SPWLA 56th Annual Logging Symposium, July 18-22, 2015
AN MICP-BASED PORE TYPING WORKFLOW – CORE SCALE TO LOG
SCALE
Paul N. Theologou, Mark Skalinski & Robert K. Mallan
Chevron Energy Technology Company, Houston, Texas
Copyright 2015, held jointly by the Society of Petrophysicists and Well Log
Analysts (SPWLA) and the submitting authors.
This paper was prepared for presentation at the SPWLA 56th Annual Logging
Symposium held in Long Beach, California, USA, July 18-22, 2015.
ABSTRACT
Carbonate reservoirs are inherently complex in their
nature. This complexity is due to a combination of
depositional rock fabric textures and diagenetic
modification of the rocks. Post-depositional processes
can modify the original petrophysical properties (e.g.
permeability, irreducible water saturation and relative
permeability) and result in a disconnection between
original depositional rock fabric and current reservoir
properties.
Pore types are a critical element of rock types since
they exert a dominant control over petrophysical
properties and fluid flow. Their proper definition is
especially important in complex carbonates with
multiple pore systems. Several papers, however, restrict
pore typing to MICP groups without transferring to log
domain necessary for reliable earth modeling.
A procedure has been developed to describe the
dominant pore types occurring within a carbonate
reservoir based on the interpretation of standard core
data, mercury injection capillary pressure data and
wireline log data. This procedure incorporates the
following components: sample selection methodology,
data acquisition, data quality control and corrections,
parameterization of the MICP curves using Gaussian
decomposition, clustering, extrapolation of MICP
derived pore types groups (PTGs) to all core plug
samples, and lastly prediction in the log-domain.
The workflow described here is unique in that it
describes the process from sample selection through
log-scale prediction, PTGs are defined independently of
the original depositional geology, parameters which
describe the whole MICP curve shape are utilized, and
objective clustering is used to remove subjective
decisions.
In this paper, we will describe the proposed workflow
1
and present a case study from a carbonate field where
characterizing PTGs in this way provided a better
understanding of controls on rock properties and fluid
flow than was achieved by looking at the depositional
facies alone.
INTRODUCTION
The combination of complex depositional rock fabric
textures and diagenetic modification of carbonate rocks
leads to complex pore systems. Post-depositional
processes can modify the original pore system
characteristics leading to changes in petrophysical
properties like permeability, irreducible water
saturation and relative permeability, resulting in a
disconnection between original depositional rock fabric
and current reservoir properties. Wu et al. (2013) study
the impact of diagenetic processes through the
quantification of pore size and connectivity, and its
impact on FZI-based rock typing.
Skalinski & Kenter (2013a, 2015) describe an
integrated workflow for petrophysical rock typing
which involves acquiring the right data to decode the
impact of depositional and diagenetic processes on the
creation of in-situ reservoir properties and property
distributions. Step 4 of that workflow is termed
“determine pore type and pore type grouping”. This
paper expands upon step 4, to the extent required to
obtain quality results using MICP data, and covers all
parts of the process from data acquisition through to
pore type propagation at the log-scale. A method has
been developed to describe the dominant pore type
groups (PTGs) occuring within a carbonate reservoir,
and their associated petrophysical properties. These
PTGs are initially determined independently of
geological facies.
LINKS BETWEEN MICP MEASUREMENTS
AND PORE CHARACTERISTICS
The relationship between the capillary pressure and the
curvature of the surface or wall is defined in the Young-
Laplace equation (Gueguen & Palciauskas, 1994)
SPWLA 56th Annual Logging Symposium, July 18-22, 2015
1 1
𝑃𝑐 = 𝜎 ( + ), (1)
𝑅1 𝑅2
where Pc is capillary pressure, 𝜎 is the interfacial
tension, and 𝑅1 and 𝑅2 are the principal radii of
curvatures of the surface.
If we assume the capillary tube model, where 𝑅1 and 𝑅2
are equal and account for the contact angle (𝜃) of the
system, we get the more common form of the equation:
2𝜎𝑐𝑜𝑠𝜃
𝑃𝑐 = (2)
𝑅 pressure associated with pore sub-systems in a sample.
With the understanding that rocks containing differing
pore systems will yield different shaped capillary
pressure curves, representing the capillary pressure
curve with a set of parameters that embodies these
differences provides a means to describe and classify
rocks according to unique combinations of these model
parameters. Furthermore, because the pore network
governs the movement of fluids, the same model should
be applicable for saturation height analysis and
permeability prediction.
Thomeer (1960) introduced the parameterization, or
modeling, of mercury injection capillary pressure data
with a function that approximates a hyperbola in
logarithmic Pc – Vp space. Thomeer’s hyperbola
𝑃
𝑉𝑃𝑐 −𝐺/log(𝑃 𝑐 )
= 𝑒 𝑑 (3)
𝑉𝑃∞
is defined by three parameters that are related to
physical characteristics of the rock’s pore structure; the
extrapolated displacement pressure (Pd), the volume of
the interconnected pore system (𝑉𝑃∞ ), and a pore
geometric factor (G).
Robinson (1966) classified different rock types using
MICP-derived curve characteristics very similar to
those proposed by Thomeer. The typical MICP
characteristics were described for certain carbonate and
clastic lithologies and correlated with some associated
reservoir properties and visual characteristics.
Al-Aruni et al. (1998) presented a methodology for
generating rock types using MICP data. Classification
of rock types was made by manual sorting of
parameters derived form the MICP, porosity and
permeability data (e.g. RQI).
Marzouk et al. (1998) used pore throat size distributions
2
to classify depositional and diagenetically altered
carbonate reservoirs into rock types for use in reservoir
characterization. This classification scheme dominantly
used relative proportions of macro/micro porosity to
partition rock types.
Clerke et al. (2008) investigated over 500 MICP curves
using a Thomeer fitting algorithm to describe the
dominant pore systems in each sample. They defined
porositons as distinct and separable frequency
distributions of Thomeer-derived displacement
pressures from MICP data representing the pore entry
Rock types were classified based on combinations of
porositons identified and the link to geological facies.
In this classification scheme, the displacement pressure
was the major controlling parameter.
MODELING MICP DATA
Thomeer hyperbolas. Fitting the capillary pressure
curve of a multi-modal pore system with a combination
of individual hyperbolas was also proposed by Thomeer
in 1960 and several authors have interpreted MICP data
using utilizing multiple Thomeer fitting methods (Gao
et al., 2011; Clerke et al., 2008 ).
Initial research by the authors of this paper focused on a
methodology incorporating multiple Thomeer
hyperbolas, however there are two features of the
Thomeer hyperbolas that the authors of this paper
believe limit its broad applicability. When applied to
multimodal pore systems, Thomeer’s formulation
produces artificial discontinuties in the capillary
pressure and pore throat size distribution cuves at the
merge pressure of the hyperbolas – this is especially
obvious when viewed in the derivative space (Figure 1).
These discontinuities are a direct result of the
superposition of one hyperbola on another, and are not
present in the measured data.
The second observation is that the Thomeer hyperbola
does not show the Gaussian nature of many pore throat
size distributions when plotted in derivative log10(Pc)
space (Figure 1). It is true that some pore throat size
distributions show a skewness to smaller throats,
however the majority of the data analyzed by the
authors of this paper could be better modeled using a
log-normal function. Misfit is also evident in the low
pressure part of MICP curve where the Thomeer
extrapolation cannot account for a variation in larger
pore throat sizes within a pore sub-system.
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015

