Seismic Image Waves
Transcription
Seismic Image Waves
Geophys. J . Int. (1996) 125,431-442 Seismic image waves Peter Hubral,’ Martin Tyge12and Jorg Schleicher’?” ‘ Geophyrrc s institute, Karlsruhe Univerrrty, HertzstraJe 16,76187 Kurlsruhe, Germany Department ofilpplred Mathematics, IMECCIUNICAMP, C P 6065,13081-970 Campinar, SP, Brazil Accepted 1995 December 8. Received 1995 November 10; in original form 1995 July 20 Key words: seismic waves, seismic reflection imaging. INTRODUCTION The aim of this paper is to contribute to an interesting theory proposed by Goldin (method of discontinuities in seismic imaging; paper presented at the SvenFest, 1994 March 6-8, Golden, CO) with close reference to the work of Fomel (1994a, b). At the present stage, this theory is restricted to simple objects, such as homogeneous acoustic media. Nevertheless, it can be looked upon as a general approach in order to formulate partial differential equations (which we call image-wave equations) for various fundamental reflectionimaging problems. By introducing and making use of the concept of a seismic image wave, a geometrically appealing explanation for the general approach can be found. An image wave is not a physical wave but behaves like one. The concept is best understood with the help of Fig. 1. Fig. l(a) depicts the familiar situation of a propagating (physical) body wave. Three wavefronts of an elementary wave are shown at three different instants of time. In Fig. l(b), another familiar situation is shown; however, this is commonly not viewed as an example of wave propagation. We see three different (purely kinematically) migrated images of one seismic reflector, obtained with three different (constant) migration velocities. By comparison with the situation depicted in Fig. l(a), it is not difficult to accept that this can be conceptually understood *Now at: Department of Applied Mathematics, IMECC/UNICAMP, CP 6065, 13081-970 Campinas, SP, Brazil. 0 1996 RAS as a certain kind of ‘wave propagation’. In this case it is the image of the seismic reflector that ‘propagates’; we can call this phenomenon the ‘image wave’. In the same way as Fig. 1(a) depicts a physical wavefront at three different instants of time, Fig. l(b) can be said to depict an image wavefront at three different ‘instants of migration velocity’. Comparing Figs l(a) and (b), we see that they have much in common. Their common features are stressed in Fig. l(c), which is easily seen to be a generalization of the situations of Figs l(a) and (b). Fig. l(c) shows a set of curves in the (a, b) domain, where for each curve a parameter c is kept constant. Identifying the parameter c with the time variable t and the coordinates (a, h) with Cartesian space coordinates (x, z), we immediately have the same situation as depicted in Fig. l(a). O n the other hand, by identifying c with the migration velocity v and the coordinates (a, b) with (x, z), we observe the same situation as depicted in Fig. l(b). Not much imagination is necessary to accept that Fig. l(c) serves to describe many other seismic-reflection-imaging problems, some of which are described in detail in this paper. For example, by identifying c with v, but (a, b) with (x, t ) , we can interpret the same set of curves as time-migrated reflector images for three different migration velocities. As an important example of seismic image waves, we examine the depth- or time-migrated reflector image of a primary reflection [e.g. a seismic horizon in a zero-offset (ZO) or common-midpoint (CMP) stacked section] taken as a function of the (constant) migration velocity. When the 43 1 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 SUMMARY The concept of a seismic image wave is introduced and explained with the aid of some examples. Seismic reflector images in various domains (e.g. depth-migrated reflections in the depth domain or common-offset reflections in the time domain) behave like snapshots of elementary body waves. These ‘propagating’ images are thus referred to as ‘image waves’. The propagation variable, however, is now not time as it is for physical waves. It can be any other parameter involved in the seismic imaging process, for example the migration velocity or the common source-receiver offset. In parallel to the acoustic wave equation, which governs true elementary physical body waves, partial differential equations (here called the image-wave eikonal equation and the image-wave equation) can be derived that describe the propagation of image waves as a function of the problem-specific propagation variable. The concept of image waves is suited to solving a variety of different imaging problems. Image waves can be propagated, for example by a finite-difference or spectral-method algorithm. 432 P. Hubral, M . Tygel and J. Schleicher X X c v l J & This describes the propagation of an acoustic pressure wavefield p ( x , z, r) in a medium with velocity v, assumed to be constant throughout. A subscript x , z, or t denotes the partial derivative of p ( x , z, t ) with respect to the corresponding variable. The kihematics of acoustic waves (i.e. the propagation of wavefronts) are described by the eikonal equation v=v 3 1 T;+ T I = - c=c3 - Figure 1. (a) Propagating wavefront at three different instants of time. (b) Depth-migrated reflector images for three different migration velocities. (c) Graph showing Cartesian coordinate axes a and b, which define a suite of curves, where the parameter c is constant for each one. migration velocity is changed continuously, the migrated reflector image propagates in a similar manner to an elementary body wave. I n the case of a time migration, it propagates in the time domain; in the case of a depth migration it propagates in the depth domain. Different depth- or timemigrated images resulting from different migration velocities represent different ‘snapshots’ of this image wave. The construction of a variety of migrated image-wave snapshots is often desirable in order to study the influence of the migration velocity o n the reflector images (e.g. in connection with a velocity analysis o r in order to find the best possible images). Another example of an image wave, which we will investigate below, is the primary-reflection