Detecting Vegetation to Open Water Transitions in a Subtropical Wetland Landscape from Historical Panchromatic Aerial Photography and Multispectral Satellite Imagery

Round 1

Reviewer 1 Report

TITLE: Detecting vegetation loss in Florida Coastal Everglades wetland land-cover from multi-spectral satellite imagery and historical panchromatic aerial photography

Overall Assessment

This paper proposes a change detection method that uses a combination of classification algorithms and minimal mapping units to achieve binary classification of vegetated wetlands and open water. However, I think this research lacks innovation because that the authors only use two traditional classification methods for historical panchromatic aerial photographs and high-resolution satellite images followed by statistical land type changes. The time span of two images used in the paper is too large, the authors do not give a detailed classification process, the producer’s accuracy is more than 100%, and the classification accuracy lacks authenticity. Therefore, I rejected the manuscript published in the journal of Remote Sensing.

Introduction

Point 1: Machine learning algorithms have been applied to monitor land cover/use change, and you should describe their application in the introduction, and provide detailed depictions of the current problems facing wetland classifications.

Materials and Methods

Point 2: How do you select the sample points? how do you make sure that the accuracy of sample points in the panchromatic image of 1940. I do know find the relevant statements in the manuscript.

Point 3: Line 175-176 “We defined vegetated wetland as any pixel that contained > 50% vegetation.” I cannot understand this statement, there is only one value for one pixel, how do you define vegetation in the image.

Point 4: Section 2.6 of Materials and Methods The 1940s panchromatic image has 5362 sample points and the 2021 WV-2 satellite imagery has 5606 sample points. Please explain why the size of sample points is not the same.

Point 5: Line 213-214 “Characteristics such as brightness, color, texture, pattern, and context, were used 213” Did you extract features such as texture and brightness, and if you extracted them, you should show the features used in this study.

Point 6: There is an error in the vegetation index formula in Table 1, please check carefully, such as missing brackets.

Results

Point 7: Line 290-291 “No user’s accuracy for either vegetated wetland or open-water classes differed significantly across all maps (Table 2)”. Please explain why there is no significant difference in user accuracy for vegetated wetland or open-water, I could not get this information from Table 2

Point 8: Line 292-293 “. For the open-water class, producer’s accuracy was low, ranging from 66.5 ± 25.0% to 86.3 ± 23.2%”. How can the producer accuracy rate be greater than 100%? Please check these statements.

Point 9: Some accuracies in Tables 2 and 4 exceed 100%, which does not exist in the classification, please check your classification accuracy.

Point 10: Please explain how to calculate the accuracy of Change maps according to Table 4.

Point 11: The title of Figure 5 should be below the figure

Point 12: In Figure 4 and Figure 5, the size of Figure A and Figure B are not the same.

Discussion

Point 13: Your discussion is too much to find the important point. Streamline your discussion and delete unnecessary descriptions.

Author Response

Please see attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Revision of the manuscript ID: remotesensing-1803352 “Detecting vegetation loss in Florida Coastal Everglades wetland land-cover from multi-spectral satellite imagery and historical panchromatic aerial photography”.

 

The article deals with methodological issues regarding the classification of two images, panchromatic and multispectral, to obtain the changes in land cover surface area extent. Peatland vegetation and water pools are the two land covers studied. The analysis of change focusses on their areal extent, but not in their distribution.

The title should be more specific, and adapted to the content, since “vegetation loss” results too general, and could have different causes (fire, urban occupation, etc.). In the introduction, the causes of vegetation loss seem to be related to the see level raise in coastal areas. The title can be more explicit.

The introduction do not relay in methods and previous works dealing with classification and image processing to separate vegetation from water, at detailed scales, and the precision and accuracy of their results.

Methodology: the aim is to quantify change in surface extent between two images. This must be clearly mentioned. The reader does not know how many data is analyzed in the 72-year period. Just until page 5.

Materials: must be explained. And Fig 1 simplified after that.

Methods: L106-123. Difficult to read. Some parts belong to discussion issues.

Study area: the description of the area must be clearer, at least at the resolution of this works: very detailed: Which type of vegetation are there? (Add a field photograph, for example) and also for the water: which is the size of the pools and also their depth? are they permanents or seasonal? Are there submerged or emerged vegetation? Etc. Moreover, the authors discuss later some features which has not been previously explained.

Figure 2: Location of the area in a National context, at least. Avoid to double the degrees on all sides of the square. Indicate the photograph at right, which is this?, date and location  (A figure is not clear). Being the sea proximity important (sea level rise), can you indicate or illustrate also the distance to the sea in this figure?

 

Methodology is very difficult to read, it is not easy to follow. Revise and try to simplify all the steps. Do not use or repeat names or processes which are not necessary. This must be revised throughout all the manuscript. 

Specific comments:

L156. You have resampled the panchromatic (1x1) to 2x2. Why not the contrary? It seems you have lost spatial resolution. Later you mention the low quality of this image. I do not understand why.

L172. This objective is not explained before. Please, be consistent.

L201. Remove “ecological questions”

L203. Do you have collected more than 5000 samples manually? How many time for this process? Is this affordable?

Figure 3. Why the threshold in the spectral imagery if it is not used for classification? The characteristics of the satellite image must be compiled previously (material section) in a table (bands and spectral features). Which means “Data Value” in the two graphics?

 

L271 and throughout all the manuscript. From my point of view, the authors must go to the point. The use of “wetland land cover” classification should be simplified.  Here the objective is to differentiate vegetation (which vegetation?) from water. (Assuming there is not bare soil).

Fig. 5. Where is the grey color of the legend? It is not identifiable. Perhaps a plate with the three MMU used could be illustrative.

L337. Dense cluster of trees. I think this is the first time it is mentioned.

L411-414. Some repetition in the sentences.

L401. This is a peatland landscape. Must be defined at the beginning of the article, even in the title.

In general Figures must be improved.

Author Response

Please see attached.

Author Response File: Author Response.docx

Reviewer 3 Report

attached

Comments for author File: Comments.pdf

Author Response

Please see attached.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Your paper has been revised with outstanding research highlights and a clear structure, so I think it has achieved the requirements of "Remote Sensing" for publication.

Reviewer 3 Report

Thank you for your correction