Vous êtes sur la page 1sur 13

PETROPHYSICS, VOL. 45, NO. 3 (MAY-JUNE 2004); P.

268280; 12 FIGURES

Well Log Normalization: Methods and Guidelines


Daniel E. Shier1

ABSTRACT

Curve normalization identifies and removes systematic


errors from well log data so that reliable results may be
obtained for reservoir evaluation, solving difficult correlation and seismic modeling problems. It is especially critical for any work involving batch-mode computer processing.
The normalization equation is a function of four variables, two of which are defined for each well and two of
which are related to regional lithologic patterns.
Well-to-well comparisons are made using histograms,
crossplots, depth plots, and statistical measurements.
Prior publications on normalization have been individ-

INTRODUCTION

Those who use well logs on a regular basis have noted


that curves from neighboring wells do not necessarily have
the same values in stratigraphic units that are believed to
have nearly identical properties throughout an area. This
observation holds true even when standard borehole- and
environmental-correction procedures have been applied.
For accurate formation evaluation, real anomalies with
implications for reservoir management must be separated
from apparent anomalies caused by inaccurate data. This is
especially important for regional studies in which only one
or two wells with regionally anomalous log values may contain vital information. In many situations, distinguishing
between these cases is made more difficult by datasets comprising logs from many vendors, and logs of widely different vintages. Well log normalization is needed before useful
quantitative comparisons can be made from such diverse
data.

ual case studies. This paper describes methods that can be


applied generally, and provides guidelines for their use in
a variety of rock suites. Also discussed are the errors that
are expected for the various curve types, and suggested
methods for correcting them.
Factors to be considered in planning a normalization
project include the rock types and compaction patterns in
the study area, hole rugosity, curve types, and the stratigraphic level at which run changes take place. Guidelines
are provided to avoid the introduction of additional inaccuracies. Even with these caveats, an irreducible random
error will remain in the data.

There are various reasons for tool inaccuracies. Wellsite


tool calibration may have been ineffective or there may
have been drift in tool response between calibrations. Not
all logging engineers are equally experienced and may have
scaled some curves improperly or may have registered them
on the depth grid incorrectly. Changes in the borehole environment may have been recorded improperly. The design of
logging tools has varied over the years, and from one company to the next. With older logs, the exact model number
for the tool is often obscure or the borehole correction chart
is not readily available (Shier, 1997). Some more recent
tools were released before the software or tool calibration
methods became entirely reliable.
Hydrocarbon exploration and production can benefit
from curve normalization in several ways. Normalized
curves may result in better correlations. This is especially
useful in intervals with converging seismic horizons,
unconformities, onlap, shifts in depositional environment,
or other complexities. Gamma ray curves in shales depos-

Manuscript received by the Editor January 29, 1999, revised manuscript received January 18, 2004.
Consultant, Golden, CO, USA.
2004 Society of Petrophysicists and Well Log Analysts. All rights reserved.
1

268

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

ited above an unconformity may be 10 API units higher


than in those below. Precise integration of the data from a
3D seismic survey and from scattered wells requires log
data that are free of the effects of differences in mud programs and other drilling contractor procedures. A poor tie
will erode confidence in the prospect.
Normalization may be performed whether or not the data
have been borehole corrected. When compensated porosity
curves are available, it is best to normalize the data without
any additional borehole-correction procedures. This is
especially true for neutron curves that usually require a
major effort to back out the original corrections (if it can be
determined which ones had been made) and then apply the
corrections again. For porosity curves in general, the
improvement in results from additional borehole-correction
work is usually not worth the effort. In datasets with rugose
holes, borehole correction of the gamma ray curve prior to
normalization can often be done with little effort by assuming some average mud weight and batch-mode processing
the data. This procedure is recommended even if the mix of
centered and eccentered tools cannot always be clarified.
Only a small number of papers have been published on
well log normalization. Previous workers have established
the basic tools of well log normalization and published individual case studies. Neinast and Knox (1973) published the
first study on well log normalization. This was followed by
three other well-documented case studies: Patchett and

Coalson (1979), Lang (1980), and Reimer (1985). These


authors established histograms and crossplots as standard
techniques for comparing curve responses between wells.
Doveton and Bornemann (1981) introduced the use of trend
surfaces for regional normalization.
The goal of this paper is to place well log normalization
in a broader framework. This is an attempt to codify knowledge gained in interpreting approximately 200 different
datasets from diverse rock suites from many different North
American basins. It is not a case study, but rather a distillation of 24 years of experience with well log normalization.
Any well log curve may be considered as the sum of a
signal (actual rock properties), random noise (including
random variations in the count rates for the nuclear tools),
and systematic errors. Well log normalization is the process
of eliminating systematic errors from well log curves.
This paper outlines the general principles of well log
normalization and defines the normalization equation. It
also includes suggested approaches for normalization when
the rocks studied are inconsistent or otherwise difficult to
deal with. In addition, this work describes the types of corrections that are generally required for the different curves
and the percentage of curves of each type that are likely to
require correction in the normalization process. Finally,
there are suggested new steps to add accuracy to porosity
interpretations from the Gamma Ray Neutron Tool and similar devices (GNT-type neutron curves). These curves are
all scaled in counts per second or equivalent units.
Each of the common log curves has unique characteristics that affect how to best handle it in the normalization
process. A curve-by-curve discussion of these characteristics and suggested normalization guidelines are provided
for selected, commonly available logs. Also included are a
few broad principles to avoid the erroneous alteration or
removal of anomalous, but real, variations in log curve
responses, reflective of real differences in rocks and pore
fluids.
NORMALIZATION EQUATION