image in a common-offset (CO) section of a subsurface reflector below a constant-velocity overburden. This reflector image also propagates when changing the CO continuously. This latter type of propagation is implicitly performed when changing a CO into a simulated ZO seismic section. This change is often done either in one step by a migration-to-zero offset (MZO) or in two steps by applying the normal-moveout (NMO) followed by the dipmoveout (DMO) correction. For all kinds of seismic imaging problems, one can specify certain image waves. For each of them, one can then formulate a problem-specific image-wave equation. This describes the image-wave propagation as a function of the problem-specific propagation variable (which is the migration velocity or the 02 ’ This comes from the above acoustic wave equation ( 1 ) after inserting the ray-theoretical ansatz (3) Here f ( t ) is a high-frequency pulse that may slowly change its shape along the wavefront t = T ( x , z). Thus, the problem of propagating a given pressure wavefield [incorporating a set of individual elementary body waves of type (3)] in (x, z) space from the instant of time to to the instant of time t > to can be expressed symbolically by P ( X > z, t o ) P ( X 3 z, t ) ’ (4) In the seismic literature, a variety of techniques [e.g. finitedifference (FD) methods and spectral or integral techniques] exists to solve the initial-value problem (4), which will not be further discussed for that reason The main purpose of this paper is to show how differential equations of the form of the wave equation (1) and the eikonal equation (2) can be designed for various seismic-reflectionimaging problems. In the next section we define some of these problems and in the ensuing section we state and interpret the solutions. All mathematical derivations can be found in the Appendix. -+ SEISMIC I M A G I N G PROBLEMS In this section we consider various reflection-imaging problems, for each of which an initial-value problem analogous to eq. (4) can be formulated. Depth remigration The imaging process required in order to construct (from a given depth-migrated seismic section, including a set of arbitrarily complicated subsurface reflector images) depth-migrated seismic sections that would have resulted for a continuum of (constant) migration velocities is subsequently referred to as depth remigration. The term ‘remigration’ is chosen as a 0 1996 RAS, G J I 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 v = Y=V2 common offset in the above two cases). In connection with each image-wave equation there exist also (1) an image-wave eikonal equation (which propagates the image wavefronts) and (2) Huygens image waves, i.e. (hypothetical) elementary pointsource image waves that are ‘excited at each point of a propagating image wavefront. These two concepts are helpful in formulating the image-wave equation for the specific imaging problem of interest. Image-wave equations can be derived using a general approach. Before applying this approach to the examples mentioned above, let us explain it by returning to the familiar situation of a propagating (physical) body wave (Fig. la) described by the homogeneous acoustic wave equation Seismic image waves A comparison with eq. (3) reveals that the function v = V ( x ,z) can be considered as an ‘image-wave eikonal.’ Each elementary image wave p(x, z, u ) in eq. ( 5 ) propagates when continuously changing the migration velocity t), in the same way as the physical elementary wave p(x, z, t ) in eq. (3) propagates when continuously changing the time t. Snapshots of the image wave ( 5 ) in the (x, z ) domain at different instants of migration velocity u thus describe the selected depth-migration reflector image for the respective value of u. The problem of transforming the snapshot of the total image wavefield p(x, z, uo), involving many individual reflector images of type ( 5 ) , to the snapshot of the total image wavefield p(x, z, u) can then, by analogy with the initial-value problem (4), be symbolically expressed by Here, u, is some arbitrary constant initial migration velocity. As shown below, there exists for this particular imaging problem a partial differential equation (the image-wave equation) for the zero-offset or post-stack depth-migrated section p(x, z , u), with u as the independent propagation variable. Time remigration Let us now assume in Fig. l(c) that all parameters are as in the previous case, apart from the parameter b, which now represents time, i.e. b = t. Let us consider again a ZO (or poststack) situation, so that a curve implicitly given by u = e ( x , t ) in the time domain describes the image wavefront of a selected time-migrated reflector image (7) This reflector image again propagates like a body wave [now in ( x , t ) space] if the constant migration velocity u changes continuously. Note that although the migration velocity u may be identical to the one used in depth migration, the shape of the migrated reflector image will be different, and thus the image-wave eikonal P(x, t ) in the time domain will be different from V ( x ,z ) in the depth domain. The image wavefield p ( x , t, u) consists of many individual image waves of type (7) and the respective image-wave equation 0 1996 RAS, G J I 125, 431-442 describes the propagation P(X, t , u,) -.