Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents an early-season field reference dataset of croplands in Goiatuba, Brazil, within the Cerrado biome. It addresses the critical gap in up-to-date in situ data for tropical agricultural remote sensing, adheres to FAIR principles, and demonstrates rigor in data collection and formatting. The dataset holds significant value for training remote sensing classifiers, benchmarking studies, and early-season agricultural mapping. Before publication, it needs some revisions to address gaps in data detail and methodological clarity.

The manuscript states "an RGB color composite mosaic derived from Sentinel-2 images acquired on 23 October 2025 and 2 November 2025" was used but does not mention preprocessing steps (e.g., atmospheric correction) or spatial resolution.

The "conservative mapping strategy" lacks operational details. It only mentions "removing internal non-crop areas" but not thresholds (e.g., minimum size of water catchment basins to exclude).

The authors are advised to supplement references, including "Heterogeneous pressure on croplands from land-based strategies to meet the 1.5 °C target", at specific gaps: first, when discussing Goiatuba’s role as a Cerrado agricultural frontier (now only supported by Reference [11]), to link local dynamics to global climate goals; second, when noting the post-2020 early-season in situ data gap, to add 1-2 recent studies on tropical early-season mapping challenges; third, when introducing FAIR principles (now only Reference [9]), to include 1 reference on FAIR’s application in agricultural remote sensing datasets.

The manuscript currently lacks explicit and systematic accuracy validation descriptions for the dataset. Specifically, it mentions using a Garmin 78S GPS receiver for coordinate collection ("we recorded geographic coordinates using a Garmin 78S GPS receiver in decimal degrees and referenced to the World Geodetic System (WGS-84) datum (EPSG: 4326)") , delineating plot boundaries in QGIS with a conservative strategy ("plot boundaries were later delimited in QGIS 4.44 software, following a conservative mapping strategy to minimize edge effects and internal heterogeneity") , and labeling fields based on in situ observations ("all reference fields were labeled based on in situ observations") , but fails to provide key metrics such as GPS horizontal positioning RMSE, boundary delineation overlap rate/average deviation, and LULC label inter-observer Kappa coefficient or image-verified accuracy, nor does it validate the dataset’s spatial representativeness against the study area's overall LULC distribution. To improve this, it is recommended to supplement GPS accuracy validation using control points (report RMSE), randomly select plots for RTK re-measurement to calculate boundary consistency, conduct inter-observer labeling and high-resolution image verification for LULC labels (report Kappa/accuracy), and use chi-square tests/nearest neighbor indices to confirm spatial representativeness, thereby enhancing the dataset’s reliability and usability.

Author Response

Please, see the atachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

your work is rather clear and I have no major comments.

Please find below some suggestions that I hope wil guide you in improving the manuscript and its impact:

• line 28: please check the version of QGis, as it sounds wrong
• section 1 title: it might help using a more common "Introduction" here, as you revise the literature
• l. 66-70: this is not really important for the article, so I suggest taking it out and moving to the Acknowledgement section
• the first section could be expanded by providing more details on why you selected this specific area, and what could be the transferability/reusability of the data (e.g., is the study area a good representation of other areas across Brazil or worldwide?)
• Fig. 1 could be of higher quality
• Please avoid numbering plots in Fig. 2, and eventually consider providing a better link with the classes reported in Fig. 1 (e.g., add the class colors along with the names in Fig. 2)
• l.132: please add some more characteristics about Sentinel images (e.g., resolution, row/path, cloud cover). The more details you provide, the better it is for the study's reproducibility
• Fig. 4b is very unclear. I suggest trying to provide a better representation
• l.161-163: it should be better described why using images covering such a long period, while your analysis accounted for just a few days. Do you expect changes in the workflow, considering a longer period?
• Can you provide some references justifying the duration of the seasons in Brazil? This could help readers in transferring your approach to other countries
• Fig. 5: I suggest changing the scale for NDVI if you focus only on the circled areas. This would help in improving the readability and in better showing spatial differences

Author Response

Please, see the atachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks for the revision.