Estimating Forest Above-Ground Biomass in Central Amazonia Using Polarimetric Attributes of ALOS/PALSAR Images

Round 1

Reviewer 1 Report

Authors established an empirical model for the AGB estimation by ALOS/PALSAR images through considering the relation between some parameters of scattering mechanisms, and the 6 parameters includes σ°HH, Pv, αS2, ФS2, ФS3 and τm. However, there are some questions need to be clarified as following: 

1-   There are many parameters can used to describe the scattering mechanism, why only the 6 parameters were used in this paper?  Authors should give reasons for this choosing.

2-   The empirical model is very dependent on the selection of samples. The author should explain the structural characteristics of the forest at the sample point, because the scattering mechanism is closely related to the structural characteristics of the forest. In this paper, volume scattering Pv has the strongest positive correlation with AGB. However, AGB mainly exists in trunk, while volume scattering is closely related to leaves, branches and other structures. For example, only when the trunk is larger and the leaves and branches are more luxuriant, AGB have a strong correlation with Pv. However, some tree species, such as eucalyptus, have fewer leaves and branches, and in this case, the relationship between volume scattering and AGB will be another feature. So, it is doubtful whether there is a strong correlation between its volume scattering and AGB are suitable for all kinds of forest characteristics.

3-   What is the role of this section “3.1. SAR, inventory data and scattering mechanisms” for the following work? If the guiding significance is not obvious, it can be deleted.

4-   Line 68, please give the full name of the abbreviation “POLINSAR”.

5-   Line 508, numbers should be avoided as subjects.

Author Response

The authors would like to thank the reviewer for the suggestions which help us to improve well the article, we appreciated it. 

 

Comments and Suggestions for Authors

 

Authors established an empirical model for the AGB estimation by ALOS/PALSAR images through considering the relation between some parameters of scattering mechanisms, and the 6 parameters includes σ°HH, Pv, αS2, ФS2, ФS3 and τm. However, there are some questions need to be clarified as following:

Question 1- There are many parameters can used to describe the scattering mechanism, why only the 6 parameters were used in this paper?  Authors should give reasons for this choosing.

Response: The model is not empirical. 9 incoherent attributes and 12 coherent attributes were generated and used for the initial composition to the model. All these variables are named in item 2.3. (from lines 211 to 217). Finally, the 6 best variables were selected as components of the final biomass model following the statistical criteria described in item 2.4 (from lines 240 to 252) of the proposed methodology, best exemplified in the new Graphical abstract.

Question 2- The empirical model is very dependent on the selection of samples. The author should explain the structural characteristics of the forest at the sample point, because the scattering mechanism is closely related to the structural characteristics of the forest. In this paper, volume scattering Pv has the strongest positive correlation with AGB. However, AGB mainly exists in trunk, while volume scattering is closely related to leaves, branches and other structures. For example, only when the trunk is larger and the leaves and branches are more luxuriant, AGB have a strong correlation with Pv. However, some tree species, such as eucalyptus, have fewer leaves and branches, and in this case, the relationship between volume scattering and AGB will be another feature. So, it is doubtful whether there is a strong correlation between its volume scattering and AGB are suitable for all kinds of forest characteristics.

Response:  About  the volume and biomass relationship, the study carried out by Narvaes et al. (2010) in a similar area proved that the volumetric scattering component (Pv) contributes with more than 90% in all analysed units for an analysed primary and secondary forest in the Amazon (which is very dense, stratified and complex ecosystem), mainly due to the high incidence angle and consequently the low penetration capacity vertical, even in L-band, which naturally leads to greater interaction with the random structure of the forest canopy and naturally inclined thick branches and trunks. Figure 5 demonstrates that a large part of the biomass contribution is greater for taller individuals, as in this case, the trunk is larger than the leaves and branches are leafier, and so the greater correlation of biomass with Pv.

  Therefore, part of the conclusion written“ The explanatory variable with the highest positive correlation in the composition of the model (Pv) demonstrated  the importance of the different forest canopy strata that composed the phytophysiognomies analysed and caused the volumetric scattering” is directly based on these results found , in turn, section 3.1 describes that the volumetric scattering is the one with the greatest contribution to the analysed units.

