Professional Documents
Culture Documents
10.1190/GEO2018-0778.1
Manuscript received by the Editor 21 November 2018; revised manuscript received 13 August 2020; published ahead of production 12 September 2020;
published online 14 December 2020.
1
China University of Petroleum, Qingdao 266555, China and Qingdao National Laboratory for Marine Science and Technology, Laboratory for Marine
Mineral Resources, Qingdao 266555, China. E-mail: zhaoyunzong@yahoo.com (corresponding author); 306174763@qq.com.
© 2021 Society of Exploration Geophysicists. All rights reserved.
R1
(Vavryčuk and Pšenčík, 1998; Grechka, 1999; Zhu and Tsvankin, tion characteristics of reflection coefficients in orthotropic media in
2007). Ursin and Tjåland (1996) implement the exact inversion of terms of incidence angle and azimuth by using the approximate re-
multiple parameters. To improve the inversion stability, the exact flection coefficient equations (Rüger, 1998a; Pšenčík and Martins,
reflection coefficient equation was derived in terms of the three 2001; Bachrach et al., 2009; Bachrach, 2014, 2015; Oh and Alkha-
novel intermediate variables; it might also be expressed in terms lifah, 2016). Bachrach (2015) introduces two novel anisotropic gra-
of the bulk modulus, shear modulus, and density (Kurt, 2007). dient parameters based on the reflection coefficient equation of the
Although the exact equations provided perfect inversion accuracy, orthotropic media. The fracture-induced anisotropy in weakly
their high algebraic complexity deeply influenced the stability and anisotropic media with orthotropic symmetry was also widely stud-
efficiency of inversion (Hao and Stovas, 2017). The rewritten and ied (Pan et al., 2017a, 2017b; Osinowo et al., 2017; Pan and Zhang,
lower dimensional reflectivity equations help to improve the effi- 2018). For example, prestack anisotropic amplitude variation with
ciency and robustness of the inversion. In addition, they are helpful angle and azimuth (AVAZ) inversion for fracture characterization,
in analyzing the influence of the variation in the medium parameters by using an orthotropic system of symmetry for analysis of VTI-
across the interface on the reflection coefficients. For isotropic and layered formation containing a suite of aligned vertical fractures,
anisotropic media, approximate reflectivity equations have been which was equivalent to orthotropic media, was conducted for oil
widely discussed in the literature. For example, in weakly anisotropic and gas prediction and fluid identification (Narhari et al., 2015;
media, P- and S-wave velocities (V P0 , V S0 ), and anisotropic param- Kumar et al., 2018; Pan and Zhang, 2018).
eters including ε, δ, and γ are used to describe the anisotropic strength Seismic AVAZ inversion is usually used to estimate the aniso-
in transversely isotropic media (Thomsen, 1986). The popular tropic parameters of the underground media (Erik Rabben et al.,
approximate reflection and transmission coefficient equations of ver- 2008; Zong et al., 2015, 2017; Minato and Ghose, 2016; Zong
tical transverse isotropy (VTI) and horizontal transverse isotropy (HTI) and Yin, 2017; Pan et al., 2017b; Pan and Zhang, 2018). The
media were introduced by Rüger (1997, 1998a, 1998b). Furthermore, reflectivity characteristics of anisotropic media and the exact or
Schoenberg and Protazio (1990) calculate the exact solution of the approximate reflection coefficient equations of HTI or VTI or
plane-wave reflection coefficient by using the submatrix of the coef- orthotropic media have been well-discussed (Rüger, 1997, 1998a,
ficient matrix of the quasi-Zoeppritz equation. Zong et al. (2015) de- 1998b; Rüger and Tsvankin, 1997; Schoenberg and Helbig, 1997;
rive the seismic reflection coefficient equation in terms of Young’s and Tsvankin, 1997; Pšenčík and Martins, 2001; Bachrach, 2015; Zong
Poisson’s impedance to avoid unstable estimation of density. Bachrach and Yin, 2017). However, current orthotropic reflection coefficient
(2014) reconstructs the P-impedance, S-impedance, and anisotropic equations are usually in terms of several parameters or variables.
parameters in anisotropic fracture media, assuming horizontal trans- For example, seven variables were used in the equation derived
verse anisotropy symmetry, using linearized analysis. by Bachrach (2014), which rendered the inversion unstable. The
With the exploration and development of shale gas and shale oil, small contribution of the orthotropic parameters to the reflectivity
the complex fractured shale reservoirs with a suite of aligned ver- made the inversion more difficult. Therefore, in this study, we de-
tical or near-vertical fractures are usually assumed to be orthotropic veloped a robust and pragmatic AVAZ inversion approach to esti-
media. Two sets of vertical orthotropic fractures embedded in an mate the orthotropic parameters for improving inversion stability
isotropic background media or a set of parallel or orthogonal and reliability. We initially derived a three-term approximate PP-
vertical fractures embedded in VTI media is also equivalent to wave reflection coefficient equation in weak orthotropic media. By
orthotropic media (Schoenberg and Protazio, 1990; Tsvankin, 1997; convolving this equation with seismic wavelets as the forward
Bachrach, 2014; Pan and Zhang, 2018). Considering the weak solver, a robust AVAZ inversion approach with smooth background
anisotropy theory (Thomsen, 1986), Tsvankin (1997) derives the constraints under the Bayesian inversion framework was then used
weak anisotropy characteristic parameters of orthotropic media. to improve the stability of the inversion.
The rock with parallel fractures or directionally arranged micro-
cracks were equivalent to a group of anisotropic media (Schoenberg THEORY AND METHOD
and Douma, 1988). Schoenberg and Helbig (1997) propose that the
horizontal layered formation with vertical fractures ere equivalent to Approximate PP-wave reflection coefficient equation in
the orthotropic media and simulated seismic wave propagation in orthotropic media
this type of media. In addition, much research discusses the varia- A suite of aligned vertical or near-vertical fractures embedded in
a VTI media can be taken as orthotropic media; a schematic dia-
gram of the above orthotropic media is displayed in Figure 1.
The proposed rewritten and reduced dimensional seismic reflection
coefficient equation in orthotropic media is (details are given from
equations A-1 to A-12f in Appendix A)
C ¼ α−K ρβ2 eηy ; where A ¼ ρα is the P-wave impedance, B and C represent the de-
1
(2c)
gree of development of fractures on different axes of the symmetry
plane, respectively, θ is the incidence angle, ϕ is the azimuth, α and
2 β are the P- and S-wave velocities in isotropic background media,
2β respectively, ρ is the density of the media, Δ represents the difference
K¼ ; (2d) between lower and upper media for each parameter, − represents the
α
average value of the two layers for each parameter, and Δα∕α is the P-
wave reflection coefficient. The combined anisotropic parameters ηx
and ηy are defined as ðεx − δx Þ∕ð1 þ 2δx Þ and ðεy − δy Þ∕ð1 þ 2δy Þ,
δx − 2Kγ x respectively, and they represent the difference between the weak
ηx ¼ − ; (2e) anisotropic parameters δx and γ x , δy , and γ y , respectively. They are
K
dependent on the degree of fracture development. The bigger the val-
ues of ηx and ηy , the greater the degree of fracture developed. The
weak anisotropic parameters are ε, δ, and γ, whereas the anisotropic
δy − 2Kγ y parameters’ subscripts x and y represent anisotropic parameter defi-
ηy ¼ − ; (2f)
K nition on axis of symmetry xz-plane and xy-plane, respectively.
We used four models to demonstrate the accuracy of the novel calculated with the proposed novel approximate-orthotropic-reflec-
approximate reflection coefficient equation 1 with different types tivity equation diverge substantially from those calculated with
of AVO. In this model, the upper layer is isotropic whereas the the Zoeppritz’s exact-reflectivity-equation at large incident angles,
lower layer is orthotropic. The model parameters are displayed they show a high degree of conformance with the approximate
in Table 1. Figures 2, 3, 4, and 5 display the comparison of the Bachrach and Chen et al. equations. We thus conclude that the novel
Zoeppritz’s equation (the curves in black) (Zoeppritz and Erdbeb- equation is feasible for use in seismic inversion.
nenwellen, 1919), the Bachrach (2014) equation (the curves in red)
(Bachrach, 2014), the Chen and Yin (2015) equation (the curves in Bayesian AVAZ inversion in
green) (Chen and Yin, 2015), and the novel equation 1 (the curves orthotropic media
in green). Figures 2–5 display comparisons of the accuracy of
the reflection coefficients for the different types of AVO models In the Bayesian inversion framework, the a posteriori probability
including class I, II, III, and IV. Although the reflection coefficients distribution of the estimated model parameters can be predicted by
a priori probability distribution of the model parameters and the Using the matrix parameterization, equation 3 is rewritten as
likelihood function. In this study, we use the Cauchy and Gaussian
distribution as an a priori probability distribution of the model Rpp ¼ AR: (5)
parameters (Alemie and Sacchi, 2011; Zong et al., 2015) and the
likelihood function (Buland and Omre, 2003; Zong et al., 2015), Supposing the number of the azimuth is n and the number of the
respectively. The decorrelated covariance matrix is constructed incidence angle is k. Equation 5 can be expressed as
using the model parameters to implement the decorrelation of
2 3 2 3
the parameters to estimate and improve the lateral continuity. In RPP ðθ1 ;ϕ1 Þ aA ðθ1 ;ϕ1 Þ aB ðθ1 ;ϕ1 Þ aC ðθ1 ;ϕ1 Þ
addition, smooth trends of the model parameters are added to 6 RPP ðθ2 ;ϕ1 Þ 7 6 aA ðθ2 ;ϕ1 Þ aB ðθ2 ;ϕ1 Þ aC ðθ2 ;ϕ1 Þ 7
6 7 6 7
the objective function to improve the inversion stability. 6 .. 7 6 .. 7
Simplifying equation 1 as 6 . 7 6 . 7
6 7 6 72 3
6 RPP ðθk ;ϕ1 Þ 7 6 aA ðθk ;ϕ1 Þ aB ðθk ;ϕ1 Þ aC ðθk ;ϕ1 Þ 7
6 7 6 7 RA
6 .. 7 6 .. 74 5
RPP ðθ; ϕÞ ¼ aA ðθ; ϕÞRA 6 . 7¼6 . 7 RB :
6 7 6 7
6 R ðθ ;ϕ Þ 7 6 aA ðθ1 ;ϕn Þ aB ðθ1 ;ϕn Þ aC ðθ1 ;ϕn Þ 7 R
þ aB ðθ; ϕÞRB þ aC ðθ; ϕÞRC ; (3) 6 PP 1 n 7 6 7 C
6 RPP ðθ2 ;ϕn Þ 7 6 aA ðθ2 ;ϕn Þ 7
aB ðθ2 ;ϕn Þ aC ðθ2 ;ϕn Þ 7
6 7 6
6 .. 7 6 .. 7
where 4 . 5 4 . 5
RPP ðθk ;ϕn Þ aA ðθk ;ϕn Þ aB ðθk ;ϕn Þ aC ðθk ;ϕn Þ
1
aA ðθ; ϕÞ ¼ ; (4a) (6)
2
The seismic synthetic model can be expressed as
K 2
aB ðθ; ϕÞ ¼ − sin θ cos2 ϕ; (4b) S ¼ WRpp : (7)
2
Combining equations 5 and 7 as
K 2
aC ðθ; ϕÞ ¼ − sin θ sin2 ϕ; (4c) S ¼ WAR ¼ GR; (8)
2
where S is the matrix of the observable azimuthal seismic reflection
data, W is the matrix of the wavelet, G is the forward solver, and R
is the matrix of the model parameter.
