Delineation of thick incised canyons using spectral-decomposition analysis, curvature and Self-

Organizing Maps in the Exmouth Plateau, Australia David Lubo-Robles* and Kurt J. Marfurt, The University of Oklahoma

Summary

Seismic attributes allow us to extract valuable information

from the seismic amplitude cube, thus making a more

complete and accurate interpretation. In this paper, we use

spectral-decomposition analysis, curvature and Self-Organizing Maps to delineate and study the internal

architecture of incised canyons located in the Mandu

Formation, Exmouth Plateau, Australia.

Introduction

Spectral-decomposition analysis is a powerful technique that

estimate the magnitude and phase components of the seismic data at time-frequency samples, thus allowing us to interpret

geologic features of interest at different scales (Chopra and

Marfurt, 2016).

Of the various spectral-decomposition analysis methods, the

Continuous Wavelet Transform (CWT) method provides

good temporal and frequency resolution using a window

analysis that varies with frequency (Chopra and Marfurt, 2015).

The seismic survey used in this research is Rose 3D located

on the Exmouth Plateau, Australia. The target area is the Mandu Formation which is characterized by incised canyons

created in Early Oligocene time when the sea level was

falling, resulting in a Low-Stand System Track (LST)

(Tellez, 2015).

Spectral decomposition is routinely use to map areal extent

of thin channels that fall below seismic resolution. In this

paper, we want to evaluate how spectral-decomposition analysis performs when interpreting very thick incised

canyons. We will compare this response to curvature

analysis and Self-Organizing Maps in order to delineate and

study the internal structure of the canyons within the Mandu Formation.

Continuous-wavelet transform (CWT) method

Using orthogonal basis function, the CWT algorithm

decomposes the input data in frequency or voice components

(Puryear et. al., 2008). These voice components are

comparable to applying a bandpass-filter to the data and they represent information at a particular frequency (Chopra and

Marfurt, 2016).

As basis function, we use the Morlet wavelet because in

general it provides stronger results than other wavelets such as Derivative of Gaussian (DOG) and the Shannon wavelet.

(Chopra and Marfurt, 2015). This mother wavelet varies

depending on the frequency that is being analyzed, thus

improving the resolution of the data (Chopra and Marfurt, 2015).

In order to analyze the detailed information provided by the

voice components, we plot the results against a red-green-blue (RGB) color scheme which represent a visual way to

interpret geologic features in the subsurface (Li and Lu,

2014). In addition to the individual components, we study

the peak frequency and peak magnitude of the data because it provides a general understanding of the depositional

system and sequence stratigraphy (Marfurt and Kirlin,

2001).

Geologic setting

The Exmouth Plateau is the most western of the structural

elements inside the Northern Carnarvon Basin, Australia. This deep-water marginal plateau is underlain by block-

faulted Paleozoic to Mesozoic sedimentary rocks, deposited

during periods of extension prior to continental breakup in

the Middle Jurassic and Early Cretaceous (Geoscience Australia, 2016).

The Rankin Platform is located on the southeast margin of

the Exmouth Plateau (Figure 1) (Geoscience Australia, 2016). During Early Oligocene time, the overall pattern of

sea level was a major low-stand resulting in a progradational

wedge of carbonate slope developing the Mandu Formation.

From the Oligocene to the middle Miocene, a group of wide and thick incised canyons originated within the Mandu

Formation (Figure 2), due to high-stand, low-stand and

transgressive intervals of sea level (Tellez, 2015).

Figure 1. Present day cross section (Picture from Stoner,

2010). The Rankin Platform is located on the southeast

margin of the Exmouth Plateau located in the western part of

the Northern Carnarvon Basin (Geoscience Australia, 2016).

Delineation of thick incised canyons using spectral-decomposition analysis, curvature and Self-Organizing Maps in the

Exmouth Plateau, Australia

Figure 3. Seismic expression of the Mandu Formation in the Rose 3D seismic survey (red rectangle). There is a group of thick and wide incised canyons. In order to decrease noise, structure-oriented filtering was applied to the data.

Figure 2. The target area is the Mandu Formation (red

rectangle) which is characterized by a group of wide and

thick incised canyons formed by high-stand, low-stand and

transgressive intervals. (Tellez, 2015). Picture from Kelman, et. al., 2013.

Data description

The Rose 3D survey was acquired from August to

September 2008 and is located in the Exmouth Plateau,

Northern Carnarvon Basin, Australia. It has a streamer

separation of 100 m and source separation of 50 m.

The Mandu Formation in the survey is shown in Figure 3.

Picking a consistent horizon slice in this Formation is almost

impossible because these incised canyons extended laterally in all the Formation. On the other hand, if we pick a reflector

above or below the zone of interest, and create phantom

horizons, these will cut different geologic ages (Wallet,

2016). For these reasons, we make our analysis using time

slices.

In addition, the incised canyons in the Mandu Formation are

approximately 0.06 - 0.1 s thick. Principal component structure-oriented filtering was applied to the data to

decrease noise.

Seismic expression of incised canyons using CWT

Spectral Analysis

Figure 4A shows the peak frequency and peak magnitude of

the data, representing a general behavior of the data in the frequency spectrum. Because the incised canyons are very

thick and they have a heterogeneous infill with complex

stacking pattern, we obtain a composite frequency response,

and cannot delineate neither the edges nor the internal structure of the incised canyons.

Because the thickness of the canyons varies in time between

0.06 s and 0.1 s, we expect to observe their frequency

response between 10 and 16 Hz. For this reason, we analyze

the 11, 13 and 15 Hz components of the data and plot them

against a RGB scheme. Figure 4B shows these spectral

components of the data at 1324 ms. Note that the canyons are poorly delineated. The highly heterogeneous canyon fill

of variable thickness stacked channels results in a composite

spectral response that provides little interpretational value.

The random noise that appears in Figure 4B is associated to destructive interference between all the different frequency

responses.

0 5 Km

Delineation of thick incised canyons using spectral-decomposition analysis, curvature and Self-Organizing Maps in the

Exmouth Plateau, Australia

' Figure 4. (A) Time slice at 1324 ms of peak frequency co-rendered with peak magnitude. We cannot delineate the incised canyons,

because we obtain a composite spectral response associated to their heterogeneous infill. (B) 11, 13 and 15 Hz frequency

components against a RGB scheme. Canyons are poorly delineated (yellow arrows). Random noise appears on the data and it is

associated to destructive interference between the frequency responses (red rectangle).

Seismic expression of incised canyons using: most-

negative curvature, most-positive curvature and Self -

Organizing Maps (SOM)

After analyzing spectral components, we thought in another

approach to try to delineate the incised canyons and reveal

their internal architecture. Most-negative curvature (𝑘2) is a seismic attribute that delineates structural valleys, thus it

should be helpful to delineate the canyons presented in the

Mandu Formation. Most-positive curvature (𝑘1) highlights

ridge features in the seismic data. Figure 5 shows most-

negative curvature (𝑘2) co-rendered with most-positive

curvature (𝑘1) at 1324 ms. Using curvature, it is possible to

delineate the canyons correctly but we cannot analyze their

internal structure.

Figure 5. 3D chair view showing the seismic inline

correlated with the co-rendered image of most-positive

curvature 𝑘1 and most-negative curvature 𝑘2. Canyons are

well delineated using most-negative curvature 𝑘2 (yellow

arrows).

In order to study the internal structure within the canyons,

we use Self-Organizing Maps (SOM), which is an

unsupervised 3D facies classification technique that, based on topological relations, reorganize the data (Zhao et. al.,

2016). As inputs, we use the most-positive curvature (𝑘1),

most-negative curvature (𝑘2) and texture attributes (GLCM).

The Gray Level Co-occurrence Matrix (GLCM) statistically represent lateral and vertical variability in seismic

amplitudes. We use homogeneity which measures the

closeness of the distribution, entropy that measures the

roughness and dissimilarity which measures intensity contrast between data samples (Haralick, et. al., 1973).

Figure 6 shows the result from the SOM analysis at 1324 ms.

From this method, we can delineate the canyons where different colors inside the canyons represent different facies,

providing insight about their internal structure.

Figure 6. (A) SOM analysis (homogeneity, dissimilarity,

entropy, 𝑘1 , 𝑘2 ). It is possible to delineate the incised

canyons (yellow arrows). Also, inside the canyons different colors are seen, these colors are associated with the different

facies.

Delineation of thick incised canyons using spectral-decomposition analysis, curvature and Self-Organizing Maps in the

Exmouth Plateau, Australia

Finally, we perform geobody extraction (Meyer et. al., 2001) to model and extract the values in 3D from our SOM

analysis. Figure 7 shows the canyons in 3D and they look

geologically feasible.

Figure 7. Incised canyons extraction from the SOM

analysis.

Conclusions

Spectral-decomposition analysis (components, peak

frequency and peak magnitude), provided disappointing

results, because the canyons are thick (0.06 - 0.1 s) and comprised of heterogenous fill with a complex stacking

pattern. The canyons are barely visible and random noise

appear in the data due to destructive interference between

different frequency responses. In contrast, curvature and Self-Organizing Maps successfully delineate the canyons.

Using Self-Organizing Maps (SOM), also detect different

changes in facies inside the incised canyons revealing their internal architecture. SOM also provides useful geobody

images of the canyons in 3D.

Acknowledgements

We thank Geoscience Australia for providing the data. In

addition, we would like to thank the sponsors of the Attribute

Assisted Seismic Processing and Interpretation (AASPI)

consortium for their support. Finally, we thank

Schlumberger for the licenses in Petrel, provided to the

University of Oklahoma.

Delineation of thick incised canyons using spectral-decomposition analysis, curvature and Self-Organizing Maps in the

Exmouth Plateau, Australia

References

Chopra, S., Marfurt, K. J., 2015, Choice of mother wavelets in CWT spectral decomposition: 85th Annual International Meeting,

SEG, Expanded Abstract, 2957-2961.

Chopra, S., Marfurt, K. J., 2016, Spectral decomposition and spectral balancing of seismic data: The Leading Edge, 35, 176-179,

doi: 10.1190/tle35020176.1.

Geoscience Australia, 2016, Exmouth Plataeu. http://petroleum-acreage.gov.au/2016/geology/northern-carnarvon-basin/exmouth-

plateau. Access on December 5𝑡ℎ, 2016.

Haralick, R. M., Shanmugam, K., Dinstein, I., 1973, Textural Features for Image Classification: IEEE Transactions on Systems,

Man and Cybernetics, Vol. SMC-3, No. 6, 610-621

Kelman, A.P., Nicoll, R.S., Kennard, J.M., Mory, A.J., Mantle, D.J., le Poidevin, S., Bernardel, G., Rollet, N., Edwards, D., 2013,

Northern Carnarvon Basin Biozonation and Stratigraphy Chart 36.

Li, F., and W. Lu, 2014, Coherence attribute at different spectral scales: Interpretation, 2, no. 1, SA99–SA106, http://dx.doi.

org/10.1190/INT-2013-0089.1

Marfurt, K. J., Kirlin, R.L., 2001, Narrow-band spectral analysis and thin-bed tuning: Geophysics, Vol. 66, No. 4, P. 1274–1283

Meyer, D. E., E. L. Harvey, T. E. Bulloch, J. C. Voncannon, and T. M. Sheffield, 2001, Use of seismic attributes in 3-D geovolume

interpretation: The Leading Edge, 20, 1377–1400, doi: 10.1190/1.1486768.

Puryear, C. I., Shenghong, T., Castagna, J.P., 2008, Comparison of frequency attributes from CWT and MPD spectral

decompositions of a complex turbidite channel model: 78th Annual International Meeting, SEG, Expanded Abstract, 393-397.

Stoner, S., 2010, Fluvial architecture and geometry of the Mungaroo Formation on the Rankin trend of the Northwest shelf of Australia: Master’s thesis, The University of Texas at Arlington.

Tellez Rodriguez, J. J. J., 2015, Seismic-sequence stratigraphy framework and architectural elements for Cenozoic strata at the

Rankin Platform sub basin, north Carnarvon basin, Australia: Master’s thesis, University of Oklahoma.

Wallet, B., 2016, Attribute expression of channel form in a hybrid carbonate turbidite formation: Interpretation, Vol. 4, No. 2,

SE75-SE86, doi: 10.1190/INT-2015-0108.1.

Zhao, T., Li F., Marfurt, K. J., 2016, Advanced self-organizing map facies analysis with stratigraphic constraint: 86th Annual

International Meeting, SEG, Expanded Abstract, 1666-1670.