FIG. 1 Gamma ray curve comparisons. The clean limestone


and the marine shale are regional in extent and should not vary
much in these three nearby wells. However, the three wells
have significantly different gamma ray response. Well A can be
adjusted to the same pattern as the type well by shifting the
entire gamma ray curve. Well B has a value of 12 API units in the
regional clean limestone just like the type well, but it has a very
different value in the regional marine shale. To adjust Well B to
the pattern of the type well, the scaling factor for the curve must
be changed.
May-June 2004

Figure 1 illustrates the normalization problem. Suppose


that stratigraphic studies predict that both a clean limestone
reservoir and a regionally distributed marine shale vary little from well to well in a field study area. Suppose also that
it has been determined that one of the wells has a correct
gamma ray response in both formations (for example, by
using the techniques discussed in later sections of this
paper). That well may be designated the type well. In Well
A, the difference in the gamma ray response, measured in
API units, between the limestone and shale is the same as in
the type well. However, in Well A, all of the gamma ray values are too high by a constant amount. Well A can be nor-

PETROPHYSICS

269

Shier

malized by simply shifting the entire gamma ray curve by


some constant number of API units. Well B has the predicted response in the limestone but has a shale response
that is higher than that in the type well. The gamma ray
curve in Well B can be adjusted using a new scaling factor,
in addition to shifting, so that it will match the type well.
The normalization equation makes linear adjustments to
log curves according to standard parameters picked in the
type well, shifting the curve across the scale and/or changing the scaling factor in one operation. While it is theoretically possible that nonlinear corrections are needed in some
cases, as a practical matter, such errors tend to be so small as
to be inseparable from random errors that cannot be
addressed by the normalization process.
In each well, the equation requires measurement of the
value of two specific lithologies using the uncorrected data.
These are usually close to the maximum and minimum values of the curve in the interval. For the gamma ray curve
shown in Figure 1, these are the clean limestone and the
marine shale. The regional value for that same lithology is
the best estimate of the correct value for that lithology at
that location.
For a log curve whose unnormalized values are designated Vlog, the normalized values of the curve Vnorm are
given by
Vnorm = R min + ( R max - R min )(Vlog - Wmin ) / (Wmax - Wmin ).
(1)

sured using the uncorrected data. Parameters Rmin and Rmax


are the regional best estimates of the correct value for the
two lithologies at that location, whether they are constant
values or are taken from trend surfaces (Doveton and
Bornemann, 1981).
These parameters are illustrated by the curves in Figure
2. Choice of the regional maximum and minimum parameters is shown in the type well, while the picks for maximum
and minimum values in Wells A and B are also shown. Figure 3 shows the effects of the normalization as related to the
original scale.
More accurate normalization results are achieved if the
rock types used for standardization are picked to maximize
the difference between Rmax and Rmin. This ensures that values are interpolated accurately over the entire range of values in the wells to be normalized. A choice of Rmax and Rmin,
such that their difference is small, for example three porosity units, is unsuitable because small errors would be
greatly exaggerated when extrapolated to higher and lower
log values outside the range of Rmax and Rmin. Unfortunately,
real rocks often prove uncooperative. Coal, for example,
should be an excellent low-density normalization rock, but
is highly vulnerable to hole enlargement and can also harbor significant impurities. Similarly, shale facies are prone
to hole enlargement, borehole breakout, and other forms of
damage, depending on the mud used.

Wmin is the value of a specific lithology in each well.


Generally, it is near the minimum value for that curve in that
interval. Wmax is the value of a different specific lithology in
each well. It is usually near the maximum value for that
curve in that interval. Parameters Wmin and Wmax are mea-

FIG. 2 Normalization parameters for the same curves shown in


Figure 1.
270

FIG. 3 Changes in the gamma ray scale after normalization for


Wells A and B in Figures 1 and 2.

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

Histograms and crossplots for a specified formation are


useful because they allow data from different wells to be
compared efficiently using overlay methods. Depth plots
need to be at a much more compressed scale than is available on blueline copies.

need to be aware that incorrectly normalized data can be


made to fit the same linear trend as the type well if offsetting errors are made on both of the crossplotted curves. If
normalization is done on the basis of nonreservoir intervals,
as a final check it is often best to make normalized
crossplots of the reservoir interval itself to make sure that
each well has reasonable porosities.

Histograms

Vertically compressed depth plots

Frequency plots of the log values for an interval are the


easiest display to use. Although good bimodal histograms
(Figure 4a) are occasionally encountered, typical histograms vaguely resemble a statistical normal distribution
(Figure 4b). In this case, the low-side shoulder may be used
as the minimum and the high-side shoulder may be used as
the maximum. In a polymodal case, several distinct modes
are present (Figure 4c). If the lithology representing one of
several modes is poorly represented in some of the wells, it
may result in confusion and incorrect normalization when
other display types are not used to check the results. If histograms like those in Figure 4b or 4c are the only means
used to compare curves, good results require that the same
lithologies be present in the same proportions.

Depth plots at 200 ft per inch (2400:1) to 1000 ft per inch


(12000:1) are useful in a number of situations. Consider, for
example, a stratigraphic sequence in which a few somewhat
inconsistent thin beds of tight limestone or other lithology
provides the best maximum or minimum. Using a vertically
compressed plot allows the simultaneous assessment of
whether a given bed has the desired typical development,
is thick enough to provide an accurate value, and has no
hole enlargement issues. Without changing the display,
other maximums and minimums for other formations can be
recorded as well. The effects of data gaps that may make
histograms and crossplots highly misleading can also be
ignored. Compressed vertical plots can help determine
exactly where a run change takes place for a given curve
and record maximums and minimums for both runs. After
completing normalization using histograms and crossplots,
it is useful to quickly check the same data as vertically compressed plots on a video display to identify unrecognized
run changes or other problems that were not evident on the
histograms and crossplots.

