Previous Article in Journal
Advancing the Prediction and Evaluation of Blast-Induced Ground Vibration Using Deep Ensemble Learning with Uncertainty Assessment
Previous Article in Special Issue
North Caucasus Underground Geophysical Observatory: Instrumentation and Observation Results
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Geosciences
Volume 15
Issue 5
10.3390/geosciences15050183
geosciences-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Seismic Facies Classification of Salt Structures and Sediments in the Northern Gulf of Mexico Using Self-Organizing Maps

by
Silas Adeoluwa Samuel
*,
Camelia C. Knapp
and
James H. Knapp
Boone Pickens School of Geology, 105 Noble Research Center, Oklahoma State University, Stillwater, OK 74078, USA
*
Author to whom correspondence should be addressed.
Geosciences 2025, 15(5), 183; https://doi.org/10.3390/geosciences15050183
Submission received: 16 April 2025 / Revised: 11 May 2025 / Accepted: 12 May 2025 / Published: 19 May 2025
(This article belongs to the Special Issue Editorial Board Members’ Collection Series: “New Horizons in Geophysics: From Theory to Applications”)
Browse Figures
Versions Notes

Abstract

:
Proper geologic reservoir characterization is crucial for energy generation and climate change mitigation efforts. While conventional techniques like core analysis and well logs provide limited spatial reservoir information, seismic data can offer valuable 3D insights into fluid and rock properties away from the well. This research focuses on identifying important structural and stratigraphic variations at the Mississippi Canyon Block 118 (MC-118) field, located on the northern slope of the Gulf of Mexico, which is significantly influenced by complex salt tectonics and slope failure. Due to a lack of direct subsurface data like well logs and cores, this area poses challenges in delineating potential reservoirs for carbon storage. The study leveraged seismic multi-attribute analysis and machine learning on 3-D seismic data and well logs to improve reservoir characterization, which could inform field development strategies for hydrogen or carbon storage. Different combinations of geometric, instantaneous, amplitude-based, spectral frequency, and textural attributes were tested using Self-Organizing Maps (SOM) to identify distinct seismic facies. SOM Models 1 and 2, which combined geometric, spectral, and amplitude-based attributes, were shown to delineate potential storage reservoirs, gas hydrates, salt structures, associated radial faults, and areas with poor data quality due to the presence of the salt structures more than SOM Models 3 and 4. The SOM results presented evidence of potential carbon storage reservoirs and were validated by matching reservoir sands in well log information with identified seismic facies using SOM. By automating data integration and property prediction, the proposed workflow leads to a cost-effective and faster understanding of the subsurface than traditional interpretation methods. Additionally, this approach may apply to other locations with sparse direct subsurface information to identify potential reservoirs of interest.

1. Introduction

Accurately characterizing reservoirs is necessary for the geologic storage of carbon dioxide or hydrogen to mitigate the adverse effects of climate change [1,2,3,4]. Conventional and high-resolution techniques such as core analysis and well log interpretation provide limited information on the spatial distribution of reservoir properties [5,6]. However, the use of seismic data provides 3-dimensional subsurface rock information. Integrating multiple seismic attributes extracted from seismic data can enhance reservoir characterization by providing valuable insights into fluid and rock properties between wells [7,8,9].
Despite the ability of seismic multi-attribute analyses to enhance reservoir interpretation, analyzing numerous attributes simultaneously can pose a challenge [10]. Hence, machine learning (ML) algorithms can be incorporated into a seismic multi-attribute analysis workflow to recognize patterns and create clusters the human eye may not readily detect, such as reservoir sands [11,12,13]. Unsupervised machine learning algorithms such as self-organizing maps (SOM) have been instrumental in conducting seismic facies analysis, reservoir characterization, and fault identification [14]. Self-organizing maps have an artificial neural network architecture that reduces the dimensionality of input data while preserving its topologic characteristics [15,16,17,18,19]. As the SOM algorithm runs, multi-attributes are mapped onto similar neighboring nodes in an unsupervised manner, thereby enabling the interpreter to identify distinct seismic facies and understand the spatial distribution of lithologic properties [14,18,20].
The study area, Mississippi Canyon 118 (MC-118), is influenced by salt tectonics and slope failure, resulting in the formation of fault systems [21,22,23,24,25,26]. In 2015, ConocoPhillips drilled a 19,400 ft exploration well to the Harrier prospect at MC-118 and found no commercial hydrocarbons [27]. Nonetheless, this location holds some potential for geologic hydrogen or CO2 storage due to existing geophysical data, the presence of geologic reservoirs, and oil and gas infrastructure. Identifying potential reservoirs and other geologic features such as salt structures, faults, and gas hydrates would provide insight into the suitability of this location for CO2 storage.
This research aims to: (1) test the efficacy of using unsupervised machine learning for seismic facies classification in a geologically complex location; (2) delineate distinct seismic facies to aid in identifying reservoirs with potential for the storage of gases; and (3) leverage automated data integration using a workflow that can save the cost and time of simultaneously analyzing multiple seismic attributes compared to conventional interpretation.
Unsupervised machine learning algorithms like SOM are beneficial in classifying seismic facies based on amplitude, frequency, and geometry without requiring predefined classifications, ensuring uniform analysis across extensive 3D seismic data [8,10,12,14]. However, SOM may face certain challenges, including sensitivity to data quality, dimensionality issues, ambiguous interpretations, complex relationships between attributes, and difficulty in estimating its accuracy [8,10,16]. Hence, applying workflows that incorporate seismic attribute selection and validation using available well logs would aid in reducing classification uncertainty.

2. Geologic Setting

Mississippi Canyon Block 118 (MC-118) is located on the northern slope of the Gulf of Mexico, approximately 120 km southeast of New Orleans, and at a water depth of about 2900 feet [21,24,26,28]. MC-118 covers approximately 84 km2 and is located on the Gulf’s shelf (Figure 1). The presence of potential geologic reservoirs, available geophysical data, and pre-existing oil and gas infrastructure presents opportunities for geologic hydrogen or CO2 storage. Salt beds were deposited in the Gulf of Mexico during the Mesozoic Era, followed by the deposition of sediments in the Late Cretaceous from materials weathered from the uplifted Rocky Mountains [29].
From the Jurassic to Recent geologic periods, the continuous movement of allochthonous salt has shaped the continental margin along the northern slope of the Gulf, including MC-118 [21,22,24,26,29,30] (Figure 2). Additionally, this has led to faulting and consequent gas leak points or seepage across the ocean floor [29,30]. A consortium was formed in 1999 by the Center for Marine Resources and Environmental Technology at the University of Mississippi to study the gas hydrates at MC-118, which were discovered in the summer of 2002 on the seafloor [21,22,23,24,26,31].
The geomorphology, past environmental conditions, potential for hydrocarbon resources, seismic facies, geochemical and biological processes at those depths, and their connection to gas hydrates as potential hazards have been investigated by many researchers [21,22,24,26,32,33].
The presence of faulting, slope failure, and gas hydrates bring complexity to the geology of this site in the shallower sections (less than 400 m below the seafloor) (Figure 3). Additionally, the reduction in seismic data quality with increasing depth also increases the difficulty in interpreting seismic data [34,35]. Hence, applying seismic multi-attribute analysis and machine learning aids in delineating seismic facies that would inform potential sites for carbon or hydrogen storage.
Geosciences 15 00183 g002
Figure 2. Stratigraphic sequence highlighting key seismic reflection events along with the associated geologic timescale modified from [36]. V.E. (Vertical Exaggeration).
Figure 2. Stratigraphic sequence highlighting key seismic reflection events along with the associated geologic timescale modified from [36]. V.E. (Vertical Exaggeration).
Geosciences 15 00183 g002
Geosciences 15 00183 g003
Figure 3. The location map, an interpreted time-migrated seismic section, and well log information of reservoirs drilled by the ConocoPhillips (Modified after [28,37]). GR = Gamma Ray; RES = Resistivity.
Figure 3. The location map, an interpreted time-migrated seismic section, and well log information of reservoirs drilled by the ConocoPhillips (Modified after [28,37]). GR = Gamma Ray; RES = Resistivity.
Geosciences 15 00183 g003

