Photosynthetically Active Radiation and Foliage Clumping Improve Satellite-Based NIRv Estimates of Gross Primary Production

Round 1

Reviewer 1 Report

Reviewer’s Comments

The paper aims to improve the gross primary production (GPP) estimates. The authors used GPP data from FLUXNET sites to test these possible improvements by comparing the performance of a model based solely on near-infrared reflectance of vegetation (NIRv) with two other models, one combining NIRv and photosynthetic photon-flux density (PPFD) and the other combining NIRv, PPFD and the foliar clumping index (CI) of each vegetation cover type. The results show that the addition of radiation and CI information to NIRv improved estimates of daily GPP, in particular in evergreen needleleaf forests.

The paper is generally well organized, clear and interesting. Some points need to be clarified:

 

Point 1: Discussion and conclusions should be improved because it is mainly a summary. What is the main contribution of the paper? Which problems did you face and how you solved them? the possible practical applications of the methodology? Further research?

Point 2: More discussions on the advantages and disadvantages of the proposed method should be provided in the result and discussion section.

Point 3: Some qualitative results should be provided in the Abstract.

Point 4: The present Conclusions part is too short. More conclusions should be listed in this part.

Comments for author File: Comments.pdf

Author Response

The paper aims to improve the gross primary production (GPP) estimates. The authors used GPP data from FLUXNET sites to test these possible improvements by comparing the performance of a model based solely on near-infrared reflectance of vegetation (NIRv) with two other models, one combining NIRv and photosynthetic photon-flux density (PPFD) and the other combining NIRv, PPFD and the foliar clumping index (CI) of each vegetation cover type. The results show that the addition of radiation and CI information to NIRv improved estimates of daily GPP, in particular in evergreen needleleaf forests.

The paper is generally well organized, clear and interesting. Some points need to be clarified:

• Thank you very much for your positive assessment about the structure and content of the manuscript, and for your comments and suggestions that have helped to improve it.

 

Point 1: Discussion and conclusions should be improved because it is mainly a summary. What is the main contribution of the paper? Which problems did you face and how you solved them? the possible practical applications of the methodology? Further research?

• We have rewritten the Conclusions sections to make it more “conclusive” in the line suggested by the referee. It now reads: “Our results demonstrate that the addition of daily radiation information (photosynthetic photon-flux density, PPFD) and the clumping index (CI) for each vegetation cover type to the NIRv algorithm improves its ability to estimate GPP. The improvement is related to foliage organization, given that the foliar distribution in the canopy (CI) affects radiation distribution and use, and radiation drives productivity. Evergreen needleleaf forest is the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information. Vegetation type is more determinant in the sensitivity to PPFD changes than latitude or seasonality. Our results highlight that foliar organization and tree structure play an important role in productivity. We advocate for the incorporation of PPFD and CI into NIRv algorithms to improve GPP estimates and GPP models.”

 

Point 2: More discussions on the advantages and disadvantages of the proposed method should be provided in the result and discussion section.

• These issues have been now addressed in the conclusions section. If you consider it necessary, we can also comment them in the discussion section, but we are afraid it could be reiterative.

Point 3: Some qualitative results should be provided in the Abstract.

• Now provided. It reads: “Monitoring of gross primary production (GPP) is necessary for quantifying the terrestrial carbon balance. The near-infrared reflectance of vegetation (NIRv) has been proven to be a good predictor of GPP. Given that radiation powers photosynthesis, we hypothesized that (i) the addition of photosynthetic photon-flux density (PPFD) information to NIRv would improve estimates of GPP and that (ii) a further improvement would be obtained by incorporating the estimates of radiation distribution in the canopy provided by the foliar clumping index (CI). Thus, we used GPP data from FLUXNET sites to test these possible improvements by comparing the performance of a model based solely on NIRv with two other models, one combining NIRv and PPFD and the other combining NIRv, PPFD and the CI of each vegetation cover type. We tested the performance of these models for different types of vegetation cover, at various latitudes, and over the different seasons. Our results demonstrate that the addition of daily radiation information and the CI for each vegetation cover type to the NIRv improves its ability to estimate GPP. The improvement was related to foliage organization, given that the foliar distribution in the canopy affects radiation distribution and use, and radiation drives productivity. The evergreen needleleaf forests are the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information likely as a result of their greater radiation constraints. Vegetation type was more determinant of the sensitivity to PPFD changes than latitude or seasonality. We advocate for the incorporation of PPFD and CI into NIRv algorithms and GPP models to improve GPP estimates.”

Point 4: The present Conclusions part is too short. More conclusions should be listed in this part.

• We have rewritten the conclusions integrating the suggestions of the five referees. It now reads: “Our results demonstrate that the addition of daily radiation information (photosynthetic photon-flux density, PPFD) and the clumping index (CI) for each vegetation cover type to the NIRv algorithm improves its ability to estimate GPP. The improvement is related to foliage organization, given that the foliar distribution in the canopy (CI) affects radiation distribution and use, and radiation drives productivity. Evergreen needleleaf forest is the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information. Vegetation type is more determinant in the sensitivity to PPFD changes than latitude or seasonality. Our results highlight that foliar organization and tree structure play an important role in productivity. We advocate for the incorporation of PPFD and CI into NIRv algorithms to improve GPP estimates and GPP models.”

Reviewer 2 Report

This paper presents a method for estimating GPP using remote sensing products and ground data. The authors present their methods and findings in a concise yet easy to follow way. I have no issue with the length, though short among most publication standards, and applaud the authors for their ability to do so. Thus, my comments are minor:

1. There are several lines where references are referred to by name rather than by number - 47, 49, 60, 129, 146, 165

2. There are several extra spaces between words in lines 43, 81

3. Acronyms go in parentheses so switch the order of SIF in lines 75/76

4. line 165: fix should read fixed

5. Conclusion is very brief. Consider expanding.

Author Response

This paper presents a method for estimating GPP using remote sensing products and ground data. The authors present their methods and findings in a concise yet easy to follow way. I have no issue with the length, though short among most publication standards, and applaud the authors for their ability to do so. Thus, my comments are minor:

• Thank you very much for your positive assessment of our manuscript as concise and easy to follow, and for your comments and suggestions that have helped to improve the manuscript.

There are several lines where references are referred to by name rather than by number - 47, 49, 60, 129, 146, 165 Jiang et al 24

• Thank you for noticing this. We have replaced the references by number.

There are several extra spaces between words in lines 43, 81

• We have corrected it.

Acronyms go in parentheses so switch the order of SIF in lines 75/76

• Thank you for making us notice it. We have now switched the order.

Line 165: fix should read fixed

• We have corrected it.

Conclusion is very brief. Consider expanding.

• We have rewritten the conclusions integrating the suggestions of the five referees. It now reads: “Our results demonstrate that the addition of daily radiation information (photosynthetic photon-flux density, PPFD) and the clumping index (CI) for each vegetation cover type to the NIRv algorithm improves its ability to estimate GPP. The improvement is related to foliage organization, given that the foliar distribution in the canopy (CI) affects radiation distribution and use, and radiation drives productivity. Evergreen needleleaf forest is the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information. Vegetation type is more determinant in the sensitivity to PPFD changes than latitude or seasonality. Our results highlight that foliar organization and tree structure play an important role in productivity. We advocate for the incorporation of PPFD and CI into NIRv algorithms to improve GPP estimates and GPP models.”

Reviewer 3 Report

The authors investigate three models based on NIRv for GPP estimation. The study overall is interesting, but the presentation should be considerably improved. 

 

General comments:

- The introduction contains a lot of information, but it is too brief. Sometimes, I had the feeling that separate sentences were just added together. The paper will benefit from rewriting this introduction section. 

- In the materials and methods section, the time period of the flux data and the remote sensing data should be mentioned. In table 1, the years for which data is available should be added. 

- statistical analysis: the authors use RMSE and mean error (bias) and state in section 2.5 that the latter is a measure of accuracy. It should be added that RMSE is a measure of uncertainty. Yet further in the results section, the term 'disperion' is used, but this is not defined in section 2.5. Dispersion is actually the standard deviation of the bias and should be added to the results. RMSE is uncertainty and as such is the combination of bias and dispersion. In addition, the bias is not reported in the results section, but only in some graphs in the supplementary material. The results of the bias should be added also in the main paper. 

Detailed comments:

- line 60: should be numbered reference

- line 92: if you use data from sites between 70°N and 30°N, you can hardly call this a global distribution

- line 93: homogeneous sites: did you check this for the entire period for which you have data. Are you sure that the land cover did not change?

- line 101: IGBP classification is based on data from the period 2001-2004. Is this still correct for the data that you use? 

- table 1: all the white space should be removed. Add the time period for which data is used in this study

- figure 2: axes are not readable. What is the line, the 1-1 line? This should be added to the caption. Add the bias and precision results.

- line 175: here you discuss dispersion, but this term is not explained. I assume you use RMSE for this, but this is not correct, you should use precision for this (standard deviation of the bias)

- line 175: R2 should be R²

- lines 193-208: you discuss only the change in bias without providing the bias for the different datasets. A negative change in bias is not always what you want, especially when the first bias was already negative. What you want is the bias to be closer to 0, and this should be discussed. I suggest to present these results differently not to allow any misunderstanding.

- line 212: in figure S2 the bias is worse after adding PPFD, but here it is only mentioned that there is an improvement. This should be reformulated.

-line 224: '... in terms of accuracy and bias.' accuracy = bias according to the definition in section 2.5. Rephrase.  

 

Author Response

General comments:

- The introduction contains a lot of information, but it is too brief. Sometimes, I had the feeling that separate sentences were just added together. The paper will benefit from rewriting this introduction section. 

• Thank you for your suggestion. We have rewritten the introduction section in order to make it flow well.

- In the materials and methods section, the time period of the flux data and the remote sensing data should be mentioned. In table 1, the years for which data is available should be added. 

• We have now added this information in the material and methods section and in the Table 1.

- statistical analysis: the authors use RMSE and mean error (bias) and state in section 2.5 that the latter is a measure of accuracy. It should be added that RMSE is a measure of uncertainty. Yet further in the results section, the term 'dispersion' is used, but this is not defined in section 2.5. Dispersion is actually the standard deviation of the bias and should be added to the results. RMSE is uncertainty and as such is the combination of bias and dispersion. In addition, the bias is not reported in the results section, but only in some graphs in the supplementary material. The results of the bias should be added also in the main paper. 

• Thank you for your explanation and suggestions. We have added that RMSE is a measure of uncertainty in the statistical analysis section, and we have removed the term dispersion in the manuscript to avoid confusion and have replaced it by uncertainty when referring to RMSE.
• Regarding the standard deviation of the bias, we have calculated it using the package blandr, but the RMSE as determined in this work yields exactly the same value as the standard deviation of the bias (Bland & Altman 2010), and because of this we have not included it.
• We have added bias information in the figure where it was lacking, Fig. 2.

Detailed comments:

- line 60: should be numbered reference

• We have changed the reference by the numbered reference.

- line 92: if you use data from sites between 70°N and 30°N, you can hardly call this a global Distribution

• We agree and have deleted these words from the sentence.

- line 93: homogeneous sites: did you check this for the entire period for which you have data. Are you sure that the land cover did not change?

• Thanks for the insightful question. The check of site homogeneity was based on information about land cover, management and disturbances provided by FLUXNET principal investigators for each selected site. This information guaranteed that for the selected sites there were not changes in land cover in the flux footprint and pixel area during the period of flux measurements.   

- line 101: IGBP classification is based on data from the period 2001-2004. Is this still correct for the data that you use? 

• Thanks for pointing it out. The land cover classification at the sites were provided by site investigators following the IGBP classification and adapted in case of land cover changes during the flux measurement period. To better clarify it in the manuscript, we have ow changed the sentence at L101-102 (Vegetation cover types were defined using the International Geosphere–Biosphere Programme (IGBP) land classification (https://fluxnet.org/data/badm-data-templates/igbp-classification/) to: ‘Vegetation cover types were provided by FLUXNET using the International Geosphere–Biosphere Programme (IGBP) land classification (https://fluxnet.org/data/badm-data-templates/igbp-classification/).’

- table 1: all the white space should be removed. Add the time period for which data is used in this study

• Thank you for your suggestion. We have added a column in the Table 1 with the time period of used data for each site.

- figure 2: axes are not readable. What is the line, the 1-1 line? This should be added to the caption. Add the bias and precision results.

• We have improved axes readability. Yes, the black line is the 1:1 line. We have also added this information to the figure caption. We have provided the bias, RMSE and R² statistics of the comparison.

- line 175: here you discuss dispersion, but this term is not explained. I assume you use RMSE for this, but this is not correct, you should use precision for this (standard deviation of the bias)

• We have replaced dispersion by uncertainty to avoid confusion.

- line 175: R2 should be R²

• We have corrected it.

- lines 193-208: you discuss only the change in bias without providing the bias for the different datasets. A negative change in bias is not always what you want, especially when the first bias was already negative. What you want is the bias to be closer to 0, and this should be discussed. I suggest to present these results differently not to allow any misunderstanding.

• We calculated the variation in the absolute value of bias. We have added this information to the caption of the figure and modified the title of the y axis accordingly.

- line 212: in figure S2 the bias is worse after adding PPFD, but here it is only mentioned that there is an improvement. This should be reformulated.

• Thank you for your suggestion. We have added to the sentence that bias increased after adding PPFD.

-line 224: '... in terms of accuracy and bias.' accuracy = bias according to the definition in section 2.5. Rephrase.  `

• We have replaced accuracy by uncertainty.

Reviewer 4 Report

This paper used GPP data from FLUXNET sites to compare the performance of a model based solely on NIRv to two other models, one combining NIRv and PPFD and the other combining NIRv, PPFD, and the CI of each vegetation cover type at various latitudes, and throughout the seasons. their findings demonstrate that incorporating radiation and CI into NIRv improved estimates of daily GPP, particularly in evergreen needleleaf forests. The paper is well organized and easy to follow. I only have a few comments.

1.      L137-141: For the description of the method, it is recommended to be more detailed, and it is better to list the exact form of the equation.

2.      I like Figure 3. It is visually appealing and depicts the impact of CI. One question: it appears from the caption that CI is not taken into account with models GPPnirppfd and GPPnirv, but the image shows that the degree of improvement has a good relationship with CI. Could you give more explanations here (despite the fact that L220 contains a relevant description, it remains unclear about the relationship between PPFD and vegetation structure CI, which is also related to the following comment.). Furthermore, there are several instances of text overlapping in the images, and improvements are suggested.

3.      It is not very clear how the PPFD information are obtained and added to the model.

4.      The conclusion part is a bit simple, including some words are not very detailed, such as the “daily radiation information”. It is suggested that this section could be a little more detailed.

5.      L41, L54, L94: CO2, km2… please check the Sub/Superscript, as well as the rest of the paper.

6.      L146: Please double-check the citation form and other locations.  

7.      Figure 2's font size could be increased to make it more visible.  

8.      Fig. S3, L23: please check the sentence “Different colours identify the vegetation cover types”.

Author Response

This paper used GPP data from FLUXNET sites to compare the performance of a model based solely on NIRv to two other models, one combining NIRv and PPFD and the other combining NIRv, PPFD, and the CI of each vegetation cover type at various latitudes, and throughout the seasons. their findings demonstrate that incorporating radiation and CI into NIRv improved estimates of daily GPP, particularly in evergreen needleleaf forests. The paper is well organized and easy to follow. I only have a few comments.

• Thank you very much for your positive evaluation of the paper organization and its clarity, and for your comments and suggestions that have helped to improve the manuscript.

 

L137-141: For the description of the method, it is recommended to be more detailed, and it is better to list the exact form of the equation.

 

• We show the detail of the variables used in each model, but not the particular parameters because they would vary with a different set of FLUXNET sites.

 

I like Figure 3. It is visually appealing and depicts the impact of CI. One question: it appears from the caption that CI is not taken into account with models GPPnirppfd and GPPnirv, but the image shows that the degree of improvement has a good relationship with CI. Could you give more explanations here (despite the fact that L220 contains a relevant description, it remains unclear about the relationship between PPFD and vegetation structure CI, which is also related to the following comment.). Furthermore, there are several instances of text overlapping in the images, and improvements are suggested.

     

• The improvement in GPP after adding radiation to NIRv was related to CI, i.e. foliage organization. This is due to the light distribution and its use inside the canopy. If leaves are more clumped on shoots, branches and crowns, more light can penetrate the canopy, making it less light-saturated and more sensitive to light changes. We now explain it in the revised discussion section. We have also corrected the overlapping in the figures.

 

It is not very clear how the PPFD information are obtained and added to the model.

•  PPFD data was obtained from FLUXNET2015 Dataset Tier 1 (https://fluxnet.org/data/fluxnet2015-dataset/). We used the daily mean radiation data estimated from average half-hourly records. This is explained in the Material and Methods section, subsection 2.1 Carbon flux. It was added to the model as an independent variable. We have now clarified it.

 

The conclusion part is a bit simple, including some words are not very detailed, such as the “daily radiation information”. It is suggested that this section could be a little more detailed.

• We have rewritten the conclusions integrating the suggestions of the five referees. The conclusion part is a little more detailed now. We have also clarified that “daily radiation information” refers to “photosynthetic photon-flux density, PPFD”. It now reads: “Our results demonstrate that the addition of daily radiation information (photosynthetic photon-flux density, PPFD) and the clumping index (CI) for each vegetation cover type to the NIRv algorithm improves its ability to estimate GPP. The improvement is related to foliage organization, given that the foliar distribution in the canopy (CI) affects radiation distribution and use, and radiation drives productivity. Evergreen needleleaf forest is the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information. Vegetation type is more determinant in the sensitivity to PPFD changes than latitude or seasonality. Our results highlight that foliar organization and tree structure play an important role in productivity. We advocate for the incorporation of PPFD and CI into NIRv algorithms to improve GPP estimates and GPP models.”

 

L41, L54, L94: CO2, km2… please check the Sub/Superscript, as well as the rest of the paper.

• We have checked and corrected when needed.

L146: Please double-check the citation form and other locations. 

• We have replaced the reference by the numbered reference and checked it for all the document.

Figure 2's font size could be increased to make it more visible. 

•  We have increased the font size in Fig 2.

Fig. S3, L23: please check the sentence “Different colours identify the vegetation cover types”.

• Thank you for noticing it. We have replaced the sentence by: “Different colours identify the latitudinal ranges.”

Reviewer 5 Report

General comments:

The manuscript entitled "Photosynthetically active radiation and foliage clumping improve satellite-based NIRv estimates of gross primary production" aims to explore how radiation as a modulator of NIRv could improve GPP estimates, especially at high latitudes and in the latter part of the year, given that photosynthetic photon-flux density (PPFD) drives the GPP of the green biomass. The method of this study is appropriate, and it has a certain practical significance.  However, there are still some minor questions to be considered as follows:

Specific comments:

1. line 84, the significance of this study should be further clarified.  

2. How reliable is the method? What are the advantages and disadvantages compared with the existing relevant research? These should also be discussed separately.

3. The application scope of this method should also be clarified in the abstract.

4. It is suggested to further summarize and enrich the conclusion.  

Author Response

General comments:

The manuscript entitled "Photosynthetically active radiation and foliage clumping improve satellite-based NIRv estimates of gross primary production" aims to explore how radiation as a modulator of NIRv could improve GPP estimates, especially at high latitudes and in the latter part of the year, given that photosynthetic photon-flux density (PPFD) drives the GPP of the green biomass. The method of this study is appropriate, and it has a certain practical significance.  However, there are still some minor questions to be considered as follows:

• Thank you very much for your positive evaluation of the method of this article and its practical significance, and for your comments and suggestions that have helped to improve the manuscript.

Specific comments:

line 84, the significance of this study should be further clarified.  

• We have now further clarified that “The improvement in GPP after adding radiation to NIRv was related to foliage organization and vegetation type, that seem to be more determinant in the sensitivity to PPFD changes than latitude or seasonality. We thus advocate for the incorporation of PPFD and CI into NIRv as a means of improving GPP estimates, and the inclusion of this structural information on foliage organization in GPP models.”

How reliable is the method? What are the advantages and disadvantages compared with the existing relevant research? These should also be discussed separately.

• These issues have been now addressed in the conclusions section. If you consider it necessary, we can also comment them in the discussion section, but we are afraid it could be reiterative.

The application scope of this method should also be clarified in the abstract.

• Done. It now reads: “We advocate for the incorporation of PPFD and CI into NIRv algorithms and GPP models to improve GPP estimates”.

It is suggested to further summarize and enrich the conclusion.  

• We have rewritten the conclusions integrating the suggestions of the five referees. It now reads: “Our results demonstrate that the addition of daily radiation information (photosynthetic photon-flux density, PPFD) and the clumping index (CI) for each vegetation cover type to the NIRv algorithm improves its ability to estimate GPP. The improvement is related to foliage organization, given that the foliar distribution in the canopy (CI) affects radiation distribution and use, and radiation drives productivity. Evergreen needleleaf forest is the vegetation cover type with the greatest improvement in GPP estimation after the addition of CI information. Vegetation type is more determinant in the sensitivity to PPFD changes than latitude or seasonality. Our results highlight that foliar organization and tree structure play an important role in productivity. We advocate for the incorporation of PPFD and CI into NIRv algorithms to improve GPP estimates and GPP models.”