DATA DISPLAY TOOLS FOR COMPARING


LOG RESPONSE

Crossplots

A more or less linear trend on a crossplot typically represents various mixtures of two lithologic components. In a
pure limestone, a linear trend on a neutron/density crossplot
represents various mixtures of calcite and fluid-filled pore
space. In shales, a linear trend on the neutron/gamma ray
crossplot may represent a mixture of clay and silt.
If the type well includes all proportions of the lithologic
components and a linear trend is established, then points
from a well with only a limited lithologic mixture will fall
on the trend when normalized. However, crossplot users

FIG. 4 Some typical gamma ray histogram patterns. Pattern a


is generated from beds of two distinct rock types. Min and Max
values are best picked on the peaks. Pattern b shows an overlapping mix of rock types. Min and Max values are best picked
on the shoulders. Pattern c, with a number of distinct peaks, is
seldom seen in actual stratigraphic sequences.
May-June 2004

ESTABLISHING CORRECT CURVE VALUES

Normalization requires that at any location, the correct


values of the maximum and minimum lithologic end members be known. Defining these correct values is the most
critical part of any studyit may also be the most
time-consuming.
The type well for a particular curve is the standard to
which the other wells must be adjusted. Generally, comparisons are made in one or more specific formations. A type
well can have either local or regional application. Even
when lateral stratigraphic change is not observed across an
area, there will usually be some. Thus, it is good practice to
select a type well near the geographic center of the study
area.
Under certain conditions, stratigraphy may be only a
minor source of regional variation as compared with sediment compaction. The Lower Miocene of the Gulf Coast is
a good example. Rocks presently buried at 6000 ft [1830 m]
have the same porosity log properties for long distances,
while the very same lithologies at a burial depth of 10,000 ft

PETROPHYSICS

271

Shier

[3050 m] have very different properties. Instead of a type


well, a type compaction curve is utilized for sediments with
normal pore pressures. Figure 5 illustrates this normalization process. Damage to the shales during the drilling process and varied regimes of abnormal formation pressure
add extra challenge to normalization in the Gulf Coast Tertiary sequences.
METHODS
Statistical normalization

All references cited in this paper assume that there is a


correct value for each part of each curve and that the correct value is the same as the value from appropriate measurements from core samples or other data sources. This is
different from normalization as defined by statisticians. As
practiced in the petroleum industry, statistical normalization uses statistical measures of curve data in the normalization of well log values. For example, a log curve (e.g.,
the gamma ray) may be adjusted so it has the same average
and standard deviation in a correlated interval (e.g., the Fort
Union formation) in all wells. Another statistical method is
to adjust the curve so that the 10th percentile and the 90th
percentile are identical in each well in a correlated interval.
Statistical normalization is based on the assumption that
the same rocks are present in the same percentages over the
study area. Practical difficulties include real changes in
rock compaction, incomplete penetration in some wells,
and null intervals.
Tampering with the standard deviation of the data from a
well ignores the fact that in some wells a greater spread of
the data may be due simply to factors such as more noise,
severe hole enlargement effects in that well, and different
thin-bed response by different tools. Brute-force statistical
warping of the data may result in entirely incorrect curve
scaling.

may be made to the dispersion of the data of individual


wells using calculated standard deviations.
In reservoir studies for unitization, an equity holder may
be especially attached to certain incorrect porosity values
that provide a financial advantage. This method remains
popular because it is the method least likely to change those
values.
There are several flaws in this method. Nothing is done
to test the implicit assumption that all stratigraphic variability is effectively random. If there is a genuine trend, it will
be removed in the normalization process. The acceptable
deviation of the mean for any one well from the all-well
mean is selected arbitrarily.
Type-well method

A representative sample of the wells is examined and one


of the wells is selected as the type well for that particular
curve. As part of that process all other curves are examined.
In general, a good type well has values that are the same as
those in many other wells and has an in-gauge hole.
The data for each of the individual wells are then compared with the type well. If adequate software is available,
the person making the comparison will be able to compare
histogram patterns, crossplot patterns, and vertically compressed depth plots simultaneously. Variables for the basic
normalization equation are adjusted until a good fit with the
type well is obtained. Many wells may be simply checked
but not adjusted at all.

Big histogram method

In a correlated interval, all data from all wells are combined into one large composite histogram. Analysis of this
composite histogram, based on a statistical Normal Distribution, yields an envelope into which the means of the
individual wells must fit. For example, it may be decided
that the envelope consists of 75% of the area under the Normal Curve. The dimension of this envelope is considered a
measurement of the real geological changes among the
wells. If the mean of an individual well does not fall within
the envelope, the data are adjusted so that they do fall
within the envelope. In some cases, similar adjustments
272

FIG. 5 Effects of sediment compaction. In this Gulf Coast


example, the density curves for Runs 1 and 3 are compatible
with regional compaction patterns. The curve for Run 2 will need
to be shifted in the normalization process. This simplified example assumes that the entire well is normally pressured. For purposes of illustration, the shale compaction curve is shown as a
straight line. In actual field examples, the density compaction
curve is slightly concave toward the low-density side.

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

This is an improvement over the big histogram method.


Some petrophysical judgement has been used to explain
and correct anomalous wells. The type-well method works
well in settings where there is little distance or geological
change between wells in the dataset. Some noise has been
removed from the data but more remains than if the trial
normalization method (below) had been used. Nothing is

learned about data variability or any directional properties


that they might have.
Neighbor comparisons

In this method, a geologist designates clusters of wells in


the area. A cluster of wells might penetrate the formation
much deeper and have different compaction than the other
clusters. Within a cluster, the formation might have a rather
different mix of rock types.
Normalization is accomplished by selecting a type well
for each cluster and applying the type-well method to each
cluster individually. Selection of a good type well for a
porosity curve is usually straightforward, as density, compensated neutron, and sonic curves, as they appear on the
bluelines, are usually accurate in more than 75% of the wells.
In the tear fault regime of coastal California or Cook
Inlet, Alaska, each of the various fault blocks may have a
very different compactional history. In such a situation,
neighbor comparisons within each fault block are the most
accurate normalization method available.
Trial normalization method

FIG. 6 Bubble map of Wmax values showing an apparent


regional gradient in rock properties.