R ¼ ½ RA RB RC ¼ ½ Δ ln A Δ ln B Δ ln C :
To keep the length of seismic records consistent with that of the
(4d) reflection coefficients, the dimension of the wavelet matrix should
2 3
be determined according to the number of sampling points k. The wð1;1Þ · aA ðθ1 ;ϕ1 Þ wð1;1Þ · aB ðθ1 ;ϕ1 Þ wð1;1Þ · aC ðθ1 ;ϕ1 Þ
wavelet matrix can be expressed as 6w wð2;1Þ · aB ðθ2 ;ϕ1 Þ wð2;1Þ · aC ðθ2 ;ϕ1 Þ 7
6 ð2;1Þ · aA ðθ2 ;ϕ1 Þ 7
6 7
6 .. 7
6 7
6 . 7
2 3 6 7
wk 0 0 0 0 0 0 6 wðk;1Þ · aA ðθk ;ϕ1 Þ wðk;1Þ · aB ðθk ;ϕ1 Þ wðk;1Þ · aC ðθk ;ϕ1 Þ 7
6 7
6 wkþ1 wk 0 0 0 0 0 7 6 7
6 7 6 .. 7
6 .. 7 ¼6 . 7
6 . wkþ1 wk 0 0 0 0 7 6 7
6 7 6 7
6 .. 7 6 wð1;nÞ · aA ðθ1 ;ϕn Þ wð1;nÞ · aB ðθ1 ;ϕn Þ wð1;nÞ · aC ðθ1 ;ϕn Þ 7
6 0 7 6 7
W ¼ 6 wi−1 . wkþ1 wk 0 0 7; 6 7
6 .. 7 6 wð2;nÞ · aA ðθ2 ;ϕn Þ wð2;nÞ · aB ðθ2 ;ϕn Þ wð2;nÞ · aC ðθ2 ;ϕn Þ 7
6 wi 0 7 6 7
6 wi−1 . wkþ1 wk 0 7 6 .. 7
6 .. 7 6 7
6 0 0 7 4 . 5
4 wi wi−1 . wkþ1 wk 5
.. wðk;nÞ · aA ðθk ;ϕn Þ wðk;nÞ · aB ðθk ;ϕn Þ wðk;nÞ · aC ðθk ;ϕn Þ
0 0 wi wi−1 . wkþ1 wk 2 3
RA
(9) 6 7
× 4 RB 5: (10)
RC
where w is the value of the seismic wavelet in different sample
points.
Equation 8 can be further written as There is a general correlation between each model parameter.
This problem can be resolved by the preconditioned-conjugate-gra-
dient optimization method (Zong et al., 2015). The covariance ma-
trix of the reflection coefficients for the model parameter is
2 3 2 3
Sðθ1 ;ϕ1 Þ aA ðθ1 ;ϕ1 Þ aB ðθ1 ;ϕ1 Þ aC ðθ1 ;ϕ1 Þ
6 Sðθ2 ;ϕ1 Þ 7 6 aA ðθ2 ;ϕ1 Þ aB ðθ2 ;ϕ1 Þ aC ðθ2 ;ϕ1 Þ 7 2 3
6 7 6 7 σ 2A σ AB σ AC
6 .. 7 6 .. 7 4
6 . 7 6 . 7 Cr ¼ σ BA σ 2B σ BC 5; (11)
6 7 6 7
6 Sðθk ;ϕ1 Þ 7 6 aA ðθk ;ϕ1 Þ aB ðθk ;ϕ1 Þ aC ðθk ;ϕ1 Þ 72 3 σ AC σ BC σ 2C
6 7 6 7 RA
6 . 7 6 .. 74 5
6 .
. 7 ¼W 6 . 7 RB
6 7 6 7
6 Sðθ ;ϕ Þ 7 6 aA ðθ1 ;ϕn Þ aB ðθ1 ;ϕn Þ aC ðθ1 ;ϕn Þ 7 RC where σ 2A is the variance of A, σ AB is the covariance of A and B, and
6 1 n 7 6 7
6 Sðθ2 ;ϕn Þ 7 6 aA ðθ2 ;ϕn Þ aB ðθ2 ;ϕn Þ aC ðθ2 ;ϕn Þ 7 more. Applying the singular value decomposition to matrix Cr ,
6 7 6 7
6 .. 7 6 .. 7
4 . 5 4 . 5 2 3
Sðθk ;ϕn Þ X σ 2A 0 0
aA ðθk ;ϕn Þ aB ðθk ;ϕn Þ aC ðθk ;ϕn Þ
Cr ¼ u uT ¼ u4 0 σ 2B 0 5uT ; (12)
0 0 σ 2C
Figure 6. The synthetic seismic traces: (a) synthetic seismic traces with no noise, (b) synthetic seismic traces when the S/N is 5, and (c) syn-
thetic seismic traces when the S/N is 2.
