Professional Documents
Culture Documents
Multicomponent AVO
Linearized AVO Inversion of Multicomponent SeismicData
Exploration and exploitation of hydrocarbon re- proposed methods are put in a general theoretical
serves is becoming increasinglycomplex. To be suc- framework in the discussion of existing methods.
cessful, an increasing amount of information is re- Next, the general framework is extended to incorpo-
quired. In recent years much attention has been rate statistical noise and multi-component measure-
devoted to the information present in reflection ampli- ments. One advantageof the statistical approach is the
tude variation with offset (AVO) (Balogh et al., 1986; possibility to calculate accuracy measures for the
results, which is illustrated in the discussion of result
Gelfand et al., 1986; Spratt, 1987; Chiburis, 1987;Ball,
1987; L6rtzer and Berkhout, 1988, 1989; Haas and analysis. Subsequently, the problems associatedwith
Berkhout, 1988, 1989;Rijssen and Herman, 1988; SEG the practical implementation are addressed and the
workshop, 1988). In principle, the offset variations of whole approach is illustrated with an example using
the P-P reflection amplitudes contain information on simulated data. Finally, some remarks and a critical
all the elastic parameter contrasts (P-wave velocity, review of the results obtained are given in the conclu-
sions.
S-wave velocity, and density), while the zero-offset
reflection amplitude contains only information on the EXISTING METHODS
acoustic impedance.
All AVO inversion methods that derive estimates In all existing methods for linearized AVO inversion
for the elastic parameters of the subsurface, can be (Smith and Gidlow, 1986, 1987; Gelfand et al., 1986;
subdivided accordingto two criteria (Figure 1): Balogh et al., 1986; Spratt, 1987) the nonlinear Zoep-
pritz equation for the elastic angle dependent plane-
(1) The type of forward model used, and wave P-P reflection coefficient d(•) is linearized.
(2) The incorporation of wavelet interference. Following Aki and Richards (1980 p. 153) d(•) is
written as a weighted sum of the relative contrasts in
With the first criterion, the true, nonlinear Zoeppritz the elastic parameters P-wave velocity V, S-wave
equation can be used to relate the elastic properties velocity W, and density D
(contrasts) to the variations in the plane wave P-P
d(q>)= gl (q>)AV/V + g2(q>)AW/W + g3(q>)AD/D,
reflection amplitudewith offset (or angleof incidence).
On the other hand, a linear approximation of that (1)
equation may be used. The second criterion concerns
where • equals the angle of the incident P-wave, AV
the wavelet interference effects such as "tuning."
equals contrast in the P-wave velocity over an inter-
These effects can be directly incorporated in the
face, and V equals the average of the P-wave velocity
inversion, left in the results to be removed by subse-
over the interface. Similar definitions apply for AW,
quent processing, or neglected altogether (i.e., work-
W, AD, and D. In the original formulation of Aki and
ing with "picked" amplitudes).
In this study we restrict ourselves to methods that Richards(1980)the functionsg • (•), g 2(•), andg 3(•>)
read:
use a linear approximationof the Zoeppritz equation
as forward model for the P-P AVO, and leave the 1 1
g• (•) = -2 + -2 tan2 (•), (2)
*Shell International Petroleum, MAT. B. V., P.O. Box
162, The Hague, NL 2501 AN, Netherlands. W2
•Delft University, Physics Dept., P.O. Box 5046, Delft,
GA 2600, Netherlands. •72(•)
= -4 •-7sin2
(•),and (3)
317
318 Linearized AVO Inversion of Multicomponent SeismicData
1 W: which minimizes the squared data mismatch
•73
(•)=2 2•- sin2(q>). (4) F(•) - rain{[d- _Gx]T[d
- _Gx]}. (8)
x
• = (_G
TG_)-I_G
Td, (7) The relative (pseudo)Poissonratio contrast (A•/•)
is claimed to be a lithology indicator. The last, the fluid
factor (AF), is claimed to be a hydrocarbon indicator.
It shows deviations from an (empirical) relation be-
Type of forward model used
tween P-wave and S-wave velocities called the
mudrock-line, see Castagnaet al. (198• p. •72), valid
I for water bearing silici-clasticdeposits
Nonlinear
/
model Linearized
'Zoeppritz' 'Zoeppritz'
model V = Vo + kW. (11)
Deviations from this relation are associated with gas.
Type of wavelet interference handling Unfortunately,
thematrixõTG_in equation
(7)will
be close to singular, meaning that the calculation of its
inverse is very unstable and sensitive to noise or errors
Remains in in theelements
of _G
TG_.Thedegreeof singularity
or
Iin
inversion
/
Incorporated results
I Completely
neglected
conditioning
ofthematrix_GTG_canbeshown withthe
singularvalue decompositionof the forward matrix _G
Fig. 1. Subdivisioncriteria for AVO inversion methods. (e.g., Menke, 1984, p. 119). Combinations of parame-
L6rtzer and Berkhout 319
ters that correspond to small singular values are called (10a) is well resolved and can be estimated correctly,
weak, or ill-resolved, and cannot be estimated accu- correspondingwith our expectations.
rately (i.e., they are too sensitive to the noise on the The methods listed differ in the way they cope with
data). It appears that even for a considerable range of this instability problem. Gelfand et al. (1986) propose
angles of incidence there is at least one (very) weak to take W/V = 1/2 (i.e., a Poisson model), while Smith
parameter
combination
of theformx = (1, c•, - 1)r, and Gidlow (1987, p. 998) use an empirical relation
with c• being a model dependent constant. If we between the P-wave velocity and density, which reads
display this in 3-D parameter space, with the three for the relative contrasts
parameters set up in a right handed coordinate system
1
(Figure 2), the weak parameter combination yields a
AD/D =- AV/V. (12)
line L of almost equally possiblesolutionsin parame- 4
ter space. Models with parameter combinationsvary-
ing along this weak direction L (A V/V + •xAW/W - The so-called Gardner relation (Gardner et al., 1974)
/XD/D = const.) do not show significantdifferencesin in terms of the relative contrasts is substituted into the
the simulated seismicdata generatedfor those models. original equation (1). This substitutionhas an adequate
However, models with parameter combinations vary- stabilization effect following directly from Figure 2'
ing accordingto A V/V + AD/D = const., in the plane The parameter combinations that satisfy equation (12)
perpendicular to this weak direction do show signifi- lie in a plane P, which has a considerable angle with
cant differences in the simulated seismic data, i.e., the line L. This type of stabilization is to be preferred
they are well resolved. This implies that the relative over a purely mathematical one, like diagonal stabili-
acoustic impedance contrast as defined by equation zation, since it is geologically more meaningful.
STATISTICAL APPROACH
AW/W
In all existing methods for linearized AVO inversion
the noise on the data and the scatter around the
stabilizing empirical relations are neglected. Thus, the
first improvement to the existing methods is to incor-
porate this kind of uncertainty. We use the Bayesian
inversion approach (Jackson and Matsu'ura, 1985;
Tarantola, 1987; Duijndam, 1988 a, b), which can
elegantly handle deterministic forward models and
noise on the measurements (seismic information), to-
gether with a sensible incorporation of any available
additional (prior) information with a corresponding
' ' "' solutions: degree of confidence (scatter). The empirical relations
are considered as additional information. Both types
L••/r xtrue
+•(1,a,-1) of information (seismic and additional) are described
in terms of probability densityfunctions (pdf's), where
the statistical properties of the seismic observations
(i.e., the offset/angledependentreflection amplitudes)
and the scatter around the empirical relation (generally
Fig. 2. Graphic interpretation of the ill-resolved direc- from regressionanalysis of cross-plots)is modeled as
tion (line L) and the solutions satisfying the Gardner an additional "noise" term n, meaning that equation
relation (plane P). The solutions of the unstabilized (6) changes to
inversion problem will all lie on line L, which is
defined by the true model vector xtrae and the weak dp= G_px+np, (13)
direction
(1, a, - 1)r witha beinga model-dependent
constant. The position of the solution on L dependson where the subscript p refers to the primary P-P
the noise on the seismic data. The line L' is the
projection of L on the A V/V-AD/D plane, making a reflectivity information. Symmetrically with equation
45ø angle with the A V/V and the AD/D axes. The (13), we write for all empirical relations
parameter combinations satisfying the Gardner rela-
tion lie in the plane P. de = G_ex-I- ne, (14)
320 Linearized AVO Inversion of Multicomponent Seismic Data
where the subscript e refers to the empirical informa- F(•) - min[(d• - G•x)t (d• - G*
-P x)
tion. Note that the noise term ne describesthe scatter x
zeromeanandcovariance
matricesC_vand•Ce respec- contain strong enough information in favor of this
tively. Note that the elements on the main diagonal of deviation. This type of soft stabilization, where the
•Ce(thesocalledvariances)
referto theabsolute
noise answer is determined by the quality of the data (noise
content in the data. The noise is generally computed level), the theoretical information (forward model),
from an estimate of the signal-to-noise ratio and the and the additional information (empirical relations) is
total energy in the data. The off-diagonal elements very elegant and more realistic.
refer to the correlation between the different noise Note that in practical situations additional informa-
samples(correlated noise). Note also that •e follows tion is always available and should always be used,
directly from the aforementioned regression analysis even if only wide ranges can be specified. The inclu-
(scatter). Under the Gaussian assumption it can be sion assures the most accurate solution, i.e., the
shown that again a (statistically weighted) least- solution with the smallest standard deviation.
squares solution is obtained (Tarantola, 1987 p. 188; Later we will propose a measure to determine to
Duijndam, 1988a p. 883) what degree the solution depends on the additional
information. In those situations where this measure
•: -- (•G•TG•ß-1 .rd • ,
t) •G! (16) indicates a large dependence, the involved seismic
data has little to add to what we knew already. The use
where•G• = C•t-1/2•Gt
is the statistically
weighted of any alternative linear inversion method does not
forward
matrixandd• = C•t-1/2d
t is thestatistically change this fundamental fact.
weighted data vector. In equation (16) the subscript t
refers to the total amount of information and the
MULTICOMPONENT EXTENSION
forward matrix õt, the covariancematrix •Ct, and the
data vector dt are partitioned as follows: Until now only P-P reflection data were considered.
However, the statistically weighted stacking formal-
ism developed in the previous section allows the
Gt = , Ct = ,dt = , (17) incorporation of multicomponent reflection data very
- G_e - Q Ce de
easily. In principle the multicomponent data includes
where Q is a null matrix with the proper dimensions. not only S-S reflection data but also P-S and S-P
Using equation (17) in equation (16) yields converted reflection data. Even if the inversion is to be
applied to CMP gather data, the inclusion of converted
waves suffersfrom a severe reflection point smear, so
_ _eG•e) G_,e •e' (18) they are not considered in the remainder of this paper
and we will concentrate on the effect of additional S-S
which minimizes the squared total mismatch reflection data. Rijssen and Herman (1988) and Haas
Lbrtzer and Berkhout 321
. ß T
Table 1. Coefficientsin !inearized expressionsfor angle G_p d•
dependentP-P and S-S reflection amplitudes.
Rel.
•:: (•G•TG•
~ q-G*T ß + G_e
_s G_s ßTG,
_e) -1 S ß
W2 - 1/2+ 2 sin2q• reflection data using only the S-S reflection data. The
•D/D 1/2
- 2-• sin2q> resulting equations are identical to equations (17) and
(19) with the subscript p replaced by s.
RESULT ANALYSIS
G_p ~ O_ O_ dp C•
.. = (G*
~a TG•a
* q-G•e
* TG•e)
* -1 , (22)
_Gt= G•s, C•t = Cs O ,dt = ds. (20) where the subscript a denotes the seismic (AVO)
Ge O_ C e de information consistingof either P-P reflection data or
S-S reflection data alone or in combination. Expres-
The P-P reflection data, the S-S reflection data, and sion (22) which is independent of the actual seismic
the empirical information are assumedto be uncorre- measurements, depends only on their statistical prop-
lated. Of courseC_p,C_• and/or C_emay contain erties, and can therefore be determined prior to data
non-zero, off-diagonal terms, to represent also corre- acquisition to determine the optimum recording pa-
lated noise (multiples, converted waves, etc.) and rameters (spread length, group interval, etc.)
correlated scatter. The full expression for the solution For an objective resolution measure of the solution,
reads similar to equation (19) we propose to use
322 Linearized AVO Inversion of Multicomponent SeismicData
o'i + '•/(G_*
a TG_*
a -t-G_*
eTG_*e)ii
-1 uncertainties
(i.e., smallstandard
deviationcri
+) and
Ki = • = (•S) primarily determined by the seismic input (i.e., large
cri V'(G*a
_ ZG_•a)ii-1 resolution ratio Ki) to prevent potential bias from
incorrect empirical information.
being the ratio of the standard deviation with empirical
As said, even though the accuracy and resolution
information
•i + overthe standard
deviationwithout
the empirical information •i-. K i is bounded to the measuresdepend on _Ga, C_a, _Ge,and C_ethey are
interval [0, 1]. The two extremes co•espond with independent of the actual seismic input (seismic and
"totally determined by the empirical input" and "to- empirical) and can be calculated before the actual
tally determined by the seismic input" respectively, acquisition. To determine the required minimum angle
with the turning point at K i = 1/2. Note that increas- range, Figure 3 is shown as an example for P-wave
ingbothq• andq• witha factorb2 doesnotalterthe velocity input. Based on a maximum requesteduncer-
value of Ki, but increases the standard deviation of tainty and a minimum requested resolution ratio, the
the parameter estimates with a factor b. In principle minimum angle range can be determined. A different
we would like an inversion solution with (very) low approach is shown in Figure 4, where uncertainty and
Standard
deviation Resolution
ratio
0.3-- 1.0
ss 0.8
ss
0.2-- lllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll/llll•lllllllllllllllllllllllll
0.6
+D
0.4
0.1 --
0.2
0.0'- 0.0
I I I I I I
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Angle range (degrees) Angle range (degrees)
Fig. 3. Standard deviation and resolution ratio as a function of the range of angle of incidence from 0ø
to maximumanglein 1øincrements
(C_p= 0.01I_p,C_s= 0.02_Isand_C
e = 0.20_Ie,whereI is theunity
matrix of the correct dimension). -
Standard
deviation Resolution
ratio
0.3-- 1.0
0.2"'
'-,\ ss
0.8-
0.6- .....
'"'"'",,
pP•PP.
+SS
0.4-
0.1 --
PP ',•',•,.
PP+ SS '",".•, 0.2- $$ ",..
0.0--
........ I .... '"'1 ' ' ''""1 ' ' ''""
0.0- ............................................
' ' '''"'l ' ' '''"'!
d..................
' ' ''""l
:.'.
.'.
.'.
.'.
.-.
.,.-.
.-.
' ' '''"'
.-.
-
0.01 o.1 1 lO lOO 0.01 o.1 1 lO 1•
Stabilization rate Stabilization rate
Fig. 4. Standard deviation and resolution ratio as a function of the stabilization rate. Same covariance
matrices as in Figure 3 and angles of incidence range from 0øto 25øwith 1øincrement. The stabilization
rate creates an additional stabilization: A rate of (x results in a decrease of the standard deviations with
a factor l/ix.
Lbrtzer and Berkhout 323
resolution ratio are displayed as a function of the example, the estimated A W/W contrast will be a band
stabilization rate, which acts as an additional scaling limited version of the actual (spiky) A W/W contrast.
factor on C•e (the covariancematrix of the empirical This is important when interpreting results from thin
information): At a stabilization rate of 1, the original layers with interfering reflections. The spiky relative
C•e is used, while for a stabilization rate of 100, contrasts can be retrieved by applying a "poststack"
1/100C_eis used. Thus a stabilization rate of 100 stratigraphic inversion procedure (Duijndam, 1987,
decreases the standard deviation (assumed scatter) by 1988a, b; Gelfand and Larner, 1984; Brac et al., 1988)
a factor of 100 and increases the importance (the to the weighted stacking results. Or in other words, a
weights in equation 18) of the empirical information. weighted stack result is the preferred input to post-
Figure 4 can be used to determine the maximum stack stratigraphic inversion.
achievable resolution K i for a certain maximum al- For sake of completeness, the following list gives
lowed uncertainty level •. With respect to a certain some of the most important factors affecting the inver-
range of angles of incidence, number of traces, noise sion results, some of which may have a more signifi-
level on the seismic data, and scatter around the cant effect than others:
empirical relation, the maximum allowed uncertainty
level correspondsin Figure 4a with an additional stabi- Errors in the offset to angle conversion (accura-
lization rate for the empirical information. For this cy of the macro model);
stabilization rate, Figure 4b gives the corresponding (2) Unwanted amplitude effects (propagation ef-
resolution. If the resolutionis not satisfactory,it may be fects, source/receiver directivity);
improved by (in decreasingorder of importance): (3) Approximation of the Zoeppritz model (linearity
assumption);
(1) Increasing the offset (range); this may result in a (4) NMO correction errors (stretch, erroneous ve-
conflict with the linearity assumption in the locities);
forward model; (5) Random noise, statics; and
(2) Decreasing the noise on the seismic input (by (6) Correlated noise (multiple reflections, converted
better processing); or waves etc.).
