You are on page 1of 8

Frequency-dependent AVO analysis

Douglas J. Foster1, Zeyu Zhao1, Dhananjay Kumar2, Danica Dralus2, and Mrinal K. Sen1
https://doi.org/10.1190/tle39020084.1

Abstract reflections are not incorporated into the series until the third-order
A method for generalizing the conventional amplitude varia- terms of the series are included.
tion with offset model from an isolated interface to a scattering The generalization of the reflectivity from an interface to an
reservoir interval is presented. The advantage of this new method interval is a complex function with real and imaginary components.
is that it can provide enhanced detection of subtle reservoir and The wavefield amplitudes are given by integrating over the interval,
pore fluid properties. First- and second-order expressions for the and the phase delays are required to properly calculate these
reflected compressional wave energy from a specified heterogenous amplitudes. This integration determines the interference pattern
interval are given. These expressions are applied to two problems set up by heterogeneities. The amplitude and phase are incorpo-
of interest for reservoir description. One application is discriminat- rated, which shows that the kinematic and dynamics of the
ing low versus higher saturations of hydrocarbons, and the other scattering process are coupled. Our first-order approximation is
is detecting the extent of vertical stratification within a reservoir. the same as the primary pulse method (Richards and Frasier,
The first-order expression is used for determining hydrocarbon 1976). Richards and Frasier assume small contrasts of material
saturations, and the second-order expression is used for detecting properties, whereas in our derivation, given in Foster et al. (2017),
the magnitude of fine-scale layering within a reservoir. Synthetic we do not make this assumption. Kumar et al. (2017, 2020) show
models and field data examples are used in demonstrating the how the first-order model is used in discriminating low-saturation
applicability of the proposed method. gas from commercial concentrations. Also, we describe the second-
order model and show how it may be used for discriminating
Introduction between thick, blocky reservoirs and thinly laminated ones.
Traditional amplitude variation with offset (AVO) analysis Simmons and Backus (1994) numerically demonstrate the impor-
assumes that a seismic compressional (P) wave reflection is gener- tance of including converted shear energy in a layered model, and
ated by an incident plane P-wave backscattered from a single this is consistent with our second-order term, which is dominated
interface (Ostrander, 1984; Castagna and Backus, 1993; Hilterman, by the contrast of the S-wave reflectivity.
2001; Foster et al., 2010). Also, there are implicit assumptions Constructive interference (tuning) from an interval at certain
that reflections from the top and base of a reservoir layer are frequencies, incident angles (ray parameter), and layer thicknesses
resolvable. The conventional AVO model is based on a linearized accentuates the seismic amplitude response. Often, seismic pro-
version of the full P-wave reflection coefficient (Zoeppritz equa- cessing and inversion attempt to mitigate this amplitude bloom
tion). The Zoeppritz equation is linearized with respect to small due to tuning. Such processes as deconvolution and inversion
contrasts of material properties (Bortfeld, 1961; Aki and Richards, (particularly spiking deconvolution and sparse spike inversion)
1980). This approximate form results in a convolutional model try to collapse the wavelet to provide a wider bandwidth repre-
that ignores transmission effects, mode conversions, and multiple sentation of the reflectivity. These processes attempt to broaden
reflections but gives a first-order (Born) approximation for the the spectrum of the data with some a priori statistical assumptions
reflected wave. of the wavelet or the underlying structure of the earth model.
Here, we describe a new AVO model that considers the Particularly, in sparse spike inversion there is a bias that a simple
response from an interval and not an isolated interface. The model is preferred (Occam’s razor), but often well logs recorded
backscattered response is modeled as an integration over a specified through reservoirs show a multitude of small-scale internal het-
interval. This model is based on scattering theory (Foster et al., erogeneities. This heterogeneity is typically upscaled or averaged
2017) that uses a Born series approach to systematically add in in some way (i.e., blocking, sparse models, etc.). Here, we take a
complexities such as energy loss due to mode conversions and different approach and utilize tuning to emphasize the desired
internal multiple reflections. A frequency dependency in our rock and fluid response. The model upscales the fine details
model adds a new degree of freedom, which we exploit. We refer contained within the reservoir interval. One example of the use
to this frequency-dependent approach to AVO analysis as FAVO. of tuning is the delineation of a hydrocarbon water fluid contact
Chapman et al. (2002) also propose a FAVO method based on a as overlying reservoir bed (top seal) tunes with the fluid contact.
formulation incorporating intrinsic attenuation, whereas our Often, the subtle response of a fluid contact is visible only as
model is based on scattering effects. In this work, we will consider tuning accentuates the amplitude as the hydrocarbon portion of
only the first- and second-order scattering effects. The first-order the reservoir pinches out laterally. Another example is spectral
approximation is a primary-only P-wave convolutional model, decomposition (Partyka et al., 1999), which utilizes tuning to
and the second-order approximates the energy loss the P-wave bring out stratigraphic details not obviously seen on conventional
experiences due to shear (S) wave mode conversions. Multiple full-fold stack seismograms. Other examples are Marfurt and

