Вы находитесь на странице: 1из 12

JOURNAL OF THE CANAOIAIN SOCIETY OF EXPLORATION tEOPH”SlClSTS

“OL 12 NO ? ,DFC 1986, P 14~55

INVERSE VELOCITY STACKING FOR MULTIPLE ELIMINATION

DAN HAMPSON’

cocfticicnlscnn hc inlet-prcrcd dil.cctlyiogivc lhccncrg)


iwoci:mxl with the vill~iou\ vcloci~ic~ iit diffcfel-cnr /crw
offset times. In this scnsc‘. the cocfficient~ form WI
‘idc;diLcd’ consli!r~l docity \blck. 11scf111for vclocil)
an:dy\i\. The derived ~modcl mxy he furthel. LISCCIfur
noise reducG0n :ind intcrpol:ltion of missing d:11:1.
An ;Ipplic;ifion that WIS n<,t discussed by Thorwil i\
that of long-pet-id mrdGple elimination. ‘I‘his xpplic:~.
lion depc~lds Ott the fxct Ih;ll prim;lry ;!nd long-period
multiple events 111:I given /cro-offv2t time will gcner;~lly
hi~vc diffcl.cnt RMS vclocitic\ and will snap ,<I diffcwuc
rcgi0n\ of the tl0mGr~ \p;mned hy lhe v3 of weighting
cocfticicnts. Invet-\ctr-anclol-min~:~ftel-rcl-oin~sclcctc~l
set\ ~fcocfficicl~t> will then iwlillc plCrn;\r\ ;III~ Imulti-
pie encrgy. :I% is c~~mm~~~lly d~~nc with F-K lillcl-ing
(liyu. I’IXO). This prowdu~-c W:IS u\cd with wmc ~IIC-
ccss by Ilamp\on and I~11nham , IYXi ,. hut lhC cw,Iprll;!-
liowl effwl ~wde lhc ;1pp~~0;1ch impl~x3ic;ll liw volume
p,-occs\l”g.
4 I;~t-gc p;~rt ufthe coniplltational cffut-1 cxr> it1 fxct hc
11.aced IO the hypcrholic f~l-m ofthc move-out equation.
Modifyin~thisequation 1,,inc~,~-pol-atc;1p;ll-;lhulic wthct
rhan II hypcrholic fu~-rn I-CSIIIIS in ;I signifiwnr simplifica-
11on d the liww by\lcm. which mu\1 hc wlvd. xrxl ii
cot-re\p0niling reduction it7 c0mp~bt:1linn;1l ellilrI. 111lhe
rcmaindet- of this p:\pcr-. I de!-ivc lhc m<dificd linc;ll
Inverts Velocity Stxcking i% Ihe rums :ipplied 10 :I sybtcm. .itlrtify ilb use on NhlO-co~~~~cc~cJ di11i1. i~ml
procedure devised hy Thorson (IYX4). and dcscr-ihcd ill dcmonsttxle it5;Ipplic;llion lothe pl-ohlem~,flon~-pul-i~,d
Thorson and Clacrhouc c 19X5). for modelling hypcr- multiple climillatiorl.
holic cwxls ori seismic wxd\. In Ihis algorithm ihe
recorded seismic pn)filc is modelled ilsil linearc~,n-,hina~
tiun of simple hypcrholic CVCII~S ~fconswni ;m~plitudc. The hxic modelling equ:ltioll pl-opovxl I>) Thorwn i\
A bet of \rcighling coefficient\ is dcrivcd such thet the
rcsulling model ;~ppn)sim;itc? the input protilc ill the dih.t) = ~jdpd~LJ~p.~);iI-;~\ t~~-p~~h-I i tGh.l) (21
lenst-squawssense. ICxh hypc~holiccvencischal-;,c~cl--
izcd by two parametrx\: the Lcro-offset lime of the whcl-c d 1h.t) = mc;~su~-cd xi\mogu-;lm xt offset h and
CVCIII. ‘T,,. :tnd the st~u.rw\\. Iw-wiy ll!llC I

p = 1:” III U (p.~) hypc[holic tr;lllrf(>rm cocfficicr~t ;~t glow


ners p :mtl /cl-id-offwl lime 7
where V is lhc KMS velocity associxcd with that cvc~~t.
As discussed by Thorson (IYX4). rhc \ct of weighting n lh.0 = mcasul-cmcnt nDisc xl offwt h ;~nd tw~w;\)
t1111ct
1vrrit;l\ S,,ftuiiw I .A,,.. ,>I’ I,K ;,\c ,111
c S,\\‘.. C’;,lga, >, .,,h c,na
14
lNVEKSE “I.I.OCI’I’Y SrAc:KINc; 4s

Therearetw~,waysofintelpretingthiseqllation. One to the hypet-bolic (p.7) domain and hack. WC do not


way is to ,view it as a modelling system. Our seismic retrieve the original data. The noise n(h.t) i\ the compo-
datasrcdCh~.t). whichcould beashot profileoracommon- nent of dCh.t) that is ‘lost’ in the transformation or.
depth-point gather. In equation (21 we we attempting to equivalently. the component of d(h,t) that is orthogonal
model thcsc data as a linear combination of functions ot to the space gcncratcd by the basis functions (3).
the form Regardless of which intcrpt-ctation WC use. the proh-
lem lo he solved is: given the mcasurcd seismogram
6 IT \,:(i’] (3) d(h.t). find the weights U(p.7) that satisfy equation (2)
whiletninimiringthetnisfit n(h.t).ThorsonandClae~hout
(19851 showed that this problem can be solved as an
The real numbers U(P.T) arc the weights associated
~)verdeterminedlincarsy,tcm. Assumingth;ltd(h.t)and
witheachofthcfunctions.Thesefilnctionsinf~tct repre-
U(P.T) are defined at discrctc values of h,t.p. and T.
sent hypcrbolae. as can be seen by imagining a situation
equalion (2) can hc roduccd to the matrix equation:
in which all the ueights are zero except for that associ-
ated with p = p,, and T = T,) :
d=Lu+n (8)
U(P.TI =I A 6 (p-p,?) 8 C-~-T,,) 141
where d is a vector containing all values in the mcusurcd
seismogram, u a vector containing all the weights. n the
Inserting this into equation (2). and ignoring the noise vector of noise ulues and L. the matrix that incorpo-
component. momcntarily.we get rates the discrete form ofthc delta-function in (2). The
standard least-squares solution to (8) is:
d(h.t) = A 6 [T,> \:t’~ p<,‘hr] IS)
u = (L.‘L)- ’ 1.~‘.d (9)
This represents a hyperbola with amplitude A and a
trajectory defined hy While the form of equation (9) suggest> that this wlw
lion is straightforward, the dimensions of the matricw
1’ = T,,’ + p,,‘h’ (6) involved make an explicit solution impractical. L is a
matrix of dimension:
In gcner:d, the double integral in equation (2) i\ a sum
(N-, x NH) by (N.,- x NP)
ovcrallpossible hyperbolae.each with itsownwcighting
fxtor.
where N.r = number of time samples in the input
The prcscnce ofthe noise term. nth.0. indicates that
seismogram
WC do not expect this modelling to he perfect; that is we
NH = number of offsets in the input sis-
do not expect the linear combination to yield prcciscly
mogram
the input data. One t-ciwn for this is that the integrals
N ,I = numherofslownessesused in Ulp.~)
will be performed over :Llimited range of slowncsses. p.
and times. T. The noise term rcprcsents the misfit between
In a typical 96.tract seismic shot of 3.second do&on.
our ‘bat model and the input data. As we shall see
1. would hc of the order of lSO,OOO hy lStJ.000. .Thorson
below. thil; ‘best- model is derived by minimizing the
and Clacrhout (1985) suggested an itcrativc solution,
cncr-gy in this noise term.
which is still prohibitively time-consuming.
The second way to interpt-et equation (2) is to view it
In order to reduce the computational load. I propose
as a translixm. analogous, for example, to the 7-p
lo replace the hyperbolic equation (2) by the parabolic
transform. In fxcl. the 7-p transform can be written as
form:
(Thorson, 19X4):

d(h,t) = IIdpdTU(p,T)6[7-(t-phZ)] + n(h,t) (IO)


d(h.t) = IIdpd~IJ(p,~)6[~-(t-ph)l + n(h.0 (7)

where p is no longer the slowness defined by equation


Just as the 7-p transform can be calculated by sum- (I). hut simply a coefficient whose significance will
mingalon~Iincnrtr~jcctories,thecalculationofU(p,~)in hecomeapparentlater. Justascquation(?)modelledthe
equation (:I) involves summing along hyperbolic trajec- input seismogramasalinearcomhinationofhyperholae,
tories. From this point of view, equation (2) is closely equation (IO) models the seismogram as a linear comhi-
analogous to the T-P transform. The presence of the nationofparaholae. Thcappropriatencssofthisfunction
noise term now implies that the transform space gcncr- will be discussed below. The important feature to note
ated by the basis functions (3) is not a complete space. here is that we can take the Fourier Transform of both
That is, if we transform R data set from the (h.t) domain sides of equation (IO) and show that each Fourier com-
.M> I, ,I \\ll’iOY

poncnI cim hc cidc~~li~lcd illdependc~111y h! v>lvitig 21


fX”l! re<lilccd \y\,en,. I .c,:

illlil

dch.li \IJltl.\ilc~“~ 1121


M

dctr.i\, 1 llip.ui c l~w”l j Illh.i\i II31


I’

