Fabric and Internal Architecture of Permian Basin Turbidites Indicated by Unsupervised Machine Learning Analysis of P-P and SV-P Images

By Bob Hardage, Tom Smith, Diana Sava, Yi Wang, Rocky Roden, Gary Jones, and Sarah Stanley | Published with permission: Interpretation Journal | November 2020

Abstract

The Permian Basin of west Texas spans two major subbasins — the Midland Basin and the Delaware Basin. Both basins contain Wolfcampian- to Leonardian-age turbidites that form thick sections of prolific unconventional reservoirs. For the past several years, the most active drilling targets in the United States have been the Wolfberry turbidite interval of the Midland Basin and the Wolfbone turbidite interval in the Delaware Basin. We have used two new technologies to examine the internal architecture and fabric of a thick interval of these unconventional drilling targets with seismic reflection seismology. First, we used seismic interpretation software that uses unsupervised machine learning (ML), so that a higher level of detail could be extracted from seismic images. Second, we complemented our P-P imaging of Wolfberry turbidites with a new seismic imaging option, that being SV-P (or converted-P) imaging. Because vertical vibrators, particularly arrays of vertical vibrators, produce downgoing P and downgoing SV illuminating wavefields, SV-P reflections can usually be extracted from the same vertical-geophone responses as are P-P reflections. The combination of these two images essentially doubles the amount of information that can be extracted from data generated by P sources and recorded with vertical geophones. SV-P imaging with P sources has been ignored by reflection seismologists for decades, so we felt an obligation to illustrate the value of this ignored seismic mode. These two new tools — SV-P imaging and interpreting P-P and SV-P images via unsupervised ML — expanded our insights into the internal architecture and fabric of Wolfberry turbidites. Our work provides interpreters a much-needed example of applying unsupervised ML technology in a joint interpretation of P- and S-wave data.

Introduction

Thirteen-year-old legacy 3D 3C seismic data acquired on the western shelf of the Midland Basin, where there is a thick stack of Wolfberry turbidites, were used in this study, so these unconventional reservoirs could be
investigated with P-wave and S-wave seismic attrib­utes. Wolfberry turbidites form a high-interest drilling target of unconventional reservoirs that extend upward from the Wolfcamp (Wolf) formation to the Sprayberry (berry) formation. This Wolfberry interval of stacked turbidite flows is approximately 3000 ft (1000 m) thick across our study area.

Three-dimensional P-P, P-SV, and SV-P data volumes were interpreted with commercial software that uses unsupervised machine learning (ML) principles. Inter­pretation focused on determining if combinations of seismic attributes extracted through joint interpretations of depth-equivalent P-mode and S-mode data provide im­proved understanding of the internal architecture and fabric of turbidite assemblages. This approach allowed specific P-wave and S-wave geobodies embedded in stacked turbidites to be isolated.

Mineral counts based on analyses of thin sections cut from Wolfberry cores allowed us to propose the hy­pothesis that these seismic geobodies represent spatial distributions of distinct mineral mixtures found in ma­trices of Wolfberry turbidite units. Turbidite matrices that have different mineral mixtures have different stiff­ness coefficients. Turbidite matrix-fabric 1 thus creates P and S reflection wavelets that have different attrib­utes and/or different numerical ranges of attributes than does turbidite matrix-fabric 2. Each searching neuron that migrates through each P and S attribute space in unsupervised ML searches identifies the x-y-z coordi­nates where specific combinations of seismic attributes and specific numerical ranges of those attributes exist. Maps of these data coordinates describe a geobody. We thus view a geobody created in this study as a region inside a turbidite assemblage that has a distinct rock fabric. This fabric may be associated with a single tur­bidite, it may be only a portion of a single turbidite, or it may span a group of several individual turbidites.

SV-P (converted-P) data provided insights into the internal architecture and fabric of turbidites that were equal to, and sometimes better than, insights provided by P-P and P-SV data. The SV-P data used in this study were generated by a traditional P source (an array of three inline vertical vibrators) and were recorded with vertical geophones. Because these 3D seismic data were acquired 13 years before this study was done, this project demonstrates that SV-P data can be extracted from hundreds of thousands of square kilometers of leg­acy seismic data generated by P sources and recorded with vertical geophones. Such P source vertical-geo­phone data are preserved in many digital data libraries around the globe. There are, in fact, several thousands of square kilometers of legacy P source vertical-geo­phone seismic data across the Permian Basin alone.

Study Area

Our study area is shown in Figure 1. Data were gen­erated by an inline array of three vertical vibrators that occupied a grid of source stations distributed across a 3 × 3 mi area of surface-source illumination. Data were recorded by a 2 × 2 mi grid of 3C geophones centered inside this source-station grid. The map scale used for Figure 1 would cause this seismic grid to appear as only a small dot on this basin-view map.

Unsupervised machine learning analysis of P-P and SV-P Figure 1
Figure 1. The study area is shown as the blue rectangle on the western shelf of the Midland Basin where a small 3D 3C seismic survey was acquired. The acquisition geometry was a 2 × 2 mi grid of 3C geophones centered inside a 3 × 3 mi grid of source stations. This data acquisition grid would only be a dot at this map scale. The San Simon Channel is shown entering the Midland Basin approximately 30 mi (50 km) northwest of the study area. Most turbidite movement in the basin is ap­proximately north to south.

An objective of this paper is to illustrate that this new concept of extracting SV-P images from vertical-geo­phone data can be included in seismic interpretation studies across other areas where P source, vertical-geo­phone, legacy seismic surveys are available. Because data used in this study were recorded with 3C geo­phones, horizontal-geophone data could have been processed to extract S-S reflections generated by the direct S illuminating wavefields produced by the P source (three inline vertical vibrators). In fact, a brute stack of S-S data was constructed and was of an accept­able quality for a brute stack. However, because only a limited amount of 3C geophone data exist across most turbidite prospect areas, S-S data processing efforts were set aside to concentrate fully on demonstrating the value of SV-P modes that can be extracted from tra­ditional vertical-geophone data. There are huge amounts of legacy vertical-geophone data across the Permian Basin and other basins that will allow SV-P studies of other turbidite systems, or of any deep geologic targets, to be repeated by other investigators.

Log data spanning the Wolfberry interval

After examining many thin sections extracted from cores in logged wells, Hamlin
and Baumgardner (2012) in their seminal report for the Texas Bureau of Econo­mic Geology conclude that the mineralogical complex­ity present in Wolfberry turbidites could be reasonably represented by the three-color rock-facies distinc­tions used in Figure 2. These three general facies clas­sifications of Wolfberry turbidites were then related to gamma-ray and resistivity log readings. In Figure 2,the left-side boundary of each colored turbidite-facies col­umn indicates gamma-ray readings and the right-side boundary indicates resistivity-log values. An examina­tion of these log responses and their simplified lithofa­cies interpretations will help investigators understand the thicknesses, lateral dimensions, areal shapes, and internal mineralogical frameworks of Wolfberry turbi­dites. The voluminous report by Hamlin
and Baumgard­ner (2012) shows that Wolfberry turbidite shapes, sizes, thicknesses, and mineralogical content vary rapidly in the vertical and horizontal directions across the Mid­land Basin. One should thus expect that most seismic reflections across Wolfberry turbidite intervals will be continuous for only short distances. Seismic inter­preters should thus assume that age-equivalent turbi­dite units will be difficult to define with seismic data.

Equivalence of P-SV and SV-P imaging

One objective of this study was to include S-mode imaging in the study of Wolfberry turbidites. Data ac­quired in a legacy P source 3D 3C seismic survey were used so that the P-P and P-SV data could be used in a joint interpretation of Wolfberry turbidite targets inside the small area spanned by this seismic grid. Unfortunately, P-SV data generated in this legacy seis­mic survey have a prominent acquisition footprint. The companion P-P image showed no hint of an acquisition footprint. This P-SV acquisition footprint caused our investigation to switch to SV-P data as a substitute for P-SV data. When stratigraphic layering is approximately horizontal and lateral changes in P and S velocities are not excessive, the P-SV and SV-P images should be identical.

SV-P data extracted from the vertical geophone data contained no evidence of an acquisition footprint. In P­SV imaging that uses data generated at source station A and recorded at geophone station B, the P-SV image point is closer to receiver station B than it is to source station A. In contrast, the SV-P reflection point for this same source-receiver pair is closer to source station A than to receiver station B. This distinction in the spatial distributions of P-SV and SV-P image points evidently can cause one converted mode (P-SV in this case) to have an acquisition footprint, but its companion conver­sion mode (SV-P in this case) to not have an acquisition footprint. People who design 3D 3C seismic acquisition programs need to be sensitive to this possibility and ensure that there is a degree of randomness in the spac­ings between adjacent source stations and/or receiver stations. Random station spacings tend to eliminate ac­quisition footprints.

P-SV and SV-P images should be identical in that part of the seismic image space where the P-SV and SV-P image points overlay each other. This principle has been established by Aki and Richards (1980, 2002) who develop the following equation:

Unsupervised machine learning analysis of P-P and SV-P equation

In this equation, RSVP is the SV-P reflection coefficient at a targeted inter­face, RPSV is the P-SV reflection coeffi­cient at that same interface, .SV is the incident angle of the downgoing SV ray-path, .P is the reflected angle of the up-going P raypath, and VS and VP are, respectively, the S-wave velocity and P-wave velocity in the medium above the interface. This equation shows that RSVP and RPSV have the same algebraic sign at a reflecting interface and, be­cause of the numerical balancing effect of the angle and velocity terms in equa­tion 1, the magnitude of RSVP should usu­ally be slightly less than the magnitude of RPSV at that interface.

The principle that SV-P and P-SV images are identical was demonstrated with real 2D 9C data by Frasier and Winterstein (1990). The downgoing SV wavefield used to construct the SV-P data in their award-winning 1990 publication was generated by a horizontal vibrator. In contrast, the downgoing SV wavefields used to generate SV-P data across Wolfberry turbidites in this study were generated by an array of three inline vertical vibrators. Presently, the SV-P image illustrated in Frasier and Winterstein’s (1990) paper is considered to not only be the first, but also to be the only, published example of an SV-P image in geophysical literature. Researchers at the Bureau of Economic Geology, the University of Texas at Austin, began providing the sponsors of their research consortium private examples of SV-P images made with P sources and recorded with vertical geo­phones in 2009 and then began revealing several years of SV-P research in open publications in 2014 (Hardage et al., 2014; Hardage and Wagner, 2014a,2014b; Li and Hardage, 2015; Hardage, 2017a, 2017b, 2017c, 2017d; Karr, 2017; Li, et al., 2017; Wagner and Hardage, 2017; Gupta and Hardage, 2017; Hardage and Wagner, 2018a, 2018b). Presently, there are no published works by other investigators on the topic of practicing S-wave reflection seismology with data generated by P sources and recorded by vertical geophones.

Examples of P-P and SV-P profiles across the study area

Examples of west-to-east P-P and SV-P profiles pass­ing through the approximate center of the 3D seismic image space used in this study are shown in Figure 3.

Unsupervised machine learning analysis of P-P and SV-P Figure 2
Figure 2. Well-log cross section traversing the seismic survey area. Log curves are gamma ray (the left-side boundary of color columns) and resistivity (the right-side boundary of colored columns). The distance between the wells is ap­proximately 5 mi (8 km). Well 1 is west of our study area, well 2 is inside our seismic image space, and well 3 is east of our study area. These logs illustrate the dramatic short-distance changes in the size, thickness, shape, and mineralogical fabric of Wolfberry turbidites. Modified from Hamlin and Baumgardner (2012).

The same coordinate profile through the companion P-SV data volume is not shown because the acquisition footprint embedded in the P-SV data volume makes those data unsuitable for analysis. This P-SV acquisition footprint will be shown later in map views of the P-P, P-SV, and SV-P amplitude behaviors.

The vertical coordinates of these P-P and SV-P pro­files start slightly above the top of the Sprayberry and extend slightly deeper than the base of the Wolfcamp. The entire Wolfberry interval of stacked turbidites is thus shown in each profile. The vertical and horizontal axes are not labeled to protect the operational activity of the data owner. Figure 1
shows that the east–west San Simon Channel enters the Midland Basin a short distance northwest of our study area. As a result, some turbidite sediment inflow can be from the west and northwest at our location on the western shelf of the Midland Basin. If any inflow has a westward compo­nent, some turbidites should progress from left to right across the seismic profiles in Figure 3. The Midland Ba­sin deepens significantly to the east, and the Wolfberry turbidite interval thickens as gravity-driven flows of sediment from the north and from the western shelf fill this deeper basin area. As shown in Figure 3, SV-P data show more evidence of stacked, downlapping units like one would expect to see in gravity-driven sediment movement than do P-P data.

Interpreting P-P and SV-P images with ML software

ML is arguably the most important technology pres­ently being introduced into the seismic interpretation community. The first ML procedure used in this study was to determine what types of seismic information should be focused on as P-P and SV-P images of Wolf-berry turbidites are interpreted. This goal was accom­plished by applying a principal component analysis (PCA) to combinations of trial attributes extracted first from the P-P image, and then from the SV-P image, to determine the ranked order of importance of infor­mation provided by selected seismic attributes in each image space. In this methodology, each image space describes the value of one specific seismic attribute at each data point in the 3D image space. There is an image space for each seismic attribute that is applied to the P-P data volume and an image space for each attribute calculated for the SV-P data volume.

Unsupervised machine learning analysis of P-P and SV-P Figure 3
Figure 3. (a) West-to-east P-P profile through the central part of the Wolfberry seismic survey. (b) SV-P image along the same profile. Each image begins slightly above the top of the Wolfberry section and ends slightly below the base of the Wolfberry. The SV-P image shows numerous, down-lapping units. The internal features in P-P data are more subtle.

The second procedure was to visualize where spe­cific combinations of these critical information-bearing attributes were positioned in the P-P and SV-P seismic-survey space. This visualization objective was achieved by constructing self-organized maps (SOMs) that show image coordinates where distinct-color neurons found clusters of specific combinations of seismic attributes. A description of PCA and SOM seismic procedures and applications has been published by Roden et al. (2015).

Step 1 — PCA (determining the relative importance of seismic attributes)

Fifteen instantaneous seismic attributes that are helpful for defining information content in P-P data were calculated throughout that part of the 3D image space occupied by Wolfberry turbidites. Following these calculations, P-P data then existed not only in their traditional 3D, x-y-z, seismic-survey space that in­terpreters have used for decades, but they also existed in 15 3D, x-y-z, instantaneous-attribute spaces, each of which is also an image of the Wolfberry turbidite sys­tem. Collectively, these P-P data and their 15 instanta­neous attributes form a 16D P-P attribute space, and each data point inside the Wolfberry turbidite interval has 16 P-P attribute values.

These same 15 instantaneous seismic attributes were then calculated throughout that part of the 3D SV-P data space that encapsulated the same Wolfberry turbidites. SV-P data then existed in the traditional 3D x-y-z seis­mic survey space and also in fifteen 3D x-y-z instan­taneous seismic-attribute spaces, which created a set of 16 SV-P attribute volumes spanning the Wolfberry turbidites.

A PCA was applied to the suite of 16 P-P attribute spaces to create the P-P principal components (some­times abbreviated as PCs) that, collectively, defined the relative importance of attribute information embedded in the P-P Wolfberry turbidite images. A similar pro­cedure was then applied to the 16 SV-P attribute spaces to create a second suite of PCs that, collectively, de­fined the relative importance of attribute information embedded in the SV-P Wolfberry turbidite images.

An important ML axiom is “if data consist of N observations that have M variations, then the number of principal components required to describe the data is min{[N-1], M}.” In this study, N would be the several thousands of data points in the P-P image space (or in the SV-P image space) and M would be the 16 attributes selected to describe the information embedded in the P-P data (and also in the SV-P data). The number of principal components that needed to be calculated to properly apply unsupervised ML to either our P-P data, or to our SV-P data, was thus 16.

For this study, a PCA of our Wolfberry P-P data re­quired that a 16D ellipsoid be fitted to the 16D P-P seis­mic-attribute space. Because attributes have different units and ranges of value (e.g., frequency units versus phase units; acceleration range versus envelope-slope range), each attribute distribution is normalized so that it has zero mean and unit variance. By doing this nor­malization, no one attribute is statistically more impor­tant than any other. All possible combinations of pairs of attributes can then be compared for their relative im­portance using a simple sample-to-sample multiplica­tion and summation — i.e., a zero-lag correlation.

The first 16D principal component (PC1) is a vector from the PCA that points in the direction where the vari­ance of all pairs of these normalized attributes is a maxi­mum. Variance in normalized attributes is analogous to the “amount of information.” An attribute with a large variance provides a large amount of information. If an attribute has no variance, it is a constant value and pro­vides no information. PC1 thus describes the largest possible amount of information in the 16D P-P attribute data. The second 16D principal component (PC2) ac­counts for the second largest variance (the second larg­est amount of information) in 16D P-P attribute data that is orthogonal to PC1, PC3 accounts for the third largest variance in 16D P-P attribute data that is ortho­gonal to PC1 and PC2, and so forth until the last prin­cipal component (PC16) is considered. PC16 accounts for the direction in 16D P-P attribute data that has the least variance and that is orthogonal to all 15 previously calculated PCs. This PCA procedure was then repeated with the 16 SV-P attribute volumes to determine the 16 PCs that collectively describe the ranked order of im­portance of attribute information embedded in our SV-P image of Wolfberry turbidites.

Figure 4 illustrates the rank order in which seismic attributes contributed information to the first three P-P, SV-P, and P-SV principal components. The construction of each PC was stopped, arbitrarily, when the informa­tion content in the sequence of PCs built to approximately 90% of the total amount of available information that existed in the 16D attribute space. Interpreters use different criteria to decide how many PCs should be used in an ML interpretation. For example, some inter­preters use only the first three PCs regardless of the amount of information that is provided by those three PCs. Even with ML technology, seismic interpretation will, and should, be practiced differently by different people.

The percent of total available information carried by each principal component in Figure 4 is indicated by the percentage number written below each instantaneous-attribute list. Even though P-SV data were not used for interpretation purposes because the data were con­taminated by an acquisition footprint, P-SV data are included in this attribute list to understand which seismic attributes provide significant amounts of P-SV information.

Readers should examine the list of attributes that builds PC1 for P-P data in Figure 4a, then the list of P-P attributes that builds PC2 in Figure 4b, and then the P-P list that builds PC3 in Figure 4c to see that a high-rank P-P seismic attribute in one P-P principal component rarely appears in a high-rank position in an­other P-P principal component. This outcome occurs because PC1 points in the direction of maximum infor­mation, PC2 points in the direction of second-most maximum information, and PC3 points in the direction of third-most maximum information. High-rank PCs thus capture huge chunks of the most critical attributes, leaving only small amounts of these key attributes to be used by lower rank PCs. This same observation also applies to the SV-P and P-SV principal components in Figure 4a–4c.

Unsupervised machine learning analysis of P-P and SV-P Figure 4
Figure 4. Rank order of seismic attributes that cause the first three principal components extracted from P-P, SV-P, and P-SV Wolfberry data to contain approximately 90% of the total available information from seismic attributes. The percent of total information content embedded in each principal component is written below each attribute list. (a) Principal component 1. (b) Principal component 2. (c) Principal component 3.

The first 9 principal components (of the 16 pos­sible PCs) captured approximately 90%, or more, of the total information embedded in the P-P and SV-P data (Figure 5). Thus, only the first 9 PCs extracted from the P-P Wolfberry turbidite data, and likewise only the first 9 PCs extracted from the SV-P data, were used in un­supervised ML analyses. This choice to ensure that 90% of the information embedded in P-P and SV-P im­ages of Wolfberry turbidites would be used was strictly an arbitrary decision. A 70% or 80% level of information can often be sufficient to illustrate critical geologic features.

Comparison of P-P and S-mode principal components

Because this study uses unsupervised ML concepts in a joint interpretation of P-wave and S-wave images, it is important to compare principal components deter- mined from companion P-P, SV-P, and P-SV data volumes. The objective of this comparison is to define which seismic attributes make major contributions to the first two or three principal components of P-P data and then determine if these same attributes make significant contributions to the first two or three principal components of S-mode data. By understanding which seismic attributes build the most dominant P and S principal components, interpreters will know if P-P data and S-mode data are affected in approximately the same way by the same seismic attributes, or if the attributes that influence S-mode data are significantly different from the attributes that influence P-P data.

The information in Figure 4 allows this type of investigation to be done. Figure 4a shows that the first four attributes that contribute to the first principal component of P-P, SV-P, and P-SV data are identical and are essentially in the same ranked order of contribution. Likewise, the first three ranked attributes that contribute information to the second P-P, SV-P, and P-SV principal component are identical, and they are essentially in the same ranked order. This consistency continues into principal component 3, where the first two ranked attributes of the third P-P, SV-P, and P-SV principal component are identical.

At least two important conclusions can be made
from these results. First, the acquisition footprint em- bedded in the P-SV data does not cause P-SV principal components to have a dramatically different information structure than the information structure contained in the PCs of P-P and SV-P data, which have no acquisition footprint. PCA results in the Wolfberry P-SV image space are not significantly different from PCAs in the P-P and SV-P image spaces. This observation leads to the conclusion that, in this case, P-SV attributes were not affected by the P-SV acquisition footprint. One should not assume, however, that attributes of a seismic wave mode will always remain stable inside and outside of acquisition-footprint areas.

Second, and probably more important, is the fact that P-P data are not influenced by a different set of seismic attributes than are SV-P converted-mode data. Our P and S data are, to first-order accuracy, influenced by the same seismic attributes in approximately the same ranked order of influence. A factor that may need to be considered is that the P and S modes used in this study were generated by the same source — an inline array of three vertical vibrators — at the same instant of clock time and at exactly the same surface coordinates. To carry forward this significant observation, a PCA now needs to be applied to 3C data where the P-P mode is generated by a vertical vibrator but the S-wave data are generated by a different source — perhaps a horizontal vibrator — at a different clock time (perhaps even on a different day) and at surface coordinates that are slightly displaced, perhaps by only 2 or 3 m, from the P-source position. Other questions can also be considered, such as, is the finding that the rank order of attributes that influence P data is identical to the rank order of attributes that influence S data an outcome that is unique to this particular seismic survey, or to the particular data window that was analyzed, or to the type of rocks that were imaged? This early effort to compare the seismic-attribute structure of P and S data is placed in the public domain for others to expand on and to address questions of this nature.

Step 2 — Displaying principal component information (profile views)

To use principal component information, it is essential to implement a procedure that takes advantage of multiple coincident attributes in an attribute space that constitutes a common-image space. The first advantage of implementing this procedure is the dimensionality reduction afforded by a PCA analysis. The list of 16 “trial attributes” used in this study was reduced by interpreting our PCA analysis and selecting a shorter list of nine “important attributes” (Figure 5). One may compute a new set of attributes based on just these nine important attributes, and a second PCA analysis of this reduced number of attributes can then be made if desired (Guo, et al., 2009; Brito, 2010; Chopra and Marfurt, 2014). A second set of important attributes can then be computed from the PCs of the first important-attribute set. An alternative is to select the important attributes from the first PCA analysis for further study.

PCA is a linear process and is a suspect procedure if input data themselves are not linearly related. Nonlinear properties do exist among seismic attributes because some attributes are quite different from others. A nonlinear extension of PCA for seismic interpretation shows promise. Although PCA is suitable for dimensionality reduction from a long list of “trial attributes” to a shorter list of “important attributes,” the reduced attribute dimension that results may still be too large for practical seismic interpretation. Each seismic image space is an image of only a single attribute, so many important relationships between attributes are difficult to recognize in a 9D attribute space, or in any attribute space with a dimensionality greater than 3. What is needed is a simultaneous analysis of all attributes (nine attributes in our case). Such simultaneous analyses are possible because of the spatial coincidence of all attribute images.

