t Technical papers

Similarity attributes from differential resolution components

Bruno César Zanardo Honório1, Ulisses Miguel da Costa Correia1, Marcílio Castro de Matos2,
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

and Alexandre Campane Vidal1

Abstract
Seismic resolution plays an important role in the delineation of structural and stratigraphic features. The
resolution improvement directly affects the seismic attributes and, consequently, the interpretation of a given
feature. However, the broadband data do not necessarily provide the best insight for seismic attribute evalu-
ation. Particularly, geologic discontinuities, such as karsts, faults, and fractures, can have different seismic ex-
pressions according to their intrinsic scales, and, therefore, they are better illuminated in a given frequency
range. To extract dissimilar characteristics in different frequency bands, we have combined a recently devel-
oped spectral enhancement method based on differential resolution (DR) and similarity attributes. The DR al-
gorithm is simultaneously used for frequency enhancement and acting as a pseudofilter, allowing us to compute
similarity attributes at different frequency bands. The similarity computation follows the reflector dip of each
DR subband and adjusts its analysis window accordingly to the dominant frequency within the subbands. Then,
the subband similarities are combined in the red-green-blue-alpha color space, allowing a more detailed view of
the geology under investigation. Although more expensive in terms of processing time because of all the steps
needed for each subband, the proposed strategy proved to be a great improvement over the conventional pro-
cedure of detecting and delineating discontinuities in fault and karst structures when treating seismic data from
an offshore carbonate field in Campos Basin, Brazil.

Introduction the quality and reliability of the seismic attributes ex-

The relatively low dominant frequency of conven- tracted from it. Chopra and Marfurt (2007) show that
tional amplitude seismic data limits the visualization the seismic attribute calculated from a data set with lim-
of structural details. In general, because greater band- ited resolution can overlook the delineation of subtle
width implies higher resolution, the application of reservoir features. On the other hand, Zhou et al.
processing techniques to broaden the seismic record (2015) demonstrate how resolution improvements have
spectrum is a common and useful approach to resolve a positive impact on seismic data when it comes to co-
subtle features, such as minor faults and karsts. Nowa- herence, energy, curvature, and frequency attributes,
days, resolution enhancement has become a hot topic in allowing a more accurate interpretation of fault and
this field of seismic research. Enhancing seismic reso- channel edges.
lution allows a more refined structural and stratigraphic Usually, the most broadband data are preferred for
interpretation. Such refinement is vital in complex geo- interpretation. However, the broadband data do not
logic settings, as the ones encountered on carbonate necessarily provide the best insight for seismic attribute
reservoirs, but it is equally true for any scenario in evaluation. In fact, the seismic response of a given geo-
which a detailed analysis is needed. logic structure and, consequently, the seismic attributes
Broadly speaking, resolution improvement algo- calculated from it, has a different seismic expression at
rithms fall into spectral balancing and inverse Q-filter- different frequency bands (Li and Lu, 2014). Al-Dossary
ing approaches (Fraser and Neep, 2004; Wang, 2006; and Marfurt (2006), for example, evaluate how long-
Puryear and Castagna, 2008; Matos and Marfurt, and short-wavelength curvature attributes impact the
2011; Braga and Moraes, 2013; Zhou et al., 2015). Re- delineation of geologic features of different scales.
gardless of the technique applied to this finality, the Sun et al. (2010) show how discrete frequency coher-
seismic resolution improvement will directly affect ence cubes can detect faults and fracture zones not

1
University of Campinas — UNICAMP, Department of Geology and Natural Resources, Campinas, Brazil. E-mail: [email protected];
[email protected]; [email protected].
2
SISMO Research & Consulting, Rio de Janeiro, Brazil. E-mail: [email protected].
Manuscript received by the Editor 11 December 2015; revised manuscript received 11 April 2016; published online 11 November 2016. This paper
appears in Interpretation, Vol. 5, No. 1 (February 2017); p. T65–T73, 8 FIGS.
http://dx.doi.org/10.1190/INT-2015-0211.1. © 2017 Society of Exploration Geophysicists and American Association of Petroleum Geologists. All rights reserved.

