Comparisons of Convolutional Neural Network and Other Machine Learning Methods in Landslide Susceptibility Assessment: A Case Study in Pingwu

Round 1

Reviewer 1 Report

The authors applied machine learning methods in landslide susceptibility assessment and compared Convolutional Neural Network with six traditional machine learning methods. The authors may add the motivation for these comparisons in the “1 Introduction” part. In the “2.2 Data” part, the authors may add more description about NDVI, like how this index is defined and derived, what the data source is for this index. The authors may add more discussion in the “3.1 Analysis of conditioning factors” on why NDVI has the highest contribution rate. In the 2.3.9 Samples construction part, the authors may specify the reason for selecting 7*7 window size. Have other window sizes been tested?

Line 305, 343: “Error! Reference source not found” should be deleted.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Review of the manuscript “Comparison of Convolutional Neural Network and Other Machine Learning Methods in Landslide Susceptibility Assessment: A Case Study in Pingwu”

The manuscript showed a study case of a landslide susceptibility assessment. The dataset includes landslide points and landslide conditioning factors.

The manuscript fits very well with the scope of Remote Sensing and it has a significant contribution in terms of using different approaches. Overall, in my opinion, this manuscript is acceptable for publication in this high-standard journal. My remarks for the authors to improve the manuscript are in the comments.

Comments

Considering the temporal resolution of the dataset, it is important to highlight the variables that could change over time.  In this case, the NDVI is prone to change. If we look into contribution analysis, NDVI is the one with the greatest influence on the landslide. Considering this the authors could:

1) Incorporate temporal tests considering seasonality. Just elaborate on another experiment changing the NDVI temporal input. Would it change the results? Is the NDVI close to the rainfall season differ from other seasons?

2)  Considering the importance of rainfall for landslides, you could improve the description of it in the study area. Add in more details, the precipitation conditions over time. Also, incorporate a better report on earthquakes considering the study area and its historical assessment and probability of occurrence; this will enhance the paper for the reader.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Well done. Very interesting paper.

Line 305: Error message must be corrected.

Line 343: Error message must be corrected.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

This research compared the performance of the Convolutional Neural Network and Other Machine Learning Methods in Landslide Susceptibility Assessment. It is benefit for disaster reduction and social stability. Nevertheless, there are still some aspects that the authors should pay attention to:

(1) Line 339: how to determine the correctly discriminated samples?

(2) Figure 6: the cut values are different for various methods. Please give the reasons that this research chooses the natural break method.

(3) There are mistakes in Line 305 and Line 343. Please check the similar problems

(4) Figure 1 shows there are large number of landslides in the middle low part, whereas only scarce landslides in the southeastern high mountains. Meanwhile, elevation is a relatively important factor (Figure 5). Please check this, it is not usual.

(5) The quality of some figures should be improved, such as Figure 8.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

The authors have made corresponding revisions according to the comments.  Now that the equation of the NDVI is given, so I suggest other equations, such as Stream Power Index (SPI). Moreover, the quality of the figures can be improved further. They are not very clear at their current form.

Author Response

Please see the attachment.

Author Response File: Author Response.docx