Professional Documents
Culture Documents
Abstract
This paper presents a simple method for seismic ray tracing in a general anisotropic medium,
which may include complex structures and compound materials, such as water, isotropic and
anisotropic rocks, fine-layers and parallel small cracked blocks. The anisotropy may be
defined by up to 21 density-normalized elastic moduli which vary with spatial position. The
method presented is a direct extension of the irregular network ‘shortest path’ method for an
isotropic solid. For this extension, we first apply analytic solutions of the wave velocities
(phase velocity and group velocity) for a general anisotropic medium as a ‘transform’ or
‘mapping’ operator to convert the elastic-moduli-described medium into the
direction-dependent group-velocity models for the three independent wave modes (qP, qS1,
qS2). We then utilize the ‘shortest path’ method to trace raypaths through such group-velocity
models for the three modes. We also give an alternative derivation of Fermat’s principle of
stationary time in anisotropic media. With this method, the travel times and ray paths of the
first arrivals emanating from a source to multiple receivers can be simultaneously obtained for
the three modes. Some 2D/3D numerical experiments are performed to show the accuracy and
applicability of the method. From these results, one can see that the method may be applied to
kinematic modelling and inversion in 2D and 3D seismic or seismological applications.
Keywords: seismic ray tracing, shortest path, anisotropy, first arrival, 2D/3D media
Ray tracing is an effective way to process and interpret dip-angle-dependent expression for handling group velocity
seismic data in practice. It may be used for seismic through its components. Qian and Symes (2002) developed
tomography (e.g. Chapman and Pratt 1992, Pratt and Chapman a finite difference method to calculate the travel time of qP
1992, Wang and Houseman 1995), seismic migration waves in anisotropic media. Alkhalifah (2002) demonstrated
(Alkhalifah and Larner 1994, Faria and Stoffa 1994a, 1994b) another travel time computation scheme by solving a linearized
and synthetic seismograms (Červeny 1972, Guest and Kendall eikonal equation for 2D and 3D VTI media. However, these
1993, Rüger and Alkhalifah 1996). The principal advantage travel time computation methods have difficulty in giving the
of ray tracing is the individual computation of the travel times ray paths due to the discrepancy between the phase-slowness
and the ray paths for the three wave modes. This provides vector (the gradient of travel time ∇τ ) and the group-velocity
the possibility of separately recovering media information vector (direction of the ray path) in an anisotropic medium.
from seismic observations. To simulate the ray paths and One cannot simply calculate the gradient of travel time to
travel times in anisotropic media, Červeny (1972) suggested obtain the ray paths for all pairs of shots and receivers in
applying a numerical method, such as the finite difference anisotropic media. In addition, the subsurface structure may be
method, to the partial differential equations he had earlier a composite assembly of VTI media with variable orientations
given. Unfortunately, the partial differential equations for of the symmetry axis due to folding, disruption and other
ray tracing in a general anisotropic medium are much more tectonic movements subsequent to deposition. This means
55
B Zhou and S A Greenhalgh
into the wave velocity (group velocity) space, in which we (θ0 , ϕ0 ) for the symmetry axis of the medium, one can easily
then adapt SPM for determining the travel times and the ray obtain the following group-velocity vector:
paths diverging from a source to all the nodes of the network.
U(m) = Uh(m) Hxn ex + Hyn ey + Uv(m) ez (6)
This ray tracing method is valid for computing the first arrival
travel time and giving the ray paths of the first arrivals for the where
three wave modes (qP, qS1, qS2). ∂c(m)
Uh(m) = c(m) sin ϑ + cos ϑ ,
∂ϑ
3. Anisotropic wave velocity model ∂c(m)
Uv(m) = c(m) cos ϑ − sin θ ,
∂ϑ
The kinematic properties of seismic waves are governed by
the Christoffel equation (Červeny 1972), ϑ = (n · ez )
det[aij kl (x)ni nl − c2 δj k ] = 0. (1)
Hxn = (n · ex ) (n · ex )2 + (n · ey )2 , (7)
Here, the fourth-order tensor, aij kl (x), denotes the density-
normalized elastic moduli that may be functions of the spatial Hyn = (n · ey ) (n · ex )2 + (n · ey )2 ,
coordinates x, and the quantities c and n = (n1 , n2 , n3 )
56
Anisotropic ray tracing
Equation (8) indicates that either the phase-slowness vector involve considering the direction-dependent group velocity
p(m) or the group velocity U (m) can be employed for the travel U (m) (x, r̂(m) ) and the manner by which we can efficiently find
time computation. Along the ray path, we have the following the minimum travel time over all possible paths linking the
equations (Červeny and Firbas 1984): two points xi and xj . In the next section, we will describe
a method for doing this, which is a direct extension of an
[c(m) ]2 = aij kl ni nl gk(m) gj(m) ,
irregular-network SPM for ray tracing in isotropic media.
dpi(m) (9)
1 ∂asj kl (m) (m)
= − nl ns g g .
dτ (m) 2[c(m) ]2 ∂xi k j 5. Ray tracing method
Note that ni are independent variables of the phase velocity.
As is well known, SPM is based on Huygens’ principle
We obtained the identity
(Musgrave 1970), which describes the following picture of
1 ∂c(m) dpi(m) wave propagation: every point on a wavefront is a new source
= − (m) (10)
c(m)
∂xi dτ of secondary waves. For the next wavefront the wave solution
by differentiating the first of equations (9). In equation is the summation of the contributions from all the ‘point
(10), we used the perpendicular relation gk(m) ∂gk(m) ∂xi = 0. sources’ on the previous wavefront. This is then combined
According to U(m) · p(m) = 1 and c(m) = n · U(m) , we have with Fermat’s principle embodied in equation (14)—the travel
57
B Zhou and S A Greenhalgh
a(x) U ( m ) (x, rˆ )
a(x) --Elastic moduli --primary nodes
U ( m ) (x, rˆ ) --Group velocity --secondary nodes
Figure 1. Model conversion from a gridded anisotropic medium (elastic moduli) into a group-velocity-gridded model field. Each cell or
element has the corner points or primary grid nodes (solid circles) and the secondary nodes (grey circles) on the six planes of the cell. The
corner points are used as the model parameter samples and the candidate points of a ray path. The other points are only used as the
candidate points for tracing a ray path through the element.
58
Anisotropic ray tracing
a11 = 15.1, a13 = 1.6, a33 = 10.8, a44 = 3.1, a66 = 4.3 (km /s) 2
qP-wave qSV-wave qSH-wave
(km)
(km) (km) (km) (km) (km)
(km) (km)
(km)
a11 = 25.7, a13 = 15.2, a33 = 15.4, a44 = 4.2, a66 = 9.0 (km /s) 2
qP-wave qSV-wave qSH-wave
Figure 2. 3D synthetic wavefronts at τ = 1s (or group velocities) for the three wave modes (qP, qS1, qS2) in Clay Shale and Green River
Shale.
(a )
qP-wave qS1-wave qS2-wave
(b )
qP-wave qS1-wave qS2-wave
Figure 3. First arrival travel time contours from 3D ray tracing for the three wave modes in Clay Shale with (a) vertical and (b) dipping
orientation (θ0 = 45◦ , ϕ0 = 0◦ ) of the symmetry axis. The solid lines are the results of the presented method and the posted triangles are
analytic solutions for comparison. The shot is located at the middle point on the surface.
the SPM are consistent with the analytic method (maximum The results are almost the same as those depicted for the xz-
relative error <1%), and the travel time contours significantly plane in figure 3.
change with the orientation angle of the symmetry axis. The As is well known, cell size and secondary node density
are important control parameters in SPM. The accuracy and
results for the qS1-wave show that the SPM method fails to
computational efficiency of the method depend on these
calculate the cusps (triplications of travel time), because we parameters (Moser 1991, Klimeš and Kvasnička 1994). In
are unable to use three values of the group velocity in a ray order to investigate the accuracy and efficiency of the method,
direction r̂(m)
ij in equation (15). This is a disadvantage of the we repeated the experiments shown in figure 1 with different
method. We also conducted the experiments with 2D models. cell sizes (dx = dy = dz = 0.25 km, 0.3 km, 0.5 km, 1.0 km)
59
B Zhou and S A Greenhalgh
(a ) (b )
Figure 4. The maximum relative errors in the travel time for the numerical 3D modelling shown in figure 3 versus the secondary node
density and the cell size.
(a )
7 100
N xe = N ze = 5
4.85
60
5 4.53
4.21 40
3.9 25.45
4 3.75
20
0.31 4.21
3 0
3 4 5 6 7 9 100x50 200x100 300x150 400x200
Secondary node density (points) 2D grid (cells)
(b )
300 180
CPU time (minutes)
266.6 161.4
250 150
Grid: 20 × 20 × 10 N =N =N =4
e
x
e
y
e
z
200 120
150 90
100 59.02
100 60
58
50 14.9
31 30 14.9
6.3 3.52
0 0
3 4 5 6 7 9 10x10x10 20x20x10 30x30x15 40x40x20
Secondary node density (points) 3D grid (cells)
Figure 5. CPU time versus secondary node density and grid dimension for 2D and 3D modelling.
and different secondary node densities Nxe = Nye = and secondary node density will benefit the accuracy and the
efficiency of the computations.
Nze = 3, 4, 5, 6, 7, 9 , recorded the computer times (CPU)
and calculated the relative errors (referred to the analytic Figure 5 gives the CPU times on a Pentium 4, 3.2 GHZ
solutions) in the 2D and 3D ranges of the models. Figure 4 PC for the 2D (a) and 3D (b) experiments. From the
shows the maximum relative error versus secondary node description of the method in the previous section, it is
clear that the CPU times should be unchanged, whether a
density (figure 4(a)) and cell size (figure 4(b)) for 3D
homogeneous or heterogeneous anisotropic model is being
models. The 2D experiments gave similar results to figure 4.
considered; computation speed depends only on the total
In figure 4(a), we used a constant cell size (dx = dy =
numbers of cells and nodes, which are given by the cell size
dz = 0.5 km) for all the secondary node densities. The plot
(dx, dy, dz) and the secondary node densities Nxe , Nye , Nze .
indicates that the accuracy is generally improved by increasing So, the CPU time shown in figure 5 may be considered as
the number of secondary nodes, but the improvement tapers a general guide to the cost of computer time for 2D and
off when the secondary node density exceeds 4 points for this 3D applications. Figure 5(a) shows that in the 2D case, the
model. In figure 4(b), we employed
a fixed secondary node CPU time gradually increases with the secondary node density
density Nxe = Nye = Nze = 4 for all cell sizes. The diagram when the cell size is fixed, and it shows a major jump with
shows that when the cell size reaches a certain level, i.e. cell number at around 80 000. Apparently, the method is
0.5 km in this case, the computational error does not applicable for most 2D situations. Figure 5(b) gives the CPU
significantly decrease with decreasing cell size. These times for the 3D numerical experiments. These diagrams show
experiments imply that an appropriate choice of cell size that the CPU time increases significantly with the secondary
60
Anisotropic ray tracing
(a )
qP-wave qS1-wave qS2-wave
(b ) qP-wave qS2-wave
qS1-wave
Figure 7. First arrival travel time contours and ray paths computed by the presented method for a compound transversely isotropic model.
The shot is located at the origin point (0, 0). The anisotropic parameters of each layer are adapted from Thomsen’s paper (1986). (a) All the
shale layers have vertical symmetry axes, and (b) the first two shale layers have a dipping symmetry axis orientation of 45◦ .
node density and the 3D grid dimension, and suggest that a 3D node density of 5 points were applied to these experiments.
application may employ 4 or 5 points for the secondary node Figure 6(a) gives the results for vertical symmetry axes of the
density, and up to several tens of thousands of cells. two rocks (VTI media) while figure 6(b) shows the change in
Figure 6 is an example of a 3D heterogeneous, anisotropic the travel time patterns when the symmetry axis of the Green
model. It shows the travel time distributions (wavefronts) of River Shale is rotated by 45◦ (gTI-medium).
the three wave modes for a model involving a vertical contact Figure 7 is an example of the method for ray tracing in
between two anisotropic media: the left side is Clay Shale a 2D heterogeneous, anisotropic model, which consists of a
and the right side is Green River Shale. Their group velocities water layer and three kinds of shale layer whose anisotropic
are given in figure 2. A cell size of 0.5 km and a secondary parameters were given by Thomsen (1986). Faria and Stoffa
61
B Zhou and S A Greenhalgh
62
Anisotropic ray tracing
Crampin S 1981 A review of wave motion in anisotropic and Mensch T and Rasolofosaon P 1997 Elastic-wave velocities in
cracked elastic media Wave Motion 3 343–91 anisotropic media of arbitrary symmetry—generalization of
Crampin S 1984 Effective anisotropic elastic constants for Thomsen’s parameters , δ and γ Geophys. J. Int. 128
wave-propagation through cracked solids Geophys. J. R. 43–64
Astron. Soc. 76 135–45 Moser T J 1991 Shortest path calculation of seismic rays
Daley P F and Hron F 1977 Reflection and transmission coefficients Geophysics 56 59–67
for transversely isotropic media Bull. Seis. Soc. Am. 67 661–75 Musgrave M J P 1970 Crystal Acoustics, Introduction to the Study
Dijkstra E W 1959 A note on two problems in connection with of Elastic Waves and Vibrations in Crystals (San Francisco:
graphs Numer. Math. 1 269–71 Holden-Day)
Ettrich N and Gajewski D 1998 Travel Time computation by Nakanishi I and Yamaguchi K 1986 A numerical experiment on
perturbation with FD-eikonal solver in isotropic and weakly non-linear image reconstruction from first arrival times for
anisotropic media Geophysics 63 1066–78 two-dimensional island structure J. Phys. Earth 34 195–201
Faria E L and Stoffa P L 1994a Finite-difference modeling in Pratt R G and Chapman C H 1992 Travel Time tomography in
transversely anisotropic media Geophysics 59 282–9 anisotropic media: II. Application Geophys. J. Int. 109 20–37
Faria E L and Stoffa P L 1994b Travel Time computation in Qian J and Symes W W 2002 Finite-difference quasi-P travel times
transversely isotropic media Geophysics 59 272–81 for anisotropic media Geophysics 67 147–55
Fryer G J and Frazer L N 1984 Seismic waves in stratified Rüger A and Alkhalifah T 1996 Efficient two-dimentional
anisotropic media Geophys. J. R. Astron. Soc. 78 691–710 anisotropic ray tracing Seismic Anisotropy ed F Erling,
Fryer G J and Frazer L N 1987 Seismic waves in stratified M H Rune and S R Jaswant pp 556–600
63