the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 07 May 2025
From Subsurface to Seafloor: Understanding Fault Activity Through Pockmark Mapping and Machine Learning-Driven Seismic Attribute Analysis in the NW Sicily Channel
Abstract. This study explores fault activity and fluid migration in the NW Sicily Channel, focusing on the Adventure Plateau, using 2D seismic data, multi-attribute seismic analysis, and machine-learning-based fault detection. We assess the structural controls on fluid escape pathways and pockmark formation by applying seismic attributes such as fault probability, similarity, dip variance, and textural entropy.
Our findings reveal that pockmarks are spatially confined near structural highs, where near-vertical faults cross-cut the Terravecchia Formation and extend to the seafloor. Additionally, polygonal faults – despite hosting gas/fluids – do not contribute to pockmark formation, suggesting that tectonic discontinuities, rather than sediment compaction alone, govern active seepage.
To enhance fault characterization, we introduce a machine-learning-derived Fault Probability Section, providing a quantitative assessment of fault likelihood, surpassing traditional seismic interpretation. Our study proposes that pockmarks serve as proxies for active faulting, offering a new approach for offshore fault characterization. This method holds implications for seismic hazard assessment, hydrocarbon exploration, and geohazard monitoring.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(1964 KB)
Interactive discussion
Status: closed
- RC1: 'Comment on egusphere-2025-1535', Anonymous Referee #1, 04 Jun 2025
-
RC2: 'Comment on egusphere-2025-1535', Anonymous Referee #2, 04 Jun 2025
Peer Review for:
From Subsurface to Seafloor: Understanding Fault Activity Through Pockmark Mapping and Machine-Learning-Driven Seismic Attribute Analysis in the NW Sicily Channel
This was a solid study which illuminated novel information on the mechanisms for pockmark formation in the Northwest Sicily Channel. The text is easy to follow and the figures are of good quality and depict the data in an easily interpretable fashion. The methods were thoroughly described and appropriately applied. The multilayer perceptron developed in this study also quite clearly brings out the locations of faults within the sub-bottom profiles. The main conclusion that fault activity is related to seabed fluid escape and thus pockmark formation was made mostly apparent from the datasets provided in this study, but some additional analysis is needed to test this hypothesis further. In conclusion, I recommend a major revision--mainly due to my recommendation for further analysis and testing of the main conclusion of this paper.
I think this paper could be improved and sufficient for publication after addressing the below issues:
Major Issues:
Something I feel that is missing is a figure illustrating bathymetry data showing the pockmarks at a finer scale than in Figure 2. I feel that some sort of figure combining a zoomed in picture of the actively forming pockmarks over top the sub-bottom profile data illustrating the presence of active faulting. Also, some analysis relating the density of pockmarks to the density of faults could provide more comprehensive and definitive evidence supporting the main hypothesis of this paper.
Minor Issues:
Line 40: define MBES acronym (multibeam echosounder—you define later on but please do so here)
Line 65: change “gently” to “gentle”
Line 120: maybe change “providing high-resolution seafloor mapping” to “providing finely resolved seafloor bathymetry”.
Line 134: Please explicitly state what Inkscape was used to do.
Figure 2: Suggest making this figure bigger and increasing the font size.
Line 146: Change “constituted broadly three parts” to “consisting broadly of three parts”
Figure 4: Also suggest making this figure bigger with larger font size.
Figure 5: Again, suggest making this figure bigger with larger font size.
Figure 10 caption: “This model supports the hypothesis”, change this to “This model illustrates the hypothesis”. This conceptual model was constructed to illustrate the main interpretation of this paper. The data provided are what support the validity of the interpretations used to construct this conceptual model.
Citation: https://doi.org/10.5194/egusphere-2025-1535-RC2 -
RC3: 'Comment on egusphere-2025-1535', Anonymous Referee #3, 13 Aug 2025
The manuscript "From Subsurface to Seafloor: Understanding Fault Activity Through Pockmark Mapping and Machine Learning-Driven Seismic Attribute Analysis in the NW Sicily Channel", submitted by Srivastava et al., aimed at assessing the structural controls on the distribution of pockmarks in the Adventure Plateau, NW Sicily Channel. However, to be considered for publication, major revisions are necessary.
