale. Fuel, 147, 82-94
Sheng, J.J., Chen, K., 2014. Evaluation of the EOR Potential of Gas and Water Injection in Shale Oil
Reservoirs. Journal of Unconventional Oil and Gas Resources 5, 1–9.
Shoaib, S., Hoffman, B.T., 2009. CO2 Flooding the Elm Coulee Field. SPE-123176-MS.
Song, C., Yang, D., 2013. Performance Evaluation of CO2 Huff-n-Puff Processes in Tight Oil
Formations. SPE-167217-MS.
Teklu, T.W., Alharthy, N., Kazemi, H., Yin, X., Graves., R.M., AlSumaiti, A.M., 2014. Phase Behavior
and Minimum Miscibility Pressure in Nanopores. SPE Reversoir Evaluation & Engineering
Vega, B., O'Brien, W.J., Kovscek, A.R., 2010. Experimental Investigation of Oil Recovery From
Siliceous Shale by Miscible CO2 Injection. SPE-135627-MS.
Wan, T., Sheng, J., Watson, M., 2014. Compositional Modeling of the Diffusion Effect on EOR Process
in Fractured Shale Oil Reservoirs by Gas Flooding. SPE-2014-1891403-MS.
Wan, T., Sheng, J.J., Soliman, M.Y., 2013. Evaluate EOR Potential in Fractured Shale Oil Reservoirs by
Cyclic Gas Injection. SPE-168880-MS.
Wang, X., Luo, P., Er, V., Huang, S., 2010, Assessment of CO2 Flooding Potential for Bakken
Formation, Saskatchewan. SPE-137728-MS.
Yanze, Y., Clemens, T., 2012, The Role of Diffusion for Nonequilibrium Gas Injection into a Fractured
Reservoir. SPE Reservoir Evaluation & Engineering
Yu, W., Lashgari, H., Sepehrnoori, K., 2014. Simulation Study of CO2 Huff-n-Puff Process in Bakken
Tight Oil Reservoirs. SPE-169575-MS.
Yu, W., Lashgari, H.R., Wu, K., Sepehrnoori, K., 2015. CO2 Injection for Enhanced Oil Recovery in
Bakken Tight Oil Reservoirs, Fuel 159, 354-363.
12
13
 Table 1. Pseudo component
C7- C11- C15- C20-
Component CO2 C1 C2 C3 iC4 nC4 iC5 nC5 C6 C30+
C10 C14 C19 C29
A* 0.39 35 15 9.5 1.2 4.6 1.3 2.4 2.9 12 6.1 3.9 3.1 2.4
Mole
Frac. B* 0.59 54.9 21.3 12.6 1.5 5.1 1.0 1.7 0.93 0.43 0.0 0.0 0.0 0.0
( %) *
C 0.58 55.8 20.1 12.3 1.4 5.0 1.0 1.7 0.93 0.43 0.0 0.0 0.0 0.0
*
A: Reservoir fluid.
*
B: Recycled gas produced by 500 days of primary depletion for kmatrix=10 D.
*
C: Recycled gas produced by 500 days of primary depletion for kmatrix=1 D.

Table 2. Fluid properties at reservoir conditions

Parameters Values Units
Oil density 40.47 lb/ft3
Oil viscosity 0.142 cp
Oil compressibility 1.29×10-5 psi-1
Saturation pressure 2664 psi
CO2 3470 psi
Minimum miscible Recycled gas (k
matrix=10 D) 3750 psi
pressure
Recycled gas (kmatrix=1 D) 3825 psi
CO2 8500 psi
First contact
Recycled gas (kmatrix=10 D) 5200 psi
miscible pressure
Recycled gas (kmatrix=1 D) 5375 psi

14
Table 3. Corey-Brooks parameters for relative permeability curves
Oil-water relative permeability
Residual water saturation, Swr 0.2
Critical water saturation, Swcrit 0.325
Residual oil saturation, Sor 0.1
Endpoint water relative permeability, krwo 1
o
Endpoint water relative permeability, kro 1
Exponent for calculating krw 2
Exponent for calculating kro 3.5

