Professional Documents
Culture Documents
Kalkomey 1997
Kalkomey 1997
T he advance of our ability to generate seismic attrib- resented by that sample. Enthusiastic practitioners can
utes and the growing emphasis on production geo- now generate 10s — even 100s — of seismic attributes.
physics has led to the widespread use of seismic attrib- An all too common practice is identifying which seis-
utes as predictors of reservoir properties. In many mic attributes to use solely on the strength of their
cases, we can show — using seismic modeling or rock observed correlations with reservoir properties mea-
physics — a physically justifiable relationship between sured at the wells. Often these correlations are esti-
a seismic attribute and the reservoir property of inter- mated by a very small number of observations. As with
est. When this is true, we are able to greatly reduce any parameter, when the sample size is small, the
the uncertainty of interwell predictions of reservoir uncertainty about the value of the true correlation can
properties. be quite large.
The simple flow diagram in Figure 1 illustrates the To get an idea of how large this uncertainty can be,
basic process. The first critically important step is accu- consider a sample correlation (r) of 0.80 calculated from
rately tying the well and seismic — both vertically and 10 data points. The 95% confidence interval estimate of
areally. We then must choose which seismic attributes the true correlation is [0.34, 0.95]. The only statement
we believe to be related to the reservoir property. Using we can make with (95%) confidence is that the true cor-
that attribute or set of attributes, we use the dense seis- relation between the reservoir property and seismic
mic data to guide the prediction of reservoir properties attribute is between 0.34 and 0.95. If we have only five
between sparse well data. A number of methods can be data points instead of 10, the 95% confidence interval
used for the prediction step — linear or nonlinear widens to [-0.28, 0.99]. Since this interval contains zero,
regression, geostatistics, or neural networks. The pur- we cannot say with confidence that there is any corre-
pose is estimating in-place hydrocarbon volumes or lation between the reservoir property and the seismic
making reservoir management decisions — such as attribute.
location and number of wells, depletion strategy, or gas Most practitioners are already aware that the small-
and water injection operations. The problem with the er the sample, the greater the uncertainty about the
process is: How do we identify which seismic attribut- true value of the correlation. A lesser known phenom-
es to use? ena is something statisticians call experiment-wise
error rates. This just means that as we generate an ever
The problem. All of the prediction methods (regres- increasing number of attributes, the greater the chance
sion, geostatistics, and neural networks) require mak- of observing at least one spuriously large sample cor-
ing inference from seismic data at a small number of relation. You can get the idea of this if you consider an
wells to a larger population which we assume is rep- experiment where you draw a small number of obser-
vations from two completely unrelated variables —
say the annual birthrate of elephants and pounds con-
sumed of a food substance. If we randomly select five
years and observe the birthrate of elephants and
pounds of sugar consumed, we would probably cor-
rectly infer that there is no relationship between the
two variables. But if we continue to test this for all pos-
sible food types, by chance alone we will likely find at
least one food type that appears, on the basis of the
sample correlation, to be related to the birthrate of ele-
phants. When this happens, we say we’ve encoun-
tered a spurious correlation.
0.4 0.50 0.25 0.14 0.08 0.05 0.02 0.00 0.00 0.00 is representative of the population under study and
0.5 0.39 0.14 0.06 0.02 0.01 0.00 0.00 0.00 0.00 that the amount of independent information available
0.6 0.28 0.07 0.02 0.01 0.00 0.00 0.00 0.00 0.00
0.7 0.19 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00
to estimate the true correlation is maximized. If in fact
0.8 0.10 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 the data are spatially correlated, the relationship gives
0.9 0.04 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 a conservative estimate of the probability of a spurious
correlation. This is because the effective sample size will
be smaller than the actual sample size; and as n
actually uncorrelated. How likely is it that this could decreases, the probability of a spurious correlation
occur? increases. Table 1 shows the probability of observing
a spurious correlation for different magnitudes of the
Probability of observing a spurious correlation. Con- sample correlation (R) and different sample sizes (n).
sider a single seismic attribute as a possible predictor. Now that we’ve quantified the probability of a spu-
The probability of observing the absolute value of the rious correlation when considering a single seismic
sample correlation (r) greater than some constant (R) attribute, we can calculate the probability of at least one
given the true correlation (r) is zero can be found from spurious correlation when considering a set of k inde-
pendent attributes. The probability of observing at least
æ R n-2ö
( )
p sc = Pr r ³ R = Prç t ³ ÷
one spurious correlation when considering a set of k
independent attributes is simply 1 minus the probabili-
è 1- R2 ø
ty that none of the sample correlations are spurious. This
where n is the sample size or number of locations with can be expanded to the summation
Tables 2-5. Probability of observing at least one spurious sample correlation from a
given number of independent seismic attributes.
