Forage Biomass Estimation Using Sentinel-2 Imagery at High Latitudes

Round 1

Reviewer 1 Report

line 19. The conclusion of the abstract is not consistent with the conclusion of the article. The article aims to evaluate various statistical methods (coupled with orbital images) to model forage production in areas of Sweden. the conclusion is that the methods are valid, however, the sampling effort was insufficient. therefore, I do not think it is valid to value the results of the regression, after all, they do not represent reality.

line 39 - spectrometers

line 87 - "obvious" is a subjective term. I suggest carrying out statistical tests of means to support comparisons between temperatures and precipitation in the study sites and point out differences or similarities.

line 99 to 105 and table 2 - I suggest that you make a better report on the sampling field. I can´t understand if the "n" of 30 from As is made of sub-samples. May to September are 5 months, in two years, 10 months, multiplied by 3, 30 samples. It doesn´t correspond with the phrase: n, 3 sub-samples were averaged as 1 observation (line 106). Please clarify this point.

line 253 and line 314 - table s1 and figure s2 - supplementary material - I´m not aware of this supplementary material. It was not in the download manuscript.

Author Response

Comment 1: line 19. The conclusion of the abstract is not consistent with the conclusion of the article. The article aims to evaluate various statistical methods (coupled with orbital images) to model forage production in areas of Sweden. the conclusion is that the methods are valid, however, the sampling effort was insufficient. therefore, I do not think it is valid to value the results of the regression, after all, they do not represent reality.

Reply: We understand the reviewer’s concern and we emphasized the uncertainty of the modelling performances derived from the relatively small observation datasets both in the abstract and the main body of the paper. The final model we selected (shown in the Figure 4, with NSE values 0.92, 0.62, and 0.84 for calibration, validation and evaluation) is promising and with a more comprehensive dataset from future observations a more robust model could be built. We modified the abstract a bit to respond to this comment. See lines 20-22.

 

Comment 2: line 39 – spectrometers

Reply: Good point, it is corrected now. See line 41.

 

Comment 3: line 87 - "obvious" is a subjective term. I suggest carrying out statistical tests of means to support comparisons between temperatures and precipitation in the study sites and point out differences or similarities.

Reply: We did the statistical tests as suggested, and modified the texts accordingly. See line 93-112.

 

Comment 4: line 99 to 105 and table 2 - I suggest that you make a better report on the sampling field. I can´t understand if the "n" of 30 from As is made of sub-samples. May to September are 5 months, in two years, 10 months, multiplied by 3, 30 samples. It doesn´t correspond with the phrase: n, 3 sub-samples were averaged as 1 observation (line 106). Please clarify this point.

Reply: According to the original protocol, each time we went to the field to take samples, we took samples from three different places inside each field, and for each place, three sub-samples with 1-2 meters distance to each other were taken. We treat the 3 nearby subsamples as 1 observation so that the observation would be representative of the place and be aligned to the Sentinel-2 image pixels. In Ås, in total, 90 sub-samples were taken in 2 years, and the total observation for Ås is 30.

Comment 5: line 253 and line 314 - table s1 and figure s2 - supplementary material - I´m not aware of this supplementary material. It was not in the download manuscript.

Reply:  We did upload the supplementary file to the system but it was not transferred to the review panel. We decided to move all of the contents (one table and two figures) to the main body of the paper.

Reviewer 2 Report

The submitted manuscript titled "Forage biomass estimation using Sentinel-2 imagery at high latitudes" is good. Authors done a good job but need to work on some issues before get it published.

1. Calculate the NDVI value and make a regressing with the ground collected data.

2. If possible calculate the dry mass value as per the market rate

3. The collected sample data is near by and make an culuster, check the auto-correlation if any

4. Refine the Dry MY calculation  methods with suitable references.

 

 

Author Response

Comment 1:  The submitted manuscript titled "Forage biomass estimation using Sentinel-2 imagery at high latitudes" is good. Authors done a good job but need to work on some issues before get it published.

Reply:  Thanks for the positive comments of this paper. We read and implement the reviewer’s comments carefully.

 

Comment 2:  Calculate the NDVI value and make a regressing with the ground collected data.

Reply:  We calculated the NDVI and made the regression analyses with the ground-collected data in Section 2.4.1. In total, five different regression analyses (linear, exponential, power, polynomial and logarithmic) were tried and we found for NDVI, exponential regression performed best with highest model accuracy among different regression analyses. We put the results as table S1 in the supplementary file, but now we moved it to the main body of the paper.

 

Comment 3:  If possible calculate the dry mass value as per the market rate

Reply:  The main purpose of the paper was to build a reliable regression model to predict forage biomass, with units of t ha^-1, because this is the measure of biomass that farmers use to make decisions, e.g. fertilization and harvest timing. Farmers do not make these decisions based on the market value of the forage. In addition, the forage is commonly ensiled and stored in bunkers, and so there is essentially no forage market – it is used almost entirely by the farmer that produces it.

 

Comment 4:   The collected sample data is nearby and make an culuster, check the auto-correlation if any

Reply:  We understand the reviewer’s concern. We averaged each three sub-samples to increase the representativeness, which should remove auto-correlation. Furthermore, we trained models based on datasets from four different field stations and different sampling days, and there was no auto-correlation for the training datasets, which was determined by the auto-correlation test using “acf” function in R. Moreover, the machine-learning algorithms, e.g., Random forest, used in the study have been proved to have strong ability to overcome the auto-correlations issues. We clarified in the paper that there was no auto-correlation issue. See line 153-155.

 

Comment 5:  Refine the Dry MY calculation methods with suitable references.

Reply: The description of dry matter yield calculation was modified and the reference was inserted. See line 124.

Reviewer 3 Report

This study uses Sentinel-2 data to estimate forage biomass with empirical algorithms. It is generally a popular and simple method that remote sensing data are employed to linked with ground measured biomass. Overall, the structure of this manuscript is reasonable, and the conclusion can be supported by Figures and/or Tables. However, there are some major concerns:

(1)  It is stated that each sample consisted of three sub-samples with a spacing of 1-2 m between quadrats. It is unclear that how to match ground measurements and Sentinel-2 reflectance data in this study. What is the spatial resolution of the image pixels?

(2)  Sentinel-2 has a spatial resolution of 10-60 m. Which spatial scale is used? Please clarify this issue.

(3)  Supplementary materials cannot be seen when reviewing this manuscript. VI selection is an important topic to derive accurate biomass estimates from remote sensing data. Also, Table S1 has been mentioned many times in the main text. Thus, it is better to put S1 in main text.

(4)  Landsat-8 is an useful satellite, and it is also free-to-use. Improved satellite observation frequency will be obtained if combining Landsat-8 and Sentinel-2 together. It is suggested that Landsat-8 data should be also used to estimate forage biomass in this study.

Author Response

Comment 1:  This study uses Sentinel-2 data to estimate forage biomass with empirical algorithms. It is generally a popular and simple method that remote sensing data are employed to linked with ground measured biomass. Overall, the structure of this manuscript is reasonable, and the conclusion can be supported by Figures and/or Tables. However, there are some major concerns:

Reply:  Thanks for the positive comments of this paper. We read and implement the reviewer’s comments carefully.

 

Comment 2:  It is stated that each sample consisted of three sub-samples with a spacing of 1-2 m between quadrats. It is unclear that how to match ground measurements and Sentinel-2 reflectance data in this study. What is the spatial resolution of the image pixels?

Reply:  we mentioned this in section 2.3 that “Reflectance information for each sub-samples from each band was extracted using the “extract” function from package “raster” in R environment”, which means we extracted the reflectance of each band for each sub-sample point (with coordinate). Then each set of three sub-samples was averaged as one observation. The image pixel size was not mentioned in the paper, which was our mistake and we apologize for it. We used the 20-m spatial resolution images from Sentinel-2 and now it is mentioned in section 2.2 the paper. See line 133.