3. Datasets

The seismic data used for this research is a time-migrated 3D post-stack seismic volume acquired in 2014 at MC-118. The data consists of a record length of 10.5 s, a sample interval of 4 ms, and a 25 m bin size (Figure 3). The available well log information was from the exploration well drilled by ConocoPhillips in 2015 to the Harrier prospect. This well contains a suite of geophysical logs (Gamma Ray, Thermal Neutron Porosity, P-wave, Density, Density Correction, Caliper, and Resistivity logs) useful for reservoir characterization and validating SOM results (Figure 3).

4. Methodology

The workflow adopted for this research to extract numerous seismic attributes for multi-attribute analysis and computing the Self-Organizing Maps (SOM) algorithm for seismic facies identification can be seen in Figure 4. The Attribute Assisted Seismic Processing and Interpretation (AASPI) software provided by the ASSPI Consortium was used in extracting seismic attribute volumes and executing the SOM algorithm, while SLB’s Petrel software was used in picking seismic horizons across continuous amplitude traces but terminated against salt and extracting the seismic attributes on each horizon.

4.1. Seismic Facies Identification

Seismic facies can be defined as a set of distinct patterns identifiable on seismic data and providing insights into specific geological features and depositional environments [38,39]. These distinct patterns that differentiate a group of seismic reflections from others are recognizable by analyzing seismic reflection amplitude, continuity, configuration, frequency, and geometry [40].
Geoscientists have effectively used seismic facies analysis to interpret geologic depositional environments, predict lithology, and describe stratigraphic features from seismic data [41,42]. This framework enables consistent interpretation of subsurface geology from seismic data, improving reservoir characterization. Applying machine learning for seismic facies classification for this research is essential due to the sparsity of direct subsurface measurements from petrophysical well logs and wellbore cores.
The observable seismic facies within the area of interest were classified into (1) a central salt diapir and a salt canopy on the northeastern portion in the shallow section of the seismic volume (less than 400 m); (2) gas hydrates in the Pliocene to Holocene section; (3) conformable sediments in shallow and deeper sections (over 500 m); and (4) geologic faults (Figure 2 and Figure 3). A salt diapir and canopy were identified and were discernable by their chaotic and low-amplitude seismic reflectors, while the conformable sediments were continuous and sub-parallel reflectors [39]. The gas hydrates were shallow, high-amplitude reflectors seen in the shallow section of the seismic volume [32,33]. The seismic data on the vertical margins and below the salt diapir and canopy were noisy due to their chaotic seismic reflectors, which are difficult to visualize, while the top and base of the salt canopy are easily identified by their strong reflectors.

4.2. Seismic Data Quality Check

Before seismic interpretation, we carried out a data quality check to map out areas with the least interpretative confidence. Poor data quality may arise from acquisition artifacts, seismic diffraction due to the presence of salt structures, and other sources of noise. This data quality check was achieved by calculating the Normalized Root Mean Square (NRMS) [43]. This method works by subtracting structurally smoothed seismic amplitude from the original seismic amplitude with no smoothing applied and classifying areas with high NRMS values as poor data quality and lower values as areas with the least noise. From equation 1, the base is the original dataset, and the monitor is the structurally smooth dataset. Noise may appear as good data, especially when it is coherent.
N R M S = 200 × R M S ( B a s e − M o n i t o r ) R M S M o n i t o r + R M S ( B a s e )
NRMS = Normalized Root Mean Square; Base = Original Dataset; Monitor = Structurally Smooth Dataset.
As observed in Figure 5, the seismic data quality is poor on the vertical margins and below the salt canopy and diapir (areas in red in the NRMS section). Areas in yellow have moderate noise, while green denotes cleaner data. Seismic amplitudes, NRMS, and the extracted dominant frequency show that below the salt canopy on the right, the seismic resolution is poor. Furthermore, seismic resolution decreases as depth increases due to seismic energy attenuation. It is important to flag these areas with poor data quality due to their impact on results from using machine learning algorithms.

4.3. Seismic Attribute Selection

Computing specific attributes of the seismic signal provides valuable information about geological structures in the subsurface, stratigraphy, and rock properties [8,44]. In this research, we specifically selected aberrancy, similarity, spectral frequencies, instantaneous phase, sweetness, curvature, and envelope to delineate geologic faults, conformable reservoirs, and gas hydrates, as well as a salt diapir and canopy. These seismic attributes can be classified broadly into the following types: geometric, instantaneous-based, amplitude-based, and spectral decomposition frequencies [8,45] (Table 1).
Numerous studies have successfully employed these seismic attributes to identify seismic facies, sub-seismic faults, and other stratigraphic/architectural elements of different kinds of reservoirs, including those in deepwater [12,13,47,49,50,66,67,68,69]. Numerous seismic attributes can be helpful in multi-attribute analysis, but the interpreter is tasked with deciding which combination of attributes is best suited for the project.

4.4. Self-Organizing Map Algorithm

Due to the challenge of analyzing numerous attributes simultaneously, machine learning (ML) algorithms can be incorporated into a seismic multi-attribute analysis workflow [11,12,13]. This research explores using Self-Organizing Maps (SOM) to improve seismic interpretation (Figure 6).
Self-organizing maps (SOM) is an unsupervised machine learning clustering algorithm developed by Kohonen in 2001 for identifying clusters [15,16,71,72]. SOM uses an artificial network architecture that applies weights, vectors, and biases to train input data, such as seismic attributes, and then create clusters by selecting the neuron that has the nearest Euclidean distance to the input sample, thereby creating a geometric structure [15,16,48,71,73,74,75,76,77].
The outcome of the SOM is a topological representation of the input data, with every sample being mapped to a particular node where neighboring nodes with similar patterns are clustered together [78,79]. This allows the high-dimensional multi-attribute seismic patterns to be visualized and interpreted as distinct seismic facies corresponding to different reservoir properties or geological features of interest [20]. SOM has been successfully applied to enhancing seismic resolution, seismic facies analysis, and multi-attribute analysis by numerous geoscience researchers [12,13,48,67,71,72,80,81]. La Marca and Bedle [12,13] applied SOM to identify deepwater channel complexes in Taranaki Basin, New Zealand, using geometric and amplitude-based seismic attributes as input data. La Marca and Bedle [12,13] observed that to effectively apply SOM to these seismic attributes, the interpreter must avoid duplicating attributes with high correlation coefficients to avoid redundancy. This redundancy could create a bias toward a particular attribute or geologic feature over another and skew the seismic interpretation by favoring one geologic feature more than another [12,13].

5. Results

Prior to utilizing seismic attributes to characterize seismic facies, we scanned the seismic volume to detect notable features such as salt canopy and diapir, gas hydrates, and faults by examining the seismic amplitudes within the studied area. The salt canopy is more defined than the salt diapir by an inward chaotic seismic response bordered by high amplitudes.
Generally, seismic reflectors were mostly more conformable in the shallower section (above 2500 ms) than in the deeper sections (below 2500 ms) due to seismic energy attenuation and diffraction from increasing depth and the presence of the salt canopy (Figure 3). The sediments on the vertical margins of the salt diapir and below the salt canopy were difficult to image due to the seismic diffraction and seismic resolution (Figure 3). Also, the sediments above the salt canopy and diapir are faulted, as seen by the disruption of continuous seismic reflectors (Figure 3).
Despite our ability to identify some of these seismic facies by merely scanning through the seismic volume, seismic attributes aid in accentuating and validating these unique geologic features, which are likely a result of stratigraphic or structural processes.
Prior to performing the SOM, each of these volumetric attributes was extracted on seismic horizons and closely examined to detect unique structural and stratigraphic characteristics (Figure 7). Geometric attributes like similarity and curvature helped identify salt structures and radial faults. Amplitude-based attributes like sweetness and envelope helped identify sand-rich bodies (validated using the ConocoPhillips well log information shown in [36]), gas hydrates, and the lateral variation in lithologic properties such as porosity. Areas with high sweetness and envelope attribute values potentially have preferable reservoir properties. The instantaneous and spectral frequency attributes also provided inputs for the SOM models in understanding changes in reservoir thickness. Spectral frequencies ranging from 10 to 60 Hz were obtained since these were the most dominant frequencies in the spectrum.
Examining numerous seismic attributes simultaneously is task-heavy and could be time-inefficient when assessing large datasets. Hence, unsupervised learning algorithms, such as SOM, have been known to successfully combine interesting geologic features within attribute maps to create an output map that incorporates all these elements.
Table 2 presents the different seismic attribute combinations used in our SOM analysis. In each model, specific seismic attributes were selected due to their ability to distinguish specific geological features. Model 1 combined geometric, spectral frequency, and amplitude-based attributes to highlight structural changes, thickness variations, sand bodies, and gas presence. We tested a combination of spectral frequencies and geometric attributes to detect structural and thickness changes in Model 2.
Model 3 modified Model 2 by substituting the spectral frequencies with an instantaneous attribute. Finally, SOM Model 4 introduced a textural attribute, GLCM Entropy, while omitting geometric attributes. We tested these combinations to identify which attribute set best achieved our objectives.
We analyzed the SOM output from all four models by comparing the clustered features. This comparison was performed by visually inspecting seismic sections and extracting the output along a selected seismic horizon. The covariance matrix of each of these attributes was calculated to provide insight into how related these attributes are and to avoid redundancy (Table 3). By examining their covariance, we removed redundancy by avoiding a combination of attributes with a relationship above 60% (0.6).
In SOM Models 1, 2, and 3, the salt canopy and diapir are represented mainly by the blue color, seen as a prominent feature in the central and northeastern portion of the generated SOM map (Figure 8A–C).
Furthermore, seismic noise or areas with artifacts also appeared blue but were less conformable and more chaotic than the salt canopy and diapir. The yellow colors represent thick sand bodies, and those with a red color within them show the presence of gas, as seen with many gas hydrates in the shallow section, while the green colors represent shales. In SOM Model 4, we see a reversal in the colors of the sediments, thick sediments, salt canopy and diapir, artifacts, and noise (Figure 8D). Here, the salt canopy and diapir are red and yellow, while the sands which are not easily resolved, are blue, and the shales are green.
Models 3 and 4 image the geometry of the central salt diapir much better than models 1 and 2 due to the different categories of seismic attributes selected. Additionally, faults are more evident in models 1 and 2 (Figure 8A,B). Models 1 and 2 were more useful in delineating the sand bodies, gas hydrates, shales, and areas with noisy data/artifacts (Figure 8A,B). Figure 9A–E compares cross-sections of an interpreted seismic with the SOM models, which generally agree with our results.
SOM Models 1 and 2 were preferable because they resolved most of the geologic features of interest. SOM Model 3 was the least preferred because of its inability to image the succession of sediments, and Model 4 produced quite reasonable results compared to Model 3.
Once these results were examined, well control was used to investigate the validity of these results. We validated the model’s accuracy by comparing sand bodies interpreted in the well logs from the ConocoPhillips well to the SOM cross-sections on some of the target horizons. Gamma ray signatures matched the seismic facies defined by the SOM models. Models 1 and 2 correlate much better with the SOM facies classification than Models 3 and 4 (Figure 10). Some discrepancies in the well and SOM model match stem from the substantial difference in vertical resolution between well logs (centimeters to meters) and seismic data (tens of meters or more).
Multiple thin reservoir sand layers visible in the higher-resolution well logs appear to be merged into single reservoir units in the seismic clustering results. Hence, upscaling can be challenging, resulting in the loss of minute details like discrete thin reservoirs (as seen at the bottom of the logs). Using these SOM model results, other locations assigned the same facies classifications as these sands can be considered potential reservoirs of interest, especially on the other side of the salt diapir where there is limited well control (Figure 9B,C). SOM was useful for discriminating between the salt canopy, noise sub-salt, and radial faults in many locations (as seen in Figure 11).

