Sei sulla pagina 1di 24

1496_Ch08 10/10/08 5:44 PM Page 193

Chapter 8
Synthetics

No rock so hard but that a little wave may beat admission in a thousand years.
— Alfred, Lord Tennyson

Introduction
We assume that a seismic trace has been corrected for amplitude decay resulting from
spherical spreading over the seismic time scale of interest (say, for example, from 0 to 6 s).
However, in reality, other effects also must be considered. One such effect is inelastic
absorption — the loss of seismic energy to frictionally generated heat. (We will treat inelas-
tic absorption in Chapter 14.) We must consider the effect of the seismic energy’s source.
In addition, effects result from the instrumentation; source and instrument effects are man-
made at or near the surface of the ground. We lump these surface effects together in the
form of a source wavelet, which we denote by s.
However, the most important effect is that of the deep earth itself. The deep earth is
represented by the sequence ε of reflection coefficients. This sequence is called the reflec-
tivity (Robinson, 1957; Ulrych, 1999). In an ideal seismic experiment, an impulsive source
(i.e., a sharp spike of energy imposed at time 0) produces the impulse response of the earth,
which we denote by h. In the elastic range, the earth is a linear system, so we can write the
synthetic trace (also called the synthetic seismogram) x as the convolution

x = s ∗ h. (1)

Equation 1 is called the dynamic convolutional model (Robinson, 1999). The model is
called dynamic because the reflection impulse response h is a highly nonlinear function of
the reflection coefficients. To deconvolve the dynamic convolutional model, dynamic
deconvolution must be used (Robinson, 1975, 1999).
The basic problem that we confront in this chapter is the discovery of adequate and
useful approximations of the impulse response h in equation 1. We arrive at two approxi-
mations, which we call 1 and 2. This chapter will give the details. Approximation number
1, which holds in the case of small reflection coefficients, says that h can be approximated
by m * ε, where m is a wavelet representing the multiple reflections and ε is the reflecti-
vity. Under this approximation, equation 1 reduces to the synthetic seismogram with mul-
tiples, as given by
x = s ∗ m ∗ ε. (2)

193
1496_Ch08 10/10/08 5:44 PM Page 194

194 Digital Imaging and Deconvolution

Approximation number 2, which holds in the case of small and white reflection
coefficients, says that h can be approximated by the reflectivity ε. Under this approximation,
equation 1 reduces to the synthetic seismogram without multiples, as given by

x = s ∗ ε. (3)

By white, we mean that the reflectivity is made up of independent random variables all
with the same distribution. In other words, by white, we mean white noise. Equation 3 rep-
resents the classic synthetic seismogram that has been in constant use for the past 50 years.

Polarity
A solid consists of many small particles. Elastic waves transfer energy from one point
to another by the movements of those small particles. The particles themselves are not
transported; they only vibrate with small displacements about their respective equilibrium
(or rest) positions. Their vibrational movements can propagate energy in the form of a
wave. A wave is an energy-transport phenomenon that transports energy through a medium
without transporting matter. Any particular small particle simply vibrates about its own
rest position while the wave is passing, and it returns to its rest position and remains there
after the wave has passed. The initial energy that created the wave spreads out with the
wave. An electromagnetic wave is nature’s way of transporting energy from one place to
another at the speed of light.
A homogeneous, isotropic rock has two critical physical parameters: its density  and
its wave velocity v. The product of these two parameters gives the acoustic impedance or
characteristic impedance Z; that is, the acoustic impedance is Z  v. Next, we shall learn
how the reflection coefficient and the transmission coefficient at an interface depend on the
respective impedances on each side of the interface.
Most seismic interpretation is done in terms of models involving the propagation of P-
waves (also called compressional waves) through a sequence of rock formations. In many
cases, but by no means in all cases, it is appropriate as a first approximation to assume that
each formation is homogeneous and isotropic.
A seismic wave represents an interchange between kinetic and potential energy. The
kinetic energy comes from the physical motion of the particles, and the potential energy
comes from their relative positions with respect to the interparticle elastic (or restoring)
forces. The rest position of each particle can be taken as the origin of a 3D coordinate
system. The particle vibrates around this rest position. At any instant, we can measure in
principle the particle’s velocity and, if we wish, the particle’s acceleration. We also can
measure the degree of compression of the particles in the form of a pressure or stress.
In the case of a homogeneous isotropic solid, physical theory tells us that we need only
two quantities for a complete specification of the wave motion resulting from the particle
motion. In seismic work, it is convenient to take the particle velocity and the pressure as
these two quantities. Particle velocity is a vector in a 3D coordinate system, whereas pres-
sure is a scalar. To simplify, we shall restrict ourselves to P-waves traveling in the vertical
direction, so particle velocity becomes a scalar too.
1496_Ch08 10/10/08 5:44 PM Page 195

Chapter 8: Synthetics 195

A seismic trace is a graph of amplitude versus time. Each trace is equal to the sum of
the downgoing wave motion plus the upcoming wave motion at the sensor. In marine work,
the hydrophone measures pressure, so the amplitude of a marine seismic trace indicates
pressure. In land work, the geophone measures particle velocity, so the amplitude of a land
seismic trace represents particle velocity. In a given homogeneous isotropic medium, let V
(which we simply call the particle-velocity disturbance) denote the solution of the wave
equation for particle velocity. Let p (which we simply call the pressure disturbance)
denote the solution of the wave equation for pressure. Let D denote the downgoing com-
ponent of the particle-velocity disturbance, and let U denote the upgoing component of the
particle-velocity disturbance. Similarly, let d denote the downgoing component of the pres-
sure disturbance, and let u denote the upgoing component of the pressure disturbance.
Thus, we have the two equations V  D  U and p  d  u for particle velocity and for
pressure, respectively.
Various conventions are used for pressure waves and particle-velocity waves. Let us
use the Berkhout convention (Berkhout, 1987). Berkhout’s first equation, d  ZD, says
that the downgoing pressure wave has the same polarity as does the downgoing particle-
velocity wave, and it says that the two are related by a scale factor given by the acoustic
impedance Z. His second equation, u ZU, says that the upgoing pressure wave has the
opposite polarity to that of the upgoing particle-velocity wave, and it says that both also
are related by the same scale factor.

