Вы находитесь на странице: 1из 38

1

Well Control
Geological Statistics

CONTENTS

1. INTRODUCTION
2. MEASURES OF CENTRAL TENDENCY
3. MEASURES OF VARIABILITY
4. DISTRIBUTIONS
5. SAMPLE SUFFICIENCY
6. MEASURES OF SPATIAL CORRELATION
7. EXERCISES

1
LEARNING OBJECTIVES:
In this Chapter, we introduce some of the various statistical techniques used in the
analysis of reservoir data. These techniques play an important role in Development
Geology in two areas:
Reservoir characterisation requires quantification of the geological description for
use in reservoir simulators.
Well data are few and far between and statistical techniques are used to populate the
interwell volumes.
The student will learn :
How to average permeability and porosity
How to explore the relationship between permeability measurements and geology
How to relate core and test permeability
How to measure permeability heterogeneity
Construct and evaluate measures of spatial correlation (variograms)
At the end of this Chapter the Petroleum Engineeer will be able to analyse core data
and prepare summary data for mapping or reservoir simulation or for general use in
petroleum engineering. The student should also be able to design a core sampling
programme for porosity and permeability.

Geological Statistics

1. INTRODUCTION
This brief introduction to statistics is designed to give an overview for industry
professionals who find themselves involved in reservoir characterisation. Reservoir
characterisation seeks to define quantitatively the input data needed to undertake
predictions of flow through permeable media (Lake, 1991). There is an obvious need
for hard numerical data and spatial information relating to the petrophysical properties. This information is needed in numerical reservoir simulation.
Before commencing any discussion of statistics, some terminology needs to be clearly
understood. The properties of a reservoir unit for which the geologist or engineer is
required to infer (or estimate or guess) values can be considered a population. This
population may be the entire reservoir (e.g., the Brent Group reservoir in the North
Sea), a subdivision of the reservoir (e.g., the Etive, Rannoch Formations of the Brent
Group) or even a sedimentological entity within the reservoir (e.g., a bedform or
lamina type). In each case, one can estimate the population parameter (e.g., mean,
standard deviation, etc.) by statistical inference from a statistic computed from some
data. In this Chapter we will learn the definitions and how to calculate the simplest
statistical parameters.
The geologist usually estimates the population parameter (e.g., the mean) by an
appropriate descriptive statistic (e.g., an average) from a sample. The sample (in the
statistical, rather than geological sense) can be a small set of measurements (e.g., core
plugs) taken from the reservoir or population of interest. The confidence with which
the sample statistic can be taken as an estimate of the population parameter can also
be quantified, expressing the uncertainty in the prediction of the property.
In the petroleum industry, the samples that are available are generally very small and
not necessarily representative. It is common to infer the parameters for an entire
reservoir (order 108-1010 m3) from a few cores (10- 102 m3) from which limited samples
are taken (10-2-10-3 m3). The wells that are cored may be drilled in unrepresentative
field areas. In addition, only the exploration and appraisal wells (under non-optimum
conditions of interval, recovery or mud chemistry) and the first few development
wells are cored. Lack of representivity may arise from the fact that these wells are
generally located in the crestal areas of the field which, possibly because of variable
diagenesis within the hydrocarbon column, are often not representative of the
reservoir population as a whole. Cores are rarely taken in the aquifer beneath
hydrocarbon accumulations, but aquifer parameter estimation can be as important as
parameters for the reservoir and the aquifer may have a different diagenetic history.
It is important to recognise that the estimates of the core population parameters (i.e.,
average horizontal permeability or porosity) should be based on sufficient samples (in
number and size) taken from that core. If the core properties are poorly estimated, one
can expect the reservoir properties to be poorly modelled. Geologists and engineers
should aim to, at least, provide good estimates of core populations. The more variable
a parameter is, the more samples are required to estimate it - permeability is commonly
very variable and therefore most difficult to estimate.

Department of Petroleum Engineering, Heriot-Watt University

1
A typical analysis of a core data set might include evaluation of:

summary numbers
the variability
how the summary numbers relate to the variability
how good the numbers are
how the variability of a property relates to location in the reservoir.

The next sections in this Chapter provide the tools to undertake such an evaluation.

2. MEASURES OF CENTRAL TENDENCY


The most commonly used descriptive statistics that are determined from a sample are
the measures of central tendency. By central tendency we mean the tendency of the
observations (measurements) in a sample to centre around a particular value rather
than spread themselves across a range of values. When required to produce a set of
summary numbers that describe our available set of variables, the average is the most
easily determined.
Firstly, we need to discuss the types of variables that occur in petroleum engineering.
These are illustrated in Figure 1.

Variables
CATEGORY
Nominal
Ordered
categories

QUANTITY

Ordinal

Discrete
(counting)

Continuous
(measuring)

Ranks

Category variables refer to any variables that involve putting individuals into
categories. These can be:
nominal (e.g., names such as make of bit, formation ) or,
ordinal (e.g., a number to represent degrees of bit wear, or stratigraphic location).
Ordinal data can be ranked (e.g., bits ranked by relative condition of wear, low= 1 to
high = 7) or ordered (e.g., stratigraphic unit 1 is older than stratigraphic unit 2).
Quantity variables have a numerical quantity that can be either discrete (e.g., fossil
specimens, numbers of channels) or continuous (e.g., permeability, porosity, etc.). It
is the latter type of variables that we are most interested in.
Quantity variables can be either extensive or intensive. Extensive variables are
additive (e.g., volume). The volume-weighted average of the porosity for two samples
is the average porosity for the combined volume and a meaningful quantity.

Figure 1
Types of variables (after
Rowntree, 1981)

Geological Statistics

Permeability, on the other hand is an intensive variable, and is generally non-additive.


The average of two plugs, e.g., a shaley core plug (1mD) and a clean sandstone
(1000mD), combined, would depend on how they were combined. Intuitively, the
combined value could be either closer to 1000mD (in the case that flow is parallel
through the two plugs) or closer to 1mD (flow through the two plugs in series). The
average permeability depends on the flow conditions and flow geometry. In
engineering the effective property is the property of an equivalent homogeneous
medium. Permeability averages are often used to estimate the effective permeability.
Whilst more than one measure of central tendency can be obtained for each type of
variable, the numbers obtained may not always be meaningful. We now discuss the
various measures and consider the relative merits of the measures in reservoir
characterisation. In this text, mean is the population parameter for central tendency
and average, the estimator of the population mean calculated by a sample statistic
(such as the arithmetic average).

