Title:

A texture-based approach for automated detection of listric faults

Author(s):

Muhammad Amir Shafiq*, Haibin Di, and Ghassan AlRegib

Center for Energy and Geo Processing (CeGP),

School of Electrical and Computer Engineering,

Georgia Institute of Technology,

Atlanta, GA, 30332 *amirshafiq@gatech.edu

Summary:

Classical interpretation of listric faults is usually performed manually by the

geophysicist, which is not only time consuming and labor intensive but also susceptible

to interpreter bias. In this paper, we propose a texture-based interpretation workflow for

the automated delineation of major listric faults in a 3D migrated seismic volume. In the

first step, we compute three-dimensional gradient of texture (3D-GoT), which describes

the texture dissimilarity between neighboring cubes around each voxel in a seismic

volume across time or depth, crossline, and inline directions. In the second step, we

calculate an adaptive global threshold and apply it to the 3D-GoT map to obtain a binary

map, which highlights the probable boundary regions of the listric faults. Finally, we

apply post-processing to the obtained binary maps, which include morphological

opening and curve fitting, to yield a delineated listric fault surface within the migrated

seismic volume. The experimental results on a real seismic dataset from the Stratton field

in the Texas Gulf coast show the effectiveness of the proposed workflow, indicating its

great potential for assisting the structural interpretation.

Introduction

The majority of the world’s petroleum and gas reservoirs are found near structural traps and fault systemsdue to their sealing properties. Therefore, detection of faults is one of the crucial steps for hydrocarbonsexploration and production. As one of the most important fault types for hydrocarbon migration andaccumulation, listric normal fault, formed during rifting, drifting, and evolution of passive continentalmargins are usually recognized either by the decreasing dip with the depth, which flattens out at thebottom of the fault, or changes in strata, and in some cases as changes in texture. Listric faults haveconcave surface curved upward and usually occur in an area where brittle rocks overlie ductile rocksin an extensional regime. As mentioned in Shelton (1984), such fault plays a key role in the formationof structural traps, rollover anticlines, and upthrown-fault-block closures. Plethora of research has beenpublished in the past decades on the listric fault (Shelton, 1984; El-Mowafy and Marfurt, 2008; Kelevitzet al., 2010; Zheng and Gao, 2013).

Experienced geophysicists interpret the migrated data by observing the intensity and texture variationsof seismic traces near different structures. The majority of fault delineations are performed interactivelyby the geophysicists with the help of various seismic interpretation and visualization softwares such asPetrel, OpendTect, and PaleoScan etc. Manual and interactive interpretation yields detailed structuralinterpretation and avoids false positives, but at the same time it is influenced by the subjective inter-pretation of the geophysicist and may vary remarkably for different geophysicists. The modern wide-azimuth and high-density seismic acquisitions have aided seismic interpreters by yielding seismic datawith higher quality and better resolution. However, these techniques have resulted in the striking growthof acquired seismic data, which in turn are causing manual interpretation extremely time consuming andlabor intensive. To aid geophysicists in the interpretation process, researchers have proposed severalfully- and semi-automated fault delineation methods based on edge detection, texture, spectral decom-position, seismic attributes, Hough transform, visual saliency, and different image processing techniques(Gibson et al., 2003; Silva et al., 2005; Jacquemin and Mallet, 2005; Pepper and Bejarano, 2005; Co-hen et al., 2006; AlBinHassan et al., 2006; Barbato, 2012; Aarre et al., 2012; Zhang et al., 2014; Hale,2013; Lawal et al., 2016; Wu and Hale, 2016; Wang and AlRegib, 2016). However, to the best of ourknowledge, such methods often fail when applied to the automated delineation of listric faults. In thispaper, we present a novel approach based on texture dissimilarity for automated delineation of majorlistric faults within Stratton field, which is a large gas-producing field in the Texas Gulf coast.

Stratton field, which is characterized predominantly by a major growth normal listric fault and associ-ated rollover anticlines is located within the Texas Gulf coast, near the northern end of Frio FR-4 gasplay along the northwestern margin of the Gulf of Mexico basin (Zheng and Gao, 2013). Gas reservoirsin this area are characterized by both structural and stratigraphic traps featured by faults, pinch-outs,and changes in facies (Barbato, 2012). The presence of major normal listric fault complemented by aseries of antithetic faults have created many small reservoirs in this area. Stratton field is further char-acterized by varying bed thickness, complex fault structures, salt-related anticlines, pillows, and turtlesstructures related to the Paluxy Formation (Barbato, 2012). In this paper, we propose a novel approachfor the delineation of listric faults within migrated seismic volumes based on the three-dimensional gra-dient of textures (3D-GoT) by Shafiq et al. (2016), which describes the texture dissimilarity in a threedimensional space. By adaptively determining a threshold within obtained 3D-GoT maps, we highlightthe probable boundary regions of the listric faults. Then, we apply some post-processing operationsto delineate the listric faults. The rest of the paper is organized as follows: the proposed fault delin-eation workflow is explained in method section followed by a section on experimental results and theirdiscussion. Finally, conclusions are given in last section.

Method

Given a 3D migrated seismic volume V of size T×X×Y , where T represents time or depth, X representscrosslines, and Y represents inlines, we want to generate a 3D plane of listric fault. The proposedworkflow for listric faults delineation is illustrated in Figure 1.

79th EAGE Conference & Exhibition 2017Paris, France, 12 – 15 June 2017

Figure 1: The block diagram of the proposed listric fault delineation workflow.

In the first step, we compute an attribute map using 3D-GoT (Shafiq et al., 2015, 2016), which measuresthe dissimilarity between neighboring cubes around each voxel in seismic volume across time or depth,crossline, and inline directions. The 3D-GoT map highlights the texture dissimilarity across listric faultby computing perceptual texture difference around fault surface. The 3D-GoT can perform robustly inthe presence of noise and requires very few parameters as compared to other algorithms. The detailedexplanation of 3D-GoT can be found in Shafiq et al. (2015, 2016), whereas a brief description is givenin this paper. To evaluate the GoT across crossline direction i.e x-direction, we calculate GoT at eachvoxel [t,x,y] in x-direction. As we move the center point as shown in Fig. 2, and its two neighboringcubes, denoted Wx− and Wx+, in the x-direction along the blue line, the texture dissimilarity along theblue line using the function d(·), we yield the GoT profile as the curve shown at the bottom of Fig. 2.Theoretically, the highest dissimilarity, and hence the highest GoT value is obtained when the centerpoint falls exactly on the texture boundary. Similarly, GoT is also calculated along x and y directions.To improve the delineation efficiency and robustness, 3D-GoT employs a multi-scale gradient, whichis the weighted average of GoT values calculated based on various cubes. The multi-scale 3D-GoT ismathematically expressed as

G[t,x,y] =

∑i∈{t,x,y}

(N

∑n=1

ωn ·d(Wn

i−,Wni+))2

12

, (1)

d (Wi−,Wi+) = E (|F {|F {|Wi−−Wi+|}|}|) , i ∈ {t,x,y}, (2)

F [µ,ν ,ω] =1L3

L−1

∑t=0

L−1

∑x=0

L−1

∑y=0

V [t,x,y]e−2πi(tµ+xν+yω)/L, (3)

where d(·) is the dissimilarity function, which computes the perceptual dissimilarity between two cubesby applying two concatenated 3D-FFT magnitude operations to the absolute difference of neighboringcubes and averages the results using expectation operation E. F {·} represents the 3D-FFT and Wn

−and Wn

+ represent the two cubes of size n in negative and positive directions, respectively, with respectto the voxel at which GoT is calculated. N represents the number of cubes, L defines the edge lengthof a cube-shaped data volume, wn is inversely proportional to n and defines the weights associated witheach cube size for GoT calculation, and [t,x,y] and [µ,ν ,ω] represent the coordinates of the spatial andfrequency domains, respectively. The GoT values near the listric fault boundary are higher as comparedto other regions because of texture dissimilarity. In the second step, we threshold the obtained 3D-GoT map using Otsu’s method to yield a binary volume B of size T ×X ×Y , same as that of V andG, highlighting the areas of strong texture dissimilarity. The white regions in B highlight the potentiallistric fault boundaries. The binary map may also highlight noisy areas and areas characterized by texturedifference away from the listric fault surface. Finally in the last step, we apply post-processing in orderto get rid of noisy areas and accurately delineate the listric faults. Post-processing comprises of binarymorphological opening to get rid of small binary patches, and curve fitting to bridge the gaps betweendisconnected points in the binary map B and to make the detected fault surface smooth.

Results

In this section, we present the effectiveness of the proposed workflow for listric faults delineation onthe real seismic dataset acquired from the Stratton field in the Texas Gulf coast. The dataset used inthis paper is a subset of the 92-BEG 3D seismic survey, which is a 7.6 mi2 public domain 3D seismicsurvey acquired in 1992 by the Texas Bureau of Economic Geology (BEG). The focus study area has a

79th EAGE Conference & Exhibition 2017Paris, France, 12 – 15 June 2017

Figure 2: Illustration of GoT along x-direction.Figure 4: The detected listric fault plane shownover the original seismic slices illustrate the effec-tiveness of the proposed workflow.

time direction ranging from 1,000ms to 3000ms, crosslines ranging from #1 to #200, and inlines rangingfrom #1 to #100. The output of the proposed workflow on four different seismic inline sections fromthe Stratton dataset, highlighted in the blue color, is shown in Figure 3. The subjective evaluation ofresults shows that the proposed workflow detects the listric fault boundary with good precision andaccuracy. The detected listric fault plane superimposed on the original seismic section is shown inFigure 4, which show that the fault surface matches the seismic reflections quite well and demonstratethe effectiveness of the proposed workflow. Our method for computer-aided 3D delineation of majorlistric faults holds the potential for assisting the seismic interpretation in various ways. First, it achievesautomatic detection and interpretation of complex listric faults, which not only is computational efficientbut also helps avoid the bias from interpreters, especially for a dataset of geologic complexities. Theproposed algorithm performs really well in the areas characterized by the strong texture difference.Second, it allows interpreters to interactively correlate the listric faults to the seismic reflectors, whichfacilitates the investigation of the influence of the faults creation on the neighboring layers and therebyimproves the accuracy of the listric-fault-related structural interpretation. Finally, it provides importantgeologic constraints for 3D facies analysis and structural modeling in the exploration areas featured withvariations in texture.

Conclusions

We have proposed a workflow to detect the major listric faults within migrated seismic volumes usingthe 3D attribute, the gradient of texture, which can describe the texture variations along time, crossline,and inline directions. The experimental results of the listric fault delineation on a real seismic datademonstrate the effectiveness of the proposed workflow. The proposed workflow augments the faultsdelineation performance by exploiting the strong coherence between seismic section and texture dissim-ilarity in a migrated 3D volume. The proposed workflow is expected to not only become a handy tool inthe interpreter’s toolbox for automatically delineating listric faults but also reduce the time for seismicinterpretation.

Acknowledgements

This work is supported by the Center for Energy and Geo Processing (CeGP) at Georgia Tech and KingFahd University of Petroleum and Minerals. We would also like to thank the Bureau of EconomicGeology at the University of Texas at Austin for providing the Stratton seismic dataset.

References

Aarre, V., Astratti, D., Al Dayyni, T., Mahmoud, S., Clark, A., Stellas, M., Stringer, J., Toelle, B.,Vejbæk, O. and White, G. [2012] Seismic Detection of Subtle Faults and Fractures. Oilfield Review,

79th EAGE Conference & Exhibition 2017Paris, France, 12 – 15 June 2017

(a) (b) (c) (d)

Figure 3: The output of the proposed workflow on four different seismic inline sections shows theprecision and accuracy of listric fault delineation. The detected listric fault boundary is labeled in bluecolor. (a) Seismic inline section #30. (b) Seismic inline section #55. (c) Seismic inline section #65. (d)Seismic inline section #90.

24(2), 28–43.AlBinHassan, N.M., Luo, Y. and Al-Faraj, M.N. [2006] 3D edge-preserving smoothing and applications.

Geophysics, 71(4), P5–P11.Barbato, U. [2012] Fault Detection Using the Phase Spectra from Spectral Decomposition. Master’s

thesis, Department of Earth and Atmospheric Sciences, University of Houston.Cohen, I., Coult, N. and Vassiliou, A.A. [2006] Detection and extraction of fault surfaces in 3D seismic

data. Geophysics, 71(4), P21–P27.El-Mowafy, H. and Marfurt, K.J. [2008] Structural Interpretation of the Middle Frio Formation using

3D Seismic and Well Logs: An example from the Texas Gulf Coast of the United States. The LeadingEdge, 27(7), 840–854.

Gibson, D., Spann, M. and Turner, J. [2003] Automatic Fault Detection for 3D Seismic Data. In: DICTA.821–830.

Hale, D. [2013] Methods to compute fault images, extract fault surfaces, and estimate fault throws from3D seismic images. Geophysics, 78(2), O33–O43.

Jacquemin, P. and Mallet, J.L. [2005] Automatic faults extraction using double hough transform. SEGTechnical Program Expanded Abstracts 2005, 755–758.

Kelevitz, K., Raáb, D., Dombrádi, E. and Spahic, D. [2010] Geoelectric Survey of a Roll-over Anticline(Austria). Extended Abstract, 72nd EAGE Conference and Exhibition incorporating SPE EUROPEC.

Lawal, A., Al-Dharrab, S., Deriche, M., Alregib, G. and Shafiq, M.A. [2016] Fault detection usingseismic attributes and visual saliency. SEG Technical Program Expanded Abstracts 2016, 1939–1943.

Pepper, R. and Bejarano, G. [2005] Advances in Seismic Fault Interpretation Automation. Search andDiscovery Article 40170, Poster presentation at AAPG Annual Convention, 19–22.

Shafiq, M.A., Wang, Z., AlRegib, G., Amin, A. and Deriche, M. [2016] A texture-based interpretationworkflow with application to delineating salt domes. Accepted in Interpretation.

Shafiq, M.A., Wang, Z., Amin, A., Hegazy, T., Deriche, M. and AlRegib, G. [2015] Detection of Salt-dome Boundary Surfaces in Migrated Seismic Volumes Using Gradient of Textures. SEG TechnicalProgram Expanded Abstracts 2015, 1811–1815.

Shelton, J.W. [1984] Listric normal faults; an illustrated summary. AAPG Bulletin, 68(7), 801–815.Silva, C.C., Marcolino, C.S. and Lima, F.D. [2005] Automatic Fault Extraction Using Ant Tracking

Algorithm in the Marlim South Field, Campos Basin. 9th International Congress of the BrazilianGeophysical Society; EXPOGEF 2005, 1378–1381.

Wang, Z. and AlRegib, G. [2016] Interactive Fault Extraction in 3D Seismic Data Using the HoughTransform and Tracking Vectors. IEEE Transactions on Computational Imaging, PP(99), 1–1.

Wu, X. and Hale, D. [2016] 3D seismic image processing for faults. Geophysics, 81(2), IM1–IM11.Zhang, B., Liu, Y., Pelissier, M. and Hemstra, N. [2014] Semiautomated fault interpretation based on

seismic attributes. Interpretation, 2(1), SA11–SA19.Zheng, Z. and Gao, D. [2013] Along-strike variation in extension and cross-strike fractures in Stratton

field, south Texas: Implications for fractured reservoir description and characterization. SEG Techni-cal Program Expanded Abstracts 2013, 1534–1539.

79th EAGE Conference & Exhibition 2017Paris, France, 12 – 15 June 2017