FIG. 7 Bubble map of residuals from a trend surface. Residuals appear to be randomly distributed, indicating that the trend
surface is a valid indication of the regional pattern.

May-June 2004

A well from the center of the area that has a relatively


low data dispersion is picked at random and used as a preliminary type well. Wells with high standard deviations
should be avoided because they usually have significant
hole-enlargement effects. Preliminary Wmax and Wmin picks
are then made for the 10 to 20 nearest neighbors for comparison with the preliminary type well. Experience gained in

FIG. 8 Bubble map of residuals from a trend surface. Residuals are clumped. Clumps may reflect different vintages of data,
structural patterns, or other factors.

PETROPHYSICS

273

Shier

this brief exercise is then used to confirm the preliminary


choice of the type well, or to select a different well from the
set of 10 to 20 wells. The Wmax and Wmin picks with reference to the final type well are determined and mapped for
all wells, either as contours or as a bubble map. Figure 6
shows a nonrandom pattern. Doveton and Bornemann
(1981) dealt with this sort of pattern by constructing a trend
surface that reflected regional geological variations and
normalizing the data with respect to that trend surface. Figure 7 is a random pattern of a trend surface residual suggesting that the trend surface removes most of the systematic
variation in the data. Figure 8 is a clumped pattern that may
be the result of historical patterns of tool vintage or logging
contractors.
The various maps will generally delineate the issues to
be faced in normalization work. Can clumped values be
explained by different tool vintages or logging contractors?
Can clumped values or regional trends be explained by
structure maps related to compaction? Are clumped values
indicating that it would be better to use the type-well
method? If a trend surface is to be used, is it better to use a
first-order or a second-order trend? After these questions
are answered, the trial normalization parameters are discarded and the final normalization parameters determined.
Results of the trial normalization method are more reliable than other methods because the additional dimensions
of stratigraphy and instrument type have been considered.
Because there are such large variations in gamma ray
response with regard to tool vintage and contractor, and
because the gamma ray is fairly insensitive to rock compaction, trial normalization is generally used for gamma ray
normalization over large areas regardless of the methods
used for the porosity curves.
POROSITY CURVES

Approximately 20% of the compensated neutron and


density curves from the 1970s have errors of two porosity
units or more. Uncompensated sonic logs from the 1950s
have a similar error rate. Fewer than 5% of the compensated
sonic logs have noticeable tool-related errors. Wells logged
in the 1990s have significantly fewer inaccuracies than
those logged in the 1970s.
In ideal reservoirs like those of the Gulf Coast Pleistocene (32% porosity and low connate water resistivity), the
effect of a change of two porosity units may not significantly impact economic decisions. In a producing province
with low porosities and high water resistivities, such as a
typical Rocky Mountain gas reservoir, a difference in two
porosity units in a shale-free sandstone may change the calculated water saturation by 10% and the hydrocarbon pore
274

volume by about 20%. These errors affect not only reserve


estimates but also completion decisions.
When picking a type well for a porosity curve, an additional constraint is required. The curves from the type well
must have the same lithology response in the reservoir rock
type as that built into the software routine that will be used
to calculate crossplot porosities. In normalizing a dataset,
the analyst may use a different lithology response for different vendors and tool models, or may simply select one of
the responses for the final batch-mode work and normalize
out the vendor and tool differences. This latter procedure
requires significantly less effort and leaves far fewer
opportunities for errors in keeping track of details. However,
some instrument types have a dolomite porosity response
that is significantly nonlinear with respect to limestone
porosity; others do not. Therefore, in dolomitic sequences, it
may be unwise to combine certain types of neutron curves.
Sonic curves

The best reference lithology for sonic curves is a regionally distributed rock that is resistant to hole damage or
enlargement during drilling. In thick clastic sequences, this
is usually a hard, silty shale. In carbonate sequences, it is a
tight limestone, tight dolomite, or an anhydrite.
Apparently, even the earliest sonic tools had highly
accurate clocks. Hence, well-to-well differences between
Wmax and Wmin are rare. Such differences, when observed,
are usually due to incorrect scaling of logs in the field or in
the digitizing of blueline prints. As a result, normalization
of sonic curves is a matter of shifting the curve without
changing the scaling factor. As many as 40% of early
uncompensated sonic curves require shifting by 2 to 5 ms/ft
in the normalization process. Any errors in compensated
sonic curves are usually small when compared to shale
damage effects, compaction effects, and editing problems
such as cycle skips and noise spikes. Frequently these
effects masquerade as normalization problems. As a rule of
thumb, if more than 5% of compensated sonic curves
appear to need adjustment in the normalization process, the
normalization is probably in error. These errors can be
accounted for and eliminated by considering other sources
such as compaction and shale damage.
In Gulf Coast Tertiary sequences, compensated sonic
curves usually require no changes in the normalization process to obtain valid values for formation evaluation in the
sandstones. However, shales are of equal importance in the
creation of accurate synthetic seismic curves. In most wells,
the sonic curve shows significant formation damage (sonic
values that are too high) in the associated shales. The
amount of formation damage is directly related to the
silt/clay ratio in a shale bed, with formation damage

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

increasing with clay content. Using the procedures


described here, it may be possible to define a Wmax and Wmin
value for each logging run in each well that will change the
scaling factor of the sonic curve and produce a fairly good
result for making synthetic seismic traces. The normalized
curve generated in this way should be designated by a special mnemonic since it may not be entirely suitable for formation evaluation in the shaly sandstones.
Neutron curves

In carbonate/shale sequences, the best minimum


lithologies are anhydrites and tight carbonates. Shales that
are not particularly subject to hole enlargement are the best
maximum lithologies.
In purely clastic sequences, there are no ideal neutron
maximum and minimum lithologies. Minimums are highly
problematic. Only in exceptional circumstances are
water-wet sandstones consistent enough to use for the maximum. Maximum porosity shales usually have varying
degrees of hole enlargement. Purely statistical methods or
methods that use only histogram displays will almost invariably produce poor results. Some methods for addressing
these problems are given in the Guidelines section below.
The basic data for any porosity interpretation from a neutron log are counts/second. Porosity is more or less
inversely proportional to the log of the counts/secondalthough in practice the logging contractor may use
dual detectors and more complex mathematical variants of
this relationship. Tool calibration relates a particular count
rate to a particular porosity. As a consequence of the logarithmic relationship, a small error in assigning the count
rate for the high porosities will produce a large error. On the
other hand, a small or moderate error in assigning the count
rate for the low porosities will result in only a small error in
porosity. As a result, substantial inaccuracies in neutron
porosities are more often found in rocks with higher
porosities than in those with lower porosities.
In normalization work, there are two practical effects of
these relationships. First, nearly all neutron normalization
involves changing the scaling factor, since (Wmax Wmin)
varies a good deal in wells requiring normalization. Second, if there is an obvious problem requiring changes to the
high-end porosities but there is no good low-end porosity
rock available, reduction of the noise level is usually better
served by assuming that zero porosity is correct. This produces better results than shifting the whole curve so the
high-porosity end fits.
Neutron curves often have the same apparent porosity in
a number of entirely different rock types. For this reason,
using crossplots in the normalization process is more
important than for any of the other curves. In formations
May-June 2004

that include shales, neutron/gamma ray and neutron/sonic


crossplots are especially useful.
Neutron curves scaled in porosity units

Generally 15% to 25% of 1970s-vintage compensated


neutron curves require adjustment in the normalization process. Stated another way, 75% to 85% of compensated neutron curves give reasonable values without any borehole-correction procedures beyond those done before the
blueline log was created. If possible, sidewall neutron curves
should be normalized separately from compensated neutron
curves. These two main types of neutron curves should be
handled separately in any batch-mode calculations.
Neutron curves scaled in counts/second

These GNT-type logs include standardized versions of


counts/second such as API units, Environmental Units,
MicroRoentgens, and Radium microgram equivalents/ton.
Schlumberger (1969) provided charts for GNT tools that
indicate that the logarithm of the counts/second is slightly
nonlinear with respect to porosity. The exact function by
which porosity departs from linearity depends on the spacing of the neutron device, hole size, temperature, water
salinity, and dolomite content. In the 5% to 20% porosity
range, departure from linearity varies from 0.5 to 2.0 porosity units. As a source of noise, this is minor compared
with other factors.
Typical sets of GNT-type neutron curves represent four
to eight different logging companies, the majority of which
are small companies for which no charts are readily available. Header and borehole-size information are frequently
inadequate to use a chart, if one even exists. As a practical
matter, the small departure from linearity of the logarithm
of the counts/second with respect to porosity can be
ignored, simplifying the whole process of removing much
of the rest of the noise from the dataset. This is essentially
the same procedure as using semilog paper recommended
by Hilchie (1979) and others.
For datasets that are entirely digital, it is best to calculate
a curve of the logarithm of the counts/second curve and use
it for all further normalization and porosity interpretation
work. Inspection and comparison of the logarithm curve
between wells easily reveal casing shoes in wells completed
open hole, changes of fluid in the hole, incorrect zero registry of the curve on the log copy, and other problems that
have to be dealt with on a well-by-well basis. When the
GNT-type tool passed from an environment that was poorly
shielded from the borehole walls to a more shielded environment (i.e., from air to mud, or from uncased hole to
cased hole), it resulted in a change in scaling factor but no
change in the zero point. As a result, these changes can be

PETROPHYSICS

275

Shier

dealt with by simply shifting the portion of the logarithmic


curve below the discontinuity so that a key lithology (such
as a tight limestone) lines up with the same lithology above
the discontinuity. Shier and Sponable (1997) point out that
baselining these curves may also be required to allow for
mudcake and other problems not seen in curves from more
modern dual-detector tools.
The type-well method is used when shales are present.
The gamma ray/neutron crossplot is the definitive data display. First, histograms and/or compressed depth plots are
used to normalize the gamma ray curve for all wells. Rmin
for the neutron curve is determined using modern logs
nearby. Anhydrites are generally assigned a porosity of
zero, and tight carbonates a porosity of 1% to 2%. The main
task is to determine a neutron porosity for the Rmax that corresponds to an index shale that has a regionally consistent
neutron response. Using modern compensated neutron
curves from the same area does not work because the more
modern neutron tools count only those gamma rays that
have energies characteristic of neutron events. The
GNT-type neutron tools count gamma rays of all energies.
The effect of the gamma rays from the shales or other radioactive rocks on the GNT-type tools may be quite large. For
example, in a West Texas study, a Grayburg feldspathic
siltstone used for the Rmax had a value of 20% porosity on a
compensated neutron curve. This same siltstone would
have an apparent neutron porosity of only 8% if the gamma
ray/neutron crossplot pattern of the GNT-type neutron
curve had been assumed identical to a nearby compensated
neutron curve.
If core porosities are available, they can be crossplotted
versus the logarithm of the counts/second for that well and a
linear trend line established as shown in Figure 9. The point
on the trend line that corresponds to the normalized gamma
ray for the index shale has a corresponding tentative value
Rmax on the porosity scale. If there is another log that gives
reliable porosities in the same well, it can be used instead of
the core data. Generally, a tentative Rmax is calculated for
most or all of the wells that have adequate supporting data,
and these are used to establish the final Rmax for all wells.
In the absence of any other source of porosity data in the
same wells, a trial Rmax is assigned, and 10 to 20 wells are
normalized as if the trial Rmax were correct. The porosity
histograms for these wells are compared with those of 10 or
more modern logs from the same area. Of course, there will
be a variety of histogram patterns, but the highest porosity
modern wells should be similar to the highest porosity older
neutron wells. If the patterns compare well, that trial Rmax is
accepted as the final Rmax determination. If not, the trial Rmax
is changed, and the results checked again until a good
match is obtained.
In a section that consists of shale and sandstone with var276

