E & P NOTES AUTHORS

A. Hurst Department of Geology and

Petroleum Geology, University of Aberdeen,
Swansons 30-40-30 rule Aberdeen, AB24 3UE, United Kingdom;
a.hurst@geol.abdn.ac.uk
A. Hurst, G. C. Brown, and R. I. Swanson Andrew Hurst received a Ph.D. in geology
from the University of Reading in 1980. From
1981 until 1990, he worked for Statoil,
Norway, in the petroleum technology group.
ABSTRACT From 1990 to 1992, he was employed as an
exploration geologist by Unocal UK. In 1992,
Evaluation of the possible range of reserves associated with a pros- he assumed the new Shell Chair of Production
pect is a key part of risk taking in hydrocarbon exploration. The Geoscience at Aberdeen University. His main
challenge of presenting a range of geologically possible models for research interest is in reservoir
a range of prospect reserve estimates is addressed using Swansons characterization, particularly of turbidite
30-40-30 rule. Swansons rule defines the mean as 0.3P10 0.4P50 reservoirs. He is the chief editor of Petroleum
0.3P90, and provides a good approximation to the mean values Geoscience.
for modestly skewed distributions. Pragmatic and mathematical jus-
G. C. Brown Department of
tifications for this rule are given. Applications of the rule to a his- Mathematical Sciences, University of
torical field size distribution and a specific prospect evaluation dem- Aberdeen, Aberdeen, AB24 3UE, United
onstrate its efficacy in handling routine problems in hydrocarbon Kingdom. Current address: Office for National
exploration, with particular reference to use with the lognormal Statistics, 1 Drummond Gate, London, SW1V
distribution. 2QQ, United Kingdom;
Gary.Brown@ons.gov.uk
Gary Brown received a B.Sc. degree (1989) in
INTRODUCTION mathematics and a Ph.D. (1995) in statistics
from the University of Birmingham. From
Translation of geological concepts and models used in exploration 1996 to 1999 he was employed as a research
prospect evaluation into a language that is amenable to economic fellow in statistics at the University of
evaluation is a challenging task (Rose, 1987). The uncertainty as- Aberdeen, where he also ran an internal and
sociated with the geological parameters used to calculate resources external statistical consultancy service. His
or reserves is difficult (if not impossible) to quantify. A traditional main research interest is in multivariate
regression, especially in the associated
approach to estimation of the most likely reserves in a prospect is
diagnostics. Since 1999 he has been
to assume that the midpoint (median) values for areal extent, pay
employed as a methodologist in the Office for
thickness, and recovery factor can be estimated and their triple National Statistics.
product calculated. Monte Carlo type or other probabilistic simu-
lation procedures can then be applied to generate ranges of reserve R. I. Swanson deceased early 2000.
estimates from these individual parameters. This approach makes Roy Swanson received a B.S. degree (1953) in
the assumption that both larger and smaller cases for each required geology from the University of Pittsburgh and
value (leading to a high reserves case and a low reserves case, re- an M.S. degree (1955) in geology from
spectively) are possible to estimate and geologically reasonable. Michigan State University. He worked for
The triple product approach may be argued to be too sim- Exxon Corporation first as a geophysicist in
plistic in some applications. For example, in situations where there Wyoming, Colorado, and Venezuela and then
is a high level of confidence in the mapped geometry of a prospect as a geologist in New Orleans. In 1969, he
entered the area of exploration economics
and, possibly, where a direct hydrocarbon indicator is present, one
and in 1981 became coordinator of Inland
may argue that it unnecessarily oversimplifies reality. Although we
Exploration analysis for Exxon. He retired from
Exxon in 1992.
Copyright 2000. The American Association of Petroleum Geologists. All rights reserved.
Manuscript received October 6, 1998; revised manuscript received January 24, 2000; final acceptance
March 27, 2000.

AAPG Bulletin, v. 84, no. 12 (December 2000), pp. 18831891 1883

 believe that these considerations may hold true for exploration in
ACKNOWLEDGEMENTS mature hydrocarbon plays, experience tends to demonstrate the
frailty of our predrill knowledge and a tendency toward overcon-
We would like to acknowledge the helpful
fidence. Assumption of an unrealistic slablike (area thickness)
comments of Bill James (Exxon), to whom we
geometry may fail to capture a realistic three-dimensional prospect
are also indebted for the neat theoretical justi-
fication of the 30-40-30 rule presented in this geometry but does allow simple generation of average values. The
paper. Peter Rose and Joseph Studlick pro- heterogeneity of reservoirs, both depositional and diagenetic, ac-
vided useful review comments for which we tually produces a range of recovery factors, even within a restricted
are grateful. Through his excellent teaching of area of a play, which is commonly confirmed by production his-
this method, Peter Rose provided the impetus tory. However, adopting a more sophisticated approach to reserve
to document the theoretical justification of the estimation where technical risk cannot be adequately constrained
method. may only serve to introduce uncertainty in estimates rather than
Our thanks to Roy Swanson for his invalu- constrain them.
able contributions to this paper, and to the If prospect sizes lie on a lognormal curve, taking a midpoint to
field of petroleum geology in general. It was represent the most likely reserves case does not yield the true me-
with great sadness we heard of his passing
dian. Equally significant is that, during the process of prospect gen-
away in March 2000.
eration, geologists and managers commonly have difficulty in en-
visaging prospects smaller than the 90th percentile or larger than
the 10th percentile. To take account of this, Swanson, in an Exxon
internal memorandum (1972, personal communication), encour-
aged the use of average prospect size rather than a subjective most
likely case. To achieve this, he proposed use of a rule when explain-
ing prospect economics. This rule, called the 30-40-30 rule, is
commonly referred to as Swansons rule and has since been widely
used in the hydrocarbon exploration industry (Megill, 1984; Rose,
1991; Rose and Thompson, 1992).

THE 30-40-30 RULE

The rule set out to select three sizes along the reserves distribution
curve, having the aim of reflecting the range of the distribution
curve and, where weighted by discrete probabilities, giving the av-
erage value of the prospect reserves.
The three sizes were proposed as

the medium (50th percentile median, P50), which half the

reserves were larger than, half smaller;
the maximum (10th percentile, P90), which only 10% were larger
than;
the minimum (90th percentile, P10), which 90% were larger
than.

Probabilities were then assigned to P10, P50, and P90 based on

the proportions of the total range 0100% appropriate to each. For
P50, the ranges to P10 and P90 were 0.40 of the total range in each
case (i.e., 1050% 0.40 of the total range). These 0.40 ranges
were then halved and one half assigned to each (i.e., 5070%
0.20 of total range assigned to P50 and 7090% 0.20 of total
range assigned to P90). By this method the probability of P50 was
0.20 0.20 0.40 (i.e., 3050% and 5070%). Following this,

1884 Swansons 30-40-30 Rule

each of P10 and P90 already had probabilities of 0.20
assigned to them. Added to these probabilities were
0.10 to represent the tails of the distribution beyond
(i.e., 90100% 0.10 of total range for P90), which
gave probabilities of 0.30 for each of P10 and P90 (i.e.,

FREQUENCY
7090% and 90100% for P90). The weightings of P10,
P50, and P90 were thus 0.30, 0.40, and 0.30, respec-
tively, summing to a value of 1 (i.e., total range 0
100%) as required for a probability distribution. This
defines Swansons 30-40-30 rule, which was noted to
work for all except highly skewed distributions.

MIN ML MAX
JUSTIFICATION PARAMETER SIZE

Figure 1. Symmetric triangular distribution. MIN minimum;

In the original memorandum, Swanson noted that hav- ML most likely; MAX maximum.
ing different choices of percentiles for the maximum
and minimum sizes (and corresponding new probabil-
ities) did not return as accurate an estimate of the av-
erage as the 30-40-30 rule. Thus no equivalent 20-60- ues to economic values. Further uses are in the plotting
20 rule would be as accurate. of possible reserves for individual prospects and com-
A test for the variate being lognormal was origi- parison of prospect size with historical field size
nally proposed as distributions.
Whether Swansons rule, or a more sophisticated
approach, is used for reserve estimation is determined
Minimum Value Most Likely Value
(1) by user confidence in the various risk parameters being
Most Likely Value Maximum Value
assessed. High confidence follows from experience and
analysis of all appropriate data using state-of-the-art
If values satisfied equation 1 then this was taken to technology. First-time participation in a play, or in the
indicate a lognormal distribution. Equation 1 was also testing of a new play, is unlikely to be accompanied by
given as a method of finding an unknown value if log- such confidence, and the triple product approach will
normality was already assumed and only two values be more attractive. Although modern computational
were known. The connection between log symmetry power makes it relatively fast and straightforward to
and lognormality in equation 1 is not mathematically generate many resource and economic scenarios, defin-
justifiable, and consequently the test was later ing the parameters necessary to generate resource es-
dropped. However, the lognormality of the reserves timates remains time intensive and subjective.
was exemplified by the reserves of known contempo-
rary fields plotting a straight line on lognormal proba- Prospect Reserve Estimation
bility paper (Figure 1). The precise percentiles and
their corresponding probability weightings, chosen to When estimating the likely range of reserves for a pros-
estimate the mean of such a positive skewed distribu- pect, an exploration geologist is always faced with
tion as the lognormal, were actually found empirically problems associated with the limitations of the data
by computer simulation. from which the prospect has been generated. A com-
A mathematical justification of the 30-40-30 rule mon misconception in estimating reserves is that the
is provided in the Appendix. minimum and maximum values for specific parameters
(e.g., closure, thickness, porosity, recovery, etc.) plot
on frequency diagrams having a value of zero (Figure
APPLICATIONS 1). This is a statistical misconception because, by plot-
ting at zero, the high and low values cannot occur. Fur-
During prospect evaluation, the main application of ther, this means that any estimates lower or higher
Swansons rule is to translate resource assessment val- than these values also cannot exist.

Hurst et al. 1885

 Therefore, to accommodate the need to include a costs. Clearly, this definition of P10 varies enormously,
range of values greater than the known values, reserves depending on local conditions and operator economics.
are plotted as percentiles. It is important to realize that Once P10 and P50 are plotted, a P90 value may be ex-
all we know is not all there is to know and that normal trapolated. At this point, it is wise to consider how
practice in subsurface studies involves dealing with in- reasonable the extrapolated P90 and P99 reserve esti-
complete data (Foley et al., 1997). mates are, both by using input parameters specific to
Additional to the misconception regarding the the prospect and by comparison with historical field
meaning of statistical distributions is a tendency for ex- size distributions for the trend or play, if available. If
ploration geologists to believe that a prospect map, the P90 and P99 estimates seem unreasonably large
commonly the product of a highly individual exercise, given the geological constraints (that must be geolog-
is the only (certainly, most likely) map for a prospect ically feasible), the most likely case should be adjusted
(Megill, 1984). Although it only takes a little consid- to a higher value (toward P60) to increase the gradient
ered thought or practical experience to dismiss this be- of the lognormal probability curve (Figure 2, line a). If
lief, human nature tends to anchor estimates to original the P90 estimate appears too small, because it is easier
concepts and interpretations (Rose, 1987). Capen et to make a substantially larger reserve estimate using
al., (1971) and Bain (1993) give examples that dem- reasonable geological input parameters, the most likely
onstrate how different exploration geologists, using case should be adjusted to a lower probability (toward
similar data, can produce a range of different assess- P40, Figure 2, line b).
ments of hydrocarbon exploration prospectivity. In the
light of this tendency to anchor onto the reserves case
perceived as most popular by the exploration geologist
involved, the generation of highside and lowside re- 99

serve cases is always challenging. Once generated,

though, use of lognormal probability plots allows sim- 95

ple extrapolation above and below an exploration ge- 90

a
b
ologists favored reserves case (Megill, 1984; Rose,
1987; Rose and Thompson, 1992). 70
Unfortunately, no rules exist for constructing pros-
PROBABILITY

pect lognormal probability plots. Equally, no rule ex- 50

ists for use of the probability axis. Therefore, P90 is 30

quoted by different users to represent a low probability
of occurrence (there is a 10% probability of a reserve
10
being greater than or equal to the P90 case) or a high Swanson's Mean
probability of occurrence (there is a 90% probability base case 60 MMBO
of a reserve being greater than or equal to the P90 case). a 37 MMBO

Here we adopt the former use (Rose, 1987; Rose and 1 b 160 MMBO

Thompson, 1992). In constructing a lognormal prob-

ability curve, Megill (1984) favors defining the P50 and
P10 (lowside estimate) values and extrapolating to the
P90 (highside). The exploration geologists favored re- 1 10 100 1000
serves case, perhaps termed the most likely technical RESERVES (MMBO)
case, is a reasonable place to start.
Intuitively, one may expect the most likely case to Figure 2. Adjusting lognormal curves for estimation of pros-
be close to the P50 value on a probability curve, though pect size. In the initial plot (base case) P10 is assigned as an
economic cutoff value, P50 is the exploration geologists most
this need not be the case. However, by plotting the
likely case, and Swansons mean values, using the 30-40-30 rule,
most likely case as P50 one can begin to generate a plot are shown. P90 is extrapolated from these values. Line a illus-
for a reserve distribution if a value for P10 can be de- trates a scenario where the original most likely case is consid-
fined (Rose, 1991). A pragmatic approach to defining ered too high and lowered to less than P50. Line b illustrates a
P10 is to assume that it corresponds to the economic scenario where the original most likely case is considered too
cutoff for a single well discovery that when converted low and moved to greater than P50. MMBO million bbl of
to a producing well would be expected to recoup all oil.

1886 Swansons 30-40-30 Rule

 This procedure should never be treated merely as Table 1. Fields and Discoveries in the Upper Jurassic, Salt-
a numbers game. Any changes to the reserves Related Play in the United Kingdom and Norwegian Central
log(probability) must be as a consequence of geologi- North Sea*
cally reasonable adjustments to mappable parameters.
Good practice involves generating several reserve dis- Field/Well Date of Discovery Reserves (MMBO)
tribution curves for a prospect until one feels comfort- Fulmar 1975 462
able that the distribution is representative of the inter- Clyde 1978 150
preters perception of reality. 21/24-1 1978 10
A common error in prospect evaluation is to assume Guillemot A 1979 50
that a lognormal reserves distribution has successfully Gyda 1980 195
captured the entire spread of reasonable reserves for a Iris 1980 10
prospect. Whether the reserves distribution has been 30/12b-2 1981 29
generated manually or by using software, visual checks 21/19-1 1981 10
should be made that high case reserves (P90) fall N7/12-5 1981 12
within the bounds of reason; prospect area and thickness Kittiwake 1981 70
do not exceed what is physically possible to accommo- 21/15a-1 1981 50
date; and reservoir properties do not exceed reasonable 21/13b-1 1982 80
values for porosity and recovery factors. Mime 1982 15
Angus 1983 8
Historical Data 21/19-2 1983 16
21/20-2 1983 20
Field size distributions in a basin or play commonly Duncan 1985 20
approximate lognormal trends (Megill, 1984; David, N15/12-5 1986 100
1996). When prospects are generated it is commonly 29/6a-3 1987 30
useful to compare reserve estimates with known re- 22/13a-1 1989 90
serves from the same trend. This practice aims to pro- N2/4-14 1990 175
vide an objective comparison of prospect reserves
*Data for fields are public domain information available in the period 1989
with known reserves. We do not discuss the issue of
1991. Other data are estimates obtained from independent studies and scout
determining the similarity of discoveries to prospects data. Prefix N denotes a Norwegian sector well.
in any further detail, except to state that combining
and comparing dissimilar data prejudices statistical
relevance.
Plotting historical field size distributions is done in 78% probability of a field of 13 MMBO or larger oc-
the same manner as prospect reserves. An example of curring, corresponding to X78 in equation 5). These
historical data from the central North Sea (United probabilities of occurrence are not probabilities of suc-
Kingdom and Norway) compiled during the period cess for the individual prospects. Mitigating circum-
19891991 is used to demonstrate their use for sup- stances may exist that will influence this simple inter-
porting evaluation of exploration risk. Field and dis- pretation of the data.
covery data are summarized in Table 1, and the full
historical trend is shown in Figure 3. Three prospects
within the trend, X, Y, and Z, are marked on the dis- CASE STUDY
tribution in Figure 3. The implications to be drawn
from the position of the prospects are that, given the Prospect X is a three-way dip closure into a major basin
historical context of exploration, X (250 million bbl of margin fault having small areas of four-way dip closure
oil [MMBO]) is relatively unlikely to occur (a 9.5% and an inferred stratigraphic pinch-out to the east that
probability of a field of 250 MMBO or larger occurring, provides substantial possible upside to reserve esti-
i.e., 250 MMBO would correspond to X9.5 in Appen- mates (Figure 4). Substantial oil accumulations are
dix equation 5); Y (45 MMBO) has a reasonable prob- known downdip and to the west of prospect X, where
ability of occurrence (a 46% probability of a field of 45 the target reservoir is thick and productive. Strati-
MMBO or larger occurring, corresponding to X46 in graphic pinch-out is key to providing closure but also
equation 5); and Z (13 MMBO) is likely to occur (a makes implicit that presence, thickness, and quality of

Hurst et al. 1887

 99

95

P 90 = 248 MMBO
90
X P 90.5
70 Swanson's mean
= 92 MMBO
P 50 = 38 MMBO
PROBABILITY

50
Y P 54
30

P 10 = 8 MMBO
Z P 22
10

X = 250 MMBO
1 Y = 45 MMBO
Z = 13 MMBO

1 10 100 1000

RESERVES (MMBO)

Figure 3. Field size distribution for the Upper Jurassic, salt-

related play (United Kingdom and Norway, North Sea) compiled
during 19891991. Prospects of the size of X, Y, and Z (Table
1) have reserve estimates of 250, 45, and 13 million bbl of oil
(MMBO), respectively.

the target reservoir are significant uncertainties. The

reservoir is known to thicken immediately downdip
and to the west, where several dry holes are located. Figure 4. Prospect X. A three-way dip closure into a basin
Further east and to the south the reservoir is not pres- margin fault. Areas A, B, and C are three possible geological
interpretations of hydrocarbon distribution (outline areas also
ent. At the time of evaluation, no similar play concept
shown as insets). Well names are indicated by paired letters.
had been tested.

Constraints
gin fault. The general thinning of the target reservoir
As no direct hydrocarbon indicator is identifiable on unit from west to east (Table 2) leads to the concept
the seismic data to aid mapping of hydrocarbon distri- of stratigraphic trapping that combines depositional
bution, the approach to reserve mapping was to map pinch-out and an erosional truncation.
closure manually, honoring well data, and to infer pos- For the sake of simplicity, only prospect areas are
sible hydrocarbon distribution. Well GG penetrated an disclosed in this example. In reality, variations in net
8 m thick oil column and an oil-water contact in the pay and recovery factor were also considered in the
target reservoir horizon; wells AA and BB failed to lo- reserve distributions (cf. Megill, 1984; Rose, 1991).
cate the target reservoir. All other wells located a
water-bearing reservoir. Possible Hydrocarbon Distribution
Well AA is located on the footwall of a major basin
margin fault. Well BB penetrated a thick unit of the The oil-water contact proved at 2373 m allows extrap-
top seal that directly overlies the target reservoir in olation of a simple reserve case limited updip by a fault
wells to the west but in BB is in faulted contact with to the east of well GG and downdip by the fluid con-
basement. This fault is interpreted to be the basin mar- tact (A, Figure 4). If the fault is considered to be non-

1888 Swansons 30-40-30 Rule

Table 2. Well Data Adjacent to Prospect X ent value equal to zero (at the time of the evaluation).
Three reserves-probability distributions, P50A, P50B,
Well Status Reservoir Thickness
and P90C (Figure 5), are generated using reserve esti-
AA Dry hole np* mates of 18 MMB, 45 MMB, and 92 MMB (associated
BB Dry hole np with areas A, B, and C, Figure 4). Curve P50A is the
CC Dry hole 21 m most conservative reserve distribution for Prospect X
EE Dry hole 28 m and curve P50B the most optimistic. One can equally
GG Oil well 17 m interpret the conservative curve as being equivalent to
HH Dry hole 36 m a high level of confidence in the outcome and the op-
timistic curve as a lower level of confidence. Selection
*np reservoir not present.
of the most appropriate reserve distribution should be
made by examination of geological evidence. Any re-
serve scenario must be geologically possible.
sealing, it is possible that the reservoir can extend far- Examination of P values less than P90 (10% prob-
ther east up to the inferred stratigraphic termination ability of occurring) are particularly useful in this con-
(B, Figure 4); the presence and quality of the reservoir text because if high reserve cases require unreasonable
are substantial risks in this case. Both these scenarios geological conditions to exist, they are probably un-
underfill the mapped closure. realistic. The greater than P90 reserves for curves P50A
Underfilling of closure presents many possible geo- and P50B are values greater than 47 MMB and 280
logical scenarios that may elucidate geological risk and MMB, respectively. The 280 MMB is more than three
make possible upgrading of reserves. The trap may times the reserves associated with area C (Figure 4),
have received too little charge or the top seal may have
insufficient capillary pressure to hold back a thicker
column (possibly because of faulting). As a thick shale
unit underlies the target reservoir, bottom seal leakage
99
is not considered to be a significant element of risk. The
direction from which oil has migrated into Prospect X

0C
P9
is significant in determining hydrocarbon distribution. 95
P90
Migration is believed to be from the northwest past 90

well EE. Given the geometry of the inferred pinch-out

0A

0B
P5

to the east and south, it is possible to generate a model 70 P5

in which a perched oil-water contact occurs (C, Figure

PROBABILITY

P50
50
4). The shallower oil-water contact in well GG and in
the south of the prospect is caused by physical trapping 30

of formation water in a segment of reservoir that is

bounded by the basin margin fault penetrated by well 10
P10

BB (assumed to be sealing) and the stratigraphic pinch-

MINIMUM COMMERCIAL RESERVE = 7MMB

out (Figure 4).

The three possible hydrocarbon distributions (A,
1
B, and C) for Prospect X are just three of an infinite
range of possible reserves cases that can be described
using a lognormal-probability crossplot. Using com-
mon sense and experience one should recognize that
the geological risk increases from A through to C, but 1 10 100 1000

how does one evaluate the most likely technical case? RESERVES (MMBO)

Following the method of Rose (1991), the reserves-

Figure 5. Reserves (log) and probability crossplot for Prospect
probability crossplot is generated by choosing to use X (Figure 4). All curves use 7 million bbl (MMB) as the mini-
the minimum commercial reserve value for Prospect mum commercial reserve value. The fine dashed curve (P50A)
X, 7 million bbl (MMB), to represent P10 (a 90% prob- is derived using area A as the P50, the broad dashed curve
ability of 7 MMB or less occurring). This assumes that (P50B) uses area B as the P50, and the solid curve (P90C) uses
discovery of 7 MMB generates an incremental net pres- area C as the P90. MMBO million bbl of oil.

Hurst et al. 1889

which is an optimistic scenario for hydrocarbon distri- ration scenarios the level of confidence associated with
bution (perched oil-water contact). Consequently, 280 prediction of prospect geometry and hydrocarbon
MMB seems likely to be much too large for a P90 value presence is high, practical experience of exploration
and quite possibly unlikely to occur at all. The P50B drilling demonstrates a record of overconfidence in
curve is rejected as representative reserves distribution prediction. Swansons 30-40-30 rule has a long history
for Prospect X because the range of reserves is impos- of application to the translation of resource assessment
sible, or at least extremely difficult, to accommodate into economic values, however the efficacy of the
geologically. In contrast, curve P50A is rejected as a method is dependent on effective representation of
representative distribution because the P90 and P99 re- geologically reasonable models throughout the range
serves (47 and 100 MMB, respectively) (Table 3) are of reserves represented.
easily accommodated geologically and the Swansons In defining an appropriate curve to represent a
mean (21 MMB) is small relative to the possible range of reserves for a specific prospect an iterative ap-
mapped hydrocarbon distributions (Figure 4). proach is used. Typically, the P10 value (90% proba-
As the results of further exploration of Prospect X bility of finding reserves of this size or lower) can be
are unavailable, validation of the reserves distribution assumed to be the minimum commercial reserves
is not possible. However, it seems likely that the ap- value, the most likely reserve case may be used as a
propriate curve lies between P50A and P50B. Geolo- provisional P50, and a reserve distribution curve be
gists working Prospect X believed that area B (Figure generated. The validity of the curve, and hence the
4) is quite likely to occur, despite reserves of 45 MMB probability of the most likely case, can then be tested
being unlikely to be a reasonable P50 value for the pros- by evaluating how geologically realistic the high re-
pect. Curve P90C is plotted in the belief that area C serve cases are. The final reserve distribution curve is
(Figure 4) is likely to be more probable as a P90 reserve one that depicts geologically reasonable reserve cases
case. Curve P90C gives a Swansons mean of 38 MMB and represents a view of confidence in their occur-
that seems reasonable in the context of the mapped rence. Curve slope provides an easy assessment of the
closure and other undisclosed variations in pay and re- level of confidence associated with estimates.
covery factor. Equally, the P99 of 255 MMB is man-
ageable as a model, having only a 1% probability of
occurring. APPENDIX: THEORETICAL JUSTIFICATION
OF SWANSON RULE

CONCLUSION The average (X) the 30-40-30 rule estimates is given by

Swansons 30-40-30 rule is able to find the mean of X 0.30X90 0.40X50 0.30X10 (2)
modestly skewed distributions. Reserve estimates for
where X represents the point of the distribution X (i.e., distribution
prospects and historical field size distributions com-
of reserves) that is exceeded % of the time. The general form of
monly approximate to modestly skewed distributions equation 2, retaining symmetry but allowing the weighting and per-
(linear trends on log-probability plots), thus the 30-40- centile parameters to vary, is
30 rule is particularly amenable to their study. Al-
though it may be argued that in some modern explo- X(x, d) xX50 d (1 2x)X50 xX50 d (3)

where x is the probability and d is the number of percentage points

difference of the maximum and minimum from the medium (50th
Table 3. Reserves Data Derived from Each Reserve percentile, P50).
The true mean of a lognormally distributed variate is known to
Distribution Curve (RD Curve) in Figure 5*
be (Hastings and Peacock, 1975)
RD Curve P10 P50 P90 P99 MSw
2 r
1 2
X m exp (4)
P50A 7 18 47 100 21
P50B 7 45 280 960 104
P90C 7 20 92 255 38 where m X50 (the median) and r is the standard deviation of
log(X). The derivation of Swansons rule thus proceeds by calculating
*All values are in MMB. where the general form in equation 3 equals the true mean in equa-
tion 4. By definition

