Fast Fusion of Sentinel-2 and Sentinel-3 Time Series over Rangelands

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Abstract:

You stated (-72% on MAE) with the abbreviation of MAE for the first time. To make the reader understand the meaning of MAE (Mean Absolute Error). Please add what MAE indicated.

The introduction:

You expressed the literature review of related research to your study, and you mentioned that you are introducing EFAST, the algorithm that you proposed. You need to state clearly your objectives at the end of the introduction.

The Materials and Methods:

·     You demonstrated that “this paper focuses on two 16km2 areas, both located in the Senegalese Louga region (Figure 1). Though our method is designed to predict high resolution phenological products at continental scale, for all African rangelands, focusing our analysis on those two areas allows to better visualize the performance of our approach over diverse small-scale features”. Do you think these two small areas can be approved using your approach on a regional and continental scale? How can that be possible with image-covering issues such as cloud cover in large-scale areas and wet seasons?

·     You stated that after calculating the NDVI for both Sentinel 2 and 3 images, you resampled Sentinel 3 to Sentinel 2 spatial resolution. Do the created images from resampling sentinel 3 to sentinel 2 spatial resolution present a sufficient spectral and radiometric resolution of 10m images that can be used in image processing and analysis to produce accurate results?

·     You mentioned that the three interpolation methods (EFAST, STARFM, and the Whittaker filter) will be compared in two experiments. Also, you added that these time series will be compared visually to the in-situ data, from 2019 until the end of the year 2022. Are you talking about processes you did in this study or will do in the future?

Results

You stated that EFAST accurately identified the regions of high and low values, but the resulting image does not present the same contrast as the Sentinel-2 image (Figure 6a): the NDVI of bare parcels is overestimated and, conversely, the vegetation index over the surrounding grasslands is underestimated. Also, it is apparent from Figure 8.3 that EFAST and STARFM overestimate the phenological variations of point number 3 (Figure 7), sometimes leading to negative NDVI values (2019 and 2022). Moreover, the Sentinel-3 signal over this bare area is corrupted by the surrounding grasslands. The temporal averaging of EFAST gives a lower weight to individual cloud-free pixels, leading to corruption of the phenological signal, even at periods of low cloud cover. All of you mentioned are a disability of EFSAT. What do you think needs to be done to solve or overcome these issues?                                 Discussion

You mentioned that the surrounding parcels corrupted the Sentinel-3 signal over the grass between the agricultural parcels (Figure 9, dot 5) in the second study area.

You stated that EFAST has limitations in heterogeneous areas, which leads the EFAST model to perform poorly. What do you suggest to solve the problem?

Conclusion

You provided a summary of your study and the findings.  Also, you stated that you have limitations on using the EFSAT method, and you got poor results in heterogeneous areas, which indicates the importance of the improvement. Elaborate on your conclusion by including some future recommended work and suggestions that could solve these issues.

Comments on the Quality of English Language

After reading the manuscript, I found some grammar mistakes in the text parts, including sentence structure, subject-verb agreement, and punctuation.

     For example,

In the introduction:

However, in seasonally dry ecosystems such as the vast savanna rangelands of Africa, vegetation growth typically coincides with periods of frequent precipitation and therefore (and, therefore,) cloud cover, leading to prolonged periods without Sentinel-2 data (Figure A1).

In Materials and Methods: Though our method is designed to predict high resolution phenological products at continental (a continental) scale, for all African rangelands, focusing our analysis on those two areas allows to (us to) better visualize the performance of our approach over diverse small-scale.

The spatio-temporal fusion algorithm STARFM [12] with the following parameters: four classes and (a) window size of 31 pixels.

Free Sentinel-2 image as input data. Comparison the (of the) performance of our method with STARFM allows us to verify whether the increase in computational efficiency.

In the results:

The next subsection documents the ability of the Whittaker 246 filter, STARFM. (STAREM,) and EFAST to reconstruct the phenological profile when no data is avail-247 able for multiple months, which is common in tropical savannas (Figure A1).

But their values were lower than measured (the measured) data.

Changes in the types of the regional (regional) ground features.

In discussion

Harmonized and uninterrupted timeseries (time series) on vegetation phenology are crucial.

Timeseries (time series) at high resolution, EFAST is therefore expected to have (a) large potential for.

short distances, driven by local environmental gradients such catenal (as catenal) sequences and the localized impacts of herbivory and fire [28].

In conclusion

STARFM is associated with a slightly lower error than EFAST on both (in both) test areas.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I found the introduction to be well-written and informative. It highlights the importance of ecosystem monitoring on a regional or continental scale, particularly for biodiversity conservation, sustainable land management, and climate change mitigation. However, it would be helpful to include examples of international works that have been done outside the study's area. The methodology section and the results are also presented in a clear and concise manner. In general, the note suggests that EFAST is a promising approach for generating large-scale phenology products efficiently, surpassing other methods. However, my main critique is that the discussion and conclusion sections do not fully address the general limitations of this methodology.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have done a very good job in presenting the topic, the reasoning and the experiments done to obtain a refined NDVI map with high spatio-temporal resolutions.

Minor remarks:

1.      Abstract: abbreviations MAE , STARFM not introduced

2.      Figures can be smaller.

3.      Figure A1: mark the position of the test sites.

4.      Line 247: comma after STARFM instead of point.

 

5.      Table 1 and Figure 7: better to clarify that you refer to NDVI values.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This study presents a rapid fusion method for Sentinel-2 and Sentinel-3 time series, which was tested in a rangeland and an irrigated cropland area. Although this method is faster than STARFM, concerns arise regarding its scientific robustness and novelty. Firstly, the process accelerates STARFM by omitting the spatial averaging (weighting) step, crucial for minimizing the impacts of heterogeneity, land cover changes, and BRDF geometry alterations. Omitting this step could compromise the results, as evidenced by your findings. Considering STARFM's performance in heterogeneous areas is not exemplary compared to other fusion algorithms like ESTAFM, a change in STARFM causing an even poorer performance in heterogeneous areas should be made with extreme caution.

Secondly, the concept of temporal averaging lacks novelty. For instance, Rao et al. (2015) implemented a temporal weighting strategy, logically incorporating weights based on the reflectance/VI discrepancy between the prediction and reference imaging dates. Imagine a scenario with two reference image pairs at times t1 and t2, where t1 is closer to the target date t*, but a land cover change occurs between t* and t1. In such instances, assigning a higher weight to t2 than t1 could make more sense.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

After reviewing the authors' responses in which they addressed and corrected the issues stated in the first review, I have no more comments.