the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Oct 2023
Regional landslide susceptibility assessment based on Inter.iamb-Tabu algorithm
Abstract. Due to the great differences in geological environment characteristics and landslide disaster mechanism in different regions, the logical structure of each mathematical model is also different. It can only be determined through comparative research. Four improved algorithms based on Bayesian networkwere verified, and the error index was introduced to determine the algorithm with the best modeling effect. The landslide susceptibility probability of 774570 grids in Boshan District was calculated, and the landslide susceptibility distribution map of Boshan District was plotted. Based on the spatial superposition and grid calculator function of GIS, the landslide susceptibility assessment results of each model were compared.
- Preprint
(1876 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed (peer review stopped)
-
RC1: 'Comment on nhess-2023-153', Anonymous Referee #1, 10 Nov 2023
The submitted manuscript presents a landslide susceptibility assessment using four improved algorithms based on the Bayesian network. The paper is poorly written and needs to be largely reworked. The whole landslide susceptibility analysis is non-representative, from the selection of landslide causal factors to the definition of the landslide sample for model training and verification. The paper lacks an objective presentation of the landslide susceptibility maps and verification results (only one AUC result is mentioned and shown in the manuscript). The comparison of the obtained maps is interesting, but it is not representative and it is not clear why one of the models was accepted as the most accurate and chosen as a reference model for the verification of the other models. The paper lacks separate discussion and conclusion chapters. In my opinion, the manuscript does not reach the required quality standard of this journal.
Citation: https://doi.org/10.5194/nhess-2023-153-RC1 -
CC1: 'Comment on nhess-2023-153', Baocheng Ma, 13 Dec 2023
This paper studied the regional landslide susceptibility assessment. It contributes Inter.iamb-Tabu algorithm scheme for landslide susceptibility assessment, which enables the assessment result much more accurate. And the proposed scheme outperforms the existing technology and can apply to the future development of landslide susceptibility assessment. However, there are some weakness need to be improved before it is considered for publication.
1. The paper contains a few grammatical errors and awkward sentence structures that hinder comprehension. Thoroughly proofread the paper for grammar and clarity.
2. In this paper, some tables have the problem of inconsistent font size, and it is suggested to modify the table format again.Citation: https://doi.org/10.5194/nhess-2023-153-CC1 -
CC2: 'Comment on nhess-2023-153', Yuwei Zhang, 26 Dec 2023
In this manuscript, the authors summarized and discussed the landslide susceptibility assessment based on Inter.iamb-Tabu algorithm. This work provides new insight and opinion into the development of finding the assessment result of the highest accuracy. The manuscript is well-organized and clearly stated. I would suggest accepting it after the following concerns are addressed.
- The author should summarize the main contributions of this paper in Section 1.
- Thereare a few types of grammar errors in this paper.
- The writing style changes significantly about half way through the manuscript. Please improve the sentence structure and refine the text.
- Conclusions need more in it. The authors are suggested to highlight important findings and include afterthought of this work.
Citation: https://doi.org/10.5194/nhess-2023-153-CC2 -
RC2: 'Comment on nhess-2023-153', Anonymous Referee #2, 19 Jan 2024
Dear editors, dear authors,
Thank you for inviting me to this review. I have read the manuscript carefully and came to the following evaluation.
General comments
In their manuscript, the authors present a case study where they compare different Bayesian network algorithms to model landslide susceptibility for a study area in Shandong Province, China.
First of all, I want to mention that landslide susceptibility modelling with machine learning techniques is a heavily published topic with hundreds or even thousands of publications in the international literature. In my opinion, any further publications in this field have to be justified by presenting either groundbreaking technical developments or particularly interesting case studies with deeper geological and geomorphological interpretations. In the current manuscript, I see neither of these points adequately addressed.
Apart from the lack of novelty, the manuscript is poorly written. Grammar mistakes and sometimes inadequate terminology hinder understanding, the structure is not straightforward with important information e.g. on the methodology and the differences between the investigated algorithms missing.
Getting deeper into the study design, it is hard to understand from the text how the “best” model was assessed. What I understand is that the authors used synthetic data from two landslide-unrelated examples to assess the performance of the different algorithms to select the best-performing one. If this is true I highly doubt that this is transferable to a completely different problem (landslide susceptibility). In my experience, the performance of a machine learning algorithm depends very much on the specific problem and dataset at hand. Also, I do not understand why the authors classified the input data. In my opinion, the whole point of using machine learning is that the input data does not have to be classified. Classification introduces user-based bias into the models and covers smaller effects in the data.
The presentation of the results is also not convincing. The authors only show zoomed-out versions of the susceptibility maps and they present ROC curves only for one model, although they compare various ones. The only part I like is the spatial comparison of landslide susceptibility in Figure 9.
To sum it up, I believe that the manuscript in its current form lacks the quality required for publication in NHESS and therefore I suggest rejecting the manuscript. Anyway, please find below some specific comments that could help the authors improve the manuscript apart from a thorough review of the language. In any case, they need to point out what is the groundbreaking novelty of their work that justifies the publication in a scientific journal.
Specific comments
Introduction
A paragraph summarizing the study as is common in scientific articles is missing at the end of the introduction.
Page 2, lines 1-3: As far as I know, random forest and Bayesian networks are not deep learning algorithms. Only convolutional ANN are.
Section 2.1
Please provide a geological map. Since it is also part of the input data it is essential to be able to correctly interpret the results.
Section 2.2
Page 3, lines 2-11: Please provide the actual sources of the data, like the specific satellites
Page 3, lines 10-11: What is “the prevention and control scheme of geological disasters”? What kind of data does this contain? It has to be described in more detail.
Page 3, lines 12-14: How were the landslides mapped? As polygons or points? How were the landslides discretized spatially for model training?
Section 3.1
Page 3, line 24: What is the data source and resolution for the NDVI?
Section 3.2:
Page 4, lines 4-7: It remains unclear how the single factor Logistic regression method works and how it was implemented here. This needs to be documented and reproducible for the readers.
Page 4, lines 12-18: This passage should be moved to section 3.1.
Section 3.3
See general comment. Why classification?
Section 4.1
Here the language suddenly reads completely different from the rest of the manuscript.
Section 4.3.1
Could the authors point out to the readers how the Inter.iamb-Tabu algorithm improves the MMHC algorithm and what are the differences between the three different variations?
Page 6, lines 22-25: The interpretation that profile curvature and NDVI “induce” landslides seems a bit wild to me. These are no triggering factors. Maybe the wording is imprecise. Also, the authors state that “the occurrence of the landslide may change the profile curvature and NDVI”. Is this actually reflected by the input data? Was the DEM generated before or after the landslides occurred?
Section 4.3.3
See general comment. Why performance evaluation on problems that have nothing to do with landslide susceptibility?
Section 4.4
Figure 8 looks exactly like figure 7. It is redundant.
Page 7, lines 13-19: This information could be visualized in a diagram.
Page 7, line 8 and others: Here the authors suddenly use the word “sensitivity”. Could it be they mean “susceptibility”?
Section 4.5
ROC graphs for the other models need to be provided.
References
There are several inconsistencies in the references. Some references are not quoted in the text (e.g. An et al. 2018, He et al. 2022, Sun et al. 2020), others are not provided in the references section (Pan 2019), and others are not clearly distinguished (Chen et al. 2018)
Citation: https://doi.org/10.5194/nhess-2023-153-RC2
Status: closed (peer review stopped)
-
RC1: 'Comment on nhess-2023-153', Anonymous Referee #1, 10 Nov 2023
The submitted manuscript presents a landslide susceptibility assessment using four improved algorithms based on the Bayesian network. The paper is poorly written and needs to be largely reworked. The whole landslide susceptibility analysis is non-representative, from the selection of landslide causal factors to the definition of the landslide sample for model training and verification. The paper lacks an objective presentation of the landslide susceptibility maps and verification results (only one AUC result is mentioned and shown in the manuscript). The comparison of the obtained maps is interesting, but it is not representative and it is not clear why one of the models was accepted as the most accurate and chosen as a reference model for the verification of the other models. The paper lacks separate discussion and conclusion chapters. In my opinion, the manuscript does not reach the required quality standard of this journal.
Citation: https://doi.org/10.5194/nhess-2023-153-RC1 -
CC1: 'Comment on nhess-2023-153', Baocheng Ma, 13 Dec 2023
This paper studied the regional landslide susceptibility assessment. It contributes Inter.iamb-Tabu algorithm scheme for landslide susceptibility assessment, which enables the assessment result much more accurate. And the proposed scheme outperforms the existing technology and can apply to the future development of landslide susceptibility assessment. However, there are some weakness need to be improved before it is considered for publication.
1. The paper contains a few grammatical errors and awkward sentence structures that hinder comprehension. Thoroughly proofread the paper for grammar and clarity.
2. In this paper, some tables have the problem of inconsistent font size, and it is suggested to modify the table format again.Citation: https://doi.org/10.5194/nhess-2023-153-CC1 -
CC2: 'Comment on nhess-2023-153', Yuwei Zhang, 26 Dec 2023
In this manuscript, the authors summarized and discussed the landslide susceptibility assessment based on Inter.iamb-Tabu algorithm. This work provides new insight and opinion into the development of finding the assessment result of the highest accuracy. The manuscript is well-organized and clearly stated. I would suggest accepting it after the following concerns are addressed.
- The author should summarize the main contributions of this paper in Section 1.
- Thereare a few types of grammar errors in this paper.
- The writing style changes significantly about half way through the manuscript. Please improve the sentence structure and refine the text.
