Sentinel-2 Poplar Index for Operational Mapping of Poplar Plantations over Large Areas

Round 1

Reviewer 1 Report

The manuscript “Sentinel-2 Poplar Index for Operational Mapping of Poplar Plantations over Large Areas” by Hamrouni et al. represents a satellite data (Sentinel-2) index based approach for mapping poplar plantations in France. The manuscript has a good introduction, interesting study question and well executed satellite data-based approach, but the manuscript should be improved considerably especially in the respect of critical discussion that considers the benefits and disadvantages of the methodology that is partly based on having database data as reference data instead of field measurements.

Major comments:

-The timing of bud burst, leaf emergence or leaf senescence needs to be discussed, since all the studied trees are deciduous and for an international audience it is not self-evident when poplars or other deciduous trees have leaf emergence in spring or leaf senescence in autumn in France. Do those phenological events coincide with the timing of the index in Fig. 4 starting to work in the spring and cease to work in the autumn?

-The User’s Accuracy (i.e. the proportion of true positive pixels in relation to the sum of true positives and false positives) would be relevant in addition to the Producer’s accuracy (the proportion of true positive pixels in relation to the sum of true positives and false negatives)

-I would suggest calculating UA, PA and F1 = 2⋅UA⋅PA / (UA + PA)

- Confusion matrices would be informative to check prediction against the reference

-The result section is unnecessarily long and therefore tedious to read. It should be shortened considerably. Only test set results should be given emphasis in the text, training set results can be found in the Tables and Figures. Empty sentences can be omitted (such as “The results of feature selection with SFFS on the pool of spectral bands (i.e. multi-band configuration) are summarized in Figure 5. These graphs provide for each number of features selected, ranging from 1 to 10, the best performance achieved in terms of average poplar producer’s accuracy (PA, based on 5-fold CV), and the best bands selected with their corresponding dates.”) and instead the references to the figures can be added at the end of the sentences that tell the result shown by the figure.

I would suggest that lines 189-190 and Figure 3 should actually be included in the results section instead of Materials and Methods even though I understand that certain decisions in made in the Mat&Met would then be more difficult to write in that case since they were made based on these results. But still, all the results should be presented in the results section.

-Did the authors test systematically all possible variation of their four indices and then decide only to report the important ones? Was for example an index of B6 – (B11+B12) tested?

-Figure 4 could include also B12 and PI₂, to show the seasonal variation for all the important ones.

-Table 3 is presented misleadingly since “Training set CV score” is named as “Validation score” and presented after the test score. The training set results should be presented first and Test set results later.

-The discussion section is the weakest one in the manuscript. The value of the results should be considered in the context of existing literature. Seasonal trends, cultivar differences, performance of indices such as SIWSI for example could be discussed in context of what was known before and what do these results add to that knowledge. 

-The approach of not having full field data with each tree measured by RTK but reference data that has been collected in an inventory database should be critically discussed. The benefits, the concerns, the disadvantages, the approaches to check by additional field visits by forest partners instead of researchers.

Minor comments:

-Add latin names for all the tree species, also the ones that currently only appear in Table 1.

-Report the bandwidths for the Sentinel-2 wavelengths in addition to the central wavelength (check the central wavelengths, I am not certain if all of them are correct,) lines 75-79

-Line 89 – Do you mean that this was done for the entire tiles of 31UEQ, 30TYT and 31TCJ?

-Line 92 – do you mean complemented instead of completed?

-Line 98-99 “all poplar polygons for both 2017 and 2018 were checked through photo-interpretation by three experts” – Do you mean that all the polygons labelled as poplar plantations were checked against aerial photos taken in 2017 and/or 2018? Each of them by three experts or the work divided between three experts?

-Line 99 ensures -> ensured

-We selected references of various stand ages and cultivars – be more specific -which age groups, which cultivars? Was this done for all the three tiles as well?

-Line 103, do you mean that “A set of 230 photo-interpreted poplar polygons from the southwestern tile (31TCJ) was validated in the field by forest partners. They found two polygons of eucalyptus plantations mislabeled as poplar plantations, representing less than 1% misidentification.” -Is this metholodology or does this contain a result that should be presented in the results section?

-Line 105 “Another set of 85 polygons from a field campaign was also used to identify the limit of detection of poplar stands according to their stage of development.” -Do you mean that As a separate test, a set of 85 field-checked polygons was also utilized to study the effect of the stage of development of the poplar stand to the detection by Sentinel-2 data?

-Lines 120-123. What is meant by the sentences “The locust tree class was removed from tile 31UEQ in 2018 because the majority of its samples were identified as outliers and the number of remaining pixels was too small to be considered representative of the whole tile. This was also the case for other species.” -Which other species? All other species? Which year?

 -Line 127 ‘the most relevant spectral regions’ – Do you mean the most important spectral features for poplar recognition? – the context in which the spectral features are “relevant” needs to be stated -check all places where this occurs either in respect of spectral features or indices

-Line 140 “the effect of the curse of dimensionality” – what does this mean? Do you mean the need to increase the numbers of reference data when the dimensionality of the analysis increases?

-Line 149 “the objective function” -what is meant by this? Do you actually mean “for poplar recognition”?

-Line 158. This was carried out until ten features were included in the subset. -Surely this cannot be true for the single-feature SFFS configuration that is mentioned in the previous sentence. -Why 10 was chosen? Did you test other options to begin with?

 -Line 161 Reference polygons (forest stands) were randomly separated into 50% for training and 50% for testing. -How many reference polygons altogether in the training and the test set? Were they distributed equally among the three tiles?

-Line 167. Surely the single-feature SFFS was tested as well?

-Line 230 spring to late summer, line 251 from early spring to late summer, line 339 between May and October, line 369 from early April to mid September, line 383 between May and October -When the authors think that it is the best time to acquire satellite data for mapping poplar plantations? -The best way to write this would be to include both generic terms such as early spring and the specific months in the same sentence since the timing of the seasons greatly varies across the earth and the readers from another part of the world cannot be expected to grasp when spring occurs in France.

-Line 312 Over-detection of poplar plantations in agricultural fields was also observed (Figure 10a). -Do you mean false positives? Agricultural fields that were mislabeled as poplar plantations?

-Lines 354-355, 1450 nm and 1950 nm do not coincide within 1563-1657 nm and 2094-2278 nm, respectively, not even roughly. However, it is known that water absorption wavelengths saturate at high water concentrations and adjacent wavelengths reflect water content more accurately than the absorption bands. (See for example in Ullah et al. 2014, Retrieval of leaf water content spanning the visible to thermal infrared spectra ISPRS Journal of Photogrammetry and Remote Sensing 93 (2014) 56–64)

-Could the authors present a comparison of water contents of poplar leaves in relation to water contents of other deciduous species during the summer season? Poplar leaves should have higher water content during the summer in their leaves than the other deciduous trees.

-Alternatively, could it be that phenolic compounds that are abundant in poplar leaves could affect the B11 band, since they have absorption at 1660 nm?

Couture, J.J.; Singh, A.; Rubert-Nason, K.F.; Serbin, S.P.; Lindroth, R.L.; Townsend, P.A. 2016. Spectroscopic determination of ecologically relevant plant secondary metabolites. Methods Ecol. Evol. 7, 1402–1412.

Kokaly, F.R.; Skidmore, A.K. 2015. Plant phenolics and absorption features in vegetation reflectance spectra near 1.66 _m. Int. J. Appl. Earth Obs. Geoinf. 43, 55–83.

-Line 363 When is dry season in France?

-Line 395 and Figure 13 All the results need to be transferred to the results section. Discussion cannot include results not presented in the results section.

-Lines 402 onwards include some results that were not present in the results section, where they should be

-Lines 440-442 This is discussion, should be in discussion section

Author Response

See the reply in attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The study deals with the large-scale classification of poplar stands in France. The work is very well written and conscientiously elaborated. Individual paragraphs are somewhat detailed and lengthy but overall it is a successful work. I doubt whether the introduction of a new poplar index is really useful, but I find the large-scale coverage of stands with S2 very exciting.

It is the first time that I could find almost no room for improvement. I congratulate you on the work.

From my point of view, the following points should be considered for the final version:

1) Data from two years is used throughout the paper and the results are compared. However, the final map was only made within one year but especially here a comparison of the products from the two years would mean an extreme added value. The authors address this point in the outlook, I would expect to implement it already in this paper.

2) A comparison with the poplar areas of the national statistics and the areas obtained in the work would be nice. 

3) Why did you not consider the other broadleaved trees in the final map and why did you not use the HRL dominant leaf type product for the final map? 

4) The importance of S2- SWIR channels has been pointed out in several papers, a little more discussion here would be nice. 

5) I would suggest in Add the Authors name whenever the citation is part of the sentence. Sentences like “[25] tested the…” looks strange to me.

6) Comparisons of %-values result in percentage points: A percentage point or percent point is the unit for the arithmetic difference between two percentages. For example, moving up from 40 percent to 44 percent is an increase of 4 percentage points, but a 10-percent increase in the quantity being measured (https://en.wikipedia.org/wiki/Percentage_point).

 

L64: S2 was already introduced

L96: and what about less than 50%? E.g. mixture of 3 species with 33% each?

L103: So you used several pixels from one polygon? I would avoid the use of neighboring pixels for the model training. 

L133: Why did you not use the bands and VIs together?

L172: 10 to 150 trees are unusually few trees for a random forest model!

Figure 3: I would prefer the use of the wavelength instead of the band number. Mainly because of the shape in RE and NIR bands…

L204: You introduce the index and some lines later you exclude it already. Not clear to me. A high correlation was expected. 

Figure 4: Please show it also for 2018

Table 2: should be placed in the methods and not in the result chapter.

L226: How was the confidence value calculated?

L243: 3 percentage points instead of %

L258: 2 percentage points

L288-289: percentage points

 

 

Author Response

See the reply in attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and suggestions：

The authors proposed a new poplar detection index using the best discriminative spectral band (SWIR bands, and Red-edge bands), and its performance was investigated and compared with other spectral indices. Abundant work has been involved in the manuscript, and extensive detailed information on data analysis have been presented. Nevertheless, the novelty of the research seemed to be not enough, since the manuscript mainly focus on the detection accuracy of poplar area. Moreover, the proposal of new poplar index SHOULE be mentioned with detailed mechanism. For example, the manuscript put forward PI1, PI2, PI3 and PI4. The difference between PI2 and PI3 was the calculation equation, however, their performance varied markedly. There should be enough explanations. Moreover, I believe the manuscript should clarify several crucial issues which are presented in the general comments below, and some minor specific comments also should be addressed for the improvement of the manuscript.

 

General comments

1. Here is a very simple question. The manuscript mentioned in the abstract that ‘PI2 was used to produce the national map of poplar plantations in 2018 with an average producer’s accuracy of 95%’. How did the PA results come from? Were there any validation results at the national scale?

 

2. from the table 3. Here is a strange phenomenon. All validation results (PA) seemed to be better than that of test results. Generally, validation results were inferior than that of training results. Here should make some reasonable explanations.

 

3. in section 2.2, the manuscript mentioned that ‘all poplar polygons for both 2017 and 2018 were checked through photo-interpretation by three experts’, detailed photo-interpretation techniques or process should be mentioned here to support that these visualized results were scientific and reliable.

 

Specific comments

1. Figure 7 and Figure 8 are not cited in the whole manuscript.

 

2. For Figure 9-11, these figures are not standard. For example, these figures are all without color ramp and plotting scale.

Comments for author File: Comments.pdf

Author Response

See the reply in attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Thank you for giving me this opportunity to read the manuscript entitled "Sentinel-2 Poplar Index for Operational Mapping of Poplar Plantations over Large Areas". The topic of this manuscript is interesting and would be a good contribution to this field. I think it could be considered for publication in Remote Sensing once the following issues are addressed.

 

1. Please replace the keywords that already appear in the manuscript's title with close synonyms or other keywords, which will also facilitate your paper to be searched by potential readers.

 

2. Scales and compass are also needed in Figure 9 and Figure 12.

 

3. “Limitation” should be added as a sub-section of “Discussion” 

 

4. Some grammatical errors exist in the manuscript. Therefore, a critical review of the manuscript language will improve readability.

 

Author Response

See the reply in attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thanks to the authors for considering the comments and suggestions. In the present version of the manuscript, the authors have answered the general comments accordingly, detailed information have been added to the new version as well. Thus the manuscript could be considered for publication in journal of remote sensing.