Semi-Automatic Extraction of Hedgerows from High-Resolution Satellite Imagery

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The study addresses a very interesting topic for rural land management using remote sensing and geospatial technologies. Some improvements are necessary before it can be considered for publication, to better clarify some points for the benefit of the readers.

The paper focuses on hedgerows; therefore, I would suggest that the title mentions hedgerows, instead of woody elements.

Connection between hedgerows mapping and hedgerows ecological importance should be described better, to clarify how hedgerows mapping can support biodiversity conservation and ecological infrastructure enhancement.

Provide more information about how the 'Small woody features' inventory released by Copernicus Land Monitoring Service is realized, to allow to better understand the added value of the proposed method.

The object based extraction of hedgerows is performed both on PlanetScope and Copernicus Sentinel-2 imagery. This is clear in the abstract, but it’s not explained in the introduction and goals, add more explanation. Moreover, the authors in the discussion compare the results obtained using sentinel 2 and planetscope with the Copernicus Small Woody Features dataset. And this should be also clarified in the goals.

Training phase is not mentioned in the workflow description.

The Nearest Neighbour classifier was trained with samples manually labelled in eCognition. The number of training samples for each class within the study area was 60: explain better how training samples were derived (validation reference data are specified, training data are not), and justify training size (60), their geographic extent and spatial distribution across the study area.

Planet's imagery is defined at a relatively low cost. Provide more information and numbers. More discussion about implementation opportunities for public institutions would be useful. Also including considerations about the software used (eCognition).

Reason for choosing two different cloud cover thresholds for Copernicus and planetscope should be explained.

Set of features extracted for the classification process should be better motivated. Some of them are poorly discussed. What other alternatives were not considered and why? Was there any comparison to select the best performing ones?

NDBI is mentioned without full explanation of the acronym.

In this study, shadows were explicitly addressed by identifying them as a distinct object class derived from the objects initially classified as "Hedgerow": clarify.

Check section 2.2.5. Validation.  Some sentences in the first part seem incomplete. Clarify why validation was done using a two-step approach, first eCognition segmentation and then visual interpretation on a subset. Advantages compared to visual interpretation? Better distribution and representativeness across the study area? Time reduction? Why it was not used to derive the training dataset?

In general, choice of features, indicators, sizes, thresholds, etc. should be better clarified.

More conclusive and comprehensive final considerations about profitability and sustainability of using the proposed approach and paid satellite data against free satellite data would be very useful.

Please discuss potential applicability to other contexts and geographic areas, other landscape types.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article presents the results of a case study concerning an attempt to develop a semi-automatic hedgerow extraction method based on satellite images (PlanetScope and Sentinel-2). The subject matter is important in the context of monitoring the implementation of the EU Common Agricultural Policy and maintaining/increasing biodiversity in Europe.

The research work was carried out correctly, although I do not understand the intentions of the Authors regarding the method of defining the validation dataset based on the aerial orthophotomap. The choice of segmentation as a method for defining the reference dataset in this case could have introduced additional errors, and consequently could have affected the correctness of the accuracy assessment of the obtained results. In my opinion, the argumentation for the adopted method of obtaining reference dataset should be better justified and discussed. The issue of whether the accuracy assessment was conducted in terms of the number of polygons of correctly detected hedges or in terms of area (pixels that were actually hedges) also requires further clarification.

Some of the conclusions drawn from this case study were obvious and predictable even before the comparative studies were conducted (e.g. better results based on PlanetScope images than Sentinel-2 images), but I understand the desire to confirm the predictions quantitatively.

In my opinion, the study would be much more valuable if it was based on more than one research area. Then it would be possible to assess the repeatability of the proposed semi-automatic hedge extraction method based on satellite images in different contexts (arable land layout, different landscape type, etc.). For example, the agricultural land configuration may be important in relation to the performance of GLDV Entropy (quick 9/11) (90°).

