the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Predicting the productivity of Alpine grasslands using remote sensing information
Abstract. Gross primary productivity (GPP) is a crucial variable for ecosystem dynamics, and it can significantly vary on the small spatial scales of vegetation and environmental heterogeneity. This is especially true for mountain ecosystems, which pose severe difficulties to field monitoring. In addition, the specificity of such ecosystems and the extreme abiotic conditions that they experience often make global and regional models unsuited to predictions. In this case, remote sensing products offer the opportunity to explore the productivity of vegetation communities in remote areas such as Alpine grasslands all year round, and empirical models can help in the challenge of modelling Alpine GPP. Along these lines, we took a hybrid approach, blending several remote sensing data sources (such as a high-definition digital terrain model and moderate- and high- resolution satellite products such as MODIS and Sentinel 2) and gridded datasets such as ERA5 with in situ measurements to implement a specific empirical model. The resulting remote-sensing-based model developed here was suited to represent the measured primary productivity in different areas within a high-altitude grassland at the Nivolet plain, in the north-western Italian Alps at 2700–2500 m amsl. A cross-validation approach allowed us to evaluate to what extent a single empirical model could represent diverse communities and different abiotic factors found in these areas. We finally identified the ratio between MCARI2 and MSAVI2 as a good predictor of light use efficiency, a key factor in the empirical model, probably due to its good correlation with the leaves phenological status, inasmuch it estimates the ratio between chlorophyll and the ensemble of leaf pigments.
- Preprint
(2274 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-2824', Anonymous Referee #1, 04 Mar 2024
General comments
The paper titled “Predicting the productivity of Alpine grasslands using remote sensing information” evaluates different approaches to estimate gross primary production (GPP) of Alpine grasslands using remote sensing data and a Light Use Efficiency (LUE) model. Several remote sensing and climate reanalysis datasets are used to empirically estimate the LUE model input variables and parameters, and in situ GPP data from 5 different plots are used as a reference ground dataset to assess model performances.
The topic is interesting and scientifically relevant, given the ecological complexity of Alpine grasslands and their vulnerability to global environmental changes. However, in its current version, the manuscript has major pitfalls. Both the abstract and the introduction fail in describing clearly the rationale behind the study, the novelty/relevance of the proposed approach with respect to published literature, and the general/specific objectives of the study. Several sentences are generic and often not supported by appropriate citations. The methodology is largely empirical, which can be okay, but would require a more detailed explanation of the reasons behind choosing a specific proxy rather than another to estimate the LUE input variables and parameters. In addition, the field dataset is not described in details and it is difficult to understand how it could be considered representative of highly heterogeneous Alpine grasslands. Finally, discussions are very limited and do not explain clearly the relevance of the results and the limitations of the study (e.g., how the challenge of the different spatial scale between field data and satellite data is addressed).
Specific comments:
l.4: what kind of global and regional models and prediction are the Authors referring to?
l.9: what do the Authors mean by ‘specific empirical model’?
l.11: 2500-2700
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specific
L.23: average values of productivity?
l.28: in this study empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.
l.36: what do the Authors mean by “the system”? Can you be more specific?
l.36-39: the sentence is unclear
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
l.52: please add references about the use of PRI for LUE estimation
l.54: spectral feature?
l.61-63: sentence is unclear. I would not describe SIF in this way.
l.65: RS drivers?
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
l.117: what is the rationale behind choosing these indices?
l.118: can the Author provide a short explanation of this application?
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
Table 1. The resolution of the different products is pretty different and, overall, it is coars,e especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
Figure 4. It should be ‘positively’ and ‘negatively’
l.215: what the Authors mean by ‘extra correlations’?
l.239: also the model base on in situ data does not reach 80-90% if I understand correctly
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).
Citation: https://doi.org/10.5194/egusphere-2023-2824-RC1 -
AC2: 'Reply on RC1', saverio vicario, 06 May 2024
l.4: what kind of global and regional models and prediction are the Authors referring to?
R:As explained in the introduction regional and global model refer to climate model that includes vegetation
l.9: what do the Authors mean by ‘specific empirical model’?R: Is the model of GPP production implemented in the article. Is empirical given that parameters are estimated from the data, and specific to the type of vegetation observed in the case study. Applying the cross validation ensure that albeit specific the model is not overfitted.
l.11: 2500-2700R:Done
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specificR: we added refereces in the introduction
L.23: average values of productivity?
R: describing model with multichilometric scale we point out that their result are a spatial average of different type of vegetationl.28: in this study empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
R: we clearly stated that that the general shape of the model would be fixed but how the macro variable are estimate do change
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
R: we added references
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.R: we removed
l.36: what do the Authors mean by “the system”? Can you be more specific?l.36-39: the sentence is unclear
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
l.52: please add references about the use of PRI for LUE estimation
R: done
l.54: spectral feature?
l.61-63: sentence is unclear. I would not describe SIF in this way.
l.65: RS drivers?
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
R: stated in the text, just after L1 statement we used sen2core to produce the level 2. We needed to run in house the script to match out snow filter ( Richiardi 2021), that is more able than standard SCL from sen2core to distinguish snow from cloud.
l.117: what is the rationale behind choosing these indices?
l.118: can the Author provide a short explanation of this application?
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
Table 1. The resolution of the different products is pretty different and, overall, it is coars,e especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
Figure 4. It should be ‘positively’ and ‘negatively’
R: corrected
l.215: what the Authors mean by ‘extra correlations’?
R: we changed in "additional significant correlations"
l.239: also the model base on in situ data does not reach 80-90% if I understand correctlyR: correct
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).R: we enlarged the discussion
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC2 -
AC4: 'Reply on RC1 second version', saverio vicario, 14 May 2024
l.4: what kind of global and regional models and prediction are the Authors referring to?