Nolc thill. in Ihi\ equation. the unknou 11coctficicnti


vcr-y lilllc cncl-gy itI ncighlx~unng vctocIIIc\. tn I;~cl. a\
1Jlp.w) 11lr;t given frcqilcncy w depend only 03 Ihc da1;1
‘l’ho~~~on p~~iclicd C)LL~.lhc conventionill (‘L’S ix prcciwly
c~,~np~~~nc~\l\~ll(h.~~I (\>rth;d l’~vxp~ncy. ‘I’hix ~dcc<lc~pting’
lhc t .‘<I ~;~cIcI~-in cquxiion (4). ~l’hr, c,cwKicicnl\ t’ip.;i
~11th~ il-c~j~~c-ni) components OCCLII\ hec:mc cctil:llion
:~re c:~lc~~l;~lcd h) nrultipl)inL: lt?l\ I;IC~OI. lh\ (t.‘l.i !
( IO/ i5 linc:ll. in both I :stld 7. t~qu;itit~n (2). on the t~thc~
wllick ctTcclivcl\ dtx01I~~1I~c~ lllc I;llcldty \illciil~cd
hitnd. i\ 1101linci~r~ itncl 170\llch dcurupling uoutd 0ccIfi
cvcnt\ ;tnd \h:~rpcn\ thc‘ir im:igie. tn Ihi\ f,~~.rn 1hcl-c i\ :I
in Ihc fvcctltcncy ilom;i,iIi.
hepawlion 01 cvcnt\ al similar timc5 xcurding ttr lhcil
~~t~~cilic~,i~ll~l~~t~;~l~i~tio~~~~t’~~l~lllipl~I’~~~~l~l
priiliilrycllc1~,g\
~lmdd lx pwrihle.

<I I.11 n , I-l)

whca~c (I i\ lhc cornplc\ \cclu~~ CI~Al ihc comtx~i~c‘l~l\


d(ll.\i / 11w il \,wil~iC Li. I1 IhC mllplc\ Vi’ClUl~ 01’ Ill~iW
c~II11,~ImIII\ 8, tl~c‘,“rnc~ \\. II lhC \CC,,II 01~ c~ln,,~~l-

ncn1\ IIl,~.U ,. illl<l I. 1hc Illilll~i\ ddincd I>\:

1t.‘t.1 II I.‘<1 (Ihi

tihhlllg ihccounl of lhc tilcl 1h:LI wc i~w nc1iv dr,;~lifig with


c~mplcy quanlilic5. ‘t~t~c~tirrli!n\i~~n~~l’lhc~~~n~plc\ ~niatrix.
t.‘t._ which ~n,u\t hc invcl.lcll. i\ I\,. I>! I\!.. nhcrc
.r,h) ‘I- / \ T‘ t,‘:\‘~’ \ T’ , t,‘:V<- 117)
N,. is Ihc nltnit,~~.otiti\cl~~l~ v;ltrle\ 01 t’o\~cr \vIbich LLL‘
pcl-lllm~ Ihc irllcyrmlic~n in / 10)~ ;A\ dixil3wd hcl~~i~. lhii
‘1 I, ~I~\ ( +~\,?/~“r-’ \ , i ,,2/\/c2’,“\ (,s)
is 115r1:1lly t>f the 01n1c.1,or 211. VI that invcrling t .‘I, i\
cq~~~v;dcnl I0 in\~cl-tin!: ;A0 lh) JO rc.i~l IlliillG\. it I;l\h I hill
i\ ~~~I~III~IIIC~~ very ~~llii‘icnlly in :in xw~!; prn,cc\\iw Fc>I~ t)cliGng the rcsidu:i/ vclcxil) VIMthy:
:I giwn \h,,l pr,,tit<:. sy\tar, ( Ihi ,,,,,‘.I the \C’I I,,> :,,,<I
,;\fI-‘ _ I;,!’ ,jL’<.’ ,I’)1
~1t~lvccl N,, lime\. v lhcw N,, i\ 111~II~IIII~CI~ oI’iti\cI.t.t~.
I’lrcqllcnc~ \.illllC\ 01 inlel.e\r.
i~ncl c\t~;In~tinrcctll;~li~lil ( I Xl in ;~Ta~Icwwric~ ill lt1\11~Tl.
tic gc,

‘l.ih! T ,,‘;2fv,~‘ I 1201

It’lldVr-f)~ ~~~ I. wearc.i~~s~ilic~l in dropping rhc highct~.


ordct~ 1cnus. .11111\. (0 lhc cxlenl lh;ll cqu:llion (201 is
INVEKSI: \‘,:I.OCI’I’Y YICKIZ~; 47
valid, WE can cxpcct the NMO-correcied events on the 2.This~cp~cscnts1lnNMO.corrcctedshot profile. Expe-
input sci:imogl-am 10 map 10 discrctc point\ on the plot ricncc has shown that. when pt-occssin$ split-qread
of UCp.v dcfincd by equakm CIO). As even,\ deviaw shot profiles. it is best to procc~ each Gdc of the spread
fromthadeal p;lraholicfortn.~ccnnexpectasmca~ing. scparatcly. Accordingly. thesynthetic in Figure? repre-
with a I-c:culkmt dcgr-adation in the ability to distinguish scnts one side of a split-spt-cad profile. Both primary
multiple from primary cncrgy. and multiple enel-gy can be xcn in a strongly intcl-fcring
pttU”.
Figure 2 dcmonstrakx the modelling intcrprctation of
MOIEI. RESLII.‘I’S
equation (IO) - the input profile is I-cprcscntcd as a
The application of equation I IO) will now he demon- weighted wrn of simple peraholx ofconstant amplilude.
stratcd WI the model data set shown on the left of I;igurc One might at first wonder how a combination of con-

