Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
GEOPHYSICS
Schemes for seismic mapping of reflectors in mula in\olving up antl downgoing \vaves. The
the presence of an arbitrary velocity model, dip mapping formula may be implemented without
ping and cur\,ed reflectors, diffractions, ghosts, undue complexity hy means of difference ap
surface elevation variations, and multiple reflec- proximations to the relativistic Schroedinger
tions are reviewed and reduced to a single for- equation.
t Presented at the Pacific Coast Section meeting of the So&t>, of Exploration Geophysicists at Bakersfield, Cali-
fornia, October 20, 19iO. Manuscript received hy the Editor L\pril 6. 1970; revised manuscript received October 5,
1970.
* Geophysics Department, Stanford University, Stanford, California 91303
@ 1971 by the Society of Exploration Geophysicists. .\I1 rights reserved.
467
468 Claerbout
s
knowledge of the interface, they will predict an erro-
neous upcoming wave lTr at E‘3, indicated I)>-dots lxfore
the arrival of the downgoing wave. This error has no
~hp(.v, z) = L.(Z) D(Z)tlw. (J)
bad effect on the reflector mapping formulas which
utilize time coincidence of up and downgoing waves.
These ideas are valid in situations where there are man! Expanding out the inverse of D(Z), we get
reflectors at many depths.
,\Iap(.v, i)
lrave. Even if multiples are included, this as- performs the deghosting correctly for all offsets if
sumption is al\vays valid ior an impulsi!,e surface the do\vngoing wave D(s, z, w) inclutles the ghost.
source’ of scalar \vaves in a layered medium. It will ;\s noted in paper 1, there is no extra computa-
usually be a reasonable assumption in the fielcl. tional effort in including a ghost in the calculation
(The principal feature of minimum phase is that 0i Dix, 2, w) ll!- difference equations. Caution is
the energy arrives in a burst which is not followed required in the frequency domain as uell as the
by a larger burst.) .I process similar to decori- time domain. Figure 2 show the nodal lines
solution is implicit in equation (3). In a sim- which \vill appear ii the ghost is as strong as the
plified \-ie\v of tlcconvolution, \re a5sumc that re- primary. Fortunately, the ghosting problem is
flection coefficients are randomI!- distributed with usually not severe in practice; and we shall sug-
depth, so that the upcoming \~a\-e is created by a gest an implementation of equation (3) \vhich ac-
convolution of some unkno\vn downgoing wave- counts ior noise and avoids division by small
form with random reflection coefficients. Be- quantities. Let the integrand Z7/11 of equation
cause of the randomness of the reflectors, \ve ma) (3) be multiplied and divided by the complex
equate a power spectral estimate of the observed conjugate of the do\vngoing Irave. Then, the de-
upcoming \va\-e to the energy density spectrum nominator is the spectral density DD* of the
of the initial do\\ngoing wave. \Ve then compute tlo\vngoing Jr-ave. A spectrum is real and contains
a minimum phase time function \rith the rc- no phase iniormation. Recalling earlier remarks
quit-cd spectrum anti call it cE(.v,2, I). Division b>- about automatic gain control, we find it natural
D(.r, e, w) in expression (3) amounts to something to omit the denominator and consider another
similar to the usual implementation of decon- map construction formula:
solution with d(s, 2, t).
time variable deconvolution is a process mo-
tivated by the kno\vletlge that as the downgoing
wave penetrates the earth, its higher frequencies
dissipate first because of the nonideal acoustic
properties of rocks. If D(.r, z, w) is computed h!- The phase of the integrand in equation (1) is the
difference approximations to the 1vaI.e equation, same as that of equation (3). The amplitude in
it iS not diffiCUlt 10 iJlCiU(lC the i~~~ChaJli51~~ Oi high equation (1) has the very tlcsirable property that
frcquenc?_ losses. \\.C could alto iJlclude loss it drops off rapidly in any region where either the
mechanisms in the tlo\vn\vartl extrapolation 0i I-. downgoing \\-al-e is weak or the upgoing wave is
;\n interesting aspect of this approach to time \vcak. Thus, equation (4) is good for superposition
variable deconvolution is that it is really space of maps of many spatially separated shots be-
variable filtering and may I)c empirically related cause each shot contributes most to the map in
to properties of earth materials. the region it best illuminates. .\ compromise
i\nother application of equation C.3) is to the het\veen equations (3) and (4) which can sho\v
operation called deghosting. .\ ghost arises \vhen some of the advantages of deconvolution with
wc are shooting \vith a rather deepI>- buried some of the signal-to-noise advantages of (1) is to
charge. -liter the \vave that travels directly from use an integrand like
the shot do\vn to the reflector arrives, a ghost
c.D*,‘(\- + DD*),
wave which has traveled from the shot to the
surface and then down to the reflector arrives. If where 1. is a constant or a slo~vly varying function
the ghost is not accounted for in the mapping of frequency and space. Some kind of optimiza-
process, a single reflecting layer \vill appear to be tion could clearly lead to an involved discussion.