where u is the eigenvector. Defining the inverse of u as The posterior probability distribution function of the estimated
2 3 model parameters is
u11 u12 u13
u−1 ¼ 4 u21 u22 u23 5: (13) N
Y
1
u31 u32 u33 pðR 0 jSÞ ∝
i¼1
1 þ R2i ∕σ 2R
Considering T time samples, the inverse of eigenvector u−1 be- −ðS − G 0 R 0 ÞT ðS − G 0 R 0 Þ
comes U−1 , · exp ; (19)
2σ 2k
2u 0 ··· u12 0 ··· u13 0 ··· 3
11
where N represents the number of sampling points, and σ 2k and σ 2R are
6 0 u11 0 · · · u12 0 · · · u13 0 7
6 7 the variance of the noise and the model parameters, respectively.
6 ··· 7
6 7 Substituting equation 17 into the marginalization equa-
6 u21 0 ··· u22 0 ··· u23 0 ··· 7
−1 6 7 tion (Downton, 2005) and maximizing the posterior probability dis-
U ¼ 6 0 u21 0 · · · u22 0 · · · u23 0 7; tribution, the objective function FðR 0 Þ is
6 7
6 ··· 7 X
6 7 N
6 u31 0 ··· u32 0 ··· u33 0 ··· 7 FðR0 Þ¼ðS−G0 R0 ÞT ðS−G0 R0 Þþ2σ 2k lnð1þR2i ∕σ 2m Þ:
4 5
0 u31 0 · · · u32 0 · · · u33 0 i¼1
··· (20)
(14)
To improve the stability of the inversion, the model smoothness
decorrelating the parameters for equation 8. The preconditioned constraint is added to the objective function. Finally, we obtain the
wavelet matrix G 0 and model matrix R 0 are objective function as
G 0 ¼ GU
; (15)
R 0 ¼ U−1 R
and
S ¼ G 0R 0; (16)
YM
1 1
pCauchy ðR 0 Þ ¼ ;
ðπσ R 0 ÞM i¼1 1 þ R2i ∕σ 2R 0
(17)
1
pðSjR 0 Þ ¼ pffiffiffiffiffi
2π σ k
−ðS − G 0 R 0 ÞT ðS − G 0 R 0 Þ
exp ; (18)
2σ 2k Figure 8. The estimated and real values of A, B, C, ηx , and ηy when the S/N is 5:1. The
curves in blue represent the estimated values, those in red represent real values, and
where σ 2k is the variance of the noise distribution. those in green represent the initial model.
X
N ηA ¼ 1∕2 × lnðmA ∕mA0 Þ; (23b)
FðR 0 Þ ¼ ðS−G 0 R 0 ÞT ðS−G 0 R 0 Þþ2σ 2k lnð1þR2i ∕σ 2m ÞþΛ;
i¼1
(21) ηB ¼ 1∕2 × lnðmB ∕mB0 Þ; (23c)
where
Λ ¼ λA ðηA − PA RA0 ÞT ðηA − PA RA0 Þ þ λB ðηB − PB RB0 ÞT ηC ¼ 1∕2 × lnðmC ∕mC0 Þ; (23d)
× ðηB − PB RB0 Þ þ λC ðηC − PC RC0 ÞT ðηC − PC RC0 Þ; (22)
and λA , λB , and λC are the constraint coefficients for the parameter to m ¼ ½ mA mB mC ¼ ½ A B C ; (23e)
be inverted A, B, and C, respectively.
Z where mA0 , mB0 , and mC0 are the initial values of the model param-
ti eters, respectively.
Pi ¼ dτ; (23a) Minimizing the objective function equation 21 yields
t0
ΓR ¼ Ψ; (24)
where
Γ ¼ G 0T G 0 þλc Qc þλA PTA PA þλB PTB PB
þλC PTC PC ;
(25a)
Ψ ¼ G 0T S þ λA PTA PA þ λB PTB PB
þ λC PTC PC : (25b)
Figure 10. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 15° (and a range of 0°–30°). The black
curve represents the (smoothed) P-wave impedance from sonic and density logs.
EXAMPLES The ranges of incidence angles are 0°–10°, 10°–20°, and 20°–30°,
and the average incidence angles used in equation 1 are 5°, 15°,
Synthetic examples and 25°, respectively. The ranges of azimuths are 0°–30°, 30°–60°,
By using a suite of well log data in a fractured field containing 60°–90°, 90°–120°, 120°–150°, and 150°–180°, and the average azi-
anisotropic information as the model parameters, we test the fea- muths used in equation 1 are 15°, 45°, 75°, 105°, 135°, and 165°,
sibility of the proposed method for estimating the anisotropic respectively. Figure 6 shows the synthetic seismic traces with the dif-
parameters. The synthetic seismic traces are simulated with the con- ferent signal-to-noise ratios (S/Ns), whose incidence angles range
volution of reflection coefficient equation 1 and seismic wavelets. from 0°–30°. The real (the curves in red) and estimated (the curves
Figure 11. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 45° (and a range of 30°–60°). The black
curve represents the (smoothed) P-wave impedance from sonic and density logs.
Figure 12. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 75° (and a range of 60°–90°). The black
curve represents the (smoothed) P-wave impedance from sonic and density logs.
in blue) A, B, C, ηx , and ηy are displayed in Figure 7. The elastic estimated values of A, B, C, ηx , and ηy with S/N are 5:1 and 2:1 are
isotropic and anisotropic combined parameters could be reasonably displayed in Figures 8 and 9, respectively. By comparison of the
estimated even with fairly smooth initial models (Figure 7). estimated and the real parameters, we can still reasonably estimate
Gaussian random noise is added into the synthetic traces to test A, B, C, ηx , and ηy even with moderate noise. These synthetic ex-
the stability of the proposed inversion method, and we implement amples demonstrate the feasibility and robustness of the proposed
the seismic inversion using synthetic traces with different S/N. The inversion approach.
Figure 13. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 105° (and a range of 90°–120°). The
black curve represents the (smoothed) P-wave impedance from sonic and density logs.
Figure 14. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 135° (and a range of 120°–150°). The
black curve represents the (smoothed) P-wave impedance from sonic and density logs.
Field data examples Then, the anisotropic combined parameters ηx and ηy can be further
estimated with equations 2e and 2f. Figure 17a-17e shows the esti-
The proposed inversion approach is further tested on a real shale mated results of the combined parameters A, B, and C, and the novel
oil data set from a work site in eastern China where vertical or near-
anisotropic combined parameters ηx and ηy , respectively, wherein
vertical fractures developed within layered formations. In this case,
the yellow and green sections in the well interpretation column il-
we treat the shale reservoirs with a single set of vertical or near-
lustrate the target layers where fractures and nonfractures develop,
vertical fractures as orthotropic media. To ensure that the strength
respectively. We can see that the estimated anisotropic combined
of the subsurface reflection interfaces is correct, we use wide azi-
parameters have good lateral continuity in Figure 17d and 17e,
muthal observed seismic reflection data that have
been processed by a contractor. We use the angle
stack seismic gathers, which are the common in-
cidence angle stacks and the common azimuth
stacks, to improve the S/N of the observed seis-
mic data. The average azimuths are 15°, 45°, 75°,
105°, 135°, and 165°, respectively. Figures 10,
11, 12, 13, 14, and 15 show the partial azimuthal
stack seismic profiles with different incident an-
gle ranges, and the average angles for the near,
middle, and far stacks are 5°, 15°, and 25°, re-
spectively. The red line demonstrates the well po-
sition, and the black curve represents the P-wave
impedance smoothed from sonic and density
logs. Figure 16 shows the results of a fracture’s
quantitative interpretation and formation micro-
scanner image (FMI) characteristics, fracture true
dip, fracture depth, and fracture tendency, frac-
ture direction, and fracture dip of the different
layers developing fractures. From Figure 16,
we can see that fractures are mainly developed
in the layers below 3 km, where the fractures
are more effective and ductile.
The terms A, B, and C are first estimated by Figure 16. Fractures’ quantitative calculation results for well A, which show the FMI,
fracture true dip, fracture depth and the fracture tendency, fracture direction, and fracture
implementing the proposed Bayesian inversion dip of different fracture types.
on the field azimuthal observed seismic data.
Figure 15. Azimuthal observable seismic data at different incidence angles, with an average azimuth of 165° (and a range of 150°–180°). The
black curve represents the (smoothed) P-wave impedance from sonic and density logs.
CONCLUSION
Orthotropic models are usually used for reservoir forecasting in
shale oil and gas reservoirs with a single set of aligned vertical or
near-vertical fractures. We derived and tested a new approximate
equation for PP-wave reflection coefficient calculation that reduces
the number of estimated parameters and improves the robustness of
the inversion. A pragmatic Bayesian AVAZ inversion method with
smooth background constraints was used to estimate the parame-
ters. The anisotropic parameters in orthotropic media were reason-
ably estimated in the synthetic and field data examples, which
demonstrates the feasibility of the proposed novel reflection coef-
ficient equation and the stability of the inversion method in frac-
tured reservoirs. The proposed approach for model and seismic
reflectivity parameterization helps to enhance the robustness and
reduce the inversion multiplicity in orthotropic parameter estima-
tion for horizontal layered formations with a single set of aligned
vertical or near-vertical fractures. This will be helpful to guide the
fracture prediction and sweet spot evaluation of shale oil and gas
reservoirs with pre-tack azimuthal seismic data.
ACKNOWLEDGMENTS
We would like to acknowledge the sponsorship of the National
Nature Science Foundation of China (41974119, U1762103), the
Science Foundation for Innovation and Technology Support Pro-
gram for Young Scientists in Colleges of Shandong Province,
and Ministry of Science and Technology of China
(2020RA2C620131).
APPENDIX A
THE APPROXIMATE REFLECTION COEFFICIENT
EQUATION IN ORTHOTROPIC MEDIA
In this appendix, we describe the derivation process of the
approximate reflection coefficient equation (as in equation 1). The
approximate reflection coefficient in orthotropic media presented
by Bachrach (2014) is
Figure 17. The estimated parameters in orthotropic media: (a) the RPP ðθ; ϕÞ ¼ Riso
PP ðθ; ϕÞ þ RPP ðθ; ϕÞ;
Aniso (A-1)
inversion result of combined parameters A, (b) the inversion result
of combined parameters B, (c) the inversion result of combined where
parameters C, (d) the inversion result of anisotropic parameters
ηx , and (e) the inversion result of anisotropic parameter ηy.
PP ðθ; ϕÞ ¼ A þ b1 sin θ þ b2 sin θ tan θ;
Riso 2 2 2
(A-2)
lously high values at the oil reservoir position where the fractures
develop. This demonstrates that the high value of the anisotropic þ ðb5 cos4 ϕ þ b6 sin4 ϕ þ b7 cos2 ϕ sin2 ϕÞ sin2 θ tan2 θ;
combined parameters may be used to characterize the fractured res- (A-3)
ervoirs. The estimated novel anisotropic parameters are reasonably
consistent with prior knowledge of the target reservoir, which where
demonstrates that the proposed method is helpful in the oil and
1 ΔI P
gas prediction of shale reservoirs with a single set of vertical or A¼ ; (A-4a)
near-vertical fractures. 2 IP
2 2
1 Δα 2β Δðρβ2 Þ 1 β
b1 ¼ − ; (A-4b) ΔΓy ¼ ðΔδy − 8 Δγ y Þ: (A-6d)
2 α α ðρβ2 Þ 2 α
ΔX
with X
≈ ΔInX,
Δε
b5 ¼ x ; (A-4f) 1
2 RPP ðθ;ϕÞ ¼ ΔlnðραÞ
2
0 2 2 1
1
Δ ln α− 2β
Δ lnðρβ 2 ÞþΔlneδx −8 β Δlneγ x cos2 ϕ
B2 α α C
Δεy B C 2
þB 2 2 Csin θ;
b6 ¼ ; (A-4g) @ 1 δ β γ
A
2 þ 2 Δ ln α− α Δ lnðρβ ÞþΔlne −8 α Δlne sin ϕ
2β 2 y y 2
(A-8)
Δδ 2
b7 ¼ z ; (A-4h) with 2β
¼ K, then
2 α