Habitat Classification Predictions on an Undeveloped Barrier Island Using a GIS-Based Landscape Modeling Approach

Round 1

Reviewer 1 Report

The manuscript is interesting, well-structured and addresses a series of relevant issues. The possibility of classifying the habitats distribution through remote sensing is certainly one of the most important frontiers for the space economy. These applications, in the coastal environment, are even more complex and significant and have important impacts on the green and blue economies.

Introduction and discussion: one of the main values of this manuscript is that it is succinct. At the same time, this synthesis has led to some limitations. In particular, the introduction and discussion paragraphs lack general references, a robust description of the state of the art. For example, it would be useful to briefly describe when and why coastal environments were firstly studied with remote sensing technologies, and which areas, authors and results have been achieved at international level through time (introduction).

In the discussion, a comparison of results reached by different case studies and an explanation of how the present work has contributed at international level to the “Habitat classification”, can be useful (also by using figures or tables).

Structure: my suggestion is to move “Study area” (paragraph 2.1. in the present form of the ms) before methodology as paragraph 2. Then, paragraph 3 can be composed by paragraph 2.1. “Data and material” (description of dataset - lines 106-120 - and table 1) and 2.2 “Geospatial Data and Processing”. In this way, the papers will be easier to follow for the readers.

Methodology: the 2000 DEM should be use to calculate elevation change (2012-2000) as well as the slope to better compare the data.

Nutrient sequestration rates were assigned to each cell (5x5 m – other studies use 2x 2 or 3x3 for the emerged part of the beach-dune system), but all values that did not meet the elevation or vegetation criteria of Table 2 were assumed to be zero. What is the approximation derived from this methodology?

In addition, in table 2 ranges of elevation (m) are indicated for both Herbaceous and Woody habitats. For each elevation range a value of Carbon and Nitrogen Sequestration is indicated. It is an important assumption and output of the present research, but it is not easy to understand. In fact, ref 23 (Rossi, A.M.; Rabenhorst, M.C. Organic carbon dynamics in soils of Mid-Atlantic barrier island landscapes. Geoderma 2019) seems to correlate carbon and nutrient sequestration to soil age and proximity of water table. Can you please explain what is the assumption made for this part of the processing? How elevation values have been chosen and how meteo-climatic condition influence vegetation and habitat distribution along the coast?

References:

Furthermore, with regards to geomorphological aspect associated with habitat distribution some reference should be added. How barrier Island morphology is related to shoreface connected bedforms (such as sand bars or shallow water sedimentary features and/or seagrass) should be considered by the authors or it should be explicitly reported as a limitation/assumption. Recently, many publication have reported similar case studies and coastal habitat classification in Asia, Australia and Mediterranean Sea (see LiDAR and spectroscopic techniques developed under the FHyL approach and applied to Sabaudia coast, Tirrenyan Sea, Central Italy on both emerged and submerged coastal environment).

Author Response

Lines 59-72: We have added a paragraph in the introduction that describes some applications of remote sensing in coastal environment.

Lines 368-380: We have included some additional text/references in the discussion that describes other habitat classification studies using remote sensing. To our knowledge, ours is the only study that explicitly tries to predict vegetation habitat using elevation metrics exclusively, while other studies combine imagery and lidar to generate more accurate classifications.

Lines 367-380:  We have included some additional text/references in the discussion that describes other habitat classification studies using remote sensing. To our knowledge, ours is the only study that explicitly tries to predict vegetation habitat using elevation metrics exclusively, while other studies combine imagery and lidar to generate more accurate classifications.

Line 95-117: We have moved study area to its own section.

We did not use slope change as a metric since we felt including it with elevation change would be redundant. However, slope and distance to shore were both considered “static” measurements and we therefore used the most recent DEM to calculate these.

Lines 187-199: Table 2 was updated with new values and the respective text in the methods was expanded to explicitly state the assumptions. We also clarified that the 5x5 cell size was the cell size we were using for this study and mentioned in the discussion that 5x5 cell size may overestimate the nutrient sequestration since the whole cell is assumed to be vegetated.

Line 204: The reference in the original version was incorrect. We have updated the reference and clarified the language to show that these were observations from the Rossi dissertation.

Lines 374-378: We agree with the reviewer regarding adding references in other regions of the world and have done so. We would like to emphasize that this approach does not really try to predict vegetation habitat using lidar only, but integrates imagery, lidar, and field observations to produce highly accurate maps. We think this method can be used to determine new metrics for predictions but has a steep learning curve relative to our approach.

Reviewer 2 Report

I read the manuscript carefully, I'm not particularly experienced with vegetation, but the research seems set up well.

At first glance, probably a couple of things can be improved:

If I understand correctly, the value of C and N sequestration are calculated based on the results obtained from the proposed model. However, as shown in Figure 4, the proposed model is not perfect: egg., it identifies areas with sparse vegetation, but it does not work very well to distinguish between grassy and tree vegetation.

In my opinion, the authors could also calculate C and N sequestration data from the observed vegetation, compare them to those obtained by model, and discuss the differences between the results.

Figure 4c is hard to read, there are too many different classes that you can't really distinguish on the map. To make it more readable the authors could make two classes: class 1 prediction=observation; class 2 = prediction <> observation. All the other details are in the tables.

Author Response

Lines 312-323: We added the sequestration results based on observations and compare them.

Line 249 (Figure 4C): We agree with the reviewer that the figure was difficult to read. We have changed the figure to show these 2 classes, but also added the original as a supplemental figure and used a more qualitative color scale. 

Reviewer 3 Report

• This paper describes an interesting methodology for barrier island habitat classification predictions. The introduction provides an appropriate insight into the field of the paper.
• The paper is well structured, but needs significant improvement for publishing in RS journal.
• Study area needs to be described in more detail. Figure 1: Please add (in the corner) small States map and mark a wider area of research. Assateague State Park is not marked on the map.
• Please add some characteristic photos of the island, which are related to this paper, and later discuss according to results.
• Figure 2: Please make your original figure, on this one it can be seen that it is poorly patched.
• Rewrite the paper according to instructions for the authors:  figure – use (a)…..
• Figure 4: in the paper text it is mentioned 4B and 4C, not 4A.  Also Fig 4C should be explained better in figure caption. It is given 7 categories, which do not differ much on a given map scale.
• On pages 6 and 7 images and tables are given: they should better fit into the paper (text).
• In my opinion the paper needs further analysis. The paper does not provide the adequate analysis of areas where the classification did not provide proper results. 1/3 of the data don’t match, are there some characteristics of those areas? It is missing some images or other field and results (errors) discussion according the situation on the field (ground)

Author Response

Line 120: The site map was updated.

Line 161: Figure 2 was updated.

We followed the guidelines for authors when revising the manuscript.

Line 249 (Figure 4C): We agree with the reviewer that the figure was difficult to read. We have changed the figure to show these 2 classes, but also added the original as a supplemental figure and used a more qualitative color scale. Additionally, references to Figure 4A have been included.

Lines 209-230: We revised this text to better integrate these figures and tables into the text

Lines 357-368: We agree with the reviewer that our original text was not detailed enough. We rewrote the section as follows:

“Conversely, the model misclassified a majority of woody cells as herbaceous. These errors primarily occurred because there was little difference between the herbaceous and woody habitat requirements reflected in the parameter distributions (Figure 3). Although the overall accuracy was comparable to other elevation-based modeling studies on barrier islands [20,46,49], woody vegetation classification could be improved by including additional parameters that can better distinguish between woody and herbaceous vegetation, such as access to freshwater lens [50,51] or soil age [52]. However, woody species, such as Morella sps., may exhibit more plasticity than the literature suggests, warranting further investigation into their ranges to improve habitat classification efforts surrounding them [18]. Vegetation height information, which can be acquired from the first return of lidar data, would help distinguish between woody and herbaceous vegetation, but requires additional processing of raw lidar point clouds [53]. “

However, we would like to emphasize we did not do any fieldwork for this study and thus do not have any field photos or on-the-ground examples. All data were publicly available.

Round 2

Reviewer 2 Report

The authors have been revised the manuscript as suggested

Reviewer 3 Report

Dear Authors,

The revised paper is suitable for publication in the RS journal. My opinion is that this paper would be much better wits fieldwork for this study, which would test the real performance of the presented methodology.

Regards