 

Comment 3:  Sentinel-2 has a spatial resolution of 10-60 m. Which spatial scale is used? Please clarify this issue.

Reply:  We used the 20-m spatial resolution images from Sentinel-2 and now it is mentioned in section 2.2 the paper. See line 133.

 

Comment 4:   Supplementary materials cannot be seen when reviewing this manuscript. VI selection is an important topic to derive accurate biomass estimates from remote sensing data. Also, Table S1 has been mentioned many times in the main text. Thus, it is better to put S1 in main text.

Reply:  We did upload the supplementary file to the system but it was not transferred to the review panel. We decided to move all of the contents (one table and two figures) to the main body of the paper.

 

Comment 5:  Landsat-8 is an useful satellite, and it is also free-to-use. Improved satellite observation frequency will be obtained if combining Landsat-8 and Sentinel-2 together. It is suggested that Landsat-8 data should be also used to estimate forage biomass in this study.

Reply:  Thanks for the suggestion, which is very good. However, after a deep thinking, we think it would be better to not include Landsat-8 data into our study at the moment. The reasons are: 1) the spatial resolution of landsat-8 images is 30 meters and involving them will reduce the capacity to relate ground observations to satellite data to a certain extend; 2) the red-edge bands are lacking in Landsat-8 data, but according to the ranking of the importance of predictor variables in the random forest modelling (figure 3 in the revised version of the paper), the red-edge bands were extremely important; 3) considering the spatial resolution and the band characteristics of Landsat-8 datasets, involving them into the study would change the overall structure of this paper to a very large extent since several issues (the effects from the different spatial resolutions and lacking of red-edge bands on the modelling) need to be addressed. Even though we decided to not include Landsat-8 data into this study, we appreciate this idea and mention the possibility in the discussion as a potential improvement. See line 378-382 in Discussion part of the paper.

Reviewer 4 Report

The paper concerns an interesting topic for the journal as it reports the estimation of aboveground biomass in forage resources from satellite imagery. Even if the research seems well conducted I found some issues that make it not ready for publication in the present form. It should be adequately revised to improve it before publication.

Some issues are listed hereafter:

- in the introduction (lines 29 and following) authors report the importance of the analyzed forage resources for Sweden but without any detail. Please add some data (land use dedicated to this kind of resources, percentage of surface covered, etc.) to better explain the importance of the study

- in the introduction (lines 37 and following) some instruments to estimate forage availability in situ such as spectrometers are cited to reduce time of field samplings of forage. Some other instruments exist to find relationships between biomass and filed parameters such as rising plate meter, which are easily to use and cheaper. Please add some information and references also on this approach

- line 39: the word “spectrometers” is badly written (there is a comma in between the letters, please remove)

- a great issue is reported about cloud presence in satellite imagery (lines 74-75). This is a pivotal point but the authors almost neglected it in the previous part of the introduction. I think authors should enhance a little bit this part, adding some references (now they are missing)

- line 93: authors report the importance of winter kill and its occurrence, bat in the results section no data are available about it. Did this condition affect the experimental fields? To which extent? Please provide some info about this topic

- in the materials and method sections a very big issue in my opinion is the complete lack of information about the vegetation under investigation. Which kind of species were present? Grasses or legumes? Both? In pure stand or in mixture? How differ in botanical composition the fields investigated (that where located in four experimental stations)? Or was the composition of the fields the same? In one location there were 5 fields: were they seeded with the same species? I think this is a very important point to be defined in the article, as different species can affect relationship between biomass and vegetation indices that are a “response” of the occurring vegetation. If the species or the proportion among them were very different, it is probable that this affected the variability of the regressions. A think authors should properly address this point

- line 115: please provide more information about Scene Classification map (SCL) and references about it. To not experts in the domain this is very difficult to understand

