Patagonian Andes Landslides Inventory: The Deep Learning’s Way to Their Automatic Detection

Round 1

Reviewer 1 Report

The title “Patagonian Andes landslides inventory: The deep learning’s way to their automatic detection” is a good quality paper. But some comments are provided to improve the current version of the manuscript shown below:

Section 2:

Please clarify the characteristics of the landslides analyzed in this study. When the landslides happened? What is the type of landslides (landslides, rock falls, avalanches or debris flows)? Since the authors analyzed the Sentinel-2 images observed from 2020 to 2021, I imagine that not only recent new landslides but also very old landslides were included in the database. I am wondering is it rational to analyze old and new landslides, and different types of landslides with same way.

Table 2:

Please show not only the accuracy percentages but also the numbers of landslides (or pixels) detected or mis-detected by the proposed method in order to show more detailed performance.

Discussion section:

I believe CNN-based automatic detection of landslides have been proposed in many previous studies. Is it necessary to compare the performances of the proposed method with the previous method?

Is the proposed method applicable only to Patagonian Andes region? If so, could you discuss the potential expansion of the method to other regions? If not, please describe the reasons of the types of the landslides.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Please find the attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

I found this an interesting paper, although somewhat peripheral to my own expertise. The computer recognition process was found to be generally good, but to highlight some inaccuracies. Well, so does air photo interpretation, and so to does on-the-spot geomorphological mapping – there are many features that look like landslides, but are not, and there are also many unrecognised ancient landslides. It does seem to me that in an active mountainous area any computer procedure was going to get better results than (say) in an urban area or an area that is heavily vegetated (although the authors do say that).

I’m going to recommend publication, but to make a small number of observations. Firstly, I think that the paper would be improved if the Authors had described the motives behind the project. Secondly, I think that repeating the exercise with a different set of images of more or less the same date and comparing the outputs would have been a useful test of the procedure. Thirdly, it would also be useful to compare runs with images of an earlier or later date, to see how landslides are evolving or new ones are generated (or how unseen ancient landslides are regenerated).

I am going to assume that English is not the first language of any of the authors, and on the basis of that assumption, to commend the command of English displayed in the text. I think that it is an heroic effort to write a paper this well in a second language. I am from the UK, and dislike elements of US spelling and grammar, but if they are the house style of the Journal, they must override reviewers’ preferences. In this case the differences are few and far between and (subject to understanding the techniques and their acronyms) do not impinge on the readability of the text.

In a similar vein, I am of the school of thought that going on site and taking a look for yourself is invaluable and beats sitting in a darkened office peering into a computer screen, so for me, a paper describing field excursions to try to resolve some of the more uncertain cases would be of great interest.

As it is, the paper is well worth publishing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors appropriately responded for my comments and suggestions.
I feel the revised manuscript should be accepted in the present form.

Reviewer 2 Report

This paper was modified properly.