Layer | Fill | Outline |
---|
Map layers
Theme | Visible | Selectable | Appearance | Zoom Range (now: 0) |
---|
Fill | Stroke |
---|---|
Collaborating Authors
Results
Speeding Up RTM Velocity Model Building Beyond Algorithmics
Ortigosa, Francisco (Repsol) | Liao, Qingbo (Repsol) | Cai, Wenying (Repsol) | Guitton, Antoine (3DGeo Inc.)
Summary. In this paper we show the unique role of Reverse Time Migration (RTM) as tool for Velocity Model Building (VMB). But for this new use of RTM we need a relative speed up of two orders of magnitude in processing. We discuss how to achieve this target beyond algorithms. Introduction Reverse Time Migration (RTM) is becoming the new standard for imaging the complex subsurface of the Gulf of Mexico (GOM). RTM yields the best possible images because it is based on the solution of the two-way acoustic wave-equation. This technique relies on the velocity model to image turning waves and prismatic waves. These turning and prismatic waves are particularly important to unravel subsalt reservoirs and delineate salt-flanks. RTM opens new frontiers in designing better velocity estimation methods, including the use of recursive iterative RTM itself as a trial an error procedure; RTM is the most accurate way to image simultaneously turning and prismatic waves. RTM and Accuracy of Velocity Models We tested the validity of RTM as a VMB with several synthetic models. The most traumatic example is depicted in Figure 1. We removed completely the salt pillar from the model, and after RTM the reflections from the salt pillar, although in the wrong position, are still imaged in the processed section. This is because these reflections are coming from the data. RTM VMB is both model and data driven. RTM acceleration . One way to accelerate the RTM is through hardware implementation. We are accelerating RTM by porting the code to the new generation of Cell /B.E. Processors (Ortigosa el al. 2008). The RTM code discussed in the cited paper performs at 83.2 GFlops in a IBM QS20 Blade (Figure 3), almost 8x faster than in a IBM JS21 Blade or almost 15x faster than in a AMD Opteron 270 blade. Another way accelerating the RTM is decimating the dataset and increasing computing grid spacing. Shot decimation. We have observed empirically that the quality of an RTM image does not increase linearly with the percentage of the shots processed (Figure 2). In fact, the quality increase is asymptotic, and therefore for VMB purposes data may well be decimated. Processing 20% of the shots represents a relative speed up per shot of 5x. Increasing Computing Grid Spacing.Figure 4 shows from empirical observations that relative speed up increases exponentially with increasing data spacing for a given aperture. In our experience, tuning these parameters for VMB purposes, an 2x performance increase is reasonably achievable. RTM Velocity Model Building. Table I shows the combination of feasible performance increments resulting from individual performance increment achieved individually from hardware, shot decimation and data spacing. The cumulative speed up discussed above ranges from x30 to x100. If we assume an inefficient generic RTM algorithm with a performance of 10 hours per shot and per AMD Opteron 64 dual core blade, the performance increment results in 5 to 20 minutes per shot per QS20 blade.
- North America > United States (0.37)
- North America > Mexico (0.25)
- Geophysics > Seismic Surveying > Seismic Modeling > Velocity Modeling (1.00)
- Geophysics > Seismic Surveying > Seismic Processing > Seismic Migration (0.95)
Summary A workflow combining (1) plane-wave migration (PWM) in tilted-coordinates and (2) tomography with an automatic picking scheme can reduce turnaround time while producing accurate velocity updates. PWM in tiltedcoordinates can image steeply dipping events which can help better constraining the velocity model and better defining the salt geometry. Introduction In tomography, reducing the turnaround time for each velocity iteration remains a major challenge. Ideally, we would like to have a fast and accurate migration technique combined with a fast and accurate tomography. Then, the interpreter can spend more time QCing the results and make geophysical decisions. With this in mind, we propose a workflow combining plane-wave migration (PWM) in tilted-coordinates and a tomographic scheme with automatic volume-based picking. PWM has long been recognized as a valuable technique to help building velocity models (Whitmore and Garing, 1993; Ji, 1997; Jiao, 2001): PWM can be very cost-effective compared to shotprofile or source-receiver migrations. Gathers indexed as a function of the surface-ray parameter exhibit a moveout whenever the velocity is not correct. These gathers (that we call p-gathers) are generated naturally at no extra cost within the migration process. Fewer plane-waves can be used during the velocity iterations than during the final migration. On top of these properties, PWM in tilted-coordinates can help imaging steeply dipping events, and thus provide better images in complex media and better velocity models (Etgen, 2002; Shan and Biondi, 2007). For the tomography, we propose using an automatic volume-based picking scheme. This approach selects the reflection points used for (1) the backprojection raypaths and (2) the residual velocities. With this approach, the picking becomes less labor intensive and less biased. It also uses much more data than is commonly used in manual picking approaches, thus increasing the robustness of the inversion (Bevc et al., 2006a; Bevc et al., 2006b). We present our workflow in the next sections and will present inversion results using the BP benchmark data (Billette and Brandsberg Dhal, 2005). PWM in tilted-coordinates PWM in tilted-coordinates is a fast and accurate imaging technique suitable for complex geology (Shan and Biondi, 2007). Whereas PWM utilizes a Cartesian mesh that does not change for each plane-wave, PWM in tilted-coordinates rotates the propagation grid according to the surface-ray parameter of the plane-wave such that the direction of propagation and extrapolation are closer to each other. This property allows the imaging of turning event when one-way propagators are used for the wavefield extrapolation. To illustrate this property, we show in Figure 1 the result of PWM in tilted-coordinates for the BP model. The salt flanks are well imaged. This property can help us better defining the salt geometry and better constraining the velocity model by using information coming from steeplydipping events. Tomography For migration velocity analysis, residual moveout in the pgathers is parameterized as a residual velocity by fitting a reflection event in a gather by semblance analysis (Jiao, 2001). The velocity model is then updated by raytracing tomography applying a global approach to calculating the residual traveltime (Ji, 1995).