ied cementation, the same procedures may be followed if a


reliable Wmin can be picked in each well. Often this is not the
case and accurate results cannot be expected.
In dolomitic sections, comparisons between GNT-type
neutron curves and modern neutron curves should use modern curves that are more or less linear with respect to porosity. These are the more modern compensated neutron curves
and the sidewall neutron curves. Compensated neutron
curves from the 1970s and 1980s have a nonlinear response
to dolomite porosity changes and should be avoided.
Density curves

Typically 15% to 20% of the compensated density


curves require adjustment during the normalization process. The best reference lithology for the density curve is a
single pervasive rock type that suffers minimal hole enlargement or damage during drilling. In thick clastic sequences,
this is usually a hard, silty shale. In carbonate sequences, it is
a tight limestone, tight dolomite, or an anhydrite.
Like the neutron porosity, density logs are based on
count-rate measurements and therefore are theoretically
subject to the same scaling factor problems as neutron logs.
Defining Wmax (low porosity) values is simple if anhydrite
or a tight carbonate is present. However, defining a suitable
Wmin (high porosity) rock type is usually difficult. The highest apparent density porosities often represent enlarged
hole, not some consistent lithology. As a practical matter, it
is usually best to simply shift density curves, without

FIG. 9 Procedure for estimation of Rmax for GNT-type neutron


curves. Core data are used to establish the regression line for
that well. The index shale has been selected for its regionally
consistent neutron response. The log10 of the counts/second is
entered on the x-axis and projected to the regression line; that
point is projected to the y-axis.

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

changing the scaling factor, making sure the prevailing rock


types have the appropriate values in each well. If the scaling
factor is to be changed, it must be based on crossplot analysis. Relying on histograms alone as illustrated in Figure 4b
is not recommended.
In Gulf Coast Tertiary formations, it is common to
encounter density curves that show the effects of shale damage, resulting in an apparent shale porosity that is too high
as compared with a nearby type well with minimal shale
damage. Wells drilled with oil-base mud have the least
shale damage. The procedure shown in Figure 5 is recommended. It is generally best to check the sandstones or hard
siltstones against the standard compaction curve and ignore
the most clay-rich shales entirely.
Photoelectric effect (PEF) curves

Generally about the same proportion (15% to 20%) of


PEF curves require adjustment as do density and neutron
curves. The PEF curve is normalized by shifting, without
any change of the scaling factor, for the same reasons as
density curve.
GAMMA RAY CURVES

Ideal rock types for gamma ray curve normalization are


similar to those for the neutron curve. Anhydrites and clean
carbonates provide the best Wmin values. In fluvial and delta
plain sequences, it is usually possible to define a channel
facies that makes a satisfactory Wmin. Shales are usually
used for Wmax. The ideal shale should be sufficiently competent that hole enlargement is not a problem but should be
rich enough in clay that ambient gamma ray values are

FIG. 10 A frequency plot of the gamma ray values for the


Dakota formation sandstone channels in the southern Powder
River basin of Wyoming. Channel sandstones contain little clay
but have a noticeable feldspar content.

May-June 2004

fairly high. The recommendation is to avoid using uranium-enriched shales like the Woodford in the Permian and
Anadarko basins for two reasons: First, many of these
shales have more local variations than casual observation
would suggest. Second, tool response in these beds will
depend on which detector was usedthe early low-efficiency Geiger-Mueller counter or the later high-efficiency
scintillator. For more ideas on dealing with difficult clastic
sequences, see the Guidelines section below.
Although there is an industry standard for the gamma ray
curve, as a practical matter, it is not possible to establish
exactly what this should be based on the usual data at hand.
For the gamma ray curve, the correct response is one that
is consistent with its neighbors and gives a reasonable volume of shale interpretation for the various types of rocks. In
other words, the gamma ray scale is relative to the other
wells in the project, not absolute. Most gamma ray
normalizations require only a scale change. Zero API units
is frequently correct, whether or not values in the shale
range require adjustment. The increasing inaccuracy of
gamma ray values with higher radioactivity is supported by
data from the Dakota formation in the southern Powder
R i v e r b a s i n o f Wyo mi n g . F i g u r e 1 0 sh o w s t h e
unnormalized gamma ray values for the Dakota channel
facies in 638 wells. Figure 11 shows the unnormalized
gamma ray values for the associated regional marine shales
for the same wells.
Of all the instrument responses, the gamma ray curve is
the most likely to be normalized using a trend surface
because clay minerals and the clay/silt mix in shales vary
according to their source rocks, distance from the sediment

FIG. 11 A frequency plot of the gamma ray values for a


regional marine shale associated with the Dakota formation in
the southern Powder River basin of Wyoming. Note the much
higher standard deviation than that for the channel sandstone
seen in Figure 10.

PETROPHYSICS

277

Shier

source, depositional environment, compaction, and other


regional patterns.
Even if some adjustments are minor, it is usually easier to
normalize all gamma ray curves except those of the type
well(s), than it is to adjust only those wells that need a large
amount of adjustment.
Spontaneous potential (SP) curves

SP curves are normalized by conversion to a Volume of


Shale curve. This involves editing by hand to remove
mechanical shifts, followed by definition of a shale baseline (a maximum) and definition of a clean sand line (a minimum). Care should be taken to allow for hydrocarbon suppression of the SP as much as possible, based on the gamma
ray, resistivity, and other curves that reflect the volume of
shale in the reservoir.
The raw SP curve (Figure 12) is not helpful. It shows the
effects of both run changes and changes in formation water.
In Figure 12, a shale baseline was defined for the SP curve,
using the caliper and resistivity curves (not shown) as additional references. The straightened SP curve has the shale
baseline set to zero. A clean sand baseline was defined for
the straightened SP curve, again using the caliper and resistivity curves for reference. This was then used to create the
Volume of Shale curve.
Cook Inlet of Alaska is a place where SP normalization is
vital; volcanic clasts frequently make the sandstones more
radioactive than shales, rendering the gamma ray curve
nearly useless for distinguishing sand from shale.

lies or to introduce artifacts. These pernicious tendencies


can be reduced if certain guidelines are followed.
In many wells, different depth intervals were logged
independently at different times. The various logging runs
frequently involved different instruments, with different
calibration histories, different logging crews, and different
service companies. This lack of standardization dictates that
each logging run should usually be normalized separately.
Shier (1997) showed that there is a tendency for the size
of the normalization corrections to be smaller with modern
instruments. This is somewhat balanced by a marked
increase in the introduction of new tools with different
response characteristics. When first introduced, new tools
may experience initial reliability problems. Normalization
is needed for both old and new datasets.
Hole rugosity affects log response. Histograms,
crossplots, and other data displays that do not allow the user
to simultaneously assess the associated caliper curve may
be misleading and lead to inaccurate normalizations. An
alternative method is to define a bad-hole curve, which may
be based on curves in addition to the caliper. The bad-hole

Resistivity curves

Resistivity curves are usually handled with standard


borehole-correction procedures. They are not normalized at
all unless there is some compelling reason to do so. If normalization is necessary, the zero is generally accepted, and
only values of Wmax are needed. Unfortunately, a sufficiently consistent Wmax is almost impossible to define in
most datasets.
Some of the early resistivity curves have negative values
because of incorrect registry of the galvanometer to the
film. The solution here is to shift the entire resistivity curve
for that run so that the minimum value is compatible with
nearby wells. However, this will produce its own inaccuracies if the galvanometer adjustment was not the same for the
entire log run but was arbitrarily adjusted at some unknown
depth.
GUIDELINES

The purpose of curve normalization is to remove systematic inaccuracies, not to remove genuine lithologic anoma278

FIG. 12 Creation of a Volume of Shale curve from an SP curve.

PETROPHYSICS

May-June 2004

Well Log Normalization: Methods and Guidelines

curve can be used as a filter prior to selection of a type well


or other normalization tasks.
Reducing the noise but not the signal

Avoid very small adjustments. If it cannot be clearly


demonstrated that an adjustment will improve the accuracy
of the data, it should not be made. Set a minimum level for
adjustments to be made for each of the curve types based on
the consistency of the lithologies that are available. An
exception is made for the gamma ray curve, the values for
which are not related to an absolute standard but are simply
relative to the other wells in the study.
Look for map patterns. As an initial test, map the adjustments that appear to be needed. If the polarity and magnitude of the adjustments are more or less random, the adjustments are likely to be legitimate. If there are nonrandom
areas, check possible causes. Are they real geologic anomalies? Are they the result of the consistent use of a particular
logging contractor in a small area? Are they clumps of older
vintage logs and younger vintage logs from successive
exploitation episodes?
Check for lithologic trends. Detect regional gradients by
making trend surface maps of the unadjusted log data.
Consider other information before finalizing normalization work. Dont pursue the normalization work in isolation
from what is known about the geology and production tests.
Methods and results need to be discussed with the project
team. Any study that covers a large area and does not
include the efforts of a geologist/stratigrapher usually contains unacceptable artifacts.
Do not overwork the data. There is little to be gained by
attempting to wring the last tiny bit of accuracy from the
data by slightly revising the normalization parameters time
after time.
Some curves cannot be repaired. Most curves can be
converted to reliable data. However, if normalization cannot give the curve the correct response characteristics, the
curve should be discarded.
Improve the data if you can. If it is clear that a particular
adjustment will reduce the noise level, make the adjustment
even if it is also clear that an undesirable amount of noise
will remain. Consider the case of a gamma ray curve
penetrating mechanically competent 15 API unit limestones
and caving 120 API unit shales. There is no caliper curve.
Normalizing the data as if they had been borehole corrected
will tend to correct for the hole enlargement in the shales in
that well.

lesser of two evils strategy. Consider the case where thin


shaly, possibly arkosic sands are developed in a section that
is 90% shale and there are no rocks in the study wells that
have a gamma ray of less than 65 API units. There will be
no problem using a gamma ray histogram to establish the
maximum gamma ray normalizer from the pervasive shale
facies. The problem lies in defining the minimum normalization rock type. Three less-than-perfect approaches are
suggested below.
Wmin can be picked from the low-side shoulder of the histogram as shown in Figure 4b. It can be argued that as long
as the Wmax and Wmin values that are used form an envelope
around the data to be analyzed, there should be no concern
about the effect of the normalization parameters on nonexistent values less than 60 API units. This approach is based
on the assumptions that each gamma ray curve has the same
thin-bed resolution characteristics and that the lowest volume of shale in typical sandstones in each well is the same.
If sandstones are actually better developed in one part of the
area than in another, use of this procedure by itself will
erase the very geologic changes that are most important!
This source of error will be reduced if the trial normalization method is used and regional trends are considered.
Zero can be assumed correct. The analyst adopts zero as
Rmin and Wmin for each well. As indicated in Figures 10 and
11, the amount of noise in the uncorrected gamma ray values decreases markedly at lower gamma ray values.
A hybrid approach uses the zero assumption initially.
Then histograms or crossplots are used to check the results
of the preliminary normalization for an anomalous low-side
data pattern. For example, check the log headers for potassium chloride muds, look at the local stratigraphy more
carefully, and adjust Wmin in anomalous wells. Experience
suggests that this hybrid approach is the most efficient way
to proceed.
Another way to handle nonideal lithologies is to start
with the most inherently reliable curves and work toward
the least reliable. For example, in a thick clastic sequence
that is only partly consolidated, it may be possible to normalize the gamma ray quite accurately but the neutron
response varies greatly. The procedure in this case would be
to first normalize the gamma ray and then crossplot the normalized gamma ray versus the unnormalized neutron curve.
Using trial and error, a Wmax is found that produces the
desired crossplot pattern for each well. Similarly, the compensated sonic curve (which rarely needs adjustment)
might be crossplotted against the density curve in a carbonate play.

