See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266671420

A Review of Alternative Methods to Classify Rock-Types from Capillary

Pressure Measurements

Article · January 2014

DOI: 10.2523/IPTC-17631-MS

CITATIONS READS

6 1,707

2 authors:

Bruno Lalanne Michel Rebelle

TOTAL TOTAL
10 PUBLICATIONS   34 CITATIONS    30 PUBLICATIONS   248 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Carbonate Petrophysical Synthesis View project

All content following this page was uploaded by Michel Rebelle on 26 May 2016.

The user has requested enhancement of the downloaded file.

IPTC 17631-MS

A Review of Alternative Methods to Classify Rock-Types from Capillary

Pressure Measurements
Bruno Lalanne, TOTAL E&P Qatar, Michel Rebelle, TOTAL E&P

Copyright 2014, International Petroleum Technology Conference

This paper was prepared for presentation at the International Petroleum Technology Conference held in Doha, Qatar, 20–22 January 2014.

This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The material, as presented, does not necessarily
reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at IPTC are subject to publication review by Sponsor Society Committees
of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the International Petroleum Technology Conference is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of
where and by whom the paper was presented. Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435

Abstract
Lab capillary pressure Pc measurements are routinely performed on core plug samples. Applications of Pc curve
measurements are mainly to derive some relationships between water saturation Sw and the height H above the Free-Water-
Level FWL. The resulting correlations can then be cross-checked vs. the Elogs-derived Sw so as to confirm the overall
consistency of the Sw model. Ultimately, the Sw - H relationship can be used as input to the 3D model.

In addition, screening of Pc – Sw relationships often allows defining robust rock-types, based on observation of Pc clusters
vs. the sample porosity, permeability, mineralogy and geological facies. In the case of MICP, the rock-type classification may
be further enhanced by looking at the Pore-Throat-Radius PTR distributions, related to the derivative of the Pc curve.

Several methods to derive Sw – Pc relationships and classify Pc clusters are possible (e.g. Leverett, Thomeer, Johnson,
Cuddy, Skelt, ..) and have been documented and published in the past.

In this paper, we make a synthetic review of the pros and cons associated to each Pc classification method. Eventually, we
propose an innovative way to take the best of several known Pc classification methods in defining a straightforward technique
to derive robust Pc clusters and equations.

Introduction
Abundant empirical equations have been produced since 1940, with Leverett's pioneering work aiming to relate the 3
parameters Sw, Pc, and petrophysics (PHI, k) into one equation predictive of Sw.

The published methods basically differ by the order in which the 2 parameters petrophysics and Pc are adressed; some
authors process the Pc 1st and the petrophysics in 2nd, others in the opposite way. In addition, the Pc curves dataset can be
processed collectively (global fit), or individually.

Based on these 2 considerations, it is possible to gather the methods in 3 main families:

• Class 1: Collective fit, priority on petrophysics then Pc,
• Class 2: Collective fit, equal priority on both petrophysics and Pc,
• Class 3: Individual fit, priority on Pc then petrophysics.
Other combinations are obviously possible but would be difficult to handle in practice.

The 9 most popular methods fall within these 3 categories (with lots of similarities per class):
• Class 1: The Wooddy – Wright / Johnson / Sondena / Heseldin family,
• Class 2:
o The Cuddy / Alger & al. family,
o And the Leverett J function,
• Class 3: The Thomeer / Skelt – Harrison family.
2 IPTC 17631-MS

The underlying idea of all these empirical equations is always the same: try to reduce the dataset to a linear trend at one
point, thanks to an appropriate transformation of variable (the domains in which the data become linear being listed for each
method in a table at the end of this paper). Then, logically, when dealing with a given field dataset to fit, it's recommended to
look 1st for the domains in which a linear trend of the data is obtained (which in turn suggests which equation is the most
appropriate), rather than applying all the equations at random and comparing their output in the 3D model.

1st approach: Wooddy – Wright / Johnson / Sondena / Heseldin family

Johnson is discussed 1st; the alternative techniques specificities are presented next.

Johnson (1987):
The method, illustrated in Fig. 1, consists in plotting the pairs of values [Sw, k] obtained for a few characteristic Pc values
picked on the curves on the left. A Pc dataset is then eligible to a Johnson-fit if it satisfies Equ.1 here below, i.e., if the [Sw, k]
pairs plotted in log – log scales:
• Post along some linear and parallel trends for given Pc values, i.e. of same constant slope A (3rd plot from the left on
Fig. 1),
• Intercepts B being variable with Pc, B being itself linear when plotted vs. Pc in log-log scales (plot on the right).
݈‫ ݃݋‬ሺܵ‫ݓ‬ሻ = −‫ ܣ‬. ݈‫ ݃݋‬ሺ݇ሻ + ‫ܤ‬ ………………………………………………………………………. Equ. 1
‫ݐ݅ݓ‬ℎ ‫ܾ = ܤ ݀݊ܽ ݐݏܿ = ܣ‬1 . ܲܿ ି௕ଶ , ܾ1 ܽ݊݀ ܾ2 = ܿ‫ݐݏ‬
With Pc in psi, k in mD, and Sw in %. As per the Johnson paper, Log (k) can be replaced by PHI, if the fit proves
better. NB: Sw must be input in %, and not v/v, this to avoid B becoming negative in some cases, which would impede
a log fit on B.

On Fig. 1 example, the resulting Sw equation would thus be: log (Sw) = - 0.213 . log (k) + 2.85 . Pc-0.083, which indeed
allows computing Sw whatever the range of k and Pc. In addition to providing a Sw equation, the method allows
differentiating rock-types RT, since distinct RT's would definitively plot on different trends, for a given Pc step (Fig. 2).

Fig.1 Johnson approach

Fig.2 rock-types discrimination with the Johnson method

Methods similar to Johnson's one: The Woody-Wright / Johnson / Sondena family (WWJS)
Johnson method is very similar to the one Wooddy & Wright published in 1955, the only difference being (cf Fig. 3) that
the trend is linear in "Sw vs. log(k)" instead of "log(Sw) vs. log(k)". Like for Johnson, the slope A is constant and independent
of Pc. In 1992, Sondena refined the Wooddy-Wright method, by introducing a variable slope A = a1 . Pc-a2 (cf Fig. 3). In many
case studies, slope A proves effectively variable, and Sondena equation is then better suited. The next logic step is to refine
IPTC 17631-MS 3

Johnson 1987 by adding a variable slope A (like Sondena did for the Woody-Wright equation), and this is precisely the
method the author is using the most (Johnson – modified Lalanne, "basic version" in the table on Fig. 3).

