Polarimetric Measures in Biomass Change Prediction Using ALOS-2 PALSAR-2 Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

As I indicated in the pdf document, the article lacks explanation in many places, making it very difficult to understand the approach. 

I have two strong remarks:

- It is not possible (for me) to build a model of AGB change using data between 2014 and 2015 in order to predict AGB change between 2015 and 2021 for two reasons: (1) the estimated AGB for SAR in 2015 is not accurate and (2) AGB change cannot be modelled using data with a difference of only one year (too small).

- What is the reference AGB change? Do you have reference AGB data at two dates? I think not because you only have AGB data in 2014 and for 2015 you have an estimated AGB. However, an estimated AGB is not a reference AGB.

Comments for author File: Comments.pdf

Author Response

First, we would like to address your strong remarks, cited here for clarity:
"It is not possible (for me) to build a model of AGB change using data between 2014 and 2015 in order to predict AGB change between 2015 and 2021 for two reasons: (1) the estimated AGB for SAR in 2015 is not accurate and (2) AGB change cannot be modelled using data with a difference of only one year (too small).

"What is the reference AGB change? Do you have reference AGB data at two dates? I think not because you only have AGB data in 2014 and for 2015 you have an estimated AGB. However, an estimated AGB is not a reference AGB.I have two strong remarks:

    Both of these remarks are based on a misunderstanding of what was done, probably indicating that we have been unclear in the description. The models of AGB change are built on reference data from both 2014 (after growth season, so matching 2025 ALOS-2 data) and 2021. No estimated AGB is used as reference data. Hopefully the modifications described below will make this more clear.

Responses below are preceded by line numbers (Lx) or other indications of the position in the manuscript when line numbers are not applicable.

Title: Your suggestion to remove "Polarizations" in the title is good, and simplifies the title wothout compromizing in informativity. Thank you, we have modified the title to "Polarimetric measures in biomass change prediction using ALOS-2 PALSAR-2 data"

Title, L2, L8: We think "prediction" ought to be a suitable term here, as we first estimate the parameters of a model using one set of data, and then, using cross-validation, use the model guess at the value of a variable that is not part of the training data. Usually, this is considered prediction.

L44-46: Thank you, we have expanded paragraph about the reported limitations of L-band in estimating AGB. The suggested paper is a relevant reference, and has been included.

L82: Thank you, we have included these relevant papers.

L135-136: The dates are given close to the comment, on these lines: "..surveyed after the growth season of 2014, and again after the growth season of 2021" The dates correspond closely to the SAR acquisition dates.

Table 2: We have adden an explanation of the subscripts. We have also clarified your point about the number of coefficients. The reason for one extra coefficient in the all models is that they all include an intercept term (equal to 1) in addition to the predictor coefficient. Alpha_0 is the coefficient of the intercept.

L162: Your assumption is correct, t1 refers to 2015, and t2 to 2021. We have modified the text to give the actual years.

L164: We have clarified by changing all references to these as "variables" to "predictors". The final models are models 1-4 in Table 2. We have modified the caption to table 2 to clarify this point.

L176,L191: You are correct, Table 2 shows only the significant predictors (the predictors selected by the procedure described here). What is desribed here are the candidate predictors of each model type before selection. We have expanded the caption to Table 2 and the main text around L193 to clarify this point. The number of coefficients in Table 2 is also related to this, plus the intercept issue adressed above.

L193: This is adressed on L190: "The candidate predictors (subsrcripted with the acquisition year)", but as noted above, we have now also clarified this in the caption to Table 2.

L198-200: I understand your confusion. Up to this point, we have modeled AGB CHANGE, but This part refers to illustrating the relationship between SAR from 2015 and the STATE of AGB for the same year (inventoried in 2014, but after the growth season, so corresponding to the SAR data from 2015). We have made a separate paragraph of this part, and tried to reword it to clarify the situation.

L209: I assume you refer to the final output pixel size of 10 m x 10 m here. While it is true that the applied window size exceeds this, and this may lead to some of the information in the pixel depending on information from nearby areas. Ideally, one would have avoided this, but the inherent noisiness of the data necessitated a large window size in order to achieve reasonable prediction quality. As such, the output image is to be considered oversampled. However, each field plot is in fact a 40 m radius circle, thereby containing multiple such oversampled pixels.

L281: Thank you. It is fixed.

Figures 2, 3, 4, including captions: Apart from one mistake here on our part ("2018" in the caption of Figure 5 should be "2014"), these comments arise from a misunderstanding I think we already addressed above, namely the reference data used. For the change prediction, reference data from 2014 (after growth season of 2014, so that it matches 2015 ALOS-2 data), and 2021. For the prediction of AGB STATE, only the reference data from 2014 was used. We have added to the years to the captions.

Reviewer 2 Report

Comments and Suggestions for Authors

The study is interesting and timely, and the results are good. The plots are large with 40m radius.

I suggest to use a title like: Estimating forest biomass from repeated polarimetric L-band data

The first three objectives do have answers in the conclusion, but not the last on forest structure.

It would be good if the term forest structure was explained, because it is a generic term and can include a wide range of forest properties.

In the instructions to authors the journal requires that there is a Results chapter and a Discussion chapter, - a Results and discussion chapter appears not to be in line with the instructions.

At the end they should discuss the representativity and reproducibility. Do they believe that these results are valid across a range of weather types?

Details:

L134: subjective selection of 46 plots could be an issue, so please clarify what is meant by forest types and spatial distribution here?

L140: The word hyposmeter must be wrong here, I assume they mean relascope.

L151: I suggest rewriting this “high-sensitive mode data in strip map format” to “high-sensitive stripmap mode”

Table 2: correct and explain the subscripts in the table heading.

L162: insert after alfa i: where i=1-k

L165: what is a pair here?

L166: Was decibel only used in model 1? If so, why?

L168: What is ‘They’ referring to here? I suggest they write cross-pol data have …

L210: what is a ‘target window’?

L233: lowercase d in discussion

L263: hig hlighted

 

L264: use the logic “large height” not “high height”. So write “largest hL”. Or maybe even better is to here use “largest Lorey’s height”, or “tallest forest”. 

Author Response

Response to reviewer 2

The study is interesting and timely, and the results are good. The plots are large with 40m radius.
-Thank you, we very much appreciate your assessment of the interest and timeliness of the paper.

I suggest to use a title like: Estimating forest biomass from repeated polarimetric L-band data
-We find that the title, as modified in accordance with a suggestion from Reviewer 1, better reflects the specific nature of this study as compared to a number of other papers which could be given the title you suggest.

The first three objectives do have answers in the conclusion, but not the last on forest structure.
-The fourth objective is stated as "4) investigate the impact of forest structure on the sensitivity of L-band polarimetric backscatter to AGB and AGB change" We are of the opinion that this is addressed in the conclusion. Specifically by the following passage:  "No clear correlation between tree species and sensitivity to AGB or AGB change was observed. The sensitivity to AGB was reduced for high values of h_L, BA and n, when predicting absolute AGB, but the best change prediction model was still sensitive to differences in changes larger than the apparent saturation point for AGB state estimates."

It would be good if the term forest structure was explained, because it is a generic term and can include a wide range of forest properties.
-While forest structure is admittedly a somewhat vague term, we are of the opinion that, by focussing on specific well defined variables relating to it, the meaning of forest structure as employed in the paper is well defined.

In the instructions to authors the journal requires that there is a Results chapter and a Discussion chapter, - a Results and discussion chapter appears not to be in line with the instructions.
-The instructions may state this, but we have experienced that in practice, Remote Sensing allows for some flexibility not only in this regard when suitable, but also in the naming of Sections. Out of the currently displayed articles on the front page of Remote Sensing, at least these these three articles have combined Results and Discussion sections: https://doi.org/10.3390/rs16050738, https://doi.org/10.3390/rs16050734, https://doi.org/10.3390/rs16050733, 

At the end they should discuss the representativity and reproducibility. Do they believe that these results are valid across a range of weather types?
-Good point, we have expanded on the part discussing limitations in the discussion to address this point.

Detailed comments:

L134: subjective selection of 46 plots could be an issue, so please clarify what is meant by forest types and spatial distribution here?
-"Forest types" refers to aiming at an approximately representative sample of tree species ..., while spatial distribution refers to the objective of sampling the estate so that the field plots are well spread out over the area. We have modified the text to clarify these points.