Constant Velocity Stack W-L PI


11500 “ELcmTY(ftkx, 6500 8.7x 10-5 mecm, 1.54x 10

Fig. 2. In the modelling interpretation, each input NMO~corrected profile is represented as a linear combination of weighted parabolic event.5
h~mt unplilutc curves ~wltl possibly rept-cscn! t-c.;11 Offset
seismic data with ils variation\ in amplitude and -
w~avcfwm. The qucslion is cnlircly cqrliulcnt 10 thow a
series of unifwm sine waves can modct 2 compte~ time
xl-its in the ~ouricr ‘II-ansl’orm. The ~II\M’CI i\ lhill KC
use i, VC,-? lar~c numbc~- OS clo\~Iy rpaccd paraholac
which consli-uclively interfcl-c 10 prod~~ce the t-c\ulling
imx~e. Ceure 3 demonslrale~ the case of four pxaholic
curves Sor each of three z~~-o-ofl’x~ time ~mplc\. In
pl-aclicc. 20 10 30 curvc‘~ are used over wmc specified
~-:m~c of rnovc-o~~Is. With an X00-m\ dal:~ window al 1-
ms s:rmplc rate. this would mean that 40(1 Y 10 X000
pnraholac xc king used 10 model the profile of I$wrc
2.
t’igurc 4 showy rhc ICSIIII ofwlvin$ the Ic;~st-~q~~xcs
1~1
system ( 16) fwthis model. t’.ach puahola i\ piwim~c~cr-
i/cd by its ~cro-~~fl’~c~ lime :lnd ils .IIIOVC-out‘. or lime
P2 T2
diffcrcnce hctwcen ta-off\er and ~c~-~~-ollet lime: P3
T3
P4

In this cusc. the variahlc. p \\a\ disclKli/cd at 20


~alucs cot-responding Iomwc-ou(\ I-;m~in~ffl~uu IO rms
to + 200 ms. ‘l‘hc right hand side ofl;iguw 4 plots the co-
cfficicnts U(p. 7). III fxzl. cqultion C16) uik wlvcd at a
wxics 0fdiscrcIc li-cqucncic\ 1‘1~~17IOtt 10 1001 II and Fig. 3. For each ~ero-011settime. T. a range of paraboIae is used.
U(P.T) wax c:~lculated hy using equarion CI I). etkctively spanning the profile with a Iarge number 01 paraboIae.

Model Calculation
INPUT PROFILE W-L P)
x(W 200 145 89 34 -10 a p: Moveoutat ‘al
offset in ms

Fig. 4. Plot of parabolic coefficicnis Uip.i) caIcuIated for input model by using equation 116).
ISVEKSL VEL.O~‘ITY ST;\(‘K,r;<; 4’)

The transform U(p.~)cnn now be interpreted in much f.enerate artifacts analogous to the Gihb‘s phenomena
the ramc way as the constant-v&city stack of Figure I, well known in F-K t’iltcring. Although it has not been
except that ‘move-out replaccs velocity on lhc hwiron- tcs~cd. it should hc possible 10 reduce [his pl-ohlcm by
tal axis. Since the input model is NMO-corrected, we tapering between the pass and reject zones in the trans-
can cxpcc’l pr-inxwy entxgy lo lnap to events at around fwm domain. Fw these enamplcs a shwp cutoff wax
O-ms mow-WI. while undercorrccrcd mulliples should xlually used without any apparent ill effects. Setting to
map to higher move-outs. This has in fact taken pl;~~e. zero all coefficients with move-outs Icss than 30 ms
and primary energy can he discerned at 7X0 ms, 1000 produced the model result in Figure 6. Comparing it
ms, and I I IO ms. Strong multiples appcwat YX ms. YXO with Figwc 5. we can xc thal the pt-imary cvcnts at 780
m\, 1(1X0 ms, and below 1200 ms. The input model was ms. IWtl ms and I I Ill ms are largely misjing and the
calculated and NM0 corrected by using exact hypcl-- multiple cvcnts arc isolated. Subtracting the multiple
holic move-outs. w the shxpnesr ofthc imaged cncl-gy model from the input data produces the final multiplc-
on the Utp.7) plot is evidence (11 lhe validity of lhe attenuated rcstllt in the right panel of l+re 6.
parabolic ;approximation. The model result in Figul-c 6 confirms that the algo-
Having calcuk~tcd the cocfficicnts UCp.7). the for- rithm is XI-Y powclful at ~~rtcnuarin~llonppcl-iod multiple\.
ward model can now lx gencratcd by using equation A very satisf:xtory feature is that. where multiples
(IO) with nth,tI set to LCTO. This result is shown in have been attcnuatcd. the attenuation is equally good at
Figure 5. The left panel of the figure shows the ot-iginal all offscrs. This is in contrast to the ux of F-K filtering
synthetic data set, the cents-e pxnel shows the model form~~lripleelimination. where it iscommonlyohscrved
result, anJ the t-ight p;mcl shows the residwd erw that the srmd-offset 01 ‘near trace\ iire handled lew
calculated hy subtracting the model I’-om the input. As than satisfactorily. Thcrc is ofcourse nu mystel-y in the
expected. the model is a close approximation to rhc fact that these two algorithms petform differently on
input. In fact, it is cay to show that the residual crt-o~ this data set. ‘~l’hc inverse velocity stacking algorithm i\
could he reduced to LWC by using :I sufficiently Iargc based on a parabolic model and is ideally suited to
range of moYc-out wlucs p. modelling parabolic orapproximatcly pat-nholic cvrnts.
For the purpose of multiple elimim~tion. it is now The F-K algorithm is b:~scd on a plane-wave model and
ncccssary to product a model containing the multiple is ideally suited to modelling Iinewevents. The trajccto-
encl-gy only. This is done by ‘filtering’ the coefficient\ I-& of long-period multiple\ aftel- NMO-cwrcction XI
of Figure&that i\. setting a \~llxec ofthc cocSficicnl\ to the primary velocity at-e mwe closely approximated hy
zero. The procedure is directly comparable to filterins pxaholnc than by straight lines.
in the F-K~ domain, and one would expect that I-&\ ot Another important fcaturc ofthc algorithm is evident
thumb appropriate to that domain should he applicable in Figute 6: while the primarics should he NMO-corrected.
here. In particular. sharpcutoffs have been ohset-ved to it is not necessary that the corl-wtion hc exactly right.

