You are on page 1of 12

GEOPHYSICS, VOL. 74, NO. 5 SEPTEMBER-OCTOBER 2009; P. U35U46, 12 FIGS., 3 TABLES.

10.1190/1.3169600

Sensitivity kernels for seismic Fresnel volume tomography

Yuzhu Liu1, Liangguo Dong1, Yuwei Wang2, Jinping Zhu2, and Zaitian Ma1

Sheriff and Geldart, 1982; Dziewonski, 1984; Nolet, 1987; Harris


et al., 1992; Pulliam et al., 1993; Billette and Lambar, 1998; Aki
and Richards, 2002. However, it cannot resolve the fine structures
of the media and is insensitive to high spatial wavenumber variations
of the media because of its infinite frequency approximation. Waveequation tomography also has been studied since the 1980s Devaney, 1984; Wu and Toksz, 1987; Pratt and Goulty, 1991; Woodward, 1992; Schuster and Quintus-Bosz, 1993. Theoretically,
wave-equation tomography should give higher resolution results
than raypath tomography. However, it has not been widely used in
practice until now because of very high computational cost, low signal-to-noise ratio, and imprecise seismic wavelets.
In seismic experiments, the recorded transmitted signals contain
broadband frequency waves. For a single source-receiver pair, not
only the points on the raypath but also those outside the ray affect
wave propagation Woodward, 1992; Snieder and Lomax, 1996;
Marquering et al., 1999; Spetzler and Snieder, 2001, 2004. In raypath tomography, the time delay or the amplitude variation is only
projected to the raypath. It does not consider the influence of the offray points on wave propagation; therefore, the inversion result is inaccurate. To take advantage of raypath tomography and consider
wave propagation, fat-ray tomography has been proposed Michelena and Harris, 1991; Vasco and Majer, 1993; Watanabe et al., 1999;
Xu et al., 2006. However, fat-ray tomography just fattens the ray on
certain principles without considering the physics of seismic wave
propagation.
The fat ray constructed on the basis of the wave equation is the
sensitivity kernel, also called the Frchet kernel Tarantola, 1987. It
reflects the sensitivity of the recorded seismic signals including amplitude and traveltime to the perturbation of the media. The region
with the highest sensitivity corresponds to the first Fresnel volume,
through which the main energy of the wave travels to the receiver
around the geometric ray erven and Soares, 1992; Woodward,
1992. This kind of tomography related to the Fresnel volume is
called Fresnel volume tomography FVT Yomogida, 1992;
Snieder and Lomax, 1996; Marquering et al., 1998, 1999. In FVT,
the time delay or the amplitude variation is mapped to the first

ABSTRACT
Fresnel volume tomography FVT offers higher resolution and better accuracy than conventional seismic raypath
tomography. A key problem in FVT is the sensitivity kernel.
We propose amplitude and traveltime sensitivity kernels expressed directly with Greens functions for transmitted
waves for 2D/3D homogeneous/heterogeneous media. The
Greens functions are calculated with a finite-difference operator of the full wave equation in the frequency-space domain. In the special case of homogeneous media, we analyze
the properties of the sensitivity kernels extensively and gain
new insight into these properties. According to the constructive interference of waves, the spatial distribution ranges of
the monochromatic sensitivity kernels in FVT differ from
each other greatly and are 1 / 8, 2 / 8, 3 / 8 and 4 / 8 periods of
seismic waves, respectively, for 2D amplitude, 3D amplitude, 2D traveltime, and 3D traveltime conditions. We also
have a new understanding of the relationship between raypath tomography and FVT. Within the first Fresnel volume of
the dominant frequency, the band-limited sensitivity kernels
of FVT in homogeneous media or smoothly heterogeneous
media are very close to those of the dominant frequency.
Thus, it is practical to replace the band-limited sensitivity
kernel with a few selected frequencies or even the single
dominant frequency to save computation when performing
band-limited FVT. The numerical experiment proves that
FVT using our sensitivity kernels can achieve more accurate
results than traditional raypath tomography.

INTRODUCTION
Seismic raypath tomography based on the asymptotic highfrequency solution of the wave equation has been used widely in
seismology, geodynamics, engineering, and petroleum exploration

Manuscript received by the Editor 17 July 2008; revised manuscript received 10 April 2009; published online 14 September 2009.
1
Tongji University, State Key Laboratory of Marine Geology, Shanghai, China. E-mail: liuyuzhu@tongji.edu.cn; dlg@tongji.edu.cn; mazaitian@tongji.
edu.cn.
2
Tongji University, School of Ocean and Earth Science, Shanghai, China. E-mail: willwangsmail@gmail.com; huayihuayu@163.com.
2009 Society of Exploration Geophysicists. All rights reserved.

U35

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U36

Liu et al.

Fresnel volume according to the sensitivity kernel. For simplicity,


we ue the term Fresnel volume to denote the first Fresnel volume unless noted otherwise.
Since Slaney et al. 1984 and Wu and Toksz 1987 applied
Born and Rytov approximations to diffraction tomography and studied their inversion abilities, diffraction tomography based on Born
or Rytov approximation has undergone extensive seismic investigation. Woodward 1992 provides a uniform expression on raypath
tomography and diffraction tomography, and he proposes the concepts of wavepath and band-limited raypath. erven and Soares
1992 present a modeling method for the Fresnel volume. Yomogida 1992 calculates the band-limited raypath according to the
traditional Fresnel volume and applies it to diffraction tomography.
Snieder and Lomax 1996 derive the sensitivity kernels of a velocity perturbation on phase and amplitude according to Born and Rytov
approximations.
Following Snieder and Lomax 1996, Spetzler and Snieder
2001, 2004 and Jocker et al. 2006 propose band-limited amplitude and traveltime sensitivity kernels for homogeneous media and
summarize some of their properties. Spetzler et al. 2007, 2008 propose the sensitivity kernels for smoothly heterogeneous media with
a Jacobi determinant and apply them to time-lapse crosswell tomography. Tian et al. 2007 also propose a method to calculate the sensitivity kernels by using dynamic ray tracing. Marquering et al. 1998,
1999 derive sensitivity kernels for finite-frequency seismic traveltimes measured by crosscorrelating broadband waveforms. Following the method of Marquering et al. 1998, 1999, many studies
Dahlen et al., 2000; Zhao et al., 2000; Hung et al., 2001; Dahlen,
2005; Zhao and Jordan, 2006; Zhang et al., 2007 investigate the
properties of band-limited traveltime sensitivity kernels in homogeneous media and analyze the influence of abnormal factors on wave
propagation in seismology and geodynamics.
In this paper, on the basis of previous work Woodward, 1992;
Snieder and Lomax, 1996; Spetzler and Snieder, 2001, 2004; Jocker
et al., 2006; Spetzler et al., 2007, 2008, we present the amplitude
and traveltime sensitivity kernels expressed directly with Greens
functions. They are suitable for homogeneous and heterogeneous
media. In the special case of homogeneous media, we extensively
analyze the properties of the monochromatic and band-limited sensitivity kernels by using the constructive interference of waves. The
numerical experiments indicate that the properties of the homogeneous case can also be used approximately for smoothly heterogeneous media. In the last part, the near-surface velocity structure of a
theoretical model is inverted using FVT with sensitivity kernels and
the corresponding properties proposed in this paper.

METHODS
For a recorded seismic signal, amplitude perturbation A is an integral of the amplitude sensitivity kernels KAr multiplied by the
slowness perturbation field sr over all points r in volume v between the source and the receiver. It can be expressed as

KArsrdr.

Similarly, traveltime perturbation can be expressed as

1a

KTrsrdr,

1b

where KTr is the traveltime sensitivity kernel Woodward, 1992;


Snieder and Lomax, 1996; Dahlen et al., 2000; Spetzler and Snieder,
2001, 2004.
These functions can be generalized to a form
vKrsrdr for all kinds of seismic tomography. How to calculate Kr is the key problem in tomography. In fact, Kr is the sensitivity of observations to media perturbation. In equations 1a and
1b, Kr is the sensitivity of amplitude or traveltime variations to the
slowness perturbation per unit area in two dimensions or per unit
volume in three dimensions. In FVT, it is restricted in the Fresnel
volume.
In wave-equation tomography, Kr is called the wavepath
Woodward, 1992. According to seismic information used in tomography, Kr corresponds to different sensitivity kernels, such as
the amplitude sensitivity kernel KAr, traveltime sensitivity kernel
KTr, and waveform sensitivity kernel KUr. The symbol corresponds to amplitude variation A, traveltime delay , or wavefield residual u, respectively. These residuals can be calculated using different theories, such as ray theory, FVT, or seismic wave theory, according to how the sensitivity kernel is constructed. In this
work, we only study KAr and KTr based on FVT.
Following Jocker et al. 2006, we substitute equation 3 into equations 2; comparing with equations 1a and 1b, we get the monochromatic sensitivity kernels equations 4 expressed with Greens functions directly:

A A0 Re

Im

u1g,s

u1
,
u0

2a

u1
u0
;

2b

2 2sr
G0g,ru0r,sdr;
V0r
v

KAr,

2 2A 0
G0g,ru0r,s
,
Re
V0r
u0g,s

4a

KTr,

2
G0g,ru0r,s
Im
.
V0r
u0g,s

4b

In equations 24, u0 denotes the unperturbed wavefield, u1 is the


first-order Born scattering wavefield, A0 is the unperturbed amplitude, is the angular frequency, G0g,r is the Greens function of
point g from a point source in r for the unperturbed media V0r, and
u0r,s is the unperturbed field in r from a point source in s.
Equations 4 are the general forms of the sensitivity kernels for homogeneous and heterogeneous media. They involve calculating the
Greens function G0 and synthetic wavefields u0, which can be calculated with wave-equation modeling or dynamic ray tracing. We calculate them directly through a finite-difference operator of the full
wave equation in the frequency-space domain Jo et al., 1996.
Considering that the recorded transmitted signals contain broadband waves, we get the corresponding band-limited sensitivity ker-

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Sensitivity kernels
nels by adding the normalization integral over frequency Spetzler
and Snieder, 2001, 2004:

U37

homogeneous media, the monochromatic sensitivity kernels can be


expressed by the equations listed in Table 1:

KAr

P KAr, d ,

i 1
G2D
0 g,s H0 k0s g,
4

5a

G3D
0 g,s

KTr

P KTr, d .

eik0sg
.
4 s g

5b

7a
7b

In equations 7, s and g denote the source and receiver positions, respectively; H1