There is a nonlinear ML technology that reduces attribute dimensionality from the “important attribute” level (nine dimensions in our case) to only two dimensions. This technique is called the self-organizing map (SOM) (Kohonen, 2001; Bishop, 2006; Haykin, 2009; Wilmott, 2019). SOM is a type of ML that preserves clusters of natural information density as a 2D SOM and simultaneously preserves each natural cluster of information in its input attribute dimensions (Haykin, 2009, p. 442–445). A 2D SOM thus contains self-organization properties that allow geologic features embedded in the dimensions of “important” seismic attributes to be displayed and examined.
Some ML algorithms, such as deep learning, are based on an attractive property that an algorithm trains on input training data so that the characteristics of these training data are preserved. Such algorithms are self-adapting. SOM, in contrast, offers an ML property in which regions of information in several reduced dimensions reflect similar regions of information in the original dimensions. Such algorithms are self-organizing, a property that is quite advantageous for seismic interpretation.

SOM ML identifies and preserves information through a network of neurons that learn characteristics of seismic data through simple training rules. In SOM procedures, training rules are refined as one training step follows another. The network itself is a 2D mesh, and the training rules define how information is shared between neurons. In seismic interpretation, training samples are usually constrained to be between two stratigraphic horizons, and each sample is a multi-attribute vector. The volume of training samples is generally confined to a specific region of interpretation interest.

When a SOM procedure is used, ML is, by definition, unsupervised learning because none of the training samples in the region of interest have classification labels. A complete set of classification labels (i.e., winning neuron numbers) are developed by the SOM process. A P-P SOM is created by neurons that have searched through P-P attribute spaces to find natural clusters of multi-attribute data samples. When all neuron training ends, each training sample is associated with one of the neurons in this SOM. After machine training, the neurons have stopped their searching and the result is called the winning neuron set. After training, each seismic sample has a winning neuron. In other words, each seismic sample has been classified. For example, if the SOM is an 8 × 8 network of neurons like we used, each seismic sample has been classified as one of the 64 possible winning neurons. These SOM classifications constitute a new seismic image based on classification numbers.

The principal reason why these SOM procedures are so valuable is that they reduce ML results from the multidimensional attribute space (a 9D space in our example) to a simpler 2D color grid that allows interpreters to see how attribute information is distributed in P-P and SV-P data spaces. The number of cells in this 2D color grid and the distribution of colors among these grid cells are defined by the person interpreting the seismic data. Interpreters should experiment with different magnitudes of neuron populations, ranging from a small population (maybe a 5 × 5 grid), to a medium-size population (perhaps an 8 × 8 grid), to a large population (maybe a 10 × 10 grid). An interpreter can then view each neuron population with different color maps and decide which color depiction of which neuron population best illustrates what they want to see.

One SOM procedure was used to visualize, through classification, where unsupervised ML positioned various classified samples of Wolfberry P-P attributes in the P-P multi-attribute space. A second SOM procedure was used to visualize where unsupervised ML placed various classified samples of Wolfberry SV-P attributes in the SV-P multi-attribute space. The resulting P-P and SV-P classification images that we show in this paper thus display natural clusters of attributes identified by unsupervised ML in the multi-attribute space.

Vertical profiles of SOM results through the central part of P-P and SV-P attribute spaces are displayed as Figure 6. The 2D SOM displays of the 64 colored neurons used to construct each of these profiles are shown later. These SOM vertical profiles are the same profiles shown in Figure 3 that use a traditional linear color scale. Details about the internal architecture and fabric of stacked Wolfberry turbidites are better expressed by the SOM displays in Figure 6, particularly by the SV-P profile, than by traditional color displays. In particular, the SV-P color section has more discontinuous, hummocky events expected of a thick stack of turbidites than do the flatter, smoother events in the P-P color section.

Step 3 — Displaying principal component information (map views)

It is particularly informative to view SOM volumes of Wolfberry turbidites in map view. Examples of horizontal slices through P-SV, P-P, and SV-P SOM volumes are displayed in Figure 7. The color at any x-y coordinate in these maps defines which of the winning neurons identified in the SOM displays resides at that x-y coordinate. The dimension of each side of these square constant-time slices is slightly more than 3 km. Note that each of the three slices through the P-SV volume reveals attribute distributions that have straight edges oriented in the same directions that source lines and receiver lines were deployed when acquiring the seismic data. These straight-edge boundaries of P-SV seismic facies are classic evidence that an acquisition footprint is embedded in the P-SV data volume. No evidence of an acquisition footprint is exhibited in either the P-P or SV-P horizon slices. Because these acquisition footprint effects contaminate P-SV data and mask geologic effects, the use of P-SV data was abandoned in this study of Wolfberry turbidites.

Traditionally, an acquisition footprint is revealed by linear map irregularities in reflection amplitude that align with source and/or receiver lines. However, the data displayed in Figure 7 are not just reflection amplitudes. The data are, instead, SOMs of winning neuron classification samples created by unsupervised ML in the P-SV multi-attribute space. Examination of Figure 4 shows that the seismic attributes embedded in PCs of P-SV data involve amplitude attributes, numerous frequency attributes, and several phase attributes. This P-SV SOM analysis thus shows that an acquisition foot- print is far more insidious than just being linear trends of overamplified reflection amplitudes. Instead, an acquisition footprint distorts and incorrectly positions all of the seismic attributes along linear trends that correspond to the azimuths of source lines and/or receiver lines. All P-SV attributes were thus abandoned in this study. Attention focused only on P-P amplitude data and attribute data and on SV-P amplitude data and attribute data.

Unsupervised machine learning analysis of P-P and SV-P Figure 6
Figure 6. SOM versions of the P-P and SV-P profiles shown in Figure 3. The colors define the locations where many (probably not all) of the 64 searching neurons used in this SOM analysis found the specific combinations of seismic attributes they sought and became winning neurons. Each neuron in this 64-neuron search group was assigned a distinct color (Figure 10). More details about the internal architecture and fabric of Wolfberry turbidites are revealed in these types of SOM displays than in traditional linear-color displays such as those in Figure 3.

Constant-time horizontal slices do not follow strati- graphic boundaries, so the displays in Figure 7 are not appropriate for seismic stratigraphy purposes. In addition, readers should not assume that side-by-side dis- plays of the P-SV, P-P, and SV-P horizons displayed in Figure 7a–7c are exactly depth equivalent. The left-to-right slices in each data panel are only approximately at equivalent depths. Approximations of depth-equivalent horizon slices are sufficient for providing a first look at depositional patterns of seismic facies provided by each imaging option (P-SV, P-P, and SV-P). These depositional patterns, in turn, provide information about the internal architecture and internal fabric of Wolfberry turbidites.

Note how the seismic facies features in SV-P slices are aligned southwest to northeast as they would be if there were episodic waves of gravity-driven sediment, often extending for lateral distances of several kilometers, spilling into the Midland Basin from the north-west where the San Simon Channel opens into the basin (Figure 1). Such episodic features exist throughout most of the SV-P Wolfberry interval. Southwest-to- northeast alignments of SV-P seismic facies begin filling the basin in early Wolfcampian time (Figure 7c), move toward the present-day shelf and become more prominent in younger geologic time (Figure 7b), and continue to regress toward present-day shelf and become even more prominent in Leonardian time near the top of the Wolfberry interval (Figure 7a).

In contrast, P-P data display large sheet-like features in the upper Wolfberry (Figure 7a) and show chaotic discontinuous behavior in the middle and lower Wolf- berry (Figures 7b and 7c). It is possible, albeit more difficult, to infer from these P-P SOM horizontal slices that Wolfberry turbidite influx from the northwest moves across the image space.

Our SV-P data provided a more confident depiction of shelf-ward movement of Wolfberry turbidite features over geologic time than did our P-P data. SV-P imaging of Wolfberry turbidites is thus quite valuable, perhaps even more valuable than P-P imaging in some respects. These SOM slices are valuable because their data are classified by a rather large number of winning neurons
(64) that allow an interpreter to see the details of the internal architecture and fabric of Wolfberry turbidites. Each neuron classification region in the attribute space has its own combination of attributes that are uniquely different from all other regions. No one attribute is the “key” to understanding how seismic reflections respond to turbidite features in the subsurface.

Why do Wolfberry P-P and SV-P images differ?

There are significant differences between P-P data displays and their companion SV-P data displays in this paper. These differences cause some interpreters who have worked with only P-P images to question the reliability of SV-P images made from the same vertical-geophone data and/or to struggle with depth registering SV-P images with their companion P-P images. A fundamental principle of joint-interpretation of P-P and SV-P data is that an interpreter must be aware that some rock boundaries will generate a P-P reflection but will not create an SV-P reflection. Similarly, some rock boundaries will generate an SV-P reflection but will not produce a P-P reflection. It takes a new mindset for interpreters to accept the concept that when P-P data produce a reflection at a rock interface, but the companion SV-P data do not, both images can be correct.

An illustration of the fundamental difference in P-P and SV-P reflectivity physics for Wolfberry turbidites is demonstrated by data displayed in Figures 8 and 9. The erratic distribution of minerals in low-porosity, low-permeability Wolfberry lithofacies is illustrated in Figure 8. These insights into the mineral fabric of Wolfberry depositional units were determined by a laborious, detailed, mineral count in thin sections cut from 182 Wolfberry cores (Hamlin and Baumgardner, 2012). Mineral mixtures observed in these thin sections are the data plotted in Figure 8. These X-ray diffraction data, coupled with the log facies in Figure 2, indicate that the mineralogical composition of Wolfberry turbidites can vary in dramatic fashion between immediately adjacent Wolfberry facies units.

This ground truth insight into the complexity of mineral mixtures that form Wolfberry rock fabric is used in the reflectivity modeling illustrated in Figure 9. Elastic coefficients of each mineral component in mineral mixtures 1, 2, 3 identified in Figure 9d were combined to construct the elastic properties of hypothetical Wolfberry facies units that form the lower layer of the two-layer model in Figure 9c. The methodology used to create composite rocks that have matrices with specific percentages of quartz, clay, and calcite was patterned after a procedure proposed by Hill (1965). The P-P and SV-P reflection behaviors in Figure 9a and 9b were then calculated for the boundary separating the two rock units shown in Figure 9c.

Reflectivity curves in Figure 9a indicate that rock boundaries 1 and 2 would be invisible, or quite weak at best, in a P-P image, but rock boundary 3 would create a robust P-P reflection. In contrast, the SV-P reflectivity curves in Figure 9b show that rock boundary 3 would not appear in an SV-P image but rock boundaries 1 and 2 would produce respectable SV-P reflections. These opposite outcomes illustrate a fundamental axiom of seismic stratigraphy that needs to be applied when analyzing Wolfberry turbidites: P-P images and SV-P images will differ in some areas of image space, yet both images can still be correct. Said another way, SV-P data provide seismic interpreters a different set of Wolfberry stratal surfaces than do P-P data, and both sets of stratal surfaces (SV-P and P-P) should be used in stratigraphic studies of the Wolfberry interval. The seismic interpretation community does not yet practice S-wave seismic stratigraphy and uses only P-P data in seismic stratigraphy studies. Perhaps with the knowledge that S-mode seismic stratigraphy can be practiced with SV-P data extracted from vertical-geophone data, attention to using P and S data in a unified seismic stratigraphy interpretation will begin to be practiced.

Unsupervised machine learning analysis of P-P and SV-P Figure 7
Figure 7. Horizontal slices through the P-SV, P-P, and SV-P SOM volumes. The color at any map coordinate identifies which of the colored winning neurons identified in SOM displays in Figure 10 resides at that map coordinate. These specific SOM displays show that P-SV data are contaminated by an acquisition footprint. No acquisition footprint appears in P-P and SV-P data. WB = the thick- ness of Wolfberry, and slices are positioned approximately (a) 1/3, (b) 1/2, and (c) 2/3 of WB below the top of the Wolfberry. The display areas are approximately 3 × 3 km.

Extracting turbidite geobodies from SOM volumes The term geobody is often used to designate a distinct definable component of a depositional system. The size, shape, and location of a geobody extracted from seismic data will change when different criteria are used to identify seismic geobodies. In our study, unsupervised ML was used to train neurons on P-P and SV-P attribute volumes so that each neuron could search for combinations of attribute coordinates where clusters of seismic attributes provided a best match with each other in attribute space. Each winning neuron found a significant number of P-P and SV-P attribute co- ordinates that satisfied that neuron’s search parameters. Families of x-y-z coordinates that constitute a distinct seismic geobody in the seismic survey space are also found in the seismic attribute space. If there are other seismic geobodies elsewhere in the seismic image space that have a specific combination of seismic attributes, these geobodies will also be found at the same x-y-z coordinates in each attribute space. These common seismic geobodies “stack” at their consistent coordinates in the attribute space.

Unsupervised machine learning analysis of P-P and SV-P Figure 8
Figure 8. Mineral percentages in matrices of Wolfberry lithofacies determined from X-ray diffraction analysis of thin sections cut from 182 Wolfberry cores. Modified from Hamlin and Baumgardner (2012).

Unsupervised machine learning analysis of P-P and SV-P Figure 9
Figure 9. Contrasting behaviors of (a) P-P and (b) SV-P reflectivity at the boundary between (c) a low-porosity, constant-minerology top layer, and a bottom layer of low-porosity rock having one of three possible lithofacies mixtures of minerals defined in (d).

The power of multi-attribute ML is that the stacking of one or more seismic geobodies occurs in a common natural cluster in attribute space. For completeness, it should be stated that not all winning neurons locate geologically important seismic geobodies. Some detected geobodies are not geologic but are clumps of random noise. Others are acquisition footprints, multiples, or other coherent noise events. An interpreter has to decide which geobodies are important and which have no geologic value.

A winning neuron is associated with a natural cluster in the attribute space. Such natural clusters represent concentrations of seismic samples that possess roughly the same combination of attribute values. In essence, a winning neuron may represent more than one seismic geobody if those geobodies share the same combination of attributes discovered by ML. Winning neurons align with these natural clusters. Not all winning neurons identify geologic geobodies, but all similar seismic geobodies do align with the same center of coherent energy revealed in attribute space. Sorting out which seismic geobodies are geologic and which are not is a new domain of ML assistance in seismic interpretation that will be continually improved.

This paper discusses how readily available ML interpretation of P-P and SV-P data can take interpreters deeper into their understanding of how Wolf- berry seismic geobodies are related to Wolfberry geologic geobodies. The starting point that we invoked was to invoke the premise that Wolfberry P-P and SV-P seismic responses are based on the mineralogy properties of Wolfberry rocks (Figure 9). When, as indicated in Figure 8, mineral mixture 1 occurs in a portion of one Wolfberry turbidite, but mineral mixture 2 occurs in a portion of turbidite unit 2, then turbidite 1 and turbidite 2 have different stiffness coefficients. When the rock stiffness coefficients change between coordinates 1 and 2, the seismic attributes also vary between coordinates 1 and 2. In unsupervised ML, searching neurons will identify the area around coordinates 1 as geobody 1 and the area around coordinates 2 as geobody 2.

A geobody identified in this study is thus probably best described as a seismic object that has a different rock fabric than the rock fabric associated with other detected geobodies. This premise caused the terminology “fabric and internal architecture of turbidites” to be used in the title of this paper. A geobody identified in this study should not be assumed to be an individual turbidite but to be a rock volume inside stacked Wolfberry turbidites in which there is a reasonably consistent rock fabric.

The number of neurons that search through seismic attribute space is defined by the person doing a seismic interpretation. In this study, 64 searching neurons were used for training, and each resultant winning neuron was identified with a specific color. Just like a display of seismic data products with a color scale has been a valuable interpretation tool for several decades, SOM winning neurons are also identified with a 2D color map. The 2D eight-cell by eight-cell color maps dis- played in Figure 10 identify the colors assigned to the 64 winning neurons that searched the P-P attribute space (Figure 10a) and the colors assigned to the 64 winning neurons that searched the SV-P attribute space (Figure 10b).

The neurons shown in Figure 10 are assigned a hexagonal shape, which allows each neuron to have two to six immediate neighboring neurons. Information passes through each of the six membranes around a neuron to its immediate neighbors. If neuron A decides to move a distance X toward natural attribute cluster C, this cross-membrane information sharing causes its immediate neighbors to move a small fraction of X toward C. This process is one training step for only neuron A and its immediate neighbors. When similar decisions and movements occur for every neuron (64 neurons in this study), one training step has occurred for the total neuron population. This neuron training continues until neuron migration efforts fail to advance a neuron a specified distance. At the end of the training, each neuron has found its natural clusters of attributes.

The choice of 64 searching neurons in this study was an arbitrary decision. Other interpreters may prefer to reduce their winning-neuron map to a 5 × 5 grid and use only 25 searching neurons or to use a 10 × 10 color grid and increase the number of searching neurons to 100. Geobodies will have different sizes and shapes as the dimensions of a winning-neuron map vary, i.e., as the number of winning neurons varies. Interpreters have to decide what population of neurons best provides the information they seek.

When each searching neuron, during its training, finds a data point where there is the combination of seismic attributes that it seeks, the x-y-z coordinates of that image point are assigned to that neuron’s colored cell. At the conclusion of all neuron migrations through the multidimension attribute space, each colored cell in each 2D winning-neuron map in Figure 10 contains several tens to several thousands of sets of x-y-z coordinates that define where each winning neuron should be distributed in the seismic attribute space.

A color display of a seismic data volume that uses only one cell from a 2D SOM winning-neuron map (e.g., cell 14 in Figure 10a) defines the spatial distribution throughout the P-P x-y-z space where winning-neuron 14 found its unique combination of P-P seismic attributes during neuron training. This spatial distribution where neuron 14 was the winning neuron may be a population of only a few x-y-z coordinates, or it may be a population of several hundreds of x-y-z coordinates.

A display of these single-color seismic coordinates identifies seismic geobodies that have similar seismic attribute properties. For example, seismic samples associated with seismic geobodies identified by SOM winning-neuron index 37 will be colored a dark blue with the color map of Figure 10a. The same seismic geobodies would be colored a yellow shade if the color map of Figure 10b is used. A seismic image created by SOM classification is a valuable interpretation tool. The choice of number of winning neurons and the choice of colors assigned to those neurons are arbitrary, but important, decisions made by each seismic interpreter.

Unsupervised machine learning analysis of P-P and SV-P Figure 10
Figure 10. Color maps designating winning neurons that migrate through the P-P attribute space and the SV-P attribute space. Color map (a) displays the 64 neurons that searched the P-P attribute space, and color map (b) shows
the 64 neurons that searched the SV-P attribute space. Each neuron is depicted as a hexagon with six straight sides and six internal angles of 120°. During neuron training, varying degrees of information sharing occur between neighboring neurons.

Examples of P-P Wolfberry geobodies are shown in map view in Figure 11. Figure 11a is a constant-time horizontal slice through the P-P data volume. This horizontal slice is positioned slightly more than one-fourth of the Wolfberry interval below the top of the Spray- berry, and all 64 colors of the P-P SOM in Figure 10a are active. It is difficult to interpret individual Wolfberry regions with this display format. In fact, based on just this one volume slice, it is challenging to even state that the imaged geology is a turbidite system.
When specific P-P geobodies are examined, however, there are better pictures of Wolfberry turbidite deposition. Figure 11b shows the P-P geobody associated with only color cell index 28 (i.e., winning neuron index 28) of Figure 10a. Now, there are indications of waves of turbidite deposition coming from the north- west. The arrow in the upper-left corner of the display depicts an inferred direction of sediment flow. The orientation of this depositional-flow arrow is approximately the azimuth direction from the mouth of the San Simon Channel to the study site (Figure 1). A tentative model of the internal fabric and architecture of Wolfberry turbidites at this location can begin to be constructed by assuming that each P-P winning neuron (e.g., neuron index 28) represents a unique mixture of inflowing turbidite minerals. This assumption is based on the variable mineralogy of Wolfberry turbidites exhibited in Figure 8 and on the P-P mineralogy-dependent reflectivity physics illustrated in Figure 9.

Unsupervised machine learning analysis of P-P and SV-P Figure 11
Figure 11. (a) A constant-time horizontal slice through the P-P SOM volume of the Wolfberry turbidite interval. This display allows all 64 P-P winning neurons to be active. (b) P-P geobody found by winning-neuron index-28 from Figure 10a. (c) Inclusion of a second P-P geobody found by winning-neuron index-35 (Figure 10a). (d) Addition of a third P-P geobody found by winning-neuron index-26 (Figure 10a). The arrow in the upper left corner of each panel proposes that turbidite flow enters the study area via the east–west San Simon Channel that separates the Northern Shelf and the Central Basin Platform (Figure 1).

In Figure 11c, the P-P geobody associated with P-P winning neuron index 35 is added, and in Figure 11d, the geobody associated with P-P winning neuron index 26 is added. These latter displays indicate that each P-P geobody could be interpreted as a turbidite influx that transports a unique combination and percentage-mixture of matrix-forming minerals. Each P-P geobody contributes to the internal architecture and fabric of Wolfberry turbidites. This methodology takes interpreters a long way toward understanding the geologic processes that built Wolfberry turbidites and the turbidite architecture and fabric resulting from these processes.

Winning neurons 28, 35, and 26 are near neighbors in a small region of the SOM, which means that self-organization will cause their associated natural clusters to be near neighbors in an SOM result. For example, Figure 10a shows that neurons 26, 28, and 35 communicate with each other through their common connections to neuron 27. Figure 11 then shows that indeed geobodies 26, 28, and 35 nestle against each other in the multi-attribute space.

SOM analysis is expanded to SV-P geobodies using the progression of winning neurons displayed in Figure 12. Figure 12a shows a horizontal time slice positioned almost one-third of the Wolfberry interval below the top of the Sprayberry. This full-color display reveals repeated rows of seismic facies aligned approximately southwest to northeast. Figure 12b shows the SV-P geobody identified by SV-P winning neuron index 40 of Figure 10b. The SV-P geobody associated with SV-P winning neuron index 30 is added in Figure 12c, and the SV-P geobody associated with SV-P winning neuron index 25 is added in Figure 12d. Figure 10b shows that neurons 25, 30, and 40 are well separated in the SV-P SOM grid and thus do not share significant information with each other. As a result, geobodies 25, 30, and 40 in Figure 12 are separated from each other, and the separation distances between these three geobodies in Figure 12d are roughly proportional to the distance between neurons 25, 30, and 40 in their winning-neuron grid (Figure 10b).

Unsupervised machine learning analysis of P-P and SV-P Figure 12
Figure 12. (a) A constant-time horizontal slice through the SV-P SOM volume of the Wolfberry turbidite interval. This display allows all 64 SV-P winning neurons to be active. (b) SV-P geobody found by winning-neuron index-40 from Figure 10b. (c) Inclusion of a second geobody found by winning-neuron index-30 (Figure 10b). (d) Addition of a third SV-P geobody found by winning-neuron index-25 (Figure 10b). The arrow in the upper left corner of each panel proposes that turbidite inflow is via the east–west San Simon Channel that separates the Northern Shelf and the Central Basin Platform (Figure 1).

All of these SV-P geobodies have an orientation that seems to conform with a consistent sediment provenance. By repeated additions of SV-P winning neurons onto the horizontal time slice, the internal architecture and fabric of Wolfberry turbidites, as defined by SV-P reflectivity, react to different mineral mixtures of incoming sediment (Figure 9). As a result, the positions and orientations of elements in the full-color SV-P winning-neuron displayed in Figure 12a differ from the positions and orientations of elements that comprise the P-P winning-neuron display in Figure 11a.

A tentative working hypothesis is that these dual P-P and SV-P SOM geobody analyses allow interpreters to not only have a broader understanding of the internal architecture of Wolfberry turbidites but to also have a possible picture of the distributions of mineral mixtures within Wolfberry turbidites.

Conclusions

Interpreting Wolfberry seismic images with unsupervised ML software exposes details about the internal architecture and fabric of Wolfberry turbidites that cannot be seen with conventional seismic interpretation procedures. Principal component and SOM analyses are invaluable for determining which seismic attributes, used in what percentage combinations, provide maximum information about Wolfberry turbidites. It is advisable, and probably essential, that principal components used in ML interpretations of Wolfberry turbidites be calculated in seismic volumes constrained to the Wolfberry turbidite interval. Principal components created to describe Wolfberry turbidites should not be influenced by data that image nonturbidite geology.

SOMs are effective tools for visualizing how natural clusters of combinations of Wolfberry seismic attributes are distributed in attribute space. Once a SOM of Wolfberry attributes, picked with the aid of principal components, is displayed with an appropriate color palette, seismic geobodies embedded in Wolfberry geologic turbidites can be extracted and analyzed. Each geobody represents the spatial distribution of a specific weighted-combination of seismic attributes found in the Wolfberry image space by unsupervised ML in the attribute space.

This study shows that SV-P data are valuable for imaging Wolfberry turbidites. SV-P images show turbidit-like Wolfberry features with quality equal to, and sometimes better than, features observed in P-P data. Because P-P reflectivity and SV-P reflectivity react to different interfaces in a propagation medium where rocks have spatially varying mineral mixtures, SV-P data contain geobodies that describe a different set of gravity-driven mineral distributions than do geobodies extracted from P-P data.

Rock-physics analysis of the effects that alterations in mineral mixtures of a rock matrix have on P-P and SV-P reflectivity introduces the concept that Wolfberry geobodies define volumes inside stacked Wolfberry turbidites where there are unique combinations of fabric-forming minerals. P-P attributes contain geobody information that respond to the effects that mineral composition has on P-P reflectivity; similarly, SV-P attributes contain geobodies that respond to the effects that mineral composition has on SV-P reflectivity. This tentative hypothesis provides two independent descriptions of the internal architecture and fabric of Wolfberry turbidites. Future work needs to focus on how these two sets of geobodies (a P-P set of geobodies and an SV-P set of geobodies) should be integrated into a coherent description of the internal fabric and architecture of a turbidite system and the provenance of turbidite-forming sediments.
Wolfberry operators should examine their legacy P-source seismic data and determine if P-source data recorded by vertical geophones contain SV-P reflections of usable quality. The image value of SV-P data is too great to continue to ignore the fact that SV-P reflections exist in thousands of square miles of legacy P-source data and are available for imaging exploration targets at zero data-acquisition cost.

Acknowledgments

We thank Leah Mann of Advertas Inc. for preparing the graphics for this publication.

Data and materials availability

Data associated with this research are confidential and cannot be released.

References

Aki, K., and P. G. Richards, 1980, Quantitative seismology — Theory and methods: W.H. Freeman and Co.

Aki, K., and P. G. Richards, 2002, Quantitative seismology— Theory and methods: W.H. Freeman and Co.

Bishop, C. M., 2006, Pattern recognition and machine learning: Springer, 561–565.

Brito, M., 2010, Principal component analysis for strati­graphic imaging improvement and facies predictions: 80th Annual International Meeting, SEG, Expanded Abstracts, 2401–2405, doi: 10.1190/1.3513333.

Chopra, S., and K. J. Marfurt, 2014, Churning seismic attrib­utes with principal component analysis: 84th Annual International Meeting, SEG, Expanded Abstracts, 2672–2676, doi: 10.1190/segam2014-0235.1.

Frasier, C., and D. Winterstein, 1990, Analysis of conven­tional and converted mode reflections at Putah Sink, California using three-component data: Geophysics, 55, 646–659, doi: 10.1190/1.1442877.

Guo, H., K. J. Marfurt, and J. Liu, 2009, Principal compo­nent spectral analysis: Geophysics, 74, no. 4, P35–P43, doi: 10.1190/1.3119264.

Gupta, M., and B. Hardage, 2017, Improved reservoir delin­eation by using SV-P seismic data in Wellington field, Kansas: 87th Annual International Meeting, SEG, Ex­panded Abstracts, 5182–5186, doi: 10.1190/segam2017­-17559574.1.

Hamlin, S., and R. Baumgardner, 2012, Wolfberry (Wolf­campian-Leonardian) deep-water depositional systems in the Midland Basin — Stratigraphy, lithofacies, reservoirs, and source rocks: Report of Investigations No. 277, Bureau of Economic Geology, The University of Texas at Austin.

Hardage, B. A., 2017a, Practicing S-wave reflection seis­mology with “P-wave” sources — Concepts, principles, and overview: 87th Annual International Meeting, SEG, Expanded Abstracts, 5152–5156, doi: 10.1190/segam2017­-17256173.1.

Hardage, B. A., 2017b, Examples of SV-P images made with P sources and vertical geophones: 87th Annual International Meeting, SEG, Expanded Abstracts, 5187– 5191, doi: 10.1190/segam2017-17430569.1.

Hardage, B. A., 2017c, Real-data comparisons of direct-S modes produced by “P” sources and “gold standard” S sources: 87th Annual International Meeting, SEG, Ex­panded Abstracts, 2481–2485, doi: 10.1190/segam2017­-17430461.1.

Hardage, B. A., 2017d, Land based S-wave reflection seis­mology with P sources — Does it work?: 87th Annual International Meeting, SEG, Expanded Abstracts, 6067–6071, doi: 10.1190/segam2017-w11-04.1.

Hardage, B. A., D. Sava, and D. Wagner, 2014, SV-P — An ignored seismic mode that has great value for inter­preters: Interpretation, 2, no. 2, SE17–SE27, doi: 10.1190/INT-2013-0096.1.

Hardage, B. A., and D. Wagner, 2014a, Generating direct-S modes with simple, low-cost, widely available seismic sources: Interpretation, 2, no. 2, SE1–SE16, doi: 10.1190/INT-2013-0095.1.

Hardage, B. A., and D. Wagner, 2014b, S-S imaging with vertical-force sources: Interpretation, 2, no. 2, SE29– SE38, doi: 10.1190/INT-2013-0097.1.

Hardage, B. A., and D. Wagner, 2018a, Direct-SV radiation produced by land-based P sources — Part 1: Surface sources: Interpretation, 6, no. 3, T569–T584, doi: 10.1190/INT-2018-0046.1.

Hardage, B. A., and D. Wagner, 2018b, Direct-SV radiation produced by land-based P sources — Part 2: Buried explosives: Interpretation, 6, no. 3, T585–T599, doi: 10.1190/INT-2018-0047.1.

Haykin, S., 2009, Neural networks and learning machines, 3rd ed.: Pearson.

Hill, R., 1965, A self-consistent mechanics of com­posite materials: Journal of the Mechanics and Physics of Solids, 13, 213–222, doi: 10.1016/0022-5096(65) 90010-4.

Karr, B., 2017, SV-P imaging compared to P-SV imaging — Analysis of statics and velocities required to create an SV-P image: 87th Annual International Meeting, SEG, Expanded Abstracts, 5177–5181, doi: 10.1190/segam2017-17728712.1.

Kohonen, T., 2001, Self organizing maps: Third extended addition: Springer, Series in Information Services.
Li, Y., and B. A. Hardage, 2015, SV-P extraction and imaging for far-offset vertical seismic profile data: Inter­pretation, 3, no. 3, SW27-SW35, doi: 10.1190/INT-2015­-0002.1.

Li, Y., D. Wang, S. Shi, and X. Cui, 2017, Gas reservoir char­acterization using SV-P converted wave mode — A case study from western China: 87th Annual International Meeting, SEG, Expanded Abstracts, 5172–5776, doi: 10.1190/segam2017-17726552.1.

Roden, R., T. Smith, and D. Sacrey, 2015, Geologic pattern recognition from seismic attributes: Principal component analysis and self-organizing maps: Interpretation, 3, no. 4, SAE59–SAE83, doi: 10.1190/INT-2015-0037.1.

Wagner, D., and B. A. Hardage, 2017, Using finite-difference modeling to understand direct-SV illumination produced by P sources: 87th Annual International Meeting, SEG, Expanded Abstracts, 5167–5171, doi: 10.1190/segam2017-17661718.1.

Wilmott, P., 2019, Machine learning — An applied mathematics introduction: Panda Ohana Publishing.

Welcome Back!

Download PDF here

OR

Request access by filling the form below to download full PDF.
Name(Required)
Most Popular Papers
Systematic Workflow for Reservoir Characterization in Northwestern Colombia using Multi-attribute Classification
Systematic Workflow for Reservoir Characterization in Northwestern Colombia using Multi-attribute Classification
A workflow is presented which includes data conditioning, finding the best combination of attributes for ML classification aided by Principal ...
Read More
Net Reservoir Discrimination through Multi-Attribute Analysis at Single Sample Scale
Net Reservoir Discrimination through Multi-Attribute Analysis at Single Sample Scale
Published in the special Machine Learning edition of First Break, this paper lays out results from multi-attribute analysis using Paradise, ...
Read More
Seismic Facies Classification Using Deep Convolutional Neural Networks
Seismic Facies Classification Using Deep Convolutional Neural Networks
Using a new supervised learning technique, convolutional neural networks (CNN), interpreters are approaching seismic facies classification in a revolutionary way ...
Read More
Shopping Cart
  • Registration confirmation will be emailed to you.

  • We're committed to your privacy. Geophysical Insights uses the information you provide to us to contact you about our relevant content, events, and products. You may unsubscribe from these communications at any time. For more information, check out our Privacy Policy

    Aldrin RondonSenior Geophysical Engineer - Dragon Oil

    Machine Learning Fault Detection: A Case Study

    An innovative Fault Pattern Detection Methodology has been carried out using a combination of Machine Learning Techniques to produce a seismic volume suitable for fault interpretation in a structurally and stratigraphic complex field. Through theory and results, the main objective was to demonstrate that a combination of ML tools can generate superior results in comparison with traditional attribute extraction and data manipulation through conventional algorithms. The ML technologies applied are a supervised, deep learning, fault classification followed by an unsupervised, multi-attribute classification combining fault probability and instantaneous attributes.

    Aldrin RondonSenior Geophysical Engineer - Dragon Oil

    Bachelor’s Degree in Geophysical Engineering from Central University in Venezuela with a specialization in Reservoir Characterization from Simon Bolivar University.

    Over 20 years exploration and development geophysical experience with extensive 2D and 3D seismic interpretation including acquisition and processing.

    Aldrin spent his formative years working on exploration activity in PDVSA Venezuela followed by a period working for a major international consultant company in the Gulf of Mexico (Landmark, Halliburton) as a G&G consultant. Latterly he was working at Helix in Scotland, UK on producing assets in the Central and South North Sea.  From 2007 to 2021, he has been working as a Senior Seismic Interpreter in Dubai involved in different dedicated development projects in the Caspian Sea.

    Bob A. Hardage

    Investigating the Internal Fabric of VSP data with Attribute Analysis and Unsupervised Machine Learning

    Examination of vertical seismic profile (VSP) data with unsupervised machine learning technology is a rigorous way to compare the fabric of down-going, illuminating, P and S wavefields with the fabric of up-going reflections and interbed multiples created by these wavefields. This concept is introduced in this paper by applying unsupervised learning to VSP data to better understand the physics of P and S reflection seismology. The zero-offset VSP data used in this investigation were acquired in a hard-rock, fast-velocity, environment that caused the shallowest 2 or 3 geophones to be inside the near-field radiation zone of a vertical-vibrator baseplate. This study shows how to use instantaneous attributes to backtrack down-going direct-P and direct-S illuminating wavelets to the vibrator baseplate inside the near-field zone. This backtracking confirms that the points-of-origin of direct-P and direct-S are identical. The investigation then applies principal component (PCA) analysis to VSP data and shows that direct-S and direct-P wavefields that are created simultaneously at a vertical-vibrator baseplate have the same dominant principal components. A self-organizing map (SOM) approach is then taken to illustrate how unsupervised machine learning describes the fabric of down-going and up-going events embedded in vertical-geophone VSP data. These SOM results show that a small number of specific neurons build the down-going direct-P illuminating wavefield, and another small group of neurons build up-going P primary reflections and early-arriving down-going P multiples. The internal attribute fabric of these key down-going and up-going neurons are then compared to expose their similarities and differences. This initial study indicates that unsupervised machine learning, when applied to VSP data, is a powerful tool for understanding the physics of seismic reflectivity at a prospect. This research strategy of analyzing VSP data with unsupervised machine learning will now expand to horizontal-geophone VSP data.

    Bob A. Hardage

    Bob A. Hardage received a PhD in physics from Oklahoma State University. His thesis work focused on high-velocity micro-meteoroid impact on space vehicles, which required trips to Goddard Space Flight Center to do finite-difference modeling on dedicated computers. Upon completing his university studies, he worked at Phillips Petroleum Company for 23 years and was Exploration Manager for Asia and Latin America when he left Phillips. He moved to WesternAtlas and worked 3 years as Vice President of Geophysical Development and Marketing. He then established a multicomponent seismic research laboratory at the Bureau of Economic Geology and served The University of Texas at Austin as a Senior Research Scientist for 28 years. He has published books on VSP, cross-well profiling, seismic stratigraphy, and multicomponent seismic technology. He was the first person to serve 6 years on the Board of Directors of the Society of Exploration Geophysicists (SEG). His Board service was as SEG Editor (2 years), followed by 1-year terms as First VP, President Elect, President, and Past President. SEG has awarded him a Special Commendation, Life Membership, and Honorary Membership. He wrote the AAPG Explorer column on geophysics for 6 years. AAPG honored him with a Distinguished Service award for promoting geophysics among the geological community.

    Carrie LaudonSenior Geophysical Consultant - Geophysical Insights

    Applying Machine Learning Technologies in the Niobrara Formation, DJ Basin, to Quickly Produce an Integrated Structural and Stratigraphic Seismic Classification Volume Calibrated to Wells

    This study will demonstrate an automated machine learning approach for fault detection in a 3D seismic volume. The result combines Deep Learning Convolution Neural Networks (CNN) with a conventional data pre-processing step and an image processing-based post processing approach to produce high quality fault attribute volumes of fault probability, fault dip magnitude and fault dip azimuth. These volumes are then combined with instantaneous attributes in an unsupervised machine learning classification, allowing the isolation of both structural and stratigraphic features into a single 3D volume. The workflow is illustrated on a 3D seismic volume from the Denver Julesburg Basin and a statistical analysis is used to calibrate results to well data.

    Dr. Carrie LaudonSenior Geophysical Consultant

    Applying Unsupervised Multi-Attribute Machine Learning for 3D Stratigraphic Facies Classification in a Carbonate Field, Offshore Brazil

    We present results of a multi-attribute, machine learning study over a pre-salt carbonate field in the Santos Basin, offshore Brazil. These results test the accuracy and potential of Self-organizing maps (SOM) for stratigraphic facies delineation. The study area has an existing detailed geological facies model containing predominantly reef facies in an elongated structure.

    Dr. Carrie LaudonSenior Geophysical Consultant - Geophysical Insights

    Carolan (Carrie) Laudon holds a PhD in geophysics from the University of Minnesota and a BS in geology from the University of Wisconsin Eau Claire. She has been Senior Geophysical Consultant with Geophysical Insights since 2017 working with Paradise®, their machine learning platform. Prior roles include Vice President of Consulting Services and Microseismic Technology for Global Geophysical Services and 17 years with Schlumberger in technical, management and sales, starting in Alaska and including Aberdeen, Scotland, Houston, TX, Denver, CO and Reading, England. She spent five years early in her career with ARCO Alaska as a seismic interpreter for the Central North Slope exploration team.

    Deborah SacreyOwner - Auburn Energy

    How to Use Paradise to Interpret Clastic Reservoirs

    The key to understanding Clastic reservoirs in Paradise starts with good synthetic ties to the wavelet data. If one is not tied correctly, then it will be easy to mis-interpret the neurons as reservoir, whin they are not. Secondly, the workflow should utilize Principal Component Analysis to better understand the zone of interest and the attributes to use in the SOM analysis. An important part to interpretation is understanding “Halo” and “Trailing” neurons as part of the stack around a reservoir or potential reservoir. Deep, high-pressured reservoirs often “leak” or have vertical percolation into the seal. This changes the rock properties enough in the seal to create a “halo” effect in SOM. Likewise, the frequency changes of the seismic can cause a subtle “dim-out”, not necessarily observable in the wavelet data, but enough to create a different pattern in the Earth in terms of these rock property changes. Case histories for Halo and trailing neural information include deep, pressured, Chris R reservoir in Southern Louisiana, Frio pay in Southeast Texas and AVO properties in the Yegua of Wharton County. Additional case histories to highlight interpretation include thin-bed pays in Brazoria County, including updated information using CNN fault skeletonization. Continuing the process of interpretation is showing a case history in Wharton County on using Low Probability to help explore Wilcox reservoirs. Lastly, a look at using Paradise to help find sweet spots in unconventional reservoirs like the Eagle Ford, a case study provided by Patricia Santigrossi.

    Deborah SacreyOwner - Auburn Energy

    How to Use Paradise to Interpret Carbonate Reservoirs

    The key to understanding Carbonate reservoirs in Paradise start with good synthetic ties to the wavelet data. If one is not tied correctly, then it will be very east to mis-interpret the neurons as reservoir, when they are not. Secondly, the workflow should utilize Principal Component Analysis to better understand the zone of interest and the attributes to use in the SOM analysis. An important part to interpretation is understanding “Halo” and “Trailing” neurons as part of the stack around a reservoir or potential reservoir. Usually, one sees this phenomenon around deep, pressured gas reservoirs, but it can happen in shallow reservoirs as well. Two case studies are presented to emphasize the importance of looking for halo or trailing patterns around good reservoirs. One is a deep Edwards example in south central Texas, and the other a shallow oil reservoir in the Austin Chalk in the San Antonio area. Another way to help enhance carbonate reservoirs is through Spectral Decomposition. A case history is shown in the Smackover in Alabama to highlight and focus on an oolitic shoal reservoir which tunes at a specific frequency in the best wells. Not all carbonate porosity is at the top of the deposition. A case history will be discussed looking for porosity in the center portion of a reef in west Texas. And finally, one of the most difficult interpretation challenges in the carbonate spectrum is correctly mapping the interface between two carbonate layers. A simple technique is shown to help with that dilemma, by using few attributes and a low-topology count to understand regional depositional sequences. This example is from the Delaware Basin in southeastern New Mexico.

    Deborah Sacrey
    Owner, Auburn Energy

    Deborah is a geologist/geophysicist with 44 years of oil and gas exploration experience in Texas, Louisiana Gulf Coast and Mid-Continent areas of the US. She received her degree in Geology from the University of Oklahoma in 1976 and immediately started working for Gulf Oil in their Oklahoma City offices.

    She started her own company, Auburn Energy, in 1990 and built her first geophysical workstation using Kingdom software in 1996. She helped SMT/IHS for 18 years in developing and testing the Kingdom Software. She specializes in 2D and 3D interpretation for clients in the US and internationally. For the past nine years she has been part of a team to study and bring the power of multi-attribute neural analysis of seismic data to the geoscience public, guided by Dr. Tom Smith, founder of SMT. She has become an expert in the use of Paradise software and has seven discoveries for clients using multi-attribute neural analysis.

    Deborah has been very active in the geological community. She is past national President of SIPES (Society of Independent Professional Earth Scientists), past President of the Division of Professional Affairs of AAPG (American Association of Petroleum Geologists), Past Treasurer of AAPG and Past President of the Houston Geological Society. She is also Past President of the Gulf Coast Association of Geological Societies and just ended a term as one of the GCAGS representatives on the AAPG Advisory Council. Deborah is also a DPA Certified Petroleum Geologist #4014 and DPA Certified Petroleum Geophysicist #2. She belongs to AAPG, SIPES, Houston Geological Society, South Texas Geological Society and the Oklahoma City Geological Society (OCGS).

    Dr. Tom Smith
    President & CEO

    Dr. Tom Smith received a BS and MS degree in Geology from Iowa State University. His graduate research focused on a shallow refraction investigation of the Manson astrobleme. In 1971, he joined Chevron Geophysical as a processing geophysicist but resigned in 1980 to complete his doctoral studies in 3D modeling and migration at the Seismic Acoustics Lab at the University of Houston. Upon graduation with the Ph.D. in Geophysics in 1981, he started a geophysical consulting practice and taught seminars in seismic interpretation, seismic acquisition and seismic processing. Dr. Smith founded Seismic Micro-Technology in 1984 to develop PC software to support training workshops which subsequently led to development of the KINGDOM Software Suite for integrated geoscience interpretation with world-wide success.

    The Society of Exploration Geologists (SEG) recognized Dr. Smith’s work with the SEG Enterprise Award in 2000, and in 2010, the Geophysical Society of Houston (GSH) awarded him an Honorary Membership. Iowa State University (ISU) has recognized Dr. Smith throughout his career with the Distinguished Alumnus Lecturer Award in 1996, the Citation of Merit for National and International Recognition in 2002, and the highest alumni honor in 2015, the Distinguished Alumni Award. The University of Houston College of Natural Sciences and Mathematics recognized Dr. Smith with the 2017 Distinguished Alumni Award.

    In 2009, Dr. Smith founded Geophysical Insights, where he leads a team of geophysicists, geologists and computer scientists in developing advanced technologies for fundamental geophysical problems. The company launched the Paradise® multi-attribute analysis software in 2013, which uses Machine Learning and pattern recognition to extract greater information from seismic data.

    Dr. Smith has been a member of the SEG since 1967 and is a professional member of SEG, GSH, HGS, EAGE, SIPES, AAPG, Sigma XI, SSA and AGU. Dr. Smith served as Chairman of the SEG Foundation from 2010 to 2013. On January 25, 2016, he was recognized by the Houston Geological Society (HGS) as a geophysicist who has made significant contributions to the field of geology. He currently serves on the SEG President-Elect’s Strategy and Planning Committee and the ISU Foundation Campaign Committee for Forever True, For Iowa State.

    Fabian Rada
    Sr. Geophysicist, Petroleum Oil & Gas Services

    Statistical Calibration of SOM results with Well Log Data (Case Study)

    The first stage of the proposed statistical method has proven to be very useful in testing whether or not there is a relationship between two qualitative variables (nominal or ordinal) or categorical quantitative variables, in the fields of health and social sciences. Its application in the oil industry allows geoscientists not only to test dependence between discrete variables, but to measure their degree of correlation (weak, moderate or strong). This article shows its application to reveal the relationship between a SOM classification volume of a set of nine seismic attributes (whose vertical sampling interval is three meters) and different well data (sedimentary facies, Net Reservoir, and effective porosity grouped by ranges). The data were prepared to construct the contingency tables, where the dependent (response) variable and independent (explanatory) variable were defined, the observed frequencies were obtained, and the frequencies that would be expected if the variables were independent were calculated and then the difference between the two magnitudes was studied using the contrast statistic called Chi-Square. The second stage implies the calibration of the SOM volume extracted along the wellbore path through statistical analysis of the petrophysical properties VCL and PHIE, and SW for each neuron, which allowed to identify the neurons with the best petrophysical values in a carbonate reservoir.

    Fabian Rada
    Sr. Geophysicist, Petroleum Oil & Gas Servicest

    Fabian Rada joined Petroleum Oil and Gas Services, Inc (POGS) in January 2015 as Business Development Manager and Consultant to PEMEX. In Mexico, he has participated in several integrated oil and gas reservoir studies. He has consulted with PEMEX Activos and the G&G Technology group to apply the Paradise AI workbench and other tools. Since January 2015, he has been working with Geophysical Insights staff to provide and implement the multi-attribute analysis software Paradise in Petróleos Mexicanos (PEMEX), running a successful pilot test in Litoral Tabasco Tsimin Xux Asset. Mr. Rada began his career in the Venezuelan National Foundation for Seismological Research, where he participated in several geophysical projects, including seismic and gravity data for micro zonation surveys. He then joined China National Petroleum Corporation (CNPC) as QC Geophysicist until he became the Chief Geophysicist in the QA/QC Department. Then, he transitioned to a subsidiary of Petróleos de Venezuela (PDVSA), as a member of the QA/QC and Chief of Potential Field Methods section. Mr. Rada has also participated in processing land seismic data and marine seismic/gravity acquisition surveys. Mr. Rada earned a B.S. in Geophysics from the Central University of Venezuela.

    Hal GreenDirector, Marketing & Business Development - Geophysical Insights

    Introduction to Automatic Fault Detection and Applying Machine Learning to Detect Thin Beds

    Rapid advances in Machine Learning (ML) are transforming seismic analysis. Using these new tools, geoscientists can accomplish the following quickly and effectively: a combination of machine learning (ML) and deep learning applications, geoscientists apply Paradise to extract greater insights from seismic and well data for these and other objectives:

    • Run fault detection analysis in a few hours, not weeks
    • Identify thin beds down to a single seismic sample
    • Overlay fault images on stratigraphic analysis

    The brief introduction will orient you with the technology and examples of how machine learning is being applied to automate interpretation while generating new insights in the data.

    Hal GreenDirector, Marketing & Business Development - Geophysical Insights

    Hal H. Green is a marketing executive and entrepreneur in the energy industry with more than 25 years of experience in starting and managing technology companies. He holds a B.S. in Electrical Engineering from Texas A&M University and an MBA from the University of Houston. He has invested his career at the intersection of marketing and technology, with a focus on business strategy, marketing, and effective selling practices. Mr. Green has a diverse portfolio of experience in marketing technology to the hydrocarbon supply chain – from upstream exploration through downstream refining & petrochemical. Throughout his career, Mr. Green has been a proven thought-leader and entrepreneur, while supporting several tech start-ups.

    He started his career as a process engineer in the semiconductor manufacturing industry in Dallas, Texas and later launched an engineering consulting and systems integration business. Following the sale of that business in the late 80’s, he joined Setpoint in Houston, Texas where he eventually led that company’s Manufacturing Systems business. Aspen Technology acquired Setpoint in January 1996 and Mr. Green continued as Director of Business Development for the Information Management and Polymer Business Units.

    In 2004, Mr. Green founded Advertas, a full-service marketing and public relations firm serving clients in energy and technology. In 2010, Geophysical Insights retained Advertas as their marketing firm. Dr. Tom Smith, President/CEO of Geophysical Insights, soon appointed Mr. Green as Director of Marketing and Business Development for Geophysical Insights, in which capacity he still serves today.

    Hana Kabazi
    Product Manager

    Hana Kabazi joined Geophysical Insights in October of 201, and is now one of our Product Managers for Paradise. Mrs. Kabazi has over 7 years of oil and gas experience, including 5 years and Halliburton – Landmark. During her time at Landmark she held positions as a consultant to many E&P companies, technical advisor to the QA organization, and as product manager of Subsurface Mapping in DecsionSpace. Mrs. Kabazi has a B.S. in Geology from the University of Texas Austin, and an M.S. in Geology from the University of Houston.

    Heather Bedle
    Assistant Professor, University of Oklahoma

    Gas Hydrates, Reefs, Channel Architecture, and Fizz Gas: SOM Applications in a Variety of Geologic Settings

    Students at the University of Oklahoma have been exploring the uses of SOM techniques for the last year. This presentation will review learnings and results from a few of these research projects. Two projects have investigated the ability of SOMs to aid in identification of pore space materials – both trying to qualitatively identify gas hydrates and under-saturated gas reservoirs. A third study investigated individual attributes and SOMs in recognizing various carbonate facies in a pinnacle reef in the Michigan Basin. The fourth study took a deep dive of various machine learning algorithms, of which SOMs will be discussed, to understand how much machine learning can aid in the identification of deepwater channel architectures.

    Heather Bedle
    Assistant Professor, University of Oklahoma

    Heather Bedle received a B.S. (1999) in physics from Wake Forest University, and then worked as a systems engineer in the defense industry. She later received a M.S. (2005) and a Ph. D. (2008) degree from Northwestern University. After graduate school, she joined Chevron and worked as both a development geologist and geophysicist in the Gulf of Mexico before joining Chevron’s Energy Technology Company Unit in Houston, TX. In this position, she worked with the Rock Physics from Seismic team analyzing global assets in Chevron’s portfolio. Dr. Bedle is currently an assistant professor of applied geophysics at the University of Oklahoma’s School of Geosciences. She joined OU in 2018, after instructing at the University of Houston for two years. Dr. Bedle and her student research team at OU primarily work with seismic reflection data, using advanced techniques such as machine learning, attribute analysis, and rock physics to reveal additional structural, stratigraphic and tectonic insights of the subsurface.

    Ivan Marroquin
    Senior Research Geophysicist

    Connecting Multi-attribute Classification to Reservoir Properties

    Interpreters rely on seismic pattern changes to identify and map geologic features of importance. The ability to recognize such features depends on the seismic resolution and characteristics of seismic waveforms. With the advancement of machine learning algorithms, new methods for interpreting seismic data are being developed. Among these algorithms, self-organizing maps (SOM) provides a different approach to extract geological information from a set of seismic attributes.

    SOM approximates the input patterns by a finite set of processing neurons arranged in a regular 2D grid of map nodes. Such that, it classifies multi-attribute seismic samples into natural clusters following an unsupervised approach. Since machine learning is unbiased, so the classifications can contain both geological information and coherent noise. Thus, seismic interpretation evolves into broader geologic perspectives. Additionally, SOM partitions multi-attribute samples without a priori information to guide the process (e.g., well data).

    The SOM output is a new seismic attribute volume, in which geologic information is captured from the classification into winning neurons. Implicit and useful geological information are uncovered through an interactive visual inspection of winning neuron classifications. By doing so, interpreters build a classification model that aids them to gain insight into complex relationships between attribute patterns and geological features.

    Despite all these benefits, there are interpretation challenges regarding whether there is an association between winning neurons and geological features. To address these issues, a bivariate statistical approach is proposed. To evaluate this analysis, three cases scenarios are presented. In each case, the association between winning neurons and net reservoir (determined from petrophysical or well log properties) at well locations is analyzed. The results show that the statistical analysis not only aid in the identification of classification patterns; but more importantly, reservoir/not reservoir classification by classical petrophysical analysis strongly correlates with selected SOM winning neurons. Confidence in interpreted classification features is gained at the borehole and interpretation is readily extended as geobodies away from the well.

    Ivan Marroquin
    Senior Research Geophysicist

    Iván Dimitri Marroquín is a 20-year veteran of data science research, consistently publishing in peer-reviewed journals and speaking at international conference meetings. Dr. Marroquín received a Ph.D. in geophysics from McGill University, where he conducted and participated in 3D seismic research projects. These projects focused on the development of interpretation techniques based on seismic attributes and seismic trace shape information to identify significant geological features or reservoir physical properties. Examples of his research work are attribute-based modeling to predict coalbed thickness and permeability zones, combining spectral analysis with coherency imagery technique to enhance interpretation of subtle geologic features, and implementing a visual-based data mining technique on clustering to match seismic trace shape variability to changes in reservoir properties.

    Dr. Marroquín has also conducted some ground-breaking research on seismic facies classification and volume visualization. This lead to his development of a visual-based framework that determines the optimal number of seismic facies to best reveal meaningful geologic trends in the seismic data. He proposed seismic facies classification as an alternative to data integration analysis to capture geologic information in the form of seismic facies groups. He has investigated the usefulness of mobile devices to locate, isolate, and understand the spatial relationships of important geologic features in a context-rich 3D environment. In this work, he demonstrated mobile devices are capable of performing seismic volume visualization, facilitating the interpretation of imaged geologic features.  He has definitively shown that mobile devices eventually will allow the visual examination of seismic data anywhere and at any time.

    In 2016, Dr. Marroquín joined Geophysical Insights as a senior researcher, where his efforts have been focused on developing machine learning solutions for the oil and gas industry. For his first project, he developed a novel procedure for lithofacies classification that combines a neural network with automated machine methods. In parallel, he implemented a machine learning pipeline to derive cluster centers from a trained neural network. The next step in the project is to correlate lithofacies classification to the outcome of seismic facies analysis.  Other research interests include the application of diverse machine learning technologies for analyzing and discerning trends and patterns in data related to oil and gas industry.

    Dr. Jie Qi
    Research Geophysicist

    An integrated machine learning-based fault classification workflow

    We introduce an integrated machine learning-based fault classification workflow that creates fault component classification volumes that greatly reduces the burden on the human interpreter. We first compute a 3D fault probability volume from pre-conditioned seismic amplitude data using a 3D convolutional neural network (CNN). However, the resulting “fault probability” volume delineates other non-fault edges such as angular unconformities, the base of mass transport complexes, and noise such as acquisition footprint. We find that image processing-based fault discontinuity enhancement and skeletonization methods can enhance the fault discontinuities and suppress many of the non-fault discontinuities. Although each fault is characterized by its dip and azimuth, these two properties are discontinuous at azimuths of φ=±180° and for near vertical faults for azimuths φ and φ+180° requiring them to be parameterized as four continuous geodetic fault components. These four fault components as well as the fault probability can then be fed into a self-organizing map (SOM) to generate fault component classification. We find that the final classification result can segment fault sets trending in interpreter-defined orientations and minimize the impact of stratigraphy and noise by selecting different neurons from the SOM 2D neuron color map.

    Dr. Jie Qi
    Research Geophysicist

    Dr. Jie Qi is a Research Geophysicist at Geophysical Insights, where he works closely with product development and geoscience consultants. His research interests include machine learning-based fault detection, seismic interpretation, pattern recognition, image processing, seismic attribute development and interpretation, and seismic facies analysis. Dr. Qi received a BS (2011) in Geoscience from the China University of Petroleum in Beijing, and an MS (2013) in Geophysics from the University of Houston. He earned a Ph.D. (2017) in Geophysics from the University of Oklahoma, Norman. His industry experience includes work as a Research Assistant (2011-2013) at the University of Houston and the University of Oklahoma (2013-2017). Dr. Qi was with Petroleum Geo-Services (PGS), Inc. in 2014 as a summer intern, where he worked on a semi-supervised seismic facies analysis. In 2017, he served as a postdoctoral Research Associate in the Attributed Assisted-Seismic Processing and Interpretation (AASPI) consortium at the University of Oklahoma from 2017 to 2020.

    Jie Qi
    Research Geophysicist

    An Integrated Fault Detection Workflow

    Seismic fault detection is one of the top critical procedures in seismic interpretation. Identifying faults are significant for characterizing and finding the potential oil and gas reservoirs. Seismic amplitude data exhibiting good resolution and a high signal-to-noise ratio are key to identifying structural discontinuities using seismic attributes or machine learning techniques, which in turn serve as input for automatic fault extraction. Deep learning Convolutional Neural Networks (CNN) performs well on fault detection without any human-computer interactive work. This study shows an integrated CNN-based fault detection workflow to construct fault images that are sufficiently smooth for subsequent fault automatic extraction. The objectives were to suppress noise or stratigraphic anomalies subparallel to reflector dip, and sharpen fault and other discontinuities that cut reflectors, preconditioning the fault images for subsequent automatic extraction. A 2D continuous wavelet transform-based acquisition footprint suppression method was applied time slice by time slice to suppress wavenumber components to avoid interpreting the acquisition footprint as artifacts by the CNN fault detection method. To further suppress cross-cutting noise as well as sharpen fault edges, a principal component edge-preserving structure-oriented filter is also applied. The conditioned amplitude volume is then fed to a pre-trained CNN model to compute fault probability. Finally, a Laplacian of Gaussian filter is applied to the original CNN fault probability to enhance fault images. The resulting fault probability volume is favorable with respect to traditional human-interpreter generated on vertical slices through the seismic amplitude volume.

    Laura Cuttill
    Practice Lead, Advertas

    Young Professionals – Managing Your Personal Brand to Level-up Your Career

    No matter where you are in your career, your online “personal brand” has a huge impact on providing opportunity for prospective jobs and garnering the respect and visibility needed for advancement. While geoscientists tackle ambitious projects, publish in technical papers, and work hard to advance their careers, often, the value of these isn’t realized beyond their immediate professional circle. Learn how to…

    • - Communicate who you are to high-level executives in exploration and development
    • - Avoid common social media pitfalls
    • - Optimize your online presence to best garner attention from recruiters
    • - Stay relevant
    • - Create content of interest
    • - Establish yourself as a thought leader in your given area of specialization
    Laura Cuttill
    Practice Lead, Advertas

    As a 20-year marketing veteran marketing in oil and gas and serial entrepreneur, Laura has deep experience in bringing technology products to market and growing sales pipeline. Armed with a marketing degree from Texas A&M, she began her career doing technical writing for Schlumberger and ExxonMobil in 2001. She started Advertas as a co-founder in 2004 and began to leverage her upstream experience in marketing. In 2006, she co-founded the cyber-security software company, 2FA Technology. After growing 2FA from a startup to 75% market share in target industries, and the subsequent sale of the company, she returned to Advertas to continue working toward the success of her clients, such as Geophysical Insights. Today, she guides strategy for large-scale marketing programs, manages project execution, cultivates relationships with industry media, and advocates for data-driven, account-based marketing practices.

    Carrie LaudonSenior Geophysical Consultant - Geophysical Insights

    Automatic Fault Detection and Applying Machine Learning to Detect Thin Beds

    Rapid advances in Machine Learning (ML) are transforming seismic analysis. Using these new tools, geoscientists can accomplish the following quickly and effectively:

    • Run fault detection analysis in a few hours, not weeks
    • Identify thin beds down to a single seismic sample
    • Generate seismic volumes that capture structural and stratigraphic details

    Join us for a ‘Lunch & Learn’ sessions daily at 11:00 where Dr. Carolan (“Carrie”) Laudon will review the theory and results of applying a combination of machine learning tools to obtain the above results.  A detailed agenda follows.

    Agenda

    Automated Fault Detection using 3D CNN Deep Learning

    • Deep learning fault detection
    • Synthetic models
    • Fault image enhancement
    • Semi-supervised learning for visualization
    • Application results
      • Normal faults
      • Fault/fracture trends in complex reservoirs

    Demo of Paradise Fault Detection Thoughtflow®

    Stratigraphic analysis using machine learning with fault detection

    • Attribute Selection using Principal Component Analysis (PCA)
    • Multi-Attribute Classification using Self-Organizing Maps (SOM)
    • Case studies – stratigraphic analysis and fault detection
      • Fault-karst and fracture examples, China
      • Niobrara – Stratigraphic analysis and thin beds, faults
    Mike DunnSr. Vice President of Business Development

    Machine Learning in the Cloud

    Machine Learning in the Cloud will address the capabilities of the Paradise AI Workbench, featuring on-demand access enabled by the flexible hardware and storage facilities available on Amazon Web Services (AWS) and other commercial cloud services. Like the on-premise instance, Paradise On-Demand provides guided workflows to address many geologic challenges and investigations. The presentation will show how geoscientists can accomplish the following workflows quickly and effectively using guided ThoughtFlows® in Paradise:
    • Identify and calibrate detailed stratigraphy using seismic and well logs
    • Classify seismic facies
    • Detect faults automatically
    • Distinguish thin beds below conventional tuning
    • Interpret Direct Hydrocarbon Indicators
    • Estimate reserves/resources
    Attend the talk to see how ML applications are combined through a process called "Machine Learning Orchestration," proven to extract more from seismic and well data than traditional means.
    Mike Dunn
    Senior Vice President Business Development

    Michael A. Dunn is an exploration executive with extensive global experience including the Gulf of Mexico, Central America, Australia, China and North Africa. Mr. Dunn has a proven a track record of successfully executing exploration strategies built on a foundation of new and innovative technologies. Currently, Michael serves as Senior Vice President of Business Development for Geophysical Insights.

    He joined Shell in 1979 as an exploration geophysicist and party chief and held increasing levels or responsibility including Manager of Interpretation Research. In 1997, he participated in the launch of Geokinetics, which completed an IPO on the AMEX in 2007. His extensive experience with oil companies (Shell and Woodside) and the service sector (Geokinetics and Halliburton) has provided him with a unique perspective on technology and applications in oil and gas. Michael received a B.S. in Geology from Rutgers University and an M.S. in Geophysics from the University of Chicago.

    Rocky R. Roden
    Senior Consulting Geophysicist

    The Relationship of Self-Organization, Geology, and Machine Learning

    Self-organization is the nonlinear formation of spatial and temporal structures, patterns or functions in complex systems (Aschwanden et al., 2018). Simple examples of self-organization include flocks of birds, schools of fish, crystal development, formation of snowflakes, and fractals. What these examples have in common is the appearance of structure or patterns without centralized control. Self-organizing systems are typically governed by power laws, such as the Gutenberg-Richter law of earthquake frequency and magnitude. In addition, the time frames of such systems display a characteristic self-similar (fractal) response, where earthquakes or avalanches for example, occur over all possible time scales (Baas, 2002).

    The existence of nonlinear dynamic systems and ordered structures in the earth are well known and have been studied for centuries and can appear as sedimentary features, layered and folded structures, stratigraphic formations, diapirs, eolian dune systems, channelized fluvial and deltaic systems, and many more (Budd, et al., 2014; Dietrich and Jacob, 2018). Each of these geologic processes and features exhibit patterns through the action of undirected local dynamics and is generally termed “self-organization” (Paola, 2014).

    Artificial intelligence and specifically neural networks exhibit and reveal self-organization characteristics. The reason for the interest in applying neural networks stems from the fact that they are universal approximators for various kinds of nonlinear dynamical systems of arbitrary complexity (Pessa, 2008). A special class of artificial neural networks is aptly named self-organizing map (SOM) (Kohonen, 1982). It has been found that SOM can identify significant organizational structure in the form of clusters from seismic attributes that relate to geologic features (Strecker and Uden, 2002; Coleou et al., 2003; de Matos, 2006; Roy et al., 2013; Roden et al., 2015; Zhao et al., 2016; Roden et al., 2017; Zhao et al., 2017; Roden and Chen, 2017; Sacrey and Roden, 2018; Leal et al, 2019; Hussein et al., 2020; Hardage et al., 2020; Manauchehri et al., 2020). As a consequence, SOM is an excellent machine learning neural network approach utilizing seismic attributes to help identify self-organization features and define natural geologic patterns not easily seen or seen at all in the data.

    Rocky R. Roden
    Senior Consulting Geophysicist

    Rocky R. Roden started his own consulting company, Rocky Ridge Resources Inc. in 2003 and works with several oil companies on technical and prospect evaluation issues. He is also a principal in the Rose and Associates DHI Risk Analysis Consortium and was Chief Consulting Geophysicist with Seismic Micro-technology. Rocky is a proven oil finder with 37 years in the industry, gaining extensive knowledge of modern geoscience technical approaches.

    Rocky holds a BS in Oceanographic Technology-Geology from Lamar University and a MS in Geological and Geophysical Oceanography from Texas A&M University. As Chief Geophysicist and Director of Applied Technology for Repsol-YPF, his role comprised of advising corporate officers, geoscientists, and managers on interpretation, strategy and technical analysis for exploration and development in offices in the U.S., Argentina, Spain, Egypt, Bolivia, Ecuador, Peru, Brazil, Venezuela, Malaysia, and Indonesia. He has been involved in the technical and economic evaluation of Gulf of Mexico lease sales, farmouts worldwide, and bid rounds in South America, Europe, and the Far East. Previous work experience includes exploration and development at Maxus Energy, Pogo Producing, Decca Survey, and Texaco. Rocky is a member of SEG, AAPG, HGS, GSH, EAGE, and SIPES; he is also a past Chairman of The Leading Edge Editorial Board.

    Sarah Stanley
    Senior Geoscientist

    Stratton Field Case Study – New Solutions to Old Problems

    The Oligocene Frio gas-producing Stratton Field in south Texas is a well-known field. Like many onshore fields, the productive sand channels are difficult to identify using conventional seismic data. However, the productive channels can be easily defined by employing several Paradise modules, including unsupervised machine learning, Principal Component Analysis, Self-Organizing Maps, 3D visualization, and the new Well Log Cross Section and Well Log Crossplot tools. The Well Log Cross Section tool generates extracted seismic data, including SOMs, along the Cross Section boreholes and logs. This extraction process enables the interpreter to accurately identify the SOM neurons associated with pay versus neurons associated with non-pay intervals. The reservoir neurons can be visualized throughout the field in the Paradise 3D Viewer, with Geobodies generated from the neurons. With this ThoughtFlow®, pay intervals previously difficult to see in conventional seismic can finally be visualized and tied back to the well data.

    Sarah Stanley
    Senior Geoscientist and Lead Trainer

    Sarah Stanley joined Geophysical Insights in October, 2017 as a geoscience consultant, and became a full-time employee July 2018. Prior to Geophysical Insights, Sarah was employed by IHS Markit in various leadership positions from 2011 to her retirement in August 2017, including Director US Operations Training and Certification, the Operational Governance Team, and, prior to February 2013, Director of IHS Kingdom Training. Sarah joined SMT in May, 2002, and was the Director of Training for SMT until IHS Markit’s acquisition in 2011.

    Prior to joining SMT Sarah was employed by GeoQuest, a subdivision of Schlumberger, from 1998 to 2002. Sarah was also Director of the Geoscience Technology Training Center, North Harris College from 1995 to 1998, and served as a voluntary advisor on geoscience training centers to various geological societies. Sarah has over 37 years of industry experience and has worked as a petroleum geoscientist in various domestic and international plays since August of 1981. Her interpretation experience includes tight gas sands, coalbed methane, international exploration, and unconventional resources.

    Sarah holds a Bachelor’s of Science degree with majors in Biology and General Science and minor in Earth Science, a Master’s of Arts in Education and Master’s of Science in Geology from Ball State University, Muncie, Indiana. Sarah is both a Certified Petroleum Geologist, and a Registered Geologist with the State of Texas. Sarah holds teaching credentials in both Indiana and Texas.

    Sarah is a member of the Houston Geological Society and the American Association of Petroleum Geologists, where she currently serves in the AAPG House of Delegates. Sarah is a recipient of the AAPG Special Award, the AAPG House of Delegates Long Service Award, and the HGS President’s award for her work in advancing training for petroleum geoscientists. She has served on the AAPG Continuing Education Committee and was Chairman of the AAPG Technical Training Center Committee. Sarah has also served as Secretary of the HGS, and Served two years as Editor for the AAPG Division of Professional Affairs Correlator.

    Thomas ChaparroSenior Geophysicist - Geophysical Insights

    Paradise: A Day in The Life of the Geoscientist

    Over the last several years, the industry has invested heavily in Machine Learning (ML) for better predictions and automation. Dramatic results have been realized in exploration, field development, and production optimization. However, many of these applications have been single use ‘point’ solutions. There is a growing body of evidence that seismic analysis is best served using a combination of ML tools for a specific objective, referred to as ML Orchestration. This talk demonstrates how the Paradise AI workbench applications are used in an integrated workflow to achieve superior results than traditional interpretation methods or single-purpose ML products. Using examples from combining ML-based Fault Detection and Stratigraphic Analysis, the talk will show how ML orchestration produces value for exploration and field development by the interpreter leveraging ML orchestration.

    Thomas ChaparroSenior Geophysicist - Geophysical Insights

    Thomas Chaparro is a Senior Geophysicist who specializes in training and preparing AI-based workflows. Thomas also has experience as a processing geophysicist and 2D and 3D seismic data processing. He has participated in projects in the Gulf of Mexico, offshore Africa, the North Sea, Australia, Alaska, and Brazil.

    Thomas holds a bachelor’s degree in Geology from Northern Arizona University and a Master’s in Geophysics from the University of California, San Diego. His research focus was computational geophysics and seismic anisotropy.

    Tom Smith
    President and CEO, Geophysical Insights

    Machine Learning for Incomplete Geoscientists

    This presentation covers big-picture machine learning buzz words with humor and unassailable frankness. The goal of the material is for every geoscientist to gain confidence in these important concepts and how they add to our well-established practices, particularly seismic interpretation. Presentation topics include a machine learning historical perspective, what makes it different, a fish factory, Shazam, comparison of supervised and unsupervised machine learning methods with examples, tuning thickness, deep learning, hard/soft attribute spaces, multi-attribute samples, and several interpretation examples. After the presentation, you may not know how to run machine learning algorithms, but you should be able to appreciate their value and avoid some of their limitations.