Assessing Leaf Biomass of Agave sisalana Using Sentinel-2 Vegetation Indices

Round 1

Reviewer 1 Report

The leaf biomass is a principal parameter in crop monitoring and assessing carbon cycle. leaf biomass of Agave sisalana was estimated using the Sentinel-2 vegetation indices and generalized additive model. The highest performance of the model can predict AGB using ratio and normalized difference VIs. In addition, the method of the ratio-based spectral index in the manuscript was not well presented.

Main comments:

1. Line 101, “Figure 2A-B” should be deleted “A-B”.
2. I would suggest that Figure 3 showed the spacing between double rows and the spacing between single rows.
3. Line 142, the literature showed not correct.
4. Lines 183, “Table 2” should be “Table 3”, and IRECI is not included in Table 3.
5. Lines 240-241, I would suggest making this sentence into the two sentences.
6. I suggest to revise the Figure 5, 7 and 8 to make it more artistic.
7. Lines 245-246, “surface reflectance, %” indicates that the Sentinel-2 spectral reflectances were increased by 100 times. Please check carefully.
8. Lines 247-248, How to combinate two band for three VI forms (RSI, NDSI, RDSI), and Figure 6 showed the explained deviances (D²) between Sentinel-2 spectral bands. Please make it clear.
9. Lines 275, “Table 3” should be “Table 4”. Please authors check the Figure and Table carefully.
10. Lines 266-267, “Figure 7 shows best two VIs … less than -0.15Mg ha^-1.” It is impossible to draw a conclusion less than -0.15Mg ha^-1 from the Figure 7. “the best RSI, NDSI, and RDSI outperformed all reference VIs” is contradictory with “Figure 7 shows best two Vis…”.
11. I would suggestion that Figure 8 should be added the prediction error.

Author Response

Response to Reviewer 1 Comments

Comments and Suggestions for Authors

The leaf biomass is a principal parameter in crop monitoring and assessing carbon cycle. leaf biomass of Agave sisalana was estimated using the Sentinel-2 vegetation indices and generalized additive model. The highest performance of the model can predict AGB using ratio and normalized difference VIs. In addition, the method of the ratio-based spectral index in the manuscript was not well presented.

- We sincerely thank you for the comments. Please find our answers below.

- We have revised the methods section and we explain the calculation of spectral indices clearer.

Main comments:

Point 1: Line 101, “Figure 2A-B” should be deleted “A-B”.

Response 1:

- Line 318: “A-B” was deleted and general description of the figure was added.

Point 2: I would suggest that Figure 3 showed the spacing between double Lines and the spacing between single Lines.

Response 2:

- Line 395: Information about spacing was added to Figure 3.

Point 3: Line 142, the literature showed not correct.

Response 3:

- Line 417: Citation was revised to journal format.

Point 4: Lines 183, “Table 2” should be “Table 3”, and IRECI is not included in Table 3.

Response 4:

- Line 517: the number of the table was corrected and IRECI was added to the table.

Point 5: Lines 240-241, I would suggest making this sentence into the two sentences.

Response 5:

- Line 628-629: Sentence was divided into two.

Point 6: I suggest to revise the Figure 5, 7 and 8 to make it more artistic.

Response 6: We revised all the figures and made small adjustments.

Point 7: Lines 245-246, “surface reflectance, %” indicates that the Sentinel-2 spectral reflectances were increased by 100 times. Please check carefully.

Response 7:

Line 653: % was corrected to 0-1.

Point 8: Lines 247-248, How to combinate two band for three VI forms (RSI, NDSI, RDSI), and Figure 6 showed the explained deviances (D2) between Sentinel-2 spectral bands. Please make it clear.

Response 8:

- Lines 475-488: We expanded the description of fitting all the bands into the formulas.

- Line 655: “VI-biomass” was added for clarification.

- Line 672-673: RDSI formula was missing from the caption but is now included.

Point 9: Lines 275, “Table 3” should be “Table 4”. Please authors check the Figure and Table carefully.

Response 9: The number of the table was corrected.

Point 10: Lines 266-267, “Figure 7 shows best two VIs … less than -0.15Mg ha-1.” It is impossible to draw a conclusion less than -0.15Mg ha-1 from the Figure 7. “the best RSI, NDSI, and RDSI outperformed all reference VIs” is contradictory with “Figure 7 shows best two Vis…”.

Response 10:

- Line 686: We had calculated mean of the prediction errors, for which the correct term would have been “mean bias error”, not “bias”. Basically, this is MAE but without absolute values. However, considering that this is not a standard error metric and the error is better conveyed with MAE and RMSE, we decided to remove this sentence, rather than adding the correct term.

Point 11: I would suggestion that Figure 8 should be added the prediction error.