R: We thank the reviewers for giving the opportunity to clarify this and the following parts of pur manucript. As explained in the introduction, regional and global models refer to climate models that includes vegetation
l.9: what do the Authors mean by ‘specific empirical model’?R: We refer the model of GPP production implemented in the article. We named it “empirical” as it is given that parameters are estimated from the data, and “specific” to the type of vegetation cover observed in the case study. Applying the cross validation ensures that, albeit specific, the model is not overfitted.
l.11: 2500-2700R: Thanks for spotting this; the correction has been reported in the manucript
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specificR: Thanks for highlighting the need to be more accurate. We have added refereces in the introduction
L.23: average values of productivity?
R: When describing models at multi-kilometric scale, we point out that their result are a spatial average of different types of vegetationl.28: in this study, empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
R: We clearly stated that that the general shape of the model would be fixed but how the macro variable are estimate do change
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
R: Thanks for noting this; we added the correct references
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.R: We removed the sentence.
l.36: what do the Authors mean by “the system”? Can you be more specific?R: We amended the sentence to make it clear.
l.36-39: the sentence is unclear
R: We amended the sentence to make it clear.
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
R: We amended the sentence to make it clear.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
R: Thanks for noting this; we added som new references
l.52: please add references about the use of PRI for LUE estimation
R: new references have been added
l.54: spectral feature?
R: I could not find the “spectral feature”
l.61-63: sentence is unclear. I would not describe SIF in this way.
R: Thanks for highlighting that the sentence was not clear enough. We moved the discussion about SIF in the “discussion” section, given that the use of SIF was an approach not used in the manuscript.
l.65: RS drivers?
R: we rephrased the end of the introduction, and clarified that we use Remote Sensing data to estimate variable parameters which are drivers of photosynthesis, as the incident radiation, and the availability of chlorophyll and water in the leaves.
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
R: We rephrased introduction clarifying the experimental set up
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
R: We reported maximum snow depth to give the sense of the amount of snow that could hit the region. We didn’t include information about the snow cover duration because in this paper we focused on the 4 plots of interest during the summertime, and a more detailed comment about the distribution of the snow cover duration across time and space within the region would have deserved more space.
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
R: we added a more detailed description of the 4 sites, including their flora and geology.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
R: yes indeed, but here we are following its remote sensing application.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
R: As stated in the text, just after L1 statement we used sen2core to produce the level 2 image. We needed to run the script in house to match out snow filter (Richiardi 2021), that is more able to distinguish snow from cloud than standard SCL from sen2core.
l.117: what is the rationale behind choosing these indices?
MSAVI and FAPAR from sen2core are indices for overall pigments in the leaves. MSAVI uses a polynomial formation, while sen2core uses an AI approach. Both have the goal to overcome the high noise of normalized division as NDVI. We chose MCARI 2 because of the similar polynomial shape to MCARI. As stated below, we clarified our choice in the introduction
l.118: can the Author provide a short explanation of this application?
R: we rephrased the introduction in order to clarify our choice
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
R: We added some details and changed from below to bottom
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
R: the local DTMs have very different resolutions. In the Piedmont side we have 10 meters resolution, while the Aosta side has a 2m resolution. We chose an intermediate resolution
Table 1. The resolution of the different products is pretty different and, overall, it is coarse, especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
R: Only the FAPAR variable is correctly covered by Sentinel2. For the meteo variables, we extensively evaluated the difference between local measures and remote estimator. Only PRI index for MODIS was not evaluated, given the lack of an in situ measure, but albeit its coarse spatial resolution it was one of the selected predictors in the final model. PERCHE’? qual’è l’importanza di tenere questo predictor?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
R: Thanks for noticing the need to clarify. We added the phrase “For several variables we were able to access to in situ measures performed at the same time of local estimation of GPP, so we implemented also a model with in situ measures for some variables listed in table \ref{tab:Variables}, in order to observe the limitation of remotely sensed parameters.”
Figure 4. It should be ‘positively’ and ‘negatively’
R: Thanks for noticing, we corrected the sentence
l.215: what the Authors mean by ‘extra correlations’?
R: Thank for noticing that the sentence was unclear. We changed in "additional significant correlations"
l.239: also the model base on in situ data does not reach 80-90% if I understand correctlyR: yes, this is correct
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).R: Thanks for suggesting to provide further insights to our results. We thereby enlarged the discussion.
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC4
-
AC2: 'Reply on RC1', saverio vicario, 06 May 2024
-
RC2: 'Comment on egusphere-2023-2824', Anonymous Referee #2, 23 Mar 2024
Review «Predicting the productivity of Alpine grasslands using remote
sensing information»
Gerneral commets
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
- While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
- Also, the aims of the study are not well defined.
- The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
- The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.
Please find my specific comments below.
Specific comments
Abstract
- Please add information on how you measured GPP in your study.
Introduction
- Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.
- Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
- Also, a specification of the aims or research questions of the study is missing.
Materials and Methods
- I suggest splitting the description of the study site and “in-situ” data
- I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
- Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
- Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
- Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
- You switch between active and passive expressions. Please homogenize.
- Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.
- Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
- Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
- Figure 1: Text is blurry.
- Table 1: The first ¦ seems to be between two lines. Which line is referred to?
- Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
- And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
Results
- Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
- What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
- Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
- Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?
- Table 4: In the line of R2: are these the values without cross-validation?
- Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
- Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
- The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
- Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
- Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
- As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
- Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
- Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
- Figure 6: Some text in the caption is blue … 😉
Discussion and conclusion
- This is neither a sufficient discussion nor a conclusion.
- Please put your results in the context of (more) other published works.
- Reflect on your methodology, what was good, where are shortcomings.
- How and to which extend are the research questions (that where not explicitly defined) answered?
- Write up a proper conclusion on the main findings of your research.
