Deep Learning for Characterizing Paleokarst Collapse Features in 3-D Seismic Images

Xinming Wu, Shangsheng Yan, Jie, Qi, and Hongliu Zeng | Journal of Geophysical Research: Solid Earth | September 2020


Paleokarst systems are found extensively in carbonate-prone basins worldwide. They can form large reservoirs and provide efficient pathways for hydrocarbon migration, but they can also create serious engineering geohazards. The full delineation of potentially buried paleokarst systems plays an important role for reservoir characterization, oil and gas production, and other engineering tasks. We propose a supervised convolutional neural network(CNN) to automatically and accurately characterize paleokarst and associated collapse features from 3-D seismic images. To avoid time-consuming manual labeling for training the CNN, we propose an efficient workflow to automatically generate numerous 3-Dtraining image pairs including synthetic seismic images and the corresponding label images of the collapsed paleokarst features simulated in the seismic images. With this workflow, we are able to simulate realistic and diverse geologic structures and collapsed paleokarst features in the training images from which the CNN can effectively learn to recognize the collapsed paleokarst features in real field seismic images. Two field examples from the Fort Worth Basin demonstrate that our CNN-based method is superior to conventional automatic methods in delineating paleokarst collapse features from seismic images. From the CNN-based paleokarst characterization, we can further automatically extract 3-Dcollapsed paleokarst systems and quantitatively measure their geometric parameters. Our CNN-based method is highly efficient and takes only seconds to classify collapsed paleokarst features a 3-D seismic image with 320×1, 024×1, 024 samples (approximately 268 km2) by using one graphics processing unit.

1. Introduction

Karst is a typical type of carbonate terrain that has gone through significant diagenetic processes, which often generate associated structural features of joints, caves, faults, and collapses (Qi et al., 2014). Buried, collapsed paleokarst systems can form carbonate petroleum reservoirs including those explored in the Fort Worth Basin (FWB) (e.g., Dou et al., 2011; Kerans, 1988; Loucks & Anderson, 1985), the Lower Cretaceous Golden Lane field in eastern Mexico (Coogan et al., 1972; Viniegra & Castillo-Tejero, 1970), the Tarim Basinin China (e.g., Desheng et al., 1996; Maoshan et al., 2011; Qi et al., 2014; Zeng et al., 2011), and other paleokarst-related fields reported in previous works (e.g., Andre & Doulcet, 1970; Loucks, 1999; Mazzullo & Mazzullo, 1970; Zhai & Zha, 1982). In addition, collapsed paleokarst systems pose potential drilling hazards due to the weakness of the paleokarst zones (Qi et al., 2014; Soriano et al., 2019; Zhao et al., 2018). Therefore, delineating subsurface collapsed paleokarst systems is important for petroleum reservoir characterization and production.

Three-dimensional seismic images are widely and commonly used to map paleokarst systems and to study paleokarst related features. To facilitate the paleokarst interpretation in 3-D seismic images, some seismic attributes including coherence (Bahorich & Farmer, 1995; Li & Lu, 2014; Marfurt et al., 1999), structural curvature (Al-Dossary & Marfurt, 2006; Di & Gao, 2016; Roberts, 2001), reflector rotation (Marfurt &Rich, 2010), and spectral decomposition (Chen, 2016; Qi & Castagna, 2013) are calculated to highlight the paleokarst features in 3-D seismic images. Automatic paleokarst interpretation based on such attributes alone, however, remains challenging because these attributes are typically sensitive to noise and other types of structural discontinuities or variations that are unrelated to the collapsed paleokarst features. Significant human interactions are still required to delineate the collapsed paleokarst systems by integrating multiple types of seismic attributes (Khatiwada et al., 2013; Qi et al., 2014; Sullivan et al., 2006; Zhao et al., 2018)

We propose a convolutional neural network (CNN) to fully detect paleokarst collapse features in 3-D seismic images and extract 3-D paleokarst chimney tubes all at once, which would provide a clear 3-D perspective of he collapsed paleokarst systems in the subsurface. CNN-based methods are powerful for multidimensional image processing tasks including image classification (e.g., He et al., 2016; Krizhevsky et al., 2012; Simonyan & Zisserman, 2014; Szegedy et al., 2015), object detection (e.g., Lin et al., 2017; Liu et al., 2016; Redmonet al., 2016; Ren et al., 2015; Sermanet et al., 2013), and image segmentation (e.g., Badrinarayanan et al., 2017; Chen et al., 2017; Long et al., 2015; Ronneberger et al., 2015). Recently, CNNs have been increasingly applied in geoscience problems (Bergen et al., 2019; Bianco et al., 2019) including the interpretation of geologic faults(e.g., Di et al., 2019; Lu et al., 2018; Wu, Liang, et al., 2019; Wu, Shi, et al., 2019; Zhao & Mukhopadhyay, 2018),horizons (e.g., Geng et al., 2019; Wu, Zhang, et al., 2019; Wu et al., 2020), salt bodies (e.g., Di & AlRegib, 2020; Di et al., 2018; Guillen et al., 2015; Shi et al., 2019), and channels (Pham et al., 2019) in seismic images.

In this paper, we consider the characterization of paleokarst collapse features in a 3-D seismic image as an image segmentation problem and solve this problem by using a supervised CNN. The architecture of our CNN is modified from the U-net (Ronneberger et al., 2015) by upgrading the convolutional layers from 2-D to 3-D and reducing the number of layers and features at each layer, which significantly saves GPU memory and makes it applicable to our task of processing 3-D large seismic images. Training a CNN typically requires a large amount of training data sets including input images and the corresponding label images. In our problem of collapsed paleokarst interpretation in seismic images, however, we lack the input seismic images, especially labeled images with interpreted paleokarst features. Manually interpreting or labeling the paleokarst in seismic images is time-consuming, which makes it hard to prepare a large amount of reliable or diverse label images for training a CNN.

To solve this problem of lacking training data sets, we propose a workflow to automatically generate numerous synthetic seismic images, where a variety of realistic structural patterns and collapsed paleokarst features are simulated with well-defined functions. The collapsed paleokarst features in these generated synthetic seismic images are well defined, and we are able to automatically obtain the corresponding label images with the ground-truth of collapsed paleokarst features. With this workflow, we automatically generate 100 pairs of 3-D synthetic seismic images and the corresponding labeled paleokarst images. These100 pairs, with further data augmentation, are proven sufficient to train our CNN for collapsed paleokarst characterization in 3-D seismic images. Although trained only by synthetic seismic images, our CNN shows powerful performance in delineating collapsed paleokarst features in field seismic images. The detected paleokarst results are consistent with previous manual interpretations and are superior to the collapsed paleokarst features characterized by conventional seismic attributes of curvature and coherence. From the images of collapsed paleokarst features computed by the trained CNN, we are able to automatically extract the boundaries of 3-D collapsed paleokarst systems, which may be extensively and complicatedly developed in a 3-D seismic image. Based on the extracted 3-D collapsed paleokarst systems, we can further automatically measure their geometric parameters for quantitative analysis of paleokarst development.

The structure of the paper is organized as follows: We start with discussing two 3-D field seismic surveys in the FWB, the related geological background, and observations of the collapsed paleokarst features in the seismic images. We then present a novel workflow to numerically simulate realistic folding structures and collapsed paleokarst systems in 3-D synthetic seismic images. We further discuss the architecture of the proposed CNN and train it with various synthetic data sets that are automatically generated by using the proposed numerical simulation workflow. We finally demonstrate the applicability of the trained CNN with two 3-D field seismic images and compare the results with those from both manual interpretation by previous researchers and conventional automatic methods of curvature and coherence.

2. Field Seismic Data and Geologic Background

As denoted by the blue and red blocks in Figure 1, the two time-migrated 3-D seismic data sets (Figures 2a and 3a) used in this paper were both acquired at the FWB but are located at different structural settings within the basin. The first seismic data (Figure 2a) was acquired by the Bureau of Economic Geology at the University of Texas at Austin and its industry partners including Arch Petroleum, Enserch, and Oxy USA (Hardage, 1996). As denoted by the blue block in Figure 1a, the corresponding seismic survey covers an area of 67 km2at the Boonsville field in the northern FWB. The Boonsville field (colored cyan in Figure 1a) is one of the largest gas fields in the United States. The second seismic image (Figure 3a) is from the central FWB by the Marathon Oil Company with 16 live receiver lines, forming a 324 km2wide-azimuth survey (red block in Figure 1a) with a fine 16×16 m2Common Depth Point (CDP) bin size (Qi et al., 2014).

General structure map of the Fort Worth Basin
Figure 1. In the general structure map of the Fort Worth Basin (modified after Wen et al., 2016), the red and cyan blocks denote the survey locations of the two 3-D seismic data sets used in this paper. The diagram in (b) shows a stratigraphic section (modified after Alhakeem, 2013) crossing the Fort Worth Basin from north to south as denoted by the magenta line in (a).

Multiple seismic images of the Fort Worth Basin
Figure 2.The 3-D seismic image (a) is from the northern Fort Worth Basin as denoted by the blue block in Figure 1a. In the paleokarst probability map (b), computed by our CNN-based method, the vertically extended paleokarst chimney tubes are highlighted by relatively high probability values (colored by red). The vertical seismic section of Line A (denoted in blue in a and b) is displayed with manual interpretation (c) by McDonnell et al. (2007), the CNN-based paleokarst probability image (d), and the boundaries (cyan curves in e) that are automatically extracted from the probability image.

The FWB is a major petroleum producing geological system which has yielded over 120 billion of gross production since 2001 (George, 2016). The FWB is formed along the Ouachita thrust front as shown in Figure 1a. It is a wedge-shaped (Figure 1a), asymmetrical, and northward deepening (Figure 1b) depression with about 12,000 feet (3,657.6 m) of strata in its deepest northeast area near the Ouachita thrust belt and Muenster Arch (Figure 1a). Previous studies suggested that the FWB was formed during the late Paleozoic Ouachita Orogeny, which is a major tectonic event of thrust-fold deformation associated with the oblique lithospheric convergence of the North American and South American plates (Walper, 1982).

Figure 1b shows a stratigraphic cross section of the FWB from north to south as denoted by the magenta line in Figure 1a. The lower Ellenburger group of carbonate strata is unconformably overlain by the Ordovician Viola and Simpson formations, which pinch out just east of the first seismic survey (Figure 2 and denoted by the blue block in Figure 1a) (McDonnell et al., 2007) and are also missing in the second seis-mic survey (Figure 3 and denoted by the blue block in Figure 1a). Therefore, in the two study areas, the lower Ellenburger group carbonate strata are unconformably overlain by the organic-rich Barnett Formation (as shown in Figure 1b), which recorded deposition of the deep-water foreland basin during the late Mississippian (Sullivan et al., 2006) and plays a critical role in forming multiple gas fields in northern Texas (Pollastro et al., 2007). The Barnett group is overlain by the Pennsylvanian formation, which includes the Bend (lower Pennsylvanian), Strawn, and Canyon groups. As marked in Figure 2c, the Bend group (Morrowan and Atokan stages of the lower Pennsylvanian) consists of Marble Falls, Vineyard, Runawary, Davis, and Caddo stratigraphic units (McDonnell et al., 2007).

3D seismic images from the central Fort Worth Basin
Figure 3. The 3-D seismic image (a) is from the central Fort Worth Basin as denoted by the red block in Figure 1a. The vertical seismic sections in (b) are extracted at Lines AA′ and BB′, which cross a large paleokarst chimney as denoted by the yellow lines in (a). From the 3-D seismic image, a paleokarst probability image (c) is computed by using our CNN-based method to highlight the paleokarst chimneys, which are mostly consistent with the manually interpreted paleokarsts in the vertical sections in (a). The 2-D probability sections (d), extracted at Lines AA′ and BB′, provide a close-up view of our CNN-based detection of the paleokarst chimney, which is consistent with the manual interpretation denoted by the dashed cyan lines.

Paleokarst caves and solution collapses are extensively generated in the Ellenburger group because it went through a series of karst events ranging between the post Ellenburger and Early Pennsylvanian (Canteret al., 1993). In addition, the strata of the Silurian and Devonian strata are missing in the two study areas, which provides additional time for generating paleokarst in the Ellenburger group (Sullivan et al., 2006; Zhao et al., 2018). Figure 4a shows a model (Kerans, 1988) of the paleokarst system and the associated developments in the Ellenburger group. As shown in this model, the collapsed paleokarst system typically contains the three basic components of a paleocave floor, fill, and roof. During burial, paleocaves often collapse and sequentially produce a collapsed zone with some unique associated features as shown in the diagram (Figure 4b) of the coalesced collapsed-paleocave system proposed by Loucks (1999). According to the diagram, a coalesced collapsed paleokarst system typically contains fill deposits (breccias and stratified deposits), a broader damage zone of crackle fractures, circular faults, and chimney or sag structures (Figure 4a) in younger formations. Such collapse chimneys or sags in our seismic images vertically extend upward from the Ellenburger group through the Mississippian (e.g., Barnett Shale) and Pennsylvanian strata (as interpreted by McDonnell et al., 2007 in Figure 2c) over a distance of almost 800 m (Hardage et al., 1996).Therefore, interpreting the collapsed paleokarst-related features, especially the sags with long vertically upward extents, is important for reducing the risks of drilling and hydraulic fracturing in exploring the main oil and gas reservoirs in the Barnett Shale (deposited during the Mississippian period) and Bend group (deposited during the Middle Pennsylvanian period).

A 3-D paleokarst model and a 2-D diagram of a single collapsed paleokarst system
Figure 4. The 3-D paleokarst model (modified after Kerans, 1988) in (a) shows the general features of a paleokarst system where the sag structures or paleokarst chimneys may vertically extend hundreds of meters. The 2-D diagram (b) of a single collapsed paleokarst system (modified after Loucks, 1999) shows features of the associated collapse and extensive brecciation, where deformations (faults, fractures, and sag structures) are developed in postkarst-deposited strata.

The proposed workflow for generating synthetic training data sets.
Figure 5. The proposed workflow for generating synthetic training data sets. In this workflow, we start with a flat reflectivity model (a) in which we then simulate realistic folding structures (b) and collapsed paleokarst features (c). We further convolve the reflectivity model with a Ricker wavelet to compute a 3-D synthetic seismic image (d). As the geometry of the collapsed paleokarst systems is well defined in the simulation, we are able to automatically obtain a binary label image (e) with the ground-truth of the collapsed paleokarst areas labeled with ones. From the labeled binary paleokarst image (e), we can automatically extract the 3-D collapsed paleokarst systems shown in (f).

Three-dimensional seismic imaging provides an effective way to interpret the paleokarst system in the 3-D subsurface. As shown in Figures 2a and 3a, collapsed paleokarst features, especially the large-scale fault and sag structures, can be extensively observed. Manually interpreting such features, however, remains a time-consuming task, while accurately and fully delineating the paleokarst-related features in the 3-D subsurface space is challenging. We propose a CNN to efficiently and accurately segment or detect the paleokarst-collapse features from the 3-D seismic images by computing a paleokarst collapse probability volume as shown in Figures 2b, 2d, 3c, and 3d. From the probability volumes, we are able to further automatically obtain the 3-D bodies of the paleokarst chimneys or sags in the 3-D subsurface space, which provides a convenient way to quantitatively analyze collapsed paleokarst systems.