Response 11:

-Line 720: model fit and prediction error were added to Figure 8.

Reviewer 2 Report

The manuscript entitled ‘Assessing leaf biomass of Agave sisalana using Sentinel-2 vegetation indices’ by Vuorinne et al., submitted to Remote Sensing, aims to assess the ability to estimate leaf biomass of Agave sisalana (sisal) via Sentinel-2 multispectral imagery. First, the authors estimate leaf biomass, based on an allometric approach, in 58 field plots on a sisal plantation in south-eastern Kenya. Secondly, different Sentinel-2 derived band combinations in different formulations of vegetation indices were input into generalised additive models to explore how well leaf biomass can be estimated. The authors found that the best-performing VI was based on green, red-edge, and near-infrared spectral bands, and conclude that the heterogeneity of ground vegetation and resulting background effects limit overall model performance.

The study is outstanding in that it focusses on the derivation of sisal succulent biomass based on remote sensing, which has not really been in the focus of the remote sensing community yet. The manuscript is well-structured, the figures are demonstrative and understandable and the text is well-written. However, I am not fully convinced by the methodology that is employed in this study, and I thus ask the authors to address the following weaknesses that I identified:

• A number of studies focussing on the estimation of aboveground biomass of various crops via means of remote sensing have been published in the past years. What those studies have in common is that they relied on ground-truth data that was collected over the course of different growing stages, locations and years, as to cover all changes in crop development and possible biotic and abiotic influences that might have an impact on the relationship of biomass and spectral signal. In this study, the authors solely relied one the estimation of sisal leaf biomass at one specific point in time (based on an allometric model, i.e. no destructive measurements), plus one (!) Sentinel-2 scene that captured spectral reflectance one month later. Despite different plant ages and thus growing conditions, the overall informational value of the model is limited in both space and time, and thus of limited value to the scientific community. Contrary to destructive measurements, the allometric model approach seems to be less time-consuming. So why did the authors include so little information only in their study?
• While the remote sensing community seems to move on to machine learning approaches, often without proper comprehension of what is actually happening in the models (just my impression), I appreciate the authors’ decision to rely on well-established, straightforward and easy-to-understand vegetation indices. I do frankly disagree, however, with the authors’ concept to establish the relationships between the biomass and vegetation indices. In line 195, the authors state ‘without prior knowledge of the VI biomass relations, this modelling approach seemed plausible’. I would like to argue that biomass (and other plant variables) – spectral signal relationships are well-documented and, with a bit of knowledge in remote sensing of vegetation, even somewhat predictable for plants that have not been in the focus of research yet. My opinion is that the authors should have looked at relationships between spectral bands and biomass first (as done in figure 5), and then ultimately start with testing different approaches. What I grasp from figure 5 that a simple regression of some form might also do the trick. Eventually, the authors could have moved on to more sophisticated models, just as GAMs. If ‘keep it simple’ is one the principles in science and statistics, why take a sledgehammer to crack a nut?

 

Some more comments:

Title: All words should be capitalized, except prepositions, articles and conjunctions.

L32: Please add a brief explanation of what a ‘succulent’ plant is. Keep in mind that not every reader is a botanist.

L42: So where on earth is sisal cultivated?

L54: I’m sure that there are more than two reason to estimate crop biomass via means of remote sensing. Please drop the numbering.

L73ff.: I suggest you re-formulate this section such that your study questions become clearer. I suggest you work with study questions and bullet points.

L81ff.: Please provide a description about the cultivation practices (are leaves cut all year around? What about fertilisation? Pesticide application?)

L113: Is ‘established’ the correct term? Rather use ‘selected’ or ‘chosen’.

L118: The description about cultivation practices (row planting etc.) should be shifted to the study area description. What is a double row, what is a single row?

L.183f: NDVI and SR can be excluded from the list of reference indices, because those combinations will be included anyway if all possible band combinations were tested

L.243: Figure 5 – is the surface reflectance given in %? Decimal-%?

Author Response

Response to Reviewer 2 Comments 

Comments and Suggestions for Authors

The manuscript entitled ‘Assessing leaf biomass of Agave sisalana using Sentinel-2 vegetation indices’ by Vuorinne et al., submitted to Remote Sensing, aims to assess the ability to estimate leaf biomass of Agave sisalana (sisal) via Sentinel-2 multispectral imagery. First, the authors estimate leaf biomass, based on an allometric approach, in 58 field plots on a sisal plantation in south-eastern Kenya. Secondly, different Sentinel-2 derived band combinations in different formulations of vegetation indices were input into generalised additive models to explore how well leaf biomass can be estimated. The authors found that the best-performing VI was based on green, red-edge, and near-infrared spectral bands, and conclude that the heterogeneity of ground vegetation and resulting background effects limit overall model performance.

The study is outstanding in that it focusses on the derivation of sisal succulent biomass based on remote sensing, which has not really been in the focus of the remote sensing community yet. The manuscript is well-structured, the figures are demonstrative and understandable and the text is well-written. However, I am not fully convinced by the methodology that is employed in this study, and I thus ask the authors to address the following weaknesses that I identified:

- We sincerely thank you for these comments. Please find our answers below.

A number of studies focussing on the estimation of aboveground biomass of various crops via means of remote sensing have been published in the past years. What those studies have in common is that they relied on ground-truth data that was collected over the course of different growing stages, locations and years, as to cover all changes in crop development and possible biotic and abiotic influences that might have an impact on the relationship of biomass and spectral signal. In this study, the authors solely relied one the estimation of sisal leaf biomass at one specific point in time (based on an allometric model, i.e. no destructive measurements), plus one (!) Sentinel-2 scene that captured spectral reflectance one month later. Despite different plant ages and thus growing conditions, the overall informational value of the model is limited in both space and time, and thus of limited value to the scientific community. Contrary to destructive measurements, the allometric model approach seems to be less time-consuming. So why did the authors include so little information only in their study?

- We agree that the generalisation of these results is limited in both space and time although we covered blocks of varying plant age. The field sampling was done during the dry season and satellite image was selected from the same period when background effects (variation in ground vegetation greenness) is at its minimum. The reason for this was to minimize seasonal effects and improve generalisation of the results (as far as possible with single image).

- We acknowledge this constrain of the study and it is now more extensively discussed in Lines 1111-1116. Since this was the first study on remote sensing of sisal and rather extensive work as such, these aspects will remain to be examined in the future studies.

While the remote sensing community seems to move on to machine learning approaches, often without proper comprehension of what is actually happening in the models (just my impression), I appreciate the authors’ decision to rely on well-established, straightforward and easy-to-understand vegetation indices. I do frankly disagree, however, with the authors’ concept to establish the relationships between the biomass and vegetation indices. In line 195, the authors state ‘without prior knowledge of the VI biomass relations, this modelling approach seemed plausible’. I would like to argue that biomass (and other plant variables) – spectral signal relationships are well-documented and, with a bit of knowledge in remote sensing of vegetation, even somewhat predictable for plants that have not been in the focus of research yet. My opinion is that the authors should have looked at relationships between spectral bands and biomass first (as done in figure 5), and then ultimately start with testing different approaches. What I grasp from figure 5 that a simple regression of some form might also do the trick. Eventually, the authors could have moved on to more sophisticated models, just as GAMs. If ‘keep it simple’ is one the principles in science and statistics, why take a sledgehammer to crack a nut?

- Line 523: We agree that the sentence about VI-biomass relationships was not well argued and we have removed it.

- We agree also that some simpler form of regression(s), such as linear regression with suitable variable transformations, could have been used for modelling the relationships. However, since we wanted to compare over 200 VIs, which, as can be seen from figures 5 and 7 had varying forms of responses, we thought GAM was advantageous due to its flexibility. Furthermore, GAMs do not require transformations that can produce bias in linear regression model predictions if not accounted for. For a large set of VIs, manual model selection would very laborious. Although GAMs have been used in some RS studies, we also think this study demonstrates their applicability for a new use-case. Now that the best VI candidates are known, suitable parametric model forms could be developed for a more limited set of VIs (if needed).

Some more comments:

Point 1: Title: All words should be capitalized, except prepositions, articles and conjunctions.

Response 1: Lines 2-3: Words in the title were capitalised.

Point 2: L32: Please add a brief explanation of what a ‘succulent’ plant is. Keep in mind that not every reader is a botanist.

Response 2: Lines 31-35: A short description of leaf structure and plant ecology was added.

Point 3: L42: So where on earth is sisal cultivated?

Response 3: Line 44: The largest producers of sisal were added.

Point 4: L54: I’m sure that there are more than two reason to estimate crop biomass via means of remote sensing. Please drop the numbering.

Response 4: Lines 76-78: Numbering was dropped.

Point 5: Point L73ff.: I suggest you re-formulate this section such that your study questions become clearer. I suggest you work with study questions and bullet points.

Response 5:

- Line 98-99: Study questions were revised and numbered to make them stand out.

Point 6: L81ff.: Please provide a description about the cultivation practices (are leaves cut all year around? What about fertilisation? Pesticide application?)

Response 6:

- Lines 311–315: Description of the cultivation practices was extended.

- Lines 326-327: Information on leaf cutting was specified and information on flower stalk length was added.

Point 7: L113: Is ‘established’ the correct term? Rather use ‘selected’ or ‘chosen’.

Response 7: Line 331: “established” was changed to “selected”.

Point 8: L118: The description about cultivation practices (row planting etc.) should be shifted to the study area description. What is a double row, what is a single row?

Response 8: Lines 311-312: Description about the rows was shifted to study area description. We also extended the description. Furthermore, we added information on row spacing to Figure 3 (Line 395).

Point 9: L.183f: NDVI and SR can be excluded from the list of reference indices, because those combinations will be included anyway if all possible band combinations were tested

Response 9:

- Line 517: SR was removed from the table.

- We excluded SR, but included NDVI. We wanted to compare best indices we found to NDVI due to its popularity and we thought it’s better to introduce it as a reference, rather than pick it up later from the results.

Point 10: L.243: Figure 5 – is the surface reflectance given in %? Decimal-%?

Response 10: Line 653: This was mistake. % was changed to 0-1.

Reviewer 3 Report

The article "Assessing leaf biomass of Agave sisalana using 2 Sentinel-2 vegetation indices" describes an application of remote sensing data to assess the biomass using some vegetation indices. 

In particular, the Authors compared the performance of all the possible combinations between the Sentinel-2 bands with the performances of some reference vegetation indices. To validate the retrieved data, the biomass was measured in 58 field plots distinguished by the year of cultivation of the plant considered. 

Overall, I found the paper interesting, because despite the methodology is known, provides some interesting information about the possibility to use remote sensing in the monitoring of the Agave and, gives some insight on what vegetation index performed better and on what are the limitations of the study. 

The Authors cite several times Vuorinne et al. (submitted) which is not published yet. At line 142, is this methodology original or it is derived from other papers? if yes I think this information should be added in the text. Also, in the discussions (line 360) this statement needs other references in addition to the submitted paper. 

In addition, I think that in the methodology the combination of all the Sentinel-2 bands should be better explained because the reader must arrive at figure 6 to understand what the band A and the band B indicated in the equations are.

Finally, the conclusion must be completely rewritten, because they are very similar to the abstract. Generally, the conclusions have to summarize the work, to remind the strength and the importance of the work, and to indicate opportunities for future research.

I have just some minor comment for the authors:

1 - at line 119, the unit measure must be uniformed with the other removing the dash.

2 - the RMSE extended name need to be indicated before line 215 because it is already mentioned in line 142.

3 - Line 186, why IRECI is not in table 3? 

4 - Add trend line in figure 5

5 - there are 2 table 3 in the text

6 - line 295 use the apex for the -1

7 - why the spectral indices indicated in lines 323-327 were not used by the Authors as a reference? 

 

Author Response

Response to Reviewer 3 Comments

Comments and Suggestions for Author

The article "Assessing leaf biomass of Agave sisalana using 2 Sentinel-2 vegetation indices" describes an application of remote sensing data to assess the biomass using some vegetation indices.

In particular, the Authors compared the performance of all the possible combinations between the Sentinel-2 bands with the performances of some reference vegetation indices. To validate the retrieved data, the biomass was measured in 58 field plots distinguished by the year of cultivation of the plant considered.

Overall, I found the paper interesting, because despite the methodology is known, provides some interesting information about the possibility to use remote sensing in the monitoring of the Agave and, gives some insight on what vegetation index performed better and on what are the limitations of the study.

The Authors cite several times Vuorinne et al. (submitted) which is not published yet. At line 142, is this methodology original or it is derived from other papers? if yes I think this information should be added in the text. Also, in the discussions (line 360) this statement needs other references in addition to the submitted paper.

In addition, I think that in the methodology the combination of all the Sentinel-2 bands should be better explained because the reader must arrive at figure 6 to understand what the band A and the band B indicated in the equations are.

Finally, the conclusion must be completely rewritten, because they are very similar to the abstract. Generally, the conclusions have to summarize the work, to remind the strength and the importance of the work, and to indicate opportunities for future research.

Response:

We sincerely thank you for the comments. Please find our responses below.

- Line 417: the cited methodology is original methodology described in detail in our other manuscript, which has been submitted but not yet published.

- Lines 406-409: We added a more detailed description of how the plants were measured.

- Lines 475-488: We revised the explanation of how we calculated the VIs.

- Line 886: Relevant reference was added.

- Lines 1167-1177: Conclusion were rewritten using the structure suggested by the reviewer.

I have just some minor comment for the authors:

Point 1: at line 119, the unit measure must be uniformed with the other removing the dash.

Response 1: Line 311: dash was removed.

Point 2: the RMSE extended name need to be indicated before line 215 because it is already mentioned in line 142.

Response 2:

- Line 417: Acronym RMSE is now written out.

- Line 604: Consequently, “root mean squared error” was removed.

Point 3: Line 186, why IRECI is not in table 3?

Response 3:

- Line 517: IRECI was added to table 3.

Point 4: Add trend line in figure 5

Response 4: Line 651: GAM smooth line and goodness of fit was added to figure 5 similar to figures 7 and 8.

Point 5: there are 2 table 3 in the text

Response 5:

-Line 517: Table number was corrected to 4.

Point 6: line 295 use the apex for the -1

Response 6:

Line 731: superscript was added to ha^-1.

Point 7: why the spectral indices indicated in lines 323-327 were not used by the Authors as a reference?

Response 7: These indices are two-band ratios and hence they were calculated when we calculated all the two band combinations with simple formulas (Figure 6). Although there are countless two band combinations in the literature, we didn’t include them as reference since they were automatically included in our modelling when we calculated all the two band combinations. The only two band ratio we added to our reference list was NDVI due to its popularity. So, other reference VIs were chosen because they are calculated with different formulas than the three formulas we used.

Author Response File: Author Response.docx

Reviewer 4 Report

This is a very well-written paper and the authors are to be commended for their work. I only have a few minor suggestions for them, otherwise I recommended accepting the paper in present form.

Line 114: spell out NDVI first time cited

Lines 180-181: RDSI acronym does not match text (would be RDVI or change vegetation to spectral)

Line 296: intra- and inter-block

Lines 303-304: second sentence of paragraph is a fragment; perhaps combine with first sentence

Line 373: Clevers et al. (period missing)

Author Response

Response to Reviewer 4 Comments

Comments and Suggestions for Authors

This is a very well-written paper and the authors are to be commended for their work. I only have a few minor suggestions for them, otherwise I recommended accepting the paper in present form.

Response from the Authors

We sincerely thank you for the comments. Answer to the suggestions can be found below.

Point 1: Line 114: spell out NDVI first time cited.

Response 1:

- Line 514: NDVI Spelling was added.

Point 2: Lines 180-181: RDSI acronym does not match text (would be RDVI or change vegetation to spectral)

Response 2:

- Line 485: Text was edited to match the acronym.

Point 3: Line 296: intra- and inter-block

Response 3:

- Line 732: “-“ was added as suggested.

Point 4: Lines 303-304: second sentence of paragraph is a fragment; perhaps combine with first sentence

Response 4:

Lines 796-798: Second and third sentences in the paragraph were combined.

Point 5: Line 373: Clevers et al. (period missing)

Response 5:

- Line 861: Period was added as suggested.

Author Response File: Author Response.docx

Reviewer 5 Report

This work on estimating leaf biomass of Agave Sisalana using Sentinel-2 satellite imagery is a well-conducted and written manuscript. I liked the concept of using 2 band indices that fits well with the biomass. That is a novelty in this paper, and other techniques are also good. I should really appreciate the authors conducting research on a large scale. Such large-scale studies are limited, and this kind of work should be supported. While reading through the manuscripts, I had a few concerns, which I have listed as comments below:

1. Line 101: Figure caption. What does Figure 2A-B indicate? I think it should be Figure 2A-F. Please check.
2. Line 106: I would prefer “contributed by” instead of “allocated to”.
3. Line 134: Is it median-sized or medium-sized? If it is median-sized, it is difficult to follow. Please rephrase the sentence.
4. Line 144: Please include units after defining each symbol from the equation. For example, where B is the dry mass of a leaf (kg, Mg, g)?
5. Line 144: If possible, please briefly state how the maximum diameter and height of the plant were measured. As far as I know, the maximum diameter can only be measured by image processing? Is that how it was done in this study? Similarly, height was measured from the ground to the topmost tip of the plant?
6. Line 150: Plot biomass of 0 Mg ha^-1? Does that mean no leaves at all? Or because of rounding off the digits, it shows 0 Mg/ha? In the case of rounding, you can indicate something like <0.01 Mg/ha.
7. Figure 4 x-axis: It will be better to show “dry leaf biomass” directly and remove the definition in the caption. If this image is shown on google search, then the figure will stand alone without a caption. Maybe include the bin width in the caption. I assume it should be 5 Mg/ha? It will be good to show the plot with a bin width of 1 Mg/ha.

8. Line 160: When was the date of field data collection? I checked the preceding paragraphs and couldn’t find the date. I am just curious to know if the Sentinel image used was ‘before’ or ‘after’ the field data collection.
9. Equation numbering mistake. Please check and revise
10. Line 173: Instead of starting a paragraph with an abbreviation, please use that to define VI.
11. What are those bands A and B in equations 4-6? Are these new indices tested specifically for this study? If yes, then emphasize that.
12. Please provide the abbreviations of the indices before the equation (line 174, 177, and 180)
13. Line 183 – This should be referred to in Table 3.
14. Expand all the indices abbreviation provided in Table 3. Maybe you can add a column to Table 3 and include the expanded name.
15. What is p in equation 6? This should be actually eq. 7.
16. Line 210: Instead of ‘inspection’, use ‘evaluation’. Inspection is to ‘see’ carefully.
17. Line 235: How is the pixel-scale uncertainty calculated? Line 236 is also confusing. Is there a better way to convey this information?
18. Sections 2.4 and 3.1 have the same headings. Please make both descriptive. Section 2.4 can be the Vegetation index calculation, and Section 3.1 can be “Relationship between biomass vs vegetation indices”.
19. Figure 5 y-axis: Please make it “leaf biomass”
20. Line 250: There is no RSI, it is only RS.
21. Line 275: This should be Table 4, not Table 3. Please check and correct the locations where they are referred to in the text.
22. Figure 8A-E: I don’t understand why the x-axis is 1-(RE2/RE3), while the best VI from table 4 is RE2/RE3.
23. Line 318: Even this study is species-specific, I think. This statement makes me think the authors have developed a universal model that works for any species. Maybe slightly rephrase this sentence so that it is not misleading.
24. Line 446: This question arises right from the methods section. Whether a high D² is acceptable or low D²? Because as the name suggests, it is a measure of deviance. So higher the number means higher the deviations? Please clarify this in the methods right after the equation.
25. D² in the conclusions can be spelled out. It will be helpful for readers.
26. Please cite the R core team and the mgcv package. These are open source packages, so the citation will help them.

Author Response

Response to Reviewer 5 Comments 

Comments and Suggestions for Authors 

This work on estimating leaf biomass of Agave Sisalana using Sentinel-2 satellite imagery is a well-conducted and written manuscript. I liked the concept of using 2 band indices that fits well with the biomass. That is a novelty in this paper, and other techniques are also good. I should really appreciate the authors conducting research on a large scale. Such large-scale studies are limited, and this kind of work should be supported. While reading through the manuscripts, I had a few concerns, which I have listed as comments below:

We sincerely thank you for these comments. Please find our answers below.

Point 1: Line 101: Figure caption. What does Figure 2A-B indicate? I think it should be Figure 2A-F. Please check.

Response 1:

-Line 318: The caption was revised to the format of the journal.

Point 2: Line 106: I would prefer “contributed by” instead of “allocated to”.

Response 2:

- Line 323: Revised as suggested.

Point 3: Line 134: Is it median-sized or medium-sized? If it is median-sized, it is difficult to follow. Please rephrase the sentence.

Response 3: The objective was to choose a representative plant to measure, with which the leaf biomass could be extrapolated for the plot. We called this plant as median-sized plant (median refers to the value separating the higher half from the lower half of the data). This is an analogy to forest inventory where median trees are commonly used in similar circumstances. However, as this can be confusing, and exact median-sized plant is impossible to determine in the field, we simply revised the sentence and call this plant “representative”. It must be also noted that as this is crop, most of the plants are very similar in size.

- Line 406: “median” was changed to “representative”.

Point 4: Line 144: Please include units after defining each symbol from the equation. For example, where B is the dry mass of a leaf (kg, Mg, g)?

Response 4:

- Line 434: All the units are now included after variables.

Point 5: Line 144: If possible, please briefly state how the maximum diameter and height of the plant were measured. As far as I know, the maximum diameter can only be measured by image processing? Is that how it was done in this study? Similarly, height was measured from the ground to the topmost tip of the plant?

Response 5: We are thankful for this comment because it made us realise a variable naming error. We had erroneously named leaf circumference as diameter. However, since there was no difference whether circumference or width was used (except in the parameters of the allometric model) we decided to use width for simplicity. Furthermore, we added a proper explanation of the measurements. A thorough explanation of this model is given in our other manuscript, which we have submitted.

- Line 434: We had erroneously named circumference as D (diameter). We changed the variable to W (width) for simplicity.

Point 6: Line 150: Plot biomass of 0 Mg ha-1? Does that mean no leaves at all? Or because of rounding off the digits, it shows 0 Mg/ha? In the case of rounding, you can indicate something like <0.01 Mg/ha.

Response 6: These fields were recently cleared from old plants to grow new plants (like in Figure 2A and cleared fields adjacent to it). Therefore, plot biomass is 0 Mg ha-1.

Point 6: Figure 4 x-axis: It will be better to show “dry leaf biomass” directly and remove the definition in the caption. If this image is shown on google search, then the figure will stand alone without a caption. Maybe include the bin width in the caption. I assume it should be 5 Mg/ha? It will be good to show the plot with a bin width of 1 Mg/ha.

Response 6:

- Line 445: Figure 4 x-axis title was changed from B to dry leaf biomass. The same was done for all the other figures as well for consistency.

- We tried bin-width of 1 Mg/ha, but with range of 0-45 and n=58 it is difficult to interpret. So, we thought 5 would be better for this sample size. Furthermore, literature seemed to support avoiding too small bin widths.

- Line 445: Information on bin width is now included in the caption.

Point 7: Line 160: When was the date of field data collection? I checked the preceding paragraphs and couldn’t find the date. I am just curious to know if the Sentinel image used was ‘before’ or ‘after’ the field data collection.

Response 7: Field data was collected one month before the image acquisition. So, there was one-month gap, due to cloud cover (obvious drawback with this approach). This information can be found from lines 331.

Point 8: Equation numbering mistake. Please check and revise

Response 8:

- Line 479: Equation numbering was corrected.

Point 9: Line 173: Instead of starting a paragraph with an abbreviation, please use that to define VI.

Response 9:

- The definition of VI is given in intro (Lines 83-84).

- Line 475: we added a sentence to describes VIs.

Point 10: What are those bands A and B in equations 4-6? Are these new indices tested specifically for this study? If yes, then emphasize that.

Response 10:

- Lines 669: We inserted all the bands with 20-m resolution into these equations. We try to explain this more precisely now in section 2.4.

Point 11: Please provide the abbreviations of the indices before the equation (line 174, 177, and 180)

Response 11:

- Lines 479-485: VI abbreviations were added as suggested.

Point 12: Line 183 – This should be referred to in Table 3.

Response 12:

-Line 488: Table number was corrected.

Point 13: Expand all the indices abbreviation provided in Table 3. Maybe you can add a column to Table 3 and include the expanded name.

Response 13: The names were too long to fit in one line and having them on two line would make this table difficult to read. Hence we expanded the abbreviations in the text instead.

- Lines 488-512: Full names of the vegetation indices were added.

Point 14: What is p in equation 6? This should be actually eq. 7.

Response 14: Degrees of freedom for the smoothing function.

- Line 526: Since we use k in the text, we use it now in the equation as well.

Point 15: Line 210: Instead of ‘inspection’, use ‘evaluation’. Inspection is to ‘see’ carefully.

Response 15: Line 537: Inspection was changed to evaluation.

Point 16: Line 235: How is the pixel-scale uncertainty calculated? Line 236 is also confusing. Is there a better way to convey this information?

Response 16:

- Lines 621-625: We tried to give better explanation how we calculated the pixel-scale uncertainty.

Point 17: Sections 2.4 and 3.1 have the same headings. Please make both descriptive. Section 2.4 can be the Vegetation index calculation, and Section 3.1 can be “Relationship between biomass vs vegetation indices”.

Response 17:

- Lines 474 and 627: Headings were revised as suggested

Point 18: Figure 5 y-axis: Please make it “leaf biomass”

Response 18: Line 652: Figure 5 y-axis was revised as suggested.

Point 19: Line 250: There is no RSI, it is only RS.

Response 19:

- Line 485: There was a typo in the equation in the methods section. It should have been RSI (ratio-based spectral index) and it is now corrected.

Point 20: Line 275: This should be Table 4, not Table 3. Please check and correct the locations where they are referred to in the text.

Response 20: Table numbering was corrected.

Point 21: Figure 8A-E: I don’t understand why the x-axis is 1-(RE2/RE3), while the best VI from table 4 is RE2/RE3.

Response 21: RE2/RE3 had an inverse relation to biomass, so it is shown as 1-(index) to make the relation positive. This makes it easier to compare with other models were the relation was positive. This information can be found from the figure caption.

Lines 686-687: The above explanation was added to the text as well.

Point 22: Line 318: Even this study is species-specific, I think. This statement makes me think the authors have developed a universal model that works for any species. Maybe slightly rephrase this sentence so that it is not misleading.

Response 22:

- Line 831: Indeed, it is species specific study and we slightly edited the sentence.

- However, considering that it’s an introductory sentence to the following sentences, which clearly state that one species was studied, we don’t consider it misleading if taken as a whole.

Point 23: Line 446: This question arises right from the methods section. Whether a high D2 is acceptable or low D2? Because as the name suggests, it is a measure of deviance. So higher the number means higher the deviations? Please clarify this in the methods right after the equation.

Response 23: D² is a measure of explained deviance (Line 533). So higher the number, the more deviance the model explains.

Point 24: D2 in the conclusions can be spelled out. It will be helpful for readers.

Response 24:

- Line 1169: Explained deviance was spelled out in conclusions as suggested.

Point 25: Please cite the R core team and the mgcv package. These are open source packages, so the citation will help them.