0 is the zeroth-order Hankel function of the first kind,
1
and k0 is a spatial wavenumber in unperturbed media. In the equaHere, P is the weight coefficient calculated from the amplitude
tions in Table 1, rs, rg, and sg are distances from spatial point r to
spectrum or a given Gauss function. It satisfies the relationship
source s, from r to receiver g, and from s to g, respectively. The value
12 P d 1. We express a Gaussian distribution in equations 6,
t is the traveltime delay between the detour path s r g and the
selected for the weight function:
direct wavepath s g, that is, t rs rg sg / V.
The sensitivity kernels in Table 1 are almost the same as those dewf
rived by Spetzler and Snieder 2004, except that the approximations
,
6a
Pf f
2
1 / rs 1 / x and 1 / rg 1 / sg x are made during their derivation Figure 1. Although the geometric spreading factors after the
wfdf
approximations are accurate enough, we do not follow them in this
f1
study because rs, rg, and sg are easy to obtain.
Fresnel volume can be interpreted physically as a region where
1
2
2
wf
ef f 0 /2 .
6b
scattered waves interfere constructively with the direct wave. It is

2
usually around the corresponding central ray. Slowness perturbations inside the Fresnel volume have the greatest effect on seismic
In equations 6, f is the circular frequency and denotes the bandwave propagation from the source to the receiver. The distribution
width of the frequencies with the highest energy near the central cirrange of the Fresnel volume can be defined by the maximum detour
cular frequency f 0.
traveltime delay tmax Spetzler and Snieder, 2001, 2004. AccordBy crosscorrelating broadband waveforms Marquering et al.,
ing to the sensitivity kernels listed in Table 1, the distribution ranges
1998, 1999, the band-limited sensitivity kernels also can be derived
of the Fresnel volumes corresponding to different conditions can be
Zhang et al., 2007. The resulting expressions are different from
calculated easily.
ours, but choosing a special weight function can equalize them.
The results are listed in Table 2, where T is the period of the monochromatic wave. From Table 2, we can see that tmax is not always
PROPERTIES OF SENSITIVITY
equal to T / 2, as some tomographic inverse schemes use Michelena
KERNELS FOR FVT
and Harris, 1991; Vasco and Majer, 1993; Watanabe et al., 1999; Xu
et al., 2006. It varies with dimensions and seismic information but
To perform FVT, the spatial distribution and characteristics of the
is independent of the source type, according to Liu and Dong
sensitivity kernels should be studied in detail. The kernels for hetero2008. Only for 3D traveltime tomography tmax 4T / 8 is the
geneous media are complicated, and it is not easy to summarize their
Fresnel volume the broadest. Liu and Dong 2008 propose the same
properties. However, they should share some properties with the
expressions and spatial distributions of the sensitivity kernels, but
sensitivity kernels for homogeneous media. So, we first analyze the
they use the far-field approximation and stationary-phase approxispecial case of a point source in homogeneous media.
mation following Snieder and Lomax 1996 during the derivation.
In homogeneous media, the unperturbed wavefield u0 in equations
Fresnel volume usually is defined geometrically as a special range
4 equals the Greens function. According to the analytical formulas
and is suitable only for monochromatic signals. Now, by combining
of the Greens functions in 2D equation 7a and 3D equation 7b
with the sensitivity kernels equations 5 derived
from wave theory, we can define the Fresnel volTable 1. Amplitude and traveltime sensitivity kernels of finite-frequency
ume for the band-limited signal.
tomography in homogeneous media.
Considering the frequency band in seismic exploration, we set the frequency f to vary from
KTr,
Dimension
KAr,
0 to 100 Hz with a frequency interval of 2 Hz
and a dominant frequency of 50 Hz. As an example, we set as 20 and f 0 as 50 Hz in this experi2D
ment. The weighting curve without normalizasg 3
sg

cos t
sin t
A0 ,g s
tion is shown in Figure 2. The band-limited sensi2 Vrsrg
4
2 Vrsrg
4
tivity kernels are illustrated in Figure 3. The color
2
scale indicates the relative sensitivity of ampli3D
sg
sg
A0 ,g s
cos t
sin t
tude or traveltime variation to slowness perturba2 Vrsrg
2 Vrsrg
tion in the Fresnel volume. To highlight the properties of the sensitivity kernels, different gains are

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U38

Liu et al.

l sg

g
l rs

l rg
r

Figure 1. The 2D definition of the geometry for a point source in homogeneous media.
Table 2. Distribution ranges of Fresnel volumes in different
conditions.

Dimension

Traveltime
Fresnel volume

tmax T / 8
tmax 2T / 8

tmax 3T / 8
tmax 4T / 8

Weight coefficient

2D
3D

Amplitude
Fresnel volume

Frequency (Hz)

Figure 2. Weight curve without normalization, calculated from


equation 6b.

applied to the plots. The white dashed line in every plot denotes the
distribution range of the corresponding Fresnel volume of the dominant frequency. To exhibit the characteristics of the sensitivity kernels clearly, the vertical cross sections at position x 500 m Figure
3 are shown in Figure 4.
To analyze the properties of the sensitivity kernels in smoothly
heterogeneous media, the band-limited sensitivity kernels in media
with a constant velocity gradient are calculated in the same way. The
kernels are illustrated in Figure 5, and the corresponding cross sections are shown in Figure 6.
From the equations in Table 1 and from Figures 36, we come to
four major conclusions. First, the side lobes of the band-limited sensitivity kernels cancel out because of the destructive interference of
the monochromatic kernels when integrating over broadband
frequencies. Meanwhile, the main lobe in the Fresnel volume is
strengthened by the constructive interference. This means the bandlimited sensitivity kernels and Fresnel volumes can be used to well
describe the seismic wave propagation between two points.
Second, we see in Figures 4 and 6 that the green lines coincide
with the red lines very well in the Fresnel volumes. Thus, it is reasonable to restrict the band-limited sensitivity kernel within the range of
the Fresnel volume of the dominant frequency. That is, when the
background of the media is supposed to be homogeneous or smoothly heterogeneous, it is practical to replace the band -limited sensitivity kernel with several frequencies or even the single dominant one
for less computation when performing band-limited FVT. Certainly,
when the background is complex, it is necessary to perform bandlimited FVT with some discrete frequencies uniformly sampled in
the frequency band.
Third, we note that under the stationary-phase approximation,
Spetzler and Snieder 2001 prove the equivalence between FVT
and ray theory in homogeneous media. The higher the frequency, the
narrower the Fresnel volume. So when the frequency is infinite, the
Fresnel volume reduces to a ray with infinitely small width. In this
case, the Fresnel volume theory converges to the ray theory; hence,

a)

c)

b)

d)

Figure 3. a Two-dimensional amplitude, b 3D amplitude, c 2D traveltime, and d 3D traveltime sensitivity kernels excited by a point source
in homogeneous media. In a and c, the source is located at 200,250 m and the receiver is located at 800,250 m. In b and d, the central
slices of the 3D sensitivity kernels in the x-z-plane; the source is located at 200,250 m, and the receiver is located at 800,250 m.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Sensitivity kernels
FVT is equal to raypath tomography. Additionally, the integral of the
traveltime sensitivity kernel along a line perpendicular to the central
ray equals one refer to Appendix A. So for a single shot-receiver
pair, the FVT updates the points in the Fresnel volume with the sensitivity kernel as a weight function.
Fourth, we conclude that, commonly, the distribution ranges of
the Fresnel volumes for amplitude tomography are smaller than
those for traveltime tomography, and the distribution ranges of the
2D Fresnel volumes are smaller than those for 3D conditions. Furthermore, considering a cross section perpendicular to the central
ray, the amplitude sensitivity kernel has its maximum value on the
geometric ray and decreases off the ray. However, the traveltime
sensitivity kernel has its minimum value on the geometric ray and increases away from the raypath. After reaching a peak at a certain
point, it gradually decreases to zero at the boundaries of the Fresnel
volume.

U39

It is important to note that, for the 3D traveltime sensitivity kernel,


the values on the geometric ray are zero. Marquering et al. 1999
call this form of sensitivity kernel a banana-doughnut. Obviously,
there is a contradiction between ray theory and the finite-frequency
wave theory. According to finite-frequency wave theory, the weight
coefficients on the central ray are zero. However, tomography based
on ray theory updates the points on the central ray with an invariable
weight coefficient of one, and zero elsewhere. Jocker et al. 2006
test the validity of finite-frequency wave theory through physical experiments and explain the banana-doughnut as the interferential result of monochromatic waves. So ray theory is just a special case of
finite-frequency theory.
As mentioned, finite-frequency theory equals ray theory when the
frequency tends to infinity. However, the frequencies of seismic signals are finite. Does this mean traditional traveltime tomography
does not work? Actually, the points with large sensitivity generally

Figure 4. Vertical cross sections of the a 2D amplitude, b 3D amplitude, c 2D traveltime, and d 3D traveltime sensitivity kernels in homogeneous media shown in Figure 3, which are the functions of sensitivity values versus distance perpendicular to the geometric ray. The curves
with thin blue lines are monochromatic components starting from 0 to 100 Hz, with a step of 2 Hz. The thick red lines are the stacked results of
the monochromatic cross sections. The thick green lines are the cross sections of the 50-Hz dominant frequency.

a)

b)

Figure 5. a Two-dimensional amplitude and b 2D traveltime sensitivity kernels excited by a point source in slowness media with a constant
gradient. The source is located at 50,50 m; the receiver is located at 1450,50 m.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U40

Liu et al.

are close to the geometric ray. That is, the velocity-updated points for
traditional tomography are very close to the points whose velocity
should be updated. That is why we can still obtain the large-scale
structure of the media by traditional traveltime tomography. Obviously, it decreases the inversion resolution. This is also why we cannot resolve fine structures by traditional traveltime tomography.
Spetzler and Snieder 2004 try to reconcile the two theories through
the Fresnel volume, i.e., the ray theory is valid when the slowness
perturbation has a geometric size greater than the width of the
Fresnel volume. Otherwise, the finite-frequency wave theory should
be used to describe seismic wave propagation.
Sensitivity kernels are related to the positions of the source and receiver, the frequency band of the wavelet, source type, wave style,
and velocity distribution. Sensitivity kernels may be very complex
in some models. Multiple paths may exist in some conditions. Figure
7a and b shows band-limited sensitivity kernels in a smoothed Marmousi model. The sensitivity kernels with high values are distributed complicatedly between the source and receiver and are surrounded by a small part with low values. If we restrict the sensitivity kernels within the corresponding Fresnel volumes according to Table 2,
the blue parts and the surrounding parts with low values in Figure 7a
and b disappear and we see the multipath phenomenon Figure 7c
and d. Although the Fresnel volumes are complicated in the Marmousi model, we still find they roughly obey the principles described above.

NUMERICAL TESTS OF FVT


The sensitivity kernel is the key problem in seismic modeling and
inversion using Fresnel volume theory. Theoretically, the inverse result of FVT should be more accurate than that of raypath tomography because the Fresnel volume describes seismic wave propagation
more accurately than ray theory. To test the validity of FVT with sensitivity kernels, we apply 2D traveltime FVT and 2D traveltime raypath tomography to invert the near-surface velocity structure of a
theoretical model. Because of practical problems such as attenuation, noise, and wavelets, amplitude FVT is not practical yet. So only
traveltime kernel is actually used in this experiment.
The complex 2D theoretical model is shown in Figure 8a. It consists of 4001 75 grid points; the spatial interval is 10 10 m. The
velocities vary from 800 to 4300 m / s. The FVT of the numerical
experiment consists of six steps.

We simulate 640 shots shot interval of 40 m with staggeredgrid, high-order, finite-difference elastic wave modeling. The first
shot is located at the surface at 5000 m; 201 receivers are located
symmetrically at both sides of the excited point, with a trace interval
of 20 m. Thus, the maximum offset is 2000 m and the minimum offset is zero. Figure 9 shows the vertical components of two synthetic
records excited at the surface at 9 and 26 km. The first-break traveltime is picked on the synthetic records.

Step 2: Construct the initial model


To invert the near-surface velocity, an initial model is needed. We
specify a constant gradient model as the initial model. The velocities
are calculated according to the equation vz 500.0 3.0*z.

Step 3: Calculate the traveltime residual


Traveltime residual of the ray is usually similar to that of the
Fresnel volume in the near offset, so we use a traveltime residual calculated by kinematic ray tracing to reduce calculation costs.

Step 4: Calculate the kernels


Equation 4b is used to calculate the monochromatic traveltime
sensitivity kernel. Greens function and synthetic wavefields are calculated with a finite-difference operator of the full wave equation in
the frequency-space domain Jo et al., 1996. To reduce the calculation cost of FVT, only the dominant frequency of 50 Hz is used. As
an example, the monochromatic traveltime sensitivity kernel on the
circled part Figure 8b of the smoothed initial model Figure 8a is
shown in Figure 10b, which is restricted within the range of the
Fresnel volume of the dominant frequency based on Table 2. For
comparison, the amplitude sensitivity kernel is also shown Figure
10a, although it is not used.

Step 5: Construct the FVT equation


Substituting equation 4b into equation 1b, we obtain the original
monochromatic FVT equation 8a. In case of instability, we perform
FVT using equation 8b:

r V0r Im

G0g,ru0r,s
sr ds , 8a
u0g,s

b)
0.06
0.04
0.02
0
0.02
0.04

Sensitivity of 2D amplitude to
slowness perturbations (m 1 )

Sensitivity of 2D amplitude to
slowness perturbations (m 1 s 1 )

a)

Step 1: Pick first-arrival traveltimes

0.005

0.005

0.06

Figure 6. Vertical cross sections of the a 2D amplitude and b 2D traveltime sensitivity kernels shown in Figures 5, which are the functions of
sensitivity values versus depth. The curves with thin blue lines are monochromatic components starting from 0 to 100 Hz, with 2-Hz steps. The
thick red lines are the stacked results of the monochromatic cross sections. The thick green lines are the cross sections of the 50-Hz dominant frequency.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Sensitivity kernels

U41

However, FVT reveals the smooth background and the high-wavenumber perturbations more accurately.

r V0r ImG0g,ru0r,su0g,s sr ds
u0g,s2 .

8b

DISCUSSION

Here, ds is the size of the area element. Considering that the Fresnel
volume is much wider than the corresponding ray, the SIRT method
Liu and Dong, 2007 is used directly to solve tomographic equation
8b without regularization.

From the red curves in Figures 4 and 6, we see that the first side
lobe of the stacked cross section remains large in spite of canceling
the monochromatic sensitivity kernels as a result of destructive interference. This means the perturbation inside the first side lobe also
affects the wave propagation significantly. The sign of the sensitivity
kernel has different physical meanings. For example, according to
equation 1b, a positive value corresponds to a delay in traveltime because of increasing slowness; a negative value corresponds to an advance in traveltime becaue of increasing slowness. So both the main
lobe and the first side lobe should be used to define the band-limited
Fresnel volume.
Figures 4 and 6 also show that the extended band-limited Fresnel
volume has a good coincidence with the second Fresnel volume of
the dominant frequency. The distribution ranges of the second
Fresnel volumes of the dominant frequency are listed in Table 3.
However, for the experimental model Figure 8, FVT with this second Fresnel volume does not obtain a better tomographic result than

Step 6: Update model


After solving the tomographic equation, the model is updated.
Steps 36 are repeated until the stop criteria are satisfied.
The final inverted results of raypath tomography and FVT are
shown in Figure 11. Figure 12 shows the velocity sections of the theoretical model and the inverted results at different depths from
0 to 160 m beneath the surface, with a depth interval of 40 m. We
can see from Figures 11 and 12 that the result of FVT is much more
accurate than that of raypath tomography. Raypath tomography can
obtain only the smooth background variations of the media. It is insensitive to the structural variations with high spatial wavenumber.
Distance (km)

a)

Distance (km)
4

c)
Depth (km)

Depth (km)

5500

Velocity (m/s)

4511
3521

b)

d)
0

Depth (km)

Depth (km)

2532

Figure 7. The 2D band-limited a amplitude and b traveltime sensitivity kernels and restricted band-limited c amplitude and d traveltime
sensitivity kernels excited by a point source in a smoothed Marmousi model. The source is located at 1275,40 m, and the receiver is located at
5000,40 m.

a)
Velocity (m/s)

10

1630

0.25

0.5

Distance (km)

30

28

29

30

Depth (km)

Depth (km)

2920

20

5500
4210

b)

Distance (km)

0.25

0.5

340

Figure 8. a The 2D complex near-surface theoretical model with topography and b the zoomed view of the circled part of a after smoothing.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U42

Liu et al.
Fresnel volume of the dominant frequency. After all, the misfit of the
first side lobe is larger than that of the main lobe in homogeneous and
heterogeneous media Figures 4 and 6. The integral of the traveltime kernel over a line perpendicular to the ray is equal to one Appendix A, which means the positive region has a stronger effect than
the negative region. So, this misfit is especially large for the traveltime sensitivity kernels.
For the amplitude kernels, the misfits are less. According to the
derivation in Appendix A, the integral of the amplitude kernel over a
line perpendicular to the ray is equal to zero, which means the posi-

FVT using the first Fresnel volume. This failure may be a result of replacing traveltime delays of the Fresnel volumes with those of the
rays.
In numerical experiments to compare the traveltime delays of the
rays and Fresnel volumes, we find they are similar in near offsets but
have larger differences in far offsets. The traveltime delays of the
rays and the first Fresnel volumes are similar, but the delays of the
rays and the second Fresnel volumes have large differences. Maybe
the traveltime delays of the rays cause the failure of FVT using the
second Fresnel volume. On the other hand, this failure might be attributed to replacing the band-limited Fresnel volume by the second

Position (km)

Time (ms)

Position (km)

Figure 9. Vertical components of two synthetic records excited at the surface at 9 and 26 km.
Distance (km)

a)
28

Velocity (m/s)

28

1630

29

30

Depth (km)

2920

Depth (km)

4210

30

5500

Distance (km)

b)

29

0.25

0.5

0.25

0.5

340

Figure 10. a Amplitude and b traveltime monochromatic sensitivity kernels of 50 Hz between the source located at the surface at 28 km and
the receiver located at the surface at 30 km.

a)

1630

0.25

0.5

Distance (km)

30

10

20

Depth (km)

Depth (km)

2920

20

5500
4210

b)

Distance (km)
10

Velocity (m/s)

0.25

0.5

340

Figure 11. Tomographic results of a raypath tomography and b FVT based on the picked traveltime of the first arrivals.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

30

Sensitivity kernels

a)

U43

Figure 12. Velocity sections of the theoretical model red, the inverted results of raypath tomography
green, and FVT blue in different depths from
0 to 160 m beneath the surface.

b)

c)

d)

e)

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U44

Liu et al.

Table 3. Distribution ranges of band-limited Fresnel volumes


in different conditions.

Dimension
2D
3D

Amplitude
Fresnel
volume

Traveltime
Fresnel
volume

tmax T / 4
tmax 2T / 4

tmax 3T / 4
tmax 4T / 4

tive and negative regions have the same important effect on the observation. So, as for the amplitude FVT, the second Fresnel volume
of the dominant frequency should be considered.

program of the University of the Education Ministry of China grant


NCET-05-0384, Chinas Ongoing National Keystone Basic Research program grant 2006CB202402, and the State Key Library
of Marine Geology of China project grant MG20080205. Most of
the figures were created using the Seismic Unix software package
from Colorado School of Mines. We are grateful to Hua-Zhong
Wang, Jiu-Bing Cheng, Kai Yang, Mei Xue, and Chun-Feng Li from
Tongji University for their constructive comments on the manuscript. We are also grateful to Bill Harlan for arranging reviews of
this paper. Jesper Spetzlers, Moritz Fliedners, and Reinaldo
Michelenas detailed reviews and comments greatly improved the
manuscript. Meanwhile, we thank the other two anonymous reviewers for comments.

CONCLUSIONS

APPENDIX A

Sensitivity kernels are important for seismic modeling and inversion with Fresnel volume theory. On the basis of Born and Rytov approximations, we directly express the amplitude and traveltime sensitivity kernels for 2D and 3D FVT with Greens functions. These
equations are suitable for homogeneous and heterogeneous media.
The final tomographic experiment verifies the validity of the sensitivity kernels we propose. Our results show that FVT using these
sensitivity kernels can achieve much more accurate inversion than
conventional raypath tomography.
Through the analytical expressions of the sensitivity kernels listed
in Table 1, we summarize their properties. Although they are for homogenous media, the numerical experiments show they also can be
used approximately in heterogeneous media Figures 27.
According to constructive interference of monochromatic waves,
we find that the spatial distributions of the sensitivity kernels in
monochromatic FVT vary widely for different conditions. They are
1 / 8, 2 / 8, 3 / 8, and 4 / 8 periods, respectively, for 2D amplitude, 3D
amplitude, 2D traveltime, and 3D traveltime conditions.
The amplitude sensitivity kernel has its maximum value on the
geometric ray and decreases off the ray. The traveltime sensitivity
kernel has its minimum value on the geometric ray and increases off
the ray.After reaching a peak at a certain point, it gradually decreases
to zero at the Fresnel volume boundary.
The band-limited sensitivity kernel can be calculated by stacking
the monochromatic kernels with a weight function similar to the amplitude spectrum of the wavelet. Under two preconditions, only a
few selected frequencies or even a single dominant frequency may
be necessary to perform FVT. First, the sensitivity kernels must be
restricted within the first Fresnel volume of the dominant frequency.
Second, the background media must be homogeneous or smoothly
heterogeneous. This replacement is a preliminary substitute. When
the media are complex or calculation cost is not a problem, a restricted band-limited kernel is preferable.
Another trade-off performed to make the FVT application practical is that, in near offset, the traveltime residuals of the rays replace
those of the Fresnel volumes. This approximation is based on numerical experiments that are not presented in this paper. For higher accuracy, the traveltime residuals should be estimated through crosscorrelation or dynamic ray tracing.

RELATIONSHIP BETWEEN FRESNEL


VOLUME AND RAY

ACKNOWLEDGMENTS
This study is supported by the National Natural Science Foundation of China grant 40804023, the New Century Excellent Talents

Spetzler and Snieder 2001 prove that in homogeneous media,


the Fresnel volume theory converges to ray theory, and FVT is equal
to raypath tomography when the frequency tends to infinity. In the
same way, we give a further relationship between the two theories.
The integral of the 2D monochromatic traveltime sensitivity kernel Table 1 over a line perpendicular to the ray can be written as
Figure 1:

KT2Dx,z, dz

2 Vrsrg

2 Vx2 z2L x2 z2

sg

sin t

dz
4

sin

x2 z2 L x2 z2 L

dz.
V
4
A-1

Here, L is the distance between source and receiver, which equals sg


as Figure 1 shows. We assume that z / x1, z / L x1; we set
x2 z2, L x2 z2 to the first-order Taylor approximation and
1 / x2 z2, 1 / L x2 z2 to the zeroth-order Taylor approximation to get equation A-2:

KT2Dx,z, dz

sin

L
2 VxL x

L z2

dz 1.
2xL xV
4
A-2

The above characteristic is independent of frequency, which

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

Sensitivity kernels
means the integral of the 2D band-limited traveltime sensitivity kernel equation 5b over a line perpendicular to the central ray is one.
In the same way, we can get the integral of the 3D traveltime sensitivity kernel over a plane perpendicular to the ray. It is not difficult to
verify that the integral also equals one.
The integral of the 2D monochromatic amplitude sensitivity kernel Table 1 over a line perpendicular to the ray can be written as
equation A-3 Figure 1:

KA2Dx,z, dz

A0 ,gs

cos t

sg 3
2 Vrsrg

dz
4

L 3

2 Vx2 z2L x2 z2

cos

x2 z2 L x2 z2 L

dz.
V
4
A-3

Under the same approximations as above, we get equation A-4:

KA2Dx,z, dz

L 3
L z2

cos

dz 0
2 VxL x
2xL xV
4
A-4

The above characteristic is also independent of frequency, which


means the integral of the 2D band-limited amplitude sensitivity kernel equation 5a over a line perpendicular to the central ray is zero.
In the same way, we can get the integral of the 3D amplitude sensitivity kernel over a plane perpendicular to the ray. It is not difficult to
verify that the integral also equals zero.

REFERENCES
Aki, K., and P. G. Richards, 2002, Quantitative seismology, 2nd ed.: University Science Books.
Billette, F., and G. Lambar, 1998, Velocity macro-model estimation from
seismic reflection data by stereo-tomography: Geophysical Journal International, 135, 671680.
erven, V., and J. E. P. Soares, 1992, Fresnel volume ray tracing: Geophysics, 57, 902915.
Dahlen, F. A., 2005, Finite-frequency sensitivity kernels for boundary topography perturbations: Geophysical Journal International, 162, 525540.
Dahlen, F. A., S. H. Hung, and G. Nolet, 2000, Frchet kernels for finite-frequency traveltimes I. Theory: Geophysical Journal International, 141,
157174.

U45

Devaney, A. J., 1984, Geophysical diffraction tomography: IEEE Transactions on Geoscience and Remote Sensing, GE-22, 313.
Dziewonski, A., 1984, Mapping the lower mantle: Determination of lateral
heterogeneity in P-velocity up to degree and order 6: Journal of Geophysical Research, 89, 59295952.
Harris, J. M., R. Nolen-Hoeksema, J. W. Rector III, M. V. Schaack, and S. K.
Lazaratos, 1992, High-resolution cross-well imaging of a West Texas carbonate reservoir, Part 1: Data acquisition and project overview: 63rd Annual International Meeting, SEG, Expanded Abstracts, 3539.
Hung, S. H., F. A. Dahlen, and G. Nolet, 2001, Wavefront healing: A bananadoughnut perspective: Geophysical Journal International, 146, 289312.
Jo, C. H., C. S. Shin, and J. H. Suh, 1996, An optimal 9-point, finite-difference, frequency-space, 2-D scalar wave extrapolator: Geophysics, 61,
529537.
Jocker, J., J. Spetzler, D. Smeulders, and J. Trampert, 2006, Validation of
first-order diffraction theory for the traveltimes and amplitudes of propagating waves: Geophysics, 71, 167177.
Liu, Y. Z., and L. G. Dong, 2007, Analysis of influence factor of first-breaks
tomography in Chinese: Oil Geophysical Prospecting, 42, 544553.
, 2008, Sensitivity kernels and Fresnel volumes for transmitted waves:
78th Annual International Meeting, SEG, Expanded Abstracts, 3234
3238.
Marquering, H., F. A. Dahlen, and G. Nolet, 1999, Three-dimensional sensitivity kernels for finite-frequency traveltimes: The banana-doughnut paradox: Geophysical Journal International, 137, 805815.
Marquering, H., G. Nolet, and F. A. Dahlen, 1998, Three-dimensional waveform sensitivity kernels: Geophysical Journal International, 132,
521534.
Michelena, R. J., and J. M. Harris, 1991, Tomographic traveltime inversion
using natural pixels: Geophysics, 56, 635644.
Nolet, G., 1987, Seismic wave propagation and seismic tomography, in G.
Nolet, ed., Seismic tomography with applications in global seismology
and exploration geophysics: D. Reidel Publishing, 124.
Pratt, R. G., and N. R. Goulty, 1991, Combining wave-equation imaging with
traveltime tomography to form high-resolution images from crosshole
data: Geophysics, 56, 208224.
Pulliam, R. J., D. W. Vasco, and L. R. Johnson, 1993, Tomographic inversions for mantle P-wave velocity structure based on the minimization and
norms of International Seismological Centre traveltime residuals: Journal
of Geophysical Research, 98, 699734.
Schuster, G. T., and A. Quintus-Bosz, 1993, Wavepath eikonal traveltime inversion: Theory: Geophysics, 58, 13141323.
Sheriff, R. E., and L. P. Geldart, 1982, Exploration seismology: Cambridge
University Press.
Slaney, M., A. C. Kak, and L. Larsen, 1984, Limitations of imaging with
first-order diffraction tomography: IEEE Transactions on Microwave
Theory and Techniques, MTT-32, 860873.
Snieder, R., and A. Lomax, 1996, Wavefield smoothing and the effect of
rough velocity perturbations on arrival times and amplitudes: Geophysical
Journal International, 125, 796812.
Spetzler, J., D. Sijacic, and K. H. Wolf, 2007, Application of a linear finitefrequency theory to time-lapse crosswell tomography in ultrasonic and numerical experiments: Geophysics, 72, no. 6, O19O27.
Spetzler, G., and R. Snieder, 2001, The effect of small-scale heterogeneity
on the arrival time of waves: Geophysical Journal International, 145, 786
796.
, 2004, The Fresnel volume and transmitted waves: Geophysics, 69,
653663.
Spetzler, J., Z. Xue, H. Saito, and O. Nishizawa, 2008, Case story: Timelapse seismic crosswell monitoring of CO2 injected in an onshore sandstone aquifer: Geophysical Journal International, 172, 214225.
Stockwell, Jr., J. W., 1997, Free software in education: A case study of CWP/
SU: Seismic Un*x: The Leading Edge, 16, 10541059.
Tarantola, A., 1987, Inverse problem theory: Elsevier Science Publ. Co., Inc.
Tian, Y., R. Montelli, G. Nolet, and F. A. Dahlen, 2007, Computing traveltime and amplitude sensitivity kernels in finite-frequency tomography:
Journal of Computational Physics, 226, 22712288.
Vasco, D. W., and E. L. Majer, 1993, Wavepath traveltime tomography: Geophysical Journal International, 115, 10551069.
Watanabe, T., T. Matsuoka, and Y. Ashida, 1999, Seismic traveltime tomography using Fresnel volume approach: 69th Annual International Meeting,
SEG, Expanded Abstracts, 14021405.
Woodward, M. J., 1992, Wave-equation tomography: Geophysics, 57,
1526.
Wu, R. S., and M. N. Toksz, 1987, Diffraction tomography and multisource
holography applied to seismic imaging: Geophysics, 52, 1125.
Xu, S., Y. Zhang, and T. Huang, 2006, Enhanced tomography resolution by a
fat ray technique: 76th Annual International Meeting, SEG, Expanded Abstracts, 33543357.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

U46

Liu et al.

Yomogida, K., 1992, Fresnel zone inversion for lateral heterogeneities in the
earth: PAGEOPH, 138, 391406.
Zhang, Z. G., Y. Shen, and L. Zhao, 2007, Finite-frequency sensitivity kernels for head waves: Geophysical Journal International, 171, 847856.
Zhao, L., and T. H. Jordan, 2006, Structural sensitivities of finite-frequency

seismic waves: A full-wave approach: Geophysical Journal International,


165, 981990.
Zhao, L., T. H. Jordan, and C. H. Chapman, 2000, Three-dimensional Frchet
differential kernels for seismic delay times: Geophysical Journal International, 141, 558576.

Downloaded 24 Mar 2010 to 222.66.175.192. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

You might also like