1
The University of Texas at Austin, John A. and Katherine G. Jackson School of Geosciences, Institute for Geophysics, Austin, Texas, USA.
E-mail: djfoster@ig.utexas.edu; zeyu.zhao@utexas.edu; mrinal@ig.utexas.edu.
2
BP America Inc., Houston, Texas, USA. E-mail: dhananjay.kumar@bp.com; danica.dralus@bp.com.

84 THE LEADING EDGE February 2020 Special Section: Reservoir characterization Part I
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
Kirlin (2001) describing frequency-dependent attributes for thin-
bed resolution and Chopra and Marfurt (2007) for the interpreta-
tion of geomorphological features. Our approach is similar in
principle to spectral decomposition, but we have an underlying
physical model to quantify the reflectivity as we look for tuning
effects at specific thicknesses, frequencies, and angles of incidence
(horizontal ray parameter). This new model suffers from nonu-
niqueness, but this is true for all inversion methods.

First-order effects
The underlying model we employ is an interval where material
properties are vertically heterogeneous. This interval is embedded
between two infinite half spaces where properties are homoge-
neous. The traditional P-P reflection coefficient is replaced with
a transfer function that represents the backscattered P-wave energy
from an interval due to an incident P-wave of unit amplitude Figure 1. This figure shows the absolute value of the first-order transfer function.
impinging at the top of the heterogenous interval. This transfer The yellow zone encased in red shows the bandwidth of the observed field data in
terms of temporal frequency and incident angle. The angles are determined by ray
function can be thought of as a filter’s impulse response where
parameter p and the P-wave velocity of the incident wave.
the earth acts as a filter. In the limit as the thickness of the het-
erogeneous interval goes to zero, the first-order transfer function
reduces to the continuous form of Aki and Richards’ (1980) linear- One way is to perform a windowed Fourier transform around an
ized P-P reflection coefficient. interpreted horizon (in either depth or time) and find the frequency
We use the Born scattering series to systematically approximate that accentuates the desired response (fixed depth/traveltime and
the full solution for the transfer function Hp+p –. The full solution variable frequency). The other way is to partition the data into
for Hp+p – involves solving a nonlinear ordinary differential equation frequency bands, find the fixed frequency band that best shows
of which no general solution exists. The first-order transfer function the strongest response, and then analyze the data with respect to
for the reflected P-wave satisfies the differential equation depth or time (fixed frequency and variable depth/traveltime).
This latter approach is more practical because it allows the use of
the usual tools of conventional AVO analysis. Kumar et al. (2020)
d (1)
H + − ( p,z, ω ) = 2iωqα H (1)
p+ p −
( p,z, ω ) − R21 ( p,z) , (1) show in more detail how this can be performed in practice. Either
dz p p of these two approaches can benefit from modeling well logs to
determine which frequencies best show the desired response. In
and the integral form of the solution is the future, a more complete solution could be to use a time fre-
quency representation (i.e., wavelet transform, Gabor transform,
L etc.) in which both are variable. This may be a way to find the
H (1)
p+ p −
( p,z, ω ) = ∫e −2iωθα ( p,zʹ ,z)
R21 ( p, zʹ)dzʹ, (2) optimal windowing in time and frequency.
z Figure 1 shows the first-order transfer function, H (1) p +p – , using
wireline measurements through an interval of shale and gas-
where the P-P reflection coefficient R 21 (equation 17, Foster et al., charged siltstone. The measurements used to calculate H (1) p +p – are
2017) is the continuous form of the first-order approximate P- and S-wave velocities and density. This interval is characterized
Zoeppritz equation (Aki and Richards, 1980). The P-wave vertical by strong impedance contrasts on a fine scale. Also, this interval
is within the wavelength of the seismic wavelet. There is no wavelet
slowness is qα = 1
− p 2, where p is the ray parameter (horizontal used in this graph as it shows the natural impulse response of this
α2 interval. The absolute value of H (1)p +p – is shown in Figure 1, but in
slowness), z is depth, ω is the angular frequency, and the super- some instances the real or imaginary components might show
scripts + and - represent downgoing and upgoing directions of more stratigraphic detail (i.e., discriminating between coarsening-
propagation, respectively. The range of integration is from the upward or fining-upward sequences, Kumar et al., 2020). Prior
observation location z to L, and the heterogeneous interval is to drilling this well, the seismic data showed a strong amplitude
[0, L]. The discrete form of equation 2 is discussed in Kumar and a pronounced class III AVO anomaly. The farthest angle used
et al. (2017). in the AVO analysis was 45°, and a strong amplitude bloom at
The transfer function, H (1)p+p – , in equations 1 and 2, is a complex the higher angles was observed. It was difficult to explain this
function with real and imaginary components. The frequency AVO anomaly by fitting Zoeppritz curves to match this observa-
dependency adds an additional degree of freedom that can be tion. This figure shows the loss of bandwidth at the higher angles
exploited but also adds complexity on how to best utilize this due to longer propagation distance and normal moveout correction.
additional parameter. H (1)
p +p – can be thought of as having a dual It appears that constructive interference at the lower frequencies
domain representation with respect to depth (or traveltime) and and higher incident angles produces a tuning effect that enhances
frequency. We have analyzed seismic data in two different ways. the amplitude and makes the AVO anomaly stronger than a

