You are on page 1of 30

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

net/publication/240743984

Toward consistency in petroleum exploration: A systematic way of


constraining uncertainty in prospect volumetrics

Article in AAPG Bulletin · October 2008


DOI: 10.1306/06040808064

CITATIONS READS

16 3,888

2 authors, including:

David Gordon Quirk


Technical University of Denmark
77 PUBLICATIONS 857 CITATIONS

SEE PROFILE

All content following this page was uploaded by David Gordon Quirk on 21 April 2021.

The user has requested enhancement of the downloaded file.


Toward consistency in AUTHORS
David G. Quirk  Maersk Oil, 50 Esplanaden,
petroleum exploration: DK-1263 Copenhagen K, Denmark;
dgq@maerskoil.com
A systematic way of Dave Quirk is an exploration geoscientist at
Maersk Oil in Denmark and has previously worked
constraining uncertainty for Shell, Oxford Brookes University, Burlington
Resources, and Hess Corporation. He received
in prospect volumetrics his B.Sc. degree from Liverpool University, U.K.,
in 1984 and Ph.D. from Leicester University,
U.K., in 1988. His current interests include salt
David G. Quirk and Richard G. Ruthrauff tectonics, the geology of the Isle of Man, and
risk and uncertainty.
Richard G. Ruthrauff  Hess Corporation,
ABSTRACT 500 Dallas Street, Houston, Texas 77002;
rruthrauff@hess.com
The results of exploration in individual plays show that, al-
though discovery sizes are highly variable, their distributions Rick Ruthrauff is a petroleum engineer working
on decision analysis at Hess. He has a diverse
have certain properties useful in assessing undrilled prospects,
background in exploration, exploitation, produc-
an example being the upper limit of P99. However, such tion, reserves, and operations at Hess and
information rarely forms an integral part of the evaluation of an Texaco, and one year of LNG projects at El Paso
undrilled prospect. Where they form part of the workflow, Corporation as a contractor. He received his
though, historical data can help define the natural distributions B.S. degree in chemical engineering from the
associated with reservoir properties and help constrain geo- University of Pittsburgh in 1978.
logical and commercial risk.
In evaluating a new prospect, deterministic volumetrics ACKNOWLEDGEMENTS
can be calculated for a high case, but the probability of encoun-
The authors thank the many colleagues at Maersk
tering this is, by itself, mostly guesswork; low volumetrics cases
Oil, Hess, and other organizations who have
are even more difficult to constrain. Consequently, probabi- contributed to the ideas developed in this ar-
listic methods have become the standard way of dealing with ticle, in particular Matt Docherty, Kenneth Nielsen,
uncertainty in exploration. However, there is a problem in Lene Madsen, Danny Schwarzer, Tim Tranter,
defining the trap size or volume of hydrocarbon-bearing gross Ross Ensley, Mark Parsley, Tim Cordingley, Steve
reservoir (hcbGRV), specifically the distribution defined by Massie, Dave Burnett, Phil Cox, John O’Connor,
Christopher Tiratsoo, John Boucher, David
the combined cumulative probabilities of the position of seals,
Boote, Dan Tearpock, Pete Rose, Gary Citron,
reservoirs, and fluid contacts because, prior to drilling, no direct and Jim Murtha. We acknowledge Craig Smalley’s
samples of these surfaces are given. Unfortunately, this prob- many suggestions and help in improving the
lem does not have a solution, so we have developed a quality article. We are grateful to Maersk Oil and Hess
assurance tool that uses deterministic inputs to check the re- for support in publishing the work although
ality of probabilistic outputs. The tool is called ‘‘real point re- the opinions expressed here are solely those of
the authors.
source iteration’’ (RPRI) and is primarily aimed at improving
consistency in volumetrics prediction.
RPRI uses objective criteria to calculate two deterministic
cases from which a full-discovery-size distribution is created.

Copyright #2008. The American Association of Petroleum Geologists. All rights reserved.
Manuscript reviewed by special issue editor; AAPG received manuscript June 4, 2008.
DOI:10.1306/06040808064

AAPG Bulletin, v. 92, no. 10 (October 2008), pp. 1263 –1291 1263
The results are then iterated with simple statistics
and information from historical data. One of the
key elements of the technique involves defining
standard hcbGRVs based on the depth of the last
closing contour relative to the culmination. The
method is quick, transparent, and repeatable and
helps avoid overoptimism using empirical obser-
vations to predict realistic low case volumes. The
outputs are analogous to those obtained from prob-
abilistic techniques, meaning they can easily be
compared and adjusted. RPRI can also be used to
produce maps and reservoir parameters for specif-
ic probabilistic outcomes, providing real cases for
input to economics calculations and planning.

INTRODUCTION

Finding significant quantities of oil and gas becomes


ever more challenging. Nevertheless, there seems
to be no shortage of outlines on maps showing pros-
pects and leads in acreage licensed for exploration.
Which ones are worth drilling, however, may not
be immediately obvious and is the subject of ex-
tensive evaluation in exploration companies. One
of the key factors in the drilling decision is the po-
tential size of the prize, which, of course, is uncer-
tain at the predrill stage.
This article presents a process aimed at helping
improve consistency and predictability in volumet-
ric assessments. It is divided into two parts, address-
Figure 1. A normal distribution (a) can be displayed as a straight
ing the following questions:
line (b) by transformation to the cumulative standard normal
domain using the formula v = (s) NORMSINV(1 Pn) + m.
1. How can historical data on discoveries be used
to improve volume prediction in exploration?
2. Is there a simple way of checking the validity of Natural Distributions
probabilistic volumetrics in undrilled prospects?
The central limit theorem (Trotter, 1959) shows
that random samples of average values from a nat-
LESSONS FROM HISTORICAL ural population, such as porosity in a reservoir, will
EXPLORATION DATA tend toward a normal distribution around a central
mean, equal to the median and mode (Figure 1a).
Before discussing the use of historical data, we sum- Theoretically and empirically, a normal distribution
marize briefly the statistical theory that will be used best represents the uncertainty associated with any
in the analyses; more detail can be found in Quirk new sample of a geological parameter, for example,
and Ruthrauff (2006). in a proposed exploration well (Murtha, 2001). A

1264 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 2. (a) Porosity data from a Tertiary
reservoir in an offshore play from Brazil dis-
played as ranked fractile probabilities in the
cumulative standard normal domain. The data lie
approximately along a straight line indicating
that the distribution is normal. (b) Reservoir net/
gross data for the same play shown in Figure 2a
displayed in the cumulative standard normal
domain. Again, this is a normal distribution
as the data (as fractile percentages) plot on a
straight line.

normal distribution can be plotted as a straight line there is an n% chance that any new sample is at
by transformation to a domain known as the cu- least as large as v). The symbols s and m are constants
mulative standard normal distribution (Figure 1b) representing the slope of the straight line and the
using the formula intercept of the y axis or median value, respectively.
With a statistically valid number of samples from
v ¼ ðsÞ NORMSINV ð1  ½Pn=100Þ þ m; a normal distribution (typically 20 or more random
data points), a natural geological population can be
where v = value belonging to the normal distribution, plotted in the cumulative standard normal domain
s = standard deviation of the distribution, m = mean (Figure 2) by distributing the ranked data in equal
of the distribution, and NORMSINV(1 – [Pn/100]) = divisions between 0 and 100% (fractile percentages).
percentage reverse cumulative probability of the In contrast to normal distributions, values used
value converted into a standard statistical function to represent the size of a prospect, such as recov-
known as the inverse of the standard normal distri- erable resources, gross rock volumes, areas, barrels
bution (reverse cumulative probability, Pn, means per acre foot, and hydrocarbon column heights (for

Quirk and Ruthrauff 1265


Figure 3. A lognormal distribution (a) can be
displayed as a straight line (b) by transforming
it to the cumulative standard normal domain
using the formula lnx = (s) NORMSINV(1 Pn)
+ m. Only two points are needed to define the
straight line for the entire distribution.