Fig.1 Modeling of MICP data using two Thomeer hyperbolas on the right and two modified gaussian error functions
on the left. The discontinuity created by the hyperbola is most obvious in the pore-throat size distribution plot.
Note also that in this sample, the skewness of the hyperbolic function is not observed in the measured data. The
modified Gaussian (left hand plots) provides for smooth transitions between pore systems, and provides a log-
normal shape as observed in the measured dataset. Black open circles are measured data, blue solid lines are the
individual modes, and the red solid lines are the final modeled result.

Gaussian error function. The proposed mathematical can be written as

formula to model a single connected pore system is
𝑉𝑃∞ 2 𝑥 2
derived from the Gaussian error function. Sun (2006)
𝑉𝑃𝑐 =
2
(1 + ∫ 𝑒 −𝑡 𝑑𝑡)
𝜋 0
(4)
proposed a Gaussian shape for fluid component √
decomposition of NMR T2 distributions. Clerke (et al.
2014) used Gaussian distributions for spectral analysis where,
of NMR T2 distrbutions. Xu & Torres-Verdin (2014)
use a Gaussian fitting function to model large and small 1 𝑃𝑐
𝑥= 𝑙𝑜𝑔 ( ), (5)
pore throats associated with tight sandstone reservoirs. 𝑆 𝑃𝑚

The key features of the proposed Gaussian model form
are that it mimics the observed character of many pore
throat systems, the implemented parameters relate to
real characteristics of the pore system, and that multiple
modes can be superimposed without the creation of
discontinuities in the model.
The pore volume for a given capillary pressure (𝑉𝑃𝑐 )
S is the pore system shape factor, Pm is the modal
pressure of the pore system, and 𝑉𝑃∞ is the bulk volume
of the pore system. It should be noted that S is similar
in nature to the G factor implemented by Thomeer, and
in contrast to the Thomeer model, Pm is related to the
pore throat size of the largest connected volume of
pores, instead of the entry pore throat size.

