Exploring the Impact of Noise on Hybrid Inversion of PROSAIL RTM on Sentinel-2 Data

Round 1

Reviewer 1 Report

I have some comments as below: 

• The conclusion must be enhanced
• Is the codes/module can be shared in the appendix for the users to use like Sen2cor
• Significant findings should be highlighted in the abstract and conclusion 
• How this approach can be applied for other data?

Author Response

Dear Reviewer,

Thank you for your comments. Other reviewers pointed out that the manuscript required an improvement in terms of language, so we requested a colleague to review it and implemented his suggestions. Therefore, we invite you to re-read the entire document again.

Please find below the responses to your comments, point by point. We have uploaded a version where changes resulting from each of the reviewer’s comments are added with different colors. Hopefully, this helps in the review process.

Reviews regarding your comments are kept in purple.

An important change that we have to point out is that we updated the values in Table 3 to have the same number of significant characters. We also noticed that we did some mistakes on the previous table and have now corrected them. These changes do not amount to significant differences in terms of interpretation from our point of view.

• Comment: The conclusion must be enhanced

We re-evaluated our conclusions section and most adjustments to make our message come through more clearly and to enhance certain aspects of future research.

• Comment: Is the codes/module can be shared in the appendix for the users to use like Sen2cor

We will add a link to a github repository that can be publicly accessible. We prefer to share it through github because we find it easier for other software developers use. We have added a repository with all the code and data: https://github.com/nunocesarsa/RTM_Inversion

• Comment: Significant findings should be highlighted in the abstract and conclusion

We re-evaluated our abstract and conclusions sections and adjusted to make our message come through more clearly to enhance certain aspects of future research.

• Comment: How this approach can be applied for other data?

There are multiple ways this approach can be applied to other data. One obvious approach is to repeat the hyperparameter tuning optimization on the new sensor every time. Other more complex, could use a meta-learning approach and store the model structure of the best performing models in Sensor X and use those configurations to warm start the hyperparameter tuning on Sensor Y. This type of approach has been shown to improve both the model performance as well as to reduce the necessary tuning time (e.g. Feurer, 2015).

Author Response File: Author Response.pdf

Reviewer 2 Report

Thanks for this nice and interesting work. Monitoring the plant functional traits by satellite technique is important for multiple ecological processes, but subject to the impacts of noise. This paper explores the possibility of using the S2-like data to retrieve several leaf biochemical and structural traits with four hybrid approaches and analyze the impacts of noise on the inversion. The methodology and results are well presented. However, extensive language revisions are still needed before acceptance. I have listed only some part of these editing errors.

L30, plant functional traits are usually classified into biophysical, anatomical, and biochemical traits, please rewrite this sentence.

L75，use the term of 'simulated S2-like data' instead of 'S2 simulated data'

The caption of figure 1, the second and third steps test

L83, can be S2 potentially detected

L85，use the term of simulated S2-like data throughout the whole paper

Table 1, solar zenith angle, sensor zenith angle, relative azimuth angle

L134，and x, y represent the input and output

L147，remove textit

L191，in this case of RTM inversion

L214,  band at wavelength i

L224， the remaining 90 % of the total data

Author Response

Dear Reviewer,

Thank you for your comments. As some reviewers pointed out that the manuscript required an improvement in terms of language, so we requested a colleague to review it and implemented his suggestions. Therefore, we invite you to re-read the entire document again.

Please find below the responses to your comments, point by point. We have uploaded a version where changes resulting from each of the reviewer’s comments are added with different colors. Hopefully, this helps in the review process.

Reviews regarding your comments are kept in green.

An important change that we have to out is that we updated the values in Table 3 to have the same number of significant characters. We also noticed that we did some mistakes on the previous table and have now corrected them. These changes do not amount to significant differences in terms of interpretation from our point of view.

• Comment: L30, plant functional traits are usually classified into biophysical, anatomical, and biochemical traits, please rewrite this sentence.

This sentence was also commented by another reviewer. Therefore, following both the comments we have changed it to: ‘’While a precise definition is somewhat contentious, functional traits of vegetation are defined as the characteristics of vegetation that contribute to its fitness and performance (Nock, 2016). Often, in remote sensing, biophysical variables or parameters of vegetation are used instead to define the physical (e.g., height, leaf mass area) and biochemical (e.g., leaf nitrogen, leaf chlorophyll) characteristics of vegetation (Asner, 1998, Weiss, 2000).’’

• Comment: L75，use the term of 'simulated S2-like data' instead of 'S2 simulated data'

We adapted this sentence as was suggested. Following the same suggestion, we adapted the Figure 2 flowchart as well.

• Comment: The caption of figure 1, the second and third steps test

The caption now reads: “The first step consists of identifying which biophysical variables can be detected by S2 while the second step uses Bayesian optimization to find the best parameters for the various models. Lastly the third and fourth step test the performance of the various algorithms for inverting simulated S2-like data using Radiative Transfer Models in pure and noisy conditions”

• Comment: L83, can be S2 potentially detected

This sentence now reads: “ (…) which biophysical traits can potentially be detected by S2 based on a global (..)”

• Comment: L85，use the term of simulated S2-like data throughout the whole paper

We have adapted and now use the suggested term throughout the whole manuscript.

• Comment: Table 1, solar zenith angle, sensor zenith angle, relative azimuth angle

We have adapted Table 1 accordingly.

• Comment: L134，and x, y represent the input and output

We have added “represent” to the sentence.

• Comment: L147，remove textit

Thank you for noticing this mistake from our side We have corrected it.

• Comment:  L191，in this case of RTM inversion

Here we adapted a bit differently, it now reads: ‘’(..) for the task of RTM inversion’’.

• Comment: L224， the remaining 90 % of the total data

The sentence now reads: ’the remaining 90% of the total data’’

Author Response File: Author Response.pdf

Reviewer 3 Report

The study describes an attempt of employing several remotely sensed models to obtained from Sentinel-2 satellite data useful to monitor biophysical plant traits. Moreover other specific objectives of this research were on identifying which biophysical traits can be measured by S-2, identifying the best machine learning algorithms to invert S-2 simulated data and explore the effect  of noise in the models.

Overall, the paper is well written and touches a topical subject in the context of remote sensing applied to plant traits detection. However, there are few issues to be addressed before the manuscript can be suitable for publication. Please following my comments and suggestions in the attached file.

Moreover, I would suggest to submit this paper on specific topics of ecosystem monitoring, given that the article is not yet assigned to a specific journal (e.g. on this special issue “Special Issue "New Insights into Ecosystem Monitoring Using Geospatial Techniques" Remote Sensing | Special Issue : New Insights into Ecosystem Monitoring Using Geospatial Techniques - mdpi.com), only having integrated the text as suggested in the last comment of the pdf file (see conclusion) and a more suitable title in the field of application of EO on environmental science.

My review response of this paper is accepted after minor revision.

All the Best!

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you for your comments. As some reviewers pointed out that the manuscript required an improvement in terms of language, so we requested a colleague to review it and implemented his suggestions. Therefore, we invite you to re-read the entire document again.

Please find below the responses to your comments, point by point. I extracted them from the pdf and added them here to facilitate our communication.

Furthermore, we have uploaded a version where changes resulting from each of the reviewer’s comments are added with different colours. Hopefully, this helps in the review process.

Reviews regarding your comments are kept in blue.

An important change that we have point out is that we updated the values in Table 3 to have the same number of significant characters. We also noticed that we did some mistakes on the previous table and have now corrected them. These changes do not amount to significant differences in terms of interpretation from our point of view.

• Comment: Page 1/ L30-31 & L31:
  • L30-31: Please add here more reference about the definitions you mentioned.
  • L31: characteristics of plant no vegetation, please reconsider it after have read carefully vegetation concept in ecology

We believe these two comments are related to each other, so our changes aim to respond to both. This section was updated according to your feedback but also from other reviewers. It now reads:

“While a precise definition is somewhat contentious, functional traits of vegetation are defined as the characteristics of vegetation that contribute to its fitness and performance (Nock,2016). Often, in remote sensing, biophysical variables or parameters of vegetation are used instead to define the physical (e.g., height, leaf mass area) and biochemical (e.g., leaf nitrogen, leaf chlorophyll) characteristics of vegetation (Asner, 1998, Weiss, 2000, Verrelst, 2016)”

• Comment: Page 2 L35: tree canopy no vegetation (please remember the ecological concept of vegetation)

Vegetation substituted by “tree canopy”

• Comment: Page 2 L46-48: Good! Explanation

This specific sentence was also commented by another reviewer, so now it reads: “(..) have larger data requirements and limited generalization (..)”.

• Comment: Page 2 L48 – 49: do not exist in botany the concept of "Vegetation Leaf"in RS "Vegetation Leaf area", please recosider it the text

The sentence was simplified and to include all possibilities of RTM models, now it simply reads: (..) physical models to simulate the interaction of electromagnetic radiation with vegetation (..)

• Comment: Page 3 L82: I strongly advise against starting a paragraph by quoting a picture (1), please rephrase the sentence here. The Figure of stepwise approach was been an good idea instead of complex workflow figure

Thank you for your recommendation, this sentence now reads: “Our approach consisted of four main steps, and a clear overview is given in Figure 1.”

• Comment: Page 9 Figure 3: please add Cab, Cw, Cm, LAI and Car on the left part of figure, corresponding on the respective graphs  to better understand it...

We have added the items to the left of the figure as suggested.

• Comment: Page 14  Conclusion (general comment): overall comment I would suggest improving in the conclusion scientific proposal on the opportunity of the approach proposed how can be applied on ecosystem monitoring. Just further potential application cab be added (only references) in the field of environmental science for example to observe changes of ecosystems due global changes, climate, pollution or similar issues.

We thank the reviewer for this comment and in relation to the suggestion of submitting to another special issue: New Insights into Ecosystem Monitoring using Geospatial techniques. This submission was already on a special issue of MDPI (Remote and proximal assessment of plant traits) which we believe fits more with our manuscript in its current form. In that sense, and still taking in consideration your suggestions, we have adapted our conclusions to include a reference to further applications besides improving biophysical variable retrievals.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The paper aims to investigate the best hybrid approach for determination of satellite-based biophysical factors from Sentinel-2 data. The author applied the hybrid-RTM (e.g., PROSAIL RTM) and machine learning algorithms as the inversion procedure and further analysis in noise effects. The model and experiment design are well explained. However, the figures and tables are not arranged well in this manuscript. The type of figures is not easy to understand the result in detail. The paper needs some major revision. I think it is necessary to figure out some key points as below:

• In part 2.6., 10% of validating data is poor. The gold ratio between training and testing data should be 80-20 or 70-30.
• Table 2 and Table 4 contribute the same information, it should be considered to integrate them together.
• Figure 6 is intricate. You should separate training and validation phases figure. It is difficult to see the result and understand the key points. On the other hand, it is better to try another way to perform well the results (e.g., boxplot with the error range)
• It seems the author the described result from the figure only. It should be used effectively from tables and figures together.

Author Response

Dear Reviewer, please find our responses on the appended word document.

Author Response File: Author Response.docx