Interpretation / February 2017 T65

 easily seen through the full-spectrum coherence data, whereas adding the three difference traces highlights
particularly when focusing on the high-frequency band higher frequencies successively. The combination of
of the data. Hardage (2015), on the other hand, demon- these components enhances the entire bandwidth of
strates a better fault illumination considering only the the data.
lowest octave of the seismic spectrum (8–16 Hz). Thus, Using the Fourier properties and theorems, we can
spectral decomposition and the red-green-blue (RGB) better understand the effect of differentiation in the DR
color stack have been proven to be useful techniques method. Denoting FðωÞ as the Fourier transform of the
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

to extract and display geologic features at different differentiable signal f ðtÞ, prime to the derivative with
spectral bands. Although there is useful information respect to time, and i as the imaginary unit, the succes-
in the instantaneous spectral attributes, the exact fre- sive derivatives of f ðtÞ are:
quency range that produces an optimal image of a target
varies according to target size, depth, thickness, and
f 0 ðtÞ ⇔ iωFðωÞ;
impedance properties (Hardage, 2009). In addition, it
is not easy for interpreters to individually evaluate all f 0 0ðtÞ ⇔ −ω2 FðωÞ;
possible isofrequency volumes, making the choice of op- ···
timum components subjective and unclear (Liu and Mar-
furt, 2007). There are techniques that circumvent this f 0 0 0 0 ðtÞ ⇔ ω4 FðωÞ; and
“problem” somehow by using mathematical approaches, ···
such as the average of three nonoverlapping spectral
000000
bands (Stark, 2006); the use of three predetermined basis f ðtÞ ⇔ −ω6 FðωÞ; (2)
functions, producing more continuous and overlapped
spectral bands (Liu and Marfurt, 2007); or the use of pro- which progressively boosts the higher frequencies ac-
jections in a multidimensional space (Guo et al., 2009; cording to the derivative order. In addition, it is straight-
Honório et al., 2014). forward to understand the negative signals in the
In this work, we demonstrate how a recently devel- subbands stacking in equation 1.
oped spectral enhancement method based on differen- The amplitude normalization for each version of the
tial resolution (DR) (Sajid and Ghosh, 2014) can be used trace X is performed according to its median of the ab-
as a preprocessing technique to compute a similarity ~
solute value jXj
attribute at different frequency bands, which we call
the DR Similarity (DRS). To do so, we first review X
Y¼ ; (3)
the concepts of the DR method. Then, we propose an ~
jXj
approach to compute and combine the similarity attrib-
ute from different spectral bands honoring their fre- where Y is the normalized seismic trace. To obtain a
quency content as a guide for windowing. Finally, we normalized DR, denoted by R, r is also normalized ac-
apply the proposed strategy to enhance the delineation cording to equation 3.
of geologic features using seismic data from an offshore The differentiation is obtained by difference opera-
carbonate field in Campos Basin, Brazil. tors for efficiency. In the algorithm proposition, a com-
bination of one forward and one backward difference is
DR and similarity done to not introduce a time shift. Although the algo-
In this section, we introduce the proposed method rithm works well in terms of computational efforts
and briefly highlight the parameters involved. The DR and precision, we have refined this procedure by per-
method adds different versions of the seismic trace forming a forward difference for the first sample, a
to the original one. To keep the main behavior of backward difference for the last sample, and the central
the signal, the first term to be added is the smoothed difference for all the intermediate samples:
version Y S of the normalized original signal Y , which
is obtained by 10 passes of a three-point smoother with yjþ1 − yj
yj0 ≅ ðj ∈ Rðj ¼ 1ÞÞ forward; (4)
weights [1 2 1]. The normalization is discussed in the Δt
sequence. Then, the normalized second-, fourth-, and
sixth-order differentiated versions of the trace, denoted yj − yj−1
by Y II , Y IV , and Y VI , respectively, are added: yj0 ≅ ðj ∈ Rðj ¼ nÞÞ backward; (5)
Δt
r ¼ Y þ Y S − Y II þ Y IV − Y VI ; (1) and
where r is the nonnormalized DR. The negative signal in yjþ1 − yj−1
yj0 ≅ ðj ∈ Rð1 < j < n − 1ÞÞ central: (6)
equation 1 is to correct the phase shift introduced 2Δt
by differentiation. The fourth-order difference has nor-
mal polarity, whereas the second- and the sixth-order The reason for this is that the truncation error for the
differences have reverse polarity. The effect of add- backward and forward differences is OðΔtÞ, whereas
ing the smoothed trace is to boost low frequencies, the truncation error for the central difference is

T66 Interpretation / February 2017

 OðΔt2 Þ, which yields a more accurate approximation. true. From checking the output of the DR algorithm,
Although the truncation error can drop to OðΔt2 Þ after we can clearly discriminate the two interfaces in IV
one backward and one forward differentiation, by ap- (separated by five samples or 10 ms thickness), and
plying the central difference we reduce by half the num- get a glimpse about the thin bed III (four samples or
ber of calculations and the computing time, which can 8 ms thickness). In addition, note the frequency shift
be significant when working with large 3D seismic vol- that occurs as we move from Y S to Y VI (Figure 1g),
umes. The nth-order difference is approximated by ap- which is the base for our algorithm.
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

plying n successive differences on the signal yðtÞ. As mentioned before, the seismic response of a given
Figure 1 shows a synthetic signal and the results of geologic structure is expressed differently at different
the main steps of the DR method. In Figure 1a, the event spectral bands. Thus, instead of reconstructing the sig-
I is a single interface transition, whereas events II–VI nal with extended bandwidth according to equations 1
represent thin beds from 6 to 14 ms thicknesses and 3 as the DR algorithm does, we extract the similar-
(2 ms thickness increments). The representation of ity attribute for each DR component. In this way, the DR
the seismic trace generated by applying a 35 Hz Ricker algorithm is used as a preprocessing procedure and acts
wavelet is shown in Figure 1b. The output of the DR as a pseudofilter to be used for similarity computation.
algorithm R is shown in Figure 1c. The corresponding Because we want to combine the attributes in the RGBα
spectra for the signals in Figure 1a and 1b are illustrated color space, we are limited to four variables. Thus, we
in Figure 1f. can combine Y and Y S as a new variable Y NS according
It is clear from Figure 1c, the resolution improve- to equation 3, what accounts to the general behavior of
ment obtained from DR algorithm. The constructive in- the data, i.e., the low-frequency content. Then, in a step
terference that occurs between the top and base of thin called dip steering, we extract the local-dip information
beds III and IV in the original seismic trace (red arrows) at every sample position for each DR component to
could suggest a single interface transition, which is not guide the attribute extraction along the reflectors.

Figure 1. Illustration of the DR method: (a) reflectivity series; (b) seismic trace; (c) DR output; (d and e) the components used for
signal reconstruction in the DR method; (f) spectra from (b and c); and (g) spectra from (e).

Interpretation / February 2017 T67

 Considering that each derivative shifts the frequency them in the RGBα color space. The α-channel (transpar-
spectrum to higher frequencies, the seismic wavelets ency) of the RGBα color space goes from zero (fully
become tighter as we move to higher order differences. transparent) to 255 (fully opaque), so the lowest simi-
Therefore, we can use a different analysis window larity or completely dissimilar feature appears as the
honoring the peak frequency of each DR component background color, which is black in our case. The pro-
to extract the similarity attribute, in a similar way as posed workflow is summarized in Figure 2.
proposed by Lin et al. (2014). In our case, we determine
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

the peak frequency f peak for each time sample inside a Results and discussion
400 ms interval, 200 ms above and below the key hori- We start our study by first analyzing the effect of ap-
zon. Then, we calculate the mean peak frequency f peak plying the DR algorithm and how the frequency content
of each DR component to determine the half-window varies within each DR component. Figure 3 displays a
height H gate for the similarity computation random seismic section and the average spectra for the
corresponding volume. The green line represents a key
β
H gate ¼ ; (7) horizon (H1) in our study. Comparing the spectrum of
2f peak the original (plus curve) and the R data (circle curve),
we can see a considerable gain in the frequencies
where β is a refinement parameter that can be used to greater than 30 Hz. In addition, note how the spectrum
adjust the actual window size, depending on data qual- of each DR component (solid lines) fit in the R spec-
ity, thus giving a flexibility to the workflow. In our sim- trum. The high-frequency gain seen in the R volume
ulations, we set the β value as one, but it can be spectrum helps to delineate some subtle features not
adjusted to greater values if the signal-to-noise ratio properly seem in the original data, as the features high-
is judged to be too low or, equivalently, if the detected lighted by the white arrows in Figure 3a and 3b. By
discontinuity is noise influenced. The final step is to evaluating the subsequent sections of Figure 3, we
compute the similarity for each subband and combine can track back which DR component or subband better
illuminates a given feature. Because the
derivatives operations make the seismic
wavelets progressively tighter, the tun-
ing effects migrate to thinner thickness,
helping to discriminate reflections previ-
ously merged in the original data, as the
one pointed by the white ellipses in Fig-
ure 3. Thus, considering that each DR
component highlights slightly different
information, we can adjust the similarity
computation accordingly and get a final
image with improved discontinuity de-
tection and delineation.
Figure 4 shows the application of
DRS to real seismic data from a Brazil-
ian carbonate offshore field in Campos
Basin. Figure 4a–4d displays the hori-
zon H1 through the similarity attribute
extracted from the DR components
(SYn ) separately, whereas Figure 4e dis-
plays its RGBα color stack. The time
window for the similarity computation
varies from 22 ms for SYVI to 40 ms
for SYNS . The red arrows in Figure 4a–
4d show features that are better seen
in the derivatives components, which
suggest that they have a smaller scale.
The green arrows, on the other hand,
show features that are more evident in
the lower frequency band (SYNS ). We
can see a slight noise increase as we
go from SYNS to SYVI , but this is justified
by the gain we have in the definition and
resolution of some subtle structures.
Figure 2. The DRS workflow. The color interpretation in the RGBα

T68 Interpretation / February 2017

 also gives us an idea about the scale of
the features. For example, note the
subtle feature highlighted by the blue ar-
row in Figure 4. In the RGBα display,
this feature is slightly red, which means
that the discontinuity is captured mostly
in the green and blue channels (see Fig-
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ure 4c and 4d), but poorly seen in the

red channel (see Figure 4b) and not seen
at all in the α-channel (see Figure 4a). In
other words, this suggests a small fea-
ture that is not detected by SYII and
SYNS , thus appearing continuous in the
red and α-channels.
For comparison, in Figure 5, we show
the similarity attribute computed di-
rectly using the original seismic volume,
which we call base model, and by apply-
ing the proposed strategy. We also
evaluate the effect of computing the sim-
ilarity directly in the bandwidth ex-
tended volume R to demonstrate the
effect of not considering the scale of
each DR subband separately as we do
in DRS. To make the comparison ac- Figure 3. Seismic sections and spectra: (a) original amplitude; (b) DR output;
cordingly, for the computation of simi- (c-f) the components Y NS , Y II , Y IV , and Y VI , respectively; and (g) the correspond-
larity attribute through the original and ing spectra from (a-f). The green line in the seismic sections represents the key
R volumes, the dip information is also horizon H1.
extracted and the window size follows
the peak frequency as described in equa-
tion 7, in accordance with what we do
in DRS.
A particular improvement occurs in
the definition of the features pointed
by the red arrows. The similarity ex-
tracted from the R volume (Figure 5c)
enables the detection of different fea-
tures slightly over the original data (Fig-
ure 5b). However, through the DRS, we
can see a much better detection and def-
inition of features in Figure 5d. The left-
most arrow, for example, shows a subtle
discontinuity that is well-captured in
DRS, poorly seen in similarity from R,
and not seen at all in the similarity from
the original data. The red arrow with
black outline in the center is another
similar example that pinpoints to a karst
structure that is clearly seen in the DRS.
A vertical section along profile A-A′
in Figure 5a crossing this structure
through the base model and DRS data
are shown in Figures 6 and 7. Note that,
although correlated, each DRS compo-
nent illuminates and detects different
structures. In the particular case of
the pointed karst, its boundary is better
detected in the SYII (Figure 6c2), a sub- Figure 4. The DRS computed for H1. (a-d) Similarity from Y NS , Y II , Y IV , and Y VI ,
band dominated by approximately 40 Hz respectively; and (e) RGBα color stack of SYII , SYIV , SYVI , and SYNS .

Interpretation / February 2017 T69

 Figure 5. Comparison of similarity attrib-
utes. Horizon H1 through (a) original seismic
amplitude and the corresponding similarity
attribute in (b); (c) similarity from R; and
(d) DRS.
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 6. Seismic section through (left) am-

plitude data, (middle) similarity, and (right)
amplitude/similarity corendered from: (a) origi-
nal and (b-e) DR components (Y NS , Y II , Y IV ,
and Y VI ), respectively. See Figure 5a for sec-
tion reference location.

T70 Interpretation / February 2017

 in the amplitude data. Also note the discontinuities To get a quantitative evaluation regarding the differ-
pointed by the blue arrows in the center column of Fig- ent procedures applied, we extracted the crosscorrela-
ure 6. In the base model, such discontinuities are quite tion between the DRS components and the base model
blurred. As we move down in the figure, we are able to (Figure 8a) and between the DRS components and the
see a progressive better definition of the two fractures, similarity from R (Figure 8b). Basically, the sample cor-
which attests the DRS ability to illuminate features at relation coefficient c denotes the quality of least-
different scales. squares regression fit between any two attribute vec-
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

tors, and it is equivalent to a windowed zero-lag cross-

correlation, in which the window length is the entire
attribute vector length. In Figure 8a, the highest corre-
lation occurs between the similarity from the base
model and SYNS , which is related to the lowest fre-
quency component in the DRS algorithm. As we move
to higher frequency components in the DRS and, con-
sequently, looking to smaller scale features, the cross-
correlation with the base model drops close to 50%. The
components not only share half of information, but also
another 50% of different information is present in SYVI .
Obviously, some noise can be accounted for in this
equation. Therefore, it is important to verify if such
new information is geologically feasible, such as the
feature pointed out by yellow arrows in Figures 6a3–
6e3 and 7, which clearly suggests a fault that is poorly
seen in the base model but is well-illuminated in DRS. A
similar reasoning is drawn to the crosscorrelation be-
tween the DRS components and the similarity from
R. Except for the low-frequency component, their cor-
relation is higher than the ones obtained between the
DRS components and the base model. Such analysis
is in accordance and is justified because of the intrinsic
Figure 7. Similarity comparison through the (a) base model relationship between R and its components, which will
and (b) DRS. impact the higher correlation between the attributes ex-
tracted from them. Note, for example,
the high correlation between SYII and
the similarity from R (c ¼ 0.94, Fig-
ure 8b2), which occurs mainly because
the corresponding amplitude data have
almost the same dominant frequency
(approximately 40 Hz; see Figure 3g)
and, consequently, the same window
size is applied to similarity computation
in both cases. This reinforces the impor-
tance of considering the scale of the
feature we are dealing with and inter-
ested in. For the other subbands, their
crosscorrelation drops close to 70%,
suggesting that they are capturing differ-
ent information.

Conclusion
If on one hand broadening the seis-
mic spectrum is a logical step for seis-
mic resolution improvement, on the
other hand we have to consider that
such image is a composite of various
geologic structures, which may have dif-
ferent scale ranges. Thus, if the scale of
Figure 8. Crossplots derived from the attribute maps of Figure 5. (a) DRS ver- observation is not appropriate to the
sus base model and (b) DRS versus similarity from R volume. scale we are interested in, we risk an

Interpretation / February 2017 T71

 improper interpretation. An interpretation in a structur- Hardage, B. A., 2015, Pitfall experiences when interpreting
ally complex setting, such as karst- or fault-related re- complex structure with low-quality seismic images:
gions, can be very challenging to corender different Interpretation, 3, no. 3, SB29–SB37, doi: 10.1190/
seismic attributes to highlight seismic features of inter- INT-2014-0118.1.
est. The scale range plays an important role in defining Honório, B. C. Z., A. C. Sanchetta, E. P. Leite, and
and identifying whether we are seeing a simple seismic A. C. Vidal, 2014, Independent component spectral
feature or a subtle geologic structure. analysis: Interpretation, 2, no. 1, SA21–SA29, doi: 10
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

In this way, we have combined a seismic spectral .1190/INT-2013-0074.1.

enhancement method based on DR and similarity attrib- Li, F., and W. Lu, 2014, Coherence attribute at different
utes to extract dissimilar features at different spectral spectral scales: Interpretation, 2, no. 1, SA99–SA106,
bands. Because each DR component has a different fre-
doi: 10.1190/INT-2013-0089.1.
quency content, we can use different analysis windows
Lin, T., B. Zhang, Z. Zhan, Z. Wan, F. Li, H. Zhou, and K. J.
to evaluate the discontinuity at a given frequency range.
Marfurt, 2014, Seismic attributes of time- vs. depth-
The combination of the DRS components in the RGBα
color space helps to group the subbands discontinu- migrated data using self-adaptive window: 84th Annual
ities, allowing a more detailed definition of the struc- International Meeting, SEG, Expanded Abstracts, 1659–
tures under investigation. 1663.
Although more expensive in terms of processing Liu, J., and K. J. Marfurt, 2007, Multicolor display of spec-
time because of all steps needed in DRS, it proved to tral attributes: The Leading Edge, 26, 268–271, doi: 10
be a great improvement in discontinuity detection .1190/1.2715047.
and delineation over the base model. Such improve- Matos, M. C., and K. J. Marfurt, 2011, Inverse continuous
ment is especially important in complex scenarios in wavelet transform “deconvolution”: 81st Annual Inter-
which the discontinuities can vary in many different national Meeting, SEG, Expanded Abstracts, 1861–1865.
scales and be associated with either reservoir compart- Puryear, C. I., and J. P. Castagna, 2008, Layer-thickness de-
mentalization or migration pathways. termination and stratigraphic interpretation using spec-
tral inversion: Theory and application: Geophysics, 73,
Acknowledgments no. 2, R37–R48, doi: 10.1190/1.2838274.
We would like to thank Statoil for financial support Sajid, M., and D. Ghosh, 2014, A fast and simple method of
and for allowing us to publish this study. We also thank spectral enhancement: Geophysics, 79, no. 3, V75–V80,
the Brazilian Council for Technological and Scientific doi: 10.1190/geo2013-0179.1.
Development (CNPq, Brazil) for the support to this Stark, J., 2006, Visualization techniques for enhancing
project by means of a Ph.D. scholarship for B. Honório stratigraphic inferences from 3D seismic data volumes:
and dGB Earth Sciences Company for providing Open- First Break, 24, no. 4, 75–85.
dTect software through an academic license agreement. Sun, D. S., Y. Ling, X. Y. Guo, J. Gao, and J. X. Lin, 2010,
We are grateful for the multiclient seismic data pro- Application of discrete frequency coherence cubes in
vided by PGS. Sinochem is also acknowledged for the fracture detection of volcanic rocks in full-azimuth
the permission of this publication. seismic data: 80th Annual International Meeting, SEG,
Expanded Abstracts, 1342–1346.
References Wang, Y., 2006, Inverse Q-filter for seismic resolution en-
Al-Dossary, S., and K. J. Marfurt, 2006, 3D volumetric multi- hancement: Geophysics, 71, no. 3, V51–V60, doi: 10
spectral estimates of reflector curvature and rotation: .1190/1.2192912.
Geophysics, 71, no. 5, P41–P51, doi: 10.1190/1.2242449. Zhou, H., Y. Wang, T. L. F. Li, and K. J. Marfurt, 2015, Value
Braga, I. L. S., and F. S. Moraes, 2013, High-resolution gath- of nonstationary wavelet spectral balancing in mapping
ers by inverse Q filtering in the wavelet domain: Geo- a faulted fluvial system, Bohai Gulf, China: Interpreta-
physics, 78, no. 2, V53–V61, doi: 10.1190/geo2011-0508.1. tion, 3, no. 3, SS1–SS13, doi: 10.1190/INT-2014-0128.1.
Chopra, S., and K. J. Marfurt, 2007, Seismic attributes for
prospect identification and reservoir characterization:
SEG. Bruno César Zanardo Honório re-
Fraser, G. B., and J. Neep, 2004, Increasing seismic reso- ceived a B.S. (2008) in physics and
lution using spectral blueing and colored inversion: an M.S. (2011) in science and petro-
Cannonball field, Trinidad: 74th Annual International leum engineering, both from the Uni-
versity of Campinas — UNICAMP,
Meeting, SEG, Expanded Abstracts, 1794–1797.
where he is pursuing a Ph.D. in geosci-
Guo, H., K. J. Marfurt, and L. Liu, 2009, Principal compo- ences. He works as a researcher at the
nent spectral analysis: Geophysics, 74, no. 4, P35–P43, Center for Oil Studies CEPETRO/UNI-
doi: 10.1190/1.3119264. CAMP. His research interests include
Hardage, B. A., 2009, Frequencies are fault finding factors: signal processing, time-frequency analysis, seismic resolu-
Looking low aids data interpretation: AAPG Explorer, tion, statistical methods, noise attenuation, and geophysi-
30, 34. cal attributes.