Reflection coefficients and transmission coefficients


We will deal with the 1D case, so we treat only variations along the depth axis, which
points straight down into the earth. As we saw in Chapter 3, a plane wave is a wave whose
wavefronts are infinite parallel planes normal to the direction of travel. Suppose a plane
P-wave travels vertically downward in a homogeneous isotropic medium. This downgoing
incident wave encounters a horizontal interface separating two media. The upper medium
has density 1 and seismic wave velocity v1, and the lower medium has density 2 and
seismic wave velocity v2. The result is that a portion of the wave energy will be reflected
at the interface, and the remainder will be transmitted.
This splitting of the incident wave into a reflected wave and a transmitted wave at the
interface is caused by the abrupt change in rock density and/or velocity. For the normal-
incidence case that we treat here, the reflected and transmitted waves have the same shape
and breadth as did the incident wave, but they differ from it in amplitude. The ratio of the
amplitude of the reflected wave to that of the incident wave is termed the reflection coeffi-
cient. Similarly, the ratio of the amplitude of the transmitted wave to that of the incident
wave is called the transmission coefficient. However, the reflection coefficient defined for a
wave in which the amplitude is measured in terms of particle velocity is different from the
reflection coefficient for a wave in which amplitude is measured in terms of pressure. The
same statement holds for the transmission coefficient. Let us now establish this difference.
A reflector is characterized by a contrast in acoustic impedance, which gives rise to
a seismic reflection (O’Doherty and Anstey, 1971). Reflectivity refers to the reflection coef-
ficient. The amplitudes that are required in the definitions of reflection and transmission
1496_Ch08 10/10/08 5:44 PM Page 199

Chapter 8: Synthetics 199

ghost spike. Figure 4a shows the a)


downgoing direct-wave/ghost- Direct pulse
0.4
wave complex with amplitude in
terms of pressure. Figure 4b 0.3 The case of pressure
shows the corresponding complex
in terms of particle velocity. In 0.2
these figures, the spike series have
been convolved with an arbitrary
0.1
waveshape.
Suppose that a receiver that is
0.0
sensitive to particle velocity is
Time
placed on the water bottom.
Would such a receiver be effec- –0.1
tive? For a rigid sea bottom (i.e.,
Z2   for the layer below sea –0.2 Ghost pulse
bottom), the reflection coeffi-
cients are
b)
0.2 Ghost pulse
Z1 − Z 2
ε pv = = −1 and
Z1 + Z 2 0.1
Z − Z1
ε pres = 2 = 1, (11)
Z1 + Z 2 0.0
Time
and –0.1
τ pv = 1 + ε pv = 1 − 1 = 0 and
τ pres = 1 + ε pres = 1 + 1 = 2. (12) –0.2
The case of
particle velocity
The coefficient pv  0 –0.3
shows that when a downgoing
wave strikes a hard bottom, the –0.4 Direct pulse
particle velocity cannot be trans-
mitted. Because the particle Figure 4. (a) The ghost complex for pressure. (b) The
velocity is continuous across the ghost complex for particle velocity.
interface, it follows that the parti-
cle velocity at the bottom must be
zero. Thus, a receiver that is sensitive to particle velocity and is dragging on the bottom
from a cable would lose its effectiveness for a rigid sea bottom.
Let us discuss a hydrophone at the water surface. Why must the hydrophone (which is
sensitive to pressure) be kept below the surface of the water? Consider the water surface
as a free surface (Z1  0 for air). The upgoing reflection coefficients are
Z 2  Z1 Z  Z2
εU , pv = = 1 and εU ,pres = 1 = −1. (13)
Z1 + Z 2 Z1 + Z 2
1496_Ch08 10/10/08 5:44 PM Page 200

200 Digital Imaging and Deconvolution

The upgoing transmission coefficients for particle velocity and pressure are

τ U , pv = 1 + εU , pv = 1 + 1 = 2 and τ U .pres = 1 + εU ,pres = 1 − 1 = 0. (14)

We see from the above two relations that the particle velocity at the surface is twice the
particle velocity of the upgoing incident wave, whereas the pressure at the surface is zero.
Thus, the pressure-sensitive hydrophone must be kept below the surface of the water
because its effectiveness diminishes as the streamer floats to the surface.
What are the Knott-Zoeppritz equations? Thus far, we have limited ourselves to
normal incidence. In the more general case of a plane wave that is incident at an arbitrary
angle, reflected P- and S-waves and transmitted P- and S-waves will be generated. The
amplitudes of all these waves can be found from the Knott-Zoeppritz equations (Aki and
Richards, 2002; Sheriff, 2002). These equations are basic to amplitude-variation-with-
offset (AVO) studies.

Layer-cake model
What is a reflectivity function? A reflectivity function is a time series intended to
represent reflecting interfaces by their reflection coefficients, which usually are defined for
the normal-incidence case. A reflectivity section is a display of the reflectivity functions
versus their locations on a seismic line.
A multiple reflection represents seismic energy that has been reflected more than once.
A primary reflection represents energy that has been reflected only once and hence is not
a multiple. All recorded seismic energy involves multiples.
An important distinction is between long-path and short-path multiples. A long-path
multiple arrives as a distinct event, whereas a short-path multiple arrives so soon after the pri-
mary that it merely adds a tail to the primary. Short-path multiples can obscure stratigraphic
detail even when structural aspects are not affected significantly. The moveout behavior of
long-path multiples might not be representative of the portion of the section associated with
their arrival times. Usually, long-path multiples have traveled more in the slower (shallower)
part of the section than have the primaries with the same normal-incidence arrival times, so
long-path multiples generally show greater normal moveout and thus can be attenuated by
common-midpoint (CMP) stacking.
The main purpose of a seismic signal model is to explain the seismic-wave propaga-
tion phenomenon. The most valuable models are three dimensional. Such models tend to
be numerical because the mathematics of a theoretical 3D model is much too involved to
produce closed-form solutions except in the simplest cases. The most pronounced varia-
tions in the earth layering are usually along the vertical direction, so a 1D vertical model
often is adequate. The foremost 1D model — the so-called stratified, or layered-earth, or
layer-cake model — is mathematically identical to the lattice model for electric transmis-
sion lines. The model is also mathematically identical both to the acoustic tube model used
in speech processing and to the thin-film model used in optics. In such a 1D model, the
earth is sliced mathematically into many thin horizontal layers that are normal to the
vertical z direction. Such a theoretical division of the earth into thin layers produces a
1496_Ch08 10/10/08 5:44 PM Page 201

Chapter 8: Synthetics 201

stratified medium characterized by the interfaces between the layers. Thus, a vector of the
discrete amplitudes of the signal thus can represent a digital signal. One vector results if
the amplitude of the signal is measured by a geophone in terms of particle velocity. Another
vector results if the amplitude of the signal is measured by a hydrophone in terms of pres-
sure. The inner (or dot) product of these two vectors gives the energy of the signal. In the
basic model treated here, there is no allowance for dissipation of kinetic energy into heat.
Thus, all the source energy imparted into the body can be accounted for, over time, in
terms of the resulting elastic wave motion.
A synthetic seismogram is an artificial seismic-reflection record made by assuming
that a wavelet travels through an assumed earth model (Anstey, 1960; Kelly et al., 1976;
Shtivelman and Loewenthal, 1989). A 1D synthetic seismogram (without multiple reflec-
tions) can be obtained by convolving a wavelet with a reflectivity function. A reflectivity
function consists of a series of spikes that indicates the sign and magnitude of the reflec-
tion coefficient of each interface. The reflectivity usually is calculated for normal-incidence
wave motion on the basis of changes in velocity and in density. Often, only velocity changes
are considered because density information frequently is unavailable. Alternatively, some
empirical relationship between density and velocity can be assumed.
The interfaces usually are identified by their normal-incidence two-way traveltime
from source to receiver and back to the source. The wavelet sometimes is an assumed
waveform such as a Ricker wavelet and sometimes is a waveform resulting from analysis
of actual seismic data. The most common type of synthetic seismogram involves primary
reflections only. Other synthetics include the contribution of short-path multiples in addi-
tion to the primaries. Even more sophisticated synthetic seismograms include only certain
selected multiples or all possible multiples. Often, a synthetic seismogram includes
various earth-filtering effects (such as attenuation resulting from geometric divergence and
attenuation resulting from absorption) as well as distortions from instrumental filtering
effects.
What are the limitations of a 1D synthetic seismogram? A 1D synthetic seismogram is
the result of a single-channel convolution of a wavelet with a computed impulse response
of a layer-cake medium. This impulse response can include either primary reflections only
or multiples as well. Only vertical travel paths are used in this model. However, the layer-
cake model can be varied in a lateral direction, and successive 1D synthetic traces can be
displayed side by side to simulate a seismic section. Such a synthetic seismic section can
be compared with an actual seismic section to help identify events and to see how varia-
tions in the model might appear on the synthetic section. Usually, the models assume that
the source and receiver are coincident, and sometimes offset-dependent effects can be
included. Some models include head waves, surface waves, and other wave modes.
Calculation of a synthetic seismogram is a direct or forward problem, as opposed to the
inverse problem, which produces estimates of the seismic wavelet along with the reflec-
tion coefficients characterizing layer interfaces and their depths in terms of their two-way
traveltimes.
Why should one study the layer-cake model at all? Certainly, stratigraphic traps can
occur in flat-lying strata, which fit the layer-cake model. In addition, many great oil fields
are found in huge, gently sloping anticlines, where the layer-cake approximation serves
1496_Ch08 10/10/08 5:44 PM Page 202

202 Digital Imaging and Deconvolution

quite well. However, exploration cannot be restricted to the use of such a simple model. It
is essential that geophysicists have the tools to model regions of highly complex geologic
structure. Reflections come from interfaces, and the observed reflections are used to reveal
structure.
The layer-cake model is the simplest expression of a sequence of interfaces. The cor-
responding 1D synthetic seismogram without multiples (equation 3), based on (1) a fixed
source wavelet and (2) only primary reflections is the simplest expression of the linear
time-invariant convolutional model. Such a relationship between geologic model and
synthetic seismogram is so basic that it cannot be disregarded. We cannot stop at this point,
however. We also must understand the corresponding 1D synthetic seismogram with
multiples (equation 2), which is based on (1) a fixed-source wavelet, (2) primary reflec-
tions, and (3) multiple reflections.
The synthetic seismogram can be represented more exactly by the dynamic convolu-
tional model (equation 1). Again, such a relationship between geologic model and syn-
thetic seismogram is so basic that it cannot be disregarded. In this respect, we owe a debt
of gratitude to the layer-cake model for being amenable to such an elegant mathematical
solution. From a learning point of view, the layer-cake model serves as an indispensable
pathway to the discovery of the intricacies of multiple reflections by mathematics and not
just by computations.
As soon as we start to ease some of the restrictions present in the layer-cake case, the
simplicity rapidly dissipates. A 2D synthetic seismogram can simulate such effects as
reflections from dipping reflectors and diffractions from sharp discontinuities. However, a
2D model has a shortcoming in that it cannot adequately handle energy coming from out-
side the plane of the model. A 3D model is required to take into account all such effects.
Attempts to find closed mathematical expressions that model the wave motion in 3D com-
plex geologic structures soon encounter insurmountable difficulties, so numerical methods
on computers must be used instead. The ingenuity displayed in geophysics in the construc-
tion of computer-based models is indeed impressive (Hilterman et al., 1998).
Geophysics has become even more exciting as it has evolved. By the 1970s, geophysi-
cists confirmed that seismic data could reveal information not only about geologic struc-
ture but also about lithology. By using such things as wave patterns, frequency content, and
strengths as well as reflection continuity and terminations, geophysicists could find impor-
tant clues about rock types and depositional environments. Thus, the disciplines known as
seismic stratigraphy (Vail, 1977) and sequence stratigraphy (Van Wagoner et al., 1988)
were born. Geology and geophysics had found common ground on which exploration
could and did achieve new heights (Sangree and Widmier, 1979; Mallick, 2007).

Synthetic seismogram without multiples


The layer-cake model gives us a means for discovering firsthand why deconvolution
works. Let us introduce an analogy. A trace with primaries and all multiples corresponds
to whole milk. A deconvolved trace, which we hope has primaries only, corresponds to
skim milk. In other words, the multiples correspond to the cream, which is bad because it
contains cholesterol. The purpose of deconvolution is to remove the cream from the whole
1496_Ch08 10/10/08 5:44 PM Page 203

Chapter 8: Synthetics 203

milk or, in other words, to remove the multiples from the whole trace (i.e., the trace with
primaries and multiples). The layer-cake model gives us an opportunity to show that the
deconvolution operation indeed can be done.
The time unit on a seismic trace is the time spacing t of the sampling — for example,
one time unit could be equal to 4 ms. This time unit sets the resolution of the seismic model
in terms of the layering. In other words, the thinnest layer that can be distinguished has a
two-way traveltime of 4 ms. Thus, the thickness of each layer is chosen so that the two-
way traveltime (downward traveltime plus upward traveltime) in each layer is equal to one
time unit. Each interface separating two adjacent layers with different impedances has a
nonzero reflection coefficient. Of course, several thin layers can be lumped together to
form a thicker uniform layer by setting the intermediate reflection coefficients equal to
zero. The greater the impedance contrast is between the two adjacent layers, the greater is
the magnitude of the reflection coefficient. For computation, the wave motion is digitized
so that a signal becomes a discrete sequence (that is, a time series with discrete values
separated by the given unit time interval).
In the layered or stratified model, the boundary layers are the air (on the top) and the
basement rock (on the bottom). Let N  1 be the number of interfaces, with interface 0 the
ground surface and interface N the deepest interface.
A plane wave is the simplest form of a propagating wave. We consider plane waves
traveling normal to the interfaces — that is, waves traveling up and down in the vertical z
direction. A pulse that is normally incident on interface i is split into a reflected pulse and
a transmitted pulse. Energy is conserved. As a consequence, the magnitude of the reflec-
tion coefficient ci must be less than or equal to one.
The magnitude of the reflection coefficient does not depend on the direction in
which the plane wave travels through the interface, but the sign of the reflection coeffi-
cient does. For a given interface, the reflection coefficient for an upgoing pulse is the
negative of the reflection coefficient for a downgoing pulse. As was pointed out above,
the sequence ε  {ε0, ε1, . . ., εN} of the downgoing reflection coefficients is called the
reflectivity function or simply the reflectivity. The reflectivity represents the internal
structure of the earth and is an unknown quantity in the remote-detection problem faced
in seismic prospecting.
We suppose that the subsurface consists of many ideal layers, each of which has a two-
way traveltime t. Let the top interface (the surface of the ground or water, as the case may
be) be called interface 0, let the next interface down be interface 1, and so on, to the bottom
interface N. Denote the reflection coefficient of interface n by εn. In practice, many of these
interfaces will be nonexistent. Any nonexistent interface has the reflection coefficient 0.
Normal incidence is assumed, so the source and receiver coincide. This source-receiver
point is chosen as a point just below the surface, so it is just barely inside the top layer
(Figure 5).
An impulsive source is activated that emits a one-way downgoing spike at time 0.
The receiver is a one-way receiver; that is, the receiver picks up all of the upgoing waves
and none of the downgoing waves. The approximate impulse response, consisting of
primary reflections only (without transmission losses), is given by the sequence of reflec-
tion coefficients (i.e., the reflectivity function ε). In Figure 5, for clarity, the raypaths have
1496_Ch08 10/10/08 5:44 PM Page 204

204 Digital Imaging and Deconvolution

Surface
(or Shot S (just Receiver R (just below surface)
interface 0) below surface) e1 e2 eN
e0

Interface 1
e1

Interface 2
e2

Interface N
eN
Basement

Figure 5. 1D layered model made up of primary reflections only. All the wave directions are
vertical, but they are depicted here as slanting lines for visual clarity.

been drawn as slanting lines, although in our normal-incidence model they are perpendi-
cular to the interfaces.
The synthetic seismic trace consists of the primary reflections from the underground
interfaces. Because the source point is below the surface of the ground, the reflection
coefficient ε0 of the surface does not generate a primary reflection. The approximate
impulse response (i.e., the reflectivity without the ε0) has the Z-transform

E ( Z ) = ε1 Z + ε 2 Z 2 + ε 3 Z 3 + ... + ε N Z N . (15)

This is the Z-transform of an Nth-order causal feedforward filter with the subsurface
reflection coefficients on the feedforward loops. For example, the filter in the case of just three
primary reflections is depicted by the pure feedforward block diagram shown in Figure 6.
Instead of a pure impulse, an arbitrary source wavelet or signature s can be used as the
input wavelet to the synthetic seismogram generator, whose output is the synthetic trace x.
The synthetic trace without multiples is the convolution of this signature with the reflec-
tivity (without the ε0 term). The result is x  s * ε, which we recognize as the multiple-
free convolutional model (equation 3) with which geophysicists are so familiar. From
downhole surveys, we can obtain the (approximate) reflection coefficient series as a function
of seismic traveltime. This signal represents the reflectivity ε.
In addition, an estimate of the seismic wavelet s must be found. An example of a
reflectivity function with a seismic wavelet is shown in Figure 7, in which we also show
1496_Ch08 10/10/08 5:44 PM Page 206

206 Digital Imaging and Deconvolution

First, take the zero-lag term and the right-hand side of this autocorrelation, and then
replace the zero-lag term by unity. The result is

(1, 0, 0, g3 ) = (1, 0, 0, ε 0 ε 3 ). (17)

As a Z-transform, this result is

1 + g3 Z 3 = 1 + ε 0 ε 3 Z 3 . (18)

We will show now that this Z-transform represents the feedback loop that produces a rever-
berating wavetrain.
The surface (the water-air interface) has the reflection coefficient ε0. Because the
surface is a strong reflector, this reflection coefficient is close to unity in magnitude. The
reflection coefficient for a wave striking the water surface from below is ε0.
The water-bottom interface also is a strong reflector, with the reflection coefficient ε3.
Thus, the water layer acts as an imperfect energy trap in which a seismic pulse is reflected
successively between its two interfaces. Energy that bounces back and forth between two
interfaces is called a reverberation.
Let us examine the dynamics of this water reverberation. Our approach is applicable
to any reverberation occurring between two interfaces, not just to the water reverberation
occurring between the water-surface layer and the water-bottom layer encountered in
marine seismic work.
In summary, the source and receiver locations are just below the water surface. Source
energy travels down to the water bottom, from which it is reflected. The reflected upgoing
energy proceeds toward the water surface, where it is recorded as a primary reflection.
However, this upgoing energy also is reflected from the surface and then continues to
bounce back and forth in the water layer. These multiple reflections within the water layer
make up the water-layer reverberation. Such reverberations are generally undesirable
because they obscure reflections from deeper horizons. Figure 8 shows the raypaths for the
water reverberation. (For clarity, the raypaths have been drawn as slanting lines, although
in our normal-incidence model, they are perpendicular to the two interfaces.)

Shot S
(just below Receiver R
Interface 0 surface) (just below surface)
e0
A C D F G
Interface 1
e1 = 0
(missing)

Interface 2 e2 = 0
(missing)

Interface 3 B E
e3

Figure 8. The downward pass through the water layer.


1496_Ch08 10/10/08 5:44 PM Page 207

Chapter 8: Synthetics 207

In our example, the integer T  3 represents the two-way traveltime parameter in the
water layer. The source is a unit impulse at A, which is a point just below the surface. The
pulse travels downward in the water and arrives at the water bottom. There, it is reflected.
The resulting upgoing pulse at B has value ε3 and arrives at the water surface C at time 3.
The receiver records this upgoing pulse as the primary reflection. However, this upgoing
pulse then is reflected downward. In the case of an upcoming incident wave, the water
surface has a reflection coefficient of ε 0, so the downgoing pulse at D has the value
ε0 ε3. This downgoing pulse travels downward and is reflected from the water bottom
at E. It returns to the surface at F. The receiver records this upgoing pulse as the first
multiple reflection, with the value ε3 (ε 0 ε3) occurring at time 2T  6. This repeats to
produce the second multiple reflection with the value ε3(ε0 ε3)2 occurring at time 3T  9.
The process keeps repeating itself, each time producing another multiple reflection.
Because we omit the shot, we write the trace as

h = (0, 0, 0, ε 3 , 0, 0, ε 3 (− ε 0 ε 3 ), 0, 0, ε1 (− ε 0 ε 3 )2 , 0, 0, …), (19)

where the first 0 is at time 0, the second 0 is at time 1, the third 0 is at time 2, the ε3 is at time
3, and so on. In other words, the shot occurs at time 0, the first nonzero value ε3 (the primary)
occurs at time T  3, the second nonzero value (the first multiple) occurs at time 6, the third
nonzero value (the second multiple) occurs at time 9, and so on. The successive nonzero
values (after the primary) are separated by the time parameter T  3. For that reason, T is
called the cycle time of the reverberation. Thus, the Z-transform of the impulsive trace is

H ( Z ) = ε 3 Z 3 + ε 3 (− ε 0 ε 3 ) Z 6 + ε 3 (− ε 0 ε 3 ) 2 Z 9 + …
ε3 Z 3 (20)
= ε 3 Z 3 [1 + (− ε 0 ε 3 ) Z 3 + (−ε 0 ε 3 )2 Z 6 + …] = .
1 + ε 0ε3 Z 3

The geometric series within square brackets is convergent and summable because the
autocorrelation coefficient g3  ε0 ε3 is less than one in magnitude. Now comes an impor-
tant point. We recognize that the multiple-generating factor ε0 ε3 that occurs in the above
expression for the trace x is the autocorrelation coefficient g3. Thus, the Z-transform becomes

ε3 Z 3
H (Z ) = . (21)
1 + g3 Z 3

We therefore have shown that the primary reflection is represented by the feedforward
term ε 3 Z 3, whereas the reverberation is represented by the feedback term 1  g3 Z3 (Figure 9).

Synthetic seismogram with multiples


Previously, we discussed the synthetic trace without multiples, and we have just looked
at multiple water reverberations in the topmost water layer. We now turn to the more
general case of a synthetic seismogram for a model which can produce multiples in any of
1496_Ch08 10/10/08 5:44 PM Page 208

208 Digital Imaging and Deconvolution

Input its layers. The details of


Output
e3 Z3 + generating a synthetic trace
with multiples were given in
Z3 –g3 Robinson and Treitel (1978),
who also described an essen-
Figure 9. Block diagram of the feedback system that tial mathematical simplifica-
generates reverberation. tion that holds for the case of
small reflection coefficients.
The term small is defined in a
relative manner that depends on the circumstances at hand. In some cases, small might
mean a magnitude less than 0.2, whereas in other cases, it might mean a magnitude less
than 0.05. Here, we present a method of generating a synthetic trace with multiples in such
a way as to bring out what the mathematics is doing, rather than in terms of a strict math-
ematical derivation. The key concepts are feedforward and feedback.
We start with the case of only two interfaces (namely, the surface and the first subsur-
face interface). They are separated by a vertical distance with two-way traveltime T  1.
The reflectivity is (ε 0, ε1). Its one-sided autocorrelation (with the zero-lag term set equal
to unity) is (1, ε 0, ε1), which we denote by (1, g1) where g1  ε 0 ε1 . . ..
We recognize that the current problem is the same as the reverberation problem treated
above, but now with two-way time T  1. Thus, the impulse response h (with multiples) in
the case of two interfaces has the Z-transform

H ( Z ) = ε1 Z + ε1 ( − ε 0 ε1 ) Z 2 + ε1 ( − ε 0 ε1 ) 2 Z 3 + …
ε1 Z (22)
= ε1 Z [ 1 + ( − ε 0 ε1 ) Z + ( − ε 0 ε1 ) 2 Z 2 + … ] = .
1 + ε 0 ε1 Z