6. Discussion

The study sought to identify suitable sites for storing CO2 and hydrogen, while also mapping structural features that could affect these storage reservoirs. A combination of seismic attributes was extracted on a seismic horizon to aid in interpreting structural and stratigraphic changes in Mississippi Canyon Block 118, Gulf of Mexico. The application of machine learning algorithms is time-effective and could go beyond the interpreter’s abilities by detecting geological reservoirs where sparse direct subsurface information like well logs are available. This study was performed to maximize the available dataset, which mainly comprises 3-D time-migrated seismic data and sparse well log information. The seismic attributes selected for this research were carefully chosen by reviewing literature where their successful use was achieved [12,13,47,49,50,66,67,68,69] (Table 1). These attributes include geometric attributes like Curvature, Similarity, and Aberrancy which are essential faulting and folding indicators. Amplitude-based attributes like Envelope and Sweetness help detect changes in lithology. Instantaneous attributes such as instantaneous frequency, instantaneous amplitude, and spectral decomposition frequencies were used for detecting changes in bed thickness and lithology. Textural attributes like GLCM Entropy are useful for seismic facies classification and fault detection.
The Machine Learning (ML) workflow implemented in this study has effectively detected data correlations that are difficult for the human eye to observe, such as potential storage reservoirs in the deeper southwest portion of the seismic section lacking well control (approximately 7 km). Using an unsupervised ML approach like SOM streamlines seismic interpretation by combining the strengths of multiple seismic attributes simultaneously. Nonetheless, due to a lack of well log information in this location, further data acquisition would be required to ascertain the validity of the SOM models in this location. This method produces more robust interpretations and reduces the time needed to analyze individual attributes separately. Likewise, these ML algorithms could detect geologic features that the interpreter may ignore or not observe during initial interpretation.
ML aids in detecting and grouping clusters in seismic attributes without predefined labels based on amplitude, frequency, and geometric patterns, thereby providing consistent classification across large 3D seismic volumes [8,10,12,14]. Despite the numerous advantages of Self-Organizing Maps (SOMs) for seismic facies classification, some of the limitations may include data quality dependency, dimensionality, lack of supervision, the ambiguity of interpretation, complex attribute relationships, and difficulty in estimating the algorithm’s accuracy [80]. Hence, the interpreter’s experience, knowledge of the pitfalls of unsupervised algorithms, and careful selection of seismic attributes would help improve its usefulness [45]. Furthermore, this automated approach faces other limitations: (1) the algorithms may not correspond to meaningful geological units; results depend heavily on data quality; validation is challenging due to limited well control; and the methods may miss subtle features crucial for interpretation [46]. The key lies in balancing automated classification with expert geological interpretation to ensure meaningful results. We applied this workflow across the complete seismic volume to achieve volumetric classification. However, processing such a large dataset presents two key challenges: reduced computational efficiency and increased risk of misclassification, as noted by [12,13].
In the case of supervised machine learning, substantial labeled data is required for supervised learning, which is often scarce in subsurface interpretation, especially with the well log data available in this research [62]. Well logs provide ground truth but only at discrete locations and well information was sparse. The training data on just well logs may introduce subjectivity to classification due to the assumption that the facies classified from the well log are sufficient to interpret the entire seismic. Hence, more well logs and cores would be more useful for training. Additionally, models trained on one geological setting may perform poorly when applied to different basins with unique geological characteristics, so this method provides a method that avoids the subjectivity of the interpreter and the location for training. Also, seismic data inherently contain ambiguity, which deterministic classifiers struggle to represent properly. Certain facies types may be underrepresented in training data, leading to biased classifiers. Hence, we have applied unsupervised machine learning, which reduces this bias while comparing to the well log available for validation. Classification results are also highly sensitive to seismic processing workflows, potentially amplifying artifacts [47]. The training of sophisticated models would require significant computational resources, especially for 3D seismic volumes, which were unavailable during this research.
We identified each seismic facies with a specific color for each SOM model. In SOM Models 1, 2, and 3, yellow and red colors were used to display potential CO2 and hydrogen storage reservoirs, the areas in blue correspond to the salt canopy and diapir, noise, or faults, and the green colors represent shales. The reverse is the case for SOM Model 4, where red denotes the salt canopy and diapir, noise, or faults. Models 3 and 4 were not suitable for identifying the thick reservoir sands or gas hydrates but were very useful in determining the outline of the salt canopy and diapir. On the top left corner of the horizon slices, Models 3 and 4 misclassify noise as gas hydrates. It is noteworthy to mention that the SOM models were not the most efficient in delineating salt boundaries, which would significantly influence the estimation of the lateral extent of promising reservoirs.
Models 1 and 2 were great for classifying thicker sands, shales, gas hydrates, and faults, as well as the salt canopy and diapir. SOM Models 1 and 2 were considered preferable because they best differentiated structural and stratigraphic changes with each seismic facies represented. Consequently, SOM Models 3 and 4 were decisive for delineating structural changes and contained a few misclassifications, which may be potentially due to computational errors and seismic data artifacts that need to be further investigated. From these results, we advise using a spectral frequency (preferably Peak Frequency or Peak Magnitude) alongside geometric and amplitude-based attributes to adequately capture all the target facies in one model. The 40 Hz Spectral Frequency extraction contained many artifacts, but Peak Spectral Magnitude and Peak Spectral Frequency were more effective in understanding thickness variations in the sediments. Aberrancy, Similarity, and Curvature were proven to be suitable geometric attributes for mapping faults and the salt canopy and diapir, while Sweetness or Envelope also produced great results in detecting changes in lithology. Seismic attribute selection and testing a combination of these attributes, both subjective processes, may vary depending on the architecture or geology of the studied area.
For future work, the various 3D facies bodies extracted as potential sands can be assessed as potential storage sites for CO2 by estimating their volumetric capacities for carbon storage using established methodologies. Additionally, by delineating the location of faults, gas hydrates, and salt, adequate plans can be made on the feasibility of drilling new gas storage wells in this location and the drilling hazards or risks associated with dealing with such geologic features. Some of these drilling hazards could include overpressured zones around salt or in gas-filled sediments, carapace on the salt, wellbore and formation instability due to faulting and hydrate dissociation, creeps, gas kicks, depth uncertainty due to poor data quality sub-salt and faulting, etc.

7. Conclusions

The goal of this study is to show how the integration of seismic attributes with ML can be incorporated into a seismic interpretation workflow to characterize potential storage reservoirs for hydrogen or CO2 storage in MC-118. Geometric, amplitude-based spectral, textural, and instantaneous seismic attributes were combined and incorporated into an unsupervised machine learning algorithm, Self-Organizing Map (SOM), to classify seismic facies and delineate faulting influenced by salt tectonics and slope failure. These techniques were applied to Miocene sediments in Mississippi Canyon Block 118, Gulf of Mexico.
Based on our results, we suggest that the right combination of seismic attributes should be selected based on how they align with the geological objectives (seismic facies classification, fault identification, etc.) for optimal performance of classification algorithms. These attributes should incorporate geometric attributes (Coherence, Curvature, and Aberrancy) that define seismic geomorphology; amplitude-based attributes (RMS, Sweetness, and Envelope) useful in identifying changes in reservoir properties; textural attributes such as GLCM entropy to differentiate facies; and instantaneous attributes to identify subtle changes in the seismic traces. Our research demonstrated that combining seismic attributes using unsupervised machine learning algorithms helped identify potential storage reservoirs in Mississippi Canyon 118. This workflow can be adopted and improved upon to maximize sparse datasets in similar geologic settings.

Author Contributions

Conceptualization, S.A.S., C.C.K. and J.H.K.; methodology, S.A.S., C.C.K. and J.H.K.; software, S.A.S.; validation, S.A.S.; formal analysis, C.C.K. and J.H.K.; investigation, S.A.S.; resources, C.C.K. and J.H.K.; data curation, S.A.S.; writing—original draft preparation, S.A.S.; writing—review and editing, S.A.S. and C.C.K.; visualization, S.A.S.; supervision, C.C.K. and J.H.K.; project administration, C.C.K. and J.H.K.; funding acquisition, C.C.K. and J.H.K. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the US Department of Energy and the Southeast Regional Carbon Storage Partnership: Offshore Gulf of Mexico led by the Southern States Energy Board (DE-FE0031557).

Data Availability Statement

The seismic and well log data in this study are owned by private entities and can be accessed by obtaining the necessary licenses from the data owners.

Acknowledgments

The authors would like to express their gratitude to TGS and ConocoPhillips for providing the seismic and petrophysical well log data used in this study. They are also thankful to SLB and the AASPI Consortium for granting access to the Petrel and AASPI software licenses, which were instrumental in data analysis and visualization. Finally, the authors acknowledge the support for this project, provided by the US Department of Energy and the Southeast Regional Carbon Storage Partnership: Offshore Gulf of Mexico led by the Southern States Energy Board (DE-FE0031557).

