Journal of Petroleum Science and Engineering 120 (2014) 102110

Contents lists available at ScienceDirect

Journal of Petroleum Science and Engineering

journal homepage: www.elsevier.com/locate/petrol

A viscosity prediction model for Kuwaiti heavy

crude oils at elevated temperatures
Osamah Alomair n, Adel Elsharkawy, Hassan Alkandari
Petroleum Engineering Department, College of Engineering and Petroleum, Kuwait University, P.O. Box # 5969, Safat 13060, Kuwait

art ic l e i nf o

a b s t r a c t

Article history:
Received 7 September 2013
Accepted 31 May 2014
Available online 11 June 2014

Viscosity is a key uid property for characterization, evaluation, management and development of
petroleum reservoirs. The accurate prediction of dynamic viscosity will be helpful for heavy oil recovery
methods including primary production, thermal production, and enhanced oil recovery (EOR). Reservoir
oil viscosity is usually measured isothermally at reservoir temperature. However, at temperatures other
than reservoir, dynamic viscosity is estimated by empirical correlations. Most of the published
correlations have been performing well at the reservoir temperature, especially for conventional crudes.
However, the published literature has lack of reliable methods for viscosity estimation due to an acute
shortage of dead oil data at elevated temperatures. These methods are essential and employed in
planning thermal recovery methods (Kuwait as well as worldwide). In this study, the API gravity and
viscosity of 50 dead crude oil samples collected from various areas of Kuwaiti oil elds were measured.
These oil samples have API gravity ranging from 101 to 201. The viscosities were determined at
temperatures ranging from 20 1C to 160 1C. Consequently the results of the heavy oil viscosity data were
used to develop a reliable model and to compare the proposed model with the published models. Both
quantitative and qualitative analytical methods were implemented using statistical parameters and
performance plot, respectively. From the general evaluation it has been shown that the proposed model
has the lowest average absolute error of 11.04% and highest coefcients of correlation of 92% for training
and 96% for the testing data. The performance of the proposed correlation has also been tested using
dead heavy crude oil data from the region as well as various parts of the world. Compositional data of
heavy oil viscosity has been used to compare predicted viscosity from the proposed correlation with that
from LorenzBrayClark (LBC) and Pederson models. These comparisons show that the proposed
correlation performed better than the other correlations, corresponding state and EOS-based methods
for the dead heavy crude oils considered.
& 2014 Elsevier B.V. All rights reserved.

Keywords:
correlations
dead oil viscosity
heavy oil viscosity
Kuwait

1. Introduction
The properties of petroleum reservoir uids are essential for
optimizing their production and transportation. Viscosity plays an
important role in the calculations of uid ow through reservoir
rock, pressure loss (with implications for the designs of tubing and
pipelines), and the design of surface facilities, reservoir simulations, and predictions of oil recovery. The thermal oil recovery of
heavy crude oils is designed to meet the industry demand for
improving oil production. Because it is necessary to consider the
variation of viscosity with temperature in engineering activities,
including piping and pipeline construction for enhanced transportation, thermal expansion is the key property for increasing the
productivity of heavy oils. Modern reservoir engineering practices
require accurate information concerning the thermodynamic and
n

Corresponding author.
E-mail address: osamah@dr-alomair.com (O. Alomair).

http://dx.doi.org/10.1016/j.petrol.2014.05.027
0920-4105/& 2014 Elsevier B.V. All rights reserved.

transport uid properties, in addition to the reservoir rock properties, to perform material balance calculations. Viscosity is often the
limiting factor in heavy oil production.
The viscosity of oils has been studied for many years. Although the
viscosity is affected by pressure and gas content, it is primarily a
function of oil gravity and temperature (Batzle et al., 2004). Accurate
prediction of the physical properties of oil is required to design
appropriate recovery, storage, transportation, and processing systems
for crude oil handling (Quail et al., 1987). Heavy oils are characterized
by high viscosities, ranging from 100 CP to 10,000 CP at the reservoir
temperature, as dened by the World Petroleum Congress, and have
low API gravity, ranging from 101 to 221, as dened by the U.S.
Department of Energy (Nehring et al., 1983). Heavy oils have a low API
gravity compared with conventional oil and are particularly known for
the difculty associated with achieving accurate measurements of
their uid properties.
The uncertainty of heavy oil uid property measurements
affects the quality of the data, which in turn affects the accuracy

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

Nomenclature
API
T
MRE

AARE
SD
RMSE
m

specic gravity @ 15.56 1C (deg)

temperature, (1C)
mean relative error (%)

of the production forecast (Zabel et al., 2008) and the recovery

processes, such as the steam-assisted gravity drainage (SAGD)
process. An optimal recovery process must be a function of oil
mobility (permeability/viscosity ratio). Therefore, a comprehensive and well-dened mapping of the reservoir oil mobility based
on viscosity is essential for the effective exploration and design
optimization of production strategies, particularly in biodegraded
heavy oil and bitumen reservoirs (Adams et al., 2008, 2009).
For live oils, the variation in dynamic viscosity with changes
in temperature and pressure is typically predicted empirically
(Sattarina et al., 2007).
The reservoir oil viscosity is typically measured isothermally at
the reservoir temperature. However, at temperatures other than
the reservoir condition, these data are estimated using empirical
correlations (Naseri et al., 2005). The viscosity of crude oil varies
depending on its origin and type, as well as the nature of its
chemical composition, particularly the polar components, for
which intermolecular interactions can occur. For this reason,
developing a comprehensive model of viscosity to include different regions of the world appears to be a difcult task. The various
approaches to addressing this issue have been the topic of
numerous studies, as reported in the following section.