Arithmetic Average
The arithmetic average of N data is obtained by adding the quantities and dividing by
the number of data in the sample. This is commonly expressed mathematically as:

1 N
kar = ki
N i =1
where k represents permeability.
The arithmetic average is an estimator of the arithmetic mean is equally sensitive to
all values. The arithmetic average (and the other averages) can be biassed. Bias is a
systematic error in the estimator and may occur because of a number of reasons:
sands may be sampled more than other lithologies leading to high values for the
arithmetic average,
it may not be possible to plug very loose sands - often the best reservoir zones leading to a pessimistic estimate of permeability,
thin low permeability baffles (faults, thin siltstones) may not be sampled leading
to optimistic estimates of permeability.

Geometric average
The geometric average is determined as the Nth root of the product of N data and is
usually written as:

N
k
kgeom = i
i=1

1/
N

The geometric average of permeability can also be considered as the exponential of


the arithmetic average of the natural log of permeability. This is easier to compute,
as the product term in the above expression rapidily exceeds the capacity of most
computers. It can be written in this form as:

Department of Petroleum Engineering, Heriot-Watt University

1
1 N

loge (ki)
kgeom = exp N

i=1

The geometric average is indeterminate in the presence of zero data values and this can
cause problems for a sandstone matrix containing shales. Very low values (0.001Md
or less) should be used instead of zeros.

Harmonic average
The harmonic average for N permeability data is given by:
N

khar = N

i=1

-1
1
ki

Like the geometric average, the harmonic average is also indeterminate in the
presence of zero values and low values should be substituted.. All rocks (possibly
except salt) have some permeability , even if the measurement of it is below the
resolution of the device in use.
The inverse of permeability (i.e., k-1) can be considered as resistance to flow. The
harmonic average is therefore the permeability that corresponds to the arithmetic
average resistance to flow. It follows that the harmonic mean is sensitive to low values
of permeability (i.e., large values of k-1). We have also seen that low permeability,
fine grained, material commonly occurs in much thinner layers (e.g., micaceous or
carbonaceous laminae) than high permeability, coarse grained material (e.g., channel
fill sandstones). This tends to result in a systematic (biassed) overestimate of
permeability by the harmonic average (often used to estimate vertical permeability)
due to undersampling of the low permeability layers.

Median
The median of a distribution is equal to the value above and below which equal
numbers of samples lie (k50 see below). The median is most easily determined by
ranking the data and determining the middle value. The median may not be unique
for a discrete distribution. The median is rather insensitive to the tails of the
distribution and, therefore, is often less sensitive to outliers (potential errors). To
illustrate this consider the following two samples of 5 permeability measurements:
A
B

120
120

133
133

210
210

220
220

350mD
1350mD

In each case the median permeability is 210mD but, in sample A, the arithmetic
average is 207mD and in sample B, 407md. In this example, the median could be
considered a more robust estimator of population central tendency.

Geological Statistics

Mode
The mode of a distribution is the most commonly occuring value. A single mode is
not appropriate in be determined in bimodal (i.e., with two peak) or multi-modal
(many peaked) distributions. The arithmetic average is not a meaningful number with
category variables (e.g., average bit wear is 1 or 2; not 1.334 for example). In this case,
the the mode is preferred.

Differences between measures of central tendency


For a perfectly normal distribution all the above measures of central tendency will
overlie (Figure 2). Differences become increasingly marked as the distributions
become more skewed. Skewness is the term used to describe a distribution which is
not symmetrical. In this latter situation, one might ask which average is most
appropriate. Skewed distributions are commonplace in permeability data and
estimating a single average measure may not be appropriate. The appropriate average
will depend on what use is to be made of the estimator..

SYMMETRICAL
DISTRIBUTION

Figure 2
Distibutions of measures of
central tendency

Average
Mode
Median

SKEWED
DISTRIBUTION

Mode
Median
Ar. Av.
Geom. Av.
Har. Av.

Differences in the arithmetic ( kar), geometric ( kgeom), and harmonic ( khar) averages
are a function of the sample heterogeneity, and are commonly observed in permeability datasets. The differences are such that always:
khar kgeom kar
The differences can be exploited for permeability as each average is appropriate for
different flow conditions (refer to Archer and Wall, 1986, for derivation). These
averages are appropriate for the following conditions:
kar
bed parallel, single phase flow (i.e., horizontal flow in a horizontally
layered, bounded system)
khar
bed series, single phase flow (i.e., vertical flow in a horizontally layered,
bounded system)
kgeom
single phase flow in a random, 2-D field
The use of the various averages to estimate mean permeability is appropriate only for
the specific flow conditions described. Often the averages are used as (poor)
Department of Petroleum Engineering, Heriot-Watt University

1
estimators under the wrong flow conditions (e.g., two phase flow, wrong dimensions,
wrong boundary conditions, etc.), so extreme care is needed here to select the
appropriate average. If the medium is homogeneous, the averages will be very similar.
Permeability measurements are often limited by measurement resolution. Frequently
plug permeability data are listed, in core contractors reports, as <0.001mD or
>6000mD or no measurement possible", (NMP). The undetermined low values will
have an affect on the harmonic average (and also the population vertical permeability)
and truncated high values on the arithmetic average (affecting the horizontal permeability estimates). Whilst there are no set procedures for handling this problem, here
are some suggestions:
If the distribution is thought to be fairly symmetrical, the median will prove to be
a better estimate of the population mean.
If the distribution is significantly skewed it is most likely that the extreme missing
values are the ones you really want! In this case, re-measure with a more accurate
device - statistics cant help you.
It is always important to consider the ultimate use in engineering for any statistic. This
information, together with knowledge of the geological structure and the degree of
variability, should indicate the most appropriate average to use in reservoir characterisation. For averaging permeability:
o
in a layered system at low dips (<20 )- use the arithmetic average for horizontal flow
and harmonic avergae for vertical flow,
o
in steeply dipping beds (>70 ), us the arithmetic average for vertical flow and
horizontal flow along the beds. Use the harmonic average for horizontal permeability
across the beds.
in a random system - use the geometric average for vertical and horizontal
permeability.

Instrument bias discussion

1000

100

10

1
1

10

100

1000

Horiz. plug perm. (mD)

Probe har. av. perm. (mD)

Probe ar. av. perm. (mD)

The Rannoch Formation in the North Sea is highly laminated and enables a test of the
averaging procedures. A series of experiments compares the average of a very small
scale measurement (taken with a probe permeameter - a device for measuring
permeability of a rock surface by injecting nitrogen through a very thin nozzle) with
the core plug measurement. The rock is laminated with the flow is parallel to the
lamination. The procedure for layered systems used to average the small scale
permeability measurements and compare the averages with the measured plug results
(Figure 3).
1000

100

10

1
1

10

100

1000

Vertical plug perm. (mD)

Figure 3
Comparison of probe
permeameter estimates of
horizontal and vertical
permeability with
measurements of core plugs

Geological Statistics

The probe permeameter cannot resolve the low permeability thin laminations. As a
result of this instrument bias, the harmonic average is a poor estimator of vertical
permeability. Estimating vertical permeability is often a problem in the industry,
because the thin low permeability layers are under-represented in the sampling or
difficilt to measure. Care should be taken when using the harmonic average.
In contrast, the arithmetic average (because the high permeability layers tend to be
thicker in these sediments) is a reasonable estimator of the horizontal permeability
(Figure 3). This a fairly widespread problem in the measurement of the properties of
rocks - thin beds are under-represented because of volume and spacing considerations. Anisotropy in permeability is usually taken to be the difference between
horizontal and vertical permeability and is estimated by the ratio, kv/kh . Sometimes
there is also anisotopy measured in horizontal permeability in orthogonal directions.
Estimates of anisotropy in sediments depends on the measurement scale. For the
measurement scale , we consider the volume over which a measurement is made or
an average is taken (Figure 4):
-7

At the probe scale (a volume


of 10 m ), kv/kh ratios range between 0.7 and 1.
-5 3
At the core plug scale (10 m ) kv/kh ratios range from 0.2 to 0.4 with lamina-induced
anisotropy in the 2.5cm samples
At the bed scale, it is difficult to take direct measurements of anisotropy but we can
use statistical estimators. Arithmetic averages of the horizontal plugs and harmonic
3
averages of the vertical plugs can be used to estimate bed-scale anisotropy (1m ).
At this scale anisotropy can be 0.01-0.5, the former in more layered and the latter
in the more massive reservoirs.
Gridblock anisotropies in the simulator will also vary with the scale of the
gridblock. In most field simulations, the blocks are larger than the scale of bedding.
In these cases, the anisotropy will be determined by the bedding and vary greatly
from that observed in core plugs. If a single gridblock is used to represent the
Rannoch Formation the anisotropy value will be as low as 0.007.
Knowledge of anisotropy is particularly critical in the performance prediction of
horizontal wells.

Rannoch anisotropy
Grain

Lamina

Bed

Parasequence

Figure 4
Estimates of scale
dependent anisotropy (kv/kh
ratios) in the Rannoch
Formation.WB, SCS and
HCS refer to the rock types
of the Rannoch Formation.
Formation referes to the
whole Rannoch interval.

kv/kh

.1
WB
SCS
HCS
Formation

.01

.001

10 -6

Probe
Plug

10 -4

10 -2

10 0

Sample volume (m3)

10 2

10 4

Plug averages
Probe average

Department of Petroleum Engineering, Heriot-Watt University

1
Comparing well test and core plug average permeabilities.
Well testing provides larger scale in-situ measurements of permeability. Largely
beyond the scope of this Chapter, well testing is defined as a perforated flowing
inteval over which permeability is estimated from the analysis of pressure data whilst
the well is flowing ar once the flow has been stopped. Permeabilities derived from
these measurements are often compared with core for ground truthing either sets of
data.
There are a number of problems or issues to bear in mind when comparing well test
and core plug average;
Scale, flow geometry and boundary conditions of well test and core plug
permeabilities are all different.
Plug data come from within the well, whereas well test data are dominated by a
region a few metres to a few 100metres from the wellbore.
Well tests are in-situ measurements, whereas core plugs are laboratory measure
ments (often at different stress conditions and conducted with different fluids)
However, even with these problems coupled with the uncertainty of the near well
region, the reservoir engineer is often required to make such comparisons.
Well testing tends to be dominated (in clastic reservoirs) by the properties of beds.
Lamina-scale features are usually undetectable or confused with wellbore phenomena. Larger scale features (e.g., formation boundaries) can be detected if the tests are
long enough.. Well test permeabilities are most closely associated with average bed
permeabilities. The appropriate average will therefore depend on the bed geometry
(Figure 5) . A case study is discussed to illustrate the problem.

kar

kgeom

khar

10

10-50ft

5-10ft

1-5ft

Figure 5
Alternative estimators for
well test permeabilities
depend on the geometry of
the geology at the bedscale.Well testing (in the
middle time region) is
essentially a bedform scale
measurement. Well tests
measure effective scale
permeabilities at that scale

Geological Statistics

The accompanying sketch (Figure 5) shows how the averaging procedure might relate
to bed scale and geometry in an imaginary reservoir. In many reservoirs, single
isolated high permeabilities can be filtered out as they represent lamina (i.e., very
local) properties.
Two wells in the same braided stream reservoir have similar permeability distributions (Fiure 6). The well test permeabilities are quite different in the two wells and
correspond to the geometric average (Well A) or the (weighted) arithmetic average
(Well B). Looking at the spatial distribution of the permeability (Figure 7) we see the
reason why. Well A is characterised by isolated high permeability plug most likely
corresponding to thin, minor fluvial channels of limited extent. Well B on the other
hand, contains relatively thick, major channels of signifcant extent. The lower
permeability matrix dominates the well test volume of investigation in well A
(geometric average more appropriate for random discontinuous high permeability
zones) and the channels, well B (arithmetic average for layer flow). This example
shows the pitfalls of applying statistics without geology - one can always determine
the arithmetic average - but using it as the estimator of horizontal permeability is
sometimes dangerous! In well B it was appropriate because it is a layered system. In
well B it would have overestimated the permeability as the layers are not continuous.

Figure 6
Permeability distribution
from two wells in the same
braided stream reservoir
left: Well (A) right: Well
(B)

Figure 7
Permeability distribution
with depth in the two wells
in Figure 6. left: Well (A)
right: Well (B)

Department of Petroleum Engineering, Heriot-Watt University

11

1
3. MEASURES OF VARIABILITY
In the previous section we reviewed measures of central tendency. The second class
of descriptive statistics that can be used to describe a sample are measures of
variability or dispersion. These are commonly used in other areas of data analysis but
tend, traditionally, to be overlooked in petroleum engineering (particularly by
geologists). As we will see in this and subsequent sections, the measures of variability
of permeability can:

define the level of heterogeneity


determine the number of samples required
indicate likely recovery process
suggest likely flow performance

Because of these reasons, it is considered that measures of variability can be equally


(if not more) useful than averages. Every estimate of central tendency (of permeability) should be accompanied by a measure of variability. The most commonly occuring
measures of variability are reviewed in this section.