WWJS Family of Wooddy - Wright / Johnson / Sondena Sw fits:

Wooddy-Wright -b2
1955 Sw = - (a1) . Log (k) + (b1 . Pc )
Sondena -a2 -b2 a1, a2,
1992 Sw = - (a1 . Pc ) . Log (k) + (b1 . Pc ) b1 & b2
Johnson -b2 =
1987 log (Sw) = - (a1) . Log (k) + (b1 . Pc ) constants
Johnson - modified Lalanne -a2 -b2
"basic version" log (Sw) = - (a1 . Pc ) . Log (k) + (b1 . Pc )
Fig.3 the WWJS family

Which of the 4 methods is the best suited to Pc modeling?

Both Wooddy-Wright and Sondena fits express that the plots Sw = f (log (k)) are linear (Fig. 3). However, this is clearly
not the case on the 2nd plot from the left on Fig. 1, and numerous case studies effectively confirm, like Johnson, a curvature of
the plot Sw = f (log (k)) toward the lowest perm k values. Since Sondena case study was mainly covering the higher range
[100, 1000 mD], the curvature almost looks negligible giving the impression the Sw vs. log(k) trends are ~ linear in 1st
approximation. This could explain why Wooddy and Sondena proposed a fit on Sw rather than on log (Sw). Nevertheless, the
Johnson's approach (and its Lalanne's modified version) fitting log(Sw), and not Sw, seems the most universal.

Perm k may be replaced by PHI or log (PHI) in all these WWJS equations. A & B coefficients may also be fitted with other
mathematical functions than the hyperbola one. For that reason, the proposed equation on Fig. 4 is the most general form of
the WWJS family.
log (Sw) = - A . X + B
X = log(k), or PHI, or log(PHI), the one that
WWJS - modified Lalanne
yields the best linear fit in log-lin scales
"log(Sw)" fit
A = f1 (Pc), f1 & f2 being whatever
B = f2 (Pc), mathematical functions.
Fig.4 the universal form of the WWJS equations

Heseldin (1974):
As can be seen on Fig. 5, the method is very similar to WWJS on Fig. 1 middle (i.e. plotting some trends for given values
of Pc), except that this is not Sw or log (Sw) which is plotted, but PHI.Shc, the hydrocarbon bulk volume. However, it is our
observation that, on many case studies, the PHI.Shc trends are much more linear (like on Fig. 6) than they are on Heseldin
original Fig. 5. It is also possible to use log(k) instead of PHI. It is then possible to derive the general equation to use for
"PHI.Shc" plots on Fig. 7, quite equivalent to the general formula to use on Fig. 4 for "log (Sw)" plots.

Fig.5 Heseldin approach Fig.6 a generalization of the Heseldin approach

4 IPTC 17631-MS

PHI.Shc = A . X + B
X = log(k), or PHI, or log(PHI), the one that
Heseldin - modified Lalanne
yields the best linear fit in lin-lin scales
"PHI.Shc" fit
A = f1 (Pc), f1 & f2 being whatever
B = f2 (Pc), mathematical functions.
Fig.7 the universal form of the Heseldin equation

Rationale behind the Heseldin method:

In many cases, it appears that a plot of Sw vs. PHI follows either an exponential fit or a power fit (brown and green fits on
Fig. 8 respectively); this remains true for any Pc value. As can be seen on Fig. 8 right, a remarkable property of these 2 fits is
that they turn to pseudo-linear when plotted in PHI.Shc rather than Sw as expressed by the real case example on Fig. 9. That
basically explains the Fig. 6 plot appearance and the rationale behind the Heseldin method. Manipulating PHI.Shc (Heseldin
method) rather than Sw (WWJS methods) is particularly advantageous when the dataset presents dispersion: in that case, the
PHI.Shc best-fit is a true unbiased arithmetic fit, which is not the case with the Sw fit, as it has to be fitted in log (Sw).
Furthermore, the PHI.Shc fit minimizes the contribution of the low PHI – k outliers which do not contribute that much to the
volumetrics.

Fig.8 rationale for the PHI.Shc fit

Sw (v/v), from MICP at Pc Hg = 1000 psi PHI.Shc (v/v), from MICP at Pc Hg = 1000 psi
1 0.3

0.8

0.2
0.6

0.4
0.1

0.2

0 0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0 0.05 0.1 0.15 0.2 0.25 0.3
PHI (v/v) PHI (v/v)
Fig.9 rationale for the PHI.Shc fit

2nd approach: Cuddy and Alger family

This 2nd group of methods present close similarities with the previous WWJS / Heseldin approach. The difference, here, is
that no plot is proposed to tentatively derive the A slopes and B intercepts values. Instead, a global multilinear regression is
done simultaneously over the whole Pc dataset. The method is extremely fast (immediate results), but accuracy depends on the
presence or not of outliers.

Alger – Luffel – Truman ("caplog") method (1989):

The multivariable linear regression (MVLR) is done on PHI.Shc, the bulk hydrocarbon volume, with:
ܲ‫ ܫܪ‬. ܵℎܿ = ‫ ܣ‬+ ‫ ܤ‬. ݈‫ ݃݋‬ሺ‫ܪ‬ሻ + ‫ ܥ‬. ܲ‫…………………………………… ܫܪ‬..………………….………. Equ.2
A, B & C are constants, precisely found by multivariable linear regression MVLR over the whole dataset. The optional
term "+ D . log (k)" may be added if necessary.

Rationale behind Alger & al. (1989) method:

As seen on Fig. 8, power and exponential fits of Sw vs. PHI data (the most common fits on Sw – PHI datasets) turn to
pseudo-linear when posted in PHI.Shc. If in addition A slopes are imposed to be constant, results are given by the plots in the
IPTC 17631-MS 5

middle of Fig. 10 and Fig. 11 (parallel trends of constant slopes for variable Pc values), for power fit and exponential fit
respectively. The equivalent exact plots in PHI.Shc are posted on the right of Fig.10 and Fig.11, suggesting the following
remarks:
• the slopes are still pseudo-linear on the PHI.Shc plots,
• with still pseudo-parallel trends (slope A = cst) for the power fit on Fig.10,
• but non-parallel slopes (A = f (Pc)) for the exponential fit on Fig.11.

Fig.10 PHI.Shc plot for a Sw - PHI power fit

Fig.11 PHI.Shc plot for a Sw - PHI exponential fit

Therefore, in the very specific case of a power fit with constant slope (Fig. 10 middle), one can see that the PHI.Shc vs.
PHI plot on the right follows almost parallel trends vs. PHI.
The equation is therefore: PHI.Shc = A + B . log (H) + C . PHI, i.e. Equ. 2, hence the rationale behind the Alger & al.
equation. In conclusion, the Alger method is extremely fast (immediate MVLR results in fact), but is only suitable for very
specific cases, like having a constant slope on a log(Sw) – log (PHI) trend, which is far from being a generality. In particular,
if the slopes are not constant (Fig. 11 right or Fig. 6), the method is not recommended.

Cuddy (1993):
The MVLR is done on log (PHI.Sw), the logarithm of the bulk water volume, with:

݈‫ ݃݋‬ሺܲ‫ ܫܪ‬. ܵ‫ݓ‬ሻ = ‫ ܤ – ܣ‬. ݈‫ ݃݋‬ሺ‫ܪ‬ሻ …………………………………….………………………………………Equ. 3

A & B are constants, precisely found by MVLR over the whole dataset.

Note that Equ. 3 can be written:

݈‫ ݃݋‬ሺܵ‫ݓ‬ሻ = ‫ ܣ‬− ‫ ܤ‬. ݈‫ ݃݋‬ሺ‫ܪ‬ሻ – ݈‫ ݃݋‬ሺܲ‫ܫܪ‬ሻ ……………………………………………….……………….. Equ. 4

As alternative to Equ. 4 (original Cuddy equation), it is then possible to add a coefficient "C" to log(PHI), and end-up with
Equ. 5 (Cuddy – modified Lalanne) instead:

݈‫ ݃݋‬ሺܵ‫ݓ‬ሻ = ‫ ܣ‬− ‫ ܤ‬. ݈‫ ݃݋‬ሺ‫ܪ‬ሻ– ‫ ܥ‬. ݈‫ ݃݋‬ሺܲ‫ܫܪ‬ሻ………………………………....…………………………. Equ. 5

that is:

݈‫ ݃݋‬ሺܲ‫ ܫܪ‬஼ . ܵ‫ݓ‬ሻ = ‫ ܤ – ܣ‬. ݈‫ ݃݋‬ሺ‫ܪ‬ሻ…………………………..……………………………………………. Equ. 6

to be compared with Equ. 3 where C = 1.
6 IPTC 17631-MS

The reason why Cuddy omitted this coefficient "C" (C set to 1 actually) is unclear, as the MVLR is equally quick and
straightforward regardless of the C coefficient value in output. Indeed, as will be seen in a while on a real case study, Equ. 5
or Equ. 6 (Cuddy modified Lalanne) allows to improve a lot the data fits compared to the basic Cuddy equation (Equ. 3 or
Equ. 4).

Rationale behind Cuddy (1993) method:

As seen on Fig. 8, the transform in the bulk hydrocarbon PHI.Shc domain can yield very convenient linear trends.
However, as posted on Fig. 12 left, this is not anymore the case for the bulk water domain (unremarkable plots on the left). In
the very specific case of a power fit though, the product PHIC.Sw is constant (cf Fig. 12 middle, with C = 1.58 as example).
This results from the fact that the power fit is a hyperbola, Sw = 0.017 . PHI-1.58 in our example on Fig. 8 and Fig. 12, i.e.
(PHI1.58.Sw) = 0.017 = cst. If in addition the power fit slopes are parallel (constant slope as on Fig. 10 middle), it results in
Fig. 12 right plot, a series of horizontal lines parallel where the intercepts are function of H, i.e. some equations log (PHIC.Sw)
= intercept, with intercept = A - B . log (H), i.e. log (PHIC.Sw) = A - B . log (H), which is Equ. 6. This is the rationale for the
Cuddy method, except that Cuddy considered C = 1 in Equ.3, i.e. a very particular case of Sw – PHI hyperbole, with a slope =
1 in log – log domain.

Therefore, one can note that both Alger & Cuddy methods stem on the very same hypothesis of a power fit with constant
slopes in the log (sw) - log (PHI) domain (Fig. 10 middle). These 2 methods are then just 2 slightly different expressions (one
in PHI.Shc, the other in PHIC.Sw) of a same underlying fit type.

C
Fig.12 rationale for the PHI .Sw fit (Cuddy – modified Lalanne)

Practical example using Alger (1989) and Cuddy (1993) methods:

One of the case study shown by Alger 1989 is based on the Heseldin data. The plots on Fig. 13 are thus, respectively, the
original Heseldin dataset on the left, and the Alger fit on the right (lines of equation: PHI.Shc = A + B . log (H) + C . PHI, i.e.
Equ. 2). The plot on Fig. 14, is the transform of the Heseldin data, from PHI.Shc into Sw. Each Sw – PHI trend can be
approximated by a log-log power fit, which average slope is around - 1.78, which is far from the base Cuddy hypothesis which
states that the slope should be - 1 (Equ. 4); without surprise, the Cuddy fit is then going to fail.

Fig.13 Heseldin vs Alger & al. methods ("caplog") Fig.14 Heseldin plot converted in Sw
IPTC 17631-MS 7

Ability of the 4 methods to modelize Heseldin dataset, respectively Alger & al., Cuddy original, Cuddy modified Lalanne,
and WWJS is assessed on the next figures Fig. 15, 16, 17 and 18. The accuracy is rated by comparing the PHI.Shc on the left
plot (prediction vs. original), and the Sw on the right plot (superimposed Sw vs. PHI plots).

Alger & al. on Fig. 15:

• The PHI.Shc fit looks acceptable, however the Sw are obviously underestimated for the high Pc in the high PHI range
(check the 400 & 700 psi curves over the [10, 14 PU] PHI range). This is due to the fact that the original Heseldin
PHI.Shc trends are not linear from the start.

Original Cuddy on Fig. 16:

• The Sw fit is obviously poor on the right (modelized slopes are significantly oblique compared to the original ones),
and this results from the arbitrary C = 1 coefficient in Cuddy's Equ. 4 (while the average is obviously rather around C
= 1.78 as seen on Fig. 14).

Cuddy – modified Lalanne on Fig. 17:

• We are simply using Equ. 5 instead of Equ. 4, that is, using a realistic C coefficient (1.78 here, the average of the
slopes on Fig. 14) and not the default Cuddy value of 1.
• As can be seen on Fig. 17, the fit between model and original data is now almost perfect.

WWJS Wooddy-Wright / Johnson / Sondena family type on Fig. 18:

• We used the actual equations on Fig. 14, i.e. with both variable slopes and intercepts function of Pc.
• The result, on Fig. 18, is also quite excellent, but not better than the Cuddy – modified Lalanne case on Fig. 17.
• The reason is simply that the difference in the slopes values on Fig. 14 is not big enough to make a difference between
a constant slope (Cuddy - modified Lalanne) and a slope = f (Pc) (WWJS modelization).

Fig.15 Heseldin vs Alger & al. methods Fig.16 Heseldin vs Cuddy methods

Fig.17 Heseldin vs Cuddy modified methods Fig.18 Heseldin vs WWJS modified methods
8 IPTC 17631-MS

3rd approach: Leverett J function

Leverett was a pioneer in the domain of Pc curve fitting, defining his method in 1940.

The base equation is:

J ሺSwሻ = ሺPc / σ . Cos ሺθሻሻ . sqrt ሺk / PHIሻ……………………….……………………….. ……….………. Equ. 7
All the Pc curves related to the same rock-type are supposed to plot on the same J function, which is expected to be an
hyperbola when plotted vs. Sw:
J ሺSwሻ = a / Swb ……………………………….…………………………………………………………...….. Equ. 8
Sw equation can be derived for the given rock-type:
Sw = [a / ሺሺPc / σ . cosሺθሻሻ . sqrt ሺk / PHIሻሻ]1/b …………………………………….……….………. Equ. 9
The Pc term in the above equations is the reservoir Pc, that is:
Pcres = ∆RHO . g . H……………………………………………………………………………………………...Equ. 10

Rationale behind the Leverett J function method (1940):

The idea of the J function is to normalize Pc curves belonging to a same rock-type. The normalization consists in
multiplying Pc by perm k (or sqrt(k) to render the expression dimensionless, cf here below), which, for a given Sw, are
antagonist one from the other. Indeed, for a given Sw:
• Higher perm k are associated with lower Pc, and reciprocally,
• Therefore, multiplying Pc by perm k will potentially make Pc curves initially separated to gather to some extent,
ideally, if belonging to a same rock-type, plotting on a unique curve, the J function.
• The Leverett J function is labeled "J" because it's resembles the letter J shape in lin-lin scales.

For historical reasons, Leverett chose his J function to be dimensionless. This imposed the square-root function of k / PHI
in Equ. 7 (but as will be seen in a while, this was not compulsory in itself). The "σ.cos(θ)" term in Equ. 7 & 9 contributes to
render the J expression dimensionless as well, but above all, allows normalizing all the Pc to the common reservoir fluids,
should different fluid pairs have been used during the Pc measurements campaign. In some cases, it may prove useful to
proceed with normalized Sw*, rather than Sw, with:
Sw* = ሺSw – Swirr.ሻ / ሺ1 – Swirr.ሻ………………… ……………………………..………………………Equ. 11

Practical case illustrating the Leverett J function method:

The synthetic case plotted in the middle of Fig. 19 illustrates a perfect Leverett J-function fit:
• The upper Pc curves have (logically) a poorer PHI – k petrophysics than the lower Pc curves.
• By multiplying the Pc value by sqrt (k/PHI) for each Sw step, one exactly obtains the plot on the right of Fig. 19, that
is, a unique J function J (Sw) for the 4 Pc curves.
Rather than modelizing each Pc curve individually, the unique J function allows to simultaneously describe the 4 Pc
curves, which yields a tremendous gain in time.

k / PHI (mD / dec) Pcres (bar) J = Pc res * pow (k / PHI, 0.5)

5 10000

1000 k = 1.5 mD Leverett exponent

PHI = 0.15 n = 0.5
4 k = 5.2 mD
1000
PHI = 0.174
k = 20 mD
100 3 PHI = 0.2
k = 250 mD
k = 1.5 mD PHI = 0.25 100
PHI = 0.15
k = 5.2 mD 2
PHI = 0.174
10
k = 20 mD
10
PHI = 0.2
1
k = 250 mD
PHI = 0.25

1 0 1
0.1 0.15 0.2 0.25 0.3 0 0.2 0.4 0.6 0.8 1 0.1 1
Sw (v/v)
PHI (v/v) Sw (v/v)

Fig.19 Leverett approach, with default exponent n = 0.5

IPTC 17631-MS 9

Advantages and drawbacks of the method:

Advantages:
• The main advantage of the method is its rapidity. Calculating and plotting the J values is indeed immediate.
• Unlike other methods, the resulting equation is never un-physical (e.g. no risk of nil or negative entry pressure Pd or
Sw like with other methods).
• By reversing the equation, synthetic Pc curves can be easily generated (which is not always the case with other
methods); that may prove useful for 3D reservoir initialization.
Drawback:
• Though very handy and easy to produce, intuiting "J" values (which are a product of petrophysics (k & PHI) by fluids
properties (∆RHO and H in the Pc term)) is indeed a challenge.
• Leverett's choice to set the J function dimensionless is empirical, i.e. the choice of the 0.5 exponent in the term sqrt(k /
PHI) = (k / PHI)0.5 is empirical. Some datasets may need alternative values of n, as shown on Fig. 20 & Fig. 21:
o On Fig. 20, the Pc dataset cannot be fitted to a unique J function when using n = 0.5 (sqrt in (k/PHI)n), but
matches a unique J function when using n = 0.3. Idem on Fig. 21, where the match is obtained for n = 3.7.
o Actually, the "exponent n" value that allows the match is related to the slope on a Sw vs. log(k) plot (cf Fig. 22):
the higher the slope, the higher the "n" value and reciprocally. It's only a relative rule of the thumb, and there is,
unfortunately, no way to derive the appropriate n value except by trials and errors.
• The use of k/PHI, as in the original equations, assumes that a good PHI – k trend is apparent, which is not always the
case. It may be more convenient to use k alone instead (with other n exponent of course) or even PHI alone (PHI in
numerator, not in denominator).
k / PHI (mD / dec) Pcres (bar) J = Pcres * pow (k / PHI, 0.5) J = Pc res * pow (k / PHI, 0.3)
5 1000 1000

1000 k = 1.0 mD Leverett exponent Leverett exponent

PHI = 0.15 n = 0.5 n = 0.3
4 k = 6.6 mD
100 100
PHI = 0.174
k = 44.3 mD
100 3 PHI = 0.2
k = 2004 mD
PHI = 0.25 10 10
k = 1.0 mD
PHI = 0.15
2
k = 6.6 mD
10 PHI = 0.174
k = 44.3 mD 1 1
PHI = 0.2 1
k = 2004 mD
PHI = 0.25
1 0 0.1 0.1
0.1 0.15 0.2 0.25 0.3 0 0.2 0.4 0.6 0.8 1 0.1 1 0.1 1
Sw (v/v) Sw (v/v) Sw (v/v)
PHI (v/v)

Fig.20 Leverett approach, alternative n = 0.3 better suited

k / PHI (mD / dec) Pcres (bar) J = Pc res * pow (k / PHI, 0.5) J = Pcres * pow (k / PHI, 3.7)
5 10000 1E+014

1000 k = 170 mD Leverett exponent Leverett exponent

PHI = 0.15 n = 0.5 n = 3.7
4 k = 227 mD 1000 1E+013
PHI = 0.174
k = 302 mD
100 3 PHI = 0.2
k = 535 mD
PHI = 0.25 100 1E+012
k = 170 mD
PHI = 0.15
2
k = 227 mD
10 PHI = 0.174
k = 302 mD 10 1E+011
PHI = 0.2 1
k = 535 mD
PHI = 0.25

1 0 1 1E+010

0.1 0.15 0.2 0.25 0.3 0 0.2 0.4 0.6 0.8 1 0.1 1 0.1 1
Sw (v/v) Sw (v/v)
PHI (v/v) Sw (v/v)

Fig.21 Leverett approach, alternative n = 3.7 better suited

Fig.22 qualitative rule of the tumb to assess Leverett exponent evolution

10 IPTC 17631-MS

4thapproach: Thomeer and Skelt-Harrison family

In the 3 previous approaches, all Pc were processed globally / simultaneously. This 4th approach is radically different as it
consists in processing each Pc individually, by basically finding the transformation of variable that allows getting a hyperbola
fit for each Pc, i.e. a linear trend after an extra log-log transform. The 2 parameters of the hyperbolae are then plotted for all
the Pc curves vs. PHI and/or k to derive a final equation Sw = f (Pc, PHI, k). The Thomeer approach is suited for mercury
MICP Pc, while the Skelt – Harrison fit is rather appropriate for lower maxi Pc (gas – liquid or liquid – liquid Pc experiments).

Thomeer (1960):
The original Thomeer equation is expressed as in Equ. 12 which can be simplified in Equ. 13. A Pc curve is then eligible
to a Thomeer fit if it follows a hyperbola (Fig. 23 middle) when log (SHg) is plotted vs. log (Pc/Pd). A linear trend of slope G
is obtained after a 2nd log transform (Fig. 23 right), when log (log (SHg)) is plotted vs. log (log (Pc/Pd)).

-G
∅ . SHg = ∅ . exp ቆ Pc ቇ …………………………………………………………………………………… Equ.12
log ቀPdቁ
Pc & Pd in psi, PHI & SHg in v/v, G: pore geometrical factor, dimensionless

ሺ- G / Ln ሺ10ሻሻ
log ሺSHgሻ = ቆ Pc ቇ…………………………………………………………..……………………… Equ. 13
log ቀPdቁ

Fig.23 Thomeer approach

Due to lack of computational capabilities in the 60', Thomeer produced an abacus (Fig. 24) filled with a large range of
Thomeer hyperbolas and their corresponding G values (ranging from G = 0.0 to G = 10.0 in the original Thomeer abacus),
with the Y-axis graduated in Pc, not anymore in Pc/Pd. The nice point about the abacus is that the hyperbolas have exactly the
same shape for a given G value, whatever the entry Pd (compare both plots on Fig. 24, with Pd = 1 and 140 psi on left and
right respectively), which is the reason why the G factor is called pore geometrical factor; to note, however, that this
"geometry" similarity is only in log-log domain. The plots on Fig. 25 are exactly the same than on Fig. 24, but plotted in lin-
lin scale: there is no more similarity, even for a same G value. The use of the abacus, plotted on transparent paper, was then
straightforward and consisted in plotting the actual Pc data in same 3 decades Y-axis vs. 2 decades X-axis, then shift the
transparent abacus over the plot to find the G value which allowed fitting an hyperbola to the data. The X-axis at the bottom of
the abacus was giving the Pd when projected on the Y-axis of the actual data plot.

Fig.24 Thomeer abacus (log-log) Fig.25 Thomeer abacus (lin-lin)

IPTC 17631-MS 11

No need of abacus with present calculating capabilities: deriving G and Pd is straightforward (Fig. 26): with transformation
of Equ. 13 in Equ. 14, a Thomeer fit plotted in lin-lin scales of log(Pc) vs. 1 / log(Sw) appears as a linear trend with a slope =
– G / Ln(10) and an intercept = log(Pd).
-G 1
log ሺPcሻ = ቀLn ሺ10ሻቁ . ቀlog ሺௌு௚ሻቁ + ݈‫ ݃݋‬ሺܲ݀ሻ ……………………………….………………………….. Equ. 14
The fit process is carried out individually on each available Pc curve (one [G, Pd] pair per Pc curve), and the set of [G, Pd]
pairs are then eventually plotted altogether vs. PHI and k (and per rock-types if available), so as to derive global relationships
G = f1 (PHI, k, rock-type) and Pd = f2 (PHI, k, rock-type), which, once input in Equ. 13, allows to derive a final global
equation Shc = f (Pc, PHI, k, rock-type).

Fig.26 criterion for a Thomeer fit Fig.27 plot Thomeer with SHg units in %

Two mathematical pitfalls to address:

• NB1: Thomeer equation has been written with saturation SHg being expressed in decimal v/v fraction of pore volume
pv, and not in % of pv. It is important to respect this unit v/v, otherwise the linearity on Fig. 26 is lost when Shc in
expressed in %, as illustrated by the plot on the left of Fig. 27. With SHg' in %, the linearity is then only restored when
plotting in X-axis 1 / (log (SHg') – log (2)), plot on the right of Fig. 27.
• NB2: With a Thomeer fit, the only true hyperbola is the one on Fig. 23 middle (plot of Equ. 13), i.e. with Pc / Pd as
Y-axis. With Pc as Y-axis now (plot on the left of Fig. 28), the equivalent log(Pc) vs. log(SHg) plot is not anymore
strictly speaking a true hyperbola (even though it's a perfect hyperbola "copy/paste" just shifted by Pd), as shown by
the plot on the right of Fig. 28: the double-log doesn't result anymore in a linear trend due to the log(Pd) term. It has
however exactly the shape of a hyperbola (~ pseudo-hyperbola), but just shifted by Pd (that's the rationale behind the
Fig. 24 abacus). In conclusion, check the linearity on Fig. 26 plot (and in v/v), never on the Fig. 28 right one.

