A Deep Learning Based Method to Delineate the Wet/Dry Shoreline and Compute Its Elevation Using High-Resolution UAS Imagery

Round 1

Reviewer 1 Report (Previous Reviewer 1)

My biggest concern with this paper from my first revision still stands, which is the High-Water Line is not a stable shoreline indicator that should be used for shoreline monitoring. The authors have attempted to rectify this by including citations that suggest otherwise, but these papers are >50 years old (McCurdy 1950 and McBeth 1956). Since then, there have been a number of papers that suggest the HWL has no "objective, quantitative control" (Boak and Turner, 2005). The technical aspect of the paper is sound, but I'm not sure the application is of use to the coastal geomorphology/remote sensing community given the preferred use of tidal-datum indicators. If the authors could provide a more convincing argument for the adoption of their HWL mapping technique using recent literature, then I would reconsider its suitability for publication in Remote Sensing.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

Please see attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

In this paper, a modified HED architecture was proposed to create a model for automatic feature identification of the wet/dry shoreline and the effectiveness of the proposed method was verified by data sets from different regions. In general, this manuscript is well-organized and the methodology is clearly explained. The reviewer's comments are as follows:

1. Line 62: The format of Ref [26] needs to be modified.

2. Figure 4 has multiple subgraphs, I think it might be better to briefly introduce them in the annotations.

3. Figure 6 and 7 must be improved, in fact, they can be better represented than just words.

4. Line 482: Please add a space between "2.2" and "cm", while the space between "44.7" and "%" (Line 427) should be deleted.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 1)

The authors' explanation for their use of the High Water Line shoreline indicator is now much clearer in this revision. They are using it as input for coastal inundation models on the timescale of hours to days, as opposed to long-term change assessments. A sufficient explanation was added to the Introduction that explains the limitations of the HWL but how it is useful for their purpose. The authors have also addressed the lack of details regarding the UAS and RTK-GPS surveys, which was also pointed out by another reviewer.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

This revised version (v2) is a well-written and clear manuscript. Overall, I think this is a good paper.  The results are very good, and worthy of publication. 

 

Only a few extremely minor comments: 

Line 119: should say ‘enables’  

 

Lines 440-442: It might be better to say something like: “This case highlights the fact that while the AP and F-1 scores are useful indicators, by themselves they are not sufficient to determine the performance and the applicability of a wet/dry shoreline prediction.”

Note that is a suggestion, and just because I think it is more clear does not mean that all readers will think that. 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This paper is focused on generating a deep learning technique for extracting the wet/dry sediment line from UAS orthomosaic imagery so that it may be used as a shoreline indicator for coastal management. The overarching problem and likely crux of this goal is that the wet/dry line in itself is not an accurate indicator of the shoreline position. As the authors themselves point out, the wet/dry line is highly influenced by a variety of processes: significant wave height, tide stage, beach sediment volume change, to name a few. The wet/dry line is a boundary that is shifting horizontally anywhere from a few meters to tens of meters per day, depending on the tide range and slope of the beachface. As such, the wet/dry line really should not be used as a shoreline indicator for conducting repeat coastal surveys as the chances of wave height/tide stage matching between all of them is incredibly low. From a remote sensing perspective, the paper does not introduce a novel image processing technique that could be used across a wider application spectrum that would be of interest to the broad readership of Remote Sensing. As such, I cannot recommend this paper for publication. Perhaps it might be better suited for a journal more narrowly focused on coastal processes, especially if there are other uses for mapped wet/dry sediment lines. Please see the document attached for line-by-line comments. 

Comments for author File: Comments.pdf

Reviewer 2 Report

Dear Authors

Compliment for the interesting paper.

I have not specific observation.

I am a final user of your results and they seem to go in the right direction.

It is very interesting to try to export analysis to stationary camera.

Good luck

 

Cordially

Reviewer 3 Report

Thank you for the opportunity to review the paper entitled: A Deep Learning Based Method to Delineate the Wet/Dry Shoreline and Compute its Elevation Using High-Resolution UAS Imagery by Marina Vicens-Miquel, F. Antonio Medrano, Philippe E. Tissot, Hamid Kamangir, Michael J. Starek, and Katie Colburn.

The paper presents a deep learning technique using a modified Holistically Nested Edge Detection (HED) architecture to extract the position and height of the wet/dry shoreline. The height of the wet/dry shoreline was extracted using the structure-from-motion derived digital surface model (DSM).

In the Introduction several statements need citation and some form of explanation since not all satellite derived shoreline extraction have positional errors of tens of meters or are without elevation. Also, I do not quite understand the Literature Review section, especially in the present format. The authors do not even mention the majority of well used methods to extract land/water from satellite imagery and focus on very few selected papers. I do not find even necessary to mention the majority of methods to extract land/water by name, but they can be grouped in few categories, and you could give citations for several methods in each category. Also, it is not necessary to explain in detail the methods used to do the land/water classification of satellite imagery. There is a very rich literature out there on this topic. All methods can be applied to georeferenced imagery so saying that one method drawback is the lack of real-world coordinates for results is not appropriate.

The datasets were collected by UAS with different cameras. It would be interesting for the reader to know the types of camera used and what parameters were set for each collection. From each UAS collect DSMs were derived, but there are no details about processing or the result RMSE or accuracy. Since from the figures used in the paper it is clear that targets (markers) were used for SfM processing, it would be nice to know at least the horizontal and vertical RMSE on targets, if no independent GPS measurements are available for validation.

The paper mentions that the beaches are different (3 in Texas and one in Florida) but nothing is mentioned except the sand color and the fact that in Texas you can drive a car on the beach and on FL beach you cannot. These are not all the differences in the beaches a reader wants to know. What about differences in geomorphology (if any exists), sand composition (FL sand beaches are predominantly quarzitic, while in TX sand beaches are made of quartz, feldspar, heavy metals, igneous, metamorphic, and sedimentary minerals), sand grain size, nourishment vs. natural beaches, etc. All of these potentially could impact how the images on the beach are recorded and how distinguishable the wet/dry shoreline actually is. Maybe the width of the AI shoreline prediction is dependent on the beach characteristics.

A part of the paper discusses the appropriateness of using AP and F1-score to judge how good the AI model predicts the wet/dry shoreline, and the mean height of the wet/dry shoreline is proposed as a better metric. For the purpose of this paper maybe a better metric would be the 3D location difference of the labeled wet/dry shoreline vs AI shoreline, or some metric that characterizes this difference. At the minimum I would like to see the distance difference (error) between the labelled and AI shoreline expressed as RMSE and MAE. For example, the 3D distance between equally distanced points on one shoreline and the nearest point on the other shoreline. Or any other metric that is more inclusive and the authors want to explore. This is important especially since the beaches have a very small slope of 1.6 to 3 degrees, so 1 cm in elevation is gained over quite considerable distances (approx. 20 to 36 cm).

For more comments, please see the pdf manuscript with highlights. I am recommending this paper for publication after major corrections.

Comments for author File: Comments.pdf