SPE 165711

Long Term Forecasting in Shale Reservoirs Using Rate Transient Analysis

D. Gunaydin, SPE, Z. He, SPE, K.G. Brown, SPE, M.A. Deasy, SPE, M.D. Porter, SPE, D.J. Corinchock, SPE,
Range Resources – Appalachia, LLC

Copyright 2013, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Eastern Regional Meeting held in Pittsburgh, Pennsylvania, USA, 20–22 August 2013.

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
Shale reservoirs have become one of the most promising natural gas resources in the energy industry. However the complex
nature of shale, combined with limited production history, make predicting recovery very difficult in these reservoirs,
especially early in the life of the well.

The hyperbolic form of the Arps equation has been modified for use in estimating reserves and forecasting production in
shale reservoirs. Although the ease and accessibility of this method make it convenient for most applications, the Arps
equation has limitations (assumes boundary-dominated flow, and assumes that flowing bottomhole pressure, drainage area,
permeability and skin factor are all constant). By analyzing production data using Rate Transient Analysis (RTA) theory,
analytic models can be developed and used to forecast production and recovery in shale gas reservoirs. Because RTA-based
models are not subject to many of the assumptions imposed in the development of the Arps equation, these models are much
more versatile than decline curves derived using the Arps modified hyperbolic equation.

In this paper, we present a practical workflow for performing RTA to determine the key performance parameters of
horizontal shale wells with multiple fractures. This methodology provides a deterministic approach to estimate long term well
performance in a shale reservoir. One important advantage over using the Arps equation is the ability to forecast production
under different operating strategies. Using this workflow, the production and economic impacts resulting from different
completion designs and operational scenarios (e.g., delayed installation of compression) can be studied.

This methodology was implemented on data sets from over 150 Marcellus Shale wells. Forecasts determined using this
workflow compare favorably to the forecasts estimated by decline curve analysis and reservoir simulation. Field examples
from Marcellus wells are included in this paper to demonstrate the workflow and results.

Introduction
The development of unconventional shale gas in North America in recent years has extensively altered the energy supply
picture for the region. The trend in US natural gas production has been rapidly increasing over the past decade due to shale
gas reservoirs. The Marcellus Shale, one of several shale reservoirs in the U.S., is the most expansive shale gas play
extending through five states in the northeastern U.S. Each shale play has its own reservoir characteristics and unique set of
recovery mechanisms. In this paper, we will focus on Marcellus Shale.

Marcellus Shale
Complex geology and fluid characteristics make the Marcellus an unconventional reservoir. One of the most important traits
of the Marcellus is its ultra-low matrix permeability which prevents significant fluid flow into an unstimulated wellbore. The
permeability of the Marcellus Shale is at least one million times lower than conventional reservoirs. Consequently, horizontal
drilling and multi-stage hydraulic fracturing are required for economic development of this play. In addition to free gas stored
2 SPE 165711

in natural fractures and intergranular porosity, gas can also be stored by sorption on organic matter (Boyer et al., 2006).

Effective reservoir management plans in the Marcellus require an understanding of rock and fluid properties, rock-fluid
interactions and the physics of fluid flow in shales. Broadly speaking, our reservoir management tool box consists of three
types of models; empirical models (decline curves), numerical reservoir simulation models, and analytical reservoir models
based rate transient analysis (RTA) theory. Although this paper’s main focus is analytic RTA models, we will briefly
summarize the empirical and numerical models to provide context.

Empirical Models
To create the empirical models, we project the future based on a match and extrapolation of the existing production data
using the original decline curve model (Arps, 1945). In Arp’s empirical method, established trends of historical production
declines are used to forecast the future well performance using the following general equation.

⁄ (1)

Which, for 0<b<1 is also called the hyperbolic decline equation. When b is equal to 0, the exponential equation is used:

(2)

In equation 2, α is a constant and is equal to the initial decline rate, Di. Lee and Sidle (2010) noted that in tight gas and shale
wells, the best fit of the historical data yields b-values greater than one, which is beyond the limit that Arps specified. Since
b-values > 1.0 can result in unreasonable estimates of long-term production, multiple segmentations of hyperbolic equation
lines are often used in shale well decline curves. Moreover, Rushing et al. (2007) noted that the best fit value of b starts to
decrease as more production data becomes available in shale wells. Early-time b-values >1.0 for the first segment of the
decline curve are the result of the well being in the transient flow regime (Fetkovich, 1987). For shale gas wells, transient
linear flow is predominant, and it is commom to observe b-factors that exceed 1.

An exponential decline can be used if the production data exhibits stabilized conditions (i.e. if boundary dominated flow
(BDF) is observed). However, transient linear flow may persist for decades before BDF is established. Therefore, when
forecasting EUR’s, a constant minimum decline rate is often used and an exponential decline is assumed for the third
segment of the decline curve representing the remaining well life. The constant minimum decline rate used in the third
segment of the decline curve is based on engineering judgement and experience from other older shale reservoirs producing
for decades from vertical wells. Figure 1 shows a typical multi-segmented decline curve for a Marcellus Shale well.

Numerical Models
Numerical models are advanced tools that quantify the complex flow mechanisms in unconventional reservoirs. They are
superior at explaining the physics of fluid flow when compared to other simplified methodologies, and they have the ability
to model heterogeneity, multi phase flow conditions, stress-dependent permeability, fracture-dependent conductivity and
changes in compositions that occur during the life of the reservoir. However, the ability to model such micro-scale details in
numerical simulation comes with a price; it requires considerable amounts of experienced manpower, time and computer
power. An alternative approach for managing daily operational decisions is the analytical models, which are simplified
versions of numerical models.

Analytical Models
Typical analytical models are mathematical solutions of the diffusivity equation, and are used to define the physics of fluid
flow through the reservoir for certain idealized situations. Unlike the numerical models, analytical models are not divided to
numerous grid blocks, which vastly simplify data input.

Anderson et al. (2010) proposed a shale gas model which is composed of a horizontal well with multiple fractures. They
extended the bounded linear flow model with a single fracture introduced by Wattenbarger et al. (1998). Anderson’s shale
gas model is a horizontal wellbore with multiple hydraulically-induced parallel planar fractures crossing the wellbore at a 90°
SPE 165711 3

angle (Figure 2). In this model, the stimulated reservoir volume (SRV) contains the hydraulic fracture network and the space
between the hydraulic fractures. An unstimulated shale reservoir surrounds the SRV. Based on the work of Bello and
Wattenbarger (2008) and Brown et al. (2009), Anderson and Liang (2011) identified six flow regimes which may be
manifested in the shale gas model. The dominant flow regime seen on these wells is the linear flow from SRV to fractures
through the cross sectional area (Ac) which is the area between the matrix and fracture surface (Figure 2). Although other
flow regimes (transitional flow, unstimulated matrix to SRV flow and boundary dominated flow (BDF)) are expected after
transient linear flow, analysis of more than 250 Marcellus wells in SW PA have not shown any evidence of BDF to date,
some after 4 years of production.

Historical data can be matched with the shale gas model’s results. Once the unknown reservoir and well properties are
determined by history matching, the calibrated model can be used to predict the future production performance. Some model
inputs (e.g., reservoir temperature, thickness and porosity) can be obtained with reliable measurements and/or log
interpretation. However, other important input parameters, such as fracture half-length (xf), must also be determined in order
to characterize long-term well performance. In order to reduce the level of non-uniqueness of the matched parameters and
increase the effectiveness of the model matching process, we analyze the production data using the shale gas analytical
model.

Production Data Analysis – Rate Transient Analysis

Anderson et al. (2010) introduced the rate transient analysis (RTA) methodology for shale gas wells based on the work of
Wattenbarger et al (1998). RTA focuses on estimating the major well/reservoir performance parameters by analyzing the
dominant flow regime, transient linear flow, on the well’s dynamic data (i.e. gas rates and flowing pressures). The equation
describing the transient linear flow of a horizontal well with multiple fractures is given below.

.
√ ∅
√ (3)

In the above equation, T is the reservoir temperature, is the number of fractures, h is the net pay thickness, is the
fracture half length, k is the permeability, ∅ is the porosity, is the gas viscosity, is the total compressibility, and m is the
slope of the square root of time plot. When there is condensate production, we calculate the recombined gravity and use it to
convert condensate volumes to gas volume, since all the calculations are being done at bottomhole conditions. This equation
has the form of y = mx + b, where the y-values are normalized delta pseudo pressure (calculated from the initial reservoir
pressure (Pi), the flowing bottomhole pressure (Pwf) and the gas rate (q)), and the x-values are √ . The intercept b’ is used to
estimate dimensionless fracture conductivity (FCD’).

As suggested by the form of this equation, during transient linear flow, a plot of normalized delta pseudo pressure versus
square root of time plot appears as straight line having slope m, and intercept b’. Using the slope of this straight line (m), the
product of fracture half length and square root of permeability ( √ ) can be estimated as follows.

.
√ (4)
∅

The area of stimulated rock volume (ASRV) encompasses the portion the reservoir containing the system of hydraulic
fractures (Figure 2), and is mathematically equal to Le * 2xf. This area can be calculated directly, but only if boundary
dominated flow within the SRV is observed on the square root of time plot. Figure 3 shows a theoretical example of a square
root of time plot where telf (the time where the data starts to deviate from the linear transient straight line), is clearly seen.
ASRV can be calculated by using the observed telf value in Equation 5 below:

. √
(5)
∅
4 SPE 165711

Knowing that is mathematically equal to Le*2xf, we can now calculate a fracture half length using the ASRV calculated
from Equation 5 in Equation 6 below:

(6)

Where is the effective length of the horizontal wellbore. Finally, using the √ product calculated from equation 4, the
permeability of the SRV system can be calculated using Equation 7 below:

√
(7)

For reservoirs with high permeability values, the time to reach boundary dominated flow is relatively short. However, in
unconventional gas reservoirs, this is often not the case. For example, in most Marcellus Shale wells, boundary dominated
flow has not manifested itself - even after 4 years of production. If boundary dominated flow is not manifested in the well
performance data, there will be uncertainty in the determination of ASRV which will make the deterministic estimation of
fracture half length impossible. In this case, multiple realizations could yield similar history matches. In this paper, we
present a practical iterative workflow for RTA that allows us to determine ASRV with less uncertainty when the effects of
boundary dominated flow is not observed on the production data.

Methodology
To develop an integrated workflow, we first identify the parameters affecting the reservoir performance. These parameters
include the inputs of the linear flow equation (Equation 3) and results of the rate transient analysis. Parameters derived from
direct measurements such as logs or core analyses are considered known parameters for the sake of this discussion, and the
remaining parameters are considered unknown (Table 1).

Of the four unknown parameters (ASRV, nf, xf, and FCD’), fracture half length and dimensionless conductivity can be
calculated once the ASRV is estimated. Since there are two unknown parameters remaining, ASRV and nf, and there is only one
equation, we use an iterative process to solve for the unknown parameters. Table 2 shows the assumed range of these two
parameters. A lower limit for ASRV can be calculated by assuming that the time associated with the last data point represents
the end of linear transient flow time in Equation 5 (i.e., assumes boundary dominated flow begins immediately after this last
data point). The existing well spacing represents an upper limit for ASRV. For number of fractures, the lower limit is estimated
to be the total number of hydraulic fracture stages. Since there might be multiple perforation clusters within each stage, the
upper boundary of nf is determined as the total number of perforation clusters.

We obtain the matrix/organic permeability value from a three dimensional gridded geologic model generated using
PETRELTM. Seismic data, petro physical log evaluation and core analysis are used to build the geologic model. Our
Geology Department provides essential rock and fluid properties for each perforation cluster of each well using the Petrel
Model. They use measured core permeabilities to calibrate their log analysis and then they map this permeability in Petrel.
Several factors impact the accuracy of these permeability measurements, including lab testing under ambient conditions,
water loss, and condensate drop out. The actual in-situ permeability is believed to be smaller than permeability measured in
the lab. Because our numerical reservoir simulation in the Marcellus suggests that estimated core permeabilities are about
twice the permeability derived from history matching, we divided core permability values by two to maintain consistency.
These aspects of our project are considered routine, and are not described in detail in this paper.

We have identifed five performance drivers influencing flow in a horizontal well with multiple fractures in a shale reservoir.
These are the permeability of the system, area of the SRV, fracture half length, number of fractures and dimensionless
fracture conductivity. Of these five parameters, we believe the permeability can be most accurately and consistently
estimated based on the PETRELTM geological model. Therefore, we seek to converge on our permeability estimate during
the iterative portion of our workflow.
SPE 165711 5

Figure 4 summarizes our proposed workflow, and demonstrates the iterative process of quantifying values of ASRV and nf, for
use in the shale RTA model. The workflow starts with traditional rate transient analysis by plotting normalized delta pseudo
pressures versus their corresponding time values in a square root of superposition time function format. Although rigorous
analysis requires the use of superposition time to account for the non-constant rate profiles typical in the producing life of a
well, superposition time plots are used in conjunction with square root of time plots in our analysis process. Superposition
time plots are used because they typically yield a more accurate slope for use in Equation 3. However, identification of flow
regimes is always done using square root of time plots, because this plot has less bias than the superposition time plots ( i.e.
BDF is more clearly identified on the square root of time plot whereas the superposition time function tends to over-linearize
the data trends). Once the straight line is established and its slope is estimated, √ can be determined using equation 4.

The iterative process to solve for √ starts with the maximum ASRV and minimum nf. Based on these inputs, fracture
half length and permeability are calculated using equation 6 and equation 7. If the calculated permeability is the same as the
permeability obtained from direct measurement and the geologic model, the iterative process is complete. If the calculated
permeability is less than measured permeability, ASRV is reduced until there is a match between two permeability parameters.
If the calculated permeability is higher than the measured permeability, the number of fractures is increased until a match is
achieved.

Once a match is obtained, finalized realizations are used to calculate the new fracture half length and dimensionless
conductivity. Then an analytical model is built and calibrated by history matching the well’s production and pressure profiles.
Finally the calibrated model is used to forecast production, study different operation scenarios, and make daily operation
decisions.

This workflow integrates all available information to characterize the reservoir-well system, and has been applied to more
than 150 Marcellus Shale wells in a consistent manner. It represents a demonstrably effective workflow to constrain the
history-matching problem and reduce its associated non-uniqueness issue.

Case Study 1
To illustrate the iterative workflow, we present a case study on a Marcellus horizontal well with multiple hydraulic fractures.
This well is located in southwest PA and produces, gas, condensate and water. The stabilized condensate gas ratio of 38
bbl/MMscf allowed us to build an analytical model based on RTA. As seen in Figure 5, boundary dominated flow is not
evident during the 20 months the well has been producing, based on the lack of deviation from the transient linear line on the
square root time plot. To obtain a match between the calculated permeability and measured permeability during the iteration
process, ASRV was reduced by 4.2% from its maximum value, and the number of fractures was increased by one. After
achieving a satisfactory permeability match, we constructed the analytical model using these new values of ASRV, nf and xf.

Multi phase flow in the wellbore, well cleanup and changing operational conditions can introduce some error in calculated
bottomhole pressures. Consequently, there may be some inaccuracy in the calculated FCD’ values. Also, the effective
system permeability inside the SRV may be enhanced by the secondary fractures induced from the fracture treatment.
Therefore, the effective permeability inside the SRV and FCD’ are varied during the history matching process in order to
obtain the best possible history match between the model-forecasted production profile and the actual production history.

Figure 6 shows the history match, which was validated using 60 days of production data. The history matching results were
then used to forecast the production for 50 years. A comparison of the cumulative productions forecasted by the analytical
model, the numerical model and the empirical model is summarized in Table 3. The observed differences fall within a range
of 1.4-2.7%.

Case Study 2
The second case study represents a dry gas horizontal well in Marcellus, and is presented here to illustrate the value of RTA
in a well having only rate-constrained data available for analysis. This well could not be analyzed using decline curves early
in its life because it was being produced under constrained operational conditions. Our objective was to predict how long this
well could maintain the constrained flowrate early in the life of the well. An analytical model was built, based on the
6 SPE 165711

workflow presented in this paper.

In this case study, a constant rate solution is used to calculate the input parameters of the model, since the analysis is done by
using only the early time (constrained) data. Figure 7 shows the square root of time plot for this case. Despite flowing under
constrained conditions, a well-defined slope is evidenced on the plot, indicating transient linear flow occurs throughout the
flowing period. Figure 8 shows the performance comparison between the analytical model’s forecast and rate history. The
model predicts the start of the decline within three days of its actual occurance. Several subsequent comparisons of actual
production vs model-predicted production yielded good agreement.

Conclusions
 Although modeling multi stage fractured horizontal wells in Marcellus is very challenging due to the complex
geology and complex fluids, empirical, numerical and analytical models can be useful in the evaluation of reservoir
performance in shale reservoirs.
 RTA models help us understand the physics of fluid flow in unconventional shale reservoirs in a way that type
curves cannot, and are therefore useful for imporved reservoir management in shale plays.
 RTA can be used to estimate the key parameters that influence well behavior in the Marcellus, and RTA results can
be use to build history-matched analytical models.
 If BDF is observed, there is less uncertainty in RTA output, but where BDF is not observed; deterministic
estimations of ASRV have more uncertainty.
 This paper presents a practical iterative RTA workflow that can help reduce the uncertainty in estimated values of
ASRV when BDF is not observed.
 Analytic models based on the RTA workflow proposed in this paper are useful, and they can accurately predict well
performance, even when all available data is collected under constrained operational conditions.
 Because analytic models based on the RTA workflow proposed in this paper can accurately predict well
performance, they are useful for assessing economic impacts associated with operational changes.
 Although analytic models are not as rigorous as numerical models and may have more limitations than numerical
models, they can be very useful and should be part of an integrated reservoir analysis approach.
SPE 165711 7

Nomenclature
Ac = Cross sectional Area, ft2
ASRV = Area of stimulated reservoir volume, acres
B = Arps’ decline exponent, dimensionless
ct = Total compressibility, 1/psi
Di = Initial decline rate, 1/d
h = Net pay thickness, ft
k = Permeability, md
Le = Effective horizontal wellbore length, ft
m = Slope of straight line on square root time plot
nf = Number of fractures
m(Pi) = Initial pseudo reservoir pressure, psia2/cp
m(Pwf) =Pseudo Bottomhole flowing pressure, psia2/cp
Q = Gas production rate, Mscf/d
qi = Initial gas production rate, Mscf/d
t = Time, d
telf = End of transient linear flow time, days
T = Reservoir temperature, oR
xf = Fracture half length, ft

Greek Symbols

∅ = Porosity, fraction
= Gas viscosity, cp

Abbreviations

BDF = Boundary Dominated Flow

i = Initial Conditions
RTA = Rate Transient Analysis
SRV = Stimulated Reservoir Volume

References

Anderson, D.M. and Liang, P. 2011. Quantifying Uncertainty in Rate Transient Analysis for Unconventional Gas Reservoirs.
Paper SPE 145088 presented at the SPE North American Unconventional Gas Conference and Exhibition,. Woodlands,
TX, 14-16 June
Anderson, D.M., Nobakht, M., Moghadam S., and Mattar, L. 2010. Analysis of Production Data from Fractured Shale Gas
Wells. Paper SPE 131787 presented at the SPE Unconventional Gas Conference, Pittsburgh, PA, 23-25 February
Arps, J.J. 1945. Analysis of Decline Curves. Trans. AIME, 160,228-247
Bello, R.O., and Wattenbarger, R.A. 2008. Rate Transient Analysis in Naturally Fractured Shale Gas Reservoirs. Paper SPE
114591 presented at the CIPC/SPE Gas Technology Symposium, Calgery, Alberta, 16-19 June
Boyer, C., Kieschnick, J., Suarez-Rivera R., Lewis, R. E., and Waters, G. 2006. Producing Gas from Its Sources, Oilfield
Review.
Brown, M., Ozkan E., Raghavan R. and Kazemi H. 2009. Practical Solutions for Pressure Transient Responses of Fractured
Horizontal Wells in Unconventional Reservoirs. Paper SPE 125043 presented at the SPE Annual Technical Conference
and Exhibition, New Orleans, LA, 4-7 October
Fetkovich, M.J., Vienot, M.E., Bradley, M.D. and Kiesow, U.G. 1987. Decline Curve Analysis Using Type Curves - Case
Histories, SPE Formation Evaluation. SPE 13169-PA
Lee, W.J. and Siddle, R. 2010. Gas Reserves Estimation in Resource Plays. SPE Economics & Management (October), 86-91
Rushing, J.A., Perego, A.D., Sullivan, R.B., and Blasingame, T.A. 2007. Estimating Reserves in Tight Gas Sands at HP/HT
Reservoir Conditions: Use and Misuse of an Arps Decline Curve Methodology. Paper SPE 109625 presented at the
Annual SPE Technical Conference and Exhibition, Anaheim, CA, 11-14 November
Wattenbarger, R.A., El-Banbi, A.H., Villegas, M.E., and Maggard, J.B. 1998. Production Analysis of Linear Flow into
Fractured Tight Gas Wells. Paper SPE 39931 presented at SPE Rock Mountain Regional/Low Permeability Reservoirs
Symposium and Exhibition. Denver, CO, 5-8 April
 8 SPE 165711

Table 1: Parameters Affecting the Reservoir Performance

Known Unknown
Initial Pressure (Pi) Area of SRV (ASRV)
Reservoir Temperature (TR) Number of Fractures (nf)
Net Thickness (h) Fracture Half Length (xf)
Porosity (Φ) Dimensionless Conductivity (FCD’)
Reservoir Fluid Properties
Completion Design
Matrix/Organic System Permeability (k)

Table 2: Unknown Parameters

Min Max
ASRV Estimated by using the last data point on √t Plot Spacing
nf Number of Hydraulic Fracturing Stages Number of Total Perforation Clusters

Table 3: Comparison of Case 1 Results

Model Gas Cum
Analytic Model QRTA
Numeric Model 1.014 x QRTA
Empirical Model 1.028 x QRTA
SPE 165711 9

Figure 1: Typical Multi-Segmented Decline Curve for a Shale Well

Figure
Figure 2: 3-D and Map Views of Shale Gas Model
10 SPE 165711

Figure 3: Example of Square Root Time Plot

Figure 4: Workflow of the Methodology

SPE 165711 11

Figure 5: Square Root Time Plot for Case 1.

Figure 6: History Match of Analytic Model. Case 1.

12 SPE 165711

Figure 7: Square Root of Time plot for Case 2.

Figure 8: Model Calibration and Verification of Case 2.