Solar-Induced Chlorophyll Fluorescence Trends and Mechanisms in Different Ecosystems in Northeastern China

Round 1

Reviewer 1 Report

In this paper, the authors have evaluated the relationships between Solar-induced chlorophyll fluorescence (SIF) from OCO-2 and vegetation indices (VIs) from MODIS. The authors have considered three ecosystems, including cropland, forest, and grassland, and performed the comparisons between VIs and SIF. By these comparisons, the author intended to understand the SIF seasonality in the ecosystems assess the relationships between the indices at regional and footprint scales. Additionally, they considered the association between meteorological factors and SIF in the different ecosystems.

The authors have drawn several conclusions that I found difficult to comprehend.

First, the authors claim that their analysis showed that “the forest, cropland, and grassland areas experience different growing seasons, and the SIF value reaches its peak at a different time of year in each”. I am not sure based on which evidence the authors have claimed this. Considering Figures 2 and 3, I do not find different peaks for the SIF products. Although a double peak likely occurs when SIF products are evaluated (as already established by other authors, e.g., https://doi.org/10.5194/bg-17-405-2020), I did not find this characteristic based on the analysis and data provided in the current version of the manuscript.

Second, the authors point out that “the relationships observed between SIF and VIs were stronger at a regional scale than at footprint scale”. First, the footprint scale is not clearly described in the manuscript. Second, the number of sampling points in the footprint and regional scales affects this conclusion. Third, even if the number of sampling points are similar at both scales, this conclusion is expected due to the spatial differences between the OCO2 and MODIS data. Overall, averaging out the data over large areas would reduce the bias. This cannot be considered as a new conclusion unless the authors refer to something else that I am missing.

Third, the authors state that their findings “suggest that dissimilar ecosystems differ in their sensitivity to meteorological factors”. This is a fact, not a conclusion. I am not sure what message the authors wanted to convey, but it is a fact that different vegetation cover, even different varieties of similar vegetation cover, respond differently to the meteorological variable. Additionally, I did not understand what the authors tried to show by correlating SIF with meteorological variables. This is also important to note that there is strong co-variance between the meteorological variables, and simply relying on a linear correlation analysis would be misleading.  

 Based on these points, I suggest that the authors carefully revisit their work and make the necessary changes particularly focusing on innovative conclusions.

Author Response

Dear Reviewer 1,

We must thank you and other reviewers for the critical feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.

Based on the comments we received, careful modifications have been made to the manuscript. All changes were marked in red text.

We hope the new manuscript will meet your magazine’s standard. Below you will find our point-by-point responses to the reviewers’ comments/ questions:

Question 1: First, the authors claim that their analysis showed that “the forest, cropland, and grassland areas experience different growing seasons, and the SIF value reaches its peak at a different time of year in each”. I am not sure based on which evidence the authors have claimed this. Considering Figures 2 and 3, I do not find different peaks for the SIF products. Although a double peak likely occurs when SIF products are evaluated (as already established by other authors, e.g., https://doi.org/10.5194/bg-17-405-2020), I did not find this characteristic based on the analysis and data provided in the current version of the manuscript.

Reply: Thanks for your advice and you are right that this conclusion is somewhat biased. From Figure 3 we can see the peak SIF value of cropland in late July and forest in the beginning of July and grassland in August. Of the three ecosystems, forest reaches the peak SIF value first and followed by cropland and then grassland and we can’t draw the conclusion that they experience different growing seasons because of different LAI and even photosynthetic modes for tree and herbaceous plant.

From this research we didn’t find a double peak of SIF as mentioned by Turner et al. (2020) which is due to two ecosystems that are out of phase with each other. Our study regions are quite different with California as we selected three nearly pure ecosystems. The forest is Greater Khingan Mountains which is the largest virgin forest in China; Cropland the Songnen Plain which is one of the ‘black soil regions’, in the world, is one of the most important grain pro-duction areas in China and this region has the largest acreage of corn; Grassland is China's largest natural pastureland. We cited this reference this time when we discuss our results.

At last, we deleted this sentence from conclusion.

Question 2: Second, the authors point out that “the relationships observed between SIF and VIs were stronger at a regional scale than at footprint scale”. First, the footprint scale is not clearly described in the manuscript. Second, the number of sampling points in the footprint and regional scales affects this conclusion. Third, even if the number of sampling points are similar at both scales, this conclusion is expected due to the spatial differences between the OCO2 and MODIS data. Overall, averaging out the data over large areas would reduce the bias. This cannot be considered as a new conclusion unless the authors refer to something else that I am missing.

Reply: This is quite a good question. Firstly, the footprint scale is the ground pixel size of 1.3 km × 2.25 km which is the spatial resolution of OCO-2. This time we added this information in section 2.2.

Secondly, yes, you are right. The sampling number must affect the correlation coefficient between SIF and VIs and the regional scales sample number is within 100 but for footprint scale is more than tens of thousands. This time we deleted the competition between regional scale and footprint scale and just compare the different relationships between SIF and MODIS for different ecosystems.

Third, yes, you are right. The spatial differences also affect the relationships. In fact, we resampled the spatial resolution of SIF and MODIS with the same resolution by using buffer rings as descripted below in section 2.6:

“The selected 100 SIF points were buffered with 2 km radius and then extracted the averaged values of EVI, FPAR and LAI within the circle the scatter plots were plotted accordingly.”

Based on your comments, many sentences have been deleted from conclusion.

Question 3: Third, the authors state that their findings “suggest that dissimilar ecosystems differ in their sensitivity to meteorological factors”. This is a fact, not a conclusion. I am not sure what message the authors wanted to convey, but it is a fact that different vegetation cover, even different varieties of similar vegetation cover, respond differently to the meteorological variable. Additionally, I did not understand what the authors tried to show by correlating SIF with meteorological variables. This is also important to note that there is strong co-variance between the meteorological variables, and simply relying on a linear correlation analysis would be misleading.

Reply: Thank you so much about this question. Yes, this is not a conclusion and this time we deleted this sentence. In fact, we just want to check the relationships between SIF and meteorological factors in different ecosystems. The different sensitivities means that we couldn’t model SIF from VIs and meteorological factors without distinguish ecosystems. Many researchers used VIs and meteorological factors as independent variable to model SIF at regional scale even global scale and our results want to indicates that to improve model accuracy, we must model SIF at each ecosystem respectively.

Yes, you are right that there is strong co-variance between the meteorological variables but we just calculated the relationships between temperature and SIF and between precipitation and SIF (no significant). We want to know how meteorological factors affect the photosynthesis activities in different ecosystems. With regards to the seasonal trend of temperature, we de-seasonalized the time serious data used the value minus the mean at this time.

Question 4: Based on these points, I suggest that the authors carefully revisit their work and make the necessary changes particularly focusing on innovative conclusions.

Reply: Thank you so much for this advice and we have carefully revisited all the manuscript and changed the conclusions based on your comments.

Reviewer 2 Report

Major issues:

1. Linear regression and correlation were used for comparing SIF and other variables. But apparently, not all relationships between SIF-EVI and SIF-FPAR were linear (Figures 4 and 5). The scatter plots in Figure 5 show mostly nonlinear features, which may fit better with exponential models. The exponential relationship indicates higher sensitivity in SIF than EVI or FPAR, when SIF is high (see Figure 5).

2. The high correlation between SIF and air temperature is mainly caused by seasonality. In other words, both SIF and air temperature time series show seasonal patterns that are typically low in the non-growing season and high in the growing season. Relationship between the two sets of time series with seasonal patterns is likely to create spurious correlation rather than cause-and-effect relationship. The correct correlation or regression analysis should use de-seasonalized time series. A simple technique for de-seasonalizing time series is to use anomaly or "departure from mean". The relative low correlations between SIF and precipitation found in this study were because precipitation did not have strong seasonal patterns like temperature.

3. The comparisons between SIF and MODIS products were conducted using two scales: OCO-2 footprint scale and regional scale. These two scales are not sufficient to explore the correlations between the two datasets. The OCO-2 footprint is about 2-3 km, while the pixel size of the MODIS data is 500 m or 1 km. There might be a problem caused by image misregistration for both products. Using sampling blocks with aggregated pixels (e.g., 3 x 3, 5 x 5, etc.) around the OCO-footprint may help steer away from the misregistration issue. At the regional scale, the analysis in the study used average values of the variables for each cover type within the entire study area. Although the regional analysis generated higher correlations, the results were less useful than the analysis at footprint scale. All local variations and heterogenous features were smoothed out by using regional averaging techniques.

Minor issues:

Title: May specify the study area in the title. The current title is too general for a research paper.

L68: “closed” should be “close”.

L106-111: In this paragraph or any previous paragraphs the study area should be specified and the reasons for selecting the area need to be stated clearly.

L121-127: Provide data source for the land cover map, either in this paragraph or in the figure 1 caption.

L122-124: Need reference(s) for the vegetation types and species.

L128-130: Break up the long sentence into two.

L130: What are other crops produced in there beside corn?

L151: Delete “was”.

L156: Change to “MODIS EVI, FPAR, and LAI products”.

L157: Add a comma after “(FPAR)”.

L162-165: Need references for the functions of LAI and FPAR.

L185: The size of the footprint should be specified.

L198-206: Explain why a gap exists from Jul 27 to Sept 29, 2017 (Figure 2).

L218-220: “water stress and temperature stress” should not be a factor since the seasonal changes are calculated from the multi-year average.

L254-259 and Table 2: Because R-values for SIF-FPAR (forest) and SIF-LAI (all cover types) are very low (<0.1), p-value can not be <0.001. Please recheck the calculation of the statistical analysis.

L283: What is “former 16 d temperature”? Does it mean “mean temperature of previous 16-day period”? 

L307-309: As GOSIF and HRGCSIF dataset are one part of the study, a brief description of the two datasets should be provided in Section 2. 

L321-327: Suggest adding scatter plots for the comparison between OCO-2 SIF data and simulated data. The numbers in Table 2 are not enough for indicating the relationship. Because the slopes are far away from 1, big biases exit between the two datasets, either for footprint or regional scale.

L431-434: These two sentences can be moved in introduction.

Figure 1: Add a scale bar in the map. Use more distinguishable colors for cropland and grassland, and for the outlines of forest and grassland.

Figure 2: Label years on the x-axis.

Figures 2&3: Use more distinguishable colors for forest and grassland.

Figures 4&5: Add tick marks on both axes.

Figure 5: Fonts of values are not consistent between x-axis and y-axis (Figure 5). To be consistent with Figure 4, put SIF on y-axis and EVI etc. on x-axis.

Author Response

Dear Reviewer 2,

We must thank you and other reviewers for the critical feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.

Based on the comments we received, careful modifications have been made to the manuscript. All changes were marked in red text.

We hope the new manuscript will meet your magazine’s standard. Below you will find our point-by-point responses to the reviewers’ comments/ questions:

Major issues:

Question 1: Linear regression and correlation were used for comparing SIF and other variables. But apparently, not all relationships between SIF-EVI and SIF-FPAR were linear (Figures 4 and 5). The scatter plots in Figure 5 show mostly nonlinear features, which may fit better with exponential models. The exponential relationship indicates higher sensitivity in SIF than EVI or FPAR, when SIF is high (see Figure 5).

Reply: Thank you so much and this is quite a good question and another reviewer also proposed the same question. This time, for each scatter plot, we tried many regression models and selected the best fit one. We added the sentence in section 3.2 as below:

“Liner, exponential, power and quadratic models were tried for each scatter plot and then selected the best fit one. F-test was assessed to test significance of the regression models.”

Question 2: The high correlation between SIF and air temperature is mainly caused by seasonality. In other words, both SIF and air temperature time series show seasonal patterns that are typically low in the non-growing season and high in the growing season. Relationship between the two sets of time series with seasonal patterns is likely to create spurious correlation rather than cause-and-effect relationship. The correct correlation or regression analysis should use de-seasonalized time series. A simple technique for de-seasonalizing time series is to use anomaly or "departure from mean". The relative low correlations between SIF and precipitation found in this study were because precipitation did not have strong seasonal patterns like temperature.

Reply: This is quite a good question and we are so sorry about this question. Based on your advice, we de-seasonalized the time serious data used the value minus the mean for both figure 6 and figure 7.

Question 3: The comparisons between SIF and MODIS products were conducted using two scales: OCO-2 footprint scale and regional scale. These two scales are not sufficient to explore the correlations between the two datasets. The OCO-2 footprint is about 2-3 km, while the pixel size of the MODIS data is 500 m or 1 km. There might be a problem caused by image misregistration for both products. Using sampling blocks with aggregated pixels (e.g., 3 x 3, 5 x 5, etc.) around the OCO-footprint may help steer away from the misregistration issue. At the regional scale, the analysis in the study used average values of the variables for each cover type within the entire study area. Although the regional analysis generated higher correlations, the results were less useful than the analysis at footprint scale. All local variations and heterogenous features were smoothed out by using regional averaging techniques.

Reply: This is quite a good question. In fact, before calculating the relationships between SIF and MODIS indices, we firstly selected 100 SIF points and then did a buffer ring with 2 km radius for each SIF point and then calculated the mean value of MODIS indices within the buffer ring. We added the method description in section 2.6 and hope it matches the standard of footprint relationships.

With regards to the regional scale, yes, you are right. Because of the averaged values of SIF and MODIS indices, the local variations and heterogenous features were smoothed out. Here, we don’t just want to give the relationships between SIF and MODIS indices and also don’t just want to compare the R² with footprint scale. We want to compare the relationships between different ecosystems. We also want to tell the readers that in different ecosystems, no matter the scales, the relationships between SIF and MODIS indices are different. This time we deleted the comparation between regional scale and footprint scale.

Minor issues:

Question 1： Title: May specify the study area in the title. The current title is too general for a research paper.

Reply: We added the study area in the title and then the title is as below:

“Solar-induced chlorophyll fluorescence Trends and Mechanisms in Different Ecosystems in Northeastern China”

Question 2：L68: “closed” should be “close”.

Reply: We revised “closed” to “close”.

Question 3：L106-111: In this paragraph or any previous paragraphs the study area should be specified and the reasons for selecting the area need to be stated clearly.

Reply: That is a good advice and we added the sentence as below:

“The purpose of this study is to explore the relationships of SIF and vegetation indices and the sensitivity of SIF to meteorological data in different ecosystems. To achieve this, three typical ecosystems in Northeastern China, forest, grassland and cropland were selected.”

Question 4：L121-127: Provide data source for the land cover map, either in this paragraph or in the figure 1 caption.

Reply: We added the reference for the data of Land cover type as below:

Zhang, X.; Liu, L.; Chen, X.; Gao, Y.; Xie, S.; Mi, J. GLC_FCS30: global land-cover product with fine classification system at 30m using time-series Landsat imagery. Earth Syst. Sci. Data 2021, 13, 2753-2776, doi:10.5194/essd-13-2753-2021.

Question 5 L122-124: Need reference(s) for the vegetation types and species.

Reply: We added the references for the vegetation types and species in Greater Khingan Mountains.

Question 6 L128-130: Break up the long sentence into two.

Reply: This time, we revised the sentence as below:

“The cropland ecosystem is located in the central Songnen Plain, which is one of the ‘black soil regions’, in the world. Songnen Plain is one of the most important grain production areas in China and this region has the largest acreage of corn and also small area of rice.”

Question 7 L130: What are other crops produced in there beside corn?

Reply: Beside corn, there is also rice but the area is not quite large and we added the sentence as “and also small area of rice”.

Question 8 L151: Delete “was”.

Reply: We deleted the word “was”.

Question 9 L156: Change to “MODIS EVI, FPAR, and LAI products”.

Reply: We revised “MODIS EVI and FPAR and LAI products” to “MODIS EVI, FPAR, and LAI products”.

Question 10 L157: Add a comma after “(FPAR)”.

Reply: We added “,” this time.

Question 11 L162-165: Need references for the functions of LAI and FPAR.

Reply: This time we added the reference of “Myneni, R., Knyazikhin, Y., Park, T. (2015). MOD15A2H MODIS Leaf Area Index/FPAR 8-Day L4 Global 500m SIN Grid V006. NASA EOSDIS Land Processes DAAC”.

Question 12 L185: The size of the footprint should be specified.

Reply: In section 2.2 we added a sentence. “Each second, OCO-2 acquires 24 data points with a footprint of 1.3 km × 2.25 km.”

Question 13 L198-206: Explain why a gap exists from Jul 27 to Sept 29, 2017 (Figure 2).

Reply: The gaps are because of SIF data missing and this time we added the description in figure 2.

Question 14 L218-220: “water stress and temperature stress” should not be a factor since the seasonal changes are calculated from the multi-year average.

Reply: This is a quite good question and this time we deleted this reason.

Question 15 L254-259 and Table 2: Because R-values for SIF-FPAR (forest) and SIF-LAI (all cover types) are very low (<0.1), p-value can not be <0.001. Please recheck the calculation of the statistical analysis.

Reply: We check the statistical data again and found that all the P-value <0.001. Maybe this is because of the larger sample number (larger than 400).

Question 16 L283: What is “former 16 d temperature”? Does it mean “mean temperature of previous 16-day period”?

Reply: Yes, that is right. Maybe “former 16 d temperature” is unintelligible. This time we changed to “mean temperature of previous 16-day”.

Question 17 L307-309: As GOSIF and HRGCSIF dataset are one part of the study, a brief description of the two datasets should be provided in Section 2.

Reply: Yes, you are right. This time we added a brief description of GOSIF and HRGCSIF in section 2.5.

Question 18 L321-327: Suggest adding scatter plots for the comparison between OCO-2 SIF data and simulated data. The numbers in Table 2 are not enough for indicating the relationship. Because the slopes are far away from 1, big biases exit between the two datasets, either for footprint or regional scale.

Reply: This is quite a good advice and this time we added the scatter plots and please see Figure 8.

Question 19 L431-434: These two sentences can be moved in introduction.

Reply: That is a good advice and we moved this sentence to introduction.

Question 20 Figure 1: Add a scale bar in the map. Use more distinguishable colors for cropland and grassland, and for the outlines of forest and grassland.

Reply: Thanks for your advice and this time we revised figure 1 based on your advice.

Question 21 Figure 2: Label years on the x-axis.

Reply: Thank you for this advice and we added the years label this time.

Question 22 Figures 2&3: Use more distinguishable colors for forest and grassland.

Reply: Thanks for your advice and this time we changed the label color of grassland.

Question 23 Figures 4&5: Add tick marks on both axes.

Reply: Thanks for your advice and this time we added the tick marks for Figures 4 and 5.

Question 24 Figure 5: Fonts of values are not consistent between x-axis and y-axis (Figure 5). To be consistent with Figure 4, put SIF on y-axis and EVI etc. on x-axis.

Reply: This is quite a good question. Based on your advice, we revised Figure 5 carefully.

Reviewer 3 Report

This paper proposes a trend analysis of Solar-induced chlorophyll fluorescence over different vegetation type northeaster China. In principle, the issue is very interesting and worth publishing. Nevertheless, there are problems in experimental design and statistical analysis. A deep major revision is required and there are several concerns that should be considered before the paper can be published as a scientific contribution.

General comments:

The main problem is related to statistical analysis about the relationship of SIF and other environmental variables.  In particular, author have assumed a linear relationship for all explored correlations. Moreover, many sentences about findings are not supported by evidence or by hypothesis tests.   

Specific comments:

1. L 138. Please add in the caption the cartographical reference system adopted. I suppose WGS84.
2. L 200- you should discuss these results in respect to recent literature. Especially the Vegetation index peak of phenological season was found different between several vegetated ecosystem in:

Sarvia, F., De Petris, S., & Borgogno-Mondino, E. (2021). Exploring climate change effects on vegetation phenology by MOD13Q1 data: The Piemonte region case study in the period 2001–2019. Agronomy, 11(3), 555.

3. L220-223. This is your hypothesis. You don’t have any evidence to support this sentence. Note that ordinarily corn (the main crop type in your study area) has high LAI values due to its higher parenchyma density per leaf. Moreover, since it is a C4 plant, therefore it has a better carbon fixing rate than other vegetation ecosystems in your study area. Therefore, we cannot be sure that SIF increasing is affected by human (fertilization) rather than other natural factors.

4. L 227. I think that MODIS- derived data are characterized by a high correlation therefore you should discuss why you selected that variable if one of them could summarized the other ones.

5. Note that in this study you have adopted the determination coefficient. The latter doesn't tell you about the relationship between variables, but the percentage of variance explained by your regressive model (in this study you suppose linear). Therefore, it is not clear how did you obtain p-value. Which test did you performed? I suppose that you tested the significance of Pearson correlation coefficient (r), but it can be referred to R2. Moreover, why did you suppose linear the regression model? why not a power model or exponential? this is a critical issue. You have to prove that the linear model is the best fitting one that could somehow describe the relationship between SIF and other variables.

6. L250-260. Using the R2 of a linear model could hide the real fitting. Maybe using other model (e.g., exponential, power, quadratic) these results could be totally and unexpectedly different.

7. L 319-320. This sentence is not supported by rigours statistical analysis. In fact, from table 3 RMSE and Slope values are quite similar. Therefore, i wonder if there are significant differences. A statistical hypothesis test is required to prove the differences between OCO-2 observed SIF and modelled values and vegetation ecosystems.

Author Response

Dear Reviewer 3,

We must thank you and other reviewers for the critical feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.

Based on the comments we received, careful modifications have been made to the manuscript. All changes were marked in red text.

We hope the new manuscript will meet your magazine’s standard. Below you will find our point-by-point responses to the reviewers’ comments/ questions:

Question 1: The main problem is related to statistical analysis about the relationship of SIF and other environmental variables. In particular, author have assumed a linear relationship for all explored correlations. Moreover, many sentences about findings are not supported by evidence or by hypothesis tests.

Reply: We are so sorry about this question and also thank you so much. This time we recalculated all the relationships between SIF and each variable and selected the best fit model to acquire the R2 value. Based on your advice we also deleted or revised some sentences that are not support by some statistic results.

Question 2: L 138. Please add in the caption the cartographical reference system adopted. I suppose WGS84.

Reply: Thanks for your comment and this time we added the reference system as “Reference frame: GCS_WGS_84, EPSG: 4326”.

Question 3: L 200- you should discuss these results in respect to recent literature. Especially the Vegetation index peak of phenological season was found different between several vegetated ecosystem in: Sarvia, F., De Petris, S., & Borgogno-Mondino, E. (2021). Exploring climate change effects on vegetation phenology by MOD13Q1 data: The Piemonte region case study in the period 2001–2019. Agronomy, 11(3), 555.

Reply: Thank you so much for this advice and also thanks for this useful reference paper. We discussed these results based on this reference as below:

Sarvia, et al. [31] analyzed how climate change affects vegetation phenological (derived from maximum annual NDVI) in different ecosystems and found that forest and agriculture show different trend of phenological.

Question 4:  L220-223. This is your hypothesis. You don’t have any evidence to support this sentence. Note that ordinarily corn (the main crop type in your study area) has high LAI values due to its higher parenchyma density per leaf. Moreover, since it is a C4 plant, therefore it has a better carbon fixing rate than other vegetation ecosystems in your study area. Therefore, we cannot be sure that SIF increasing is affected by human (fertilization) rather than other natural factors.

Reply: This is quite a good question and this time we deleted this sentence.

Question 5:  L 227. I think that MODIS- derived data are characterized by a high correlation therefore you should discuss why you selected that variable if one of them could summarized the other ones.

Reply: Yes, you are right that MODIS-derived indices have high correlation. In this research, we just want to know which are the most suitable indices that could be used to model SIF. In fact, the relationships between SIF and many other MODIS-derived indices, such as NDVI, GPP and NPP were also calculated, but no significant correlation were found. So, in this study, we just selected EVI, FPAR and LAI.

This time we added some sentence in section 2.3 as below:

“There are closed relationships between MODIS derived indices, but these indices reflect plant photosynthesis at different aspects and for different aims, different MODIS indices were used to model SIF. MODIS GPP was not considered in this study because of no significant relationship between MODIS GPP and OCO-2 SIF in forest.”

Question 6:  Note that in this study you have adopted the determination coefficient. The latter doesn't tell you about the relationship between variables, but the percentage of variance explained by your regressive model (in this study you suppose linear). Therefore, it is not clear how did you obtain p-value. Which test did you performed? I suppose that you tested the significance of Pearson correlation coefficient (r), but it can be referred to R². Moreover, why did you suppose linear the regression model? why not a power model or exponential? this is a critical issue. You have to prove that the linear model is the best fitting one that could somehow describe the relationship between SIF and other variables.

Reply: Yes, in this manuscript, we adopted liner regression and F-test were also adopted and obtained the P values.

Based on your advice, this time we tried many models and selected the best fitting one for each scatter plot. We found that in some scatter plots, the R2 value increased obviously.

We also added some sentences in section 3.2 as below:

“Liner, exponential, power and quadratic models were tried for each scatter plot and then selected the best fit one. F-test was assessed to test significance of the regression models.”

Question 7:  L250-260. Using the R² of a linear model could hide the real fitting. Maybe using other model (e.g., exponential, power, quadratic) these results could be totally and unexpectedly different.

Reply: Thanks for your advice and this time we tried many models and selected the most fitted one and then calculated the R² values. From table 2 we see that some R2 values increased.

Question 8:  L319-320. This sentence is not supported by rigors statistical analysis. In fact, from table 3 RMSE and Slope values are quite similar. Therefore, I wonder if there are significant differences. A statistical hypothesis test is required to prove the differences between OCO-2 observed SIF and modelled values and vegetation ecosystems.

Reply: Thanks for your comment and this time based on you and another reviewer’s advice, we replaced table 3 as scatter plots. We also deleted the sentence of “RMSE values calculated at the regional scale were lower than at the footprint scale; however, the calculated slope value was higher than in the footprint scale.” based on your advice.

Round 2

Reviewer 1 Report

I have no major concerns. some minor suggestions:

The introduction section needs to be edited to better reflect the motivation of the study. 

what are the negative values in Figure 3? please correct this. 

Figure 4. use another color for the regression line. 

Author Response

Dear Reviewer 1,

We must thank you and other reviewers again for the critical feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.

Based on the comments we received, careful modifications have been made to the manuscript. All changes were marked in red text.

We hope the new manuscript will meet your magazine’s standard. Below you will find our point-by-point responses to the reviewers’ comments/ questions:

Question 1: The introduction section needs to be edited to better reflect the motivation of the study. 

Reply: This time we revised the introduction and all the manuscript text according to you and another reviewer’s comment.

Question 2: What are the negative values in Figure 3? please correct this. 

Reply: This is quite a good question. Yes, OCO-2 SIF contain minus value (but the number is limit) because it is not a true measured value but an inversion result. Just as in many other papers, SIF contain minus value in this research. Although the minus SIF value is meaningless, but we think that delete the minus values will increasing the trend of monthly mean value.

Question 3: Figure 4. use another color for the regression line. 

Reply: Thanks for your advice and we revised the regression line to red color.

Reviewer 2 Report

I reviewed the earlier version of the manuscript. The revised version has been improved after the revision. The authors addressed appropriately my major comments.  However, there are still two important issues in the revised manuscript which I should mention:

1. In my last review (major issue #2), I suggested using de-seasonalized time series. The authors accepted my suggestion and then used “value minus the mean” (see authors’ reply #2) to deal with it. But here, “multi-year annual mean value” is used incorrectly to represent “mean value” by the authors. This is not right. A correct method is to use “multi-year mean for a given period”, it can be expressed using the following equation:

Temperature departure (i) = temperature (i) – multi-year temperature mean (i)

where i is a given period (16-day period in this case).

Also, SIF needs to be de-seasonalized accordingly. The “departure from mean” is a simple way for de-seasonalization. Other more complicated methods can be used also.

Another option for the authors is elimination of entire Section 3.4 “relationships between SIF and meteorological data” and all relevant contents in the paper. Because the method used in the current version is flowed, elimination of the section does not affect the major value of the study.

2. Nonlinear models are not necessary for the comparison between GOSIF and OCO-2 SIF (Figure 8). Because the two sets of SIF have the same data scale and unit, using linear relationship is good enough to examining the agreement or disagreement between the two.

Minor issues:

L23: Add a comma after “cropland”.

L42-44: May cite a few publications about the topic.

L108: Add a comma after “grassland”.

L114: Add “resolution” after “30m”.

L156 “Closed” should be “close”.

L156-158: Break up the sentence “There are … model SIF” into two sentences; Or add “;” after “aspects”.

L228: Add a comma after “cropland”.

L232: What caused the data missing? May specify.

L240: May delete “and also because”.

Fig 3: On the y-axis, change (0.50) to -0.50 ?

Fig 8: Caption of the figure is missing.

Author Response

Dear Reviewer 2,

We must thank you and other reviewers again for the critical feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.

Based on the comments we received, careful modifications have been made to the manuscript. All changes were marked in red text.

We hope the new manuscript will meet your magazine’s standard. Below you will find our point-by-point responses to the reviewers’ comments/ questions:

Question 1. In my last review (major issue #2), I suggested using de-seasonalized time series. The authors accepted my suggestion and then used “value minus the mean” (see authors’ reply #2) to deal with it. But here, “multi-year annual mean value” is used incorrectly to represent “mean value” by the authors. This is not right. A correct method is to use “multi-year mean for a given period”, it can be expressed using the following equation:

Temperature departure (i) = temperature (i) – multi-year temperature mean (i)

where i is a given period (16-day period in this case).

Also, SIF needs to be de-seasonalized accordingly. The “departure from mean” is a simple way for de-seasonalization. Other more complicated methods can be used also.

Another option for the authors is elimination of entire Section 3.4 “relationships between SIF and meteorological data” and all relevant contents in the paper. Because the method used in the current version is flowed, elimination of the section does not affect the major value of the study.

Reply: Thanks for this question again and this time we revised figure 6 and figure 7 according to your advice for both meteorological data and SIF.

We think that section 3.4 is quite important because firstly, many SIF products (such as GOSIF) was modeled using MODIS products and meteorological data and secondly, meteorological factors have directly relation with plant photosynthesis and can be used to explain the mechanism of SIF changes.

Question 2. Nonlinear models are not necessary for the comparison between GOSIF and OCO-2 SIF (Figure 8). Because the two sets of SIF have the same data scale and unit, using linear relationship is good enough to examining the agreement or disagreement between the two.

Reply: Thanks for this question and this time we revised to liner regression.

Minor issues:

Question 3. L23: Add a comma after “cropland”.

Reply: OK. We added a comma after cropland this time.

Question 4. L42-44: May cite a few publications about the topic.

Reply: Thanks for your question and this time we added two references.

Question 5. L108: Add a comma after “grassland”.

Reply: OK. We added a comma after grassland.

Question 6. L114: Add “resolution” after “30m”.

Reply: This time we revised “30m” to “30m resolution”.

Question 7. L156 “Closed” should be “close”.

Reply: We revised “closed” to “close”.

Question 8. L156-158: Break up the sentence “There are … model SIF” into two sentences; Or add “;” after “aspects”.

Reply: This time we add “;” after “aspects”.

Question 9. L228: Add a comma after “cropland”.

Reply: Based on your comment, we add a comma after “cropland”.

Question 10. L232: What caused the data missing? May specify.

Reply: The data missing is because of OCO-2 orbit apart from the study area.

Question 11. L240: May delete “and also because”.

Reply: Thanks for your comment about this and this time we deleted “and also because”.

Question 12. Fig 3: On the y-axis, change (0.50) to -0.50?

Reply: OK. This time we revised “(0.50)” to “-0.5”.

Question 13. Fig 8: Caption of the figure is missing.

Reply: OK. This time we added the caption for figure 8.

Reviewer 3 Report

Authors have improved their manuscript according to the majority of my suggestions. Therefore, i think that now the paper is ready for publication. 

Author Response

Dear Reviewer 3,

We must thank you and other reviewers again for the critical and careful feedback. We feel lucky that our manuscript went to you as the valuable comments from you not only helped us with the improvement of our manuscript, but suggested some useful ideas for future studies.