Table 2. Five independent seismic attributes Table 3. Ten independent seismic attributes
Table 4. Twenty independent seismic attributes Table 5. Forty independent seismic attributes
a) b)
Figure 2. Probability of observing at least one spurious sample correlation of magnitude (a) Æ 0.6, and (b) Æ 0.4
when actually the well data and seismic attributes are uncorrelated.
a) b)
Figure 4. Scatter plot of porosity versus seismic ampli-
tude at the 10 well locations. Figure 5. True spatial distributions. (a) True distribu-
tion of porosity ranging from 0 (dark blue) to 20%
(red). Well locations are indicated are indicated by
black dots. (b) Seismic attribute over the same area.
p
o
r
o
s
i
t
y
a) b)
seismic attribute
Figure 8. Estimation errors. (a) Areas overestimated by
Figure 6. (above) more than five porosity units. (b) Areas underestimat-
Scatter plot of ed by more than five porosity units.
porosity versus
seismic amplitude requires additional assumptions about the true magni-
at all tude of the correlation between the well property and
locations. the seismic attribute.) To quantitatively assess risk, we
would have to assign the cost or economic consequence
of each of the four possible outcomes. The costs will
obviously be situation dependent; however, we can gen-
Figure 7. (right) eralize the cost in a qualitative sense.
Estimate of poros- The cost of a Type I Error (using a seismic attribute
ity generated to predict a reservoir property when actually uncorre-
using collocated lated) is:
cokriging with the • Inaccurate prediction biased by the attribute.
seismic attribute • Inflated confidence in the inaccurate prediction —
as the covariate. apparent prediction errors are small.
associated with choosing seismic attributes to predict a The cost of a Type II Error (rejecting a seismic attribute
reservoir properties. for use in predicting a reservoir property when in fact
A Type I Error occurs if no relationship truly exists they are truly correlated) is:
between the seismic attribute and the reservoir proper- • Less accurate prediction than if we’d used the seismic
ty of interest, yet we select the seismic attribute as a pre- attribute.
dictor. A Type II Error occurs if a physical relationship • Larger prediction errors than if we’d used the attribute.
does exist between the seismic attribute and the reser- I believe that for most cases the economic conse-
voir property of interest, but we fail to use the seismic quences of making highly confident but inaccurate pre-
attribute as a predictor. dictions are more severe than the consequences of a Type
The previous section quantified the probability of a II Error.
Type I Error occurring when the selection criteria is
based solely on the magnitude of the sample correlation A simple example. To illustrate the consequences of a
between a seismic attribute and a reservoir property. Type I Error, consider a simple example. Suppose you
(Calculating the probability of a Type II Error occurring have measurements of porosity and a seismic attribute
zero. By happenstance, the field has been drilled in Secondly, when the probability of a spurious cor-
locations which makes it appear that there is a strong relation is large, then selection of seismic attributes
relationship between this attribute and porosity when based solely on empirical evidence is risky — it can
in fact there is none. We have observed a spurious sam- lead to highly confident, but highly inaccurate predic-
ple correlation. tions and thus poor business decisions. Therefore, it is
If we use this attribute for prediction, our resulting strongly recommended, especially when the number
estimate of porosity will bear a strong resemblance to of wells is small, that only those seismic attributes
the seismic attribute — regardless of whether we used that have a physically justifiable relationship with
regression, geostatistics or neural networks. Figure 7 the reservoir property be considered as candidates for
shows such an estimate, generated using collocated predictors. LE
cokriging.
This prediction results in large actual estimation
Cynthia Kalkomey received a doctorate in statistics from Southern
errors. Figure 8a highlights areas where the porosity is
Methodist University (1991). She joined Mobil in 1979 and is cur-
overestimated by 5-12 porosity units. Figure 8b shows rently manager of Reservoir Characterization, Mobil Exploration and
areas where the porosity is underestimated by 5 -12 Producing Technical Center.
porosity units. As you can imagine, drilling locations
chosen on the basis of this prediction will be less than Corresponding author: Cynthia Kalkomey, Mgr., Reservoir Char-
optimal. This could also be disastrous for other reser- acterization, Mobil Exploration and Producing Technical Center,
voir management decisions. For example, if an injec- 13777 Midway Rd., Dallas, TX 75244; phone 972-851-8598; fax
tion strategy was designed on the basis of this predic- 972-851-8703.