SPE 93125

Quantifying Petrophysical Uncertainties
S.J. Adams, WellEval.com Ltd.

Copyright 2005, Society of Petroleum Engineers Inc.
users know how reliable the numbers produced actually are,
This paper was prepared for presentation at the 2005 Asia Pacific Oil & Gas Conference and and what range of properties is physically realistic. Such work
Exhibition held in Jakarta, Indonesia, 5 – 7 April 2005.
also allows the key contributions to uncertainty to be defined
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
and targeted if overall volumetric uncertainty must be reduced.
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at Introduction
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
Petrophysical evaluations are carried out for a number of
for commercial purposes without the written consent of the Society of Petroleum Engineers is different purposes, including operational decision-making,
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous volume in place estimation and reservoir modeling. In all
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
cases, the uncertainty in the deliverables of net reservoir,
porosity, permeability, water saturation and contact locations
Abstract are critical. However, these data are usually provided without
Typical petrophysical deliverables for volumetric and quantitative determination of their uncertainties.
modeling purposes are net reservoir, porosity, permeability, This paper will highlight the ease with which uncertainties
water saturation and contact locations. These data are usually can be derived using Monte-Carlo simulation. It will also
provided without quantitative determination of their illustrate how flexible this technique is when it comes to
uncertainties. working with different interpretation models, which is not
Current computing power renders it now feasible to use commonly done. The largest source of uncertainty in
Monte-Carlo simulation to determine the uncertainty in petrophysical interpretation may be the interpretation model
petrophysical deliverables. Unfortunately, quantitative itself.
uncertainty definition is more than just using Monte-Carlo Given the large number of possible interpretation models
simulation to vary the inputs in your interpretation model. The for all the different petrophysical deliverables, this paper will
largest source of uncertainty may be the interpretation model only use the most basic petrophysical deliverable, being
itself. porosity, to illustrate the relationship between uncertainty and
This paper will use a variety of porosity interpretation the log interpretation model selected.
models to illustrate how the impact of each input on the It will also be shown that verification of log porosity using
uncertainty varies with the combination of input values used in an independent measure such as core porosity can also provide
any given model. It will show that use of the incorrect model quantitative uncertainties allowing comparison with the log
through oil and gas zones may give porosity estimates with derived uncertainties.
Monte-Carlo derived uncertainty ranges that exclude the
actual porosity. The State of Uncertainty in Petrophysics
Core data provides the best means of quantifying actual The requirement for quantification of petrophysical
uncertainty in the petrophysical deliverables. Methodologies uncertainty is not a recent development. Many papers are in
for deriving uncertainties quantitatively by comparison with the literature describing functions for uncertainty definition
core data will be presented. In the absence of core data, and how to use Monte-Carlo modeling for the same purposes.
interpretation models should have been tested against core Although work such as that of Amaefule & Keelan (1989),
data through the same or similar formations nearby. Monte- Chen & Fang (1986) and Hook (1983) provides an excellent
Carlo simulation can then be used as an effective means of foundation on which to calculate uncertainties, the
quantifying petrophysical uncertainty. Comparisons between methodologies are both time consuming to program and
the core comparison and Monte-Carlo techniques will be inflexible with regard to interpretation model.
made, showing that similar results are achieved with the With the computing power available on desktop machines
appropriate interpretation models. today, engineers no longer have to use these analytical
The methodologies described in this paper are techniques to derive uncertainty. Monte-Carlo models are
straightforward to implement and enable petrophysical straightforward to build and no longer time consuming to run.
deliverables to be treated appropriately in volumetric and The literature contains a number of examples of Monte-Carlo
modeling studies. In addition, quantification of petrophysical simulation being used to characterize petrophysical
uncertainty assists in operational decision-making by letting uncertainty, such as the work of Spalburg (2004).
2 SPE 93125

However, none of these examples highlight the uncertainty 0.24
owing to the interpretation models being used. P90
P50
0.23
P10

Monte-Carlo Modeling & Assumptions 0.22

In order to understand the problems associated with Monte-

porosity values (v/v)
0.21
Carlo simulation and how best to overcome them, a simple
explanation of the technique is warranted. 0.20

Monte Carlo methods use random numbers and probability 0.19
descriptors for the input variables to investigate problems
expressed as mathematical formulae. As an example, if a 0.18

simple density porosity calculation is considered, as shown in 0.17

the equation below:
0.16
10 100 1000 10000 100000

φd = (ρma – ρ)/( ρma – ρfl) number of scenarios

Figure 2 – The P90, P50 and P10 statistics derived from the
where ρma is the matrix density, ρ is the density log Monte-Carlo density porosity distributions vary with the
measurement and ρfl is the density of the fluid in the pore number of scenarios modeled.
space of the zone investigated by the density tool and φd is the
log-derived density porosity. Figure 2 shows how the statistics derived from the
The input values (ρma, ρ, ρfl) all have uncertainties porosity distributions do not approach the correct values until
associated with them, so the resulting output (φd) will also at least 500 scenarios have been run.
have an uncertainty. With Monte-Carlo simulation, the Note that the required number of scenarios for statistical
uncertainty in the output is determined by randomly selecting accuracy will increase with the number of input variables used
input values from their uncertainty distributions and in any particular model.
calculating the output value. The output value is stored then
the input selection and calculation processes are repeated a Porosity Uncertainty Using Monte-Carlo in
large number of times. Finally all the output values are Theoretical Cases
examined statistically to determine the uncertainty in the To illustrate the significance of the assumptions and models
output value. used for uncertainty quantification, the basic petrophysical
Monte-Carlo modeling is very flexible, allowing different deliverable of total porosity is used.
interpretation models to be built and the uncertainties tested Monte-Carlo models have been built for density porosity,
quickly. Dependencies between input variables may also be sonic porosity and density-neutron porosities. The
accounted for in the input value determination. uncertainties in these three different methods are compared
The downside to Monte-Carlo simulation is that a large through water, oil and gas bearing sand models.
number of cycles (>500) are typically required for meaningful Figure 3 compares the porosities calculated for the same
statistics to be developed. This point is illustrated below where density log measurement in known water, oil and gas systems.
the equation for density porosity above has been modeled This Figure serves to illustrate that not correcting for the
through a water-bearing sand. In Figure 1 the distribution of presence of hydrocarbons will result in significant errors in
porosities does not begin to approach a reasonable (“normal”) density porosity estimates. Indeed, failure to correct for
shape until 500 scenarios or more are run. hydrocarbons in a gas-bearing zone will result in most likely
0.25 porosity estimates that do not include the actual porosity value
50
100 in the P90 to P10 uncertainty range.
500
0.20 10000 0.5
30000 water
normalised frequency

oil
normalised frequency

0.15 0.4
gas

0.3
0.10

0.2
0.05

0.1
0.00
0.16 0.17 0.18 0.19 0.2 0.21 0.22 0.23 0.24 0.25 0.26 0
density porosity (v/v) 0.1 0.15 0.2 0.25 0.3

Figure 1 – Histogram of the density porosities derived using density porosity (v/v)

Monte-Carlo simulation through a water-bearing sand. Only
Figure 3 – Histogram of the density porosities derived using
the number of scenarios (from 50 to 30000) changes between
Monte-Carlo simulation through water, oil and gas-bearing
each curve.
sands. Note that these porosity estimates are corrected for the
presence of hydrocarbons.
SPE 93125 3

Figure 4 compares the porosities calculated from the To better illustrate that the various porosity models do
density-neutron log combination using the same scenarios as actually give different results and different uncertainty
modeled for the density porosity. In this case, the reduced distributions Figures 6, 7 and 8 compare the water, oil and
neutron response through the gas-bearing sand results in some gas-bearing sand porosities for the three porosity interpretation
correction for the hydrocarbons. However, this “correction” techniques.
would not be so apparent were the sands shaley. The porosities In the water-bearing sands (Figure 6), all three techniques
estimated through the oil sands show little correction for the give similar porosity values, but the uncertainty distributions
presence of hydrocarbons. are slightly different, as should be expected. In the oil-bearing
sands (Figure 7), the best estimate of the porosities differs
0.5
water significantly between the different interpretation techniques,
oil yet the range of the uncertainties for each technique remains
normalised frequency

0.4
gas similar to that observed for the water-bearing case. And in the
0.3 case of gas-bearing sands (Figure 8), the differences in the
best estimate of porosity increase even more between the
0.2 interpretation techniques, while the uncertainty ranges remain
virtually unchanged from the water-bearing case.
0.1 It is clear from the foregoing that even porosity estimation
from wireline log data can give different answers depending
0 on the method used i.e. on the logs used and whether or not
0.1 0.15 0.2 0.25 0.3
hydrocarbon corrections are carried out. Accordingly the
density-neutron porosity (v/v) uncertainty estimates also differ depending on the models
used.
Figure 4 – Histogram of the density-neutron porosities
derived using Monte-Carlo simulation through water, oil and 0.5
density
gas-bearing sands. Note that no assumptions have been made density-neutron
normalised frequency

about the presence of hydrocarbons. 0.4
sonic

Figure 5 compares the porosities calculated from the sonic 0.3

log using the same scenarios as modeled for the density
0.2
porosity. Again, the porosity estimated for the water-bearing
sands is a match with the density log-based estimate.
0.1
However, the porosities for the oil and gas-bearing sands are a
little too high unless corrections for the presence of 0
hydrocarbons are made. Of course, in the case of this 0.1 0.15 0.2 0.25 0.3
theoretical model, such corrections are possible using the work porosity (v/v)
of Batzle and Wang (1992), but in real cases, the porosities
through the hydrocarbon-bearing intervals must be derived Figure 6 – Histogram of the porosities derived using Monte-
from another source before quantitative corrections for Carlo simulation through water-bearing sands.
hydrocarbons can be made.
0.5
density
0.4 density-neutron
water
0.4
normalised frequency

oil sonic
normalised frequency

0.3 gas
0.3

0.2 0.2

0.1
0.1

0
0 0.1 0.15 0.2 0.25 0.3
0.1 0.15 0.2 0.25 0.3 porosity (v/v)
sonic porosity (v/v)
Figure 7 – Histogram of the porosities derived using Monte-
Figure 5 – Histogram of the sonic porosities derived using Carlo simulation through oil-bearing sands.
Monte-Carlo simulation through water, oil and gas-bearing
sands. Note that no assumptions have been made about the
presence of hydrocarbons.
4 SPE 93125

density
0.5 0.5
density density HC corr.
density-neutron density-neutron

normalised frequency
0.4 0.4
normalised frequency

sonic sonic
core
0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.1 0.15 0.2 0.25 0.3 0.1 0.15 0.2 0.25 0.3
porosity (v/v) porosity (v/v)

Figure 8 – Histogram of the porosities derived using Monte- Figure 9 - The log-derived porosity statistics for water-
Carlo simulation through gas-bearing sands. bearing sands in a medium porosity shaley sand reservoir.
0.5
Porosity Uncertainty Using Monte-Carlo in Real density
Cases density HC corr.

normalised frequency
0.4 density-neutron
To allow the conclusions drawn from the theoretical models
sonic
presented in the previous section to be verified, real data has 0.3 core
been selected from cored wells through water, oil and gas
columns. Figures 9, 10 and 11 show the data. Note that only 0.2
the “density HC corr” data has been hydrocarbon corrected
(using invaded zone resistivity logs). 0.1
In the water-bearing sands (Figure 9), the log-derived
density and sonic porosities have similar average values, close 0
0.1 0.15 0.2 0.25 0.3
to those from the core data. While the density-neutron
combination actually overestimates porosities in these sands. porosity (v/v)
The data also confirm that the uncertainty ranges are different
Figure 10 - shows the log-derived porosity statistics for oil-
for each measurement type. Note too that the log-derived
bearing sands in a similar medium porosity shaley sand
porosities have not been calibrated to the core data.
reservoir.
What is particularly interesting about Figure 9 is that the
actual average porosity value confirmed by both the core and 0.4 density
hydrocarbon corrected density porosity is less than the P90 density HC corr.
normalised frequency

estimate from the density-neutron combination. 0.3
density-neutron
sonic
In the oil-bearing sands (Figure 10), the porosity estimates
core
are much closer together, with all the uncertainty ranges (P90
0.2
to P10) including the actual best estimate porosity from the
core data. What is significant here is that the density porosities
should be corrected for the density difference due to even oil 0.1
being lighter than water i.e. the non-hydrocarbon corrected
density porosities are already 0.5 p.u. too high. The density- 0
neutron porosities are also on average 1.0 p.u. too high. 0.1 0.15 0.2 0.25 0.3
The largest differences are observed in the gas-bearing porosity (v/v)
sands (Figure 11), with the sonic and non-hydrocarbon
corrected density porosities being more than 2.0 p.u. too high. Figure 11 - The log-derived porosity statistics for gas-bearing
Indeed the P90 to P10 uncertainty ranges for these porosity sands in a similar medium porosity shaley sand reservoir.
models do not include the actual best estimate porosity from
the core data. Even the density-neutron combination A second set of examples from a different region are
overestimates porosity by 0.5 p.u. presented as Figures 12, 13 and 14.
In the water-bearing reservoir (Figure 12), the log-derived
density and sonic porosities have similar average values, close
to those from the core data. While the density-neutron
combination again overestimates porosities in these sands.
In the oil-bearing reservoir here (Figure 13), the porosity
estimates still show divergent most likely values. In fact the
density-neutron based porosity estimate again does not
encompass the actual best porosity estimate from core within
SPE 93125 5

the P90 to P10 range. In this example too, the value of Porosity Uncertainty By Core Comparison
correcting the density porosities for even oil is apparent. Since the real cases presented in the previous section also had
The largest differences are again observed in the gas- core data acquired over the same logged intervals, it is
bearing reservoir (Figure 14), with the sonic and non- possible to compare the log-derived total porosities with those
hydrocarbon corrected density porosities being more than 5.0 measured on the equivalent piece of core.
p.u. too high. Here too, the P90 to P10 uncertainty ranges for When making such a comparison, there are two factors to
these porosity models do not include the actual best estimate bear in mind. Firstly, the depth match between log and core
porosity from the core data. Even the density-neutron data must be excellent so that the same intervals are in fact
combination does a poor job, underestimating porosity by 2.3 being compared. Secondly, the porosity resolution at the log
p.u. scale is not the same as that derived from the core plug scale.
If the core porosity data is not “filtered” back to a similar
0.4
density resolution to the log-derived data, then the variability (or
density HC corr. uncertainty) implied by the comparison will be larger than it
normalised frequency

0.3 density-neutron should be.
sonic
Figures 9, 10, 11, 12, 13 and 14 all show very similar most
core
0.2 likely porosity estimates and uncertainty distributions for the
hydrocarbon-corrected density and the core porosities. This
observation implies that the best match to the core porosities is
0.1
using these hydrocarbon-corrected density porosities.
Although the other porosity estimation techniques can provide
0 reliable porosities in some circumstances, provided the
0.1 0.15 0.2 0.25 0.3
hydrocarbon influence on the log measurements being used
porosity (v/v) are taken into account.
Figure 12 - The log-derived porosity statistics for water-
Overall Porosity Uncertainty Interpretation
bearing sands in a medium porosity limestone reservoir.
From the foregoing, it is apparent that the uncertain ranges
0.4 estimated using Monte-Carlo simulation are interpretation
density
density HC corr.
model dependent. It is still possible for calculated uncertainty
normalised frequency

0.3 density-neutron ranges not to include the actual reservoir porosities, if an
sonic inappropriate porosity interpretation model is used.
core The best way to ensure that the appropriate interpretation
0.2
model is selected is by comparison with core data. If no core
data is available, then the work presented herein suggests that
0.1 hydrocarbon-corrected density porosities should be used. If it
is not possible to carry out these calculations, then whatever
0 model is selected should either include hydrocarbon correction
0.15 0.2 0.25 0.3 0.35 or model the likely range of hydrocarbon densities in the
porosity (v/v) uncertainty analyses.

Figure 13 - The log-derived porosity statistics for oil-bearing Uncertainty in Other Petrophysical Deliverables
limestone in the same unit as Figure 12. Of course the techniques discussed and conclusions drawn
from the work presented in this paper are equally valid for
0.4
density other petrophysical properties such as water saturation,
density HC corr. permeability, net reservoir and contact locations.
normalised frequency

0.3 density-neutron Although not detailed in this paper, since the impact of the
sonic
porosity uncertainties illustrated is sufficient to illustrate the
core
0.2 value of model uncertainty quantification, it is good practice to
derive uncertainties in all petrophysical deliverables so that
users are aware of any limitations in the data presented.
0.1

Conclusions
0 Monte-Carlo simulation is well suited to uncertainty
0.15 0.2 0.25 0.3 0.35 quantification in the current petrophysical environment.
porosity (v/v) However, simply calculating uncertainty is insufficient unless
it can be shown that the interpretation model applied is
Figure 14 - The log-derived porosity statistics for gas-bearing appropriate. This conclusion is true for all petrophysical
limestone in the same unit as Figures 12 and 13. deliverables, not just porosity as presented in this paper. Good
quality core data provides an excellent basis on which to
determine the appropriate interpretation model.
6 SPE 93125

With Monte-Carlo modeling, care should also be taken to
ensure that sufficient scenarios are run to determine valid
statistics on the output values. Generally a few tens of
scenarios are insufficient. Typically greater than 500 runs are
required.
Petrophysical evaluation should attempt to determine
uncertainties in at least the critical items of porosity and water
saturation. Knowledge of the possible range of values enables
Operators to make better data gathering and completion
decisions. Reservoir modeling studies are also more likely to
include scenarios approaching the real reservoir.

Acknowledgements
The author would like to acknowledge the feedback received
from many clients over the years that have seen the value of
uncertainty quantification in their petrophysical deliverables
once the data was made available to them.

References
Amaefule, J.O. & Keelan, D.K.: “Stochastic Approach to
Computation of Uncertainties in Petrophysical Parameters,”
Society of Core Analysts, Paper No. SCA-8907, 1989.

Batzle, M. & Wang, Z.: “Seismic Properties of Pore Fluids,”
Geophysics, Vol. 57, No. 11 (November 1992); P. 1396-1408.

Chen, H.C. & Fang, J.H.: “Sensitivity Analysis of the
Parameters in Archie’s Water Saturation Equation,” The Log
Analyst, pp39-44, September-October 1986.

Hook, J.R.: “The Precision of Core Analysis Data and Some
Implications for Reservoir Evaluation,” SPWLA 28th Annual
Logging Symposium, June 27-30, 1983.

Liu, N. & Oliver, D.S.: “Evaluation of Monte Carlo Methods
for Assessing Uncertainty," SPE Journal (2003) 8.

Spalburg, M.R.: “Bayesian Uncertainty Reduction for Log
Evaluation,” SPE 88685, 11th Abu Dhabi International
Petroleum Exhibition and Conference held in Abu Dhabi,
U.A.E., 10-13 October 2004.