Appendix:
Caption of figure a6 is confusing. Also the figures lack a label on the y axis.
Citation: https://doi.org/10.5194/egusphere-2023-2824-RC2 -
AC1: 'Reply on RC2', saverio vicario, 06 May 2024
Gerneral commets
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
Also, the aims of the study are not well defined.
The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.
Please find my specific comments below.
Specific comments
Abstract
Please add information on how you measured GPP in your study.
Introduction
Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.
Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
Also, a specification of the aims or research questions of the study is missing.
R: we reworked the introduction
Materials and Methods
I suggest splitting the description of the study site and “in-situ” data
I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
R: we rewrote the introduction following this advice
Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
R: We removed the sentence.
Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
R: as stated below we run locally sen2core algorithm to obtain level 2 data. This also was relevant to implement our in house snow filter
You switch between active and passive expressions. Please homogenize.
Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.R: the details of the snow cover pipiline is given in Richiardi 2021. In that analysis pipeline care is given to distiguis
Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
R: yes we moved at the end of the sentences the nivolet name connected to study site.
Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
R: we add reference to section 2.6
Figure 1: Text is blurry.
changed fron png to pdf to get crisp text
Table 1: The first ¦ seems to be between two lines. Which line is referred to?
R: we replaced "|" with "OR" and "+" with "AND" both in bold, so that is clear that they do apply to all line of the section of the table were they appear.
Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
R: we added the size of the different crossvalidation used in the table 2
And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
R: we cited the problem of the limited sample size as the main reason to stick to linear model.
Results
Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
R: we removed figure 3, while figure 2 is relevant to show the were about of the sites. We replaced the old figure 5, now called figure 4 ( due elimination of figure 3) with an expected vs observed GPP graph with our best model from remote data.
Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?
R: in section 2.5 we stated that we used interpolator to obtain measure fitting the date of in situ measure
Table 4: In the line of R2: are these the values without cross-validation?
R: yes indeed, we changed legend to explicit the fact.
Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
R: no we didn't. the article of Metz has a time series of in situ measure, so probably it could appreciate even small change in correlation, while in our case the VPd was estimated from ERA5 data. Further in our case vpd had a strong informative power: alone it has 5% of variance explained and adding incident radiation and fapar estimator the vpd add 7% of variance explained to the model.
Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
R: sorry for the confusion. In the 73 observations data set we have both in situ and remote (ERA5) meteo data. In figure 4 we use this subset to compare the correlation among variable when measured in situ and estimated from remote ( with the model ERA5). So on the left the meteo variable are from in situ while on the left they are from ERa5. To make the figure more clear, we moved all upper label near the sign Remote on the bottom rigth. In this manner title and label of the left side are togheter as those of the rigth.
The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
R: in the table 4 already the first line state the source of predictors. We updated that line to be more explicit on all sources.
Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
R: as already stated with replaced this figure with an expected versus observed GPP. Indeed this figure was relevant in a discussion that was removed at last minute of the submission.
As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
R:sorry, we rewrote hopefully in a more clear structure
Figure 6: Some text in the caption is blue … 😉
R: sorry, corrected
Discussion and conclusion
This is neither a sufficient discussion nor a conclusion.
Please put your results in the context of (more) other published works.
Reflect on your methodology, what was good, where are shortcomings.
How and to which extend are the research questions (that where not explicitly defined) answered?
Write up a proper conclusion on the main findings of your research.
R: yes we rewrote the conclusionCitation: https://doi.org/10.5194/egusphere-2023-2824-AC1 -
AC3: 'Reply on RC2', saverio vicario, 14 May 2024
Second version
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
Also, the aims of the study are not well defined.
The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.R: We thank the reviewer for appreciating our work and for the encuragment to improve it. We did the suggested amendment, as explained in the replies to the specific requests below.
Please find my specific comments below.
Specific comments
Abstract
Please add information on how you measured GPP in your study.
R: starting at line 150 within paragraph 2.2 “in situ data”, we already detailed the accumulation chamber method. We added for clarity a reference in the abstract
Introduction
Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.R: We thank the reviewer for spotting the unclear sentence.
We changed from “Both sources are (complicated) fractions of the top-of-atmosphere radiation, while this latter can be easily calculated from astronomical and geographical information”
To “Both sources are fractions, that depends from the interaction between sun position and details in the topography, of the top-of-atmosphere radiation. This latter variable on the contrary can be easily calculated from astronomical and geographical information.”
Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
Also, a specification of the aims or research questions of the study is missing.
R: We thank the reviewer for giving us the opportunity to clarify the introduction - we substantialy changed it
Materials and Methods
I suggest splitting the description of the study site and “in-situ” data
I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
R: we amended the introduction following all the three suggestions above
Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
R: We removed the sentence.
Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
R: as stated below, we run locally the sen2core algorithm ourselves to obtain level 2 data. This also was relevant to implement our “in house” snow filter
You switch between active and passive expressions. Please homogenize.R: we reworded introduction and conclusion making the style homogenous
Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.R: We thank the reviewer for giving us the opportunity to clarify. The filtering method is explained in detail in the reference “Richiardi 2021”. The filter is able to distinguish clouds from snow based on addition of swir from Sentinel2 and DTM information to fmask and SCL evalution. As briefly described, we marked “data not available” for pixels covered with clouds, as we are not able to infere the reflectance values, while the FAPAR is zero under the snow, as there is no photosynthesis.
Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
R: We thank the reviewer for this suggestion: we moved the name “nivolet “, connected to study site, at the end of the sentences.
Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
R: we added new references to section 2.6
Figure 1: Text is blurry.
Thanks for noticing it! We changed the image from png to pdf to get crisp text
Table 1: The first ¦ seems to be between two lines. Which line is referred to?
R: thanks for noticing it! We replaced "|" with "OR" and "+" with "AND", both in bold, so that it is clear that they do apply to all the line of the section of the table were they appear.
Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
R: we added the size of the different crossvalidation used in the table 2
And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
R: Thanks for giving us the opportunity to clarify. We cited the problem of the limited sample size as the main reason to stick to a linear model.
Results
Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
R: We removed figure 3, while we deem that figure 2 is relevant to show the were about of the sites. We replaced the old figure 5, now called figure 4 ( due elimination of figure 3) with an “fit between modelled and observed GPP values” graph with our best model from remote data.
Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?NOTA: la risposta non è chiara: va bene riscritta cosi’? Ho capito bene?
R: Thanks for giving us the opportunity to clarify this. In section 2.5, we stated that we used an interpolator to obtain thevalues, by fitting the measures obtained in the specific dates of the field campaigns
Table 4: In the line of R2: are these the values without cross-validation?
R: yes indeed, we changed legend to make this explicit.
Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
R: no we didn't. the article of Metz has a time series of in situ measure, so probably it could appreciate even small change in correlation, while in our case the VPD was estimated from ERA5 data. Further in our case vpd had a strong informative power: alone it has 5% of variance explained and adding incident radiation and FAPAR estimator the VPD adds 7% of variance explained to the model.
Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
R: sorry for the confusion. In the 73 observations data set, we have both in situ and remote (ERA5) meteo data. In figure 4, we use this subset to compare the correlation among variables measured in situ and estimated from remote (with the model ERA5). So, on the right the meteo variable are from in situ data, while on the left they are from ERA5. To make the figure clearer, we moved all upper label near the sign “Remote” on the bottom right. In this manner, title and label of the right side are togheter, similarly to those on the left side.
The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
R: in table 4, we list best models using meteo variables with in situ or with ERA5 estimates. In both case FAPAR is estimated from Sentinel 2 data. So no model is entirely informed with local measure given that no spectral reading of red and nir band was performed in situ.
Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
R: In table 4, the first line state the source of predictors. We updated that line to be more explicit on all sources.
Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
R: Thanks for this comment that gave us the opportunity to improve the clarity of the manuscript. As stated aove, we replaced this figure with one reporting modelled versus observed GPP. Indeed, this figure was relevant in a discussion that was removed in the final version of the manuscript.
As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
R: thanks for the note, we revised the legend try to balance readability and brevity.
Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
R: Thanks for your feedback on the clarity of the reading. This gave us the opportunity to re-shape this part in a more clear structure
Figure 6: Some text in the caption is blue … 😉
R: sorry, corrected
Discussion and conclusion
This is neither a sufficient discussion nor a conclusion.
Please put your results in the context of (more) other published works.
Reflect on your methodology, what was good, where are shortcomings.
How and to which extend are the research questions (that where not explicitly defined) answered?
Write up a proper conclusion on the main findings of your research.R: Thanks for your feedback on the discussion and the conclusions. This gave us the opportunity to make a deep review and change substantially the discussion and the conclusions.
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC3
Status: closed
-
RC1: 'Comment on egusphere-2023-2824', Anonymous Referee #1, 04 Mar 2024
General comments
The paper titled “Predicting the productivity of Alpine grasslands using remote sensing information” evaluates different approaches to estimate gross primary production (GPP) of Alpine grasslands using remote sensing data and a Light Use Efficiency (LUE) model. Several remote sensing and climate reanalysis datasets are used to empirically estimate the LUE model input variables and parameters, and in situ GPP data from 5 different plots are used as a reference ground dataset to assess model performances.
The topic is interesting and scientifically relevant, given the ecological complexity of Alpine grasslands and their vulnerability to global environmental changes. However, in its current version, the manuscript has major pitfalls. Both the abstract and the introduction fail in describing clearly the rationale behind the study, the novelty/relevance of the proposed approach with respect to published literature, and the general/specific objectives of the study. Several sentences are generic and often not supported by appropriate citations. The methodology is largely empirical, which can be okay, but would require a more detailed explanation of the reasons behind choosing a specific proxy rather than another to estimate the LUE input variables and parameters. In addition, the field dataset is not described in details and it is difficult to understand how it could be considered representative of highly heterogeneous Alpine grasslands. Finally, discussions are very limited and do not explain clearly the relevance of the results and the limitations of the study (e.g., how the challenge of the different spatial scale between field data and satellite data is addressed).
Specific comments:
l.4: what kind of global and regional models and prediction are the Authors referring to?
l.9: what do the Authors mean by ‘specific empirical model’?
l.11: 2500-2700
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specific
L.23: average values of productivity?
l.28: in this study empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.
l.36: what do the Authors mean by “the system”? Can you be more specific?
l.36-39: the sentence is unclear
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
l.52: please add references about the use of PRI for LUE estimation
l.54: spectral feature?
l.61-63: sentence is unclear. I would not describe SIF in this way.
l.65: RS drivers?
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
l.117: what is the rationale behind choosing these indices?
l.118: can the Author provide a short explanation of this application?
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
Table 1. The resolution of the different products is pretty different and, overall, it is coars,e especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
Figure 4. It should be ‘positively’ and ‘negatively’
l.215: what the Authors mean by ‘extra correlations’?
l.239: also the model base on in situ data does not reach 80-90% if I understand correctly
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).
Citation: https://doi.org/10.5194/egusphere-2023-2824-RC1 -
AC2: 'Reply on RC1', saverio vicario, 06 May 2024
l.4: what kind of global and regional models and prediction are the Authors referring to?
R:As explained in the introduction regional and global model refer to climate model that includes vegetation
l.9: what do the Authors mean by ‘specific empirical model’?R: Is the model of GPP production implemented in the article. Is empirical given that parameters are estimated from the data, and specific to the type of vegetation observed in the case study. Applying the cross validation ensure that albeit specific the model is not overfitted.
l.11: 2500-2700R:Done
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specificR: we added refereces in the introduction
L.23: average values of productivity?