two closely spaced layers. ‘1‘0 a first approxima- Equation (4) may easily be expressed in the
tion, .deghosting amounts merely to deconvolv- time domain as
ing the data with the appropriate double pulse.
More precisely, the time separation of the t\ro
Mnp(a, z) = J ZL(.Y, z, t)d(.r, 0, t)tlt. (5)
pulses is a function of shot-to-geophone offset.
If a dcghost filter for normal incidence is applied
too far off from normal incidence, ghosts will he The mapping formula (5) will be most accurate
generated rather than destroyed. Equation (3) when the downgoing wave is an impulse, but it is
470 Claerbout
clearly a reasonable approximation if (t(.r, 2, t) is (1969), and Trorey (1968, 1070). This second
a wavelet of short duration. method has been called “impulse holography” by
Fontanel and Grau. The third method is the
PROJECTING WAVES DOWNWARD
method of difference equations [Jroposedin paper
To the first approximation, projecting the 1. First let us review- ray methods and impulse
downgoing wave downward is purely a theoretical holography to see how they ma! be understood in
operation. All that is needed is the shot time and terms of the mapping formulas oi the last section.
position and a velocity model. The upcoming Normal moveout correctiotl is the simplest
wave, however, is what we observe on the surface. form of migration or reflector mapping. Let us see
If we project this wave back downward using a how it may be regarded as an implementation of
linear scheme (scaling and shifting), the mapping the mapping formula (5). As sl~own in Figure 3,
formulas of the previous section will all be linear the source is taken to be an impulse s(t) at the
functions of the observations. This is important coordinate origin (0, 0). The wave reflects from a
because it means we can operate with confidence horizontal planar interface at die point (x,2). The
at low signal-to-noise ratios. wave bouncing off from the refh,ctor is of the form
There are at least three operationally different 6(t-r/c), wherer= (.P+~‘)r~~i: the distance along
schemes for projecting waves from the surface a ray from the shot to the interface and c is the
back down into the earth. The first method, which (constant) velocity of the material. The wave
is perhaps the only method in currently popular arrives at the surface at the point (as, 0), where
data processing programs, is ray theory. The it should be of the form 6(t--Zr c). In fact, if ob-
second is based on the diffraction integral and has servations arc made at the point (2.r, 0), a compli-
been used by Peterson (1969), Fontanel and Grau cated function u(2.r, 0, t) is obxt~rvedthat is likely
Reflector Mapping 471
to be only a crude approximation to the impulse. create a depth map. ;\long these lines, Hagetloorn
However, from ray theory \re might expect the 11054) presents a geometrical method for migra-
wave at the surface to be a delayed verhion of tion of time domain seismograms:.
the wave at the reflector; that is, u~L.v., 0, I) r\ll oi these ideas may be reexpressed in the
ZEu(.r, 2, l-~/c). In other words, the \va\‘e at frequency domain. In that domain, different ideas
depth is given in terms of the surface obs5rvation present different advantages or ditliculties in
by U(S, z, L)=u(~.v, 0, t+r c-j. Thus ior a single concept and coml~utation. Since the impulse ho-
shot, equation (5) becomes lography method and the tlitTerencc equation
method are expressed in the frequency domain,
it may be useful to the reader to see that normal
moveout correction may be clone in the frequenq
domain. .I single irequenc!- component 0i the
= 11(2X, 0, 27 C’)
imp&i\-e ~va\‘e emitted by the source is rcpre-
= ?1(2S, 0, 2(.? + 5’))’ ” c). hented by O(O, 0, w) =6(r) cxp( - iwf). The do\vn~
going \\-al-e at the refcctor is D(.v, 2, w)
Hence, the rctlector map is mule from all the
=exp-kLJif-Y c-)1,‘/,.
seismograms recorded at all .v-positions on t ht’
.is with the time domain situation, \ve espect
surface bv resealing the time axes in terms 0i thv
the wave at depth to bc derivable iron1 the sur-
two-way traveltime 011 the raypath to the depth
iacc observation by mran5 0i a time shift (ant1 a
2. In concept, \ve get a depth map \vith just 011~’
scale factor bccau~ 0i 1 r). \\.v ha\-t~
shot. In practice, more shots are used, allowing
illumination of a greater region 0i the subsuriacc, I-(X, 2, W) e\p (- i0l)
statistical averaging (common depth point stack-
= 2(.(2.v, 0, w) exp (-iwit + r c)).
ing), and estimation of the vclocitv function
(residual normal moveout). It is not difficult to Ignoring the scale iactor5, \vc wc’ that equation
generalize the procedure to a situation \vhere the (4) bxollle~
velocity function c(z) is a function 0i depth. \\.e
may trace rays to find the appropriate time a&
resealing function \Vhat ih more tlilficult is to
extend the process to a more general \.clocitJ.
function for example, the function correspontlin::
=S1’(2x,
0,w)
to iolded and dipping layers. The problem is that
the surface point (2.r, 0) will then not be associ- e\;p (- i2wr r)dw.
ated with a depth point (I., z). In this case, some
type oi scan or search procedure is required to
The result must be the same as the one lve had
in the time domain. 11.e could actually do normal
moveout corrections in the frequency domain b!
this method, but it \I-ould turn out to be mereI>
an awkward wal- of rcscaling the time axis oi the
seismograms.
Complicatetl reflector shapes may also be
mapped by frequency domain procedures. Ho-
lography illustrates an application 0i equation (4)
to the curved reflector situation. There are t~vo
tlasic approaches: (1) focusing the image point of
the source in the reflector and (2) Rayleigh-
Sommerfeld aplanatic mapping. The first of these
t\vo methods encounters difficulties with curved
reflectors, so \\-e \vill restrict attention to the
second. Refer to Figure 4. .I monochromatic
seismic wave source is located at S, a seismic wave
I,‘r c. 3. Ckometrj- for rc.llection from ;I planar interface
(normal moveout correction geomelr~-). receiver is located at K,,, anal the image field scan-
472 Claerbouf
v
earlier in terms of perturbation theory. Subse-
quent efforts have put the method on a firmer
b %n base, a base which allows clearer exposition, ex-
tended region of validity, and gc,neralization to
other physical problems. As compared to ray
P methods and the diffraction integral, the differ-
ence equation method offers certain advantages.
I:lc. 4. Geometry for reflection or scattering from an
arbitrary depth point p at (.Y,a). I;rom Peterson (1969), The principal advantage is that die velocity func-
1’. 58. tion may be prescribed as a rather arbitrary func-
tion of space. Thus, computer inll)lcmentation is
ning point [the point at (s, z)] is denoted by p. not complicated when the geological situation is
‘I’hc field observed at K, is taken to be A,, complicated. This arbitrariness in the velocity
exp( - &). The Rayleigh-Sommerfeld aplanatic
model would be especially advantageous in off-
mapping equation as given by Peterson (1969),
shore prospecting, where the thpth is rapidly
(who specialized to 24 surface receivers to study
variable, or if the computing were to be done in
resolution with a finite spread length) is
an interactive environment, wlrcre a laterally
variable velocity model is devclol)cd iteratively.
Map(x,z,w)= 2 A,Lexp (-i&) The wave equation includes 110th upgoing and
n-1 (6) downgoing waves in its solution5. ‘I’0 implement
‘esp (--i(r,, + rPJw/‘c). the mapping formulas, we need not only the solu-
tion to the wave equation but also a means ol
We have ignored the so-called obliquity factor separating the solution into tlrc~ upgoing and
and an amplitude :vhich is- slow! v xvariable w i ! h downgoing components: iVirt!n I\ c Ue working
space. With the associations with analytic solutions in a constant velocity
medium, the separation is a trivial choice of
sign of a square root. For exarnllle, in hologra-
x=24
phy, extensive use is made oi the fact that
: : C A, esp (- i$,, - ir,,&c)
exp[-iiw(l-v/c)],‘r is a solutiorl to the scalar
wave equation. The distance Y= ~.P+z~)~~~ from
and the source (or focus) to the oljservation point
is given by a square root. Choice or the root sign
D(X, z, w) : : esp (+ ir,,w/f),
gives the wave direction. In nunlckrical work, the
we may compare equation (6) with the integrand separation is not so simple. For the following
of equation (4). Elimination of the plane layer practical reason, separation is essc,ntial: Consider
assumption is made possible by the creation of a pulse observed at the surface oi a homogeneous
U(X, z, w) from all the surface observations rather earth. Let the wave equation be USYL to determine
than just the observation at (2x, 0). This con- the waveforms seen at depth. :It any depth, we
struction of U(N, z, W) is firmly based in diffraction see two pulses, the upgoing wa\ c and later the
theory and may be intuitively regarded as a sum- downgoing wave. At greater dey)ths, the pulses
mation of Huygen’s secondary wavelets. Thus, are more widely separated. 7’11~. more widely
holographic principles enable us to migrate dif- separated the pulses are, the more’ lrequencies are
fractions from complicated reflector shapes in a required to represent the double i)ulse. The num-
medium of constant velocity. For a homogeneous ber of required frequencies quickly exceeds a
medium, the diffraction integral represents an practical limit. Hence, we will scl)arate the wave
important analytical solution to the difference equation into two parts, one for the upgoing
equations to be developed in the following section. waves and one for the downgoing waves. Where
As an analytic solution, it should be of further there are velocity gradients, one leave type scat-
value in shedding light on questions of resolving ters into the other. Thus, it will be sensible to
power of the difierence equation method. organize the computation so that Ilrimary arrivals
Reflector Mapping 473
are calculated first anti multiply reflected waves Sotice that the velocity oi the material is given
are considered later if necessary. by
Among those who work Gth field data, no
c = (K/‘p)“‘.
apology need be made for using the acoustic wave
equation to describe elastic waves. Let U and II’
It is convenient to define the \vavenumber uz by
denote displacement in the s and z directions and
let P denote pressure. Let p denote density and K Wl = w ‘r.
denote incompressibility. Consider an exp(- iwt)
time dependence. Since the product of mass den- Ii we temporarily take ~2 to be intlepenclent of .x,
sity and acceleration equals the negative of the 171commutes \vith D,,; and we may use the b-
pressure gradient, we have nomial theorem to define the square root oh.
[:;.I=[;bJ[p]> (11)
(d,‘dz) ;,
[I
and the inverse relation, a transform of ro\j
(8) eigenvectors,
0 w?p P
=[ - 1,/K-DD,,,‘(w’p) 0 I[IT’
1
It is convenient to define (12)
(1? = W”P (‘kt)
~‘IG. 5. \V;tves impinging on a Imrieti IAock of slogsvelocity material. \Vaves enter at the top of the I~lock and are
com~~letel~- internall~~ retlectetl from the sitle of the I~lock. This leaves a shadow on the out4de of the I~lock.
FIG. 6. A sloxvvelocit); l~lock is illuminated from the side. There is partial reflection from the side of the I~lock and
mterference hetwcen naves entering the I~lock through different faces.
476 Claerbout
or
aD,/as = i(a*b*)‘!D
‘. (16)
= Ka2b’
_a;2]
0
- f[12_I]aiJaz going wave.
be homogeneous,
its usual form
Temporarily taking
we can get the wave equation
by differentiating
the medium
equation (lo)
to
in
with respect to z.
I:IG. 8. Expanding cylindrical wave. An analytical solution for a point source in the ix, x) l)l:~ne is taken at the
top boundary and projected doqnyard with a difference approximation Lo equation (17) Given (heapproximations
which went into equation (17), It 1ssomewhat surprising that the solution looks as good as it ~CIVSfor ravs between
45 degrees and the horizontal. The principal deviations from the correct true cylindrical WV&VP are phase fronts
which are not quite circular and the anomalously slow amplitude decay for rays traveling ahoot 60 degrees from
the vertical.
Reflector Mapping 477
V’IC. 9. kVaves coming to a focus. The calculation includes the first relativisticapproximation, that is, the D,, .D,
term. This accounts for the much better result than those show] in Claerlwut (19iOa).
Ii the square root in equation (16) is approxi- in the present paper tar oif-axts waves (See
mated by the first two terms of the binomial ex- Figures 8 and 9.) Inclusion of the extra term
pansion, equation (16) becomes the ordinar> causes no significant increase in the amount of
Schroedinger equation (with seismic z correspond computation, since the term fits on the same
ing to time t in quantum mechanics). ;\s it stands, 6-point (3 points in s, 2 points in z) star. ;\ more
equation (16) is the relativistic Schroedinger thorough analysis of various approximations to
equation. (The positive square root prevents the square root for various grid spacings at con-
present changes in quantum potential from intlu- stant velocity is given in Claerbout (lOSOb). In
cncing past values of the probability function.) .\ that paper, the wave equations are anal>-zed from
very useful first approximation to the relativistic a filter theory point of view. An improved approx-
Schroedinger equation is obtained by replacing imation, when the velocity is the space variable,
the square root in equation (1.5) by the first two is found by observing that operating t\vice \vith
terms of the binomial expansion. Specifically, the operator
with this truncation, equation (15) may be nrit-
m + D,, ‘(2~2) - (D,m) ‘(2m?)D, (18)
ten
gi\w
9
PM- + D,, + error. (19)
The D,,D, term \vas not included in paper 1; its
inclusion accounts for the sonic\vhat I)cttcr results Thus relation (18) is the &sired square root of
478 Claerbout
relation (10) except for the error. ‘I’hc error terms D(s, 2, w). Hence, wc can make 8 first approxi-
are of two types. Some are neglected, since they mai ion fIO to II by extrapolation Ilownward with
contain D, and higher terms which are signifi-
iabfiD,,/& = (-a”b” + s11)Do. (21)
cant only for waves farther from the z-axis. Other
terms are neglected, since they do not increase 1: may be projected downcvard from surface data
with frequency as do the terms retained. WC have with a similar equation. On the other hand, we
not yet worked on examples where the added may desire to construct LTfrom II and a velocity-
accuracy of the last term in expression (18) ap- density model. In this case, we construct a first
pears to have made a notable difference. approximation UO to U in the iollowing way:
‘The right hand matrix in equation (14) is the below all the reflectors, ZTomay I)c taken to be
matrix which causes scattering from upgoing zero. Then knowing ZA, every\\ here, we may
waves to downgoing waves and vice versa. Let us extrapolate CT0upward with
introduce a simplifying definition for this matrix.
p h-j (22)
= (--u”b” + Srw) I’,, + SyD,,.
Is,, s,,J At the fret surface, the upcominfi wave bounces
(20)
back down into the medium, starting off a second
= -;E; _~]d/az[yb J.
downgoing wave UI. Since pressure vanishes at
the surface WC have D, = Va. Actually, WCmight
If this scattering matrix S is small, there is a v?sh to take account of an UIIC\‘I’II surface or a
simple iterative procedure for solving the ximul- static correction at the surlacc. Ii the effective
taneous equation set (l-1). Below a surface source surface elevation is h(s) and near wrface velocity
in a medium of modest inhomogeneity, U(s, z, w) is C, an al)proximation \:alid near vertical inci-
may be expected to be small compared to dence for small /r is
I“IG. 10.b’aves in a medium whosevelnci(y increasesI)y a factor oi 3 from t(q) to l)otlom. .\K;Linthe apparent
result for ra~‘sfar from vertical is much I~ct~erthorncould IX e\-pecte~l.In fact, one can see r:r!s which appear to
I)otLom~outand Iqin a return to the xuriacc.Suchrays, alLhoughI)h~~sicall~ l)retlictetl,arc ctrmlllrtel>~
outsidethe
realm ol validity of the qq>roximations.
Reflector Mapping 479
480 Claerbout
Reflector Mapping 481
Dl(x, 0, CO)exp (i/w/c) multiply by the inverse matrix of (23), and re-
arrange. 1Vc get
- l!“(n-, 0, W) esp (-&W/C) = 0.
[I [
I’= 0 esp(--4) I[ 1 H ’
(23) near-surface layer effects from seisnric records: Geo-
physics, v. 26, p. 754-760.
Hagedoorn, J. G., 19.54, A process of seismic reflection
internrelation: geophys Prow.. v. 2. o. 85-127.
where A and B are complex constants. If the Peters&, R. A., 1969; Seismography I(j?O, the writing
medium is slightly inhomogeneous and waves are of the earth waves: preprint for presentation at SEG
not exactly plane waves traveling vertically, we symposium, 1,os Angeles.
Ilichtmyer, I<. D., and Morton, K. W.. 1967, Difference
still suspect that A or B will vary much more methods for initial value problems: New York, Inter-
slowly with z than does D or U. To find a differen- science.
Trorey, A. W., 1068, personal communication.
tial equation with A and H as solutions, we sub-
- 190, A simple theory for seismic diffractions:
stitute equation (23) into equation (11), pre- Geophysics, v. 35, p. 762-784.