example, Niemann, 1998) are products of natural and downside of uncertainty and can therefore can
populations whose multiplicative distributions are be rewritten as f(logeP10/logeP90). Likewise, m rep-
skewed so that the mean is larger than the median resents the median value or P50, which is equivalent
p
and the median is larger than the mode (Figure 3a). to (P10/P90). In other words, with values for P10
However, natural logarithms of the values form and P90 (or, for that matter, any other two cumu-
normal distributions and these are consequently lative probabilities), this formula can easily be solved
known as lognormal (Aitchison and Brown, 1957). (Quirk and Ruthrauff, 2006; Figure 3b).
Like normal distributions, any lognormal distribu-
tion can be plotted as a straight line in the cumu-
lative standard normal domain (Figures 3b, 4) Discovery Size Distributions
using the formula
The sizes of in-place or recoverable resources dis-
loge v ¼ ðsÞ NORMSINV ð1  ½Pn=100Þ þ m: covered in mature plays are lognormally distributed
(for example, Figure 5) although some discrimina-
Because s represents the slope of the straight line, tion in the sampling of very small traps is present.
it is a function of the difference between the upside Unfortunately, it is easy to produce invalid statistics

1266 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 4. Average millions of
barrels of oil per square kilo-
meter for discoveries in an off-
shore play from north Gabon
displayed on a logarithmic
scale (loge[mmbbl/km2]) in the
cumulative standard normal
domain. The data lie along a
straight line defined by m and
s indicating that the distribu-
tion is lognormal. The parame-
ter mmbbl/km2 can be used
for quality checking exploration
prospects within specific mature
plays (for example, Table 4c).

by wrongly sampling the distribution (Quirk and although all data are needed to determine fractile
Ruthrauff, 2006). To avoid this, it is important to percentage and ranking (Figure 6).
(1) include data from only one play; (2) use all dis- We have analyzed more than 100 different plays
coveries in the play not just commercial fields; and across the world in this way, and all those with more
(3) define the lognormal trend only with reliable than 20 discoveries show a good lognormal trend
data, namely, those discoveries constrained by ap- between P99 and P1. To prevent the statistics being
praisal, development, or production information, biased by extreme upsides, as well as to ensure that

Figure 5. Lognormal discovery


size distribution for Miocene-
Pliocene gas play in Nile Delta.
These data are plotted as the
natural logarithm of recover-
able resource in the cumulative
standard normal domain. PoSc =
chance of commercial success;
PoSg = chance of geological
success.

Quirk and Ruthrauff 1267


Figure 6. Discovery size dis-
tribution for the Permian gas
play in Eromanga Basin, Austra-
lia, showing the unreliability of
data on noncommercial finds.
The best fit line defining the log-
normal distribution and related
statistics was made using only
commercial (reliable) reserves
data. Note that the recoverable
resources quoted for noncom-
mercial discoveries are usually
underestimates, falling below
the discovery trend line. PoSc =
chance of commercial success;
PoSg = chance of geological
success.

the mean of the distribution is similar to the aver- discovered divided by number of discoveries) and,
age discovery size, it is recommended that the log- unlike the mean for the untruncated distribution,
normal trend is truncated at P99 and P1. Empir- typically has a narrow range of cumulative prob-
ically, the mean of the distribution between P99 ability, around P25–P20 (Table 1), equating with
and P1 (meanP99 – P1) is generally within 20% of the a 1-in-4 to 1-in-5 chance that a discovery is larger
average discovery size (total recoverable resources than meanP99 – P1, a useful fact for quality-checking

Table 1. Summary of Typical Limits, Ranges, and Averages in Risk and Uncertainty for Oil and Gas Plays Analyzed by Quirk and
Ruthrauff (2006)*

1268 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


purposes. The untruncated mean is larger than fied, the best of both worlds can ensue (for example,
meanP99 – P1, sometimes significantly larger. Doré and Robbins, 2005).
Quirk and Ruthrauff (2006) have shown that Although the size of discoveries decreases over
meanP99 – P1, the P10 to P90 ratio, the total number time, the discovery rate in many plays does not
of discoveries, and the chance of making a discov- change significantly (see Figure 9), which may in
ery are unique to each play. Nevertheless, most part explain why the industry seems to be relatively
plays show certain common limits to P99, P50, good at estimating the chance of geological success
risked meanP99 – P1, chance that a discovery is com- in proven fairways (for example, Otis and Schnei-
mercial, and total amount of recoverable resource dermann, 1997).
between P99 and P1 (Table 1), some of which can All in all, it appears that it is important for ex-
be used as reality checks for estimations of recov- ploration companies to invest at least as much tech-
erable resource in new prospects. Further analysis nical effort in volumetrics evaluation as they do in
shows that exceptions in a positive sense may oc- risk assessment (cf., Hesthammer and Boulaenko,
cur in new plays (less than 20 discoveries) and plays 2005).
with direct hydrocarbon indicators (DHIs), where
small traps have effectively been screened out. Ex-
ceptions at the other end of the spectrum occur METHODS OF CALCULATING
in certain onshore plays with good infrastructure, PROSPECT RESOURCES
where traps significantly smaller than usual have
been targeted (Quirk and Ruthrauff, 2006). Background
Differences between plays in many of the key
metrics such as P99, meanP99 – P1, and the P10 to Exploration programs usually underperform because
P90 ratio are related primarily to the range of size of overoptimistic assessments of the size of pros-
in individual traps, true also at different stages in pects (Rose, 1987; Capen, 1992; Campbell, 2001;
the maturity of a single play as the bigger prospects Rose et al., 2003). Uncertainty is an inescapable
are commonly drilled first. part of exploration, but with realistic predrill as-
sessments, there would be an equal chance that a
Exploration Risk prospect improves or deteriorates with additional
data and further interpretation. Instead, volumes
One of the more sobering statistics to come out of estimated for an exploration prospect usually de-
the analysis of historical data is that, on the aver- crease over time as technical work continues, rarely
age, only a little more than half of all discoveries are increasing. Does exploration fail to deliver on pre-
commercial in a typical exploration play (Figure 7; dictions because uncertainty in the subsurface is so
Table 1). In other words, although small subeconom- great or because evaluation processes tend to have
ic discoveries are never sought, they are an inescap- a bias toward high-side models? In either case, this
able reality of exploration and therefore need to be section of the article aims to show that, although
acknowledged in prospect evaluations (Capen, 1992). the potential ranges of geological properties in un-
Quirk and Ruthrauff (2006) have shown that drilled targets may at first seem dauntingly large,
very small discoveries are avoided during the early realistic assessments of prospect size are achievable
stages of exploration in some successful frontier using a systematic quality control process.
plays because, initially, only the biggest features Our experience is that volumetric assessments
are selected, i.e., P99 for certain immature plays of the same predrill prospect made by different inter-
can be significantly larger than usual, also reflected preters and by different companies commonly vary
in a smaller P10 to P90 ratio (Figure 8; Table 1). by more than a factor of two, despite internal quality
The flip side of exploring new plays is that the geo- assurance processes. The problem seems to revolve
logical risk is generally much higher, although some- around a few human characteristics: (1) a subjec-
times when a new play in a proven basin is identi- tive approach to probabilities, (2) an inclination

Quirk and Ruthrauff 1269


Figure 7. Historical
chance of making a geo-
logical discovery (POSg)
versus chance of making a
commercial discovery
(POSc) for a number of
(a) oil plays and (b) gas
plays. Note that almost
half of all discoveries are
too small to be devel-
oped, meaning that P50
in many discovery size
distributions lies close to
the minimum economic
field size. SNS = southern
North Sea.

toward the optimistic, (3) a preference for narrow of tests), are valued correctly, and have been con-
rather than broad ranges, and (4) a tendency to ducted efficiently and in a timely manner. Assum-
stick with and embellish single scenarios. The latter ing this is true, then the straightforward question
two points are in part compounded by ever im- ‘‘how big is the prospect?’’ is valid, but as will be
proving technology and increasing amounts of data discussed now, the preferred answer is ‘‘there is a
that may lead to a certain degree of overconfidence probability of X% that a discovery will contain at
in one particular model. Despite advances in seis- least Y million barrels of oil (or gas equivalent).’’
mic imaging and sophisticated stochastic analyses,
the degree of uncertainty involved in subsurface Probabilistic versus Deterministic
prediction is such that relatively simple concepts Volumetric Estimations
and wide distributions generally describe prospects
best because they contain fewer biases and encom- Exploration volumetrics can be estimated either de-
pass the greatest range of possibilities. terministically or probabilistically (Kaufman, 1963;
Reliable exploration decisions depend on eval- Smith, 1970; Davidson and Cooper, 1976; Megill,
uations that are internally consistent and objective, 1984; Capen, 1992; Murtha, 2001). Advantages and
deliver what they predict (based on a valid number disadvantages of each method are summarized in

1270 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 8. Discovery size
distribution for an im-
mature play, the subsalt
play in deep-water Gulf
of Mexico. The unusually
high value of P99 occurs
because only the largest
structures have been tar-
geted at this early stage
of exploration.

