NANJING INSTITUTE OF GEOPHYSICAL PROSPECTING AND INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF GEOPHYSICS AND ENGINEERING

J. Geophys. Eng. 2 (2005) 213–221 doi:10.1088/1742-2132/2/3/005

Reservoir characterization using seismic

data after frequency bandwidth
enhancement
Yongsheng Ma1, Xuan Zhu1, Tonglou Guo1, Tony Rebec2
and Kostia Azbel2
1
Sinopec Southern E&P Company, No 201 Guan-Shang-Zhong-Lu Road, Guan-Du District,

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Kunming City, Yunan Province 650200, China
2
Paradigm Geophysical, Houston, TX 77024, USA

Received 8 April 2005

Accepted for publication 8 June 2005
Published 1 August 2005
Online at stacks.iop.org/JGE/2/213

Abstract
Reservoir characterization is a crucial prerequisite to predict the economic potential of a
hydrocarbon reservoir or to examine different production scenarios. Unfortunately, it is
impossible to determine the exact reservoir properties at the required scale. The most abundant
seismic data have a resolution of around 30 m. Wells resolve the reservoir down to the
centimetre scale, but only at some points in the vertical direction. This paper presents a
method, referred to as frequency bandwidth enhancement (FBE), for enhancing the frequency
bandwidth and restoring high frequencies. The method increases the resolution of seismic data
by padding the frequency spectrum of seismic wavelets, thus pushing the notch corresponding
to the time limit of resolution to a higher part of the spectrum. This approach (patent pending)
results in sharper wavelets capable of identifying thinner beds. A by-product of extending the
frequency spectrum is the elimination of the tuning effect of beds thinner than the new limit of
resolution. When tuning curves before and after the process are compared, it is observed that
although different, the differences are minuscule and insignificant compared to the benefits
coming from being able to resolve thinner beds. The procedure enunciated here is robust and
helps define trends better, leading to more confident interpretations. Such applications could
redefine prospects, which in some cases may have been declared unsuccessful on the basis of
interpretation of seismic data with the original bandwidth. The reservoir characterization is
realized with the aid of coherence cube processing, which is an extremely powerful tool to
efficiently exploit the wealth of structural and stratigraphic information encapsulated in the
seismic waveforms of 3D seismic data volumes.

Keywords: reservoir characterization, frequency bandwidth enhancement, coherence

cube processing

1. Introduction with contrasting velocity. Low-porosity carbonate bodies

associated with thin shales and encased in shaly carbonate
Reservoir characterization comprises determining reservoir rocks having a narrow frequency bandwidth may not be
architecture, establishing fluid-flow trends, constructing seen on seismic data. Often we come across examples
reservoir models and identifying reserve growth potential. where the initially processed 3D seismic volume results in
However, geophysicists are often frustrated at their inability to interpretations that sometimes are geologically suspect, e.g.,
extract and understand the subtle stratigraphic detail contained cases involving complex faulted patterns or subtle stratigraphic
in 3D seismic volumes. Seismically, stratigraphic bodies plays. Similarly, post-mortem analysis may cite small fault
with definitive shapes show up if they are encased in rocks displacements or obscure seismic data as reasons for dry wells.

1742-2132/05/030213+09$30.00 © 2005 Nanjing Institute of Geophysical Prospecting Printed in the UK 213

Y Ma et al

In such cases, the usual practice is to create a new version of the wavelet’s characteristics (breadth or frequency information).
3D volume with some target-oriented processing to improve The interference between wavelets as beds become thinner
imaging in the zone of interest that will, in turn, lead to a more (wedge model) results in thickness and amplitude deviations
accurate interpretation. In some cases this helps, but in others from the real ones. These differences are graphically displayed
some questions still remain unresolved. and analysed using tuning curves where apparent versus real
In the latter, more often than not, more accurate thickness and amplitudes are compared.
stratigraphic interpretation is needed but the available Ricker (1953), Widess (1973) and Kallweit and Wood
bandwidth of the seismic data is inadequate to image or resolve (1981) quantified the time limit of resolution, but gave no
the thickness of many of the thin targets seen in the wells. This insight into the possible ramifications of doing it in the
problem can be addressed by having data of reasonable quality spectral domain. Our method described below deals with
and augmenting it by some frequency restoration procedure the expression of a thin bed in the frequency domain and a
that would improve the vertical resolution (Wang 2002, methodology for extending the relevant part of the spectrum
2003). Frequency restoration is necessary because seismic and, as a consequence, decreases the limit of resolution.
waves propagating in the subsurface are attenuated and this Moreover, we show that the expression, in the frequency
phenomenon is frequency dependent—higher frequencies are domain, of two spikes convolved with a wavelet is a notch in

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

absorbed more rapidly than lower frequencies. Consequently, the amplitude spectrum at a frequency that is a function of the
the highest frequency recovered on most seismic data is spikes’ separation (in time) and events’ polarity. Furthermore,
usually about 80–100 Hz. This enables confident mapping the notch is present in the spectrum for beds thinner than
of subsurface horizons of interest, clarifies detailed geological the time limit of resolution, implying that conducting the
settings and eventually leads to more profitable seismic analysis in the frequency domain will allow us to push the
exploration programmes. limit of resolution to a higher degree, i.e. we can resolve
The coherence cube analysis (Bahorich and Farmer 1995, smaller thickness of beds. The basic technique described
Marfurt et al 1998, 1999) provides accurate maps of the spatial below increases the limit of resolution (thinner beds in the time
change in the seismic waveform that can readily be related to domain) by extending the frequency spectrum of the wavelet
geologic features and depositional environments. Faults and such that the (frequency) value of the highest identifiable
fracture systems can be spatially imaged and directly mapped frequency notch corresponds to the limit of resolution of
from the coherence cube without the tedious task of drawing the extended spectrum. The resultant seismic data are of
faults on each vertical section and proceeding blindly without much higher frequency and allow us up to a 30% increase
the knowledge of their spatial position in the early crucial phase in resolution. The effect of padding the frequency spectrum
of the interpretation. Stratigraphic features can be readily where no seismic information is recorded in the first place is
detected in the volume, relieving the interpreter of the tedious also analysed.
task of locating them, thus saving time for detailed analysis We will use a Kallweit and Wood (1981) resolution criteria
(Maione 1999). to illustrate the technique since they used a broadband wavelet,
similar to that used in our analysis, to quantify resolution. The
2. Frequency bandwidth extension analysis, done for spikes of the same polarity and amplitude,
can be easily extended to include spikes of different polarities
We are aware that seismic waves propagating in the subsurface and amplitudes. It is stated without proof that such an analysis
get attenuated. Since this loss is frequency dependent, seismic would result in conclusions similar to those observed in the
signals with higher frequencies are absorbed more rapidly than test case. The practical time limit of resolution, as defined
those of lower frequencies (Wang and Guo 2004). Attempts by Kallweit and Wood (1981), for a broadband wavelet with
are usually made to determine the subsurface attenuation a maximum frequency fU and a white spectrum occurs at
in terms of the attenuation constant α or the quality factor a one-quarter-wavelength condition and approximates TR =
Q (Wang 2004). The conventional methods used for their 1/(1.4fU ), provided that the wavelet’s band ratio exceeds two
determination from surface seismic data not only have their octaves.
limitations but also have a certain level of uncertainty. The frequency representation of two spikes of the same
The frequency bandwidth enhancement (FBE) approach polarity separated at a time interval t corresponds to a
to enhance the frequency bandwidth results in sharper wavelets notch at frequency fN = 1/(2t). The thinner the bed
capable of identifying thinner beds. Resolving thin beds from being analysed, the larger the value of fN . The notch in
seismic data implies identifying individual reflections from the frequency spectrum for the time limit of resolution will
the top and bottom of a bed, and the limit of resolution is fall at a frequency fR = 1/(2TR ) or fR = 0.7fU ; that is,
defined as the smallest bed separation that can be identified the notch falls at 70% of the maximum frequency. As the
as two distinct events in seismic data. Ricker (1953), Widess bed thickness decreases, the representation in time is an
(1973) and Kallweit and Wood (1981) studied the limit of increase in the amplitude of the wavelet where upper and
resolution more than two decades ago by convolving wavelets lower limits cannot be distinguished (the events are in tune).
of known characteristics with two spikes, the distance (time The representation in the frequency (spectral) domain
interval) between the spikes representing the top and bottom corresponds to a notch in the amplitude spectrum at
of the structure under study. Through their analysis they frequencies higher than fR (0.7fU ) and up to nearly fU . The
quantified the thinnest bed to be resolved as a function of the technique presented here is based on the fact that thinner

214
 Reservoir characterization using seismic data after frequency bandwidth enhancement

O rig in a l FBE

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Figure 1. Stacked section before (left) and after (right) FBE processing. Many detailed subtle stratigraphic features can be seen on the FBE
data (the green line interpretation).

