Linking Terrestrial LiDAR Scanner and Conventional Forest Structure Measurements with Multi-Modal Satellite Data
, Mathieu Decuyper 1,2, Benjamin Brede 1, Lammert Kooistra 1, Johannes Reiche 1, Brice Mora 3 and Martin Herold 1Round 1
Reviewer 1 Report
The authors present the relationship of satellite remote sensing variables with terrestrial lidar and conventional forest measurements in the article. However, the materials and methods section has to improve to make the design of study clear. The results that show some moderate and acceptable results are expectable.
Introduction
The introduction could have more descriptions of the significance of the study to tell readers why the study is important for future study works.
Materials and methods
Please have contents to describe the design of TLS in the field and how you process the TLS data.
After 2003, Landsat 7 have the SLC-off issue. Please have a brief description about how you solve the issues.
What is the final spatial resolution of the satellite remote sensing images you use to find the relationship with TLS and conventional measurements?
Did you split your data to training and testing data for the stepwise regression?
How do you validate your results?
When the multicollinearity happened, what is the threshold to include or eliminate the variables?
Figure 2 does not present how you validate your fianl models.
Results
You can remove the titles of TLS and conventional in figure 4.
You mention you use logarithmic transformation in the study, could you show the results?
The model variables in table 5 are not clear.
Discussion
The first paragraph could move to conclusion.
In the figure 6, you have both univariate and multivariate predictors, what predictor you recommend to use?
Author Response
Response to Reviewer 1 Comments
The authors present the relationship of satellite remote sensing variables with terrestrial lidar and conventional forest measurements in the article. However, the materials and methods section has to improve to make the design of study clear. The results that show some moderate and acceptable results are expectable.
We thank the reviewer for their positive comments on our paper and on the suggestions for improvement
Introduction
1. The introduction could have more descriptions of the significance of the study to tell readers why the study is important for future study works.
- We would like to thank the reviewer for the valuable comment. We agree with the reviewer that the background and significance of the study was not sufficiently described. We have now addressed this comment by re-writing the introduction. We have also included relevant references that indicate the existing gaps, and thus the relevance of this study (L:60-114). Our study investigates the usability of remote sensing derived variables to estimate and produce wall-to-wall information on field measured forest structure parameters.
Materials and methods
2. Please have contents to describe the design of TLS in the field and how you process the TLS data.
- We agree with the reviewer that there should be more information presented on the TLS data used. We have updated the material and methods section with details on the TLS data acquisition and analysis (L:141-144, L:150-153). We now also made a clear reference to our previous paper that fully addresses the details on field protocols and analysis.
3. After 2003, Landsat 7 have the SLC-off issue. Please have a brief description about how you solve the issues.
- We did not have this issue with the Lansat-7 images as we were working on a plot level (9-pixels per plot) and for these plots no scanlines were missing.
4. What is the final spatial resolution of the satellite remote sensing images you use to find the relationship with TLS and conventional measurements?
- We used the original resolution of each image (table 2) for both correlation analysis and regression model. We now explicitly mentioned this on Line 195-196
5. Did you split your data to training and testing data for the stepwise regression?
How do you validate your results?
- We did not split our data to training and validation dataset as we used a small number of observation (25 plots) for the study. Thus, our interest was to make an exploratory study on the applicability of SRS derived variables, instead of doing the actual upscaling. We now have explicitly mentioned our aim in the introduction (Line118-123) and also discussed the issue of sample sizes and analytical algotithms on the discussion (L:388-392)
6. When the multicollinearity happened, what is the threshold to include or eliminate the variables?
- We now have added information (L223-224-231) on the threshold value we used based on correlation of predictors (<0.6) and variance inflation factor ( < 2)
7. Figure 2 does not present how you validate your final models.
- We have not used a validation dataset (see also under point 5 in the Material and Methods part of the rebuttal).
Results
8. You can remove the titles of TLS and conventional in figure 4
- Corrected!
9. You mention you use logarithmic transformation in the study, could you show the results?
- We log transformed the data before using them in the linear model. Some field measured parameters (AGB, mean gap, total basal area, number of gaps, maximum gap, PAVD at 10m) data distribution were skewed because of structural diversity across forest types
10. The model variables in table 5 are not clear.
- We have updated the caption of table 5 with a description of abbreviations.
Discussion
11. The first paragraph could move to conclusion.
- We agree with the reviewer. Moved!
12. In the figure 6, you have both univariate and multivariate predictors, what predictor you recommend to use?
- We agree with the concern of the reviewer, but as discussed (L:393-412), the choice of predictors will depend on (i) the complexity of the structural parameter being estimated, on (ii) the type of SRS dataset used, on (iii) the forest ecosystem of the study area etc. Therefore, instead of making a general recommendation, we are indicating on the possible integration of SRS dataset based on our specific case.
- We thank the reviewer for the critical comments towards the improvement of our manuscript.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The paper is well written and organized and reads very easily.
However, I really don't like statistical studies like this without some explanation of the underling physics (basically, light interaction with matter) that is the reason for the observed correlations. As it is this study shows fairly weak correlations (Figure 4) for this particular set of observations. I am left with many questions that I believe the authors could answer to make the paper more general:
What are the physical premises for choosing the particular suite of observational instruments?
Are these measurements applicable to other ecosystems, forest types or structures?
Are the measurements seasonal, that is will these same correlations be observed in different parts of the growing season?
Is the spatial resolution of the the observational instruments sufficient to correlate with the ground-based measurements? (Nyquist's theorem)
Does the fact that the satellite measurements are not time-coincident matter?
I could come up with more questions like this, but I think you get the general idea of what I am talking about. I have seen too many remote sensing papers where a a particular algorithm works for some measurement but only for one particular scene at one time of day and time of year and so is strictly a fortuitous measurement without any scientific merit. I don't think this work is wihtout merit, but I urge the authors to demonstrate this before the paper is ready to be published.
Author Response
Response to Reviewer 2 Comments
The paper is well written and organized and reads very easily.
However, I really don't like statistical studies like this without some explanation of the underling physics (basically, light interaction with matter) that is the reason for the observed correlations. As it is this study shows fairly weak correlations (Figure 4) for this particular set of observations. I am left with many questions that I believe the authors could answer to make the paper more general:
We thank the reviewer for their positive comments on our paper and on the suggestions for improvement
1. What are the physical premises for choosing the particular suite of observational instruments?
- We would like to thank the reviewer for the valuable comment. We have now updated the paragraphs in the introduction (L: 60-114) to discuss the expected link between the information obtained from the sensors (both satellite and the three-dimensional TLS data) and vegetation.
2. Are these measurements applicable to other ecosystems, forest types or structures?
- We agree with the concerns of the reviewer. We have now added more information on the applicability of SRS data in different forest ecosystem types, and their findings in comparison to our study (L:83-91, L:375-383)
3. Are the measurements seasonal, that is will these same correlations be observed in different parts of the growing season?
- We have now discussed issues of seasonality in detail (L: 345-347, L: 383-388). Seasonality will affect data acquisition means. Otherwise, our plots are located in an evergreen mountain forests, thus the correlations are expected to be consistent throughout seasons.
4. Is the spatial resolution of the observational instruments sufficient to correlate with the ground-based measurements? (Nyquist's theorem)
- We now have introduced conditions about the spatial resolution of datasets in the introduction (L:86-89) and about possible mismatches in the discussion (L:337-338, 354-358).
5. Does the fact that the satellite measurements are not time-coincident matter?
- We agree with the concern of the reviewer. Since the change in forest vertical structure is often a slow process, we do not assume the time-lag between field data collection and SRS data acquisition will affect the study (L:169-171).
6. I could come up with more questions like this, but I think you get the general idea of what I am talking about. I have seen too many remote sensing papers where a a particular algorithm works for some measurement but only for one particular scene at one time of day and time of year and so is strictly a fortuitous measurement without any scientific merit. I don't think this work is wihtout merit, but I urge the authors to demonstrate this before the paper is ready to be published.
- We agree with the reflections of the reviewer. We have extended the introduction (L: 60-114) and discussion section (L:363-392) to answer general but critical questions on usability, repeatability, comparison with other eco-systems, limitations and so on.
- We thank the reviewer for the critical comments towards the improvement of our manuscript.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear authors,
Thank you very much for the manuscript. The theme of the manuscript is interesting because the terrestrial laser scanner enabled the characterization of detail object structures and its application for various non-uniform structures in forestry could benefit the science and knowledge in the field. Utilization of the satellite remote sensing data would enabled coverage of large areas, suitable for cost effective monitoring and assessment.
However, despite the amount of work that has been done, particularly the laborious field data sampling there some major issues which in my opinion need to be addressed before this manuscript could be consider for publication. Kindly find my comments below. I wish that my comment is useful to increase the quality of this manuscript.
1. In my opinion, the problem statement and the hypothesis are not present or unclearly presented. Usually, this critical element of a study can be found in the introduction section. Hence, I think the introduction is too general and contains least information regarding the scientific basis of the study.
Such critical question is like;
· What is your scientific hypothesis/reasons of using various indices from various SRS to establish relationship with the forest derived structure parameters?
For instance, how the different polarization of radar backscatter helps you to characterize different forest structure parameters.
And how TLS which provides 3D physical structure information could relate with the spectral based indices which is influenced by the leaf biochemical and biophysical properties?
· TLS provides you with 3D structure parameters, while SRS provide 2D parameters. Technically, how this could be related? Because there is no methodological explanation on the derivation of the TLS derived parameters, I think the reader will find it difficult to understand the concept which you tend to introduce.
2. I also find that the literature review of this manuscript is inadequate, quite misleading with the objective of the study and lack depth. I suggest the authors to emphasize on the theory of spectral vegetation indices, SAR radar-based polarization and forest structure parameters derived TLS, and how scientifically this could relate each other.
3. I have major concern of the methodology and also its corresponding results.
a) The authors performed correlation with different parameters between the SRS vs. the field and TLS measurement. I found that most correlation are weak, except very few parameters such as canopy openness, min and its maximum gap. Could you explain why this happen (including why the correlation are weak for others).
b) Due to this low correlation and coefficient of determination (R-squared), I think the derived model have low predictive power and is not robust. The use of stepwise regression (one type of mixture linear model that based on linear prediction) clearly could not compensate and increasing the predictive power of the model, although the degree of freedom and samples are statistically increased. I suggest the authors to employ non-linear mixture model, as clearly the results showing non-linear relationship.
c) Do the correlation between the predicted and observed parameters in table 5 are computed based on similar plot?
d) Again, I wish the authors could explain how the physical forest structure 3D variable that derived from TLS can be relate with the 2D derived variables by the SRS. Kindly remember that most of the SRS variables especially one from the optical satellites are based on the leaf biophysical and biochemical properties.
I think all the major concerns should be address therefore to increase the quality of scientific soundness of this manuscript.
Thank you very much,
Good luck
Author Response
Response to Reviewer 3 Comments
Dear authors,
Thank you very much for the manuscript. The theme of the manuscript is interesting because the terrestrial laser scanner enabled the characterization of detail object structures and its application for various non-uniform structures in forestry could benefit the science and knowledge in the field. Utilization of the satellite remote sensing data would enabled coverage of large areas, suitable for cost effective monitoring and assessment.
However, despite the amount of work that has been done, particularly the laborious field data sampling there some major issues which in my opinion need to be addressed before this manuscript could be consider for publication. Kindly find my comments below. I wish that my comment is useful to increase the quality of this manuscript.
1. In my opinion, the problem statement and the hypothesis are not present or unclearly presented. Usually, this critical element of a study can be found in the introduction section. Hence, I think the introduction is too general and contains least information regarding the scientific basis of the study.
- We would like to thank the reviewer for the valuable comment. We agree that the introduction was too general. We have now made an extended introduction where we presented the basic principles on forest structure estimation using remote sensing. We have included relevant references that indicate the existing gaps, and thus the relevance of this study (L:60-123). Our study investigates the usability of remote sensing derived variables to estimate and produce wall-to-wall information on field measured forest structure parameters.
Such critical question is like;
What is your scientific hypothesis/reasons of using various indices from various SRS to establish relationship with the forest derived structure parameters?
- We agree with the reviewer that the scientific reasoning were not clearly stated. We now updated the introduction to show the expected interaction between sensors and vegetation, and what that means in relation to the three-dimensional information (L:74-106)
For instance, how the different polarization of radar backscatter helps you to characterize different forest structure parameters.
- We have now addressed this by introducing the relevance of different polarizations and bands in estimating structural parameters (L:92-106, L:347-351). For instance the use of L-band and HV polarization has been discussed in terms of its capability to penetrate through dense vegetation and indicate on backscatters from stems and large branches.
And how TLS which provides 3D physical structure information could relate with the spectral based indices which is influenced by the leaf biochemical and biophysical properties?
- The use of SRS for structural assessment of forest environments is based on the distinct characteristics expected from forest canopies when in contact with solar radiation and/or with signals from active satellites. We have now discussed this for both optical (L:74-91) and SAR (L:92-106) remote sensing.
TLS provides you with 3D structure parameters, while SRS provide 2D parameters. Technically, how this could be related? Because there is no methodological explanation on the derivation of the TLS derived parameters, I think the reader will find it difficult to understand the concept which you tend to introduce.
- We agree with the reviewer. We have updated the material and methods section with details on the TLS data acquisition and analysis (L:141-144, L:150-153). We also made a clear reference to our previous paper that fully addresses the details on field data acquisition and analysis
2. I also find that the literature review of this manuscript is inadequate, quite misleading with the objective of the study and lack depth. I suggest the authors to emphasize on the theory of spectral vegetation indices, SAR radar-based polarization and forest structure parameters derived TLS, and how scientifically this could relate each other.
- We have addressed this comment by adding an extensive review in the introduction (L:60-114). In addition, we have rephrased our objectives and added the hypothesis to bring across our ideas clearly (L:115-123)
3. I have major concern of the methodology and also its corresponding results.
a) The authors performed correlation with different parameters between the SRS vs. the field and TLS measurement. I found that most correlation are weak, except very few parameters such as canopy openness, min and its maximum gap. Could you explain why this happen (including why the correlation are weak for others).
- We agree with the comments and concerns of the reviewer. We have addressed this comment by first, adding a section in the introduction (L:60-123) regarding the expected relationship between SRS variables and field measurements, especially in the case of tropical forests. We then addressed this issue by extending the discussion on our findings (L:363-368), in relation to saturation of indices/backscatters as well as non-linear relationships.
b) Due to this low correlation and coefficient of determination (R-squared), I think the derived model have low predictive power and is not robust. The use of stepwise regression (one type of mixture linear model that based on linear prediction) clearly could not compensate and increasing the predictive power of the model, although the degree of freedom and samples are statistically increased. I suggest the authors to employ non-linear mixture model, as clearly the results showing non-linear relationship.
- We agree with the comments and concerns of the reviewer. However, we argue that, as also indicated on other studies, relatively lower correlations and predictive power of models are expected in tropical forests in regards to the complexity of forest structure in such ecosystems. Regarding the modelling approaches, we first explored random forest, but learned that the linear models suits better to the small number of observations we used on this study. We have now justified the reason why for using linear models (L:371-374). We also discussed non-linear relationships on Line 367-374.
Below, we present the information on residuals to justify the suitability of the linear models used in this study.
Mean gap
Maximum gap
Canopy openness
Number of gaps
PAVD@10m
PAVD total
Average height
AGB
Tree density
Total basal area
Number of species
c) Do the correlation between the predicted and observed parameters in table 5 are computed based on similar plot?
- Yes.
d) Again, I wish the authors could explain how the physical forest structure 3D variable that derived from TLS can be relate with the 2D derived variables by the SRS. Kindly remember that most of the SRS variables especially one from the optical satellites are based on the leaf biophysical and biochemical properties.
- We agree with the comments of the reviewer. We believe that with the substantial changes made in the manuscript (L:60-106, L:363-392), we have now addressed the major issues related to explaining the theory behind the measurement of 3D structure with the use of 2D image derivable.
I think all the major concerns should be address therefore to increase the quality of scientific soundness of this manuscript.
- We thank the reviewer for the critical comments towards the improvement of our manuscript.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I appreciate the efforts that the authors completed all of the concerns from reviewers. However, I still concern about using SRS with different spatial resolutions to build models. The coverage of a 30 m spatial resolution pixel is equivalent to 9 10 m spatial resolution pixels or 100 3 m spatial resolution pixels. For instance, when the model has 30 m and 10 m spatial resolution data, which is the corresponding pixel of the 9 10 m spatial resolution pixels? Therefore, I anticipate the authors to answer the SRS spatial resolution problem in the methods and materials section.
Author Response
Response to Reviewer 1 Comments
I appreciate the efforts that the authors completed all of the concerns from reviewers.
We thank the reviewer for their positive comments on the revision made and on the suggestions for improvement
Materials and methods
However, I still concern about using SRS with different spatial resolutions to build models. The coverage of a 30 m spatial resolution pixel is equivalent to 9 10 m spatial resolution pixels or 100 3 m spatial resolution pixels. For instance, when the model has 30 m and 10 m spatial resolution data, which is the corresponding pixel of the 9 10 m spatial resolution pixels? Therefore, I anticipate the authors to answer the SRS spatial resolution problem in the methods and materials section.
- We have now added a sentence in the Materials and Methods section to justify why we used the mean value of vegetation indices and backscatter values of different number of pixels across different datasets representing our field plots, to build our linear regression model.
L:223-L:225: “The SRS variables used in the multiple linear regression models were derived using the original pixel size of the high to medium spatial resolution images (table 2), so as to capture the possible detailed information on the corresponding field measured structural variables from the high-resolution SRS images”
In addition to the added sentence, there are other statements (L:207-212, L:195-196) across the section that elaborates how the field measured plots are represented on the high and medium resolution images.
Author Response File: Author Response.docx