 

Question 3 -What is the role of this section “3.1. SAR, inventory data and scattering mechanisms” for the following work? If the guiding significance is not obvious, it can be deleted.

Response: One of the points raised by the editor of the journal was to show in the introduction what is the differential of this research compared to other local studies and the main one (along with the use of the Touzi decomposition) and the generation of parallel polarisation responses in order to establish a correlation of forest inventory data, to understand the main scattering mechanisms involved in the scene and its relationship with biomass. We understand that this is one of the differentials of our analysis and it is an important detail. To reduce the paper size, we chose to insert them as a supplementary document join with the Table 1 and Figure 3.

Question 4-   Line 68, please give the full name of the abbreviation “POLINSAR”. –

Response: The correction was carried out (Line 67).

 

Question 5-   Line 508, numbers should be avoided as subjects. 

Response: We agree with the suggestion and we replaced it with the following wording “Forest biomass using X-band interferometric and radargrammetric techniques that successfully predicted models with high levels of biomass saturation point and good precision for forests with less floristic-structural complexity such as Spruce forests was studied by [22,23]. However, they presented superior errors in comparison with the models developed using polarimetric and PolInSAR techniques in tropical forests, which have a more complex structure”. Lines 515 - 520

Reviewer 2 Report

See attached file.

Comments for author File: Comments.pdf

Author Response

The authors would like to thank the reviewer for the suggestions that were very useful to improve the article's quality.

 

Question 1 “The paper is inconsistent and too wordy. It is entitled “Modelling of forest above-ground biomass in central Amazonia using polarimetric attributes of ALOS/PALSAR images” and in abstract states “We developed a method for estimating above-ground biomass (AGB) from power and phase-radar attributes in L- band images.”

Response – We agree and change the expression “Modelling” to “Estimating”, which was carried out along the paper. (Line = 2)

 

Question 2- “Then it is not clear why section 3.1 with results of polarimetric synthesis is needed?”

Response: We believe that this concept is different from other studies because it helps to clear up the interaction between the radar signal and the forest structure as a way to solve doubts about the statistically chosen parameters of the model.

To make the article less extensive and easier to read, and also provoked by the other reviewer, we condensed the table on this theme.

 

Question 3 - “The parameter ‘pedestal height’ given in Table 1 is not used further to estimate the AGB.”

Response: These parameters were used to additionally explain the main mechanism involved in the target after the model parameters were chosen by described statistical method. Furthermore, the pedestal value indicates the occurrence of volumetric scattering and in turn, indicates a higher AGB. It’s can be seen in lines 413-415.

To make this clear this pattern is corroborated to the AGB values measured in the forest inventory (Appendix A)

 

Question 4 - “For some reason, the method of estimation AGB itself is placed in section 3. Results, and not section 2.”

Response:  The biomass calculation follows the statistical precepts presented in section 2, which are widely used in the scientific literature. Consequently, the result of the best model for estimating aboveground biomass was presented in section 3 (3.1).

 

Question 5 - “Materials and Methods (paragraph 2.5). You declared a lot of parameters being analyzed (polarimetric coherence and polarimetric phase difference; entropy, anisotropy, and mean alpha angle; angle orientation phase, double bounce and surface component, parallel and cross-polarisation ratio, and total power; biomass index), but where the results?”

Response:  All of the above variables, except the parallel polarisation responses, were candidates for choosing the model with the greatest adherence to field data (with the smallest difference between the inventoried biomass and the model generated by the selected SAR attributes), according to the methodology established in [50]. Item 2.4 details the statistical criteria established for choosing the best model (with only the best-fit parameters) for the AGB calculation. In this regard, we accept the observation and reference all paragraphs in order to be clearer about this item with the literature used (Neter’s statistical book). Lines = 238; 240, 244 and 251.

 

Question 6- “Section 4 looks like a review of various methods of studying the forest rather than a discussion of the results of your work”.

Response: In this case, we partly agree because the review of various methods (Table 3) used reference to some points of the discussion as a way of comparing the quality of the result generated by us, relating it to different methods/parameters of analysis inserted in different structural and physiognomic contexts and different analysis scales (local to global).

 

Question 7 - Poor quality of all figures! For example, in Figure 2, it is impossible to read the text in some places due to jpeg compression artefacts.

 

Response: We agree, as for figure 2, we changed the font size and saved it in *.TIFF with at least 300 dpi format too, in order to improve its quality. The other figures also changed to conform to the norms established by the MDPI.

Other remarks.

Question 8 - “Kindly define all abbreviations during their first appearance in the abstract and main text and use them consistently throughout the paper. Do not duplicate definitions of abbreviations”.

 

Response – We adopted this procedure from the introduction because the word number limitation rules in the abstract made it impossible to describe all the abbreviations.

 

Question 9 - “Kindly use subscripts correctly.”

Response - The corrections were carried out.

 

Question 10 - “Line 25: The attributes of Touzi decompositions designated sometimes…”

Response – We agree and standardized all of them.

 

Question 11 - “Lines 28-29: Kindly define all abbreviations during their first appearance. What are R², Syx?”

Response - For the same reasons as mentioned above, we adopted the same criteria as previously described due to restrictions on the number of words in the abstract. These variables were described in the final model choice (section 2.4) as one of the selection techniques (lines 242 and 250, respectively). 

 

Question 12 - “Line 41: “... forests has changed from 31.6 to 30.6% ...” In what period of time?”

Response – The corresponding period of time was inserted in line 41.

 

Question 13 - “Line 70: Reference [30] is a duplicate of reference [14]”.

Response – The mistake was correct with new [14] reference (Lines 64 and 69).

 

Question 14 - “Figure 1: Why you don’t use Pauli RGB color composite, which is commonly used for polarimetric data?”

Response – It would be a possibility, but we generated the HH/HV and VV RGB composition for better visualization and your sample locations. It is important to highlight that we opted to use only the parameters calculated in our analyses.

 

Question 15: “Line 147: You specify the resolution of fully polarimetric ALOS-1/PALSAR-1 image as 3.56x9.37m. However, as noted in https://www.eorc.jaxa.jp/ALOS/en/alos/sensor/palsar_e.htm the resolution is not better than 24m. Also, after multilook processing in line 193, you wrote “... images with approximate dimensions of 22.85 m and 24.92 m (azimuth and range, respectively)” which contradicts your first statement (should be 24.92x9.37m).

Response – We agree with the observations and corrected some concepts adopted. The full polarimetric mode was initially distributed to the principal investigators to aim for the potential of these images in a wide range of applications worldwide.

In this way, the initial parameters for this kind mode of high resolution and incidence angle are adequate for forest analysis, with phase data differing from https://www.eorc.jaxa.jp/ALOS/en/alos/sensor/palsar_e.htm). The multi-look processing applied in [S] image with original line and pixel spacing (line 147) was a step used to become the cell resolution “almost squared” (22.85 m and 24.92 m) whilst preserving their phase intact and mainly hitting it to the ground (slant to ground range processing) afterwards filtered (Line 196 to 199).

 

Question 16: “Lines 154-156: Please specify at what time the forest inventory was made. Is it synchronized with satellite flight?

Response – This relevant information has inserted a compliment in the end of the indicated paragraph (Line 158)

 

Question 17: Lines 182-188: Why “The above-ground biomass (AGB in Mg ha-1) of each tree was ...” has dimension Mg ha-1? For example, the equation from [55] has the dimension of AGB in kg, not in kg ha-1.

Please, explain this and check the dimension of equations (1) and (2).

Response – We agree it was a mistake. The equations were used to calculate the individual biomass, however, we inserted “both later weighted by area (to Mg ha-1) according to the sample size used” at the end of this paragraph and remove the biomass area in the equations (Lines – 189 and 190; Equations 1 and 2 (lines 191 and 192).