My main concern about the manuscript is related to the actual pockmark mapping. While it is in the title of the manuscript, there is no description of the characteristics of these pockmarks or their geometry. There is no seismic line showing evidence of pockmarks in the study area, and the only bathymetric profile shows the interpretation drawn over it. It would be interesting to see the DEM with no interpretation, followed by the interpreted image.
Instead, the authors concentrated mainly on the application of the OpendTect workflow for dip-steering the seismic data for fault interpretation. This is the largest section of the manuscript, and I would be fine with that if it were a description of a novel workflow for generating seismic attributes to enhance fault identification. However, it is not the case. All the workflows for the application of Structurally-Oriented Filters (SOF) are widely distributed in the OpendTect documentation page (https://doc.opendtect.org/7.0.0/doc/dgb_userdoc/Default.htm#dip-steering.htm) or in YouTube hands-on tutorials provided by dGB. The same can be said of attributes such as Similarity, Dip Line Polar, Dip Variance, Energy, and Textural Entropy. None of these attributes could be used to highlight the gas chimneys in the seismic lines?
The authors also mention the 3D visualisation of the structures, both in the OpendTect and Move environments. However, I did not see any attempt at providing a 3-dimensional view of the interpreted structures. Inserting two strike and two dip seismic lines in the 3D viewer is not a 3D view of the study area. A 3D view of the structures would entail the generation of fault surfaces (fault meshes) from interpolation methods. In addition, Figure 9 shows the fault probability in the 2D lines, with the purple/pink colours indicating a "yes". One can understand that the basement of the easternmost strike line and the central dip line is entirely faulted, whereas the western line is not faulted at all (even at the intersection with the central dip line).
Moving to the section discussing the use of Neural Networks to characterise the faults, the authors mention (page 17, 405): "Neural network-based fault detection in 2D seismic data lacks dip-line direction information, restricting the model’s ability to capture 3D fault continuity, leading to increased misclassification and fluctuating error convergence" as a source of uncertainty and instability, something that solely would raise a concern for people well-versed in seismic interpretation. But then the authors continue by saying (page 17, 410-415): "While manual interpretation can provide reasonable insights, it is inherently time-consuming and labour-intensive, often confined to a limited number of seismic slices due to practical constraints. In contrast, this study incorporates a comprehensive dataset of 40 2D seismic lines, allowing for a more extensive fault mapping approach that enhances structural interpretation. Additionally, manual fault picking is subjective, struggles with fault continuity and subtle fault detection, and is prone to misinterpretation, particularly for small-scale fractures. These limitations emphasize the need for automated, attribute-assisted, and machine-learning approaches to improve accuracy and efficiency." While I completely agree and recommend the use of seismic attributes to aid in the identification of discontinuities, faults, and fractures, machine learning approaches (as also stated by the authors) are still not able to capture the continuity of the structures from different 2D lines, something that an interpreter with a minimum of geological knowledge of the study area can do with no difficulty. Have the authors attempted to manually interpret some of the faults highlighted after the application of the SOF and attributes, and then compare with the faults interpreted through the machine-learning algorithm?
While the pockmarks return in the discussion and are stated as "proxies for active faults", none of this discussion is related to the results presented in the manuscript (fault network characterisation, neural computation aiding interpretation, attributes extraction, and neural computation pitfalls). Again, what we see is a block diagram with a drawing of pockmarks.
I believe that the authors need to consider the above-mentioned matters to improve the current manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-1535-RC3
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-1535', Anonymous Referee #1, 04 Jun 2025
The study presented by Dr. Srivastava and co-authors, titled “Fault Activity Through Pockmark Mapping and Machine Learning-Driven Seismic Attribute Analysis”, focuses on a very small region of the NW Sicily Channel. Although the topic is of considerable interest, the manuscript displays significant weaknesses.