T72 Interpretation / February 2017

 Ulisses Miguel da Costa Correia He was a visiting scholar at the University of Oklahoma
received a B.S. (2012) in geology from from January 2008 to January 2010. He is currently coinves-
the Faculty of Sciences at the Univer- tigator of the Attribute-Assisted Seismic Processing & Inter-
sity of Lisbon and an M.S. (2015) from pretation Research Consortium at the University of
the Institute of Geoscience at the Oklahoma, and he is the principal of SISMO Signal Process-
University of Campinas. He is a Ph.D. ing Research, Training & Consulting. His research interests
candidate in the Department of Petro- include applied seismic analysis, digital signal processing,
Downloaded 12/03/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

leum Engineering at the University of spectral decomposition, and seismic pattern recognition.
Campinas — UNICAMP. His Ph.D. re-
search focuses on structural geologic analysis and seismic
interpretation. Alexandre Campane Vidal received
a B.S. (1993) in geology from the Uni-
versity of São Paulo — USP, an M.S.
Marcílio Castro de Matos received (1997) in reservoir geoengineering
a B.S. (1988) and an M.S. (1994) in from University of Campinas — UNI-
electrical engineering from Instituto CAMP, and a Ph.D. (2003) in regional
Militar de Engenharia (IME) and a doc- geology from Universidade Estadual
toral degree (2004) from Pontificia Paulista — UNESP. He is a professor
Universidade Catolica do Rio de Ja- at the Institute of Geosciences, UNI-
neiro. From 1989 to 1999, he served CAMP. Postdoctoral studies were performed from 2002
as a military engineer at the Brazilian to 2003 in the Departamento de Geologia Aplicada at
Army Test Center and worked as a sig- UNESP. He has experience in geology, with emphasis
nal processing military professor at IME from 1999 to 2010. on reservoir geology.

Interpretation / February 2017 T73