In the discussion it would be worth presenting an in-depth analysis regarding the comparison of hedgerow extraction results based on Sentinel-2 data obtained by the Authors and the Copernicus SWFs dataset. Apart from that, the discussion of results made little reference to the results of other authors.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

1. The paper lacks a thorough analysis of the research motivations and innovation. The explanation of the core problem addressed in this work remains underdeveloped and requires further elaboration.
2. A succinct summary of the paper's primary contributions should be explicitly provided in the introduction or conclusion section.
3. The current Figure 2 appears overly simplistic and fails to illustrate the technical details of the proposed method. It is recommended to include a comprehensive network architecture diagram to enhance clarity.
4. The authors should carefully revise the manuscript to eliminate minor errors. For instance, the workflow description in Section 2.1 ("four main steps") contradicts the actual three-step structure listed, which may confuse readers.
5. The "Image Segmentation" subsection (Section 2.2.1) appears to focus solely on parameter tuning. If novel methodological improvements have been implemented, the authors must explicitly clarify how their approach differs from existing methods.
6. The specific method of feature extraction in section 2.2.2 is not clear enough. A step-by-step explanation of the implemented techniques and algorithms is strongly recommended.
7. Systematic ablation experiments are required to validate the effectiveness of individual modules in the proposed framework. Quantitative analysis of each component's contribution should be included.
8. The experimental section lacks rigorous comparisons with recent SOTA approaches. Expanded benchmarking against relevant baselines is necessary to demonstrate the method's competitiveness.
9. The manuscript does not fully comply with the formatting guidelines of Remote Sensing. The authors must rigorously revise the document structure, citation style, figure captions, and table layouts to meet journal standards.

Comments on the Quality of English Language

The manuscript is generally written in comprehensible English, though minor grammatical inaccuracies and occasional awkward phrasing are present (e.g., inconsistencies in verb tense). While these issues do not severely hinder readability, thorough proofreading by a native English speaker or professional editing service is strongly recommended to refine syntax, improve clarity, and ensure adherence to formal academic writing conventions.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

1. Lines 59-72 elaborate on the importance of small woody features and the data provided by the Copernicus Land Monitoring Service but do not cite relevant references to support the arguments. It is recommended to include representative references to substantiate the claims.
2. Is “[12]1” in line 78 a formatting error?
3. Is there an error on the left side of Figure 1 in line 134? Additionally, the font size of the latitude and longitude in the figure is too small, making it less intuitive. It is recommended to improve this aspect.
4. In line 156, it is stated that there are four steps, but only three are listed. Was one omitted?
5. It is recommended to briefly describe the structure of the paper at the end of the Introduction.
6. In line 256, the font of [30, 33, 34] is inconsistent with other related reference numbers. Please modify it and check the relevant content to ensure uniform formatting.
7. In the table at line 323, does the use of commas in “8,21” and “52,01” comply with the standard format? Please verify and check the relevant content in other tables to ensure consistency.
8. In Table 7 at line 375, please standardize the format of "Count" and "count." Additionally, in the "Total" row, if certain items are not needed, please use "-" or "NA" as appropriate markers.
9. Please adjust the numbering of the references to match the appropriate citation format and conduct a thorough review of the entire paper.

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

the authors have addressed the comments and improved the paper 

Author Response

We thank the Reviewer for the positive feedback. We are glad to know that the revisions have improved the quality of the manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

I thank the Authors for taking my suggestions into account. In its improved form, the paper contains broader justifications and explanations of the methodology used, as well as an in-depth discussion of the results obtained by the Authors.

Author Response

We thank the Reviewer for the positive feedback. We are glad to know that the revisions have improved the quality of the manuscript. 

Reviewer 3 Report

Comments and Suggestions for Authors

The revised manuscript has addressed some of the concerns, but two critical issues remain inadequately resolved in the authors' response:

1. Insufficient ablation studies: Quantitative analyses of individual modules remain absent in the manuscript; The weights mentioned in Section 2.2.1 lack empirical validation in the experimental analysis.

2. Deficiency in comparative experiments: The experimental section fails to provide a systematic comparison demonstrating the superiority of the proposed classification  method over existing approaches.

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf