t Technical papers

Density log correction for borehole effects and its impact on

well-to-seismic tie: Application on a North Sea data set
Isadora A. S. de Macedo1, Jose Jadsom S. de Figueiredo1, and Matias C. de Sousa1
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Abstract
Reservoir characterization requires accurate elastic logs. It is necessary to guarantee that the logging tool is
stable during the drilling process to avoid compromising the measurements of the physical properties in the
formation in the vicinity of the well. Irregularities along the borehole may happen, especially if the drilling
device is passing through unconsolidated formations. This affects the signals recorded by the logging tool,
and the measurements may be more impacted by the drilling mud than by the formation. The caliper log in-
dicates the change in the diameter of the borehole with depth and can be used as an indicator of the quality of
other logs whose data have been degraded by the enlargement or shrinkage of the borehole wall. Damaged well-
log data, particularly density and velocity profiles, affect the quality and accuracy of the well-to-seismic tie. To
investigate the effects of borehole enlargement on the well-to-seismic tie, an analysis of density log correction
was performed. This approach uses Doll’s geometric factor to correct the density log for wellbore enlargement
using the caliper readings. Because the wavelet is an important factor on the well tie, we tested our methodology
with statistical and deterministic wavelet estimations. For both cases, the results using the real data set from the
Viking Graben field — North Sea indicated up to a 7% improvement on the correlation between the real and
synthetic seismic traces for well-to-seismic tie when the density correction was made.

Introduction as the requirement of minimum phase. The main factors

The well-to-seismic tie is an important tool in seismic controlling the accuracy of the well-to-seismic tie are
inversion and interpretation. It can join the vertical res- the seismic wavelet, a coherent time-depth relationship,
olution of the well logs with the horizontal resolution and the accuracy of the elastic well-log data. The time-
of the seismic data, thus enabling the identification depth relationship is used to generate a reflectivity
of stratigraphic markers on the seismic section and series in the time domain. Accurate elastic well-log data
the correct estimate of the wavelet. Consequently, with produce a reliable reflectivity series in depth.
the proper wavelet, the real seismic data can be in- An accurate well tie for the characterization of any
verted to reflectivity or impedance. In general, the basic reservoir requires accurate elastic logs. The compre-
principle behind a well tie is to compute a synthetic hensive information provided by wireline procedures,
seismic trace and compare it with the real seismic trace. when tied properly to land seismic data, allows inter-
The convolutional model is the basis for most well tie preters to verify whether their geologic conclusions
procedures because it establishes a relationship be- about the seismic background are suitable to the ob-
tween the reflectivity function calculated from the served lithologic parameters (White and Hu, 1998). A
well-log data, the seismic wavelet, and the synthetic central matter to the well profile interpretation are
seismic trace. According to White and Simm (2003), the log corrections, which can be challenging (Serra,
the methods to estimate the seismic wavelet are divided 1994). The change in well diameter along the drilling
into two categories: deterministic and statistical. The is a factor that can affect the measurements made by
deterministic method requires direct measurements the tool; therefore, corrections must be applied to pre-
of the source wavefield or the use of the well-log data serve the meaning of the log values (Ellis and Singer,
(Oldenburg et al., 1981; Yilmaz, 2000). Statistical meth- 2007). Although the stability of the wellbore is con-
ods estimate the wavelet from the seismic trace itself, trolled during the drilling process, nonpredicted oc-
and depending on the approach used, they require currences of formation collapse change the distribution
assumptions about the characteristics of the wavelet of the physical properties around the wellbore. The
(Buland and Omre, 2003; Lundsgaard et al., 2015), such consequence are anomalous values observed in the

1
Federal University of Pará (UFPA), Faculty of Geophysics, Faculty of Geophysics, Petrophysics and Rock Physics Laboratory, Belém, Pará,
Brazil. E-mail: [email protected]; [email protected]; [email protected].
Manuscript received by the Editor 8 January 2019; revised manuscript received 23 August 2019; published ahead of production 07 October 2020;
published online 6 December 2019. This paper appears in Interpretation, Vol. 8, No. 1 (February 2020); p. T43–T53, 9 FIGS.
http://dx.doi.org/10.1190/INT-2019-0004.1. © 2020 Society of Exploration Geophysicists and American Association of Petroleum Geologists. All rights reserved.

Interpretation / February 2020 T43

caliper log in the regions where the formation collapse where i represents the index of a sample in depth, Rc is
occurred. the reflectivity in the depth domain, v is the P-wave
Borehole enlargement implies log distortions, includ- velocity, and ρb is the bulk density. The reflectivity, cal-
ing incorrect density measurements that result from culated in-depth, needs to be placed in the time domain
mixing the formation components with the drilling mud by the proper establishment of the time-depth relation-
(Liu and Zhao, 2015). This interaction of the drilling ship, which can be achieved by check-shot surveys or
fluid and the formations around the wellbore is a rel- vertical seismic profile (VSP) data.
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

evant factor to be concerned for in-depth acquisitions, Correlation is a measure of coherence that can be
especially regarding the density log, whose precision used to compare the real and synthetic seismic traces
is directly related to the well-to-seismic tie response. for well tie purposes. Correlations between variables
In a situation in which the mud density is lower than can be measured with the use of different indices
the formation bulk density, an expanded borehole or (coefficients). The three most popular are Pearson’s
an irregular wellbore wall affect the density log curve coefficient, Spearman’s coefficient, and Kendall’s tau
so markedly that the curve drops precipitously, and coefficient (Hauke and Kossowski, 2011). The choice of
the measured density value is much lower than the true each correlation coefficient depends on the type of data
density value (Yong and Zhang, 2007). On the other being analyzed. Spearman’s and Kendall’s correlation
hand, in cases in which the mud density is greater than coefficients are used for interval or ordinal data. The
the formation bulk density, the opposite happens: The Pearson correlation coefficient, the one used in this
density curve would increase rather than drop, and the work, is appropriate only for interval data (Chok, 2010).
apparent density would be greater than the formation The Pearson correlation coefficient (r) measures
bulk density. the strength and direction of the linear relationship
Macedo et al. (2017) analyze the influence of the sta- between two independent variables, giving a sense of
bility of the borehole diameter during acquisition on the degree to which one variable is linearly related to
the well-to-seismic tie and show that anomalies on the another. Its values lie between 1 and are calculated
caliper logs can directly affect the quality of the tie according to the following equation:
and, consequently, the estimated wavelet. Within this
scenario, the present paper aims to analyze the well-to- PN
i¼1 ðxi − x̄Þðyi − ȳÞ
seismic tie response when the proper corrections on r ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ; (3)
2 2
the density log for the wellbore enlargement are made. i¼1 ðxi − x̄Þ i¼1 ðyi − ȳÞ
These corrections are based on Doll’s geometric factor.
To verify the feasibility of our proposed methodology, where N is the total number of samples, i is the index
we perform well-to-seismic tie on the real data set from of samples, and x̄; ȳ are the means. For the case of the
the Viking Graben field — the North Sea with and with- well-to-seismic tie, vector x represents the real seismic
out density log correction — and with two different trace, whereas vector y represents the synthetic seis-
wavelet estimation methods: a classical deterministic mic trace.
approach and a statistical approach through the predic-
tive deconvolution. Density log measurements
The bulk density curve is crucial for calculation of
Theory reflectivity. The density estimated by the logging tool
Well-to-seismic tie procedure is based on back-scattering (from Compton scattering)
The well-to-seismic tie procedure requires the for- gamma radiation emitted by a radioactive source such
ward modeling of the synthetic seismic trace. It is per- as Cs137 or Co60 . The gamma rays emitted from a source
formed by the convolution of the reflectivity series will interact with the electrons of the formation. The
created directly from the sonic log and bulk density higher the electron density of the formation, the higher
curves with the seismic wavelet. The synthetic seismo- is the number of collisions of the gamma rays with the
gram is constructed through the following equation: electrons and, consequently, the lower is the intensity
of gamma-rays detected by the sensor. The intensity of
sðtÞ ¼ wðtÞ rðtÞ; (1) gamma rays detected is expressed by

where sðtÞ represents the synthetic seismic trace, wðtÞ I ¼ I o e−μρe L ; (4)
represents the seismic wavelet, and rðtÞ is the reflectiv-
ity converted to the time domain. Before converting where I is the intensity of gamma rays detected by the
the reflectivity series to the time domain, one must cal- scintillometer, I o is the intensity of gamma rays at the
culate it in depth from the density and velocity logs. source, μ is a constant that depends on the geometry
Mathematically, the discretized reflectivity Rc in depth of the tool, ρe is the density of electrons of the forma-
is represented by tion, and L is the distance between the source and
the detector. The electron density ρe and the bulk den-
ρbiþ1 viþ1 − ρbi vi
Rc ðiÞ ¼ ; (2) sity ρb of the formation are related through the follow-
ρbiþ1 viþ1 þ ρbi vi ing equation:

T44 Interpretation / February 2020

Z effects of severe borehole enlargement along with the
ρe ¼ ρb N; (5)
A use of the caliper readings.

where Z is the atomic number, A is the atomic mass,

and N is the Avogadro number (6.023 × 1023 ). Because Methodology
for most of the formations ðZ∕A ≈ 1∕2Þ, equation 5 Correcting the density log for the borehole
becomes enlargement
To correct the density log for the borehole enlarge-
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

N ment through the caliper log readings, we used Doll’s

ρe ¼ ρb : (6)
2 geometric factor. Doll (1949) develops the apparent
geometric factor theory for the induction logging, which
By substituting equation 6 into equation 4, the intensity
was created to measure formation resistivity in bore-
of detected gamma rays becomes
holes containing oil-based muds and in air-drilled bore-
ρb N holes because electrode devices could not work in
I ¼ I o e−μ 2 L : (7) these nonconductive boreholes. For the induction log-
ging, the theory states that the voltage at the receiver is
Taking the logarithm of both sides of equation 7, we
the sum of the contribution of a large number of infini-
have the density of formation ρb in terms of the intensity
tesimal rings of the Foucault current. The geometric
of the gamma ray detected by the sensor in the logging
factor of each coaxial cylindrical area would represent
tool:
the fraction of contribution of this single area to the
2½LnI o − LnI entire signal, assuming a uniform conductivity within
ρb ¼ : (8) each zone.
μNL
The presence of unconsolidated formations com-
The lower the source-detector distance, the lower is promises the signals recorded by the logging tool. When
the depth of investigation, which makes the density the formation in the vicinity of the well is homogeneous
formation values more influenced by the borehole and there are no washout zones, the density measured
enlargement. Moreover, the depth of investigation by the logging tool is similar to the correct density of
also decreases with the increase in the density of the the formation. However, when the formation is hetero-
formation. geneous, the apparent density measured by the tool rep-
The first single-detector density tool measurements resents a combination of the densities of the different
suffered from the effects of mud-cake and were mar- formations that exist around the well. The influence of
keted in the mid-1950s (Labo, 1987; Serra, 1994). The each formation can be considered separately, and the
second-generation tool uses a two-detector system that signal measured is represented by the sum of each sig-
compensates for near-borehole problems (Labo, 1987). nal generated by each formation.
According to Serra (1994), the effect of the borehole is To properly use the apparent geometric factor theory
more severe for uncompensated density logs but less so for the density logging, it is necessary to establish the
for the compensated ones. The author also states that if geometry of the borehole environment we are assuming
the borehole wall is not smooth, the formation density in our methodology. The following assumptions are
compensated pad is not correctly applied to the for- made: (1) different zones around the well are concen-
mation and isolates zones full of mud, which strongly tric and, therefore, have a rotation symmetry, (2) the
affect the measurement. Therefore, even if the tool has geometry of the borehole is a cylinder of radius similar
a compensation system to mitigate the effect of the to the drill bit, and (3) the diameter of the borehole
rugosity of the borehole wall, if the borehole enlarge- maintained its size during the drilling procedure.
ment is severe, it will still affect the density log values. The vicinity of the well is composed of the rock
It can be assumed that the medium where the density formation and the drilling mud. As proposed by Doll
logging tool operates is composed of rock formations (1949) for induction logging, the density measurement
and drilling mud. The density measured by the tool can also be obtained from a weighted average of mud
is therefore derived from the formation density and and formation densities in equation 10 as a conse-
the drilling mud density. The traditional correction quence of the apparent geometric factor that satisfies
schemes for density has some limitations because be- the condition
yond some mudcake thickness (≈1in), depending on Gb þ Gmud ¼ 1; (9)
the tool design details that control, in part, the depth of
investigation, the compensation scheme breaks down, and
and the estimate of the bulk density will be in doubt
(Ellis and Singer, 2007). According to Ellis and Singer ρa ¼ Gb ρb þ Gmud ρmud ; (10)
(2007), the auxiliary measurement that is most helpful
to indicate suspicious density readings is the caliper. where Gb is the coefficient for the formation rock,
Motivated by those reasons, Doll’s geometric factor was 0 ≤ Gb ≤ 1; Gmud is the coefficient for the mud,
used as an attempt to correct the density log for the 0 ≤ Gmud ≤ 1; ρa is the apparent density ðg∕cm3 Þ; ρb

Interpretation / February 2020 T45

is the bulk density ðg∕cm3 Þ; and ρmud is the mud density levels. This means that on the portions of the density log
ðg∕cm3 Þ. If there is significant borehole diameter expan- where the enlargement of the borehole occurs, the den-
sion, all values are represented by the mud density. sity log values are underestimated. Hence, the correc-
According to equations 9 and 10, Gmud ¼ 1, Gb ¼ 0, tion for the geometry of the wellbore should increase
ρa ¼ ρmud ; in contrast, if the logging tool keeps contact the values of the density log on those portions. To cor-
with a regular wellbore wall, then Gmud ¼ 0, Gb ¼ 1, rect the density log for the geometry of the borehole, we
ρa ¼ ρb . Therefore, in terms of error estimation due to calculate the slope and intercept of the line of Figure 1:
borehole expansion, Gmud and ρmud are the fundamental
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

parameters that needed to be investigated. Calmax − Calmin

m¼ ; (12)
To determine the true values of density that repre- Gmax − Gmin
sent the subsurface formations, we derive from equa-
tions 9 and 10 the following expression: and

ρa − Gmud ρmud b ¼ Calmin − mGmin ; (13)

ρb ¼ ; (11)
1 − Gmud
where Calmax and Calmin represents the maximum and
which indicates the corrected value of bulk density, minimum value of the caliper log, and Gmax and Gmin
in terms of apparent density, mud density, and apparent represents the maximum and minimum value of the
geometry factor of mud. The method we used to analyze apparent geometric factor of the mud Gmud , with the
how the borehole effect affects the well-to-seismic tie minimum value being zero and the maximum value de-
results consists of search for a value for Gmud and pending upon the geology and the logging tool. For the
ρmud using equation 11 to investigate the proper values Viking Graben dataset (the real data used in this paper),
of formation density ðρb Þ. we set Gmax ¼ 0.4.
We performed the correction on the density log at Accordingly, the Gmud log is created through the
each point in depth by creating a linear relationship equation
between the minimum and maximum values of the cal- Cal − Calmin
iper log with the minimum and maximum values of the Gmud ¼ þ Gmin ; (14)
m
apparent geometric factor of the mud Gmud . From equa-
tion 10, one can note that when Gmud ¼ 0, the corrected where Cal represents each value of the caliper log in
bulk density is equal to the measured bulk density. The depth and Gmud is each value of Gmud created along the
choice of the maximum value of Gmud , related to the depth axis. By applying the relation 14 in equation 11,
maximum value of the caliper log, must be based on we generate the relation that we used to correct the
the depth of investigation of the density logging tool. bulk density value for each point in depth 15:
To create a log of Gmud values in-depth to proceed with
the correction of the bulk density log at each point in ρa − ½ðCal − Calmin Þ∕m þ Gmin ρmud
ρb ¼ : (15)
depth, a linear relationship between the caliper log and 1 − ½ðCal − Calmin Þ∕m þ Gmin
the Gmud values was established according to the equa-
tions described below and the graphic from Figure 1. In the situation where there is no enlargement of the
As the diameter of the borehole increases, the den- borehole (when the caliper log is stable), m ¼ 0 and
sity log value decreases; on the contrary, density log equation 15 becomes
values increase as borehole diameters approach normal
ρa − Gmin ρmud
ρb ¼ : (16)
½1 − Gmin

Also, in this situation, the minimum value of the ap-

parent geometric factor of the mud is Gmin ¼ 0. Hence,
when the caliper log is stable and the diameter of the
borehole maintains its bit size during the drilling, there
is no need to correct the measured bulk density be-
cause ρb ¼ ρa . The procedure to perform the correction
on the density log is shown in the flowchart of Figure 2,
and the complete procedure to perform the correction
and the well-to-seismic tie are described in the follow-
ing steps and in charts in Figures 2 and 3:

Figure 1. The linear relationship between the apparent geo- 1) Edit the sonic and density logs to remove noisy
metric factor Gmud and the caliper log, according to its maxi- spikes.
mum and minimum values. 2) Establish the range of possible values of ρmud .

T46 Interpretation / February 2020

3) Establish the value of Gmax and set Gmin ¼ 0. of the corrected density log, we used the same wavelet
4) Create the Gmud log in depth according to for the well ties with and without correction on the den-
equation 14. sity log. For the traditional deterministic wavelet estima-
5) Perform the correction on the measured density tion, different wavelets are extracted when different
log using the Gmud log and equation 15. density logs are generated. The results of the application
6) Calculate the reflectivity series with the corrected of the predictive deconvolution to estimate the wavelet
density log using equation 2. were also used by Macedo et al. (2017).
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

7) Estimate the seismic wavelets through the deter- The predictive deconvolution is a special case of
ministic and statistical approaches. the Wiener filtering, which requires the solution of the
8) Convolve the reflectivity series with the estimated normal equations:
wavelets to calculate the synthetic seismic trace. 0 10 1 0 1
9) Compare the synthetic seismic trace with the real r0 r1 r2 · · · r n−1 a0 g0
seismic trace through cross-correlation. B r1
B r0 r1 · · · r n−2 C B C B
CB a1 C B g1 C
C
10) Compare the correlation of the well-to-seismic tie B r2 r r · · · r C
n−3 CB
B a C
2 C¼B
B g C
2 C;
B 1 0
B .. .. .. .. .. C B . C B . C
. A@ .. A @ .. A
without correction on the density log with the cor- @ . . . .
relation of the well-to-seismic tie with the correc-
tion on the density log. r n−1 r n−2 r n−3 ··· r0 an−1 gn−1
(17)

Estimation of the wavelet where r i represents the autocorrelation lags of the in-
The estimation of the wavelet is a key feature of put wavelet, ai is the Wiener filter coefficients, and gi
the well-to-seismic tie. Deterministic methods to esti- are the crosscorrelations lags of the desired output with
mate the wavelet uses seismic and well-log data. The the input wavelet. The prediction process is assumed
deterministic methods based on the Wiener filtering when the desired output on the normal equations is a
technique, for instance, are based on the classic as- time-advanced form of the input series. In the case of
sumptions of the convolutional model, in which the the predictive deconvolution, given an input series
limiting factors are the premises of a
random process reflectivity and a
minimum-phase wavelet. Deterministic
Inputs

methods based on least-squares mini- Caliper and Set G mud(min) = 0

Set ρmud
mization circumvent these objections. density logs and G mud (max)
Statistical methods to estimate the
wavelet use only the seismic data. One
possible approach is through the predic-
tive deconvolution, which also implies
the assumption of a random process
reflectivity and a minimum-phase wave- G mud (max)
Log edition
let. It is possible to perform a statistical Priori information
estimate of the wavelet without these
limitations, by using, for example, the
homomorphic deconvolution. To main-
tain both estimates of the wavelet under
the same premises, we used a determin- Create G mud log
istic and a statistical estimation based
on the classical assumptions of the con-
volutional model. By fixing it, it is pos-
sible to analyze the well-tie response
only due to the effect of the correction
on the density log for the borehole en-
Output

Corrected
largement and analyze its behavior ac- density log
cording to the two general techniques
to estimate the wavelet.
The wavelets in this study were esti-
mated by two different approaches: a Figure 2. The flow diagram indicating the procedure to correct the density log
statistic estimation through predictive for the borehole enlargement. The first steps are the log editions for despiking,
and to remove null values, set a usual value for the ρmud , set the minimum value
deconvolution using the optimum Wiener of Gmud as zero, and set the maximum value of Gmud according to the prior in-
filter, and a deterministic extraction by formation and depth of investigation of the tool. With this information, it is pos-
building a filter, using the seismic trace sible to create the Gmud log that will be used to correct the density log for the
and well-log data. To study the influence borehole enlargement for each point in depth.

Interpretation / February 2020 T47

xðtÞ, the goal is to predict its value at some posterior Therefore, we selected a segment of the trace that we
time xðt þ αÞ, where α is the prediction lag. Using the believed to be a result of a reflector and deconvolved
normal Wiener equations to build a filter to estimate that segment to separate the white components from
xðt þ αÞ (the new desire output that must be cross-cor- the minimum-phase component, by choosing a predic-
related with the input xðtÞ), it gives tion lag and operator length. For a single segment in
0 10 1 0 1 time of the real seismic trace, we selected a range of
r0 r1 r 2 · · · r n−1 a0 rα prediction lag and a range of operator length and per-
B r1 r 1 · · · r n−2 C B C B C
r0 CB a1 C B r αþ1 C
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

B formed the deconvolution to estimate the wavelet. The

B r2 r1 r 0 · · · r n−3 CB a2 C ¼ B αþ2 C
C B C B r
B C: algorithm returns the prediction lag and operator length
B .. .. .. .. .. CB .. C B .. C
@ . . . . . A @ . A @ . A that produces the best wavelet that matches the syn-
r n−1 r n−2 r n−3 ··· r0 an−1 r αþn−1 thetic trace and the real trace.
The deterministic extraction of the wavelet is made
(18)
by the building of a filter that when convolved with
That is the case for an n-long prediction filter and an the reflectivity produces the best match between the
α-long prediction lag. The prediction filter requires only synthetic trace (provided by the convolution of the
the autocorrelation of the input series, which config- filter coefficients and the reflectivity), with the real
ures a statistical estimation of the wavelet. According seismic trace. The inputs to the deterministic wavelet
to Robinson and Treitel (2008), the canonical represen- extraction are the reflectivity series, the seismic trace,
tation of the seismic trace is the convolution of an all- the length of the filter (wavelet), and the increment in
pass filter — with a flat magnitude spectrum — and a time, which corresponds to the best shift between the
minimum-delay wavelet. The trace and the wavelet reflectivity series and the real seismic trace. For a given
have the same nonflat magnitude spectrum. The predic- range of length of filter and increment, the algorithm
tive deconvolution separates the components of the returns the best parameters that produce the wavelet
trace on the basis of the criteria of minimum-delay that best matches the synthetic trace and the real seis-
and white: The error series of the predictive deconvo- mic trace.
lution yields the white components of the seismic trace
(all-pass filter and the reflectivity), and the predic-
tion filter yields the predictable components of the Results on the Viking Graben data set
seismic trace, which is the minimum-delay component Geologic background
that constitutes the minimum-phase estimated wavelet. The Viking Graben is located in the North Sea Basin,
and it is the product of a rifting event during the late
Permian to the Triassic. The formation
Statistical of the oil and gas of the North Sea is
wavelet related to an extensional episode at
the beginning of Jurassic, which caused
Real Pangea to break into the continents
seismic trace Gondwana and Laurasia. Details regard-
ing the stratigraphy of the Viking Graben
Deterministic can be found in Madiba and McMechan
wavelet (2003), Keys and Foster (1998), and
Macedo et al. (2017). The data set used
in this study was the subject of a
Outputs
I nputs

Deterministic Statistical
Sonic log synthetic trace synthetic trace 1994 SEG workshop on comparison of
seismic inversion methods (Keys and
Foster, 1998). The seismic data consist
Corrected of a 2D seismic line, oriented east–west,
reflectivity
with 2142 CMPs separated by 12.5 m,
each CMP with 1501 samples and a sam-
Corrected ple rate of 0.004 s. The well-log informa-
density log tion is from two wells named well A and
well B. Macedo et al. (2017) perform
well-tie procedures using these logs
Figure 3. The flow diagram indicating the procedure to perform the well- and the same seismic section, verifying
to-seismic tie after the correction on density log. The corrected reflectivity is that the best match to the well data is
calculated, and two wavelets are estimated: a statistical wavelet through the pre- the CMP 809 for well A, which we also
dictive deconvolution, which uses only the real seismic trace; a deterministic
wavelet, from the corrected reflectivity and the real seismic trace. The convo- used as a real seismic trace. The well
lution of those wavelets with the corrected reflectivity will produce the statistical logs were edited to remove noisy spikes
synthetic trace and the deterministic synthetic trace, respectively. and null values.

T48 Interpretation / February 2020

Numerical experiments and discussion depth of investigation, which means that there is a
In this section, we present our results of the well-to- strong sensitivity of the tool to the near-borehole zone
seismic tie applied on the Viking Graben data set with presumed to be invaded by the drilling mud. Because of
the correction on the density log for the borehole that, it is necessary to add more weight to the zones
enlargement. We used two different ways to estimate farther from the borehole, where the true value of den-
the seismic wavelet: the deterministic approach and sity could be measured, to obtain a proper measure.
the statistical approach, through the predictive decon- Thus, the geometric factor of the formation Gb will
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

volution (Macedo et al., 2017). For the deterministic vary from 0.6 to 1. These weights are always higher than
seismic wavelet estimation, we performed the correc- the established mud weights, which configures an
tion on the density log in two ways: The first is cor- approach to compensate for the higher sensibility of
recting the entire density log, and the second is the density tool to the near-borehole zone. Moreover,
performing the correction only on the segments of by establishing those weights, we ensure that the cor-
the density log where the corresponding caliper log rected density log maintain its geologic consistency:
is unstable. We did that because the deterministic wave- The fixed density log measurements are not underesti-
let estimation is dependent on the reflectivity series. mated nor overestimated.
Therefore, different corrections pro-
duce different reflectivity series and, 3
Density (kg/m ) VP (m/s) Reflectivity Caliper (cm)
therefore, produce different determinis-
1000 2000 3000 0 5000 10000 −0.1 0 0.1 20 40 60
tic estimated wavelets. It was important 0 0 0
to verify when the correction is most ef- 1000
fective for the well tie: whether a correc- 2000 2000 2000 2000
tion to the entire density log or only in
3000
anomalous caliper segments. The statis-
tical wavelet estimation is dependent 4000 4000 4000 4000
Depth (m)

only on the seismic wavelet; therefore, 5000

different types of corrections on the 6000 6000 6000 6000
density log will not change the wavelet 7000
estimation — only the reflectivity 8000 8000 8000 8000
series to be used on the tie, which does
not have a meaningful impact (less than 9000
1%) on the final correlation between the 10000 10000 10000 10000
synthetic seismic trace and the real seis- 11000
mic trace. 12000 12000 12000
Figure 4 shows the well logs from the Original Despike
real data set from the Viking Graben
field used for the well-to-seismic tie. Figure 4. Real well-log data from the Viking Graben field used on the well-to-
The despiking procedure was applied seismic tie without the proper corrections on the density log and only with the
to remove the noise spikes. Figure 5 despiking to remove the noisy spikes.
shows the caliper readings, the original
and corrected density log, and the cor-
rection applied on the original density a) Caliper (cm) b) Density (g/cm3 ) c) Correction (g/cm 3 )
log, which was produced through our 1200
20 40 60
1200
1 2 3 4 5
1200
0 0.5 1

methodology using Doll’s geometric

1400 1400 1400
factor.
We test different values for the maxi- 1600 1600 1600

mum value of Gmud and set the maxi- 1800 1800 1800
mum value of Gmud in 0.4 for this
2000 2000 2000
case. By establishing that, the Gmud val-
Depth (m)

2200 2200 2200

ues will range from 0 to 0.4; conse-
quently, the Gb values will range from 2400 2400 2400

0.6 to 1. The value of the geometric fac- 2600 2600 2600

tor is related to the depth of investiga-
2800 2800 2800
tion of the logging tool. In the case of
the density tool, the depth of investiga- 3000 3000 3000

tion is small due to the short penetration 3200 3200 Corrected density log 3200
Original density
of the gamma ray into the formation. Ac-
cording to Ellis and Singer (2007), 90% Figure 5. (a) Caliper log used to calculated the correction to be applied on the
of the response of the density logging density log. (b) Original and corrected density log for the borehole enlargement.
tool is influenced by the first 4 in of (c) The calculated correction applied to the original density log.

Interpretation / February 2020 T49

Well tie without correction on density log—statistical wavelet
Amplitude (a.u.) 1 Figures 6 and 7 show the results of
0
the well-to-seismic tie for the CMP 809
of the Viking Graben field, with and
–1
1.4 1.6 1.8 2 2.2 2.4 without the correction on the density
Time (s) Real trace log. The results are divided into three
Synthetic trace
Correlation = 0.641492
sections: (1) well-to-seismic tie using a
statistical wavelet estimation and a cor-
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

rection on the density log only where

Real trace
Synthetic trace the caliper indicates a severe borehole
Correlation = 0.716044 enlargement, (2) well-to-seismic tie us-
Well tie with correction on density log—statistical wavelet ing a deterministic wavelet estimation
1
Amplitude (a.u.)

and a correction on the entire density

0
log, and (3) well-to-seismic tie using a
–1
1.4 1.6 1.8 2 2.2 2.4
deterministic wavelet estimation and a
Time (s) correction on the density log only where
the caliper indicates a severe borehole
Figure 6. Well-to-seismic tie using a statistical wavelet estimation and a correc- enlargement.
tion on the density log only where the caliper indicates a severe borehole The results show that for the three
enlargement. The correlation improved from 0.64 to 0.71 when performing cases, the correlation improved after
the correction.
the correction. The best improvement
a) was achieved for the statistical case:
Well tie without correction on density log—deterministic wavelet an increase of 7% on the correlation be-
Amplitude (a.u.)

1
tween the real and synthetic seismic
0 traces. In this case, because the statisti-
–1 cal estimation of the wavelet through
1.4 1.6 1.8 2 2.2 2.4
Time (s)
the predictive deconvolution does not
Real trace
Synthetic trace require the reflectivity series as an input,
Correlation = 0.714135
the same wavelet is used for the original
well-to-seismic tie and the corrected
Real trace well-to-seismic tie. The wavelet was es-
Synthetic trace
Correlation = 0.750478 timated by the predictive deconvolution
Well tie with correction on entire density log—deterministic wavelet of the segment from 1.38 to 1.6 s of the
Amplitude (a.u.)

1
real seismic trace, using the optimum
0 parameters returned by the semiauto-
–1 matic algorithm (Macedo et al., 2017),
1.4 1.6 1.8 2
Time (s)
2.2 2.4
with a prediction lag α ¼ 13 and filter
length N ¼ 14.
b) The well-to-seismic tie using the
Well tie without correction on density log—deterministic wavelet
deterministic estimation of the wavelet
Amplitude (a.u.)

1
show that the best correlation was
0
achieved when the correction on the
–1 density log is performed only on the
1.4 1.6 1.8 2 2.2 2.4
Time (s) Real trace segment where the caliper indicates a
Synthetic trace
Correlation = 0.714135 severe borehole enlargement and the
rest of the density log remains as the
Real trace original, with an improvement of 6% on
Synthetic trace
Correlation = 0.776533 the correlation between the real and the
synthetic seismic traces. As in this case,
Well tie with correction on segmented density log—deterministic wavelet
a different density log (because of the
Amplitude (a.u.)

0
correction) results in different reflec-
tivity series, and the deterministic esti-
–1
1.4 1.6 1.8 2 2.2 2.4 mation of the wavelet requires the
Time (s) reflectivity series as an input, wherein
different wavelets are extracted. The
Figure 7. (a) Well-to-seismic tie using a deterministic wavelet estimation and a wavelets used for the deterministic case
correction on the entire density log. The correlation improved from 0.71 to 0.75
when performing the correction. (b) Well-to-seismic tie using a deterministic
are shown in Figure 8.
wavelet estimation and a correction on the density log only where the caliper Although the density of the drilling
indicates a severe borehole enlargement. The correlation improved from 0.71 mud is a known factor during the com-
to 0.77 when performing the correction. pletion of the well, and its density varies

T50 Interpretation / February 2020

according to the conditions of the borehole, we ana- It is observed that the greatest influence of ρmud on
lyzed how the ρmud affects the correlation between the well-to-seismic tie was for the case where a correc-
the real and synthetic seismic traces after the cor- tion on the anomalous segment of the caliper log is
rection on the density log. We used a range from ρmud ¼ made, and using the deterministic wavelet, where a
1.10 g∕cm3 to ρmud ¼ 1.30 g∕cm3 to contemplate the change from ρmud ¼ 1.10 g∕cm3 to ρmud ¼ 1.30 g∕cm3 ,
typical values for the density of the drilling mud. The causes the correlation between dropping approxi-
results are shown in Figure 9. mately 1.5%. For all of the other cases, the change in
ρmud causes a change of less than 1%
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

on the correlation between the real

a) 0.5
Original Deterministic Wavelet
and synthetic seismic traces.
0
Conclusion
–0.5 We have proposed a method to cor-
0 0.5 1 1.5
b) Deterministic wavelet estimated with the correction on the entire density log rect the density log for the borehole
0.5
enlargement at each point in depth using
Amplitude (a.u.)

0 Doll’s geometric factor theory to obtain

a more accurate well-to-seismic tie for
–0.5
0 0.5 1 1.5
the real data set from the Viking Graben,
c) Deterministic wavelet estimated with the correction on the segmented density log North Sea. The first step is to set the
0.5
minimum and maximum values of Gmud
0 and associate them to the caliper read-
ings to generate the Gmud profile accord-
–0.5
0 0.5 1 1.5 ing to prior information and the depth of
Time (s) investigation of the logging tool. The
Gmud log represents the geometric factor
Figure 8. The extracted deterministic wavelets used on the well-to-seismic tie
in Figure 7. (a) The original wavelet. (b) The wavelet used in the case of cor-
relative to the density of the drilling mud
rection on the entire density log. (c) The wavelet used in the case of correction that interacts with the formation on the
on the segment where the caliper is unstable. washout zones. This procedure makes
it possible to associate the geometric
factor of the mud with the segments in which there
Influence of ρmud on the correlation for the different wavelets is enlargement of the borehole diameter. Those specific
0.78
areas have their density log corrected, and we then per-
0.76 formed the well-to-seismic tie. We estimate the seismic
wavelet through a statistical and a deterministic ap-
0.74
proach: For the statistical case through the predictive
Correlation

0.72 deconvolution, only the seismic trace is used, whereas

0.7
for the deterministic case, the seismic trace and the
Deterministic Correc. 1
Deterministic Correc. 2
reflectivity — calculated with the corrected density
0.68 Deterministic Original
Statistical Correc.
log — are used. Our analysis agrees with previously
0.66
Statistic Original
published works that an anomalous segment of caliper
contributes to lower the correlation on the well-to-
0.64
1.1 1.15 1.2 1.25 1.3
seismic tie. The results showed that both ties had their
ρ (g/cm ) 3 correlations increased with the correction on the den-
mud
sity log.
Figure 9. The influence of ρmud on the correlation between However, it is recommended to test different values
the real and synthetic seismic traces. Deterministic correction for the maximum Gmud according to the depth of inves-
1 (solid light-gray line) indicates the correlation for the case in tigation of the logging tool: It is necessary to establish a
which the correction is made on the entire density log using higher weight for the physical property that the logging
the deterministic wavelet; deterministic correction 2 (the solid
dark-gray line) indicates the correlation for the case where the
tool has the lowest sensitivity. Because the density log-
correction is made on the segment where the caliper is unsta- ging tool is more sensitive to the near-borehole zones,
ble using the deterministic wavelet; the deterministic original it is recommended that the geometric factor of the for-
(dashed black line) indicates the original correlation of the mation far from the borehole be higher than the weight
well-to-seismic tie when no corrections are made using the of the washout zone to avoid unrealistic densities val-
deterministic wavelet; statistic correction (the solid black ues. The limiting assumptions of this methodology is
line) indicates the correlation for the case in which the cor- that the geometric factor G varies linearly with the mud-
rection is made on the segment where the caliper is unstable
using the statistical wavelet; statistic original (the dashed cake thickness and that it does not depend on formation
light gray line) indicates the original correlation of the well- properties.
to-seismic tie when no corrections are made using the statis- For further work, it is worthwhile to apply this meth-
tical wavelet. odology of correction on density log on another data set

Interpretation / February 2020 T51

that also has an anomalous caliper due to severe en- Macedo, I. A. S., C. B. da Silva, J. de Figueiredoa, and B.
largement of the borehole. Moreover, we also suggest Omoboyac, 2017, Comparison between deterministic
verifying how this methodology behaves with different and statistical wavelet estimation methods through pre-
methods of wavelet estimation on the well-to-seismic dictive deconvolution: Seismic to well tie example from
tie, especially the ones that do not imply a minimum- the North Sea: Journal of Applied Geophysics, 136,
phase wavelet. A comparison of this approach of 298–314, doi: 10.1016/j.jappgeo.2016.11.003.
correction with the traditional spine-rib plot is also Madiba, G. B., and G. A. McMechan, 2003, Processing, in-
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

recommended, as well as a deeper study of how to version, and interpretation of a 2D seismic data set from
circumvent the limitations of this methodology. the North Viking Graben, North Sea: Geophysics, 68,
837–848, doi: 10.1190/1.1581036.
Acknowledgments Oldenburg, D., S. Levy, and K. Whittal, 1981, Wavelet esti-
The authors would like to thank Exxon Mobil for mation and deconvolution: Geophysics, 46, 1528–1542,
providing the Viking Graben data set. This work was doi: 10.1190/1.1441159.
kindly supported by the Brazilian agencies INCT-GP Robinson, E. A., and S. Treitel, 2008, Digital imaging and
and CNPq and the Geophysics Graduate Program at deconvolution: The ABCs of seismic exploration and
the Federal University of Pará. We would also like to processing: SEG.
thank the reviewers and editor for the useful sugges- Serra, O., 1994, Fundamentals of well logging interpre-
tions and corrections. tation — 1. The acquisition of logging data: Elsevier,
Development in Petroleum Science 15A.
Data and materials availability White, R., and R. Simm, 2003, Tutorial: Good practice in
Data associated with this research are available well ties: First Break, 21, 75–83.
and can be accessed via the following URL: Note: A White, R. E., and T. Hu, 1998, How accurate a well tie can
digital object identifier (DOI) linking to the data in a be?: The Leading Edge, 17, 1065–1071, doi: 10.1190/1
general or discipline-specific data repository is strongly .1438091.
preferred. Yilmaz, O., 2000, Seismic data analysis: Processing, inver-
sion, and interpretation of seismic data: SEG.
References Yong, S. H., and C. M. Zhang, 2007, Logging data processing
Buland, A., and H. Omre, 2003, Bayesian wavelet esti- and interpretation: Petroleum University Press, 59–61.
mation from seismic and well data: Geophysics, 68,
2000–2009, doi: 10.1190/1.1635053.
Chok, N. S., 2010, Pearson’s versus Spearman’s and Ken- Isadora A. S. de Macedo received a
dall’s correlation coefficients for continuous data: Ph.D. B.S. (2014) from the Federal Univer-
thesis, University of Pittsburgh. sity of Pará (UFPA) with a year
Doll, H. G., 1949, Introduction to induction logging and abroad at the University of California
application to logging of wells drilled with oil base mud: - Riverside (UCR); an M.S. (2015) and
Journal of Petroleum Technology, 1, 148–162, doi: 10 a Ph.D. (2019) in geophysics from the
.2118/949148-G. Federal University of Pará (UFPA).
Currently, she is a postdoc researcher
Ellis, D. V., and J. M. Singer, 2007, Well logging for earth
in geophysics also at the UFPA. She is
scientists, 2nd ed.: Springer. a member of SEG and SBGf-Brazil. Her research interests
Hauke, J., and T. Kossowski, 2011, Comparison of values of include geophysical modeling and inversion (particularly
Pearson’s and Spearman’s correlation coefficients on seismic and electromagnetic methods), petrophysics, and
the same sets of data: Quaestiones Geographicae, 30, well-to-seismic tie.
87–93, doi: 10.2478/v10117-011-0021-1.
Keys, R. G., and D. J. Foster, 1998, Comparison of seismic
inversion methods on a single real data set, open file José Jadsom S. de Figueiredo re-
publications: SEG. ceived a B.S. (2006) in physics from
Labo, J., 1987, A practical introduction to borehole geo- the Federal University of Paraiba
physics: An overview of wireline well logging principles (UFPB), an M.S. (2008) in physics,
for geophysicists: SEG, Density Log Principles 8. and a Ph.D. (2012) in petroleum sci-
Liu, Z., and J. Zhao, 2015, Correcting hole enlargement ence and engineering from the State
University of Campinas (UNICAMP).
impacts on density logs for coalbed methane reservoirs:
While working toward his Ph.D., he
The Open Petroleum Engineering Journal, 8, 72–77, doi: spent one year at Allied Geophysical
10.2174/1874834101508010072. Laboratories at the University of Houston. In October
Lundsgaard, A. K., H. Klemm, and A. J. Cherrett, 2015, Joint 2012, he joined the Faculty of Geophysics at the Federal
Bayesian wavelet and well-path estimation in the University of Pará as a permanent professor. He is a
impedance domain: Geophysics, 80, no. 2, M15–M31, member of SEG, EAGE, and SBGfBrazil. His research
doi: 10.1190/geo2014-0378.1. interests include seismic imaging methods (particularly

T52 Interpretation / February 2020

diffraction imaging), physical modeling of seismic phe-
nomena, anisotropy, rock physics, petrophysics, and
well ties.

Matias Costa de Sousa received

a B.S. (2017) and a master’s degree
(2019) in geophysics from UFPA. He
Downloaded 12/22/19 to 128.114.34.22. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

is currently a Ph.D. student of the

Postgraduate Programme in Geophys-
ics (CPGf), where he studies method-
ologies of correlating seismic and
well-log data, especially for geopres-
sure analysis. His research interests
include seismic imaging, electromagnetic methods, bore-
hole geophysics, and rock physics.

Interpretation / February 2020 T53