Special Section: Reservoir characterization Part I February 2020 THE LEADING EDGE 85
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
one-interface model would suggest. This example shows how expectations can provide constraints on the viability of a given
frequency dependence can complicate the AVO response. attribute as a predictive tool. The morphology of attributes should
One application of FAVO is distinguishing between low gas be consistent with the geometry of a given structural setting,
saturation and commercial concentrations. This problem has hydrocarbon effects should be in plausible structure positions,
plagued amplitude analysis because a small amount of gas is stratigraphic variations should be consistent with the expected
indistinguishable from higher levels. Investigations by Castagna depositional environments, etc.
et al. (2003), Goloshubin et al. (2014), and Ahmad et al. (2019)
relate low-frequency seismic anomalies and the presence of gas.
Figure 2 shows a plot of gas saturation versus seismic amplitude.
Here we see that the 5 Hz response shows more sensitivity to gas
saturation than the 10 Hz response.
Another way of looking at the sensitivity of saturation is by
crossplotting the AVO attributes intercept (A) and gradient (B).
Figure 3 demonstrates that the frequency-domain amplitude at
the 20 Hz response shows greater sensitivity to changes in satura-
tion than the time-domain amplitude.
This model is derived from properties given by a well log
containing a relatively thick, porous sandstone layer, and the
saturations are based on Gassmann fluid substitution.
The application of new techniques on field data is a crucial
test, but it can be hard to completely evaluate and verify the Figure 2. This figure (after Kumar et al., 2020) shows the real part of the modeled
amplitude of a 20° centered angle stack versus gas saturation. Equation 2 is used
results. Nevertheless, it is informative to judge how a new method in the seismic FAVO modeling, and Gassmann fluid substitution is used to estimate
behaves under realistic conditions. The results should be con- the effects of different gas saturations on earth model parameters. Red squares
sistent with expectations for a given geologic setting. These indicate the response at 10 Hz, and blue squares indicate the response at 5 Hz.

Figure 3. Four crossplots between AVO intercept and AVO gradient are displayed (after Zhao et al., 2018). Panels (a) and (b) show time-domain conventional crossplots
for low and higher saturation levels. (a) The diamonds are for low saturation as a function of time windowed around the event of interest, and (b) the squares are the
same window but with a higher saturation level. (c) and (d) The same as (a) and (b) except the frequency attributes are displayed.

86 THE LEADING EDGE February 2020 Special Section: Reservoir characterization Part I
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
Figure 4. After Kumar et al., 2020. (a) A conventional rms amplitude extracted from a window around a horizon at the top of gas-bearing sandstone reservoir. (b) The FAVO
attribute is derived from the 10 Hz imaginary component of the midangle seismic (centered around 20°). High gas saturation is present in wells 1, 2, and 3 in this reservoir unit.