1890 Swansons 30-40-30 Rule

 tion 9, and solving for x, yields
P(X X) (5)
100

2 r 1
where is the percentile. Because making a transformation does not 1 2
exp
alter the relative probability
x (10)
50 d 50 d

exp rU1 100 exprU 1
100 2
P(log(X) log(X)) (6)
100
With the choice of d 40 (as in Swansons rule), then 1(0.1)
The mean of the log of the lognormal variate X is log(median), 1.282 1(0.9). Substituting this into equation 10, because
and because (by definition) the log of a lognormal variate is normally r tends to log(1), gives a value of x 0.304 (to 3 decimal places).
distributed, then normalization can proceed. This entails transform- This agrees closely with the 30-40-30 rule proposed by Swanson: x
ing the variate log(X) in equation 6 into a normal variate Z having increases with increasing r, becoming 0.35 at r log(2.864).
mean 0 and variance 1

log(X) log(X50) log(X) log(X50)

P r

r REFERENCES CITED
log(X) log(X50)
P Z r

100 (7) Bain, J. S., 1993, Historical overview of exploration of Tertiary plays
in the U.K. North Sea, in J. R. Parker, ed., Petroleum geology
of northwest Europe: proceedings of the 4th conference: Lon-
Now, from tables of the normal distribution, P(Z 1), notated (1), don, Geological Society, p. 513.
is 0.1587. Capen, E. C., R. V. Clapp, and W. M. Campbell, 1971, Competitive
For the normal distribution, probability P(Z 1) is commonly bidding in high-risk situations: Journal of Petroleum Technol-
notated (1) and is the probability that a random selection of a vari- ogy, v. 23, p. 641651.
able Z, whose distribution is normal having mean 0 and variance 1, David, M. J., 1996, History of hydrocarbon exploration in the Moray
is greater than 1. For example, if the mean height of men in a town Firth, in A. Hurst, H. D. Johnson, S. D. Burley, A. C. Canham,
was 1.8 m with standard deviation 0.2 m (i.e., variance 0.04 m), and D. S. Mackertich, eds., Geology of the Humber Group:
then the variable Z (height 1.8)/0.2 would have mean 0 and Central Graben and Moray Firth, UKCS: London, Geological
Society Special Publication 114, p. 4780.
variance 1, and (1) would be equivalent to the probability that a
Foley, L. M., L. Ball, A. Hurst, J. Davis, and D. Blockley, 1997,
randomly chosen male would have height greater than 2.0 m.
Fuzziness, incompleteness and randomness: classification of un-
Substitution of (1) 0.1587 into equation 7 yields certainty in reservoir appraisal: Petroleum Geoscience, v. 3, p.
203209.

X
X15.87 Hastings, N. A. J., and J. B. Peacock, 1975, Statistical distributions:
r log(X15.87) log(X50) log (8)
50 London, Butterworth, 84 p.
Megill, R. E., 1984, An introduction to risk analysis, 2d ed.: Tulsa,
With r and m now known, the mean in equation 4 is completely Pennwell Publishing, 274 p.
Rose, P. R., 1987, Dealing with risk and uncertainty in exploration:
determined, and equation 7 leads to
how can we improve?: AAPG Bulletin, v. 71, p. 116.
Rose, P. R., 1991, Exploration, economics, risk analysis, and prospect


X X50 exp rU1 100 (9) evaluation: Austin, Texas, Peter R. Rose, Telegraph Explora-
tion, Inc.
Rose, P. R., and R. S. Thompson, 1992, Economics and risk assess-
where 1 is the inverse of , (i.e., {1(a)} a and 1(0.1587) ment, in D. Morton-Thompson and A. H. Woods, eds., Devel-
1). Setting equation 3 equal to equation 4, substituting for equa- opment geology reference manual: AAPG Methods in Explo-
ration Series 10, p. 2156.

Hurst et al. 1891