Supervised Classifications of Optical Water Types in Spanish Inland Waters
Round 1
Reviewer 1 Report
The minor comments include
- Line 33 – 36, what does the word “previous” refers to?
- Line 33 – 47, specially lines from 44 – 47, requires citations
- Line 50 -53 requires citations
- Line 65, what does “these” refers in this line?
- Line 76 – 78, how do OWTs provide useful information…... Please cite any previous work you could ?
- Line 79, How could using the so-called supervised learning classifiers makes it innovative?
- Line 81 – 86, inappropriate use of “, “ & ” ;”. You could rather split the statement in to several statements.
Major comments
- Line 103 – 111; there were no any description of the selected water bodies out of the 10 water bodies you listed. Could you add more information on this line about the lakes?
- In your methodology section, you stated (lines 203 – 205) that you manually labelled all the pixel values based on knowledge of the area, reflectance values and derived concentrations. The question is, wasn’t it difficult to manually analyse and label 8740 pixel reflectance values and classify it into; clear, moderate, turbid, very turbid as the reflectance values overlap specially for moderate (560 – 700) and turbid (500 -565)? overlap. It doesn't look realistic
- How do you present the result of these manually classified pixel values?
- Again, in the methodology section most of the pixel value were from a single water body (Albufera lagoon) as in line 214-215. If so, how could this result from the study extrapolate to other lakes(water bodies)?, how do you justify? In line with your problem statement in line 32 -33 ("A single method for obtaining quality variables in all lakes, even in the same lake, may not work well [4])".
- The result in Table 1 shows that KNN has the best performance in both (all bands and only Rrs) performance however,
o you stated (lines 381 -383) Random Forest classifier (RFC) were the best
o most of your results were mapped (figure 7 and figure 9) using RF classifier and
o Concluded RFC were the one selected (line 538),
why are your conclusions not supported by your results?
Author Response
Reply is in atached PDF file.
Author Response File: Author Response.pdf
Reviewer 2 Report
Review of “Supervised Classifications of Optical Water Types in Spanish Inland Waters”
By Pereira-Sandoval et al.
This manuscript describing the application of a satellite-derived optical classification approaches for inland Spanish waters. The manuscript is generally well-written and clear in its presentation and organization. It provides a nice example of application of different machine learning methods for optical water type classification. The authors provide a thoughtful and thorough analysis of the different approaches and their associated statistics and performance.
The manuscript could benefit from a more careful editing for grammar, spelling, and syntax. It also utilizes a large number of statistical terms and abbreviations. The addition of an Appendix or table summarizing the statistics, along with their abbreviation and brief definition would be helpful. Some of the equations could use more explanation and citations as appropriate.
The authors chose to use the C2RCC neural net algorithm to process the S2-MSI imagery. There are numerous other atmospheric correction algorithms and it would be useful to have some additional discussion about the choice of this approach over others (e.g., C2X, Polymer, ACOLITE, etc.).
I believe the manuscript is worthy of publication after a careful revision. Additional comments are provided below.
Specific comments:
Page 4, line 120: Perhaps replace “avoid” with “remove”.
Page 4, line 128: Is the C2RCC algorithm inverting the spectrum for atmospheric correction, applies the atmospheric correction.
Page 5, lines 203-205: How reproduceable are these manual water type assignments? Are they user-dependent?
Page 6, line 236: Change “easiness” to “ease”.
Page 7, lines 272-274: Please explain the notation for Eqn 1 or provide an appropriate citation. I assume this means that pixels correctly assigned as compared to a reference classification are given the value 1. This differs somewhat from other expressions for overall accuracy that I have come across. Note also that the Kappa statistic has been criticized by some in the literature (Pontius, 2011, Int. J. Remote Sens.), although it is still used widely for describing the performance of classification methods in remote sensing. The authors may wish to elaborate more on potential limitations of the Kappa method.
Page 7, lines 299-300: Please explain how the confusion matrix was determined.
Page 7, lines 301-303: Including all the bands would seem to introduce redundancy in the training dataset, since the derived IOP and concentration products use information from the Rrs bands. Is the improved performance over using just the Rrs bands significant or simply an artifact of the redundancy in the training set?
Pages 8-9, lines 322-323: Please explain the permutation method in more detail. By permuting the feature, are your randomly reordering the pixel assignments within a given feature?
Figure 6: This figure requires more explanation in legend. What do the red rectangle areas in the time-series plots represent? The images were from 11 May, but the red rectangle encompasses multiple days. Were the images binned over multiple days? Please clarify.
Figure 9: Why is the land boundary included in the color scale in the left panel of Figure 9 and white in the right panel?
Page 15, line 505: Change "phycocianyn" to "phycocyanin".
Author Response
Reply is in atached PDF file.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors proposed classier algorithms (SVC, Random Forest, k-NN,..) using Sentinel-2 MSI imagery datasets to evaluate the water quality of lakes and reservoirs. They presented well-organized and good work. the paper it is ready for publication. My alone comment
line 546. please revise this sentence ''For doing this we could dedicate more man hours to manual labeling, continuing with the method done...''
Author Response
Reply is in atached PDF file.
Author Response File: Author Response.pdf
Reviewer 4 Report
This manuscript is well written, and the result can benefit the scientific community.
TRANSLATE with x English
Arabic | Hebrew | Polish |
Bulgarian | Hindi | Portuguese |
Catalan | Hmong Daw | Romanian |
Chinese Simplified | Hungarian | Russian |
Chinese Traditional | Indonesian | Slovak |
Czech | Italian | Slovenian |
Danish | Japanese | Spanish |
Dutch | Klingon | Swedish |
English | Korean | Thai |
Estonian | Latvian | Turkish |
Finnish | Lithuanian | Ukrainian |
French | Malay | Urdu |
German | Maltese | Vietnamese |
Greek | Norwegian | Welsh |
Haitian Creole | Persian |
TRANSLATE with EMBED THE SNIPPET BELOW IN YOUR SITE Enable collaborative features and customize widget: Bing Webmaster Portal Back
TRANSLATE with x English
Arabic | Hebrew | Polish |
Bulgarian | Hindi | Portuguese |
Catalan | Hmong Daw | Romanian |
Chinese Simplified | Hungarian | Russian |
Chinese Traditional | Indonesian | Slovak |
Czech | Italian | Slovenian |
Danish | Japanese | Spanish |
Dutch | Klingon | Swedish |
English | Korean | Thai |
Estonian | Latvian | Turkish |
Finnish | Lithuanian | Ukrainian |
French | Malay | Urdu |
German | Maltese | Vietnamese |
Greek | Norwegian | Welsh |
Haitian Creole | Persian |
TRANSLATE with EMBED THE SNIPPET BELOW IN YOUR SITE Enable collaborative features and customize widget: Bing Webmaster Portal Back
TRANSLATE with x EnglishArabic | Hebrew | Polish |
Bulgarian | Hindi | Portuguese |
Catalan | Hmong Daw | Romanian |
Chinese Simplified | Hungarian | Russian |
Chinese Traditional | Indonesian | Slovak |
Czech | Italian | Slovenian |
Danish | Japanese | Spanish |
Dutch | Klingon | Swedish |
English | Korean | Thai |
Estonian | Latvian | Turkish |
Finnish | Lithuanian | Ukrainian |
French | Malay | Urdu |
German | Maltese | Vietnamese |
Greek | Norwegian | Welsh |
Haitian Creole | Persian |
Author Response
Reply is in atached PDF file.
Author Response File: Author Response.pdf