Obstacles in the analysis of azimuth information from prestack seismic data Anat Canning* and Alex Malkin, Paradigm.

Transcription

1 Obstacles in the analysis of azimuth information from prestack seismic data Anat Canning* and Alex Malkin, Paradigm. Summary The azimuth information derived from prestack seismic data at target layers is very sensitive to many aspects of the analysis process, making reliability a serious issue. This is true for both AVAZ and VVAZ analysis. In this paper we analyze some of the major factors affecting the reliability of the results and show our approach to dealing with the main obstacles. Introduction Azimuthal seismic analysis is becoming increasingly important, due to the growing interest in unconventional shale plays, where stress direction and facture orientation are the most sought-after information. The challenge is to extract maximum information about the nature of the azimuthal variations from wide-azimuth, migrated, prestack data. In this paper we refer to azimuthal variations which can be characterized by three parameters ( 1, 2, ) and (G1,G2, ). 1 and 2 are the primary axis of a velocity ellipse, G1 and G2 are the primary axis of an AVAZ gradient ellipse, and is the orientation angle (Grechka and Tsvankinm, 1998; Ruger, 1998). The source of these azimuthal variations can be HTI, TTI or orthorhombic anisotropy. Obviously, there are specific layers, typically carbonates or shales, which exhibit such azimuthal variations and are of particular interest. The common practice is to generate a horizon oriented map for the target layers displaying attributes of interest, e.g. orientation angle, ( 1-2), similar to the Eagle Ford example displayed in Figure 1. The workflow for extracting azimuthal information along target layers normally involves picking the geometry of the horizon and displaying one or more of the above azimuthal anisotropy parameters as the color value of the horizon map. This produces horizon maps such as the one displayed in Figure 1. The main problem with this method is reliability. Like many geoscientists who perform these workflows, we noticed that the picture map obtained from such an analysis is very sensitive to workflow and parameter changes, and one can easily obtain very different pictures from the same data. We are used to methodologies in which establishing the quality and robustness of the result (e.g. velocity analysis or migrations), is relatively straightforward. However, analyzing attributes along a horizon oriented map does not provide clear insight that can help users establish the reliability of their data. This statement is generally true for every horizon oriented attribute map, but in our case, the lack of such clear insight starts with the data itself, even before it is extracted onto a horizon slice. This becomes the main obstacle in the analysis of azimuthal information and is dangerous because it may lead to the interpreter depending on unreliable data. Figure 1: AVAZ azimuthal intensity along Eagle Ford shale layer displayed in combination with a structural attribute. Our main claim in this paper is that the azimuth information from prestack seismic data at target layers is very sensitive to many aspects of the analysis process, making reliability a serious issue. On the other hand, analyzing the reliability of the results is not a natural aspect of this workflow, at least not the way it is commonly performed in the industry today, so the interpreter is unaware of the reliability issues involved. The anisotropic seismic phenomena we are discussing here are both Velocity versus Azimuth (VVAZ) and Amplitude versus Azimuth (AVAZ). Theory and Method Horizon tracking: The first question we ask is, Does the order of the operations matter? Does it matter if we first extract prestack information along the target horizon and then perform azimuthal anisotropy analysis, or should we first perform the analysis, calculate the anisotropic parameter on the full seismic volume, and then extract those parameters onto the horizon map? This may seem like a trivial question, but its implication for the workflow is very significant. It seems that it is most common to use the first approach and perform the azimuthal anisotropy analysis directly along the target horizons or along a narrow window around them. We have taken a different approach: We first process the full volume and only then extract information along the target horizon. For VVAZ analysis the differences can be significant, for a number of reasons: 1) It is more robust to stabilize residual NMO when analyzing the full geological column, 2) interpretation of a horizon on the stacked data does not correctly represent the offset/angle azimuth aspect of the geometry of that horizon, and 3) the amplitude extraction SEG New Orleans Annual Meeting Page 427

