Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at

Transcription

1 t Technical papers Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at Relative P-impedance estimation using a dipole-based matching pursuit decomposition strategy Xiaotao Wen 1, Bo Zhang 2, Wayne Pennington 3, and Zhenhua He 1 Abstract The P-impedance is one of the most important elastic parameters of rocks, and it is commonly used for reservoir characterization. Conventional P-impedance inversion merges a low-frequency log-based model with a high-frequency seismic-derived model. We have proposed a method to estimate the P-impedance by employing dipole-based matching pursuit (DMP) decomposition. The matching pursuit decomposes the seismic traces into a superposition of scaled wavelets, and the associated scalar information represents the reflectivity series, which can be integrated for P-impedance estimation. Unfortunately, DMP analysis is usually performed trace by trace, resulting in a poor lateral continuity. Applying conventional lateral smoothing through mean or median filtering improves the lateral continuity but typically decreases the vertical resolution. We have evaluated an adaptive smoothing strategy that required the filtering to follow bed boundaries in an automated manner, sharpening the boundaries while maintaining the high quality of inversion. We have determined the effectiveness of our algorithm by first applying it to a synthetic wedge model and then to a real seismic data set. Introduction The P-impedance is the product of rock density and compressional (P-wave) velocity and is commonly related to porosity and/or lithology of subsurface formations and to provide information for reservoir characterization. A suite of methods have been proposed to derive P-impedance from seismic data, usually constrained by borehole data (e.g., Oldenburg et al., 1983). The poststack seismic trace can be regarded as the convolution of seismic wavelet with sparse-spike reflectivity series and noise. (To relate this reflection series to changes in P-impedance, the stacked trace is considered equivalent to a zero-offset trace.) This in turn relies on the assumption that we can obtain the reflectivity series through a processing sequence. Deconvolution is one method that can be used to achieve this goal, whether explicitly or implicitly. Unfortunately, obtaining the spike reflectivity series is beyond the capability of deconvolution, due to the limited bandwidth of seismic data. The procedure of deconvolution is equal to placing a scaled wavelet at those locations where nonzero reflectivity is found; the scaled wavelet is the multiplication of that reflectivity and a user-defined wavelet. This suggests that we can obtain the spike reflectivity series by removing the corresponding wavelet. We use a dipole-based matching pursuit (DMP) algorithm to obtain the sparse-spike reflectivity series, drawing from a family of wavelets appropriate for the data set under consideration. Matching pursuit (MP) inversion is developed by Mallat and Zhang (1993) to obtain a sparse signal from a time series. Many geophysicists apply this approach to the decomposition of the seismic trace into a series of scaled wavelets, in which the wavelets are selected from a userdefined redundant wavelet library. As currently used, MP usually includes four steps: (1) define a wavelets library that should include all possible wavelets, (2) correlate each wavelet in that library against the seismogram, moving along the time axis, and determine the best-fit wavelet for the strongest reflection, (3) subtract that best-fit wavelet while maintaining a record of the reflection coefficient, and (4) repeat steps 2 and 3 on the residual trace until the residual energy falls below a user-defined threshold. Nguyen (2000) demonstrate the usefulness of MP to perform 2D prestack seismic data filtering. Liu and Marfurt (2007) perform time-frequency analysis on inverted results obtained through MP to map a channel system. Wang (2007) obtain the 1 Chengdu University of Technology, College of Geophysics, Chengdu, China and Michigan Technological University, Department of Geological and Mining Engineering and Sciences, Houghton, Michigan, USA. wenxiaotao@cdut.cn; hzh@cdut.edu.cn. 2 Michigan Technological University, Department of Geological and Mining Engineering and Sciences, Houghton, Michigan, USA. bzhang9@mtu.edu. 3 Michigan Technological University, Department of Geological and Mining Engineering and Sciences, Houghton, Michigan, USA and Michigan Technological University, College of Engineering, Houghton, Michigan, USA. wayne@mtu.edu. Manuscript received by the Editor 4 February 2015; revised manuscript received 17 May 2015; published online 24 July This paper appears in Interpretation, Vol. 3, No. 4 (November 2015); p. T197 T206, 8 FIGS Society of Exploration Geophysicists and American Association of Petroleum Geologists. All rights reserved. Interpretation / November 2015 T197

2 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at time-frequency spectrum by MP decomposition and then use that spectrum for detection of gas reservoirs. In most studies, MP is performed trace by trace, which may result in lateral discontinuity associated with data quality rather than geologic features. To avoid this, Nguyen and Castagna (2010) stabilize the inverted results using a priori information in the first pass for MP decomposition. The a priori information of reflectivity is either generated from a well log or a pseudolog. They obtain the pseudolog by averaging the inverted reflectivities from the first-round MP. Wang (2010) develops multichannel MP, which exploits lateral coherence as a constraint to improve the uniqueness of the solution. Zhang and Castagna (2011) obtain high-resolution reflectivity series by decomposing the seismic traces using a basis pursuit strategy. Specifically, their dipole model considers the interference of pairs of reflection events and thus can generate an accurate reflectivity series for thin beds. However, they use one wavelet for inversion, which limited the quality of their results because the characteristics of wavelets, such as dominant frequency, change as the seismic wave propagates through the subsurface. Their inversion is also limited by their trace-by-trace approach, which typically will demonstrate poor lateral continuity. Structure-oriented filtering is commonly used to improve the quality of a seismic image (Luo et al., 2002) and to enhance the lateral continuity of seismic attributes (Weickert, 1999), while preserving the important discontinuities such as faults or channels. This approach was initially used for smoothing of photographic images while preserving edges. The technique was developed because conventional smoothing or diffusion algorithms, while improving the image quality, may blur the edges of objects (e.g., Perona and Malik, 1990; Catte et al., 1992; Wen et al., 2011). The steps of structure-oriented filtering usually include first an estimation the orientation of reflector and second the application of a filter along that orientation. Common algorithms used for estimating orientation of reflectors are based on Figure 1. Cartoon showing the dipole model for scaled reflecting events. Any reflector pair can be represented as the weighted sum of an even pair and an odd pair (modified from Zhang and Castagna, 2011). the structure tensor of the seismic image (Fehmers and Höcker, 2003) and on multiple-window-based scanning (Marfurt, 2006). Common smoothing strategies include filtering based on the mean, the median, and the principal component. We use the multiple-windowbased method proposed by Marfurt (2006) for our estimate of the orientation (dip and azimuth) of reflectors. We apply an adaptive structure filtering technique to improve the lateral continuity of the inverted reflectivity profile. We begin by estimating the dip of reflectors using a structure tensor derived locally within the stacked seismic data, whereas estimating the likelihood that an event is true reflectivity from a continuous reflector, or not. We then apply 1D diffusion to the truereflectivity samples along local structure, and we apply 2D diffusion to the other samples, representing background nonchanging values. We first test our workflow on a synthetic wedge mode to validate its effectiveness. We then apply it on a real seismic data and compare the estimated P-impedance with that from conventional inversion. Matching pursuit based on a dipole model A typical workflow of current MP algorithms begins with the identification of seismic events. Each picked event is assumed to have a corresponding wavelet, stored in a library. We then determine the amplitude, phase, and dominant frequency of the wavelets for each of the events. We next subtract a synthetic seismic trace created using the appropriate wavelet for a recognized event, and we repeat the procedure for each event, until finally the residual seismic trace falls below a desired threshold. However, current MP algorithms do not consider the interference of thin beds. The conventional wavelet library consists of user-defined wavelets wðf;φþ where f;φ are the dominant frequency and phase of wavelets (Liu and Marfurt, 2007), respectively, but this works poorly in the presence of thin beds. We use a dipole model, in which each wavelet is the result of a pair of reflecting events. Our wavelet library T198 Interpretation / November 2015