2. An overview of published oil viscosity correlations

This study conducted a detailed literature review of various
correlations to estimate the viscosity of crude oil. These correlations can be divided into three categories, including dead, saturated, and under saturated. For the purpose of this research, only
dead, heavy od (oil with no gas in the solution) correlations will
be discussed.
Table 1 explains the different published correlations that can be
considered for the purpose of comparing Kuwait dead-heavy oils.
These correlations were divided into three groups based on their
range of API and/or temperature values. Group A comprises the
correlations that are outside the range of the viscositytemperatureAPI gravity data described in this study and includes the
Glaso (1980), Labedi (1992), Petrosky and Farshad (1995) and
Table 1
Comparison of the range of application of the dead-heavy oil correlations.
Developer

API range
(deg)

Temperature range
(1C)

Group A

Glaso (1980)
Labedi (1992)
Petrosky and Farshad ( 1995)
Elsharkawy and Alikhan (1999)

20.148.1
32.248
25.446.1
19.948

10149
38152
46142
38149

Group B

Standing (1981)
Egbogah and Ng (1990)
De Ghetto et al. (1995)
Bennison (1998)
Hossain et al. (2005)

10.152.5
558
1022.3
1022
7.121.8

38149
1580
24146
10121
093

Group C

Beggs-Robinson (1975)
Al-Khafaji et al. (1987)
Kartoatmodjo (1990)
Naseri et al. (2005)

1658
1551
14.458.9
1744

24146
16149
24160
41246

103

absolute average relative error (%)

standard deviation
root mean square error
dynamic viscosity (CP)

Elsharkawy and Alikhan (1999) correlations. Group B comprises

the correlations that fully cover the range of the viscosity
temperatureAPI gravity data and includes the Beal (modied
form as Standing, 1981), Egbogah and Ng (1990), De Ghetto et al.
(1995), Bennison (1998) and Hossain et al. (2005) correlations.
Group C includes the correlations that partially cover the range of
these parameters from the entire databank and includes the Beggs
and Robinson (1975), Al-Khafaji et al. (1987), Kartoatmodjo (1990)
and Naseri et al. (2005) correlations.
Many correlations utilize the oil API gravity and temperature to
determine dead oil viscosity. Beal (1946) presented a dead oil
dynamic viscosity correlation as a function of API gravity and
temperature using 655 viscosity data points collected from 492 oil
elds in the United States at a temperature of 38 1C and 98 data
points at temperatures above 38 1C.
Using the same input variables, Beggs and Robinson (1975)
published a similar correlation that was developed with 460 data
points from 93 different oil samples.
Glaso (1980) realized that parafnic crudes and naphthenic
crudes have the same API gravity but not the same viscosity at a
given temperature. For the same reason, the author published a
dead oil viscosity correlation with the suggestion of an adjustment
to the API gravity.
Al-Khafaji et al. (1987) presented a modied form of Beal's
correlation for predicting the viscosity of dead crude oil from
the Middle East. Egbogah and Ng (1990) presented two different correlations for predicting the viscosity of dead oil. The rst
correlation was a modied form of the Beggs and Robinson (1975)
correlation using 394 data points, and in the second correlation,
Egbogah and Ng introduced the pour point (Tp) as a new parameter for estimating the dead oil viscosity (Egbogah and Ng, 1990;
De Ghetto et al., 1995). Additionally, Svrcek and Mehrotra (1988)
presented a one-parameter viscosity equation for bitumen.
The application of dead oil viscosity correlations to crude oils
from different sources results in signicant errors. These deviations are attributed to the difference in asphaltic and parafnic oils
and/or the mixed nature of the oils (Sattarina et al., 2007). Using a
large databank of crude oils from different parts of the world,
Kartoatmodjo presented a modied form of Glaso's correlation
(Kartoatmodjo 1990). Labedi (1992) presented a correlation for
African dead oils, in particular, from Libya. Puttagunta et al. (1988)
published a viscosity correlation for Athabasca and Cold Lake
heavy oil and bitumen that depends on a single-point viscosity
measurement at a temperature of 30 1C and atmospheric pressure.
Another correlation was later developed for heavy oil viscosity
with dissolved gas (Puttagunta et al., 1993). Petrosky and Farshad
(1995) presented a correlation for estimating the viscosity of dead
oil recovered from the Gulf of Mexico. De Ghetto et al. (1995)
published a set of two modied correlations for dead oil (extra
heavy oil and heavy oil) viscosity predictions, and Bennison (1998)
presented a new correlation for the viscosity of heavy dead oil
from the North Sea. Elsharkawy and Alikhan (1999) presented
crude oil viscosity correlations for Middle East crudes. Argillier
et al. (2001) analyzed the rheology of heavy oil with contents of
asphaltene and resin by dividing heavy oil crudes into two groups
as a non-colloidal liquid (the maltene) and a dark brown powder
(the asphaltene); their rheological experiment with the mixture of

104

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

maltene and asphaltene demonstrated that the viscosity of heavy

oil increased signicantly above a critical asphaltene concentration. Dindoruk and Christman (2001) developed an empirical new
dead oil viscosity correlation using data from the Gulf of Mexico.
Naseri et al. (2005) published a correlation for the prediction of
the viscosity of Iranian dead oil viscosity. Hossain et al. (2005)
established a heavy oil databank based on three heavy oil datasets
to assess and develop correlations for each set. The dataset
provided by Chevron also included the composition, shear rate,
and saturates, aromatics, resins, and asphaltenes (SARA) analysis
of the oils. Bergman and Sutton (2007) developed a new correlation using a large database from conventional PVT reports, crude
oil assays, and the literature. Additionally, these authors improved
their proposal by introducing the Watson characterization factor
(Kw), together with the oil API gravity and temperature, in the
viscosity correlation.
It is apparent from the aforementioned survey that the literature
contains an abundance of information and a database for viscosity
measurements, although mainly for saturated oils. However, there
are still several methods, such as corresponding state (CS) models
(Little and Kennedy, 1968; Ely and Hanley, 1981) and equation-ofstate-based (EOS) models (Lohrenz et al., 1964; Pedersen et al., 1984)
that are used to calculate the viscosity of heavy oil. However, the
prediction of heavy oil viscosity is limited at elevated temperatures.
The viscosity of heavy oil is rarely measured at a temperature higher
than the formation temperature.
In this study, the dynamic viscosity data for several dead heavy
oil samples are measured at the original formation temperature
and at elevated temperatures, that is, the temperatures that are
typically anticipated in hot water, cyclic-steam stimulation (CSS),
SAGD, and steam ooding. Considering the conguration of
published dead-heavy-oil correlations, the main objective of this
work is to develop a dead-heavy-oil viscosity correlation for
Kuwaiti crude oils as a function of API gravity and temperature.
Further, this research aims to develop a correlation that can be
employed with condence by engineers (design, production, and
reservoir) for the design of good ow and surface facility, with a
special emphasis on application at elevated temperature.

3. Experimental methods
3.1. Materials and preparation
Fifty heavy Kuwaiti crude oil samples of different API gravity
values were collected in specially designed glass-stoppered bottles
of 0.0025 m3 in capacity and stored at 20 1C. Before analysis, each
sample was degassed, and the bottles were shaken vigorously
using open-air platform shakers to achieve homogeneity. The
homogenized samples were transferred into a separating funnel
and stored for 24 h while waiting for gravity settling. The commercial demulsier Nalco product (USA) was mixed with each
crude oil sample in centrifuge tubes with a volume of 0.0001 m3;
the samples were spun at 400650 rpm for 30 min at 40 1C in a
K60002 automatic oil test centrifuge (Koehler Instrument, USA) to
remove any traces of basic sediments and water. The water
content of each sample was veried using a GD-2122 petroleum
oil water content tester (Karl Fischer Titrator), and the results were
satisfactory, with an average range of 0.010.03%. These puried
crude oil samples were transferred to clean dry bottles, each with
a unique identication tag.

Cambridge Viscosity, Inc., USA). The main purpose of this stateof-the-art equipment is to measure the strength of an electromagnetic eld generated from two magnetic coils inside a stainless steel body. This structure allows the stainless steel piston
inside the measurement chamber to move by magnetic force back
and forth in the uid. The time required for the piston to move a
xed distance (approximately 0.508 cm) is then very accurately
related to the dynamic shear viscosity of the uid in the chamber.
Instrument calibration was performed by a triplicate measurement
of the two reference samples supplied by the manufacturer
in the temperature range of interest, that is, up to 160 1C, with a
reproducibility of 0.92% of the measuring range, and the
estimated uncertainty in the dynamic viscosity measurements
was observed to be no larger than 9  10  3 mPa s, with a condence interval of 95% for all measurements. The densities were
measured at temperature intervals between 20 and 160 1C using a
densimeter (mPDS 2000, Anton paar, GmbH), which functions
according to the oscillating U-tube techniques. The calibration was
performed using dry air and ultra-pure water (S. No. 78169, S.H.
Kalibrier, GmbH products) at the temperature of interest. The
measurement cell is thermostatic with a solid-state thermostat
and two integrated Pt 100 measuring sensors, with a temperature
reproducibility of 10  2 K. Triplicate density measurements
were performed for all samples. The results were averaged,
and the estimated uncertainty of the measurements was within
0.5 kg m  3. After verication, only 41 heavy crude oil samples
with a calculated API gravity ranging from 101 to 201 were used for
the study. Each sample was tested for approximately 12 temperature steps from 20 to 160 1C, and a total of 492 data points were
collected in the viscosity range from 1.784 to 4867 CP.

4. Proposed model
Fig. 1 illustrates the general trend of the experimental viscosity
measurement data at different temperatures, with the API of the
samples ranging from 11.81 to 20.51, corresponding to the entire
viscosity databank. Experimentally, it is observed that most of
the viscosity measurement population is approximately 1000 CP,
which is in agreement with published studies (Oskui and Al Naqi,
2009; Tirtharenu and Al-Sammak, 2011).
The viscosity measurements at different temperatures for the
crude oil samples in this study were used to develop a heavy oil
viscosity correlation. The entire dataset was divided into a training
set composed of 374 viscosity measurements and a testing set
containing 118 measurements. The training and testing sets were

3.2. Viscosity and density measurements

The viscosity of each crude oil sample was measured using
viscosity monitoring and control electronics (VISCO lab 3000,

Fig. 1. Experimental viscosity measurements (API from 101 to 201, Kuwaiti

heavy oils).

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

105

selected randomly. Note that the 118 data points were not used to
develop the proposed models or the correlations considered in
this paper.
Multiple non-linear regressions (the least-squares minimization technique) were employed in a surface tting (3D-type curve)
software to obtain an optimum viscosity model. The following
proposed model is able to simulate the viscositytemperatureAPI
relationship for heavy Kuwaiti crude oil with a correlation coefcient of 95%. The new proposed viscosity model has the following
form:
5:76588
 0:001011:8T 32
lnAPI
 ln1:8T 32

lnln 0:07547

Where is the dynamic viscosity in cp, API is the crude oil API
gravity, and T is the formation temperature in 1C. The model covers
a reasonable API gravity range, from 101 to 201, and is quite
suitable for the anticipated temperatures of Kuwaiti reservoirs
because it covers a wide range, from  4 1C to 177 1C.

Fig. 2. Cross plot of the Kuwaiti heavy crude oil viscosity data of the training set.

5. Results and discussion

In this section, the accuracy of the proposed model used to
simulate the experimentally measured viscosity data of the heavy
crude oil was studied. The viscosity data predictions using the
newly proposed model, in addition to many of the published
correlations, were compared with experimentally measured viscosity data. Quantitative and qualitative analyses were performed
as a part of the comparison. The quantitative analyses include
calculations of relative and absolute errors, standard deviations,
and correlation coefcients. The qualitative analysis includes
viscosity temperature plots and cross plots. The viscosity temperature plots were used to determine whether the model
captured the physical trend of the viscosity change as a function
of both temperature and API gravity. For comparison purposes,
many of the correlations considered in this study (Table 1) have
several limitations in terms of both the API gravity and temperature ranges.
A comprehensive assessment of all of the correlation groups (A,
B, and C) revealed that none of the existing correlations are able to
properly match the Kuwaiti dead-heavy oil viscosity measurements over the entire range of temperatures with a reasonable
degree of accuracy. Therefore, it was necessary to propose a new
correlation capable of predicting the viscosity behavior of heavy oil
at the formation and elevated temperatures with an appropriate
degree of accuracy.
5.1. Assessment of the proposed model
Fig. 2 presents a cross-plot of the measured viscosity data of
the training set versus the corresponding predictions. This gure
indicates that the data points are uniformly distributed along the
unit slope line.
Fig. 3 presents the performance of the proposed model on the
118 data points of the testing set. This gure indicates that most of
the predicted data are signicantly closer to the unit slope line.
5.2. Statistical performance
As mentioned above, many of the correlations considered in
this study have limitations, either for the associated API gravity
range or the temperature range, or were not developed for heavy
crudes. A detailed descriptive statistical analysis was performed
to evaluate the performance of each correlation. The statistical
analysis tools employed for this purpose were the percentage

Fig. 3. Validation viscosity cross-plot of the proposed model (testing data).

means relative error (MRE), absolute average relative error (AARE),

standard deviation (SD), coefcient of correlation (r), coefcient of
determination (r2), and relative mean square error (RMSE). Table 2
presents a comparison of the error analysis obtained for the
training data described in this study for the proposed correlation
versus the published existing correlations.
Table 2 demonstrates that the proposed empirical model has
the smallest average absolute error (25%) and mean relative error
(11%) and the highest correlation coefcient (0.96), followed by
the Bennison correlation. Some of the well-known correlations
that have been frequently used in the petroleum industry to
predict dead oil viscosity, such as the Standing (modied form of
Beals) and Kartoatmodjo correlations, yield exceptionally high
errors and low correlation coefcients. The same error analyses
were applied to the testing dataset.
Table 3 demonstrates that the proposed model produces the
lowest errors and standard deviations and the highest correlation
coefcients of all of the correlations considered in this study. This
result indicates that the proposed model performed signicantly
better than the other models.
5.3. Graphical performance
To further investigate the performance of our proposed correlation, detailed graphical comparisons between the different
published correlations (Groups AC) and the proposed correlation
here were prepared and are discussed in the following sections.

106

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

Table 2
Statistical analysis of the training data.
Model

AARE
(%)

MRE
(%)

SD

r
(Correlation)

r2
(Determination)

RMSE

Group A

Proposed
Glaso
Labedi
PetroskyFarshad
ElsharkawyAlikhan

25.29
103.86
649.00
139.36
93.91

 11.04
 90.22
 628.82
 103.92
 86.24

129.97
121.29
121.33
59.20
238.09

0.96
0.87
0.20
0.67
0.93

0.92
0.76
0.04
0.45
0.86

16.80
53.51
118.34
40.80
84.50

Group B

Standing
Bennison
Hossain
Ghetto
EgbogahNg

217.25
62.90
111.11
72.75
85.55

 180.19
 53.87
 107.15
 63.24
 57.86

53.38
387.90
448.17
203.80
84.59

0.46
0.92
0.92
0.94
0.95

0.21
0.84
0.85
0.88
0.91

47.33
156.09
175.82
70.78
26.05

Group C

BeggsRobinson
Khafaji
Kartoatmodjo
Naseri et al.

68.30
85.28
145.55
77.87

 16.23
 68.74
 131.55
 30.00

811.62
306.61
116.73
47.04

0.83
0.75
0.84
0.87

0.68
0.56
0.70
0.76

457.85
203.03
63.84
22.94

Table 3
Statistical analysis of the testing data.
Model

AARE
(%)

MRE
(%)

SD

r (correlation)

r2 (determination)

RMSE

Proposed
Glaso
Labedi
PetroskyFarshad
ElsharkawyAlikhan

28.08
113.54
783.45
157.21
97.18

 11.81
 101.46
 765.64
 126.72
 89.98

124.42
124.89
143.58
65.01
227.02

0.96
0.87
0.20
0.67
0.93

0.92
0.76
0.04
0.45
0.86

15.70
61.29
140.54
48.11
84.98

Group B

Standing
Bennison
Hossain
Ghetto
EgbogahNg

238.75
60.17
105.86
75.59
90.78

 208.08
 48.90
 102.51
 64.87
 64.54

60.82
380.48
440.01
190.06
82.08

0.44
0.92
0.93
0.91
0.95

0.20
0.84
0.86
0.84
0.89

54.52
166.35
175.82
77.05
26.76

Group C

Beggs-Robinson
Khafaji
Kartoatmodjo
Naseri et al.

66.41
98.62
167.01
87.41

 9.87
 83.68
 154.20
 42.24

683.06
373.03
124.64
47.97

0.80
0.63
0.79
0.84

0.64
0.40
0.62
0.71

408.88
290.01
76.55
25.99

Group A

5.3.1. Comparison with Group A correlations

The performance of the proposed model is compared with that
of the Group A correlations in Fig. 4a, b, and c for low, medium,
and high APIs, respectively. The proposed model performs very
well compared with the Group A correlations, with the measured
viscosities being in good agreement with the proposed model
predictions. The performance curves for the Group A correlations
are plotted for three different APIs (low, medium, and high). For
the low API in Fig. 4a, the Labedi viscosity predictions are most
likely the largest outliers, whereas the remaining Group A correlations, comparatively, overestimate the viscosity of heavy Kuwaiti
crude oils. However, for both the medium and high APIs in Fig. 4b
and c, respectively, the Glaso and PetroskyFarshad predictions are
relatively close to the experimental measurements for some
higher temperatures. This result is not surprising, as these correlations are applied to light crudes. Overall, no single correlation was
able to completely capture the viscosity measurements for the
entire temperature range.

5.3.2. Comparison with Group B correlations

The performance of the proposed model is compared with the
Group B correlations in Fig. 5a, b, and c for low, medium, and high

APIs, respectively. The proposed model clearly outperforms the

commonly published Group B correlations. Fig. 5a demonstrates
that the Bennison and Hossain correlations overestimate the
viscosity predictions because both the Bennison and Hossain are
dened for the same API gravity range as that of the current study,
but if we observe Table 1, these API gravity ranges are not
sufciently compatible with the temperature range. The Standing
correlation exhibits insensitive behavior for the majority of the
temperature range for this particular low-API crude oil. The Group
B correlations are in good agreement with the measured data,
mainly at elevated temperatures. The Egbogah and Ghetto correlations underestimate the viscosity.
Fig. 5b presents the viscosity predictions for medium APIs for
the same group; however, the predictions are poorly dened at
relatively low temperatures. However, after a certain high temperature value, all of the predictions exhibit similar behavior. Both
the Bennison and Hossain models overestimate the viscosity at
temperatures below 66 1C with a similar trend. The Standing and
Ghetto models also overestimate the viscosity, whereas the
Egbogah model underestimates the viscosity. The same behavior
can be observed in Fig. 5c, even for temperatures greater than
82 1C. However, below this temperature, both the Ghetto and
Hossain models yield overestimated predictions. The Egbogah

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

Fig. 4. (a). Comparison of the performance of the Group A correlations for low-API
(13.01) crude oils. (b). Comparison of the performance of the Group A correlations
for medium-API (16.51) crude oils. (c). Comparison of the performance of the Group
A correlations for high-API (19.81) crude oils.

model underestimates the viscosity at relatively low reference

temperatures. A comparison between the proposed model performance and the other correlations for medium crude is presented

107

Fig. 5. (a). Comparison of the performance of the Group B correlations for low API
(13.01) crude oils. (b). Comparison of the performance of the Group B correlations
for medium-API (16.51) crude oils. (c). Comparison of the performance of the Group
B correlations for high-API (19.81) crude oils.

in Fig. 5b. The Group B correlations fail to predict the actual

viscosity at low temperature but predict almost the same viscosity
at temperatures greater than 71 1C. However, the correlation

108

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

proposed in this paper performs better and is in good agreement

with the measured viscosity for the entire temperature range.
The same comparison is considered for high-API crude oils in
Fig. 5c. This gure demonstrates that the proposed model, as well
as the Ghetto and Bennison correlations, was able to match the
measured viscosity for the entire temperature range. However, the
Standing and Egbogah correlations do not match the experimental
data well at low temperatures, which is typical of heavy oil
reservoirs.

5.3.3. Comparison with Group B

The performance of the proposed model is compared with the
Group C correlations in Fig. 6a, b, and c for low, medium, and high
APIs, respectively. The proposed model again exhibits an excellent
t with the measured data. For low API, the Group C correlations
either over- or underestimate the viscosities, whereas the proposed model performs well over the entire temperature range.
However, the Beggs model is in good agreement over the majority
of the temperature range. Fig. 6a demonstrates that for low API,
only the Beggs correlation exhibits a good match with the
measured heavy oil viscosity for the entire temperature range,
whereas both the Khafaji and Kartoatmodjo models display poor
agreement with the measured data. This anomaly can be understood by reviewing Table 1 for the parameter ranges. The viscosity
prediction based on the Naseri correlation is similar to the
experimental measurements over the entire temperature range.
For a medium API, Fig. 6b demonstrates that only the Kartoatmodjo correlation is a good match with the pertinent data range,
whereas both the Beggs and Khafaji correlations overestimate the
experimental viscosity data. The Naseri model is in good agreement only at higher temperatures.
Most of the time, the proposed model performs well compared
with the Group C correlations. Additionally, the Group C correlations are in good agreement for a portion of the elevated
temperature range. Fig. 6c presents the performance of the Group
C correlations for the high API range. The Kartoatmodjo and
Khafaji correlations are in good agreement for temperatures above
54 1C and exhibit the same viscosity prediction trend. The Beggs
and Naseri correlations overestimate and underestimate the viscosity predictions, respectively. The proposed correlation displays
good agreement with the experimentally measured viscosities.
All of the Group C correlations are in an underestimating locus.
However, in a portion of the elevated temperature range, this
group agrees well with the proposed model. In summary, the
reservations and concerns that were raised in the rst qualitative
phase of this study (Alomair et al., 2011) have now been
addressed. An overall assessment of the correlation proposed in
the current study reveals that although the correlation is not
generalized, it still performs relatively better for the group of
Kuwaiti dead-heavy oils. The robustness of the current model can
be visualized from Figs. 2 and 3 for the training and testing
datasets, respectively. Both datasets indicate that the model performs well at elevated temperatures.