Gas-liquid relative permeability

Residual gas saturation, Sgr 0.03
Critical gas saturation, Sgcrit 0.07
o
Endpoint gas relative permeability, krog 1
Endpoint liquid relative permeability, krgo 0.85
Exponent for calculating krog 5.75
Exponent for calculating krg 3.15
Reservoir depth 9500
Formation compressibility 1×10-6
Injected gas CO2
Longitudinal dispersitivity l 0.033
Transverse dispersitivity t 0.0033
Depletion period 500
Injection period 5000

15
Table 4. Domain properties and reservoir conditions of the base models.
Property Value Units
Matrix permeability kmatrix 10 or 1 μD

Hydraulic fracture width (real/model) 0.005/2 ft

Hydraulic fracture permeability (real/model) 10,000/25 mD
Hydraulic fracture porosity (real/adjusted) 43/0.11 %

Natural fracture width (real/model) 0.003/2 ft

Natural fracture permeability (real/model) 100/0.15 mD
Natural fracture porosity (real/adjusted) 100/0.15 %

Hydraulic fracture spacing S 160 ft

Initial water saturation 20 %
Initial reservoir pressure 6840 psi
Production pressure Pout 1000 psi
Injection pressure Pin 7000 psi
Reservoir temperature 241 °F
Reservoir depth 9500 ft
Formation compressibility 1×10-6 psi-1
Injected gas CO2 N/A
Longitudinal dispersitivity l 0.033 ft
Transverse dispersitivity t 0.0033 ft
Depletion period 500 days
Injection period 5000 days

16
 Table 5. Recovery data for the base cases
Properties Value Unit
Original oil in place 10.80 Mstb
Original gas in place 13.15 MMscf
Original water in place 4.679 Mstb
Hydrocarbon pore volume 102.98 Mrft3

Base case kmatrix=10D Day 1000* / 5500

Cumulative CO2 injected 9.68 / 119.0 MMscf
Cum. oil recovered (CO2 injection) 2.62 / 3.54 Mstb
Cum. oil recovered (depressurization) 1.26 / 1.83 Mstb
Incremental oil recovered 1.36 / 1.71 Mstb
Recovery factor (CO2 injection) 24.3 / 32.8 % OOIP
Recovery factor (depressurization) 11.7 / 17.1 % OOIP
Incremental recovery factor 12.6 / 15.7 % OOIP

Base case kmatrix=1D Day 3600* / 5500

Cumulative CO2 injected 6.14 / 9.045 MMscf
Cum. oil recovered (CO2 injection) 1.89 / 2.37 Mstb
Cum. oil recovered (depressurization) 0.89 / 1.02 Mstb
Incremental oil recovered 1.00 / 1.35 Mstb
Recovery factor (CO2 injection) 17.5 / 21.9 % OOIP
Recovery factor (depressurization) 8.2 / 9.4 % OOIP
Incremental recovery factor 9.3 / 12.5 % OOIP
*Day when major CO2 breakthrough occurred and production declined

Table 6. Gas injection rate in reservoir domains with different hydraulic fracture spacing

Hydraulic Fracture Spacing S (ft) 80 120 160 200

Injection rate kmatrix=10D 20 30 40 50

(ft3/day) kmatrix=1D 1.5 2.25 3 3.75

17
 Prod. Well

Inj. Prod Inj. Prod. Inj. Prod Inj. Well

. .

(a) (b)
Figure 1. Two viable schemes for effective fracture-to-fracture injection in shale (Dombrowski et al.,
2015). (a) Fractures along the same well are used alternatively for injection and production. (b) One
well is used for injection while the other for production when the well spacing is close enough.

Injection Production
fracture fracture

(b)
Figure 2. (a) A well completion scheme with two horizontal wells and intersected hydraulic fractures.
The computational domain (red and shaded) is 1000 ft in the y direction and 160 ft in the x direction.
(b) Upper half of the computational domain where finer meshes are implemented. Hydraulic fractures
are shown in red, and natural fractures in blue. Discretization around the one of the hydraulic fractures
is shown in the magnified inset.