3
SPWLA 56th Annual Logging Symposium, July 18-22, 2015

When more than one pore system is present in the we must sample our reservoir in both geological and
formation, Equations 4 & 5 can be written in a more petrophysical space. This means we should obtain a
general form by summing the contribution of each representative set of samples from the observed
individual system to the total pore volume, where p is petrophysical (porosity-permeability) domain of each
the number of pore systems present, so that DRT. This raises the question of how many samples do
we need.
𝑝 𝑉𝑃∞(𝑖) 2 𝑥(𝑖) 2
𝑉𝑃𝑇 = ∑𝑖=1 (1 + ∫ 𝑒 −𝑡 𝑑𝑡)
2 𝜋 0
√
(6)

and,

1 𝑃𝑐
𝑥(𝑖) = 𝑙𝑜𝑔 ( ). (7)
𝑆(𝑖) 𝑃𝑚(𝑖)

Figure 2 shows the impact of varying the three curve

fitting parameters individually. Varying 𝑉𝑃∞ will result
in a change of the fraction of the total porosity occupied
by that pore system (this changes Swirr for a given
sample). This parameter is exactly the same as defined
in the Thomeer model. Varying Pm will change the
pressure at which the non-wetting phase will enter the
pore system (equivalent impact to changing 𝑃𝑑 in the
Thomeer model). Changing the S parameter changes
the slope of the saturation curve or the width of the
derivative curve (similar impact to changing G in the
Thomeer model). Thus the three optimized parameters
are both practical in their relationship to observable
characteristics of real MICP data, and they should be
familiar to readers who have previously used the
Thomeer hyperbola approach.

MICP PORE TYPING METHODOLOGY

The process described here consists of 7 steps:

1. Sample selection
2. Acquire MICP & Porosity/Permeability data
on same 1-inch core plug
3. Edit and quality control data
4. Curve fit MICP data using Gaussian error
function
5. Cluster the curve fit parameters
6. Extrapolate MICP pore types to core data
using K-Phi KNN method Fig.2 Impact of varying the curve fitting parameters on
7. Extrapolate pore types to log domain using just the shape of the implemented Gaussian error function.
log data
Corbett & Jensen (1992) describe a method for
Step 1 - Sample Selection. Successful execution of the estimating the number of samples required to
clustering and extrapolation steps require gathering of characterize a rock unit based on the coefficient of
sufficient and representative data. At this stage of the variation and the desired precision. This and most other
PRT process, we do not have a known link between
geological processes and petrophysical properties, thus
4
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015

Fig.3 Automated sample selection results applied to a synthetic data set consisting of three facies having different
correlation and variation in porosity-permeability space. A sample is selected from each grid cell that contains two
or more values in it. Open symbols are the original data, red dots are the selected samples, black line is the
regression from the original dataset and red line regression from the selected sample set.

methods are based on the ability to estimate the mean or
expected value of a characteristic of the population. In
this workflow we are trying to capture and describe the
heterogeneity of the formation and acquire sufficient
data to be able to detect links between geological
processes and pore characteristics. For this reason we
propose moving away from the typical methods of
defining sample number requirements and instead use a
more robotic approach to sampling which lets the
character of the existing data inform us about the
number of samples required.
We will use existing routine core analysis (RCA) plugs
as the basis for making our MICP sample selections.
The underlying concept is that we need to take more
samples from DRTs that exhibit poor correlation in
porosity-permeability (P-K) space, and less samples
from DRT’s that exhibit good correlation in P-K space.
Likewise, we need more samples when there is a large
range in porosity and/or permeability, and less samples
for smaller ranges. To achieve this, for each DRT the
RCA data is gridded into a pre-defined number of
porosity and permeability bins based on the range and
number of all the available data. MICP samples are
selected from the center (or as close to as possible) of
each of the grid cells that contain existing RCA data.
We exclude outliers by limiting our sample selection to
be from grid cells containing two or more samples and
by looking at the content of neighboring cells. This
method does not intend to capture the statistical
frequency of samples in P-K space but the rather the
heterogeneity of the dataset.
5
In figure 3 the sample selection method is applied to a
synthetic dataset consisting of 3 depositional facies of
varying correlation and variability. The number of
samples selected from each facies is a function of the
correlation of porosity and permeability and the
variability of those properties. The total number of
samples per facies or the frequency of samples within a
given porosity/permeability range both have little
impact.
Step 2 – Data acquisition. Most routine core analysis
data is acquired using 1.5 x 2 inch core plugs and
historically MICP data has been acquired on end trims
from those plugs. For homogenous samples this will
produce reasonable results, however for heterogenous
formations (like many carbonates) the end trim pore
system may not be representative of the whole 1.5 inch
diameter plug. Figure 4 shows an example where a CT
scan of a core plug allows us to investigate the how
representative an end trim is of the whole plug. In this
example the size and quantity of the pores located at
each end of the plug are quite different.
Sampling the plug via a 1 x 1 inch sub-core provides a
sample big enough to be able to re-measure porosity
and permeability prior to the MICP analysis, meaning
we will have the data acquired from exactly the same
piece of rock. This is very important when we are
trying to relate pore structure to measured permeability.
While the measurements acquired can be extended to
beyond those listed below, the following is the
minimum required core analysis sequence in support of
this process:
SPWLA 56th Annual Logging Symposium, July 18-22, 2015
1. Acquire plug CT scan, white light photograph,
and description on selected samples to identify
internal heterogeneities and surface features
prior to sub-coring
2. Sub-core representative 1x1 inch plug
3. Undertake suitable cleaning (taking into
account mineralogy, fluids and previous
cleaning)
4. Measure porosity, grain density and
permeability using accepted API techniques
5. <undertake any other additional methods such
as 2D-NMR> - optional
6. Undertake additional cleaning if required to
remove salts
7. MICP measurement
Fig.4 Traditional methods use offcut trims for analysis
such as MICP which may not be presentative of the
whole core plug in heterogeneous reservoirs such as
carbonates. This method proposes acquiring all data
on the same bit of (smaller) rock volume to ensure
equivalency of the various measurements.
Step 3 – Data editing and quality control. Only quality
data should be progressed beyond step 3, so the purpose
of this step is to filter out data of poor or suspect quality
and make corrections to the remaining data where
appropriate. The following corrections (while not
always applicable) should be considered in every case:
 Blank corrections (machine calibrations)
 Closure (conformance) corrections
 Merging of low and high pressure chamber
data
 Removal/interpolation of bad data points
 Smoothing or rejection of noisy data
6
Blank corrections are needed to account for the
compression of components within the MICP testing
equipment, and failure to account for this can
significantly impact the interpreted results (Shafer &
Neasham, 2000). Closure or sample conformance
corrections account for low pressure conformance of
the mercury to the surface topography of the sample
prior to mercury entering the connected pore system of
the sample. At some pressure, there is sufficient force
achieved to cause the mercury to enter into the largest
pore throats. All intrusion data recorded up to this
initial pore entry pressure is assigned as a conformance
volume and subtracted from the total intrusion volume
(Shafer & Neasham, 2000).
Fig.5 An example of the correction needed for
adjustment of poor low to high merged data. The black
data are the from the original measurement, and the red
data after removal of the bad point.
In modern high pressure mercury porosimeters, the data
is acquired in two pressure stages: low pressure (0 – 30
psi) and high pressure (30 – 60,000 psi). Between the
two stages the sample must be moved between
chambers and re-oriented 90 degrees. Poor handling or
failure to make appropriate corrections can result in
data merging problems of the low and high pressure
intrusion data. Figure 5 shows an example of a data set
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015

that has had low-high pressure merging problems. In
some cases the problem can be solved by interpolating
across the bad data in absolute incremental volume
space and re-computing the cumulative volume, while
in some cases the sample may need to be rejected.
Samples will be rejected for which the MICP data is
noisy, or for which the difference between the He-
injection and MICP porosity is large. Samples for
which the plug permeability measurement failed will be
rejected from the initial interpretation workflow, but
can be re-included later in the workflow using a MICP-
derived permeability. At the conclusion of step 3 the
accepted dataset (i.e. excluding any rejected samples)
should be re-cycled to step 1 to ensure that geological
and petrophysical representativeness still holds true. If
not then further sampling should be undertaken and
those additional samples taken through steps 2 & 3.
Step 4 – Curve-fit MICP data using Gaussian error
function. A method is developed to mathematically
describe (mercury injection) capillary pressure
measurements of rock samples having multimodel pore
systems. An inversion algorithm, driven by the
modified Gaussian error forward model, composes a set
of capillary pressure models characterizing respective
pore sub-systems into a cumulative capillary pressure
model that approximates the measured capillary
pressure data.
saturation (𝑉𝑃𝑇 /𝜑) for a given pressure step k.
The steps applied in the inversion process are as
follows (Figure 6):
1. An initial model is chosen with a large number
of pore systems, say p=4.
2. The inversion is executed and a solution with p
systems is obtained that minimizes the
difference between the measured and
predicted, or modeled, (mercury injection)
capillary pressure data.
3. If an individual system does not satisfy an
acceptance criteria (e.g. the pore system
volume is relatively small, and thus considered
negligible), then it is stripped from the model,
and the remaining systems from the solution
form a new starting model with p=p-1 pore
systems.
4. Steps 2-3 are repeated until a solution is
obtained where all pore systems meet the
acceptance criteria.

The goal is to model, or parameterize, capillary

pressure data in such a way as to translate the
characteristics of the capillary pressure curve (and thus
the characteristics of the pore network) into a few
parameters that may be easily grouped into families of
similar combinations of parameters. The inversion
algorithm decomposes (mercury injection) capillary
pressure data into a set, or combination, of pore sub-
system models, where each pore sub-system model is
uniquely described by three parameters. The inversion
Fig.6 Flow diagram showing the inversion process.
algorithm solves for the optimum combination of pore
sub-system models that fit the measured capillary
pressure data within a desired tolerance. A least
squares minimization of the objective function (),
Figure 7 and 8 present examples of the inversion
applied to real mercury injection capillary presure
𝜀 = ∑𝑛𝑘=1(𝑆𝑤𝑘 − 𝑆̂𝑤𝑘 )2 (8) measurements conducted on carbonate rock samples
bearing characteristics of bimodal and trimodal pore
is performed by varying the curve fitting parameters systems, respectively. Figure 7a and Figure 8a show
and the number of pore systems, where n = number of the measured and modeled capillary pressure curves,
pressure measurements for a given sample, 𝑆𝑤𝑘 is the while the measured and modeled pore throat size
measured wetting phase saturation for a given pressure derivatives are shown in Figure 7b and Figure 8b.
step k and 𝑆̂𝑤𝑘 is the re-constructed wetting pahse

7
SPWLA 56th Annual Logging Symposium, July 18-22, 2015
Fig.7 Example (a) mercury injection capillary presure
curve and (b) the derived pore throat size distribution
showing the model fit to measured data using the new
modified Gaussian error function. The models of the
individual pore modes (blue lines) combine to form the
final multimodal pore model (red line).
The solid blue lines are the pore sub-system models,
described by Equations 4 & 5, that combine to form the
final model that approximates the measurements.
These examples show excellent agreement between the
measured data and the models predicted by the
inversion. The developed inversion algorithm provides
fast and accurate quantification of pore networks of
rock samples subjected to (mercury injection) capillary
pressure measurements. This provides a means to
accurately and efficiently compare large data sets, thus
enabling the classification of many rock samples
according to combinations of modeled parameters.
Step 5 – Clustering of MICP data to create PTGs.
Once the parameters have been derived for all samples
having quality MICP, porosity and pemeability data,
selected curve fit parameters and other related data are
8
Fig.8 Example (a) mercury injection capillary presure
curve and (b) the derived pore throat size distribution
showing the model fit to measured data using the new
modified Gaussian error function. The models of the
individual pore modes (blue lines) combine to form the
final multimodal pore model (red line). These data are
characteristic of a system with three dominant pore
sizes (trimodal pore system).
used to cluster the samples into groups with similar
pore type characteristics i.e. PTGs. The most effective
clustering variables may change from reservoir to
reservoir. Generally we have found that the best
clustering results come from using the parameters from
the largest pore systems, along with total porosity,
permeability and the fraction of micro-porosity. This
process uses the MRGC clustering method (Ye &
Rabiller, 2000) as it does not require the user to define
the number of clusters, but rather identifies the number
of clusters (at a given resolution) that are naturally
present in the dataset.
Figure 9 shows a crossplot of porosity and permeability
regions for MICP samples, from an example carbonate
reservoir, colored by the cluster results from MRGC
analysis. The inset shows the PTR derivative shape that
is typical of each region. This regionalized space is the
result of the clustering step.
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015
Fig.9 Example of porosity permeability regions
coloured by pore type groups derived from clustering of
MICP-derived parameters. The inset shows the MICP
PTR derivative shape that is typical of the PTG for the
interpreted dataset.
Step 6 – Extrapolate MICP-PTGs to available core
data. With the final objective of log-propagation in
mind, the purpose of step 6 is to increase the statistical
basis for log-domain training, through extrapolation of
the MICP-PTGs to all available routine core data
(Figure 10) to generate core-PTGs. This step is
especially important for fields where limited MICP data
is available. The assumption made during this
extrapolation step is that there exists a singular
relationship between porosity, permeability, and grain
density (PKG) with the pore type group. This allows us
to use the PKG values for each routine core analysis
plug to estimate the associated core-PTG. The method
of nearest neighbor log prediction (F-NN & K-means,
Rabiller, 2011) is employed to perform the prediction,
as it is non-linear and does not rely on any statistical
relations in the dataset.
Step 7 – Propagate core-PTGs to log domain. The
core-PTGs, in combination with a set of selected log
measurements and calculated log curves, are used to
build a training dataset for the puporse of PTG
prediction in the log domain (log-PTG). The optimum
set of log curves may be derived using one of the
various methods available in commerical software
packages (such as stepwise discriminant analysis as
9
applied by Skalinski & Kenter (2015). During this step,
successful propagation at the log scale requires careful
attention to core-log depth matching, correction and
quality control of log data, appropriate inclusion of
DRT information, and may require some careful
lumping of PTGs depending on their respective log
characteristics.
Fig.10 In this depth plot, the significant increase in
data is observed when MICP pore types are
extrapolated to all core plugs with PKG measurements.
Track 1 is depth (in feet), track 2 is permeability of
MICP samples, track 3 is the MICP-PTG from
clustering, track 4 is core plug permeability and track 6
is the predicted core-PTG. Color shading represents
different PTGs.
Any log prediction should include an objective measure
of success. We choose to build a blind dataset
(consisting of 20% of the original data) on which to
perform a cross-validation of the propagation model.
For the test process, we use the remaining 80% of the
data to build the test propagation model and measure
success by applying the test model to the blind dataset.
If the test is successful, 100% of the data is used to
build the final propagation model. At completion of the
process, log-PTGs based on clustering of parameterized
SPWLA 56th Annual Logging Symposium, July 18-22, 2015

MICP data (which is representative of the petrophysical
and geological variation within the reservoir) will be
available. These log-based predictions are at the right
scale to be integrated with other log-based geological
information to generate petrophysical rock types as
described in Step 5 of the PRT workflow (Skalinski &
Kenter, 2013). PTGs are an integral part of the
defintion of PRT’s, the other critical component being
geological processes (depositional and/or diagenetic).
In the PRT workflow, the geological processes are used
to define the geometry of resevoir model building
blocks, and the PTGs define the associated
petrophysical properties.
heterogenity observed at the core scale. This method
also ensures internal consistancy between the PTG,
porosity, permeability and water saturation calculations.
Figure 11 shows an example of synthetically generated
MICP curves. The data was produced using 100
realizations based on the distribution of curve fitting
parameters shown in Figure 12. The curve fitting
parameter distributions are derived from actual MICP
datasets and include any inherent cross-correlations in
the parameters. Figure 13 is the calculated
permeability based on a calibrated Winland R35
(Kolodzie, 1980).

MICP-PTG, WATER SATURATION &

PERMEABILITY MODELS

Having defined MICP-PTGs as the basis for

propagation, we can also calculate petrophysical
properties such as permeability and water saturation for
each of the PTGs. We can make these calculations
using averaged MICP data from each of the PTGs or we
can use a Monte-Carlo approach to build probabilistic
representations of the expected variability within each
PTG. Both methods are described below.

Permeability prediction calibration. Several models

exist for predicting single phase permeability from
MICP data (Purcell, 1949, Thomeer, 1960, Swanson,
1981, Kolodzie, 1980 and Buiting & Clerke, 2011). In
our workflow we test the predictive capability of
calibrated versions of these permeability models and Fig.11 Synthetically generated MICP derivative
select the model that works best for the reservoir under created using a Monte-Carlo process applied to a multi-
investigation. The permeability prediction with the modal modified Gaussian function using a distribution
highest r-squared for a given dataset is selected as the of fitting parameters. Color represents PTG and
preferred method. corresponds to color in Figures 12 & 13.

Monte-Carlo characterization of permeability and
water saturation. Each PTG exhibits a range of
petrophysical properties related to the variation in pore
systems present. For each PTG, the number of pore
systems and the properties of each pore system have
been characterized, providing us with the information
we need to synthetically re-create the observed
variability with a Monte-Carlo approach. Each
realization will generate a synthetic MICP curve based
on the range of observed pore system parameters for a
given PTG, and that MICP curve will be used to
generate a calibrated permeability and an associated
water saturation curve. Assuming that spatial
variability of petrophysical properties within each PRT
cannot be predicted, we can use this method to populate
any 3D model in a way that preserves the PTG
10
Global optimization for unique Sw-Ht function. An
alternative method for deriving PTG specific Sw-Ht
functions is to perform a global optimization of the
curve fitting parameters with measured properties such
as porosity and permeability. In this methodology, the
curve-fitting parameters are defined as functions of
porosity and/or permeability,
𝑃𝑚(𝑖) = 𝑎1𝑖 + 𝑏1𝑖 log(𝑘) + 𝑐1𝑖 𝜑
𝑉𝑝∞ (𝑖) = 𝑎2𝑖 + 𝑏2𝑖 log(𝑘) + 𝑐2𝑖 𝜑
𝑆(𝑖) = 𝑎3𝑖 + 𝑏3𝑖 log(𝑘) + 𝑐3𝑖 𝜑 (9)
where a1, a2, a3, b1, b2, b3, c1, c2 and c3 are regression
fitting parameters globally optimized to build the Sw-
Ht function for each PTG.
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015
Fig.12 Frequency distributions of MICP fitting
parameters (from pore system 1) used to generate the
synthetic data presented in Figure 11. Color represents
PTG and corresponds to color in Figures 11 & 13.
Fig.13 Porosity-permeability derived from the
synthetic MICP data shown in Figure 11 based on
Winland permeability estimates. Color represents PTG
and corresponds to color in Figures 11 & 11.
Ultimately Sw will be described as a function of
capillary pressure (or height above free water level),
porosity and permeability. The results from this
methodology are best applied at the well for
11
reconciliation with log interpretations as they provide
the average saturation for a given PTG having specific
porosity and permeability values. The main limitation
in applying this methodology in multi-modal pore
structures is that prediction of the shape of all but the
largest pore sub-systems is difficult.
CASE STUDY – WEST TEXAS FIELD
McElroy Field, located in the Permian Basin (Figure
14), produces oil from the Grayburg Formation.
Discovered in 1926, McElroy Field has been under
waterflood since the early 1960’s, however maximizing
oil recovery is still a major challenge in this field. The
PRT workflow was applied to an existing dataset from
the McElroy Field, and its outcome was validated using
the available dynamic data (Saneifar et al 2015). An
extensive database of 850 MICP measurements (plug
trim offcut samples) from 62 wells were available as
the basis for MICP pore typing in the McElroy field.
After careful quality control only 501 were accepted
(59%) with rejections due to excess porosity difference
greater than 3 porosity units or due to lack of a
permeability measurement. While this fraction of
acceptance seems low, it is similar to that observed by
Clerke et al (2014). MICP measurements were
characterized by fitting the multi-modal Gaussian
function into each sample measurement. Clustering
(and subsequent lumping) of the MICP parameters of
the two pore sub-systems produced three main MICP-
PTGs (Figure 15) - PTG1 (highest porosity and
dominated by macro-pores), PTG2 (bimodal pores), and
PTG3 (low porosity and dominated by micro-pores).
Fig.14 Location map of the McElroy Field (Dehghani
et al., 1999).
SPWLA 56th Annual Logging Symposium, July 18-22, 2015

Fig.15 Mercury intrusion vs pore-throat radius for the three identified pore types in the McElroy field.
PTG1 is dominantly macro-porous, samples in PTG2 exhibit strong bimodality, while samples in PTG3
are dominantly micro-porous and generally have low total porosity. The red line indicates the separation
between macro and micro pores in the McElroy field.

Saneifar (2015) found that when analyzed by

depositional rock type, there was little separation
in porosity-permeability space. However, each of
the MICP-PTGs occupy relatively unique space on
the core plug porosity-permeability crossplot
(Figure 16). Core-PTGs were generated by
making use of this separation in porosity-
permeability space, assuming that for any given
porosity-permeability pair there is a unique
corresponding PTG. KNN log prediction was used
with an 80% prediction score to extrapolate the
PTGs to the core domain (Figure 17).

Following the core extrapolation process, the PTG

database had 7,400 samples which were used as a
basis for log propagation (Figure 10). The eight
best log predictors were used to build a KNN log
prediction model to generate log-PTGs away from
the cored intervals. The best predictor logs were
picked by comparing prediction scores for all
combinations of a selected list of 15 measured and Fig.16 Cross plot of permeability and porosity
interpreted logs. The logs selected for prediction measurements, colored by the identified pore types
were: illite volume, gamma-ray, bulk density, based on the MICP data in the McElroy field. Pore
neutron porosity, compressional slowness, shallow types exhibit distinct ranges of porosity and
resistivity, photoelectric factor, and neutron- permeability. PT1 has the highest porosity and
density separation. permeability, while PT2 and PT3 have
intermediate and low porosity and permeability,
The quality of the log prediction of PTG was respectively.
cross-validated using a blind dataset which
comprised 20% of the original dataset and showed
a 73% prediction score.
12
 SPWLA 56th Annual Logging Symposium, July 18-22, 2015
Fig.17 Cross plot of permeability and porosity
measurements, colored by the identified pore types
after propagation of pore types to all routine core
analysis plugs.
CONCLUSIONS
A method to derive MICP-based pore type groups and
propagate them through to the log scale is proposed.
The method exploits the link between MICP data and
the petrophysical characteristics of rock samples to
derive self-consistent predictions of PTG, porosity,
permeability and water saturation. The focus is on
acquiring sufficient amounts of good quality
representative data, applying objective clustering
principles, and carrying the interpretation through to the
log scale.
The method is not sufficient in itself for the purpose of
reservoir modelling but is a critical component of the
previously published petrophysical rock typing
workflow of Skalinski and Kenter (2013). While the
focus of this paper has been the petrophysics -
geological integration, in terms of understanding the
geological processes that lead to PTGs and their spatial
distribution is very important for successful application
to reservoir modeling.
ACKNOWLEDGMENTS
The authors would like thank management of Chevron
Energy Technology Company for their continued
13
support of the PRT workflow project. We would also
like to thank Mehrnoosh Saneifar for her work in
performing the pore typing for the case study and Cliff
Cuffey along with Mid Continent Business Unit for
permission to publish the results.
REFERENCES SECTION
Al-Aruni, A, Ali, F. B., Ahmad, H. A. and Samad, S. A.
(1998): Rock type and permeability prediction from
murcery injection data: application to a heterogeneous
carbonate oil reservoir, offshore Abu Dhabi, (UAE).
SPE 49556 ADIPEC Conference.
Buiting, J.J.M. & Clerke, E.A. (2013): Permeability
from porosimetry measurements: Derivation for a
tortuous and fractal tubular bundle. Journal of
Petroleum Science and Engineering.
Clerke, E. A., Mueller III, H.W., Phillips, E.C.,
Eyvazzadeh, R.Y., Ramamoorthy, R and Srivastava, A.
(2008): Application of Thomeer hyperbolas to decode
the pore systems, facies and reservoir properties of the
Upper Jurassic Arab D Limestone, Ghawar Field, Saudi
Arabia: a “Rosetta Stone” approach. GeoArabia, vol
13, no. 4.
Clerke, E., Allen, D.F., Crary, S.C., Srivastava, A.,
Ramamoorthy, R., Aldungaray, P., Savundararaj, P.,
Heliot, D. and Bordakov, G. (2014): Wireline spectral
porosity analysis of the Arab Limestone – from Rosetta
Stone to Cipher. SPWLA 55th Annual Logging
Symposium. Abu Dhabi, U.A.E. May 18-22.
Corbett, P.W.M. & Jensen, J.L. (1992): Estimating the
mean permeability: how many measurements do you
need? First Break, vol 10, no. 3 March.
Gao, B., Wu, J., Chen, S., Kwak, H. & Funk, J. (2011):
New method for predicting capillary pressure curves
from NMR data in carbonate rocks. SPWLA 52nd