Conflicts of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Abbreviations

The following abbreviations are used in this manuscript:
MLMachine Learning
SOMSelf-Organizing Maps
MC-118Mississippi Canyon, Block 118

References

  1. Sambo, C.; Dudun, A.; Samuel, S.A.; Esenenjor, P.; Muhammed, N.S.; Haq, B. A review on worldwide underground hydrogen storage operating and potential fields. Int. J. Hydrogen Energy 2022, 47, 22840–22880. [Google Scholar] [CrossRef]
  2. Askarova, A.; Mukhametdinova, A.; Markovic, S.; Khayrullina, G.; Afanasev, P.; Popov, E.; Mukhina, E. An overview of geological CO2 sequestration in oil and gas reservoirs. Energies 2023, 16, 2821. [Google Scholar] [CrossRef]
  3. Osahon, U.; Efetobore, M.O.G. Reservoir Characterization: Enhancing Accuracy through Advanced Rock Physics Techniques. J. Geosci. 2023, 11, 67–78. [Google Scholar] [CrossRef]
  4. Vodopić, F.; Vulin, D.; Karasalihović Sedlar, D.; Jukić, L. Enhancing Carbon Capture and Storage Deployment in the EU: A Sectoral Analysis of a Ton-Based Incentive Strategy. Sustainability 2023, 15, 15717. [Google Scholar] [CrossRef]
  5. Jia, A.; He, D.; Jia, C. Advances and challenges of reservoir characterization: A review of the current state-of-the-art. In Earth Sciences; BoD - Books on Demand GmbH: Norderstedt, Germany, 2012; Volume 1. [Google Scholar]
  6. Samuel, S.A.; Zhang, R. Total Organic Carbon Content Estimation of Bakken Formation, Kevin-Sunburst Dome, Montana using Post-Stack Inversion, Passey (DLogR) Method and Multi-Attribute Analysis. In Proceedings of the SPE/AAPG/SEG Unconventional Resources Technology Conference, Houston, TX, USA, 26–28 July 2021; URTEC: Tulsa, OK, USA, 2021; p. D021S047R003. [Google Scholar]
  7. Hampson, D.P.; Schuelke, J.S.; Quirein, J.A. Use of multiattribute transforms to predict log properties from seismic data. Geophysics 2001, 66, 220–236. [Google Scholar] [CrossRef]
  8. Satinder, C.; Marfurt, K.J. Seismic attributes—A historical perspective. Geophysics 2005, 70, 3SO–28SO. [Google Scholar]
  9. Oumarou, S.; Mabrouk, D.; Tabod, T.C.; Marcel, J.; Ngos, S., III; Essi, J.M.A.; Kamguia, J. Seismic attributes in reservoir characterization: An overview. Arab. J. Geosci. 2021, 14, 402. [Google Scholar] [CrossRef]
  10. Roden, R.; Smith, T.; Sacrey, D. Geologic pattern recognition from seismic attributes: Principal component analysis and self-organizing maps. Interpretation 2015, 3, SAE59–SAE83. [Google Scholar] [CrossRef]
  11. La Marca, K. Seismic Attribute Optimization with Unsupervised Machine Learning Techniques for Deepwater Seismic Facies Interpretation: Users vs Machines (2020). Available online: https://shareok.org/items/2de83777-843d-4ea0-a0c7-332351c7fc0a (accessed on 10 May 2025).
  12. La Marca, K.; Bedle, H. Deepwater seismic facies and architectural element interpretation aided with unsupervised machine learning techniques: Taranaki basin, New Zealand. Mar. Pet. Geol. 2022, 136, 105427. [Google Scholar] [CrossRef]
  13. La Marca, K.; Bedle, H. User vs. machine-based seismic attribute selection for unsupervised machine learning techniques: Does human insight provide better results than statistically chosen attributes? In Advances in Subsurface Data Analytics; Elsevier: Amsterdam, The Netherlands, 2022; pp. 3–30. [Google Scholar]
  14. Zhao, T.; Jayaram, V.; Roy, A.; Marfurt, K.J. A comparison of classification techniques for seismic facies recognition. Interpretation 2015, 3, SAE29–SAE58. [Google Scholar] [CrossRef]
  15. Kohonen, T. Self-organized formation of topologically correct feature maps. Biol. Biol. Cybern. 1982, 43, 59–69. [Google Scholar] [CrossRef]
  16. Kohonen, T.; Kohonen, T. The basic SOM. In Self-Organizing Maps; Springer: Berlin, Heidelberg, 2001; pp. 105–176. [Google Scholar]
  17. Rauber, A.; Merkl, D.; Dittenbach, M. The growing hierarchical self-organizing map: Exploratory analysis of high-dimensional data. IEEE Trans. Neural Netw. 2002, 13, 1331–1341. [Google Scholar] [CrossRef]
  18. Saraswat, P.; Sen, M.K. Artificial immune-based self-organizing maps for seismic-facies analysis. Geophysics 2012, 77, O45–O53. [Google Scholar] [CrossRef]
  19. Astudillo, C.A.; Oommen, B.J. Topology-oriented self-organizing maps: A survey. Pattern Anal. Appl. 2014, 17, 223–248. [Google Scholar] [CrossRef]
  20. Zhao, T.; Li, F.; Marfurt, K.J. Constraining self-organizing map facies analysis with stratigraphy: An approach to increase the credibility in automatic seismic facies classification. Interpretation 2017, 5, T163–T171. [Google Scholar] [CrossRef]
  21. Christianson, P.A.; Murray, P.E. Investigating active deep-water sedimentation near a gas hydrate system with high-resolution seismic data. In Proceedings of the 2007 SEG Annual Meeting, San Antonio, TX, USA, 23–28 September 2007. [Google Scholar]
  22. McGee, T.; Macelloni, L.; Lutken, C.; Bosman, A.; Brunner, C.; Rogers, R.; Woolsey, J.R. Hydrocarbon gas hydrates in sediments of the Mississippi Canyon area, Northern Gulf of Mexico. Geol. Soc. Lond. Spec. Publ. 2009, 319, 29–49. [Google Scholar] [CrossRef]
  23. Macelloni, L.; Simonetti, A.; Knapp, J.H.; Knapp, C.C.; Lutken, C.B.; Lapham, L.L. Multiple resolution seismic imaging of a shallow hydrocarbon plumbing system, Woolsey Mound, Northern Gulf of Mexico. Mar. Pet. Geol. 2012, 38, 128–142. [Google Scholar] [CrossRef]
  24. Ingram, W.C.; Meyers, S.R.; Martens, C.S. Chemostratigraphy of deep-sea Quaternary sediments along the Northern Gulf of Mexico Slope: Quantifying the source and burial of sediments and organic carbon at Mississippi Canyon 118. Mar. Pet. Geol. 2013, 46, 190–200. [Google Scholar] [CrossRef]
  25. Pizzi, M.; Macelloni, L.; Lutken, C.B.; D’Emidio, M. Temporal Evolution of MC118 Woolsey Mound Seep Activity: Constraints from Analysis of Small-Scale Salt-Induced Sediment Deformation 2012. Available online: https://www.researchgate.net/publication/295920286_Temporal_Evolution_of_MC118_Woolsey_Mound_seep_activity_constraints_from_analysis_of_small-scale_salt-induced_sediment_deformation (accessed on 10 May 2025).
  26. Salazar, J.A.; Knapp, J.H.; Knapp, C.C.; Pyles, D.R. Salt tectonics and Pliocene stratigraphic framework at MC-118, Gulf of Mexico: An integrated approach with application to deep-water confined structures in salt basins. Mar. Pet. Geol. 2014, 50, 51–67. [Google Scholar] [CrossRef]
  27. Offshore Engineer. ConocoPhillips Misses in GoM. 2005. Available online: https://www.oedigital.com/news/452843-conocophillips-misses-in-gom (accessed on 10 February 2025).
  28. Simonetti, A.; Knapp, J.H.; Sleeper, K.; Lutken, C.B.; Macelloni, L.; Knapp, C.C. Spatial distribution of gas hydrates from high-resolution seismic and core data, Woolsey Mound, Northern Gulf of Mexico. Mar. Pet. Geol. 2013, 44, 21–33. [Google Scholar] [CrossRef]
  29. Fisher, C.; Roberts, H.; Cordes, E.; Bernard, B. Cold seeps and associated communities of the Gulf of Mexico. Oceanography 2007, 20, 118–129. [Google Scholar] [CrossRef]
  30. Bouma, A.H.; Bryant, W.R. Physiographic features on the northern Gulf of Mexico continental slope. Geo-Mar. Lett. 1994, 14, 252–263. [Google Scholar] [CrossRef]
  31. Sleeper, K.; Lowrie, A.; Bosman, A.; Macelloni, L.; Swann, C.T. Bathymetric mapping and high-resolution seismic profiling by AUV in MC 118 (Gulf of Mexico). In Proceedings of the Offshore Technology Conference, Houston, TX, USA, 1–4 May 2006; p. OTC-18113. [Google Scholar]
  32. Alam, S.; Knapp, C.C.; Knapp, J.H.; Simonetti, A. Temporal and spatial characterization of a thermogenic, fault-controlled gas hydrate system, Woolsey Mound, Gulf of Mexico. Interpretation 2024, 12, T221–T232. [Google Scholar] [CrossRef]
  33. Alam, S.; Knapp, C.; Knapp, J. Three-Dimensional Amplitude versus Offset Analysis for Gas Hydrate Identification at Woolsey Mound: Gulf of Mexico. GeoHazards 2024, 5, 271–285. [Google Scholar] [CrossRef]
  34. Biondi, B.L. 3D Seismic Imaging; Society of Exploration Geophysicists: Houston, TX, USA, 2006. [Google Scholar]
  35. SEG Wiki. Gulf of Mexico 2019. Available online: https://wiki.seg.org/wiki/Gulf_of_Mexico (accessed on 1 November 2023).
  36. Samuel, S.A.; Knapp, C.C.; Knapp, J.H. Volumetric carbon storage capacity estimation at Mississippi Canyon Block 118 in the Gulf of Mexico using Post-Stack Seismic Inversion. Int. J. Greenh. Gas Control 2025, 141, 104319. [Google Scholar] [CrossRef]
  37. Cilenti, F.; Oppo, D.; Macelloni, L. Controls on fluid discharge at cold seep-hydrate systems: 4D seismic monitoring of Woolsey Mound, Gulf of Mexico. Earth Planet. Sci. Lett. 2024, 648, 119087. [Google Scholar] [CrossRef]
  38. Mitchum, R.M., Jr.; Vail, P.R.; Sangree, J.B. Seismic stratigraphy and global changes of sea level: Part 6. Stratigraphic interpretation of seismic reflection patterns in depositional sequences: Section 2. Application of seismic reflection configuration to stratigraphic interpretation. AAPG Mem. 1977, 26, 117–133. [Google Scholar]
  39. Salazar Florez, D.; Bedle, H. Study on the parameterization response of probabilistic neural networks for seismic facies classification in the Gulf of Mexico. Interpretation 2022, 10, T1–T23. [Google Scholar] [CrossRef]
  40. Sangree, J.B.; Widmier, J.M. Interpretation of depositional facies from seismic data. Geophysics 1979, 44, 131–160. [Google Scholar] [CrossRef]
  41. Roksandić, M.M. Seismic facies analysis concepts. Geophys. Prospect. 1978, 26, 383–398. [Google Scholar] [CrossRef]
  42. Brown, A.R. Interpretation of Three-Dimensional Seismic Data; Society of Exploration Geophysicists and American Association of Petroleum Geologists: Houston, TX, USA, 2011. [Google Scholar]
  43. Umoren, E.B.; George, N.J. Time lapse (4D) and AVO analysis: A case study of Gullfaks field, Northern North Sea. NRIAG J. Astron. Geophys. 2018, 7, 62–77. [Google Scholar] [CrossRef]
  44. Hart, B.S. Definition of subsurface stratigraphy, structure and rock properties from 3-D seismic data. Earth-Sci. Rev. 1999, 47, 189–218. [Google Scholar] [CrossRef]
  45. Taner, M.T.; Schuelke, J.S.; O’Doherty, R.; Baysal, E. Seismic attributes revisited. In SEG Technical Program Expanded Abstracts; Society of Exploration Geophysicist: Houston, TX, USA, 1994; pp. 1104–1106. [Google Scholar]
  46. Barnes, A.E. A tutorial on complex seismic trace analysis. Geophysics 2007, 72, W33–W43. [Google Scholar] [CrossRef]
  47. Chopra, S.; Marfurt, K.J. Seismic Attributes for Prospect Identification and Reservoir Characterization; Society of Exploration Geophysicists and European Association of Geoscientists and Engineers: Houston, TX, USA, 2007. [Google Scholar]
  48. SEG Wiki. Self-Organizing Map and Multi-Attribute Analysis 2018. Available online: https://wiki.seg.org/wiki/Self_Organizing_Map_and_Multi-attribute_Analysis (accessed on 1 November 2023).
  49. Chopra, S.; Marfurt, K. Seismic Curvature Attributes for Mapping Faults/Fractures, and Other Stratigraphic (2007b). Available online: https://csegrecorder.com/articles/view/seismic-curvature-attributes-for-mapping-faults-fractures-and-other (accessed on 10 May 2025).
  50. Chopra, S.; Marfurt, K. Curvature attribute applications to 3D surface seismic data. Lead. Edge 2007, 26, 404–414. [Google Scholar] [CrossRef]
  51. Subrahmanyam, D.; Rao, P.H. Seismic Attributes-A Review. In Proceedings of the 7th International Conference and Exposition on Petroleum Geophysics, Hyderabad, India, January 2008; Available online: https://spgindia.org/2008/398.pdf (accessed on 10 May 2025).
  52. Gao, D. Integrating 3D seismic curvature and curvature gradient attributes for fracture characterization: Methodologies and interpretational implications. Geophysics 2013, 78, O21–O31. [Google Scholar] [CrossRef]
  53. Hutchinson, B. Application and Limitations of Seismic Attributes on 2D Reconnaissance Surveys 2016. Available online: https://mcee.ou.edu/aaspi/AASPI_Theses/2016_AASPI_Theses/2016_Hutchinson_Bryce_Thesis.pdf (accessed on 10 May 2025).
  54. Qi, X.; Marfurt, K. Volumetric aberrancy to map subtle faults and flexures. Interpretation 2018, 6, T349–T365. [Google Scholar] [CrossRef]
  55. Bhattacharya, S.; Verma, S. Application of volumetric seismic attributes for complex fault network characterization on the North Slope, Alaska. J. Nat. Gas Sci. Eng. 2019, 65, 56–67. [Google Scholar] [CrossRef]
  56. Verma, S.; Bhattacharya, S. 3D seismic attribute visualization and analysis for fault characterization. In Developments in Structural Geology and Tectonics; Elsevier: Amsterdam, The Netherlands, 2024; Volume 6, pp. 37–55. [Google Scholar]
  57. James, G.; Partyka, G. Processing and interpretational aspects of spectral decomposition. SEG Tech. Program Expand. Abstr. 1997, 1997, 1055–1058. [Google Scholar]
  58. Partyka, G.; Gridley, J.; Lopez, J. Interpretational applications of spectral decomposition in reservoir characterization. Lead. Edge 1999, 18, 353–360. [Google Scholar] [CrossRef]
  59. Partyka, G. The birth of spectral decomposition. Lead. Edge 2007, 26, 1624. [Google Scholar] [CrossRef]
  60. Partyka, G.A.; Bush, M.D.; Garossino, P.G.; Gutowski, P.R. Spectral Decomposition. In Interpretation of Three-Dimensional Seismic Data; American Association of Petroleum Geologists: Tulsa, OK, USA, 2011; pp. 295–308. [Google Scholar]
  61. White Roy, E. Properties of instantaneous seismic attributes. Lead. Edge 1991, 10, 26–32. [Google Scholar] [CrossRef]
  62. Chopra, S.; Marfurt, K.J. Emerging and future trends in seismic attributes. Lead. Edge 2008, 27, 298–318. [Google Scholar] [CrossRef]
  63. Laughlin, K.; Garossino, P.; Partyka, G. GCSpectral Decomposition for Seismic Stratigraphic Patterns. 2003. Available online: https://www.searchanddiscovery.com/pdfz/documents/geophysical/2003/laughlin/images/laughlin.pdf.html (accessed on 10 May 2025).
  64. Naseer, M.T.; Asim, S. Application of instantaneous spectral analysis and acoustic impedance wedge modeling for imaging the thin beds and fluids of fluvial sand systems of Indus Basin, Pakistan. J. Earth Syst. Sci. 2018, 127, 97. [Google Scholar] [CrossRef]
  65. Nanda, N.C. Analysing seismic attributes. In Seismic Data Interpretation and Evaluation for Hydrocarbon Exploration and Production: A Practitioner’s Guide; Springer International Publishing: Cham, Switzerland, 2021; pp. 205–222. [Google Scholar]
  66. Silver, C.; La Marca-Molina, K.; Bedle, H. Seismic geomorphology of deep-water channel systems in the southern Taranaki Basin. In SEG Technical Program Expanded Abstracts; Society of Exploration Geophysicists: Houston, TX, USA, 2019; pp. 2018–2022. [Google Scholar]
  67. La Marca-Molina, K.; Silver, C.; Bedle, H.; Slatt, R. Seismic facies identification in a deepwater channel complex applying seismic attributes and unsupervised machine learning techniques. A case study in the Taranaki Basin, New Zealand. In SEG Technical Program Expanded Abstracts; Society of Exploration Geophysicists: Houston, TX, USA, 2019; pp. 2059–2063. [Google Scholar]
  68. Verma, S.; Scipione, M. The early Paleozoic structures and its influence on the Permian strata, Midland Basin: Insights from multi-attribute seismic analysis. J. Nat. Gas Sci. Eng. 2020, 82, 103521. [Google Scholar] [CrossRef]
  69. Olaleye, O.K.; Enikanselu, P.A.; Ayuk, M.A. Characterization of reservoir sands using 3D seismic attributes in the coastal swamp area of Niger Delta Basin. J. Pet. Explor. Prod. Technol. 2021, 11, 3995–4004. [Google Scholar] [CrossRef]
  70. SuperDataScience Team. Self-Organizing Maps (SOM’s)—How Do Self-Organizing Maps Work? 2018. Available online: https://www.superdatascience.com/blogs/self-organizing-maps-soms-how-do-self-organizing-maps-work (accessed on 27 October 2023).
  71. Matos, M.C.D.; Marfurt, K.J.; Johann, P.R. Seismic interpretation of self-organizing maps using 2D color displays. Rev. Bras. Geof. 2010, 28, 631–642. [Google Scholar] [CrossRef]
  72. Smith, T.; Treitel, S. Introduction to Self-Organizing Maps in Multi-Attribute Seismic Data; Geophysical Society of Houston: Houston, TX, USA, 2011. [Google Scholar]
  73. McMullen Patrick, R. A Kohonen self-organizing map approach to addressing a multiple objective, mixed-model JIT sequencing problem. Int. J. Prod. Econ. 2001, 72, 59–71. [Google Scholar] [CrossRef]
  74. Liu, Y.; Weisberg, R.H.; Mooers, C.N. Performance evaluation of the self-organizing map for feature extraction. J. Geophys. Res. Ocean. 2006, 111. [Google Scholar] [CrossRef]
  75. Tasdemir, K.; Merényi, E. Exploiting data topology in visualization and clustering of self-organizing maps. IEEE Trans. Neural Netw. 2009, 20, 549–562. [Google Scholar] [CrossRef] [PubMed]
  76. Singh, N.; Javeed, A.; Chhabra, S.; Kumar, P. Missing value imputation with unsupervised kohonen self organizing map. In Emerging Research in Computing, Information, Communication and Applications: ERCICA 2015; Springer: New Delhi, India, 2015; Volume 1, pp. 61–76. [Google Scholar]
  77. Gu, F.; Cheung, Y.M. Self-organizing map-based weight design for decomposition-based many-objective evolutionary algorithm. IEEE Trans. Evol. Comput. 2017, 22, 211–225. [Google Scholar] [CrossRef]
  78. Vesanto, J.; Alhoniemi, E. Clustering of the self-organizing map. IEEE Trans. Neural Netw. 2000, 11, 586–600. [Google Scholar] [CrossRef]
  79. Kohonen, T. Essentials of the self-organizing map. Neural Netw. 2013, 37, 52–65. [Google Scholar] [CrossRef]
  80. Köhler, A.; Ohrnberger, M.; Scherbaum, F. Unsupervised pattern recognition in continuous seismic wavefield records using self-organizing maps. Geophys. J. Int. 2010, 182, 1619–1630. [Google Scholar] [CrossRef]
  81. Du, H.K.; Cao, J.X.; Xue, Y.J.; Wang, X.J. Seismic facies analysis based on self-organizing map and empirical mode decomposition. J. Appl. Geophys. 2015, 112, 52–61. [Google Scholar] [CrossRef]
Geosciences 15 00183 g001
Figure 1. A location map of the study area (Mississippi Canyon Block 118) modified from [28] [Longitude: 88°28′15.4″ W; Latitude: 28°51′0.6″ N].
Figure 1. A location map of the study area (Mississippi Canyon Block 118) modified from [28] [Longitude: 88°28′15.4″ W; Latitude: 28°51′0.6″ N].
Geosciences 15 00183 g001
Geosciences 15 00183 g004
Figure 4. Multi-Attribute Analysis and Self-Organizing Maps (SOM) Workflow adopted for this research.
Figure 4. Multi-Attribute Analysis and Self-Organizing Maps (SOM) Workflow adopted for this research.
Geosciences 15 00183 g004
Geosciences 15 00183 g005
Figure 5. A comparison between the original seismic amplitudes, NRMS result, and Dominant Frequency. Areas in red within the NRMS section are areas with the highest amount of noise, yellow corresponds to moderate noise, and green is the area with the best data quality. (VE = Vertical Exaggeration).
Figure 5. A comparison between the original seismic amplitudes, NRMS result, and Dominant Frequency. Areas in red within the NRMS section are areas with the highest amount of noise, yellow corresponds to moderate noise, and green is the area with the best data quality. (VE = Vertical Exaggeration).
Geosciences 15 00183 g005
Geosciences 15 00183 g006
Figure 6. Self-Organizing Maps (SOM) algorithm architecture [70].
Figure 6. Self-Organizing Maps (SOM) algorithm architecture [70].
Geosciences 15 00183 g006
Geosciences 15 00183 g007aGeosciences 15 00183 g007b
Figure 7. Various Categories of Seismic Attributes extracted on seismic horizon showing key geologic features labeled in legend with colored arrows at the bottom of the images.
Figure 7. Various Categories of Seismic Attributes extracted on seismic horizon showing key geologic features labeled in legend with colored arrows at the bottom of the images.
Geosciences 15 00183 g007aGeosciences 15 00183 g007b
Geosciences 15 00183 g008aGeosciences 15 00183 g008b
Figure 8. SOM results for each Model (1–4) showing distinct geologic features such as radial faults, gas hydrates, and the extent of thick reservoirs and salt domes.
Figure 8. SOM results for each Model (1–4) showing distinct geologic features such as radial faults, gas hydrates, and the extent of thick reservoirs and salt domes.
Geosciences 15 00183 g008aGeosciences 15 00183 g008b
Geosciences 15 00183 g009
Figure 9. Top-most image (A) showing seismic section annotated with key geologic features. Images (B–E) show the results of the four SOM Models and how well they image these geologic features of interest.
Figure 9. Top-most image (A) showing seismic section annotated with key geologic features. Images (B–E) show the results of the four SOM Models and how well they image these geologic features of interest.
Geosciences 15 00183 g009
Geosciences 15 00183 g010aGeosciences 15 00183 g010b
Figure 10. Comparison between SOM Model results and the Gamma Ray log from the available well log information, with the preferred model shown at the bottom (SOM Model 1).
Figure 10. Comparison between SOM Model results and the Gamma Ray log from the available well log information, with the preferred model shown at the bottom (SOM Model 1).
Geosciences 15 00183 g010aGeosciences 15 00183 g010b
Geosciences 15 00183 g011
Figure 11. Comparison between Seismic Amplitude, NRMS, and SOM Model results showing SOM’s ability to detect the salt canopy (top-right corner), noise below the salt body, and faults (left).
Figure 11. Comparison between Seismic Amplitude, NRMS, and SOM Model results showing SOM’s ability to detect the salt canopy (top-right corner), noise below the salt body, and faults (left).
Geosciences 15 00183 g011
Table 1. Summary of Seismic Attribute Categories and Applications in Geosciences.
Table 1. Summary of Seismic Attribute Categories and Applications in Geosciences.
ATTRIBUTE CATEGORYSEISMIC
ATTRIBUTE		PRINCIPLEAPPLICATION IN GEOSCIENCES REFERENCES
Amplitude-based
Attributes		SweetnessDependent on instantaneous amplitude and frequencyIndicate the presence of gas hydrates and hydrocarbon.
Direct hydrocarbon indicators like bright spots, flat spots, and amplitude versus offset (AVO) anomalies.		[12,13,46,47,48]
EnvelopePhase-dependent instantaneous amplitude
Geometric
Attributes		AberrancyMeasures the degree of curvatureDetect faults and delineate stratigraphic discontinuities.
Detect distortions in lateral waveforms caused by faults, channels, or pinchouts.		[8,12,13,45,48,49,50,51,52,53,54,55,56]
CurvatureMeasures the degree of bending in the seismic waveform to identify folds
SimilarityMeasures the consistency of adjacent traces
Spectral
Decomposition		Peak
Magnitude		Seismic traces broken down into their integral frequencies by the continuous wavelet transform (CWT)Identify stratigraphic discontinuities.[57,58,59,60]
Peak
Frequency
25 Hz and 40 Hz Spectral Frequencies
Instantaneous AttributeInstantaneous PhaseEmphasize the continuity of seismic events influenced by bed thickness, sequence boundaries, and unconformitiesHighlight subtle stratigraphic features and tuning effects, direct indicators of acoustic impedance contrasts.[61,62,63,64,65]
Textural
Attribute		Grey Level Co-Occurrence Matrix (GLCM) EntropyMeasures the textural complexity of seismic dataDifferentiates between seismic facies and the detection of faults.[12,13]
Table 2. Selection of Seismic Attributes used for each Self-Organizing Map Model.
Table 2. Selection of Seismic Attributes used for each Self-Organizing Map Model.
SOM Model 1SOM Model 2SOM Model 3SOM Model 4
Attribute 1Total AberrancyPeak Magnitude SweetnessSweetness
Attribute 2Similarity Peak FrequencyDip MagnitudeGLCM Entropy
Attribute 3Peak FrequencyCurvatureInstantaneous PhaseSpectral CWT 40 Hz
Attribute 4EnvelopeSimilarity Curvature Spectral CWT 25 Hz
Attribute 5Curvature Similarity Similarity
Table 3. Covariance matrix showing the relationship between the five seismic attributes selected for SOM Model 1 results.
Table 3. Covariance matrix showing the relationship between the five seismic attributes selected for SOM Model 1 results.
SOM Model 1 Scaled Covariance Matrix
Attribute NameAttribute 1Attribute 2Attribute 3Attribute 4Attribute 5
Total Aberrancy Azimuth (Attribute 1)1.000−0.0310.0280.0220.003
Energy Ratio Similarity (Attribute 2)−0.0311.000−0.4340.4740.024
Frequency At Peak Magnitude (Attribute 3)0.028−0.4341.000−0.210−0.026
Instantaneous Envelope (Attribute 4)−0.0220.474−0.2101.000−0.016
Most-Positive Curvature (Attribute 5)0.0030.024−0.026−0.0161.000
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Samuel, S.A.; Knapp, C.C.; Knapp, J.H. Seismic Facies Classification of Salt Structures and Sediments in the Northern Gulf of Mexico Using Self-Organizing Maps. Geosciences 2025, 15, 183. https://doi.org/10.3390/geosciences15050183

AMA Style

Samuel SA, Knapp CC, Knapp JH. Seismic Facies Classification of Salt Structures and Sediments in the Northern Gulf of Mexico Using Self-Organizing Maps. Geosciences. 2025; 15(5):183. https://doi.org/10.3390/geosciences15050183

Chicago/Turabian Style

Samuel, Silas Adeoluwa, Camelia C. Knapp, and James H. Knapp. 2025. "Seismic Facies Classification of Salt Structures and Sediments in the Northern Gulf of Mexico Using Self-Organizing Maps" Geosciences 15, no. 5: 183. https://doi.org/10.3390/geosciences15050183

APA Style

Samuel, S. A., Knapp, C. C., & Knapp, J. H. (2025). Seismic Facies Classification of Salt Structures and Sediments in the Northern Gulf of Mexico Using Self-Organizing Maps. Geosciences, 15(5), 183. https://doi.org/10.3390/geosciences15050183

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop