technical article first break volume 26, January 2009 Practical issues in reverse time migration: true amplitude gathers, noise removal and harmonic source encoding Yu Zhang1* and James Sun2 Abstract Reverse time migration (RTM) is the method of choice for imaging complex subsurface structures. In this paper, we show that slightly modifying the conventional formulation, plus implementing an appropriate imaging condition, yields a true amplitude version of RTM that provides the correct amplitude-versus-angle relation. We also discuss different ways to suppress the migration artifacts and show how noise attenuation can be handled naturally in the common reflection angle domain. Finally, we introduce a harmonic source phase-encoding method to allow a relatively efficient delayed shot or plane wave reverse time migration. Taken together, these techniques yield a powerful true amplitude migration method that uses the complete two-way acoustic wave equation to image complex structures. Introduction Recently, reverse time migration (RTM) has drawn a lot of attention in the industry. Unlike one-way wave equation migration, RTM does not need to deal with the theory of singular pseudo-differential operators. A straightforward implementation of RTM correctly handles complex velocities and produces a complete set of acoustic waves (reflections, refractions, diffractions, multiples, and evanescent waves). The RTM propagator also carries the correct propagation amplitude and imposes no dip limitations on the image. In the past, strong migration artifacts and the intensive computational cost have been two major problems that prevented RTM from being used in production. In this paper, we first formulate RTM based on inversion theory and then we address some solutions to suppress the low frequency migration artifacts. At the end, we propose harmonic source migration as a way to improve the efficiency of RTM. True amplitude reverse time prestack depth migration We first discuss how to formulate a true amplitude RTM. To migrate a shot record Q(x, y; xs, ys; t), we have to compute the wavefield originating at the source location (xs, ys, zs = 0) and observed at the receiver locations (x, y, z = 0). Because the source wavefield expands as time increases and the recorded receiver wavefield is computed backwards in time, we denote them by pF and pB, respectively, in the following two-way wave equations: (1) and (2) where v=v(x, y, z) is the velocity, ƒ(t) is the source signature, and is the Laplacian operator. To obtain a common shot image with correct migration amplitude, we need to apply the deconvolution imaging condition (Zhang et al., 2005a), where is defined as the inverse of the wavefield i.e., in frequency domain (3) , (4) This imaging condition is simple to apply in the frequency domain for one-way wave equation migration. However, it is difficult to implement in the time domain for RTM. In practice, the cross-correlation imaging condition, (5) 1 CGGVeritas, 10300 Town Park Drive, Houston, TX 77072, USA. CGGVeritas, 9 Serangoon North Avenue 5, Singapore, Singapore 554531. *Corresponding author, E-mail: yu.zhang@cggveritas.com. 2 © 2009 EAGE www.firstbreak.org 29 technical article first break volume 26, January 2009 is often preferable for reasons of stability. Although this imaging condition does not appear to be consistent with true amplitude migration, Zhang et al. (2007a) proved that the imaging condition specified in Equation (5) is a proper choice to obtain true amplitude angle-domain common image gathers from wave equation-based migration provided that Equation (1) is modified accordingly as (6) Equation (6) is different from Equation (1), the conventional wave equation for the forward wavefield, because the source at the surface is treated as a boundary condition instead of a right hand side forcing term in the equation. An integral is applied to the source wavelet ƒ(t) to guarantee that the migrated phase and amplitude are correct. In summary, the following algorithm allows RTM to output true amplitude angle domain common image gathers from RTM: 1. Compute forward and backward wavefields pF and pB by solving the two-way wave equations, Equations (6) and (2). 2. Apply the cross-correlation imaging condition, Equation (5), during the migration. 3. Use an existing method, e.g., Sava and Fomel (2003), to output angle domain common image gathers. The migration output then provides the angle-dependent reflectivity in the sense of the high frequency approximation. To show how true amplitude angle domain RTM works, we apply it to a 2D horizontal reflector model. The input comprises shot records over five horizontal reflectors (Figure 1). The shot is at the centre of the section and the receivers cover the surface across an aperture of 15 km on each side of the shot. The amplitude variation across traveltime and lateral distance is due only to the geometrical spreading loss. We firstly migrated the shot records using the conventional common shot RTM algorithm, Equations (1) and (2), with the cross-correlation imaging condition, Equation (5). At an image location, we stacked all the migrated common image shot gathers to generate the subsurface offset gathers, and then converted them to the subsurface reflection angle gathers shown in Figure 2a. The peak amplitudes along the five migrated reflectors are shown in Figure 2b. We can see that in this RTM implementation, the migrated amplitude-versusangle curves are not correct. The amplitudes at far angles are overestimated, especially for the shallow reflectors. We then migrated the shot records using Equations (2) and (6) with the cross-correlation imaging condition, Equation (5). The angle domain common image gather is shown in Figure 3a and its normalized peak amplitudes along the reflectors are shown in Figure 3b. It is clear that the amplitudes in the A B Figure 1 Shot record over five horizontal reflectors in a medium with velocity v, in units of metres per second, given by v=(2000+0.3z), where z is the depth in metres. 30 Figure 2 (a) An angle domain common image gather from RTM using Equations (1) and (2). (b) The corresponding curves of amplitude versus angle. www.firstbreak.org © 2009 EAGE first break volume 26, January 2009 angle domain recover the reflectivity accurately over a large angular range, aside from the edge effects. Compared with the formulations of true amplitude Kirchhoff migration (Bleistein, 1987; Bleistein et al., 2001) and one-way wave equation migration (Zhang et al., 2005a, 2007a), implementing a true amplitude RTM algorithm is much simpler because the propagator itself naturally carries correct propagation amplitude, if we assume the observed seismic wave is well approximated by solving the acoustic wave equation. Therefore, from the amplitude-versus-angle point of view, RTM is superior to other existing migration methods and provides a stable and reliable way for seismic A technical article inversion. However, to take advantage of wide azimuth acquisition, we need to generate five-dimensional commonimage gathers from prestack depth migration, retaining both reflection angle and azimuth angle information. This leads to a dramatic increase in both computation and input/output costs, by two orders of magnitude, when compared to a stack output. Efficiently outputting 3D angle gathers from wide azimuth data remains a challenge to the industry. Noise removal from true amplitude migration point of view It has been observed that the conventional cross-correlation imaging condition of Equation (5) produces strong low frequency migration artifacts in RTM. Figure 4 shows a direct application of RTM to the 2004 BP 2D dataset (Billette and Brandsberg-Dahl, 2005). Migration artifacts appear mainly at shallow depths but also above strong reflectors, and severely mask the migrated structures. They are generated by the cross-correlation of reflections, backscattered waves, head waves, and diving waves. Figure 5 illustrates how the migration noise was generated. For any migration algorithm, if the sum of ts, the traveltime from the source to a subsurface location, and tr, the traveltime from the same location to the receiver, is equal to the recorded two-way traveltime t, then such a location is considered as a possible imaging point, i.e., a point where the reflection can occur (Figure 5a). However, due to the two-way propagation nature of RTM, it also generates reflections above a hard interface in the velocity model. The imaging relation (7) is satisfied for any point along the reflection raypath (Figure 5b). Therefore, in addition to the real reflection point at the hard interface, RTM also produces many unreal imaging points which give the low frequency migration artifacts (Figure 5c). B Figure 3 (a) An angle domain common image gather from RTM using Equations (6) and (2). (b) The corresponding curves of amplitude versus angle. © 2009 EAGE www.firstbreak.org Figure 4 Direct output of RTM applied to the 2004 BP 2D data set. 31 technical article A first break volume 26, January 2009 B C Figure 5 (a) For most of the migrations, the imaging relation ts + tr = t is used to generate an imaging point. (b) The propagation in RTM also generates reflections above a hard interface in the velocity model. As a result, the imaging relation ts + tr = t is satisfied for any point (white triangle) along the reflection ray path. (c) In addition to the real reflection point (black triangle), RTM also produces many unreal imaging points (white triangles) which give the low frequency migration artifacts. A B Figure 6 (a) Some angle domain common image gathers from the 2004 BP dataset. No artifacts show up on 0° to 60° gathers. (b) A stacked image for 2004 BP 2D dataset using reflection angles of 0° to 60°. 