18
Figure 3. Oil recovery from the base cases with kmatrix = 10 D (black) and kmatrix = 1 D (red). The
solid curves represent CO2 injection after 500 days of primary depletion. The dash curves represent
primary depletion. The breakthrough points are shown by the circles.

Figure 4. (a) Oil recovery for different injection pressure Pin. The curves correspond to depletion (dash
black) and CO2 injection at Pin = 3500 psi (solid red), 5000 psi (solid blue) and 7000 psi (solid black,
base case). The reservoir domain has kmatrix = 10 μD. (b) Effect of injection pressure, Pin on average
daily oil rate before breakthrough .

19
 Pin 3500psi 5000psi 7000psi

160 ft
Figure 5. Global mole fraction of CO2 (a) at the end of CO2 injection (day 5500) at 3500, 5000 and
7000 psi. The reservoir domain has kmatrix = 10 D. Only the upper half of the computational domain is
shown.

Figure 6. Oil recovery versus pore volume of CO2 injected when kmatrix = 10 μD. The curves
correspond to CO2 injection pressure at 3500 psi (red), 5000 psi (blue) and 7000 psi (black). The curves
are cut off at Vpi = 0.18.

20
Figure 7. (a) Oil recovery versus time for different injection pressures Pin. The curves correspond to
depletion (dash black) and CO2 injection at Pin = 3500 psi (solid red), 5000 psi (solid blue) and 7000 psi
(solid black). The reservoir domain has kmatrix = 1 μD. (b) Average daily oil rate before breakthrough
versus Pin in the same reservoir domain.

y k (mD)

160 ft
Figure 8. Heterogeneous permeability in the common logarithm scale for Dykstra-Parsosn coefficient
VDP = 0.5 in the upper half of the computation domain. Correlation length is 40 ft.

21
Figure 9. (a) Oil recovery versus time for different Dykstra-Parsons coefficient VDP. Solid lines
correspond to CO2 injection after 500 days of primary depletion. Dash lines are primary depletion.
(b) Incremental recovery at day 5500 versus Dykstra-Parsons coefficient.

a b c d e

Figure 10. Natural fractures (blue) as imposed on the stimulated region of the reservoir domain.
Hydraulic fractures are shown in red. Heterogeneity of the reservoir is not shown to avoid complexity.
The natural fracture probabilities are (from left to right): p = 0.1, 0.2, 0.5, 0.68 and 0.63 (unevenly
distributed).

22
Figure 11. Effect of natural fractures on recovery for (a) oil recovery versus injection time for the case
of evenly distributed fractures, (b) oil recovery versus cumulative gas injected and (c) comparison of oil
recovery versus injection time between evenly distributed (black and red) and propped natural fractures
(blue).

23
Figure 12. Oil recovery versus time for reservoir domains with kmatrix = 10 D (a) and kmatrix = 1 D (b)
and different hydraulic fracture spacing S at constant pressure injection. In both figures, the red curves
represent S = 80 ft, and the black curves represent S = 160 ft (base case). The cases for CO2 injection
after 500 days of depletion are shown by the solid lines. Primary depletions are shown by the dash lines.

Figure 13. Production versus S-1 in reservoir models with matrix permeability of 10 D (black) and
1 D (red).

24
Figure 14. Oil recovery versus time for reservoir domains with kmatrix = 10 D (a) and kmatrix = 1 D (b)
and different hydraulic fracture spacing S at constant rate injection.

Figure 15. (a) Effect of pore space on the phase behavior of the crude oil. (b) Effect of pore size on oil
recovery. The ultimate recovery for the cases of rp = 5 nm and 3 nm is extremely small. Therefore,
those results are not presented.

25
Figure 16. Oil recovery versus time in reservoir models with matrix permeability of 10 D (a) and 1 D
(b). The curves correspond to primary depletion (dash black), CO2 injection (black) and injection of
hydrocarbon gas (red).
Highlights
 Compositional modeling of fracture to fracture miscible gas injection
 Enhanced oil recovery in shale
 Substantial improvement in oil recovery by CO2 or hydrocarbon gas injection
 Effect of natural fracture density on oil recovery

26