Journal of African Earth Sciences 131 (2017) 105e116

Contents lists available at ScienceDirect

Journal of African Earth Sciences

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

Integrated petrophysical and reservoir characterization workow to

enhance permeability and water saturation prediction
Meshal Al-Amri a, Mohamed Mahmoud b, c, *, Salaheldin Elkatatny b, d, Hasan Al-Yousef b,
Tariq Al-Ghamdi a
a
Saudi Aramco, Dhahran, Saudi Arabia
b
King Fahd University of Petroleum&Minerals, Dhahran, Saudi Arabia
c
Suez University, Egypt
d
Cairo University, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: Accurate estimation of permeability is essential in reservoir characterization and in determining uid
Received 27 December 2016 ow in porous media which greatly assists optimize the production of a eld. Some of the permeability
Received in revised form prediction techniques such as Porosity-Permeability transforms and recently articial intelligence and
5 April 2017
neural networks are encouraging but still show moderate to good match to core data. This could be due
Accepted 13 April 2017
to limitation to homogenous media while the knowledge about geology and heterogeneity is indirectly
Available online 21 April 2017
related or absent. The use of geological information from core description as in Lithofacies which in-
cludes digenetic information show a link to permeability when categorized into rock types exposed to
Keywords:
Petrophysical
similar depositional environment.
Reservoir characterization The objective of this paper is to develop a robust combined workow integrating geology and pet-
Permeability rophysics and wireline logs in an extremely heterogeneous carbonate reservoir to accurately predict
Lithofacies permeability. Permeability prediction is carried out using pattern recognition algorithm called multi-
Water saturation resolution graph-based clustering (MRGC). We will bench mark the prediction results with hard data
from core and well test analysis.
As a result, we showed how much better improvements are achieved in the permeability prediction
when geology is integrated within the analysis. Finally, we use the predicted permeability as an input
parameter in J-function and correct for uncertainties in saturation calculation produced by wireline logs
using the classical Archie equation. Eventually, high level of condence in hydrocarbon volumes esti-
mation is reached when robust permeability and saturation height functions are estimated in presence of
important geological details that are petrophysically meaningful.
2017 Elsevier Ltd. All rights reserved.

1. Introduction permeabilities, and water/hydrocarbon saturations. Common

In any reservoir engineering studies, the main objective of
integrating petrophysical properties is to develop a reliable reser-
voir static model that is vital in characterizing reservoir matrix and
improving reservoir simulation history matching for improved eld
strategies and enhanced reservoir engineering best practices. These
models consisted of various reservoir petrophysical parameters
which include and not limited to porosity, permeability, relative
* Corresponding author.
E-mail addresses: meshal.alamry@aramco.com (M. Al-Amri), mmahmoud@
kfupm.edu.sa (M. Mahmoud), elkatatny@kfupm.edu.sa (S. Elkatatny), hyousef@
kfupm.edu.sa (H. Al-Yousef), tarek.ghamdi.3@aramco.com (T. Al-Ghamdi).
permeability modeling is randomly constructed using conventional
core analysis (CCA) (porosity/permeability relationship) to form a
non-liner relationship that will ultimately make permeability
prediction as porosity transformation. However, it does not take
into account other important rock/reservoir properties such as
grain size, sorting, tortuosity and digenesis. Reservoir permeability
variation provides information on reservoir rock heterogeneity. In
addition, pore throats play a major role in identifying accurate
permeability values; this is a major challenge in carbonate envi-
ronment as there is no robust pore size identication of the pore
system. Moreover, the complexity in predicting permeability is the
wide variety of pore system in terms of geometry (Intergranular,
Intragranular, Intercrystalline, Vuggy and Fracture) and pore sizes