R: describing model with multichilometric scale we point out that their result are a spatial average of different type of vegetationl.28: in this study empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
R: we clearly stated that that the general shape of the model would be fixed but how the macro variable are estimate do change
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
R: we added references
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.R: we removed
l.36: what do the Authors mean by “the system”? Can you be more specific?l.36-39: the sentence is unclear
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
l.52: please add references about the use of PRI for LUE estimation
R: done
l.54: spectral feature?
l.61-63: sentence is unclear. I would not describe SIF in this way.
l.65: RS drivers?
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
R: stated in the text, just after L1 statement we used sen2core to produce the level 2. We needed to run in house the script to match out snow filter ( Richiardi 2021), that is more able than standard SCL from sen2core to distinguish snow from cloud.
l.117: what is the rationale behind choosing these indices?
l.118: can the Author provide a short explanation of this application?
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
Table 1. The resolution of the different products is pretty different and, overall, it is coars,e especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
Figure 4. It should be ‘positively’ and ‘negatively’
R: corrected
l.215: what the Authors mean by ‘extra correlations’?
R: we changed in "additional significant correlations"
l.239: also the model base on in situ data does not reach 80-90% if I understand correctlyR: correct
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).R: we enlarged the discussion
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC2 -
AC4: 'Reply on RC1 second version', saverio vicario, 14 May 2024
l.4: what kind of global and regional models and prediction are the Authors referring to?
R: We thank the reviewers for giving the opportunity to clarify this and the following parts of pur manucript. As explained in the introduction, regional and global models refer to climate models that includes vegetation
l.9: what do the Authors mean by ‘specific empirical model’?R: We refer the model of GPP production implemented in the article. We named it “empirical” as it is given that parameters are estimated from the data, and “specific” to the type of vegetation cover observed in the case study. Applying the cross validation ensures that, albeit specific, the model is not overfitted.
l.11: 2500-2700R: Thanks for spotting this; the correction has been reported in the manucript
l.20: what do the Authors mean by “many types of studies”. I would suggest to be more specificR: Thanks for highlighting the need to be more accurate. We have added refereces in the introduction
L.23: average values of productivity?
R: When describing models at multi-kilometric scale, we point out that their result are a spatial average of different types of vegetationl.28: in this study, empirical models are used to estimate the input variables and parameters of the LUE model (fAPAR, PAR, ε). As such, it seems incorrect to state that different regressive models are used to estimate GPP.
R: We clearly stated that that the general shape of the model would be fixed but how the macro variable are estimate do change
l.29: Rossini et al. 2012 cannot be used as the main citation for the Monteith model. Please use the original reference from Monteith.
R: Thanks for noting this; we added the correct references
l.34: please remove ‘complicated’ as it seems subjective. In addition, if ‘both sources are complicated’, what is the latter source that can be simply calculated? The sentence is unclear.R: We removed the sentence.
l.36: what do the Authors mean by “the system”? Can you be more specific?R: We amended the sentence to make it clear.
l.36-39: the sentence is unclear
R: We amended the sentence to make it clear.
l.47: another approach. What is the first approach? The sentence at lines 39-46 describes LUE, not a retrieval approach.
R: We amended the sentence to make it clear.
l.49: Can the Authors provide references about the method they describe? My understanding is that potential LUE can be estimated for each PFT and that specific functions can be used to consider the LUE reduction caused by non-optimal conditions.
R: Thanks for noting this; we added som new references
l.52: please add references about the use of PRI for LUE estimation
R: new references have been added
l.54: spectral feature?
R: I could not find the “spectral feature”
l.61-63: sentence is unclear. I would not describe SIF in this way.
R: Thanks for highlighting that the sentence was not clear enough. We moved the discussion about SIF in the “discussion” section, given that the use of SIF was an approach not used in the manuscript.
l.65: RS drivers?
R: we rephrased the end of the introduction, and clarified that we use Remote Sensing data to estimate variable parameters which are drivers of photosynthesis, as the incident radiation, and the availability of chlorophyll and water in the leaves.
l.65-69: the objectives and motivations of the study are unclear and very generic. In addition, what do the Authors mean by stating that they try to balance spectral, temporal and spatial resolution?
R: We rephrased introduction clarifying the experimental set up
l.81: why the Authors report the maximum snow depth for an extreme? Wouldn’t be more relevant to indicate the average snow depth and duration of snow cover?
R: We reported maximum snow depth to give the sense of the amount of snow that could hit the region. We didn’t include information about the snow cover duration because in this paper we focused on the 4 plots of interest during the summertime, and a more detailed comment about the distribution of the snow cover duration across time and space within the region would have deserved more space.
l.97-102: The description of the plots and Figure 2 do not help understanding the ecological representativeness of the study plots. Among others, more information about grassland composition in each plot would be important. Figure 2 is not very informative. It would be valuable to add a general overview of the location of the study area within Aosta Valley and to improve the terrain representation to better capture contours, slope, aspect, shade reliefs, etc.
R: we added a more detailed description of the 4 sites, including their flora and geology.
l.107: The basic formula refers to the Monteith LUE model, not to a remote sensing approach.
R: yes indeed, but here we are following its remote sensing application.
l.116: why level 1? What type of level 1? So atmospheric correction was not performed?
R: As stated in the text, just after L1 statement we used sen2core to produce the level 2 image. We needed to run the script in house to match out snow filter (Richiardi 2021), that is more able to distinguish snow from cloud than standard SCL from sen2core.
l.117: what is the rationale behind choosing these indices?
MSAVI and FAPAR from sen2core are indices for overall pigments in the leaves. MSAVI uses a polynomial formation, while sen2core uses an AI approach. Both have the goal to overcome the high noise of normalized division as NDVI. We chose MCARI 2 because of the similar polynomial shape to MCARI. As stated below, we clarified our choice in the introduction
l.118: can the Author provide a short explanation of this application?