Annual logging symposium. Colorado Springs, USA,
14-18 May.
Gueguen, Y. & Palciauskas, V. (1994): Introduction to
the physics of rocks. Princeton university press,
Princeton, New York, 292p.
Kolodzie, S. Jr. (1980): Analysis of pore throat size
and use of the Waxman-Smits equation to determine
OOIP in Spindle Field, Colorado. 55th Annual technical
Conference and Exhibition of the Society of Petroleum
Engineers of AIME. Dallas, texas, September 21-24.
SPE paper 9382.
SPWLA 56th Annual Logging Symposium, July 18-22, 2015
Marzouk, I., Takezaki, H. & Suzuki, M. (1998): New
classification of carbonate rocks for reservoir
characterization. 8th Abu Dhabi Internation Petroleum
Exhibition and Conference. Abu Dhabi, U.A.E. 11-14
October. SPE paper nunber 49475.
Purcell, W.R. (1949): Capillary pressures – their
measurement using mercury and the calculation of
permeability therefrom. Journal of Petroleum
Technology, 1 (2), AIME Transactions 186.
Rabiller, P. (2011): Facimage: its main tools and their
purpose. Unpublished training notes.
Robinson, R.B. (1966): Classification of reservoir
rocks by surface texture. Bulletin of the American
Association of Petroleum Geologists, Vol 50 no 3.
Saneifar, M., Skalinski, M., Theologou, P., Cuffey, C.,
Kenter, J. & Salazar-Tio, R. (2015) – Integrated
petrophysical rock classification in the McElroy field,
west Texas, USA. SWPLA 56th Annual Logging
Symposium, Long Beach, CA, July 2015.
Shafer, J. & Neasham, J. (2000): Mercury porosimetry
protocol for rapid determination of petrophysical and
reservoir quality properties. International symposium
of the Society of Core Analysts. SCA paper number
2021.
Skalinski, M. and Kenter, J. (2013a) Integrated
workflow or method for petrophysical rock typing in
carbonates. United States Patent Application
Publication US 2013/0179080 A1.
Skalinski, M. and Kenter, J. (2015): Carbonate
petrophysical rock typing: integrating geological
attributes and petrophysical properties while linking
with dynamic behaviour. From (eds) Agar, S.M. &
Geiger, S., Fundamental controls on fluid flow in
carbonates: Current workflows to emerging
technologies. Geological Society Special Publication
406, pages 229-259.
Swanson, B.F. (1981): A simple correlation between
permeabilities and mercury capillary pressures.
Journal of Petroleum Technology, December.
Thomeer, J. H. M., (1960): Introduction of a pore
geometrical factor defined by the capillary pressure
curve, 34th Annual Fall Meeting of SPE, paper 001324-
G.
14
Wu, K., Katz, D., Skalinski, M. & Salazar-Tio, R.
(2013): Investigation of the impact of diagenesis on the
petrophysical properties in the complex porosity
systems in carbonates. SPWLA 54th Annual Logging
Symposium. New Orleans, Louisiana, United States,
June 22-26.
Xu, C. & Torres-Verdin, C. (2014): Petrophysical rock
classification in the Cotton Valley tight-gas sandstone
reservoir with a clustering pore-system orthogonality
matrix. Interpretation, Vol. 2, No. 1.
Ye, S. & Rabiller, P. (2000): a new tool for electro-
facies analysis: multi-resolution graph-based clustering.
SPWLA Annual Logging Syposium, Dallas, Texas, June
4-7.
ABOUT THE AUTHORS
Paul Theologou received his Ph.D. in applied geology
from the University of South Australia (1997). He
worked as a petrophysicist at Santos Ltd., A.C.S.
Laboratories, and Mincom Ltd. before starting a
petrophysical consulting company (the Saros Group) in
2000. He is currently a research petrophysicist at
Chevron in Houston, Texas, where he has been
developing petrophysical interpretation workflows
since 2008. Paul is a member of SPWLA, SCA and
PESA.
Mark Skalinski is currently a Senior Research
Consultant in Petrophysics in Chevron ETC. He has
M.Sc. (1971) and Ph.D. (1979) degree in Geophysics
from Mining University in Cracow. His previous
assignments include: Tengizchevroil in Atyrau,
Kazakhstan, Chevron Canada Resources in Calgary,
CABGOC in Angola, Husky Oil in Calgary,
ONAREP in Morocco and University of Mining &
Metallurgy in Cracow Poland. Mark’s interests
includes: rock typing methodology, petrophysical
multi-mineral modeling, application of statistical
methods for facies and permeability prediction and
integrated petrophysical field studies. Mark is a
member of SPE, SPWLA and AAPG.
Robert K. Mallan is currently a research petro-
physicist in Chevron ETC. He has a M.Sc. (1995) in
Geophysical Engineering from the University of
Arizona, Tucson, and a Ph.D. (2010) in petroleum
engineering from the University of Texas at Austin.