Exploring Distributed Scatterers Interferometric Synthetic Aperture Radar Attributes for Synthetic Aperture Radar Image Classification

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

The following is the review of the article titled “Exploring DS-InSAR Attributes for SAR Image Classification” submitted to Remote Sensing by Wei et al.

 

The study evaluates the usage of multiple SAR imagery attributes for land cover classification. While many studies used polarimetric parameters along with SAR backscatter for land cover classification, this study uses interferometry properties (coherence and phase) and pixel similarity properties to achieve higher classification accuracy. I appreciate the authors’ effort in bringing out the study. I have the following questions or suggestions to further improve the papers’ understanding.

 

 1. The study mainly focuses on using SAR backscatter and coherence properties. While I see the majority of the classification is achieved with backscatter, the additional parameters coherence and SHPs are only adding to it. I wonder if the authors have tested multiple polarizations to achieve better classification than using coherence. I feel the same methods will achieve better results if other polarization information is added, since polarization of the SAR wave is more sensitive to the scattering/surface properties than coherence. Lower coherence will be similar over water and large vegetated surfaces. Polarization is more sensitive to the structure and can better capture the difference between road and building, where as they exhibit similar coherence.

 

2.  In line 150-151, it is mentioned that NxN coherence matrix has amplitude information in the diagonal elements. In my opinion they should be simply unity as it is coherence between the same images. Please address that.

 

3. It is not clear, what attribute the authors are using as input to the RF algorithm when they refer to “number of homogenous pixels”. Is it the number of homogenous pixels that they found in the window (ex:13x13)? Please illustrate it better.

 

4. In section 2.1.4, it is mentioned that equation (5) holds good for PS pixels. In my opinion, the difference between the left- and right-hand sides is close to zero for PS. It is not entirely equal. Please correct it [refer to Ferretti et al. 2011; Zheng, Fattahi, Agram et al. 2022, TGRS].

 

5. In calculating the ensemble coherence (2.1.4), please clarify whether the phase used is wrapped phase or unwrapped phase. If the phase is unwrapped then, unwrapping method errors can propagate into the ensemble coherence estimation.

 

6. In section 4.5, it seems the analysis conducted to determine the optimal window size is lacking enough sampling. While the influence of the window size is clear, I suggest the authors explore further sizes to determine an optimal window size for SHP selection. If it has to be limited at 15x15, kindly provide an explanation for it. From the results in table 4, there seems not a big difference between 11-, 13- and 15-pixel wide windows.

 

7. While the results look fine on the study area, can the authors explain more on how this method can effectively be extended to other urban areas. I suggest if an example from another city in a different climate can be added.

8. The method seems to be lacking in explaining the influence of seasonality in the data, since coherence and backscatter can vary seasonally in many areas. Does the method works only if one has a large temporal dataset or if a smaller time span dataset (ex 3-4 months) can work too?

 

9. Please add more citations to section 2.1.1 where scattering mechanism of different surface types are discussed.

 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This paper explores the application of Distributed Scatterers Interferometric Synthetic Aperture Radar (DS-InSAR) attributes for land cover classification in urban areas. It investigates the utility of coherence matrix, statistically homogeneous pixels (SHP), and ensemble coherence, along with traditional backscatter intensity, to enhance classification accuracy. Using a random forest classifier, the study evaluates the performance of these features on a dataset from Shenzhen, China, comprising TerraSAR-X satellite images.

 

1：The paper does not clearly state its hypotheses or specific research questions.  Clearly articulate the hypotheses or research questions in the introduction. For example, "This study hypothesizes that the integration of DS-InSAR attributes will significantly enhance the accuracy of land cover classification in SAR imagery compared to traditional methods."

2：The methodology section is thorough but lacks detail in the description of the random forest model parameters and the criteria for selecting training and test sets. Please include specific details about the random forest parameters (e.g., number of trees, max depth). Clarify how the training and test sets were selected and whether cross-validation was used to ensure robustness.

3：The feature selection process using the FaSHPS algorithm is not clearly explained. Please provide a more detailed explanation of the FaSHPS algorithm and its role in selecting statistically homogeneous pixels. Include any assumptions or limitations associated with this method.

4：The interpretation of results could benefit from a deeper statistical analysis, particularly regarding the significance of the observed improvements.  Include statistical tests to validate the significance of the observed differences in classification accuracy. Discuss potential sources of error and how they might have impacted the results.

5：Some figures (e.g., confusion matrices) are not clearly labeled, making it difficult to interpret the results.  Ensure all figures and tables are clearly labeled with titles, axes, and legends. Consider adding more detailed captions to explain the content and significance of each figure/table.

6：Include the following references to strengthen the discussion on range ambiguity suppression and landslide analysis, which are relevant to the study's focus on SAR image classification.

—An advanced scheme for range ambiguity suppression of spaceborne SAR based on blind source separation

Reason: This reference provides insights into advanced techniques for improving SAR image quality, which can be beneficial for enhancing classification accuracy in DS-InSAR applications.

— Identification and Analysis of Landslides in the Ahai Reservoir Area of the Jinsha River Basin Using a Combination of DS-InSAR, Optical Images, and Field Surveys

Reason: This study demonstrates the application of DS-InSAR in identifying and analyzing landslides, highlighting the technique's versatility and potential applications in different geological contexts.

Comments on the Quality of English Language

1： There are several grammatical errors and awkward phrasings throughout the manuscript.

Recommendation: Carefully proofread the manuscript or use a professional editing service to correct grammatical errors. Examples include:

Page 1, Line 11: "However, due to the unique imaging methodology, interpreting SAR images presents numerous challenges." Consider rephrasing for clarity.

Page 4, Line 81: "This implies that pixels representing the same land feature will have similar statistical properties in a time series." Rephrase to improve readability.

 

2： Some references are missing publication details.

Recommendation: Ensure all references are complete and formatted according to the journal's guidelines. For example:

Reference [1]: Provide full citation details, including authors, title, journal, volume, pages, and year.

 

3：The narrative flow in some sections is disjointed, making it difficult to follow the argument.

Recommendation: Revise sections to improve coherence and logical flow. Use transition sentences to connect ideas and guide the reader through the argument.

Author Response

Please see the attachment

Author Response File: Author Response.docx