3 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at consists of the convolution result of wavelets wðf ; φþ with the wedge model (cδðtþ þ dδðt nδtþ), where c and d are the reflection coefficients of the top and base reflectors, respectively, Δt is time interval, and nδt is the time thickness of the model. The number of samples n varies from zero to N with NΔt representing the maximum time thickness of the wedge model. The size of our wavelet library is ðn þ 1Þ times larger than that of an equivalent MP library, and the computation cost of our algorithm is also ðn þ 1Þ times that of the current MP algorithm. The dipole model (Figure 1) (Zhang and Castagna, 2011) decomposes the reflectivity pair (cδðtþþ dδðt nδtþ) of a wedge model into even (subscript e) and odd (subscript o) parts as cδðtþ þ dδðt nδtþ ¼ ar e þ br o ; a¼ cþd ; 2 (1a) c d ; 2 (1b) d ¼ a b; (1c) b¼ and c ¼ a þ b; where r e ¼ δðtþ þ gðtþ þ δðt nδtþ and r o ¼ δðtþ gðtþ δðt nδtþ. Then, our library of wavelets is decomposed into even (e ¼ w r e ) and odd (o ¼ w r o ) parts. The seismic response of reflectivity pair (cδðtþ þ dδðt nδtþ) of a wedge model can be expressed as f ¼ a w re þ b w ro (2a) f ¼ a e þ b o; (2b) and where stands for the scalar product and stands for the convolution operator. At each decomposition iteration k, we compute the inner product IP k;j;l between the residual seismic trace RK 1 s and each element el ; ol, from our wavelet library at every time index j as IP k;j;l ¼ kak;j;l hrk 1 s; el ik2 þ kbb;j;l hrk 1 s; ol ik2 ; (3) where l is the index of elements in our wavelet library. We obtain the parameters ak;j;l and bk;j;l of each selected element by using a simple least-squares approach. The element, eblðkþ and oblðkþ, which corresponds to the maximum value ðip k;j;l Þmax of IP k;j;l, is considered to be the best element for the current time index j; lðkþ is the index of the best-matched element in our wavelet library for the decomposition iteration k. We then compare the value of ðip k;j;l Þmax along the time axis j, and the time index with the maximum value of ðip k;j;l Þmax is regarded as the picked event for the decomposition iteration k. The seismic trace after the K iterations of decomposition can be expressed as Figure 2. A model consisting of two reflecting horizons showing the results of smoothing. (a) The model before smoothing in which the continuous curves represent reflectivity and the isolated dots represent noise. (b) Smoothed reflectivity using 2D anisotropy diffusion. (c) Smoothed reflectivity using the proposed algorithm. Note that although the 2D anisotropy diffusion improves the lateral continuity and suppresses the noise, it blurs the edges. In contrast, the new smoothing strategy preserves the reflectivity very well. Interpretation / November 2015 T199

4 s¼ K X fabk eblðkþ þ bbk oblðkþ g þ RK 1 S ; (4) Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at k¼1 where eblðkþ and oblðkþ are the even and odd parts of the best element in our wavelet library for the k decomposition iteration, respectively, abk and bbk and are the scalar coefficients for the even and odd parts, respectively; and RK s is the residual trace after Kth decomposition. The reflectivity series rðtþ after K times decomposition can be expressed as Figure 3. Cartoon showing the strategy of judging whether an inverted reflectivity at the analysis point is noise or a reflecting event. We exam the product of inverted reflectivity between points parallel (r 4 r 5 r 6 ) and perpendicular (r 2 r 5 r 8 ) to the local dip of structure. The inverted reflectivity is considered to be noise if the difference between (r 4 r 5 r 6 ) and (r 2 r 5 r 8 ) is smaller than a designed threshold. Figure 4. Wedge model illustrating the procedure of reflectivity inversion using DMP. (a) The P-impedance wedge model used for generated synthetic seismograms. (b) The noise-free synthetic seismic section obtained by convolution of the reflectivity model with a Ricker wavelet having a dominant frequency of 30 Hz. (c) The synthetic seismogram with 10% noise added. We obtain the relative P-impedance of noise-free seismic sections using (d) conventional MP and (e) DMP. Using DMP to invert the 10%-noise seismic section, we obtain the relative P-impedance (f) without smoothing, and (g) after filtering with our structure-oriented method. T200 Interpretation / November 2015 rðtþ ¼ K X ðabk r be þ bbk obe Þ; (5) k¼1 where r bk and r bk are the reflectivity model corresponding to eblðkþ and oblðkþ, respectively. We repeat the above decomposition procedure until the energy of the residual trace falls below a desired threshold. Improve lateral continuity by adaptive smoothing We apply an adaptive smoothing strategy to improve the lateral continuity of the inverted reflectivity. The images shown in Figure 2 illustrate the procedure of smoothing a model consisting of two reflecting events. Figure 2a shows the model prior to smoothing; the curves represent the reflectivity signal, whereas the isolated dots represent noise. Figure 2b shows the results after applying conventional 2D anisotropy diffusion. It is obvious that we suppress the noise, but we also blur the reflectivity. To avoid this situation, we apply a 1D diffusion to samples along local structure obtained from seismic image if we consider they are true reflectivity and 2D diffusion to other samples (Figure 3), the result of which is shown in Figure 2c. To judge whether an inverted reflectivity is noise or a reflecting event, we examine the products of inverted reflectivity for three points at the analysis point parallel to the local dip of the structure (r 4 r 5 r 6 ) and for three points perpendicular to it (r 2 r 5 r 8 ). In our example (Figure 3), we extract nine samples centered at the analysis point. Usually, the reflectivity value of most points surrounding a sample point containing noise will be close to zero, resulting in products of inverted