Figure 5. After Kumar et al., 2020. Similar to Figure 4, but the attribute extracted is from a deeper formation. Another difference is the FAVO attribute is derived from the
5 Hz real component. High gas saturation is present in wells 2 and 3, but low gas saturation was encountered in well 1.

For the field data test, three wells and a marine 3D seismic Ultimately, well 3 was drilled in the anomaly and encountered
volume recorded with a conventional air gun array and recorded commercial concentrations of gas.
on towed streamers are used. These data come from the Columbus Figure 5 shows the conventional and FAVO attributes for a
Basin offshore Trinidad. The reservoirs of interest are of Pliocene deeper formation. In this stratigraphic interval, well 2 has com-
and Pleistocene age with porosities in the 20%–30% range with mercial quantities of gas, but well 1 is wet with low-saturation
high net-to-gross of sandstone to shale. Figure 4 shows a con- gas. Overall, both of the conventional and FAVO attributes look
ventional rms amplitude and a FAVO attribute. relatively similar. At well 1, neither one clearly shows any anoma-
The same stratigraphic interval is shown in Figure 4. All three lous behavior, which is the desired response. At well 2, the FAVO
wells in this sequence are gas bearing with commercial concentra- attribute is more intense in the reservoir interval than the con-
tions. Wells 1 and 2 were used as calibration, and well 3 was a ventional attribute. The predrill prediction for this interval was
proposed well location that was used as a blind test. Overall, the high gas saturation, and the well confirmed this prediction. As
FAVO attribute is more coherent and appears to better delineate shown in Figure 4, the bounding faults are seen more clearly on
the known gas-bearing zones. At well location 1, both attributes the FAVO attribute. The interpreted gas-water contact is apparent
show a small anomaly with the FAVO attribute slightly more on both attributes at well 3. Also, both anomalies at well 3 show
coherent. Both are brighter updip against a fault and dim downdip, relatively good conformance to structure, but one difference is
which gives confidence that both attributes are showing a hydro- the FAVO attribute changes polarity downdip, which can aid in
carbon effect, which is expected in these relatively porous reser- highlighting the fluid contact.
voirs. At well location 2, again both show anomalous amplitudes,
but the FAVO attribute shows a more consistently coherent Second-order effects
amplitude. The FAVO attribute more clearly shows amplitudes In elastic scattering, due to wave mode conversion and mul-
terminating against bounding faults. At well 3, similar to well 1, tiples, mathematical expressions quickly become intractable. These
both attributes are stronger in the updip trapping configuration expressions can be simplified to a manageable form using the
and dim downdip, but the FAVO attribute shows more downdip Born series. Although the Born series is not exact, it can provide
conformance. The predrill prediction was a narrow gas pay with useful approximations. One advantage of this approach is higher
the possibility of downdip low saturation at the well location. order effects such as mode conversions and multiples can be

Special Section: Reservoir characterization Part I February 2020 THE LEADING EDGE 87
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
systematically added. Traditionally, only the first term is used, Figure 6 shows there is no apparent difference between the two
and this is equivalent to the convolutional model. The second different scenarios. This result is expected because equation 2 simply
term of the Born series approximates the effects the P-wave integrates the entire interval. No loss of energy due to mode conver-
experiences from energy losses due to S-wave conversions. Multiple sions is accounted for in the first-order term of the approximation.
reflections are not considered until the third and higher order Higher order terms will provide a more complete approximation
terms are included. correcting the first-order term. The second-order term only corrects
One application is the use of the second-order gradient attri- for forward and backscattered S-wave conversions.
bute for discriminating between thick reservoirs and thinly strati- Figure 7 displays the first and second terms, H (1) (2)
p +p – + H p +p – ,
fied ones. Ideally, one hopes to encounter thick, clean sands. This for the same model shown in Figure 6. Foster et al. (2017) give
is often not the case, however, and a tool to determine the extent the expression for H (2)
p +p –. Figure 7a shows distinct difference from
of fine layering can be useful. Also, many times, very thinly layered Figure 7b around 30° and 10 Hz. Also, Figure 7b shows a neg-
sandstone hydrocarbon-bearing reservoir units encased in thicker ligible difference between that seen in Figures 6a and 6b. The
shale can be missed and can lead to pay being bypassed. reason for the difference between Figure 7a and Figure 7b is that
Conventional well measurements, particularly resistivity measure- in Figure 7b there are only two interfaces where shear conversion
ments, also can miss significant quantities of hydrocarbons in this will occur, but in Figure 7a there are 200 interfaces for mode
type of environment (Henderson et al., 2000). conversion. Although these are second-order effects, the accu-
Synthetic models can illustrate the potential of this second- mulated effect over this interval is noticeable at the higher incidence
order attribute. We use a couple of simple models of an idealized angles. Another observation is that the effects are seen at certain
200 ft thick reservoir. One scenario is 100 one-foot-thick layers temporal frequencies (harmonics). Here, 10 Hz is the first har-
of clean sandstone periodically embedded in 100 one-foot shale monic, and it clearly shows a difference.
layers. The other scenario is a single 200 ft block of sand with Zhao et al. (personal communication, 2019) give the second-
40% volume of shale (Vsh). Both scenarios are modeled as gas order AVO gradient term, B (2), and show how to calculate it from
bearing. Figure 6 shows the first-order transfer functions, H (1) p +p – , the first-order term. For brevity, we do not show the second-order
for both scenarios. gradient term from an interval but from an isolated interface. This

