SPE 116688

Production Data Analysis of Shale Gas Reservoirs

Adam M. Lewis* and Richard G. Hughes, Louisiana State University

Copyright 2008, Society of Petroleum Engineers

This paper was prepared for presentation at the 2008 SPE Annual Technical Conference and Exhibition held in Denver, Colorado, USA, 2124 September 2008.

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

Abstract
Unconventional shale gas reservoirs have become a very important part of the resource base throughout the world, but
especially in the United States. Production data analysis techniques using various forms of material balance time originally
put forth by Palacio and Blasingame (1993) have been proposed and validated in recent publications for a variety of problems
(including shale gas and coalbed methane systems) using an adjusted system compressibility function similar to that proposed
by Bumb and McKee (1988) to account for adsorbed gas. These modified material balance solutions allow for "type curves"
(rate or pressure solutions) to be used in a conventional analysis manner.
The significant challenge in the application of production data analysis for shale gas systems is to determine what
the parameter values (analysis results) represent within the context of the inherent complexity of these systems. In this work
we propose a slight (but substantive) modification to material balance time and apply the technique to synthetic and field data
to assess the capability of this approach for the analysis of production data from gas shales. The formulation we use is that of
Cox, et al. (2002) which provides a means for evaluation production data as an equivalent "well test", where we are able to
observe characteristic flow regimes in the data and constrain several of the key parameters of interests in shale gas systems.
While the field results are difficult to corroborate, the relative magnitudes of the parameters we obtained appear to be
reasonable.
Introduction
The vast majority of gas production in the United States comes from what are known as conventional hydrocarbon reservoirs.
However, these conventional reservoirs are becoming increasingly difficult to find and exploit. In an era of rising prices for
crude oil and natural gas, the ability to produce these commodities from unconventional reservoirs becomes very important.
The United States Geological Survey states that, among other things, an unconventional reservoir must have
regional extent, very large hydrocarbon reserves in place, a low expected ultimate recovery, a low matrix permeability and
typically has a lack of a traditional trapping mechanism (Schenk, 2002). In particular, shale gas reservoirs present a unique
problem to the petroleum industry in that they may contain natural gas in the pore spaces of the very tight reservoir rock, in
the pore spaces of natural fractures in the formation and on the surface of the rock grains themselves which is referred to as
adsorbed gas (Montgomery, et al., 2005). This sorbed gas presents difficulties in that desorption time, desorption pressure,
and volume of the adsorbed gas all play a role in how this gas affects the production of the total system. Adsorption can allow
for significantly larger quantities of gas to be in place and possibly produced.
Historically, the first commercially successful gas production in the U.S. came from what would now be considered
an unconventional reservoir in the Appalachian Basin in 1821. Currently, some of the largest gas fields in North America are
unconventional, shale gas reservoirs such as the Lewis Shale of the San Juan Basin, the Barnett Shale of the Fort Worth
Basin, and the Antrim Shale of the Michigan Basin. In addition, gas production from unconventional reservoirs accounts for
roughly 2% of total U.S. dry gas production (Hill, et al., 2007).
Shale gas reservoirs present numerous challenges to analysis that conventional reservoirs simply do not provide. The
first of these challenges is the dual porosity nature of these reservoirs. Similar to carbonate reservoirs, shale gas reservoirs
almost always have two different storage volumes for hydrocarbons, the rock matrix and the natural fractures (Gale, et al.,
2007). Because of the plastic nature of shale formations, these natural fractures are generally isolated or closed due to the
pressure of the overburden rock (Gale, et al., 2007). Consequently, their very low matrix permeability, usually on the order of
hundreds of nanodarcies (nd), makes un-stimulated, conventional production difficult, if not impossible. Therefore, almost
every well in a shale gas reservoir must be hydraulically stimulated (fractured) to achieve economical production. These
hydraulic fracture treatments are believed to re-activate and re-connect the natural fracture network (Gale, et al., 2007).
*
Now with ConocoPhillips, Anchorage, AK
2 SPE 116688

Another key difference between conventional gas reservoirs and shale gas reservoirs is the adsorbed gas. The nature
of the solid sorbent, temperature, and the rate of gas diffusion all affect the adsorption (Montgomery, et al., 2005). Currently,
the only method for accurately determining the adsorbed gas in a formation is through core sampling and analysis. The
amount of adsorbed gas is usually reported in SCF/ton of rock or SCF/ft3 of rock. Depending on the situation, adsorbed gas
can represent a large percentage of the gas in place and can have a dramatic impact on production.

Theoretical Discussion
There have been many methods for analyzing production data put forth over the years by many talented contributors. These
authors have built upon each others work in much the same way that this work builds upon theirs. This work will focus on a
methodology for analyzing production data of shale gas reservoirs in an attempt to answer the questions of what can be
seen? and when can it be seen?.
Building on the work of Fraim and Wattenbarger (1987), Palacio and Blasingame (1993) derived material balance
pseudotime for gas reservoirs producing at variable rates and bottomhole pressures. Starting with the material balance
equation for dry gas reservoirs, they incorporated pseudopressure and pseudotime to yield
p
c
gi ti Gz i p
ta dp (1)
qg pi pi
z g

Equation 1 is scaled on initial values for the pressure, gas deviation factor, system compressibility, gas viscosity, and changes
in the real gas pseudopressure from the initial pressure to the average reservoir pressure at the time of the flow rate value. It
also is scaled by the ratio of the original gas in place to the production rate value. This allows a system produced at variable
rates and/or pressures to be analyzed as either an equivalent constant pressure system or an equivalent constant rate system.
Agarwal, et al. (1999) verified that the formulation of material balance time was a valid method for transforming variable
rate/pressure systems to their constant rate/constant pressure counterparts. Hager and Jones (2001) and Cox, et al. (2002)
have also used this validity in their work. This is a very important finding because it allows the vast library of well known
analytical solutions for constant rate systems to be utilized in production data analysis.
Since most real shale gas systems have adsorbed gas (Montgomery, et al., 2005), it is necessary to account for this
component when analyzing production data or constructing simulation models. Typically, adsorbed values are obtained from
laboratory work in the form of Langmuir Pressure (pL) and Langmuir Volume (vL). An isotherm can be constructed using the
Langmuir equation:
vL p
v ads (2)
pL p
In Eqn. 2, vL is the maximum volume of adsorbed gas (in scf/ft3 for field units) and pL (in psia for field units) controls what
portion of that gas desorbs. In production data analysis, the desportion of adsorbed gas can be handled using a modified
system compressibility like that proposed by Bumb and McKee (1988). Their approach incorporates pL and vL to account for
adsorption. Spivey and Semmelbeck (1995) present this modified form of the total system compressibility as
p sc v L p L Tz
ct* S w c w (1 S w )c g cf (3)
Tsc z sc ( p L p ) 2
When analyzing production data or pressure transient data it is often convenient to use dimensionless variables.
Dimensionless time, dimensionless pressure and dimensionless flowrate are the three variables commonly used. Typical
definitions for these dimensionless variables are (Lee and Wattenbarger, 1996)
0.00638kt
td (4)
c t g rw2
kh[m( p i ) m( p wf )]
p wd (5)
1422Tq g
141.2q g g Bg
qd (6)
kh( pi p wf )
For the case of production data analysis of shale gas systems, one must incorporate both material balance time and the
adjusted system compressibility proposed by Spivey and Semmelbeck (1995) into Eqn. 4. When using the standard definition
for material balance time in place of the time variable in Eqn. 4, compressibility terms are usually eliminated by assuming
that cg ct. For systems with adsorbed gas, this assumption does not hold. The formulation of material balance time to
account for the adsorbed gas is
gi c ti Gi z i t p cg dp
ta dt (7)
qg p i 0 z g ( p )ct ( p) dt
SPE 116688 3

This equation leaves the compressibility terms inside the pseudopressure integral which means that it also remains in the
equation for dimensionless time. This has the same effect as the modified z-factor proposed by Clarkson, et al. (2007). For
production data analysis, the dimensionless time function can also be scaled by the drainage area rather than the wellbore
radius. The resulting expression for dimensionless time in a single-porosity, single-phase, shale gas system is then
0.00638 z i kG [mct ( pi ) mct ( p )]
td (8)
2 pi rw2 qg
where [ m ct ( p i ) mct ( p ) ] is the pseudopressure integral which includes the ratio (cg/ct). Note that the psudopressure
difference in Eqn. 5 also should use the revised definition.
Converting rate and pressure data obtained at various times throughout the life of a well using Eqns. 5 and 8 should
then be equivalent to a constant rate solution to the diffusivity equation when all of the parameter values (or more accurately
when the relative ratios of all of the parameter values) are correct. Note also that there is a direct tie between the assumed or
measured value for the initial pressure and the average reservoir and flowing bottomhole pressures assocoiated with each rate
used in the analysis.
Analysis of a simulated data set
For this discussion, a simulated dataset will be used so that data reliability and accuracy do not cause problems. An analyst
rarely has a complete understanding of the reservoir in question when the analysis of the production data is performed. So,
the same set of production data will be analyzed using different analytical models (single porosity, dual porosity and dual
porosity w/hydraulic fracture). The effect of these different analysis styles will be noted in the match parameters. The
important question being asked is whether or not any additional insight is gained or lost from using a more complicated
model.
Presently, reservoir descriptions of common shale gas reservoirs are considered to be dual porosity reservoirs with
adsorbed gas in the matrix portion only. In addition, shale gas wells are typically stimulated by hydraulic fracturing.
Therefore, the simulation model used for this section is a representation of these features. Production comes from a single
well located in the center of the model which can be seen in Figure 1. Local grid refinement was used to create the hydraulic
fracture of width 4 inches (the smallest value allowed). A list of the reservoir parameters used in the simulation can be found
in Table 1.
Fracture Well

40 ft

5280 ft

Figure 1: Reservoir schematic for simulation data set

Table 1: Simulation parameters
Parameter (unit) Value Parameter (unit) Value
Height/Net Pay (ft) 40 Initial Pressure, (psig) 1500
Length (x & y) (ft) 5280 Water Saturation, Sw 0.05
Well Radius, rw (ft) 0.25 Rock Dens., (g/cm^3) 2.45
Porosity, m 0.10 pL (psia) 635
Porosity, f 0.001 ( = 0.01) vL (scf/ton) 89
xf (ft) 480 (Cdf = 0.03) Temperature, (oF) 225
Permeability, km (md) 0.01 Gas Gravity 0.56
Permeability, kf (md) 1.875 ( = 1x10-5) Gas Composition 100% CH4
Permeability, kfh (md) 2500 (Cdf = 0.3) Initial BHP (psig) 900

The constant flowing bottomhole pressure was arbitrarily set so that starting rates close to 1 MMSCF/day were obtained. The
model was run for 10 years with daily production data being recorded for the first 3 months followed by monthly data points
for the remaining time.
4 SPE 116688

For the first two models, a dual porosity system with adsorbed gas and a hydraulic fracture is being simplified since
it is being viewed as a single or dual porosity porosity system without the hydraulic fracture. All the properties for both
natural and induced fractures influence the rates obtained from the flow simulation, but do not appear in the conversion of the
data into dimensionless format. Three analysis methods were initially used to evaluate the resulting data. Standard qd vs td
(Ehlig-Economedes and Ramey, 1981) and Fetkovich (1980) type curves proved to have a high degree of uncertainty as the
late-time data values appeared to be right at the cross-over between the transient flow period and the boundary dominated
flow period. Using techniques suggested by Cox, et al. (2002) proved to be significantly easier to interpret as the classic flow
regimes from pressure transient diagnostic (derivative) curve analysis were observed. This allows flow regimes to be
identified more easily and clearly shows the presence of boundary dominated flow when the other two methods could not.
Also, if one analyzes Eqn. 8 and assumes that rates and bottomhole pressures are accurately known, the uncertainty lies in the
parameter values for permeability, gas-in-place, porosity, and initial pressure. Using the group kG to match the data lumps
most of this uncertainty into one term.
The match is achieved by varying reservoir parameters, chiefly permeability, porosity, and drainage radius, and then
converting the rate or pressure data into dimensionless format to obtain a match. So, every time one of these parameters is
changed the plotting functions are recalculated to reflect this new data. The new plotting functions are then plotted against the
analytical dimensionless solution for the model of interest with the same parameter values until a match is attained. To
achieve the match, we assumed that the initial pressure and porosity values would be the most accurately known parameters.
For shale gas systems, these parameters may in fact be highly uncertain. Once the initial pressure (1500 psig) and porosity
(10%) values were set, values for the remaining matching parameters were estimated until a reasonable shape to the
derivative curve could be seen. Rodriguez (2005) has shown that, for tight gas systems, reasonable derivative shapes for high
quality data occur when the ratios of the system parameters are approximately correct and discontinuities appear when
parameter values are incorrect. The vertical location of the derivative data was then adjusted by altering the system
permeability value used in both Eqns. 5 and 8. The horizontal position was then adjusted by varying the drainage area of the
system.
Figure 2a shows the result for analyzing the data as a single porosity system with no wellbore storage or skin.
Shown are the results for assuming that the system had adsorption and one that ignored the adsorbed gas. Note that the
dimensionless pressure values nearly overlay each other and the derivative curves are also very close to one another until late
time. For this particular case, ignoring the effects of adsorption would cause an analyst to estimate a distance to the outer
boundary that is a bit more than 10% higher than the value estimated using the adsorbed gas. The estimate of the free gas in
place is overestimated by about 50%. In addition, higher values for vL and/or the initial pressure will cause larger errors in the
drainage area and hence the free gas volume estimate will increase substantially. For the case presented here, neglecting the
adsorbed gas underestimates the total gas in place by only about 5%. In addition, the permeability value that best matches the
data is the value associated with the natural fracture system. Thus we learn very little about the porosity of the natural
fracture system or the permeability of the matrix system from this analysis.
Figure 2b shows the result for analyzing the system as a dual porosity system with adsorbed gas. Since the late-time
data in Figure 2a was well matched, the parameter values that were adjusted were the values for the dual porosity parameters
and (the interporosity flow coefficient and the storativity ratio as defined by Spivey and Semmelbeck, 1995). The result
shown uses the transient interporosity flow formulation of the analytic solution; however an almost identical match could be
obtained for this problem assuming the pseudo-steady state formulation. If the effective porosity for the system is assumed to
be that of the matrix and the permeability obtained from the analysis is assumed to be the permeability of the natural fracture
system, then the fracture porosity can be obtained from the value for and the matrix permeability can be obtained from .
Figure 2c shows the result for analyzing the system as a hydraulically fractured, dual porosity system (Aguilera,
1989 and Olarewaju and Lee, 1989) with the adsorbed gas correction. For this case, the transient interporosity flow model did
provide a better result and the adjustment values were the hydraulic fracture width, the fracture half-length and the fracture
permeability. For this case, we assumed a value for the fracture permeability and for the fracture width that were the input
parameters for the simulation and the unknown variable was then the fracture half-length.
Table 2 provides a listing of the values that were assumed for the analyses. Since we had the actual values from the
simulation input deck, these values can be determined; for real data, these values would need to be constrained in some
manner. We will discuss ideas as to how these values might be estimated for field cases. Table 3 then provides the
unknown variable results along with the values from the simulation input deck. For the single porosity result, the effective
drainage radius is slightly larger than the actual value and the effective permeability is more closely related to the fracture
permeability than the matrix permeability. While we expected the differences in the permeability value observed in the
system, the fact that we did not find the correct drainage area was not something we observed in our preliminary studies
using the method (Lewis, 2007). Another way we tried to adjust the data to match the input drainage area was to adjust the
value for rw to account for the hydraulic fracture. However, the skin factor necessary to shift the data to obtain the correct
drainage area was a very small negative value which was not representative of the modeled fracture. We believe that this
discrepancy is most likely due to the way we set the simulator up to model the flow behavior of the hydraulic fracture. Again
we did not observe the discrepancy when we modeled an unstimulated well for a variety of single and double porosity
systems with adsorption (Lewis, 2007).
SPE 116688 5

100.0
Pwd
Pwd'
Pwd (no correction)
Pwd' (no correction)
Correct Analytical Solution
Incorrect Analytical Solution

10.0
pwd or pwd'

1.0

0.1
1.00E+05 1.00E+06 1.00E+07 1.00E+08 1.00E+09
td

(a)
100 100
Analytic Model Analytic Model
Data
Data

10 10
pwd '

1
pwd'

0.1
0.1

0.01
0.01 1.00E+00 1.00E+01 1.00E+02 1.00E+03 1.00E+04 1.00E+05
1.00E+03 1.00E+04 1.00E+05 1.00E+06 1.00E+07 1.00E+08 1.00E+09
td/Cdf
td

(b) (c)
Figure 2: Diagnostic curves for the analysis of the simulated data set

Likewise the dual porosity parameters are not exactly what the input data suggests they should be, but they are
reasonably close. Our input value for was 0.01 while the match value is 0.008 and our value for was found to be 5 10-6
rather than the input value of 1 10-5. Finally, using the input values for the fracture permeability and fracture width, our
match result suggests that there is only a 10 foot effective fracture half-length for this system. Given that our input value was
480 ft, this result is less than satisfying; however there is very little data during this part of the flow period to work with. This
would suggest that data values for a system like this one might need to be more frequent than daily. Data at such a high
frequency during the early flow stages of a well would need to be planned up front so that data quality upon initial startup of
a well could be maximized. Higher pressure drawdown values than what we simulated also may help to provide the necessary
early-time data necessary to evaluate fracture parameters.
6 SPE 116688

Table 2: Constant values used in the parameter matching

Parameter (unit) Value Parameter (unit) Value
Height/Net Pay (ft) 40 Initial Pressure, (psig) 1500
Well Radius, rw (ft) 0.25 Water Saturation, Sw 0.05
Temperature, (oF) 225 Rock Dens., (g/cm^3) 2.45
Gas Gravity 0.56 pL (psia) 635
Gas Composition 100% CH4 vL (scf/ton) 89
Initial BHP (psig) 900

Hyd frac perm, kfh (md) 2500 Fracture width (ft) 0.333

Table 3: Matched parameter values

Effective Drainage Effective Effective
xf (ft)
Radius, re (ft) Porosity ( ) Permeability, k (md)
-5
Simulated 2979 0.10132 kf = 1.875; km = 0.01 0.01 1 10 480
Single Porosity Analysis 3250 0.1 1.125 N/A N/A N/A
Dual Porosity Analysis 3250 0.1 1.875 0.008 5 10-6 N/A
Dual Porosity w/fracture 3250 0.1 1.875 0.008 5 10-6 10

Field Example 1
Now, the insights gained from the simulation example cases will be applied to a field case. Production data for 32 months
from a well in the Barnett Shale, Texas were obtained from a database provided by Devon Energy Corporation. Daily rates
for the gas, oil and water production were recorded for the entire 32 months. Daily tubing and casing pressures were recorded
for the first two months, but weekly tubing and casing pressures were recorded for the final 30 months (Figure 3). An
important aspect to note about this well is that it was reported to have some interference from neighboring wells. Also, the
shift to weekly tubing head pressures will definitely pose a problem if this shift turns out to coincide with any important flow
regime changes.

Data for Barnett Shale Example Well #1

10000
Gas Rate (MSCF/day)
Water Rate (STB/day)
GLR (MSCF/BBL)
1000 FBHP (psia)
Gas Rate, Water Rate, GLR

100

10

0.1
0 200 400 600 800 1000 1200
Time (days)

Figure 3 Combined data plot for Barnett Shale Example Well #1

SPE 116688 7

The first step in analyzing this data was to remove any production data entries that did not have an associated tubing
pressure. This resulted in 165 data points that had complete data (i.e. rates and pressures). Next, bottomhole pressures were
estimated using the Cullender and Smith flow correlation (Brill and Beggs, 1988). Then, both the rate and pressure data were
smoothed using an exponential smoothing algorithm. The algorithm uses a dampening factor ( ) to smooth the data. The
equation used in the smoothing was
x smth (t ) x(t ) (1 ) * x smth (t 1) (9)

For both the rate and pressure streams, the value of used was 0.25. The smoothed rate and pressure profiles can be seen in
Figure 4.

1200 400
380
1000 Smooth Pressure
360
Gas Rate (mscf/day)

Smooth Rate 340 Orig Pressure

800

FBHP (psi)
Orig Rate 320
600 300
280
400 260

200 240
220
0 200
0 200 400 600 800 1000 0 200 400 600 800 1000
time (days) time (days)

(a) (b)
Figure 4 Original and smoothed rate and pressure values for Barnett Shale Example Well #1
The smoothed rate and pressure profiles were used in the analysis with no further modification. The smoothed rate
profile has 103% of the cumulative production of the original rate profile. For the first evaluation of this well, it was assumed
that this is a single porosity system with adsorbed gas. This minimized the complexity of the analysis and provided a solid set
of base parameters with which to do more complicated analysis for our simulation study. There is no Langmuir data available
for this well. Langmuir parameters were estimated based on data provided to us by Devon Energy Corp. for another part of
the field. The Bumb and McKee (1988) correction was applied with these initial estimates of the Langmuir parameters. To
test the robustness of the match, 3 values of initial pressure were used (4000 psi, 5000 psi, and 6000 psi). This parameter is
perhaps the most difficult parameter to obtain in a low permeability shale-gas system, and it is the most critical parameter in
the material balance time approach to production data analysis. In addition, values for the adsorption parameters were
increased in one trial to show the effect of using different adsorption parameters. Table 5 shows other properties that were
used when matching this data to the analytical models.
Table 4 Initial Property Estimates for Barnett Shale Example Well #1

Parameter Value Parameter Value

g 0.6 Sw 0.3
re 625 ft 0.06255
rw 0.33 ft Tres 180o F
A 28.2 acres pi 4000 psia
h 433 ft pL 635 psia
-8 3
vb 5.31 10 ft vL 89 scf/ton

Individual match plots for each pressure are not shown here, but the results were similar for all three initial pressure
cases with the 4000 psia case shown in Figure 5. The match plots for the other initial pressures can be seen in Appendix A of
Lewis (2007). The first thing that should be noted is the poor match to the early-time or radial flow portions of the data. What
could be considered a radial flow section appears to be 2 to 3 times the typical value for radial flow of 0.5. This seems to
indicate that there are multiple (2 or 3) boundaries of some sort affecting this system. In this case, both the dimensionless
pressure and the dimensionless pressure derivative were used to constrain the match. At late-time, both pwd and pwd overlay
the type curve. If an attempt is made to match the early-time portion of the data, it is not possible to get both pwd and pwd to
simultaneously overlay the type curve. This was evidence that the early-time data could not be fit using a single porosity type
curve. This could be because the data does not come from a single porosity system or because the data even though smoothed
are not consistent enough to provide an appropriate derivative.
8 SPE 116688

Matching Barnett Shale Example Well #1

(Single Porosity)
1000
Data (Pwd) Data (Pwd')
Analy. Model (300ft) Analy. Model (300ft)
Analy. Model (500ft) Analy. Model (500ft)
100 Analy. Model (1000ft) Analy. Model (1000ft)

10
pwd or pwd'

0.1

0.01
10 100 1000 10000 100000 1000000 10000000 1E+08 1E+09
td or tda_new

Figure 5: Match for Barnett Shale Example Well #1 assumed to be a single porosity system.

Next, the data from this well was analyzed as a dual porosity system to see if any additional information can be
obtained. Figure 6 shows the match assuming a dual porosity system. In this case, both the dimensionless pressure and the
dimensionless pressure derivative were used to constrain the match. Figure 6 shows a better overall match to the entire
dataset rather than just late-time, especially for pwd. There are what appear to be two flat portions on the derivative curve

Matching Barnett Shale Ex Well #1

(Dual Porosity)
1000
Analy. Model (Pwd)
Analy. Model (Pwd')
Data (Pwd)
100
Data (Pwd')

10
pwd or pwd'

0.1

0.01
10 100 1000 10000 100000 1000000 10000000 1E+08 1E+09
td or tda_new

Figure 6 Match for Barnett Shale Example Well #1 assumed to be a dual porosity system
SPE 116688 9

of the analytic model. These flat portions are separated by sections of unit-slope. These flat portions could be interpreted as
two distinct radial flow periods separated by boundary dominated flow. The first flat portion would then represent radial flow
through a high permeability (fracture) system and the second would represent radial flow through a low permeability (matrix)
system. The first section of boundary dominated flow occurs when the boundary is seen through the high permeability
system. The second section occurs when a boundary is seen through the low permeability system. While the second radial-
flow portion of this dataset does not overlay the type curve, the trend is clearly seen. An equivalent, alternative interpretation
is that this data comes from a multilayered reservoir. However, information to confirm this interpretation could not be
obtained.
Table 5 shows a summary of the matching results for the Barnett Shale Example Well #1. If the permeability value
found is assumed to be the matrix permeability, the value found is much higher than would be expected for a shale gas.
However, as noted previously, the permeability returned by this type of evaluation is an effective system permeability. This
system is most likely a dual porosity system, and the well is hydraulically fractured. Therefore, it is reasonable to assume that
it would behave as if it had a larger effective system permeability. The drainage radius obtained is smaller than the initial
information about the system indicated. This could be due to the interference that was suggested to affect the performance of
this well. It could also be reality. A reasonable scenario in this situation is for the drainage area to represent the stimulated
region. Shale gas reservoirs have very low permeability; it is not unreasonable to assume that little or no production or
pressure support is coming from areas not directly connected to the hydraulic fracture system. This is somewhat supported by
the fact that the analysis indicates that the average reservoir pressure has dropped by a bit more than 50% of the initial value.
Changing the adsorption parameters does have an effect on the results of the analysis because the correction for
adsorption put forth by Bumb and McKee (1988) was used. However, to obtain a noticeable shift in the position of the data
on the type curve plot that could not be confused with natural data scatter, a value for vL that was 8 times larger than the
original value assumed was required. As can be seen in Table 5, the effect is to have a larger match term ( kG A ). This is
attributed to the smaller drainage area that is required to match the data since a larger value for vL shifts the data to the left on
the type curve plot. Adsorption effects the system compressibility in this analysis. With a larger system compressibility due
to adsorption, the value calculated for td will be smaller at any given point (recall the definition of td in Eqn. 8). To
compensate for this shift, either a smaller drainage radius or a smaller porosity would be required in the calculation of the
analytic solution and the dimensionless plotting functions for the production data. Note that in Table 5 the G/A values are
computed using only the free gas in place and not the adsorbed gas. In addition, consistent with our simulation analysis, the
permeability value in the kG A term is the effective permeability for the single porosity cases, but is likely the fracture
permeability for the dual porosity analyses.
Table 5 Summary of Matching Results for Barnett Shale Example Well #1
Model Drainage Porosity Effective Match Term
Perm. (nd) or G/A ( kG A ) *
Type radius, re (ft) ( ) or
Single Porosity
300 5% 8000 nd 3.55 0.57
(4000psi)
Single Porosity
300 5% 5000 nd 4.16 0.42
(5000psi)
Single Porosity
300 5% 3000 nd 4.66 0.28
(6000psi)
Lrg. Adsorption 200 5% 8000 nd 3.55 0.57
Dual Porosity 300 0.1 2.5 x 10-7 0.82 2.44
The varying of initial pressure does not have a large effect on the outcome of the matching process. While the match
term does decrease in value, a change in initial pressure does not significantly affect td because any increase in initial pressure
is offset by an corresponding increase in the estimated gas in place. Thus, an increase in the initial pressure results in a
positive vertical shift of the data on the type curve plot because it affects the dimensionless pressure term and not the
dimensionless time term. This vertical shift is best counteracted by decreasing the permeability or the reservoir thickness.
This is reflected in the changes to the match term kG A . Permeability is also in the td term, so there will be a slight
horizontal shift in the data as well. It is conceivable that a very large change in initial pressure could cause a significant
increase in the estimated permeability that would have to be offset by changing other match parameters. But, that was not the
case here. However, we want to stress the importance of knowing or estimating the initial pressure as accurately as possible.
Any uncertainty in this parameter effects every parameter returned from the analysis.
An important thing to note about the data from Table 5 is the true nature of the matching/output parameters obtained
from production data analysis. The match for the dual porosity system will return a value of , , and dimensionless radius
(rd) used to generate the analytical type curve. The production data will return a value for the match term kG A . If the
value of any specific parameter is desired, further assumptions will need to be made. If one wants fracture permeability from
, values for matrix permeability, wellbore radius, fracture spacing, and fracture geometry must be provided. These values
can in no way be determined from the match without additional information and/or assumptions.
10 SPE 116688

Field Example 2
Data from another Barnett Shale well encompassing 17.5 years of production were obtained from a database provided by
Devon Energy Corporation. Daily gas rate, oil rate, water rate, flowing tubing pressure, and casing pressure were recorded for
the first 4 months. After that, all rates were recorded daily, but flowing tubing and casing pressures were recorded weekly.
No initial estimates for reservoir parameters were available for this well so the initial set of reservoir parameters from the
previous example was used. Only the adsorption properties were changed to reflect estimated average values for the
reservoir. For this well, pL was assumed to be 1400 psia and vL was 89 scf/ton (6.8 scf/ft3). The production data obtained can
be seen in Figure 7. There appears to be 4 stimulations performed on this well starting at about day 4000. These are most
likely hydraulic re-fracture operations. However, only the first appears to have multiple points where both rate and pressure
were recorded.

Barnett Shale Example Well #2

10000
Rate (Vol./day), GLR (MSCF/BBL)

1000
or FTHP (psig)

100 Gas Rate

Water Rate
GLR
FTHP
10

0.1
0 1000 2000 3000 4000 5000 6000 7000
Time (days)

Figure 7 Combined data plot for Barnett Shale Example Well#2

Production data entries that did not have an associated tubing head pressure were removed. This resulted in 732 data
points that had complete data (i.e. rates and pressures). Next, bottomhole pressures were estimated using the Cullender and
Smith flow correlation (Brill and Beggs, 1988). Then, both the rate and pressure data were smoothed using the exponential
smoothing algorithm (Eqn. 9) with a value for of 0.15 for both rate and pressure smoothing. The smoothed rate and
pressure profiles can be seen in Figure 8.
2500 1000
Orig Pressure
Orig Rate
800
Gas Rate (mscf/day)

2000 Smooth Pressure

Smooth Rate
FBHP (psig)

1500 600

1000 400

500 200

0 0
0 1000 2000 3000 4000 5000 6000 7000 0 1000 2000 3000 4000 5000 6000 7000
time (days) time (days)

Figure 8 Original and smoothed rate and pressure values for Barnett Shale Example Well #2
SPE 116688 11

The smoothed rate and pressure profiles for the first 4050 days (up to the first stimulation job) were used in this
analysis with no further modification. The smoothed rate profile has 105% of the cumulative production of the original rate
profile. While this data is a bit noisy, the data appeared to line up quite well for a dual porosity analysis. This analysis is
shown in Figure 9. Again, three values for the initial pressure were evaluated. This time the values were 4000 psia, 5000 psia
and 7500 psia. Shown in Figure 9 is the match for the 4000 psia value.

Matching Barnett Shale Ex Well #2

(Dual Porosity)

1000

100

10
td*dln(pd)/dtd

Lam1E-7 Omg=0.05
Data
0.1 Lam 1E-7 Omg=0.085
Lam1E-8 Omg=0.05
Lam1E-8 Omg=0.1

0.01
1.00E+04 1.00E+05 1.00E+06 1.00E+07 1.00E+08 1.00E+09
tda

Figure 9 Match for Barnett Shale Example Well #2 assumed to be a dual porosity system
The match is of good quality especially at early-time. The smooth and distinct shape of the early-time data reduces
the uncertainty of this match. Interestingly, this match has the characteristics of a dual permeability system. The first flat
portion of the analytic solutions (solid lines) is a radial flow. Each of the analytic solutions also exhibits a second flat period
of radial flow just after a unit-slope that could be interpreted as boundary-dominated flow. This is followed by an extended
unit-slope indicating boundary-dominated flow. This could be interpreted as multiple layer behavior or as two different
porosities, matrix and fracture, in a dual porosity system. Finally, the late-time data does not quite overlay any of the type
curves.
The exercise to determine the sensitivity of the match to initial pressure was performed on the dual porosity cases
because the data exhibits dual porosity behavior. The dual porosity parameters ( and ) did not initially change with
pressure because initial pressure did not affect the shape of the data curve; it only affected its position on the type curve plot.
As initial pressure was increased, matrix porosity was decreased in an attempt to match the late-time portion of the analytic
type curve. Fracture porosity was also changed resulting in a similar value of . However, for the case of the 7500 psi initial
pressure, matrix porosity became small enough that a slightly larger was thought to be a better match. Figure 10 is the
match for Barnett Shale Example Well #2 using an initial pressure of 7500 psi. When performing log-log production data
analysis, it is very important to achieve a very good match at late-time. It was necessary to increase the initial pressure to
7500 psi which is a higher pressure than expected for this field. This suggested a smaller porosity needed to be used without
material balance indicating an unreasonably low reservoir pressure at the end of this production period. The reservoir
pressure at the end of this production period was estimated to be about 600 psi. Flowing bottomhole pressures at the end of
this production period are about 250 psi and gas production rates are about 100 MSCF/day.
The value of the adsorption parameters had to be increase dramatically to have any effect on the outcome of the
analysis. Increasing these adsorption parameters causes an overcorrection for adsorption using the adjusted total system
compressibility that results in the data being shifted to the left on the type curve plot. Therefore, porosity or drainage radius
must be changed to bring the data back. In this case, the drainage radius is very small and is very small; this makes the two
interconnected. Recall that represents the amount of connection or interaction between the natural fracture permeability and
the matrix permeability. With this being small, dual porosity effects last until boundary-dominated flow. Therefore, drainage
radius cannot be changed in this instance without changing and . It was chosen to lower porosity (therefore ) slightly to
bring the data back to its original position. This can be seen in the lower G/A for this scenario in Table 6.
12 SPE 116688

Matching Barnett Shale Ex Well #2

(Dual Porosity)

1000

100

10
td*dln(pd)/dtd

Lam1E-7 Omg=0.05
Data
0.1 Lam 1E-7 Omg=0.09
Lam1E-8 Omg=0.05
Lam1E-8 Omg=0.1

0.01
1.00E+04 1.00E+05 1.00E+06 1.00E+07 1.00E+08 1.00E+09
tda

Figure 10 Match for Barnett Shale Example Well #2 with an initial pressure of 7500 psi.

