Landslide Risk Assessment in Eastern Kentucky, USA: Developing a Regional Scale, Limited Resource Approach

Round 1

Reviewer 1 Report

Dear Authors,

It was a peasure to read interesting and well-presented paper.

I recommend to accept it.

Jusy one moment -  Line 243. The susceptibility  model is based on geomorphic ...

Author Response

Dear reviewers,

Thank you for the thorough and thoughtful comments, edits, and suggestions. We have addressed your feedback through organizational changes (removal and addition of sections/subsections) and by adding clarifying text to bolster central topics. We think the changes make this manuscript significantly clearer and demonstrate an advancement of landslide risk mapping, considering the many approaches and data limitation challenges.

Reviewer 1 response included here. Please see the uploaded document that preserves the tracked changes to the manuscript.

REVIEWER 1

Corrected line 243.

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript intends to assess landslide risk for a landslide-prone region of eastern Kentucky, by considering formal risk components including hazard, vulnerability, and consequences. This is a typical application study. The applied methods are not novel, and had been used in several published studies. The use of 30 m slope-angle map as the hazard input is somehow interesting. I encourage the authors to consider the more detailed comments below as major revisions and to improve the manuscript before publication in Remote Sensing.

1. Abstract. The abstract is not well structured. The to-be-solved question showed be defined and answered in the abstract. The title of manuscript is “Landslide Risk Assessment in Eastern Kentucky, USA: Developing a Regional Scale, Limited Resource Approach”. I assume “data-limited approach” is the key point. However, the development and application of a “data-limited approach” have not been sufficiently illustrated in the abstract. Similar problems exist in the Introduction.

2. Introduction. It is suggested to complement the literature review about the subject and applied methods. Some up-to-date and review papers discussing impacts of landslide hazards on society, economy and engineering projects, and those discussing methods of landslide risk assessment, should be cited.

Rossi, M., Guzzetti, F., Salvati, P., et al. (2019). A predictive model of societal landslide risk in Italy. Earth-Science Reviews, 196, 102849.

Lan, H., Tian, N., Li, L., et al. (2022). Kinematic-based landslide risk management for the Sichuan-Tibet Grid Interconnection Project (STGIP) in China. Engineering Geology, 308, 106823.

3. Landslide susceptibility assessment approach. How do you select landslide locations for testing and training? The description of applied methods is not enough for readers to understand, and should be extended. As mentioned above, publications about landslide susceptibility assessment approaches should be cited.

4. Results. The “Results” section needs a revision. The descriptions of processing models and results are inadequate.

5. Discussion. This section should justify and demonstrate the novelty of your work over previous studies. A discussion about the advantages and disadvantages of the used approaches should be added in this section.

6. Conclusions. The conclusions of this manuscript must be more concise and straightforward. The obtained results should not be simply considered as conclusions. Please show readers that you have solved the to-be-solved problem.

Author Response

Dear reviewers,

Thank you for the thorough and thoughtful comments, edits, and suggestions. We have addressed your feedback through organizational changes (removal and addition of sections/subsections) and by adding clarifying text to bolster central topics. We think the changes make this manuscript significantly clearer and demonstrate an advancement of landslide risk mapping, considering the many approaches and data limitation challenges.

Reviewer 2 responses are included here. Please see the uploaded document that preserves the tracked changes.

REVIEWER 2

This manuscript intends to assess landslide risk for a landslide-prone region of eastern Kentucky, by considering formal risk components including hazard, vulnerability, and consequences. This is a typical application study. The applied methods are not novel, and had been used in several published studies. The use of 30 m slope-angle map as the hazard input is somehow interesting. I encourage the authors to consider the more detailed comments below as major revisions and to improve the manuscript before publication in Remote Sensing.

1. Abstract. The abstract is not well structured. The to-be-solved question showed be defined and answered in the abstract. The title of manuscript is “Landslide Risk Assessment in Eastern Kentucky, USA: Developing a Regional Scale, Limited Resource Approach”. I assume “data-limited approach” is the key point. However, the development and application of a “data-limited approach” have not been sufficiently illustrated in the abstract. Similar problems exist in the Introduction.

Thank you for this critical feedback. We expanded the concept of limited data in two ways, by adding a sentence in the abstract and by adding additional text throughout the manuscript (see tracked changes) that emphasizes the application of a data limited approach and how it is the central issue.

2. Introduction. It is suggested to complement the literature review about the subject and applied methods. Some up-to-date and review papers discussing impacts of landslide hazards on society, economy and engineering projects, and those discussing methods of landslide risk assessment, should be cited.

Rossi, M., Guzzetti, F., Salvati, P., et al. (2019). A predictive model of societal landslide risk in Italy. Earth-Science Reviews, 196, 102849.

Lan, H., Tian, N., Li, L., et al. (2022). Kinematic-based landslide risk management for the Sichuan-Tibet Grid Interconnection Project (STGIP) in China. Engineering Geology, 308, 106823.

Thank you for these high-quality citation suggestions. Rossi et al. combines several robust hazard and exposure datasets, focusing on fatalities. Lan et al. emphasizes landslide kinematics and a specific infrastructure type. Both papers are good examples of quantitative landslide risk assessments that also acknowledge the limitations of modeling and communicating risk.

We added these citations and reworded several lines in the introduction to reference this work. The subsection titles were removed from the introduction to highlight the organization and manuscript flow by background and methods. We made the Study Area its own section that includes Geology, and Study Area Impact.

3. Landslide susceptibility assessment approach. How do you select landslide locations for testing and training? The description of applied methods is not enough for readers to understand, and should be extended. As mentioned above, publications about landslide susceptibility assessment approaches should be cited.

Based on this suggestion, we added text in the Section 3.1.1 to better define the methodology and landslide mapping (This section has a new number because we added Study Area as a separate section).

The landslides in the susceptibility study were mapped using lidar and lidar-derivative maps. We mapped over 1,000 landslides in a Kentucky county. Most landslides are thin, translational colluvial landslides, however specific type was not taken into consideration with the modeling. We conducted a two-step machine-learning approach, first using bagged trees and second using logistic regression. The bagged trees (ensemble decision tree classification) binary landslide data was divided into training (75%) and testing (25%) datasets. The bagged-trees model estimates feature importance from the entire statistical dataset of 36 geomorphic variables and guides the logistic regression. We also validated the logistic regression results on a separate landslide inventory.

We cite the landslide susceptibility study in the paper [43].

Crawford, M.M., Dortch, J.M., Koch, H.J., Killen, A.A., Zhu, J., Zhu, Y., Bryson, L.S., and Haneberg, W.C. Using landslide-inventory for a combined bagged-trees and logistic regression approach to landslide susceptibility in eastern Kentucky, United States. Quarterly Journal of Engineering Geology and Hydrogeology, 2021, http://doi.org/10.1144/qjegh2020-177.

We also acknowledge the limitations with landslide susceptibility modeling. For example, data accuracy and resolution of terrain models is a limitation with statistical and geomorphic-based modeling. For physics-based approaches, particularly at a regional scale, limitations include requiring specific knowledge of soil properties, hydrological conditions, and geotechnical inputs into slope-stability models. These data are typically not available on a regional basis or at the catchment scale, as is the case in this study.

4. Results. The “Results” section needs a revision. The descriptions of processing models and results are inadequate.

To address this feedback, we added a new subsection in the Materials and Methods section called Risk Model Estimation. Pertinent text was moved from Results to the new Risk Model Estimation section. This separates the modeling details from the results. We think this also emphasizes the comparison between risk estimations and resulting maps, justifying the changes in model estimation inputs.

5. Discussion. This section should justify and demonstrate the novelty of your work over previous studies. A discussion about the advantages and disadvantages of the used approaches should be added in this section.

To emphasize the advantages and disadvantages of this approach we added text to first paragraph of the Discussion about the primary drawback for all type of landslide risk. This sets up the discussion section to revolve around limited resources and the novelty of that approach.

We also discuss the specific disadvantages with each approach. For example, in using a static, geomorphic-based landslide susceptibility input does not include information about landslide type or runout (lines 589-590). We also discuss disadvantages to our elements at risk which did not include critical elements such as powerlines, water lines, and sewage lines (lines 593-594). These data, if available, would certainly influence the outcome of the risk estimation. We think the main disadvantage relevant to all landslide risk approaches (but particularly our socio-economic approach) is accounting for changing cultural, physical, and atmospheric conditions. Many additional data sets should be considered in future risk assessments (lines 611-613).

Recognizing limitations and producing risk maps based on limited resources can still be advantageous and provide an important framework or baseline of information for communities in need (last paragraph in Discussion).

6. Conclusions. The conclusions of this manuscript must be more concise and straightforward. The obtained results should not be simply considered as conclusions. Please show readers that you have solved the to-be-solved problem.

Thank you for highlighting this distinction. To address this, we moved text to beginning of section to emphasize how this approach advances the science and addresses  challenges of landslide risk. We also removed some text so that the conclusion section is more concise.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

the manuscript titled " Landslide Risk Assessment in Eastern Kentucky, USA: Developing a Regional Scale, Limited Resource Approach" presents an interesting study investigating the risk assessment with different quality inputs. A particular attention is given to the comparison between a Landslide susceptibility (1.5m lidar) and Slope map (30m SRTM).

Literature is broad and complete, but the structure of the manuscript should be modified. Discussion and conclusion may be more extensive, and they could consider the effects and impact of this specific research.

The subject is of interest to Remote Sensing, but the manuscript needs some major revisions prior to be accepted to be published.

I have marked some corrections and I penned my comments on the attached pdf copy of the manuscript.

Please consider the attached pdf file.

Regards

Comments for author File: Comments.pdf

Author Response

Dear reviewers,

Thank you for the thorough and thoughtful comments, edits, and suggestions. We have addressed your feedback through organizational changes (removal and addition of sections/subsections) and by adding clarifying text to bolster central topics. We think the changes make this manuscript significantly clearer and demonstrate an advancement of landslide risk mapping, considering the many approaches and data limitation challenges.

Reviewer 3 responses are included here, but also in the uploaded annotated pdf. Please see the uploaded Word document that preserves the tracked changes in the entire manuscript.

REVIEWER 3

Major changes listed here (these are also documented in the annotated .pdf)

• Removed subsections in the Introduction
• Added Study Area as its own section, with Geology and Impact as subsections
• Moved Equation 1 to beginning of Materials and Methods section. We also created a new Risk Model Estimation section (3.3). This clarifies the calculations and separates the modeling details from the results. We think this also emphasizes the comparison between risk estimations and resulting maps, justifying the changes in model estimation inputs.
• Added (a), (b), and (c) to Figure 1, changed the caption
• Added text in section Landslide Susceptibility Approach section (new section 3.1.1).
• Added text to expand on the landslide susceptibility modeling.

Most landslides in the area are thin (<3m thick) translational slides. Slide type, age, and potential extent was not considered in the modeling. We conducted a two-step machine-learning approach, first using bagged trees and second using logistic regression. The bagged trees (ensemble decision tree classification) binary landslide data was divided into training (75%) and testing (25%) datasets. The bagged-trees model estimates feature importance from the entire statistical dataset of 36 geomorphic variables and guides the logistic regression. We also validated the logistic regression results on a separate landslide inventory.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Dear Authors,

The new version of the manuscript has much better clarified the wording and the questions proposed. The authors took into account my comments and requests, and the structure of the manuscript has benefited from it. 
Overall, I found the manuscript improved.