Figure 6. This figure shows the imaginary part of H (1)


p p for two different scenarios.
+ – Figure 7. This figure shows imaginary parts H (1) (2)
p p + H p p for same models shown
+ – + –

Both models are 200 ft thick but in (a) there are 100 one-foot-thick sands evenly in Figure 6. In (a), the 100 layer clean sand and shale is seen, and in (b) the one
embedded in one-foot shales and in (b) there is a single 200 ft sand with 40% 200 ft shaley sand model is displayed. The Vsh in (b) is 40%.
shale Vsh content.

88 THE LEADING EDGE February 2020 Special Section: Reservoir characterization Part I
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
is given by taking the limit as the interval thickness goes to zero. equations 3 and 4 show, the shear velocity changes play a more
This expression is
1
prominent role. Equation 3, or equation 4 when ε ≈ , can be
2 2
2
⎛ ρ′ ⎞ ρ′ β ′ ⎛ β′ ⎞
B ( 2) = −2ε (1 − 4ε 2 ) ⎜ + 32ε 3 + 32ε 3 ⎜ , (3) used directly when analyzing seismic data in different frequency

⎝ 2ρ ⎠ 2 ρ 2β ⎝ 2β ⎟⎠
bands (fixed frequency and variable depth/traveltime).
To further show the effects of thin beds, we consider five
β background
where ε = . Assuming ε ≈ 1 , which is a realistic assump- different models with varying degrees of layering (Figure 8).
α background 2 Figure 9 shows the B (2) attribute for the models shown in Figure 8.
tion for clastic rocks, particularly shales, we have In this figure, from left to right as the individual layers become
thicker and fewer in number, we see that the amplitude of B (2)
decreases. This result shows that constructive interference of
β ʹ ⎛ ρʹ β ʹ ⎞
B ( 2) = 4 ⎜ + ⎟, (4) fine-scale layering may be useful in differentiating vertically
2β ⎝ 2 ρ 2β ⎠
heterogeneous reservoirs from thicker, blockier reservoirs. This
suggests that B (2) may be useful as an attribute that can provide
ρʹ βʹ
where and are relative changes of the density insight to the internal stratigraphic detail within a reservoir.
2ρ 2β
Again, it is difficult to validate the results on field data tests,
and shear velocity, respectively. The derivatives are with respect but qualitative information might provide useful insights into the
to depth (or traveltime) and are denoted by the prime, ' . These practical issues of implementation. Figure 10 shows an amplitude
derivatives can be replaced by differences. Notice there is no extraction of B (2) from two different intervals with varying degrees
contribution to B (2) from changes of P-wave velocity contrasts, of layering. These extractions come from a 3D volume of seismic
only density and shear velocity changes contribute. As data from offshore Trinidad. The amplitudes from the thin sands,
seen in Figure 10b, show a more coher-
ent anomaly in the center around well 1.
The anomaly in Figure 10a is weaker
overall and not as compelling as the one
on 10b. Both of these extractions come
from the 10 Hz calculated B (2) volumes,
and the 10 Hz response is stronger than
ones derived at other frequencies. This
supports one of our main premises —
that different frequencies can be used
to emphasize the desired response.