- the heading 2.4.1 and 2.4.2 are not typographically formatted in the same way. Please format them coherently

- line 189: why in one location the dry matter yield was significantly higher than the others? Due to climatic conditions? Or to a different botanical composition (see my concern previously raised)

- line 198: authors report in figure 2 the presence of an outlier in one location. Was this outlier removed from the analysis? Please specify

- line 202: different VIs were analysed to find out different regressions among them and biomass. They are reported only in supplementary material. I think that some information and references about them should be reported also in material and methods section (synthetically), maybe grouped for similar kind of indices

- line 243: in the text figure 2 seems formatted as a legend (in bold). Please verify and remove bold characters

- line 264 and following: another main issue in my opinion is that in many parts of the paper the authors state that maybe the number of the available samples used “have not been sufficient” (line 265). Is it acceptable? If the authors are not sure that the number of samples is sufficient to assess the variability of the data how they can manage it? Please enhance this part in order to define if the conducted survey is still acceptable under a statistical point of view or not to describe the relationship under investigation. This issue is reported also in the discussion section (lines 327 and following). I think that this concern (of authors themselves) should be addressed properly

- line 339: bibliographic reference number 51 is not present in the reference list. The reference list should be adequately revised in total as two number “1” references are present at the beginning of the list and I have the doubt that ALL references are shifted of one place and not properly related to the text (My personal opinion is that this way to report reference with “numbers” instead of “surnames, year” is really a waste of time….)

- line 352: it seems that this line begins with an index of a list (ii) but I don’t find the first point, that should be identified by a i) at the beginning of the conclusions. Please verify

Author Response

Comment 1:  The paper concerns an interesting topic for the journal as it reports the estimation of aboveground biomass in forage resources from satellite imagery. Even if the research seems well conducted I found some issues that make it not ready for publication in the present form. It should be adequately revised to improve it before publication.

Reply:  Thanks for the positive comments of this paper. We read and implement the reviewer’s comments carefully.

 

 

Comment 2:   in the introduction (lines 29 and following) authors report the importance of the analyzed forage resources for Sweden but without any detail. Please add some data (land use dedicated to this kind of resources, percentage of surface covered, etc.) to better explain the importance of the study

Reply:  The relevant data were added. See lines 31-33.

 

Comment 3:   in the introduction (lines 37 and following) some instruments to estimate forage availability in situ such as spectrometers are cited to reduce time of field samplings of forage. Some other instruments exist to find relationships between biomass and filed parameters such as rising plate meter, which are easily to use and cheaper. Please add some information and references also on this approach

Reply:  The information and reference were added. See lines 41-45.

 

Comment 4:   line 39: the word “spectrometers” is badly written (there is a comma in between the letters, please remove)

Reply:  The word has been corrected. See line 41.

 

Comment 5:    a great issue is reported about cloud presence in satellite imagery (lines 74-75). This is a pivotal point but the authors almost neglected it in the previous part of the introduction. I think authors should enhance a little bit this part, adding some references (now they are missing)

Reply:  The references have been added. See line 81.

 

Comment 5:  line 93: authors report the importance of winter kill and its occurrence, bat in the results section no data are available about it. Did this condition affect the experimental fields? To which extent? Please provide some info about this topic

Reply:  The winter kill mainly affected the clover portion of the forage. Timothy is very cold resistant, whereas red clover can be damaged by ice. The fields used were not badly affected by winter kill. The differences in DMY are more reflective of the timing of the sampling, keeping in mind that the timing of the sampling was not necessarily done to reflect typical harvest times; instead we wanted a wide range of biomass values, to build more robust models. For avoiding misunderstanding, we decided to remove the issue of the winter kill. See lines 111-112.

 

Comment 6:   in the materials and method sections a very big issue in my opinion is the complete lack of information about the vegetation under investigation. Which kind of species were present? Grasses or legumes? Both? In pure stand or in mixture? How differ in botanical composition the fields investigated (that where located in four experimental stations)? Or was the composition of the fields the same? In one location there were 5 fields: were they seeded with the same species? I think this is a very important point to be defined in the article, as different species can affect relationship between biomass and vegetation indices that are a “response” of the occurring vegetation. If the species or the proportion among them were very different, it is probable that this affected the variability of the regressions. A think authors should properly address this point

Reply:  We did consider the effect from botanical composition on the modelling. According to our examination, the botanical composition did not significantly affect the modelling, which was robust to different botanical compositions. We added some relevant information about the botanical composition to the paper.  See lines 90-92, 120-122, 124-126, 221-224, 292-294 and 340-345.

 

Comment 7: line 115: please provide more information about Scene Classification map (SCL) and references about it. To not experts in the domain this is very difficult to understand

Reply:  More explanation and references were added. See lines 137-140.

 

Comment 8:  the heading 2.4.1 and 2.4.2 are not typographically formatted in the same way. Please format them coherently

Reply:  They were re-formatted.

 

Comment 9:  line 189: why in one location the dry matter yield was significantly higher than the others? Due to climatic conditions? Or to a different botanical composition (see my concern previously raised)

Reply:  The reason could be that in 2019, there was more intense solar radiation in Öjebyn. The explanation was added in the paper, see lines 218-220. However, as we explained previously the variation in DMY does not so much reflect differences in the growing conditions, but that we were aiming to collect samples with a wide range of biomass.

 

Comment 10:  line 198: authors report in figure 2 the presence of an outlier in one location. Was this outlier removed from the analysis? Please specify

Reply:  Yes, this outlier was removed. And the explanation was added in lines 230-231.

 

Comment 11:  line 202: different VIs were analysed to find out different regressions among them and biomass. They are reported only in supplementary material. I think that some information and references about them should be reported also in material and methods section (synthetically), maybe grouped for similar kind of indices

Reply:  The table contains was put into the main body of the paper from supplementary material.  See lines 243-248.

 

Comment 12:   line 243: in the text figure 2 seems formatted as a legend (in bold). Please verify and remove bold characters

Reply:  The issue has been addressed.  See line 289.

 

Comment 13:   line 264 and following: another main issue in my opinion is that in many parts of the paper the authors state that maybe the number of the available samples used “have not been sufficient” (line 265). Is it acceptable? If the authors are not sure that the number of samples is sufficient to assess the variability of the data how they can manage it? Please enhance this part in order to define if the conducted survey is still acceptable under a statistical point of view or not to describe the relationship under investigation. This issue is reported also in the discussion section (lines 327 and following). I think that this concern (of authors themselves) should be addressed properly

Reply:  We understand the reviewer’s concern.  The original idea we wanted to express in line 265 and following was that we thought the datasets we had were from several different locations and years, so it might be difficult for the single vegetation index to build the robust model to cover all of the conditions from different locations and years. We modified the sentences a bit.  Although this type of work can always benefit from more sampling, we think that we have sufficient data to build and compare models. See lines 314-318, 397-398, amd 428-430.

 

Comment 14:   line 339: bibliographic reference number 51 is not present in the reference list. The reference list should be adequately revised in total as two number “1” references are present at the beginning of the list and I have the doubt that ALL references are shifted of one place and not properly related to the text (My personal opinion is that this way to report reference with “numbers” instead of “surnames, year” is really a waste of time….)

Reply:  The issue has been solved. See References from line 447.

 

Comment 15:   line 352: it seems that this line begins with an index of a list (ii) but I don’t find the first point, that should be identified by a i) at the beginning of the conclusions. Please verify

Reply:  The issue has been addressed. See line 416.

Round 2

Reviewer 3 Report

I have no further comments on this manuscript.

Reviewer 4 Report

I think that all my previous issues were taken into account. The paper was highly improved and in my opinion is now suitable for publication.