Predicting Landslide Susceptibility Using Cost Function in Low-Relief Areas: A Case Study of the Urban Municipality of Attecoube (Abidjan, Ivory Coast)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper investigates landslide susceptibility in the Attecoube municipality of Abidjan, Ivory Coast, a low-relief urban environment, using logistic regression and frequency ratio methods combined with GIS techniques. It further introduces a cost-curve function to predict landslide susceptibility for the year 2050. The study holds significant theoretical and practical implications for disaster prevention and mitigation in this region.

While the study presents a compelling approach with valuable contributions, several areas require further refinement and clarification before it can be considered for publication. My comments for improvement are detailed below:

Major Concerns:

1. In the prediction part, the costs of false positives and false negatives are set based on experience. While this is an innovative aspect of the study, it also introduces considerable subjectivity. It is recommended that the authors more explicitly emphasize the hypothetical nature of these cost assignments and their potential impact on the results.
2. Similar to point 1, adjusting the event/non-event sample ratio to 1:50 and assuming that the future landslide occurrence frequency will be 20 times the current rate are very critical assumptions. Although the authors provide an explanation, the uncertainty is extremely high. It is advisable to discuss in more detail how the selected scenarios cover this sensitivity, which would enhance the robustness of the conclusions.
3. Figure 7 in the results section shows different optimal thresholds based on these scenarios (scenarios 1 and 3 correspond to x=0.04, and scenario 2 corresponds to x=0.0541). It is recommended to more clearly explain the logic behind these scenarios and threshold selections in the text and to specify which scenario/threshold-based 2050 susceptibility map is ultimately recommended for decision-making reference or to explain the application scenarios for different threshold maps.

Minor Concerns:

1. The study found that the "urban-related layer" (URL) has no significant impact on landslide occurrence and provided explanations for this. It is suggested that the discussion could briefly mention whether future research might consider other ways to quantify urban impact to more comprehensively assess the influence of urbanization on landslides.
2. There are several instances of expression issues throughout the manuscript. For example, the specific calculation sources for the two results mentioned in the abstract, "61.2% and 20.2," are not explained. Additionally, there seems to be inconsistent use of "URL" and "urban-related layer" throughout the text. Please carefully check the expressions in the paper to ensure clarity and avoid ambiguity.

Author Response

Reviewer 1

We appreciate your review and helpful suggestions, which improve the quality of the paper. Please find below the modifications done. 

Major concerns:

1. In the prediction part, the costs of false positives and false negatives are set based on experience. While this is an innovative aspect of the study, it also introduces considerable subjectivity. It is recommended that the authors more explicitly emphasise the hypothetical nature of these cost assignments and their potential impact on the results.

We acknowledge that the costs assigned to false positives and false negatives in this manuscript are based purely on reasoned assumptions, particularly in the absence of concrete socio-economic data. To clarify this point, we have revised Section 3.7.2 of the manuscript to explicitly highlight the hypothetical nature of these costs.

2. Similar to point 1, adjusting the event/non-event sample ratio to 1:50 and assuming that the future landslide occurrence frequency will be 20 times the current rate are very critical assumptions. Although the authors provide an explanation, the uncertainty is extremely high. It is advisable to discuss in more detail how the selected scenarios cover this sensitivity, which would enhance the robustness of the conclusions.

We rebalanced the samples of events and non-events for several reasons (section 3.7.1). Initially, we worked with a 1:5 ratio. However, we believe the imbalance between events and non-events is likely much greater, as is often the case with rare events. That said, we did not apply a rare event modelling approach, as the sample size available was sufficient to train a standard logistic regression model. We also note that the 1:5 sample ratio is entirely arbitrary.

To maintain a realistic approach, our sample covers several years (approximately 10 years), during which the landslide inventory in this urban area was not systematically compiled. In contrast, the prediction map we have produced is expected to remain valid for several decades, possibly up to 50 years, during which landslide hazards will likely increase due to climate change and the intensification of human activity on the slopes.

The dataset available to us shows that only one in every 1000 pixels experienced a landslide, which is a very low proportion. However, when we account for the longer time horizon of the prediction map, as well as the anticipated impacts of climate change and human interventions, a significantly higher frequency of landslides is expected in Attecoubé. We estimate that the actual frequency could be up to 20 times greater than what is currently observed. Therefore, instead of 1 in 1000, we might reasonably expect 20 in 1000 pixels, or 2%, resulting in an effective ratio of 1:50 between landslide and non-landslide pixels.

3. Figure 7 in the results section shows different optimal thresholds based on these scenarios (scenarios 1 and 3 correspond to x=0.04, and scenario 2 corresponds to x=0.0541). It is recommended to more clearly explain the logic behind these scenarios and threshold selections in the text and to specify which scenario/threshold-based 2050 susceptibility map is ultimately recommended for decision-making reference or to explain the application scenarios for different threshold maps.

In this section of the manuscript, we tested the sensitivity of the threshold to variations in the cost ratio. Initially, the test was carried out using a cost of FN = 300 and FP = 1. Subsequently, a series of tests was conducted with FN values of 250 and 350, and FP values of 1 in both cases. These thresholds yielded more or less similar results and will be taken into account to revise this part of the manuscript.

Minor Concerns:

1. The study found that the "urban-related layer" (URL) has no significant impact on landslide occurrence and provided explanations for this. It is suggested that the discussion could briefly mention whether future research might consider other ways to quantify urban impact to more comprehensively assess the influence of urbanization on landslides.

Indeed, the layer associated with urbanisation (URL) did not show a significant impact on landslide occurrence in this urban area. In future research, the construction of this anthropogenic variable will be improved to better capture its influence on the occurrence of landslides.

From a research perspective, additional variables (as mentioned in the discussion section) will be quantified and integrated into the predictive model to analyse the impact of urban development on slope instability. These may include more dynamic indicators such as the temporal evolution of built-up areas, the density of impervious surfaces, and changes in drainage networks. We believe that such approaches could enrich the analysis in future studies focused on the complex effects of urbanisation. (see lines 934-942)

2. There are several instances of expression issues throughout the manuscript. For example, the specific calculation sources for the two results mentioned in the abstract, "61.2% and 20.2," are not explained. Additionally, there seems to be inconsistent use of "URL" and "urban-related layer" throughout the text. Please carefully check the expressions in the paper to ensure clarity and avoid ambiguity.

We have thoroughly reviewed the entire manuscript to correct any imprecise expressions and improve overall clarity. In particular, we have clarified in the abstract the calculation method that led to the values of 61.2% and 20.2, and added references to the corresponding sections of the main text.

Regarding the inconsistency in the use of the terms “URL” and “layer related to urbanisation,” we have standardised the terminology throughout the document by systematically introducing the abbreviation “URL” at its first occurrence and using it consistently thereafter. These adjustments aim to enhance readability and avoid any confusion for the reader.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Reviewer 2 Report

Comments and Suggestions for Authors

This study focuses on the evaluation of landslide susceptibility (LS) using a logistic regression approach as a statistical method in this area. The dataset is composed of 54 scarps of landslides and 16 thematic layers, including altitude, slope, aspect profile curvature, plan curvature, drainage area, distance to drainage network, normalized difference vegetation index (NDVI), and urban-related layer. A 5-m resolution of DEM produced using several data sources is used in this analysis. Landslides were randomly divided into 80%-20% for model calibration and validation, respectively. After optimising, thanks to tests, thematic layers were integrated and processed to estimate a susceptibility map. The current manuscript lacks sufficient originality and requires major revisions. The following are my views on this study:

1. In the second paragraph of the Introduction section, the classification of landslide susceptibility modelling approaches appears to be rather limited, with relatively few types of models discussed. It is recommended that the authors expand this section by incorporating a broader range of representative modelling techniques.
2. Directional indicators are missing in subfigures (C), (D), and (E) of Figure 1, which may affect the reader's understanding of spatial extent and orientation. It is recommended to add directional indicators in the legend.
3. The coordinate annotations in Figure 2 are incomplete. The vertical coordinate does not clearly indicate the number of landslides. It is suggested that the author add clear coordinate axis labels to enhance the clarity and interpretability of the map.
4. In Figure 3, the north arrow and scale bar in the central subfigure appear too dark, which may hinder visual recognition. It is recommended that the authors adjust the color or contrast to enhance the readability and visual clarity of the figure.
5. The Logistic regression model used in Section 2.6 is a traditional linear statistical method. Although it has a certain application foundation in landslide susceptibility studies, it has limitations in handling complex nonlinear relationships and high-dimensional data. It is recommended that the authors further elaborate on the applicability and advantages of this model in the context of the present study.
6. The landslide susceptibility map presented in Section 3.5 is based on a single model, lacking horizontal comparison with other commonly used methods, which limits the assessment of the model's generalization ability and robustness. It is recommended that the authors include comparative analyses with other models.
7. Some details in Figure 7 are not clearly visible, which may hinder the accurate interpretation of the information presented. It is recommended that the authors improve the image resolution or enhance the visual presentation of key elements such as legends, boundaries, and color distinctions to improve the overall clarity and readability of the figure.
8. Figure 8 presents only two maps, whereas the manuscript refers to three different scenarios. The omission of one scenario is not explained, which may cause confusion for readers. It is recommended that the authors either include the missing map or clearly explain in the caption or text why only two scenarios are shown.

 

Author Response

We appreciate your review and helpful suggestions, which improve the quality of the paper. Please find below the modifications done. 

1. In the second paragraph of the Introduction section, the classification of landslide susceptibility modelling approaches appears to be rather limited, with relatively few types of models discussed. It is recommended that the authors expand this section by incorporating a broader range of representative modelling techniques.

We acknowledge that the landslide modelling approaches mentioned in the introduction appear somewhat limited. We will expand this paragraph to include a more representative range of models.

We have added several landslide modelling methods, such as Artificial Neural Networks (ANN) and Support Vector Machines (SVM), in the second paragraph.

2. Directional indicators are missing in subfigures (C), (D), and (E) of Figure 1, which may affect the reader's understanding of spatial extent and orientation. It is recommended to add directional indicators in the legend.

We acknowledge that the absence of directional indicators in subfigures (C), (D), and (E) of Figure 1 may have limited the readability of the maps. Clear directional markers will be added to enhance the clarity of the legend.

The coordinate annotations in Figure 2 are incomplete. The vertical coordinate does not clearly indicate the number of landslides. It is suggested that the author add clear coordinate axis labels to enhance the clarity and interpretability of the map.

We thank the reviewer for this helpful observation. We have revised Figure 2 to clearly indicate the number of landslides on the vertical axis.

3. In Figure 3, the north arrow and scale bar in the central subfigure appear too dark, which may hinder visual recognition. It is recommended that the authors adjust the color or contrast to enhance the readability and visual clarity of the figure.

We thank the reviewer for this pertinent observation. To improve the readability of Figure 3, we have adjusted the colour and contrast of the arrow indicating north and the scale in the central sub-figure.

4. The Logistic regression model used in Section 2.6 is a traditional linear statistical method. Although it has a certain application foundation in landslide susceptibility studies, it has limitations in handling complex nonlinear relationships and high-dimensional data. It is recommended that the authors further elaborate on the applicability and advantages of this model in the context of the present study.We believe that logistic regression is a traditional linear statistical approach, which may have limitations in modelling complex non-linear relationships and managing high-dimensional data.

However, in the context of our study, the application and advantages of this model are based on a well-defined methodological choice. The implementation of this model was based on a series of considerations, including the analysis of multicollinearity (Pearson's coefficient, evaluation of statistical indicators such as VIF and TOL). This approach aims to reduce the number of explanatory variables and allow for easier interpretation of the coefficients of these predictive variables. The data set at our disposal is not high-dimensional (n variable = 16 and n observation high), which allows for effective training of logistic regression in this manuscript.

5. The landslide susceptibility map presented in Section 3.5 is based on a single model, lacking horizontal comparison with other commonly used methods, which limits the assessment of the model's generalization ability and robustness. It is recommended that the authors include comparative analyses with other models.

We agree that a comparison with other commonly used modelling approaches would strengthen the assessment of the generalisability and robustness of our model. However, this manuscript is not a comparative study of models. The study is based solely on a case study in the municipality of Attecoube. For future studies, we plan to include a comparative analysis of several susceptibility models.

6. Some details in Figure 7 are not clearly visible, which may hinder the accurate interpretation of the information presented. It is recommended that the authors improve the image resolution or enhance the visual presentation of key elements such as legends, boundaries, and color distinctions to improve the overall clarity and readability of the figure.

 We have improved the resolution of Figure 7 and enhanced the visual presentation of key elements, including captions, boundaries, and colour distinctions, to ensure greater clarity and ease of reading of the information presented.

7. Figure 8 presents only two maps, whereas the manuscript refers to three different scenarios. The omission of one scenario is not explained, which may cause confusion for readers. It is recommended that the authors either include the missing map or clearly explain in the caption or text why only two scenarios are shown.

 Clarifications were made to Figure 8 to better illustrate the scenarios.

 

Reviewer 3 Report

Comments and Suggestions for Authors

Landslide is one of the most harmful natural disasters in greater Abidjan, causing serious economic and social losses. In this study, based on the hazards of landslides in greater Abidjan, the authors used the statistical method of logistic regression to evaluate the landslide susceptibility(LS) in greater Abidjan. The results of this study are helpful for land use planning, landslide risk reduction and scientific decision-making in the surrounding areas. Moreover, the author selected the proportion of event samples and non event samples after demonstration, and predicted the new landslide sensitivity in 2050 by using the cost-curve function. However, the manuscript still needs to be further revised before it is officially published.

 

Major comments:

1. It can supplement the comparative analysis with other methods to enhance the effectiveness of the method.
2. The model in the article is idealized and the error analysis is not enough. It is suggested to supplement theoretical analysis and error analysis.

 

Minor comments:

1. The format of the table can be further optimized. For example, each row of ranks in the second column of table 7 is not aligned.
2. There seems to be no uniform format of titles in the paper. Some are top grid, while others are not.
3. The format of the interval symbol "-" in line 354 is incorrect.
4. Some similar investigations on the landslides have been overlooked in the literature review (e.g., Modelling landslide susceptibility prediction: a review and construction of semi-supervised imbalanced theory; On the effects of rheological behavior on landslide motion and tsunami hazard for the Baiyun Slide in the South China Sea; Overcoming the data limitations in landslide susceptibility modeling). These closely related works should be properly reviewed and commented.

Author Response

We appreciate your review and helpful suggestions, which improve the quality of the paper. Please find below the modifications done. 

Major comments:

1. It can supplement the comparative analysis with other methods to enhance the effectiveness of the method.

 We agree that a comparison with other commonly used modelling approaches would strengthen the assessment of the generalisability and robustness of our model. However, this manuscript is not a comparative study of models. The study is based solely on a case study in the municipality of Attecoube. For future studies, we plan to include a comparative analysis of several susceptibility models.

2. The model in the article is idealized and the error analysis is not enough. It is suggested to supplement theoretical analysis and error analysis.

We thank the reviewer for this constructive comment. We do not agree with the simplistic nature of our susceptibility model mentioned in this comment. Indeed, we consider that our dataset (N=443 escarpment pixels) is sufficient and has enabled us to train the logistic regression model to obtain the results presented in the manuscript. In this regard, our results have been compared to other studies (Conforti and Ietto, 2021; Nhu et al., 2020), and it appears that the performance of our models (AUC>0.90) is similar. These results demonstrate the rigour and robustness of our research in this study area. 

Minor comments:

1. The format of the table can be further optimized. For example, each row of ranks in the second column of table 7 is not aligned.

We have formated the table appropriately to improve the manuscript.

2. There seems to be no uniform format of titles in the paper. Some are top grid, while others are not.

We have formated the titles appropriately to improve the manuscript.

3. The format of the interval symbol "-" in line 354 is incorrect.

A correction has been made.

4. Some similar investigations on the landslides have been overlooked in the literature review (e.g., Modelling landslide susceptibility prediction: a review and construction of semi-supervised imbalanced theory; On the effects of rheological behavior on landslide motion and tsunami hazard for the Baiyun Slide in the South China Sea; Overcoming the data limitations in landslide susceptibility modeling). These closely related works should be properly reviewed and commented.

We would like to highlight that the studies mentioned make significant contributions to the field of landslide susceptibility modelling. We would like to point out that our study was conducted in an intertropical zone of West Africa. This zone is characterised by low relief and significant human activity. We will enrich the literature review by including a targeted discussion of these studies, highlighting their specific contributions and their link to our approach, in order to better situate our study in the context of existing research.

 

 

Reviewer 4 Report

Comments and Suggestions for Authors

I found interesting the manuscript "Manuscript ID earth-3686671" entitled "Predicting landslide susceptibility and cost functions analysis in low-relief areas: A case study of urban environments municipality of Attecoube (Abidjan, Ivory Coast)" has been submitted to Earth (ISSN 2673-4834). The authors focused on the evaluation of landslide susceptibility (LS) using a logistic regression approach as a statistical method in this area. The dataset is composed of 54 scarps of landslides and 16 thematic layers, including altitude, slope, aspect profile curvature, plan curvature, drainage area, distance to drainage network, normalized difference vegetation index (NDVI), and urban-related layer. A 5-m resolution of DEM produced using several data sources is used in this analysis. Landslides were randomly divided into 80%-20% for model calibration and validation, respectively. After optimizing, thanks to tests, thematic layers were integrated and processed to estimate a susceptibility map. 6.3% or 0.7 km2 of the study area has a very high susceptibility, with values of 61.2% and 20.2 attributed to the proportion of landslides and frequency ratio, respectively. LS mapping results indicated that altitude, slope, SE, S, NW, and NDVI positively impacted landslide occurrence in this urban area. The model's performance is evaluated using the Area Under the Receiver Operating Characteristic curve (AUC). However, there are still major deficiencies, and it is recommended to be accepted after modification:

1. While the abstract is well-structured, a few elements could enhance its completeness and clarity:

• The abstract is dense. While technically detailed, it could benefit from breaking up long sentences and simplifying language for better readability.
• Indicating the specific time period during which the data was collected (e.g., "from January 2020 to December 2021") would enhance clarity.
• While the results are presented, the abstract doesn’t end with a strong concluding sentence summarizing the broader impact or potential future applications of the proposed method. For example, “introducing a future susceptibility map for 2050 near the end feels sudden. Either briefly explain how it's derived or omit it if it’s not a major contribution”.

2. The introduction is thorough and provides a detailed background about Predicting landslide susceptibility and cost functions analysis, their significance, and the methodologies used in this study. However, there are still a few areas that could be enhanced to ensure completeness and clarity:

• While the introduction details the methodologies, it should more explicitly connect these methods to the research objectives. Explain how these methods specifically help address the research questions or gaps identified.
• Briefly mention the expected outcomes or implications of the study. How will the results contribute to seismic risk assessment and mitigation strategies in the study area? What practical applications might emerge from this research?

3. I will recommend to add study area in the introduction section or a separate section, not in the Materials and Methods section.
4. Figure 1 should provide more detailed context. It’s very important for readers to understand the figure easily.
5. The captions of all tables are not up to the mark and should be explained further.
6. Please correct “Tableau 12” to “Table 12”
7. The Results and discussion provided is detailed and comprehensive, but it can be improved by Comparing your findings with those of previous studies to highlight the contributions and advancements your research has made. Explicitly state any limitations or potential sources of error in your study and their impact on the findings. Suggest specific areas for further investigation based on the results and limitations of your study.
8. The conclusions provided summarize the main findings and their implications well but can be enhanced by including a few additional elements to provide a more comprehensive closure to the study.

• Conclusion section is very dense, make it more precise and easier for readers.
• End with a strong concluding remark that reinforces the importance of study and its contributions to the field.

9. Further, on numerous occasions I had trouble with the language and I also found the repetition of sentences. A careful reading of the paper is mandatory to remove typos and to soften the language hardships. So, it is my opinion that the paper should undergo major revision for a better explanation and for a complete revision of the language.
10. Check all references and include recent studies to ensure the literature review is up to date.

Author Response

We appreciate your review and helpful suggestions, which improve the quality of the paper. Please find below the modifications done. 

1. While the abstract is well-structured, a few elements could enhance its completeness and clarity:

• The abstract is dense. While technically detailed, it could benefit from breaking up long sentences and simplifying language for better readability.
• Indicating the specific time period during which the data was collected (e.g., "from January 2020 to December 2021") would enhance clarity.
• While the results are presented, the abstract doesn’t end with a strong concluding sentence summarizing the broader impact or potential future applications of the proposed method. For example, “introducing a future susceptibility map for 2050 near the end feels sudden. Either briefly explain how it's derived or omit it if it’s not a major contribution”.

• We have taken into account your suggestions that we find relevant. Short, effective sentences have been used in the abstract to make it easier to read.
• The period during which the escarpment data was collected has been mentioned in the abstract.
• This comment has been taken into account at the end of the summary.

2. The introduction is thorough and provides a detailed background about Predicting landslide susceptibility and cost functions analysis, their significance, and the methodologies used in this study. However, there are still a few areas that could be enhanced to ensure completeness and clarity:

• While the introduction details the methodologies, it should more explicitly connect these methods to the research objectives. Explain how these methods specifically help address the research questions or gaps identified.
• Briefly mention the expected outcomes or implications of the study. How will the results contribute to seismic risk assessment and mitigation strategies in the study area? What practical applications might emerge from this research?

To address the gaps identified in this research, a series of steps was taken. The first step was to conduct an inventory for the period from 2015 to 2023, providing a reliable set of data. The second step involved collecting factors that could influence the occurrence of landslides in the area. This data set was then used in GIS-supported models to assess landslide susceptibility.

3. The results obtained will be very useful for authorities involved in urban planning decisions, enabling them to better plan the terrain and reduce potentially vulnerable areas through concrete actions.

I will recommend to add study area in the introduction section or a separate section, not in the Materials and Methods section.

Suggestion taken into account, see text

4. Figure 1 should provide more detailed context. It’s very important for readers to understand the figure easily.

Suggestion taken into account see text

5. The captions of all tables are not up to the mark and should be explained further.

Suggestion taken into account see text

6. Please correct “Tableau 12” to “Table 12”

            Suggestion taken into account see text

7.  The Results and discussion provided is detailed and comprehensive, but it can be improved by Comparing your findings with those of previous studies to highlight the contributions and advancements your research has made. Explicitly state any limitations or potential sources of error in your study and their impact on the findings. Suggest specific areas for further investigation based on the results and limitations of your study.   We thank the reviewer for this pertinent comment. We agree that a comparison with other commonly used modeling approaches would strengthen the assessment of the generalisability and robustness of our model. However, this manuscript is not a comparative study of models. The study is based solely on a case study in the municipality of Attecoube. In a future study, we plan to include a comparative analysis of several susceptibility models.  We intend to apply this logistic regression model in the municipality of Abobo, an area located in the north of Greater Abidjan. This area is also affected by landslides similar to those triggered in the municipality of Attécoube.

8. The conclusions provided summarize the main findings and their implications well but can be enhanced by including a few additional elements to provide a more comprehensive closure to the study.

• Conclusion section is very dense, make it more precise and easier for readers. End with a strong concluding remark that reinforces the importance of study and its contributions to the field.

Suggestion taken into account, see text

9.  Further, on numerous occasions I had trouble with the language and I also found the repetition of sentences. A careful reading of the paper is mandatory to remove typos and to soften the language hardships. So, it is my opinion that the paper should undergo major revision for a better explanation and for a complete revision of the language.

Suggestion taken into account, a correction of English has been done, see text

10. Check all references and include recent studies to ensure the literature review is up to date.

Suggestion taken into account, see text

 

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have adequately addressed all of my previous comments and concerns. The revised manuscript is significantly improved and now meets the standards for publication in Earth. I therefore recommend acceptance of the manuscript in its current form.

Reviewer 2 Report

Comments and Suggestions for Authors

I agree to accept after minor revisions.

Reviewer 3 Report

Comments and Suggestions for Authors

The revised paper is well revised, and could be accepted.

Reviewer 4 Report

Comments and Suggestions for Authors

The suggested changes to the text were partially adopted and the text was improved in its structure and citations. From this second evaluation, I consider that the article has the quality to be published.