+P ( X > t, 0) (8) from the instant of uo to the instant of u. In other words, given (1) the (ZO or post-stack) time-migrated seismic section, p(x, t, u,), i.e. the snapshot of the total image wavefield (computed, say, for the wrong migration velocity u,) and ( 2 ) the particular image-wave equation, yet to be found, one can derive any other time-migrated section, p ( x , t, u), from p(x, t, u,) for a continuous range of migration velocities (for u > u , and/or u < u,). Migration to zero offset (MZO) This particular reflection-imaging problem combines the familiar N M O and D M O corrections into a single step. Let us consider Fig. l(c) with u = <, b = t and c = h, where h signifies half the common offset and 5 the midpoints of the shot-receiver pairs moved along the seismic line. Each curve in Fig. 1(c) describes, then, the reflection traveltime of a selected CO reflector image, resulting from the same depth reflector but recorded with a different offset h. The recorded image wavefield can be represented, as above, by (9) Each curve h = H(5, t ) is defined by a different constant value of h. To perform the MZO imaging process implies solving the following initial-value problem: P ( 5 , t, h = h,) --* p ( 5 , t, h = 0 ) . This can be understood as a propagation in (t,t ) space of the total image wavefield p ( & t, h), i.e. the CO section incorporating a set of (elementary) image waves of type (9), from the instant at which h = h, to the instant at which h = 0. In other words, for each value of h, the function p ( 5 , t , h) describes a snapshot of the total image wavefield in the (5, t ) domain. Note that this transformation depends on the (constant)medium velocity u, assumed to be known. Dip-moveout (DMO) correction As indicated before, the M Z O image transformation ( 10) can be subdivided into two familiar seismic imaging processes. These are the N M O and DMO corrections. Applying the NMO correction to a CO section p ( 5 , t, h = h,) involves performing the following time-domain stretching operation (which cannot be conceived of as an image-wave propagation): P(5, 4 h = ho) -+ 8(5,z,h = M . Here, the time stretch by T~ = t 2 - 4h2/u2, T (11) (i.e. the N M O correction) is defined (12) where u is the medium velocity. After this operation, the NMOcorrected CO section can be represented as consisting of elementary image waves To construct the desired simulated ZO section p(<, T, h = 0 ) = p(<, t , h = 0) from this section, i.e. the DMO correction, we Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 generalization of the term ‘residual migration’ (Rothman, Levin & Rocca 1985; Larner & Beasley 1987), which generally implies construction of a new depth-migrated image directly from an old one, by making a small change to the migration velocity. A remigration as discussed here, however, allows not only for one small change, but for a continuum of arbitrarily large changes of the migration velocity. Let us again consider Fig. 1( b), i.e. assume that in Fig. 1 (c) we have u = x , b = z , and that c represents the migration velocity u. The curves in Fig. l(c) describe, then, the locations of one searched-for subsurface reflector image, implicitly given by functions u = V ( x ,z), for three different values of u. These reflector images result from a depth migration of the reflection data acquired with a certain measurement configuration. We choose here a ZO configuration that is also representative for a CMP-stacked section. The signals attached to the migratedreflector location u = V ( x ,z ) can be interpreted as one particular elementary seismic image wave (i.e. a depth-migrated reflector image) described by 433 434 P. Hubral, M . Tygel and J . Schleicher have to solve the initial-value problem P(5,7, h = ho) P(5, 7, h = 0 ) . (14) + GENERAL APPROACH In this section, we describe how the image-wave equations for the above seismic imaging problems are derived by a procedure hereinafter referred to as the general approach. We start by using the ‘design’ of the acoustic wave equation (2), briefly presented in the Introduction, as an example. Let us outline the general approach with the help of formulas (1) to ( 3 ) .It is well known that the wave equation (1) provides, with the ray ansatz (3), the eikonal equation (2). Its solution for a homogeneous medium with velocity u, t = T ( x ,z)= to + V1 J ( x - - xoy + (z -z0y, (15) is nothing more than the kinematic description of a classical elementary Huygens wave (here simply called the Huygens wave). In (x,z) space, this is a circle with radius u ( t - to), centred at ( x o , zo). In (x,t ) space (e.g. for z = O ) , it is a hyperbola. Let us assume that a continuous set of Huygens waves originates at all points along a wavefront to = T(x,z). For this purpose, assume in Fig. l(c) that a = x , b = z and c = t. Then, according to Huygens principle, each wavefront t = T(x,z)for a constant value t > to is nothing more than the envelope of all elementary Huygens waves (15) placed into the wavefront to= T ( x ,z). The procedure described above represents the classical steps taken to arrive at the most fundamental results of geometrical seismics (optics) from the wave equation (1). This familiar procedure represents, on account of being physically justified, the forward process, or, metaphorically speaking, water flowing from the tap into the tub. But can the chain of steps described above be reversed, i.e. can eq. (1) be derived from eq. (15), thus making the water flow from the tub into the tap? The answer is, yes, it can be done. The basic step necessary in this respect is to find the eikonal equation (2) from formula (15). This is achieved by substituting the partial derivatives of the eikonal T(x, z), T, = VJfX (x- xo) x0y (z - zo)2 - + 3 r,= u J ( x ( z - zo) - x0y + (z- zo)2 ’ (16) back into formula (15). Now that the eikonal equation (2) is derived from the Huygens wave (15), we have to ask whether we can also obtain the wave equation (1) from the eikonal equation (2). This is indeed possible when certain assumptions are made. It is easy to accept (Goldin 1990) that any partial differential equation of the form Depth remigration To design the image-wave equation for this problem let us look at Fig. 2. It shows a continuous subsurface reflector X o , shown as a string of diamonds at the instant of the (incorrect) migration velocity uo. From each diamond an elementary Huygens image wave emerges, which is plotted at the instant of migration velocity u > uo. Admittedly, these Huygens image waves no longer resemble true classical Huygens waves, as they are not ‘surrounding’ their origins. What they have in common with true Huygens waves, however, is that they provide the subsurface reflector image C, i.e. the propagated image wavefront a t the instant of v as their common envelope. The envelope (bold curve in Fig. 2) will correspond to the target reflector if u corresponds to the exact migration velocity. The Huygens image waves are constructed as follows: points P(xo, zo) on the reflector Zo are (kinematically) ZO demigrated with the velocity uo. This provides a set of diffraction hyperbolas (also called Huygens curves) in the time domain of the form 1 Pxx + P z z - SP,,+ F ( P n P z * P t > P> x , z,t ) = 0 , (17) will give the same eikonal equation (2) as a result of ansatz (3), irrespective of the particular form of the function F , which depends on the listed arguments only. In other words, all wave equations of type (17) share the same kinematic behaviour of These curves have the ZO reflection-time curve (from which the reflector X o was constructed) as their common envelope. ZO migration of each diffraction hyperbola with the new migration velocity u provides a Huygens image wave that appears to have its origin at P(xo,zo) on the incorrectly 0 1996 RAS, G J I 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 Wave propagation wave propagation. The simplest of these equations is surely the one with F = 0. So, why not accept eq. (17) without the function F , i.e. choosing F = O? We will then have found an equation that kinematically propagates one or many elementary waves of the form (3), say, from the instant to to the instant t > t o , provided the initial propagation direction of all waves is also specified. In mathematical terms, we can then solve the initial-value problem (4) by involving the wave equation (17) in its simplest form, i.e. without the lowerorder terms. Note, however, that an image-wave equation ‘derived’ in this way will be, in principle, incorrect with respect to the propagation of amplitudes. To design a dynamically correct image-wave equation for a certain problem, one must also consider the corresponding transport equation. In this respect, the function F plays a relevant role. It is only accidental that for the true wave equation, F = O is also the dynamically correct choice. As we consider only kinematic properties of image-wave propagation, we have no criterion to decide whether this may also be the case for any of the image-wave equations derived below. The above design strategy of the wave equation has shown that reversing the classical chain of steps from eq. (1) via eq. (2) to eq. (15) enables one to ‘derive’ the wave equation (1) from the Huygens wave equation (15). It would appear that this inversion can be regarded only as a rough and ready method of arriving at the wave equation. However, if we regard the wave equation (1) as an image-wave equation, i.e. if we demand from it no more than the propagation of image waves with desirable kinematic properties (ignoring their dynamic characteristics), we are not doing anything illegal. With this understanding, the design procedure outlined above can now be applied to the aforementioned imaging problems. Seismic image waves 435 Offset-0 m,vo=3000m/s,v-3500m/s 0 I Q --- I I I I Wrongly migrated reflector C, Depth remigration Huygens image waves -200 -600 -400 -200 0 400 200 600 Distance [m] Figure 2. An incorrectly migrated subsurface reflector constructed for the migration velocity uo is indicated by a string of diamonds. From each diamond an exploding Huygens image wave emerges, which is shown at the instant u > uo. The envelope of all Huygens image waves forms the migrated target reflector (bold curve) constructed for the migration velocity u. migrated reflector image Zo. In fact, if v is continuously increased, such that v > vo, or continuously decreased, such that v < v,, the elementary Huygens image waves will propagate away from each point on the (frozen) reflector image C,. They behave just like the classical true Huygens waves, which would propagate away from a fixed true (frozen) wavefront to = T(x,z ) and form an advanced wavefront t = T(x,z) as their envelope for t > to. The Huygens image wave for depth remigration that emanates from point P ( x o ,zo) on Zo is given by z=-Jz;+-i-(x-x,)2 V vi for v # vo (see Appendix). It results from (kinematically) ZO migrating each point on the diffraction hyperbola (18) from the (x, t ) domain into the (x, z ) domain with the migration velocity v. This provides a set of lower half-circles in the (x, z) domain with formula (19) as the common envelope. We now replace the migration velocity v in eq. (19) by the image-wave eikonal V(x, z) and take partial derivatives of this modified equation with respect to x and z (see Appendix). Substituting these partial derivatives again into eq. (19) and performing an appropriate elimination, we obtain the imagewave eikonal equation V v:+ v;--v,=o. Z ( 20) The same image-wave eikonal equation obviously results from 0 1996 RAS, GJI 125, 431-442 V PXX + P z z + ZP”. =0 (21) with the ansatz (5). With the help of this equation, the initialvalue problem (6) can now be solved. For this purpose, the initial value p(x, z, vo) should be obtained beforehand by a depth migration with any algorithm (e.g. the algorithm of Stolt 1978) for the incorrect migration velocity vo. Time remigration v -v; VO the image-wave equation A point p ( x o ,to) in a time-migrated section p ( x , t , v), which was constructed from a ZO (or CMP-stack) section for the migration velocity v, relates to a point P ( x o , zo) in the corresponding depth-migrated section p ( x , z, v) by the following depth-to-time stretch relationship: (XO, 2z*/vot --* (XO, to)‘ (22) This allows the transformation of the depth-remigration Huygens image wave (19) from the (x, z) domain into the ( x , t ) domain, thus providing the time-remigration Huygens image wave 4(x - xo)2 t = Jt;+ v; - u2 In the same manner as before, from this Huygens image wave (23) (see Appendix) one can derive the time-remigration Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 -600 P. Hubral, M . Tygel and J . Schleicher 436 to < t , < t. It also describes an imploding Huygens wave for point P(x, z ) on t l = T ( x ,z) in the (x,,, z,,) space and in the time range t , > t > to. Thus a derivation of the eikonal equation image-wave eikonal equation t vef - 4 =0 , which describes the kinematics of image waves of type (7). The image-wave equation that propagates time-migrated reflector images of the form po(x, t ) f [ u - %(x, t ) ] directly in the (x, t ) domain, as a result of continuously changing the (constant) migration velocity v, is (Fomel 1994a) 4p,, + vtp,, =0. (25) Exploding versus imploding Huygens image waves All of the above remigration problems can be solved (see Appendix) by formulating the respective remigration Huygens image waves. These can be looked upon as generalizations of classical ‘exploding Huygens waves’, which we simply called Huygens waves above. In fact, the remigration problems could also be solved with ‘imploding Huygens image waves’, as well as for other elementary wavefronts. Let us define a n imploding Huygens wave in connection with the propagation of the true wavefront t = T(x, z ) controlled by the eikonal equation (2) in the range to < t < t l . The classical design strategy to propagate a wavefront is to place exploding Huygens waves of the type (15) into all points P(x,,, zo) of the frozen wavefront to = T(x,z ) and construct the new propagated wavefront t = T(x,z) as the envelope of all exploding Huygens waves (15) defined for t > to. However, the wavefront propagation in the indicated range could also be described by placing the origins of so-called imploding Huygens waves into all points P ( x , z ) of the advanced (frozen) wavefront t , = T ( x ,z). An imploding Huygens wave is then defined by formula (15) for t < to. In fact, imploding Huygens waves are simply constructed from exploding ones due to an inherent duality. Both are in fact described by one and the same formula. Eq. (15) describes an exploding Huygens wave for point P ( x o , z o ) on the wavefront to = T(x,z ) in the (x, z ) space and in the time range Migration to zero offset (MZO) Let ( and t denote the coordinates of a point P ( ( , t ) in the C O section. To solve the M Z O problem, let us first consider the inverse problem. This is the simulation of the C O section p ( ( , t, h = h,) from the Z O section p ( 5 , t , h = 0). For brevity, we will refer to this problem as a migration to CO (MCO). An exploding M C O Huygens image wave is identical to an imploding M Z O Huygens image wave. This latter wave, in 0 1996 RAS, G J I 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 Note that formula (25) differs from the corresponding equation derived by Fomel (1994a) by a factor 4 in the first term. This is because Fomel used V = v/2 instead of v in his work. Given a time-migrated section, p ( x , t, uo) say, for the (incorrect) migration velocity vo, all other time-migrated sections, p ( x , t, v), can be obtained by applying the image-wave equation (25) in order to solve the initial-value problem (8). One may expect that, by analogy to the depth remigration problem, a previously migrated image (obtained with a certain migration velocity uo) is necessary as an initial value. It is, however, interesting to observe that this is not the case. The ZO or CMP-stacked section can be taken to represent the time-migrated image wavefield for zero migration velocity, p ( x , t, uo = 0) (Chun & Jacewitz 1981 ). Hence, this section can serve directly as the initial value in the remigration imagewave equation (25). The initial-value problem ( 8 ) is then to transform the initial image wavefield p ( x , t , u,, = 0) directly into the desired image wavefield p ( x , r , u ) . In fact, by a suitable implementation (e.g. in the form of an FD scheme), p(x, t, v) will not only be provided for one single value of v, but also for the whole range of migration velocities from u,, = 0 up to a certain limiting value. In this way, a n optimal migration velocity u can be found. Application of the stretch relationship (22) to p ( x , t, u ) provides the depth-migrated image p ( x , z, v) in a second step. (2) with the help of an imploding Huygens wave would formally be achieved by taking the partial derivatives of the eikonal T ( x ,z ) in eq. (15) with respect to x,, and z, not with respect to x and z. What has been said with respect to the duality of the true Huygens wave (15) also applies to any Huygens image wave. For instance, the depth-remigration Huygens image wave ( 19) propagates in the variables (x, z, v) away from point P(xo,z,,), i.e. it is exploding. For a point P ( x , z), it is imploding in the variables (x,,, yo, v). Imploding (image) Huygens waves can, therefore, be just as useful in arriving at the (image-)wave equation as exploding ones. In this connection, we wish to emphasize one interesting aspect involving imploding and exploding Huygens waves. This aspect involves their role in the relationship between a forward and an inverse seismic-imaging problem. Let us refer to all imaging problems discussed above (which can be looked upon as transforming an input image into an output image) as direct imagiug problems. Then, solving the respective inverse problem means transforming the output image back into the input image. It is easily observed that the roles that exploding and imploding Huygens (image) waves assume in a forward problem are exchanged when addressing the inverse problem. In other words, exploding Huygens waves turn into imploding ones, and vice versa. Imploding and exploding Huygens image waves are very characteristic functions of the specific seismic-reflectionimaging problem considered. In fact, in previous works (Hubral, Schleicher & Tygel 1996; Tygel, Schleicher & Hubral 1996) the importance of the duality that exists between imploding and exploding Huygens image waves has already been emphasised, without, however, referring to them as such, and without recognizing their potential for deriving image-wave equations. In the cited works (where similar problems to those presented here were solved in a 3-D laterally inhomogeneous medium with the help of integral operators), the snapshot of an exploding Huygens image wave was called an outplanat and the snapshot of a n imploding Huygens image wave, a n inplanat. Inplanats, then, describe the stacking curves in problem-specific Kirchhoff-type integral operators, while outplanats describe the corresponding smear-stack curves. In the MZO problem, considered in the next section, we will make use of the duality of Huygens image waves for the development of the image-wave eikonal equations. To solve the M Z O problem, we will use imploding Huygens image waves. However, we find it easier to construct exploding Huygens image waves for the respective inverse problem. From the discussion above, both types of Huygens image waves are identical. Seismic image waves Substituting the image wavefront h = H ( & t ) into eq. (26), taking partial derivatives with respect to [ and t, and performing the elimination procedure (see Appendix) leads to the MZO image-wave eikonal equation ( tH l + - ui4H : )-( t 2 + -24): Ht-tHH:=O, which, by ansatz (9), also results from the MZO-wave equation (Fomel 1994b) This, in turn, can be employed to solve the initial-value problem (10). DMO correction Suppose that an NMO correction has been applied to the CO section p(x, t, h = h,) according to formula (11). What needs to be done to obtain the ZO section p(x, t , h = 0) from the NMO-corrected CO section p(x, z, h = h,) is called the DMO correction. The exploding Huygens image wave for point P(C,, to) of the inverse DMO problem is z=t, J 1-- (5 This is obtained by applying the NMO stretching operation ( 1 2 ) to each point P(5, t ) of the exploding MCO (or imploding MZO) Huygens image wave ( 2 6 ) . The (velocity-independent) NMO image-wave eikonai equation for the image-wave eikonal h = &([, z) is then found from eq. ( 2 9 ) as (see Appendix) HH? -H + zH, = 0 . (30) This image-wave eikonal equation corresponds to the DMO image-wave equation (see also Fomel 1994b) hp”<<- hp“hh- Zp”hr = 0 . which is used to solve the initial-value problem (14). (31) Offset- 1OOOm, Velocity- 3500 m/s 500 I I I 1 1 I 400 600 400 w 300 Q E Y !i F 200 100 --- Imploding MZO Huygens image waves Q 0 -600 Zero-offset reflection-time c u m 400 -200 0 200 Distance [m] Figure 3. A reflection-timecurve in a zero-offset section is indicated by a string of diamonds. An imploding Huygens image wave, which is shown at the instant h > h, = 0, pertains to each diamond. The envelope of all Huygens image waves forms the common-offset reflection-timecurve (bold curve) with offset h. 0 1996 RAS, GJI 125,431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 turn, serves as a point of departure to design the searched-for MZO image-wave eikonal and image-wave equation. To visualize imploding Huygens image waves with the help of an example, we have constructed Fig. 3. This shows a continuous ZO (h = 0) reflection-time curve, To, as a string of diamonds. For each point on this string, exploding MCO (i.e. imploding MZO) Huygens image waves are displayed (dashed curves) at the instant h = h,. They provide the CO reflection-time curve (bold curve) as their common envelope. The exploding MCO Huygens image wave for a point P(5,, to) on the ZO reflection-time curve, h = H ( l , t o ) = 0, is constructed as follows. First, migrate point P(5,, t o ) into the (x, z ) domain with the ZO configuration. This provides a lower half-circle with centre at ({,,O) and radius vt,/2 for point P(co, to). Then, demigrate this isochron with the CO configuration, defined by the value h, by treating it as a reflector. This provides the exploding MCO Huygens image wave (see Appendix) in the variables l and t at the instant of h, 437 438 P. Hubral, M . Tygel and J . Schleicher C O N C L U D I N G REMARKS + ACKNOWLEDGMENTS The authors thank Sergey Goldin for making his presentation at the SvenFest, which was the main source of inspiration for this work, available to them, and for many fruitful discussions with respect to the subject. We also thank Makky S. Jaya, and F. Liptow for useful discussions. The research has been supported in part by the Commission of the European Union in the framework of the JOULE 11 programme, by the Deutsche Forschungsgemeinschaft (DFG, Germany), by the National Council of Technology and Development (CNPq, Brazil), and the Silo Paulo State Research Foundation (FAPESP, Brazil). Jorg Schleicher gratefully acknowledges additional support by the Alexander von Humboldt foundation in the framework of the Feodor Lynen programme. The responsibility for the content remains with the authors. This is Karlsruhe University, Geophysical Institute, Publication No. 677. REFERENCES Bleistein, N., 1990. Born DMO revisited, 60th Ann. Int. Mtg., SOC. Expl. Geophys., Expanded Abstracts, 1366-1 369. Chun, J.H. & Jacewitz, C.A., 1981. Fundamentals of frequency domain migration, Geophysics, 46, 717-733. Fomel, S.B., 1994a. Method of velocity continuation in the problem of seismic time migration, Russian Geol. Geophys., 35, 100-111. Fomel, S.B., 1994b. Kinematically equivalent differential operators for offset continuation of seismic sections, Russian Geol. Geophys., 35, 146-160. Fomel, S.B., 1994c. Amplitude preserving offset continuation in I: The offset continuation equation, Stanford theory-Part Exploration Project, Rept 84, 1-18. Goldin, S.V., 1990. A geometrical approach to seismic processing: The method of discontinuities, Stanford Exploration Project, Rept 67, 171-209. Hubral, P., Schleicher, J. & Tygel, M., 1996. A unified approach to 3-D seismic reflection imaging-Part I: Basic concepts, Geophysics, in press. Lamer, K. & Beasley, C., 1987. Cascaded migration: Improving the accuracy of finite-difference migration, Geophysics, 52, 618-643. Rothrnan, D.H., Levin, S.A. & Rocca, F., 1985. Residual migration: Applications and limitations, Geophysics, 50, 110-126. Stolt, R.H., 1978. Migration by Fourier transform, Geophysics, 43, 23-48. Tygel, M. & Hubral, P., 1989. Constant velocity migration in the various guises of plane-wave theory, Suru. Geophys., 10, 331-348. 0 1996 RAS, GJ1 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 In this paper, we have provided a conceptual foundation for deriving differential equations for seismic-reflection-imaging problems. Some examples of such differential equations describing time remigration and offset continuation have already been discussed by Fomel ( 1994a, b). Differential equations of this form can be solved numerically, in a fully analogous way to the wave equation, by means of integral, spectral, o r F D techniques. By introducing the concept of a seismic image wave, together with the corresponding image-wave eikonal equation, imagewave equation, and exploding and imploding Huygens image waves, we have provided a rigorous design strategy, by which such differential equations, termed image-wave equations, can be derived for a variety of imaging problems. In order to explain the newly introduced concepts, we related them to classical ones, connected with the familiar acoustic-wave equation. In a very natural and systematic way, we were able to employ the new strategy to design very useful new partial differential equations for a variety of seismic-reflection-imaging problems. We believe that with the help of our geometrically motivated concepts, all the formal mathematical considerations, as found in Fomel (1994a, b), are now much easier to understand. As the derivations in this paper are based on kinematic consideration only, no criterion is provided with respect to the dynamic qualities of the image-wave equations obtained. Note, however, that Fomel (1994~)proves that the M Z O imagewave equation (28) is also dynamically correct, i.e. it leads to an amplitude-preserving MZO, in the sense of Bleistein (1990) and Tygel et al. (1996). The way in which a n amplitudepreserving time-migration (by which reflection coefficients can be estimated) can be achieved with the help of eq.(25) is discussed in Fomel ( 1994a); the topic of amplitude-preserving imaging has not been addressed in this work. It has been shown that the point of departure to derive problem-specific image-wave equations involves an analytical formulation of the problem-specific Huygens image wavefronts. The formulation of a Huygens image wave requires no more than chaining a purely kinematic migration and demigration of a point in either the (x, t ) or (x, z ) domain. In this two-step sequence, different measurement configurations o r velocities (or possibly even some anisotropy parameter) can be considered in each step. In this way, a problem-specific propagation variable can be defined that describes the propagation of image waves. Of course, the specification of a Huygens image wave may not always lead to a n operationally simple image-wave eikonal or image-wave equation. In such cases, one has to find valid approximations that, in turn, will yield simple approximate image-wave equations, the latter being suitable only within certain parameter ranges. There are, of course, other important concepts (like rays, ray amplitudes, caustics, etc.) related to image-wave equations. These concepts can also be considered to be generalizations of familiar ones connected with the wave equation (1) and eikonal equation (2). For instance, there exist harmonic plane image waves that, if substituted into the image-wave equations, lead t o an image-wave dispersion relation. These would be a generalization of the dispersion relation, which for the case of the wave equation ( 1) is k; k; = w2/v2,where k, and k, are the components of the wave-number vector and w is the circular frequency. Starting from harmonic planar image waves, one can formulate solutions for the above initial-value problems in the spectral domain (Fomel 1994a) by following, for example, a similar strategy to that described in Tygel & Hubral (1989) in connection with a time-to-depth migration based on the classical wave equation (1). How to generalize the concept of image waves to inhomogeneous media remains an unsolved problem. At first sight, the restriction t o constant velocity seems intrinsic to the concept. However, at least in media consisting of constantvelocity layers, an application of the remigration image-wave equations seems possible using a layer-stripping approach. Moreover, and most intriguingly, even a reflector image below an inhomogeneous overburden behaves formally like a wavefront if the overburden is continuously changed, even though we do not know how to construct an appropriate image-wave equation that describes this propagation at present. Seismic image waves The envelope of this ensemble is the desired exploding Huygens image wave at the instant of migration velocity v. The condition for the envelope curve is Tygel, M., Schleicher, 3. & Hubral, P., 1996. A unified approach to 3-D seismic reflection imaging-Part IT: Theory, Geophysics,in press. A P P E N D I X A: DERIVATION OF THE IMAGE-WAVE EIKONAL EQUATIONS aF -=0, a5 This Appendix is devoted to the derivation of all image-wave eikonal equations dealt with in this paper, as derived from Huygens image waves. For the remigration problem we use an exploding Huygens image wave, and for the MZO problem we use an imploding Huygens image wave. which, after some algebra, yields the stationary 5= In order to get a better understanding of this function, we have constructed snapshots of this exploding Huygens image wave for point P(x,,, zo) in the ( x , z ) domain for different velocities u in Fig. Al. It is now our aim to find the imagewave eikonal eq. (4) for the depth-remigration image-wave eikonal u = V ( x , z ) . For that purpose, we substitute this expression into eq. fA6) and differentiate both sides of eq. (A6f with respect to x and z. Setting, for convenience, the simplifying notation, M ( x , z ) = V ( x ,z)/vo, we find Then, we proceed to depth migrate this curve using the same ZO configuration, but a different constant migration velocity, v, i.e. v # uo. To accomplish this task, we initially construct the isochrons for each point, M = M ( [ , T), on the Huygens curve (A1). These isochrons are the lower half-circles O=M,{[z6+= r=2V JW. + Squaring and equating expressions (Al) and (A2) yields the following ensemble of lower half-circle isochons, parametrized by 5: + z2 x - (dvo)2x,, 1 - (v/vo)’ ’ -(V/V~)’[(X,, - 5)’ + z;] = 0 . (x - x,,)’ 1 l’’ (1-M’)’ (x - Xo)’ - (A3) I’ x [kml 0.5 1 / / 1.5 2 2.5 3 3.5 \ Figure A l . Snapshots of Huygens image waves lor depth remigration. The source point is at x = 2 km, z = 2 km. The original migration velocity was uo = 2.5 km. Curves with increasing maxima at x = 2 km are Huygens image waves for decreasing velocities (u = 2.4 km s-’ to u = 1.8 km s - l in steps of 100 m s-I), and curves with decreasing minima are Huygens image waves for increasing velocities (u = 2.6 km s-’ to u = 3.2 km s-’ in steps of 100 m s-l). All displayed Huygens image waves are exploding ones for a remigration from u,, to u or imploding ones for a remigration from u to uo. 0 1996 RAS, GJI 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 Let us consider a point Mo(xo,z o )in the (x, z)domain. Locating the coincident source-and-receiver pairs of a ZO configuration o n the x-axis at x = 5 we demigrate the point Ma with the constant migration velocity v,, to find the so-called Huygens curve t = T(5)in the (5, t ) domain, 5 ) = (x-5)’ 5 value which, if substituted back into eq. (As), yields the exploding Huygens image wave Depth remigration F(x, Z , 439 P. Hubral, M . Tygel and J. Schleicher 440 ( 1 ) specify a point P(t0,to) in the ZO section (i.e. the input space); (2) construct its ZO depth-migrated image; (3) CO demigrate each point on this image with the CO configuration into the (5, t) space (i.e. the output space) and find the resulting envelope. and 1 =M , { [z: + + ( x - Xo)Z 1 (1 - M2)’ ZO depth-migration of point Po(<,, to) yields the image (i.e. the half circle) Since from eq. (A6), 1 t = - (’4’12 which, if substituted into eq. (AS) yields, after a little algebra, the eikonal equation for M ( x , z), Returning to V ( x , z)= v o M ( x , z ) , we find the desired imagewave eikonal equation (20). Time remigration 4(x - xo)2 u:- vz ’ Replacing the velocity u in eq. (A12) by the image-wave eikonal P(x, t), and taking, as before, partial derivatives with respect to the variables x and t, we find, [ t;+ V;- P2 and [ - ‘iZ 4(x P2 4(x - xo)2 1 = &C t; 1 11 4 ( x - X0)Z 0;- + 4 v;( x - xPZo y - - Xo)Z x-XO uo - - v; - &2 ’ u24(x - x o y (u: - P Z ) Z . C‘ Squaring of eq. (A15) and further substitution of the result into eq. (A14) yields the time-remigration image-wave eikonal equation fi v: - 4 (‘418) where A , = z2+ ( x - 5 h)’. A cascaded migration and demigration is realized by insertion of (A17) into (A18). This leads to 1 t =-(BY u +Byz), (A191 where B , is given by ’>O ;( +(to-5 T h)’+ 2(~-50)(50- 5 T h). (A20) The envelope of all functions (A19) for all points on the half circle (A17) is given by atlax = 0. From this equality, we can determine and eliminate the parameter x in eq. (AIO). We find Inserting expression (A21) into (A19) finally provides the exploding MCO Huygens image wave in the (t,t ) space, which, at the instant h, is given by 4 (x - x o ) - -tK 2 + A !p), (0; - t Z ) 2 The combination of the above three equations immediately yields the intermediate result 2=,- V B+ = - Substitution of z = vt/2 and zo = u0t0/2 into expression (A6) leads to the exploding time-remigration Huygens image wave o = VV- - ( x - toy. CO demigration of a point P(x, z ) on this image gives x-XO z M, ~ - M ~ - M ~ M , ’ t = Jt;+ J( = =0 . (A161 In agreement with the discussion above, formula (A22) describes an exploding MCO Huygens image wave in ( 5 , t ) space, when going from h = O to h = h,. It is equal to the imploding MZO Huygens image wave, which we require in (5, t) space, when going from h = h, to h = 0. In Fig. A2 we have constructed exploding MCO (imploding MZO) Huygens image waves for a point P(C0, to) at different instants of h. The next step in our aim to derive the MZO image-wave eikonal equation (11) consists of inserting the image-wave front h = H ( 5 , t) into eq. (A22) and taking derivatives with respect to 5 and t . Prior to doing this, we introduce, for convenience, in eq. (A22), the notations B = hZ - ( 5 -to)’, and (A23b) Migration to zero offset (MZO) To arrive at the image-wave equation for the MZO problem, we have to proceed as follows. First, we construct the imploding MZO Huygens image wave (i.e. the exploding MCO Huygens image wave) by the following steps: (A23a) from which we observe that (A23c) 0 1996 RAS, G J l 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 z we have from eqs (AS) and (A7), Seismic image waves 441 2.3 t [SI 7 20 -- 10 - 5 2.05 [kTi h=lOOm i 1.4 1.6 1.8 2 5 2.2 2.4 2.6 [kml Figure A2. Snapshots of imploding MZO Huygens image waves. The source point is at 5 = 2 km, t = 2 s. The Huygens image waves pertain to offsets or h = 100 m to h = 700 m in steps of 100 m. In the upper right corner, the region near the apices is enlarged. 1.95 1.4 1.6 2.2 2 1.8 5 2.05 2.4 2.6 [kml Figure A3. Snapshots of imploding DMO Huygens image waves. The source point is at 5 = 2 km, t = 2 s. The Huygens image waves pertain to offsets of h = 100 m to h = 700 m in steps of 100 m. In the upper right corner, the region near the apices is enlarged. Note the difference between this and Fig. A2, indicating the effect of the NMO correction. We again insert h = H ( [ , t ) into the above formulae and take the partial derivatives of eq. ( A 2 2 ) with respect to 4 and t. We find 2 2H 1 uto 1 l=-HA--U ' v A(2)SHHt' 0 1996 RAS, G J I 125,431-442 and 2 2H 1 vt, 1 2H 1 vt, 0 = - H A--- -HHI+--(~) v v A ( 2 ) B 2 v A * 1 ~(t-t,,). Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 1.95 P. Hubral, M . Tygel and J . Schleicher 442 From eq. (A24) we obtain, upon the use of eqs (A23), H, = section z = t o J 1 - (to - 5)2/h2, 1 t H - -[ t 2 - ( ~ H / v ) ~ ] H Bt (A311 where z is the NMO corrected traveltime defined by so that B= t(l H , - H ) Correspondingly, from eq. (A25),we obtain H<= -(2H/U)2)(5- 4 0 ) t H - - -[ t 2 - (2H/u)2] H Bt -(tZ (A281 Substituting B from eq. (A27) into formula (A28),we arrive at the following expression for 5 - to: 4-50=- -H~H, tH,-H. and where W, 1s given according to the above by We finally use eq. (A23a)together with eqs (A27)and (A29)to arrive a t the M Z O image-wave eikonal equation H;- 4 H; + ( t2 + ( 2H/u)2)H , = 1 . Ht The corresponding MZO image-wave equation (27) results from this. Eq. (A33)can be readily inverted for 4 - To; we find _5 - _5 0_ _- 1 B H,‘ Using t = to/WN,from eq. (A34) we obtain DMO correction To solve the DMO problem, we again use the imploding M Z O Huygens image wave given in eq. (A22).However, as the initial condition for a DMO correction is represented by a fixed point P N with coordinates (5, z), we have in the NMO corrected where we have made use of eq. (A36).By comparing eqs (A35) and (A37), we immediately obtain the DMO image-wave eikonal equation -tH, = AH; - A . (A381 0 1996 RAS, G J I 125, 431-442 Downloaded from http://gji.oxfordjournals.org/ at BIBLIOTECA CENTRAL UNIV ESTADUAL CAMPINAS on August 20, 2014 where t is the CO traveltime, as before. Eq. (A31) describes imploding DMO Huygens image waves which are depicted in Fig. A3. Inserting h = A(<,z) into formula (A31) and partially differentiating the resulting equation with respect to ( and z leads to H , H ~ ( ~ ~ - ( ~ H / ~ ) ~ )