http://dx.doi.org/10.1016/j.jafrearsci.2017.04.014
1464-343X/ 2017 Elsevier Ltd. All rights reserved.
106 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116
classes (Macro-Porosity, Meso-Porosity and Micro-Porosity). Pore
systems vary with rock type which is related to mineralogical
composition and related to certain specic uid ow characteris-
tics. In addition, porosity, which is a simple term, refers to the
mathematical ratio of pore to bulk volumes that is altered by
several factors and most importantly pore-size, grain density, and
grains sorting that make up the rock itself. Porosity can be occupied
by different uids that are called uid saturation dened as the
percentage of pore size that is lled with a specic uid phase.
Saturation is a function of porosity, permeability, capillarity,
wettability and water salinity. Most of the oil reservoirs are initially
water saturated. Migration of oil into the reservoir causes drainage
of water. Therefore, as the distance above the Free-Water Level
(FWL) increases progressively, smaller pores are lled and oil
saturation increases. There various methods in estimating water
saturation and one of the known methods is derived from logs
resistivity measurements. Archie (1952) came up with equations
using resistivity of a completely brine-saturated rock (Ro), re-
sistivity of brine (Rw), and rock actual resistivity when lled with
reservoir uid (Rt). He concluded that the formation resistivity
factor (F) is correlated with porosity and certain reservoir co-
efcients (cementations and tortuosity). Other investigators esti-
mated water saturation using saturation height functions (SHF).
One of the most popular models to calculate water saturation that
depends on capillarity is Leverett J-function. Leverett estimated
water saturation by combining different rocks representing the
reservoir into a single model (J-function) that converts all capillary
curves to a universal one which introduces a unique model t for
each reservoir layering. J-function is then constructed and plotted
using the following power law equation:
J aSw b (1)
where; J is the J-function; a and b are the model parameters; Sw is
the water saturation.
The model parameters (a and b) are obtained for similar rock
types and layers to dene the saturation height model above the
free water level. Since he recommended that layering the reservoir
would greatly assist in obtaining accurate model, we will utilize this
conclusion to correlate same lithofacies capillarity to examine
water saturation model for each depositional environment. For
condent permeability and water saturation characterization, cor-
ing campaign and physical core description should be planned in
order to get the data from each well which is quite time consuming
and need expensive laboratory measurements.
2. Literature review
A lot of work/progress evolved around the permeability pre-
diction as well as water saturation using various techniques and
methods.
The importance of pore texture on the physical properties in
carbonates was rst realized by Archie (1952). He rst classied
carbonate rocks by their pore size to get an approximation on the
permeability. Porosity is subdivided into visible macro-porosity and
matrix micro-porosity. Dunham (1962)'s classication focuses on
depositional textures created during deposition. His classication
divided carbonate rocks into two major groups: (1) mud-supported,
(2) grain-supported. Lucia (1983) introduced a qualitative carbonate
pore type classication which is based on pore size and pore con-
nectivity and its effect on the petrophysical properties. For example,
moldic and interparticle porosity fall both in the same fabric se-
lective pore classication after Choquette and Pray's (1970), but they
show very different porosity-permeability trends. They subdivided
porosity into touching vugs porosity, separate vugs porosity and
interparticle pore space. All three groups have differences in the
quality of connected pores. As a result, all three groups show distinct
porosity-permeability relationships. Gunter et al. (1997) described a
technique that combines basic reservoir properties i.e. bed
thickness, porosity and permeability information for ow unit cal-
culations. They applied Modied Lorenz plots (MLP) for character-
ization. This method of ow unit determination is quite useful
because it only requires routine porosity and permeability data
(from logs and/or core). Morris and Biggs (1967) developed an
empirical correlation to predict permeability at initial water satu-
ration. They dened the correlation of logs calculated porosity and
resistivity based saturation to estimate permeability by:
cf3
k1=2 (2)
Swi
where; k is the rock permeability; c is a constant which is a function
of reservoir uid density, 4 is the rock porosity, and Swi is the initial
water saturation.
Jennings and Lucia (2003) concluded sequence-stratigraphic
framework would be more systematically organized using rock-
fabric classication instead of using the direct relationship of
porosity and permeability. They introduced a new parameter called
rock-fabric number (l) which is a function of two reservoir pa-
rameters above capillary transition zones: initial water saturation
and porosity. A generalized permeability-porosity-rock fabric
relationship is given by:
bl
k eal fip (3)
al a0  a1 lnl (4)
bl b0  b1 lnl (5)
where; k is the rock permeability; 4ip is the rock porosity; al and
bl are rock fabric coefcients.
Fleury (2002) proposed a model that included the non-Archie
behavior which accounts for resistivity measurements on double
DPC porosity (Dual Porosity Conductivity) or triple porosity TPC
(Triple Porosity Conductivity) micritic and oolitic carbonates. The
proposed model for DPC is:
1a
RI Sn 1
; Sw  Sc (6)
w1
1 aSn
w1
1
1a
RI Sn 1

n2 ; Sw  Sc (7)
w1
1 aSn
w1 Sw2
1
where: RI: is the resistivity index, n1 is the saturation exponent at
high Sw , n2 is the saturation exponent at low Sw, a is the ratio
between micropores to macropores fraction and Sc is the average
water saturation at which micropores invaded. In this model (DPC),
4 parameters have to be identied; Sc which can be determined
from MICP data with some degree of uncertainty while other pa-
rameters are tted. Shokir (2006) proposed a novel model for
permeability estimation in uncored wells called Fuzzy model. Fuzzy
model proves best way to model heterogeneous formations,
nonlinear and multivariable petrophysical reservoir properties.
Creusen et al. (2007) showed that, the ratio of different rock types
used and the spatial organization of high and low permeable rock
have a big effect on both ow and recovery which should be
included in simulation models. Taware et al. (2008) clustered
similar log responses and linked them to petrophysics (elec-
trofacies) accordingly to predict permeability in one well in
 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116 107

