Landslide Susceptibility Evaluation Based on the Combination of Environmental Similarity and BP Neural Networks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editor and Authors,

I have revised the manuscript entitled: Landslide Susceptibility Evaluation Based on the Combined with Environmental Similarity and BP Neural Networks.  In this study the Authors proposes a landslide susceptibility evaluation method that integrates environmental similarity with BP neural networks (Environmental Similarity Model-BP Neural Network, ESM-BP). The topic of the manuscript is certainly suitable for the journal and of potential international interest, but unfortunately, the paper has several critical points that hamper the publication in the present form and a major revision is request. The main problem of the paper is that it appears confused and low fluid and in many part the methodology applied should be better described. No interesting details about the data specification and study area that show the uniqueness of the problem statement with the hypothesis and findings no in depth discussion, only presenting the results in general manner and often unclear and repetitive. Many data are presented, which is either not clear how was obtained and/or discussed in depth. The discussion failed to present a clear story but rather includes several issues, which were discussed in a superficial way.

In the following, the authors can find several comments and suggestions.

1) In the introduction section, it is challenging to see the justification for this research. The literature review is insufficient. The paper needs to clearly state the problems with existing works and specify what problem(s) this paper addresses. Also, the research background about the proposed subject and methods has not been appropriately mentioned, especially about the applied suggested models in previous studies. Furthermore, the main goal of the current research should clearly and concisely explain the main scientific contributions of this work. I would recommend add some recent research articles to reconstruct this section with extensive literature:

-Wang, L.J.; Guo, M.; Sawada, K.; Lin, J.; Zhang, J. A comparative study of landslide susceptibility maps using logistic regression, frequency ratio, decision tree, weights of evidence and artificial neural network. Geosci. J. 2016, 20, 117–136.

- Reichenbach P, Rossi M, Malamud BD, Mihir M, Guzzetti F (2018) A review of statistically-based landslide susceptibility models. Earth Sci Rev 180:60–91;

- Conforti M., Borrelli. L., Cofone G., Gullà G. 2023. Exploring performance and robustness of shallow landslide susceptibility modeling at regional scale using different training and testing sets. Environmental Earth Sciences 82:161.

These papers can be useful also to comment and discuss the data obtained.

2) I suggest to add some information, related to geological, geomorphological and climatic setting of the study area.

3) I suggest also to add in the text a brief description of the landslide type. The authors should be add in the text some information about landslide movement, according to their types (for more see: Cruden DM, Varnes DJ (1996) Landslide types and processes. In: Turner AK, Schuster RL (eds) Landslides investigation and mitigation. Transportation research board, US National Council, Special Report 247, Washington, DC, Chapter 3:36-75, and Hungr O, Leroueil S, Picarelli L (2014) The Varnes classification of landslide types, an update. Landslides 11:167-194).

4) The Authors, should explain why choice only these eight environmental factors for landslide susceptibility analysis. In literature, many papers, for assess the landslide susceptibility which using a lot of others geo-environmental factors.

5) The authors should be evaluate the relative importance of each environmental factor used in the landslide susceptibility analysis. This analyses it is important to identify the environmental factors and their contribute to perform a landslide susceptibility map. Not all factors have an equal importance on landslide occurrences; thus, the most and least important factors must be identified and the least effective factors should be removed from modelling, as they can reduce the prediction capability of the susceptibility model.

6) The methods applied for mapping the landslide susceptibility should be adequately described.

7) Why do the authors you have not validate the susceptibility maps by evaluate also the AUC ROC curve value?

 8) Communally, for the validation of a landslide susceptibility model, the landslide inventory is splitted in two subsets, e.g. training set and testing set, why in this study in not made? Specify please.

9) A statistical significance test should have been applied to conduct a rigorous comparative analysis, especially in cases where the differences are small.

10) The method used to classify the landslide susceptibility maps in 5 classes needs to be described.

11) In the discussion section, the reasons for the results obtained were not examined, on the contrary, an extensive literature review was given. Please also refer to issues such as method comparison, a summary of results, and study limitations for further research.

Author Response

Comments 1: In the introduction section, it is challenging to see the justification for this research. The literature review is insufficient. The paper needs to clearly state the problems with existing works and specify what problem(s) this paper addresses. Also, the research background about the proposed subject and methods has not been appropriately mentioned, especially about the applied suggested models in previous studies. Furthermore, the main goal of the current research should clearly and concisely explain the main scientific contributions of this work. I would recommend add some recent research articles to reconstruct this section with extensive literature:

-Wang, L.J.; Guo, M.; Sawada, K.; Lin, J.; Zhang, J. A comparative study of landslide susceptibility maps using logistic regression, frequency ratio, decision tree, weights of evidence and artificial neural network. Geosci. J. 2016, 20, 117–136.

- Reichenbach P, Rossi M, Malamud BD, Mihir M, Guzzetti F (2018) A review of statistically-based landslide susceptibility models. Earth Sci Rev 180:60–91;

- Conforti M., Borrelli. L., Cofone G., Gullà G. 2023. Exploring performance and robustness of shallow landslide susceptibility modeling at regional scale using different training and testing sets. Environmental Earth Sciences 82:161.

These papers can be useful also to comment and discuss the data obtained.

Response 1: Thank you very much for your comments, We agree with this comment; therefore, we have reconstructed the introduction section and added some recent research articles that illustrate the existing issues in landslide susceptibility assessment. This paper addresses the current problem of insufficient landslide samples in complex mountainous areas and explains the significant importance of conducting this research (specifically referring to lines 54 to 92 of the introduction section).

Comments 2：I suggest to add some information, related to geological, geomorphological and climatic setting of the study area.

Response 2: Thank you very much for your comments, According to your suggestion, the relevant information regarding the research area has been improved in the profile section of the study area, with additional details on the geomorphological environment, climate conditions, and other information (specifically in section 3.1, lines 246 to 260).

Comments 3: I suggest also to add in the text a brief description of the landslide type. The authors should be add in the text some information about landslide movement, according to their types (for more see: Cruden DM, Varnes DJ (1996) Landslide types and processes. In: Turner AK, Schuster RL (eds) Landslides investigation and mitigation. Transportation research board, US National Council, Special Report 247, Washington, DC, Chapter 3:36-75, and Hungr O, Leroueil S, Picarelli L (2014) The Varnes classification of landslide types, an update. Landslides 11:167-194).

Response 3: Thank you very much for your comments, the description of landslide disasters in the region, including types of landslides and mechanisms of landslide movement, has been added to the overview section of the research area (specifically in section 3.1, lines 246 to 260).

Comments 4: The Authors, should explain why choice only these eight environmental factors for landslide susceptibility analysis. In literature, many papers, for assess the landslide susceptibility which using a lot of others geo-environmental factors.

Response 4: Thank you very much for your comments, In the selection of environmental factors in the study, it depends on the environmental characteristics of the research area; having more factors is not necessarily better. Relevant environmental factors with a strong correlation to the causal factors of local landslide disasters should be selected to ensure the scientific validity of the experiments and the accuracy of the results. Therefore, this study selected 8 environmental factors for susceptibility assessment based on their importance.

Comments 5: The authors should be evaluate the relative importance of each environmental factor used in the landslide susceptibility analysis. This analyses it is important to identify the environmental factors and their contribute to perform a landslide susceptibility map. Not all factors have an equal importance on landslide occurrences; thus, the most and least important factors must be identified and the least effective factors should be removed from modelling, as they can reduce the prediction capability of the susceptibility model.

Response 5: Thank you very much for your comments, We greatly agree with your viewpoint that determining the importance of environmental factors is a crucial basis for selecting these factors, and supplementing this content makes the article more comprehensive. During the selection process of environmental factors, the weights of 11 factors were calculated, and those with lower weights were excluded. Due to the limited scope of the study area, some factors exhibited negligible differences within that region (such as rainfall), leading to a lower impact of these factors (as specifically reflected in section 3.2.2 and Table 1).

Comments 6: The methods applied for mapping the landslide susceptibility should be adequately described.

Response 6: Thank you very much for your comments, In the theoretical method section, it is described that the environmental similarity of historical landslide points is calculated, and thresholds are set based on normal distribution. Through environmental similarity analysis, the similarity ranges corresponding to different susceptibility levels are determined. Within each level interval, additional sample points are generated according to the similarity values. A landslide susceptibility assessment model is utilized to create landslide susceptibility maps (as detailed in the theoretical method in Section 2).

Comments 7: Why do the authors you have not validate the susceptibility maps by evaluate also the AUC ROC curve value?

Response 7: Thank you very much for your comments. The analysis of the AUC ROC curve has been added (specific details are provided in the model accuracy validation section, lines 382 to 388).

Comments 8: Communally, for the validation of a landslide susceptibility model, the landslide inventory is splitted in two subsets, e.g. training set and testing set, why in this study in not made? Specify please.

Response 8: Thank you very much for your comments, the study has supplemented the explanation of the dataset partitioning method in the model accuracy validation section: the sample data was divided into training and testing sets in a ratio of 7:3 (specific details are provided in section 2.5, lines 204 to 210).

Comments 9: A statistical significance test should have been applied to conduct a rigorous comparative analysis, especially in cases where the differences are small.

Response 9: Thank you very much for your comments, the current experiment supplemented the comparative analysis of the landslide susceptibility evaluation model with a t-test for significance testing (specific details are explained in section 2.5, lines 238 to 243 and section 4.2, lines 389 to 393).

Comments 10: The method used to classify the landslide susceptibility maps in 5 classes needs to be described.

Response 10: Thank you very much for your comments, this study is conducted through the application of normal distribution methods combined with the similarity values of known landslide sample points. Based on the theory of environmental similarity, it calculates the environmental similarity values for each pixel point. By analyzing the similarity values of historical landslide sample points at the corresponding pixel points, it summarizes distribution patterns and categorizes different intervals of landslide susceptibility levels. Consequently, landslide susceptibility is classified into five categories (specific details are elaborated in sections 2.3 and 4.1).

Comments 11: In the discussion section, the reasons for the results obtained were not examined, on the contrary, an extensive literature review was given. Please also refer to issues such as method comparison, a summary of results, and study limitations for further research.

Response 11: Thank you very much for your comments, The discussion section has been restructured, describing the innovations and limitations of this research as well as the challenges and research directions encountered in future studies (specifically in section 5.1, lines 444 to 483).

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Review of manuscript "Landslide Susceptibility Evaluation Based on the Combined with Environmental Similarity and BP Neural Networks" (land-3539810)

This manuscript tries to present a landslide susceptibility evaluation method that integrates environmental similarity with BP neural networks (Environmental Similarity Model-BP Neural Network, ESM-BP). Based on the theory of environmental similarity, the Baihetan reservoir area is selected as the study region. After reading, my detailed concerns are presented in the following:

- . The manuscript expands sample data through environmental similarity, but does not fully demonstrate the authenticity and rationality of the generated samples. The determination of the σ value in the normal distribution model lacks theoretical explanation or experimental verification, which may lead to similarity calculation deviation.

- . It is necessary to supplement the basis for selecting the σ value and verify whether the distribution of the generated samples is consistent with the characteristics of the real landslide environment.

- . The manuscript uses the normal distribution method to set the similarity threshold, but this method is too simplified and cannot accurately reflect the actual situation of the landslide. In addition, the setting of the threshold is subjective, which affects the rationality of sample expansion and the accuracy of the model.

- . Key influencing factors, such as rainfall, soil type, and seismic activity, are not included in the model, resulting in an incomplete calculation of environmental similarity.

- . Only accuracy, Kappa coefficient, and RMSE are used to evaluate model performance, without considering the problem of class imbalance. High accuracy may mask the model's insufficient prediction ability for minority classes.

- . It is necessary to supplement the F1 score, AUC-ROC curve or confusion matrix, and analyze the recall and precision of the model for each category (especially high-risk areas) to more comprehensively reflect the model performance.

- . The mechanism of BP neural network being superior to SVM and RF when the samples are insufficient is not explained. Does the nonlinear fitting ability match the environmental similarity characteristics?

- . In addition, it does not mention how to avoid overfitting or optimize the training process. Therefore, it is necessary to supplement the model structure details (such as the number of hidden layers, activation function), the hyperparameter tuning process, and show the training stability through the loss curve.

- . The threshold division method based on normal distribution does not take into account the differences in regional geological conditions, which affects the generalization ability of the method.

- . It is necessary to discuss the applicability of this method to other landform types, or propose a dynamic threshold adjustment mechanism.

- . The landslide susceptibility map in Figures 5–6 lacks contrast, making it difficult to intuitively distinguish different risk levels.

- . In the discussion part, the manuscript fails to fully compare and analyze relevant research results at home and abroad to highlight the innovations and limitations of this study. When comparing different models, the manuscript lacks systematic comparative analysis and experimental verification to prove the advantages and effectiveness of the proposed model.

- . The discussion of the limitations of the model is very limited, for example, the applicability of the environmental similarity method under extreme climate events is not analyzed, and other sample expansion methods are not compared.

- . The geological and climatic conditions in other regions may be significantly different from those in this region. Therefore, the applicability of the research results in other regions needs further verification.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Comments 1: The manuscript expands sample data through environmental similarity, but does not fully demonstrate the authenticity and rationality of the generated samples. The determination of the σ value in the normal distribution model lacks theoretical explanation or experimental verification, which may lead to similarity calculation deviation.

Response 1: Thank you very much for your comments, this study validates the effectiveness of the generated samples through model verification, demonstrating that the generated samples can enhance the model's generalization ability, indicating that they effectively supplement the deficiencies of the dataset. The method for determining the σ value is elaborated in the section on sample dataset construction (specific details are discussed in section 2.3).

Comments 2: It is necessary to supplement the basis for selecting the σ value and verify whether the distribution of the generated samples is consistent with the characteristics of the real landslide environment.

Response 2: Thank you very much for your comments, the method for determining the σ value has been supplemented in the section on the construction of the sample dataset, which uses the environmental similarity distribution characteristics of known landslide sample points to determine the σ value, ensuring that the expanded samples are consistent with the environmental characteristics of historical landslide samples. (specific details are discussed in section 2.3).

Comments 3: The manuscript uses the normal distribution method to set the similarity threshold, but this method is too simplified and cannot accurately reflect the actual situation of the landslide. In addition, the setting of the threshold is subjective, which affects the rationality of sample expansion and the accuracy of the model.

Response 3: Thank you very much for your comments, the method has been further described in the section on the construction methods of the sample dataset, and the limitations of the existing threshold division methods are discussed in the discussion section (specific details are provided in sections 2.3, 4.1, and 5.1).

Comments 4: Key influencing factors, such as rainfall, soil type, and seismic activity, are not included in the model, resulting in an incomplete calculation of environmental similarity.

Response 4: Thank you very much for your comments, during the process of selecting environmental factors, the weights of eleven environmental factors were calculated, including rainfall, elevation, slope, aspect, curvature, lithology, normalized vegetation index, distance to roads, distance to rivers, distance to faults, and land use types. Factors with lower weights were eliminated. Due to the limited scope of the study area, some factors exhibited insignificant differences within this region (such as rainfall), resulting in a lower influence of these factors (details are explained in section 3.2.2).

Comments 5: Only accuracy, Kappa coefficient, and RMSE are used to evaluate model performance, without considering the problem of class imbalance. High accuracy may mask the model's insufficient prediction ability for minority classes.

Response 5: Thank you very much for your comments, we fully agree with your suggestion and have supplemented the explanation in the sample expansion section regarding the method of threshold division, categorizing landslide susceptibility into five classes. We have expanded the same number of landslide sample data under different susceptibility levels to ensure the balance of sample quantities across different categories. Additionally, we have included the AUC-ROC curve and t-tests for further model accuracy verification (detailed information is provided in section 2.5).

Comments 6: It is necessary to supplement the F1 score, AUC-ROC curve or confusion matrix, and analyze the recall and precision of the model for each category (especially high-risk areas) to more comprehensively reflect the model performance.

Response 6: Thank you very much for your comments, the analysis of the AUC ROC curve reflecting model performance has been added to the model accuracy validation section, along with the addition of the t-test for statistical significance testing, demonstrating the differences in the model (specific details are explained in section 2.5).

Comments 7: The mechanism of BP neural network being superior to SVM and RF when the samples are insufficient is not explained. Does the nonlinear fitting ability match the environmental similarity characteristics?

Response 7: Thank you very much for your comments, when training using only historical landslide data, the parameters of the three models could not be sufficiently optimized, making it difficult to accurately capture the patterns between landslide characteristics and their environmental factors, all demonstrating signs of underfitting. Due to the calculation of environmental similarity involving various environmental factors and the presence of complex nonlinear relationships among these factors (such as the probability of landslide occurrence first increasing and then decreasing with the increase of slope), the BP neural network is better suited to capture these nonlinear relationships.

Comments 8: In addition, it does not mention how to avoid overfitting or optimize the training process. Therefore, it is necessary to supplement the model structure details (such as the number of hidden layers, activation function), the hyperparameter tuning process, and show the training stability through the loss curve.

Response 8: Thank you very much for your comments, in order to optimize the training process of the landslide susceptibility evaluation model and avoid overfitting, this study employed cross-validation, using a ratio of 7:3 to divide the sample data into training and testing sets (details are outlined in section 2.5).

Comments 9: The threshold division method based on normal distribution does not take into account the differences in regional geological conditions, which affects the generalization ability of the method.

Response 9: Thank you very much for your comments, this study analyzes the similarity values of known landslide sample points through the normal distribution method, taking into account the environmental similarities within the region, emphasizing that different areas should consider different environmental characteristics. This has been described in the discussion section, and future research can continue to explore methods for threshold division (specific details are explained in sections 2.3 and 5.1).

Comments 10: It is necessary to discuss the applicability of this method to other landform types, or propose a dynamic threshold adjustment mechanism.

Response 10: Thank you very much for your comments, it has been described in the discussion section: Due to differences in geological characteristics and landslide mechanisms across different regions, the existing threshold division methods still have certain limitations. Therefore, in the future, it will be possible to further optimize the threshold division methods by incorporating multidimensional environmental factors and regional differences (as specifically detailed in section 5.1).

Comments 11: The landslide susceptibility map in Figures 5–6 lacks contrast, making it difficult to intuitively distinguish different risk levels.

Response 11: Thank you very much for your comments, the classification color scheme in Figures 5-6 has been modified to increase contrast, allowing for a more intuitive differentiation of various risk levels (specifically amended to Figures 5 and 6).

Comments 12: In the discussion part, the manuscript fails to fully compare and analyze relevant research results at home and abroad to highlight the innovations and limitations of this study. When comparing different models, the manuscript lacks systematic comparative analysis and experimental verification to prove the advantages and effectiveness of the proposed model.

Response 12: Thank you very much for your comments, the discussion section has been reconstructed, describing the innovations and limitations of this research. When comparing different models, an AUC ROC curve comparison has been added to demonstrate the effectiveness of the models. Additionally, a t-test was employed to assess the statistical significance of the models, proving that there are significant differences in classification performance (specific content can be found in sections 5.1 and 2.5).

Comments 13: The discussion of the limitations of the model is very limited, for example, the applicability of the environmental similarity method under extreme climate events is not analyzed, and other sample expansion methods are not compared.

Response 13: Thank you very much for your comments, the discussion continues to describe the limitations of the research methods, such as the fact that the current experimental results have not yet conducted a thorough analysis of the reasons behind the model's predictive performance, and there is a lack of systematic comparative analysis with other sample expansion or transfer learning methods (as specifically outlined in Section 5.1).

Comments 14: The geological and climatic conditions in other regions may be significantly different from those in this region. Therefore, the applicability of the research results in other regions needs further verification.

Response 14: Thank you very much for your comments, the environmental similarity method quantifies the similarity between evaluation points and historical landslide samples by comprehensively considering multiple environmental factors, with the core assumption being that the historical landslide samples are consistent with the current environmental conditions. Therefore, it is discussed that suitable environmental factors should be chosen for experiments based on the characteristics of different regions (as specifically outlined in Section 5.1).

The English could be improved to more clearly express the research.

Thank you for your suggestion. We will carefully review and refine the English expression to enhance clarity and readability.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editor and Authors,

I have revised the new version of manuscript entitled: Landslide Susceptibility Evaluation Based on the Combined with Environmental Similarity and BP Neural Networks. My opinion is that the quality of the manuscript respect to first version, has been greatly improved considering all the suggestions of the reviewers. The Authors have discussed with great attention and wealth of details the items of the research. The paper appears well constructed, well documented and the results encourage the application of the methods in other areas with different geo-environmental features to verify the reliability of the methodology. Hence, it is recommended that the manuscript now can be accepted for publication in its present form.

Author Response

Dear Reviewer,
We sincerely appreciate your time and effort in reviewing our manuscript. We are grateful for your valuable comments and suggestions, which have significantly contributed to improving the quality of our work. Once again, thank you for your constructive review and support. We look forward to the publication of our work and hope it will contribute meaningfully to the field.
Best regards,
On behalf of all the authors

Reviewer 2 Report

Comments and Suggestions for Authors

Some problems remained to be tackled:

• Historical landslide data are mainly obtained through field measurements and manual visual interpretation, which may introduce subjectivity and error. In addition, the number of historical landslide data is limited, and despite sample expansion through environmental similarity methods, the inadequacy of the original data may still have a certain impact on the accuracy and generalization capabilities of the model.
• Although the paper compares the ESM-BP method with SVM and RF models, the scope of comparison is not wide enough. For example, more types of machine learning models (such as deep learning) may be included for comparison to more comprehensively evaluate the pros and cons of the ESM-BP method.
• The paper selected eight factors to calculate the similarity, but these factors were not sufficient to fully reflect the complexity of landslide occurrence. Other important factors (such as soil type, rainfall, seismic activity, etc.) were not considered.
• The occurrence of landslides is affected by a variety of factors, including geological conditions, climatic conditions, human activities, etc. The paper still hasn't fully captured all the complex mechanisms of landslides.

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

Comments 1: Historical landslide data are mainly obtained through field measurements and manual visual interpretation, which may introduce subjectivity and error. In addition, the number of historical landslide data is limited, and despite sample expansion through environmental similarity methods, the inadequacy of the original data may still have a certain impact on the accuracy and generalization capabilities of the model.

Response 1: Thank you very much for your comments, we fully agree with your viewpoint that historical landslide data is compared and validated in conjunction with field measurements and manual visual interpretation, along with existing geological survey data, to further enhance the available landslide data. Geological survey data utilizes disaster census data, which is further verified through additional measurements to minimize errors. Insufficient original data constrains the accuracy of similarity calculations within the region, potentially lacking certain special landslide types, which results in inadequate generalization capability of the model. On this basis, it is discussed that subsequent research could supplement the original landslide data and types through methods such as transfer learning to improve the model's generalization capability (specific content is in section 5.1, lines 485 to 494).

Comments 2: Although the paper compares the ESM-BP method with SVM and RF models, the scope of comparison is not wide enough. For example, more types of machine learning models (such as deep learning) may be included for comparison to more comprehensively evaluate the pros and cons of the ESM-BP method.

Response 2: Thank you very much for your comments, We also considered this issue during the experimental process. In terms of model selection comparison, the number of samples has a significant impact on model selection and performance. Complex models, such as deep neural networks, require a large amount of training data to prevent overfitting and ensure the model's generalization capability. Conversely, with a smaller sample size, simpler models may be more suitable. The study area only pertains to the reservoir range, which is relatively small. After similarity calculations, the number of landslide samples with extremely high susceptibility levels is limited. Considering the principle of sample balance, it can only be expanded to 4,500 samples. While this may be considered moderate in certain applications, it may still be insufficient for deep learning models. Deep learning models typically require a large amount of data for comprehensive training to avoid overfitting and achieve good generalization performance. Given the existing data scale, it is considered to use relatively simple models, such as random forests, support vector machines, or shallow neural networks. These models have lower sample size requirements and tend to perform more robustly under limited data conditions. The discussion section also elaborates on the potential to expand the research scope in the future, or to combine transfer learning methods to further increase the number of sample points for comparison across more models (specific content can be found in section 5.1, lines 485 to 494).

Comments 3: The paper selected eight factors to calculate the similarity, but these factors were not sufficient to fully reflect the complexity of landslide occurrence. Other important factors (such as soil type, rainfall, seismic activity, etc.) were not considered.

Response 3: Thank you for pointing out the shortcomings in the research. The factors you mentioned, such as soil type, rainfall, and seismic activity, are indeed important environmental factors affecting landslide occurrence. In this study, a weight analysis method was used to filter the environmental factors. However, the research area is located within the Baihetan Reservoir, and according to previous studies, the soil type there is primarily red soil, which is relatively homogenous, and including it in the research would impact the accuracy of the experiments; hence, the soil type factor is not considered. The area is predominantly characterized by bank-type and rainfall-type landslides, so factors such as seismic activity were not considered. Moreover, since the study area only involves the reservoir range, the differences in rainfall within the region are not significant. After calculating the weight factors, it was found that rainfall is assigned a lower weight as an environmental factor; consequently, it was excluded.

Comments 4: The occurrence of landslides is affected by a variety of factors, including geological conditions, climatic conditions, human activities, etc. The paper still hasn't fully captured all the complex mechanisms of landslides.

Response 4: Thank you for your profound insights on the multi-mechanism coupling problem of landslides. This study serves as a preliminary exploration of landslide susceptibility assessment, focusing on the feasibility and foundational role of environmental similarity theory in spatial speculation, while acknowledging that complex disaster-causing mechanisms require ongoing in-depth study, which will be further explored in subsequent research.

Author Response File: Author Response.docx