Landslide Assessment Classification Using Deep Neural Networks Based on Climate and Geospatial Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

For the following reasons, my decision is to reject the manuscript:
1. The abstract is not well written. Materials and results must be added.
2. The reason for choosing the algorithms in this research is not clear.
3. The difference between the previous research and the innovation of this research is not well expressed.
4. The case study is not well defined.
5. Factors affecting landslides are not well explained. How to prepare them and their plans...
6. The division of training and test data is not stated.
7. The research methodology is not well defined.
8. The genetic algorithm is not clear which hyperparameters of the deep learning model it optimizes?
9. What is the objective function in genetic algorithm?
10. The final landslide risk maps have not been prepared.
11. The discussion part is not well expressed.

Comments on the Quality of English Language

Minor editing of English language required

Author Response

Dear Reviewer,

Thank you for your thorough and thoughtful review of our work. Your valuable comments and recommendations have significantly improved the quality of our research. We greatly appreciate the time and effort you invested in reviewing our manuscript. Please find attached file with point-by-point answers on your commnets.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Reviewer Confidential Comments to Editor:

This paper proposes a method to optimize deep neural networks using a genetic algorithm for the classification of landslide triggering factors and scale based on climate and geospatial data, demonstrating its effectiveness in landslide prediction and disaster risk management. However, there are some deficiencies in the introduction and method sections. Specific comments are as follows.

Reviewer Blind Comments to Author:

1. The introduction can be appropriately simplified. Use the simplest language to outline the current situation and existing problems and explain the gaps in current research.

2.     The author should briefly introduce the development of deep learning in the introduction and its wide applications, such as image classification, semantic segmentation, and landslide information extraction in computer vision remote sensing images Such as, Landslide extraction from aerial imagery considering context association characteristics et al. This will help readers better understand the universality of deep learning applications.

3. It is recommended to add an overall technical framework diagram in the methods section to provide readers with a general understanding of the method.
4. There are some formatting errors in the paper, such as inconsistent image sizes in Figure 1. Maps should include a scale and a north arrow.
5. The discussion section should further refine the content, discussing different results obtained from the experiments. Additionally, the outlook section should not appear in the discussion.

 

Comments on the Quality of English Language

No

Author Response

Dear Reviewer,

Thank you for your thorough and thoughtful review of our work. Your valuable comments and recommendations have significantly improved the quality of our research. We greatly appreciate the time and effort you invested in reviewing our manuscript. Please find attached file with point-by-point answers on your commnets.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper develops a deep neural network based on climate and geospatial data to classify triggering events for landslides. The paper supports using a genetic algorithm to optimize the neural network for this application. Overall, the paper is well-written and addresses an interesting research question.I invite other to provide a revision of their valuable work and consider the followings: 

 

(a) why did the authors use PCA for dimensionality reduction?  while they reduced the number of features for considering climatic features, it is difficult to understand how DNN is deriving the prediction when the features are now expressed in terms of principal components. 

(b) Genetic algorithms (GA) are computationally expensive. What were the computational resources used to perform GA to optimize the neural network? Does the advantage of GA in this application overcome this computational cost? 

(c) The authors used accuracy for evaluation, but it is well known that landslide databases are highly imbalanced. I invite the authors to discuss data imbalance and provide other metrics such as precision, recall and F1 for a better comparison. 

(d) I'm unsure if the authors' model can capture seasonality and temporal patterns of landslides. Please comment on this important shortcoming of the model. 

(e) The authors must discuss how the geospatial features were selected and incorporated in the model. additional details on this section will improve the paper discussion. 

(f) there are some recent work in the area of using deep neural networks with more advanced architecture in hazard engineering, particularly earthquake engineering. It might add to the value of the introduction to briefly mention some works in this domain, particularly regarding the promising algorithm of LSTM and CNNs. For example, see the following: 

1. Esteghamati, M. Z. (2024). Leveraging machine learning techniques to support a holistic performance-based seismic design of civil structures. In Interpretable Machine Learning for the Analysis Design Assessment and Informed Decision Making for Civil Infrastructure (pp. 25-49). Woodhead Publishing.

2. Kikuchi, T., Sakita, K., Nishiyama, S., & Takahashi, K. (2023). Landslide susceptibility mapping using automatically constructed CNN architectures with pre-slide topographic DEM of deep-seated catastrophic landslides caused by Typhoon Talas. Natural Hazards, 117(1), 339-364.

 

 

Author Response

Dear Reviewer,

Thank you for your thorough and thoughtful review of our work. Your valuable comments and recommendations have significantly improved the quality of our research. We greatly appreciate the time and effort you invested in reviewing our manuscript. Please find attached file with point-by-point answers on your commnets.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The latest articles on landslide information extraction should be included.

I think the author's response is insufficient and should be explained in a separate document

Comments on the Quality of English Language

No

Author Response

Dear Reviewer,

Thank you very much for your thorough review of our manuscript. We appreciate your valuable feedback and suggestions. We have addressed your comments and made the necessary revisions, including updating the manuscript with the latest literature and providing a detailed response in a separate document.

Thank you again for your time and effort in reviewing our work.

1. The latest articles on landslide information extraction should be included:

   We have reviewed and incorporated the latest articles on landslide information extraction into the manuscript. The revisions have been updated accordingly in lines 486-496.

2. I think the author's response is insufficient and should be explained in a separate document:

   We acknowledge that the initial response may have been insufficient. We have prepared a separate document that provides a more detailed explanation and addresses your concerns comprehensively. Please find it attached.

Author Response File: Author Response.pdf