Cases without optimum lithologies

Study areas with nonideal lithologies are frequently


encountered. In these cases, it may be necessary to pursue a
May-June 2004

Software considerations

Effective normalization software has different impera-

PETROPHYSICS

279

Shier

tives than software designed specifically for log analysis. It


must be possible to view normalized and unnormalized
curve data as histograms, crossplots, and depth plots on the
screen simultaneously. The effect of any changes in the
normalization parameters should be displayed almost instantaneously and easily compared with the data patterns of
nearby wells, without requiring paper copies. All normalization information must be integrated with formation tops and
available for mapping. The system must be capable of handling different normalizers for different log runs seamlessly.
All of the work discussed here was performed using a
software system that can store normalization parameters
and apply them to log data as required. The parameters
themselves are available for statistical analysis or mapping.
It has a complete capability for comparing crossplot, histogram, and depth-plot data displays between wells, and with
regional norms. Data displays can be confined to any stratum that is delimited by formation tops or other stratigraphic markers. All operations except looking at the data
and making decisions, including printing of standardized
screen displays, can be batch-processed as needed.
CONCLUSIONS

Approximately 20% of neutron, density, and uncompensated sonic curves require normalization before a dataset is
consistent enough to produce reliable results in batch-processed log analysis. Virtually all SP, gamma ray, and
GNT-type neutron curves require normalization. Reasons
for the need for normalization include miscalibrated instruments, instruments of different design, varying borehole
effect, and other factors. Curve normalization is accomplished using a standard linear equation. The most critical
part of the normalization process is determination of reasonable curve values at various points in a study area, taking into account any stratigraphic or compaction trends.
Techniques that combine data from multiple wells into one
histogram or crossplot tend to blur any real stratigraphic
and/or fluid differences across the area. Also, these methods may not highlight incorrect data if they are present in
the majority of the wells in the study.
Normalization is best regarded as a process that reduces
systematic noise within the dataset. The noise contains
some random elements, including the statistical nature of
the nuclear tools and small changes in the rocks. A major
challenge in normalization work is the recognition of which
data inconsistencies are systematic (and correctable) and
which are random (and better left alone). Not all curves can
be handled in the same way. Each curve has its own optimum set of normalization techniques.

280

ACKNOWLEDGEMENTS

I thank W. David Kennedy for his extensive work with


this paper and for his helpful suggestions.
REFERENCES
Doveton, J. H. and Bornemann, E., 1981, Log normalization by
trend surface analysis: The Log Analyst, vol. 22, no. 4, p. 39.
Hilchie, D. W., 1979, Old (pre-1958) electrical log interpretation,
published by the Author in Boulder, Colorado, USA., 164 p.
Lang, W. J., 1980, SPWLA Ad hoc calibration committee report:
The Log Analyst, vol. 21, no. 2, p. 1419.
Neinast, G. S. and Knox, C. C., 1973, Normalization of well log
data, Paper I in 14th Annual Logging Symposium
Transactions: Society of Professional Well Log Analysts.
Patchett, J. G. and Coalson, E. B., 1979, The determination of
porosity in sandstone and shaly sandstone, Part 1Quality
control: The Log Analyst, vol. 20, no. 6, p. 312.
Reimer, J. D., 1985, Density-neutron cross plot discrepancies
through the Swan Hills Member, Swan Hills Unit #1, Alberta,
in Transactions of the 10th Formation Evaluation Symposium:
Canadian Well Log Society and in Vol. 3 of the CWLS Reprint
Volumes, Canadian Case Histories.
Schlumberger, 1969: Log Interpretation Charts, Schlumberger
Ltd., 76 p.
Shier, D. E., 1997, A comparison log response between logging
companies and different vintages of tools: The Log Analyst,
vol. 38, no. 3, p. 4761.
Shier, D. E. and Sponable, D. M., 1997, Correcting drift in GNT
neutron log data from West Texas: The Log Analyst, vol. 38, no.
3, p. 1619.

ABOUT THE AUTHOR


Dr. Daniel E. Shier is a consulting geologist and
petrophy sicist. He was employ ed by Shell Oil as a
micropaleontologist and stratigrapher from 1969 to 1974. During
his tenure for Scientific Software from 1974 to 1981, he worked
with sets of digital well logs that encompassed entire exploration
plays and developed the normalization methods needed to obtain
accurate mapping parameters for regional studies. He was associated with Energy Data Services from 1981 to 2001. There he
developed and taught a workshop on log normalization. He
received his BA degree in geology from Rice University and his
PhD degree in geology from Florida State University. Dr. Shier is a
member of SPWLA, AAPG, Rocky Mtn. Assoc. of Geologists and
Denver Well Log Society.

PETROPHYSICS

May-June 2004

Vous aimerez peut-être aussi