Table 6 shows a summary of the matching results for the Barnett Shale Example Well #2. The values for the G/A
and the kG A terms are assumed to be based on the free gas in the system. Using the total gas in the system the G/A values
change to 3.11, 2.98 and 2.65 for the 4000, 5000 and 7500 psia initial pressure values. The permeability values for the match
term are assumed to be based on the fracture permeability similar to the simulation study. A reasonable scenario in this
situation is for the drainage area to represent the region contacted by the hydraulic fracture. Shale gas reservoirs have very
low permeability; it is not unreasonable to assume that no little or no production or pressure support is coming from areas not
directly connected to the hydraulic fracture system. This very small drainage radius might very well be an effective drainage
volume which is connected by the hydraulic fractures.
Table 6 Summary of matching results for Barnett Shale Example Well #2
Dual Porosity Drainage Match Term
G/A ( kG A )
Models radius, re (ft)
Pi = 4000 psia 500 0.085 1 x 10-7 1.12 1.12
Pi = 5000 psia 500 0.085 1 x 10-7 1.07 1.0
Pi = 7500 psia 500 0.095 1 x 10-7 0.36 0.95
Lrg. Adsorption 350 0.07 1 x 10-7 0.97 3.28

Since this well had what appeared to be multiple stimulation jobs performed on it after day 4000, an analysis was
performed on the data from one of these operations. The same method of handling the production data that was used with the
first two field cases was used again here. Figure 11 shows the data following the first stimulation job that were used for this
analysis. This stimulation job was performed at around day 4050. All previous data was discarded, and only data from day
4050 up to the beginning of the next stimulation job was used. Since the previous analysis showed dual porosity behavior, it
was assumed that this data would also exibit dual porosity behavior. The analysis can be seen in Figure 12.
This data does not appear to have the same quality of a match as does the first set of data from this well. However, it
does display similar dual permeability/dual porosity behavior that was seen in Figure 9. There was really no discernable
boundary-dominated flow data for this stimulation job. So, it is really not possible to provide a definitive drainage radius.
However, drainage radius does slightly impact and if is small enough to cause dual-porosity effects to last until late-
time. Therefore, this match is somewhat indicative of drainage radius. The dimensionless radius obtained from this match is
less than what was obtained previously for this well. A summary of the results can be seen in Table 7.
SPE 116688 13

1000 1000
Rate (MSCF/day) FTHP (psig

Orig Rate Orig FTHP

800 800
Smooth Rate Smooth FTHP

FTHP (psig)
600 600

400 400

200 200

0 0
4000 4200 4400 4600 4800 4000 4200 4400 4600 4800
Time (days) Time (days)

Figure 11: Original and smoothed rate and pressure values for the stimulation job at day 4050

Matching Barnett Shale Ex Well#2 (Stimulation)

(Dual Porosity)

1000
Analytical Model
Data

100
pwd'

10

0.1
100 1000 10000 100000 1000000 10000000 100000000
td or tda_new

Figure 12 Match for the first stimulation job performed on Barnett Shale Example Well #2

Table 7 Summary of matching results for the first stimulation job performed on Barnett Shale Example Well #2
Model Drainage Match Term
G/A ( kG A )
Type radius, re (ft)
Dual Porosity 225 0.075 9 10-8 3.52 2.49

This match was achieved with the roughly the same dual porosity parameters and the same initial pressure as was
used on the first section of data from this well. The small drainage radius is the most notable difference between the results of
this match and the previous match. However, if one assumes that this well was hydraulically fractured before and then re-
fractured at day 4050, this can make sense. Let us assume that the only producing portions of the reservoir are those directly
connected to the hydraulic fracture by the natural fracture matrix. This assumption is reasonable if one also assumes that any
hydraulic fracture in the shale reservoir merely props open the natural fracture matrix (Gale, et al. 2007). This area that is
connected to the hydraulic fracture is very small and is drained rapidly. This is evident when matching the first data section
of this well; the average reservoir pressure estimated at for the last data point was 700 psi. This is more than a 3000 psi drop
14 SPE 116688

in reservoir pressure from the initial reservoir pressure of 4000 psi. As production declines, it becomes necessary to contact
new reservoir. Another fracture operation is performed and the well resumes producing at its original high rate. According to
the analysis here, reservoir pressures seen after the fracturing are at or near original reservoir pressure. Each time the well is
fractured additional, apparently new reservoir area is contacted. This can be seen by comparing the initial rates of the all of
the stimulation jobs that can be seen in Figure 7 and the fact that both the first data section and the stimulated data section can
be matched using the same initial pressure and the same dual porosity parameters.
Conclusions
In this work we reiterate the framework for analyzing production data from shale gas systems and provide a slightly modified
version of material balance time to account for adsorbed gas. We applied the method of Cox, et al. (2002) to develop
solutions to simulated and field case histories for shale gas wells. For the simulated cases, completion effects were seen to be
problematic to evaluate due to the lack of data at very early times for the drawdown value simulated. For the field cases,
better resolution of the completion effects were observed, but the many parameter values inherent in the models make the
analysis both difficult to interpret and corroborate results. This work has shown that expected flow regimes used to constrain
many of the model parameters do appear and parameters estimated seem reasonable (or at least there appear to be
explanations for unusual or unexpected results). High frequency, highly accurate rate and pressure data along with detailed
petrophysical and seismic measurements provide some hope that parameter values can be further constrained and/or
determined.
Acknowlegements
The authors would like to thank Devon Enegy Corp. for access to the data for this work and especially to Jerry Youngblood.
We would also like to thank Richard Sigal, Chandra Rai, Carl Sondergeld and Roger Slatt at the University of Oklahoma and
Tom Blasingame at Texas A&M for fruitful discussions on this topic.
NOMENCLATURE
A area
B formation volume factor
c compressibility
Cdf dimesionless fracture conductivity
D decline rate
G free gas in place
Gtot free gas + adsorbed gas
h height
k permeability
m(p) pseudopressure
p pressure
q flow rate
r radius
S saturation
T temperature
t time
ta material balance time
tdA dimensionless material balance time based on drainage area
tdA_new dimensionless material balance time based on drainage area with ct modification
tdxf dimensionless time base on xf
vads adsorbed gas concentration
w fracture width
x half-length
z real gas deviation factor
damping coefficient parameter
gas correction factor
viscosity
density
porosity
storativity ratio
Subscripts
d dimensionless
e external or drainage
f fracture
f+m matrix + fracture
fh hydraulic fracture
SPE 116688 15

