Refining European Crop Mapping Classification Through the Integration of Permanent Crops: A Case Study in Rapidly Transitioning Irrigated Landscapes Induced by Dam Construction

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript aims to refine the classification of the EU crop map to include permanent crops. The study has certain significance. However, the overall innovation is insufficient, particularly in the methodology section, where the classification methods and feature evaluation methods used are conventional. Specific issues are as follows:

Major Concerns:

1. The manuscript was not written according to the journal's template. It includes a Highlights section, and the text in this section is too long. Generally, each item in the Highlights should not exceed 25 words.

2. The logic between the various parts of the methodological technical route is not clear enough. The methods for each step are not fully presented.

3. There are multiple instances of tracked changes in the manuscript, such as in Line 499 and Line 703.

4. The Discussion section is too simplistic. It is recommended to structure it with subsections.

5. Figure 4 is missing the x-axis and y-axis labels.

Author Response

Reply provided in the pdf attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Summary of the Paper

This study addresses a significant limitation of the EU Crop Map, which groups permanent crops into the "woodland and shrubland" category. Using the Alqueva irrigation area in Southern Portugal - a rapidly changing agricultural landscape following the construction of a large dam - as the case study area, the authors employ Sentinel-1 and Sentinel-2 imagery with a Random Forest classifier to improve parcel-level crop type classification. The model has the ability to distinguish between permanent crops such as vineyards, almond groves, and different olive grove cultivation systems (traditional, high-density, super-high-density). Overall accuracy is 91%, with good performances for olive and almond groves, fair for vineyards, and poor for other permanent crops. The paper also evaluates the impact of reducing training data size on model accuracy, showing that while some classes are resilient to less data, others (e.g., vineyards) perform poorly.

Major comments

Lines 144-151: the objectives are clearly stated, but novelty compared to previous work should be more explicitly emphasized. For example, how does this method improve EU Crop Map 2022 beyond adding permanent crops? A further paragraph on methodological advances would strengthen the introduction.

Lines 251-338: the methodology section is too verbose and reads more like a technical report than a scientific article. Consider summarizing some of the operational details (e.g., S1/S2 acquisition parameters) and moving long descriptions to Supplementary Material. This would make the reading easier.

Lines 341-380 (and also 353-357): hyperparameter tuning description is not detailed. Please Explain the actual parameter ranges and ultimate selected values for the Random Forest model.

Lines 402-518: the analysis is good, but the discussion could be further developed. For instance, explain more explicitly why some classes deteriorate more with smaller training samples. This would strengthen the discussion.

Lines 552-588: while the intersection analysis is informative, the authors should provide a quantitative accuracy assessment (e.g., Kappa or F1-scores) for this comparison, despite the fact that COS 2018 is not an official reference. This would provide more concrete evidence of model reliability.

Lines 592-712: greater critical assessment of shortcomings would be appreciated in the text. For example, potential transferability of the model to other years or areas is not discussed, nor are implications of class imbalance for future operational mapping.

Minor comments

Line 29: "It is worth to build…" should be "It is worth building…."

Line 412: typo "Mead Decrease Accuracy" should be "Mean Decrease Accuracy".

Line 630: the term 'peanuts' appears - this is likely a printer's mistake or an incorrect reference to the crop; please check.

Author Response

Reply provided in the pdf attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript presents a case study focused on improving the CORINE Land Cover (CLC) classification by integrating permanent crop classes, particularly in the context of rapidly transitioning agricultural landscapes affected by irrigation expansion due to dam construction. Using a site in the Baixo Alentejo region of Portugal as a testbed, the authors compare the CLC with a refined classification derived from Sentinel-2 time-series data and Random Forest modeling, with particular emphasis on the role of permanent crops (e.g., olive, vineyard, orchard).

The topic is timely and relevant, particularly given ongoing discussions on updating land use datasets in the EU to better support agricultural policy and climate adaptation planning. The paper also contributes to the methodological advancement of crop-type mapping in fragmented Mediterranean landscapes.

However, I think the current manuscript needs several improvements in framing, justification of methodology, interpretation of results, and connection to broader policy relevance.

Major comments

1. While the manuscript presents a compelling use case, the framing in the Introduction is currently too narrow, based on my observations. The reader would be left unclear about the broader scientific and policy stakes of improving permanent crop classification. I suggest clarifying (1) the limitations of current CLC product in supporting agricultural monitoring, (2) the implications of dam-induced land use change, and (3) the novelty of this study compared to existing efforts like LUCAS, S2Agri, or LULUCF reporting systems.
2. The Random Forest approach is reasonable and well-established in crop classification. However, the rationale behind variable selection, time period, and specific classifier tuning is not clearly explained. What are the temporal features used from Sentinel-2 (NDVI, EVI, others)? Was any phenology-based feature engineering applied? Please describe the input features and preprocessing pipeline more explicitly.
3. The ground truth data appears to be drawn from 2021, but it is unclear how many permanent crop parcels were used for validation and how representative they are. Is there any temporal mismatch between CLC reference years and classification years that could influence performance? Clarifying the sample design and validation strategy would improve confidence in results.
4. The results show improved discrimination of permanent crops, but interpretation remains largely descriptive. For instance, the study does not explore confusion patterns (between orchard and vineyard), nor does it link land-use change trends to irrigation expansion explicitly with spatial or temporal evidence.
5. The discussion mentions EU-level CLC modernization efforts but does not convincingly connect the findings to those processes. What are the implications of including permanent crops in continental-scale mapping? Could this be generalized to other semi-arid Mediterranean basins with irrigation expansion (e.g., Spain, Greece)?
6. The manuscript’s structure can be improved for flow. Sections 2 and 3 are dense and difficult to follow without diagrams or schema of workflow. Consider restructuring to include a clear stepwise Methods flow (Data → Preprocessing → Classification → Validation → Comparison with CLC).

Major comments

Abstract: Consider quantifying improvements over CLC (e.g., overall accuracy gain, F1 score of permanent crops).

English usage: While mostly readable, the manuscript would benefit from light language editing for flow and clarity (e.g., avoiding redundancy, improving transitions between sections).

Citation of relevant literature could be strengthened, especially on previous attempts to classify Mediterranean perennial crops using Sentinel-2.

Author Response

Reply provided in the pdf attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

the authors have revised all the comments

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for sending in a revised version of your manuscript together with a detailed cover letter addressing the comments.

I have carefully reviewed your revisions and am now very happy to inform you that the changes you have addressed do indeed satisfy all the comments I raised.

Congratulations and good luck on the publishing of your work.

Best regards.

Reviewer 3 Report

Comments and Suggestions for Authors

All my concerns in the first round have been well addressed.

Thanks for your efforts.