Identifying Alpine Lakes in the Eastern Himalayas Using Deep Learning

Round 1

Reviewer 1 Report

Identification of small lakes over large areas is a difficult task not only for alpine lakes, so the automated lake identification method presented in this article is very interesting and useful for remote sensing. The method based on deep learning combined with satellite data from several sources certainly holds promise and its effectiveness shown in this study is a compelling evidence.

This is an interesting, good study describing the methodological approach and the dataset used. In my opinion, the presented manuscript is of high quality and can be published without significant modifications.

I have only one individual question:

200 - It would be interesting to know how Binary Dice Loss and sample weights were set up, or provide some publication.

Author Response

Please see the attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

Authors,

A very well researched and written paper with only minor points to consider before publication.

First the minor point;

Table 5 is confusing. It is not clear in the Table that Alpine Lakes refers to the sum of Glacial and Non-glacial. Perhaps place the Alpine line first with a break underneath it and then Glacial and Non-glacial. The reader looks at this table and sees three categories Glacial, Non-Glacial and Alpine. I realize that this is explained in the text but just looking at the table it looks like three categories.

A more major point. Unless I missed it somewhere in the paper did you compare your results directly with what might have been done using only image processing algorithms? I realize that the technique worked quite well but as a reader I might ask- What does this get me over standard image processing? Why should I waste my time building training sets, etc. when I could apply off the shelf tools and get comparable results? I ask this question as an author of publications using similar ML methods who is constantly asked "Is all that you have done worth the effort? 

Even with the above, overall I really like what you have done and the clarity of how you presented your work. 

Very well done. Recommendation: Accept in present form with the exception of briefly dealing with the two points mentioned above

Author Response

Please see the attached file.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

 

The research topic of your study is among the most interesting now-a-day for the people who study the high mounting and high latitude regions where glacial lakes are dominated. To my understanding, the most attempts were focused on developing the automatic algorithm for identifying the following important characteristics of the lakes: the number of lakes, their surface area, the lake type (glacial and non-glacial) and elevation. The remote sensing observations (Sentinel-1 and 2, Google Earth, etc) were used to produce the most recent regional inventory of the lakes in the Eastern Himalayas. 

In my opinion, the manuscript is not acceptable in its present form. I would encourage you to consider the future re-submission of the manuscript only after the serious revision by: (a) changing the structure of the manuscript; (b) extending the explanation of the figures and tables given in the text; (c) including the comparison with the in-situ observations (i.e. the surface area and type of the biggest lakes); (d) providing the numbers showing the uncertainties of the traditional methods and new algorithms for each characteristic (including surface area and elevation); and (i) excluding the technical (non-scientific) details (i.e. preparing the data and computation environment). I would also suggest refraining from using too many abbreviations, especially if they are not explained in the previous text. I would also suggest adding the information on the contribution of each co-author to the manuscript and to include the inventory of  the lakes as the supplement.

with the best regards,

Anonymous Referee

 

Hereinafter are the specific comments to the manuscript:

Line 25: SAR, the abbreviation is not explained;

Lines 26-27: I would suggest adding the numbers to show how significant is the improvement for each of the characteristics (number of lakes, their surface area, lake’s type and elevation). 

Lines 82-97: I would suggest moving this text to the section with the methods.

Lines 98-100: In my opinion, it would need to explicitly state the aim of the study here. 

Line 106: I would suggest labeling this section as the Study Area (2).

Lines 108-118: I would suggest adding what climate type (see Beck et al., 2018) is dominated in the region of the study together with the information on the average annual amount of precipitation and annual air temperature. It is also important to add how many (and their type) lakes have been found in the region of the study in the previous narratives.  

Line 121: I would suggest starting the section with the Data and Methods (3); the section includes only two subsections with the data (3.1) and methods (3.2). 

Line 127-138: it is not clear how each of three datasets (Sentinel 1, 2 and PALSAR) were used? I would suggest adding details about what the characteristics of the lakes (surface area, type, elevation) were defined from which dataset? If you have applied the SAR data (see the abstract), why is it not included in Table 1? 

Line 141: Table 1, please add the explanation of the abbreviation used in the table. It may be the full name of the variable or so.

Lines 142-152: In my opinion, this text is the part of the subsection with the Methods. 

Line 156 and Line 179: Please, how the visual analysis (expertise) was used and how substantial the role of an expert in the algorithm you applied in the lakes’ identification?

Line 160-163: I would suggest moving this part to the section with the Results (4). 

Line 168: Figure 4 is not explained in the text, the labels are too small (not readable). 

Line 178 and line 187: The figure 5 and 6 are very similar (and both poorly discussed in the text). I would suggest excluding one of the figures and adding more explanation on the modification of the deep learning method made by the authors. 

Lines 191-208 and also Lines 209-217: Too many abbreviations in the text are given without the explanation; the text seems to be the description of the software configuration which is technical (not scientific) information which can be given in the Annex. 

Lines 223-232. In my opinion, here the main results of the study are given, and I would suggest extending their representation by (a) adding the min/max for the lake's surface area and elevation; (b) estimations of the errors for each lake’s characteristics given in Table 5. 

Lines 233: I would suggest adding the table with the estimations of the surface area, lake type and lake elevation for several biggest lakes in regions. The characteristics of these lakes should be known from the in-situ surveys.

Lines 234-261: I would suggest shorting the text. 

Line 252 and 271: The labels used in these figures are too small, and it makes it difficult understanding the information given in the figures. Both figures are poorly discussed in the text.

Line 278: It is not clear what you mean by the “Ground Truth”. Where is the reference?

Lines 335: In my opinion, the lake’s surface area and elevation estimated by the deep learning algorithm need to be compared with the surface area and elevation which are known from the in-situ observation for the several lakes (i.e. small, middle and big size). 

Line 349: It is not clear how the SAR data was used in this study? It is not mentioned in the section of data and methods.

Line 354-358: I would suggest moving this text to the section of introduction to explain the motivation behind the study.

Line 369: I would suggest adding the explanation with the contribution provided by each author. It is also important to include the inventory of the lakes as the supplement to this study. 

Line 371: missing numbering.

Line 431: I would suggest not referring to preprints. 

 

Author Response

Please see the attached file.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Dear authors,

In my opinion the manuscript has improved substantially since after the revision. I would recommend accepting it after the minor revision. 

In particular, Table 5 shows the area and elevation of the alpine lakes which were divided into the glacial and non-glacial lakes. The minimum elevation for non-glacial lakes is over 1500 m, and it is in contradiction to the definition for “the alpine lakes” which were given in the first sentence of the introduction. 

I would also suggest adding some sentences about the algorithm limitations, and whether the algorithms can be applied in other regions, i.e. Greenland or Antarctica where the glacial lakes are a lot in the coastal area.

Anonymous referee.

Author Response

Please see the attachment.

Author Response File: Author Response.docx