g gas
i initial
L Langmuir
m matrix
o oil
p produced
sc standard conditions
t total
w water
wd dimensionless wellbore flowing
wf wellbore flowing
REFERENCES
1. Agarwal, Ram G., David C. Gardner, Stanley W. Kleinsteiber, and Del. D Fussell: Analyzing Well Production
Data Using Combined-Type-Curve and Decline-Curve Analysis Concepts, SPE 57916 presented at the SPE Annual
Technical Conference and Exhibition, New Orleans, September 27-30, 1999.
2. Aguilera, R.: Well Test Analysis of Dual-Porosity Systems Intercpeted by Hydraulic Vertical Fractures of Finite
Conductivity, SPE 18948 presented at the SPE Rocky Mountain Regional/Low-Permeability Reservoirs
Symposium, Denver, March 6-8, 1989.
3. Brill, James P. and H. Dale Beggs: Two-Phase Flow in Pipes, 6th Ed, (1988).
4. Bumb, A. C. and C. R. McKee: Gas Well Testing in the Presence of Desorption for Coalbed Methane and
Devonian Shale, SPE Formation Evaluation, Vol. 3, No. 1 (1988, March): 179-185.
5. Clarkson, C.R, C.L. Jordan, R.R. Gierhart, and J.P. Seidle: Production Data Analysis of Coalbed-Methane Wells,
SPE Reservoir and Evaluation Engineering, April, 2008: 311-325.
6. Cox, Stuart A., John V. Gilbert, Robert P. Sutton, and Ronald P. Stoltz: Reserve Analysis for Tight Gas, SPE
78695 presented at the SPE Eastern Regional Meeting, Lexington, October 23-25, 2002.
7. Ehlig-Economides, Christine A. and H.J. Ramey, Jr.: Transient Rate Decline Analysis for Wells Produced at
Constant Pressure, Society of Petroleum Engineers Journal, February, 1981.
8. Fetkovich, M. J.: Decline Curve Analysis Using Type Curves, Journal of Petroleum Technology, Vol. 32, No. 6
(1980, June): 1065-77.
9. Fraim, M. L. and R. A. Wattenbarger: Gas Reservoir Decline-Curve Analysis Using Type Curves with Real Gas
Pseudopressure and Normalized Time, SPE Formation Evaluation, Vol. 2, No. 6 (1987, December): 671-682.
10. Gale, Julia F. W., Robert M. Reed, and Jon Holder: Natural Fractures in the Barnett Shale and their Importance for
Hydraulic Fracture Treatments, AAPG Bulletin, Vol. 91, No. 4 (2007, April): 603-622.
11. Hager, C.J. and J.R. Jones: Analyzing Flowing Production Data with Standard Pressure Transient Methods:, SPE
71033 presented at the SPE Rocky Mountain Petroleum Technology Conference, Keystone, CO, May 21-23, 2001.
12. Hazlett, W.G., W. J. Lee, G. M. Nahara, and J. M. Gatens III: Production Data Analysis Type Curves for the
Devonian Shales, SPE 15934 presented at the SPE Eastern Regional Meeting, Columbus, November 12-14, 1986.
13. Hill, David G., John B. Curtis, and Paul G. Lillis: An Overview of Shale Gas Exploration and Current Plays,
Proceedings of the 26th Annual International Conference on Offshore Mechanics and Arctic Engineering, San Diego,
June 10-15, 2007.
14. Lee, W. J. and R. A. Wattenbarger: Gas Reservoir Engineering. SPE Textbook Series Vol. 5 (1996).
15. Lewis, A.M.: Production Data Analysis of Shale Gas Reservoirs, Louisiana State University MS Thesis, December,
2007.
16. Montgomery, Scott L., Daniel M. Jarvie, Kent A. Bowker, and Richard M. Pollastro: Mississippian Barnett Shale,
Fort Worth Basin, North-Central Texas: Gas-Shale Play with Multi-Trillion Cubic Foot Potential, AAPG Bulletin,
Vol. 89, No. 2 (2005, February): 155-175.
17. Olarewaju, J. S. and W. J. Lee: New Pressure-Transient Analysis Model for Dual-Porosity Reservoirs, SPE
Formation Evaluation, Vol. 38, No. 3 (1989, September): 384-390.
18. Palacio, J. C. and T. A. Blasingame: Decline-Curve Analysis Using Type Curves-Analysis of Gas Well Production
Data, SPE 25909 presented at the SPE Joint Rocky Mountain Regional Low Permeability Reservoirs Symposium,
Denver, April 26-28, 1993.
19. Rodriguez, L.C.: Early Prediction of Reserves in Tight Gas Reservoirs, University of Oklahoma MS Thesis,
October, 2005.
20. Schenk, Christopher J.: Geologic Definition and Resource Assessment of Continuous (Unconventional) Gas
Accumulations - the U.S. Experience, AAPG 66086 presented at the Ancient Oil-New Energy Conference, Cairo,
October 27-30, 2002.
21. Spivey, J. P. and M. E. Semmelbeck: Forecasting Long-Term Gas Production of Dewatered Coal Seams and
Fractured Gas Shales, SPE 29580 presented at the SPE Rocky Mountain Regional/Low-Permeability Reservoirs
Symposium, Denver, March 20-22, 1995.