We recall that the impulse response (without multiples) is generated by a purely feed-
forward system (with the reflection coefficients on the feedforward loops). Here, the
expression for the Z-transform shows that the impulse response (with multiples) is gener-
ated by a feedforward-feedback system (with the reflection coefficients on the feedforward
loops and the autocorrelation coefficients on the feedback loops). Equation 22, which is
for the case of two layers, is exact. In other words, equation 22 gives the Z-transform of
the impulse response for the dynamic model (equation 1).
Next let us consider the case of three interfaces. What is the impulse response (with
multiples) for three interfaces? The reflectivity is then (ε0, ε1, ε2). Its one-sided autocorre-
lation (with the zero-lag term set equal to unity) is

(1, g1 , g2 ) = (1, ε 0 ε1 + ε1 ε 2 , ε 0 ε 2 ). (23)

By analogy with the result for two interfaces given above, the synthetic impulse
response h (with multiples) in the case of three interfaces has the Z-transform
ε1 Z + ε 2 Z 2
H (Z ) = . (24)
1 + g1 Z + g2 Z 2
1496_Ch08 10/10/08 5:44 PM Page 209

Chapter 8: Synthetics 209

However, expression 24, obtained by analogy, is not exact but instead is an approxi-
mation to the true dynamic model. Equation 24 leads to the synthetic trace with multiples
(equation 2).
Now let us consider the case of four interfaces. What is the impulse response (with
multiples) for four interfaces? The reflectivity is then (ε0, ε1, ε2, ε3). Its one-sided auto-
correlation (with the zero-lag term set equal to unity) is

(1, g1 , g2 , g3 ) = (1, ε 0 ε1 + ε1 ε 2 + ε 2 ε 3 , ε 0 ε 3 ). (25)

By analogy, the synthetic impulse response (with multiples) in the case of four interfaces
has the Z-transform

ε1 Z + ε 2 Z 2 + ε 3 Z 3
H (Z ) = . (26)
1 + g1 Z + g2 Z 2 + g3 Z 3

However, expression 26, obtained by analogy, is not exact but is an approximation to the
true dynamic model. Expression 26 can be diagrammed as the feedforward-feedback system
shown in Figure 10.
By analogy, the synthetic impulse response (with multiples) for N interfaces has the
Z-transform

ε1 Z + ε 2 Z 2 + … + ε N Z N
H (Z ) = . (27)
1 + g1 Z + g2 Z 2 + … + gN Z N

As before, this expression is an approximation to the dynamic model. Let us define E(Z)
and G(Z) as

E ( Z ) = ε1 Z + ε 2 Z 2 + … + ε N Z N and G ( Z ) = 1 + g1 Z + g2 Z 2 + … + gN Z N . (28)

Equation 27 can be written as

E(Z ) 1
H (Z ) = = E ( Z ) M ( Z ) where M ( Z ) = = 1 + m1 Z + m2 Z 2 + …. (29)
G(Z ) G(Z )

For equation 29 to be true, the denominator polynomial G(Z) must be a minimum-


delay polynomial. In such a case, we have the approximation h  m * ε, which gives x  s
* m * ε. Thus, equation 2 holds for the synthetic seismogram with multiples.
Let us now discuss the approximation that is required. It is called the small-reflection-
coefficient approximation, and it was introduced by Robinson and Treitel (1978) and
further developed by Robinson (1982, 1999). The approximation requires that the reflection
coefficients be small enough to make the denominator polynomial G(Z) a minimum-delay
1496_Ch08 10/10/08 5:44 PM Page 210

210 Digital Imaging and Deconvolution

e1 polynomial. Thus, before


+ Z
Input Output this approximation can be
used, the denominator poly-
Z nomial must be tested for
minimum delay. If the
e2 + – g 1 denominator polynomial is
minimum delay, polynomial
division can be used to find
Z the coefficients m  (1, m1,
m2, m3, . . .) of M(Z). If the
e3
denominator polynomial is
+ – g2
not minimum delay, the
approximation fails, and
the exact dynamic expres-
Z – g3
sion (i.e., the expression
obtained without the small-
Figure 10. The feedforward-feedback filter for a four- reflection- coefficient app-
interface case. roximation) must be used.
In fact, it is always prudent
to use the exact dynamic model (equation 1) in computations (Robinson, 1999).
If the small-reflection-coefficient approximation holds, then the approximation
H(Z)  E(Z) M(Z) holds, and the multiples M(Z) can be removed by ordinary (i.e., linear
time-invariant) deconvolution. In other words, the small-reflection-coefficient approxima-
tion justifies the use of ordinary deconvolution. For this reason, a sequence of time gates
is chosen on the actual seismic trace. The choice is made on the supposition that within
each gate, the small-reflection-coefficient approximation holds. A deconvolution operator
is computed for each gate. The deconvolved trace is made up of all the deconvolved gates
along with appropriate interpolation between any two adjacent gates.

Examples
Let us now discuss a model consisting of a surface interface overlying two buried
interfaces, with the three interfaces separated by arbitrary two-way layer traveltimes. Let the
reflection coefficients be given by a, b, c. Let the two-way traveltime between the surface and
the first buried interface be S, and let the two-way traveltime between the first buried inter-
face and the second buried interface be T. In other words, the surface reflection coefficient is
ε0  a, the reflection coefficient for the first buried interface is εS  b, and the reflection
coefficient for the second buried interface is εTS  c. The Z-transform of the reflectivity is

a + b Z S + c Z S+T . (30)

The right-hand side of the autocorrelation of the reflectivity is

g0 + gS Z S + gT Z T + gS+T Z S+T = g0 + a b Z S + b c Z T + a c Z S+T . (31)


1496_Ch08 10/10/08 5:44 PM Page 211

Chapter 8: Synthetics 211

If we replace g0 by 1, we obtain the feedback loop

1+ a b Z S + a c Z T + a c Z S+T . (32)