Question 18: “…Also indicate that DBH measured in centimeters whereas TH in meters”

Response – We also adjusted the DBH units (lines 176 and 179).

Question 19: “Line 193: “approximate dimension” should be replaced by “approximate resolution”. See also comment to line 147.”

Response – We changed the expression “dimension” to “resolution” as suggested (lines 146-147 and 197).

Question 20: “Figure 2: The flowchart is too overloaded. The sequence of flows is not always clear. Some processes are drawn but not used (BMI, CSI, Claude-Pottier decomposition etc.).”

Response- The flowchart indicates the complete procedures performed in results generating, but some changes were carried out to be clearer (Figure 2).

We emphasize that all polarimetric attributes generated were used in the statistical analysis as elucidated in Neter et al. (1996).

Question 21: “Is radiometric terrain correction (RTC) applied?”

Response – This term can vary according to the software used. In this research, the procedures Slant to ground range and orthorectification were carried out.

 

Question 22: “What are autovalues, autovectors? Did you mean eigenvalues and eigenvectors?”

Response – Both terms are the same as eigenvalue and eigenvector because it is only a mathematical process for the matrix’s transformation, we decided to suppress them from the flowchart and thus make the results generation process clearer, as requested by the reviewer.

Question 23: “The final result should be estimated AGB but not Biomass/Geometry Accuracy”.

Response – We agree and removed Geometry Accuracy from the flowchart because, although it is fundamental for the extraction values of polarimetric attributes coinciding with the inventoried area, this evaluation was not taken into account in our analysis.

Question 24: “Line 214: The reference to (Shimada et al., 2009) is not correctly written and is missed in references.”

Response- The reference was redone according to MDPI citation standards. (line 218)

Question 25: “Formula 3: If you refer to the paper (Shimada et al., 2009), then you should give that formula unchanged. Because I can be misinterpreted as a real part of a complex number (see your own Figure 2). log10 should be written as log₁₀ or lg. cf is expressed in dB”.

Response - We agree that in this case, his concept is more appropriate. In this way, we made the requested corrections (lines 218 to 223) and adopted the original formula referred to by Shimada et al. (2009) so that it also coincides with what is described in the flowchart (Figure 2).

Question 26: “Lines 266, 277: “parallel polarization response / synthesis” You refer to paper [63], so please use the terminology of that work. For example, in caption of Figure 3 instead of “Examples of parallel polarisation syntheses in the samples of primary forest ... ” should be “Examples of copolarized signatures for the samples of primary forest... “. And not “syntheses”, but synthesis.

Response - Since 1987 when van Zyl (1987) started this new method being that other expert researchers in this area converged to polarimetric response (in this investigation parallel polarimetric response) because of not having a specific signature to each target (also explained in this article – Lines: 429-434) which depend on many factors, such as wavelength, target type (forest structure), forest layers, between others. Therefore, the term signature, in this case, it becomes ambiguous therefore we decided to use the new terminology for the same concept, widely used in the current scientific literature.

Question 27: “Line 284, Table 1: Why so many duplicates in second and third columns of table? You can simply define multiple scattering (HH<VV or HH»VV) as MS, and multiple and volumetric scattering (HH>VV) as MV and your table shrinks more than twice”.

Response - We agree with the reviewer's observations, so Table 1 has been organized according to the scattering mechanism by plots in the same typology. line 276).

 

Question 28: “Lines 302-306: Excessive definitions”.

Response - Although there is a general description of the attributes chosen by the statistical method (as a way of quoting the reader), they were described in detail for the first time in Table 2, so we believe that the descriptions should be maintained (lines 304 to 308).

 

Question 29: “Line 314: You should describe the formula in details. Here the dimension of AGB is missed. What values should the parameters have?” Should HH 0 be in decibels? What about Pv? What is the range of it changes? From zero to one or from 0% to 100%? The Touzi’s components are in degrees or in radians? Knowing of this is crucial for using your formula.”