sandstones reservoir. Amabeoku et al. (2008) presented an aston- This resulted in an excellent depth match between core and wire-
ishing approach for water saturation predication based on new line logs.
saturation/height model. He started to develop the model by link
both depositional and digenetic rock texture to HU. After that, he 5. Core description
combined the obtained HU to the zones with same capillary pres-
sure relation. He calculated water saturation values from three 17 lithofacies have been identied that represent the various
different models: Leverett J-function, FOIL and Modied FOIL depositional environments of the understudy reservoir. Table 1
functions. Sutadiwirya et al. (2008) concluded that especially in shows the description of each lithofacies. Every lithofacie has a
clustering modeling training data values can be grouped based on range of permeability values depending on the depositional
specied associated parameters. One of them is multi-resolution cyclicity. Depositional cyclicity is a common characteristic of car-
graph-based clustering (MRGC). It solves dimensionality problems bonate platform sequences, which is evident in this reservoir at a
that usually occur when log data is relatively constrained with few variety of scales ranging from centimeter and decimeter scale bed
clusters. This will merge large number of clusters into a small forms to larger scale packages of sediment. Recognition and inter-
cluster that was assigned from the geological characterization. It pretation of these larger scale cycles provides the basis for dening
also reduces several drawbacks that come from conventional chronostratigraphic surfaces, identifying time-equivalent facies,
method. and constructing a depositional and sequence-stratigraphic
The objective of this paper is to implement an integrated pet- framework for the reservoir section. This unied framework de-
rophysical workow for carbonate reservoirs to improve geolog- scribes the three-dimensional geometry of the reservoir and pro-
ical/static models in predicting permeability and saturation vides a means for evaluating historical uid movement and overall
distribution. Reservoir permeability variation provides information reservoir performance. Detailed facies mapping within each cycle
on reservoir rock heterogeneity. This will be carried out using of deposition allows the sequential development of the reservoir.
quantied geological facies model to better represent permeability
family to each facie. The proposed model will be vital in charac-
6. Data ltration
terizing reservoir matrix and improving reservoir simulation his-
tory matching for improved eld strategies and enhanced reservoir
After performing the depth shift, a systematic approach for data
engineering best practices. A reliable saturation model for un-cored
ltration was followed to dene which data set is considered for
wells will be developed from the constructed permeability model
this study to eliminate introducing unknown factors in the
which will use available special core analysis SCAL (Normal Capil-
modeling. Using standard deviation of corrected porosity differ-
lary Pressure and Mercury Injection Capillary Pressure, MICP).
ence, it is then applied to the absolute difference between the
corrected core porosity and wireline log porosity to lter the data
3. Data preparation
that exceed the standard deviation of the absolute difference. The
absolute difference should not exceed the standard deviation of the
In this study, over 30 vertical wells have been considered from
corrected core porosity to include all relevant data at this specic
carbonate reservoir for permeability and water saturation predic-
depth whereas high absolute difference of more than the standard
tion. Conventional Core Analysis (CCA) was performed for all core
deviation is considered as a bad data point which in this context is
samples beforehand to measure porosity and permeability in lab-
neglected to ensure high level of condence on the nal selected
oratory conditions. In addition, lithofacies description has been
set of data. This does not mean that removed data set is outliers
carried out for cored wells to identify the number of facies in the
however, more investigation needs to be carried out to assess the
reservoir of study as well as build a robust geological lithofacies
model to be propagated in reservoir section and then improved
using wireline openhole logs for uncored wells. Moreover, wireline
logs were prepared as input parameters for permeability models.
Intuitively, three main input sources of data are used in this study to
come-up with a permeability model for each lithofacie and hence
water saturation is derived which include core plug samples
measurements, wireline logs measurements and lithofacies model
inputs.

4. Core measurements correction

In this section, quality control of CCA data is performed to cor-

rect for laboratory conditions which includes applying overburden
corrections for porosity values, Klinkenberg effect of permeability
values and depth shift between core data and wireline logs. After
performing the necessary overburden and Klinkenberg effects
corrections, a depth shift of data is carried out by plotting core data
with wireline logs. Usually, spectral gamma-ray (SGR) reading is
used as a reference in depth-shifting however dealing with car-
bonate reservoir making it difcult to distinguish for signatures and
markers although sandstone is a perfect formation to use SGR as
shale sections are differentiated by high GR. Therefore, corrected
core porosity (shown as black dots) is plotted against wireline
corrected porosity from neutron-density cross plot (shown as red
curve) to obtain the best depth match. Fig. 1 shows the data before
and after the depth shift in the highlighted sections for two wells. Fig. 1. Well-B depth shift technique for data preparation.
108 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116

Table 1
Depositional environment of the reservoir lithofacies.

Lithofacies# Depositional Environment lithofacies

F-1 Coarse to very coarse-grained lithocodium grainstones

F-2 Cemented very coarse to coarse-grained lithocodium grainstones
F-4 Cemented and microporous coarse-grained lithocodium grainstones
F-7 Fine to medium skeletal grainstones
F-8 Cemented ne to medium grainstones
F-9 Microporous ne to medium grainstones
F-10 Cemented and microporous medium grainstones
F-11 Oolitic skeletal grainstones
F-15 High interlayered coarse and ne peloidal grainstones
F-16 Cemented high interlayered coarse and ne peloidal grainstones
F-17 Microporous cemented high interlayered coarse and ne peloidal grainstones
F-18 Cemented and microporous coarse to ne grainstones
F-19 Moderately interlayered ne grainstones to packstones
F-20 Cemented moderately interlayered ne grainstones to packstones
F-21 Microporous moderately interlayered ne grainstones to packstones
F-22 Cemented and microporous moderately interlayered ne grainstones to packstones
F-23 Skeletal packstones

measurements deviation, which is not part of the study scope,
which could be due to different measurement sizes between logs
and core data. In addition, some of the porosity measured in the
laboratory exhibited low grain density which in turn results in
wrong porosity calculations. Fig. 2 shows an example of data points
fall outside the standard deviation of the absolute difference of
porosities. Standard deviation is high when more scattered the
data.
7. Microporosity identication
The microporosity identication is interpreted using sonic logs
which are usually responds to the porous media volume (porosity)
that is controlled by several factors such as: formation lithology,
rock texture, overburden/pore pressure and fractures. In the un-
derstudy reservoir, lithology is relatively uniform mainly Calcite,
scanty fractures and no gas trapped. Therefore, the two remaining
factors are porosity and rock texture (lithofacies). Therefore, a
technique using wireline logs to detect microporosity presence is
included in this study to distinguish low permeability rocks from
high quality ones when both have the same high porosity range
which is a common phenomenon in carbonates (Fig. 3). This
technique increased our condence in detecting microporosity
zones from logs which agreed to core measurements. Core
description (lithofacies determination) also proved to be in agree-
ment with log-derived microporosity prediction. Hence, quality
rocks typically lead to a satisfactory agreement between sonic and
density-neutron porosities whereas the difference is an indication
of poor quality rocks. Fig. 4 demonstrates that the higher the sep-
aration between sonic porosity and density-neutron porosity the
more microporosity is encountered. The addition of this technique
as input parameter will strengthen permeability prediction. As
shown in Fig. 4, F-4 shows excellent porosity zone with low
permeability which needs to be captured to improve uncored wells
permeability modeling.
8. Prediction workow
The main intent of this paper is to come up with a robust
modeling workow using lithological and petrophysical inputs to
enhance the permeability prediction accuracy in uncored wells.
This involves incorporating lithological facies and wireline logs in
addition to developed techniques that supports linking logs and
pores network systems. Lithofacies, wireline logs and developed
moldic porosity identier tools are included in the training of data
against core permeability. These major inputs are then imple-
mented in the geological/reservoir modeling. We will use Facimage

Fig. 2. Well-C showing process of data removal using grain density measurement. (6th
and 5th tracks show porosity before and after data ltration, respectively). Fig. 3. Lithofacies-1 and 4 porosity-permeability relationship.
 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116 109

which are KNN 2, 3 and 4 respectively. To support selecting the

best number of nearest neighbor, two visualization techniques well
level investigation and cross-plots of the whole data are utilized to
ensure that the most representative KNN value is selected to model
propagation on data set level for all wells. The rst technique is to
compare the model for different KNN values on well level to check
how the model is preserving permeability trends. Fig. 6 shows that
KNN 2 is the best model as it captures both low and high sides of
permeability ranges whereas KNN 3 and 4 are unable to capture
high permeability ends, although all KNN show excellent correla-
tion coefcient on a reservoir level. The second technique is to use
the cross-plot of permeability values obtained from the trained
model for each KNN value and corrected core permeability and
then select the best correlation coefcient for reservoir level as
demonstrated in Fig. 7 that show all KNN values are good however
KNN 2 with R2 0.91 is the best number as KNN increases the
model starts to degrade.

10. Statistical assessment

To show the model accuracy, error statistical assessment has

been carried out to conrm that the model is preserving the core
permeability with good match. Therefore and since the t coef-
Fig. 4. Two wells microporosity interpretation using cores, sonic and neutron/density
cient (R2 0.91) of KNN 2 is the best which is excellent in car-
data.
bonate reservoirs, all above mentioned techniques suggest a KNN
value of 2 in the model propagation to the remaining cored and
technique which is one of the latest cutting edge tools of workow uncored wells. The statistical formulas of error measurements are
that employs neural network and pattern recognition algorithm dened below including error average (Ex ), error standard devia-
known as Multi-resolution graph-based clustering (MRGC) that has tion (ESt.D)., Average percent relative error (ARE), Average absolute
been known as a statistical non-parametric technique that solves percent relative error (AARE), Correlation Coefcient (R) and root
dimensionality problems. It also derives valuable information means of squares (RMS).
about the geological lithofacies from the structure of the data itself. Pn
One of the most important advantages of non-parametric tech- i1 Ei
Ex (8)
niques is it can predict without the need to provide any information n
about the data distribution. Ye and Rabiller (2000) MRGC is a
where Ei kc i  kp i
combination of two well-known techniques: K-nearest neighbor
(KNN) and graphical representation of the data which takes ad- !2
vantages of both methods of capturing data classes or clusters for 1 X n  
ESt:D Ei  E (9)
any set of data structure. n  1 i1
As discussed in the previous sections, cored wells are used
Pn
initially to build the model and train the data set to provide a where E 1n i1 Ei
recognized form from set of logs. For applying the MRGC method in  
this study, GEOLOG in Linux environment is used. The workow 1X n
kc i  kp i
ARE *100 (10)
(Fig. 5) of permeability prediction is simply consisted of: n i1 kc i

 Input sets which include: Lithofacies, density, neutron, sonic  

and difference of (density-neutron and sonic). 1X n
kc i  kp i
AARE *100 (11)
 Associated set of corrected permeability from core n i1 kc i
measurements.
 Output sets which denes the predicted permeability.   
Pn  
i1 kc i  kc kp i  kp
R s
    (12)
Pn Pn  
9. Model training i1 k c i  k c
2
i1 k p i
 k p
2

Once all of input parameters correlations are validated and su- Pn Pn

where kc 1n i1 kc i and kp 1n i1 kp i .
pervised, permeability prediction starts using cluster analysis
which assigns the number of nearest neighbors. The prediction is v
u
then applied on 50% of cored wells just to test the quality of pre- u 1 X n
RMS t E2 (13)
dicted permeability against core data. If log prediction is good then n  2 i1 i
apply the model to the wells of interest either cored or uncored
wells. In order to come up with the best nearest neighbor number Table 2 shows the statistical assessment between actual core
(KNN), the model is tested against three different values of KNN permeability and model predicted permeability.
110 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116
Fig. 5. A systematic workow used in this study to predict permeability with the training data set.
Fig. 6. Well-J predicted permeability (black curve) over the corrected core perme-
ability (red dots) for KNN 2, 3 and 4. (For interpretation of the references to colour in
this gure legend, the reader is referred to the web version of this article.)
11. Permeability model validation
The above systematic approach of predicting permeability is
also supported by transient pressure buildup/falloff analysis to
obtain the kh value for a specic well. This information assists in
achieving a higher level of condence when used in uncored wells.
This validating mechanism is used to compare kh value from core
data, modeling and well transient test data. A good match is
observed as shown in Fig. 8 between all three measurements of
ow capacity. For uncored wells, Fig. 9 also exhibits good match
Fig. 7. Cross-plot of predicted model permeability versus corrected core permeability for KNN 2, 3 and 4.
Table 2
Statistical Assessment for predicted permeability at a KNN 2.
Parameters Value
n 3916
Average Error Difference from Permeability Predicted 15.57
Standard Deviation (S) Error from Permeability Predicted 160.08
ARE from Permeability Predicted 38.91
AARE from Permeability Predicted 74.69
Correlation Coefcient (R) from Permeability Predicted 0.85
Root Mean Squares (RMS) from Permeability Predicted 160.86
between buildup/falloff test and predicted permeability. This is an
additional tool to conrm a valid model even when compared to
well test results.
Dynamic permeability from pressure buildup/falloff analysis is
utilized and integrated with ow meters analysis (from production
logging tool -PLT-) to convert static permeability to dynamic in each
lithofacies which in current practices achieved by using multiplier
to be used in simulation models. The provided kh from well test
provide information about reservoir quality and potential in both
injector and producer wells. The static permeability provided by
the model is linked to the dynamic permeability via a multiplier for
a particular zone. Only open-hole completed wells is used in this
section to establish a relationship due to limitations of cased-hole
completion accessing only a certain zone or some zones. A soft
code is used here to translate the ow meter contribution per 0.5
foot and multiply it by the magnitude of kh provided from well test.
The resultant multiplication provides a continuous log that can be
compared to the one from static permeability (modeling).
The equation to be solved is dened after Dennis Schmitt, Saudi
Aramco, Reservoir Description Division which then converted to a
soft code:
X
i
1  Qi kh aj kj hj ; i 1; 2; ; N (14)
j1
where Qi is the percentage ow obtained from PLT data, kh is the
 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116 111

Fig. 8. Validating modeling permeability using kh (horizontal permeability) value from modeling (black curve), buildup test (blue curve) and core measurements (red curve) in
cored well which demonstrate a good match among three different validation mechanisms. (For interpretation of the references to colour in this gure legend, the reader is referred
to the web version of this article.)

permeability thickness over the interval of the interest. In other elements Bim of the B matrix are given by:
words, kh can be expressed as ow capacity for a given interval
which its unit is mD-ft. the kj is the permeability obtained or 8 Bim 0 ci < Nm1
deduced from cores and aj is the multiplier that needs to be >
>
>
> X
i
determined. hj is the thickness of the layer/interval within which kj >
< Bim kj ci < Nm1 ; Nm 
is dened. jNm1 1 (17)
When using the rst equation, hj is in fact the depth sampling >
>
>
> XNm
>
: Bim kj ci < Nm
rate and is thus constant. Also, the Qm is initially dened only at a
few points, irregularly spaced. The regularly sampled vales Qi is jNm1 1
obtained by linear interpolation in between the actual Qm . Keeping
Only the last two equations apply to the rst column of B.
this in mind, the rst equation can be re-written as:

X
i
aj kj Gi ; i 1; 2; ; N (15)
j1

where Gi 1Qhi kh. Note that Q1 100. In order to avoid Gi 0, Q1

is set to 0.99998.
To solve this system, let M be the number of different multipliers
am with M < N. M us given by the original data points of the Qi
between which linear interpolation has been performed. It is
assumed that am belongs to the interval [Nm1 ; Nm ] where the Nj is
the index with respect to the analyzed interval. With such as-
sumptions, the second equation leads to an over determined sys-
tem of the form:

B$ a G
where B is an N  N matrix (M < N, a is an M vector and G is an N
vector. For m within the interval [Nm1 ; Nm ] (assuming N0 1), the
(16)
Fig. 9. Validating modeling permeability using kh value from modeling (black curve)
and buildup test (blue curve) in uncored oil producers which demonstrate a good
observed match. (For interpretation of the references to colour in this gure legend,
the reader is referred to the web version of this article.)
112 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116
There are many ways, more or less sophisticated to solve such a
system and may were tried. At the end, the simplest straight for-
ward solution works perfectly well, i.e.,
T 1 T p
a B$B $B$ G (18)
T 1
where the superscripts and indicate the transpose and the
inverse, respectively. Of course, the solution is not unique but no
signicant variation was observed between the results coming
from different algorithms. Fig. 10 shows that minimal multiplier is
needed to convert static permeability to dynamic permeability. This
tool can also be used as a validation mechanism of well transient
testing as it provides a tool to revisit well test analysis although
permeability across second and third perforations are almost
matching as illustrated in Fig. 11.
12. Permeability model prediction
After a thorough study of the modeling parameters, a quality
model is then propagated to all cored and uncored wells. In addi-
tion, the model is compared to another model using the same
approach but without incorporating lithofacies inputs which shows
a very good enhancement in permeability modeling against cor-
rected core permeability. Fig. 12 shows an example of one well with
an improved permeability prediction by introducing geology
(depositional environment) into the modeling.
Fig. 13 shows permeability prediction in uncored oil producer
which clearly follow the same permeability trend for each lithofacie
e.g. F-1 high permeability range which is a characteristic of this
facie. In addition, Fig. 14 represents the permeability prediction in
uncored water injectors. Having accurate prediction enables
reservoir engineers to selectively choose the preferred zones of
production/injection. This is the most important tool, permeability,
especially in reservoir modeling and simulation. The model in turn
has a limitation in providing accurate permeability modeling per
cluster/lithofacies when no enough data is provided for a specic
facie as not all lithofacies are uniformly distributed in the reservoir.
13. Saturation height function (SHF) modeling
Modeling of water saturation height functions is crucial in
determining the hydrocarbon reserves in the reservoir. The change
in water saturation due to production of the reservoir is continu-
ously assessed via well logging. These changes are due to depletion
of some reservoir zones that are most permeable and thus contain
considerable hydrocarbon reserves. We aim in this chapter to
Fig. 10. Converting static permeability (modeling) to dynamic permeability in two
wells with almost minimal required multiplier (5th track black curve).
model the saturation height of the reservoir prior to any effects
using Leverett J-model to estimate lithofacies based water satura-
tion which is simply expressed as:
0:2166  Pc  k=4
J (19)
s  cos q
where; J is the J-function; Pc is the capillary pressure; k is the rock
permeability, 4 is the rock porosity; s is the interfacial tension; and
q is the contact angle.
Saturation height functions (SHF) for the understudy carbonate
reservoir provide a robust saturation prole for all lithofacies to
capture saturation calculations for the heterogeneous reservoir
environment which will be bench marked against the saturation
calculated using Archie equation from wireline logs. Ultimately,
hydrocarbon in place volumes will be more accurate represented by
these lithofacies. Typically, high quality rocks will tend to have
lower initial water saturation and Swirr will elevate as quality of the
rock degrades (Leverett, 1941).
In order to come up with representative saturation height
models, we need to acquire representative capillary pressure (Pc)
measurements on these various types of lithofacies. The only
available source of data that has Pc curves is the mercury injection
capillary pressure (MICP) which is conducted on more than 36 core
plugs covering most of the reservoir lithofacies.
We utilized all inputs from well levels (predicted permeability,
log porosity, height, and interfacial tension data) to calculate J-
function based on MICP curves for each lithofacies and compare it
with water saturation from Archie equation. It is recommended
that only pre-production wells are used when applying modeling
saturation from logs to capture the original water saturation that is
not altered by production. The model from J-function is mainly
controlled by the permeability and height above FWL/OWC (oil-
water contact). On the other hand, Archie equation is given by:
a  Rw
Snw (20)
4m  Rt
where; a is the rock tortuosity, Rw is the brine resistivity, 4 is the
rock porosity; m is the cementation factor; and Rt is the true re-
sistivity of the rock.
For J-function, we initiate the saturation modeling by calculating
the height above free water level which is simply obtained using
this equation:
0:433  Drx
HAFWL (21)
Pc
where HAFWL is the height above free water level.
Following that, the saturation models built for all rock type/
Lithofacies are executed accordingly. As a result, lithofacies based
saturation height models are built and used for saturation calcu-
lations. J-function value is calculated and compared against the
MICP saturation conditioned for each lithofacies to wells above
FWL on logarithmic scale and a regression is established to obtain
the correlation between the two parameters. This power law cor-
relation is used to calculate SW as a function of J for all wells which
dene the oil column and thus OOIP.
a
Sw (22)
Jb
Fig. 15 illustrates the followed methodology to predict water
saturations.
Mercury injection is utilized in this study to come up with a
distinct capillary pressure curves for each lithofacies. Common
 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116 113

Fig. 11. Converting static permeability (modeling) to dynamic permeability in a well with almost matching model and PLT Kh however well test Kh value require further
investigation.

reservoir rock typing (RRT) methods recommended the use of MICP

data to categorize different petrophysical behaviors for certain
reservoirs (Alameri and Shebl, 2011). In this study, we used geology
to group MICP curves in accordance to their lithofacies. The Pc
obtained from MICP experiments is corrected after converting it to
reservoir conditions using the below equation to eliminate the
closure effect of entry pressure at the beginning of mercury injec-
tion which is dened as the pressure at which mercury starts to
enter the core plug.

IFTres
Pc res Pc lab (23)
IFTlab

where IFT is the interfacial tension expressed by s  cos q.

Fig. 16 illustrates the correction required for Pc form MICP ex-
periments. The summary of MICP Pc curves correction is carried out
as the following:

 Convert laboratory condition to reservoir equivalent conditions

 Entry pressure effect removal
 Cap Pc to the maximum value at the reservoir pressure which
represents the highest point from FWL at the crest.

Pc curves are capped to the maximum value at the reservoir

condition using the below equation:

Fig. 12. Comparison between with and without lithofacies inputs predicted Pc Drgh 0:433  Dr  HAFWL (24)
permeability.
114 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116
Fig. 13. Permeability prediction of an oil producer, from the model.
Fig. 14. Permeability prediction in two water injectors, from the model.
Once all curves are corrected for closure effects, J value is
calculated using Leverette J-function which uses the core plug
porosity and permeability, Pc value from the above equation, and
IFT which is known for this reservoir. The J value is generated for all
core plugs per lithofacies which are plotted versus water saturation
obtained from MICP conversion to reservoir condition (oil-brine
system). This relationship will provide the SHF parameters for each
lithofacies which will be used to calculate the initial water satu-
ration. Many researchers utilized MICP data to calculate water
saturation (Wunderlich, 1985; Tomutsa et al., 1990; Smith et al.,
2002; Seth and Morrow, 2006). However, Greder et al. (1997)
suggested that MICP uids (mercury-air system) are equivalent to
oil-brine system when converted to reservoir conditions for purely
water wet system. The below equation is used to derive the Pc at the
reservoir condition from laboratory tests and a mathematical
relationship is obtained for each lithofacies which is in the power
law form to calculate J values at various water saturations.
IFTres
Pc res Pc lab (25)
IFTlab
a
Sw (26)
Jb
Water saturation is calculated using the obtained Leverette J-
function parameters for all lithofacies which we have generated a
distinct a and b per lithofacie. This information is related to the rock
quality as more initial water saturation observed the more the
quality degraded. In other words, high quality lithofacies represents
the minimum initial water saturation. These results show the op-
timum utilization of the Leverette lithofacies based J-function as it
was initially (Leverett, 1941) developed for certain a similar rock
dynamic behavior which is now linked to geology in this study. We
believe that introducing geological lithofacies, we are able to cap-
ture the heterogeneity of different types of lithofacies which is
illustrated in Fig. 17. It is clearly noticed that when reservoir quality
degraded in lithofacies 4 when compared with the same family of
depositional environment without diagenesis effect lithofacies 1,
the water saturation starts to increase in this interval which honor
the reservoir quality in the water saturation calculations. These
 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116 115

Fig. 15. Water saturation modeling workow used in this study.

Fig. 17. Two wells water saturation predicted (red curve 4th track) compared to Archie
water saturation (blue curve, 4th track), 6th track shows the difference between
product of predicted water saturation and porosity for our method and Archie in the
shaded pinkish area. (For interpretation of the references to colour in this gure
legend, the reader is referred to the web version of this article.)

Fig. 16. MICP curves which show the closure effect that is corrected for this study.
irreducible water saturation in the pre-production stage of the
reservoir.
results are compared to Archie equation based calculations of water
 Saturation height model using J-function shows a good agree-
saturation in lithofacies 4, which is derived from resistivity mea-
ment with Archie saturations in good quality lithofacies
surements, which show big inconsistency at this interval of water
whereas J-function shows higher saturation in low quality
saturation difference that exceeds 7.5%. In most cases and especially
lithofacies and thus accurate oil column calculations.
in giant reservoirs, this 7.5% could lead to difculty in history
matching in reservoir simulation initialization runs.
Acknowledgment
14. Conclusions
The authors would like to thank KFUPM and Saudi Aramco
managements for their support in publishing this paper.
In this study, a new workow is introduced to incorporate
lithofacies geological information into petrophysical modeling.
Lithofacies inputs helped in achieving an excellent match between Abbreviation
core permeability and model predicted permeability which ne-
cessitates the integration of geology with reservoir static and dy- HCPV Hydrocarbon pore volume
namic parameters. This study summarizes the following ndings: MRGC Multi-resolution graph-based clustering
KNN K-nearest neighbor
 Geological integration requires a comprehensive knowledge on PLT Production logging tool
the geological events sequence that resulted in different car- MICP Mercury injection capillary pressure
bonates depositional environments which alters rock petro- FWL Free water level
physical properties specially diagenesis effects. kr Relative permeability
 We concluded that the use of MRGC method has provided RRT Reservoir rock typing
improved correlation coefcient as this method doesn't require CCA Conventional core analysis
priori knowledge of the training data. NMR Neutron magnetic resonance
 Statistical assessment has been carried out which shows mD Milli-Dacry
excellent model. Q Flow Rate
 The model is validated using kh from transient well testing and DP Pressure drop in the reservoir
production logging data, the model relatively matches with K Absolute permeability
good accuracy the prediction model. L Length
 In addition, 36 MICP samples are used to generate capillary A Cross sectional Area
pressures in the reservoir for each lithofacies. These Pc curves m Viscosity
are analyzed and converted to J-function to estimate original Ro Completely brine saturated rock
116 M. Al-Amri et al. / Journal of African Earth Sciences 131 (2017) 105e116