(3) Increasing the subsurface coverage within the
same offset/angle range. Obviously, the effects of these factors (especially 1,
2, 4, 5, and 6) should be minimized as much as possible
PRACTICAL ASPECTS
prior to the inversion by appropriate preprocessing
The weighted stacking for the relative contrasts in (see, Berkhout and Wapenaar, 1990).
the elastic parameters is not to be done on the AVO of Table 2 lists the required additional processing to
a singleinterface but on a whole seismicdata set, e.g., arrive at aligned, angle-dependent, reflection ampli-
in approximately 1-D media a CMP-gather and in tudes for a CMP or CDP-gather.
stmcturally complex media a CDP-gather (Berkhout
and Wapenaar, 1990). If the data in the CMP-gather EXAMPLE
have been NMO corrected, this implies we have for We now show the statistically weighted stacking
every zero offset time a reflectivity vector as defined results in an experiment on simulated multicomponent
by equation (5). Inversion, i.e., applying the weighted seismic data, allowing us to compare the inversion
stacking procedure, yields estimates of the relative results with the true model. Figure 5 shows schemat-
parameter contrasts for every zero offset time. ically this true model, which consist of a sandstone
If we assume that
layer encased in identical shales. The fluid-fill in the
sandstonechangesfrom brine to gas going from left to
(1) the seismic wavelet does not vary with offset,
right and is simulated by a linearly decreasingP-wave
(2) the NMO corrected CMP-gather can be modeled
velocity V while the S-wave velocity W and density D
by convolution of the wavelet and aligned spiky
remain almost the same (Figure 6). The shooting
reflectivity traces, and
geometry is shown in Figure 7, with offsets ranging
(3) the weights for the weighted stacking vary little
from 0 to 2400 m, resulting in a maximum angle of
over the effective length of the wavelet,
incidence of approximately 35ø. At each of the 6
then it can be shown that the linearly estimated indicated positions (Figure 5) a simulated NMO cor-
relative parameter contrasts will be a convolution of rected CMP has been generated, using the following
the estimated spiky contrasts and the wavelet. So, for scheme, where a perfect S-wave two-way traveltime
324 Linearized AVO Inversion of Multicomponent Seismic Data
(TWT) conversion to P-wave TWT is assumed, using Table 2. Additional processingfor AVO recovery.
the following forward scheme.
CMP oriented approach
ß Filtering (multiple attenuation and ground roll removal)
(1) Offset to angle conversion using raytracing. ß Correction for source-receiver directivity
(2) Angle dependent reflectivity calculation using a ß Spherical divergence correction
ß NMO correction
nonlinear forward model (i.e., the Zoeppritz
ß Offset to angle conversion (based on macro model).
equations for incident P-waves). Note that no
NMO stretch is applied. CDP oriented approach
ß Decomposition into P- and S-waves (for multicomponent
(3) Band limitation by convolution with a 20Hz zero data),
phase Ricker wavelet. ß Surface related multiple elimination,
(4) Addition of Gaussian distributed uncorrelated ß Prestack migration (based on macro model).
random noise in the seismic band (5-100 Hz).
The noise samples for the P-P reflectivity sec-
tion have a standard deviation [•r] of 0.01 (S/N
ratio = 10dB) and those for the S-S reflectivity
section have a •r of 0.02 (S/N ratio = 12dB). Shale
In the statistically (soft) stabilized inversion both the
Gardner relation equation (12) and the mudrock rela-
tion equation (15) will be used as empirical informa-
tion. In case of the deterministically (hard) stabilized
inversion only the Gardner relation will be used, as ........................................................
Gas Sand
hard substitution of equation (15) will yield a fluid :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
5: ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
1000 - - 1000
, • • , - 1000 1000- • • • •
2 3 4 5 6 I 2 3 4 5 6
Surface location Surface location
h h
1800 1800
1900 1900
20O0 2OOO
21O0•, 2100
•,
t t
h h
1800 1800
i ,
1900 1900
2000 i rI 1 2000
i
I• • (
21 O0
2100
•,
t
Fig. 8. Noisy sectionsof P-P reflectivity at locations 2 thru 5. The offset h increases from left to right.
The section at location 1 is similar to that of location 2, they only differ in noise realization. A similar
statement can be made for the sectionsat locations 5 and 6. Note the dramatic increase of the reflectivity
with offset in the gas-bearingpart, which is absent in the brine-filled part.
326 Linearized AVO Inversion of Multicomponent Seismic Data
tions 2-5. The data at location 1 only differ in noise coefficients in Table 1, the zero-offset reflectivity is a
realization with those at location 2, and similarly for scaled version of the relative impedance contrasts.
locations 5 and 6. Figure 9 shows the S-S sections at The zero-offset P reflectivity is a scaled version of the
the correspondingfour locations. In the inversion all relative acoustic impedance contrast AZ/Z and, like-
31 channels of each P-P section and/or S-S section wise, the zero-offset S reflectivity is a scaled version
will be used. Close examination of the NMO-corrected of the relative shear impedance contrast A Y/Y. Fig-
CMP gathers shows that in the gas part the P-P ures 12 and 13 show the estimated P and S reflectivi-
section exhibits a significantly increasing AVO, and ties as obtained from the P-P and S-S sections,
the S-S sections are approximately the same, but respectively. Comparison with the results of Figures
contain a polarity flip at approximatelyhalf the maxi- 10b and l lb, shows dramatic improvement in the
mum offset. Therefore, significanterrors will be intro- amplitudes of the zero-offset reflectivity estimates
duced in the interpretation of conventionally stacked obtained with statistically weighted stacking over
data, as is shown in Figures 10 and 11. those obtained with conventional stacking in the case
For the acquisition geometry shown in Figure 7 and of significant AVO variations.
the statisticalparameters listed in Table 4, the follow- For the fluid factor contrast (as a gas indicator) we
ing results are discussed: will first show in Figure 14 the results obtained with
deterministically weighted stacking of P-P data only.
Zero-offset reflectivity: Figures 12 and 13 The effect of a highly ill-resolved relative density
Fluid factor contrast: Figures 14 to 17. contrast is removed by hard substitution of the Gard-
ner relation. The relative S-wave contrast is not sta-
All figures are plotted with the same scalingfactor bilized at all. Hence, its sensitivity with respect to the
for all output traces (no AGC). As follows from the noise in the data is not taken into account. In this
1800 18oo
1900 1900
2000 2000
2100
•, 2100
,[,
t t
h h
1800 1800
1900 1900
2000 2000
21 O0
2100
•,
t
Fig. 9. Noisy sectionsof S-S reflectivity at locations2 thin 5. As in Figure 8, the sectionsat locations
1 and 2 differ only in the noiserealizationand similarlyfor locations5 and 6. Note that the sectionsshow
very similar behavior with offset for all locations. All sectionsshow a polarity flip at about half offset
which will have a devastating effect on the conventionally stacked results.
L0rtzer and Berkhout 327
2oo0 2000
22oo 2200
2ooo
22oo
2200
2000
2200
Estimated
contrast: 2000
Fluid factor
2200
1800
(b)
Error trace 2000
2200
Fig. 14. Fluid factor resultsobtained with deterministicallyweighted stackingof P-P data only. (a) The
estimates do show a significantnegative over positive contrast in the gas-bearingpart of the section.
