You are on page 1of 4

First-break tomography for near-surface velocity model building Tianfei Zhu* (Veritas GeoServices Ltd.

) 715 Fifth Avenue SW, Suite 2200, Calgary, Alberta T2P 5A2 Tianfei_Zhu@veritasdgc.com Scott Cheadle, Allan Petrella and Sam Gray (Veritas GeoServices Ltd.) Introduction Determining near-surface velocity structures is a crucial first step in seismic data processing and imaging. An inadequate near-surface velocity model will result not only in incorrect statics corrections, but also in deviation of rays predicted by the model from their actual raypaths. As this raypath deviation is cumulative, a modest inaccuracy in the near-surface velocity model can introduce large errors in both raypath and traveltime calculation, and can significantly deteriorate the seismic image at depth, especially in the areas with large lateral velocity variations. Many methods have been developed for near-surface velocity determination, primarily for the purpose of statics solutions (e.g., Marsden, 1993). Most of these are refraction methods, largely due to the fact that the shallowest portion of a seismic record is often dominated by source-generated noise, and accurate identification of reflections is difficult. The first arrivals, on the other hand, can be clearly identified and often represent the best data available for near-surface velocity estimation. The traditional refraction methods, however, have a number of limitations in velocity model building: These methods assume that the near-surface structure can be represented by a layered model, and first arrivals can be treated as refractions from the models interfaces. As a result of these simplifying assumptions, the refraction methods cannot model vertical velocity variations in each layer, and fail to accommodate strong lateral velocity variations. By treating first arrivals as refractions, they are also unable to determine first layer velocity v0. This results in a trade-off between interface depths and layer velocities in a derived velocity model, making it inadequate for imaging processes such as depth migration which are more sensitive to local velocity variations than the statics solution is. To overcome the limitations of the traditional refraction methods, we have developed a tomographic inversion method for determining near-surface velocity structures from first arrivals. Based on a newly developed grid raytracing technique, the method is robust and capable of recovering velocity structures in geologically complex areas. In this presentation, we first describe the inversion method and then demonstrate its applications with synthetic and real data. Tomographic inversion method Model and Formulation Different from the traditional refraction methods, the velocity structure is represented in our inversion method by a grid model, with each node of the grid assigned a node velocity. As the grid spacing is small and the node velocities can vary in an arbitrary fashion, the method can model strong velocity variations in both vertical and horizontal directions. Also, first arrivals are now treated as direct body waves propagating along turning rays, enabling the method to determine the first layer velocity as well. The velocity inversion is formulated as an iterative, constrained, least-squares problem, and the node velocities are determined by minimizing the differences between the observed traveltimes of first breaks and those predicted by the grid model. As the inversion method requires intensive raytracing at each iteration, an accurate and efficient algorithm for traveltime and raypath calculation is essential for practical applications; the algorithm must also be robust and devoid of the shadow-zone problem which can severely reduce the number of observations usable in a tomographic calculation. Grid raytracing (GRT) method A number of traveltime calculation methods have been developed over the past decade to avoid the shadow-zone problem. These include the wavefront construction (e.g., Vinje et

First-break tomography al., 1993), and fast marching (e.g., Popovici and Sethian, 1997) methods. The wavefront construction methods are accurate in describing both traveltimes and raypaths, but require expensive global wavefront construction and traveltime interpolation from these wavefronts to grid points. Employing a fast wavefront-tracking algorithm, on the other hand, the fast marching methods are efficient, but their accuracy is unsatisfactory for modeling first arrivals. We have recently developed a grid raytracing (GRT) method that combines the advantages of both wavefront construction and fast marching methods (Zhu and Cheadle, 1999). The method calculates both traveltimes and wave propagation vectors by tracing rays locally within a grid cell and consists of two major components: wavefront tracking and local traveltime and propagation vector updating. Our algorithm for wavefront tracking is similar to that used by Popovici and Sethian (1997): Traveltimes and propagation vectors at nodes immediately surrounding a given source are first computed by assuming straight raypaths; traveltimes at remaining nodes are then assigned an infinite value. After this initialization, the entire nodes are divided into three groups: accepted, in-band, and untimed. The accepted group consists of the nodes which have been used as propagation points to update their neighboring nodes, and the untimed group of those which have not been touched by wave propagation. Separating these two groups is a narrow band of in-band nodes, which have been touched by wave propagation but are yet to be used as propagation points. As the wavefront intersects the narrow band at the node with the smallest traveltime, wavefront tracking reduces simply to finding from this band the minimum-time node. This, in turn, can be carried out by a highly efficient heap sorting technique. After the minimum-time node is identified, it is accepted and used to construct a local wavefront for updating traveltimes and propagation vectors at its immediate neighboring nodes. Only those nodes belonging to untimed and in-band groups will be updated. As the dimension of grid cells is small, and the wavefront can be approximated locally by a spherical surface. The center of this local spherical wavefront can be determined using traveltimes and propagation vectors of the minimum-time node and a neighboring node selected from the accepted or inband groups. For illustration, Figure 1 shows a simple 2D model where a local circular wavefront is constructed using the minimum-time node A and neighboring node B. The neighboring nodes D and F are to be updated. To compute the traveltime at node D, for example, we first locate the intersection point E between the local wavefront and the radius from center C to the node, and then compute the traveltime by t = t0 + 0.5d(1/vE + 1/vD)

