Combination of Sentinel-1 and Sentinel-2 Data for Tree Species Classification in a Central European Biosphere Reserve
Round 1
Reviewer 1 Report
Topic of integration Sentinel-1 and Sentinel-2 imagery for tree species discrimination is not new, I cannot clearly see what is the novelty of this study. The methods used are also very common. The results show that S-1 does not significantly improve classification accuracy, and multi-temporal S-2 imagery provides the highest accuracies in species classification, which was already concluded in many publications before. In addition, the accuracy assessment which is extremely important in such studies is not described, so the results are not fully reliable. Therefore, in my opinion the manuscript in its current form cannot be published in Remote Sensing.
Some further comments:
Line 17: scenarios = classification/models?
L44: but you don’t perform analysis for a large area, why is it important in your case then?
Direct references should be with authors names, not only reference number in the whole text (e.g. someone et al. [1] studied)
Line 85 - what exactly forest inventory?
Line 82-83 - actually the fact that it is located on the area of overlapping orbits is a plus rather, because imagery is obtained once every 2-3 days (and it seems that the whole Park area is covered also by orbit 79), therefore I would not consider it as a challenge.
Table 1 - why is azimuth and zenith angle information important in this context? I suggest removing
Line 96 - add how many images
Line 97 - why did you use band 9? It is not used in land applications.
Line 101 - 30 vegetation indices? why? Why so many? How did you choose them? So many of them are highly correlated, I would actually remove that part/reduce it or at least justify why the indices were used.
Line 105-106 how many acquisitions?
There are so many variables/additional metrics that you calculated that it is hard to follow. Maybe you could provide table graph (and/or workflow)
Tables 2-4 S2 = S1
Line 169 and further - I don’t think scenario is a good term, I would say “models” or something like that
Line 169 Table 4 = Table 5
How was the accuracy assessment performed? Did you use the reference samples described in 2.1.? Did you split them into independent training and validation data? This step is very important in order to obtain reliable accuracy metrics so should be performed (and described) carefully.
In addition, the Kappa coefficient is not currently used in classification studies anymore - see for example: Foody 2020 “Explaining the unsuitability of the kappa coefficient in the assessment and comparison of the accuracy of thematic maps obtained by image classification”
Due to the abundance of tables in the manuscript, I am not sure if Table 8 is necessary, maybe just put that information in the text.
Line 327 - Why?
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
The continuous changes in the state of the forest with its successive modifications require more and more new and sophisticated classification systems. This manuscript analyzes the performance of Sentinel 1 and Sentinel 2 in different scenarios.. Twelve tree species, seven deciduous and five coniferous of the Austrian Biosphere Reserve Wienerwald were classified using Breiman's Random Forest. This study is very important to classify the temporal evolution of the forest. The greatest increase in accuracy has been achieved by using multitemporal Sentinel-2 data. Seasonal Sentinel-2 composites have advantages over monotemporal classifications.
The manuscript presents an analysis which deserves to be published
Author Response
Dear Reviewer
Thank you very much for your feedback.
Best regards
Markus Immitzer
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
I find your study very interesting and scientifically sound. Comparing different methods of the Sentinel-1 and Sentinel-2 data applications for the tree species classification in forested areas has great significance. Your main conclusion is that multi-temporal Sentinel-2 data outperform Sentinel-1 data in tree species classification, and that the added value of microwave data is only marginal.
I have only a few minor remarks:
Line 169: You should refer to Table 5 instead of Table 4
Lines 287-288: Figure 4. (a) should be: “F1-scores of Scenarios 1 to Scenario 3 and (b) should be: “F1-scores of all expressions of Scenarios 6 to Scenario 9”.
Lines 300-303: “The fact that the species within the conifer group were separated with satisfactory accuracies using microwave data – but not the deciduous species – is possibly related to larger structural differences between conifers (as compared to structurally more homogeneous deciduous species).”
As an ecologist, I question this statement. Deciduous species are more structurally diverse than conifers. For example, conifers Pinus nigra and Pinus sylvestris, two structurally very similar species, were successfully discriminated in your study. The possible explanation could be in the structure of the forest itself (species composition, dominant species) and the size of the pixel. Is the Biosphere Reserve Wienerwald natural forest and if yes, what are the main plant communities? It is expected that dominant species (with higher coverage) are better discriminated than others.
I wish you all the best in your future work.
Kind Regards
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The article brings another interesting study of the usability of Sentinel data to identify the tree species composition of forest stands in Central
European conditions. Compared to other similar studies, it benefits from the large number of derived parameters used in the classification. The research
design is logical and the presentation of results is clear and concise. The individual parts of the article are well written.
The following information should be added to the article:
1. The description of the study area should provide information on altitude. Wienerwald is a hilly area. Has altitude been taken into account? What is
the distribution of tree species according to altitude? Could including this parameter improve the results for this type of area?
2. A table at the beginning of the results, providing a summary of the accuracies of all scenarios, would help
The text also contains a few minor technical inconsistencies:
Tables 2-4 provide summary of S1 scenes, not S2 (correct the table names)
There are 2 tables 8
Author Response
Please see the attachment
Author Response File: Author Response.pdf