A Simple, Robust Algorithm For Seismic Net Pay Estimation: P C, BP, Sunbury-On-Thames, U.K

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

A simple, robust algorithm for seismic net pay estimation

PATRICK CONNOLLY, BP, Sunbury-On-Thames, U.K.


T
Downloaded 06/27/22 to 125.209.89.130. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

here has been a long history of using seismic attributes


to estimate net-to-gross or net pay (Brown et al, 1984). The
type of approach discussed here employs map-based com-
pensation of tuning effects followed by direct calibration to
the reservoir property rather than via an intermediate esti-
mate of absolute impedance. This article describes a process
based on band-limited impedance data updated to incor-
porate more recent advances in inversion and AVO.
The objective for this algorithm is to estimate the total
net pay between two seismic picks made on zero-crossings
of band-limited impedance data. The method is based on
the assumption that the integral of band-limited impedance
between zero crossings is proportional to net pay if we first
detune the response of the gross interval. In broad terms,
I’ll first relate the average band-limited impedance to net-
to-gross and then the integral of band-limited impedance
to net pay. This process will have two main steps; detuning Figure 1. (top) Wedge model consisting of a wavelet convolved with a
and calibration. Using this framework, I derive a simple and thickening boxcar impedance profile. Horizons picked on top and base
robust algorithm with application over a wide range of zero crossings. (bottom) Average band-limited impedance and the
gross interval values. This article describes the method and apparent thickness between picks.
a companion article in this issue (Connolly and Kemper)
investigates the uncertainties of the process.

Wedge model revisited. Let’s examine the detuning step by


first considering a 100% net-to-gross reservoir. Figure 1 (top)
DOI:10.1190/1.2794386

shows a band-limited impedance wedge formed by apply-


ing a trapezoidal filter to a simple low impedance wedge
model. I’ve picked horizons along the zero crossings at the
top and base of the unit. The time separation between zero
crossings, the apparent thickness, converges on a minimum
value as the wedge gets thinner (Figure 1, bottom) while
the average band-limited impedance, measured between
the picks, exhibits a tuning response.
Usually we won’t know the true gross interval so Figure
2 redisplays the average band-limited impedance tuning
curve as a function of the apparent thickness. The reservoir
model has a net-to-gross of unity for all thicknesses so, at
first sight, it may appear that to detune this attribute to allow
it to be used as a basis for net-to-gross prediction we would
want to multiply by it’s reciprocal to make it independent
of the apparent thickness. In other words, transform it to a
horizontal line. Figure 2. Tuning curve: Average band-limited impedance for 8-12-60-
80 Hz trapezoidal-filtered wedge model plotted against apparent thick-
The shape of the tuning curve is of course dependent ness.
on the wavelet. Because the tuning curve is changing rapidly
for small apparent thicknesses, it looks as if the detuning
would be an error prone process, highly sensitive to the approach a minimum value; hence the seismic net-to-gross
wavelet estimation. This “full” detuning is equivalent to will fall. We can derive a detuning transform to get from
attempting to estimate the true gross interval something we average band-limited impedance (Figure 2) to seismic net-
know is very difficult (Fervari and Luoni, 2006). However, to-gross (Figure 3a) by dividing the latter by the former.
the objective we set ourselves was to estimate the total net Because, for low apparent thickness, the shape of these two
within the apparent thickness. To obtain this, what we need curves is very similar, the shape of the detuning correction
is the ratio of total net to the apparent thickness, which I’ll function turns out to be a fairly simple curve (Figure 3b).
refer to as seismic net-to-gross. The product of seismic net-
to-gross and (depth converted) apparent thickness will then Variable net-to-gross. Let’s now consider what happens
provide net pay. As we’ll see, estimating seismic net-to- when the net-to-gross is less than unity. We’ll assume our
gross is much easier than true net-to-gross. reservoir broadly conforms to a binary impedance model;
Figure 3a shows the seismic net-to-gross for our 100% pay has one impedance value and non-pay a different
true net-to-gross reservoir as a function of apparent thick- impedance value. Given this model, for full bandwidth data
ness. For large gross intervals, seismic net-to-gross will be (i.e., logs), net-to-gross is proportional to average imped-
the same as true net-to-gross because apparent thickness is ance. The key assumption of this net pay estimation proce-
equal to true thickness. However, as the gross thickness dure is that, for the band-limited case, the average of the
reduces, the net will reduce but the apparent thickness will band-limited impedance over the apparent thickness is pro-

1278 THE LEADING EDGE OCTOBER 2007


Downloaded 06/27/22 to 125.209.89.130. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 3. (a) Seismic net-to-gross for 100% net-to-gross wedge model. (b) Detuning correction curve to transform the average band-limited
impedance tuning curve (Figure 2) to seismic net-to-gross.

portional to seismic net-to-gross for a fixed apparent thick-


ness and that the dependence on apparent thickness can be
then removed using the detuning algorithm outlined in the
previous section. The proportionality assumption will be
examined in the companion article.
Figure 4 illustrates some of these concepts. It shows
DOI:10.1190/1.2794386

three models; model A has a net of 4ms of low impedance


sand with a true net-to-gross of 1; model B also has a net of
4 ms but is interbedded with shale giving a true net-to-gross
of 0.57; model C has 2 ms of net and a true net-to-gross of
0.5. Superimposed are the band-limited impedance traces.
The synthetics for models A and B are very similar in shape
and amplitude demonstrating the point made earlier that,
for these subtuning intervals, the seismic responds to the
total net and not the net-to-gross.
The apparent thicknesses of these models are 16, 17, and
16 ms, giving seismic net-to-grosses of 0.25, 0.24, and 0.125
respectively. As the apparent thicknesses are just about the
same, this provides a test of the assumption that seismic net-
to-gross is proportional to average band-limited impedance.
The table in Figure 4 gives the ratios of these two for each
model; they’re consistent within a few percent confirming
the validity of the assumption for these cases.
This, and other examples, shows that we should be able
to estimate net pay from arbitrarily thin reservoirs. What is
the upper limit of the gross interval for which average band-
limited impedance and seismic net-to-gross remain pro-
portional? It turns out that as apparent thickness increases,
the relationship begins to break down because the seismic
response increasingly depends on how the net is distributed
within the reservoir. This will be examined in more detail
in the companion paper in which we explore the uncer-
tainties of this relationship, but a good rule of thumb is that
the upper limit for usefully accurate results is for an appar-
ent thickness equal to the half-cycle of the lowest frequency
component. For the 8-12-60-80 Hz trapezoidal filter, this
occurs at about 60 ms.
The binary impedance assumption is a reasonable
approximation for many environments, for example, Tertiary
marine sediments where porosity tends to follow a com-
paction curve. As depth increases, porosity variations
become more significant. Extended elastic impedance the-
ory can often be used to get closer to the binary impedance Figure 4. Three impedance models and their band-limited impedance
traces.

OCTOBER 2007 THE LEADING EDGE 1279


calibration wells.
There can, however, be many difficulties tying seismic
to wells; even determining the correct relative positioning
of the well and seismic data can be difficult. In some depo-
sitional environments, particularly channelized reservoirs,
Downloaded 06/27/22 to 125.209.89.130. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

a change of location of a trace or two can make a very sig-


nificant difference to the calibration. Furthermore, because
average band-limited impedance is only approximately pro-
portional to seismic net-to-gross, it follows that even a per-
fect well tie can bias the calibration. This will be further
discussed in the companion paper. Given these factors, a well
tie for net pay calibration may not provide the solid refer-
ence point that we would hope for.
Because of these shortcomings, any geologic insight
should also be brought to bear to guide the calibration. For
example, deep marine turbidite systems almost invariably
seem to have a maximum seismic net-to-gross of unity when
measured over a 20–30 m thickness. And of course we have
a hard constraint because seismic net-to-gross can’t exceed
Figure 5. Detuning correction curves for 6-10-60-80 Hz and 8-12-60- unity (actually it can a little because just above the tuning
80 Hz wavelets. Also shown is the % error that would occur if the
second wavelet was correct but the first had been estimated. This will point the apparent thickness is less than true thickness). One
be partially compensated for by the “recalibration” that would follow of the reasons for going through the intermediate step of
from fitting to well data and reducing the rms error over 0-60 ms from estimating seismic net-to-gross rather than going straight
20% to 13%. to net pay is to be able to apply these additional constraints.
In practice, calibration remains a somewhat delicate process
that should ensure consistency with all available well data
and provide a geologically sensible net-to-gross map.

Sensitivity. The detuning correction (Figure 3b) increases


with apparent thickness with the slope being dependent
DOI:10.1190/1.2794386

upon the low-frequency wavelet parameterization. Hence,


for large apparent thickness, the detuning factor is very sen-
sitive to an estimation of the low frequencies in the data.
Figure 5 shows detuning correction curves for 6-10-60-80 Hz
and 8-12-60-80 Hz wavelets. If the correct wavelet had a low-
end of 8–12 Hz but had been estimated as 6–10 Hz then the
figure shows a resultant under-correction of over 30% at 60
ms and an rms error over all thicknesses up to 60 ms of 20%.
However, in practice the calibration step will partially com-
pensate for this; to fit the well data the calibration factor
Figure 6. Section through West African deep marine reservoir. This is based on a 6–10 Hz wavelet would need to be increased
a far-offset, colored inversion data set. The lower unit has been picked resulting in the “recalibrated” 6-10-60-80 Hz curve. This
on top and base zero crossings.
could result in an rms error as low as 13%. The partial inter-
model ideal by choosing a chi angle projection that max- changeability between the low-frequency estimation and cal-
imises the correlation between the EEI log and the corre- ibration factor results from the simplicity of the detuning
sponding Vshale curve (Whitcombe et al., 2002; Neves et curve and makes for a more robust algorithm.
al., 2004). The companion article describes a way to deter- Unlike attributes that respond to the total energy of the
mine the affect of departures from the binary impedance seismic waveform, such as rms, the average value of a por-
model on the accuracy of net pay estimates. tion of a seismic trace, whether impedance or reflectivity, is
highly sensitive to the position of the picks. To obtain accu-
Calibration. In addition to the detuning, we must also rate average band-limited impedance values, it is therefore
include a calibration factor to transform from the arbitrary important that the picks are snapped on to the correct zero
band-limited impedance values to seismic net-to-gross. If crossings at the top and base of the reservoir. In practice,
the depth range of the reservoir is not too large, it should determining what the correct zero crossing is may not be
be reasonable to assume spatially invariant rock property straightforward. As we’ll see in the next paper, even in the
relationships and hence a constant calibration factor. noise-free case there is ambiguity. This is therefore a source
However, if we have adequate calibration data, a depth- of additional uncertainty but is within the framework inves-
dependent or spatially varying calibration can be performed. tigated in the companion article.
The process to obtain an optimum calibration proceeds
iteratively by first determining a calibration factor that Algorithm summary.
results in a reasonably sensible seismic net-to-gross map for 1) Pick top and base of reservoir on zero crossings of band-
example, by setting the maximum seismic net-to-gross to limited impedance data.
unity. This is then converted to net pay by multiplying by 2) Extract average band-limited impedance.
the depth converted apparent thickness. If well data are 3) Estimate wavelet.
available, an updated calibration factor can then be deter- 4) Calculate detuning curve (Figure 3b).
mined by minimizing the misfit between the seismic net pay 5) Apply detuning curve to average band-limited imped-
map and the petrophysical net pay values measured at the ance.

1280 THE LEADING EDGE OCTOBER 2007


Downloaded 06/27/22 to 125.209.89.130. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 7. (a) Cross-plot of average band-limited impedance against apparent thickness for the lower reservoir unit shown in Figure 6.
Superimposed is a modeled tuning curve. (b) Cross-plot of the predicted net pay values against the actual net pays for three wells after applying
the final calibration.

6) Apply first guess calibration and calculate net pay. ent thickness for the entire reservoir. From the discussion
7) Adjust calibration using well data. in previous sections our assumption is that, for any given
apparent thickness, the points at the top of the cloud will
Band-limited impedance can be obtained using colored have a seismic net-to-gross of about one and lower points
inversion (Lancaster and Whitcombe, 2000). The user should will have proportionally lower seismic net-to-gross values.
be guided by the rule-of-thumb that the separation between Inspection of the data shows that outliers above or to the
picks should not be greater than the half-cycle of the low- left of the curve are mostly caused by mis-picks.
est frequency component of the wavelet. Results for greater A modeled tuning curve calculated from an estimated
DOI:10.1190/1.2794386

thicknesses should be used with caution. The minimum wavelet is superimposed at the top of the envelope of points.
apparent thickness is dependent on the bandwidth of the This provides an initial self-calibration of the data and also
data. The algorithm is accurate for arbitrarily small true a check on the data quality by ensuring the top of the enve-
thickness (although signal-to-noise will decrease for thin lope has a similar shape to the modeled response, although
reservoirs). Multiple events can be included within the picks. this will also depend on the reservoir geometry. The tuning
In other words, both positive and negative data values. (But correction curve is next applied and the calibration adjusted
even so, if the reservoir corresponds to negative imped- to fit the available well data. In this case, there are three well
ance, the average band-limited impedance will always be penetrations and, after adjustment, a consistent calibration
negative.) The reservoir must be isolated; in other words the is obtained (Figure 7b). The reservoir had low relief so a sin-
wedge model must apply. First guess calibration can be gle calibration factor was used. Figure 8 illustrates succes-
based on scaling the seismic net-to-gross map maximum to sive stages of the process.
one.
Discussion: low frequencies. The approach described above
Example. Figure 6 shows a section through an offshore West is based on the average value of an attribute, the band-lim-
African Tertiary deep marine oil field. This is a far-offset, ited impedance, across the full reservoir interval. This is in
colored inversion data set. The far-offset impedance logs contrast to some previous methods that use trough and
have a good correlation with the Vshale curves above the peak amplitudes from reflectivity data at the top and base
oil-water contact. The lower reservoir unit has been picked of the reservoir. Peak and trough methods can only be an
on the zero crossings at top and bottom. The upper unit is effective basis for net pay estimation for thin beds. The
mostly far enough away for it not to affect the amplitudes approach described here aims to provide useful results over
of the lower unit and so adheres to the
requirement that the reservoir must be
isolated.
The minimum frequency in the data
is about 6 Hz, which would imply a
maximum thickness of about 80 ms for
applicability of this algorithm. The bulk
of the reservoir lies well within this limit
although a few locations exceed it so
these results should be flagged as unre-
liable. As can be seen in Figure 6, in
places the picks encompass both posi-
tive and negative data values; this does
not violate the assumptions of the algo-
rithm.
Figure 7a is a crossplot of average
band-limited impedance against appar-