Fig.28 Thomeer pseudo-hyperbola Fig.29 Thomeer original case study

Regarding the historical form of the Thomeer equation:

The original Thomeer equation is not exactly Equ. 12; the term "PHI.Shg" on the left was originally labeled "(Vb)Pc" by
Thomeer, while the term "PHI" on the right was labeled "(Vb)P∞", which stand for "volume of mercury injected expressed as
fraction of bulk volume", at the given Pc and at infinite Pc respectively. However, once corrected for the compressibility
biases (blank correction, which is done with modern MICP equipment), the simplification on Equ. 12 then Equ. 13 is fully
justified. The plot on the left of Fig. 29 illustrates the original 1960 Thomeer case study expressed as fraction of bulk volume
(PHI.SHg), and on the right of Fig. 29, how it looks like in simplified form as fraction of pore volume (SHg), knowing that
using the Fig. 24 abacus with X-axis expressed as PHI.SHg or SHg doesn't make any difference on the result.
12 IPTC 17631-MS

Skelt - Harrison (1995):

The Skelt – Harrison (SH) fit presents a lot of analogies with the Thomeer fit, except that it's more suitable for rather low
maxi Pc values (centrifuge and porous plate Pc's), while the Thomeer is more suitable for MICP's. The condition to match a
Skelt – Harrison (SH) fit is that the hydrocarbon Shc follows Equ. 15:
஻ ஼
Shc = 1 - Sw = A . exp [ - ቀሺு ା ஽ሻ ቁ ]……………………………………………………………...………..Equ.15
With Shc and Sw in frac v/v, and H in ft in the original SH paper.

In practice, C is set = 1 as a starting point (C is only used as a very fine-tuning final parameter if needed, with C always
remaining close to 1 anyway), and D is set = 0 during the Pc curve fit (it is only used while comparing the Pc Shc to Elogs
Shc, as a tuning parameter on "H above FWL", should the FWL be uncertain on the field). Equ. 15 can then be simplified in
Equ. 16:
ି஻
Shc = A . exp ቀ ு ቁ ……………………………………………………………………………………...……..Equ.16
Each Pc curve is then processed individually, so as to find the A & B values that allows to match the Pc curve with a SH fit.
The process is done through trials and errors, knowing that the A & B are parameters very intuitive to handle. Indeed, as
plotted on Fig. 30:
• The A parameter just allows to translate the curve along the Shc X-axis,
• While the B parameter just allows translating the curve along the H Y-axis.

Fig.30 Skelt-Harrison approach

It's therefore very easy to find the A & B values that allows the SH equation to match the original Pc curve (if the Pc curve
is eligible to a SH fit of course, the criterion being reviewed in a while). The process is repeated for all available Pc curves.
Like for Thomeer's [G, Pd] pairs, the individuals [A, B] pairs are eventually plotted altogether vs. PHI and k (and per rock-
types if available), so as to derive global relationships A = f1 (PHI, k, rock-type) and B = f2 (PHI, k, rock-type), which, once
input in Equ. 16, allows to derive a final global equation Shc = f (H, PHI, k, rock-type). To note, in Equ. 16, that when H
tends toward infinity, exp (-B/H) tends toward exp (0) = 1, i.e. Shc tends toward A. A is somehow an "irreducible" Shc irr.,
with A ~ 1 – Sw irr. The ratio (Shc / A) is therefore a normalized Shc* saturation, varying between [0, 1]. As illustrated on
Fig. 31 right, this allows a gas – brine Pc curve tangenting Shc* = 1 at Pc maxi, like for a MICP (Fig. 31 left).

Fig.31 Skelt-Harrison Shc normalization

IPTC 17631-MS 13

In order to compare SH with Thomeer, H can be replaced by Pd (changing coefficient B in B') in Equ. 16, and the
normalized Shc* equation becomes:
– ஻ᇲ
Shc* = exp ቀ ௉௖ ቁ …………………………………………..…………………………………..………..……..Equ.17
Which is equivalent to Equ. 18:
ି ஻ᇲ / ௅௡ ሺଵ଴ሻ
݈‫ ݃݋‬ሺShc*ሻ = ቀ ቁ …………………………………………….…………….……………..……Equ.18
௉௖

The previous Thomeer fit equation is Equ. 13:

ሺ- G / Ln ሺ10ሻሻ
log ሺSHgሻ = ቆ Pc ቇ…………………………..……………………………………………………….. Equ. 13
log ቀPdቁ

We have shown that Pd can be removed from Equ. 13, as in Equ. 19, and still plot as a pseudo-hyperbola.
- G / Ln ሺ10ሻ
log ሺShcሻ = ቀ log ሺPcሻ
ቁ………………………………………………..……………………………...…….. Equ. 19

Finally, the analogy between Skelt – Harrison fit in Equ. 18 and the Thomeer fit in Equ. 19 is striking, they just differ by
log(Pc) instead of Pc. Otherwise, B and G have obviously the very same impact of translating the curves vertically.

In practice, how to know whether a Pc curve is eligible to a Thomeer fit or a Skelt – Harrison fit:
As seen on Fig. 26, the Thomeer fit transforms to a linear trend when log(Pc) is plotted vs. the inverse (1 / log(Shc)) in
both lin – lin scales. From Equ. 18, we conclude that a Skelt – Harrison fit transforms to a linear trend when Pc (and not
anymore log(Pc)) is plotted vs. the inverse (1 / log(Shc*)) in both lin – lin scales.
The 2 plots on Fig. 32 illustrate the differences:
• On both plots, the blue curve is the Thomeer fit, and the red one the SH fit.
• The plot on the left is the plot which renders a Thomeer fit linear,
• And the plot on the right, the one that makes the SH fit linear.
Furthermore, the plots on Fig. 33 are exactly the same than on Fig. 32, just with the X-axis set back to regular linear Shc:
• As can be seen on the plots on the right of Fig. 32 and above all Fig. 33, the difference is tenuous between Thomeer
and SH fits, and using one fit or the other would not make that much difference (all the more that the 1st part of the
curve is always dubious due to conformance effects).
• On the other hand, the difference becomes significant for higher Pc values, as can be seen on the plots on the left of
Fig. 32 and Fig. 33.

Fig.32 Skelt-Harrison vs Thomeer approaches Fig.33 Skelt-Harrison vs Thomeer approaches

In conclusion:
• The Skelt – Harrison fit is rather valid for low Pc experiments.
• While the Thomeer equation has been developed to deal with mercury MICP, with Pc maxi up to 10,000's of psi.
• The Thomeer fit, however, also performs practically as good as the SH does for the low Pc values case (right plot on
Fig. 33).
The Thomeer fit is therefore a more general solution (larger range of Pc validity) to Pc fitting than the Skelt – Harrison
one.
14 IPTC 17631-MS

Conclusions
4 main families of Pc curve matching published in the literature have been reviewed, encompassing 9 popular methods.
Actually, and without exception, all these methods try to relate the 3 parameters Pc / Sw / petrophysics (PHI or k) and rely on
the fact that it is possible, at one point (through some appropriate transformation of variable), to get a hyperbola fit between
some data. A linear trend is then obtained through log – log transform of the hyperbola, from which 2 characteristic parameters
(the slope and the intercept of the linear trend) can be derived.

This can be achieved in 2 ways, and this is the basis of the 4 classes distinction:
• The hyperbola is defined on Petrophysics (PHI, k) and then the slope and intercepts are tentatively correlated vs. Pc:
case of Wooddy-Wright / Johnson / Sondena / Heseldin family,
• The hyperbola is defined on Pc and then the slope and intercepts are tentatively correlated vs. Petrophysics: case of
Thomeer and Skelt – Harrison family.
• Giving equal weight to Pc and Petrophysics: Leverett plus Alger / Cuddy family. Actually, Leverett and Cuddy / Alger
may be split in 2 sub-classes, according to the way the data are processed (MVLR multivariable linear regression with
Cuddy / Alger, normalization with Leverett).

In all cases, an equation Sw = f (Pc, PHI, k, and rock-type if applicable), is produced, which allows deriving Sw values for
whatever position in the 3D model. Some equations are expressed in Shc rather than Sw, or H rather than Pc, but conversions
are straightforward (Shc = 1 – Sw, Pc = ∆RHO.g.H).

Furthermore, although empirical, it has been demonstrated that in each case there is a sound rationale behind the equations
(to the exception of Cuddy which has to be modified to accommodate a variable coefficient), and that these are not just fancy
mathematical functions fitting dataset empirically. In other words, each of these 9 equations has the potential to match a given
field case study at some point.

Which method(s) to choose for a given case study?

This is partly a matter of balance between accuracy and time consumption (the most accurate methods are the most time
consuming and vice versa). The pro and cons of the 4 families methods are summarized on Fig. 34.
• Categories 2a and 2b (Cuddy / Alger & Leverett respectively) are fast to extremely fast, with quite simple formulation of
the equations, but stem on very strong underlying hypotheses, which, if they are not met, yield very poor accuracy of the
results.
• Conversely, categories 1 and 4 (WWJS and Thomeer / Skelt – Harrison respectively), are very accurate, but quite long
(require multiple pickings and X-plots) and with long mathematical formulas.
Regarding the fastest methods: (Cuddy / Alger, & Leverett)
• Beware that the 2 fastest methods (Cuddy & Alger) are the only ones that cannot differentiate distinct rock-types RT
(while the 3 remaining families do). This is a point to take into account if complex pore topologies are expected, as these 2
methods will then both yield inaccurate results and not deliver any RT clustering.
• Leverett J function is fast as well; beware with the arbitrary exponent though (n = 0.5); it may have to be refined.
Regarding the most accurate methods: (WWJS, & Thomeer / Skelt – Harrison)
• Both WWJS and Thomeer / SH are potentially equally accurate (if the dataset is eligible to the equation in question).
• WWJS is a bit more flexible in term of mathematical fitting equations.
• The X-plots Sw = f (PHI, k) at various Pc steps are a prerequisite of the WWJS method, and are indeed very intuitive and
powerful for 3D model QC. It is an adavantage of WWJS over Thomeer / SH to have these plots right from the beginning.

As a good practice and whatever the method used, we recommend upon completion of a study to plot a few synthetic Pc –
Sw curves for a few characteristic PHI – k values (in particular, those of a few actual Pc data). This allows checking there are
no un-physical Pc curves (which may occasionally occur through fancy extrapolation).

Since all 9 equations can be twisted to a linear trend through appropriate variable transform at some point, the most logic
workflow is to find to which equation an actual dataset is eligible, by looking in which domains it gets linear. This is much
more effective than testing all methods in the 3 D model and try to see the one that best matches the data in 3D (indeed, if the
dataset doesn't qualify to another equation because it doesn't get linear in the domains it should, there is no point to test this
method in real case in the 3D model). Fig. 35 illustrates in which domains each fit is supposed to become linear. Then, by
checking into which domains a given dataset gets linear, one derives which equation qualifies as being the most appropriate to
describe the actual field data.
• NB1: Saturations units (either v/v or %) must be respected for Johnson and Thomeer.
• NB2: Some saturations are in Sw, or Shc or PHI.Shc.
As already mentioned, "PHI.Shc" method (Heseldin, Alger & al.) can become very advantageous when there is some data
dispersion (unbiased arithmetic fit, which minimize the contribution of the low PHI – k outliers).
 distribution of the Pc
class advantages limitations
classification methods
IPTC 17631-MS

- Very powerful and universal methods. Equations allow to derive very - Time consuming (counterpart of accuracy).
accurate fits and is applicable on all fields without exception (the - Length of the equations expression.
counterpart of accuracy being that deriving the equations is time
Wooddy - Wright / - Mathematic extrapolations: plot the synthetic Pc - Sw curves for a
consuming, with need for various successive X-plots).
1 Johnson few characteritic steps of PHI - k to make sure the curve is physical
/ Sondena / Heseldin - The x-plots Sw = f (PHI, k) plotted for several steps of Pc are very intuitive over the full Pc range.
and very useful for practical QC in the 3D model.
- Different rock-type clusters are clearly differentiated.