5 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at Figure 5. (a) A seismic section showing the seismic-well tie based on the calibration through a synthetic seismogram (b). We identify the top and bottom of a carbonate layer, which is consistent with low gamma ray zone. The leftmost panel is the extracted statistical Ricker wavelet obtained from seismic traces (the second panel), and this is used for generating the synthetic seismogram (the third panel). The fourth through ninth panels show the logs of P-velocity, density, the difference between the deep and shallow resistivity, gamma ray, reflectivity, and P-impedance, respectively. The low values of the gamma ray log identify the top (green line) and bottom (blue) of the carbonate reservoir. High values of the RILD-RILS log identify the oil zone, indicated by the thickness of the translucent red line. (c) Quality-control inverted reflectivity by comparing the real seismic trace (green) and synthetic (red) computed from reflectivity, and their difference (blue). Interpretation / November 2015 T201

6 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at reflectivity with nearly identical low values, parallel and perpendicular to the local structure. But because the reflectivity changes gradually along a true reflecting event, there should exist a large difference between the product computed parallel to structure (r 4 r 5 r 6 ) and that computed perpendicular to it (r 2 r 5 r 8 ). When the difference in products is large, we apply a 1D filter, following the dip of the event; when the difference is small, we apply a 2D filter. Figure 2c shows the result after applying our new smoothing algorithm. Note we simultaneously suppress noise, improve lateral continuity (along dip), and avoid blurring edges. Application To validate the proposed algorithm, we first apply it to a synthetic wedge model (Figure 4a) and then to real seismic data (Figure 5a). Figure 4a shows the P-impedance wedge model we used. We obtain the noise-free (Figure 4b) synthetic seismogram through convolution of the reflectivity model with a Ricker wavelet with dominant frequency of 30 Hz. We add 10% noise to Figure 6. Comparison of estimated reflectivity before and after smoothing. (a) The inverted reflection coefficients obtained using DMP. We apply (b) conventional 2D diffusion and (c) adaptive smoothing to improve the lateral continuity. Note that the new smoothing algorithm improves lateral continuity and preserves vertical resolution. T202 Interpretation / November 2015 obtain the image shown in Figure 4c. Using the noise-free section of Figure 4b, we obtain the estimated P-impedance shown in Figure 4d using conventional MP. When we use the DMP method, we obtain the estimated P-impedance shown in Figure 4e. The estimated relative P-impedance obtained using DMP shows many of the same features as the original Pimpedance model, whereas the estimated obtained using MP exhibits problems because the beds thin in the wedge model. To demonstrate the effect of the structure-oriented filter, we show results from DMP-based inversion only. The estimated relative P-impedance is then obtained for the noisy seismic section by trace integrating the unsmoothed (Figure 4f) and smoothed (Figure 4g) DMP-inverted reflectivity using our structure-oriented filter. The P-impedance generated from the structure-oriented filtered reflectivity is more continuous and more closely matches the original model as the beds thin. Figure 5a is a seismic section taken from real data that had been used for a seismic-well tie. The target stratum is

7 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at a carbonate reservoir indicated by an arrow. Using the calibration of a synthetic seismogram (Figure 5b), we identify the top (red line) and bottom (blue line) of the carbonate stratum, corresponding to a low-valued gamma ray (the seventh panel in Figure 5b) zone. We obtain a very good correlation coefficient of 0.82 between the synthetic and real seismic traces. The wavelet illustrated in the leftmost panel is a statistical Ricker wavelet extracted from seismic traces near the borehole location. The second and third panels show the original seismic traces near the borehole location and the synthetic seismogram, respectively. The fourth and fifth panels are the P-wave velocity and density, respectively. The difference (RILD-RILS) between the deep induction log resistivity (RILD) and shallow induction log resistivity (RILS) is shown in the sixth panel where a high value is an indication of oil. The seventh, eighth, and ninth panels are, respectively, the gamma ray, reflectivity, and P- impedance logs. The result of well testing shows that the reservoir zone, indicated by the red band, contains oil, below which lies a fully water saturated zone. Determination of the oil-water contact (OWC) is critical for the development strategy, and we attempt to resolve it from the seismic data. Figure 5c illustrates the real seismic trace (green), the synthetic computed from inverted reflectivity (red), and their difference (residual trace, blue). Note that we obtain a very good match between the real and synthetic traces. Figure 6a shows the inverted reflection coefficients (not P-impedance) using the DMP method. Note that although the inverted results have high resolution, we still lack sufficient lateral continuity to be useful for seismic-based reservoir characterization (for example, the zone indicated by the arrow). After smoothing using conventional 2D diffusion, we obtain the reflectivity image shown in Figure 6b, and after smoothing using the proposed structure-oriented approach, we obtain the image shown in Figure 6c.Note that although the conventional 2D diffusion does improve lateral continuity, it blurs the reflectivity (Figure 6b). In contrast, our smoothing strategy improves lateral continuity while preserving vertical resolution (Figure 6c). We next investigate the ability to identify the oil zone and OWC on inverted seismic data. Figure 7a shows inverted impedance from conventional sparse-spike inversion after seismic-well tie. The filled well track shows the P-impedance, using the same color scale as the inverted seismic data. To eliminate the effect of the well tie (to simulate seismic-only inversion without low-frequency control), we compute the reflectivity from this inverted impedance and then obtain the relative P-impedance (Figure 7b) by trace integrating that reflectivity section; the filled well track in Figure 7b now shows the relative impedance computed from the impedance from well logs. It can be seen that although the lithologic interface (indicated by the arrow) is clear, the vertical resolution is poor and the OWC below that interface is not clearly identifiable. Figure 8a 8c shows the relative impedance we obtained by trace integrating the inverted reflectivity sec- Figure 7. (a) Estimated P-impedance from conventional poststack sparse-spike inversion using seismic-well tie. We then calculate the reflectivity from the estimated P-impedance and obtain (b) the relative P-impedance by trace integrating the reflectivity to simulate a result obtained without a well. Note that although the lithologic interface (indicated by an arrow) is clear, the vertical resolution is low and the OWC is not clear. Interpretation / November 2015 T203

8 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at tions shown in Figure 6a 6c, respectively; the filled well track is the relative impedance computed from the impedance from well logs. The accumulation of errors associated with trace integration results in noise that appears as vertical bars, indicated by the arrows Figure 8. Comparison of relative P-impedance by trace integrating the inverted reflectivity shown in Figure 6. (a) The estimated P-impedance from the unsmoothed reflectivity in Figure 6a; the arrows point to zones of noise. The estimated P-impedance is shown after (b) 2D diffusion and (c) adaptive diffusion, from the reflectivities shown in Figure 6b and 6c, respectively. Note that the adaptive smoothing mitigates the noise identified in (a) and preserves vertical resolution. (d) The residual seismic section between the real seismic and synthetic computed from inverted reflectivity. T204 Interpretation / November 2015 in Figure 8a. In contrast, Figure 8b and 8c shows more continuous layering after 2D diffusion (Figure 8b) and after our structure-oriented filtering (Figure 8c). To better see improvements with the new smoothing algorithm, we zoom into the target reservoir in the insets in

9 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at Figure 8b and 8c. Note the conventional 2D diffusion blurring the extent of the oil zone (indicated by the arrow in Figure 8b) whereas our new smoothing matches the extent of the oil zone very well (Figure 8c). Figure 8d shows the residual seismic section between real and synthetic data. Note that the energy of residual seismic traces is negligible compared to that of the real seismic traces. Conclusion A high-resolution sparse-spike-like reflectivity of a seismic profile can be obtained using DMP. Unfortunately, the inverted reflectivity can lack sufficient lateral continuity due to the fact that DMP is performed trace by trace. Diffusion is commonly used to improve the continuity of image while preserving the edges, but applying the same diffusion strategy to the entire image tends to blur the edges, resulting in lower resolution. We have proposed an adaptive strategy to simultaneously improve the continuity and preserve vertical resolution of inverted reflectivity, using a structure-oriented filtering approach in which the noisy portions are smoothed by 2D diffusion, but continuous events are smoothed along their dip. The estimated P-impedance from our proposed algorithm is shown to improve upon conventional sparse-spike inversion. Acknowledgments We appreciate the financial support provided for the research in this paper from the National Science Foundation of China (grant no ) and to Michigan Technological University for hosting the visit of the first author. Our gratitude also goes to editor Y. Sun, associate editor W. Abriel, and three anonymous reviewers for their constructive suggestions. References Catte, F., P. L. Lions, and J. Morel, 1992, Image selective smoothing and edge detection by nonlinear diffusion: SIAM Journal on Numerical Analysis, 29, , doi: / Fehmers, G. C., and C. F. W. Höcker, 2003, Fast structural interpretations with structure-oriented filtering: Geophysics, 68, , doi: / Liu, J., and K. J. Marfurt, 2007, Instantaneous spectral attributes to detect channels: Geophysics, 72, no. 2, P23 P31, doi: / Luo, Y., S. Al-Dossary, and M. Alfaraj, 2002, Edge-preserving smoothing and applications: The Leading Edge, 21, , doi: / Mallat, S., and Z. Zhang, 1993, Matching pursuits with timefrequency dictionaries: IEEE Transactions on Signal Processing, 41, , doi: / Marfurt, K. J., 2006, Robust estimates of 3D reflector dip: Geophysics, 71, no. 4, P29 P40, doi: / Nguyen, T., 2000, Matching pursuit of two dimensional seismic data and its filtering application: 70th Annual International Meeting, SEG, Expanded Abstracts, Nguyen, T., and J. Castagna, 2010, High resolution seismic reflectivity inversion: Journal of Seismic Exploration, 19, Oldenburg, D. W., T. Scheuer, and S. Levy, 1983, Recovery of the acoustic impedance of reflection seismograms: Geophysics, 48, , doi: / Perona, P., and J. Malik, 1990, Scale-space and edge detection using anisotropic diffusion: IEEE Transactions on Pattern Analysis and Machine Intelligence, 12, , doi: / Wang, Y., 2007, Seismic time-frequency spectral decomposition by matching pursuit: Geophysics, 72, no. 1, V13 V20, doi: / Wang, Y., 2010, Multichannel matching pursuit for seismic trace decomposition: Geophysics, 75, no. 4, V61 V66, doi: / Weickert, J., 1999, Coherence-enhancing diffusion filtering: International Journal of Computer Vision, 31, , doi: /A: Wen, X., Z. He, and D. Huang, 2011, Highlighting display of geologic body based on directivity filter: Applied Geophysics, 8, , doi: /s Zhang, R., and J. Castagna, 2011, Seismic sparse-layer reflectivity inversion using basis pursuit decomposition: Geophysics, 76, no. 6, R147 R158, doi: / geo Xiaotao Wen received B.S. (1997), M.S. (2003), and Ph.D. (2006) degrees in geophysics from Chengdu University of Technology. He is currently the chair of the School of Earth Detection and Information Technology where he has worked since He worked for BGP Inc., as a geophysics engineer from 1997 to His current research interests include high-resolution seismic inversion, reservoir characterization, and direct hydrocarbon identification. Bo Zhang received a B.S. (2002) in geophysics from the China University of Petroleum, an M.S. (2006) in geophysics from the Institute of Geology and Geophysics, Chinese Academy of Sciences, and a Ph.D. (2009) in geophysics from an industry-supported consortium (AASPI) at the University of Oklahoma. In 2014, he joined the Technological University as a visiting assistant professor. His current research activities include broadband seismic data processing, development and calibration of new seismic attributes, pattern recognition of geologic features on 3D seismic data, and shale resource play characterization. Interpretation / November 2015 T205

10 Downloaded 09/23/15 to Redistribution subject to SEG license or copyright; see Terms of Use at Wayne D. Pennington received degrees in geology and geophysics from Princeton University, Cornell University, and the University of Wisconsin-Madison. He is currently the dean of the College of Engineering at Michigan Technological University, where he has worked since Prior to that, he was at Marathon Oil Company s Petroleum Technology Center in Littleton, Colorado, following several years on the faculty at the University of Texas at Austin. He has served as a first vicepresident of SEG, as the president of the American Geosciences Institute, and as a Jefferson Science Fellow at the U.S. Agency for International Development. T206 Interpretation / November 2015