Table 2, the main differences being that probabilis- the apparent drawback that the outputs are some-
tic techniques attempt to incorporate uncertainty what removed from real cases, whereas determin-
into the analysis using distributions as inputs, with istic techniques use actual geological-engineering

Figure 9. Historical chance of exploration success in a selection of plays over four intervals of time, from onset of exploration to
present day. There are significant differences between basins, but interestingly, the average discovery rate does not vary significantly
over time even though average discovery size decreases. The average for all exploration wells is 62.7% in the deep-water lower
Congo, 39.2% in the southern North Sea (SNS) Permian play, 35.0% in the Bengal Basin, 34.3% in north Gabon, and 19.3% in Moray
Firth. Note that these all represent single plays except for the Moray Firth data that covers multiple plays.

Quirk and Ruthrauff 1271


scenarios but without a simple means of calculat- Table 2. The Advantages and Disadvantages of Using
Deterministic and Probabilistic Methods for Calculating the
ing the chance of finding a certain size of resource.
Size of Exploration Prospects
Despite the ascendance of probabilistic meth-
ods in resource estimations, it commonly appears Deterministic Volumetrics
that the estimated range in prospect size is biased
Disadvantages Advantages
to the upside (Capen, 1992; Rose, 2001), with the
No cumulative probabilities Simple and understandable
downside effectively truncated (Figure 10), such Low reserves cases are usually Reserves tied to maps
that exploration programs frequently underperform overoptimistic
(Rose, 1987; Campbell, 2001). Indications that the Tendency to focus on single Inputs easy to review
probabilistic evaluation of a new prospect may be model
problematic include the following:
Probabilistic Volumetrics
1. The distribution of potential recoverable resource
is symmetric (meanP99 – P1 P50) instead of log- Disadvantages Advantages
normal (skewed) as should result from the multi- Complex (black box approach) Uncertainty acknowledged
plication of geological distributions. Biases may be hidden Probabilities calculated
2. Mean resource is greater than the historic aver- Tendency to focus on Outputs easy to quality
age for the play, assuming that the largest traps extremes check
were targeted first (Root and Drew, 1979). Hard to tie commercial cases Commonest approach
3. P99 is larger than historical data supports.
4. Most of the distribution is economic, assuming
that similar commercial conditions prevail over ken issue in exploration is that there are no direct
time. samples of geological populations defining trap
size, specifically the volume of gross hydrocarbon-
In fact, the common observation that smaller bearing reservoir. The probabilities and distribu-
than expected volumes are encountered in an ex- tions of surfaces defining the top and base of the
ploration well can be caused by a multiplicity of reservoir interval and the position of hydrocarbon
reasons (Figure 11). Generally, only in cases where contacts are unknown, a problem compounded by
a prospect has a large unequivocal DHI and a high uncertainties in the accuracy and precision of seis-
degree of seismic resolution are very small discov- mic picks, seal position relative to reservoir top,
eries avoidable. depth conversion, and oil or gas versus mixed charge
Disappointing drilling outcomes can lead to a lack (Figures 12, 13). In other words, hydrocarbon-bearing
of confidence in probabilistic volumetrics among de- gross rock volume (hcbGRV) is not only highly
cision makers whereas deterministic techniques can uncertain but also unconstrained in probabilistic
more easily be related to a geological model and a terms, leading to significant differences in the mode,
map. Hence, it is useful if probabilistic and determin- median, and mean values associated with trap size.
istic techniques are not treated as mutually exclusive. This issue is serious because variance in this pa-
rameter usually has a far larger effect on prospect
The Fundamental Difficulty in Assessing Prospect Size size than any specific reservoir property, as will be
shown below.
Although probabilistic methods have been adopted
by most oil companies, it is still possible to intro- Example of Sensitivity of Prospect Evaluation to
duce overoptimism by biasing realistic input ranges Trap Size
with unrealistic distribution models. For reservoir
parameters, statistical information from existing To illustrate the trap size problem, a series of probabi-
well data can serve as a useful guide (for example, listic volumetric calculations were run for the same
Figure 2). However, the main and mostly unspo- non-DHI oil prospect (V) illustrated in Figure 14.

1272 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 10. Typical but unrealistic
probability density curves for pros-
pects where the downside has been
truncated (solid line). (a) Symmetric
distribution centered around the inter-
preter’s most likely resource value.
(b) Pseudoprobabilistic distribution
where an upside tail has been incor-
porated, but the downside is still cut
out. In both cases, a density curve re-
flecting the actual uncertainty in the
prospect is shown as a stippled line.

The top and base of the reservoir interval are clearly based on the average of a poll of 20 interpreters (the
imaged on high-quality 3-D seismic data, and 21 range was actually P50 to P0.5). A P90 hcbGRV
wells have already tested the play, meaning that was defined by a contour one third of the depth
reservoir parameters are relatively well constrained between culmination and the LCC, again based on
(for example, Figure 14c), and the presence of a gas the average choice of 20 interpreters. Normal dis-
cap is considered unlikely. tributions were used for all reservoir parameters
A base case Monte Carlo simulation was run with P90 and P10 values based on statistical in-
on prospect V assuming a lognormal distribu- formation from existing wells in the play (for ex-
tion for hcbGRV. On the most likely depth map ample, Figure 14c).
(Figure 14a), the contour representing maximum Note that depth contours are used here as a
closure, here called the last closing contour or LCC, convenient way of referring to different potential
was taken to correspond to P5 hcbGRV (Figure 14b) sizes of trap fill so that changes shown in, say, the

Quirk and Ruthrauff 1273


Figure 11. Reasons
why discoveries can be
much smaller than ex-
pected. Often more than
one of these conspires
to produce a low case that
is worse than predicted.
GRV = hydrocarbon-
bearing gross rock vol-
ume; KV = vertical perme-
ability; KH = horizontal
permeability; RF = recov-
ery factor; FVF = forma-
tion volume factor; SH =
hydrocarbon saturation;
hcs = hydrocarbons; HP =
high pressure.

cumulative probability of LCC represent variations made to the input distributions or cumulative prob-
in the distribution of gross rock volume lying above abilities of either hcbGRV or a single reservoir pa-
the hydrocarbon water contact, be these because of rameter to show its relative importance in quanti-
differences in the interpretation of seismic reflec- fying the size of the prospect (see Table 3).
tions, time to depth conversions, amount of charge, The biggest change in prospect size from the
thickness or lateral extent of the gross interval, po- base case occurs with a switch in the model of trap
sition of the top seal relative to the top of the res- size or hcbGRV from lognormal to another type
ervoir, capacity of the seals, or likelihood of strat- of distribution (Table 3). Thus, using the same con-
igraphic trapping (Figures 11, 12, 14b). tours to define P5 hcbGRV and P90 hcbGRV, a
The meanP99 – P1 probabilistic outcome for the normal distribution produces a meanP99 – P1 recov-
base case in prospect V was a recoverable resource erable resource almost twice that of the base case
of 46.4 million bbl. Individual changes were then (84.8 million bbl compared to 46.4 million bbl)

Figure 12. Schematic illustration of a tilted fault block prospect showing how parameters used in assessing the volume of hydrocarbon-
filled trap cause inconsistency in probabilistic estimates of prospect size. There is no correct way of assigning probabilities and a
distribution to the surfaces defining the volume of gross reservoir that might contain hydrocarbons within a trap. The uncertainties
associated with these surfaces (base of seal, position of lateral and base seals, top and sides of reservoir, base of reservoir, and position
of hydrocarbon contacts) are indicated by the arrows and reflect issues such as accuracies and precision of picks, depth conversion,
amount of charge, growth of structure since fill, presence of a gas cap (gray shading) or oil leg, strength of seals, facies changes, and
presence and position of thief zones.

1274 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


largest P90 value is five times, whereas the range
in P10 and mean values is only three times, the ex-
tremes once more associated with differences in
trap size (or hcbGRV). The main effect this has is
on the chance of discovering an economic field size,
apparently much lower in a probabilistic evaluation
using a lognormal distribution of hcbGRV than
other distributions documented in Table 3.
Although the depth of the contour correspond-
ing to P50 hcbGRV varies relative to culmination
by no more than ±23%, the recoverable resource
differs by a factor of 0.6 smaller to 1.9 larger than
the base case. This is because a small change in depth
has a much larger effect on hcbGRV (Figure 14b),
a fact that cannot be avoided, but one that may be
overlooked, usually out of necessity, there being no
statistics to constrain the probabilistic values as-
sociated with trap size.
These examples underline the dominant effect
that hcbGRV has relative to other parameters in
prospect volumetrics and the difficulties one has in
Figure 13. Illustration of how different assumptions in only the
distribution and cumulative probabilities of potential hydrocarbon assigning or quality-controlling distributions and
water contacts will compromise consistency in trap size estima- cumulative probabilities of hcbGRV. In compari-
tion and severely affect probabilistic resource estimations. The son, differences caused by distributions and prob-
culmination and last closing contour correspond to the example abilities of reservoir parameters are relatively minor
shown in Figure 12. Although the mode or median or mean value and are easily checked against statistics from geo-
of hydrocarbon-bearing gross rock volume can be similar for
logical data for the play or from analogs. In other
different types of distributions, all three cannot be the same and
consequently there are significant variations in the recoverable words, trap size models are commonly the key rea-
resource calculated using different distributions. son for discrepancies in the size of individual pros-
pects between different interpreters and different
companies, a situation commonly evident in farm-in
and a smaller P10 to P90 ratio (23 versus 43); a situations. Certain distributions of hcbGRV pro-
triangular distribution shows an even bigger dif- duce more positive results than others (Table 3),
ference. Furthermore, if the probability associat- but unfortunately, it cannot be verified which as-
ed with a depth contour is changed, even if only sumptions are best nor is it possible to determine
slightly, a significant change in the resultant re- directly whether probabilities assigned to depth
coverable resource occurs. For example, the differ- contours are reasonable. With this in mind, individ-
ence in the mean between assuming that LCC ual companies do need to standardize their approach
represents P1 and assuming that LCC represents in trap size definition to ensure that evaluations
P10 is 38.0 million bbl (from 29.2 million bbl to of recoverable resources are relatively consistent
67.2 million bbl). In contrast, changes in distribu- across a portfolio of prospects.
tions of reservoir parameters make no more than
15% difference (Table 3). Quality Control Using a Different
Considering the entire distributions repre- Volumetrics Approach
sented in Table 3, the biggest relative difference in
probabilistic estimations occurs in low-side cases, Noncommercial discoveries commonly occur when
for example, the variation between the smallest and the exploration model fails because of one or

Quirk and Ruthrauff 1275


Figure 14. Data illustrating
prospect V, a real case but
slightly disguised. (a) Depth map
at top reservoir. Pink hatched
area indicates location of salt
wall. (b) Distribution of reservoir
gross rock volume for prospect V
(solid black line) with closures
used for base case calculations
in Table 3 indicated. Some po-
tential causes of uncertainty
are also shown, crosses indicat-
ing a more optimistic depth con-
version (or more optimistic top
reservoir pick), stippled line for
more pessimistic gross reservoir
interval thickness, black circles
for 25-m-thick waste zone. LCC
= last closing contour. (c) Net/
gross distribution for wells in
the same play as prospect V.

more negative outcomes (Figure 11), the combined ity of low case volumes directly. Instead, probabi-
chances of which are difficult to predict. It is there- listic techniques attempt to deal with the problem
fore virtually impossible to calculate the probabil- by randomly sampling a range of values, including

1276 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 14. Continued.

downsides. However, as shown above, consisten- ogy in the form of a quality control tool, overcoming
cy is difficult to achieve, and the outputs are more some of the disadvantages of either technique when
difficult to visualize without referring to specific used alone (Table 2).
deterministic cases. Therefore, in an effort to con-
strain the standard probabilistic estimates, a qual-
ity control tool has been developed that combines REAL POINT RESOURCE ITERATION
both volumetrics techniques, creating a full proba-
bilistic distribution by extrapolation from specific As has been shown previously, the sizes of discov-
deterministic high-side cases. The constructed dis- eries in individual plays tend to form lognormal dis-
tribution is then checked and iterated using histor- tributions and so too should the probabilistic range
ical information, particularly to improve the lower in potential size of recoverable resource in an un-
half of the distribution. drilled prospect (Capen, 1992; Murtha, 2001; Quirk
An advantage of the technique is that it par- and Ruthrauff, 2006). Furthermore, lognormal dis-
tially sidesteps the issue of assigning a distribu- tributions can be defined with only two values of
tion to hcbGRV, providing an independent check known probability. Therefore, if two sizes of re-
on probabilistic methods where an assumption on coverable resource can be determined, each with
the distribution has to be made. The method orig- a specific cumulative probability, then the entire
inally arose from the realization that, while a reli- distribution can be reconstructed (for example,
able way of quantifying uncertainty is needed, most Figure 15). This idea has been developed into a
people actually have a deterministic mindset, often volumetric estimation technique called real point
thinking in specific scenarios instead of broad ranges. resource iteration (RPRI) (Table 4).
Hence, it was found useful to integrate determin- Two requirements have to be met before the
istic calculations with the probabilistic methodol- idea can be successfully applied. The first is ensuring

Quirk and Ruthrauff 1277


1278
A Systematic Way of Constraining Uncertainty in Prospect Volumetrics

Table 3. Probabilistic Inputs and Consequent Results of Recoverable Resource Estimations for an Oil Prospect Illustrated in Figure 14*
Cumulative Probabilities of Size of
Hydrocarbon-Bearing Trap as Assigned Change in Change in
to Specific Depth Contours Change to Reservoir Distribution from that Assumed in Base Case Recoverable Resource (million bbl) Mean from P10/P90
Assumed Trap Base (%) from (%)
Size Distribution Shallow P50 (m) Deep Comment Parameter Distribution P90 P50 P10 Comment** Mean P99 P90 P50 P10 P1 Case Base

Lognormal hcbGRV P90 – 4375 m 4429 P5 – 4575 m base hcbGRV none (no change to base case; normal distributions from well info) 46.4 0.3 2.6 19.0 110.7 426.6 no change
(base case)
Lognormal hcbGRV base case 4416 P1 – 4575 m upside reduced none unchanged unchanged unchanged unchanged base case 29.2 0.3 2.4 14.7 68.6 223.9  37  33
res parms
Lognormal hcbGRV base case 4438 P10 – 4575 m upside expanded none unchanged unchanged unchanged unchanged base case 67.2 0.4 2.7 23.0 158.5 698.2 45 38
res parms
Normal hcbGRV base case 4496 base case symmetric distrib. none unchanged unchanged unchanged unchanged base case 84.8 0.7 8.5 55.3 193.9 433.2 83  46
res parms
Stretched b hcbGRV base case 4489 base case mode at 4475 m none unchanged unchanged unchanged unchanged base case 80.2 0.6 7.9 51.2 185.1 421.2 73  45
res parms
Uniform hcbGRV P100 – 4275 m 4502 base case even distribution none unchanged unchanged unchanged unchanged base case 87.3 0.5 6.8 55.0 207.6 448.6 88  28
res parms
Triangular hcbGRV P100 – 4275 m 4498 base case mode at 4475 m none unchanged unchanged unchanged unchanged base case 88.5 1.2 11.3 59.9 198.2 434.5 91  59
res parms
Lognormal hcbGRV base case 4429 base case base hcbGRV net/gross unchanged unchanged 0.45 0.70 P10 down 42.1 0.3 2.5 17.4 99.8 383.5 9 6
from 0.80
Lognormal hcbGRV base case 4429 base case base hcbGRV net/gross unchanged unchanged 0.54 0.90 P10 up 49.6 0.3 2.8 20.2 118.6 455.6 7 1
from 0.80
Lognormal hcbGRV base case 4429 base case base hcbGRV net/gross lognormal unchanged 0.39 unchanged P50 down 39.6 0.4 2.4 16.3 93.6 358.0  15 8
from 0.50
Lognormal hcbGRV base case 4429 base case base hcbGRV net/gross stretched b unchanged 0.44 unchanged mode of 0.35 43.5 0.4 2.6 17.9 103.7 394.5 6 6

Lognormal hcbGRV base case 4429 base case base hcbGRV porosity lognormal unchanged 0.14 unchanged P50 down 44.1 0.4 2.6 17.9 105.3 399.4 5 5
from 0.17
Lognormal hcbGRV base case 4429 base case base hcbGRV recovery base case 0.10 0.30 0.50 P90 and P10 40.9 0.2 2.1 16.2 98.4 385.7  12 10
down 0.05
Lognormal hcbGRV base case 4429 base case base hcbGRV recovery base case 0.20 0.40 0.60 P90 and P10 51.9 0.5 3.1 21.7 122.8 459.4 12 7
up 0.05
Lognormal hcbGRV base case 4429 base case base hcbGRV recovery lognormal base case 0.28 base case P50 down 40.4 0.4 2.5 16.7 96.5 155.9  13 9
from 0.35
*The first evaluation is the base case, the input parameters for which are shown here. Only one parameter is changed at a time from this base case in the other evaluations. hcbGRV = hydrocarbon-bearing gross rock volume; FVF
that different interpreters define input values in a
consistent way, particularly concerning volume of
potential hydrocarbon-bearing rock. The second is
assigning and standardizing the cumulative prob-
ability of specific deterministic outputs, a process
24.5
1277.8

1.2
P10

80

90

55
that involves extensive testing of different prob-
abilistic and deterministic volumetrics in a variety
of exploration prospects.

RPRI Applied to Structural Prospects

Interpreters can usually come close to agreeing on


two depths in a structural trap (1) the culmination,
crest, or highest point of the trap and (2) the last
closing contour (LCC, maximum closure, or appar-
ent spill point). Even with different interpretations
and depth conversions, the area enclosed by LCC
16.5

1.5
303.9
P50

and the difference in height between this and the


50

70

35

culmination (i.e., the closure height) rarely vary


more than ±25% between interpreters. Therefore,
one way of consistently defining the size of the trap
is to use the LCC on the most reasonable depth
map. In the case of a prospect where the top seal is
a distance above the top reservoir, the chance that
a discovery is so small that only the waste zone
contains hydrocarbons can be accounted for in the
prospect risk; also the depth of LCC should be
defined by the spillpoint at the base of the top seal
(Figure 12). In prospects where the height of the
hydrocarbon column is limited by seal effective-
ness instead of structural closure, the maximum
reasonable column height defines the LCC, and in
prospects with unequivocal DHIs, the maximum
reasonable outline of the DHI should be used.
1.8
8.5
72.3
P90

20

50

15

With culmination and LCC identified, there


are various ways of standardizing contours to de-
fine hcbGRV, which can then be used as inputs to
deterministic base cases. We have tested many dif-
ferent trap size scenarios in deterministic and prob-
abilistic models and have found that simple assump-
**res parms = reservoir parameters.

tions tend to give fairly consistent results. One that


= formation volume factor.

works well for a structural prospect is a contour


lying at five ninths (56%) the vertical distance from
Recovery factor (%)
Oil saturation (%)

culmination to LCC to define an hcbGRV called


Base-Case Inputs

Net to gross (%)


hcbGRV (106m3)

Shrinkage factor

P50 hcbGRV (Figure 16). This appears to imply


Porosity (%)

that a uniform probability of encountering a hydro-


carbon water contact between culmination and a

Quirk and Ruthrauff 1279


Figure 15. Construction
of a full discovery size
distribution for an explo-
ration prospect in the
cumulative standard nor-
mal domain using two
deterministic cases (in
this case, P50 and P5) to
solve the equation log ev =
(s) NORMSINV(1 (Pn/
100)) + m. The distribu-
tion can be cross-checked
against historical data
and other deterministic
numbers shown here as
discrete points.

contour 10% deeper than LCC is present. How- 2002). In this case, P50 hcbGRV is used for trap
ever, it is actually just a convenient way of stan- size, and P50 reservoir properties can easily be es-
dardizing how uncertainty is dealt with in the size timated from statistical analysis of wells in the fair-
of a hydrocarbon-bearing trap caused by variations way (for example, Figure 14c) or, in an immature
in the amount of fill, top, or base of reservoir, etc., play, from analogs. The result of multiplying these
which, on an empirical basis, produces volumetrics deterministic inputs is a median recoverable re-
that are reasonably consistent in a large spectrum source, here called deterministic P50 (DP50), which
of undrilled structures. In other words, the stan- may be of marginal economic value based on the
dard contours are not based on any particular as- previous observation that only about half of the
sumed distribution of potential gross pay, but in- discoveries are large enough to have been economic
stead P50 hcbGRV serves as a suitable median point before drilling (Figure 7).
for trap volume defined by two fairly unequivocal A second calculation can be made for a high-
reference points: culmination and LCC. side recoverable resource case, which is often the
Nonetheless, in the method described in Table 4, commercial justification for exploration. For sim-
the implicit assumption is that LCC delineates an plicity, it has been assumed that the product of
hcbGRV close to the maximum potential trap size P50 hcbGRV and P10 reservoir parameters pro-
and that the distribution is not significantly skewed. duces a deterministic resource number that is ap-
Before using RPRI, a company would need to de- proximately equivalent to a probabilistic P5 value,
fine its own standard contours, cumulative prob- termed ‘‘deterministic P5’’ or DP5, an unlikely size
abilities, and distributions according to its own of discovery but still feasible. We have made nu-
models. The one suggested here is no better or no merous tests of this calculation and have found that
worse than many other assumptions on hcbGRVs DP5 generally falls in the range P10 to P1 when
or closure areas but has the advantage that it is similar inputs are used in probabilistic volumetrics,
simple to apply, thus facilitating consistency. so the assumption that the DP5 is similar to prob-
After defining LCC and P50 hcbGRV, the next abilistic P5 is valid based on empirical evidence.
step is to define at least two standard deterministic With these two deterministic values, equiva-
values of recoverable resource. The most straight- lent to P50 and P5, a full distribution can be con-
forward to calculate is a P50 volume, which is sim- structed in the cumulative standard normal do-
ply the product of P50 input parameters (Murtha, main (Figure 15). However, as a check, it is useful to

1280 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Table 4. Summary of Methodologies for Real Point Resource Iteration (RPRI)*

(a) Structural prospect without DHI or stratigraphic prospect with good control on column height

Identify culmination of structure and last closing contour (LCC) on best technical depth map at top of gross reservoir interval or,
if not coincident, at base of top seal
Define P50 hcbGRV by contour representing 56% (5/9th) depth from culmination to LCC
Estimate P50 and P10 values for reservoir parameters based on real/predicted ranges
Determine DP50 resource from product of P50 hcbGRV (or area) and P50 reservoir parameters
Determine DP5 resource from product of P50 hcbGRV (or area) and P10 reservoir parameters
Determine DP20 resource from product of LCC hcbGRV (or area) and P50 reservoir parameters
Use values of DP50 and DP5 to solve lnx = (s) NORMSINV(1 –Pn) + m and check fit of DP20
From resultant distribution, calculate P99, P90, Pmefs and meanP99 – P1, make reality checks, compare with other resource
estimations and iterate where necessary

(b) Stratigraphic prospect with limited constraint on column height or poorly imaged prospect

Determine reasonable maximum hydrocarbon column or spillpoint to define LCC


then continue as for structural prospect (a) above

(c) Prospect with unequivocal DHI

Determine depth of absolute minimum possible extent of anomaly (DHI minimum)


Determine maximum extent of anomaly to define LCC
Define P50 hcbGRV with contour representing 50% depth between DHI minimum and LCC
Estimate P50 and P10 values for reservoir parameters based on real/predicted ranges
Determine DP50 resource from product of P50 hcbGRV (or area) and P50 reservoir parameters
Determine DP5 resource from product of P50 hcbGRV (or area) and P10 reservoir parameters
Use values of DP50 and DP5 to solve lnx = (s) NORMSINV(1 –Pn) + m
From resultant distribution, calculate P99, P90, Pmefs, meanP99 – P1 and million bbl/km2, compare with other discoveries, analogs
and resource estimations and iterate, if necessary lowering P50 reservoir parameters as far as possible
*In most cases, a map of base of the top seal will be the same as top reservoir, but the instructions are written to also allow for prospects where a waste zone is
expected. Note, also, that in a seal-limited structural trap, LCC represents the best technical estimate of a maximum hydrocarbon column height before leakage,
assuming this is shallower than the mapped spillpoint. MeanP99 – P1 is the average recoverable resource of the distribution between P99 and P1; Pmefs is the reverse
cumulative probability of finding at least the minimum economic field size; hcbGRV is the hydrocarbon-bearing gross rock volume; NORMSINV is the inverse of the
standard normal distribution (a function in Microsoft Excel1).

determine a third recoverable resource lying some- using areas enclosed by the contours associated
where between DP50 and DP5, a size of discovery with P50 hcbGRV and LCC hcbGRV multiplied
that is not unreasonable if the original exploration by the P50 or P10 net reservoir thickness and P50
model holds true. In this respect, multiplying LCC or P10 shape factor. In fact, we strongly recom-
hcbGRV with P50 reservoir parameters empirically mend that volumetric estimations for any prospect
produces a result that lies somewhere close to P20 in should try both methods to check for inconsisten-
reliable probabilistic estimations and is therefore cies (Table 5). It is straightforward to resolve any
called deterministic P20 (DP20). DP20 is usually differences with RPRI but usually difficult when
fairly close to the meanP99 – P1 resource value for the only pure probabilistic methods are used.
prospect as this commonly has a cumulative prob-
ability somewhere between P35 and P20 (Table 5).
Note that although the discussion here con- Reality Checks on Volumetrics Results
centrates on volumetrics determined by the prod-
uct of standard hcbGRVs and the appropriate net Now that a distribution can be calculated using
to gross ratio, the calculations can instead be made RPRI, reality checks can be made similar to those

Quirk and Ruthrauff 1281


Figure 16. (a) Standard contours (LCC and P50)
and (b) hydrocarbon-bearing gross rock volumes
for the example shown in Figure 14. The P50 contour
defining the median value of hydrocarbon-bearing
gross rock volume (P50 hcbGRV) occurs at five ninths
(56%) the depth between culmination and LCC.
This is just one simple way of standardizing trap size
but has been found to cause the least dissension
among interpreters while producing what seem to
be realistic resource estimates.

listed in Table 5, based mostly on analyses of dis- distribution for that play. A large untested trap may
covery sizes from different plays made by Quirk legitimately have a P99 value higher than the his-
and Ruthrauff (2006). In particular, as the low-side torical value for the play, but it should be checked
part of the curve has been created by extrapolation that the smallest discoveries in the distribution
from high-side cases, one needs to check that P99 is were made on proportionally smaller prospects.
reasonable. The importance of historical data in this Usually, one or two iterations of the param-
quality check (QC) process comes from the fact eters used to calculate DP50 and DP5 are required
that any new discovery belongs to a discovery size before the results are in line with the checks shown

1282 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Table 5. Recommended Checks on the Validity of the Risk inconsistencies when reality checks were made,
and Size Distribution of an Exploration Prospect* specifically P99 appeared too large compared to
historical data from the play, the chance of making
General an economic discovery seemed too high and the
Distribution is lognormal (P50/P90 = P10/P50) probability of DP20 was too low. Therefore, the
MeanP99 – P1 lies between P35 and P20 (except for DHI plays inputs and results were iterated such that LCC
or very mature plays) was picked 25 m shallower, the depth of the cul-
P99 is <2 mmbbl or <20 bcf (except for prospects in mination was better constrained resulting in a slight
frontier areas) change in P50 hcbGRV, minor adjustments were
POSc/POSg (Pmefs) is < 70% (except for onshore plays made in reservoir parameters, and a new mini-
close to infrastructure and market) mum economic field size was calculated based
Volumetrics calculated using hcbGRV x reservoir net/gross on engineering scenarios for DP50, DP20, and
are consistent with volumetrics from area x net reservoir
DP5 (Figure 18a). The final results are shown in
thickness x shape factor
Figure 18b, where all checks and QCs are satisfied.
Compared to discovery data for play The exercise helped constrain probabilistic anal-
yses of the same prospect (Table 3) such that the
MeanP99 – P1 is not larger than historic average
base case using a lognormal distribution of hcbGRV
bbl/km2 or bcf/km2 is consistent with historical data
P99 is not significantly larger than historical P99 appears to be too pessimistic, primarily because
POSg is not significantly different from historic success rate it significantly skews the distribution of trap size.
Thus, the P50 depth contour is significantly shal-
*Based largely on information in Quirk and Ruthrauff (2006), also partly
summarized in Table 1.
lower than in other distributions, which in turn has
a disproportionately negative effect on recoverable
resource.
Table 6 compares three of the probabilistic
in Table 5, such as reviewing the probability of analyses from Table 3 with the resource estima-
reservoir net to gross or the exact depth of LCC. tions from RPRI (Figure 18). Within the range
Parameters such as P99, meanP99 – P1, cumulative P90–P10, RPRI provides resource numbers lying
probability of meanP99 – P1, and the chance that a somewhere between probabilistic analyses using a
discovery is commercial are important to check for lognormal distribution of hcbGRV and those using
consistency and identify possible exceptions. Note a symmetrical distribution such as normal or uni-
that DHI-constrained prospects, frontier plays, and form. Both P99 and P1 are larger in the RPRI ex-
onshore targets close to infrastructure may fall out- ample than in the probabilistic analyses, but they
side the usual constraints because of different ex- do fall in the range of current statistics for the play
ploration selection and sampling criteria. (P99 of 1.8 million bbl) and P1 values estimated
using other hcbGRV assumptions (Table 3). Fur-
ther fine tuning of the RPRI estimation is possible,
RPRI Example but as a quality checking of the probabilistic anal-
yses, it is adequate as is.
Figure 17 is a worked example of the methodol- Depending on which assumptions are recom-
ogy, prior to quality checking, applied to the struc- mended in a company for resource estimation of
tural prospect shown in Figures 14 and 16. P50 exploration prospects, an interpreter may wish to
hcbGRV and LCC hcbGRV were defined as de- review probabilistic inputs on the basis of a compar-
scribed above and in Table 4a, P50 and P10 res- ison of the outputs with RPRI results, particularly if
ervoir parameters were estimated using statistics this helps think about the depth contours used to
from the 21 wells that have tested the play, and the define trap size. The advantage of using RPRI is that
minimum economic field size was based on a pre- the quality assurance process can be applied quickly
liminary estimate. The initial results showed some and at an early stage in the evaluation.

Quirk and Ruthrauff 1283


Figure 17. (a) Example of inputs used in real point resource iteration (RPRI) for estimating recoverable resource for a prospect
shown in Figure 16, in this case before iteration. (b) Example of outputs from RPRI, before iteration, based on inputs shown in Figure 17a
and method described in Table 4a. The data are also displayed graphically in the cumulative standard normal domain, where the
lognormal distribution shown by the black trend line was defined by DP50 and DP5. The values circled in red are reality checks that
do not fit normal prospect ranges (Table 5), for example, P99 is higher than expected, the minimum economic field size (MEFS) has
an unusually high probability, and the probability and DP20 seems inappropriate (larger than P10). These issues were the reason
that the inputs were reassessed and slightly changed (Figure 18).

Application of RPRI to Other Situations calculated as before (Table 4). Usually, the mean
size of such a prospect can be large (albeit bal-
RPRI can also be applied to stratigraphic or poorly anced by a lower chance of success), so the defini-
imaged prospects and those with an unequivocal tion of what is a reasonable maximum depth needs
DHI using adaptations to the way hcbGRVs are to be objective, not optimistic.
calculated (Table 4b, c). In a prospect with an unequivocal DHI, the
In the case of a stratigraphic prospect or poorly recommended RPRI procedure to standardize trap
imaged prospect, the interpreter can decide on a rea- size is slightly different. Assuming it is robust and
sonable maximum depth below which it is unlikely well resolved, then the absolute shallowest possi-
hydrocarbons are trapped because of seal constraints, ble contour for the anomaly (DHI minimum) can be
closure limits, or dry wells. This maximum contour defined and treated in the same way that the culmi-
can then be used as an equivalent to LCC, and nation is in a structural prospect. With the same
thereafter, other contours associated with hcbGRVs logic, a reasonable maximum outline for the DHI
can be calculated in the same way as those of a equates with the LCC, and thereafter, P50 hcbGRV
structural prospect. Thus, the P50 hcbGRV may be is determined from the outline or contour equiva-
defined by the contour 56% of the depth between lent to a depth halfway between the DHI min-
culmination and LCC. Using P50 hcbGRV, LCC imum and LCC. From these, DP50 and DP5
hcbGRV, and estimates of P50 and P10 reservoir resources can be calculated in the same way as
parameters, then DP50, DP20, and DP5 can be structural and stratigraphic prospects (Table 4). It

1284 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 17. Continued.

has, however, proved more difficult to get consis- reducing P50 reservoir parameters such as net to
tency in DP20, probably because the distribution is gross, hydrocarbon saturation, and recovery fac-
so much narrower, so it is not used by us in DHI- tor, during the iteration process. One reality check
defined prospects. that helps improve confidence in the results is to
We have not tested RPRI on many DHIs and compare the yield parameter million barrels per
improvements may emerge in the method to define km2 (or bcf per km2) for the DP50 resource esti-
trap size probabilities (cf., Citron and Rose, 2001). mation to that of discoveries in the same play or
For example, with the method described in Table 4c, analogs and then iterate where it is necessary to get
it is sometimes difficult to build realistic low-side results consistent with the historical information
cases on DHI-defined prospects without severely (Table 5).

Quirk and Ruthrauff 1285


Figure 18. (a) Example
of RPRI inputs used for
same prospect evaluated
in Figure 17 but after
iteration. The values that
have been changed are
encircled. (b) Example of
outputs from RPRI, after
iteration, based on inputs
shown on Figure 18a and
the method described in
Table 4a. Note that real-
ity checks (red text) now
all fall within normal
ranges. The 16 million-bbl,
P70, minimum economic
field size case is referred
to in Figure 19.

Similar probabilities have purposely been used to calculate in-place volumes as an intermediate
in Table 4a–c to keep things as simple as possible stage in the RPRI process to help build engineering
while remaining aligned with results from pure scenarios before a final recoverable resource cal-
probabilistic techniques. It is worth reiterating that culation is completed similar to that outlined in
enough latitude is present in RPRI for different as- Table 4.
sumptions in probabilities to be used so long as
consistency is maintained. The essence of the tech-
nique is to produce a minimum of two standard Advantages of RPRI
deterministic volumetric calculations with agreed
probabilities that can then be used to construct Should RPRI replace pure probabilistic techniques?
an entire prospect size distribution (Figure 15) for RPRI produces distributions of recoverable resource
quality controlling results from pure probabilistic for a prospect by extrapolation from realistic high-
techniques (Table 6). The validity of any probabil- side cases and helps develop discipline in the way
ity or distribution can be tested by repetition and the sizes of prospects are defined. One benefit of
comparison with other estimations on many pros- this is that maps and reservoir parameters specific
pects and by reality checks against historical data to any probability of recoverable resource are a
(Table 5). direct and transparent product, meaning scenarios
So far, we have discussed the application of are available on which economics and commercial
RPRI to recoverable resources, but it is also pos- decisions can be based. However, the assumptions
sible to use the technique for calculating in-place made to get these are still assumptions, even if the
volumes, leaving consideration of recovery for results are consistent. Furthermore, it is not diffi-
later when dealing with specific discovery scenar- cult to standardize ways of defining inputs to prob-
ios. The disadvantage of this approach is that it is abilistic estimates. All things considered, RPRI is
more difficult to integrate uncertainty in recovery better applied as a control instead of a replacement
when it forms part of a separate step. We prefer for other techniques.

1286 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 18. Continued.

To summarize, using RPRI as a check on proba- whilst allowing them to check other more
bilistic outputs leads to the following advantages: subjective calls on probabilistic distributions
and probabilities (Figures 17a, 18a).
1. It incorporates some of the advantages of 4. It fosters consistency in prospect evaluations by
deterministic-only techniques (Table 2). standardizing the way uncertainty is dealt with,
2. It can be easily understood, is straightforward particularly in the size of hcbGRV (Figure 16).
to apply, and does not require extra training or 5. It mitigates against overoptimism by statistical-
highly prescriptive corporate rules (Table 4). ly accounting for low-side outcomes (Figure 15)
3. It helps interpreters focus on parts of the pros- and formalizing the integration of reality checks
pect model that they have most confidence in into volumetric assessments (Figures 17b, 18b).

Quirk and Ruthrauff 1287


Table 6. Comparison of Recoverable Resource Estimates from RPRI and Probabilistic Methods for Prospect V (Figures 14, 16)*

Recoverable Resource (mmbbl)

Resource Estimation Method Mean P99 P90 P50 P10 P1 P10/P90

RPRI (see text and Figures 16, 18) 59.2 1.7 6.3 31.0 153.3 563.6 24
Probabilistic, lognormal hcbGRV (Table 3) 46.4 0.3 2.6 19.0 110.7 426.6 43
Probabilistic, normal hcbGRV (Table 3) 84.8 0.7 8.5 55.3 193.9 433.2 23
Probabilistic, uniform hcbGRV (Table 3) 87.3 0.5 6.8 55.0 207.6 448.6 31
*Note that the probabilistic estimation using a lognormal hcbGRV (hydrocarbon-bearing gross rock volume) was the base case used in Table 3.

6. It makes it easy to iterate results and allows for probabilities and parameters to compare with what
quick resolution of differences between volumet- was actually found in a successful well. In the long
rics calculated using gross reservoir times net to run, this should help improve predictions in prob-
gross and volumetrics calculated using area times abilistic volumetrics, particularly in a play likely to
net reservoir. be tested multiple times.
7. It helps build real maps and reservoir data for
direct use in engineering studies, project deci- Comparison of RPRI with Probabilistic Estimations
sions, commercial analysis, and in selecting well
locations (Figures 16, 18). Several hundred RPRI calculations have been run
8. It allows individual prospects in a portfolio to by the authors in a wide range of different play
be quality checked rapidly (Table 5) and can be types and matched directly against probabilistic
used to produce specific cases for peer review. volumetrics produced from similar input values
9. It facilitates rapid construction of probabilistic (for example, Table 6). The results are compara-
resource distributions from limited data, for ex- ble, particularly with respect to high-side resource
ample, during farm-in evaluations (Figure 15). numbers (for example, meanP99 – P1, DP20-P20, and
10. It can easily be integrated with other procedures DP5-P5). However, a difference is usually detected
such as economics programs. in the low-side volumes such that P99 calculated
11. It produces a limited number of standard mea- from RPRI lies somewhere between probabilistic
sures of size for each prospect (Figure 18b). P99 and probabilistic P90, and the P10 to P90 ratio
is consequently smaller. The main reason for the
Expanding on point 7, because of the fact that discrepancy is that, even if hcbGRV is fixed with
simple distributions can be created from determin- the same LCC and culmination constraints, the trap
istic input values, it is possible to backcalculate size distributions assumed in the two approaches
standard parameters for any cumulative probabil- are usually different. The difference is regarded as
ity or size of recoverable resource. This means that insignificant but nonetheless underlines the im-
a simple map and reservoir properties can be pro- portance of using reality checks such as historical
duced for specific probabilistic cases for engineer- P99 for the play in question. It also shows that it is
ing, economics, and planning purposes such as a worth experimenting with different distributions
minimum economic field development (Figure 19). and probabilities before RPRI is adopted as a stan-
This process is more complex with pure probabilis- dard quality-checking tool.
tic evaluations, as any specific volume or proba- Here, it is worth noting that RPRI can be ap-
bility can be associated with numerous potential plied in a slightly different way using only one de-
scenarios. terministic value, such as DP5, to fix the upside and
One further advantage of RPRI is that the va- historical P99 for the play to fix the downside. This
lidity of predictions can easily be checked by using method can be applied rapidly in meetings or farm-
postdiscovery volumetrics to backcalculate predrill in situations as a quick check on resource numbers.

1288 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Figure 19. Trap and reservoir parameters
for any probability or size of discovery can
be obtained from RPRI inputs using assumed
simple and linked distributions. As an example,
a 16-million bbl resource case for prospect V
(Figures 16 and 18), equivalent to the minimum
economic field size (P70), is produced using
P55 input values. This information, including the
depth contour map shown, can be used directly
by engineers for development scenarios and
economics prior to drilling. Also, after drilling a
successful exploration well, an actual discovered
volume can be used to backcalculate probabil-
ities and the predicted trap size and reservoir
parameters in an effort to try and improve future
evaluations.

However, with more time, it is still recommended calculations are made for quality check proba-
that at least one other deterministic number is cal- bilistic estimates. Alternatively, in companies that
culated, for example, DP50, to ensure that consis- have not adopted probabilistic techniques, RPRI
tency is present in the trap size and reservoir pa- can be used to attach probabilities to deterministic
rameters estimated specifically for the prospect. numbers.
This article does not advocate that existing vol-
umetric methodologies are dropped in favor of Examples of Successful RPRI Application
RPRI, but in an exploration company looking to
improve its predictions of prospect size, it is rec- Although still in its early stages, two companies
ommended that certain standard deterministic are currently testing its application. It is too early

Quirk and Ruthrauff 1289


to report an improvement in the accuracy of pre- cific reservoir information are tied directly to cu-
diction, but there is certainly more confidence in mulative probabilities of recoverable resource, of
the consistency of resource assessments in pros- use in predrill commercial analysis and postdrill
pect portfolios where RPRI has been used. review.
In summary, and to answer a question posed
previously in the article, provided that the uncer-
tainty is dealt with in a simple and transparent way REFERENCES CITED
and then rigorously cross-checked against histor-
ical data, consistent and reliable estimates of the Aitchison, J., and J. A. C. Brown, 1957, The lognormal dis-
sizes of future exploration discoveries should in- tribution: Cambridge, Cambridge University Press,
deed be possible. 87 p.
Campbell Jr., J. M., J. M. Campbell Sr., and R. A. Campbell,
2001, Analyzing and managing risky investments: Nor-
man, Oklahoma, John M. Campbell Publishing, 486 p.
Capen, E., 1992, Dealing with exploration uncertainties, in
CONCLUSIONS R. Steinmetz, ed., The business of petroleum explora-
tion: AAPG Treatise of Petroleum Geology, Handbook
of Petroleum Geology, chapter 5, p. 29 – 61.
From the analysis of exploration discovery data in Citron, G. P., and P. R. Rose, 2001, Challenges associated
individual plays, several useful constraints can be with amplitude-bearing, multiple zone prospects: The
applied to prospect evaluations: Leading Edge, August 2001, p. 830 – 838.
Davidson, L. B., and D. O. Cooper, 1976, A simple way
of developing a probability distribution of present val-
1. All plays with more than 20 discoveries show ue, Journal of Petroleum Technology, v. 28, p. 1069–
lognormal distributions of discovered volumes 1078.
Doré, G., and J. Robbins, 2005, The Buzzard field, in A. G.
2. Most of these plays show predictable upper lim-
Doré and B. A. Vining, eds., Petroleum geology: North-
its to P99 recoverable resource west Europe and global perspectives — Proceedings of
3. Most of these plays also show similar cumula- the 6th Petroleum Geology Conference: Geological So-
tive probabilities of minimum economic field ciety (London), v. 1, p. 241 – 252.
Hesthammer, J., and M. Boulaenko, 2005, The offshore EM
size and of mean resource between P99 and P1 challenge: First Break, v. 23, p. 59 – 66.
(meanP99 – P1) Kaufman, G. M., 1963, Statistical decision and related tech-
niques in oil and gas exploration, Englewood Cliffs, New
Because of the inherent uncertainties in as- Jersey, Prentice-Hall, 307 p.
Megill, R. E., 1984, An introduction to exploration risk anal-
sessing the distribution of hcbGRVs in undrilled ysis, 2d ed., Tulsa, Oklahoma, PennWell Publishing Co.,
prospects, there is value in having a quality control 273 p.
tool for probabilistic resource estimates. We ob- Murtha, J., 2001, Risk analysis for the oil industry: Supple-
ment to Harts E&P, September 2001, 31 p.
serve that information from deterministic models
Murtha, J., 2002, Sums and products of distributions: Rules
can help constrain the upside end of a resource dis- of thumb and applications, SPE Paper 77442, Society of
tribution and historical data can help constrain the Petroleum Engineers Annual Technical Conference and
downside. A distribution of recoverable resource Exhibition, San Antonio, Texas, September 29 – Octo-
ber 2, 2002), 5 p.
can be constructed for any prospect using two de- Niemann, J. C., 1998, Statistical distributions of hydro-
terministic values of known probability (for exam- carbon column heights for Gulf of Mexico trap types
ple, P50 and P5) with a method known as RPRI. and seals (abs.), AAPG Hedberg Research Conference —
Statistical information is used to check the validity Integration of Geological Models for Understanding
Risk in the Gulf of Mexico, Galveston, Texas, Septem-
of the lower half of the distribution, which is iter- ber 20 – 24, 1998, unpaginated.
ated where necessary. The main application of Otis, R. M., and N. Schneidermann, 1997, A process for
RPRI is in obtaining consistency in the evaluation evaluating exploration prospects: AAPG Bulletin, v. 81,
p. 1087 – 1109.
of prospect size. It is simple and quick to apply and
Quirk, D. G., and R. Ruthrauff, 2006, Analysis of reserves
helps avoid overoptimism. One particular advan- discovered in petroleum exploration: Journal of Petro-
tage of the technique is that real maps and spe- leum Geology, v. 29, p. 125 – 146.

1290 A Systematic Way of Constraining Uncertainty in Prospect Volumetrics


Root, D. H., and L. J. Drew, 1979, The pattern of petro- link between chronic E&P underperformance and the
leum discovery rates: American Scientist, v. 67, p. 648 – prospector myth (abs.): AAPG International Conference,
652. Barcelona, September 21 – 24, 2003 (presentation and
Rose, P. R., 1987, Dealing with risk and uncertainty: How can notes available at www.roseassoc.com).
we improve?: AAPG Bulletin, v. 71, p. 1– 16. Smith, M. B., 1970, Probability methods for petroleum in-
Rose, P. R., 2001, Risk analysis and management of petro- vestment decisions: Journal of Petroleum Technology,
leum exploration ventures: AAPG Methods in Explora- v. 22, p. 543 – 550.
tion 12, 164 p. Trotter, H. F., 1959, An elementary proof of the central limit
Rose, P. R., G. P. Citron, and M. A. McClane, 2003, The theorem: Archiv der Mathematik, v. 10, p. 226– 234.

Quirk and Ruthrauff 1291


View publication stats

You might also like