3. Seismic Simulation of Paleokarst Collapses

To train a CNN for segmenting the paleokarst-related features from a 3-D seismic image, we propose a workflow (as shown in Figure 5) to automatically generate many 3-D synthetic seismic images where the folding structures and collapsed paleokarst features are diverse and realistic. The well-defined collapsed paleokarst features can be automatically and perfectly labeled to obtain the target images for training the CNN.

3.1. Simulating Folding Structures

In this workflow, we begin with an initial 3-D reflectivity model r0(x,y,z) with flat layers (as shown in Figure 5a), where the reflectivity values at each layer are randomly generated and smoothly vary in space. We then randomly generate folding structures by vertically shearing the flat model, where the shearing field s(x,y,z) is defined as a combination of two functions:

where the first function s1 is a combination of N Gaussian functions scaled by a vertically damping function as suggested by Wu, Liang, et al. (2019) and Wu et al. (2020):

In defining this function, the parameters are all randomly chosen from some predefined ranges as discussed by Wu et al. (2020). By using the damping scalar function 1.5/Zmax, we gradually decrease the curvature (or bending extent) vertically upward from bottom to top in the model. The second function is simply a linear function as follows:

where the parameters p, q, and c0 are constant values that are randomly chosen for each reflectivity model. This linear function is used to generate the purely dipping structures in the reflectivity model. The slopes of the dipping structures in the x− and y− directions, respectively, are defined by the parameters p and q, which are randomly chosen from a limited range of [−0.25, 0.25] to avoid generating layers with extremely high slopes after combining with s1 (Wu et al., 2020). We set to ensure that the center trace of the reflectivity model is not shifted.

With the defined vertical shearing field we are able to compute a folded reflectivity model r(x,y,z) from the initial model r0(x,y,z) as follows:

where a classic sinc interpolation is applied to map the reflectivity model from the initially flat space to the folded or sheared space. As the parameters for defining the shearing functions are randomly chosen in generating the folded reflectivity model, the folding structures in each model are unique. In addition, each parameter has many options; therefore, we have numerous possible combinations of parameters for generating numerous folded reflectivity models with various structures.

3.2. Simulating Collapsed Paleokarst Structures

After generating the folded reflectivity model, the next step is to further simulate the collapsed paleokarst structure features in the model. Based on the observations from the field seismic images in Figures 2 and 3, the most dominant features of the collapsed paleokarst structures apparent in the seismic images are the chimneys or sags, the fractures or faults within the chimney zones, and the steep circular or cylindrical faults bounding the sags. As shown in the vertical seismic sections in Figures 2 and 3, each collapse chimney or sag is apparent as a vertically elongated tube with an approximately ellipsoidal shape. The vertically extended chimney or sag structures are cut by horizontal seismic slices, where we observe obvious circular features or onion rings (as denoted by the red and magenta arrows in Figures 2a and 3a). Such circular features indicate lateral changes in reflection time or depth due to the sag structures. Within some of the sag zones, the seismic reflections are highly chaotic or discontinuous (as denoted by the magenta arrows in Figure 3 and magenta ellipses in Figure 3b), which indicates that the deposits within these sag zones may be highly fractured or faulted.

To simulate a collapse chimney tube in the 3-D reflectivity model, we construct a 3-D vertically elongated ellipsoid

with which we define the 3-D tube area of a chimney as follows:

In the ellipsoidal function (Equation 5), p=(x,𝑦,z) represents the coordinates of a point in the 3-D reflectivity model. c=(cx,c𝑦,cz) represents the center of the ellipsoid, which is randomly chosen within the 3-D space of the model. A is a diagonal matrix defined by the three radii of the ellipsoid as follows:

where the values of rx,ry, and rz, respectively, are randomly chosen from the predefined ranges [1, 12], [1, 12],and [4, 80] to construct ellipsoids of different sizes. Here, we set a wider range for rz to obtain primarily vertically elongated ellipsoids. In addition, we set rx>0.1rz and ry>0.1rz to avoid generating extremely longbut narrow ellipsoids that are unrealistic in practice. R is a rotation matrix that rotates the ellipsoid around the x− and y− axes as follows:

where the rotation angles 𝛼 and 𝛽 are randomly chosen from a narrow range [−10◦,10◦] to construct as lightly dipping or vertically aligned ellipsoid. By randomly choosing all the parameters in Equation 5, we are able to construct numerous possible ellipsoids with various shapes, sizes, orientations, and locations. However, in reality, a paleokarst chimney tube is not a perfect ellipsoid; we therefore randomly add some smooth perturbations to obtain irregular ellipsoids as shown in Figure 5f.

After defining the 3-D tube areas of the paleokarst chimneys, we then define the sag structures within the chimney cubes. The sag structures are typically downward bending. We therefore vertically shift the reflectors (inside a chimney cube) downward as follows to obtain a reflectivity model rk(x,y,z):

where the vertical shifts sk(x,y,z) are defined as

In this equation, f(x,y,z) is the ellipsoidal function defined in Equation 5. 𝛾 is a positive scalar that is randomly chosen from the range [10, 20]. 𝜖(x,y,z) is a random perturbation field implemented to simulate potential fractures or faults that dislocate the reflectors within the chimney tube. The perturbation field is relatively small compared to the first term of 𝛾(f(x,y,z)−1). When the perturbation field is close to 0, the shifts sk(x,y,z)≈𝛾(f(x,y,z)−1) will be nonpositive and smoothly decrease from 0 at the chimney bound to the most negative at the center of the chimney tube. In this case, the shifts will vertically shear the reflectors so that they are smoothly curved downward, as shown between the cyan dashed lines (cylindrical faults) in Figure 5c. Such generated sag structures will produce clear circular features or onion rings on the horizontal slice as denoted by the red arrows in Figure 5c. If the perturbation field is significant, the shifts sk(x,y,z) will no longer be smooth but noisy, resulting in the generation of curved and dislocated (fractured or faulted) reflectors within the chimney tubes as denoted by the magenta arrows and dashed ellipses in Figure 5c. By using this numerical method, we are able to automatically simulate various collapsed paleokarst features (as in Figure 5c) that look realistic and are comparable to the real collapsed paleokarst features in the field seismic images (Figures 2 and 3).

In summary, we have proposed a general workflow (Figure 5) to numerically simulate realistic folding and collapsed paleokarst structure features in a reflectivity model. The parameters used to model the folding and collapsed paleokarst features are summarized in Figure 6. A specific set of such parameters yields a reflectivity model with unique folding and collapsed paleokarst features. These parameters are not fixed but are randomly chosen to form numerous possible combinations, which allows us to generate numerous reflectivity models with various structural and collapsed paleokarst features. This is critical for our next step of training a CNN for collapsed paleokarst characterization which requires rich training data sets.

A summary of the parameters used to simulate folding and collapsed paleokarst structural features.
Figure 6. A summary of the parameters used to simulate folding and collapsed paleokarst structural features.

3.3. Training Data Sets

After constructing a reflectivity model (Figure 5c) with simulated folding structures and collapsed paleokarst features, our next step is to simulate a synthetic seismic image (Figure 5d) by convolving a Ricker wavelet with the reflectivity model. Instead of convolving in the vertical direction as in conventional methods, we compute the convolution here in the direction perpendicular to the reflector structures as suggested by Wu, Liang, et al. (2019) and Wu et al. (2020). The peak frequency of the Ricker wavelet is also randomly chosen from a predefined range (10–35 Hz) to compute a synthetic seismic image with various frequency components. As field seismic images are typically not as clean as the synthetic seismic image in Figure 5d, we further add random noise or the noise from field images to make the synthetic one more realistic. Figure 5eshows the corresponding label image that is automatically computed by labeling the paleokarst chimney cubes with ones while the nonkarst areas with zeros. From the label image, we can further automatically obtain the 3-D bodies of the paleokarst chimney cubes (Figure 5f) by simply calculating the isosurfaces (with isovalue 0.5) of the label image using the method of marching cubes (Lorensen & Cline, 1987).

In training a CNN for detecting the collapsed paleokarst features, we need to prepare many training dataset pairs consisting of input seismic images (like the one in Figure 5d) and the corresponding label images (like the one in Figure 5e). Fortunately, in our workflow, the parameters for simulating the synthetic seismic images and the collapsed paleokarst features are not fixed. By randomly choosing the parameters, we are able to obtain numerous combinations of the parameters to calculate numerous pairs of training image with various and diverse structures and collapsed paleokarst features. In this work, we generate 120 pairs of synthetic seismic and label images, 100 pairs for training and the rest for validation. Figure 7 shows eight pairs of the automatically generated training data sets, where the first and third rows show the input seismic images, while the second and fourth rows show the corresponding label images of collapsed paleokarst features overlain with the seismic images. Notice that noise has been added to the synthetic seismic images. The noise in the first row of images is extracted from various areas of the field seismic images. The noise in the third row of images is randomly generated. In adding the noise, the noise-to-signal-ratio is randomly selected from the range [0, 0.6]. The collapsed paleokarst zones are randomly scattered in the synthetic seismic image, some are cut at the image boundaries. This is consistent with the prediction for a large field seismic image, where we often need to cut a large volume into small subvolumes and make the prediction for each subvolume.

To further increase the diversity of the training and validation data sets, we apply two types of simple data augmentation to the original 100 pairs of training data sets. The first type of data augmentation involves rotating the images around the vertical axis by 0◦,90◦, 180◦, and 270◦, which increases the number of training data sets by a factor of 4. The second type of data augmentation involves randomly cutting smaller subimages from the original images and use the subimages to train the CNN. The dimension size of each of the originally generated images is 256×256×256 (samples). During the training process, we randomly cut 128×128×128 subimages from the larger images, which ensures that the training data sets are mostly different for different training epochs. By doing this, we are able to significantly increase the number and diversity of the training data sets. In addition, training the CNN with smaller images can greatly reduce GPU memory and computational costs during the training.

 eight pairs of synthetic seismic images (first and third rows) and the corresponding binary label images (second and fourth rows)
Figure 7. By using the proposed workflow (Figure 5), we are able to automatically generate numerous synthetic training data set pairs to train our CNN for recognizing collapsed paleokarst features in seismic images. Shown here are eight pairs of synthetic seismic images (first and third rows) and the corresponding binary label images (second and fourth rows) with the collapsed paleokarst features denoted by ones.

4. Deep learning for Detecting Collapsed Paleokarst Features

We design a CNN, as shown in Figure 8, to detect the collapsed paleokarst features in a 3-D seismic image. The architecture of the designed CNN consists of a U-shaped network followed by a residual block (Figure 9). The architecture of the 3-D U-shaped network is modified from the original 2-D U-net proposed by Ronneberger et al. (2015) for 2-D medical image segmentation, making it applicable to our 3-D problem. Compared to the original U-net, we reduce the number of layers and features at each layer to significantly save memory and computational costs which is especially important for our 3-D seismic image segmentation problem. In addition, we reduce the number of downsamplings or poolings from four to three because the dimension size (128×128×128) of our 3-D training images is relatively small and the downsampled images after four poolings would be too small (only 8×8×8) to effectively preserve the spatial or structural features in the original images.

Similar to the original U-net (Ronneberger et al., 2015), the modified U-shaped network still preserves the general architecture of a bottleneck at the middle and symmetric left contracting and right expansive paths, each of which contains three steps (reduced from four steps in the original U-net). In the left contracting path, each step consists of two 3×3×3 convolutional layers (each followed by a ReLU activation function), and a 2×2×2 max-pooling layer with stride 2. In each convolutional layer, multiple 3×3×3 filters are applied to convolve with the input image to obtain multiple output feature maps. The ReLU activation function is a nonlinear function (𝑓(x)=max(0,x)) applied to the feature maps to increase the nonlinearity of the network (Krizhevsky et al., 2012). The max-pooling layer reduces the size of the feature map by a factor of 2 and keeps only the maximum value with a 2×2×2 sliding window. At the first, second, and third steps, the numbers of feature maps at the conventional layers are 16, 32, and 64, respectively, which are significantly reduced with respect to those of the original U-net. The bottleneck at the middle contains two 3×3×3 convolutional layers, each of which contains 512 features and is, again, followed by a ReLU activation function.

CNN with the architecture of a U-net followed by two residual blocks
Figure 8.To detect the collapsed paleokarst features from an input seismic image, we design a CNN with the architecture of a U-net followed by two residual blocks (Figure 9).

The three steps in the right expanding path gradually upsample the downsampled features back to the original size. Each step consists of a 2×2×2 upsampling operation with stride 2, a concatenation to combine the features from the left path, and two 3×3×3 convolutional layers. Each convolutional layer is followedby a ReLU activation function. The upsampling operation is implemented by using the UpSampling3D layer defined in Keras (Chollet, 2015), which increases the size of an input image by a factor of 2. After the U-shaped network, we further add another two residual blocks whose architecture is shown in Figure 9. Each residual block contains two 3×3×3 convolutional layers, each of which contains 16 features and is followed by a ReLU activation function. The final output layer is a 1×1×1 convolutional layer followed by a sigmoid activation function that generates a 3-D probability image with values in the range [0, 1] as shown on the right of Figure 8.

In constructing the network (Figure 8), the parameters of each layer are randomly initialized, which are required to be further updated to create a good mapping of an input seismic image (left of the network) toa n output paleokarst image (right of the network). Updating the network parameters involves a training process that uses an optimization algorithm to iteratively update the parameters until the error between the outputs and the label images converges on the training data set. As we consider the detection of collapsed paleokarst features a binary image segmentation problem, we use the following loss of binary cross-entropy to measure the error between the outputs and the label images:

The structure of a residual block.
Figure 9. The structure of a residual block.

The training (orange) and validation (blue) curves over 25 epochs.
Figure 10. The training (orange) and validation (blue) curves over 25 epochs. The loss (a) for both the training and validation data sets quickly converges to a small value (nearly 0.03), while the prediction accuracy (b) significantly increases to nearly 0.99.

where N is the number of samples in a 3-D output or label image and yi and pi, respectively, represent the binary label and predicted values at the ith image sample. In the training process of optimizing the network parameters, we use the Adam method (Kingma & Ba, 2014) with a learning rate of 0.0001. We iteratively train the network with 25 epochs. One epoch involves passing the entire training data set (400 pairs, including the original and rotated images) forward and backward through the neural network once. At each epoch, the training images with a smaller size 128×128×128 are cut from larger generated images (256×256×256 as in Figure 7) at random locations. Therefore, the training images at different epochs are mostly different.

In addition, each seismic image is subtracted by its mean and then divided by its standard deviation to obtain a normalized image. This normalization process is applied to ensure the amplitudes of all the seismic images are consistently distributed. Such normalizations modify the seismic amplitudes but do not change the geo-metric features or structural patterns (relative amplitude variations), from which the collapsed paleokarst features are distinguished from the nonkarst features. At each epoch of the training process, we evaluate the updated network by computing the losses and the accuracies over both the synthetic training and validation data sets. The accuracy 𝛼 is defined as follows:

where, again, N is the number of samples in a 3-D output or label image and yi represents the binary label value at the ith image sample. ⌊pi⌉ represents the nearest integer to the predicted value pi(0≤pi≤1). We stop the training at the 25 epoch as both the training and validation losses converge to nearly 0.33, while the prediction accuracies over both the training and validation data set increase to nearly 0.99 (Figure 10).

5. Applications

To verify the effectiveness and generalization of the CNN trained with 25 epochs over the synthetic datasets, we apply the trained CNN to both synthetic and field 3-D seismic images. Although the CNN is trained by images of size 128×128×128, the size of the image input into the CNN for prediction is not fixed. This means that the trained CNN can be flexibly applied to images of arbitrary dimension sizes, so long as eachof the dimensions can be divided by 8 due to the three downsamplings included in the CNN architecture (Figure 8).

5.1. Synthetic Examples

Figure 11 shows the application of the coherence-based, curvature-based, and our CNN-based methods to four synthetic seismic images (first column in Figure 11) for collapsed paleokarst characterization. These four images (each with 256×256×256 samples) were generated by using the proposed numerical workflow (Figure 5) and are different from the images in the training data set. The first two images contain noise that is extracted from the field seismic images (Figures 2a and 3a). The second two images contain heavier random noise. The second row of Figure 11 shows the paleokarst probability images pc(x) computed from seismic coherence (Marfurt et al., 1999) as follows:

Characterization of paleokarst collapse features in four synthetic validation seismic images
Figure 11. Characterization of paleokarst collapse features in four synthetic validation seismic images (first row) by using coherence-based (second row), curvature-based (third row), and our CNN-based (fourth row) methods. Our CNN-based method shows the best performance in obtaining accurate paleokarst characterizations that are highly consistent with the ground-truth paleokarst features (last row).

where c(x) represents a seismic coherence map that typically highlights seismic discontinuities with relatively lower values. The value of the exponent, 6, is somewhat arbitrarily chosen and is used to increase the contrast between image samples with low and high coherences. The coherence-based paleokarst probability images (second row in Figure 11) highlight most seismic discontinuities, including those in the collapsed paleokarst zones and noise.

The third row of Figure 11 shows the paleokarst probability images computed from the most positive seismic curvature Al-Dossary and Marfurt (2006) as follows:

In this equation, 𝜅(x) represents the most positive curvature map computed from an input seismic image. The structures within paleokarst collapse zones are typically curved downward and accordingly show negative values in the curvature map 𝜅(x); we therefore clip 𝜅(x) to obtain𝜅c(x) with only negative values. We further define the curvature-based paleokarst probability by using the normalized and clipped curvature map 𝜅c(x) as shown in the above equation, where the power 6, again, increases the contrast between image samples with low and high curvatures. The curvature-based paleokarst probability images (third row of Figure 11) highlight some of the collapsed paleokarst zones as well as some downward folding structures that are unrelated to the paleokarst collapses.

(a) Precision-recall and (b) ROC curves
Figure 12. (a) Precision-recall and (b) ROC curves are calculated to quantitatively evaluate paleokarst detection on the synthetic validation volumes shown in Figure 11. Our CNN-based method (red curves) provides much more accurate results than the coherence-based (black curve) and curvature-based (blue curve) methods.

paleokarst probability images by using coherence-based (b), curvature-based (c), and our CNN-based (d) methods.
Figure 13. From a 3-D field seismic image (a), we calculate the paleokarst probability images by using coherence-based (b), curvature-based (c), and our CNN-based (d) methods. The paleokarst probability image computed by using our CNN-based method displays much cleaner and more accurate collapsed paleokarst features than those computed by using the other two methods.

(a) Precision-recall and (b) ROC curves
Figure 14. (a) Precision-recall and (b) ROC curves are calculated to quantitatively evaluate collapsed paleokarst detection on the 2-D vertical sections of the field seismic image in Figure 3. Our CNN-based method (red curves) provides much more accurate results than the coherence-based (black curve) and curvature-based (blue curve) methods. In calculating the evaluation curves, all the collapsed paleokarst detection results are compared to the ground-truth collapsed paleokarsts that are manually interpreted on the two vertical sections in Figure 3a.

The fourth row of Figure 11 shows that our CNN-based paleokarst probability images visually display much more accurate collapsed paleokarst characterizations than the coherence-based and curvature-based prob-ability images. The collapsed paleokarst features highlighted (colored by red) in our CNN-based probability images are consistent with collapsed paleokarst ground truth in the label images (last row of Figure 11). In addition, our CNN-based method is highly efficient. It took only milliseconds to compute such a probability image from an input seismic image with 256×256×256 samples by using a Quadro P6000 GPU.

In order to quantitatively evaluate the collapsed paleokarst characterization results, we further compute precision-recall (Martin et al., 2004) and receiver operating characteristic (ROC) (Provost et al., 1998) curves as shown in Figure 12. From the precision-recall curves (Figure 12a), we observe that our CNN-based method shows the highest precision for all choices of recall. From the ROC curve, our CNN-based method shows the highest true positive rates but the lowest false positive rates.

5.2. Field Examples in the FWB

In addition to the synthetic examples in Figure 11, we further use two field examples acquired from the FWB (Figure 1a), to verify the generalization of the trained CNN to field seismic images. Compared to those in the synthetic seismic images (Figure 11), the structures and features in the field images may be significantly different.

Figure 13a shows the first 3-D seismic image (the same one in Figure 3a), which was acquired at the sur-vey denoted by the red block in the FWB (Figure 1a). In this example, the paleokarst probability image (Figure 13b) computed by using the coherence-based method (Equation 13) highlights some collapsed paleokarst zones but is highly sensitive to noise. The curvature-based method (Equation 14) yields a better paleokarst probability image that detects more collapsed paleokarst zones but still contains some noisy features (e.g., the area denoted by the cyan circle) and fails to completely delineate some true paleokarst chimneys (as denoted by the magenta arrows in Figure 13c). Compared with the coherence-based and curvature-based methods, our CNN-based method yields the best probability image (Figures 3c and 13d), where the highlighted collapsed paleokarst zones are most consistent with the manual interpretation in Figure 3a.

As shown in the vertical sections in Figure 3c, the boundaries of the highlighted paleokarst chimneys are consistent with the manually interpreted cylindrical faults (cyan dashed lines) bounding the chimney tubes. In Figure 3d, we display two seismic Sections AA′ and BB′ that vertically slice a large paleokarst chimney tube, where we are able to more clearly view the details of the result. In these two sections, we observe that the highlighted chimney tube starts appearing in the Ellenburger group (below the blue horizon) and continuously extends upward into the Pennsylvanian strata at the top, which is consistent with the conclusions of previous studies (e.g., Hardage et al., 1996; McDonnell et al., 2007) on the collapsed paleokarst systems in the FWB. In addition, the boundaries of the detected paleokarst chimney are highly consistent with the manually interpreted bounding cylindrical faults as denoted by the cyan dashed lines in Figures 3b and 3d. The precision-recall and ROC curves in Figure 14 show quantitative evaluations of the CNN-based (red), curvature-based (black), and coherence-based (blue) paleokarst detections compared to the manual interpretation in the two vertical sections in Figure 3a. Our CNN-based method achieves the best performance while the coherence-based method shows the worst performance on this field example.

3-D paleokarst chimney tubes
Figure 15. The 3-D paleokarst chimney tubes in (a)–(c) are automatically extracted from the paleokarst probability image (Figure 13b), the most negative curvature image (Figure 13c), and the most positive curvature image (Figure 13d), respectively.

3-D seismic images
Figure 16. The image in (a) shows a different view of the 3-D seismic image in Figure 2a. From this 3-D seismic image, we compute a 3-D paleokarst probability image overlain with the seismic image in (b) as well as in Figure 2b. From the paleokarst probability image, we are able to automatically reconstruct the 3-D paleokarst chimney tubes (c) by simply extracting the isosurfaces (with an isovalue of 0.5) of the probability image.

From the paleokarst probability image (Figures 3c and 13b), we can further extract the 3-D paleokarst chimney tubes (Figure 15a) by simply extracting the isosurfaces (with an isovalue of 0.5) using the method of marching cubes (Lorensen & Cline, 1987). As comparisons, Figures 15a and 15b, respectively, display the chimney tubes extracted from the coherence-based (Figure 13b) and curvature-based (Figure 13c) probability images. We observe that the tubes extracted from the coherence-based and curvature-based images contains many noisy and irregular blocks, most of which are not true paleokarst chimney tubes. In addition, the shapes of the extracted tubes are highly noisy as well, which makes it difficult to quantitatively analyze their geometries. In contrast, the 3-D chimney tubes extracted from our paleokarst probability images are geologically more reasonable and the boundaries of the tubes are more clearly presented. Most of these extracted chimney tubes narrow upward, which is consistent with the observations by McDonnell et al. (2007). From each of these extracted chimney tubes, we can automatically and accurately measure the3-D geometric parameters (e.g., diameter and vertical length) of each tube and analyze the 3-D distribution of the collapsed paleokarst systems.

The second field seismic image shown in Figures 16a and 2a is from the survey denoted by the blue block in the FWB (Figure 1)a. The collapsed paleokarst systems in this seismic image have been extensively analyzed and well interpreted by previous studies (e.g., Alhakeem, 2013; Hardage et al., 1996; McDonnell et al., 2007). In this work, we compare our automatically interpreted results using the deep learning method with collapsed paleokarst systems that are manually interpreted by McDonnell et al. (2007). Figures 2b and 16b show the automatically estimated paleokarst probability images obtained by using the our CNN. We observe that the major paleokarst chimneys are clearly highlighted in red, which correspond to relatively high probabilities.

To closely visualize the details of the detected collapsed paleokarst features and compare them with the manual interpretation results, we extract two seismic sections marked by Line A and Line B in Figures 2a and 2b. As shown in Figure 2b, Line A passes through five circular anomalies with high paleokarst probability. The extracted seismic section at Line A is shown in Figure 2c, where the paleokarst chimneys (sags)and the steep concentric or cylindrical faults were manually interpreted by McDonnell et al. (2007). Based on the manual interpretation, the paleokarst chimneys originate within the Ellenburger Group (the top of Ellenburger is marked by the blue line in Figure 2c) and extends vertically upward through the lower Atokanto the Caddo Pool Formation (McDonnell et al., 2007). Figure 2d shows the corresponding paleokarst probability section (overlain on the seismic section of Line A), where the areas of the five paleokarst chimneys are clearly highlighted. The contours (cyan curves), extracted from the probability image at the probability value of 0.5, delineate the boundaries of the paleokarst chimneys, which are consistent with the manually interpreted concentric faults (dashed black lines in Figure 2c) that bound the chimneys. Line B passes one circular anomaly with high paleokarst probabilities as shown in Figure 2b. Figure 17 shows the seismic section of Line B with manual and our automatic interpretation of the paleokarst chimney in the middle. Again, our automatic interpretation of the paleokarst chimney (Figures 17b and 17c) in this seismic section is consistent with the manual interpretation (Figure 17a) by McDonnell et al. (2007).

Vertical seismic section displayed in various interpretations
Figure 17. The vertical seismic section at Line B (denoted by the magenta line in Figures 2a and 2b) is displayed with the manual interpretation (a) by McDonnell et al. (2007), our CNN-based paleokarst probability image (b), and the paleokarst boundaries (cyan curves in c) that were automatically extracted from the probability image.

Figure 18a shows a horizontal coherency slice of the 3-D seismic image, where the chimney structures are indicated as relatively small coherence anomalies (colored in dark gray). On this coherence slice, 10 circular chimney structures are manually marked as white circles according to McDonnell et al. (2007). Figure 18b shows the corresponding horizontal slice of the paleokarst probability map overlain with the coherence slice. The circular red areas in Figure 18b denote all the automatically detected chimney structures including the 10 manually interpreted chimneys in Figure 18a. Of course, some of these detected paleokarst chimneys still need to be verified although most of them look reasonable as shown in the vertical sections in Figures 16b and 16c. By extracting the contours (with a probability value 0.5) from the paleokarst probability slice, we are able to automatically obtain the boundaries (cyan curves in Figure 18b) of all the chimneys, which provide a convenient way to quantitatively analyze the geometry of each chimney, including its shape and size.

A coherency slice and the CNN-based paleokarst probability slice at the same location
Figure 18. A coherency slice (a) of the 3-D seismic image is extracted at 1,140 ms TWT, where the circular paleokarst features are characterized as dark gray colors and the boundaries of ten circular paleokarsts (white curves) are manually interpreted by McDonnell et al. (2007). The CNN-based paleokarst probability slice (b) at the same location shows a much clearer characterization of the circular paleokarst features (highlighted by red colors), from which we can automatically extract the boundaries of all the circular paleokarst features as denoted by the cyan curves. The automatically extracted boundaries of the 10 paleokarsts (denoted by yellow numbers in b) are quite consistent with the manual interpretation (denoted by black numbers in a).

Subseismic image (a), subpaleokarst image (b), and the extracted paleokarst tube (c)
Figure 19. To closely view our CNN-based paleokarst detection, we extract a small subseismic image (a) containing the third circular paleokarst denoted by the cyan arrow in Figures 18b and 2b. The corresponding subvolume of the paleokarst probability image is overlain with the seismic image in (b), from which we automatically extract the 3-D paleokarst chimney tube (c) and automatically calculate the geometric parameters of the paleokarst. Horizontal slices of the subseismic image (a), subpaleokarst image (b), and the extracted paleokarst tube (c) are shown in (d)–(f), respectively.

Figure 19 shows a subvolume containing the third paleokarst chimney denoted by the cyan arrows in Figures 18b and 2b. From the subvolume of the paleokarst probability map (Figure 19b), we can automatically extract the paleokarst chimney structure as a vertically extended 3-D tube, as shown in Figure 19c. Figures 19d–19f, respectively, show horizontal slices of the seismic amplitude, paleokarst probability, and extracted chimney tube. From the extracted 3-D chimney tube, we can automatically estimate the geometric parameters of the paleokarst system such as the length and width as marked in Figures 19c and 19f, which are consistent with the manual measurements provided by McDonnell et al. (2007).

6. Conclusions

We have proposed a CNN-based method to automatically delineate collapsed paleokarst features in 3-D seismic images, where the paleokarst detection is considered a problem of 3-D binary image segmentation. The architecture of the CNN is modified from the U-net, which was originally proposed for 2-D medical image segmentation and is today the most widely used network for common image segmentation problems. Specifically, we upgraded the original 2-D convolutional layers into 3-D layers and reduced the number of layers and features at each layer to make it applicable to our problem of collapsed paleokarst interpretation in 3-D seismic images.

The main challenge or limitation of applying supervised CNNs to geoscience problems, including our collapsed paleokarst interpretation, is the lack of training data sets, especially the label images or interpretation results. To solve this problem, we proposed an efficient and effective workflow to automatically generate reflectivity models with realistic folding structures and collapsed paleokarst features and then simulated the corresponding synthetic seismic images by convolving the reflectivity models with Ricker wavelets and adding some extent of random noise or real noise from field seismic images. As the folding structures and collapsed paleokarst features are well defined by functions in generating the reflectivity models, we were able to automatically obtain the corresponding label images with ground-truth collapsed paleokarst features. In addition, the parameters of functions are not fixed but randomly chosen, which makes it possible to generate numerous synthetic training data sets with various folding and collapsed paleokarst features.

Although trained with only synthetic data sets, the CNN can be applied to characterize the collapsed paleokarst features in field seismic images that are not included in the training data sets. The application of two field examples demonstrated that the collapsed paleokarst interpretation of our CNN-based methods is consistent with manual interpretations and is significantly superior to conventional automatic methods. From the CNN-based collapsed paleokarst interpretation results, we were able to automatically extract the3-D boundary of each paleokarst chimney tube in the seismic images, which enabled us to automatically measure the geometric parameters of the paleokarst tubes for quantitative analysis of the paleokarst system development.

The features of collapsed paleokarsts formed in different areas can be highly variable. However, the current trained CNN model may be limited to characterizing collapsed paleokarst systems in 3-D seismic images as we mainly simulated the collapsed paleokarst features in our synthetic training data sets. Fortunately, the current CNN model can be easily extended to delineate any type of paleokarst system by simply including more training data sets that contain the corresponding paleokarst features.

Data Availability Statement

These two seismic images are available through Hardage et al. (1996) and Qi et al. (2014). The 120 pairs of synthetic data sets, used for training and validating our CNN, are uploaded to Zenodo and are freely available through the DOI link ( The codes related to this work is available through GitHub ( or Zenodo (

Al-Dossary, S., & Marfurt, K. J. (2006). 3D volumetric multispectral estimates of reflector curvature and rotation. Geophysics,71, P41–P51.

Alhakeem, A. A. (2013). 3D seismic data interpretation of Boonsville Field, Texas (PhD thesis), Missouri University of Science andTechnology.

Andre, P., & Doulcet, A. (1970). Rospo Mare Field, Italy, Apulian Platform, Adriatic Sea, Stratigraphic traps II: AAPG Treatise of petroleum geology(pp. 29–54). Rospo Mare, Italy: Atlas of Oil and Gas Fields.

Badrinarayanan, V., Kendall, A., & Cipolla, R. (2017). Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, 2481–2495.

Bahorich, M., & Farmer, S. (1995). 3-D seismic discontinuity for faults and stratigraphic features: The coherence cube. The Leading Edge,14, 1053–1058.

Bergen, K. J., Johnson, P. A., de Hoop, M. V., & Beroza, G. C. (2019). Machine learning for data-driven discovery in solid Earth geoscience. Science,363, eaau0323.

Bianco, M. J., Gerstoft, P., Traer, J., Ozanich, E., Roch, M. A., Gannot, S., & Deledalle, C.-A. (2019). Machine learning in acoustics: Theory and applications. The Journal of the Acoustical Society of America,146, 3590–3628.

Canter, K. L., Stearns, D. B., Geesaman, R. C., & Wilson, J. L. (1993). Paleostructural and related paleokarst controls on reservoir development in the Lower Ordovician Ellenburger Group, Val Verde Basin, Paleokarst related hydrocarbon reservoirs: SEPM Core Workshop(Vol. 18, pp. 61–99). Tulsa, OK: SEPM Society for Sedimentary Geology.

Chen, Y. (2016). Probing the subsurface karst features using time-frequency decomposition. Interpretation,4, T533–T542.

Chen, L.-C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2017). Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFS.IEEE Transactions on Pattern Analysis and Machine Intelligence,40, 834–848.

Chollet, F. (2015). Keras.

Coogan, A., Bebout, D., & Maggio, C. (1972). Depositional environments and geologic history of Golden Lane and Poza Rica trend, Mexico, an alternative view. AAPG Bulletin, 56, 1419–1447.

Desheng, L., Digang, L., Chengzao, J., Gang, W., Qizhi, W., & Dengfa, H. (1996). Hydrocarbon accumulations in the Tarim Basin, China. AAPG bulletin,80, 1587–1603.

Di, H., & AlRegib, G. (2020). A comparison of seismic saltbody interpretation via neural networks at sample and pattern levels. Geophysical Prospecting,68, 521–535.

Di, H., & Gao, D. (2016). Efficient volumetric extraction of most positive/negative curvature and flexure for fracture characterization from3D, seismic data. Geophysical Prospecting, 64, 1454–1468.

Di, H., Shafiq, M. A., Wang, Z., & AlRegib, G. (2019). Improving seismic fault detection by super-attribute-based classification. Interpretation, 7, SE251–SE267.

Di, H., Wang, Z., & AlRegib, G. (2018). Deep convolutional neural networks for seismic salt-body delineation. Presented at the AAPG Annual Convention and Exhibition.

Dou, Q., Sun, Y., Sullivan, C., & Guo, H. (2011). Paleokarst system development in the San Andres Formation, Permian Basin, revealed by seismic characterization. Journal of Applied Geophysics, 75, 379–389.

Geng, Z., Wu, X., Shi, Y., & Fomel, S. (2019). Relative geologic time estimation using a deep convolutional neural network. 88th Annual International Meeting, SEG, Expanded Abstracts, 2238–2242.

George, M. C. (2016). The Muenster Uplift of North Texas: The Easternmost Expression of the Pennsylvanian Ancestral Rockies (PhD thesis), The University of Texas at Dallas.

Guillen, P., Larrazabal, G., González, G., Boumber, D., & Vilalta, R. (2015). Supervised learning to detect salt body, SEG Technical Program Expanded Abstracts 2015 (pp. 1826–1829). Tulsa, OK: Society of Exploration Geophysicists.

Hardage, B. A. (1996). Boonsville 3-D, data set. The Leading Edge,15, 835–837.

Hardage, B., Carr, D., Lancaster, D., Simmons, J., Elphick, R., Pendleton, V., & Johns, R. (1996). 3-d seismic evidence of the effects of carbonate karst collapse on overlying clastic stratigraphy and reservoir compartmentalization. Geophysics, 61, 1336–1350.

He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV (pp. 770–778).

Kerans, C. (1988). Karst-controlled reservoir heterogeneity in Ellenburger group carbonates of West Texas. AAPG bulletin,72, 1160–1183.

Khatiwada, M., Keller, G. R., & Marfurt, K. J. (2013). A window into the Proterozoic: Integrating 3D seismic, gravity, and magnetic data to image subbasement structures in the southeast Fort Worth Basin. Interpretation,1, T125–T141.

Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. CoRR,

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In F. Pereira, C. J. C. Burges, L. Bottou, & K. Q. Weinberger (Eds.), Advances in Neural Information Processing Systems (pp. 1097–1105). New York: Curran Associates, Inc. Retrieved from

Li, F., & Lu, W. (2014). Coherence attribute at different spectral scales. Interpretation,2, SA99–SA106.

Lin, T., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. In 2017 IEEEConference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI (pp. 936–944).

Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016). SSD: Single shot multibox detector, European conference on computer vision (pp. 21–37). Springer.

Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA (pp. 3431–3440).

Lorensen, W. E., & Cline, H. E. (1987). Marching cubes: A high resolution 3D surface construction algorithm. ACM Siggraph Computer Graphics, 21, 163–169.

Loucks, R. G. (1999). Paleocave carbonate reservoirs: Origins, burial-depth modifications, spatial complexity, and reservoir implications. AAPG bulletin, 83, 1795–1834.

Loucks, R. G., & Anderson, J. H. (1985). Depositional facies, diagenetic terranes, and porosity development in Lower Ordovician Ellenburger Dolomite, Puckett Field, West Texas, Carbonate petroleum reservoirs (Vol. 18, pp. 19–38). Berlin: Springer-Verlag.

Lu, P., Morris, M., Brazell, S., Comiskey, C., & Xiao, Y. (2018). Using generative adversarial networks to improve deep-learning fault interpretation networks. The Leading Edge,37, 578–583.

Maoshan, C., Shifan, Z., Zhonghong, W., Hongying, Z., & Lei, L. (2011). Detecting carbonate-karst reservoirs using the directional amplitude gradient difference technique. In SEG Technical Program Expanded Abstracts 2011(pp. 1845–1849). Tulsa, OK: Society of Exploration Geophysicists.

Marfurt, K. J., & Rich, J. (2010). Beyond curvaturevolumetric estimates of reflector rotation and convergence. In SEG Technical Program Expanded Abstracts 2010(pp. 1467–1472). Tulsa, OK: Society of Exploration Geophysicists.

Marfurt, K. J., Sudhaker, V., Gersztenkorn, A., Crawford, K. D., & Nissen, S. E. (1999). Coherency calculations in the presence of structural dip. Geophysics, 64(1), 104–111.

Martin, D. R., Fowlkes, C. C., & Malik, J. (2004). Learning to detect natural image boundaries using local brightness, color, and texture cues. IEEE Transactions on Pattern Analysis and Machine Intelligence, 26, 530–549.

Mazzullo, S. J., & Mazzullo, L. J. (1970). Paleokarst and karst-associated hydrocarbon reservoirs in the Fusselman Formation, West Texas, Permian Basin. In Paleokarst, karst related diagenesis and reservoir development: Examples from ordoviciandevonian age strata of west Texas and the mid-continent (Vol. 92-33, pp. 110–120). Tulsa, OK: Permian Basin Section, SEPM Publication.

McDonnell, A., Loucks, R. G., & Dooley, T. (2007). Quantifying the origin and geometry of circular sag structures in northern Fort Worth Basin, Texas: Paleocave collapse, pull-apart fault systems, or hydrothermal alteration?AAPG Bulletin, 91, 1295–1318.

Pham, N., Fomel, S., & Dunlap, D. (2019). Automatic channel detection using deep learning. Interpretation, 7(3), SE43–SE50.

Pollastro, R. M., Jarvie, D. M., Hill, R. J., & Adams, C. W. (2007). Geologic framework of the Mississippian Barnett Shale, Barnett-Paleozoic total petroleum system, Bend arch-Fort Worth Basin, Texas. AAPG Bulletin,91, 405–436.

Provost, F. J., Fawcett, T., & Kohavi, R. (1998). The case against accuracy estimation for comparing induction algorithms. ICML, 445–453.

Qi, J., & Castagna, J. (2013). Application of a PCA fault-attribute and spectral decomposition in Barnett Shale fault detection. In SEG Technical Program Expanded Abstracts 2013(pp. 1421–1425). Tulsa, OK: Society of Exploration Geophysicists.

Qi, J., Zhang, B., Zhou, H., & Marfurt, K. (2014). Attribute expression of fault-controlled karst—Fort Worth Basin, Texas: A tutorial. Interpretation,2, SF91–SF110.

Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV (pp. 779–788).

Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster R-CNN: Towards real-time object detection with region proposal networks. Advances in Neural Information Processing Systems,39, 91–99.

Roberts, A. (2001). Curvature attributes and their application to 3D interpreted horizons. First Break,19, 85–100.

Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In InternationalConference on Medical image computing and computer-assisted intervention (pp. 234–241). New York: Springer.

Sermanet, P., Eigen, D., Zhang, X., Mathieu, M., Fergus, R., & LeCun, Y. (2013). Overfeat: Integrated recognition, localization and detection using convolutional networks. arXiv preprint arXiv,1312, 6229.

Shi, Y., Wu, X., & Fomel, S. (2019). Saltseg: Automatic 3D salt segmentation using a deep convolutional neural network. Interpretation,7(3), SE113–SE122.

Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv,1409,1556.

Soriano, M. A., Pocoví, A., Gil, H., Perez, A., Luzón, A., & Marazuela, M. Á. (2019). Some evolutionary patterns of palaeokarst developed in Pleistocene deposits (Ebro Basin, NE Spain): Improving geohazard awareness in present-day karst. Geological Journal,54, 333–350.

Sullivan, E. C., Marfurt, K. J., Lacazette, A., & Ammerman, M. (2006). Application of new seismic attributes to collapse chimneys in the Fort Worth Basin. Geophysics,71, B111–B119.

Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., et al. (2015). Going deeper with convolutions. In2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston, MA (pp. 1–9).

Viniegra, F., & Castillo-Tejero, C. (1970). Golden Lane Fields, Veracruz, Mexico. In Geology of giant petroleum fields (Vol. 14, pp. 309–325). Tulsa, OK: AAPG Memoir.

Walper, J. L. (1982). Plate tectonic evolution of the Fort Worth Basin. In Petroleum geology of the Fort Worth Basin and Bend arch area (pp. 237–251). Dallas, TX: Dallas Geological Society.

Wen, T., Castro, M. C., Nicot, J.-P., Hall, C. M., Larson, T., Mickler, P., & Darvari, R. (2016). Methane sources and migration mechanisms in shallow groundwaters in Parker and Hood Counties, Texas—A heavy noble gas analysis. Environmental science & technology,50,12,012–12,021.

Wu, X., Geng, Z., Shi, Y., Pham, N., Fomel, S., & Caumon, G. (2020). Building realistic structure models to train convolutional neural networks for seismic structural interpretation. Geophysics,85, WA27–WA39.

Wu, X., Liang, L., Shi, Y., & Fomel, S. (2019). FaultSeg3D: Using synthetic datasets to train an end-to-end convolutional neural network for 3D seismic fault segmentation. Geophysics,84(3), IM35–IM45.

Wu, X., Shi, Y., Fomel, S., Liang, L., Zhang, Q., & Yusifov, A. Z. (2019). Faultnet3D,: Predicting fault probabilities, strikes, and dips with a single convolutional neural network. IEEE Transactions on Geoscience and Remote Sensing,57, 9138–9155.

Wu, H., Zhang, B., Lin, T., Cao, D., & Lou, Y. (2019). Semiautomated seismic horizon interpretation using the encoder-decoder convolutional neural network. Geophysics,84, B403–B417.

Zeng, H., Wang, G., Janson, X., Loucks, R., Xia, Y., Xu, L., & Yuan, B. (2011). Characterizing seismic bright spots in deeply buried, Ordovician Paleokarst strata, Central Tabei uplift, Tarim Basin, Western China.Geophysics,76, B127–B137.

Zhai, G., & Zha, Q. (1982). Buried-hill oil and gas pools in the North China Basin, in The deliberate search for the subtle trap. AAPG Memoir,32, 317–335.

Zhao, T., Li, F., & Marfurt, K. J. (2018). Seismic attribute selection for unsupervised seismic facies analysis using user-guided data-adaptive weights. Geophysics,83, O31–O44.

Zhao, T., & Mukhopadhyay, P. (2018). A fault detection workflow using deep learning and image processing. In SEG Technical ProgramExpanded Abstracts 2018(pp. 1966–1970). Tulsa, OK: Society of Exploration Geophysicists.

Welcome Back!

Download PDF here


Request access by filling the form below to download full PDF.
Most Popular Papers
Case Study: An Integrated Machine Learning-Based Fault Classification Workflow
Using machine learning to classify a 100-square-mile seismic volume in the Niobrara, geoscientists were able to interpret thin beds below ...
Case Study with Petrobras: Applying Unsupervised Multi-Attribute Machine Learning for 3D Stratigraphic Facies Classification in a Carbonate Field, Offshore Brazil
Using machine learning to classify a 100-square-mile seismic volume in the Niobrara, geoscientists were able to interpret thin beds below ...
Applying Machine Learning Technologies in the Niobrara Formation, DJ Basin, to Quickly Produce an Integrated Structural and Stratigraphic Seismic Classification Volume Calibrated to Wells
Carolan Laudon, Jie Qi, Yin-Kai Wang, Geophysical Research, LLC (d/b/a Geophysical Insights), University of Houston | Published with permission: Unconventional Resources ...
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

    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.

    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.

    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.


    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
    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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.

    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

    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.

    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.

    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.

    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.

    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.

    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).

    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.

    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.

    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.