FBE
FBE Original
O iginal
Figure 2. Direct merge of the seismic section before (right) and after (left) FBE processing. Additional small reflectors can be identified in
the red circled areas on the FBE data.

beds can be better resolved in the frequency domain than in the usual case. Namely, amplitudes will not increase for
in the time domain. Based on Kallweit’s model, the time wavelets convolved with thinner beds than the new limit of
limit of resolution can be reduced by 30% by extending the resolution.
wavelet’s maximum frequency to a value such that the notch
at the highest usable frequency (close to fU ) before extension 3. Coherence cube processing
corresponds to 1/(1.4fUE ) where fUE corresponds to the new
(extended) maximum frequency. We have used several coherence formulations (Bahorich and
In practice, the extension of the frequency bandwidth by Farmer 1995, Marfurt et al 1998, 1999) within this research
padding is imperfect and results in wavelets that are not white project. These algorithms are based on the following basic
and, for this reason, do not honour Kallweitt’s time limit of mathematical solutions: correlation (C1), semblance (C2) and
resolution; nevertheless, the time limit of resolution is still eigen-decomposition (C3).
reduced by 30% by making fUE larger than 1.4fU . By padding Numerous hybrids of these algorithms have been
amplitudes between fU and fUE , where there is no seismic developed and some will be discussed here.
signal, the wavelet becomes sharper with the resultant decrease The correlation algorithm (C1). The basic correlation
in the limit of resolution. Since no real signal exists from fU approach has been around since the early days of digital signal
to fUE , it will be impossible to see the effect of beds thinner processing. When applied to two seismic traces it can be
than TRN on the extended spectrum and the resultant tuning defined as the degree of linear relationship between them, a
curves will have different characteristics than those observed measure of how much they look alike or the extent to which

215
Y Ma et al

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Figure 3. Amplitude spectra of the seismic data before FBE processing. The peak frequency is about 35 Hz with bandwidth from 10 Hz to
80 Hz.

Figure 4. Amplitude spectra of the seismic data after FBE processing. The peak frequency is about 70 Hz with bandwidth from 10 Hz to
130 Hz.

216
 Reservoir characterization using seismic data after frequency bandwidth enhancement

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Original Stack
Original Stack FBE
FBE

Figure 5. Comparison of a seismic section from 500 ms to 2000 ms before (left) and after (right) FBE processing.

Before FBE After FBE

Figure 6. Time slide (1000 ms) of seismic data before (left) and after (right) FBE processing.

one can be considered a linear function of the other. It is The eigen-decomposition algorithm (C3). This is a multi-trace
interesting to note that the frequency domain equivalent of eigen-decomposition process that is more robust with higher
correlation, in the time domain, is coherence. resolution than previous algorithms. Consider two seismic
traces whose amplitudes are crossplotted sample by sample
The semblance algorithm (C2). The introduction of the on the Cartesian coordinate system. The distribution of the
semblance algorithm by Barhorich and Farmer (1995) to the general shape of the plotted points can be represented by an
coherence cube methodology added a significant contribution, ellipse and the pattern formed by these points is governed by
especially for revealing detailed stratigraphic features. This the coherence of the two input traces plotted. The ellipsoidal
is made possible in part due to the shorter temporal apertures shape is not a measure of the individual samples but more a
which can now be employed. Semblance is the energy of a measure of the overall waveform shape being input.
summed trace divided by the mean energy of the components The coherence algorithms used are typically referred to
of the sum which is the energy of the stacked trace normalized as C1, C2 and C3. The methodology originally patented by
by the energy of the components of the stack. It is this Bahorich and Farmer (1995) describes a correlation technique
technique which forms the basis of the C2 algorithm. as a part of the approach for providing the numerical similarity

217
Y Ma et al

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Semblance Eigen

Figure 7. Time slide of coherence processing results comparison of C2 (semblance algorithm, left) and C3 (eigen-decomposition solution,
right) methods.

Figure 8. Comparison of a time slide (1680 ms) of seismic data (left) and coherence cube data (right).

of a cube of seismic traces. This resulted in the C1 algorithm. by introducing an eigen gradient term which produces a
Further work performed by Bahorich and Farmer (1998, 2000) significant lift in the sensitivity resulting in higher resolution.
produced the superior C2-based semblance algorithm. In 1996 This capability, called high-resolution eigen, makes even the
Marfurt et al announced the eigen structure algorithm (C3) most subtle waveform changes visible to the eye with both
with, in most cases, a significant response improvement over faulting and stratigraphic detail.
the semblance-based formulation. The eigen solution (C3) has The ability to measure three-dimensional spatial
proved to be highly successful, with increased robustness, in variations in the seismic waveform with full dip and azimuth
revealing both subtle faults and stratigraphy in one execution. comprehension is extremely powerful in exploiting 3D data
These results are far superior to the correlation (C1) and volumes. The basic seismic waveform contains a measure
semblance (C2) solutions used in the past and available on of time, amplitude, frequency, phase and absorption. Spatial
workstations. An additional improvement was made which variations in these measurements are the seismic response to
gives an option to remove the structural effect from the lateral variations in the physical and geometric properties and
technique caused by the instability of the zero crossing on the characteristic of the rocks. Measuring these combined changes
seismic trace. This is achieved by a higher fidelity dip/azimuth in the seismic response allows the interpreter to physically
search. Recently, a major breakthrough was accomplished locate and map these changes in space and time thus giving a

218
 Reservoir characterization using seismic data after frequency bandwidth enhancement

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Before FBE After FBE

Figure 9. Time slide (1680 ms) of the coherence cube data before (left) and after (right) FBE processing.

Before FBE After FBE

Figure 10. Time slide of adaptive eigen: comparison before (left) and after (right) FBE processing.

better understanding of the subsurface model. The coherence technique. Results here indicate that the illumination of the
cube methodology as applied here is an attempt to capture subsurface is a function of offset. The imaging of fault and
these changes. fraction systems is clearly visible using the coherence cube.
3D seismic coherence is computed by measuring
waveform similarity within an aperture which includes traces 4. Real data examples
and time samples within a user-specified space and time
window control. The waveform similarity is measured along The frequency restoration procedure described above has been
all possible planes within the dips specified. Faults can run on a 3D seismic data set from an onshore oilfield in
be identified by their low similarity measurement when the southwest China, and convincing and promising results have
aperture is straddling the fault location. Subtle changes in the been obtained. Figures 1 and 2 show the application of FBE
seismic waveform which show the extent and internal details on the migrated 3D seismic data using the approach discussed
of stratigraphic features can also be identified by using the above (padding the frequency spectrum of seismic wavelets to

219
Y Ma et al

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

Before FBE After FBE

Figure 11. Time slide (1680 ms) of bright spot volume display before (left) and after (right) FBE processing.

Before FBE After FBE

Figure 12. Time slide of relief-enhanced eigen: comparison before (left) and after (right) FBE (right) processing.

push the notch corresponding to the time limit of resolution be obtained by convolving (cascading) lower order filters with
to a higher part of the spectrum). The FBE process has one another. The process is most effective on zero phase data.
increased their bandwidth from 40 Hz or less to over 100 Hz Figure 5 compares the FBE processing result from the
(figures 3 and 4) using this patent-pending new technique. 500 ms to 2000 ms section. The FBE processing has
Geologists can redefine thin beds, pinchout, small faults, and significantly increased the frequency bandwidth, sharpened the
new prospects using these high frequency seismic sections wavelet and enhanced the reflection image for much easier and
with higher confidence. more confident interpretation. Figure 6 shows the time slide at
Figures 3 and 4 show the frequency bandwidth of the 1000 ms with significant high frequency features and great
3D seismic data before and after the FBE processing. The detail after FBE processing.
cascaded dipole differentiation filters (e.g. 1, −1) of various Figure 7 shows the time slide (1680 ms) of coherence
orders have been used during the FBE processing, which processing results using the C2 semblance method (left) and
can shift and increase the dominant frequency of zero phase C3 eigen method (right). The eigen-decomposition method
seismic trace data. Higher orders (and strengths) of filters can gives much higher resolution, more clear, detailed and sharp

220
 Reservoir characterization using seismic data after frequency bandwidth enhancement

features than that of the C2 method on the same seismic redefine prospects, which in some cases may have been
data, which will provide detailed structure interpretation and declared unsuccessful on the basis of interpretation of seismic
reservoir characterization. data with poor bandwidth. Running the coherence cube
Figure 8 shows a comparison of the time slide (1680 ms) technique on seismic data shows that improved stratigraphic
of seismic data (left) and the coherence data (light) using the results can be achieved on good quality seismic data.
adaptive high resolution eigen method. There are fracturing
zones on the top of the structure and small faults developed Acknowledgments
along the major fault zone on the coherence cube time slide,
which cannot be identified from the time slide seismic data. This project was made possible by the dedication and effort
Figure 9 shows a comparison of the coherence cube time of many professionals at the Southern E&P Company of
slide (1680 ms) before (left) and after (right) FBE processing. SINOPEC, and Paradigm Geophysical. The support of
FBE processing has significantly improved and increased the SINOPEC and Southern E&P Company management in
resolution of the seismic data. Much more detailed geological facilitating the whole project is specifically acknowledged. We
features can be identified and interpreted (the area outlined in also acknowledge the Southern E&P Company of SINOPEC
red) from the coherence data after FBE processing. Figure 10 and Paradigm Geophysical’s permission to publish this work.

Downloaded from https://academic.oup.com/jge/article-abstract/2/3/213/5127600 by guest on 21 May 2020

shows the coherence results using the adaptive eigen method
before (left) and after (right) FBE processing.
The bright spot technique and relief-enhanced eigen References
method are displayed in figures 11 and 12. The ‘bright spot’ Bahorich M S and Farmer S L 1995 3D seismic coherency for faults
volume was created by running the coherence technique on and stratigraphic features The Leading Edge 14 1053–8
seismic inversion data and numerically merging the inversion Bahorich M S and Farmer S L 1998 Apparatus for seismic signal
volume with the coherence volume. The relief-enhanced processing and exploration US Patent 5838564
eigen solution provides a very novel product which enhances Bahorich M S and Farmer S L 2000 Seismic signal processing and
exploration US Patent 6092026
both structural and stratigraphic features to be highlighted. Kallweit R S and Wood L C 1981 The limits of resolution of zero
These plots have significantly helped geologists and reservoir phase wavelets Geophysics 47 1035–46
engineers to understand and interpret fluid distribution, special Maione S J 1999 Revealing the geology of the Gulf of Mexico
reservoir structure, and conduct reservoir characterization with recorded by 3D seismic Offshore Technology Conference
greater confidence. (Houston, Texas) paper 11055
Marfurt K J, Kirlin R L, Farmer S L and Bahorich M S 1996
Method and apparatus for seismic signal processing and
5. Conclusions exploration US Patent 5563949
Marfurt K J, Kirlin R L, Farmer S L and Bahorich M S 1998
The FBE method described in this paper differs from 3D seismic attributes using a semblance-based coherency
algorithm Geophysics 63 1150–65
conventional industry methods. It reduces the limit of Marfurt K J, Sudhaker V, Gersztenkorn A, Crawford K D and
resolution of seismic data by padding the frequency spectrum Nissen S E 1999 Coherency calculations in the presence of
of seismic wavelets, thus pushing the notch corresponding to structural dip Geophysics 64 104–11
the time limit of resolution to a higher part of the spectrum. Ricker N 1953 Wavelet contraction, wavelet expansion and the
These result in sharper wavelets capable of identifying thinner control of seismic resolution Geophysics 18 769–92
Wang Y 2002 A stable and efficient approach of inverse Q filtering
beds. A by-product of extending the frequency spectrum is the Geophysics 67 657–63
elimination of the tuning effect of beds thinner than the new Wang Y 2003 Quantifying the effectiveness of stabilized inverse Q
limit of resolution (TRN ). When tuning curves before and after filtering Geophysics 68 337–45
the process are compared, it is observed that although different, Wang Y 2004 Q analysis on reflection seismic data Geophys. Res.
the differences are minuscule and insignificant compared to the Lett. 31 L17606
Wang Y and Guo J 2004 Modified Kolsky model or seismic
benefits of being able to resolve thinner beds. The procedure attenuation and dispersion J. Geophys. Eng. 1 187–96
enunciated here is robust and helps define trends better, leading Widess M B 1973 Quantifying resolving power of seismic systems
to more confident interpretations. Such applications could Geophysics 47 1160–73

221