0% found this document useful (0 votes)
14 views18 pagesCap02 Chopra
Uploaded by
ben ahmed AmiraCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
14 views18 pagesCap02 Chopra
Uploaded by
ben ahmed AmiraCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 18
Chapter 2
Volumetric Dip and Azimuth
Chapter Objectives
After reading this chapter, you will be able to
lateral resolution
plane of view
+ evaluate alternative algorithms for calculating volumetric dip and azimuth in terms of accuracy and
+ interpret shaded-relief and apparent-dip images to delineate subtle structural features
+ apply composite dip/azimuth/seismic images to determine how a given reflector dips in and out of the
Introduction
‘Second to time-structure and amplitude-extraction maps,
dip and azimuth maps of interpreted seismic reflectors ar-
zguably are the most important product in the interpretation
of 3D seismic data. Originally described by Dalley et al.
(1989), dip and azimuth maps, along with closely related
shaded-relief maps (Bames, 2008), can highlight subtle faults
that have throws of less than 10 ms, as well as stratigraph
‘features that manifest themselves through differential com-
paction or through subtle changes in the seismic waveform.
‘AS we discussed in Chapter 1, Lisle (1994) and Hart et al.
(2002) demonstrated the relationship between reflectoreurva-
ture and fracture density. Unfortunately, variability in reflec
tor waveform, as well as seismic noise, can cause difficul-
ties with attribute extractions made along picked horizons
(Hesthammer and Fossen, 1997)
Because of recent advances in algorithm development,
now we can calculate 3D cubes of reflector dip and azimuth,
without explicitly picking a given horizon. The carlest
published work we have found on estimating dip for inter-
pretation purposes, working directly from seismic data, is
by Picou and Utzman (1962). They used a 2D unnormal-
ized crosscorrelation scan over candidate dips on 2D seis-
‘ic lines. Marfurt et al. (1998) generalized a later sem-
blance-based scan by Finn (1986) to a true 3D scan. Barnes
(4996, 2000a) presented an alternative approach based on
3D complex-trace analysis that Scheuer and Oldenburg,
(1988) originally had applied to velocity analysis, whereas
Bakker et al (1999) presented an estimate based on the gra-
dient structure tensor (GST)
27
Dip and azimuth volumes can be very valuable inter
pretation tools. Currently their most important use isto de=
fine a local reflector surface upon which we estimate some
measure of discontinuity, or conversely, along which we
filter the data to extract their continuous component. Ex-
amples of the former include the various coherence and
edge-detection measures (e.g, Marfurt ct al, 1998; Marfurt
etal, 1999; Luo etal, 1996, 2001). Examples ofthe later
include conventional f-x-y deconvolotion, and structurally
ordered filtering (Hoecker and Fehmers, 2002) (which al-
tcrnatively is called edge-preserving smoothing [Bakker et
al, 1999; Luo et al., 2002).
Because of velocity distortions, estimates of reflector
ip and azimuth from time-migrated seismic cubes are only
loosely related to tre dip and azimuth at depth. Even est
‘mates calculated from prestack depth-migrated data will
suffer from errors in the background-velocity model. Nev-
ertheless, because they are differential rather than absolute
measures of changes in reflector depth, dip and azimuth maps
ae less sensitive to long-wavelength errors in the velocity
‘model than are reflector depth measurements. Furthermore,
‘most interpretations of dip and azimuth calculations are
‘done on changes in dip and azimuth — through color dis-
play (Marfurt etal, 1998; Lin etal, 2003), through visual-
ization tools such as shaded-reliet projections (Barnes,
2003), or through explicit calculation of higherorder de
rivatives (Marfurt and Kirlin, 2000; Luo et al, 1996; Al-
Dossary andl Murfurt, 2003) that are sensitive to reflector
curvature oF rotation. Recently, Barnes (20000) developed
suite of computer-generated textures similar to those used 28 Seismic Atibutes for Prospect Identification and Reservoir Characterization
in traditional interpreter-driven seismic stratigraphy that
‘measure reflector convergence, divergence, parallelism, and
disorder, and that are based on an underlying estimate of
dip and azimuth,
Definition of Reflector
Dip and Azimuth
‘Mathematically, a planar element of a seismic reflector
‘ean be defined uniquely by a point in space, x = (2). and
a unit normal (0 the surface, m= (nym) , Where Me. Ny
‘and n, denote the components along the x, y, and z axes, r¢-
spectively, and are chosen such that n,> 0 (Figure 1).
Geologically, we define a planar interface such as a
formation top or internal bedding surface by means of ap-
parent dips 6, and 6,, or more commonly, by the surface’s
true dip, @, and its strike, y (Figure 1). Apparent dip 8, is
the angle measured in the vertical (x, 2) plane from the hori-
zontal x-axis to the interface. Similarly, apparent dip 6, is
the angle measured in the vertical (2) plane from the hori-
zontal y-axis to the interface. The strike, , is the angle be
‘ween north (the y-axis in the SEG-Y trace header conven
tion) and the intersection between the reflector and the
horizontal (x,y) plane, The true dip, 6, is always greater
than or equal to apparent dips 6 and @, and is the angle
‘measured in a vertical plane perpendicular to the strike be-
tween the horizontal plane and the interface. Geologic dips
have no sign and always are measured downward from the
Figure 1. Mathematical, geologic, and seismic nomenclature
used in defining reflector dip, By convention, n = unit veetor
normal to the reflector; a = unit vector dip along the reflec-
tor 6
apparent dip inthe xz plane; and @, = the apparent dip in the
ve plane.
horizontal plane to the surface. Because the strike defines a
line (¢.g., northeast-southwest) rather than 2 vector, we
need to state, or more commonly to post on map, the di-
rection of downward dip (e.g., northwest or southeast).
In seismology, we avoid such mathematical ambiguity
and commonly define a reflector by its dip and azimuth,
Dip, which is more explicitly called dip magnitude, 8, is
‘identical to that used in the above geologic definition. Azi-
uth, @, sometimes called dip azimuth, is measured either
from the north, or for convenience, from the inline seismic
survey axis. Azimuth is perpendicular to the geologie strike
and is measured in the direction of maximum downward
dip. In this book, we will use the seismic notation of dip
and azimuth, as well as apparent dips along the survey axes,
@, and 8, to define the reflector dip unit vector, a, where
a,= sind cosy, Qua)
nO sing, Q.1b)
and
v0.98. Qc)
Although theoretically the different measures of the plane
are equivalent, using its normal, n, its dip and strike, @ and
'p, and its apparent dips, & and 6, its dip and azimuth, @
and @. respectively, or is vector dip, a, such equivalent mea
sures can differ when stored with finite accuracy in an in
terpretation workstation. In particular, azimuth and stike
are undefined for a horizontal reflector. In contrast, the re
Aector normal and its components alway’ are defined
‘Without knowing the velocity ofthe earth, we often find
it convenient to measure the apparent seismic (two-way)
time dips, p and g, where pis the apparent dip measured in
sim (or s/t) in the inline, or x direction, and q is the appar-
‘ent dip measured in s/m (or s/t) in the crossline, or y direc
tion. Ifthe earth can be approximated by a constant veloci-
4, vs the relationships between the apparent time dips p and
4.and the apparent angle dips and 8, are
p=2tandlv, (22a
and
q=2 tanOyo. (2.26)
Alternative Means of Computing
Volumetric Dip and Azimuth
‘There are three popular methods of estimating volumet-
ric dip and azimuth from uninterpreted seismic data volumes.
‘These methods require (1) aligning the phase derived from
‘complex-trace analysis, (2) discretely scanning for the most
coherent planar reflector, or (3) crosseorrelating the gradient
ofthe data and forming a gradient structure tensor, Calculation of vector dip using
complex-trace analysis
Luo et al. (1996) and Barnes (1996) deseribed a meth-
od of estimating vector dip based on @ 3D extension of the
analytic trace (or complex-trace) attributes described in
Chapter 1. They began with Taner et al's (1979) instanta-
neous frequency,
au nu
ceed an a. 02)
Be? ae ar wre
where ® denotes the instantaneous phase, u(t») denotes
the input seismic data, w(x.) denotes its Hilbert trans-
‘Sem with respect to time, ¢, and where ATAN? denotes the
‘=ctancent function whose output varies between —7t and
‘== The derivatives of w and w" are obtained cither by using
Site Wihreraes wim a route wansrorm, win oe eon
rier transform approach being particularly convenient be-
‘cause this is the domain in which the Hilbert transform
typically is calculated,
‘Next, we calculate the instantaneous wavenumbers Ay
and ky:
a)
and
(2.4b)
For very large 3D-input seismic data cubes, it is more
convenient to estimate the spatial derivatives, 24 24" aie
a ay” ay "ay
and 2, ig ee tlle ot les
short Fourier transform, thereby circumventing the need to
keep the entire data cube in memory. Alternatively, itis
convenient to transpose the cube prior to calculating the de-
rivatives given in equation 2.4. Then, the instantaneous
1500)
ae
Volumetric Dip and Azimuth 29
time dip (p.q) is obtained by calculating the ratio of ky and
ky tow:
p=kio. (25a)
and
(2.5b)
and the azimuth , measured from the y-axis, and true time
dip, s, are given by
= ATAN2G.p), (26a)
ne (2.6b)
If te input data are inthe depth domain rather than in
the time domain, we would calculate rather than a
ae yy
a
a
kanye oe,
ee ae a ee
and
Qn
where w” now is the Hilbert transform with respect to depth,
<,allowing us to estimate angular dips 8, and 0:
6, = tan ek, 28a)
6,= tan ksfk). (2.80)
8 = can (2 +k), 28)
and
@ =ATAN2y ky). (2.80)
In Figure 2, we show vertical and horizontal slices
through a seismic survey acquired over a salt dome. In Fig-
ure 3, we show the instantaneous dip magnitude and azi-
ruth given by equations 2.8¢ and 2.8d. Taner et al. (1979)
‘warmed that the estimate of instantaneous frequency given
by equation 2.3 suffers from singularities when reflector
events interfere with each other. To remedy this inaccuracy,
‘Taner et al, (1979) suggested replacing equation 2,3 with
an envelope-weighted average. Indeed, such singularities
form the basis of the recently introduced SPICE (spectral
imaging of correlative events) algorithm (Liner et a, 2004),
which we will discuss in Chapter 6.
Amplitude
Positive
o
i.
Figure 2. A vertical slice and a
horizontal depth slice through a
seismic data Volume gathered over
asalt dome. Note the characteris-
tic “cutconion” appearance on the
depth slice, After Barnes (2000s)
<—_75km ——> 30 Seismic Attivutes for Prospect Identification and Reservoir Characterization
‘Barnes (2000a) proposed smoothing the calculation of
(© ky and k, over 25 oF more adjacent traces prior to esti-
‘mating dip and azimuth, thereby obtaining improved stabil-
ity at the expense of some loss of lateral resolution. We
show the results of such smoothing in Figure 4. The speck-
led nature of the instantaneous dip magnitude and azimuth
images is diminished, and we obtain a smoother (albeit
Iower-resolution), more-realistic image.
Calculation of vector dip
by discrete scans
Marfurt et al, (1998) generalized Finn's (1986) sem-
lance scanning method to 3D data, to generate a more ro-
bust means of estimating reflector dip (Figure 5):
29)
me
1
+
LS Luar pas, ~ea I}
Figuto 3. Vertical and horizon
tal slices corresponding to those
shovin in Figure 2 through (a)
Instantaneous dip magnitude,
given by equation 28c, and (b)
instantaneous dip azimuth, given
by equation 2.84. The speckled
appearance of adjacent conflicting
‘ip magnitude and dip azimuth
are the result of singularities in
the complex-trae analysis est-
ration of ky, ky, and e. After 4500
Bares (20003)
a) 1800
where p and q are given by equation 2.2, x; and y; denote
the local coordinates of the jth trace measured from an ori-
ain at the analysis point, J denotes the total number of trac-
cs in the analysis window, and K, and K, denote the first
and last temporal sample, respectively, in the analysis win-
dow. Other amplitude-normalized measures include princi-
ppal-component colierence (Gerszienkorn and Marfurt, 1999),
lateral variance normalized by the energy, and a least-squares,
fitto the data (Bednar, 1998), Details on coherence calcula-
tions, including equation 2.9, will be discussed in Chapter
3. A disadvantage of dip scan approaches is that they dis-
cretely sample 8, and 6,, such that we may miss subtle fea-
tures that might be discerned by the continuous phase esti
‘mated using the complex-trace analysis method given by
equation 2.7.
Calculation of vector dip using
the gradient structure tensor
‘The third method of estimating vector dip is based on
the gradient structure tensor (GST) and is used by Bakker et
al, (1999) and by Hoecker and Fehmers (2002) in their struc
‘ture-oriented filtering work. The goal ofthis method is to de-
fine, within a small analysis window, the direction in which
the seismic data vary the greatest. If our window contains a
constant-amplitude planar reflection, that direction will be
the normal to the plane. There willbe two other axes that are
perpendicular to the axis having the most variation. We ad- Jjust one of those two axes to be aligned with the direction of
minimum variation, The remaining axis is perpendicular to
‘the other two. Mathematicians would call those three axes
the prineipal axes defining the geometric structure.
To perform the calculation, first we generate a measure
‘of waveform variability along cach of three Cartesian axes,
au/ax, Gu/ay, and du/dz, Bakker (1999) recommended cal-
culating these derivatives by convolving the seismic data
with the derivatives of a Gaussian filter 4G/ax, 0G/0y, and
8G/ac, where
Gx jy, 03,50) = expla; +; +
+ 10)
and where x, yj, and zjare the distances along the x-.¥-, and
=-axes of the jth trace from the point at which the derivative
is being analyzed, and o isthe scale parameter. Because we
apply this derivative in each ofthe three directions and con-
volve it with the original seismic data, we now have three
seismic-amplitude-gradient volumes. If the data are sam-
pled optimally, Bakker suggested using a value of o that is
three times the sample, CMP, and line spacing.
Next, we define an analysis window, much as Bames
(2000a) did for his weighted-average estimates of dip and
azimuth and Marfurt etal, (1998) did for their discrete sem-
blance search for dip and azimuth. If we had a perfectly flat,
plane, we would expect the values of 31/02 at each point in
the analysis window to be identical to the ones laterally
next to it. The other two gradients would be zero. If we had
a vertically planar, constant-amplitude reflector that is per-
a) 1500
ie
€
é
4500
Volumetric Dip and Azimuth 31
pendicular to the axis, we would expect cach value of
‘awax to be identical 0 the values above and below it.
Mathematicians would call each triplet of dx, au/ay, and
‘waz a sample vector of the space being analyzed. To ex-
tract any consistent trend, we compare each triplet with all,
other triplets falling within the analysis window. We do this,
by simple crosscorrelation, resulting in the following cova-
riance matrix:
1
Ta" jaKeD
where the analysis window includes £K samples and J trac-
ces, We call this 3 x 3 covariance matrix, Tes, the gradient
structure tensor: The gradient structure tensor contains in-
formation on the waveform variability within the analysis
‘window, Because we have formed the matrix with Carte-
sian components of amplitude variability, we turn to cigen-
Figure 4. Vertical and horizontal
slices corresponding to those shown in
Figure 3, here through (a) a smoothed
instantaneous dip magnitude and (b) a
‘smoothed instantaneous dip azimuth
‘Smoothing was performed using an
cenvelope-weighted running mean fil-
ter (composed of five inlines, five
crosslines, and seven depth samples)
‘on each of the constituent ky ky, and @
‘components. Note that the singulari-
ties seen in Figure 3 are diminished,
‘After Bares (20008).
Dip”)
[
0
aimuth
360
180
° oe
32 Seismic Attrioutes for Prospect Identification and Reservor Characterization
decomposition and write
(2.12)
‘where Vm are the three eigenvectors and Jy, are the corre
sponding three eigenvalues of the 3 x 3 gradient structure
tensor. By convention, the first eigenvector, vs, defines the
direction of maximum variability and thus measures the di-
tection normal to the best plane representing our seismic
data-analysis window. For a perfect plane, 22 = j= 0, ¢-
gardless ofthe planar orientation. We will discuss eigenval-
tues and eigenvectors in greater detail in Chapter 3 on co-
hetence. For this chapter, you need to know that its vertical
‘and lateral resolutions are dependent first on the size of the
Figure §. (a) A schematic diagram showing 2D search-
‘based estimate of coherence. First, the algorithm estimates
coherence using semblance, variance, principal component,
fr some other statistical measure (such as that given by equa-
tion 2.9) along a discrete number of candidate dips (shown
in magenta and green). In this example, the maximum coher-
cence is calculated along the dip (shown in dark green). Next,
the algorithm passes an interpolation curve through the
‘coherence measures estimated by the peak value and two or
‘more neighboring dips (shown here in light green). The peak
value of this curve gives an estimate of coberence, whereas
the dip value of tis peak gives an estimate of instantancous
dip, (b) A schematic diagram showing a 3D search-based
cstimate of coherence, in which p indicates the inline and ¢
the crossline components of vector time dip. The technique is
analogous to that shovin in (a). After Marfurt etal. (1998).
derivative calculation (a function of the value of @ in equa-
tion 2.10) and second on the size of the analysis window.
Multiwindow Estimates
of 3D Vector Dip
Regardless of how such estimates of reflector dip are
implemented, they all fail when the analysis window spans
4 fault In that situation, they provide at best an estimate of
the apparent dip across the fault, rather than of the true re~
flector dip. Such estimates of apparent dip across faults are
excellent edge detectors (e.g., Lu et al, 1996; and shown.
in Figure 3 in Chapter 5 of this book). We will discuss more
details of these aspects in our discussion of computing co-
hotence for highlighting faults and stratigraphic features in
3D seismic data volumes, the topic of Chapter 3. Smoothed
estimates of the vector components of dip using mean or
‘median filters can improve estimates of coherence (Marfurt
et al., 1999), but they do so at the expense of eliminating
details of interest in the reflector dip volumes themselves.
‘The discrete scan given by equation 2.9 is performed
fon a regularly sampled grid. Instead of simply assuming
that the refleetor dip and azimuth correspond to the maxi-
mum discretely scanned value, we fit a quadratic surface
through the semblance values at the neighboring apparent-
dip pairs (0, 6,), of the form
6(0,.0,)= a0) +2,0,0, 40,0, +4,0, +08, +4. 2.13)
and solve for the coefficients cy in a least-squares sense.
We then calculate an improved estimate of the vector dip
by solving
6,008, are
000) 2, ,40,6,40,=
and €4(0,,0,) io
1,6,+2a,6/+a,=0+ (2.14)
or 05), where (6,03) is the apparen-dip pai come-
Sponding to the maximum of the inferpolited semblance
surface, (61.8).
Even if our semblance surface is interpolated, our analy-
sis window may inadvertently span fault dividing domains
having different ips Figure 6). To obtain an improved
timate of veetor dip, we exploit the multpl-anaysi-win-
dow construct described by Kuwahara etal. (1976) and, t0
tur knowledge, first generalized to seismic amplitudes by
Tuo etal. (2002) in ther edge-preserving smoothing alg0-
ithe, In the latter work, Luo etal, (2002) seanned a suite
‘of noncentered, overlapping analysis windows in addition to the centered window, all of which contained the analysis
point of interest (Figure 7a). They then calculated the am-
plitude variance over the J traces that fall within the ith
window:
15)
where denotes the average value of ty within the ith
analysis window. The window having the smallest variance
svill be assumed to best represent a coherent reflector and will
be used in subsequent structure-oriented filtering, which we
‘will discuss in Chapter 8 on seismic image enhancement,
Because we expect amplitudes themselves to change
‘across a discontinuity, we use an energy-normalized coher-
fence estimate (such as the semblance measure given by
‘equation 2.9) rather than the unnormalized variance given
by equation 2.15. Although Luo et al. (2002) did not state it
explicitly, clearly we also need to search over a suite of
candidate dips and choose the window with the maximum
coherence using equations 2.9, 2.13, and 2.14. Finally, we
generalize the original concept of Luo et al. (2002) by also
searching over candidate uncentered-vertical-analysis win-
dows (Figure 7b) that include the analysis point. Figures 8
and 9 display a comparison of the instantaneous dip, the
‘envelope-weighted smoothed dip, and the multiwindow
calculations on a vertical slice and a time slice from Vinton
Dome, Louisiana, U.S.A. The multiwindow-