However, in deterministically weighted stacking the relative S-wave velocity contrast is not stabilized
at all and its sensitivity with respect to the noise in the seismic input is not taken into account. (b) The
actual errors in the estimated contrasts obtained by subtraction of the true bandlimited contrasts from
the estimates.
Estimated
contrast: 2OOO
Fluid factor
22OO
1 2 3 4 5 6
1800
Stand. Dev.
+ 2OOO
Error trace
_ 1 2 3 4 5 6
22OO
1800
Conditioning 2000
22OO
Fig. 15. Fluid factor results obtained with statistically weighted stackingof P-P data only. (a) The
estimates are much better than the estimates of Figure 14a. (b) The actual error in the estimated
contrasts,togetherwith the standarddeviation. (c) The conditioningof the estimatedcontrasts,which
are approximately1/2. So the resultsare determinedby the seismicinput for 50 percent. (The displayed
range between two traces runs from 0 to 1, going from left to right).
L6rtzer and Berkhout 329
experiment a high value of k is used in equation (9d), deviations of the estimates are available as an accu-
resulting in a large contribution of the (relatively) racy measure. Due to a slowly decreasingangle range
noisy S-wave contrasts to the fluid factor, which with depth, the standard deviation slowly increases
explains the noisy fluid factor results. Note that hard with increasing depth. The error in the estimates is on
stabilization of the relative S-wave velocity contrasts the average well below standard deviation. So in
using the mudrock relation, would yield a fluid factor practice, where we can not determine the actual errors
result equal to zero (see also Figure 17). Because we in the estimates, the one standard deviation limit may
have the true model available, we can calculate the serve as an estimator for the errors one can expect
true contrasts and the corresponding errors in the with respect to the anticipated noise levels on the
estimate, but in reality this, of course, is not possible. seismic data and 'the scatter around the empirical
If we look at the results in Figure 15, obtained with relations. This is the first major advantage of the
statistically weighted stacking of P-P data only, we statistical approach over the deterministic one. The
see that the actual estimates are less noisy, than those secondadvantageis the resolution ratio, or condition-
of the previous figure (obtained with the deterministi- ing, curves as shown in Figure 15. These curves
cally weighted stacking). In the statistically weighted indicate whether the estimates are primarily deter-
stacking, the sensitivity of the relative S-wave con- mined by the empirical input or by the seismic input.
trast with respect to noise in the seismic input is taken Since in this case the conditioning is about one-half,
into account and a stabilized S-wave contrast estimate the fluid factor contrasts are still determined by the
is used in the calculation of the fluid factor. seismic input for approximately 50 percent. The de-
Moreover, much more information on the estimates creasing angle range causes the conditioning to be-
is available. First, as shown in Figure 15, the standard come less for deeper parts of the target.
Fluid factor
2200
1 2 3 4 5 6
(b) 1800
Stand. Dev.
+ 2000
Error trace
2200
1 2 3 4 5 6
1800
(c)
Conditioning
2200
Fig. 16. Fluid factor results obtained with statistically weighted stacking of P-P and S-S data
combined. (a) The estimateslook significantlyless noisy than the results obtained with P-P data only
(b) The actual error in the estimated contrasts, together with the standard deviation much smaller than
those for P-P data only. (c) The conditioningof the estimatedcontrastsis much better (K i • 1) than
obtained with P-P data alone.
330 Linearized AVO Inversion of Multicomponent SeismicData
Figure 16 shows the fluid factor contrast results relation. This deviation is exactly, what we observe in
obtained by statistically weighted stacking of both P-P Figure 18 in the gas filled part of the section: negligible
and S-S data. Comparison with Figure 15 shows that contrasts where there should have been significant
the incorporation of S-S data resulted in more accu- contrasts. The importance of correct prior (empirical)
rate estimates for/•F, due to a more accurate relative information is shown for those cases where it is
S-wave velocity contrast estimate. We also see that needed. In the case of P-P data alone or P-P and S-S
the standard deviation of the estimates has decreased, data, incorporation of the erroneous empirical infor-
to indicate the higher accuracy. On the average the mation caused no harm, since the answer was deter-
actual error is well below this one-standard deviation mined (mainly) by the seismic input. However, in the
limit. Also the conditioning has improved. case of S-S data alone, it becomes of key importance.
For the sake of completeness, Figure 17 shows the Note that the necessity of caution is clearly indicated
fluid factor contrast results obtained from inversion of by the conditioning curves (Figure 17). However, even
S-S data alone. Since, the S-S data contain absolutely in this case we see that the (very much larger) standard
no information on the P-wave velocity, the fluid factor deviations in Figure 17 are still indicative of the errors
contrast result will rely completely on the empirical one might expect in practice, where the true contrasts
input as indicated by the conditioning curve with zero are not available.
amplitude. The empirical input consists,in part, of the
mudrock line relation between /•V/V and /•W/W for
CONCLUSIONS
water filled sandstones. Results satisfying the
mudrock relation will show a negligible fluid factor In conclusion we can say that (deterministically or
contrast since the factor indicates deviations from this statistically)weighted stacking shows an improvement
Estimated
contrast ß 200O
Fluid factor
220O
1 2 3 4 5 6
(b) 1800
Stand. Dev.
+ 200O
Error trace
220O
1 2 3 4 5 6
1800
(c)
200O
Conditioning
220O
Fig. 17. Fluid factor results obtained with statistically weighted stacking of S-S data alone. (a) The
estimates show almost no contrasts. (b) The actual error in the estimated contrasts together with the
standard deviation; even for these large errors they are (on the average) well below the one-standard
deviation limit. (c) The conditioning of the estimated contrasts indicates that the result is completely
determined by the empirical information.
L6rtzer and Berkhout 331
Conventional
Structure
stack
Nonlinear Quantitative
prestack spiky
elastic amplitude
inversion information
over normal stacking for a qualitative analysis of ment of correspondingreflectivity information on P-P
multioffset amplitudesin the case of significantAVO and S-S sections.
(e.g., strong increasing AVO or polarity flip). The
statistically Bayesian based extension allows a more ACKNOWLEDGMENT
sensible stabilization than plain hard substitution,
since the solution is allowed to deviate from the This research was carried out within the DELPHI-
data, and amplitude variation-with-offset(AVO) mod- cemented sands from porous, gas-bearingreservoirs,
eling strategies to differentiate gas reservoir seismic each of which produces high-amplitude reflections on
anomalies from those formed by cement-induced im- conventional P-wave seismic data. An experimental,
pedance contrasts. three-componentline was shot with P vibrators from
The Zamora gas field is located approximately 60 north to south across several bright spot anomalies
miles northeast of the San Francisco Bay area, in Yolo that have been tested by wells along the line of section
County, California (Figure 1). Gas production in the (Figure 2). The availability of good quality geologic
Zamora area is from late Cretaceous and Tertiary age control made this an ideal area to test the potential of
deltaic sandstones that occur at depths of approxi- using P-SV-wave data as a prospecting tool.
mately 3500 to 4000 fi (Figure 2). Structure maps and The primary purpose of the field trial-was to process
cross-•ectionsindicate gas accumulationis controlled andjointly interpret the P-P section (using the vertical
in large part by structural trapping mechanismswhere componentdata) and the P-S V section(from the inline
closure is provided by subtle, low-relief, anticlinal horizontal componentonly). The motivation for this is
structures. Well site selection and risk evaluation in that the P-wave reflectivity is sensitive to changes in
this area can be complicatedby variable stratigraphy, P-wave impedance, whereas the reflectivity of P-SV-
cementation patterns, and fluid saturationwhich cause wave reflection data is sensitive to changes in S-wave
reservoir quality to vary significantly both laterally impedance (Figure 3). Although not obvious convert-
and vertically in the section. ed-wave reflection coefficients at non-zero offset scale
One difficult problem faced by interpreters in the approximately like S-wave reflection coefficients at
Zamora area is differentiating low porosity, carbonate- normal incidence. This is shown in the Appendix using
Sacramento
Valley, ..---•3-C Line-
:
California
.Sacrament"o.
o 12 Mile •:•'Stockton
o 20 Km
N
S-wave reflectivity is insensitive to variations in the
Field Trial Line gas/fluid mixture saturating a potential reservoir. The
30
/T.p '....
Tp,..,V
TS effect of that processing is to eliminate the severe
depth point smearing caused by gathering the asym-
metric rays by common midpoint (CMP).
ß P-P Reflection Coefficients • SH-SH Reflection Coefficients
e 2Tp Interval Times • Tp + Ts Interval Times
WAVELET EXTRACTION AND SYNTHETIC
SEISMOGRAM MODELS
Reflection Anomaly P-Sv Reflection Anomaly
Present Absent
The processedP-wave seismic data are very high
quality and display two pronounced bright-spots, one
Fig. 3. Schematic diagram illustrating P-P and P-SV
anomalies from a gas-sand. See the Appendix for the at approximately 0.6 s and another at approximately
relation between P-SV and SH-SH reflection 1.1 s, both of which have been tested by wells that
strength. have good quality sonic and density logs (Figure 4).
336 Correlation and Interpretation of P-P and P-SV Data Zamora Gas Field
Wavelet extraction and zero offset models were pre- In contrast, the deep bright spot anomaly has rela-
pared to test the reliability of well log ties with the field tively slow sonictraveltimesof approximately160ms/ft.
trial data. P-wave velocities for this interval are on the order of
Wavelets were calculated from the P-P section 5500-6000 ft/s. Well log analysis and formation tests
using a sparse reflectivity statistical model without indicate that this interval is a porous, gas-producing
sonic log control. These wavelets are zero-phase sand with saturationsof greater than 80 percent.
wavelets with little side lobe energy (Figure 5). Other significant reflections to note on the P-wave
The extracted wavelet was convolved with the re- section (Figure 6) are those from the top of the Capay
flectivity series calculated from the Knaggs-Wallace and Sacramento shales. The Capay shale displays a
# 1 sonic/density logs and other wells along the line of relatively continuous, low-amplitude reflection at ap-
sectionto produce a syntheticseismictrace. An excep- proximately 0.920 s. The Sacramento shale consistsof
tionally good match between the filtered synthetic and a series of high continuity and amplitude reflections at
the field trial P-P data (Figure 5) was obtained. The 1.480 s to approximately 1.600 s. The reflections
results of this modeling provided confidence generated from each of these formations will be key to
that reliable correlations could be made between avail- correlating and interpreting the P-P and P-SV data.
able lithologic control and at least the P-P seismic
data.
CONVERTED-WAVE EVENT CORRELATIONS
Posted along with the synthetic trace in Figure 5 is
the sonic log from the Knaggs-Wallace No. 1 Well. Perhaps the most important part of the interpreta-
This data indicates that the bright-spot shallow in the tion procedure involved the correlation of P-wave
section has very fast sonic traveltimes on the order of events with their equivalent reflections on the convert-
105 ms/ft which translates to P-wave velocities in ed-wave data (Figure 7). It seemed apparent early on
excess of 9500 ft/s. Log analysis and well cuttings that the high reflectivity converted (P-SV) events at
indicate that this interval representsa tight, carbonate- 2.800 s tied with Sacramento shale reflections on the
cemented sand bed with porosities on the order of P-wave data (Figures 6 and 7). As a first approxima-
10 percent. tion, the high reflectivity P-SV events at 1.900 s
...........
........ ...............................................
:-•...•• '• .•.•'"""--•'
'"••'•i•?.i•:
'"" •" .•'-•:•i;:i•i'""
"•'""•':•:'
..'"•••••':-..-•::-.•:•"•
••' '- '"""•""•
....
- .........::•:..
............... ....•...•:.•
.............
•.•:2•:.-z
.........
.:•....•
........................
....•..
...............
•.. •.• -•':7'
'"•"•'• .......
•:.•. •....:.--•::•,
•
• '"':""'-):'7"...;"2•."•'."2
• .•• :;•..:.:.•-•--•
'•:-• ....
•' '-'•'•-'"'-•-"
"'••'"•• •
....
•1;:;:•,,;:25,.•
:--•--------••:.:•:.+':....
-:.•-•":'
...... •'":":":'•::
--"'"?:'"- :::;:%:;•:,--
'-L::-.•:.. '-:•':::-:•:-;;:::&:.>-.r::•'::.•"...,.
.-•.-....•-•*:-'-'-•.::-
........
•'.•:,-,'::*:*•-.'-:
..........................
'-*:."
:.':'•;-,,:-
- ""]'.:•----•...-'.-•.2a•x.x";
:7•:;•"•q':•"""':':'"•"•'..•::½'/'•}?"
................
. .....
•:..'•-' :-::.':•i
.i.;i;5;;•.:-.:
......
-:•..--:?:-'::s'?
....•%•-•*-'x...•.•
.....
::::::::::::::::::::::::::::::::::
..•••••;:::•...:._.&,:,:
• ...•.•.-•..•.•••.....• :.•.j:4:•;.•..S.•:.•:•.
::...':;"-'
...•;-.:
..............
...•....._.......••.•
...... -••••
•'i'i••X•:
...... 3::.*":•-•'"•"-•
.............
:-:•;:.•,•:-.- .... ..............
:..:---•?
:•:•':•;;?•'•':'x:::•,•.½;$•::...:5
..... '.:..
.........";..½-•,-•i:•-:*'"-'"'""
'•-' ' '•'•'•--•.:.':'.-::.
.........'.'
.•,.--•....:::.-•.<*-.--•,•
',..
ß-•,-•-•.•:-*-;'•*-•,*'•
......•.....
-.....
...........
::::::::,•,
::• -
.................
•:':•;z•::•;.•4.•,•-•i..-':
-.½-:..
--•-•-:'"•.•.•:•>½,,:
............
.-•::•:-.-•:•.--•
::.•..-•..•-,•:
4•:s. ;,;-•.•:•.-:•.....,•:-•....:•:.:•:.•-....:•:• -•-:•-
-.-•..
'•'""..:.:.....•x
-• -'- ..--..•-::.---•-•::
::,:**:.:.,.-:,•.,•,•,..:•.;;•;;::::•:,:.:.:,.::::::•:....•
-.-..-.-'-".",. *::'.
?•,•,:...........
,:*-•-,*::•--.::'
ß....... ...... ß
...:...:;;..;.'.'.'....
,...
'"'"'
.iP-Wave Seismic
.............
-:-.......:;
.........
:,•,•
...............
•"•.-.•*•H';7:'•5•;.:•:;:.•:::;;::,;;;:;;;:"•:;
.......;'
".--':'.
' ",_J2'Z:'";:L-.:a_..4"*•;•
-& ..Z' Anomalies
'•Z:.......:•2•4;'•z-•',•:•.
;..:;:
:, :;
.......
...... ..,...+
....................... .,.•..,
.......... •..,...•."•:,•,.:,::.....:..:.•,:..•z•:.
•,::**?:• •:•:, ,:::•,...::.•,:•-.•.
;¾•. ----.....:
... _-•---•--••••••;•::.;• ..:•:•5•:.::;•,:::•?_.;.,:.•
---------•,•..--•••
'*••'"'"'
............. :•::'-•.. '..-:::'"'
'"".
' '"•:':?.;:5:•:?:.3'.,•'.'5'"
":'
•.•'"
';.**•'-'"";'•.'.•-,::•:•;'
":'"•-•;;•;;•
••....•••••.••;;,:...::::::::•::
............ ::::::::::::::::::::::::
"• ;::*•-•-•.............
';'•::':'•
.... '..... .:**..
..... ' ,.•----•;;-•
*;•;•Z;•
..............
'"'""•
..... •"• "'"'"'•"'•'•;:t¾z;;
•,:.--•'
appeared to tie with Starkey gas sand P-P anomalies. the Sacramento shale occur at roughly the same
However, when P-P and P-SV traveltimes were used P-wave times (1.5 s). Confident that this correlation
to estimateVp/Vs ratios,the valuesobtainedwere was a realistic one, it became clearer how other
less than 2.0, which is extremely low for sand/shale reflections correlated from section to section.
lithologies at this depth in the Sacramento Valley. The low-porosity, cemented sandstone shallow in
In order to correlate the events between the two the section (approximately 0.6 s) displays high reflec-
sections a "low-frequency compression" of P-SV tivity events on both the P-P and P-SV sections. The
time to P-P time was made using the ARCO mudrock presence of well developed events on both the P-P
line. The logic for using this strategy is that the ARCO and P-SV sections is interpreted to be due to "hard
mudrock line generally provides a good approximation streaks"thatresultin a largeincrease
in bothVp and
of Vp/Vs ratios,especiallyin brine-saturated,
rela- Vs velocity.
tively unconsolidated sand and shale sequenceslike The top of the Capay shale has a much sharper
those in the Sacramento Valley. In addition, rock P-SV than P-P reflection due to a much larger de-
physicsmeasurements
of Vp and Vs for sandsand creasein V• thanVp velocity.Thisis oneof thefew
shales collected at Chevron display a very good cor- cases where we've seen event reflectivities and conti-
respondencewith this empirical function (Figure 8). nuity much stronger on the P-SV data than on the
P-wave data.
COMPARISON OF P-P AND "COMPRESSED"
The Starkey producing sand is represented by a
P-SV SECTIONS
strongP-P anomaly and a barely visible P-SV seismic
When the P-wave section is compared with the reflection. The absence of a bright-spot on the P-SV
compressed P-SV section (Figures 9 and 10), it is section indicates that the S impedances are nearly
evident that deep reflections associatedwith the top of balanced across the interface at the top of the sand.
Zero Phase
Wavelet
15-50 Hz
Fig. 5. (a) Zero-phasewavelet extractedfrom the Zamora field trial data by a statisticalmethodwithout
well control. (b) Synthetic seismogramand sonic log trace from the Knaggs-WallaceNo. well.
338 Correlation and Interpretation of P-P and P-SV Data Zamora Gas Field
........
-.•,....•.•.•=•:•;...
.•..:.•:•:r•...:•:-. '........... • ::::•;..•=•;:•
• • •,•.,•;-• •:-•-•=,._•.:
•c-•..•.•.•,.•::•.• Ce-ment'..ed Sand
-•. :.-•.. • .....................
":' ' "' 'F::::.:::;.'.:.:-•i+•11'11(.
": %1":%•
"'1:•::
;•....•: :::::::::::.:
•:- ':•':
....... • .• ' :•:-
:•::'.;..•;•-:
......... ":":":
..............................
'..'.•'•.?
.•'.,•::.:•:...
':;%•--
....... ,;.7,
:&......].]•;
?'•;:•;;
'"•••-'•••:•
......
'•
"••••:'•"•'•'•;•:•....:'"'::•:•'•;'•;•:•;•.i•2:.•;] •".;::.%?•'
'• "•'*"c:;::--,47•:?•:•
••- '•'
•-•'•'•!-'-•'•::
'• ..:.3.-.:•
.......
:•,;'.::--•:.,,-:::,::::.-•
........
.-'•:':r"%:42-•-:-•::•
.......... •. "-'• .'.'L
•':•'::'. •':•...•
..•,:•.:,..•..•--,
-•:-•- '"•.•:?;-"
-•-•-',:-•.•;•:.,.;;'.'..•i•:.:•
'.::.½'Z•:-•-..
:'•:•':'"'•;:*-:-,•,,•
:..•',.•i•
..•4; :.::'•.•::::"•
•.-_:•:•;
:•;'•M•.., .•:.:•;•,::..:m_•:::::•%:•:;:*:-::•F•':2'•....j•::•i:g-•.?::;;%:•-•
:•.-
•,:..•.••.:*:;•.:(:..•.:•...•.2t?•E•;i.:•::.•::•.•...•.;:.•..•.•.:...`•.:::...2•:;..•.
....•
....
4'.•:i•':'&.•":'"
'"•"•"•'*:.:'::':.'7%
' .;½•
.. .-.;•....'•,'•::-:•::•*•'•
•'.•"• :• ::"'-:•:•:-•'•;-./"-•'::•'•;.::2.
•:.&•
•:{•*•*""..•:,•,
'•;•;•:::•'•;'f::•
'•:*'•'---•:•:-:•:
,•;•:•;f':A.•¾:•'"::•
•:• "-;'
........
?•.7•.
?• .'.i:%,'•
'•'•::•":":•
-.•
'•:•,i ';:"'•:...-.....•.•Ca
;.-::•:• .payS::hate
1' -:-4z::•.,,...:•.•-i•:;-•,...•---'
-•*;•.•:•,•---
•-•• -•--•;•;;:•----
----••••••,iC•i:•-•-.•_•.•••::.•.,.•--•-•j•.-.•i:::Z..:•::;-:
- '.•:•.-.•....
:•'•
........
.,..,..•.
........................
-.....................
•:•?,
...................
.. --;;;.-
::...-•::--
•::•:•:•....`.....•:*•`•..•*.•*??•::•:::•p•:;•.•:•..•z•`.`.:.•:•;:.•:;•;;..•;•.•
:-•::-•
%.•'.." --•:&:•;;;;;;::.;:-....:;
............................
i::.
•:-•:• ..•..•,•••.;...:•::•.•..!..:•.•.•
-•---•--
. .::•.-•.............• .............
-:.:...•.•.:•:,•:
........ .............
•::.•:,-,.;•i..
:%.•. ...•,.,.,...•e..•.•:.•
...... . ....... ::--....:.•iii:-•--•ii---•;:.•
...-: .......... •.. .. =======================
............ ...::: :-Starkey
GasS"'.'":and
.........................................................
. .................
:.:•:.• ........ :,:.i:i::
....... .. ;
:.
.................
;•;-.•:::•;•:::•-Z-:;
::•?½.:
•,
.....;:.Z?•
:.•:.:•,•.:,•:.:•.::z%,•.,.•
::::.•..:.• :.-c•:.•
::,•:•½•!•::;:•!•!•½::.::::;.,•.:•....:•:.j:.:;:;.•....•:.-•:;•)•:.•:.::•:....-•
............ ::,•.,:.:.
•;;•
................
...,..•:•;...:•,.,,:::...•:=
.:.:-;•½:..-..•
....
.............
•..:.:•;;•..:::.
;..:•-•..•½•.,::•::::,:•::;.•,•:•::.•::..
......
,•..................................
•::•.
.....................
. ......
...............
' •<-----•.-'•:;•--z:Z•;•.•
...... ---...:-
' ..............................
..................
•*':•-
........ '........
' .:,•'.....
';•...'.'.•)'i.
...... '•"•'
ß
........
*":':'"':•:
:'•:"•'•:'•:.
:
......
•';:'::•;:";;;;•::•"'":'-"•'
•*::*::.ii::*.•
.....
........... -::.;•,::•i•;;'i•E%..
• •/•a•:•.:•.::;•;•.•::*ii-•'
•.........
•"•'
:::'?.':'•::•'".%•L.
":•"•::•:•:•':•'""•
.....•:..•,.,.• '""•*•:: :.•'•7•.:.:.•.:.
..........
'"'"'• ..........................................
:' L.' •-•z•.-•!•:-•:•:.:.;•..•:•:.....•.: Sacra•nto Shale
:.;•i;•:•.•
.......
....._•::1%:•:..•,-•::...z.•x..,,:•.•;•k•,.•,•,
......... •...•...•..•..<::•:•........•.•-,.
....... ... .; .............. •.....:.:...:•::::•;..::...
................
::•:.....•.iz-.
.........•..._....•.•.••...•.•••.r..••:••:..•
••:•••':• ß .....
•-.•:• .......
•: -'"'"'•
•-"
....
:•...•:':• i.'•]•'.::':..; ..... :•:•:
:'?i:!'!:!i:;!!:.!:.•:E:;;"r
•':' '•::-•,i;•;::;• •'•':;'c11F%;;;,;,:•;•
:;•::.:•:...•:•:•:::::g:'"•.
'•.•:i.i. ][5
'.•:E;;.;i;;.
".•
...................
:::.:-:-----'•-'-';;':.'::t:?-:':-:.-':;;L
;•;'
'"..
'".:;'"';.•:":-
'-:-::*•g•:-:;:::•i•:•:•:•'•;::•-:-?'•""::;i-.::?'
•:!:.•::•:½:•½•::'
•:•'
'%."'L
.....
":.L
.....:::......-;::;
•:" ':"::'•-•." ;;;.Z'5".[.;•..•.
........ •.•;ii!
:•;•?:.:.;::;'c;':•
..;:.;..½•::•:...
•./•::
•:•
..... 7;:: .:.:.:..:..;i,-.•,:
.•:.•:•
.•?..%:•..
..........
:'::.:•..•.;•:•.....
.... :-•:.:.:•..•..,:.::•::•::.•;•½:&;j:.,•:•,,:•:•:i•:•
.......
.•..•...•.•............................
•;::• . .•...
'-••:.........::.•:...:::.:...:..;
:•: •.. ..............
::.:
...........................
.....•..:....•
...................
:•-:,•;•:,•
....................
•;•:%.•;•%%,:• :
..... •. ::•:-- .:.:::,•
':•-'•-•••% ................................
...•:•;,..:•---
---?•...
Fig. 6. Processed
P-wave seismicsectionfrom the Zamorafield area. Correlationsbetweenseismic
eventsand the top of important stratigraphicintervalsare indicated.
•.
':;•..••-•.-•-•-•
":• ....•. .':•-•.•,•,•.•,,•.... • •.••'"•-,-.
....:•. .'.p.........
P.... • '..
....
• • - ::•-...........
::.-.:• •..,
.... •.............. •- .....
.
...........
....•
---• .... .• •
....................
..........
•....•:•.
..................
.......
..... -..........
..•. ;•- • ---•:..• .::;:
..................
.................................. ..•:.<• :•..:
..........
.•,---•---• CapayShale
.:...:':-•
.........- .T¾:•::•
=.• ....
-¾'•::•-:-.••••::•:.:.•'-:
.... ••.•: StaCey
GasSand • .... :•.• .... :• ...•.;.,; .. ,• ...... • ......... ..-.•.:j-•
...
,•. ...:•..
..........
• ..... -...........
.
:• ........ . .. ....
Fig. 9. Processed P-wave seismic section from the Zamora field area. Correlations between seismic
events and the top of important stratigraphic intervals are indicated.
340 Correlation and Interpretation of P-P and P-SV Data Zamora Gas Field
sedimentsas they are more deeply buried and become The sonic log is converted to an approximate
better lithified. This conclusion is supported by the S-wave soniclog by applyingthe ARCO mudrock line,
dramaticdifferences
in V•,/V• ratiobetweenthe Ca- fromwhicha syntheticV•,/V• log is derived.Super-
pay and Sacramentoshales,both of which are marine posedonthislastlogaremeasured Vp/Vs ratiosfrom
shaleswith similar bulk compositions.Presumablythe correlated events picked near the well on Figure 11.
major difference between the two units is their degree The goodagreement
betweensyntheticVp/Vs logs
of lithification. and picked events suggeststhat the ARCO mudrock
The V•,/Vs ratiosdetermined
independently
from line is a good "first-order" approximation of V s for
seismic event correlations provide a "higher frequen- brine-saturated sand-shalesequencesin the absenceof
any real in-situ S velocities.
cy" profileof V•,/V• with depth,whereasthe low
frequency compressionof P-S V time to P-P time was
a first correction to approximately align the entire sec-
AMPLITUDE VARIATION WITH OFFSET
tions in time so that correlations could be interpreted.
MODELS
Granted that there is artistic license in this correlation,
it is interesting
to comparethe V•,/Vs ratiosfromthe In order to test the validity of event correlations and
picked events with expected ratios obtained from the interpretationsmade from the stackeddata, ampli-
applying the ARCO mudrock line to the Knaggs- tude variation with offset (AVO) reflectivity models
Wallace No. 1 well sonic log as shown in Figure 12. were prepared. Sonic and density logs from the
•:;.•.
:.•.•.......•.•;.•i:?•:.•::!.•.:.•::•..•/&•..>•:4.:.::..•..•*.•..;•.:•.•.•::(..•:..•.:•:::•::..:•
:.
..•Ti•.{i•.•:•;:-k.:•;•:::.-::
ß..•,:,:::::: . - •..• ..................
.
.....••;::•:;:"-".'." "••:..•':':-•,•...•.-:C'--:::•::•:'•?"•
':•...._•'•.•::.::•..•..2.-""•'
...... "••••-••2'•:•m. ' -•-•-•::m•
:•
:: -.•;.::
:.•:•' •:•.•?-
>:.•
'..- ..::
.. '-::-•'•"-'-'•'--•-'-':,•;::
"• . "••}:•'"'•...--'--'
'"...-:-•.......•-::•'•:::-•-•:•:•:::::•--------•:•
...•:;:•....
:'•½r•:::::•..._i•;;:•'
...... '•"••
.•:;•:='•:•::•:•;•:•;•'•-:::•'•-•
..................
;:::•:?.•.
...........
.......
======================
.:½:•::•.::•.:;?;:?•:.;•).;:•..:•-:.Z.D:•5•:;
-•.;'"• -•.•-•.•.
":;.&:::"'
....:
....................
:•-.............
:.:...•...:'
...........
:::'•;
.........
•:.••:•:•.•r;::--•.:.•::,.:•
.:;.•:•
........
::•......:;:•
...........
:,•.:•-•
........:-,:'-:-•::•<:.:"-:-:?.•.:::•:•..•-:-•-'•::;•?::::•.;}•:.:..
::.;•:•..;;•;•,::;;;?...:•;•:½.
..... .............
... ß ................. •.•::..... 5• :.:;•;"'%.•.'?':-.•
....•:.:•.........• ..................
:•-,•:•:
.... •.::•:.:•::.:.•:•:"
.:•:•,•;"
....:'•
.......
.....•::::::?:•
.......
•.•:.•:•.-•:
............
................
...•..:,..•
•:......
:•,::
............
........ .......:.:.:•::..•,•::.::?:
.•:•:...2:•<
?.............
•.•...•..•.
.•.--.•-•--•:•:::.•:.•
.......
'•-•::•.:.•:•.•
..:•.,r•;,,•:•.•,,;:•**-•...•h
...................
...... ?.•.•...:•
:.-..•s:..•..-•:-.-.-.•:
............
..-•..•:..•<•:.-•...,.
:....•;..-.•.•.-?•
.....?•::•.:•:::;::;::
.•:>½•...:::..•
- :;.?•.•-•.•.•
..S:::•
.•.-•,.::•;•,:•
:.'?.•ey Gas Sand
..... ......... ... .:....
.................. :..•.•:..•..•..•..............
•..-.m..::>,.•:.-:•::•:•..•'-::...../•
::•.,•
........
.:• •;,•::..:?:• -:•::;•..•.•;::•':;-:...:•.•
.....
•:::,.,•)•4;:-:•:.::::•.:
::•:::•.?
::'.-•:j
•.b:?,
..... •"'................
•:•:•"<:•:•*:::'
.•<4•:•,:...:•
'•:'•':::•'"
...... •..•:•:.•..•g.-:::
.:•:•: .......
..:•. ............
•.............
................•................
::;..•. . .... . . .:".:
..................
. ..............................
.•:•:::....•:•.•::.•:•4•
•
.....
...................•:•::.::;..•:•;--:•::•:::•:.
:•-•-::•-.-•
•:•.:'•-•::.,•.:.:-•:•:
::•,•:::.•
.....
-•-• ...........
:•?
.......
•:•.•.:...•::::•::•---•---•:::..•:::
..............
:::::::::::::::::::::
•':::,•:•:•:•'•"?:::'•
.......................... ......................
--.--.• -•-•••--:..
--•---:•'•.•::'"'-• .... •........ .Sacmine'hi:O: Shale
,•.•:•.•:•::::•:•::::•5•:':•:'•:•'
'•'• ... . .•..•-:.--'
...•
...............
:.:•:"'--•"'-•'
-•: .:........ :-:•%.....):
:'•-::.:;•
..................
..........•:•.•:..•-•:::.•:
.::::;•:-:•:•::.>:...-•:•:5:•-•::;•-•
'"':"••:'--•':::::::•:":.
.... •:½:--:-----•
.....
•"•'•'*½•':::::;:•:'•*'"'
•.•:•
.•:•.•::•:•½•:.•:.•
.. .:•:;
...r•: ::-•-
..-•:•:::::::::•:"
.•.½.: •:•:?::
:::•½,:-
:.•...
2 .........
:•-•..........::-•.:
.:•.•:.•:,-:;•.•::.....•.•.•-:
......
-,..:.:.::•..•:,•
...................... :.•::.::::..:.•5.. '....-...
Knaggs-Wallace Well No. 1 well provided the infor- A second model was prepared to increase the S
mation necessary for calculating P-wave velocities impedance contrast between overlying shale and Star-
and subsurface reflectivity. Rock physics programs key sand. This was done by increasingthe V s veloci-
were used to estimate V s for sand beds, while V s for ties for the sands (+ 10 percent) and lowering it for the
shaly, brine-saturated units were estimated using the shales (-10 percent). The modeled result (Figure 14)
ARCO mudrock line (Castagna, et al., 1985). does give a detectable increase in AVO with Poisson's
ratios on the order of .25 for the sand, but again falls
short of that observed in the real data.
STARKEY GAS SAND AVO MODELS A third model was prepared to see if a better
correspondencecould be achieved by varying the Vs
AVO models for the Starkey gas sand were prepared
for the sands (+20 percent) and shales (-20 percent).
to evaluate the effects of fluid saturation on AVO
The modeled result (Figure 15) gives a very detectable
response for P-P reflections.
increase in AVO that comes very close to the field trial
Well log analysis and formation tests indicate that
data. The Poisson's ratio for gas sands in this model
productive Starkey reservoirs have porosities on the
are on the order of. 15-: 18, which is within the scatter
order of 32 percent with gas saturations of over 80
percent. The first AVO example (with Poisson'sratios of rock physicspredictions. However, uncertaintiesin
on the order of .30 to .32; Figure 13) does a goodjob of offset gain corrections applied to data make more
matching events in time, however, it falls short in precise conclusionsunwarranted. In this data I s AEC
matching the increase in AVO that is seen in the actual functions were used to trace equalize the data before
seismic data. offset modeling was done.
Cemented Sand
Capay Shale
Starkey Sand
Sacramento Shale
2000
3000
Capay
Starkey
4000
5000
5300
656 /•S/m 164 2625 /•S/m 656
Fig. 12.S-wavetraveltime
andVp/Vsratiologsestimated
fromtheKnaggs-Wallace
No. 1P-wavesonic
log usingthe ARCO mudrocl•line and rock physicsprograms.Vp/Vs ratios estimatedusing
Knaggs-Wallace No. 1 P-wave sonic log using the ARCO mudrock line compare favorably with those
calculated independentlyfrom 3-C seismicevent correlations.
Vp Vs p O
Top 1.0
• = 32%
Sw- 20%
1.2
.1 .5
Fig. 13. Starkey gas sand AVO model for gas-saturatedcase (porosity = 32 percent; gas saturation =
80 percent).
Zaengle and Frasier 343
A fourth model was prepared for a brine-saturated cementation are controlling the amplitude of seismic
sand case with the same porosity of 32 percent. In the events in this interval.
example (Figure 16), the absenceof gas results in very
low reflectivity events that mimic those observed on
the stacked data where the reservoir sand is increas-
CONCLUSIONS
ingly shaly and water wet.
An integrated interpretation strategy that uti-
lizes P-P and P-SV seismic reflection data, rock
SHALLOW BRIGHT-SPOT AVO MODELS
physics relationships, and AVO modeling strate-
AVO models were prepared for the shallow carbon- gies can provide extra evidence to differentiate
ate-cemented sand that displays a high reflectivity between high-amplitude events from gas sands
P-wave event (Figure 9). Well log analysis and well and low-porosity, well-cemented sands. This
cuttings indicate that this interval has porosities on the seems to be useful in the Sacramento Valley in
order of 10 percent, with the remaining pore space areas where no shear log data are available.
occluded by calcite cement. The synthetic (Figure 17) In brine saturated sand-shale sequences the
does a goodjob of matching the timing and reflectivity ARCO mudrock line provides a good scaling
of events that are seen in the field trial data. These data function to compress P-SV reflection times to
support the conclusion that variations in carbonate P-P times so that events on the two sections can
Vp V s /3 O
1.0
1.2
.1 .5
Fig. 14. Starkey gas sand AVO model for gas-saturatedcase (porosity = 32 percent; gas saturation =
80 percent) using alternate V s estimatesfor sands(+ 10 percent) and shales(-10 percent).
Vp Vs /3 O
1.0
Top
cb=32%
Sw = 20%
1.2
.1 .5
Fig. 15. Starkey gas sand AVO model for gas-saturatedcase (porosity = 32 percent; gas saturation =
80 percent) using alternate V s estimatesfor sands(+20 percent) and shales(-20 percent).
344
Correlation and Interpretation of P-P and P-SV Data Zamora Gas Field
.1 .5
Top
•-10% 0.6 •
cement = 25%
0.8
o.1 0.5
sin0 (r-2) gp
ACKNOWLEDGMENTS
Rps(0) = +4 r
Rss - 8
(2)
The authors thank Chevron for permission to pub- where Rss is the zero offset approximate reflectivity
lish this paper. In particular, we thank Fred Herken- for SH waves given by
hoff of Chevron's Western Region for support of
multi-component recording, and A1 Hardin and Don
Winterstein of Chevron Oil Field Research Company Rss 2 Vs'
for valuable recording and processinghelp.
In the case of older, consolidated sediments where r =
APPENDIX--RELATION BETWEEN CONVERTED
2 we see that the second term inside the brackets in
AND SHEAR-WAVE REFLECTIVITIES
equation(2) vanishes.Then the Rps reflectivityat a
This appendix shows why the converted P-SV re- given offset will scale up or down proportional to the
flectivity at non-zero offset mimics SH-SH reflectivity Rss reflectivity. As shown in this paper and in previ-
at zero offset, within a scale factor, and is insensitive ous studies (Ensley, 1984; Robertson and Pritchett,
to P impedance changes across interfaces. The sim- 1985) r in the north Sacramento Valley can be 3 or
plest way to show this is with the modified Bortfeld higher.If r = 3 thenRps doesn'tquitescalepropor-
reflection coefficientsgiven in Aki and Richards (1980). tionally to R ssbut is close to it sincethe extra density
These coefficientsare good approximationsto the exact contrast effect inside the brackets of equation (2) is
reflectionformulasin casesof smallimpedancecontrasts small, i.e., (1/8)gp/p.
acrossan interface and at small incident anglesless than For seismicdata this simplerelationbetweenRps
critical. The formulas are much simpler than the exact and R ss can be useful for interpreters looking at
reflectioncoefficientsand are ideal for interpretersto use amplitude changesor anomalies in reflections without
for modelingand analysisof data. detailed processing or modeling analysis.
The reflection coefficients are expressed in terms of
reflectivitycomponents
gVp/ Vp, • Vs/ Vs and•9/9 at
an interface. These terms are the fractional changesin REFERENCES
P-wave velocity, S-wave velocity, and density across
the reflecting interface. These fractional changesequal Aki, K., and Richards, P. G., 1980, Quantitative seismology,
the actual contrasts from above to below the interface Theory and methods, Volume I, W. H. Freeman and
Company.
divided by the average parameter value of the two Castagna,J., Batzle, M., Eastwood, R., 1985, Relationships
layers. between compressional-waveand shear-wave velocities in
If we take the formula for P-SV reflectivity from clastic silicate rocks: Geophysics, 50, 571-581.
Ensley, R. A., 1984, Comparison of P- and S- wave seismic
Aki and Richards (1980) and take the near offset data: A new method for detecting gas reservoirs: Geo-
approximation, keeping only linear terms in sin 0, the physics, 49, 1420-1431.
incident angle, we obtain, after some algebra, Frasier, C. W., and Winterstein, D. F., 1990, Analysis of
conventional and converted mode reflections at Putah
Sink, California, using three-component data: Geophys-
Rps(O) = -4
wherer = Vp/Vs.
sin
0[(2
r +r)•91•Vs
8 p
+
2 Vs
(1) ics, 55, 646-659.
Robertson, J. D., and Pritchett, W. C., 1985, Direct hydro-
carbon detection using comparative P-wave and S-wave
seismic sections: Geophysics, 50, 383-393.