MODEL OUTPUT RESIDUAL


INPUT ,p = -mm3 to 200m*, (INPUT- MODEL 0ulPul-J

Fig. 5. Input synthetic. model caIculated by applying iorward equation (101 ~~5thcoefficients displayed in previous figure. and residuaI noise
resulting from subtracting model from input data.
50 I). ,I,\111’SON

‘l-he cvcnt al IOOI~11,s i\ ohviouily und~~c~,l-t-~~~~(l, ‘fhc ‘I’hc lrigh! side of t;igurc 7 \how\ the IOI~ ~mmlcl
pcrli~rmance or the dgoriltml i\ in IW way dcgmdcd lh) g!cnualcci hy \l~lv~~r~eqri;llil)n (10) f’or~Ihi\ tl;~la wt. .Ihc
Ihis cmwas tang :I\ the clegrcc o111tlilerc~)l.l-ccti,,n is not diwl-ete WI of pal-ahotac uwd was iclcnlic:ll 10 thal in
such a\ 10 GIIISC the pr-imary (0 appwr in Lhc mut[iple the pwccding n~~del rc\utt. A timi~ed \~indow of input
/one: i.~.. ~reaie~~th;ln ill-ms inlovc-0LII. II ix now pcr\si- d;lt;l. fNlm 650 ,114 10 1500 111,. \\il\ ;In;lty/ed t<r KdllCC
htc (~1pcrt’o~-m II residual vcloci( anidyis on I~hcoulphl! ccm~ptc~- ruwlinlc. t’igrare 7 shows the 1.cwl1 ol’m~dct-
<>f~igurc 6 and refine the vclocit! CSIIIII:~~~ wflhout Ihc ling hclth primal-y and multiple cm!-g!. A\ cxpl;~incd
strong inlcrfcl-ins muttiptc~. ahow. each dc d’lhc ymxd i\ nrdellcd icdcpcndcnlly
A Sinal poinl \houtd hc ndc ahoul lhc ;dgorilhnl. which lm~li~cc\ 2 dighI diswnlinuily itt lhc ccnln~ of
which is nc,l p:~r~icul:~rly ohviow in I;igwc 6. Since the Ihc ~n~odcl protitc. ‘1’0 ev:~lu:~le how well thih model tit?
L~matt-ixdcllncd hycquation( t Slc~~l,t;lin~l”zci~cinl;)l-~ 1hc inpul dilti~. 17iglla~cX \tlon\ lhc rc5utI ~~l‘~i~hllacting
nmation cmwxning1hc 11111nhc1~am1 di\ll-ihulim ~I~‘~I~‘~‘YCI\. one Ir~om !he t,lher. Ohviolr\ly. ;~ny wg!mc‘nt~ ~~Iwllec-
no xtifkc!\ will he genera~cd cithw h) the houndarics IOP 01~coh\,rcnl energy that the eye perce~\c\ as signal
ofthc model or- hy uneven imcc >pxinp. Thi\ i\ anolt]cI~ on lhc input prolilc have lhcer~ mdcllcd. iI\ cvidcnccd
feature in contuse b\ith 1:-K algorilhm\. which a\sumc hv lhc (<)I;11lath of cohcrcnl Ggnal within lhc ;maty+
periodic h0uncla1.y i~clndilions :md Ilniti)ml \p;lcing. It n’indou of the r-csidual lp~~~l’ile 01 I;iylw X. M’h:l1 nou
makes the pxahulic ;~lgor-ilhm lpavticululy clllr:ictivc hti~nclh out is lhc I~~u-frcq~~cncy gl-c~uncl roll. which has
for u\c 011 cl.(,oke~l~\pl~~;,,l or 1-I) d:11;1 \c,\. IXXII ctl’cctivcly ignored th! Ihc ;Ilgorilhcn heca~~x dims
di\\imil;ll-il); 10 rhc rricxlclliug p;\m:ltxllac.
l:or ihi\ pal.ticuI:ll. dxt:~ xI. Ihe ml~t~iplc cu-ol’~ ha\
hecn SCI a! HO 111snic~vc-c~ul. and lhc mllttiptc xflzn~~i~~cd
RIL\l, I):\,‘:\ til:sr~l.l~s
vc\utt i\ \ho\vn i/l Fig~l-u ‘I. ‘I’he ni:~.j,rt.illlllliple event :II
‘l’hc invc‘rsc vcl<>cily SI:IC~ :dgorilhnl will he :lpplicd IO?0 I,,\ hii\ lhccn Iwgely :~ftcnui~tcd. ill have \oinc
to two real data sets. :I shot pl-olitc t’rom the first data \haltc,wcl- ~m~~l~iplc~ ;I, ;m~~~nd 900 in\. ‘I‘IIc pl~inl;ir)
XI i\ shcnvn on lhc tct’t of t’ifllrc 7. ‘I’lli\ pr~rfitc i\ c\cnI iit I llll~~~~sh;~\ lhccn l~~~cwr~ed;~nd ill t’zlcl cnh;!nced
N MO-correcled :md lhc /WIG ol’inlercst CCIIIICY :II.OIIII~ h); xuppl-coxing lhc inlerfcrin~ rmb~lliple enel-gi) ;,n the 1;u
lhe rctlector at ahout I I00 imh. .A sIl-c,ng multiple with o(tsels. II appc.i~bihi~l Irc\idllill mrllliplccncrg) I~C~V.C.CII
ahout IOIl ,I,\ rc\idu;d II,,IVC-o,,, appear\ ill ;I /cl-c~-olt\cl I IO0 and 1200 113sha3 hecn lell. p~w\r~~n;~hl) thix.:~lIx ilx.
lime ot’ 1020 m\ xnd cut\ thrc~ugh bhc tprimmy even\ at lmovco~~t i\ lcs\ th;ln X0 In>\.
an c,lT5~21II~ ;IppI-o*im;licli; Iu cl-third\ 01 Lhc spt~cid
length. .Thc \hot pl-ol.ilc II;IY had dcconvol~lli,>n ilnd
1lwl--3ul-face \t;liic cowcclionx appticd. and i5 f111ly p10-
ce~wd with the cxc<~plion c,Ilhi‘ fin:ll filler.

MULTIPLE MODEL FINAL OUTPUT


INPUT ,p = mm* to 200ms, (INPUT-MULTIPLE MODEL,
Ins

180

s*n

696

,ss

ISS

IBA

611

5BI

baa

Fig. 6. lnpul synlhetic, muiti~le model calculaled by applynq forward equatiur il Oj wttk coeff~r:~enls lb parabolai: with move-outs from 30 ms 10
200 ms. and rnul,iple a,,enuateil ,CS”!I
INVl-KSI,~ Vb, OC’, I Y S,~~\rh,Nc; il

common~dcpth~point gnthcl-> and stacking twx~ \~ith the \hot profiles as &XI-ihcd above. ;md after applymg
identical ,,rsinlil:11-\orlrcc-rcceivcl-ofl\cls. The res~111is the I:-K muttiplc elimination p~rnxlurc dRyu (IYXW IO
an’aven~fc‘conlmon-dcplh-poinl gather-when-ethe stxk- thcshot p~otilcs.Thctatterproccdul-econsistsofNMO-
ing hasenhanced the signal-lo-noix Iratio while p,-exl-\;- cwrecting the data at a velocity intemediare between
ing offset-dcpcndcnl chiwactcristic~. The thwe p;mcl\ that of lhc pl-imat-y and multiple vclocilics. imd uing an
of Figure IO I-especrively \hovv lhe common offset st:xk\ F-K fitlcr ro renwv~ down-dipping cncrg);. lloth procr-
over the SUTIL d:rta tracts before mulliplc ctimination. dures have heen very effective in \uppre\binf the strong
atic~~ applying the invcrsc docity stack atgot~irhm 10 IllLllliptc cvcnts hC1MCCIl 1000 md 1200 ,115. In I;lCl. 1hc

TOTAL MODEL
INPUT PROFILE (I) = -10mslo 20cms,

Fig. 7. In~sut real-data Shot profile ww ma, mode, calculated within data window from 650 ms to ,500 ms

RESIDUAL
INPUT PROFtLE (INPur TOTIlL MODEL,
<I Il. ,,,\\II’%iN

two mstltts are surpl-isingl!, 5Inlitiw ~11the t’zr ol’l’wt~ M hew the I:-K ;dgorilhm has lcfl significant nruttiplc
Mlh~I.Cpl.itlliil.?illldllillltiplCCIICI.~\ ~nte~~~~~cJc\il~ac~i\~~l\. cnc~gy. bhilc 111~InvcInc vclocilk 5lach ;,lg~,riihm ha5
‘lhc rn;(ior~ diffwcncc\ 0cc111~011 I the iwwoffw~ mui,\ tx~‘un etYccli\‘c at xl1 clff\cl\

FINAL RESULT
INPUT PROFILE

Common Offset Stacks


AFTER AFTER
INVERSE VELOCITY STACK F-K MULTIPLE
INPUT MULTIPLE ATTENUATION ATTENUATION
.~, ~.~.~...----&..&
I,,,, ..,.
,, ,4,:/
i,il::L:i:!,
“““:;q!gJg