- Extremely fast (immediate MVLR multivariable linear regression on the - Stems on the 2 very strong hypotheses that the plot Sw - PHI qualifies for a
whole dataset taken globally). power fit, with constant parallel slopes in log - log scales (not very accurate
- Formulation of the equation is simple. method if these 2 hypothesis are not met).
Alger & al. - Does not differentate rock-type clusters since this is a global MVLR
2a
/ Cuddy modified Lalanne - Method applicable on E-Logs data as well (beware of possible statistical regression. If distinct RT's are present, accuracy may be very poor indeed.
bias between irreducible and transition zones though).
- Not very accurate if the original Pc dataset presents some dispersion, noise
or uncertainty (even if only 1 RT present).

- Rather fast (simultaneous combination over the whole dataset). n

-Stems on the very strong default hypothesis that the exponent of (k / PHI)
- Formulation of the equation is simple. is n = 0.5 (square root), while the J normalization may be achieved with
- Never delivers unphysical equations (eg negative Pd). quite distincts n values on some case studies.
2b Leverett J function
- Long trials and errors steps to derive the appropriate n in such cases.
- Inacurate Sw modelization results if the correct n is not input.
- J is not an intuitive parameter to handle (product of k by Pc).

- Quite accurate if the dataset is effectively eligible to a Thommer or - Time consuming (counterpart of accuracy).
Skelt - Harrison fit. - Length of the equations expression.

Fig.34 comparison of the 9 most popular Pc classification techniques

- Different rock-type clusters are clearly differentiated. - Mathematically less flexible than the WWJD family.
4 Thomeer / Skelt Harrison
- Mathematic extrapolations: plot the synthetic Sw - PHI - k curves for a
few characteritic steps of Pc to make sure the curve is physical
over the full PHI - k range.
1 Collective fit, priority on petrophysics then Pc
2 Collective fit, equal priority on both petrophysics and Pc
3 Individual fit, priority on Pc then petrophysics
15
 16

Pc classification methods
fluid saturation linearity ( in linear - linear scales ) warning
(9 most popular methods)
with X-plots:
Wooddy - Wright For several steps of given Pc:
Sondena "Sw" vs "log(k)", or vs "PHI" or vs "log(PHI)"
Sw
Johnson For several steps of given Pc: better to express Sw in % of pore volume (not v/v)
WWJS - modified Lalanne "log(Sw)" vs "log(k)", or vs "PHI" or vs "log(PHI)" to avoid equation failing (log of negative values)
For several steps of given Pc:
Heseldin PHI.Shc
"PHI.Shc" vs "PHI", or vs "log(PHI" or vs "log(k)"

Leverett J function Sw For all Pc together:

"log(J)" vs "log(Sw)", with J = (Pc/σ.cos(θ)) . sqrt(k/PHI)

Thomeer For each Pc curve:

better to express Sw in v/v fraction of pore volume
Shc "log(Pc)" vs "1 / log(Shc)"
(not %) to avoid hyperbola criterion failing
For each Pc curve:
Skelt - Harrison
"Pc" vs "1 / log(Shc*)", with Shc* = normalized Shc
Without X-plots:
Assumed by the MVLR regression:
Alger & al. PHI.Shc
PHI.Shc = A + B . log(H) + C . PHI

Fig.35 comparison of the 9 most popular Pc classification techniques

Assumed by the MVLR regression:
Cuddy modified Lalanne Sw
log(Sw) = A - B . log(H) – C . log(PHI)
IPTC 17631-MS
IPTC 17631-MS 17

Acknowledgements
The authors wish to thank TOTAL for permission to publish this paper. They acknowledge the important contributions of
their TOTAL colleagues who worked in the past on this topic of Pc – Sw reconciliation, in particular FM.Pellerin and
M.Bennes.

Nomenclature
a, b, c, d, A, B, C, D: generally some slopes or intercepts; may be constant or not
BV: bulk volume, no specific unit here
f: a mathematical function
FWL: Free Water Level
G: Thomeer pore geometrical factor, dimensionless
H: height above FWL, in ft or m, with Pc = ∆RHO.g.H
hc: hydrocarbon
Hg: mercury
irr.: irreducible
J: Leverett J function
k: permeability, in mD
lin: linear
log: logarithmic
MVLR: multivariable linear regression
Pc: capillary pressure, in psi or bar
Pd: displacement or entry pressure, in psi or bar
PHI, or Φ: porosity, in fraction of bulk volume BV (v/v or %)
PHI.S, or Φ.S: fraction of bulk volume BV occupied by a fluid at saturation S (v/v or %)
PV: pore volume, no specific unit here
RT: rock-type
SH: Skelt - Harrison method
S: saturation, as fraction of pore volume PV, in % or v/v depending on the method
S*: a normalized saturation
w: water
WWJS: Wooddy-Wright / Johnson / Sondena methods
∆RHO: difference of fluid densities between wetting and non-wetting phases

References:
Johnson A., "Permeability Averaged Capillary Data: A Supplement To Log Analysis In Field Studies", SPWLA Twenty-Eighth Annual
Logging Symposium, June 29-July 2, 1987

Sondena.E, "An Empirical Method For Evaluation Of Capillary Pressure Data", paper SCA 1992-07, 1992

Alger A.P., Luffel D.L., Truman A.B., "New Unified Method Of Integrating Core Capillary Pressure Data With Well Logs", SPE
Foundation Evaluation. June 1989

Cuddy S., Allinson G., Steele.R, "A Simple, Convincing Model For Calculating Water Saturations In Southern North Sea Gas Fields",
SPWLA 34th Annual Logging Symposium, June 13-16, 1993

Leverett M.C., "Capillary Behavior In Porous Solids", in Petroleum Technology August 1940, Tulsa Meeting, October 1940

Thomeer J. H. M., "Introduction Of A Pore Geometrical Factor Defined By The Capillary Pressure Curve", AIME technical note 2057,
March 1960

Skelt C., Harrison B., "An Integrated Approach To Saturation Height Analysis", SPWLA 36th Annual Logging Symposium, June 26-
29,1995

Harrison.B, Jing X.D., "Saturation Height Methods and Their Impact on Volumetric Hydrocarbon in Place Estimates ", paper SPE 71326,
SPE Annual Technical Conference and Exhibition held in New Orleans, Louisiana, 30 September–3 October 2001

View publication stats