 

Response – According to the literature, these parameters (included in this article proposal), the units and the range relating to different scattering mechanisms in the scene [41,42,58,59,60, among others]. Furthermore, according to statistical methodology when these different values and ranges are inserted into the chosen model (Best R², Syx, i.e) they generate the appropriate result.

Question 30: “How do the different terms in the formula relate to each other, what contribution do they make?”

Response – To be clearer, we change the first paragraph in the 3.2 section the following: “The model did not indicate multicollinearity problems (Table 2), in other words, there was a low non-significant correlation between the explanatory variables in the regression model (lower values of variance inflation value -VIF).” Line 298-300, supported by description methodology in section 2.4 (lines 239-251).

Question 31: “For example, let us compare the terms with Pv
and… i.e. in order to give the same contribution as the term with P_v, Tau_m should be in 176 times greater than Pv!”

 

Response – The relationship between these variables is non-direct and it relate to the own mathematical description linked to the scattering mechanism involved in the scene. Only after establishing this relationship it is possible to understand the scattering mechanism being that our importance in the model will indicate “how strong” this variable is to biomass calculation, that is how the same scattering mechanism in comparing to the other one is more relevant to the model. This can be seen in Neter et al. (1996).

 

Question 32: “Remembering that we consider tropical forest with great values of volumetric scattering, I wonder if parameter Tau_m makes valuable contribution. Why some parameters enclosed in parenthesis and some are not?”

 

Response– We believe that this issue is well reported in the results and also discussed. According to the AGB model results, the Tau_m had the third higher contribution to calculating biomass (higher values of Bk indicate its importance – Table 2, 2^nd column/last line and Equation 4). This demonstrates a valuable contribution due to its being invariant to the polarisation basis, typical of the natural forests “evidencing the secondary importance of multiple scattering” report on the Conclusion (Lines 593-597). The parenthesis in the equation also was reviewed.

Question 33: “Figure 6: Please indicate the dimension of biomass on axis, not in caption.”

Response – Based on the reviewer's request, Figure 6 changed the size and font according to MDPI standards.

Question 34: “Line 376: What r is?”

Response – Relevant observation. "The r is Pearson's correlation coefficient. This measures the degree of linear correlation between two quantitative variables", in this case, observed data versus estimated data. The concept it's at 2.4 item end (Line 251).

 

Question 35: “Lines 462-464: Unnecessary sentence. You do not discuss C-band early.”

Response - Interesting observation, however, we choose to keep it because it discusses the issue of signal interaction by randomly oriented theoretical dipoles present in the canopy and/or different strata of the tropical forest, demonstrating the contribution of multiple interactions (ɸS3 values found in our analysis) also found in our analysis.

Question 36: “Line 465: Tau_m, not Taum.”

Response – We standardized all variables as previously requested by the reviewer.

Question 37: “Table 3: You use designation “Satellite/Instrument” for SAR sensor. So, be consistent and write ALOS-1/PALSAR-1 and ALOS-2/PALSAR-2 in order to avoid ambiguity.”

Response – As requested, we have made changes to Table 3 including in section 2.2. Line 145

Question 38: “Line 567: Where is the Table 4?”

Response – It was a mistake, as we believe the best alternative would be to insert Table 3 (Line 496) because it is comparative data for understanding the accuracy and methods difference among different approaches in local and global biomass evaluations and consequently its limitations and strengths applied at forests in different typologies.

Question 39: “References should be formatted according Authors Guidelines. Some of paper titles written in CAPS (see 47, 53, 57, 59, 69). Journal names should be written in short form but you used both short and full forms (see 29, 42-44, 66, etc). For references 1, 53 you should provide hyperlinks at least.”

Response – We carried out the corrections suggested. However, the [1] is an article and following MDPI (instruction for authors) doesn’t provide the hyperlink. The reference [52] link has been provided.

Round 2

Reviewer 2 Report

Figure 3, parts (A)-(C): axis label should be ellitpicity not elipticity.

Right here: figure parts labeled as (A), (B), (C), (D), but in figure caption (a), (b), (c), (d) are used.