Standard deviation
In statistics, a deviation is a distance from the mean. The mean deviation is thus the
average deviation for a sample. The variance is the average squared deviation. The
standard deviation (SD) is given as the positive square root of variance:

SD =

i=1

(ki - k)2
N

0.5

SD =

i=1

k2i
N

0.5

k2

Standard deviation has the units of the measurement being considered (e.g., mD in the
case of peremeability). The lower the standard deviation the less the dispersion or
spread of a distribution about the mean. 68% of all the observations in a normal
distribution lie within one standard deviation (SD) either side of the mean (2SD and
3SD include 95% and 99.7% of the observations, respectively). For small sample
sizes the N in the denominator should be replaced by (N-1) to account for fewer
degrees of freedom caused by the less than optimum estimation of the mean.

Coefficient of variation
The standard deviation often tends to increase as the mean increases. An SD of 80mD
is a high dispersion for a mean of 100mD, but a low dispersion if the mean is 1000mD.
A more useful (in reservoir characterisation) absolute measure of dispersion is given
by the coefficient of variation (Cv), or normalised standard deviation:
Cv = SD /kar

12

Geological Statistics

For small samples (N < 10), the standard deviation needs to be multiplied by a
correction factor. The correction factor accounts for the fact that tha mean is not well
estimated with few samples, and therefore the SD is broader than if there were many
samples and is given by:

1+ 1
4(N-1)
The coefficient of variation is becoming more widely encountered in reservoir
description, particularly in probe permeameter studies (Figure 8), and has been used
to define the level of heterogeneity:
0.0 < Cv < 0.5 Homogeneous
0.5 < Cv < 1.0 Heterogeneous
1.0 < Cv
Very heterogeneous

Figure 8
Coefficient of variation for
a range of geological
materials.(from Jensen et
al, 1997)

Carbonate (mix pore type) (4)


S.North Sea Rotliegendes Fm (6)
Crevasse splay sst (5)
Sh. mar.rippled micaceous sst
Fluv lateral accretion sst (5)
Dist/tidal channel Etive ssts
Beach/stacked tidal Etive Fm.
Heterolithic channel fill
Shallow marine HCS
Shall. mar. high contrast lam.
Shallow mar. Lochaline Sst (3)
Shallow marine Rannoch Fm
Aeolian interdune (1)
Shallow marine SCS
Lrge scale x-bed dist chan (5)
Mix'd aeol. wind rip/grainf.(1)
Fluvial trough-cross beds (5)
Fluvial trough-cross beds (2)
Shallow mar. low contrast lam.
Aeolian grainflow (1)
Aeolian wind ripple (1)
Homogeneous core plugs
Synthetic core plugs
Cv 0

INCREASING SORTING

Normal distributions have Cv < 0.5, for Cv > 0.5 the distributions become increasingly
skewed. Even under the latter conditions, the Cv appears to be a useful statistic.

Very heterogeneous

Heterogeneous

Homogeneous
1

Department of Petroleum Engineering, Heriot-Watt University

13

1
Dykstra-Parsons coefficient
A further measure of variability, developed by the oil industry, recognises that
permeability is often log normally distributed. For permeability that is log-normally
distributed, the Dykstra-Parsons coefficient was defined in the 1950s as:

VDP = 1 -

k
k0.5

k0.5 k
k0.5

where k is the permeability one standard deviation below the median permeability
(k0.5) for a distribution of the logarithm (usually base 10) of permeability. These
parameters are best determined by plotting a probability plot for log(k) and reading off
the 50 and 84th percentiles. This graphical procedure for the determination of VDP (for
which probability paper is required) is illustrated in Figure 9. VDP is useful because
of correlations with waterflood performance,enhanced oil recovery (EOR) potential
and its common occurence in the petroleum engineering literature. Note that estimates
of VDP for distributions that are not log normal can be unrepresentative.

Figure 9
Graphical solution of the
Dykstra-Parsons coefficient
(from Willhite, 1986)

Lorenz coefficient
A third way of expressing heterogeneity in reservoirs is through the use of the Lorenz
Coefficient. The Lorenz Coefficient is defined as a specific area on a Lorenz Plot, a
useful plot which involves the relationship between of core plug porosity () and
permeability(k) as a fraction of total reservoir porosity and permeability. The Lorenz
plot is modelled on a mutch earlier plot (early 1900's) that was used to relate wealth
(flow capacity) to population (storage capacity) - wealth is not evenly distributed
across people, neither is flow capacity in rocks!
To generate the Lorenz Plot, arrange the core data in decreasing order of k/ and
calculate the partial sums

14

Geological Statistics

Fj

Cj

J
j =1

k j hj

J
j =1

kh

i =1 i i

j hj

i =1 i i

where 1 J I and there are I data points. Plot Fj (known as the flow capacity) versus
Cj (the storage capacity) on a linear plot (Fig. 10). The Lc can be calculated from the
shaded area and varies from 0 - 1. Homogeneity is expressed on this plot by the
diagonal (Lc= 0).

Fraction of total
flow capacity
(permeability x
thickness)
Figure 10
Lorenz plot showing
determination of the Lorenz
Coefficient, Lc= 2A where A
is the shaded area

Fj

Cj

0
0

Fraction of total
storage capacity
(porosity x thickness)

The plot can also be useful in well testing as it represents the balance between flow
capacity (or tranmissivity) and storage capacity (storativity) and can indicate the
possibility for double porosity behaviour. In a fractured reservoir, the transmissive
elements (the fractures) are often separate from the storage elements (the matrix) these would have Lorenz Coefficients close to 1.
A modification to the Lorenz Plot is to plot the data with their natural ordering. The
Modified Lorenz Plot ( Figure 11) can then give a good idea on the degree of layering
in the reservoir. The modified Lorenz Plot can be compared with the inflow log
measured during production logging.

Department of Petroleum Engineering, Heriot-Watt University

15

1
1

Fraction of total
flow capacity
(permeability x
thickness)

Fj
Figure 11
Modified Lorenz Plot
showing stratigraphic
layering.

Cj

0
0

Fraction of total
storage capacity
(porosity x thickness)

4. DISTRIBUTIONS

Frequency

A distribution is a graphical representation of a set of frequencies (observed distribution) or probabilities (theoretical distribution). Frequencies are presented on a bar
chart (histogram) in which the width of the bars are proportional to the class interval
and the height of the bars is proportional to the frequency it represents (Figure 12). The
class interval is the interval between boundaries selected to subdivide the range into
a number of (usually equal) windows. Points falling at the boundaries are
systematically included in the class interval below or above. As a guideline, the
number of class intervals should be between 7 and 25.

Variable

Probability is a measure of the relative frequency of occurrence of an event.


Probability (P) is a number between 0 and 1. Probability 0 means impossibility, 1 is
certainty. Values can be derived from a theoretical distribution or by observation.
For a discrete distribution, probability is defined as:
16

Figure 12
Simple histograms

Geological Statistics

number of required outcomes


total number of possible outcomes
13

Thus the probability of picking a spade from a pack of cards is 52 or 0.25

Pdf

For a continuous variable, the probability is the relevant area under the graph of its
probability density function (pdf). The total area under the graph is 1, i.e., a random
variable will lie within the range of its pdf. The pdfs for the variables in the sample
histograms above can be derived as the sample size approaches infinity and the class
interval approaches zero (Figure 13).

Figure 13
Probability distribution
functions underlying the
sample histograms

Variable

If there are sufficient observations in the sample, the sample histogram can be thought
of as an estimate (or approximation) to the underlying variable pdf. For this reason,
sample histograms are often referred to as pdfs (strictly, pdf is a population
parameter) (Figure 6).
The function that gives the cumulative probability or cumulative frequency (i.e., the
frequency with which a varible has a value less than or equal to a particular value) of
the random variable is known as the cumulative distribution function (cdf) (Figure
14).

Figure 14
Cumulative distribution
functions associated with
the above pdfs

Cdf

1
.5

k50

Variable

Department of Petroleum Engineering, Heriot-Watt University

k50

17

1
Cdfs are the form of distributions that are commonly used as the input to Monte Carlo
simulation. Random numbers between 0 and 1 are used to derive a number of
realisations of the variable cdf (e.g., for porosity, volume, shale length, channel width,
etc.). The pdf of the random variable will, with enough realisations, assume the
sample pdf.
There are major benefits in identifying the form of the underlying pdf:
the pdf is the only statistical parameter that defines the extreme values and the
probability of their occurrence.
non-normal distributions can be transformed to normality if the underlying distri
bution is known. Parametric methods are appropriate and regression is enhanced
for normally distributed variables.
parametric (i.e., sensitive to the underlying distribution) statistical tests are more
powerful. Procedures where we dont know the form of the pdf are called nonparametric.
Distibutions that are not symmetrical are known as skewed. Consider the two pdfs
in Figure 15, the one on the left is symmetrical whereas the one on the right is
positively skewed (i.e., tail - queue in French - to the positive side of the mode).
There are a set of power (p) transformations for 1 > p > -1 which will transform
P
skewed distributions to normality. For k where p = 1, the distribution is already
normal, for p = 0.5, root normal and for p = 0, log normal. These three distributions
are common for permeability within reservoir rocks.
There are other distributions more specific to small sample probabilities (e.g.,
binomial, poisson) or truncated data sets (e.g., exponential, gamma) which will not be
covered here.

5. SAMPLE SUFFICIENCY
The issue of sample sufficiency is not usually covered in basic statistical texts or even
considered in petroleum engineering. Core plugs, for example, are taken every foot,
regardless - because thats the way it has always been done! In fairness to the core
contractors, geologists and engineers, this has, historically, been the practical (in
terms of cost, core preservation, etc.) sample limit. More recently, we have considered
the appropriate number of measurements for our needs and the representatitivity of
the sample population
With the development of probe permeameters, we are able to reconsider sample
sufficiency and, because probe measurements are relatively cheap and non-destructive, ensure that sufficient samples for our requirements are obtained.
A practical method for determining sample sufficiency comes from the central limit
theorem which states that, if independent samples of size N are drawn from a parent
18

Geological Statistics

population with mean and standard deviation SD, then the distribution of their
means will be approximately normal (regardless of the population pdf) with mean
and standard deviation, SD/N. From this, the probability that the sample mean ( ks)
of N observations lies within a certain range of the population mean () can be
determined for a given confidence interval. For a 95% confidence level (i.e., only a
maximum of five times in 100 will the population mean lie outside that range) the
range is given by t SE, where the standard error (SE) is given by SD/N. ( The
greater the sample number, N, the more confident we can be about estimates of the
mean).
Standard error is the standard deviation (SD) of the sample mean, drawn from a parent
population, and is a measure of the difference between sample ( ks) and population ()
means. Students t is a measure of the difference between estimated mean, for a single
sample, and the population mean, normalised by the SE. For normal distributions the
t value varies with size of sample (Ns) and confidence level (95%), and these values
are well known (standard t tables in any basic statistics text). We can write the above,
mathematically, as:

Prob (ks = t

SD

) = 95%

Ns

We require a sample such that the average of the optimum sample size k0 with
tolerance P%, satisfies the predetermined confidence interval, or:

Prob (k0 =

P k0
) = 95%
100

When this condition is satisfied, Ns= N0 , and:

P k0
SD
=
t

100
N0 .
Rearranging this gives an expression for the optimum number of specimens, N0:

t SD 100 2
N0 =

P k0
Now, for N > 30, t ~ 2 and with a 20% tolerance (i.e., the sample mean will be within
20% of the parent mean, which we consider an acceptable limitation), the expression
reduces to:

N0 =

2 Cv 100 2
20

where Cv = SD /k0

N0 = ( 10 Cv)

Department of Petroleum Engineering, Heriot-Watt University

19

1
This rule of thumb is a very simple way of determining sample sufficiency. Although
derived for the estimate of the arithmetic mean from uncorrelated samples by normal
theory, it has been found to be useful in designing sample programs in a range of core
and outcrop studies. Having determined the optimum number of samples (N0), the
domain length (D) will determine the optimum sample spacing (D0) as:
D0 = D / N0
An initial sample of 25 measurements, evenly spaced over the domain, which can be
a lamina, bedform, formation, outcrop, etc. can be used to determine an initial estimate
of variability. If the Cv, determined from this sample, is less than 0.5, sufficient
samples have been collected. If more are required, infilling the original with 1, 2 or
n samples, will give 50, 75 or 25n samples. In this way, sufficient samples can be
collected . When insufficient in very variable rock types (e.g., carbonates) the
tolerance may have to be increased to reach achievable sample numbers). Note that
the N-zero technique has really only been tested for permeability sufficiency.
The answer to the original question, posed in the opening paragraph of this section,
can now be explained. Because formations contain facies of differing variability,
some facies will be adequately sampled with 1ft core plugs, but some thin, highly
variable and, possibly, significant facies can be under-sampled. This happens in the
Rannoch Formation (Middle Jurassic Brent Group, North Sea) where the critical
facies at the Rannoch/Etive boundary in some wells is only 10ft thick with Cv = 1.
Over 100 samples, therefore, are needed in such an interval and 10 core plugs are
obviously insufficient (Figure 15).
When insufficient samples are known the sample tolerance(Ps) can be calculated:

Ps = 200 Cv
Ns

Core depth (ft)

-10050
-10051

Core Plugs

-10052

Probe
Arith av

-10053
PLUGS: N=9
Cv = 0.74, Ar. Av. = 390mD

-10054
-10055

PROBE: N = 274,
Cv = 0.99, Ar. av. = 172mD

-10056
-10057
-10058
10

20

100
Permeability (mD)

1000

Figure 15
Highly variable Rannoch
(M. Jur.) interval showing
the concept of optimum
sample density. N0 for this
interval is 100 - satisfied by
the probe but inadequately
measured with core plugs.
(From Jensen et al 1997)

Geological Statistics

6. MEASURES OF SPATIAL CORRELATION


We have encountered, in this geoscience course, correlation in the geological sense
of drawing lines of equal stratigraphic significance between wireline logs. In this
section we consider spatial correlation (i.e., autocorrelation, correlation of a variable
with itself as a function of separation in space or time). In reservoir engineering, two
autocorrelation functions, the correlogram and the semivariogram (or variogram), are
commonly encountered (Figure 16) and are best defined by their equations below.
The former tends to be used to measure the degree of similarity between neighbouring
values (e.g, grid blocks in a numerical simulator) and the latter to examine the
differences between neighbouring values (e.g., permeability in outcrop or core
studies). The latter is also used in a mapping procedure known as kriging which has
been adopted from the mining industry and has been used (with some success) in the
petroleum industry.
The correlogram function (r) is given by:

(h) =

1
[(k(x) -k)(k(x+h) -k)]
(N-h) (SD)2

where k(x) and k(x+h) are the permeabilities of any two points seperated by lag (the
distance between sample points, h) and N is the number of data points. As h tends to
zero the correlation function tends to unity. A plot of the function against lag is the
correlogram.

CORRELOGRAM

SEMIVARIOGRAM

Figure 16
Characteristic shapes of
autocorrelation funcions in
the presence of correlation

Lag distance

For comparison, the semivariogram function (g, referred hereafter and most commonly as the variogram, the "semi-" is there because the function is symmetrical
about the origin and only positive lag distances are shown by convention) is given by:
(h)=

1
[k(x) k(x + h)]2

2N

at a lag distance h. In this case, N is the number of pairs of data points. As h approaches
zero the variogram (i.e., variance) approaches zero. Note that the variogram doesnt
require an estimate of the mean and is, therefore, more precise than the correlogram.
Department of Petroleum Engineering, Heriot-Watt University

21

80

80

60

60

SILL

40

40

RANGE

Semivariogram (meas. units^2)

IDEAL SEMIVARIOGRAM

20

20

NUGGET
0

0
0

Lag (distance units)

Figure 17
Variogram terminology

The variogram has some additional features (Figure 17). At some lag separation
(known as the range) the variogram often approaches the variance of the data (the sill)
and the statistical correlation between points at this and greater separation(s) is zero.
If the variogram at the closest separation is away from the origin, a nugget is said to
exist, often indicative of measurement inaccuracy. If the variogram at the closest
separation approaches the sill, the data are said to be uncorrelated (Figure 18, right).
On a correlogram, uncorrelated data show the correlation function at or near zero from
the shortest separation (Figure 18, left).

CORRELOGRAM

SEMIVARIOGRAM

Lag distance

It is important to determine the correlation in a data set, as correlation effectively


reduces the amount of "information" carried by each observation. This might result
in either additional samples being required. There is a paradox here, because we have
seen (section on sample sufficiency) that N0 samples (derived for uncorrelated
samples) can give appropriate estimates of mean properties, even though permeability
measurements can be seen to be correlated. Although the reason for this paradox is
not clear at the present time, it can be demonstrated that correlation in sedimentary
rocks exists at several scales. These scales are marked by significant decreases of the
variogram at some positive lag distance (holes).

22

Figure 18
Characteristic shapes of
autocorrelation functions
for random (i.e.,
uncorrelated) samples

Geological Statistics

Core plug kh
Minipermeameter
-10136
-10137
-10138
-10139
-10140
-10141
-10142
-10143
-10144
-10145

Figure 19
Probe (Minipermeameter)
and core plug profiles from
the Rannoch Fm North Sea

-10146
10
100
1000
Core 1
depth
Permeability (mD)
(ft)

Semivariogram magnitude

1.5

Figure 20
Variogram for probe data
shown in Figure 19

1.0

0.5

0.0
0

Lag (m)
The semivariogram can sometimes reveal average periodicities that are represented
by a significant reduction in variance at some lag separation greater than the range. An
example interval from the Rannoch formation shows well developed cyclicity (fig 20)
and this is captured as a hole at approx. 4ft. (1.3m) in the accompanying variogram
(Figure 20). This periodicity is related to the (hummocky cross-stratified) bedform
thickness. There is evidence that the periodicity in the sediment can impact fluid flow
and that the holes can therefore be used as a diagnostic engineering tool. This decrease
in variance at certain separations reflects increased correlation and corresponds to the
wavelength of a lamina or bedform.
Department of Petroleum Engineering, Heriot-Watt University

23

1
It can also be seen in Figure 21 that each of these scales requires a tailor-made
sampling plan, which may require more than N0 samples.

Figure 21
Multiple correlation scales
in sedimentary rocks, as
shown by the variograms
(from Jensen et al 1997)

It has been oberved that many natural phenomena exhibit variograms that never attain
a sill, suggesting correlation on a never-ending scale (Figure 22). Phenomena which
exhibit this phenomena are commonly known as fractals. Fractals exhibit variation
at all scales, the closer you look the more you see. Variograms in a fractal medium
should, therefore, exhibit the characteristic shape at all scales. Fractal behaviour over
limited length scales has been observed in rocks.

24

Lag distance

Figure 22
Fractal behaviour in a
variogram

Geological Statistics

Use of variograms in reservoir modelling


The main use of variograms in reservoir modelling is in the automated generation of
permeability fields for simulation. A field in this sense is a grid of permeability values.
With a few parameters (nugget, sill, range) and the variogram model can be used to
many equiprobable realisations from the same permeability distribution. These fields
will all look different, but will have the same correlation structure as the input
variogram. The input variograms can be generated from outcrop studies. These fields
are called correlated random field (CRF). The CRF technique involves two steps:
(i)

Randomly distributing permeabilities to all grid nodes.

(ii) Swapping the permeabilities around until their spatial correlation structure
conforms with the given variogram structure.
The generation of many equiprobable permeability fields with millions of cells would
be very difficult manually! The are equiprobable because they are all equally likely
to represent reality. An example of correlated random fields for various isotropic
fields is given in Figures 23.

Figure 23
Correlated random fields
generated from isotropic
variograms (Yuan and
Strobl, 1991)

Department of Petroleum Engineering, Heriot-Watt University

25

Figure 24
Correlated random fields
generated from anisotropic
variograms (Yuan and
Strobl, 1991)

SUMMARY
Presented with some porosity and permeability data, the engineer should be able
(using the material in this section) to carry out the following:
generate a suite of summary numbers - averages, mode, median, SD, CV - and
discuss their engineering implications,
evaluate the variability in the data and decide the engineering and sampling
implications,
assess how the summary numbers relate to the varianbility,
determine how reliable the summary numbers are, and,
determine the correlation structure of the property in the reservoir.
The exercises at the end of this Chapter allow the student to try out these skills.

26

Geological Statistics

EXERCISES
1 Determine the arithmetic, geometric, harmonic averages, mode and median for the
porosity and permeability in following set of core plugs, from an North Sea well. To
help determine the geometric and harmonic averages you may wish to determine
ln(perm) and the inverse of permeability first.

Depth

Porosity

Horiz. Perm - k.
(mD)

(m)

(%)

3834.9

23.8

105

3835.2

24.8

140

3835.5

27.4

297

3835.8

26.4

236

3836.1

23.6

106

3836.4

24.6

140

3836.7

24.2

157

3837.0

24.8

144

3837.3

26.0

189

3837.6

24.8

111

3837.9

29.4

577

3838.2

27.6

318

3838.5

14.6

nmp

3838.8

22.5

52.4

3839.1

22.1

54

nmp = no measurement possible

Department of Petroleum Engineering, Heriot-Watt University

27

1
2. If asked to provide an average of these data:
for reservoir simulation grid block, or,
for comparison with a well test?
What would you need to consider? The variation of data with depth is given below.

-3834

Core depth (m)

-3835
-3836
-3837
-3838
-3839
-3840
10

15

20
Porosity (%)

25

30

-3834

Core depth (m)

-3835
-3836
-3837
-3838
-3839
-3840
0

100
200
300 400 500
Horizontal Permeability (mD)

600

3. Plot histograms of permeability and porosity. Plot a Lorenz Plot. What can you
determine from these plots?

28

Geological Statistics

4. From the plots of porosity and permeability vs depth given in Question 2,what can
you say about variability of porosity and permeability?
Calculate the coefficient of variation for both porosity and permeability.
For comparison, the Vdp for the plug permeability is given as 0.46

5. Do we have enough plugs to estimate the arithmetic mean permeability of this


interval within 20%? To what tolerance can we estimate the arithmetic mean from
these data?
6. Over this same 4m interval we also have probe permeameter measurements.
These are plotted vs depth and displayed for comparison with the core plugs. Can
you see the permeability (geological) structure more clearly?

-3835

Depth (m)

-3836

PROBE

-3837

PLUGS

-3838

-3839
0

100

200

300

400

500

600

Permeability (mD)

Given the following analysis of the probe data, can we expect an average within 20%
of the arithmetic average? What is the true heterogeneity level of this interval and
what do you think controls it? What are the implications for the grid block
permeabilities?

Department of Petroleum Engineering, Heriot-Watt University

29

1
Probe permeability
No. of meas.

320

Arith. av. (mD)

151

Geom. av. (mD)

93

Harm. av. (mD)

39

Cv
N0

0.82

Vdp

0.69

7.Determine the (semi)variogram function for the core plug permeability and
compare with that for the probe permeability. What is your interpretation of the two
variograms and of the differences between them?

2.0

1.5

1.0

0.5

0.0
0

20

40 60

8 0 100 120 140 160 180 200

(NB: You can plot your variogram on the above plot by dividing the function by the
variance. In this way you can compare variograms structure from various data sets.
If a sill is present it should appear at the value one)

CORE 5
COARSE GRID
3835.0-3839.0m

Dim. exper.semi-var.
(md^2/Variance)

2.0

1.5

VARIANCE
15831(md^2)

1.0

0.5
dz = 1cm
0.0
0

20

40

60

8 0 100 120 140 160 180 200


Lag (cm)

30

Geological Statistics

SOLUTIONS
1.
N
Arith. Av
Geo. Av
Har. Av
Mode
Median

Por
15
24.44
24.18
23.86
24.80
24.80

Perm
14
188
153
126
140
142

2.
To consider which average is the right one - for porosity it doesnt matter for
permeability there are differences which may be significant. A simulation grid block
may be several 100m long so the possible layering of the interval would have to be
taken into consideration (i.e., the scale of application).
If you look at the variation with depth plots, there is no strong evidence for layering,
so one might choose the geometric average (appropriate for random geology). If you
think some of the data are outliers the mode or median might be a better estimate.
3. Histograms
The histogram for porosity shows a more symmetrical distribution than that for
permeability.

Histograms

Count

Count

4
3

4
3

1
0

0
12 16 18 20 22 24 26 28 30

Porosity

Department of Petroleum Engineering, Heriot-Watt University

100 200 300 400 500 600


Permeability

31

32

k(mD)
105
140
297
236
106
140
157
144
189
111
577
318
52
54

Totalk
2626

Phi (%)
23.80
24.80
27.40
26.40
23.60
24.60
24.20
24.80
26.00
24.80
29.40
27.60
22.50
22.10

To tal Phi
352

Unordered

k/phi
4
6
11
9
4
6
6
6
7
4
20
12
2
2

Phi (%)
29.40
27.60
27.40
26.40
26.00
24.20
24.80
24.80
24.60
24.80
23.60
23.80
22.10
22.50
k(mD)
577
318
297
236
189
157
144
140
140
111
106
105
54
52

Ordered

k/phi
20
12
11
9
7
6
6
6
6
4
4
4
2
2
Phi/t otal Phi
0.08
0.08
0.08
0.08
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.07
0.06
0.06
k/Total k
0.22
0.12
0.11
0.09
0.07
0.06
0.05
0.05
0.05
0.04
0.04
0.04
0.02
0.02

Cumphi
0.00
0.08
0.16
0.24
0.31
0.39
0.46
0.53
0.60
0.67
0.74
0.81
0.87
0.94
1.00
Cumk
0.00
0.22
0.34
0.45
0.54
0.62
0.68
0.73
0.78
0.84
0.88
0.92
0.96
0.98
1.00

Lorenz plot calculation

k(mD)
105
140
297
236
106
140
157
144
189
111
577
318
52
54
Totalk
2626

Phi (%)
23.80
24.80
27.40
26.40
23.60
24.60
24.20
24.80
26.00
24.80
29.40
27.60
22.50
22.10

Total Phi
352

Unordered

k/phi
4
6
11
9
4
6
6
6
7
4
20
12
2
2
Phi/total Phi
0.07
0.07
0.08
0.08
0.07
0.07
0.07
0.07
0.07
0.07
0.08
0.08
0.06
0.06
k/Total k
0.04
0.05
0.11
0.09
0.04
0.05
0.06
0.05
0.07
0.04
0.22
0.12
0.02
0.02

Cumphi
0.00
0.07
0.14
0.22
0.29
0.36
0.43
0.50
0.57
0.64
0.71
0.79
0.87
0.94
1.00

Cumk
0.00
0.04
0.09
0.21
0.30
0.34
0.39
0.45
0.50
0.58
0.62
0.84
0.96
0.98
1.00

Geological Statistics

Department of Petroleum Engineering, Heriot-Watt University

Modified Lorenz plot


calculation

33

1
Lorenz Plot
1.00
0.90
0.80
0.70

kh

0.60
0.50
0.40
0.30
0.20
0.10
Ordered Plug Data
0.00
0.00

0.20

0.40

0.60

0.80

1.00

Phih

Modified Lorenz Plot


1.00
0.90
0.80
0.70

kh

0.60
0.50
0.40
0.30
0.20
0.10
Ordered Plug Data
0.00
0.00

0.20

0.40

0.60

0.80

1.00

Phih

The Lorenz Plot shows moderate heterogeneity, the unordered data (stratigraphically
ordered data) shows the occurrence of a high permeability zone.
4. They way the scales are drawn the plots of porosity and permeability appear to have
the same variability. This is because the scales relative to the mean are quite different

34

Lorenz Plot
(A) ordered
(B) unordered

Geological Statistics

sd (STDEV; n-1)
Variance
Cv

Por
3.35
11.23
0.14

Perm
138
18907
0.73

The Cv is a normalised variability and shows the permeability variability to be much


greater (heterogeneous) than the porosity (homogeneous). A Vdp of 0.46 for
permeability also suggests moderate heterogeneity.
5
Using N0 = (10Cv)2 the optimum number of samples (N0) = (7.3)2 53.
There are 14 plugs so the answer is "no".
The 14 plugs will estimate to

200 Cv 200 0.73


=
Ns
14
i.e. 39%
6
The probe permeameter picks out more distinct layering and low permeabilities in the
region where the core plugs showed no measurement possible (3838.5m). The
variation can be related to the geological structure.
No = 67
P =9
Number of samples (320) is well in excess of the No (67) therefore the probe data
average will be within the 20% tolerance. In fact the average will be within 9%
tolerance. This suggests the arithmetic average 151mD or geometric average 93mD
might be more appropriate. Note that the variability has increased with the probe data
(this often, but not always, happens) and the differences between the averages has
become more pronounced than determined by the core plugs. The extra heterogeneity
is due to the geological structure, some of which wasnt originally sampled by the plugs.
7
Semivariogram calculation
Note the effect the missing data point (nmp) has on the numbers of pairs determined.
The normalised gamma is determined by dividing the gamma by the variance of the
permeability (refer to Question 4).

Department of Petroleum Engineering, Heriot-Watt University

35

1
Permeability

Lag 1

Lag 2

L ag 3

Lag 4

Lag 5

Lag 6

105
140
297
236
106
140
157
144
189
111
577
318

1225
24649
3721
16900
1156
289
169
2025
6084
217156
67081

36864
9216
36481
9216
2601
16
1024
1089
150544
42849

17161
1156
24649
6241
1444
2401
2116
187489
16641

1
0
19600
8464
6889
841
176400
30276

1225
289
23409
2209
25
190969
25921

2704
16
11664
15625
221841
31684

52.4
54

70543
2.56

275205
69696

3434
273529

18660
3249

8391
18225

Count
Sum
Gamma
No rmalised
gamma

12
340458
14186
0.75

11
360443
16384
0.87

11
604199
27464
1.45

10
519434
25972
1.37

9
265956
14775
0.78

8
310150
19384
1.03

Porosity

Lag 1

Lag 2

Lag 3

Lag 4

Lag 5

Lag 6

23.8
24.8
27.4
26.4
23.6
24.6
24.2
24.8
26.0
24.8
29.4
27.6
14.6
22.5
22.1

1.0
6.8
1.0
7.8
1.0
0.2
0.4
1.4
1.4
21.2
3.2
169.0
62.4
0.2

13.0
2.6
14.4
3.2
0.4
0.0
3.2
0.0
11.6
7.8
219.0
26.0
56.3

6.8
1.4
7.8
4.8
1.4
2.0
0.4
21.2
2.6
104.0
47.6
30.3

0.0
0.0
10.2
2.6
5.8
0.0
27.0
7.8
130.0
5.3
53.3

0.6
0.4
6.8
0.2
1.4
23.0
11.6
104.0
12.3
7.3

0.2
0.0
2.0
2.6
33.6
9.0
92.2
5.3
15.2

14
277.0
9.9
0.88

13
357.5
13.8
1.22

12
230.3
9.6
0.85

11
242.1
11.0
0.98

10
167.5
8.4
0.75

9
160.0
8.9
0.79

Count
Sum
Gamma
Normalised
gamma

These results are plotted below:

36

Geological Statistics

The permeability data show a very high nugget (0.65) suggesting that, whilst there is
some correlation, the plug data may be interpreted as uncorrelated random.
Note that the porosity semivariogram is not the same as the permeability variogram
however the same interpretation might be made.
Comparison with the probe data shows the effect of the closely spaced data. The
nugget has dropped to 0.25 and there is a clear hole, indicating a repetitative structure
(bedding). This suggests the interval is more layered, and that the arithmetic average of
the probe data (151mD) might be more appropriate. Note that this is also close to the
geometric average of the plug data (153mD).

Department of Petroleum Engineering, Heriot-Watt University

37

1
REFERENCES AND BIBLIOGRAPHY
Archer, J. S., and Wall, C.G., 1986, Petroleum Engineering: Principles and Practice,
Graham and Trotman, Newcastle, 362p.
Jensen, J.L., Lake, L.W., Corbett, P.W.M., and Goggin, D.J., 1997, Statistics for
Petroleum Engineers and Geoscientists, Prentice-Hall, NJ, 390p
Journal, A. G., and Huijbregts, C. J., 1978: Mining Geostatistics : Academic Press,
London, 600p.
Lake, L. W., 1989, Reservoir Characterisation, vols 1 and 2, SPE Reprint No 27,
SPE,
Linville, W., 1993, Reservoir Characterisation III, Penn Well Publishing Company
1008p.
Willhite, G.P., 1986, Waterflooding, SPE Textbook Series Volume 3, Richardson
Tx. 326p.
Porkess, R., 1988, Dictionary of Statistics: Collins, London, 267p.
Rowntree, D., 1981, Statistics Without Tears, Penguin, London, 199p.

38

Вам также может понравиться