Figure 1. A schematic diagram illustrating local wavefront construction for a 2D velocity model

where t0 is the traveltime of the local wavefront, vD and vE are the velocities at points D and E, and d is the distance between the two points. If the computed traveltime is smaller than the nodes previously assigned time value, this value is updated and the exit propagation vector of the node is calculated by local grid raytracing. In the case where the updated node is previously untimed, it is removed from this group and added to the in-band group stored in the heap structure. This tracking and updating process continues until all the nodes are timed. A comparison of the GRT method with a 2D wavefront construction and a fast marching method shows that the GRT is about 8 times faster than the wavefront construction method and about two orders of magnitude more accurate than the fast marching method (Zhu and Cheadle, 1999). The efficiency can be further enhanced by a combination of the GRT and fast marching method. Our numerical experiments also show that the GRT is robust and capable of modeling turning rays in complex media.

First-break tomography Data examples The first-break tomographic inversion method has been successfully applied to both synthetic and real data to estimate near-surface velocity structures for statics calculation and depth migration. We present in Figures 2 and 3 the results from a synthetic data study. Figure 2a shows the shallow 1400 m section of a 2D Foothills velocity model. The entire model is about 32.6 km long and 10 km deep. Seismic response from this model was simulated by a finite-difference method with 252 surface sources, each recorded by up to 150 receivers on either side of the source. The source and receiver intervals are, respectively, 100 and 25 m. A total of 75,836 first arrivals were picked from the synthetic data for near-surface velocity estimation. The velocity was first estimated with a conventional refraction method with the first-layer velocity v0 set to its exact value of 2500 m/s. The resulting velocity model is displayed in Figure 2b; it has been smoothed with a 100 m long Hanning window in both vertical and horizontal directions. This smoothed model is used as an initial model for the tomographic inversion described in the previous section; the grid model consists of 1305 x 56 cells with a grid spacing of 25 m in both vertical and horizontal directions. The final velocity model is shown in Figure 2c, which is obtained after five iterations of tomographic calculation. A comparison of Figures 2c and 2a shows a close agreement between the inverted and true velocity model, even in some small details, indicating that our method is indeed capable of recovering near-surface velocity structure in geological complex areas. To demonstrate the effects of near-surface velocity model on prestack depth migration (PSDM) of the synthetic data, we constructed two velocity models by merging the near-surface models from Figures 2b and 2c respectively with the exact velocity model at depth 1400 m. Figure 3a shows the PSDM section obtained with the exact velocity model while Figures 3b and 3c display respectively the sections produced by the models constructed with the near-surface models from Figures 2b and 2c. The results clearly show that inaccuracy in a near-surface velocity model can severely deteriorate the seismic image at depth and that the velocity model determined by our tomographic inversion method has resulted in significant improvements in depth imaging. Conclusions We have developed a tomographic method for inverting first breaks of seismic records for near-surface velocity structures. Based on a newly developed grid raytracing technique, the method is efficient and can model large variations in both velocity and topography. Experiments with synthetic and real data have shown that near-surface velocity model building is a crucial first step in seismic data processing and imaging, and that velocity models estimated by our method have resulted in significant improvements over the traditional refraction methods for both time statics and depth imaging. Acknowledgments We thank Amoco Canada for the data set used in our data examples. References Marsden, D., 1993, Static corrections a review, Part 2: The Leading Edge, 12, No. 2, 115 120. Popovici, A. M., and Sethian, J. A., 1997, Three dimensional traveltime computation using the fast marching method, 67 th Annual Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 1778-1781. Vinje, V., Iversen E., and Gjoystdal H., 1993, Traveltime and amplitude estimation using wavefront construction, Geophysics, 58, 1157-1166. Zhu, T. and Cheadle, S., 1999, A grid raytracing method for near-surface traveltime modeling, 69th Annual internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, 1759 1763.

First-break tomography Biographic note for the presenter Tianfei Zhu obtained his Ph.D. in geophysics from Cornell University in 1986 and M.Sc. in Computer Science from University of Toronto in 1994. He worked as a Research Seismologist/Lecturer at University of Toronto from 1988 to 1997 and jointed Veritas GeoServices as a Sr. Research Geophysicist in 1997. His research interests include signal processing, wave propagation, and seismic tomography.
Figure 2a. Foothills velocity model used for generating synthetic data. Only the top 1400 m of the model is shown. Figure 2b. Near-surface velocity model obtained with a refraction method. The model is smoothed and used as the initial model for the tomographic inversion. Figure 2c. Final velocity model from tomographic inversion.
Figure 2 (a)

Figure 2 (b) Figure 2 (c)

Figure 3a. PSDM obtained with the exact velocity model. Figure 3b. PSDM obtained with the near-surface velocity model from Figure 2b merged with the exact model below 1400 m. Figure 3c. PSDM obtained with the near-surface velocity model from Figure 2c merged with the exact model below 1400 m.
Figure 3 (a)

Figure 3 (c ) Figure 3 (b)

You might also like