L140: The word hyposmeter must be wrong here, I assume they mean relascope.
-Hypsometer is correct, referring to the height measurement here. However, the sentence structure perhaps made it a bit unclear which quantity was measured with the hypsometer, so we have reformulated it.

L151: I suggest rewriting this “high-sensitive mode data in strip map format” to “high-sensitive stripmap mode”
-Thank you, suggestion accepted.

Table 2: correct and explain the subscripts in the table heading.
-We have corrected them and added an explanation of them in the heading. 

L162: insert after alfa i: where i=1-k
-"k" in equation 1 refers to the number of predictors, and is thus constant for a specific model. "i" is here used only as an extra variable to refer to an arbitrary subscript. However, as alpha_1 is the coefficient of the intercept term (p0 = 1 by definition), we have modified the equation to reflect this. We have also modified the text to clarify that an intercept parameter is also estimated for each model, it was a bit unclear as written.  

L165: what is a pair here?
-It refers to two measurements of the same kind of polarimetric variable (say for example gamma_HV), one from each of the two dates. This was vaguely expressed, and we have tried to clarify it.

L166: Was decibel only used in model 1? If so, why?
-No, it was supposed to be a general statement. This was confusingly placed, thank you for noticing. We have moved the remark into the description of the general linear regression model.

L168: What is ‘They’ referring to here? I suggest they write cross-pol data have …
-Cross-pol data is correct, and we have rephrased accordingly.

L210: what is a ‘target window’?
-It refers to the size of window used to detect stable scatterers in the adaptive filtering process. However, we have noticed that the naming of the filter is incorrect in the SNAP software. It should be "Extended Lee-Sigma", not "Improved Lee-Sigma" ("Improved" is not developed for polarimetric filtering. "Extended" is). We have changed this and added a reference to the paper by Lee et al. defining the filter.

L233: lowercase d in discussion
-We think it ought to be like it is, in so called "title case", like "Methods" in "Materials and Methods". We have also adjusted the subsection titles to conform to title case.

L263: hig hlighted
-Fixed, thank you.
 
L264: use the logic “large height” not “high height”. So write “largest hL”. Or maybe even better is to here use “largest Lorey’s height”, or “tallest forest”.
-We have taken your suggestion and used "largest" and "smallest". This logic was has now been applied not just here, but in all places applicable.

-Finally, thank you very much for your review. Your comments certainly improved the readability of the manuscript, and we hope that you will find our modification satisfactory.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors aimed to establish correlations between synthetic aperture radar polarimetric signatures and above-ground biomass. Overall, the manuscript is well-crafted, with clear organization and design. However, I observed a lack of robust and significant relations that can be effectively conveyed to readers. To enhance clarity, inclusion of polarimetric images like Freeman-Durden decomposition, Pauli, or Co-Cross-pol RGB could illustrate variations in signatures with different biomass types. Additionally, the dataset’s limited size, around 40 data points, may not sufficiently persuade readers; expanding the dataset could strengthen the paper’s impact.

Comments on the Quality of English Language

Well written

Author Response

Response to reviewer 3

Thank you for your review of our manuscript. We have cited your remarks and responded to them below.

"The authors aimed to establish correlations between synthetic aperture radar polarimetric signatures and above-ground biomass. Overall, the manuscript is well-crafted, with clear organization and design."
 -Thank you, we are very happy to hear this, especially that you found the organization and design of the manuscript clear.

"However, I observed a lack of robust and significant relations that can be effectively conveyed to readers. To enhance clarity, inclusion of polarimetric images like Freeman-Durden decomposition, Pauli, or Co-Cross-pol RGB could illustrate variations in signatures with different biomass types."
 -The kind of images you are referring to could have been useful for illustration purposes if we had corresponding biomass maps to juxtapose them with, but as we only have the 40 field plots with biomass data, we do not believe that such a juxtaposition would contribute with much clarity.

"Additionally, the dataset’s limited size, around 40 data points, may not sufficiently persuade readers; expanding the dataset could strengthen the paper’s impact."
 -While 40 field plots is not a very large amount, they are of large size and hign quality, as all plots were resurveyed to produce direct reference measurements of change. Ideally We would also prefer a larger data set, but the data set used is the largest one (of high quality) we had access to in connection to the ALOS-2 data in question.

Thank you again for contributing your time to to do this review, hopefully you will find our responses satisfactory.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper has been improved and my comments have been taken into account.