the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Jul 2024
Analysis of Borehole Strain Anomalies Before the 2017 Jiuzhaigou Ms7.0 Earthquake Based on Graph Neural Network
Abstract. On August 8, 2017, a strong magnitude 7.0 earthquake occurred in Jiuzhaigou, Sichuan Province, China. To assess pre-earthquake anomalies, we utilized variational mode decomposition to preprocess borehole strain observation data and combined it with a graph wavenet graph neural network model to process data from multiple stations. We obtained one-year data from four stations near the epicenter as the training dataset and data from January 1 to August 10, 2017, as the test dataset. For the prediction results of the variational mode decomposition-graph wavenet model, the anomalous days were extracted using statistical methods, and the results of anomalous day accumulation at multiple stations showed that an increase in the number of anomalous days occurred 15–32 days before the earthquake. The acceleration effect of anomalous accumulation was most obvious in the 20-day period before the earthquake, and an increase in the number of anomalous days also occurred in the one to three days post-earthquake. We tentatively deduce that the pre-earthquake anomalies are caused by the diffusion of strain energy near the epicenter during the accumulation process, which can be used as a signal of pro-seismic anomalies, whereas the post-earthquake anomalies are caused by the frequent occurrence of aftershocks.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(5743 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(5743 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-2025', Anonymous Referee #1, 30 Jul 2024
The manuscript entitled “Analysis of Borehole Strain Anomalies Before the 2017 Jiuzhaigou Ms7.0 Earthquake Based on Graph Neural Network” presents the results of analyzing strain meter data from four sites prior to a large magnitude earthquake aiming to identify pre-seismic signal using Neural network technique.
My general feeling reading the manuscript is that is well organized, interesting work providing some new useful information of exploiting graph neural network approach to estimate/define pre-seismic signal on strain time series. I think that the manuscript require some small modifications and some of its parts to be improved in order to be more explanatory and understandable but in any case, it is considered, in my opinion, as a nice work. However, not being very familiar with the neural networks I would like to see some more detail information concerning the analysis in some of the manuscript paragraphs.
As a first remark I would like to mention that although I am not a native English person, in several cases the text has to be reformatted, to be “easier” for the reader.
Another general comment that I have is that the text and the processing focus on the pre-seismic period and ignores completely the post seismic period. Of course, it is well explained by the authors, that the manuscript examines the possible anomalies on the preparatory stage of a strong earthquake, but since the earthquake occurred in 2017, and a long time passed since this event, it would be interesting to see if the area, and the strain data from the same stations, shows any similar behavior as during the pre-seismic period. In other words, as a validation of this processing it could be interesting to examine the years after the earthquake if there was another period that these 4 stations show anomalous days (as in Figure 9) without the occurrence of an earthquake. The authors used data only 1 year before the earthquake … using longer period is it possible that there is and another “anomalous” “S” shape period? Of course I can understand that the authors focus on the technique as a tool for extracting pre-seismic signals of an earthquake, but I would like to see some comment on this issue.
Some more detail comments:
- Line 16 “pro-seismic” , I think that is more common the use of the term “pre-seismic”
- Line 19 “…such as volcanic eruptions…” I am not so sure that earthquakes trigger volcanic eruptions, if so please add a reference.
- Line 41 “…the use of a GNN can mine additional hidden information between nodes…” this is a very general statement, please provide some examples.
- Lines 49-51. Please describe a bit more analytical the meaning of a “node” how a node is defined? What is its characteristics and/or its physical meaning.
- Figure 1. It would be nice to be added a map of the broad area (as inset) indicating the position of your area so to be easier for readers not familiar with the area to orientate themselves.
- Line 146 “ .. preprocessing of the surface strain Sa” Why the authors choose only the Sa and not S13 or S24? Please explain. A comment on this issue (selection) could be added maybe in the end of Section 3.1 line 88.
- Figure 4. It would be helpful the figure caption to be more analytic. Especially as what it is presented in the final diagram.
- Sections 4.2.1 and 4.25.2 I would like (personally) to see some more detail description of this part. Actually, I would like to be more explicable and defines (maybe with examples) some of the terms used on it … “layers”, “gating mechanisms” etc.
- Equation 9. Define the parameter “T”
- Line 237 “…75% of the samples and labels. “ Please define what are the samples and what are the labels.
- Section 5.2 and Figure 9. Concerning the results presented in this section, could the authors comment on why the Haiyuan station shows this strong S shape anomalies, with many points, while the other stations (Linxia and Guza), although it appear that there are closer to the epicenter do not reveal a similar “strong” anomaly.
Citation: https://doi.org/10.5194/egusphere-2024-2025-RC1 -
AC1: 'Reply on RC1', Chenyang Li, 13 Aug 2024
Response to Reviewer:
I am very grateful to your comments for the manuscript. Thank you for your advice. All your suggestions are very important. They have important guiding significance for our paper and our research work. We have revised the manuscript according to your comments.
For specific responses to reviewers' comments, see "Responses to Reviewers."
-
RC2: 'Comment on egusphere-2024-2025', Anonymous Referee #2, 27 Aug 2024
Dear authors,
Thank you for the opportunity to review your manuscript on analysing the occurrence of borehole deformation anomalies before (and after) the 2017 Jiuzhaigou earthquake. Your work addresses an important aspect of earthquake prediction, and the methods you used are intriguing. However, I have some concerns and suggestions that I think could improve the clarity and impact of your study.
The manuscript presents a method for analysing anomalous stress accumulations in borehole data to predict potential earthquakes. This approach combines signal decomposition techniques to preprocess strain time series with a Graph Neural Network (GNN) for prediction. Although the goal of the study is ambitious and the overall objective is clear, anomaly detection for a single event, although scientifically valuable, does not necessarily correlate directly with the occurrence of an earthquake.
While their manuscript provides a comprehensive overview of the techniques used, such as Variational Mode Decomposition (VMD) and Graph Wavenet Neural Network (GWN), it lacks a detailed justification for the choice of these specific methods over more traditional approaches. The manuscript would benefit from a clearer explanation of why these complex techniques were chosen and what specific advantages they offer when analysing borehole strain data, especially compared to simpler methods such as band filtering or traditional statistical models. For example, the use of VMD could be replaced by simpler signal processing methods, and the rationale for using a neural network for prediction, which could be achieved using conventional signal processing methods, is not convincingly presented.
Furthermore, while you mention that VMD was used to pre-process the data by removing annual trends and tides, there is no explanation of how this pre-processing specifically improves the performance of the GWN model. This gap makes it difficult to assess the necessity and effectiveness of VMD within your analysis pipeline. A more detailed description of the role and impact of VMD could help clarify its contribution to your results.
In addition, while you report an increase in anomalous days 15–32 days prior to the earthquake, with a significant acceleration observed in the 20 days prior, the manuscript does not provide detailed statistical analyses or margins of error for these observations. Such information is crucial for understanding the robustness of your results. I suggest adding confidence intervals or error bars to make the reliability and statistical significance of your results clearer.
I also noticed that you observed an increase in anomalous days one to three days after the earthquake and attributed this to aftershocks. While this observation is interesting, it seems to have little to do with the main focus of your study on earthquake prediction. It would be helpful to clarify how this post-seismic analysis relates to the main goal of earthquake prediction and to discuss its significance in the context of your overall results.
In addition, the use of an S-shaped function to fit the cumulative results of anomalous days is mentioned, but the manuscript does not adequately explain why this particular fitting method was chosen or how it compares to other models. A more detailed discussion of this choice and the associated findings would improve the reader’s understanding of your analytical approach.
Finally, the manuscript suggests that the pre-earthquake anomalies are due to strain energy diffusion near the epicentre. This claim appears to have been made without a solid empirical or theoretical basis in the text. It would be beneficial if you could provide additional evidence or references to support this assumption or discuss alternative explanations for the observed anomalies.
The manuscript is generally well written, but there are some areas where the English could be improved (e.g., "This unique geographic location makes earthquakes a common occurrence"). Also, some typographical errors need to be corrected (e.g., "sevesral" and "pro-seismic"). The figures, especially Figures 3, 4 and 7, are too small and difficult to read, which makes them difficult to understand.
I hope that these comments will be helpful in revising your manuscript. Clarifying these points will not only strengthen the scientific rigour of your study, but will also make your results more accessible and meaningful to the research community.
Best regards.
Citation: https://doi.org/10.5194/egusphere-2024-2025-RC2 -
AC2: 'Reply on RC2', Chenyang Li, 20 Sep 2024
Response to Reviewer:
I am very grateful to your comments for the manuscript. Thank you for your advice. All your suggestions are very important. They have important guiding significance for our paper and our research work. We have revised the manuscript according to your comments.
For specific responses to reviewers' comments, see "Responses to Reviewers 2.pdf."
-
AC2: 'Reply on RC2', Chenyang Li, 20 Sep 2024
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2024-2025', Anonymous Referee #1, 30 Jul 2024
The manuscript entitled “Analysis of Borehole Strain Anomalies Before the 2017 Jiuzhaigou Ms7.0 Earthquake Based on Graph Neural Network” presents the results of analyzing strain meter data from four sites prior to a large magnitude earthquake aiming to identify pre-seismic signal using Neural network technique.
My general feeling reading the manuscript is that is well organized, interesting work providing some new useful information of exploiting graph neural network approach to estimate/define pre-seismic signal on strain time series. I think that the manuscript require some small modifications and some of its parts to be improved in order to be more explanatory and understandable but in any case, it is considered, in my opinion, as a nice work. However, not being very familiar with the neural networks I would like to see some more detail information concerning the analysis in some of the manuscript paragraphs.
As a first remark I would like to mention that although I am not a native English person, in several cases the text has to be reformatted, to be “easier” for the reader.
Another general comment that I have is that the text and the processing focus on the pre-seismic period and ignores completely the post seismic period. Of course, it is well explained by the authors, that the manuscript examines the possible anomalies on the preparatory stage of a strong earthquake, but since the earthquake occurred in 2017, and a long time passed since this event, it would be interesting to see if the area, and the strain data from the same stations, shows any similar behavior as during the pre-seismic period. In other words, as a validation of this processing it could be interesting to examine the years after the earthquake if there was another period that these 4 stations show anomalous days (as in Figure 9) without the occurrence of an earthquake. The authors used data only 1 year before the earthquake … using longer period is it possible that there is and another “anomalous” “S” shape period? Of course I can understand that the authors focus on the technique as a tool for extracting pre-seismic signals of an earthquake, but I would like to see some comment on this issue.
Some more detail comments:
- Line 16 “pro-seismic” , I think that is more common the use of the term “pre-seismic”
- Line 19 “…such as volcanic eruptions…” I am not so sure that earthquakes trigger volcanic eruptions, if so please add a reference.
- Line 41 “…the use of a GNN can mine additional hidden information between nodes…” this is a very general statement, please provide some examples.
- Lines 49-51. Please describe a bit more analytical the meaning of a “node” how a node is defined? What is its characteristics and/or its physical meaning.
- Figure 1. It would be nice to be added a map of the broad area (as inset) indicating the position of your area so to be easier for readers not familiar with the area to orientate themselves.
- Line 146 “ .. preprocessing of the surface strain Sa” Why the authors choose only the Sa and not S13 or S24? Please explain. A comment on this issue (selection) could be added maybe in the end of Section 3.1 line 88.
- Figure 4. It would be helpful the figure caption to be more analytic. Especially as what it is presented in the final diagram.
- Sections 4.2.1 and 4.25.2 I would like (personally) to see some more detail description of this part. Actually, I would like to be more explicable and defines (maybe with examples) some of the terms used on it … “layers”, “gating mechanisms” etc.
- Equation 9. Define the parameter “T”
- Line 237 “…75% of the samples and labels. “ Please define what are the samples and what are the labels.
- Section 5.2 and Figure 9. Concerning the results presented in this section, could the authors comment on why the Haiyuan station shows this strong S shape anomalies, with many points, while the other stations (Linxia and Guza), although it appear that there are closer to the epicenter do not reveal a similar “strong” anomaly.
Citation: https://doi.org/10.5194/egusphere-2024-2025-RC1 -
AC1: 'Reply on RC1', Chenyang Li, 13 Aug 2024
Response to Reviewer:
I am very grateful to your comments for the manuscript. Thank you for your advice. All your suggestions are very important. They have important guiding significance for our paper and our research work. We have revised the manuscript according to your comments.
For specific responses to reviewers' comments, see "Responses to Reviewers."
-
RC2: 'Comment on egusphere-2024-2025', Anonymous Referee #2, 27 Aug 2024
Dear authors,
Thank you for the opportunity to review your manuscript on analysing the occurrence of borehole deformation anomalies before (and after) the 2017 Jiuzhaigou earthquake. Your work addresses an important aspect of earthquake prediction, and the methods you used are intriguing. However, I have some concerns and suggestions that I think could improve the clarity and impact of your study.
The manuscript presents a method for analysing anomalous stress accumulations in borehole data to predict potential earthquakes. This approach combines signal decomposition techniques to preprocess strain time series with a Graph Neural Network (GNN) for prediction. Although the goal of the study is ambitious and the overall objective is clear, anomaly detection for a single event, although scientifically valuable, does not necessarily correlate directly with the occurrence of an earthquake.
While their manuscript provides a comprehensive overview of the techniques used, such as Variational Mode Decomposition (VMD) and Graph Wavenet Neural Network (GWN), it lacks a detailed justification for the choice of these specific methods over more traditional approaches. The manuscript would benefit from a clearer explanation of why these complex techniques were chosen and what specific advantages they offer when analysing borehole strain data, especially compared to simpler methods such as band filtering or traditional statistical models. For example, the use of VMD could be replaced by simpler signal processing methods, and the rationale for using a neural network for prediction, which could be achieved using conventional signal processing methods, is not convincingly presented.
Furthermore, while you mention that VMD was used to pre-process the data by removing annual trends and tides, there is no explanation of how this pre-processing specifically improves the performance of the GWN model. This gap makes it difficult to assess the necessity and effectiveness of VMD within your analysis pipeline. A more detailed description of the role and impact of VMD could help clarify its contribution to your results.
In addition, while you report an increase in anomalous days 15–32 days prior to the earthquake, with a significant acceleration observed in the 20 days prior, the manuscript does not provide detailed statistical analyses or margins of error for these observations. Such information is crucial for understanding the robustness of your results. I suggest adding confidence intervals or error bars to make the reliability and statistical significance of your results clearer.
I also noticed that you observed an increase in anomalous days one to three days after the earthquake and attributed this to aftershocks. While this observation is interesting, it seems to have little to do with the main focus of your study on earthquake prediction. It would be helpful to clarify how this post-seismic analysis relates to the main goal of earthquake prediction and to discuss its significance in the context of your overall results.
In addition, the use of an S-shaped function to fit the cumulative results of anomalous days is mentioned, but the manuscript does not adequately explain why this particular fitting method was chosen or how it compares to other models. A more detailed discussion of this choice and the associated findings would improve the reader’s understanding of your analytical approach.
Finally, the manuscript suggests that the pre-earthquake anomalies are due to strain energy diffusion near the epicentre. This claim appears to have been made without a solid empirical or theoretical basis in the text. It would be beneficial if you could provide additional evidence or references to support this assumption or discuss alternative explanations for the observed anomalies.
The manuscript is generally well written, but there are some areas where the English could be improved (e.g., "This unique geographic location makes earthquakes a common occurrence"). Also, some typographical errors need to be corrected (e.g., "sevesral" and "pro-seismic"). The figures, especially Figures 3, 4 and 7, are too small and difficult to read, which makes them difficult to understand.
I hope that these comments will be helpful in revising your manuscript. Clarifying these points will not only strengthen the scientific rigour of your study, but will also make your results more accessible and meaningful to the research community.
Best regards.
Citation: https://doi.org/10.5194/egusphere-2024-2025-RC2 -
AC2: 'Reply on RC2', Chenyang Li, 20 Sep 2024
Response to Reviewer:
I am very grateful to your comments for the manuscript. Thank you for your advice. All your suggestions are very important. They have important guiding significance for our paper and our research work. We have revised the manuscript according to your comments.
For specific responses to reviewers' comments, see "Responses to Reviewers 2.pdf."
-
AC2: 'Reply on RC2', Chenyang Li, 20 Sep 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 212 | 103 | 291 | 606 | 13 | 13 |
- HTML: 212
- PDF: 103
- XML: 291
- Total: 606
- BibTeX: 13
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Chenyang Li
Changfeng Qin
Jie Zhang
Yu Duan
Chengquan Chi
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(5743 KB) - Metadata XML