Response 25:

- Line 520: citation to R core team was added.

- The other citation is the one recommended in the mgcv documentation.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I think the revised version meet the requirements of your journal 

Author Response

We sincerely thank you for the review.

Reviewer 2 Report

The authors incorporated most of my suggestions in the revised manuscript. I am, however, not convinced by the answer that was provided with regard to the method employed to establish the VI-biomass relationship:

'We agree also that some simpler form of regression(s), such as linear regression with suitable variable transformations, could have been used for modelling the relationships. However, since we wanted to compare over 200 VIs, which, as can be seen from figures 5 and 7 had varying forms of responses, we thought GAM was advantageous due to its flexibility. Furthermore, GAMs do not require transformations that can produce bias in linear regression model predictions if not accounted for. For a large set of VIs, manual model selection would very laborious. Although GAMs have been used in some RS studies, we also think this study demonstrates their applicability for a new use-case. Now that the best VI candidates are known, suitable parametric model forms could be developed for a more limited set of VIs (if needed).'

The ARTMO Toolbox (https://artmotoolbox.com/), especially the Spectral Indices Retrieval Toolbox, was designed to enable the analysis and assessment of the accuracy of an indefinite number of SI models. Simple approaches make research easier reproducible.

 

Other comments:

Lines 174-175: Should be ‘All bands with at least 20m spatial resolution were used in this study’. So 10m-bands were aggregated to 20m resolution? This should be included in the text.

Figures 5-7-8: Shouldn't the models have an intercept? If there is no leaf biomass, the satellite senses reflectance from the soil background (or ground vegetation), so the calculation of the VI will return a value in any case.

 

 

 

Author Response

Response to Reviewer 2 Comments (2^nd round)

The authors incorporated most of my suggestions in the revised manuscript. I am, however, not convinced by the answer that was provided with regard to the method employed to establish the VI-biomass relationship:

'We agree also that some simpler form of regression(s), such as linear regression with suitable variable transformations, could have been used for modelling the relationships. However, since we wanted to compare over 200 VIs, which, as can be seen from figures 5 and 7 had varying forms of responses, we thought GAM was advantageous due to its flexibility. Furthermore, GAMs do not require transformations that can produce bias in linear regression model predictions if not accounted for. For a large set of VIs, manual model selection would very laborious. Although GAMs have been used in some RS studies, we also think this study demonstrates their applicability for a new use-case. Now that the best VI candidates are known, suitable parametric model forms could be developed for a more limited set of VIs (if needed).'

The ARTMO Toolbox (https://artmotoolbox.com/), especially the Spectral Indices Retrieval Toolbox, was designed to enable the analysis and assessment of the accuracy of an indefinite number of SI models. Simple approaches make research easier reproducible.

Reply: We agree with the reviewer that the modeling method is important. However, several approaches can be equally good and suitable for predictive modeling. In this study, we needed a method that enable relatively simple model fitting to compare large number of vegetation indices, and finally, provide predictions to produce a biomass map. GAMs with a small smoothing factor provided such method with good model fit for various types of relationships without any signs of overfitting or other problems. As our objective was to make a model for this single point in time, we did not aim to make a model that could be generalized in space or time. Parametric approach would be more justified for such case as parametric models are easy to apply based on the published model parameters. Furthermore, we have used R software environment and well-known packages and functions for fitting GAMs, which enable reproducibility of our results using the information given in the article. In summary, we consider our approach justified and suitable for the research we have conducted.

We also thank the reviewer for informing us about ARTMOT Toolbox. It seems very useful and we will consider using it in our future studies.

 

Other comments:

Lines 174-175: Should be ‘All bands with at least 20m spatial resolution were used in this study’. So 10m-bands were aggregated to 20m resolution? This should be included in the text.

Reply: Yes, we used 10-m bands which have been downsampled to 20-m.

- Line 466: This information was added to the manuscript.

 

Figures 5-7-8: Shouldn't the models have an intercept? If there is no leaf biomass, the satellite senses reflectance from the soil background (or ground vegetation), so the calculation of the VI will return a value in any case.

Reply: Our models have intercept although it might look that intercept is zero in some cases. The reviewer is right that when leaf biomass is zero, vegetation index (VI) have typically non-zero value corresponding to soil or ground vegetation. Therefore, it is justified to include intercept in the model. However, intercept tells the VI value corresponding to zero biomass only if model is fit as VI as y-variable and biomass as x-variable. This makes sense if one wants to study relationship between biomass and VIs because biomass is driving factor for reflectance, not the opposite. However, if one wants to make also predictions, the model must be fit as biomass as y-variable and VI as x-variable. As we wanted to make predictions, the models were fit using the latter approach.

Author Response File: Author Response.docx