Using a Vegetation Index as a Proxy for Reliability in Surface Reflectance Time Series Reconstruction (RTSR)

Round 1

Reviewer 1 Report

In this manuscript, a new spectral Reflectance Time Series Reconstruction (RTSR) method was proposed and its performance is evaluated to solve the problem of missing optical remote sensing data and noise. But there are a few issues that need to be addressed:

1. Introduction, it is suggested that the author supplement the current research progress and methods of reflectance time series reconstruction. Then, there are too many paragraphs in this part and the key points are not prominent. Hence, it is suggested to further adjust and process.

2. 2.1. Test sites, the contents of the study area were incorrectly described. Five European countries were mentioned in the manuscript, but the names of six European countries were listed in the paper, suggesting further verification.

3. 3.1. Masking, there is mention of creating a 5-pixel (50m) buffer based on the distance to avoid missing clouds. Is there any scientific basis for this? Please provide the necessary foundation or explanation.

4. 3.2. Adaptive smoothing, it is mentioned that smooth values should be used for the next iteration, and high NDVI values are used as reliable values. How to determine the number of iterations? What is the high NDVI value that can be considered reliable? Is there any scientific basis? Please provide the necessary foundation or explanation.

5. Methods, the introduction of core algorithmic ideas is too general and hard to understand. It is suggested to supplement the core idea of the algorithm with more information, and use more figures, and tables to clearly show the core of the algorithm.

6. Figure 6, the legend style of each subgraph is suggested to be uniform.

7. Figure 9, the content is not clear and it is suggested to enlarge the font size.

8. Discussion, the content should focus on discussing the results rather than restating them. It is suggested to supplement the discussion of comparative analysis with other relevant studies and explain the results more comprehensively.

9. There is no conclusion in the manuscript and it is suggested to be supplemented.

10. In the manuscript, only the RTSR method was used to reconstruct B₂, B₃, B₄, B₈, and B₁₂. Why are these bands selected for reconstruction? What about other bands being reconstructed? Could this approach apply to other data sources?

11. Are there limits to the amount of RS data required for the RTSR reconstruction method proposed in this manuscript? If so, what is the number of upper and lower limits?

12. In the manuscript, there are very few indicators for the reconstruction evaluation effect of the RTSR method, which is not credible. It is suggested to add more indicators for evaluation or select some new areas for evaluation.

13. The manuscript writing layout is not reasonable, and the theme is not prominent. In each section, there is a list of multiple paragraphs, and there is a lack of logic between all paragraphs. It is suggested that the manuscript be rearranged into a whole and refined the title.

Author Response

Please see the attachment in PDF format

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors proposed a new reconstruction method for spectral reflectance time series based on adaptive smoothing.

The MS is of great importance because it deals with gaps in optical remote sensing time series which are highly affected by atmospheric perturbations.

 

I recommend accepting the MS in the Remote Sensing Journal due to the clarity of the text, the quality, and the importance of the content.

Author Response

Please see the attachment in PDF format

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors propose a spectral reflectance time-series reconstruction method, where smoothed values are adjusted to approach trustworthy observations. The method was tested on 100 sites. The manuscript is very well written and meticulous in its explanations. I would not characterise the proposed method novel but is definitely of value and the methods description, results and discussion are complete, thorough and convincing. Below are listed some minor comments. 

• Authors should briefly mention the limitations of the Sen2Cor SCL. Why didn’t the authors use more sophisticated cloud masking methods?

• Why was 7 selected for the SG smoothing window?

• I suppose the SG parameterization is dependent on the application of interest. Could the authors please discuss for which settings the chosen parameterization is pertinent and why? And how someone should think if another application is of interest (e.g., time-series vs single image based tasks). 

• Why does DTS perform this badly in Figure 3? Seems strange.

• The issues on evaluating the performance are helpful to set the scene. However, a lot of information is based on conclusions from the literature. It would be easier to follow if the authors mentioned a few more things about the papers they cite and as to how their conclusions are relevant - example:  lines 221 (why?), line 235 (how?)

Author Response

Please see the attachment in PDF format

Author Response File: Author Response.pdf

Reviewer 4 Report

Evaluation of the paper entitled:

 "On adaptive smoothing for reconstructing reflectance time series for vegetation monitoring” by Kempeneers et al.

General Comments:

• The paper explores the reconstruction of time series of NDVI reflectances from Sentinel 2 data. The results show good consistency and the presented reconstruction algorithm correlates well against reference data. 

Specific comments: 

• Plenty of small grammatical and spelling errors such as; missing brackets, missing points at the end of the sentence. Figures need visual improvements as many axis titles are missing or their font is too small to read. In some cases, there is missing specific information such as the sites of interest described in a bit more in detail or the “case with Ukraine” is introduced but not elaborated.

• The adapted time series smoothness index is a valuable addition to the evaluation of temporal reconstruction methods.

• The Whittaker smoother offers many advantages over the Savitzky-Golay, however the authors choose the Savitzky-Golay. The usage of Whittaker could have led to better reconstruction profiles.

Abstract: 

• Abstract is well written. Even though they introduce their reconstruction method, it would be informational to mention the specific filter used in their reconstruction algorithm.

Introduction: 

• DKD: In general, this section provides the fundamental information on this topic with providing the underlying necessities for temporal reconstruction.

Methods: 

• Generally I feel that the wording and information they provide is mainly part of the discussion. They evaluate and mention other studies whereas it is only the methods.

• Lack of explanation about the 100 sites of study. The authors say it's 2560x2560m but could have included a bit more spatial/pixel info…

Discussion:

• The discussion is generally well written, although I think several paragraphs from “Methods” belong in this section. Also, there is a lack of intercomparison with other studies. The authors only cited 4 references in this section.

Conclusion: 

This entire part is missing. Now, I understand that it could be a combined section with the Discussion, however I do not feel that the article and findings were adequately summarized/concluded. Furthermore, they bring in the processing of Ukraine which is a completely new info and hasn't been mentioned before.

 

Please also see suggestions in the pdf.

Comments for author File: Comments.pdf

Author Response

Please see the attachment in PDF format

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

After the author's revisions, the quality of this manuscript has significantly improved, but there are still some minor issues remaining.

1. In the adaptive smoothing method used in the study, high NDVI values are considered reliable. It would be beneficial to provide additional explanatory descriptions and supporting references for this assumption.

2. I suggest that the author add subheadings to both the Results and Discussion sections in order to enhance the logical structure of the article.

3. The paper deals with the creation of five-pixel buffer zones (50 m), and although the author provides some explanation in the discussion section, it is suggested that the author include a brief summary of previous studies in the Methods section to clarify why a five-pixel buffer zone was chosen and to demonstrate that using larger buffer zones is not effective. This will provide context for the reader and enhance the logical flow of the paper.

4. The RTSR method proposed in the study mainly evaluates the agricultural areas in Europe. As a spectral reflectance time series reconstruction method, it is expected to have ideal reconstruction accuracy for other background features such as buildings, vegetation, bare soil, water bodies, etc. However, this paper does not mention the effectiveness of this method for other land cover types. It is suggested that the author discuss the potential application and performance of the RTSR method for different land cover types to provide a more comprehensive evaluation of its applicability.

5. Finally, I agree with the reviewer's suggestion to use a Whittaker smoother instead of the S-G filter, as it can potentially lead to better results and keep up with current research trends. This will allow for more meaningful and informative contributions to the field.

Author Response

Dear reviewer,

Thank you for the suggestions and comments. Please see the attachment.

Pieter Kempeneers.

Author Response File: Author Response.pdf

Reviewer 4 Report

Thanks for considering my suggestions!

 

Author Response

Dear reviewer,

Thank you again for your suggestions and comments.

Pieter Kempeneers.