5.4. Accuracy using regional and crude oils from different regions
Table 4 shows the experimental dead crude oil viscosity for six
sets of crudes from various parts of the world. The rst set is
regional oil from the Middle East. This oil is with 13.9 API degrees
gravity. The proposed correlation was able to predict the dead
oil viscosity for this set with an average absolute error (AAE) in
the order of 32%, and the second accurate correlation was
BeggsRobinson with an AAE in the order of 68%. The next set is
Maya Mexico-2004 that has API gravity of 20.21. The proposed

Fig. 6. (a) Comparison of the performance of the Group C correlations for low-API
(13.01) crude oils; (b) Comparison of the performance of the Group C correlations
for medium-API (16.51) crude oils; (c) Comparison of the performance of the Group
C correlations for high-API (19.81) crude oils.

correlation was able to predict the dead oil viscosity for this crude
with AAE of 29% followed by Beal's correlation (58%).
The third oils set are extremely heavy crudes from Venezuela
with API gravity of 8.31. None of the dead oil viscosity correlations

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

109

Table 4
Testing accuracy of proposed and other correlations using regional and world heavy crudes.
Crude sample

Tested parameter

Correlation

API

T
(1F)

Beal's

BeggsRobinson

Middle East

13.9
13.9
13.9
13.9

85
150
250
350

1458.5
162
11
1.3

1642.6
44.8
7.3
3.2

778.9
154.7
36.2
13.9

Maya Mexico-2004

21.8
21.8
21.8

32
59
77

1112
229
205

275
134
87

8,562,624
2529
313

462
125
71

8.2
8.2
8.2
8.2

32
59
68
86

789
623
548
515

2.77E 08
12,819,495
4,940,100
804,696

2.612E 08
2,583,980
273,107
13,710

California Gail-Ca

20.6
20.6
20.6

32
59
77

1393
406
196

411
194
122

21,436,699
3969
436

Boscan Crude

10.5
10.5
10.5
10.5

77
104
149
212

61,080
11,052
1202
204.4

34,131
7205
716
45

16,851
953
90
19

Alberta crudes

18
18
18
18
18
18
18
18
18
18
18
18
18

73.6
124
151
160
177
196
213
232
259
290
311
332
348

319
86
46
38
26
18
13
9
6
3
2
2
1

1383
51
22
18
13
9
7
6
4
3
3
3
2

Venezuela Orimulsin-100

Measured viscosity
(cp)
575
58.53
17.02
3.02

2152
321.3
163.4
132
58.43
39.33
30.86
18.89
10.66
7.92
6.94
5.54
4.4

Glaso

Labedi

Kartoatmodjo

ElsharkawyAlikhan

701.6
171.2
48.1
20.9

1217.6
121.1
28.4
13.6

724.4
91.3
10.9
3.4

82.6
54.7
45.7

441
123
70

6613
320
121

523.7
242.8
147.7

1,356,311
128,832
74,609
30,225

8195.3
5426.4
4931.3
4209.5

478,539
74,375
48,285
23,631

265,455,745
336,769
102,906
18,359

3063861.9
489839.1
269036.3
84590.4

680
175
97

107.8
71.4
59.7

619
168
95

12,489
486
172

654.6
295.3
176.5

7003
2558
767
235

1418.3
1158.3
909
716.7

5518
2403
889
335

10,598
1872
358
100

7580.1
2183.6
364.2
53.4

234
66
41
36
28
22
18
15
11
8
7
6
5

116
81.7
71.6
68.9
64.2
60
56.7
53.6
49.7
46.1
43.9
42.1
40.8

219
68
44
38
30
24
20
17
13
10
9
7
7

483
74
43
37
28
22
19
16
12
10
9
8
7

317.1
76.8
40.3
33.1
23.1
16.3
12.2
9.2
6.4
4.5
3.7
3.1
2.8

354.9
242
171.5
136.7

Proposed

Table 5
Comparison between measured and calculated viscosity from the proposed correlation, corresponding state and EOS methods.
Sample 1 (API 17.51)

Sample 2 (API 19.844)

Comp

mol%

wt%

MW

SG

Comp

mol%

wt%

MW

SG

C1
C2
C3
i-C4
n-C4
C5
i-C5
n C5
C6
C7
T, 1F
176

0.00
0.00
0.00
0.17
0.29
0.00
0.14
0.35
1.95
97.10
Measured m, cp
23.51

0.00
0.00
0.00
0.03
0.05
0.00
0.03
0.07
0.47
99.36
Proposed
25.4

16.04
30.07
44.10
58.12
58.12
72.15
72.15
72.15
86.18
367.35
LBC
3.39

0.300
0.356
0.506
0.562
0.583
0.624
0.630
0.685
0.668
0.951
Pedersen
47.2

C1
C2
C3
i-C4
n-C4
C5
i-C5
n C5
C6
C7
T, 1F
104

0.00
0.00
0.00
0.13
0.12
0.07
0.11
1.39
3.94
94.24
Measured m, cp
96.460

0.00
0.00
0.00
0.02
0.02
0.02
0.02
0.31
1.03
98.58
Proposed
96.420

16.04
30.07
44.10
58.12
58.12
72.15
72.15
72.15
85.73
343.10
LBC
3.550

0.300
0.356
0.506
0.562
0.583
0.624
0.630
0.685
0.666
0.940
Pedersen
150.290

Sample 3 (API 13.504)

Sample 4 (API 15.069)

Comp

mol%

wt%

MW

SG

Comp

mol%

wt%

MW

SG

C1
C2
C3
i-C4
n-C4
C5
i-C5
n C5
C6
C7
T, 1F
194

0.00
0.00
0.00
0.01
0.04
0.00
0.01
0.04
0.39
99.52
Measured m, cp
28.3

0.00
0.00
0.00
0.00
0.01
0.00
0.00
0.01
0.09
99.90
Proposed
34.35

16.04
30.07
44.10
58.12
58.12
72.15
72.15
72.15
86.18
384.46
LBC
4.06

0.300
0.356
0.506
0.562
0.583
0.624
0.630
0.685
0.671
0.975
Pedersen
66.52

C1
C2
C3
i-C4
n-C4
C5
i-C5
n C5
C6
C7
T, 1F
176

0.00
0.00
0.00
0.00
0.15
0.00
0.04
0.28
1.37
98.16
Measured m, cp
29.95

0.00
0.00
0.00
0.00
0.02
0.00
0.01
0.06
0.32
99.59
Proposed
37.19

16.04
30.07
44.10
58.12
58.12
72.15
72.15
72.15
86.18
369.29
LBC
3.84

0.300
0.356
0.506
0.562
0.583
0.624
0.630
0.685
0.668
0.966
Pedersen
63

110

O. Alomair et al. / Journal of Petroleum Science and Engineering 120 (2014) 102110

were able to estimate the dead oil viscosity for this extra heavy
crude. The fourth group is from California that has API gravity of
20.61. Again the proposed correlation has the smallest error (30%)
followed by Glaso's correlation (AAE 53%). The fth oil is Boscan
crude which has API gravity 10.51. The API gravity of this oil is
below the practical limit of most of the correlation. Therefore, only
Beal's and Glaso's correlations were able to predict its dead oil
viscosity with AAE in the order of 49% and 54% respectively. The
last crude oil is from Alberta that has an API gravity of 181.
Kartomtodjo's correlation was able to predict its dead oil viscosity
with AAE of 34% followed by the proposed correlation (42%).
Compositional data as well as measured viscosity for four
heavy crude oil samples are shown in Table 5. This table also
shows a comparison between the predicted viscosities from the
proposed model, corresponding state and EOS based model. The
corresponding state model by Lohrenz, BrayClark method (LBC)
and the EOS model by Pederson are considered in this study. The
LBC model severely underestimated the dead heavy oil viscosity of
all the four samples as shown in Table 5. Therefore, it is suggested
not to use the LBC model to predict dead heavy oil viscosity. The
proposed model shows a better accuracy than the Pederson model
for all the four compositional data considered.
The correlation proposed in this study should be carefully used
outside the range of data. It should not be used to predict dead oil
viscosity for extra heavy crudes having API gravity below 101 or
light crudes having API gravity above 221. It should not also be
used at extremely low temperature of 20 1C (68 1F). For predicting
the effect of gas injection upon viscosity of heavy crudes or to
predict the saturated or undersaturated oil viscosity the design
engineers should consider other correlations, corresponding states
methods, or equation of state based methods.

6. Conclusions
This study considered 492 viscosity measurements of heavy
Kuwaiti crude oil samples at both the formation and elevated
temperatures. The data were also used to develop a new heavy oil
viscosity model for Kuwait crudes. These data were used to
evaluate the performance of published correlations in addition to
the well-known correlations that have been considered benchmarks for the petroleum industry. Using regional as well as world
heavy crude oil data, the performance of the proposed model was
compared with various correlations as well as Corresponding
State, and EOS-based methods. The comparison revealed that the
proposed model has better accuracy and acceptable performance
relative to the other published methods with respect to the
viscosity prediction of heavy Kuwait crude oils. This correlation
should be used in case experimental data is unavailable or
unreliable. It should be carefully used outside the range of data.
It should not also be used to predict dead oil viscosity for extra
heavy crudes having API gravity below 101 or light crudes having
API gravity above 221, or at extremely low temperature of 20 1C
(68 1F).

Acknowledgments
This work was supported by Kuwait University, Research Grant
no. [EP 02/08]. The authors also acknowledge the support received
from the General Facility Research Grant [GE 01/07]. The authors
extend their appreciation to the Research and Technology (R&T)
Group at the Kuwait Oil Company (KOC) for assistance with the
crude oil samples.