2 process can be greatly improved when done postmortem, when various averaging or tracking algorithms can be tested, without having to redo the entire workflow. Figure 2 compares the two approaches and shows dramatic differences in the results, supporting our claim that reliability is a serious factor in the workflows discussed here. a) Full-volume VVAZ followed b) VVAZ on data limited by attribute extractions by horizon Figure 2: Horizon maps of velocity anisotropy intensity for a shale layer. Each map represents a different workflow applied to the same 3D gathers. 3D gathers vs. sectored data: While it is common to use a limited number of sectors to perform azimuthal anisotropy analysis, we normally use full-azimuth 3D reflection angle gathers (Koren and Ravve, 2011). These two practices may result in significantly different outputs. From 3D gathers we can create any sectors and test the effect of sectoring. Figure 3 compares maps of (azimuthal anisotropy intensity measured from VVAZ Analysis) created from full-azimuth 3D gathers with a dataset composed of five angle sectors. The dark color marks the unreliable zones which result from our practice of including the analysis of reliability in the computational algorithms (see below). Note how different these maps are. Note also that this comparison uses the exact same migrated dataset organized in two different ways. Migrating each sector independently generally results in even more significant differences. a) From full-azimuth 3D b) From five azimuth sectors reflection angle gathers. of reflection angle gathers Figure 3: Eagle Ford formation - horizon maps of extracted from a volume created using automatic VVAZ analysis. Stability of anisotropic analysis: One of the main problems in azimuthal anisotropy analysis is the instability of the isotropic data points. In azimuthal analysis, for both VVAZ and AVAZ the procedure involves fitting an ellipse to the seismic offset/angle azimuth data. When the samples exhibit large anisotropy, the fitting procedure is robust, but when the samples are isotropic, fitting an ellipse becomes impossible. Figure 4 illustrates this problem: When the data is isotropic, the ellipse becomes a circle and it is impossible to reliably estimate the azimuth ( ). In automatic algorithms this fact is normally ignored and the resulting data maps are contaminated with unreliable data values, without any warning to the interpreter. This artifact is also well illustrated in Kowalski et al, a) b) c) Figure 4: Illustration of the anisotropy ellipse. In a) there is clear anisotropy, and the ellipse is elongated, making it easy to estimate the azimuth. b) and c) represent isotropic samples where ellipse becomes a circle and it is not possible to estimate azimuth. Reliability of amplitude information: Conventional AVO relies on good amplitude recovery and is therefore very sensitive to the quality of processing and prestack migration. AVAZ analysis is even more sensitive, because the mathematical problem we solve is inherently less stable. On the other hand, the amplitude recovery for many of the datasets used in wide-azimuth analysis is significantly worse than conventional seismic data for various reasons. An extreme example of poor amplitude calibrations between sectors which were processed separately is presented in Figure 5. Again, this type of reliability problem is not recorded in the final horizon map, so the result can be misleading to the interpreter. Aliasing and azimuth oriented noise: In theory, to succeed with wide-azimuth analysis we need to use a good wide-azimuth survey. In practice, we often deal with an acquisition situation which is less than ideal. For this reason, azimuthal analysis often suffers from poor sampling, resulting in aliasing artifacts. This problem is SEG New Orleans Annual Meeting Page 428

3 even greater when working with full 3D gathers, as illustrated in Figure 6. Aliasing noise and migration swipes may be seen in Figure 6a. This noise is not random. It is organized noise which appears to be correlated with some azimuths, and it masks the real azimuthal residual velocity variations that we are looking for. This is a common artifact because most 3D surveys are not uniform and some azimuths are sampled better than others. Therefore, artifacts that are correlated with azimuths are natural in these circumstances. Re-binning the gathers can help reduce some of this noise by increasing sampling per bin, thus reducing aliasing artifacts. One of the advantages of working with 3D gathers is the flexibility to re-bin the data after migration and select optimal offset/angle azimuth bins. Figure 5: Three azimuth sectors at one CRP location. This data was used for azimuthal anisotropy analysis. Each sector was processed separately. Note the amplitude variability between the sectors and between traces within each sector. The aliasing artifacts affect AVAZ even more. Figure 7 shows an example of how re-binning can reduce aliasing artifacts and dramatically improve the results of AVAZ. In this example, the anisotropy reliability attribute is displayed along the Eagle Ford formation horizon. Acquisition footprints: In azimuthal analysis, acquisition footprints are among the most significant artifacts negatively affecting the reliability of the analysis. It is easy to see why: In wide-azimuth surveys, the acquisition is not uniform in all azimuths. If unaddressed, this problem can often lead to erroneous results - the azimuth maps show acquisition footprints instead of azimuthal anisotropy information. These azimuthal acquisition footprints are normally more dominant than the real layer anisotropy, so their patterns are picked by automated algorithms. The result is often the anisotropy of the acquisition instead of intrinsic subsurface anisotropy. Reflection angle Reflection angle a) 3D gather b) Re-binned 3D gather using larger angle bins Figure 6: 3D gathers. The horizontal axis increases with the reflection angle, while the azimuth is circulating within each angle bin. Artifacts which are correlated with the azimuth are seen as periodic noise in this data. a) Is a typical 3D gather exhibiting significant aliasing artifacts and migration smiles. Note how rebinning (b) reduces the dominant artifacts and noise. a) From 3D gathers b) From re-binned data Figure 7: AVAZ analysis: anisotropy intensity maps This is dangerous in that the user interpreting the maps has no indication of this. The data displayed on the horizon maps seems to be the real anisotropy of the target layer. Unlike the familiar acquisition footprints resulting from conventional migrations, it is difficult to spot this problem in azimuthal anisotropy maps just by visual inspection. It does not resemble the effects we are used to, it is not a grid -like artifact; rather, it behaves very much like real anisotropy. VVAZ analysis methodology: VVAZ analysis is based on detecting oscillations of moveouts as functions of azimuth. Since moveout differences between azimuths grow as the offset/angle increases, some users prefer to do the analysis on a single offset/angle bin instead of analyzing the full gather, hoping to obtain maximum effect at the large offsets (angles). This may work well with perfect data, but with real data we notice significant differences in the results when comparing two offset/angle bins. This may be seen in Figure 8, which compares two different angle bins, each containing azimuths from created from the same 3D gather. Ideally the anisotropic parameters extracted SEG New Orleans Annual Meeting Page 429

4 independently from each bin will be the same, because it is the same gather. However, from what we see in Figure 8, we may expect that the anisotropic parameters extracted from each of the bins will be different. Which is the correct one? Using the full 3D data will generally improve the stability of the procedure. a) Angles b) Angles Figure 8: Angle sector from a 3D angle gather displaying traces for all azimuths. Note the oscillations of the event, suggesting azimuthal velocity anisotropy. Lateral consistency: VVAZ and AVAZ are performed gather by gather. Incorporating a lateral consistency procedure can improve result reliability. Our approach to interpolation and filtering of the 3D vector data (G1,G2, ) or ( 1, 2, ) is discussed in detail by Weiss & Canning (2015) Filtering azimuthal anisotropic velocity field, submitted to the 85 SEG Annual International Meeting, and by Canning & Malkin (2013). Additional considerations: There are, of course, many other factors affecting the reliability of azimuthal analysis. For the sake of completeness, we mention here the effects of the anisotropy model (HTI, Orthorhombic, Tilted Orthorhombic, etc.) that is used in the analysis, and the accuracy of the procedure for estimating azimuth and reflection angles. Both factors can have a significant effect on the reliability of the result. Including reliability in azimuthal anisotropy analysis: Our approach to the reliability problem is to include reliability estimates inside our VVAZ and AVAZ analysis. Points that are marked unreliable are assumed to be isotropic, and there we perform isotropic instead of anisotropic analysis. These points are marked as unreliable in the azimuthal anisotropy attribute volume, and when extracted onto a horizon map, the unreliable tags are maintained. This is shown in Figure 3 and Figure 7, where unreliable points are colored dark red. Reliability is defined as a combination of amplitude, semblance and anisotropy intensity criteria, and is included in both VVAZ and AVAZ algorithms. Figure 9 shows an example of an anisotropic AVAZ gradient using an Eagle Ford dataset. This picture looks at cross-sections and compares an anisotropic gradient (Ruger, 1998) estimated in the conventional way at every subsurface point (b), with an anisotropic gradient in which the reliability estimation is included in the AVAZ algorithm. (c). Figure 9a) is the isotropic AVA gradient presented for comparison. The conventional calculation (b) provides strange and noisy results, reflecting the instability of the process, while c) is simple to interpret, shows only clear anisotropic data, and enables reliable interpretation. a) AVO gradient b) Anisotropic AVAZ c) Anisotropic gradient, reliability gradient not included reliability included Figure 9: Anisotropic AVAZ gradient cross-section from an Eagle Ford dataset Figure 10 provides another perspective on the effect of reliability on the result. 10b) shows AVAZ anisotropic intensity where reliable criteria were include in the computational algorithm. The result is significantly better than in the straightforward approach 10a). b) No reliability criteria b) Reliability included Figure 10: AVAZ antistrophic intensity maps along the Eagle Ford formation Conclusions The reliability of the analysis of azimuth information from prestack seismic data is very important. The issue is complex and results from multiple problems in data generation, calculation algorithms and workflows. We have demonstrated this complexity and its significance with the examples above. Including reliability analysis in VVAZ and AVAZ algorithms is a new approach which results in significantly improved reliability. Acknowledgement The Eagle Ford data used to prepare this paper is courtesy of Seitel, Inc. SEG New Orleans Annual Meeting Page 430

5 EDITED REFERENCES Note: This reference list is a copyedited version of the reference list submitted by the author. Reference lists for the 2015 SEG Technical Program Expanded Abstracts have been copyedited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCES Canning, A., and A. Malkin, 2013, Extracting azimuthal information from 3D full-azimuth gathers using automatic RMO analysis and AVAZ: 83rd Annual International Meeting, SEG, Expanded Abstracts, Grechka, V., and I. Tsvankin, 1998, 3D description of normal moveout in anisotropic inhomogeneous media: Geophysics, 63, Koren, Z., and I. Ravve, 2011, Full-azimuth subsurface angle domain wavefield decomposition and imaging: Part I Directional and reflection image gathers: Geophysics, 76, no. 1, S1 S13. Kowalski, H., P. Godlewski, W. Kobusinski, W. Makarewicz, M. Podolak, A. Nowicka, Z. Mikolajewski, D. Chase, R. Dafni, A. Canning, and Z. Koren, 2014, Imaging and characterization of a shale reservoir onshore Poland, using full-azimuth seismic depth: First Break, 32, Rüger, A., 1998, Variation of P-wave reflectivity with offset and azimuth in anisotropic media: Geophysics, 63, SEG New Orleans Annual Meeting Page 431