- Conclusions need more in it. The authors are suggested to highlight important findings and include afterthought of this work.
Citation: https://doi.org/10.5194/nhess-2023-153-CC2 -
RC2: 'Comment on nhess-2023-153', Anonymous Referee #2, 19 Jan 2024
Dear editors, dear authors,
Thank you for inviting me to this review. I have read the manuscript carefully and came to the following evaluation.
General comments
In their manuscript, the authors present a case study where they compare different Bayesian network algorithms to model landslide susceptibility for a study area in Shandong Province, China.
First of all, I want to mention that landslide susceptibility modelling with machine learning techniques is a heavily published topic with hundreds or even thousands of publications in the international literature. In my opinion, any further publications in this field have to be justified by presenting either groundbreaking technical developments or particularly interesting case studies with deeper geological and geomorphological interpretations. In the current manuscript, I see neither of these points adequately addressed.
Apart from the lack of novelty, the manuscript is poorly written. Grammar mistakes and sometimes inadequate terminology hinder understanding, the structure is not straightforward with important information e.g. on the methodology and the differences between the investigated algorithms missing.
Getting deeper into the study design, it is hard to understand from the text how the “best” model was assessed. What I understand is that the authors used synthetic data from two landslide-unrelated examples to assess the performance of the different algorithms to select the best-performing one. If this is true I highly doubt that this is transferable to a completely different problem (landslide susceptibility). In my experience, the performance of a machine learning algorithm depends very much on the specific problem and dataset at hand. Also, I do not understand why the authors classified the input data. In my opinion, the whole point of using machine learning is that the input data does not have to be classified. Classification introduces user-based bias into the models and covers smaller effects in the data.
The presentation of the results is also not convincing. The authors only show zoomed-out versions of the susceptibility maps and they present ROC curves only for one model, although they compare various ones. The only part I like is the spatial comparison of landslide susceptibility in Figure 9.
To sum it up, I believe that the manuscript in its current form lacks the quality required for publication in NHESS and therefore I suggest rejecting the manuscript. Anyway, please find below some specific comments that could help the authors improve the manuscript apart from a thorough review of the language. In any case, they need to point out what is the groundbreaking novelty of their work that justifies the publication in a scientific journal.
Specific comments
Introduction
A paragraph summarizing the study as is common in scientific articles is missing at the end of the introduction.
Page 2, lines 1-3: As far as I know, random forest and Bayesian networks are not deep learning algorithms. Only convolutional ANN are.
Section 2.1
Please provide a geological map. Since it is also part of the input data it is essential to be able to correctly interpret the results.
Section 2.2
Page 3, lines 2-11: Please provide the actual sources of the data, like the specific satellites
Page 3, lines 10-11: What is “the prevention and control scheme of geological disasters”? What kind of data does this contain? It has to be described in more detail.
Page 3, lines 12-14: How were the landslides mapped? As polygons or points? How were the landslides discretized spatially for model training?
Section 3.1
Page 3, line 24: What is the data source and resolution for the NDVI?
Section 3.2:
Page 4, lines 4-7: It remains unclear how the single factor Logistic regression method works and how it was implemented here. This needs to be documented and reproducible for the readers.
Page 4, lines 12-18: This passage should be moved to section 3.1.
Section 3.3
See general comment. Why classification?
Section 4.1
Here the language suddenly reads completely different from the rest of the manuscript.
Section 4.3.1
Could the authors point out to the readers how the Inter.iamb-Tabu algorithm improves the MMHC algorithm and what are the differences between the three different variations?
Page 6, lines 22-25: The interpretation that profile curvature and NDVI “induce” landslides seems a bit wild to me. These are no triggering factors. Maybe the wording is imprecise. Also, the authors state that “the occurrence of the landslide may change the profile curvature and NDVI”. Is this actually reflected by the input data? Was the DEM generated before or after the landslides occurred?
Section 4.3.3
See general comment. Why performance evaluation on problems that have nothing to do with landslide susceptibility?
Section 4.4
Figure 8 looks exactly like figure 7. It is redundant.
Page 7, lines 13-19: This information could be visualized in a diagram.
Page 7, line 8 and others: Here the authors suddenly use the word “sensitivity”. Could it be they mean “susceptibility”?
Section 4.5
ROC graphs for the other models need to be provided.
References
There are several inconsistencies in the references. Some references are not quoted in the text (e.g. An et al. 2018, He et al. 2022, Sun et al. 2020), others are not provided in the references section (Pan 2019), and others are not clearly distinguished (Chen et al. 2018)
Citation: https://doi.org/10.5194/nhess-2023-153-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 301 | 78 | 21 | 400 | 13 | 11 |
- HTML: 301
- PDF: 78
- XML: 21
- Total: 400
- BibTeX: 13
- EndNote: 11
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1