Estimation of Water Coverage in Permanent and Temporary Shallow Lakes and Wetlands by Combining Remote Sensing Techniques and Genetic Programming: Application to the Mediterranean Basin of the Iberian Peninsula

Round 1

Reviewer 1 Report

The authors have to consider these comments in the updated version

Line 44: The Ramsar Convention

Put a reference for the convention, and check that for the whole manuscript.

Line70: Copernicus is the name of the program …

Put a reference for that program.

Page 3 Line 109: to assess the performance of the model

It’s better to give a brief hint about the model (e.g.: to assess the performance of the model that was developed by (…)), as you mention it at the start of the discussion part.

Line 136: European Habitats Directive

Put a reference.

Figure 1: It is better to add coordinates to the figure  

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This study evaluates the accuracy of a Genetic programming algorithm used for surface water detection in high resolution Landsat and Sentinel-2 optical data within a diversity of ecosystems of Mediterranean wetlands.

The originality of this article relies on the GP techniques applied to remote sensing, that has been already used by the same authors 5 years ago.

The paper should rather focus on the remote sensing part than on the description of ecosystems. Why the standard surface water extraction methods like RandomForest supervised classification or Indices thresholds are less accurate on such shallow intermittent lakes compare to GP?

Too many details about the carbon aspects of the project and ecosystem details, that do not serve the objectives of this remote sensing paper, should be simplified to increase the paper quality.

Authors should better explain their choice for GP technique as it seems to give the same Kappa and accuracy than the other numerous surface water extraction methods already published. I found the Kappa values very low, that is not suitable, contrary to what the authors claim. It is not clear to me why the authors have tested this method firstly used in doña et al., 2016. Because shallow non perennial lakes? Saline? Better explain in a shorten version of the manuscript.

In the same idea, the discussion is a two pages block that should be split into entitled paragraph with the main idea developed in it

In the end, what is the take home message? The Remote sensing community should shift to GP methods? As stated in the conclusion: "Kappa values above 0.7 and total error values lower than 20%" is not so different from what we can get from other methods (for example the first link I found in google: https://www.mdpi.com/1424-8220/18/8/2580/html).

The time to review is short in MDPI journal, so I could not provide any specific comments. I hope my general comments will help improving the manuscript readability.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Thank you for the manuscript submission. The question of water detection and monitoring for shallow water bodies remains of interest and importance.

As I read the manuscript, I am uncertain as to why you selected two GP threshold values. Why not look at each site and evaluate the optimal threshold value that best matches the reference image? The question of optimal threshold with respect to water conditions (depth, emerging vegetation density, etc.) could then be considered in more detail. 

I am also concerned about the use of Google Imagery as a source. This imagery is known to be of random dates and random image processing for appearance (as opposed for information). Google imagery is a visualization tool, and the imagery source and date should be confirmed to make sure no select image(s) were subject to uncommonly dry or wet conditions relative other imagery and relative to the S2 and L imagery acquisition dates. Could you comment on the satellite imagery acquisition dates, and the dates of reference data used for comparison? Also, are there any near-coincident S2 and L7 acquisitions?

For Figures 4 & 5, could it be made more visually apparent which three types were acquired by both S2 and L? It would be nice to more easily make the comparison between the two sensors. Also in these figures, do I properly understand that the reference water pixels are manually extracted by some spatial resampling of the Reference Data (Section 2.2)? Could you expand this a bit to provide more confidence to the reader about these numbers in blue?

You make an excellent observation that a 5 day repeat cycle and 10m spatial resolution shows promise in characterizing such water bodies. Given the efforts you placed in getting at least one acquisition per site during your study period, in your ecological region, what you forecast the success of multiple images per year over a wetland? How would you describe the temporal dynamics of the wetlands (what type of revisit frequency is required to capture the water dynamics) and can existing satellite imagery provide this?

Wetland type 1.3.2.7.2 indicates spatial scale may be important, did you consider resampling the S2 imagery to 30m (L7) spatial resolution for at least one site to see the impact on spatial resolution alone? 

Author Response

Please see the attachment

Author Response File: Author Response.pdf