Landslide Recognition Based on Machine Learning Considering Terrain Feature Fusion

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

Please see attached.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

please see attached.

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

 

The manuscript has been considerably improved from the initial version. I acknowledge the effort to improve it. This new version has a better introduction section, containing a more complete state of the art and references, the aims and novelty of the study. The discussion section and conclusions have been improved too. All those improvements help to understand the study and to identify the contribution of authors.

Some minor editions are needed:

·         Figure 15 boundary instead boundry

·         Use lowercase “m” for meters.

·         In order to clarify what is the P/N ratio used for training data, please, include in figures 15 and 16 (or in its caption) this information. For example: LightGBM_20_5 (ratio 1:5)

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

The authors have revised the manuscript thoroughly and incorporated all the suggestions as per the previous review.

So no further comments from my side.

Comments on the Quality of English Language

Good

Author Response

Please see the attachment.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Paper Review: “landslide recognition considering terrain features supported by multiple machine learning methods and convolutional neural network” ijgi-3030597

 

Comments: The paper summarizes an interesting study of using several machine learning methods to train landslide data.  The paper’s contribution is important and   

 

Some suggestions are listed below:

 

1.      In the Abstract and Introduction sections, the authors have mentioned about one of the deficiencies of landslide classification as: “In addition, the ratio of positive and 15 negative samples in training data for landslide identification is often 1:1…” (line 15). Do you mean the sample data are differentiated into Y/N types? Please clarify a little more.

2.      The authors emphasize the performance of two particular training methods and claimed that: “Among them, LightGBM performed the best with evaluation 26 indicators of 97.47%, 85.40%, 76.95%, 80.95%, and 71.28%, respectively, while RF performed the worst with 27 evaluation indicators of 97.11%, 82.50%, 74.45%, 78.27%, and 67.50%, respectively.” (line 27-28)  It does not strike me as the differences can substantiate a strong argument about the different method performances (97.47% versus 97.11% to me is almost equally well).  The authors may want to discuss about the sensitivity of their modeling techniques.

3.      “After the occurrence of landslide geohazards, the overlying surface features of the landslide 131 body will change to varying degrees.” (Line 131), varying degrees as in “slope”? Please clarify.

4.      Tables 3 to 7, may be better to list the lower ranking results first than the higher ranking data. For example, CNN_14 before CNN_20.

5.      If the authors can highlight on Figures 8 to 12 how the best result images different from the poorer images (such as identify the groundtruth landslide locations), would greatly enhance reader appreciation of their contributions. “A good picture is worth a thousand words.” The authors can either describe the differences within the figure label or use indicators in the images to show the differences between good and bad results.

6.      Figure 13 is very important, but the labels in the figures are too small to read.

7.      “Salt and pepper images” (line 363) may be better described as “speckled” or “grainy” images.

8.      Because the outcomes from all the test techniques are all very close to each other, it is not clear to the reviewer the significant contribution from this paper.  How about comparing the computational efforts from each method? 

9.      The authors are recommended to think hard what has been accomplished here?

Comments on the Quality of English Language

.

Reviewer 2 Report

Comments and Suggestions for Authors

 

The manuscript with title “Landslide recognition considering terrain features supported by multiple machine learning methods and convolutional neural net-3 work " exposes the comparison of results for landslides recognition using several maching learning methods. This comparison is done in order to evaluate the impact of terrain feature indicators on the efficacy of this identification; and on the other hand, to investigate the effect of varying ratios of landslide and non-landslide samples.

The content of the document needs to be improved significantly because of lacks scientific information on the applied methodology (concepts and descriptions). For example, the spatial resolution of all layers (only is mentioned for the DEM), the information of the relief layer is not clear, definitions of both components of the curvature are not quite accurate. And the meaning of some acronyms is missing (TP, FP, IoU, etc).

The comments and discussion of results are very brief. Furthermore, this information is not well structured in the document. I find that the results obtained by authors related to the use of additional terrain information was quite obvious and it not represent any novelty, only the effect of varying ratios of landslide and non-landslide samples in different methods could be interesting for scientific community and creation of inventories at large scale. However, authors do not provide enough information to identify similar previous studies in this topic. In my opinion the article should clearly cite previous publications and highlight its originality compared to previous ones.

Moreover, all figures in general, and specifically the maps are very poor and with very low resolution and it is difficult to read on them the legend or the axis labels.

In general, the number of references provided is very low, and most of them are focused on work in Asia, and for very specific cases. More relevant references about the methods and to justify the novelty of the study of the article would be desirable.

Based on the previous exposed issues, I consider this should be rejected to be published in the ISPRS International Journal of Geo-Information

Reviewer 3 Report

Comments and Suggestions for Authors

Comments for author File: Comments.pdf

Comments on the Quality of English Language