Conclusions
Figure 8. Stratigraphy Vsh curves for the five synthetic models are shown. Thicknesses and the numbers of thin A new frequency-dependent gen-
sand and shale layers in each of the models are variable. The values of P-wave velocity, S-wave velocity, and eralized AVO model is introduced, and
density values are for typical shales and gas-charged 25% porosity sands. two practical applications are discussed.
The new approach is based on a Born
scattering series approximation that
allows for the complexity of wave propa-
gation to be systematically included.
Frequency-based attributes have proven
useful in a variety of applications for
detecting subtle features on seismic data.
Our FAVO model, in a quantified way,
allows us to include tuning effects to
accentuate features of interest in reser-
voir description. Two different applica-
tions are explored: one discriminating
low versus commercial saturations of
gas and the other discerning blockier
reservoirs from small-scale thinly strati-
fied reservoirs.
Low versus high saturation of
hydrocarbons has long been a challenge
Figure 9. B (2) attributes for the five models shown in Figure 8. when using seismic amplitudes for

Special Section: Reservoir characterization Part I February 2020 THE LEADING EDGE 89
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
Figure 10. This figure shows the amplitude extraction of the B (2) attribute from two different stratigraphic intervals. This attribute comes from 10 Hz angle stack seismic
sections. Panel (a) is from a shallower interval with blockier sands, and (b) is from a deeper formation with thinner sands. The B (2) attribute is a real function.

predrill predictions. Here, we used the first-order term in the analysis, the one-interface physical model has served us well,
Born series to address this challenge. The constructive interference but we demonstrate that there is more information to be dis-
of the in-situ reservoir properties enhances the differences of the covered when generalizing the model to include backscattering
response between low and higher saturations in synthetic models. from an interval. Most spectral attributes are qualitative, and
The field data example is not conclusive, but the frequency-based here we provide a physical model that can serve as a way to
attribute is consistent with the wells within the area of study. This quantify results.
attribute uses tuning to accentuate the subtle difference of satura-
tion levels and may be useful in this long-standing problem. Acknowledgments
Often, reservoirs are not thick, blocky, clean sandstone but We thank BP for permission to publish. Special acknowledg-
consist of many small-scale vertical heterogeneities. The second- ments are to Rosemarie Geetan, John Etgen, Margarita Corzo,
order term of the series is used to distinguish between these two Kevin Wolf, and Tara Rampersad for various technical discussions
types of reservoirs. The first-order term does not conserve energy and support. We thank Dave Lane and Chuan Yin for their
of the backscattered wavefield, but the second-order term attempts assistance in establishing the underlying theoretical basis,
to approximate the energy loss that the P-wave experiences due Vaughn Ball for his insights into physical models used in AVO
to S-wave mode conversions. The more laminated the reservoir, analysis, and Paul Richards for his astute inquiry about what is
the more mode conversions. This is seen on B (2) at the larger being recorded on reflection seismograms — specular reflections
incidence angles (approximately 30°) and at certain tuning frequen- versus coherently backscattered energy. Finally, Tad Ulrych is
cies. Synthetic modeled data show that the intensity of the second- acknowledged for inspiring this work by stressing the information
order gradient attribute increases with increased layering. The field contained in seismic data is amplitude and phase: one should not
data application shows this attribute is recovered stably from seismic solely analyze amplitudes without considering phase.
data and provides an interpretable estimate of the degree of reservoir
stratification. This attribute may be useful for detecting bypassed Data and materials availability
pay as well as providing additional reservoir properties. Data associated with this research are confidential and cannot
This interval-based model and its applications presented here be released.
are to be considered as complementary to existing conventional
types of analysis and attributes. In terms of conventional AVO Corresponding author: djfoster@ig.utexas.edu

90 THE LEADING EDGE February 2020 Special Section: Reservoir characterization Part I
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user
References
Ahmad, S. S., W. W. Weibull, R. J. Brown, and A. Escalona, 2019, Observations and
suggested mechanisms for generation of low-frequency seismic anomalies: Examples
from the Johan Sverdrup Field, central North Sea Norwegian sector: Geophysics, 84,
no. 1, B1–B14, https://doi.org/10.1190/geo2018-0144.1.
Aki, K., and P. G. Richards, 1980, Quantitative seismology: W. H. Freeman.
Bortfeld, R., 1961, Approximation to the reflection and transmission coefficients of plane
longitudinal and transverse waves: Geophysical Prospecting, 9, no. 4, https://doi.
org/10.1111/j.1365-2478.1961.tb01670.x.
Castagna, J. P., and M. M. Backus, 1993, Offset-dependent reflectivity — Theory and
practice of AVO analysis: SEG, https://doi.org/10.1190/1.9781560802624.
Castagna, J. P., S. Sun, and R. W. Siegfried, 2003, Instantaneous spectral analysis:
Detection of low-frequency shadows associated with hydrocarbons: The Leading Edge,
22, no. 2, 120–127, https://doi.org/10.1190/1.1559038.
Chapman, M., S. V. Zatsepin, and S. Crampin, 2002, Derivation of a microstructural
poroelastic model: Geophysical Journal International, 151, no. 2, 427–451, https://
doi.org/10.1046/j.1365-246X.2002.01769.x.
Chopra, S., and K. J. Marfurt, 2007, Seismic attributes for prospect identification and
reservoir characterization: SEG, https://doi.org/10.1190/1.9781560801900.
Foster, D. J., R. G. Keys, and F. D. Lane, 2010, Interpretation of AVO anomalies:
Geophysics, 75, no. 5, 75A3–75A13, https://doi.org/10.1190/1.3467825.
Foster, D. J., F. D. Lane, and Z. Zhao, 2017, A systematic approach for quantifying wave
propagation in vertically inhomogeneous media: Geophysical Journal International,
210, no. 2, 706–730, https://doi.org/10.1093/gji/ggx186.
Goloshubin, G., Y. Tsimbalyuk, I. Privalova, and P. Rusakov, 2014, Low-frequency
amplitude analysis for oil detection within the Middle Jurassic sediments in the southern
part of Western Siberia: Interpretation, 2, no. 4, SP35–SP43, https://doi.org/10.1190/
INT-2014-0040.1.
Henderson, K., H. J. Rose, and R. Winter, 2010, Identifying and quantifying thin-bedded
pay (part A): Log characteristics and reservoir quality: Trinidad and Tobago Energy
Resource Conference, SPE 133534-MS, https://doi.org/10.2118/133534-MS.
Hilterman, F. J., 2001, Seismic amplitude interpretation: SEG/EAGE Distinguish
Instructor Series, no. 4, https://doi.org/10.1190/1.9781560801993.
Kumar, D., Z. Zhao, D. J. Foster, D. Dralus, and M. K. Sen, 2020, Frequency-dependent
AVO analysis utilizing scattering response of a layered reservoir: Geophysics, https://
doi.org/10.1190/geo2019-0167.1.
Kumar, D., Z. Zhao, D. Foster, and M. K. Sen, 2017, Frequency-dependent AVO analysis
based on scattering series: 87th Annual International Meeting, SEG, Expanded Abstracts,
534–538, https://doi.org/10.1190/segam2017-17677300.1.
Marfurt, K. J., and R. L. Kirlin, 2001, Narrow-band spectral analysis and thin-bed tuning:
Geophysics, 66, no. 4, 1274–1283, https://doi.org/10.1190/1.1487075.
Ostrander, W. J., 1984, Plane-wave reflection coefficients for gas sands at nonnormal
angles of incidence: Geophysics, 49, no. 10, 1637–1648, https://doi.org/10.1190/1.1441571.
Partyka, G., J. Gridley, and J. Lopez, 1999, Interpretational applications of spectral decom-
position in reservoir characterization: The Leading Edge, 18, no. 3, 353–360, https://
doi.org/10.1190/1.1438295.
Richards, P. G., and C. W. Frasier, 1976, Scattering of elastic waves from depth-dependent
inhomogeneities: Geophysics, 41, no. 3, 441–458, https://doi.org/10.1190/1.1440625.
Simmons, J. Jr., and M. Backus, 1994, AVO modeling and the locally converted shear
wave: Geophysics, 59, no. 8, 1237–1248, https://doi.org/10.1190/1.1443681.
Zhao, Z., D. J. Kumar, D. Foster, D. Dralus, and M. K. Sen, 2018, A new frequency dependent
reflectivity model and estimating seismic AVO attributes: 88th Annual International Meeting,
SEG, Expanded Abstracts, 416–420, https://doi.org/10.1190/segam2018-2997403.

Special Section: Reservoir characterization Part I February 2020 THE LEADING EDGE 91
Downloaded from https://pubs.geoscienceworld.org/tle/article-pdf/39/2/84/4938377/tle39020084.pdf
by University of Cambridge user

You might also like