Introduction
The introduction fails to properly frame the main subject of the paper, pockmarks as these features are not even described. Additionally, many of the cited references are not representative of the topics discussed (see the annotated PDF for details). There is also confusion regarding the eastern and western sectors of the Sicily Channel. The paper’s aims are neither clearly stated nor logically followed throughout the manuscript. The research question remains unclear, and the results and discussion sections do not effectively address the stated objectives. At one point, the authors mention “hazard” as a focus, yet the manuscript does not develop this aspect at all.Geological and Oceanographic Setting
This section requires a complete rewrite due to significant confusion in the description of the structures and morphologies of the Sicily Channel. The narrative lacks coherence, and topics are presented in a disorganized manner. The quality of the figure in this section is poor and does not convey any useful information. It is unclear why the authors chose to plot earthquakes from the past five years. what relevance does this have?Dataset and Software
The reported lateral resolution of 0.5 m raises concerns, is such resolution realistic for this bathymetric range? Moreover, despite claiming to have high-resolution data, the manuscript does not present any detailed imagery of the pockmarks. For instance, Figure 2 appears uninformative. The methodology section, which should be the core of this study, lacks sufficient referencing to support the techniques applied.Results
This section is extremely short, and the “Neural Computation Aiding Interpretation” portion reads more like a discussion than a presentation of results. Once again, there is virtually no focus on the pockmarks themselves.Discussion
Aside from a few sentences, this chapter reads more like an introduction. It should instead critically discuss the results and directly address the research aims. General commentary on fluid flow morphologies is insufficient in this context.Conclusions
The conclusion "Our findings indicate that pockmarks in the study area are spatially correlated with fault networks, particularly near structural highs, where near-vertical faults cross-cut the Terravecchia Formation and extend to the seafloor.".. it is not supported by the analysis presented in the manuscript. The pockmarks described are very small and shallow; the supposed correlation with the Terravecchia Formation is not demonstrated or discussed.General Observations
It appears that AI tools were used inappropriately, resulting in incorrect references and awkward terminology (e.g., “tectonism”).
Many references are incorrect or inappropriate (e.g., Conte et al., 2014; Maiorana regarding the geology of the Sicily Channel).
Figure captions are excessively long and introduce concepts not addressed in the main text.
The manuscript does not clearly explain the spatial relationship between gas chimneys and pockmarks.
The term “cluster of individual pockmarks” is unclear. does this imply multiple levels of clustering? The use of geological terminology throughout the paper is often imprecise.
Recommendation
In light of the above issues (and as detailed further in the attached PDF), I recommend a thorough and substantial revision of the manuscript. I can only suggest a major revision. The authors should also carefully review and correct the reference list, as some citations are missing or incorrect. -
RC2: 'Comment on egusphere-2025-1535', Anonymous Referee #2, 04 Jun 2025
Peer Review for:
From Subsurface to Seafloor: Understanding Fault Activity Through Pockmark Mapping and Machine-Learning-Driven Seismic Attribute Analysis in the NW Sicily Channel
This was a solid study which illuminated novel information on the mechanisms for pockmark formation in the Northwest Sicily Channel. The text is easy to follow and the figures are of good quality and depict the data in an easily interpretable fashion. The methods were thoroughly described and appropriately applied. The multilayer perceptron developed in this study also quite clearly brings out the locations of faults within the sub-bottom profiles. The main conclusion that fault activity is related to seabed fluid escape and thus pockmark formation was made mostly apparent from the datasets provided in this study, but some additional analysis is needed to test this hypothesis further. In conclusion, I recommend a major revision--mainly due to my recommendation for further analysis and testing of the main conclusion of this paper.
I think this paper could be improved and sufficient for publication after addressing the below issues:
Major Issues:
Something I feel that is missing is a figure illustrating bathymetry data showing the pockmarks at a finer scale than in Figure 2. I feel that some sort of figure combining a zoomed in picture of the actively forming pockmarks over top the sub-bottom profile data illustrating the presence of active faulting. Also, some analysis relating the density of pockmarks to the density of faults could provide more comprehensive and definitive evidence supporting the main hypothesis of this paper.
Minor Issues:
Line 40: define MBES acronym (multibeam echosounder—you define later on but please do so here)
Line 65: change “gently” to “gentle”
Line 120: maybe change “providing high-resolution seafloor mapping” to “providing finely resolved seafloor bathymetry”.
Line 134: Please explicitly state what Inkscape was used to do.
Figure 2: Suggest making this figure bigger and increasing the font size.
Line 146: Change “constituted broadly three parts” to “consisting broadly of three parts”
Figure 4: Also suggest making this figure bigger with larger font size.
Figure 5: Again, suggest making this figure bigger with larger font size.
Figure 10 caption: “This model supports the hypothesis”, change this to “This model illustrates the hypothesis”. This conceptual model was constructed to illustrate the main interpretation of this paper. The data provided are what support the validity of the interpretations used to construct this conceptual model.
Citation: https://doi.org/10.5194/egusphere-2025-1535-RC2 -
RC3: 'Comment on egusphere-2025-1535', Anonymous Referee #3, 13 Aug 2025
The manuscript "From Subsurface to Seafloor: Understanding Fault Activity Through Pockmark Mapping and Machine Learning-Driven Seismic Attribute Analysis in the NW Sicily Channel", submitted by Srivastava et al., aimed at assessing the structural controls on the distribution of pockmarks in the Adventure Plateau, NW Sicily Channel. However, to be considered for publication, major revisions are necessary.
My main concern about the manuscript is related to the actual pockmark mapping. While it is in the title of the manuscript, there is no description of the characteristics of these pockmarks or their geometry. There is no seismic line showing evidence of pockmarks in the study area, and the only bathymetric profile shows the interpretation drawn over it. It would be interesting to see the DEM with no interpretation, followed by the interpreted image.
Instead, the authors concentrated mainly on the application of the OpendTect workflow for dip-steering the seismic data for fault interpretation. This is the largest section of the manuscript, and I would be fine with that if it were a description of a novel workflow for generating seismic attributes to enhance fault identification. However, it is not the case. All the workflows for the application of Structurally-Oriented Filters (SOF) are widely distributed in the OpendTect documentation page (https://doc.opendtect.org/7.0.0/doc/dgb_userdoc/Default.htm#dip-steering.htm) or in YouTube hands-on tutorials provided by dGB. The same can be said of attributes such as Similarity, Dip Line Polar, Dip Variance, Energy, and Textural Entropy. None of these attributes could be used to highlight the gas chimneys in the seismic lines?
The authors also mention the 3D visualisation of the structures, both in the OpendTect and Move environments. However, I did not see any attempt at providing a 3-dimensional view of the interpreted structures. Inserting two strike and two dip seismic lines in the 3D viewer is not a 3D view of the study area. A 3D view of the structures would entail the generation of fault surfaces (fault meshes) from interpolation methods. In addition, Figure 9 shows the fault probability in the 2D lines, with the purple/pink colours indicating a "yes". One can understand that the basement of the easternmost strike line and the central dip line is entirely faulted, whereas the western line is not faulted at all (even at the intersection with the central dip line).
Moving to the section discussing the use of Neural Networks to characterise the faults, the authors mention (page 17, 405): "Neural network-based fault detection in 2D seismic data lacks dip-line direction information, restricting the model’s ability to capture 3D fault continuity, leading to increased misclassification and fluctuating error convergence" as a source of uncertainty and instability, something that solely would raise a concern for people well-versed in seismic interpretation. But then the authors continue by saying (page 17, 410-415): "While manual interpretation can provide reasonable insights, it is inherently time-consuming and labour-intensive, often confined to a limited number of seismic slices due to practical constraints. In contrast, this study incorporates a comprehensive dataset of 40 2D seismic lines, allowing for a more extensive fault mapping approach that enhances structural interpretation. Additionally, manual fault picking is subjective, struggles with fault continuity and subtle fault detection, and is prone to misinterpretation, particularly for small-scale fractures. These limitations emphasize the need for automated, attribute-assisted, and machine-learning approaches to improve accuracy and efficiency." While I completely agree and recommend the use of seismic attributes to aid in the identification of discontinuities, faults, and fractures, machine learning approaches (as also stated by the authors) are still not able to capture the continuity of the structures from different 2D lines, something that an interpreter with a minimum of geological knowledge of the study area can do with no difficulty. Have the authors attempted to manually interpret some of the faults highlighted after the application of the SOF and attributes, and then compare with the faults interpreted through the machine-learning algorithm?
While the pockmarks return in the discussion and are stated as "proxies for active faults", none of this discussion is related to the results presented in the manuscript (fault network characterisation, neural computation aiding interpretation, attributes extraction, and neural computation pitfalls). Again, what we see is a block diagram with a drawing of pockmarks.
I believe that the authors need to consider the above-mentioned matters to improve the current manuscript.
Citation: https://doi.org/10.5194/egusphere-2025-1535-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 377 | 126 | 17 | 520 | 17 | 34 |
- HTML: 377
- PDF: 126
- XML: 17
- Total: 520
- BibTeX: 17
- EndNote: 34
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The study presented by Dr. Srivastava and co-authors, titled “Fault Activity Through Pockmark Mapping and Machine Learning-Driven Seismic Attribute Analysis”, focuses on a very small region of the NW Sicily Channel. Although the topic is of considerable interest, the manuscript displays significant weaknesses.
Introduction
The introduction fails to properly frame the main subject of the paper, pockmarks as these features are not even described. Additionally, many of the cited references are not representative of the topics discussed (see the annotated PDF for details). There is also confusion regarding the eastern and western sectors of the Sicily Channel. The paper’s aims are neither clearly stated nor logically followed throughout the manuscript. The research question remains unclear, and the results and discussion sections do not effectively address the stated objectives. At one point, the authors mention “hazard” as a focus, yet the manuscript does not develop this aspect at all.
Geological and Oceanographic Setting
This section requires a complete rewrite due to significant confusion in the description of the structures and morphologies of the Sicily Channel. The narrative lacks coherence, and topics are presented in a disorganized manner. The quality of the figure in this section is poor and does not convey any useful information. It is unclear why the authors chose to plot earthquakes from the past five years. what relevance does this have?
Dataset and Software
The reported lateral resolution of 0.5 m raises concerns, is such resolution realistic for this bathymetric range? Moreover, despite claiming to have high-resolution data, the manuscript does not present any detailed imagery of the pockmarks. For instance, Figure 2 appears uninformative. The methodology section, which should be the core of this study, lacks sufficient referencing to support the techniques applied.
Results
This section is extremely short, and the “Neural Computation Aiding Interpretation” portion reads more like a discussion than a presentation of results. Once again, there is virtually no focus on the pockmarks themselves.
Discussion
Aside from a few sentences, this chapter reads more like an introduction. It should instead critically discuss the results and directly address the research aims. General commentary on fluid flow morphologies is insufficient in this context.
Conclusions
The conclusion "Our findings indicate that pockmarks in the study area are spatially correlated with fault networks, particularly near structural highs, where near-vertical faults cross-cut the Terravecchia Formation and extend to the seafloor.".. it is not supported by the analysis presented in the manuscript. The pockmarks described are very small and shallow; the supposed correlation with the Terravecchia Formation is not demonstrated or discussed.
General Observations
It appears that AI tools were used inappropriately, resulting in incorrect references and awkward terminology (e.g., “tectonism”).
Many references are incorrect or inappropriate (e.g., Conte et al., 2014; Maiorana regarding the geology of the Sicily Channel).
Figure captions are excessively long and introduce concepts not addressed in the main text.
The manuscript does not clearly explain the spatial relationship between gas chimneys and pockmarks.
The term “cluster of individual pockmarks” is unclear. does this imply multiple levels of clustering? The use of geological terminology throughout the paper is often imprecise.
Recommendation
In light of the above issues (and as detailed further in the attached PDF), I recommend a thorough and substantial revision of the manuscript. I can only suggest a major revision. The authors should also carefully review and correct the reference list, as some citations are missing or incorrect.