Comparison Study of Hydraulic

Fracturing Models-Test Case:

GRI Staged Field Experiment No. 3
N.R. Warpinski, SPE, Sandia Natl. Labs; Z.A. Moschovidis, SPE, Amoco Production Co.;
C.D. Parker, SPE, Conoco Inc.; and 1.5. Abou-Sayed, Mobil E&P Technical Center

Summary. This study is a comparison of hydraulic fracture models run using test data from the GRI Staged Field Experiment No.
3. Models compared include 2D, pseudo-3D, and 3D codes, run on up to eight different cases. Documented in this comparison are
the differences in length, height, width, pressure, and efficiency. The purpose of this study is to provide the completions engineer
with a practical comparison of the available models so that rational decisions can be made as to which model is optimal for a given
application.
Introduction
Hydraulic fracturing, one of the most important stimulation techniques available to the petroleum engineer, is being used extensively in tight gas sandstones, 1-5 coalbed methane,6 high-permeability sandstones in Alaska,7 very weak sandstones off the U.S.
gulf coast, 8 horizontal wells in chalks, 9.10 and many other applications from waste disposal to geothermal reservoirs. Because of
this diversity of application, hydraulic fracture design models must
be able to account for widely varying rock properties, reservoir
properties, in-situ stresses, fracturing fluids, and proppant loads.
As a result, fracture simulation has emerged as a highly complex
endeavor that must be able to account for many different physical
processes.
The petroleum engineer who must design the fracture treatment
is often confronted with the difficult task of selecting a suitable hydraulic fracture model, yet there is very little comparative information available to help in making a rational choice, particularly
on the newer 3D and pseudo-3D models. The purpose of this paper
is to help provide some guidance by comparing many of the available simulators.
The Fracture Propagation Modeling Forum held Feb. 26-27,
1991, near Houston provided the origin for this paper. This forum,
sponsored by the Gas Research Inst. (GRI), was open to all known
hydraulic fracturing modelers. Participants were asked to provide
fracture designs based on the Staged Field Experiment (SFE) No.
3 fracture experiment. After the fracture designs presented at this
meeting were compared, a final, revised data set was given to all
participants. The results presented in this paper are derived from
that data set. To publish the results, a four-member committee (the
authors) was chosen from forum participants. In assembling this
comparison, committee members purposely attempted to avoid judging the relative values of the different models. Only the results and
quantifiable comparisons are given.
Background-Basic Modeling Discussion
In recent years, fracturing simulators used in the oil industry have
proliferated. This proliferation was intensified by the availability
of personal computers and the need for fast design simulators for
use in the field. Applying these models as "black boxes," without
knowledge of the underlying assumptions, may lead to erroneous
conclusions, especially for unconfined fracture growth.
Hydraulic fracturing is a complex nonlinear mathematical problem that involves the mechanical interaction of the propagating fracture with the injected slurry. Several assumptions are commonly
made to render the problem tractable: plane fractures, symmetric
with respect to the wellbore; elastic formation; linear fracture
mechanics for fracture propagation prediction; power-law behavior
of fracturing fluids and slurries; simplification of fracture geometry and its representation by few geometric parameters; etc. Ref.
11 gives a detailed description of the governing equations. Although
the models predict' 'trends" of treating pressure behavior, they may
Copyright 1994 Society of Petroleum Engineers

SPE Production & Facilities. February 1994

not always reliably predict the observed behavior for a given treatment. This discrepancy has been attributed to many complex interactions between the injected fluids and the formation that are
not well understood.
An attempt to characterize phenomenologically some of these
complex processes occurring within the fracture (e.g., mUltiple fractures and increased frictional losses) and near the fracture tip (e.g.,
nonlinear formation behavior, microcracking, formation plasticity, dilatancy, and plugging) was made in various simulators by the
introduction of additional ad hoc parameters ("knobs"). The choice
of values for these parameters is based only on the modeler's experience. These knobs, used to match model predictions with fieldobserved behavior, result in the lack of a standard model response
for a given physical problem. This issue was addressed in the forum by having different participants (discussing several different
models) simulate common test cases derived from the actual SFE
No.3 fracturing treatment. These models can be categorized in order
of decreasing complexity as follows.
1. Planar 3D models: TerraFrac of TerraTek Inc. 12-16 run by
Arco and HYFRAC3D by S.H. Advani of Lehigh U.17
2. GOHFER, a unique finite-difference simulator by Marathon
Oil Co. 18.19
3. Planar pseudo-3D models.
A. "Cell" approach: STIMPLAN of NSI Inc., ENERFRAC
of Shell,20,21 and TRIFRAC of S.A. Holditch & Assocs. Inc.
B. Overall fracture geometry parameterization: FRACPRO of
Reservoir Engineering Systems (RES) Inc. 2225 and MFRAC-II of
Meyer & Assocs. 26 -29
4. Classic Perkins-Kern-Nordgren (PKN) and Geertsma-deKlerk
(GDK) models 30-35 : PROP of Halliburton, 3436 the Chevron 2D
model, the Conoco 2D model, the She1l2D model, and pseudo-3D
models run in constant-height mode.
A discussion of the basics of these models is given to provide
some insights on the model assumptions and their expected effect
on results.
Planar 3D Models. The TerraFrac 12-16 and the HYFRAC3D 17
models incorporate similar assumptions and formulate the physics
rigorously, assuming planar fractures of arbitrary shape in a linearly
elastic formation, 2D flow in the fracture, power-law fluids, and
linear fracture mechanics for fracture propagation. Their difference is in the numerical technique used to calculate fracture opening. TerraFrac uses an integral equation representation, while the
Ohio State model uses the finite-element method. Both models use
finite elements for 2D fluid flow within the fracture and a fracturetip advancement proportional to the stress-intensity factor on the
fracture-tip contour.
Planar 3D Finite-Difference Model (GOHFER). Besides the numerical technique used, this model 18,19 is different from the previous models in two fundamental ways: (1) the fracture opening is
calculated by superposition using the surface displacement of a half7

space under normal load (Boussinesq solution); (2) the fracture propagates when the tensile stress normal to the fracturing plane exceeds the tensile strength of the formation at some distance outside
the fracture by enforcing the tensile criterion at the centroid of the
cells "outside" the fracturing contour. This model predicts higher
treating pressures and shorter, wider fractures than the previous
3D model predictions.
Pseudo-3D Models. These models were developed from the PKN
model by removing the requirement of constant fracture height.
They use equations based on simple geometries (radial, 2D, and
elliptical) to calculate fracture width as a function of position and
pressure and to apply a fracture propagation criterion to length and
height. Furthermore, they assume 1D flow along the fracture length.
These models can be divided into two categories: (1) models that
divide the fracture along its length into "cells" and use local cell
geometry (2D crack or penny crack) to relate fracture opening with
fluid pressure and (2) models that use a parametric representation
of the total fracture geometry. As a result of these assumptions,
each class is expected to have a different fracture geometry, even
for the simple case of a confined fracture.
The pseudo-3D simulators are used extensively for fracture design because of their efficiency and their availability on personal
computers. However, they are directly applicable only for the geometries that are not significantly different from the basic model
assumptions (e.g., models based on a PKN geometry should have
large length/height ratios to be appropriate). Thus, for relatively
unconfined fracture growth in a complex in-situ stress profile, a
3D model is more accurate in predicting "trends" of fracture geometry. To avoid this problem, some pseudo-3D models attempt
to include truly 3D fracture behavior in terms of "history" matching or "lumped" parameters determined from fully 3D solutions
of simpler problems or determined from simulations with 3D
models.
Classic PKN and GDK Models. The difference in treating pressure behavior and fracture geometry of the PKN and GDK models
is well documented 11.37 and is not repeated here.

Fracture Models
This section describes the individual fracture models in this comparison. The modelers or the companies that ran commercially available models provided short descriptions of the models.
Marathon (GOHFER). Marathon Oil CO.'s Grid Oriented Hydraulic Fracture Extension Replicator (GOHFER) 18,19 is a planar
3D fracture geometry simulator with coupled multidimensional fluid
flow and particle transport. As the name indicates, the model is
based on a regular grid structure used for the elastic rock displacement calculations and as a planar 2D finite-difference grid for the
fluid flow solutions. The areal pressure distribution obtained from
the fluid flow equations, including proppant transport, is iteratively coupled to the elastic deformation solution. Using the finitedifference scheme for fluid flow allows modeling of mUltiple discrete fluid entry points, representing perforations at various locations. Each grid node can be assigned an individual value of net
stress, pore pressure, permeability, porosity, wall-building coefficient, rock strength, Young's modulus, and Poisson's ratio, as well
as variables describing fracture-wall roughness and tortuosity. Displacement of the fracture face at each node is determined by integration of the pressure distribution over all nodes, including the
computed tensile stress distribution in the unbroken rock surrounding
the fracture. The fracture width equation used is the general formula given by Boussinesq for displacement of a semi-infinite halfspace acted upon by a distributed load. The solution is general
enough to allow modeling of multiple fracture initiation sites simultaneously and is applicable to any planar 3D geometry from
perfect containment to uncontrolled height growth.
Halliburton (PROP). The PROP program 34-36 is a 2D fracture
design model based on Daneshy's34,35 numerical solution. Its numerical nature makes the model much more flexible than most an8

aiytical models. For example, the program was recently modified

for use of multiple fluids and rates within a single treatment, each
fluid with its own set of time- and temperature-dependent rheological parameters. In addition to the power-law model normally used
to characterize gelled fracturing fluids, PROP uses the threeparameter Herschel-Bulkley model for fluids containing a nitrogen
or carbon dioxide phase. The program's proppant transport calculations are of similar capability. Although the model originally
presented by Daneshy was based on the Khristianovic-Zheltov width
equation (designated GDK in this paper), the PROP program has
since been expanded to include a similar numerical solution of PKNtype geometry with a width profile based on calculated local pressures. The results presented here are for the GDK-type solution only.
S.A. Holditch & Assocs. (TRIFRAC). TRIFRAC is a pseudo3D model of fracture propagation and proppant transport that computes created and propped fracture dimensions with a finitedifference numerical approach. It can handle multiple nonsymmetric
stress layers with unique values for Young's modulus, Poisson's
ratio, fracture toughness, permeability, porosity, and fluid-Ieakoff
coefficients for each layer. Currently, properties for a maximum
of 22 layers can be entered. The apparent viscosity of the fracturing fluid is computed from the shear rate inside the fracture and
changes in flow-behavior index, n' , and consistency index, K', owing to variations of temperature and time. Thus, a temperature calculation model is part of TRIFRAC. The choice of initiating the
hydraulic fracture from 10 different layers simultaneously is available. Special options are available to enter the pump schedule for
nitrogen-foam treatments. The created geometry computation module is coupled with a rigorous finite-difference proppant transport
simulator that solves simultaneously for proppant distribution, transport, and settling, along with fracture growth. Depending on the
fluid velocity along the fracture height and the proppant settling
rate, the model computes the proppant profIle at each timestep during
the job. TRIFRAC also has the simpler 2D geometry computational
finite-difference GDK and PKN models. Horizontal fracture geometry calculation with the GDK method is also available. All these
models are coupled with proppant-transport-calculation modules.
RES (FRACPRO). FRACPR022-25 uses measured values of flow
rate, proppant concentration, and fluid rheology parameters to calculate the pressure drop down a wellbore of variable deviation and
diameter. The time histories of the fracture growth and the net fracture pressure are calculated. The wellbore model handles nonNewtonian fluids and corrects for the effects of nitrogen foam, carbon dioxide, and proppant phases. The model also accounts for friction variation from entrained proppant. The fracture model is 3D
in that spatial variations in reservoir stress, modulus, pressure, and
flow distribution are taken into account.
However, FRACPRO does not need to calculate the variations
at specific points in the fracture. Instead, the effects are integrated
into functional coefficients of governing differential equations, greatly simplifying the calculation of fracture dimensions. The module
can therefore run many times faster than real time, as required for
on-site history matching. The coefficients necessary to calculate
the spatial variations are calculated from a fully 3D model and
checked against experimental and field test data. FRACPRO handles up to 3 modulus zones, up to 50 stress zones, and up to 50
permeable (Ieakoff) zones. Fluid loss is modeled as 1D flow perpendicular to the fracture face, following Darcy-law behavior, including spurt loss, filter-cake buildup on the fracture face, and a
compressible reservoir-fluid region. The rise in confining stress owing to poroelastic effects (backstress) is included. Heat transfer
modeling assumes that there is a cubic-fit temperature distribution
between the fracture and the end of the heat transfer region. FRACPRO models proppant convection and settling in a fracture. In proppant convection, heavier treatment stages (e.g., proppant stages)
displace rapidly downward from the perforations to the fractury
bottom. Then the pad or low-concentration proppant stages replace
those stages. FRACPRO also can be used to model proppant settling.
The proppant is carried with the fracturing fluid and settles. The
model takes into account the effects of non-Newtonian fluids, hindered settling rates, and settled bank buildup.
SPE Production & Facilities. February 1994

TABLE 1-ROCK AND RESERVOIR DATA

Interval

Depth
(ft)

Zone
Thickness
(ft)

9,170 to 9,340

170

1
2
3

8,990 to 9,170
9,170 to 9,340
9,340 to 9,650

180
170
310

1
2
3
4
5

8,990
9,170
9,340
9,380
9,455

180
170
40
75
195

---

In-Situ
Stress
(psi)

--

Poisson's
Ratio

Young's
Modulus
(million psi)

Fracture Toughness
(psi/Jin. )

8.5

2,000

6.5
8.5
5.5

2,000
2,000
2,000

6.5
8.5
5.4
7.9
4.0

2,000
2,000
2,000
2,000
2,000

Single-Layer (2D) Case

5,700

0.21

Three-Layer (3D) Case

7,150
5,700
7,350

0.30
0.21
0.29

Five-Layer (3D) Case

to
to
to
to
to

9,170
9,340
9,380
9,455
9,650

7,150
5,700
7,350
5,800
8,200

0.30
0.21
0.26
0.20
0.30

Chevron 2D Fracturing Simulator. This model can predict the, ing approach adopted. MFRAC-II was run in two different modes
propagation of constant-height, hydraulically induced, vertical frac- to demonstrate the effects of some of these parameters. In one case,
tures for a power-law fluid. The simulator also includes a prop- the base model using system defaults was run (designated MEYER1); in a second case (MEYER-2), additional parameters (such as
pant transport model with proppant settling and a production model.
The simulator can predict the created fracture geometry based on greater friction drop in the fracture) were applied. In both cases,
the PKN and GDK models. It is most suitable to design fractures the viscous thinning assumption was made as a default. Without
where the geologic conditions restrict height growth. In fracture viscous thinning, the effective friction factor would have increased,
resulting in higher net pressures, greater widths, and shorter lengths.
propagation models, the equations describing conservation of mass,
conservation of momentum, continuity of fluid flow, and linear elas- In addition, the fully implicit coupled model for height growth (Vertic deformation of the rock in plane strain are used to calculate mass sion 7.0) results in increased development of fracture height and
flux, fracture width, pressure, and length as functions of time. Given net pressure for certain multilayer formations.
a settling velocity, the proppant transport model calculates the final propped concentration, width, and bank height. It also can Advani (Lehigh U. HYFRAC3D). The three- and five-layer model
results (Cases 5 through 8) are obtained from the HYFRAC3D
predict possible problems caused by proppant bridging or screenout.
code. 17 This finite-element code is based on a set of coupled mass
Shell (ENERFRAC). ENERFRAC20,21 is a hydraulic fracture conservation, fluid momentum, constitutive elasticity, and fracture
model that predicts fracture dimensions for uncontained (circular) mechanics equations governing planar hydraulic fracture propagaand contained (rectangular) fractures. ENERFRAC incorporates tion in a multilayered reservoir. A mapping technique of the basefracture-tip effects and the other interacting processes of viscous line mesh (88 triangular elements representing one-half the fracture)
fluid flow, elastic rock deformation, and fluid loss. Fracture-tip defined in a unit circle to arbitrarily shaped fracture geometries is
effects are accounted for through direct input of the rock's appar- used in the numerical scheme to track the moving fracture front.
ent fracture toughness or the fracture-tip net pressure (overpres- The PKN model results (Cases 1 and 2) also are based on a 2D
sure). This overpressure, defined as the instantaneous shut-in finite-element model simulator with standard PKN model equations,
pressure minus the closure pressure, can be determined in the field including vertical stiffness and 1D fluid flow. These simulation refrom a microfracture or minifracture test. Shell also provided 2D sults are obtained with 20-line elements for the normalized, timePKN and GDK model results. The ENERFRAC results provided dependent fracture half-length.
a useful comparison of the effect of free model parameters (knobs)
on the results. Shell provided results for typical fracture toughness NSI (STIMPLAN). STIMPLAN is a state-of-the-art 3D hydrauvalues measured in laboratory tests (designated ENERFRAC-1) and lic fracture simulator for fracture design and analysis in complex
for a tip overpressure of 1,000 psi (ENERFRAC-2). This compar- situations involving height growth, proppant settling, foam fluids,
ison shows the effect of fracture-tip overpressure on fracture ge- tip screenout, etc. The model has complete fluid/proppant tracking that allows optimum fluid selection and scheduling based on
ometry and net pressure.
time and temperature histories. Fracture height growth is calculatMeyer & Assocs. (MFRAC-II). MFRAC-II26-29 is a pseudo-3D ed through multiple layers and includes proppant settling and bridghydraulic fracturing simulator. MFRAC-II also includes options ing calculations. A fracture analysis/history matching module
for the penny-, GDK- and PKN-type 2D fracturing models. This provides history matching of measured net treating pressures to yield
study was run with MFRAC-II, Version 6.1. MFRAC-II accounts the most accurate possible estimation of actual fracture geometry
for the coupled parameters affecting fracture propagation and prop- and behavior. Also, simulations during fracture closure (pressure
pant transport. The major fracture, rock, and fluid mechanics phe- decline) aid in pressure-decline analysis for fluid loss in complex
nomena include (1) multilayer, asymmetrical confining stress geologic situations.
contrast, (2) fracture toughness and tip/overpressure effects,
(3) rock deformation, (4) variable injection rate and time-dependent Arco (Using TerraFrac). TerraFrac Code 12-16 is a fully 3D hyfluid rheology properties, (5) multilayer leakoff with spurt loss, draulic fracture simulator. Initiated at Terra Tek in 1978, its comand (6) 2D proppant transport. The fracture propagation model cal- mercial availability was announced in Dec. 1983. The model's
culates fracture length, upper and lower heights, width, net pres- overall approach is to subdivide the fracture into discrete elements
sure, efficiency, and geometry parameters as functions of time. The and to solve the governing equations for these elements. These
width variation as a function of height and confining stress also governing equations consist of (1) 3D elasticity equations that reis calculated. To provide applicability over the broadest range of late pressure on the crack faces to the crack opening, (2) 2D fluid
circumstances, MFRAC-II offers numerous options. These options flow equations that relate the flow in the fracture to the pressure
and other free parameters (knobs) allow customization in the model- gradients in the fluid, and (3) a fracture criterion that relates the
SPE Production & Facilities, February 1994

Texaco (Using FRACPRO). Texaco also ran FRACPRO for six

different cases. These include single-layer PKN and GDK models,
a three-layer case with constant fracture-fluid viscosity, and fivelayer cases for constant fluid viscosity, power-Iaw-fluid behavior,
and power-Iaw-fluid behavior with the tip-dominated rheology behavior not operating. The five-layer runs provide a good comparison of tip-dominated vs. conventional rheology results with
FRACPRO.

TABLE 2-TREATMENT DATA

Bottomhole temperature, of
Reservoir pressure, psi
Spurt loss
Fluid-Ieakoff height
Fluid-Ieakoff coefficient, tt/.JiTiTr1
Viscosity-Case A, cp
Viscosity-Case B

246
3,600
0.0
Entire fracture height
0.00025
200
0.5
0.06
10,000
50
None

n'

K'
Fluid volume, bbl
Injection rate, bbl/min
Proppant

intensity of the stress state ahead of the crack front to the critical
intensity for Mode 1 fracture growth. TerraFrac provides many
distinctive features: 2D fluid flow for both proppant and temperature distribution; multiple stages having different fluids, proppants,
and rates, with fluid and proppant properties as functions of temperature if desired; multiple layers, each having different in-situ
stresses, Young's moduli, fracture toughnesses, Poisson's ratios,
and leakoffs; poroelastic and thermoelastic capabilities for waterflooding and other applications; a robust mesh generator to handle
a wide variety of fracture geometries and a quasi-Newton method
to solve the nonlinear system of equations for the fluid pressures
(this approach provides fast convergence and high accuracy); and
a post-shut-in calculation capability for which no additional assumptions are made (only the injection rate changes).

Conoco. Conoco's fracture design program is a constant-height (2D)

model where either PKN or GDK geometry can be selected. 38 It
has single inputs for n', K', and leakoff coefficient. However, the
model can calculate the positions and concentrations of progressive fluid/proppant stages. Fracture area can be calculated by either
the Howard and Fast model or by Crawford's39 extremely accurate
simplification.

SFE3 Formation and Treatment Data

The input data for the fracture modeling comparison are based on
the results obtained at the GRI-sponsored SFE-3 experiment. 3,40
Well SFE-3 was drilled as the Mobil Cargill Unit No. 15 well in
the Waskom field, Harrison County, TX. The well was spudded
in Sept. 1988 and drilled to a total depth of 9,700 ft. Of particular
interest was the Cotton Valley Taylor sand, which was perforated
between 9,225 to 9,250 ft and 9,285 to 9,330 ft. An extensive log
program and detailed core analyses were done on this well. Both
pre fracture well testing and postfracture production testing were
performed. Two minifracs and one full-scale treatment were conducted as part of the stimulation program.

TABLE 3-20 RESULTS AT END OF PUMP

Length
Model
SAHt (GOK)
SAH (PKN)
Marathon
Meyer-1 (GOK)
Meyer-1 (PKN)
Meyer-2 (GOK)
Meyer-2 (PKN)
Shell (GOK)
Shell (PKN)
Texaco-FP (GOK)
Texaco-FP (PKN)
Chevron (GOK)
Chevron (PKN)
Advani
Halliburton
Conoco (GOK)
Conoco (PKN)
ENERFRAC-1
ENERFRAC-2

Height

~ ~
2,542
4,855
2,584
2,659
4,507
2,288
3,803
2,724
4,039
2,480
4,157
1,347
2,029
4,595
2,212
2,716
3,986
3,866
3,556

170
170
204
170
170
170
170
170
170
200
200
170
170
170
170
170
170
170
170

2,542
4,629
2,516
2,098
4,118
1,808
3,395
2,142
3,347
4,046
2,031
2,304
3,656
3,396
3,155

170
170
204
170
170
170
170
170
170
170
170
170
170
170
170

200 cp
Pressure Maximum Width
(psi)
(tt)
0.848
62
1,094
0.502
1,685
0.91
70
0.79
1,188
0.55
97
0.94
1,474
0.68
0.78
53
1,377
0.59
71
0.74
925
0.50
81.9
0.77
1,380
0.63
1,182
0.54
82
0.98
1,595
1,684

0.627
0.704

b'
--

E, *

Efficiency
(%)

0.849
0.394
0.76
0.79
0.43
0.94
0.53
0.78
0.46

0.605
0.289
0.73
0.62
0.32
0.74
0.4
0.61
0.37

0.77
0.43
0.98
0.767
0.554
0.492
0.553

0.6
0.36
0.32
0.77
0.6
0.37
0.387
0.434

85.5
72.3
93
83.1
72.2
85.4
76.6
84
75
86
77
81.9
73
73.8
85.9
82.5
74.4
75
78

0.85
0.42
0.82
1.04
0.5
1.24
0.64
1.03
0.59
0.53
1.07
0.933
0.622
0.58
0.641

0.6
0.28
0.75
0.82
0.36
0.97
0.46
0.81
0.47
0.38
0.84
0.733
0.415
0.456
0.504

61.8
73.6
93
86.4
74.3
88.3
79
89
79
76.9
86
85.2
76.5
78
81.7

n', K'

SAH (GOK)
SAH (PKN)
Marathon
Meyer-1 (GOK)
Meyer-1 (PKN)
Meyer-2 (GOK)
Meyer-2 (PKN)
Shell (GOK)
Shell (PKN)
Advani
Halliburton
Conoco (GOK)
Conoco (PKN)
ENERFRAC-1
ENERFRAC-2

61.8
1,167.5
1,824
117
1,397
161
1,774
89
1,754
1,474
97

0.85
0.54
0.98
1.04
0.64
1.24
0.831
1.03
0.75
0.68
1.07
0.933

1,880
1,986

0.738
0.817

'b = average width at the wellbore.

"b, = overall average fracture width.
t S.A. Holditch & Assocs. Inc.

10

SPE Production & Facilities, February 1994

TABLE 4-THREE-LAYER RESULTS AT END OF PUMP

200 cp
Length

Height

Pressure
(psi)

Model

Maximum Width
(ft)

SAH
NSI
RES
Marathon
Meyer-1
Meyer-2
Arco-Stimplan
Texaco-FP
Advani

3,408
3,750
1,744
1,360
3,549
2,692
3,598
1,938
2,089

318
903
544
442
291
360
306
435
357

1,009
283
1,227
1,387
987
1,109
992
1,132
1,113

0.65
0.56
0.9
1.04
0.58
0.72
0.57
0.72
0.66

SAH
NSI
RES
Marathon
Meyer-1
Meyer-2
Arco-Stimplan
Advani

3,259
3,289
902
1,326
2,915
2,120
3,235
2,424

371
329
596
442
337
413
353
435

1,093
1,005
1,428
1,433
1,094
1,212
1,083
1,171

b
--

b,

Efficiency
(%)

0.35
0.32
0.54
0.68
0.35
0.41
0.31

0.3
0.25
0.36
0.64
0.29
0.34
0.25

0.33

0.25

77
66
80
96
70.3
74.3
67
68
43

0.38
0.35
0.74
0.71
0.4
0.48
0.33
0.34

0.31
0.26
0.49
0.66
0.32
0.4
0.26
0.21

77.6
68
62
96
72.7
76.9
69
47

b,

Efficiency
(%)

0.42
0.38
0.6
0.7
0.36
0.46
0.36

0.31
0.25
0.4
0.65
0.28
0.35
0.24

0.45

0.36

0.46
0.42
0.9
0.71
0.46
0.57

0.31
0.25
0.6
0.66
0.37
0.42

0.47

0.34

n', K'

0.75
0.67
1.1
1.08
0.69
0.86
0.65
0.74

TABLE 5-FIVE-LAYER RESULTS AT END OF PUMP

Model

Length
(ft)

Height
(ft)

SAH
NSI
RES
Marathon
Meyer-1
Meyer-2
Arco-Stimplan
Texaco-FP
Advani

2,905
3,709
1,754
1,224
2,962
2,407
3,399
2,011
1,594

394
361
501
476
328
327
394
428
438

SAH
NSI
RES
Marathon
Meyer-1
Meyer-2
Arco-Stimplan
Arco-Terrafrac
Texaco-FP
Texaco-FPNOTIP
Advani

2,642
2,765
1,042
1,156
2,535
1,980
2,926
3,124
1,125
2,636
1,870

430
388
600
476
330
349
405
449
602
391
458

200 cp
Pressure Maximum Width
(psi)
(ft)
0.72
960
852
0.63
1,119
0.83
1,250
1.03
0.5
669
768
0.6
0.64
944
1,008
0.68
0.81
1,129

--

80.1
66
82
97
70.5
74.8
68
69
58.1

n', K'
1,035.5
935
1,358
1,262
766
891
968
1,160
1,270
934
1,151

The SFE-3 data set was specifically chosen to ensure that the
model comparison would be performed with actual field data, not
for a contrived data set that might favor one type of model. In addition, the SFE-3 data set is one of the most complete sets of well
information available. It includes stress, rock, and reservoir and
well-performance results.
Table 1 shows the relevant rock and reservoir information for
this initial study. As described in the next section, three different
physical configurations were considered: a single layer, three layers, and five layers. Stress and rock property measurements were
averaged over the appropriate depths for each interval to yield the
physical data in Table 1. Most importantly, the stress contrasts range
from 1,450 to 1,650 psi, although the lower barrier is only 40 ft
thick for the five-layer configuration. Young's modulus and Poisson's ratio were obtained from sonic measurements, thus accounting for the elevated values of Young's modulus.
The actual SFE-3 treatment was a 13-stage procedure using
primarily a 40-lbm/l ,OOO-gal crosslinked gel with sand stages varying from 1 to 8 Ibm/gal. For this comparison, the treatment was
simplified to a single, constant-property fluid with no proppant,
SPE Production & Facilities, February 1994

0.82
0.71
1.18
1.04
0.6
0.75
0.7
0.74
1.11
0.49
0.85

81.8
70
87
93
73.7
77.8
70
62
76
62
64

primarily because changes in fluid properties owing to temperature or proppant addition cannot be quantified easily and any resulting comparisons would be of questionable value.

Test Cases
As noted, most models can accommodate and process a much broader range of complex data than presented in this data set (i.e., multiple rock properties, leakoff coefficients, n', and K'). Table 1 and
Table 2 give the complete set of data input. However, the data set
was arbitrarily restricted to limit as many discretionary inputs as
possible to allow more direct comparison of model performance.
The input also should not be construed as optimum design parameters. As mentioned, the data for the cases approximates that from
SFE No.3.
Each participant could model a total of eight possible cases. These
were GDK, PKN, three-layer, and five-layer cases, with separate
runs for a constant Newtonian viscosity and a constant n/ and K'
power-law fluid. The PKN and GDK cases were run with a constant height (2D) set at 170 ft. The three- and five-layer cases were
run with a 3D or a pseudo-3D model, allowing fracture height to
11

--

CONSTANT HEIGHT MODELS

,- ~-o-'~-"~-""-~-;-'-'~-G-_-~-\.
._-_-----._-.. .-2-'--11'---..'-----,'"

::: r7a'"

g
I 3,000
IC}

2,000

oJ

200cp

---.-_.

n'=0.5, K'=O.06

GDK - n'=O 5, K'=O 06

---.---

PKN -200cp

1>"'-' ' '

lr

--

3-LAYER MODELS

GDK -200cp

~~

PKN - n'=0.5, K'=O.06

--0-"

01H':~~.~CP

"-,--,-

OTHER 20 - 0'=0.5, K'=O.06

.. . ---_...
'

,/'

---{~--

1,000

MODEL

Fig. 1-Fracture length-2D models.

.CONSTANT HEIGHT MODELS

Fig. 4-Fracture height-three-Iayer models.

--

;u-c. 1,400

...... 8 ..

II:

3-LAYER MODELS
1,600 , - - - - - - - - - - - - - - - - - - - - - ,

GDK-200cp

'r;;

---.-_.

;u-c. 1,500

PKN-200cp

II:

=>

~ 1,000

Il.

OTHER 20 - n'=0.5, K"=O.06

- ...,:.,....-

500

.,

W
II:

_ .. -6.-..-

Il.

---.---

~ 1,200

--0-OTHER 20 - 200 cp

-200cp

n'=0.5, K'=O.06

=>

PKN - n'=O.S, K'=O.06

W
II:

Iii

'r;;

GDK - n'=0.5, K'=O.06

.........

I- 1,000
W

Fig. 2-Net pressure-2D models.

Fig. 5-Net pressure-three-Iayer models.

3-LAYER MODELS: n', K'

3-LAYER MODELS
4,000 , - - - - - - - - - - - - - - - - - - ,

I-

oJ

~MARATHON

-+-ARCO(STlM)
""'*- MEYER-l
--%-MEYER-2
___ ADVANI

2500

,a

LL

2,000

.:::. 2000

I-

<.:>

g; 1,000

z 1500

OU-__

LL

f#v"I

L-~

___ L_ _

~~

_ __ L_ _J __ _

~"I 9,,?-O"l [c"'~ ,.J-~ .1<-<' ",<'I- s>-.).t-~ v,<,?-O"l

~ ~~'< ~'t-0 ~o-?l ~~'< ~<C-'" ~<t~ ~?'
-;r e}'i? <'-"'~'- o~'1: o'i?
o~'0

'"

-e-SAH
----&- NSI
-IS-RES

3000

---.---

3,000

;;t

-200cp

n'=0.5, K'=O.06

C}

3500~---------~-----,

'"

,?-v
/

'(?'

..J

1000
500

,-<'J-

11-0

?'

MODEL

o~-,

__~__~__~__~~__~~

25

50

75
100 125
TIME (min)

150

175

200

Fig. 3-Fracture length-three-Iayer models.

Fig. 6-Length vs. tlme-three-Iayer models.

be determined by the model. Of particular interest was whether the

fracture broke through Zone 4 in the five-layer case.

2D Results. Considering first the 2D results in Table I, the final

half-lengths for all the 2D models are shown in Fig. 1. The wellknown difference in length estimates between the PKN and GDK
models is evident in these results, but some differences between
different models in each group become apparent. Presumably, tRis
difference results because other options are included in some models,
The effect of the different rheologies is generally small. Besides
the PKN and GDK models, GOHFER and ENERFRAC-l and-2
are also shown,

Model Results
Tables 3 through 5 show the complete set of results for the final
fracture geometries from these model runs for the 2D, three-layer,
and five-layer cases, respectively, Most of the results are based
on the data in these tables, In addition, some time-dependent results will be given in figures, All the data from this comparison
are available in a final report, 41
12

SPE Production & Facilities, February 1994

3-LAYER MODELS: n', K'

600~------------------~--------~

500
~----+------------il

--

5-LAYER MODELS

-e-SAH
--&-NSI
---!'l--RES
---+-- MARATHON
- + - ARCO(STIM)

200cp

0'=0.5, K'=O.06

---.---

,II,

0,

-><--- MEYER- 1

,../

-E-MEYER-2
___$_ ADVANI

\...
;.

fI

"~ 300
200

100L-__~__~__,-~,-~,-~__~__~
o 25 50 75 100 125 150 175 200
TIME (min)

MODEL

Fig. 7-Height vs. time-three-Iayer models.

Fig. 10-Fracture height-five-Iayer models.

3-LAYER MODELS: n', K'

1600,--------------------------------,

5-LAYER MODELS

I -e-SAH

r--=:::::=:::===l'l===fr=~=-~1..............!NSl
---!'l--RES

1400

---+-- MARATHON

'C;;

_-7---te----::=:::; - + - ARCO(STIM)

1200

--::::::::~::~::j ---*MEYER-1
-E-MEYER-2

Q.

~ 1000

___$_ ADVANI

0::

=>
Vl
Vl

t',

en

D-

; ; 1,200

:"..

a:

::>

~ 1,000 .~~~
W

a:
a.

800

l:ii

0::

"-

1,400 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - , , - - - - - - - - ,

,/

".i

\...
\,/

800

600

fW

400
200

o __--~--~--~--~--~--~--._~

25

50

75 100 125
TIME (min)

150

175

200

Fig. 11-Fracture pressure-five-Iayer models.

Fig. a-Net pressure vs. time-three-Iayer models.

5-LAYER MODELS: n', K'

5-LAYER MODELS
4,000 ,-------------------------------------,

I-

-200cp

---.---

3,000

UJ

2500

..J

LL

<t 2,000

I
UJ

2000
I

f-

g; 1,000

~
LL

-e-SAH
-h-NSI
---!'l--RES
---+--MARATHON
-+-ARCO(STIM)
---*-ARCO(TERR)
-E-MEYER-1
___$_MEYER-2
-+-ADVANI
___ TEXACO(FP)
___ TEX -FPNOTP

3000

n'=0.5, K'=O.06

C}

3500,--------------------------------,

OLL__L _ - L_ _

~~ ~
__

_ _L _ _ L_ _L _ _ L_ _

#c,\,:/" ~O\ ~<,;.~ ,:/" #0'. !?<' <,;.!?<'l. N~ ~O, ",,'?\

~~ ""~'? #c,'? 00>(' 6.~'?
~# ~~ 'f-<J #V'? '?~O

#-

~y>o"":Vo ~o <$"'?'

",'f-Y> ...."" >,?-<,;.'"

~'f-~

.4'f-C,0

'f-'?'

~'f-c,0 ~

"zw

1500

...J

1000
500

-<,0/'-<,0/'

MODEL

25

50

75 100 125
TIME (min)

150

175

200

Fig. 9-Fracture length-five-Iayer models.

Fig. 12-Length vs. time-five-Iayer models.

The reduction in length between ENERFRAC-I and -2 results

from increased tip overpressure. Likewise, the reduction in length
between MEYER-l and -2 is caused by options included in
MEYER-2 that reflect the designers' incorporation of more complex physics into the fracturing process.

The net pressures for the 2D models follow a similar pattern to

length, with the GDK models giving low pressures and the PKN
models providing high net pressures (see Fig. 2). GOHFER is different in that it predicts short lengths, like the GDK models, but high
pressures, like the PKN models.

SPE Production & Facilities, February 1994

13

5-LAYER MODELS: n', K'

700

600

----e-SAH
----NSI
---e--RES
---+-MARATHON
--+-ARCO(STIM)

500

-----E- MEYER-1

f-

400

,,-....

~ARCO(TERR)

______ MEYER- 2

.....

'-"

:r:

---+-ADVANI
_______ TEXACO(FP)
_______ TEX-FPNOTP

0
w

:r:
300

200

100~__._--~---r---.----r---.---~--~
75
100 125 150 175 200
25
50

TIME (min)
Fig. 13-Height vs. time-five-Iayer models.

Three-Layer Results. The three-layer results (Table 4) show considerably more variability than results from the 2D cases. In Fig.
3, the fracture half-length varies from < 1,000 ft for FRACPRO
to >3,000 ft for the conventional pseudo-3D models. The differences between MEYER -1 and -2 again show that the options available to the analyst can significantly affect results. Many such options
have probably been used in the other models but were not identified for this comparison.
The favorable comparison between Arco and NSI running Stimplan and a similar favorable comparison between Texaco and RES

running FRACPRO show that consistent results can be obtained

from a given model even if run by different organizations.
The fracture height comparison in Fig. 4 shows that much greater height growth is obtained by FRACPRO than by other models.
Net pressures (Fig. 5) are particularly high in FRACPRO and GOHFER. Efficiencies vary from 40% to >95%, as given in Table 4.
Also of interest are the length, height, and pressure development
with time, as Figs. 6 through 8 show for the case with nonNewtonian rheology. Height growth is extremely fast in FRACPRO but much better contained in most other models.

5- LAYER MODELS: n' , K'

1600~--------------------------------~

----e-SAH
----NSI

1400

I~~:::::!~~~~::::::===::===+---e--RES
---+-MARATHON

--+-ARCO(STIM)
~'----=" ~ARCO(TERR)

,,-.... 1200
(f)

0..
'-"

w
a::

,;::::===~;:;::::;!::=:t====:e::-::t=l-----E- MEYER-1

1000

~*-~~~~~~~~$:i:2::=~ -----MEYER-2
---+-ADVANI
_______ TEXACO(FP)
~

::>
(/)
(/)
w

800

0...

600

,,",-_~----:l~-t

_______ TEX - FPNOTP

a::

fW

400
200

O__--~--~---r--~----r---._--._--~

25

50

75

100

125

150

175

200

TIME (min)
Fig. 14-Pressure vs. time-five-Iayer models.
14

SPE Production & Facilities, February 1994

Five-Layer Results. The five-layer results (Table 5) are similar

to the three-layer comparison, except that fractures in some models
are shorter because the height breaks through the lower barrier.
Fig. 9 shows the half-lengths; Fig. 10 shows the fracture heights.
Net pressures range from'" 700 to '" 1,400 psi (Fig. 11). Efficiencies range from about 60 % to 97 %.
Fig. 12 shows fracture lengths as a function of time. The length
development in this case is not uniform because height breakthrough
into the lower barrier limits growth in some models. The height
growth is shown in Fig. 13 and the net pressure in Fig. 14. By
comparing these results with the three-layer results of Figs. 6
through 8, we can see the effect of breakthrough into the lower
low-stress region.

Discussion and Conclusions

The completion engineer now has a wide array of hydraulic models
available for both design and analysis of hydraulic fracture treatments. However, these models calculate widely different fracture
geometries for the same input parameters, and choosing a model
that meets the engineer's needs becomes important. We hope that
this comparison study provides sufficient information to make a
studied choice.
Some models clearly predict results that are significantly different from the majority. Considering the five-layer cases shown in
Figs. 9 through II, FRACPRO calculates very short fractures, high
net pressures, and large heights. GOHFER also predicts short fractures and high net pressures, but the height growth is not as severe. TRIFRAC, STIMPLAN, TERRAFRAC, and MFRAC-II are
all in general agreement, with longer fractures, less height, and
somewhat lower net pressures. Advani's model is midway between
the two end cases.
MFRAC-II (in 2D, three-layer, and five-layer cases), ENERFRAC (in 2D cases), and Texaco's FRACPRO cases (five-layer)
were run in two different modes and thus provide a useful assessment of the importance of the options available to the fracture
designer. In the original formulation of this study, the modelers
were asked to run their models in both a base mode (no options)
and then with a best-option mode-i.e., a mode that reflected their
expectations of the options needed to provide the closest simulation of true fracture behavior. Such options may have included tip
effects, higher frictional pressure drops in the fracture, multiple
fracture strands, or enhanced toughness.
In the three cases mentioned above, the modelers provided such
a comparison, and these results can be used to estimate how significantly the engineer can modify the fracture design by incorporating his or her estimate of the best physics possible for a given
reservoir. Presumably, such an estimate would be guided by experience with the reservoir. For the five-layer case with nonNewtonian viscosity, best-physics results for fracture length differed
by about 22 % for MFRAC-II and 57 % for FRACPRO run by Texaco. For the 2D case with non-Newtonian rheology, ENERFRAC
results differed by about 7 %. Because many models have such options, these results should be a useful guideline for estimating the
differences in model designs that can be obtained.
The 2D models, both PKN and GDK, generally provide selfconsistent results (Figs. I and 2), and the differences between these
types of models are discussed elsewhere. 11 ,37 Chevron's 2D
model, however, yields considerably shorter lengths than the other
PKN and GDK models. GOHFER is also of note because it yields
a length typical of the GDK models with the net pressure typical
of the PKN models. Other differences in these 2D models are minor.
This particular case was chosen because it was a realistic field
situation for which detailed data were available. The committee and
modelers recognize that other formations, with different stress and
lithology data, may provide a considerably different comparison.
Good examples would be cases with either minimal or extremely
large stress contrasts. It would be beneficial if future model comparisons investigated those cases as well.
Finally, in assembling this comparison, we have purposely attempted to avoid making any value comparisons between the various models. Only the results and quantifiable comparisons (e.g.,
SPE Production & Facilities, February 1994

Model A fracture length is greater than Model B fracture length)

are given. It would take a committee with greater powers than this
one has to truly know how the fracture is evolving in the subsurface and thus to decide which model is better.

Acknowledgments
We thank the GRI for its support of this modeling endeavor. We
also thank our respective companies for allowing us to perform this
comparison. The modelers who participated in the forum and prepared data for this paper also deserve special thanks for their efforts. Most importantly, Steve Holditch of S.A. Holditch & Assocs.
Inc. should be singled out for special mention as the prime mover
of the forum and this follow-up paper. He is not an author because
his firm submitted a model, but this study would never have happened without his efforts. Thanks also go to Bill Whitehead of S.A.
Holditch & Assocs. for his work in the forum and in data gathering.

References
1. Holditch, S.A. et al.: "The GRI Staged Field Experiment," SPEFE
(Sept. 1988) 519.
2. Robinson, B.M., Holditch, S.A., and Peterson, R.E.: "The GRI's Second Staged Field Experiment: A Study of Hydraulic Fracturing," paper
SPE 21495 presented at the 1991 SPE Gas Technology Symposium,
Houston, Jan. 22-24.
3. Robinson, B.M. et al.: "Hydraulic Fracturing Research in East Texas:
Third GRI Staged Field Experiment," JPT (Jan. 1992) 78.
4. Saunders, B.F. et al.: "Hydraulic Fracturing Research in the Frontier
Formation through the GRI's Fourth Staged Field Experiment, " paper
SPE 24854 presented at the 1992 SPE Annual Technical Conference
and Exhibition, WashingtoI), Oct. 4-7.
5. Northrop, D.A. and Frohne, K-H.: "The Multiwell Experiment-A
Field Laboratory in Tight-Gas Sandstone Reservoirs," JPT(June 1990)
772.
6. Cramer, D.D.: "The Unique Aspects of Fracturing Western U.S. Coal
Beds," JPT (Oct. 1992) 1126; Trans., AIME, 293.
7. Martins, P.J. et ai. : "Deviated Well Fracturing and Proppant Production Control in the Prudhoe Bay Field," paper SPE 24858 presented
at the 1992 SPE Annual Technical Conference and Exhibition, Washington, Oct. 4-7.
8. Monus, F.L. et al.: "Fracturing Unconsolidated Sand Formations Offshore Gulf of Mexico," paper SPE 24844 presented at the 1992 SPE
Annual Technical Conference and Exhibition, Washington, Oct. 4-7.
9. Owens, K.A., Andersen, S.A., and Economides, M.J.: "Fracturing
Pressures for Horizontal Wells," paper SPE 24822 presented at the
1992 SPE Annual Technical Conference and Exhibition, Washington,
Oct. 4-7.
10. Meehan, D.N.: "Stimulation Results in the Giddings (Austin Chalk)
Field," paper SPE 24783 presented at the 1992 SPE Annual Technical
Conference and Exhibition, Washington, Oct. 4-7.
11. Recent Advances in Hydraulic Fracturing, J.L. Gidley et al. (eds.),
Monograph Series, SPE, Richardson, TX (June 1989) 12.
12. Clifton, R.J. and Abou-Sayed, A.S.: "On the Computation of the ThreeDimensional Geometry of Hydraulic Fractures," paper SPE 7943
presented at the 1979 SPEfDOE Low Permeability Gas Reservoir Symposium, Denver, May 20-22.
13. Clifton, R.J. and Abou-Sayed, A.S.: "A Variational Approach to the
Prediction of the Three-Dimensional Geometry of Hydraulic Fractures,"
paper SPE 9879 presented at the 1981 SPEfDOE Low Permeability
Reservoir Symposium, Denver, May 27-29.
14. Clifton, R.J. and Wang, J.J.: "Multiple Fluids, Proppant Transport,
and Thermal Effects in Three-Dimensional Simulation of Hydraulic Fracturing," paper SPE 18198 presented at the 1988 SPE Annual Technical Conference and Exhibition, Houston, Oct. 2-5.
15. Clifton, R.J. and Wang, J.J.: "Modeling of Poroelastic Effects in Hydraulic Fracturing," paper SPE 21871 presented at the 1991 SPE Rocky
Mountain Regional/Low Permeability Reservoir Symposium, Denver,
April 15-17.
16. Clifton, R.J. and Wang, J.J.: "Adaptive Optimal Mesh Generator for
Hydraulic Fracturing Modeling," Proc., 32nd U.S. Rock Mechanics
Symposium (1991).
17. Advani, S.H., Lee, T.S., and Lee, J.K.: "Three-Dimensional Modeling of Hydraulic Fractures in Layered Media: Part I-Finite Element
Formulations," ASME J. Energy Res. Tech. (1990) 112, 1-9.
18. Barree, R. D.: "A Practical Numerical Simulator for Three Dimensional
Fracture Propagation in Heterogeneous Media," paper SPE 12273
presented at the 1983 SPE Reservoir Simulation Symposium, San Francisco, Nov. 15-18.

IS

Authors

Warpinski

Moschovidis

Parker

Norm Warplnskl is a distinguished member of the technical staff at Sandia National Laboratories in Albuquerque,
where he specializes in fluid and rock mechanics with interests in hydraulic fracturing, in-situ stresses, geomechanics, and natural fracture systems. He holds a PhD degree in
mechanical engineering from the U. of Illinois and has been
with Sandia since 1977. Warpinski was a 1990-92 Review
Chairman for JPT and was a member of the Editorial Review
Committee. Zlssls Moschovidis is currently a research associate at the Amoco Research Center in Tulsa. A member
of the Wellbore Stability Team, he works primarily in well bore
stability, cuttings injections, hydraulic fracture propagation,
modeling, and rock mechanics. Moschovidis received a diploma in civil engineering from the Natl. Technical U. of Athens,
Greece, an MS degree in structural engineering from the Imperial C. of Science and Technology in London, and a PhD
degree in engineering mechanics from Northwestern U. in
Evanston, 1L. Cecil D. Parker is a senior staff engineer for
Conoco in the Well Completions Group in Houston. He has
spent 25 years designing and executing stimulation treatments worldwide, including the North Sea, Middle East, and
North America. He holds a degree in chemistry/biology from
Abilene Christian U., Abilene, TX. Photograph and biographical sketch of 1.5. Abou-Sayed are unavailable.
19. Barree, R.D.: "A New Look at Fracture-Tip Screenout Behavior,"
1PT (Feb. 1991) 138; Trans., AIME, 291.
20. Shlyapobersky, J.: "Energy Analysis of Hydraulic Fracturing," Proc.,
26th U.S. Symposium on Rock Mechanics, Rapid City (June 1985).
21. Shlyapobersky, J., Wong, G.K., and Walhaug, W.W.: "Overpressure
Calibrated Design of Hydraulic Fracture Simulations, " paper SPE 18194
presented at the 1988 SPE Annual Technical Conference and Exhibition, Houston, Oct. 2-5.
22. Cleary, M.P.: "Analysis of the Mechanisms and Procedures for Producing Favorable Shapes of Hydraulic Fracturing," paper SPE 9260 presented at the 1980 SPE Annual Technical Conference and Exhibition, Dallas,
Sept.
23. Cleary, M.P.: "Comprehensive Design Formulae for Hydraulic Fracturing," paper SPE 9259 presented at the 1980 SPE Annual Technical
Conference and Exhibition, Dallas, Sept. 21-24.
24. Cleary, M.P., Wright, C.A., and Wright, T.B.: "Experimental and
Modeling Evidence for Major Changes in Hydraulic Fracturing Design and Field Procedures," paper SPE 21494 presented at the 1991
SPE Gas Technical Symposium, Houston, Jan. 22-24.
25. Cleary, M.P. and Fonseca, A.: "Proppant Convection and Encapsulation in Hydraulic Fracturing: Practical Implications of Computer Laboratory Simulations," paper SPE 24825 presented at the 1992 SPE
Annual Technical Conference and Exhibition, Washington, Oct. 4-7.

16

26. Meyer, B.R.: "Design Formulae for Two- and Three-Dimensional Vertical Hydraulic Fractures: Model Comparison and Parametric Studies,"
paper SPE 15240 presented at the 1986 SPE Unconventional Gas Technology Symposium, Louisville, May 18-21.
27. Meyer, B.R.: "Three-Dimensional Hydraulic-Fracturing Simulation on
Personal Computers: Theory and Comparison Studies," paper SPE
19329 presented at the 1989 SPE Eastern Regional Meeting, Morgantown, Oct. 24-27.
28. Meyer, B.R., Cooper, G.D., and Nelson, S.G.: "Real-Time 3D Hydraulic Fracturing Simulation: Theory and Field Case Studies," paper
SPE 20658 presented at the 1990 SPE Annual Technical Conference
and Exhibition, New Orleans, Sept. 23-26.
29. Hagel, M. and Meyer, B.: "Utilizing Mini-Frac Data To Improve Design and Production," paper CIM 92-40 presented at the 1992 Annual
Technical Conference of the Petroleum Soc. of CIM, Calgary, June.
30. Kristianovic, S.A. and Zheltov, Y.P.: "Formation of Vertical Fractures by Means of Highly Viscous Liquid," Proc., Fourth World Pet.
Cong., Rome (1955) II, 579-86.
31. Perkins, T .K. and Kern, L.R.: "Widths of Hydraulic Fractures," 1PT
(Sept. 1961) 937; Trans .. , AIME, 222.
32. Geertsma, J. and deKlerk, F.: "A Rapid Method of Predicting Width
and Extent of Hydraulically Induced Fractures," 1PT (Dec. 1969) 1571;
Trans., AIME, 246.
33. Nordgren, R.P.: "Propagation of Vertical Hydraulic Fractures," SPEJ,
(Aug. 1972) 306; Trans., AIME, 253.
34. Daneshy, A.A.: "On the Design of Vertical Hydraulic Fractures," 1PT
(Jan. 1973) 83; Trans., AIME, 255.
35. Daneshy, A.A.: "Numerical Solution of Sand Transport in Hydraulic
Fracturing," 1PT (Jan. 1978) 132.
36. Poulsen, D.K. and Lee, W.S.: "Fracture Design With Time- and
Temperature-Dependent Fluid Properties," paper SPE 12483 presented at the 1984 SPE Formation Damage Control Symposium, Bakersfield, Feb. 13-14.
37. Geertsma, J. and Haafkens, R.: "A Comparison of the Theories for
Predicting Width and Extent of Vertical Hydraulically Induced Fractures," ASME 1. Energy Res. Tech. (March 1979) 101, 8.
38. McLeeod, H.O.: "A Simplified Approach to Design of Fracturing Treatments Using High-Viscosity Cross-Linked Fluids," paper SPE 11614
presented at the 1983 SPE Low Permeability Symposium, Denver,
March 14-16.
39. Crawford, H.R.: "Proppant Scheduling and Calculation of Fluid Lost
During Fracturing," paper SPE 12064 presented at the 1983 SPE Annual Technical Conference and Exhibition, San Francisco, Sept. 5-8.
40. "Staged Field Experiment No.3," GRI-9110048, GRI final report,
Chicago, IL (Feb. 1991).
41. Warpinski, N.R.: "Hydraulic Fracture Model Comparison Study: Complete Results," GRI-93/0109, GRI topical report , Chicago, IL (Feb.
1993) 163.

51 Metric Conversion Factors

E-Ol
bbl x 1.589873
ep x 1.0*
E+OO
E-01
ft x 3.048*
OF (OF-32)/1.8
E-03
gal X 3.785412
E+OO
in. x 2.54*
E-01
Ibm x 4.535924
psi x 6.894757
E+OO
Conversion factor is exact.

m3
mPa's
m

C
m3
em
kg
kPa

SPEPF

Original SPE manuscript received for review April 26, 1993. Revised manuscript received
Nov. 17, 1993. Paper accepted for publication Nov. 16, 1993. Paper (SPE 25890) first
presented at the 1993 SPE Rocky Mountain Regional/Low Permeability Reservoirs Sym
posium held in Denver, April 12-14.

SPE Production & Facilities, February 1994

Discussion of Comparison Study of Hydraulic

Fracturing Models-Test Case: GRI Staged Field
Experiment No. 3
Michael P. Cleary, SPE, Massachusetts Inst. of Technology

Warpinski et al. deserve commendation for selflessly undertaking

the task of compiling their paper and maintaining an independent
approach to the process, which was initiated and supported by the
GRI to assist them (and others) in evaluating the relative performance of various "models" -e.g., in making decisions for their future
funding choices. Because I was at least partly responsible for one
or more of their nominations to this relatively thankless effort, I
hope that they will bear with me now as I make some comments.
I will try to be dispassionate in my comments, but it should be clearly
understood by readers that I do have a vested interest in one of the
modeling approaches (contained in the FRACPRO system).
My comments may be divided into three categories: (1) the limited
comparisons offered by the material in the paper alone; (2) the issues
of matching data from related field operations; (3) the relevance/impact for overall hydraulic fracturing technology. I will treat these
somewhat individually, and with limited space, I will try to make
simply some important points.

1. Limited Comparison of Models in Paper

The approach here was based on an "established" (SPE) approach
of defining a limited "test problem" and comparing the predictions
of the various models. This was deemed especially necessary (e.g.,
by GRI) because there had been such an extraordinary proliferation
of "models" for hydraulic fracturing during the 1980's, most of
them claiming some kind of "3D" capability. I feel partly (perhaps
especially) responsible for this because I "opened Pandora's box"
with my 1980 papers, 1-4 in which I ruminated/formulated different
approaches, including most relevant modeling methods.
Although I understand all the obvious reasons for this comparison
exercise-apart from common observations that some models might
not even satisfy basic requirements (like mass conservation)-I had
to invent for myself some analogy/metaphor for the process. I guess
I stayed as professor a little too long and have gotten the habit of
teaching my students in metaphors/parables. The most immediate
comparison that popped into my head was the gizmo/robot test:
we run design competitions at the Massachusetts Inst. of Technology
(MIT) (which you may see on public TV every year, including international competing teams) in which the objective is to perform some
simple function-e.g., he/she who collects the most ping-pong balls
wins. Such a simple, well-defined test allows us to eliminate the
most obvious failures (e.g., machines that do not work) and even
to declare a clear winner, in most cases.
The objective of Warpinski et at.'s paper was, obviously, not
to declare such a winner (although I would have welcomed such
an objective, as discussed in Points 2 and 3). But could it have been
possible to eliminate the obvious failures? The answer is yes (e.g.,
see Point 2), but not when based on the "rules of engagement,"
which merely report the model calculations for the specified input
parameters. The readers must decide for themselves which model(s)
they believe. This situation generates my only major objection to
the phraseology: because of the artificial test environment, the paper
does not "provide the completions engineer with a practical comparison of the available models," because no way is provided for
engineers to judge the results in any way, unless they resort to the
kind of divine inspiration (translate: personal predisposition, or what
they have been telling their company for years), which still seems
all too pervasive in this industry. So the general question arises:
how should the engineer decide on a model choice?

2. Matching Data From Related Field Operations

The best way to check these models would be to compare their predictions with definitive physical observations. We began this process
about 15 years ago, starting in the laboratory at MIT5-9 and continuing with extensive field work over the past 10 years. 10-20 The
laboratory tests served mainly to check the physical and mathematiSPE Production & Facilities. February 1994

cal consistency of our models, but they also served to eliminate

some specUlative models. For instance, the protuberant shapes calculated by the pseudo-3D Hydrafrac (P3DH) models, which Ioriginally proposed as a tentative simplistic approach, 1 were shown to be
completely wrong, even under ideal circumstances where all the associated physical assumptions should have been justified (if ever).
Although we then (1984) publicly disavowed these models and refused to publish associated papers (including Refs. I and 2) injournals, as a result perhaps achieving pariah status (successfully
achieved later with other publications), many companies continued
to use them and some still even sell them commercially. At least
three of the models compared in the subject paper (and one other
commercial model, not included) are of this kind and are demonstrably wrong at the most basic level. (This is not often made so
obvious: they generally avoid publicly displaying their seductively
long protuberances along the pay zone and/or the associated net
pressure behavior).
What is more often ignored (and even omitted in Warpinski et
al. 's paper, despite extensive data, 10) is the behavior of the true
net fracturing pressure, which can be measured on every job, with
different injection volumes and different fluids. 11-20 We have indeed found field measurements to show dramatically different
response (vs. laboratory behavior and also vs. all conventional
models like those in the subject paper). We also have identified
the major probable causative reasons, establishing the credibility
of the resulting models by matching very many data sets for many
complex and variable reservoir environments. We have done this
without resorting to "knobs" like fracture-fluid rheology or even
without very questionable manipulation of stress. In fact, we have
established such credibility, beyond subjective evaluation, by predicting the results of fracture treatments and successfully executing
the jobs without screenouts and with excellent matching of pressure
predictions, as recently reported by numerous users at the first GRI
Real-Time Conference (Houston, Oct. 1993).14-20
We did not even insist on the application of the latter stringent
condition (i.e., prediction) to the evaluation process now reported
in the subject paper. Although such a blind experiment would truly
have been the proper approach to the overall undertaking, we had
already correctly predicted the response to the job in question (GRI
SFE No.3, Ref. 10), and our work on this was already provided
or available to the other modeling participants (along with all of
the postjob data) before the "contest" was held. However, matching
ofthis available data was not required of the participants, so many
of the models' calculations did not (even with such hindsight) match
the net pressure data (e.g., Figs. 9 and 15 of Warpinski et al. 's
paper). This explains many of the major differences between models.
There are even two different results by two FRACPRO users: RES
used the physically realistic default model in FRACPRO, while Texaco presumably used one of the many conventional model options,
which are actually provided in FRACPRO to clearly demonstrate
that such models do not match the pressure data.
At risk of (being libelled for) excessively belittling expensive software sold by certain vendors, I may point out that the general story
of most models in the paper may be summarized in a simple equation
for fracture-wing length, L, and half-height, h:
U+ 1h 2 - 1 =gEVI/(PI-uc ), . . . . . . . . . . . . . . . . . . . . . . (D-la)
where VI = volume of fluid remaining in the fracture (efficiency,
e X volume pumped, V), E = Young's modulus, PI = fracture pressure (e.g., determined carefully from instantaneous shut-in pressure), uc=closure stress, and g (of order 0.25)="garbage
collector" (which depends on fracture geometry, with ridiculuously
high values for aforementioned protuberances). The variable 1 allows for variable fracture geometries (/=0 for L>h and/or PKNtype geometries, 1= I for L<h and/or GDK-type geometries, and
1=2 for circles).
17

This equation is simpler when reduced to oilfield units:

2 - 1 2V
l+ l h 100
net .................. (D-Ib)
L 100
== e 100 E 106 / PlOO'

in which L lOO and hloo=length and height in l00-ft units and

V lOO =pumped volume in l00-bbl units (while PI?:} =net pressure
in units of 100 psi and E I06 =Young's modulus in units of 10 6 psi).
To achieve ultimate simplicity, we can plug in actual data from
SFE No.3 (in round terms: Plog ==16; E I06 ==6; V loo ==I(0) to
get a very approximate mnemonic result:
2Llooh l00 2 == 100 ............................... (D-lc)
Clearly, variation of p net (which in turn, along with artificial
stress profiles, should greatly affect h in all the models in Warpinski
et al.'s paper) will affect the resulting length. For example, Figs.
9 and 15 show that FRACPRO and the only other considered 3D
model (Marathon) calculate higher net pressures (closer to the excellent agreement achieved with actual data), greater fracture
heights, and therefore shorter fracture lengths.
Ironically, despite the incredulous reaction to our early work,
matching of production data shows even shorter fracture lengths,
(e.g., Ref. 10, which typifies the reality of postfracture production
matching on many jobs in low-permeability reservoirs). This reality
is finally dawning on some in the industry. The "ballpark" expressed in Eq. D-lc may be optimistic, even with nonoptimistic
geometry. For instance, Llh == 3 - L - 800 ft should produce greater
production than actually observed, if proppant was placed effectively
opposite pay zones. 13 The latter comment represents the crux of
the whole matter: all the models calculate fracture lengths that are
excessively optimistic in terms of the actual production, unless
ridiculous games are played with kh, damage, etc. , and a more
realistic estimate of actual production could have been obtained with
the simplest possible assumptions: L==h-L-400 ft.
Indeed, this comment applies more generally to most jobs pumped
in this industry: if all treatments were just able to achieve an effectively circular fracture with the proppant placed effectively (Ref.
13), the overall average production throughout the industry would
probably be much better and overall job costs would generally be
substantially reduced.

3. Relevance to Overall Hydraulic

Fracturing Technology
Practically speaking, what matters to the completions engineer is
how he/she can reduce cost and make better wells. Any effort that
claims to be a (supporting) technology must contribute to one or
both of these goals. Our efforts, over the past 5 years at least, have
demonstrated that dramatic cost reductions and greatly improved
production (beyond conventionally designed jobs) can be achieved
confidently only with the use of appropriate real-time technology.
An intrinsic part of such a system is a reliable general physical model
that allows accurate predictions, rapid on-site evaluation, and desigp.
or redesign and execution of the most effective fracturing treatments.
Such a capability provides a reliable, cost-effective, and credible
approach that may be contrasted with costly and misleading efforts
commonly used to justify continued inappropriate procedures. Extreme examples are the use of fracture-height logs (e.g., sonic, temperature, and/or tracer), which we need not even condemn for the
thoughtful reader. Neither need we explain limitations of shallow
(e.g., "big dig") or near-wellbore observations.
What we do need to point out is that the kind of data-isolationist
modeling represented by the subject paper has been used by vested
interests to mislead the industry for many years. This process continues at some companies. Rather than going out on each individual
job, finding out what is happening (with instant analysis and feedback in real time), and pumping the appropriate job, some personnel
still insist on rendering a grave financial disservice to their company
by using such models or surveys as a crutch to support their own predispositions (e.g., about fracture dimensions). Such "blind-man's
bluff" (BMB) hurts all of us, including its practitioners, as careful
evaluation on the long term shows; it renders vast resources uneconomical, with associated negative effects on the U.S. reserve base.
However, I am happy to report that the tide has turned: sensible
practical field-oriented personnel at many companies have grown
tired of BMB and have demanded and obtained change. Spurious
18

modeling and capability claims have been exposed (and discarded)

because of continuous failure in the field, and most companies are
now turning to real technology. It may be hoped that Warpinski
et al. 's paper might serve to accelerate rather than retard that
process.

References
1. Cleary, M.P.: "Comprehensive Design Fonnulae for Hydraulic Fracturing," paper SPE 9259 presented at the 1980 SPE Annual Technical
Conference and Exhibition, Dallas, Sept. 21-24.
2. Cleary, M.P.: "Analysis of Mechanisms and Procedure for Producing
Favorable Shapes of Hydraulic Fractures, " paper SPE 9260 presented at
the 1980 SPE Annual Technical Conference and Exhibition, Dallas,
Sept. 21-24.
3. Cleary, M.P., Keck, R., andMear, M.: "Microcomputer Models for the
Design of Hydraulic Fractures," paper SPE 11628 presented at the 1983
SPE Symposium on Low Permeability Gas Reservoirs.
4. Cleary, M.P., Kavvadas, M., and Lam, K.Y.: "Development ofa Fully
Three-Dimensional Simulator for Analysis and Design of Hydraulic
Fracturing," paper SPE 11631 presented at the 1983 SPE Symposium
on Low Permeability Gas Reservoirs, March.
5. Cleary, M.P.: "Theoretical and Laboratory Simulation of Underground
Fracturing Operations," MIT UFRAC Annual Reports, Cambridge,
Sept. 1981-87.,
6. Crockett, A.R., Okusu, N.M., and Cleary, M.P.: "A Complete Integrated Model for Design and Real-Time Analysis of Hydraulic Fracturing Operations," paper SPE 15069 presented at the 1986 SPE California
Regional Meeting, Oakland, April.
7. Cleary, M.P., Barr, D.T., and Willis, R.M.: "Enhancement of RealTime Hydraulic Fracturing Models With Full 3D Simulation," paper
SPE 17713 presented at the 1988 SPE Gas Technology Symposium,
Dallas, June 13-15.
8. Cleary, M.P.: "The Engineering of Hydraulic Fractures-State of the
Art and Technology," JPT (Jan. 1988) 13.
9. Johnson, D.E. and Cleary, M.P.: "Implications of Recent Laboratory
Experimental Results for Hydraulic Fractures," paper SPE 21846 .
presented at the 1991 SPE Rocky Mountain Regional Meeting/Low Permeability Symposium, Denver, April.
10. Gas Research Institute Staged Field Experiments (SFE) I through 4;
No.3 finally published as Report No. GRI-9110048, Feb. 1991.
11. Wright, T.B., Johnson, D.E., and Cleary, M.P.: "Real-Data/On-Site
Analysis of Hydraulic Fracturing and Procedures for Design Optimization," paper presented at the IntI. Gas Research Conference, Orlando,
FL, Nov. 1992.
12. Cleary, M.P., Wright, C.A., and Wright, T.B.: "Experimental and
Modeling Evidence for Major Changes in Hydraulic Fracturing Design and Field Procedures," paper SPE 21494 presented at the 1991
SPE Gas Technology Symposium, Houston.
13. Cleary, M.P. and Fonseca, A. Jr.: "Proppant Convection and Encapsulation in Hydraulic Fracturing: Practical Implications of Computer and
Laboratory Simulations," paper SPE 24825 presented at the 1992 SPE
Annual Technical Conference and Exhibition, Washington, DC, Oct. 4-7
14. Cipolla, c.L., Meehan, D.N., and Stevens, P.L.: "Hydraulic Fracture Perfonnance in the Moxa Arch Frontier Formation," paper SPE
25918 presented at the 1993 SPE Rocky Mountain Regional Meetingl
Low Permeability Reservoirs Symposium, Denver, April 26-28.
IS. Martinez, A.D., Wright, C.A., and Wright, T.B.: "Field Application
of Real-Time Hydraulic Fracturing Analysis," paper SPE 25916 presented at the 1993 SPE Rocky Mountain Regional Meeting/Low Penneability
Reservoirs Symopsium, Denver, April 26-28.
16. Wright, T.B. et al. : "Identification and Comparison of True Net Fracturing Pressures Generated by Pumping Fluids With Different Rheology
Into the Same Formations," paper SPE 26153 presented at the 1993
SPE Gas Technology Symposium, Calgary, June 28-30.
17. Johnson, D.E. etal.: "On-Site Real-Time Analysis Allows Optimal
Propped Fracture Stimulation of a Complex Gas Reservoir," paper SPE
25414 presented at the 1993 SPE Production Operations Symposium,
Oklahoma City, March.
18. Johnson, D.E. et al.: "Real-Data On-Site Analysis of Hydraulic Fracturing Generates Optimum Procedures for Job Design and Execution," paper
SPE 25920 presented at the 1993 SPE Rocky Mountain Regional Meeting/Low Permeability Reservoirs Symposium, Denver, April 26-28.
19. Cleary, M.P. et al.: "Field Implementation of Proppant Slugs To Avoid
Premature Screenout of Hydraulic Fractures With Adequate Proppant
Concentration," paper SPE 25892 presented at the 1993 SPE Rocky
Mountain Regional Meeting/Low Penneability Reservoirs Symposium,
Denver, April 26-28.
20. Cleary, M.P. et al.: "Critical Issues in Hydraulic Fracturing of High Permeability Reservoirs," paper SPE 27618 to be presented at the 1994 SPE
European Production Operations Conference, Aberdeen, March 15-17.
(SPE 28158)

SPEPF
SPE Production & Facilities, February 1994