Rw Water resistivity Presented at the SPE Annual Technical Conference and Exhibition, 29
4 Porosity September-2 October, San Antonio, Texas.
Greder, H.N., Cordelier, G.V.P., Laran, D., Munez, V., d'Abrigeon, O., 1997. Forty
Pc Capillary pressure comparisons of mercury injection data with oil/water capillary pressure mea-
SCAL Special core analysis surements by the porous plate technique. In: SCA Annual Technical Conf. -
PDF Probability distribution functions SCA1997e10.
Gunter, G.W., Finneran, J.M., Hartmann, D.J., Miller, J.D., 1997. Early determination of
R2 Correlation coefcient reservoir ow units using an integrated petrophysical method. In: Paper SPE
Kh Flow capacity 38679 Presented at the SPE Annual Technical Conference and Exhibition, San
LP Lorenz Plot Antonio, Texas, 5e8 October.
Jennings, J.W., Lucia, F.J., 2003. Predicting permeability from well logs in carbonates
IFT Interfacial tension with a link to geology for interwell permeability mapping. SPE Reserv. Eval. Eng.
s Surface tension 6 (4), 215e225. http://dx.doi.org/10.2118/84942-PA.
q Contact angel Leverett, M.C., 1941. Capillary behavior in porous solids. SPE-941152-G Trans. AIME
142 (1), 152e169. http://dx.doi.org/10.2118/941152-G.
Swirr Irreducible water saturation Lucia, F.J., 1983. Petrophysical parameters estimated from visual descriptions of
OWC Oil-water contact carbonate rocks; a eld classication of carbonate pore space. J. Pet. Technol. 35
OOIP Original Oil in Place (3), 629e637. http://dx.doi.org/10.2118/10073-PA.
Morris, R.L., Biggs, W.P., 1967. Using log-derived values of water saturation and
Pc lab Capillary pressure at the laboratory conditions (air- porosity. In: SPWLA 8th Annual Logging Symposium, 12e14 June, Denver,
mercury system) Colorado.
Pc res Capillary pressure at the reservoir conditions (water-oil Seth, S., Morrow, N.R., 2006. Efciency of conversion of work of drainage to surface
energy for sandstone and carbonate. In: Paper SPE 102490 Presented at the SPE
system)
Annual Tech. Conf. And Exhibition, San Antonio, Texas. http://dx.doi.org/
Ex Error average 10.2118/102490-MS.
ESt.D Error standard deviation Shokir, E.M., 2006. A novel model for permeability prediction in uncored wells. SPE
Reserv. Eval. Eng. J. 9 (3), 266e273.
EMAX maximum absolute percent relative error Smith, J.D., Chatzis, I., Ioannidis, M.A., 2002. A new technique for measuring the
AARE Average absolute percent relative error breakthrough capillary pressure. J. Can. Pet. Technol. 44 (11) http://dx.doi.org/
RMS Root mean sq 10.2118/05-11-01.
Sutadiwirya, Y., Abrar, B., Henardi, D., NuGRoho, B.H., Wibowo, R.A., 2008. Using
MRGC (multi resolution GRaph-based clustering) method to InteGRate log data
References analysis and core facies to dene electrofacies, in the benua eld. Central
sumatera basin, Indonesia. In: International Gas Union Research Conference,
Alameri, M.B., Shebl, H., 2011. Reservoir rock typing of a giant carbonate eld. In: IGRC, Paris, France, vol. 1. Currans Associates, Inc., Red Hook, NY, pp. 733e744.
Paper SPE 148073, Presented at the SPE Reservoir Characterization and Simu- Taware, S.V., Taware, A.G., Sinha, A.K., Jamkhindikar, A., Talukdar, R., Datta-Gupta, A.,
lation Conference and Exhibition, 9e11 October, Abu Dhabi, UAE. 2008. Integrated permeability modeling using wireline logs, core and DST data
Amabeoku, M.O., Kersey, D.G., BinNasser, R.H., Al-Belowi, A.R., 2008. Relative in a deepwater reservoir. In: Paper SPE 113599 Presented at the SPE Indian Oil
permeability coupled saturation-height models based on hydraulic (Flow) units and Gas Technical Conference and Exhibition, 4e6 March, Mumbai, India.
in a gas eld. SPE Reser. Eval. Eng. J. 11 (6), 1013e1028. http://dx.doi.org/ http://dx.doi.org/10.2118/113599-MS.
10.2118/102249-PA. Tomutsa, L., Mahmood, S.M., Brinkmeyer, A., Honarpour, M., 1990. Application of
Archie, G.E., 1952. Classication of carbonate reservoir rocks and petrophysical integrated pore-to-core image analysis to study uid distribution in reservoir
considerations. AAPG Bull. 36, 278. rocks. In: Paper SPE 20478 Presented at the SPE Annual Technical Conference
Choquette, P.W., Pray, L.C., 1970. Geologic nomenclature and classication of and Exhibition, 23e26 September, New Orleans, Louisiana. http://dx.doi.org/
porosity in sedimentary carbonates. AAPG Bull. 54, 207. 10.2118/20478-MS.
Creusen, A., Maamari, K.K., Tull, S., Vahrenkamp, V., Mookerjee, A., Van Rijen, M.F.J., Wunderlich, R.W., 1985. Imaging of wetting and nonwetting phase distributions:
2007. Property modelling small scale heterogeneity of carbonate facies. In: application to centrifuge capillary pressure measurements. In: Paper SPE 14422
Paper SPE 111451, Presented at the SPE/EAGE Reservoir Characterization and SPE Presented at the SPE Annual Technical Conference and Exhibition, 22e26
Simulation Conference, 28e31 October, Abu Dhabi, UAE. September, Las Vegas, Nevada, 1985. http://dx.doi.org/10.2118/14422-MS.
Dunham, R.J., 1962. Classication of carbonate rocks according to depositional Ye, S.-J., Rabiller, P., 2000. A new toolfor electrofacies analysis: multi-resolution
texture. In: Ham, W.E. (Ed.), Classication of Carbonate Rocks, AAPG Memoir 1, graph-based clustering. In: Paper SPWLA-2000-p Presented at the SPWLA
pp. 108e121. 41st Annual Logging Symposium, 4e7 June, Dallas, Texas.
Fleury, M., 2002. Resistivity in carbonates: new insights. In: Paper SPE 77719