Wetland Mapping in Great Lakes Using Sentinel-1/2 Time-Series Imagery and DEM Data in Google Earth Engine

Round 1

Reviewer 1 Report

This paper proposed a method that involves two classification steps. Firstly, they identified and masked non-wetland classes like Barren, Cropland, and Open Water using radar and optical remote sensing observations and a Random Forest model. Secondly, they identified wetland classes, including Fen, Bog, Swamp, Marsh, and Forest and Grassland/Shrubland, as non-wetland classes. The method proposed in this paper improves the accuracy of wetland category classification. I have the following suggestions for authors to consider.

(1) I suggest adding some graphs in section 2.3 to describe the satellite data.

(2) In Section 3.1 the authors discuss the process of data preparation. The article should introduce the specific data preprocessing method in more detail.

(3) Section 3.2 is too concise, and the authors should focus on the specific details of the classification model.

(4) I recommend using different colors in Figure 7 to distinguish what each column represents.

(5) I wonder if the author has tried or considered using deep learning methods to solve such problems.

 

Except for some minor tensor errors, I cannot find any issues.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper describes classification of the wetlands using Sentinel-1/2 data and object-based Random Forest algorithm in Google Earth Engine. As a result, the authors compiled a wetland map of the Great Lakes), including such classes as Marsh, Swamp, Fen, and Bog.

This topic is relevant in the remote sensing data classification field, and interesting results are obtained. This research shows the difficulties of classifying different types of wetlands, and ways to solve them.

Compared to other published material, this research suggests two-step object-based supervised RF classification algorithm and the most recent wetland map of the Great Lakes as a result.

As improvements to the description of the methodology and results, I suggest to the authors:

1. To add description of the features of Marsh, Swamp, Fen, and Bog. How do these classes differ (presence of grass, trees, etc.) in nature?

2. In the Table 1 try to somehow indicate that 4611 is the number of samples of five classes, and not just one Fen. This is written in the text of the paper, but visually it is not clear in the table.

3. When the training and test samples were separated, how was their independence ensured?

4. The resulting bands importance in RF training should be added to the Results.

5. The statement in line 457 is not entirely true. Training RF and mapping can be done using QGIS plugins and Python/R scripts. If you work with one Sentinel-2 tile, the power of the one computer is enough. When processing large areas and large sets of bands, you will need a server (or several) with a GPU and a good amount of RAM and HDD - and only in the absence of such equipment can this task be called impossible, but it is not impossible at all. 

The authors describe and discuss the obtained results in detail, and the conclusion is consistent with the arguments on the main research question.

The references are appropriate but the format of references in the text (last name, year) does not meet the requirements of the journal ([number]) as far as I know.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The article deals with a very interesting topic regarding 'Wetland Mapping in Great Lakes using Sentinel-1/2 and DEM Data in Google Earth Engine" in Colombia.

Overall, it is a comprehensive article, well referenced and the findings provided indicate that a great deal of effort was put in. Suggestions for some improvements that could be performed to the manuscript prior to its publication are the following:

- Introduction: Try to focus a bit more on the goals of this research, as well as the research questions and and findings that add to the topic (the novelties that this research offers). Avoid parentheses, abbreviations and numerical references in this part of the manuscript.

-Methodology: It would be interesting to add a comparative table regarding the several methods used.

-Accuracy assessment: It is suggested to elaborate more on the visual interpretation that was performed and the corresponding factors that were taken into account.

- Conclusions: It would be interesting to comment more on the restrictions/challenges that may arise and the corresponding solutions that could be followed.

In which way the type of the wetlands may affect the proposed workflow and how this factor may affect the study?

Are there any other factors/limitations that should be taken into account (when for example, the method is implemented in a different area of interest?)

What about the general synergy of Sentinel-1/2 and DEM Data in Google Earth Engine?

Could the proposed methods be implemented in a different area and what would be the restrictions?

 

Minor editing of English language is required

Author Response

Please see the attachment

Author Response File: Author Response.pdf