~. .‘~:‘.!,.,.,.~?.J.;~~~ ~, ‘.~*Wi,..P,;&
__._~
),,) ~
,,“I. !?!T
:y;,,; ,,,, .,, ~, ,:~,;;w,:f:!;,
.,,/*:T,;,,:::I:; ~*~~!;:,:.>,,..W
“Bn..-e
,I,.t..~.*Pl.>t*,,,
-hmrrmnm
::::1,:,1::::::::.:::
,:;;t;T;,::::;::!i:i
.“;&~;~;;;:-~;~‘;
, ‘c::x:::<::JAx, f_m
,,,..,.,, ,, “*y!l!::
~..,.,,pe.......,,m..,
Il.,r**rm.,.*,.,.,,....
‘,” -‘~-~‘--.~-~~~;~~~~~~
,.,..,.**:,
I)...“l,l ,,,,)),.I ..*..,,,
,,I,,, ... “‘.,“‘),,:
,IIIutI,
2 .J.,umcw fi,,rur
.,,.., ,;I:!!.~!..!:.~‘;:~.~“““”
I.,,*... ,,,*,,,, ‘WW.I...
.l,..lllll
.,1..,+: ),)),.,,,.,,,~“,l,;l,i;!iLsguL
>,,,;I,i;~
,J;l~;;;;ii;j;,~,~,‘,,,,,,,,W,,~
,J;l~**launu,,i,~,‘,,,~,,,w,,-
e
‘;!:::::!:,:~.,:,,..,,,ru*,,~~~~~,~~~;~,
, ,),,) p.,,; )),,,, j,,>,...~l’,‘.‘~‘.‘,)
,I~‘~..,~,.~~,‘,.! ..,.... “n,,,,,I,,IIIL~~
“‘II! >l.,,.,I,,. .,,,.,,b>,,,.,,,,,),,.).,)(l,I,
,Y>1b,>~~ #,I,
- -.
S...c*rrttr*rnrl,.‘““~,“,~~,,,
<, .
bl~~w~~,,~b~, ‘)V “““’ .” ,*
.,.,.), ;,,b,,> .+TTyx$x~:~i~~
n,~,,!~.,,!~,:;!:.:~;,!,~,:l,.,!”””.”””
:.‘:I:,, :-....,,,,J!l~~~~,!~‘~:1:~,”

Fig. 10. Common dkcr viiwi :wr:x I’,%(,


,, i.(;,l’ ,,a,ii rx?,,Ae PiI,,,,,Jlf: ;ii,rn,la,,<m at,?, ,,,“e‘SP “Plocily 5111Ck
rllllllipk allenUa,i”l,. and aflel F-K
multiple aNenuamil
The second re;&data resmelt i\ shown in Figures I I to xrcxs of the zone of interest. which i\ consista with
14. The first two. Figures I I and 12. show comn~on Lhc changes olxcrvcd on the common ol’fset stacks of
offset stacks hcforc and ;ifter~ multiple eliminaiiow The Fig~11-c5 I I and 12. Expel-iencc has \hou.n Ihal Ihis level
zone ofintercst. from approximately 700 to 1000 my. is of inrpr”vement on the ~l:~chcd section i\ not achicvcd
strongly contaminated with undcrcorr-cctcd multiple in every cazc ~ stacking is itself a powerful multiple
energy as in the prrviou\ example. Again the inwnc attcnuxlion ~mcchani~n~. H~nrcvc~~. it i\clu~-that whcl-c
velocity stack :dgorithm M’:IS :lpplicd 10 :I limited time multiple\ d~minxle Ihe ~-ecol-ded seiw~ogl-am\ 10 the
window: from hO0 to 1200 ms in this cast. The ran~c 01 cxtcnt visihlc in IGgurc I I. an :dgwithm such as inverse
pat-abolic mwc-011t~ win ~51 fi-om IO ms 10 + 200 m5. velocity stacking can ?ubstanti;~lly impr-o\,c the imaging
with the cut-off move-out sepnraling mulliple (~-or pri- of weak pl.imar\; reflectors.
mary energy scl at + 30 ms. Cornpal-ing FiglIt-cs I I and Both the ~nodcl and twill dala cxamplcs have shown
IL!. WC xc a very significant change in the reilcc~n~ the pcl-form;mcc ofthc xlgwithm on \ho-wdered profiles.
quality as a rewlt of multiple attenuation. It should he Thcorctically Ihi\ approach should he applicable only
noted that each common ot&ct stack was sc:dcd with :I to plane-l:~ycux. rcrwdip d:~l:~. I’l-aclical experience
short wilndow ,ZC;c‘. This accoutres for the inwcarc in has shown that. in fact. the algorithm is quite rohus! in
amplitude ohscrved on pCmary cvcm~. particularly the pi-c\cncc ol’~~~odc~~a~c\~rw~:t~~rwl devi;llion\ from ihc
between 700 and YOO ms. ideal. :!s c:~n he wen in the previolls cxamples. One
Figures Ii and I4 compar-c the final stacked section reason SOILthis is that. lhy modelling the multiple and
hcl’orc and ilflcr multiple elimination. The pl-occaing subtracting it fl-om Ihc input I%rofitc. SII-IIC~~~I-~~ anonx-
tlow w::n identical in each USC except for the invcrsc lie\ \uch :IS I;~r!lls will lend 10 lx pw~rly modt~llcd :md
velocity stack step. which was applied to the shot PI-“- hence left in the data as noise. In wrnc (it-cumslanccs.
filcsjust hcforc final stack. The NMOand stilticscowcc- howc\;c~. ~I~n~cl~~~~iil;Inomillic,; like difl‘filction patterns
tions MCI-C idcntica. as UCI’C lhc twcc sculling\ hcforc may he modelled ah milltiplc\ and suppressed hy the
stack. The final stacked traces were xalcd hy deriving ;~lgwithm. In ~lvzxca~cs. I hi% ;Ilgol-ithm should hc applied
a single mulliplicr fat- each trace IO equlize the RMS in the C‘DI’ domain. whc.~.c alI pt-imar-!; c’nwgy should
ampliludle over the zone of inrere\t. III fenu;~l. MC WC a id~~~lly;~ppc;~l-a~~cl-~~-dipcv~nlsaftcl-n~~~~m~~lmo~c-or~t.
very signific:ml improvement in continuity 0\ic1 n~an)

)iBO
Zone of
Interest
‘,BB
54 ,i~ ,,\Al,‘>‘i;U

?nR~-7 TV--- ;:~r:;;,, “~~ ~y:~,---~-s~: :~:~~~I~ ~:;:,/;,,


:yb,:I.,, ~~ !, ,;,,i’,,,ic
!,!i y,!yr;~ :!!r!‘n ”
~BLI lJ~];,p +‘,~I
,**:,:l./,L>,/~.

7RR~__
z
BnB,,,’,,.,i‘P,~‘,~
v,yp,’ iih8
(I,,,,b
,,,q
r~o~g;,;>l&&!& 1vu
Zone of
Interest

Fig. 12. Sccmd rcal~~c!ata :;e: com:r:m ““3 L-k aim #,11,:‘~:,;:\-ol’v lb<’!S!RW lW,!d.,
d/se. a:, f 1)Jo~,,,.r,i,iit,,l~,

‘+f
HYI
Zone of
Interest
SilM

nun- +

iBI:

ina
4
Zone of
Interest

,188 +$+;i(!;e
: ;:,,,;,l;i;,ir~i*,.i),;,~‘,,,;,~ii,l; ,,,)I,,) ~,,!,.,“,;;‘:;,,::;,;,:;;i” / ‘))I”‘,“’ ’
~llil~l~‘~ill~!!~~~liI:~;~~ ,,,,,,,, * ,,,, , ,,...,,,.,,, ..,. .,., ,, ,.,ab.db ~, ,,,I ,‘~:,,!,,‘:L,I)‘,,,.‘I.Y,,” .!!,n,,,r,:,,,!!
‘)I ) I”“,‘:,“:‘;ilr..;;!l;t::;,l!i!~~~,~,!,,~~,!~~ “:,;;,;‘,!,,,,) ” (,lW.I,,,,,,
‘8,) bb’,,‘“‘,q)I,), *, II /be: ))
,i~o~:;.~l)l,~/~i~l,l,,,~“~~~~~“l~ ‘la+*, ,mm,,,,, ,,),,),, ‘,“ill’[~?,.!,, b’Lsill,,
‘I, ,,‘.i,i,.,,‘, #II .,!,fW,,

Fig. 14. Second real~data set stack after inverse velocity stack multiple alenuation

this paper. hut cxpcrience has xhown Ihat. Mhile vcrq


fine discrimination may bc achieved on ~modcl data. I-eal
This paper has shown how the Inverw Velocity Stack
data. with their variable amptitudcs and w~vcfol-ms and
algorithm u1‘Thot-son can be modified to incorporate :I
parabolic z~pproximation. and thus provide :I useful fool their additive ~wix. demand at least 30-111s mow-o~~~
St-orn nwr tr-xc to Sal- tr~c to he el’l’ccrivc.
for multiply.: climinatiw. Some ofthc features that ~makc
31 As with 1:-K filtering. \h:lrp cut-offs in the trans.
It particularly attractive we:
I) It achieves multiple attenuation at all offsets equally. Swm domain may pl-oducc arlifactr a\ a result ol’Gibh’r
2) It rcquir-cr no knuwlcdg ofthc mllltiplc~~cnci-ati~~~ pt1c110111c11011.
mechanism.
3) It requires no detailed knowledge of the multiple
and primat-y velocities.
4) It will attenuate 11 w#ide range of mulliplcs with
variable movco~~~s.
5) It will xcommodate vat-iahlc or nonunifwm acqu-
s11wn geometry. while minimiring ‘edge cl’l’ccts‘xssoci-
aled with lilnitetldala;~pertul-e. This m;~kesthealgorirhm
attractive lix application to 3-D data sets.
Limitntiotn associated with the met hod arc:
II Computation lime is significantly gvstct- than for
standard techniq\lc\.
2) Multiples must have xdficirnr IIIOVC-,wt di\cGln-
nalion to hu attenuated. This issue wx not xldt-essed in

Вам также может понравиться