Landslide Prediction Validation in Western North Carolina After Hurricane Helene

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors

few comments for this report.

Thanks

1) Fig. 13. Put in caption meaning of LE and LWE

2) Comment if any mutual orientation of mountain slope  vs burden structure has represented a condition favourable for surface slip.

3) In Abstract and also in Conclusion recall intensity of rainfall (magnitude) for studied zone. Consider that values of rainfall are today given in mm in scientific papers (metric system).

Author Response

Thanks

• 13. Put in caption meaning of LE and LWE.

Reply:

Thank you for the suggestion. We put the caption meaning in Figure 13 caption.  Also updated captions on other figures as well.

• Comment if any mutual orientation of mountain slope  vs burden structure has represented a condition favourable for surface slip.

Reply:

Yes, we have observed mountain slope and burden structure orientation plays a critical part in the Aspect parameters.  However, because of the ‘black box” approach using Random Forest and Linear Regression modeling, the effects are not visible.  We are planning to investigate further – but it is beyond the scope of the current paper.

• In Abstract and also in Conclusion recall intensity of rainfall (magnitude) for studied zone. Consider that values of rainfall are today given in mm in scientific papers (metric system).

Reply:

Thank you for the great suggestion.  This has been added in the Conclusion.  We also converted all the reported values in metrics elsewhere in the paper.

Reviewer 2 Report

Comments and Suggestions for Authors

Hurricane Helene caused widespread landslides and bridge damage in western North Carolina due to heavy rainfall and flooding. The study by Lin et al. uses machine learning to analyze landslide susceptibility, highlighting the importance of comprehensive hazard mapping for infrastructure resilience.

 

Machine learning has a wide range of applications in geosciences and has also been used in disaster prevention and mitigation. As a reviewer, I believe the authors' approach enhances the efficiency of landslide prediction, which is highly significant. However, I hope the authors can further discuss the advantages of machine learning methods compared to traditional approaches.

 

Other minor comments are as follows:

 

1) The different points (Categories) in Figure 1 represent the intensity of the hurricane at different moments. I suggest the authors label the specific times to better reflect its development process.

 

2) In the introduction, the authors could briefly describe the applications of machine learning in disaster prevention for natural hazards, such as tsunami prediction. For reference, the following articles can be referred.

https://doi.org/10.1186/s40623-023-01912-6

 

3) In Section 2.2, how does the nested multi-hazard modeling approach account for potential interactions between contributing factors, such as rainfall and wildfire, which might amplify the risk of landslides?

 

4) Why is the impact of wildfires discussed separately? Are there other factors that also warrant individual analysis?

 

5) Line 285, how to differentiate the impacts of landslides and scour on bridge damage, and what methods are used to accurately distinguish between these two types of damage in the study?

 

6) As the authors mentioned, one of the challenges in machine modeling for landslide susceptibility assessment is the issue of sampling. Could satellite remote sensing be used to increase the number of observations, thereby making the sampling more accurate?

 

Please consider my review comments as the encouragement. Look forward to your response.

Author Response

Comments and Suggestions for Authors

Machine learning has a wide range of applications in geosciences and has also been used in disaster prevention and mitigation. As a reviewer, I believe the authors' approach enhances the efficiency of landslide prediction, which is highly significant. However, I hope the authors can further discuss the advantages of machine learning methods compared to traditional approaches.

Reply:

Thank you for the comment.  Advantages of ML versus traditional approaches are still a much-debated topic in Geosciences.  Instead of engaging in a deep discussion, we offer to add a statement in the Introduction section (lines 90-94): “ML is a rapidly developing field for handling large data sets, especially in the fields of classification and feature identification that are too  tedious for human operations.  Geospatial modeling has increasingly leveraged machine learning techniques to predict extreme events, such as landslides, wildfires, floods, and earthquakes.”

Other minor comments are as follows:

 1) The different points (Categories) in Figure 1 represent the intensity of the hurricane at different moments. I suggest the authors label the specific times to better reflect its development process.

 Reply:

Thank you for the great suggestion. We put the specific times label into figure 1.

2) In the introduction, the authors could briefly describe the applications of machine learning in disaster prevention for natural hazards, such as tsunami prediction. For reference, the following articles can be referred.

https://doi.org/10.1186/s40623-023-01912-6

Reply:

Thank you for your comment. We have included some discussion on the application of machine learning in disaster prevention for natural hazards in the introduction. The content is “Geospatial modeling has increasingly leveraged machine learning techniques to predict extreme events, such as landslides, wildfires, floods, and earthquakes”.

3) In Section 2.2, how does the nested multi-hazard modeling approach account for potential interactions between contributing factors, such as rainfall and wildfire, which might amplify the risk of landslides?

Reply:

In our nested multi-hazard modeling approach, rainfall and wildfire can trigger landslide occurrences. However, the potential interaction between rainfall and wildfire is not considered in this paper. In future work, we plan to incorporate real-time data and account for this potential interaction.

4) Why is the impact of wildfires discussed separately? Are there other factors that also warrant individual analysis?

 Reply:

Post-wildfire events have been found to be increasingly important for triggering landslide occurrences at several parts of the world and a great concern in North Carolina. Several studies have highlighted this situation, including similar research conducted by the USGS. Therefore, we discussed this aspect in our paper.

5) Line 285, how to differentiate the impacts of landslides and scour on bridge damage, and what methods are used to accurately distinguish between these two types of damage in the study?

 Reply: Thank you for your comment and this is an important topic but still poorly understood – We have not been able to identify any literature on this subject matter.  We hope soon to develop techniques to differentiate these two.

6) As the authors mentioned, one of the challenges in machine modeling for landslide susceptibility assessment is the issue of sampling. Could satellite remote sensing be used to increase the number of observations, thereby making the sampling more accurate?

Reply:

Thank you for your comment.  In the case of Helene, unfortunately in September, tree cover is still an issue for remote sensing. However, remote sensing such as satellite imaging will increase the number of observations and accuracy.

Reviewer 3 Report

Comments and Suggestions for Authors

This study used machine learning (ML)-generated multi-hazard landslide susceptibility maps. They were compared to the documented landslides from Helene. The landslide models used the North Carolina landslide database, soil survey, rainfall, USGS digital elevation model (DEM), and distance to rivers to create the landslide variables. The scinetific content is all good. The serious problem is the writing of the manuscript, as it does not follow the convention very well and many essential parts are missing. Current manuscript is more like a student's report than a paper. My major concerns are as follows:

1. The Introduction part is not well organzied. There is too much background infomration. The literature review and contribution can not be found.

2. The methods of random forest and logistic regression are missing. What werr your inputs and ouputs? Did you perform cross validation? Why did you mention the previous studies in Methods part? It seems that Bjånes et al. among others have already conduct the alike studies. Why did not you choose CNN as suggested by previous study? 

3.How do you calculate the probabiity of hazard? How can you verify your results of susceptibility probabilities?

4. Why does field investigation of the landslides are included in your manuscript? Does it help to demonstrate the topic of this study?

Author Response

Reviewer 3:

Comments and Suggestions for Authors

1. The Introduction part is not well organzied. There is too much background infomration. The literature review and contribution can not be found.

Reply: Thank you for your candid comment.  We follow the reporting format of several USGS publications on landslides where descriptions should follow a geological description of the site condition.  Since a major hurricane is involved in this major disaster, an elaboration on the hurricane genesis should be included.  The formatting may be confusing because we are a team from multi-disciplines including hurricane disaster investigation, landslide, geology, bridge engineering and machine learning.  We made an attempt to format the introduction more systematically.

2. The methods of random forest and logistic regression are missing. What werr your inputs and ouputs? Did you perform cross validation? Why did you mention the previous studies in Methods part? It seems that Bjånes et al. among others have already conduct the alike studies. Why did not you choose CNN as suggested by previous study?

Reply: The focus of this paper is on the verification of our modeling outcomes and the extreme disaster event. Details about random forest and logistic regression have been reported in our previous publication in Geohazards: Lin, S., S.-E. Chen, W. Tang, V. Chavan, N. Shanmugam, C. Allan, and J. Diemer, Landslide Risks to Bridges in Valleys in North Carolina. GeoHazards, 2024. 5(1): p. 286-309.

3.How do you calculate the probabiity of hazard? How can you verify your results of susceptibility probabilities?

Reply:

Thank you for your comment.  In the previous work we discussed how we calculated the probability of hazard using historical landslides, which is consistent with other literatures. Our work has been reported in previous publication: https://www.mdpi.com/2624-795X/5/1/15

To verify our hazard prediction, we used all the reported 1,792 landslides and the damaged bridges, based on data from the USGS, NCDOT.   Our prediction is quite accurate.

4. Why does field investigation of the landslides are included in your manuscript? Does it help to demonstrate the topic of this study?

Reply:

Thank you for your comment. The field investigation of landslides can validate the results of our susceptibility map, as areas with higher probability are more likely to contain landslide sites.  As we stated in the paper, rarely has landslide susceptibility studies get to validate their results with actual landslides, most research work relied on matching historical landslides. 

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has been sufficiently improved to warrant publication in Geotechnics.