Early-Season Crop Classification Based on Local Window Attention Transformer with Time-Series RCM and Sentinel-1

Round 1

Reviewer 1 Report

The authors described important issues in the examined area of early-season crop classification by integrating the local window attention transformer with time-series RADARSAT Constellation Mission data and Sentinel-1 data. The effective application of the LWAT with the time-series data constitutes the strong and original point of the paper, as well as the concept of comparative analysis of the RCM and Sentinel-1 performance, and determination of the optimum date and the corresponding phenological stage with respect to improving the early-season crop mapping. I have to emphasize importance of that part of the paper in which the Authors compared the LWAT with traditional machine learning methods and deep learning models (SVM, RF, VGG-16, AlexNet, 2DCNNs, ResNet-50, and ST in order to manifest the effectiveness of LWAT in early-season crop mapping.

The approach is grounded on solid knowledge supported with the adequate and comprehensive literature review, and the methods were selected accordingly to the objective of the paper. The Authors convincingly showed practical importance of developing the framework comprising two components – a feature extractor and a local window attention module for the early-season crop mapping. 

Several issues should be addressed by the authors before it is printed:

·       185-188 “In our study area, the typical revisit time for RCM data is around 8 days, allowing us to capture changes and variations of crops at a relatively high frequency. The spatial resolution of the RCM data used in our study is 5 m, providing fine details for crop identification” – please, provide better explanation concerning the choice of 8 days and 5 m.

·       204-205 “To mitigate the influence of speckle noise, we employed multilook processing” – good approach, please, expand the procedure description.

·       213-214 “terrain correction was applied to rectify topographical distortions in the SAR imagery” – as above, please, expand the description.

·       268-270 “Moreover, to compare the performance of the Sentinel-1 and RCM for early-season mapping, the incremental procedure was applied to as applied to their combination as well as the respective time series” – please, clarify meaning of the sentence.

·       424 “due to it lacks the co-polarization band” – please, give more profound explanation.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General comments

Leveraging advanced technologies such as RADARSAT Constellation Mission (RCM) and Sentinel-1 synthetic aperture radar (SAR), the research introduces the effective local window attention transformer (LWAT) for time-series SAR data analysis. The comprehensive evaluation in Southwest Ontario, Canada, reveals the superior performance of LWAT over various machine learning and deep learning methods, achieving an impressive F1-score of 97.96% and a Kappa coefficient of 97.08%. The study not only highlights the excellence of LWAT but also identifies crucial phenological stages for accurate early-season crop mapping, showcasing its potential for in-season production forecasting and decision-making. Hence, this manuscript is publishable after below given minor ammendments.

 Specific comments

Please avoid use of ‘we’ in whole manuscript

Tables 2 and 3, Use Percent (%) only once in the column headings

Conise the conclusion section

The English language is fine.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript compared the performance of LWAT with other machine learning and deep learning methods on early-season crop classification by using time-series RCM and Sentinel-1. The results and discussion were clear and comprehensive. The manuscript could be expressed more clearly after some minor revision.

1.Why RCM and Sentinel-1 data were used in this study? Some background information should be included in the introduction part.

2.The comparison results of LWAT and other 7 methods showed suddenly in “Experiment and results” part. The information of all methods should be described in the methods part.

3.What did the values in Table 2 and Table 3 stand for? The accuracy? What is the meaning of the numbers in bold? The detailed explanation are needed for the table.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf