IPTC-19596-MS

Compositional Tracking: Predicting Wellstream Compositions in Tight

Unconventionals

Mathias Lia Carlsen and Mohamad Majzoub Dahouk, Whitson AS; Stian Mydland and Curtis Hays Whitson,
Whitson AS, NTNU

Copyright 2020, International Petroleum Technology Conference

This paper was prepared for presentation at the International Petroleum Technology Conference held in Dhahran, Saudi Arabia, 13 – 15 January 2020.

This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submitted by the author(s).
Contents of the paper, as presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The
material, as presented, does not necessarily reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at
IPTC are subject to publication review by Sponsor Society Committees of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial
purposes without the written consent of the International Petroleum Technology Conference 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 where and by whom the paper was presented.
Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435.

Abstract
Wellstream compositions are needed for a range of critical engineering applications but it is impractical
and too expensive to acquire these compositions frequently (e.g. daily) for every well in an unconventional
basin. Detailed PVT studies that provide wellstream compositions can easily cost 25,000–50,000 USD and
are typically performed on less than 1% of the wells in a "shale" basin. This paper provides methods to
predict wellstream compositions daily in wells without detailed PVT analysis by using readily available
production and well test (fluid sampling) data.
The three different methods presented all leverage a field-wide equation of state (EOS) model together
with a varying degree of available production and well test data. The EOS model used should honor all the
available fluid data in the relevant field, region, or formation, and should be developed based on as many
detailed PVT studies as possible. The difference between the three different methods is the amount and type
of data typically available, ranging from production data only to extensive compositional well test data.
This paper shows that accurate wellstream compositions can be predicted by using a field-wide EOS
model together with production and well test data. The accuracy of the different methods depends on (1)
the amount of available measured data, (2) the quality of the available measured data and (3) how well the
EOS model captures the phase characteristics of the relevant fluid. The proposed methods provide a tool to
consistently and accurately calculate daily wellstream compositions for all wells in a cost-efficient manner.
EOS models enable convenient and flexible calculations for describing phase behavior of petroleum
fluids. However, to use an EOS model, three input parameters are needed: temperature, pressure and
composition. In unconventionals, temperature and pressure are almost always available, but wellstream
composition is not. The purpose of this work is to make wellstream compositions readily available for
unconventional fields with thousands of wells, and thereby unlock the potential for EOS model utilization.

Introduction
Since the first accurate PVT correlation was developed in the 1940's, a wide range of PVT correlations have
been developed. Common to them all is that they rely on simple relationships between PVT properties (Bo,
2 IPTC-19596-MS

Rs, μo) and readily available field data (γapi, p, Tres). The most known correlations are the ones developed
by Standing (Whitson and Brule 2000). Alternative methods to predict PVT properties have also been
suggested, such as analyzing mud gas data as proposed by Yang et al. (2014, 2019).
In general, these PVT correlations are only realiable for "simple" fluid systems such as black oils
(Rsi<1500scf/STB) and dry gases (Rsi→∞). The shortcomings of these correlations become apparent for
more "complex" fluid systems such as volatile oils (1500<Rsi<3000scf/STB), near critical fluids (3000scf/
STB<Rsi<5000scf/STB) and gas condensates (5000scf/STB<Rsi<50,000 scf/STB). For such complex fluid
systems, the industry tends to rely on equation of state (EOS) models.
An EOS model is a mathematical equation that models the relationship between a fluid's pressure,
volume, temperature, and composition – which constitutes the state of the fluid. In addition, an EOS also
models a fluid's phase behavior, predicting when and how the fluid will partition into equilibrium phases
(typically gas and liquid), giving the relative amounts, compositions, and properties of those phases (Zick
2013). To apply an EOS for a range of fluids producing from a field – an EOS fluid characterization must
be developed and tuned to field data, as laid out in detail in the work by Younus et al. (2019) and others
(Martinsen et al. 2010, Singh et al. 2011).
In this paper, we are going to discuss how an EOS model can be used to predict wellstream compositions
over time – referred to as compositional tracking. We are also going to narrow the discussion to focus on
tight unconventionals. Compositional tracking is especially interesting in unconventionals because, i) the
produced fluid properties from a "shale" well tend to change significantly over time (Carlsen et al. 2019b),
and ii) most "shale" basins tend to contain very complex (near-critical) in-situ fluid systems (Younus et al.
2019). Production from tight unconventional basins, such as the Eagle Ford, Permian, and Montney, are
characterized by depletion performance in which the producing GOR is a strong function of the flowing
bottomhole pressure and the degree of undersaturation (Whitson and Sunjerga 2012; Jones 2017). This is
contrary to conventional reservoirs in which the depletion performance will have the producing GOR as a
strong function of the average reservoir pressure.

Compositional Tracking: Literature Review

The open literature is sparse when it comes to documenting methods that leverage an EOS model to predict
wellstream compositions over time (compositional tracking).
Whitson and Sunjerga (2012) were the first to suggest a method to predict wellstream compositions
by using readily available welltest data and an EOS model. The Whitson–Sunjerga method requires, (1)
separator-gas compositions (yi), (2) stock-tank liquid API (γAPI), (3) separator conditions (Tsep, psep), and (4)
separator producing GOR (Rp,sep), in addition to an appropriate EOS model to make an accurate prediction
of the associated producing wellstream.
Thereafter, an alternative method was suggested by Hoda and Whitson (2013). The Hoda–Whitson
method requires, (1) an appropriate EOS model, (2) an estimate of the wellstream composition (i.e. a "seed
feed"), (3) test-separator volumetric rates, and (4) test-separator conditions. This method was improved in
2017 (Hoda, Hoffmann and Kuntadi) by incorporating procedures that also match the measured separator-
gas specific gravity (γg,sp) and stock-tank liquid API (γAPI), and not only the separator GOR at the associated
separator conditions.
Our experience in "shale" suggests that separator-gas compositions are always measured together
with separator-gas specific gravity, as the specific gravity is most often calculated from the separator
compositions. In addition, if separator-oil compositions are measured, they are typically measured in tandem
with a stock-tank liquid API (after flash down to standard conditions). This practical observation provides
the basis for a third, alternative method presented in this paper.
IPTC-19596-MS 3

The purpose of this paper is to summarize and compare the three methods introduced above. It is
important to emphasize that these are not competing methods, but that each represent the "best one can do"
with the associated amount of data.

Wellstream Compositions
EOS models enable convenient and flexible calculations for describing phase behavior of petroleum fluids.
However, to use an EOS model, a composition is needed in addition to pressure and temperature. In
unconventionals, temperature and pressure are almost always available, but wellstream compositions are
not. The purpose of this work is to make wellstream compositions readily available for unconventional
fields with thousands of wells, and thereby unlock the potential for EOS model utilization.
Compositions define the relative amounts of different components that make up a fluid. A wellstream
composition represents the composition a well produces at one point in time. This should not be confused
with in-situ representative fluid compositions. In-situ representative fluid compositions represent the
original fluid(s) in place – i.e. before the well is put online. A wellstream composition may be in-situ
representative but is more likely to be reservoir representative. A reservoir representative fluid composition
is any uncontainminated sample produced from the reservoir (Fevang and Whitson 1994). This paper does
not address whether a predicted wellstream composition is in-situ representative or not, as this requires
additional information such as flowing bottomhole pressures, when the well was drilled (parent-child
depletion), and the sampling program itself (representative GOR or not). On the other hand, this paper
addresses how to computationally obtain an accurate wellstream-composition prediction for one specific
point in time.
There are several reasons why engineers would like to know variations in wellstream compositions over
time. Wellstream compositions can be used to, for example:
1. indicate when the well is i) producing at BHPs below the in-situ saturation pressure, ii) producing
from several layers with different in-situ fluid compositions, or iii) experiencing gas coning in the
perforated interval (most relevant for conventional reservoirs).
2. assist in history matching and production performance forecasting.
3. compare well-to-well production behavior throughout a field in a consistent, "surface-process
insensitive" manner – indicating differences in well performance and/or in-situ fluid spatal variations.
4. allocate oil and gas rates and/or components to individual wells.
5. assist in condensate tracking – study relative contributions from a condensate gas cap.
6. assist in fluid initialization and well classification exercises.
7. understand compositionally sensitive processes such as gas enhanced oil recovery.
8. study the sensitivity of surface process on rates, liquid yields and GORs.
9. perform separator train optimization.
10. normalize for changing separator conditions. Separator conditions might vary considerably over time,
hence, part of the GOR variation seen over time is due to changing separator conditions. By knowing
wellstream compositions over time one can express GOR in terms of a fixed surface process and
remove "noise" in production data ("GOR normalization")
The wellstream compositions together with pressure and temperature are input to an EOS model that
again can be used to predict PVT properties. Historically, the most common method to determine wellstream
compositions in the petroleum industry has been to either gather separator samples (and recombine them
at some GOR), or less common, obtain bottomhole samples. After these samples are gathered in the field,
they are typically analyzed in the laboratory by gas chromatography (GC) and/or true boiling point (TBP)
distillation (e.g. ASTM D2892) in order to determine the compositions. Modern PVT laboratories typically
measure compositions up until C30+ or C36+, and the compositional measurements normally come in tandem
4 IPTC-19596-MS

with traditional PVT experiments (i.e. a CCE experiment and some type of depletion experiment (DLE
or CVD)). However, detailed PVT studies that provide wellstream compositions can easily cost 25,000–
50,000 USD; thus, performing this type of analysis daily is both expensive and impractical. Therefore, when
detailed laboratory measurements are lacking, engineers need an accurate, cost-efficient, and consistent way
to predict these properties with the data that is readily available.

Predicting Wellstream Compositions

In this paper, we will study three different methods based on the three most common fluid sampling data
sources available in tight unconventionals. The methods differ based on what fluid sampling data that is
available and are summarized in Table 1.

Table 1—Different Production and/or Welltest Data Typically Available for Most Unconventional Wells

Method 1. Production Data Only 2. Limited Welltest Data 3. Extensive Welltest Data

Separator Gas Compositions Separator Gas Compositions

Separator GOR Stock-Tank Liquid API Separator Oil Compositions
Data
Separator Pressure Separator GOR Separator GOR
Available
Separator Temperature Separator Pressure Separator Pressure
Separator Temperature Separator Temperature

Cost - ~ 200-800 USD ~1000-3000 USD

We analyze a range of available data sources because the well-test data vary significantly from area to
area, and company to company. Sometimes, basic fluid sampling data are required by local governments.
This is the case in Canada, where separator oil and gas compositions must be reported for all new wells.
In the US, on the other hand, fluid sampling data is frequently used for allocation and product marketing
purposes.
Method 1: Production Data Only (Hoda and Whitson 2013). Fluid sampling is typically only performed
a few times (1–20+) over a well's lifetime. However, production data and the associated surface process
(#stages, psep, Tsep) are available for all wells, every single day. Hence, it is important to have a consistent
and practical method to convert production data only to compositional wellstreams, as this represents the
majority of the datapoints.
The requirements for this method are, (1) a properly tuned EOS model, (2) an estimate of the wellstream
composition – i.e a "seed feed", (3) test-separator volumetric rates, and (4) test-separator conditions. The
"seed feed" is usually the most recently determined wellstream composition from an earlier test.
1. First, the "seed feed" is flashed to the test separator conditions using a properly tuned EOS model,
computationally obtaining separator equilibrium phases (gas and liquid), giving the relative amounts,
compositions and properties of those phases.
2. Secondly, these phases are computationally recombined at the separator producing GOR. The molar
component rate ni is found by scaling the seed-feed separator-gas mole fraction by the ratio of
measured gas rate qgm to calculated seed-feed gas-molar volume vg, adding to the computed seed feed
oil mole fractions scaled by the ratio of measured oil rate qom to seed feed calculated molar volume
vo, as given by Eq. 1

(1)
where zi = ni/n, and n = Σ ni = the well test molar rate. The molar volumes are evaluated at the
conditions at which the rate is measured – gas rate at standard conditions and oil rate at separator
conditions. The separator-oil rate must be measured at separator conditions (no shrinkage applied),
IPTC-19596-MS 5

and the gas rate should not include gas coming out of solution when separator oil is processed to stock-
tank conditions (no flash factor accounted for). Note that this method does not require any regression.
Method 2: Limited Welltest Data (Whitson and Sunjerga 2012). The requirements for this method are,
(1) a properly tuned EOS model, (2) separator-gas composition, (3) test-separator volumetric rates, (4) test-
separator conditions, and (5) stock-tank liquid API. To honor the material balance, it is important that all
these measurements are taken at the same point in time.
This method relies on finding a wellstream composition that when input in an EOS model, exactly
reproduces the welltest data (i.e separator-gas compositions, separator GOR, and stock-tank liquid API).
The separator-gas composition defines the relative amounts of the lighter wellstream components (N2, CO2,
C1, …, n-C5, and C6). The separator GOR reflects the C7+ amounts, whereas the stock-tank liquid API defines
the C7+ characterization (distribution of components). The proposed routine uses the N+2 known data (yi,
Rp,sep, γAPI) to solve the system of equations with N+2 unknowns . The system of equations
is solved in the EOS-based software PhazeComp (Zick Technologies). C7+ of the wellstream is modeled
with a gamma model (Whitson 1983) with a fixed shape factor, α, and lower bound, η common for all the
fluids in the respective field. The C7+ molecular weight of the wellstream is a regression variable. The C7+ of
the separator gas is modeled with an exponential molar distribution function (α=1), and the C7+ molecular
weight of the separator gas is often reported together with the separator-gas compositions.
Method 3: Extensive Welltest Data.. The requirements for this method are, (1) a properly tuned EOS
model, (2) separator-gas composition, (3) separator-oil composition, (4) test-separator volumetric rates, and
(5) test-separator conditions. To honor the material balance, it is important that all these measurements are
taken at the same point in time.
To ensure that the separator oil and gas compositions are in equilibrium, the Hoffman K-value plot can
be created (Hoffman, Crump and Hocott, 1953). It should be noted, however, that even small discrepancies
from the Standing correlation in this plot can indicate that the samples are not in equilibrium.
First, gamma modeling of the C7+ separator oil and gas is performed. C7+ is typically the heaviest
component measured during a welltest for both separator-oil and separator-gas compositions. C7+ of the
separator oil is modeled with a gamma model (Whitson and Brule, 2000, Chapter 5.3.2) with a fixed
shape factor, α, and a common lower bound, η, for all the fluids in the respective field. The C7+ molecular
weight of the separator oil, Mo,C7+, is typically reported together with the separator-oil compositions. If
Mo,C7+ is not reported then the liquid API can be used as a surrogate; in such case, the method will be
slightly altered as there will be regression on the Mo,C7+ to match the stock-tank liquid API. The C7+ of the
separator gas is modeled with an exponential molar distribution function (α=1). The C7+ molecular weight
of the separator gas, MWg,C7+, is often reported together with the separator-gas compositions. As a quality
check, the saturation pressure of the separator oil should be calculated with the relevant EOS model and
the predicted value should be close to the reported separator pressure. The separator oil composition is
recombined with the separator gas composition at the measured separator GOR and associated separator
conditions.

Examples

Example 1: Compositional Reservoir Simulation.. First, to test and compare the different methods
summarized above, we

• Utilized a 3D compositional reservoir simulation model identical to what was presented by Carlsen
et al. (2019b). The key model parameters are given in Table 2. The models were run on a BHP
profile that mimicks the "typical" BHP behavior seen in tight unconventionals: rapid decline from
6 IPTC-19596-MS

initial reservoir pressure (7500 psia) until some minimum, constant, bottomhole pressure (500
psia).
• Studied the four different in-situ reservoir fluids presented in Table 3; black oil, volatile oil, near-
critical oil and gas condensate. The associated phase envelopes (p-T diagrams) are presented in
Fig. 1 while the simulated GOR performance versus time is presented in Fig. 2. This is important
to understand the relative importance of different fluid systems as most unconventional basins, e.g.
Permian, Eagle Ford, and Montney, span a wide range of fluids.
• Leveraged the common EOS model presented in Table 4. Hence, the only difference between each
of the different runs is the insitu reservoir fluid composition (zRi), which allows for consistent
apples-to-apples comparison.

Figure 1—Phase envelope (p-T diagrams) for the different fluid systems referenced in this paper.

Figure 2—pwf and GOR versus time for the different fluid systems referenced in this paper.
Constant 2-stage surface process, psep,1 = 300 psia, Tsep,1 = 100 F, psep,2 = 1 atm, Tsep,2 = 60 F.
IPTC-19596-MS 7

Table 2—Reservoir simulation model assumptions

Variable Unit Value

Reservoir pressure, pR psia 7500

Minimum flowing bottomhole pressure, pwf psia 500
Reservoir temperature, Tr F 250
Fracture half length, xf ft 325
Frac-to-Frac distance ft 50
Thickness, h ft 150
Matrix permeability nd 200
Porosity, φ - 0.03
Initial water saturation, Swi - 0
Residual oil saturation in gas-oil system, Sorg - 0.4
Critical gas saturation, Sgc - 0.1
Corey oil exponent, no - 2
Corey gas exponent, ng - 2
krg at maximum Sg - 0.7
kro at maximum So - 1
Fracture permeability md 12005
Rock pore volume compressibility, cf 1/psi 4E-06
Separator pressure (stage#1) psia 300
Separator temperature (stage#1) F 100
Separator pressure (stage#2) psia 14.7
Separator temperature (stage#2) F 60
Number of grid cells per fracture # 24000
8 IPTC-19596-MS

Table 3—Different fluid systems analyzed in this paper

Near-Critical
Component Near-Critical Volatile Oil Volatile Oil Black Oil
Gas Condensate

H2S 0.00 0.00 0.00 0.00

N2 0.31 0.56 0.21 0.34
CO2 2.37 1.30 0.93 0.02
C1 73.19 69.44 58.77 34.62
C2 7.80 7.88 7.57 4.11
C3 3.55 4.26 4.09 1.01
I-C4 0.71 0.89 0.91 0.76
N-C4 1.45 2.14 2.09 0.49
I-C5 0.64 0.90 0.77 0.43
N-C5 0.68 1.13 1.15 0.21
C6 1.09 1.46 1.75 1.61
C7 1.13 1.07 2.19 4.51
C8 1.12 1.09 2.25 4.70
C9 0.85 0.86 1.78 3.81
C10 0.70 0.73 1.52 3.31
C11 0.58 0.63 1.32 2.94
C12 0.49 0.55 1.16 2.64
C13 0.42 0.49 1.03 2.39
C14 0.36 0.43 0.92 2.17
C15 0.31 0.38 0.82 1.99
C16 0.27 0.34 0.74 1.82
C17 0.23 0.31 0.67 1.67
C18 0.20 0.28 0.60 1.54
C19 0.18 0.25 0.55 1.42
C20 0.16 0.22 0.49 1.31
C21 0.14 0.20 0.45 1.22
C22 0.12 0.18 0.41 1.13
C23 0.10 0.17 0.37 1.05
C24 0.09 0.15 0.34 0.97
C25 0.08 0.14 0.31 0.91
C26+ 0.68 1.58 3.83 14.89

Total 100 100 100 100

Properties

psat, psia 6086 6044 4542 1985

GOR, scf/STB 5714 3938 1589 326
OGR, STB/MMscf 175 254 629 3067
Yapi 49.0 47.0 44.4 39.3
C7+ 8.2 10.0 21.8 56.4
*Based on a 2-stage separator process with T sep,1 = 100 F, psep,1 = 300 psia and T sep,2
= 60 F and psep,2 = 14.7 psia (standard conditions). Reservoir temperature is 250 F.
IPTC-19596-MS 9

Table 4a—Component properties of the "generic" Soave-Redlich-Kwong (SRK)

equation of state (EOS) model used for example simulations in this paper.

Component Mol Wt Tc (R) Pc (psia) Crit Z Vol Tran AF A B Tb (R) Visc Zc LMW

H2S 34.1 672.1 1300.0 0.29 0.1015 0.0900 0.42748 0.08664 382.4 0.283
N2 28.0 227.2 492.8 0.29 -0.0009 0.0370 0.42748 0.08664 139.4 0.292
CO2 44.0 547.4 1069.5 0.29 0.2175 0.2250 0.42748 0.08664 333.3 0.274

C1 16.0 343.0 667.0 0.29 -0.0025 0.0110 0.42748 0.08664 201.6 0.286
C2 30.1 549.6 706.6 0.29 0.0589 0.0990 0.42748 0.08664 332.7 0.279
C3 44.1 665.7 616.1 0.29 0.0908 0.1520 0.42748 0.08664 416.2 0.276
I-C4 58.1 734.1 527.9 0.29 0.1095 0.1860 0.42748 0.08664 471.1 0.282
N-C4 58.1 765.2 550.6 0.29 0.1103 0.2000 0.42748 0.08664 491.1 0.274
I-C5 72.2 828.7 490.4 0.29 0.0977 0.2290 0.42748 0.08664 542.4 0.272
N-C5 72.2 845.5 488.8 0.29 0.1195 0.2520 0.42748 0.08664 557.0 0.268

C6 83.6 921.8 475.0 0.29 0.1343 0.2507 0.42748 0.08664 608.7 0.269 76.9
C7 96.7 983.9 439.7 0.29 0.1452 0.2859 0.42748 0.08664 660.5 0.264 90.1
C8 110.4 1038.7 404.8 0.29 0.1568 0.3256 0.42748 0.08664 709.5 0.260 103.9
C9 123.8 1087.9 372.5 0.29 0.1744 0.3675 0.42748 0.08664 755.8 0.256 117.5
C10 137.1 1131.7 344.6 0.29 0.1913 0.4088 0.42748 0.08664 798.5 0.252 130.8
C11 150.3 1171.4 320.3 0.29 0.2074 0.4496 0.42748 0.08664 838.5 0.248 144.0
C12 163.4 1207.6 299.0 0.29 0.2227 0.4898 0.42748 0.08664 875.9 0.245 157.2
C13 176.5 1240.9 280.3 0.29 0.2370 0.5293 0.42748 0.08664 911.0 0.242 170.3
C14 189.4 1271.6 263.8 0.29 0.2505 0.5681 0.42748 0.08664 944.1 0.239 183.3
C15 202.3 1300.0 249.1 0.29 0.2630 0.6062 0.42748 0.08664 975.3 0.236 196.2
C16 215.1 1326.5 236.0 0.29 0.2746 0.6436 0.42748 0.08664 1004.7 0.233 209.0
C17 227.7 1351.2 224.3 0.29 0.2854 0.6803 0.42748 0.08664 1032.6 0.230 221.7
C18 240.3 1374.4 213.8 0.29 0.2953 0.7163 0.42748 0.08664 1059.1 0.227 234.3
C19 252.8 1396.2 204.3 0.29 0.3045 0.7516 0.42748 0.08664 1084.2 0.224 246.8
C20 265.1 1416.7 195.7 0.29 0.3129 0.7863 0.42748 0.08664 1108.1 0.222 259.2
C21 277.4 1436.1 187.9 0.29 0.3206 0.8203 0.42748 0.08664 1130.9 0.219 271.5
C22 289.6 1454.4 180.8 0.29 0.3277 0.8536 0.42748 0.08664 1152.6 0.216 283.7
C23 301.6 1471.9 174.3 0.29 0.3342 0.8863 0.42748 0.08664 1173.3 0.214 295.8
C24 313.5 1488.5 168.4 0.29 0.3402 0.9183 0.42748 0.08664 1193.1 0.212 307.8
C25 325.4 1504.3 163.0 0.29 0.3457 0.9498 0.42748 0.08664 1212.1 0.209 319.7

C26p 381.5 1571.8 141.9 0.29 0.3663 1.0943 0.42748 0.08664 1293.9 0.199 331.5
10 IPTC-19596-MS

Table 4b—BIPS of the "generic" SRK EOS model used for example simulations in this paper. Values not presented in this table is 0.

Comp H2S N2 CO2 C1

H2S 0
N2 0 0
CO2 0 0 0
C1 0.08 0.02 0.12 0
C2 0.07 0.06 0.12 0
C3 0.07 0.08 0.12 0
I-C4 0.06 0.08 0.12 0
N-C4 0.06 0.08 0.12 0
I-C5 0.06 0.08 0.12 0
N-C5 0.06 0.08 0.12 0
C6 0.05 0.08 0.12 0
C7 0.03 0.08 0.1 0.0149
C8 0.03 0.08 0.1 0.0173
C9 0.03 0.08 0.1 0.0197
C10 0.03 0.08 0.1 0.0219
C11 0.03 0.08 0.1 0.024
C12 0.03 0.08 0.1 0.026
C13 0.03 0.08 0.1 0.0278
C14 0.03 0.08 0.1 0.0295
C15 0.03 0.08 0.1 0.0311
C16 0.03 0.08 0.1 0.0325
C17 0.03 0.08 0.1 0.0339
C18 0.03 0.08 0.1 0.0352
C19 0.03 0.08 0.1 0.0364
C20 0.03 0.08 0.1 0.0374
C21 0.03 0.08 0.1 0.0385
C22 0.03 0.08 0.1 0.0394
C23 0.03 0.08 0.1 0.0403
C24 0.03 0.08 0.1 0.0411
C25 0.03 0.08 0.1 0.0419
C26p 0.03 0.08 0.1 0.045

Method 1.
This method is highly dependent on having a good estimate of the wellstream composition beforehand (seed
feed). This is shown in Fig. 3 where the wellstreams are predicted using this method for the 4 reservoir-
fluid systems studied. It is assumed that accurate seed feed estimates were available (updated) only once
per month, at days 1, 31, 61, 91 and 121. In a practical context, this would represent the days where fluid
sampling data is available (ref. Method 2 and Method 3). For the days in which seed feeds are not available
the method uses the latest available seed feed. This method matches the separator GOR every single day
and one can see from Fig. 3a–b that both wellstream C1 and C7+ are predicted well over time; because the
split between C1 and C7+ in the wellstream is a very good representation of the GOR. However, the predicted
wellstream composition does not reflect the changes in characterization of the heavier components (MWC7+
IPTC-19596-MS 11

and API) which are only predicted accurately when a good seed feed estimate is available. This is specially
important for wells/reservoirs in which the produced stream will change significantly over the lifetime of
the well, i.e. gas condensates and near-critical fluids, as seen in Fig. 3c–d. These results also suggest that
in practice it is important to collect extensive fluid sampling data more frequently in time periods were the
produced fluid properties change rapidly (i.e. rapidly changing GORs).

Figure 3a—Wellstream C1 composition comparison, actual (simulated) vs predicted wellstream with method 1.

Figure 3b—Wellstream C7+ composition comparison, actual (simulated) vs predicted wellstream with method 1.
12 IPTC-19596-MS

Figure 3c—Wellstream molecular weight of C7+ vs time, comparison

of actual (simulated) vs predicted wellstream with method 1.

Figure 3d—Stock-tank liquid API vs time, comparison of actual (simulated) vs. predicted wellstream with method 1.

Method 2.
To utilize this method, separator-gas compositions and stock-tank liquid APIs should be available in addition
to separator conditions and separator volumetric rates. The calculations for this method are performed
assuming a single seed feed, meaning that there is only one accurate estimate of the wellstream composition
throughout the life of the well, but with liquid API and separator gas compositions available daily. It is
unlikely that this type of data is available every single day in the field; e.g. separator-gas compositions tend
to be only available for a few points in time. However, the purpose of these calculations is to analyze and
compare the accuracy of this method when applied to the different reservoir fluid systems presented in this
paper. The predictions using method 2 are shown in Fig. 4. Again, this method matches the separator GOR
every single day and one can see from Fig. 4a–b that both wellstream C1 and C7+ are predicted correctly
over time. Because the liquid API is available and regressed on every single day, these predictions are also
spot on. This also results in a better estimation of the wellstream MWC7+ compared to Method 1, as stock-
tank liquid API and MWC7+ are tightly coupled. This scheme relies on non-linear regression, which makes
it slower than Method 1. However, since it is typically only a few datapoints with this data available over
the lifetime of a well, speed is not necessarily that important.
IPTC-19596-MS 13

Figure 4a—Wellstream C1 composition comparison, actual (simulated) vs predicted wellstream with method 2.

Figure 4b—Wellstream C7+ composition comparison, actual (simulated) vs predicted wellstream with method 2.

Figure 4c—Wellstream molecular weight of C7+ vs time, comparison

of actual (simulated) vs predicted wellstream with method 2.
14 IPTC-19596-MS

Figure 4d—Stock-tank liquid API vs time, comparison of actual (simulated) vs predicted wellstream with method 2.

Method 3.
This method requires both measured separator-oil and separator-gas compositions, in addition to separator
conditions and separator volumetric rates. It is very important that when this method is used, the C7+
molecular weights of the separator fluids are reported, especially for the separator oil. If this is not reported,
then the method will not work properly unless there is additional measured data characterizing the C7+
content of the oil, e.g. liquid API. The calculations for this method are performed assuming that the
separator-oil and separator-gas compositions are available daily. This method honors the material balance
and contains all the information needed to fully reconstruct the wellstream without any regression involved,
just by recombining the separator fluids at the measured separator GOR. The predicted wellstreams by this
method are a proof of the method's accuracy. These results are summarized in Fig. 5 where it is clear that
all the predictions are whitin line thickness of the actual values.

Figure 5a—Wellstream C1 composition comparison, actual (simulated) vs predicted wellstream with method 3.
IPTC-19596-MS 15

Figure 5b—Wellstream C7+ composition comparison, actual (simulated) vs predicted wellstream with method 3.

Figure 5c—Wellstream molecular weight of C7+ vs time, comparison

of actual (simulated) vs predicted wellstream with method 3.

Figure 5d—Stock-tank liquid API vs time, comparison of actual (simulated) vs predicted wellstream with method 3.

Example 2: Field Example with Three Available Well Tests. Fig. 6 shows a field example were
compositional tracking has been applied. As visualized in Fig. 6a, only three days through time have
"extensive fluid sampling data" available (Rp,sep, xi, yi, MWO,C7+, Tsep, psep). For these three days, Method
16 IPTC-19596-MS

3 has been applied. For all the remaining data points, in which "only production data" is available (Rp,sep,
Tsep, psep), Method 1 has been applied. The three "welltest" days represent points in time where the seed
feed is updated for Method 1.

Figure 6a—Separator temperature and pressure vs. time for Example 2.

Figure 6b—Producing GOR vs. time for Example 2.

Operationally, this field example shows behavior commonly observed in tight unconventionals. First,
the separarator conditions change significantly over time, as seen in Fig. 6a. Here it is worth noting how
separator pressure decreases until it (about) stabilizes and how separator temperature changes with the
seasons. Secondly, the producing separator GORs increase significantly with time; behavior observed in
many "shale" basins in North Amercia. The GOR of the well starts at ~10,000 scf/sep.bbl and stabilizes
at around 25,000–30,000 scf/sep.bbl before it increases to 35,000 scf/sep.bbl at the end of the available
time series. The predicted C1 and C7+ wellstream compositions and how these values change over time are
presented in Fig. 6c, while Fig. 6d shows the associated, predicted liquid API over time. Note the "stepwise"
liquid API predictions, further discussion on this item will come in the next section.
IPTC-19596-MS 17

Figure 6c—Predicted wellstream C1 and wellstream C7+ vs. time for Example 2.

Figure 6d—Calculated liquid API vs. time for Example 2. Note the
"stepwise" change implying three different seed feed compositions.

Discussion
Seed Feed Dependency in Tight Unconventionals.
Hoda and Whitson (2013) argue that because equilibrium K-values (Ki ≡ yi/xi) are not very dependent on the
overall composition at separator conditions Ki (p, T, zi), ≈ Ki(p, T), the seed feed estimated composition is not
important for the accuracy of the method. The seed feed can represent any reasonable composition produced
from a well. As the equilibrium-phase compositions are mainly dependent on the separator conditions (p,
T), it is most important to have an accurate record of pressure and temperature together with the rates (at
the same time). However, they mention that the seed feed composition can be important in compositionally
sensitive processes (e.g. gas EOR) or when modeling large process facilities. The results presented in this
paper suggest that accurate seed feed compositions are very important when produced fluid properties
change significantly with time – i.e. for near-critical and gas condensate fluid systems. For these fluid
systems, seed feed compositions need to be updated regularly, requiring frequent measurement of fluid
properties.

Liquid API and its Dependency on Seed Feed.

Each seed feed composition is associated with an implicit stock-tank liquid API. In other words, if a constant
seed feed is used between two fluid sampling points, a constant stock-tank liquid API is implicitly assumed
for the associated time interval (for a given, fixed surface procees, i.e. #stages, psep, Tsep). Even though all
18 IPTC-19596-MS

the actual measured data is honored (Rp,sep, Tsep and Psep) between two fluid sampling points, we know from
experience (field data and simulation) that there is a relationship between the producing GOR and stock-
tank liquid API. Hence, if the GOR is changing significantly between two production data points, it will
not be possible to predict a liquid API (or C7+ distribution) that makes sense for that time interval, using
a single seed feed.
This "issue" can be observed in the calculations presented in Fig. 6d and Fig. 3d, where the different
"steps" on the calculated liquid API plot each represent a given associated seed feed. In the field example
(Fig. 6d) the small variations in liquid API within each "step" are due to changing separator conditions. Only
three actual stock-tank API measurements were available over time, each forming the basis for the next seed
feed. Even though all the measured values are honored, this type of "produced" liquid API performance
is not expected. In real-life, one would expect a smoother stock-tank liquid API performance over time as
exemplified by the simulated results, labeled "actual" in Fig. 3d, where one can also note how stock-tank
liquid API changes become more significant for fluids with higher GOR.

Rp-γAPI Curves.
One way to remove the "stepwise" liquid API behavior is to develop a producing GOR (or CGR) vs.
liquid API relationship for a well and its associated fluid system. The GOR is used to pick an estimate for
liquid API, which again is an indirect measurement of MWC7+. The strong relationship between liquid API
and wellstream MWC7+ is exemplified in Fig. 7. MWC7+ is yet again an indirect measurement of the C7+
component distribution and can be regressed on to match an appropriate liquid API using a gamma model
with fixed shape (α) and bound (η).

Figure 7—Simulated relationship between producing GOR (Rp) and stock-

tank liquid API for different fluid systems referenced in this paper.

There are several ways of generating GOR (or CGR) vs. liquid API curves in practice. One way is to
run a compositional reservoir simulator on BHP control, as exemplified in Fig. 8. For gas condensate wells,
one can generate this curve with a simulated CVD experiment, as condensate that drops out of solution in
the reservoir remains immobile and not produced. Hence, one would only need the solution CGR in the
gas removed from each stage in the CVD experiment vs. the liquid API for a specified surface process.
Another alternative would be to use field measured values of liquid API and producing GOR. However,
this would not necessarily be very practical, as these would need to be normalized to a consistent surface
process (#stages, psep, Tsep). It is not important to generate such a relationship for low GOR oils (Rs<1000
scf/STB) as liquid API is not changing significantly with GOR, as seen for the black oil cases presented
in Fig. 7 and Fig. 8.
IPTC-19596-MS 19

Figure 8—Simulated relationship between stock-tank liquid API and Wellstream

Molecular Weight of C7+ for different fluid systems referenced in this paper.

Prediction Improvements with Rp-γAPI Curves.

One way to study the benefits of using Rp-γAPI curves in practice, is to analyze how it changes the predictions
presented for Example 1 (Method 1). In this case only a predicted stock-tank liquid API is available in
addition to production data, so there is more data than what Method 1 uses, but less than what is needed
for the regression in Method 2 (missing separator gas compositions). In such a case, the seed feed can
be adjusted by altering its C7+ molecular weight in an iterative process to match the measured API. The
proposed regression routine uses the 2 known data (Rp,sep, γAPI) to solve the system of equations with 2
unknowns . The system of equations is solved in the EOS-based software PhazeComp (Zick
Technologies). The results from this method are shown in Fig. 9, where the calculations are performed
assuming a single seed feed (at day 1) and API data available daily. The results show that there is an
improvement in the prediction of the wellstream when compared to Method 1, because the predicted
wellstream compositions correctly predict API and GOR. However, the MWC7+ is still not predicted perfectly
(when compared to the "actual" values from the reservoir simulation), mainly because the seed feed is not
changing over time.

Figure 9a—Wellstream C1 composition comparison, actual (simulated) vs predicted wellstream with alternative method 1.
20 IPTC-19596-MS

Figure 9b—Wellstream C7+ composition comparison, actual (simulated) vs predicted wellstream with alternative method 1.

Figure 9c—Wellstream molecular weight of C7+ vs time, comparison of

actual (simulated) vs predicted wellstream with alternative method 1.

Figure 9d—Stock-tank liquid API vs time, comparison of actual (simulated) vs predicted wellstream with alternative method 1.

Practical Use of Wellstream Prediction.

In practice, all the methods presented will be used throughout the lifetime of a well, because at different
points in time, different sources of data are available. What this paper is showing is that one can and should
IPTC-19596-MS 21

make use of all the data that is available, as any and all data available will assist in the correct prediction
of the wellstream composition. The accuracy of the methods increases as more data becomes available;
but even the least accurate of the methods will give a reasonable wellstream prediction. It is also crucial
to "guide" Method 1, by always providing the best estimate of seed feed compositions available. This best
estimate would come from a wellstream composition calculated using any of the other methods.
Another key observation is that more detailed fluid sampling data (Method 2 and Method 3) should be
collected in time periods where the producing GOR is changing rapidly. Rapid change in producing GOR
implies that the produced fluid properties (γAPI, Mo,sep, γg) are changing fast.

Conclusion
1. This paper shows that accurate wellstream compositions can be predicted by using a field-wide EOS
model together with production and well-test data. The accuracy of the different methods depends on
(1) the amount of available measured data, (2) the quality of the available measured data, and (3) how
well the EOS model captures the phase characteristics of the relevant fluid.
2. The proposed methods provide a tool to consistently, accurately, and cost-efficiently calculate daily
wellstream compositions for all wells. These methods make wellstream compositions readily available
for unconventional fields with thousands of wells.
3. The presented methods are not competing procedures, but each method represents the "best one can
do" with the associated available data. This paper summarizes and compares these methods:
i. Method 1 requires separator rates and separator conditions. This procedure predicts wellstream
C1 and C7+ accurately. However, if GOR changes rapidly, the method requires frequent seed
feed updates (fluid sampling) to predict wellstream MWC7+ and liquid API accurately over
time.
ii. Method 2 requires separator rates, separator conditions, separator gas compositions, and stock-
tank liquid API. This procedure matches wellstream C1 and C7+ accurately, in addition to liquid
API (as this is available). The predicted wellstream MWC7+ is not perfectly matching actual
values, but the relative changes are captured with the regression approach controlled by the
API.
iii. Method 3 requires separator rates, separator conditions, and separator-gas and oil
compositions. All properties are available for a direct recombination, and all the wellstream
compositions are predicted perfectly when compared to actual (reservoir simulation) data.
4. Operationally, more detailed fluid sampling data (Method 2 and Method 3) should be collected in time
periods where the producing GOR is changing rapidly. Rapid change in producing GOR implies that
the produced fluid properties (γAPI, Mo,sep, γg) are changing fast.
5. Relationships between producing GOR and liquid API can be introduced to achieve a "smooth" liquid
API prediction through time, even in cases in which only production data is available.

Acknowledgements
We would like to thank Innovation Norway and the Norwegian Research Council for continuing support of
our technology development related to compositional analysis.

Nomenclature
Bo = formation volume factor (FVF), res.bbl/STB
fg = molar gas fraction = (zi-xi)/(yi-xi), fraction
Ki = equilibrium K-values/ratios (Ki = yi/xi)
γAPI = stock-tank liquid API
22 IPTC-19596-MS

γg = gas specific gravity

ρo,sep = density of separator oil (lbm/ft3)
Mo,sep = molecular weight of separator oil (g/mole)
ni = molar component rate
pavg = average reservoir pressure, psia
psat = saturation pressure, psia
pb = bubblepoint pressure, psia
pd = dewpoint pressure, psia
pwf = flowing bottomhole pressure, psia
pi = initial reservoir pressure, psia
qom = metered oil rate
qgm = metered gas rate
Sgc = critical gas saturation
Rsi = initial solution GOR, scf/STB
Rp = producing GOR, scf/STB
Rp,sep = separator producing GOR, scf/STB
Tres = reservoir temperature, F
vg = EOS calculated seed feed separator gas molar volume
vo = EOS calculated seed feed separator oil molar volume
xi = separator oil composition, mol%
yi = separator gas composition, mol%
zi = wellstream composition, mol%
zsi = "seed feed" composition – estimate of actual wellstream composition, mol%
zRi = in-situ reservoir composition, mol%

References
Carlsen, M.L., Whitson, C. H, Dahouk, M. M, Younus, B., Yusra, I., Kerr, E., Nohavitza, J., Thuesen, M., Drozd, J.H.,
Ambrose, R. and Mydland, S. 2019a. Compositional Tracking of a Huff-n-Puff Project in the Eagle Ford. Presented
at the Unconventional Resources Technology Conference held in Denver, Colorado, USA, 22-24 July. URTeC-539.
Carlsen, M.L., Whitson, C. H, Alavian, A., Martinsen. S.Ø., Mydland, S., Singh, K. Younus, B. and Yusra, I. 2019b. Fluid
Sampling in Tight Unconventionals. Presented at the SPE Annual Technical Confernce and Exhibition held in Calgary,
Alberta, Canada, held in Calgary, Canada, 30 Sep– 2 Oct. SPE-196056-MS.
Carlsen, M.L., Dahouk, M.M., Mydland, S. and Whitson, C.H. 2020. Predicting Wellstream Compositions in Tight
Unconventionals. Presented at 2020 International Petroleum Technology Conference held in Dhahran, Kingdom of
Saudi Arabia, 13-15 Jan. 20IPTC-19596-MS.
Ebbrell, H.K., Rimmer, C.P., Boote, D.N., Royce, P.W., Sutton, A.F. and Macknay, M.C. 1984.
The Composition of Statoil (Norway) Gas Well Troll 31/6-6, July. https://factpages.npd.no/pbl/
wellbore_documents/127_05_31_6_6_The_Composition_of_Statoil_Gas_Well.pdf
Fevang, Ø. And Whitson, C. H. 1994. Accurate Insitu Compositons in Petroleum Reservoirs. Presented at the European
Petroleum I, London, UK, 25-27 October. SPE-28829-MS. https://doi.org/10.2118/28829-MS
Hoda, M. F. & Whitson, C. H. 2013. Well Test Rate Conversion to Compositional Wellstream. Society of Petroleum
Engineers. SPE-164334-MS, March 10. Doi: 10.2118/164334-MS.
Hoda, M. F., Hoffmann, A., & Kuntadi, A. 2017. Advances in Molar Wellstream Computation from Well Test
Measurements. Society of Petroleum Engineers, May 5. SPE-185988-MS. Doi: 10.2118/185988-MS.
Jones, R.S. 2017. Producing-Gas/Oil-Ratio Behavior of Multifractured Horizontal Wells in Tight Oil Reservoirs. SPE
Reservoir Evaluation & Engineering, Volume 20, Issue 03, August. SPE-184397-PA. https://doi.org/10.2118/184397-
PA.
Martinsen, S.Ø., Castiblanco I.A., Osorio, R. and Whitson, C.H. 2010. Advanced Fluid Characterization of Pauto
Complex, Colombia. Presented at the SPE Annual Technical Conferance and Exhibition held in Florence, Italy, 19-22
September. SPE-135085-MS. https://doi.org/10.2118/135085-MS.
IPTC-19596-MS 23

Singh, K., Mantatzis, K., Whitson, C. H., & Benjemia, R. (2011, January 1). Reservoir Fluid Characterization and
Application for Simulation Study. Society of Petroleum Engineers. doi:10.2118/143612-MS
Whitson, C. H. (1983). Characterizing Hydrocarbon Plus Fractions. Society of Petroleum Engineers Journal, 23 (04).
Doi: 10.2118/12233-PA
Whitson, C.H. and Brulé, M. R. 2000. Phase Behavior, Monograph Series, Society of Petroleum Engineers.
Whitson, C.H. and Sunjerga, S. 2012. PVT in Liquid-Rich Shale Reservoirs. Presented at the SPE Annual Technical
Conference and Exhibition, San Antonio, Texas, 8-10 October. SPE-155499-MS. https://doi.org/10.2118/155499-MS.
Yang, T., Arief, I. H., Niemann, M., & Houbiers, M. (2019, July 31). Reservoir Fluid Data Acquisition Using
Advanced Mud Logging Gas in Shale Reservoirs. Unconventional Resources Technology Conference. doi:10.15530/
urtec-2019-383
Yang, T., Basquet, R., Callejon, A., Van Roosmalen, J. J., & Bartusiak, B. (2014, August 25). Shale PVT
Estimation Based on Readily Available Field Data. Unconventional Resources Technology Conference. doi:10.15530/
URTEC-2014-1884129
Younus, B., Whitson, C.H., Alavian, A., Carlsen, M.L., Martinsen, S.Ø. and Singh, K. 2019. Field-wide Equation of State
Model Development. Presented at the Unconventional Resources Technology Conference held in Denver, Colorado,
USA, 22-24 July. URTeC-551.
Zick, A.A. 2013. Equation of State Fluid Characterization and Analysis of the Macondo Reservoir Fluids. Prepared on
behalf of the United States as a part of an Expert Report US vs. BP Exploration and Production Inc. et al. USA, 22
March 2013. http://www.mdl2179trialdocs.com/releases/release201311071200013/TREX-011490R.PDF