There are three reverberations. The reverberation between the surface and the first
buried interface contributes gS  a b ZS, the reverberation between the second and third
buried interfaces contributes gT  b c ZT, and the reverberation between the surface and
third buried interface contributes gTS  a c ZST. The synthetic trace now is given by the
feedforward-feedback filter (Figure 11):

b Z S + c Z S+T
. (33)
1 + a b Z S + b c Z T + a c Z S+T

Suppose that S  2, T  5. Then the synthetic trace is given by

b Z2 + c Z7
, (34)
1+ a b Z2 + b c Z5 + a c Z7

which is

shot, 0, b, 0, −(ab 2 ), 0, a 2 b 3 , c − b 2c, − (a 3 b 4 ), 2ab(−1 + b 2 )c, a 4 b 5 ,


3a 2 b 2 (1 − b 2 )c, − (a 5 b 6 ) − bc 2 + b 3c 2 , 4 a 3 b 3 (−1 + b 2 )c,
a(a 5 b 7 − c 2 + 4 b 2c 2 − 3b 4 c 2 ), 5a 4 b 4 (1 − b 2 )c,
a 2 b(−(a 5 b 7 ) + 3c 2 − 9b 2c 2 + 6b 4 c 2 ), b 2 (−1 + b 2 )c(6a 5 b 3 − c 2 ), …. (35)

The shot occurs at time 0, the first primary b occurs at time 2, the first-surface–first-
interface multiple ab2 occurs at time 4, the second-surface–first-interface multiple a2b3
occurs at time 6, the second primary occurs at time 7, the third-surface–first-interface
multiple (a3b4) occurs at time 8, the peg-leg multiple 2ab(1  b2)c occurs at time 9,
and so on. Figure 11a shows the reflectivity for the case a  0.8; b  0.4; c  0.7. The
leading portion of the corresponding impulsive synthetic trace is shown in Figure 11b.

Small and white reflection coefficients


Finally, let us consider the case in which the reflection coefficients are not only small
but also are white. In such a case, the autocorrelation coefficients g1, g2, g3, . . . are each
approximately zero. As a result, the denominator polynomial 1  g1 Z  g2 Z2 . . .  gN ZN
reduces to 1, and thus M(Z) 1. This equation says that in effect, the multiples cancel each
other out. Thus, when the reflection coefficients are small and white, the trace becomes
1496_Ch08 10/10/08 5:44 PM Page 212

212 Digital Imaging and Deconvolution

a) 0.8 x  s * ε, which is the


multiple-free trace given
0.6
by equation 3, the syn-
0.4
thetic seismogram (with-
out multiples). In other
0.2 words, in the case of
small and white reflection
0 5 10 15 20 coefficients, the dynamic
–0.2
convolutional model (equa-
tion 1) reduces to the
–0.4 time-honored synthetic
seismogram (without mul-
b) 600 tiples), which is by far the
most used type of syn-
400 thetic.
The synthetic seismo-
200 gram (without multiples)
corresponds to a sequence
of wavelets each weighted
5 10 15 by a reflection coefficient.
–200
It is interesting that the
amplitude of each reflec-
–400
tion coefficient is pristine
because the primary reflec-
Figure 11. (a) The reflectivity for the case a  0.8, b  0.4, tion appears to suffer no
and c  0.7. (b) The resulting impulsive synthetic trace. transmission losses on the
trip down to the respective
interface and then back up.
In other words, under the small-and-white hypothesis, the effects of all the multiples are only
good. That is, the multiples jack up the primary reflections on the seismogram to their proper
strengths, and yet the multiples are otherwise invisible.
In all science, there might not be any better example of randomness causing a beautiful
result. This result has had practical value because it made possible much of the oil discov-
ered by seismic means in the period from 1930 to 1960. During that period, the seismic
prospects that could be interpreted with good results either approximately fitted this model
or else benefited from some other fortuitous circumstance. Many of these prospects
occurred at shallow depths.
The vast majority of oil-bearing regions, both in extent and in depth, including
virtually all offshore regions, could not be explored for a long time because interfering
multiples and reverberations made the seismic record uninterpretable. To explore those
regions, the oil industry turned to the digital computer, and in so doing, it was the
first industry to do so. The digital computer and associated instrumentation are responsi-
ble for most of the oil discovered after 1965. If the digital computer were to disappear,
no additional significant oil discoveries would be made. No further space exploration
1496_Ch08 10/10/08 5:44 PM Page 213

Chapter 8: Synthetics 213

would be done. Promising avenues in science and medicine would be abandoned. Business
and finance would be crippled. The Internet would disappear.
In the words of Dawson (2008, p. 117), the seismic “industry’s tools and methods have
become increasingly more accurate; however, target objectives have become much more
obscure. Consequently, the future will require considerably more thought and effort.”
The computer should not be used as an excuse for not learning the multiplications table.
Instead, the computer is the reason that the most advanced methods of geophysics, math-
ematics, physics, and electrical engineering must be mastered.
Let us now give an example of the small-and-white hypothesis. This example is made
purposely simple to illustrate what the mathematics is doing. We have a water layer with
the surface-reflection coefficient ε0  0.9 and the bottom-reflection coefficient ε1  0.7.
The reflection coefficients within the time gate from time 2 to time 9 are

(ε 2 , ε 3 , …, ε 9 ) = (−0.014, 0.056, 0.017, 0.016, − 0.005, − 0.0056, 0.041, 0.043), (36)

which are small and white. The model for the small-reflection-coefficient hypothesis is

ε1 Z + ε 2 Z 2 + … + ε 9 Z 9
H (Z ) = . (37)
1 + g1 Z + g2 Z 2 + … + g9 Z 9

Because the reflection coefficients are small and white within the gate, we can make
the approximations

g1 = ε 0 ε1 , g2 = g3 = … = g9 = 0. (38)

Thus, we have the approximation

ε1 Z + ε 2 Z 2 + … + ε 9 Z 9
H (Z ) = . (39)
1 + g1 Z

This approximation is not dynamic, but it is stationary. In the equation h  m * ε, the mul-
tiple wavelet m has the Z-transform

1
M (Z ) = (40)
1 + g1 Z

which, as we know, is a reverberation. We illustrate this model by Figures 12, 13, and 14.
Each figure is made up of two touching bars. The gate is from time 2 to time 9. As we see
in the figures, deconvolution gates should be chosen between major reflecting horizons,
where the reflection coefficients are more likely to be small and white.
1496_Ch08 10/10/08 5:44 PM Page 214

214 Digital Imaging and Deconvolution

Figure 12. The trace (dark bars of


each pair) within the gate for the 0 1 2 3 4 5 6 7 8 9
small-reflection-coefficient model. 0.0
The trace (the light bars of each
pair) within the gate for the small-
and white-reflection-coefficient
model. The two bars of each pair −0.1
are almost the same. This result
demonstrates that the
approximation is good. −0.2

−0.3

−0.4

0 1 2 3 4 5 6 7 8 9 56 10−3
0.0
48

40
−0.1
32

24
−0.2
16

−0.3 8

0
0 1 2 3 4 5 6 7 8 9
−0.4 −8

−16

Figure 13. The reflection coefficients (dark Figure 14. The reflection coefficients (dark
bars) within the gate. The trace (light bars) bars) within the gate. The deconvolved trace
within the gate for the small-reflection- (light bars) within the gate for the small-
coefficient model. The two bars of each pair reflection-coefficient model. The two bars of
are far different. This result demonstrates that each pair are almost the same. This result
the recorded trace fails to show the reflection demonstrates that the deconvolved trace shows
coefficients. the reflection coefficients. Notice the vertical
scale change from Figure 13 to Figure 14.
1496_Ch08 10/10/08 5:44 PM Page 215

Chapter 8: Synthetics 215

Appendix H
Exercises
The breaking waves dashed high
On a stern and rock-bound coast.
—Felicia Dorothea Hemans (1793–1835)

1. Identify the multiples beyond time 9 that are shown in Figure 11b.
2. Figure A-1 shows a zero-phase wavelet with its amplitude spectrum and a minimum-
delay wavelet with its amplitude spectrum. The two amplitude spectra are almost the
same. Which wavelet is better for processing and which for interpretation?

Amplitude spectrum
Zero-phase wavelet of zero-phase wavelet

0.4 0.35
0.3
0.2 0.25
0.2
0.15
0.1
0.05 0.1 0.15 0.2 0.25
0.05
–0.2
–100 –50 50 100
Amplitude spectrum of
Minimum-phase wavelet minimum-phase wavelet
0.4 0.35
0.3
0.2 0.25
0.2
0.15
0.05 0.1 0.15 0.2 0.25 0.1
–0.2 0.05

–100 –50 50 100

Figure H-1. Two wavelets that have approximately the same amplitude spectrum.

References
Aki, K., and P. G. Richards, 2002, Quantitative seismology: University Science Books.
Anstey, N. A., 1960, Attacking the problems of the synthetic seismogram: Geophysical
Prospecting, 8, 242–259.
Berkhout, A. J., 1987, Applied seismic wave theory: Elsevier.
Dawson, L. D., 2008, Sixty years with SEG: The Leading Edge, 27, no. 1, 117.
1496_Ch08 10/10/08 5:44 PM Page 216

216 Digital Imaging and Deconvolution

Hilterman, F., J. W. C. Sherwood, R. Schellhorn, B. Bankhead, and B. DeVault, 1998,


Identification of lithology in the Gulf of Mexico: The Leading Edge, 17, no. 2, 215–222.
Kelly, K., R. Ward, S. Treitel, and R. Alford, 1976, Synthetic seismograms, a finite differ-
ence approach: Geophysics, 41, 2–27.
Mallick, S., 2007, Amplitude-variation-with-offset, elastic-impedance, and wave-equation
synthetics — A modeling study: Geophysics, 72, no. 1, C1–C7.
O’Doherty, R., and N. Anstey, 1971, Reflection on amplitudes: Geophysical Prospecting,
19, 430–458.
Robinson, E. A., 1957, Predictive decomposition of seismic traces: Geophysics, 22, 767–778.
———, 1975, Dynamic predictive deconvolution: Geophysical Prospecting, 23, 779–797.
———, 1982, Spectral approach to geophysical inversion by Lorentz, Fourier, and Radon
transforms: Proceedings of the IEEE, 70, 1039–1054.
———, 1999, Seismic inversion and deconvolution: Handbook of geophysical explo-
ration, 4B: Elsevier.
Robinson, E. A., and S. Treitel, 1978, The fine structure of the normal incidence synthetic
seismogram: Geophysical Journal International, 53, no. 2, 289–309.
Sangree, J. B., and J. M. Widmier, 1979, Interpretation of depositional facies from seismic
data: Geophysics, 44, no. 2, 131–160.
Sheriff, R. E., 2002, Encyclopedic dictionary of exploration geophysics, 4th ed.: SEG
Geophysical References Series No. 13.
Shtivelman, V., and D. Loewenthal, 1989, Construction of the generalized one-dimensional
synthetic seismograms by a three-step extrapolation procedure: Geophysics, 54, no. 8,
1050–1053.
Ulrych, T. J., 1999, The whiteness hypothesis: Reflectivity, inversion, chaos, and Enders:
Geophysics, 64, no. 5, 1512–1523.
Vail, P. R., R. M. Mitchum, and S. Thompson, 1977, Seismic stratigraphy and global
changes of sea level, part 3: Relative changes of sea level from coastal onlap, in C. E.
Payton, ed., Seismic stratigraphy — Applications to hydrocarbon exploration: AAPG
Memoir 26, 63–81.
Van Wagoner, J. C., H. W. Posamentier, R. M. Mitchum, P. R. Vail, J. F. Sarg, T. S. Loutit,
and J. Hardenbol, 1988, An overview of the fundamentals of sequence stratigraphy and
key definitions, in C. K. Wilgus, C. A. Ross, and H. Posamentier, eds., Sea-level
changes: An integrated approach: SEPM Special Publications 42, 39–46.

Potrebbero piacerti anche