References
Adams, J., Jiang, C., Bennett, B., Huang, H., Oldenburg, T., Noke, K., Snowdon, L.R., Gates,
I., Larter, S.R., 2008. Viscosity Determination of Heavy Oil and Bitumen, World
Heavy Oil Conference (WHOC), Paper-443. Edmonton, AB, Canada, March 1012.
Adams, J., Jiang, C., Bennett, B., Snowdon, L., Gates, I., Larter, S.R., 2009. Heavy Oil
and Super Heavy Oil Viscosity Measurement and Estimation: Getting Representative Samples, Frontiers Innovation CSPG CSEG CWLS Convention.
Calgary, Alberta, Canada, May 48.
Al-Khafaji, A.H., Abdul-Majeed, G.H., Hasoon, F.S., 1987. Viscosity correlation for
dead, live and undersaturated crude oils. J. Pet. Res. 6 (2), 116.
Alomair, O., Elsharkawy, A., Alkandari, H., 2011. Viscosity Predictions of Kuwaiti
Heavy Crudes at Elevated Temperatures. Paper SPE 150503, SPE Heavy Oil
Conference. Kuwait City, Kuwait, December 1214.
Argillier, J.F., Henaut, I., Barre, L., Brucy, F., Bouchard, R., 2001. Rheological and
Structural Properties of Heavy Crude Oils in Relation with Their Asphaltenes
Content. Paper SPE 65020, SPE Conference. Houston, Texas, February 1316.
Batzle, M., Zadler, B., Hofmann, R., De-hua, H., 2004. Heavy OilsSeismic Properties, SEG
Int'l Exposition and 74th Annual Meeting. Denver, Colorado, October 1015.
Beal, C., 1946. The viscosity of air, water, natural gas, crude oil and its associated
gases at oil eld temperature and pressures. Trans. AIME 165 (1), 114127.
Beggs, H.D., Robinson, J.R., 1975. Estimating the viscosity of crude oil systems. J. Pet.
Technol. 27 (9), 11401141.
Bennison, T., 1998. Prediction of Heavy Oil Viscosity, IBC Heavy Oil Field Development Conference. London, UK, December 24.
Bergman, D.F., Sutton, R.P., 2007. A Consistent and Accurate Dead-Oil-Viscosity
Method. Paper SPE 110194, Annual Technical Conference and Exhibition.
Anaheim, California, November 1114.
De Ghetto, G., Paone, F., Villa, M., 1995. PressureVolumeTemperature Correlations
for Heavy and Extra Heavy Oils. Paper SPE 30316, SPE International Heavy Oil
Symposium. Calgary, Alberta, Canada, June 1921.
Dindoruk, B., Christman, P.G., 2001. PVT Properties and Viscosity Correlations for
Gulf of Mexico Oils. Paper SPE 71633, SPE Annual Technical Conference and
Exhibition. New Orleans, LA, September 30October 3.
Egbogah, E.O., Ng, J.T., 1990. An improved temperatureviscosity correlation for
crude oil systems. J. Pet. Sci. Eng. 4 (3), 197200.
Elsharkawy, A.M., Alikhan, A.A., 1999. Models for predicting the viscosity of middle
east crude oils. Fuel 78 (8), 891903.
Ely, J.F., Hanley, H.J.M., 1981. Prediction of transport properties. I. Viscosity of uids
and mixtures. Ind. Eng. Chem. Fund. 20, 323332.
Glaso, ., 1980. Generalized pressurevolume temperature correlations. J. Pet.
Technol. 32 (5), 785795.
Hossain, M.S., Sarica, C., Zhang, H.Q., Rhyne, L., Greenhill, K.L., 2005. Assessment
and Development of Heavy Oil Viscosity Correlations. Paper SPE 97907, SPE
International Thermal Operations and Heavy Oil Symposium. Calgary, Alberta,
Canada, November 13.
Kartoatmodjo T., 1990. New Correlations for Estimating Hydrocarbon Liquid
Properties (MS thesis). University of Tulsa.
Labedi, R.M., 1992. Improved correlations for predicting the viscosity of light
crudes. J. Pet. Sci. Eng. 8 (3), 221234.
Little, J.E., Kennedy, H.T., 1968. Calculating the viscosity of hydrocarbon systems
with pressure temperature and composition. SPEJ J. 8 (2), 157162.
Lohrenz, J., Bray, B.G., Clark, C.R., 1964. Calculating viscosity of reservoir uids from
their composition. J. Pet. Technol. 16 (10), 11701176.
Naseri, A., Nikazar, M., Dehghani, S.A.M., 2005. A correlation approach for prediction of crude oil viscosities. J. Pet. Sci. Eng. 47 (34), 163174.
Nehring, R., Hess, R., Kamionski, M., 1983. The Heavy Oil Resources of the United
States. Rand Corporation, Santa Monica, CA.
Oskui, G.P., Al Naqi, A., 2009. Screening Potential Production Technologies for the
Lower Fars Heavy Oil Asset in Kuwait. Paper SPE 126268, Kuwait International
Petroleum Conference and Exhibition. Kuwait City, Kuwait, December 1416.
Pedersen, S.K., Fredenslund, A.A., Christensen, P.L., Thomasen, P., 1984. Viscosity of
crude oils. Chem. Eng. Sci. 39 (6), 10111016.
Petrosky, G.E., Farshad, F.F., 1995. Viscosity Correlations for Gulf of Mexico Crude
Oils, Production Operations Symposium. Oklahoma City, OK, April 24.
Puttagunta, V.R., Miadonyea, A., Singh, B., 1993. Simple concept predicts viscosity of
heavy oil and bitumen. Oil Gas J. 91 (9), 7173.
Puttagunta, V.R., Singh, B., Cooper, E., 1988. A generalized correlation for alberta
heavy oils and bitumens. In: Proceedings of the 4th UNITAR/UNDP Conference
on Heavy Crudes and Tar Sands, No. 2, pp. 657659.
Quail, B., Hill, G.A., Jha, K.N., 1987. Correlations of Viscosity, Density and Gas
Solubility for Saskatchewan Heavy Oils, First Annual Technical Meeting of the
South Saskatchewan Section. Regina, October 68.
Sattarina, M., Modarresi, H., Bayata, M., Teymori, M., 2007. New viscosity correlations for dead crude oils. Pet. Coal 49, 3339.
Standing, M.B., 1981. Volumetric and Phase Behavior of Oil Field Hydrocarbon
Systems. Society of Petroleum Engineers of AIME, Dallas, TX.
Svrcek, W.Y., Mehrotra, A.K., 1988. One parameter correlation for bitumen viscosity.
Chem. Eng. Res. Des. 66 (4), 323327.
Tirtharenu, S., Al-Sammak, I., 2011. Analysis of the First CHOPS Pilot for Heavy Oil
Production in Kuwait. Paper SPE 148966, Canadian Unconventional Resources
Conference. Alberta, Canada, November 1517.
Zabel, F., Law, D.H.S., Taylor, S., Zuo, J., 2008. Impact of uncertainty of heavy oil uid
property measurements, Paper SPE 134000. In: Proceedings of the 9th Canadian
International Petroleum Conference. Calgary, Alberta, Canada, June 1719.