32 The first published work to suppress the artifacts is attributable to Baysal et al. (1984), who introduced a nonreflecting wave equation to remove the normal incidence reflected energy from an interface for post-stack depth migration. However, this technique is not effective for prestack depth RTM because the underlying mechanisms of noise generation are different. Other techniques have been proposed in the literature, such as a velocity smoothing high-pass filter (Mulder and Plessix, 2003), Poynting vectors (Yoon et al., 2004), and a directional damping term at the interface (Fletcher et al., 2005). In practice, we find they are either difficult to implement properly or have the drawbacks of distorting the spectrum or amplitude of the migrated images undesirably. Liu et al. (2007) proposed a new imaging condition to address the problem: decompose the wavefields into one-way components and only crosscorrelate the wave components that occur as reflections. In 3D, fully decomposing the wavefield into eight directions (up-left-forward, up-left-backward, up-right-forward, upright-backward, down-left-forward, down-left-backward, down-right-forward and down-right-backward) is computationally intensive, while an incomplete decomposition may inadvertently remove some steeply dipping reflectors in complex structures. Here we point out that suppressing migration artifacts is simple if we output angle gathers. The migration artifacts have the common feature that the source wavefield correlates to the receiver wavefield propagating in the opposite direction, which implies that the reflection angle is 90º. Therefore the artifacts can be removed by stacking the migrated angle gathers with a far angle mute (Figure 6). While outputting angle gathers is still expensive, a simple and popular way to remove the migration artifacts is to apply the Laplacian filter to the stacked migrated image. It removes the migration artifacts effectively with- www.firstbreak.org © 2009 EAGE technical article first break volume 26, January 2009 the number of shots, thus reducing the project cycle time and cost. Combining shots by line source synthesis in the inline direction (delayed shot migration) or in both inline and crossline directions (plane-wave migration) produces satisfactory results if enough p-values are used (Whitmore, 1995; Duquet et al., 2001; Liu et al., 2002; Zhang et al., 2005b; Etgen, 2005). One may consider applying similar techniques to RTM to improve its efficiency. Taking delayed shot migration for example, we need to apply a τ-p transform to the input data to synthesize the line source response, i.e., A B Figure 7 (a) A proposed processing flow to remove migration artifacts without distorting the spectrum or amplitude in the migrated image. (b) The image after application of a Laplacian filter plus proper pre- and post-migration processing, as suggested in (a). For migrations performed in the frequency domain, the time delays in delayed shot migration can be implemented as phase shifts, avoiding the need for time padding of the input traces. Since we perform RTM in the time domain, delayed shot RTM requires long time padding for long sail lines and/or large values of p, which can slow down the computation considerably. To avoid this problem, Zhang et al. (2007b) introduced a new phase-encoding scheme, called harmonic source migration, which is theoretically equivalent to delayed shot migration but does not suffer from the long time padding problem. For harmonic source migration, the phase-encoding function in the time domain is out hurting steep dips. To see how this technique works, we recall the relation (8) where θ is the reflection angle and v is the local interval velocity. Equation (8) says that applying a Laplacian filter to the stacked image is equivalent to applying a cos2 θ weight to the angle gathers. According to Equation (8), to utilize this technique correctly without distorting the migrated spectrum and amplitude, we have to apply a 1/w2 filter to the input data and rescale the migration output by a v2 factor. The proposed processing flow is summarized in Figure 7a. Figure 7b shows the result of applying this technique to the 2004 BP 2D dataset. As we have discussed, such a technique is equivalent to stacking the common image gathers with an angle domain taper, although no output in the form of angle domain common image gathers is required. Delayed shot, plane-wave, and harmonic source RTM For common shot migration, the cost equals the cost of migrating a single shot times the number of shot migrations. Various approaches have been proposed to reduce © 2009 EAGE www.firstbreak.org (9) (10) so there is negligible time padding to apply during the spatial transform. The number of k-values can be determined by a similar analysis to the number of p-values in delayed shot migration (Zhang et al., 2005b, 2006). We have applied this migration to both one-way wave equation migration (Soubaras, 2006) and RTM. For a typical production project in Gulf of Mexico, the speed of harmonic source migration versus common shot migration could be greater by a factor of 2–3. Figure 8 compares the results of one-way wave equation migration to RTM, using harmonic source encoding in both cases, for a deep water dataset from Mississippi Canyon, Gulf of Mexico. In general, RTM gives better images of the steeply dipping salt flanks. The sediments underneath the salt overhangs are extended closer to the salt flank boundaries. We attribute these improvements mainly to the high angle or turning wave propagation absent in the one-way wave equation migration. Figure 9 shows another comparison, for a Gulf of Mexico dataset in the area of Keathley Canyon, between results from oneway wave equation migration and from RTM. Although there are no complicated salt bodies and steeply dipping salt boundaries, thanks to the more accurate propagators, RTM still gives superior images for the structures in the subsalt areas. 33 technical article first break volume 26, January 2009 A A B B Figure 8 A real data example from Mississippi Canyon, Gulf of Mexico, with images from (a) one-way wave equation migration, and (b) RTM. RTM gives better images of the steeply dipping salt flanks and sediment termination. Figure 9 Real data from Keathley Canyon, Gulf of Mexico, with images from (a) one-way wave equation migration, and (b) RTM. RTM better delineates the subsalt. Conclusions and Sam Gray for their help with this paper and CGGVeritas US Seismic Imaging for providing the real data examples. We have shown that by slightly modifying the formulations, reverse time migration can be calibrated as a stable seismic inversion technique and provides correct information on angle dependent reflectivity. We have also discussed different ways to suppress the low frequency migration artifacts, and proposed a processing flow to remove migration artifacts without distorting the migrated amplitude and spectrum. Finally, we have introduced a harmonic source phase-encoding method which allows a relatively efficient implementation of delayed shot or plane-wave RTM. Taken together, these yield a powerful true amplitude migration method that uses the complete two-way acoustic wave equation to image complex structures. References Baysal, E., Kosloff, D.D. and Sherwood, J.W.C. [1984] A two-way nonreflecting wave equation. Geophysics, 49, 132-141. Billette, F.J. and Brandsberg-Dahl, S. [2005] The 2004 BP velocity benchmark. 67th EAGE Conference & Exhibition, Extended Abstracts, B035. Bleistein, N. [1987] On the imaging of reflectors in the earth. Geophysics, 52, 931-942. Bleistein, N.J., Cohen, K. and Stockwell, A. [2001] Mathematics of Multi-Dimensional Seismic Inversion. Springer-Verlag, New York. Duquet, B., Lailly, P. and Ehinger, A. [2001] 3D plane wave migration Acknowledgements We acknowledge BP and Frederic Billette for providing the 2004 BP 2D synthetic dataset. We thank Tony Huang, Mike Wolf 34 of streamer data. 71st SEG Annual Meeting, 1033-1036. Etgen, J.T. [2005] How many angles do we really need for delayedshot migration? 75th SEG Annual Meeting, 1985-1988. www.firstbreak.org © 2009 EAGE technical article first break volume 26, January 2009 Fletcher, R.F., Fowler, P., Kitchenside, P. and Albertin, U. [2005] Suppressing artifacts in prestack reverse time migration. 75th SEG Annual Meeting, 2049-2051. Yoon, K., Marfurt, K.J. and Starr, W. [2004] Challenges in reverse-time migration. 74th SEG Annual Meeting, 1057-1060. Zhang, Y., Zhang, G. and Bleistein, N. [2005a] Theory of true amplitude Liu, F., Stolt, R.H., Hanson, D.W. and Day, R.S. [2002] Plane wave source composition: an accurate phase encoding scheme for prestack migration. 72nd SEG Annual Meeting, 1156-1159. one-way wave equations and true amplitude common-shot migration. Geophysics, 70, E1-10. Zhang, Y., Sun, J., Notfors, C., Gray, S., Chernis, L. and Young, J. [2005b] Liu, F., Zhang, G., Morton, S. and Leveille, J. [2007] Reverse-time migration using one-way wavefield imaging condition. 77th SEG Annual Meeting, Delayed-shot 3-D prestack depth migration. Geophysics, 70, E21-E28. Zhang, Y., Sun, J., Gray, S. and Young, J. [2006] Sampling issues in delayedshot migration and common-shot migration. 68th EAGE Conference & 2170-2174. Mulder, W.A. and Plessix, R.-E. [2003] One-way and two-way wave equation migration. 73rd SEG Annual Meeting, 881-884. Exhibition, G012. Zhang, Y., Xu, S., Bleistein, N. and Zhang, G. [2007a] True amplitude angle Sava, P.C. and Fomel, S. [2003] Angle-domain common-image gathers by wavefield continuation methods. Geophysics, 68, 1065-1074. Soubaras, R. [2006] Modulate-shot migration. 76th SEG Annual Meeting, domain common image gathers from one-way wave equation migrations. Geophysics, 72, S49-58. Zhang, Y., Sun, J. and Gray, S. [2007b] Reverse-time migration: amplitude and implementation issues. 77th SEG Annual Meeting, 2145-2149. 2426-2429. Whitmore, N.D. [1995] An imaging hierarchy for common angle plane wave seismograms. PhD thesis, University of Tulsa. Received 24 September 2008; accepted 25 November 2008. 1/2 AD © 2009 EAGE www.firstbreak.org 35