OCTOBER 2007 THE LEADING EDGE 1281


Downloaded 06/27/22 to 125.209.89.130. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 8. Maps of the apparent thickness, average band-limited impedance and the estimated seismic net-to-gross and net pay for the lower reser-
voir unit from Figure 4. Low values are in blue, high values in red.

a larger range of reservoir thicknesses. number of parameters to select, each with well defined
The rate at which the average band-limited impedance selection criteria (chi angle for EEI and bandwidth and spec-
values decrease as the apparent thickness increases depends trum exponential for CI) so each may be regarded as rela-
on the low frequency component of the data. The detuning tively robust. Because the processes are modular, in that each
DOI:10.1190/1.2794386

process is effectively just boosting the contribution of low step can be carried out independently, the robustness of the
frequencies to the average band-limited impedance mea- complete sequence is retained. Reduced subjectivity, and
surement so, as previously discussed, there will be an upper hence robustness, is a prerequisite for quantitative uncer-
limit to the reservoir thickness for this method. For most tainty estimation which will be discussed in the compan-
seismic data, the low-frequency signal usually drops below ion paper.
the noise level between 5 and 10 Hz which would translate
to a maximum thickness of 50–100 ms. If we can obtain even Conclusion. I’ve described a simple and robust algorithm
lower frequencies by better acquisition, as appears possible for estimating net pay from band-limited impedance data.
with sea bed systems, or by stabilization during the pro- It has broad range of applicability from arbitrarily thin reser-
cessing (Whitcombe and Hodges, 2007), we could potentially voirs up to a thickness equal to the half cycle of the mini-
extend the applicability of the method to 150 ms or more. mum frequency in the seismic. The companion paper
The procedure I’ve outlined makes full use of the low- describes a method to estimate the accuracy of the net pay
frequency content of the seismic data without the need to predictions.
add low frequencies from another source with all the (prob-
ably unquantifiable) additional uncertainty that that entails. Suggested reading. “Interactive seismic mapping of net pro-
ducible gas sand in the Gulf of Mexico” by Brown et al.
Robustness. A process may be regarded as robust if simi- (Geophysics, 1984). “Net pay estimation from seismic attributes”
lar results would be obtained independently by different, by Connolly (EAGE 2005 Extended Abstracts). “Statistical esti-
competent practitioners. Robust processes are usually char- mation of reservoir characterization uncertainty” by Connolly
acterized by having a fairly small number of parameter and Kemper (TLE, 2007). “Quantitative characterization of seis-
selections with the method of their optimization being well mic thin beds: A methodological contribution using conven-
defined. Most components of a conventional seismic data tional amplitude and seismic inversion” by Fervari and Luoni
processing sequence could probably be described as robust (First Break, 2006). “Fast-track ‘colored’ inversion” by Lancaster
(with the possible exception of depth migration). Robustness and Whitcombe (SEG 2000 Expanded Abstracts). “Pseudo-gamma
is a highly desirable quality but it’s probably fair to say that ray volume from extended elastic impedance for gas explo-
seismic reservoir characterization processes generally ration” by Neves et al. (TLE, 2004). “Extended elastic imped-
haven’t attained a similar level of robustness. ance for fluid and lithology prediction” by Whitcombe et al.
The net pay process outlined here separates the detun- (G EOPHYSICS , 2002). “Stabilizing the low frequencies” by
ing and calibration steps allowing the key parameters Whitcombe and Hodgson (TLE, 2007). TLE
(wavelet specification and calibration factor) to be deter-
mined sequentially and, as we’ve seen, an error in the first Acknowledgments: The author thanks BP for permission to publish this
can be partially compensated for by the second. This algo- article and special thanks to Steve Hill for his help.
rithm can be viewed as one step within a longer sequence
which includes extended elastic impedance analysis and Corresponding author: patrick.connolly@uk.bp.com
colored inversion. These other components also have a small

1282 THE LEADING EDGE OCTOBER 2007

You might also like