R: we rephrased the introduction in order to clarify our choice
Figure 1. The figure is very unclear and not self-explanatory. Acronyms should be listed in the caption. Symbols should be also explained. BOA generally refers to Bottom of Atmosphere.
R: We added some details and changed from below to bottom
l.142. Can the Authors explain what is the rationale behind using a spatial resolution of 6.6m?
R: the local DTMs have very different resolutions. In the Piedmont side we have 10 meters resolution, while the Aosta side has a 2m resolution. We chose an intermediate resolution
Table 1. The resolution of the different products is pretty different and, overall, it is coarse, especially for mountain areas. How this can be comparable with a plot of 10 meters, considering the high heterogeneity and topographic complexity of the study region?
R: Only the FAPAR variable is correctly covered by Sentinel2. For the meteo variables, we extensively evaluated the difference between local measures and remote estimator. Only PRI index for MODIS was not evaluated, given the lack of an in situ measure, but albeit its coarse spatial resolution it was one of the selected predictors in the final model. PERCHE’? qual’è l’importanza di tenere questo predictor?
l.193: the approach based on in situ data was not mentioned explicitly earlier. Perhaps it would help to add in the “methods” a short explanation of the different approaches that are compared.
R: Thanks for noticing the need to clarify. We added the phrase “For several variables we were able to access to in situ measures performed at the same time of local estimation of GPP, so we implemented also a model with in situ measures for some variables listed in table \ref{tab:Variables}, in order to observe the limitation of remotely sensed parameters.”
Figure 4. It should be ‘positively’ and ‘negatively’
R: Thanks for noticing, we corrected the sentence
l.215: what the Authors mean by ‘extra correlations’?
R: Thank for noticing that the sentence was unclear. We changed in "additional significant correlations"
l.239: also the model base on in situ data does not reach 80-90% if I understand correctlyR: yes, this is correct
l.236-251: the discussion of the results is very limited. There is no comparative analysis with similar studies nor any discussion of the limitations of the methodology proposed (e.g. issues of resolution, issues of subjective selection of variables, issues of limited representativeness of the dataset, etc.).R: Thanks for suggesting to provide further insights to our results. We thereby enlarged the discussion.
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC4
-
AC2: 'Reply on RC1', saverio vicario, 06 May 2024
-
RC2: 'Comment on egusphere-2023-2824', Anonymous Referee #2, 23 Mar 2024
Review «Predicting the productivity of Alpine grasslands using remote
sensing information»
Gerneral commets
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
- While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
- Also, the aims of the study are not well defined.
- The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
- The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.
Please find my specific comments below.
Specific comments
Abstract
- Please add information on how you measured GPP in your study.
Introduction
- Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.
- Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
- Also, a specification of the aims or research questions of the study is missing.
Materials and Methods
- I suggest splitting the description of the study site and “in-situ” data
- I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
- Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
- Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
- Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
- You switch between active and passive expressions. Please homogenize.
- Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.
- Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
- Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
- Figure 1: Text is blurry.
- Table 1: The first ¦ seems to be between two lines. Which line is referred to?
- Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
- And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
Results
- Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
- What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
- Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
- Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?
- Table 4: In the line of R2: are these the values without cross-validation?
- Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
- Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
- The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
- Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
- Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
- As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
- Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
- Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
- Figure 6: Some text in the caption is blue … 😉
Discussion and conclusion
- This is neither a sufficient discussion nor a conclusion.
- Please put your results in the context of (more) other published works.
- Reflect on your methodology, what was good, where are shortcomings.
- How and to which extend are the research questions (that where not explicitly defined) answered?
- Write up a proper conclusion on the main findings of your research.
Appendix:
Caption of figure a6 is confusing. Also the figures lack a label on the y axis.
Citation: https://doi.org/10.5194/egusphere-2023-2824-RC2 -
AC1: 'Reply on RC2', saverio vicario, 06 May 2024
Gerneral commets
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
Also, the aims of the study are not well defined.
The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.
Please find my specific comments below.
Specific comments
Abstract
Please add information on how you measured GPP in your study.
Introduction
Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.
Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
Also, a specification of the aims or research questions of the study is missing.
R: we reworked the introduction
Materials and Methods
I suggest splitting the description of the study site and “in-situ” data
I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
R: we rewrote the introduction following this advice
Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
R: We removed the sentence.
Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
R: as stated below we run locally sen2core algorithm to obtain level 2 data. This also was relevant to implement our in house snow filter
You switch between active and passive expressions. Please homogenize.
Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.R: the details of the snow cover pipiline is given in Richiardi 2021. In that analysis pipeline care is given to distiguis
Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
R: yes we moved at the end of the sentences the nivolet name connected to study site.
Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
R: we add reference to section 2.6
Figure 1: Text is blurry.
changed fron png to pdf to get crisp text
Table 1: The first ¦ seems to be between two lines. Which line is referred to?
R: we replaced "|" with "OR" and "+" with "AND" both in bold, so that is clear that they do apply to all line of the section of the table were they appear.
Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
R: we added the size of the different crossvalidation used in the table 2
And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
R: we cited the problem of the limited sample size as the main reason to stick to linear model.
Results
Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
R: we removed figure 3, while figure 2 is relevant to show the were about of the sites. We replaced the old figure 5, now called figure 4 ( due elimination of figure 3) with an expected vs observed GPP graph with our best model from remote data.
Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?
R: in section 2.5 we stated that we used interpolator to obtain measure fitting the date of in situ measure
Table 4: In the line of R2: are these the values without cross-validation?
R: yes indeed, we changed legend to explicit the fact.
Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
R: no we didn't. the article of Metz has a time series of in situ measure, so probably it could appreciate even small change in correlation, while in our case the VPd was estimated from ERA5 data. Further in our case vpd had a strong informative power: alone it has 5% of variance explained and adding incident radiation and fapar estimator the vpd add 7% of variance explained to the model.
Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
R: sorry for the confusion. In the 73 observations data set we have both in situ and remote (ERA5) meteo data. In figure 4 we use this subset to compare the correlation among variable when measured in situ and estimated from remote ( with the model ERA5). So on the left the meteo variable are from in situ while on the left they are from ERa5. To make the figure more clear, we moved all upper label near the sign Remote on the bottom rigth. In this manner title and label of the left side are togheter as those of the rigth.
The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
R: in the table 4 already the first line state the source of predictors. We updated that line to be more explicit on all sources.
Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
R: as already stated with replaced this figure with an expected versus observed GPP. Indeed this figure was relevant in a discussion that was removed at last minute of the submission.
As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
R:sorry, we rewrote hopefully in a more clear structure
Figure 6: Some text in the caption is blue … 😉
R: sorry, corrected
Discussion and conclusion
This is neither a sufficient discussion nor a conclusion.
Please put your results in the context of (more) other published works.
Reflect on your methodology, what was good, where are shortcomings.
How and to which extend are the research questions (that where not explicitly defined) answered?
Write up a proper conclusion on the main findings of your research.
R: yes we rewrote the conclusionCitation: https://doi.org/10.5194/egusphere-2023-2824-AC1 -
AC3: 'Reply on RC2', saverio vicario, 14 May 2024
Second version
Dear authors,
Thanks a lot for presenting this interesting study to the scientific community. You address a very important topic and the in-situ observations that – I guess – took a lot of effort to collect are very valuable to address the topic of your manuscript.
Still, based on several shortcomings of the presentation of your manuscript, I think that the manuscript is not yet ready for publication.
While the introduction nicely describes the challenges related to estimating GPP from EO data, it lacks insights into the body of literature on monitoring grasslands with remote sensing and in general, studies on estimation of GPP in the alpine.
Also, the aims of the study are not well defined.
The results part is confusing. This might result from the fact that also in the methods it is not very comprehensible which data were used to create which results. As a starting point, maybe an overview figure would help here.
The discussion and conclusion part are very short and miss some important aspects.
Thus, I suggest major revisions of the manuscript to allow the readership to fully comprehend the exciting research the authors undertook.R: We thank the reviewer for appreciating our work and for the encuragment to improve it. We did the suggested amendment, as explained in the replies to the specific requests below.
Please find my specific comments below.
Specific comments
Abstract
Please add information on how you measured GPP in your study.
R: starting at line 150 within paragraph 2.2 “in situ data”, we already detailed the accumulation chamber method. We added for clarity a reference in the abstract
Introduction
Line 34 “while this latter can” typo? I don’t understand what this sentence tries to convey.R: We thank the reviewer for spotting the unclear sentence.
We changed from “Both sources are (complicated) fractions of the top-of-atmosphere radiation, while this latter can be easily calculated from astronomical and geographical information”
To “Both sources are fractions, that depends from the interaction between sun position and details in the topography, of the top-of-atmosphere radiation. This latter variable on the contrary can be easily calculated from astronomical and geographical information.”
Moreover, parts of the introduction already sound like the methods part, e.g. line 60 – 69.
Also, a specification of the aims or research questions of the study is missing.
R: We thank the reviewer for giving us the opportunity to clarify the introduction - we substantialy changed it
Materials and Methods
I suggest splitting the description of the study site and “in-situ” data
I suggest splitting the description of “Remote sensing GPP estimates” into multiple parts.
Line 114: “The introduction lacks insights into the body of literature on monitoring grasslands with remote sensing studies on estimation of GPP in the alpine”. Please cite this literature.
R: we amended the introduction following all the three suggestions above
Line 115: “We disregarded this approach given that light is unlike to be the limiting factor in wild plants”. Please provide evidence or a citation for this statement
R: We removed the sentence.
Line 116: Why did you use the Sentinel2 level 1 product instead of the l2?
R: as stated below, we run locally the sen2core algorithm ourselves to obtain level 2 data. This also was relevant to implement our “in house” snow filter
You switch between active and passive expressions. Please homogenize.R: we reworded introduction and conclusion making the style homogenous
Line 122: “Then, using a cloud filtering pipeline, pixels identified as cloud were marked as "data not available", while snow pixels were marked as "zero", given that fAPAR is null under the snow cover.” Please provide more detail on that since it is a sensitive part of the process.R: We thank the reviewer for giving us the opportunity to clarify. The filtering method is explained in detail in the reference “Richiardi 2021”. The filter is able to distinguish clouds from snow based on addition of swir from Sentinel2 and DTM information to fmask and SCL evalution. As briefly described, we marked “data not available” for pixels covered with clouds, as we are not able to infere the reflectance values, while the FAPAR is zero under the snow, as there is no photosynthesis.
Line 123: “Nivolet time-series features” Including the name Nivolet in the description of the time series is a bit confusing since one might think of it as a special method (if the name of the study site is forgotten. Thus, I suggest removing the name here.
R: We thank the reviewer for this suggestion: we moved the name “nivolet “, connected to study site, at the end of the sentences.
Line 129: “Anci are the predictors of LUE” please mention that these are spelled out below.
R: we added new references to section 2.6
Figure 1: Text is blurry.
Thanks for noticing it! We changed the image from png to pdf to get crisp text
Table 1: The first ¦ seems to be between two lines. Which line is referred to?
R: thanks for noticing it! We replaced "|" with "OR" and "+" with "AND", both in bold, so that it is clear that they do apply to all the line of the section of the table were they appear.
Nice approaches for the cross-validation. Still, could you state how many observations were used for each train/test split with each cross-validation approach, please.
R: we added the size of the different crossvalidation used in the table 2
And more generally: Why did you choose to go for a linear fixed model for your estimation of GPP? Have you tried a more flexible model such as random forest or are the number of observations to limited for such approaches? Please provide your reasoning.
R: Thanks for giving us the opportunity to clarify. We cited the problem of the limited sample size as the main reason to stick to a linear model.
Results
Figure 2 and 3: For me these figures show mainly the input data and not results from your research. I thus suggest moving them to a supplementary material section.
What I would rather like to see are some scatterplots for your results instead of only correlation coefficients, i.e. the data behind table 4.
R: We removed figure 3, while we deem that figure 2 is relevant to show the were about of the sites. We replaced the old figure 5, now called figure 4 ( due elimination of figure 3) with an “fit between modelled and observed GPP values” graph with our best model from remote data.
Along the same lines: I would like to know the spread of results in the cross-validation in addition to the single R2 values (e.g. in table 4, but also generally).
Also, I would like to see some more information on the extracted vegetation indices. You state that “In total, 394 images from the beginning of 2017 … were stacked”. I am missing some information on how this time series data is used. In the presentation of the results, it seems that individual, non continues data was used in the models. Is this correct?NOTA: la risposta non è chiara: va bene riscritta cosi’? Ho capito bene?
R: Thanks for giving us the opportunity to clarify this. In section 2.5, we stated that we used an interpolator to obtain thevalues, by fitting the measures obtained in the specific dates of the field campaigns
Table 4: In the line of R2: are these the values without cross-validation?
R: yes indeed, we changed legend to make this explicit.
Figure 4 and the methodology behind it. In a study by Merz et al. (2022) it was shown that some parameters such as Temperature and VPD had a lag time compared to GPP estimates of a flux tower when looking at the data in high temporal detail. Did you investigate such effects?
R: no we didn't. the article of Metz has a time series of in situ measure, so probably it could appreciate even small change in correlation, while in our case the VPD was estimated from ERA5 data. Further in our case vpd had a strong informative power: alone it has 5% of variance explained and adding incident radiation and FAPAR estimator the VPD adds 7% of variance explained to the model.
Figure 4: I get a bit confused from this figure. In the text you mention somewhere that you also measure meteo variables along with the GPP in your in-situ dataset. Then you say that these are only complete for 73 observations (line 191). Then I read the statement somewhere that ERA5 and in-situ measurements do not correlate to well (line 195 ff). But then in figure 4 you have meteo variables on the remote and in situ site – but as far as I understand these are the once from ERA5, right? For me, this is a confusing way of presenting things. I suggest at least improving the caption of figure 4 to clearly state that all meteo variables are from ERA5 (if this is true).
R: sorry for the confusion. In the 73 observations data set, we have both in situ and remote (ERA5) meteo data. In figure 4, we use this subset to compare the correlation among variables measured in situ and estimated from remote (with the model ERA5). So, on the right the meteo variable are from in situ data, while on the left they are from ERA5. To make the figure clearer, we moved all upper label near the sign “Remote” on the bottom right. In this manner, title and label of the right side are togheter, similarly to those on the left side.
The confusion continues in table 4: Here I wonder if now the meteo variables used in the models are still the ERA5 data or in part the in-situ meteo data is used. This confusion is fuelled by your statement somewhere in the text that you have build models based solely on the in-situ data (at least if I got this right?).
R: in table 4, we list best models using meteo variables with in situ or with ERA5 estimates. In both case FAPAR is estimated from Sentinel 2 data. So no model is entirely informed with local measure given that no spectral reading of red and nir band was performed in situ.
Table 4: Please highlight which of the models use in-situ data and which not. It can be comprehended by switching back and forth between text and table, but it would that a considerable overhead for the reader given that this information is essential to interpret the results.
R: In table 4, the first line state the source of predictors. We updated that line to be more explicit on all sources.
Is Figure 5 referred to in the text? I could not find it. Also, is that plot essential for the message the manuscript wants to convey? If yes, please elaborate. If not, move the figure to the appendix. Moreover, is air temperature measured?
R: Thanks for this comment that gave us the opportunity to improve the clarity of the manuscript. As stated aove, we replaced this figure with one reporting modelled versus observed GPP. Indeed, this figure was relevant in a discussion that was removed in the final version of the manuscript.
As a general comment to the captions of the figures and tables: In my opinion it makes sense to spell out abbreviations in the captions of the figure to easy their interpretation by the reader. Less thinking on such things gives more capacity to focus on the science. But I leave it to the authors to decide (and look up the guidelines of the journal if there are any on this).
R: thanks for the note, we revised the legend try to balance readability and brevity.
Lines 220 -234: I find that part hard to follow. Maybe a figure would help to explain the combinations and which data was used where to generate which results.
Overall, I find the presentation of the results confusion and advice to rewrite / eventually restructure it in a more comprehensible way.
R: Thanks for your feedback on the clarity of the reading. This gave us the opportunity to re-shape this part in a more clear structure
Figure 6: Some text in the caption is blue … 😉
R: sorry, corrected
Discussion and conclusion
This is neither a sufficient discussion nor a conclusion.
Please put your results in the context of (more) other published works.
Reflect on your methodology, what was good, where are shortcomings.
How and to which extend are the research questions (that where not explicitly defined) answered?
Write up a proper conclusion on the main findings of your research.R: Thanks for your feedback on the discussion and the conclusions. This gave us the opportunity to make a deep review and change substantially the discussion and the conclusions.
Citation: https://doi.org/10.5194/egusphere-2023-2824-AC3
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
340 | 101 | 33 | 474 | 14 | 13 |
- HTML: 340
- PDF: 101
- XML: 33
- Total: 474
- BibTeX: 14
- EndNote: 13
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1