Increasing the Thematic Resolution for Trees and Built Area in a Global Land Cover Dataset Using Class Probabilities

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript is well-structured and written. It introduces a timely method for plant bio-characteristics extraction from high-resolution satellite image. I enjoyed reading this paper. I only have one quick comment to help improve the rigor of this work. Have the authors ever considered including the tree/plant phenology information into the classification process? e.g., you can explore using the TimeSat software to extract such data and how do you think this could enhance your research?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript explores the potential of enhancing thematic details in land cover classification using class probabilities from the Dynamic World CNN model within Google Earth Engine. The study applies this technique to subclassify forest and built-up areas in temperate and tropical regions and evaluates accuracy using reference datasets and transferability to untrained watersheds. The work aligns well with the scope of the journal Remote Sensing, as it leverages derived remote sensing data products to advance land cover modeling applications. While the manuscript is technically sound and relevant, its unique contribution and broader implications are not fully emphasized. As a reviewer, I believe the manuscript can be reconsidered for publication once the authors address all major revisions to clearly articulate its novelty and impact.

 

Certain concerns need to be addressed to improve this manuscript. The authors need to include the revisions in the manuscript file.

1. In section 1, the authors should address the gaps in existing studies and illustrate how this study bridges those gaps. Rather than tuning this section for the case study area, the authors should highlight the main relevance of this kind of work in the global research community.
2. In section 2, a flowchart of work is recommended that details how the study methods are implemented to achieve the objectives of the study.
3. Based on section 2 (2.3 and 2.4), it is highly recommended to include a table showing the spatial, temporal, and spectral resolution of the datasets used, their sources, description, and span of analysis.
4. Sections 2.3 and 2.4 do not look comprehensive. Please include sub sections highlighting the methods and hypothesized result plans from each section.
5. Tables 1 and 2 indicate variations in land cover area discretization from the 2 sources, NLCD and DW. However, clear justification for this is missing. Please explain.
6. Figure 5 explains the land cover classification from three different methods. Please discuss well the implications of each of the methods through which the authors can validate the best employed methodology.
7. Is there any methodical validation done in this study? In section 3.2, three methods of land cover classification trends are analyzed. However, there is no mention of ground truth estimation or real data validation. Please clarify.
8. Section 4 lacks strong discussion of the interpretation of human classified DW against the benchmarked DW model. Please justify.
9. The figures in the manuscript look good.
10. A section ‘Limitations of the study’ is recommended.
11. The reference and citations for sections 4 and 5 need to be enriched and formatted according to the reference formatting style of the journal.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The article presents the methodology to automatically classify the landscapes using convolutional neural networks. The article has a good structure and the presented problem is interesting. The conclusions answer the objectives and the quality of the presentation is good. 
However, the article has several major defficiencies, such as:
- the article lacks the methodology section for CNN architecture (hyperparameters - number of filters, layers, neurons);
- data preprocessing for the model is not described;
- metrics (equations) for the evaluation of results (how the values in table 1 and table 2 were obtained) are not defined.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors provide equations for 4 metrics: User's accuracy, producer's accuracy, overall accuracy, and proportion area. However, in the tables 1 and 2, only user's accuracy is given. What is the purpose of the other metrics and where are they used? 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf