Improved Cellular Automaton for Stand Delineation
Round 1
Reviewer 1 Report
Overall comments:
This paper presents research that will be of interest to forest’s readers. The work is both useful and technically sound. The writing needs some revision in order to best highlight and frame the value of the technical work presented. The bulk of my comments address writing structure, proofreading, and readability.
In the title, abstract, and introduction, the authors have framed their work, beginning by emphasizing that they use Airborn Laser Scanner data (ALS), and a cellular automata algorithm in conjunction with those data. In highlighting the source data and algorithm first, they (inadvertently?) obscure the primary point of their work, which is to illustrate a useful and flexible method of automated stand delineation. We recommend that they revise their title, abstract and introduction, mentioning the primary goal of their work (improved, automated stand delineation), before highlighting the data and algorithm. For all three of these sections, re-arranging the information already present, would serve to improve the paper’s communication in this respect, and also possibly raise its’ profile.
Specific comments follow:
Introduction:
Line 98 and others: You describe R2 statistics. Do you mean R-squared? If so, please format the ‘2’ as a superscript throughout the paper.
Line 130: Number format could be easier to read as 15,503 than 15503, and 4,438 rather than 4438.
Materials:
Section 2.1: An overview map showing the location of your study area would be helpful.
Line 157: ‘existing forest map’ Please provide a citation for external data sources. This is for reproducibility’s sake, and also to appropriately credit the external data set’s originators.
Methods:
This section reads very well. Clearly, much care has gone in to section 3.1. The use of figures throughout is especially helpful. However, proofreading is still needed. I have listed the minor writing errors that easily caught my eye. Please review the entire text for other errors that I may have missed.
Line 171: Please define the term ‘stand number’ before you use it, or refine your text so that the term is not needed.
Line 188: Proofreading -- “The sum is the weights was equal to one.” Did you mean “The sum of the weights was equal to one.” ?
Line 190: Proofreading – “… the border length, is was assumed…” Should probably read: “ … the border length, we assumed…”
Line 195: Proofreading – “… the more the further the cell …” replace with: “… the further the cell …”
Line 215: Proofreading – “… sigmoid-type of relationships …” replace with “ … sigmoid-types of relationships …”
Line 243: The naming convention used here (“Shape 1” and “Shape 2” describing two shape-descriptor variables) is unnecessarily, and problematically redundant with the optimization case names. Please modify either the optimization case names, or the shape descriptor variable names so that they more clearly indicate two separate things.
Line 267, Figure 6 caption: The inclusion of abbreviated definitions for D, A, B and S parameters would ease figure interpretation for readers. Referring back to Equation 1 is adequate, but cumbersome.
Discussion:
Overall, this section reads well, and is informative, although some of the material presented here should perhaps be considered results (e.g., Figures 9, 10 and 11).
Lines 330 – 331, Figure 11: The use of only two subsets of data to demonstrate that the approach is robust and consistent in new situations is not particularly convincing. A larger sampling of the segmented landscape (with consistently high statistics) would be more convincing, if it is possible. Additional figures would not be needed, but statistics highlighting the stability of the R2 statistics across additional areas are needed to make the authors’ assertion of robustness in new areas convincing.
After reading through the discussion, I am especially curious to hear more of the authors’ thoughts on whether their approach could be applied to larger spatial extents, although perhaps this is a topic best addressed in another study.
Author Response
This paper presents research that will be of interest to forest’s readers. The work is both useful and technically sound. The writing needs some revision in order to best highlight and frame the value of the technical work presented. The bulk of my comments address writing structure, proofreading, and readability.
Point 1: In the title, abstract, and introduction, the authors have framed their work, beginning by emphasizing that they use Airborn Laser Scanner data (ALS), and a cellular automata algorithm in conjunction with those data. In highlighting the source data and algorithm first, they (inadvertently?) obscure the primary point of their work, which is to illustrate a useful and flexible method of automated stand delineation. We recommend that they revise their title, abstract and introduction, mentioning the primary goal of their work (improved, automated stand delineation), before highlighting the data and algorithm. For all three of these sections, re-arranging the information already present, would serve to improve the paper’s communication in this respect, and also possibly raise its’ profile.
Response 1: The abstract and the introduction were restructured as suggested by the reviewer (automated stand delineation first, data later). The title was also modified to emphasize methodological improvement and automated stand delineation
Point 2: Line 98 and others: You describe R2 statistics. Do you mean R-squared? If so, please format the ‘2’ as a superscript throughout the paper.
Response 2: ‘2’ is now superscript. The formulas to calculate R2 were added.
Point 3: Line 130: Number format could be easier to read as 15,503 than 15503, and 4,438 rather than 4438.
Response 3: Modified as suggested
Point 4: Section 2.1: An overview map showing the location of your study area would be helpful.
Response 4: An overview map was added
Point 5: Line 157: ‘existing forest map’ Please provide a citation for external data sources. This is for reproducibility’s sake, and also to appropriately credit the external data set’s originators.
Response 5: A reference to existing forest map was added.
Point 6: This section reads very well. Clearly, much care has gone in to section 3.1. The use of figures throughout is especially helpful. However, proofreading is still needed. I have listed the minor writing errors that easily caught my eye. Please review the entire text for other errors that I may have missed.
Response 6: The minor writing errors spotted by the reviewer and ourselves were corrected
Point 7: Line 171: Please define the term ‘stand number’ before you use it, or refine your text so that the term is not needed.
Response 7: It is now explained that stand number is the ID number of the stand.
Point 8: Line 188: Proofreading -- “The sum is the weights was equal to one.” Did you mean “The sum of the weights was equal to one.” ?
Response 8: Corrected
Point 9:Line 190: Proofreading – “… the border length, is was assumed…” Should probably read: “ … the border length, we assumed…”
Response 9: Corrected as suggested
Point 10:Line 195: Proofreading – “… the more the further the cell …” replace with: “… the further the cell …”
Response 10: Corrected as suggested
Point 11:Line 215: Proofreading – “… sigmoid-type of relationships …” replace with “ … sigmoid-types of relationships …”
Response 11: Corrected as suggested
Point 12:Line 243: The naming convention used here (“Shape 1” and “Shape 2” describing two shape-descriptor variables) is unnecessarily, and problematically redundant with the optimization case names. Please modify either the optimization case names, or the shape descriptor variable names so that they more clearly indicate two separate things.
Response 12: The shape descriptors are now called form indicators to avoid using the same name for two different things
Point 13:Line 267, Figure 6 caption: The inclusion of abbreviated definitions for D, A, B and S parameters would ease figure interpretation for readers. Referring back to Equation 1 is adequate, but cumbersome.
Response 13: Explanation of the letters was added to the figure caption.
Point 14: Overall, this section reads well, and is informative, although some of the material presented here should perhaps be considered results (e.g., Figures 9, 10 and 11).
Response 14: Figures 9 and 10 (10 and 11 in the revised version) belong to the results section. Only the last figure is presented in discussion.
Point 15:Lines 330 – 331, Figure 11: The use of only two subsets of data to demonstrate that the approach is robust and consistent in new situations is not particularly convincing. A larger sampling of the segmented landscape (with consistently high statistics) would be more convincing, if it is possible. Additional figures would not be needed, but statistics highlighting the stability of the R2 statistics across additional areas are needed to make the authors’ assertion of robustness in new areas convincing.
Response 15: Five additional samples were taken and the CA was run for all samples with the same parameters (delineations for four areas are shown in Figure 12). The results suggest that the R2 statistics are consistently high, and the CA works well in different places with the same parameters
Point 16:After reading through the discussion, I am especially curious to hear more of the authors’ thoughts on whether their approach could be applied to larger spatial extents, although perhaps this is a topic best addressed in another study.
Response 16: Since we did not analyze very large areas we are unwilling to add discussion about this issue. However, since the CA is based on local improvement, the computing time is expected to be linearly related to the size of the grid (not exponentially as in some optimization methods).
Reviewer 2 Report
Please, find my comments in the attached pdf.
Comments for author File: 
Comments.pdf
Author Response
Reviewer 2 noted several minor writing mistakes. They were corrected.
Point 1: You are doing segmentation. I would name the used method Markov-random-Field. Please, compare the CA to MRF and the show the benefit of the CA.
Response 1: The method has been called cellular automaton in all the references mentioned in our study. Therefore, we prefer to keep that name. Markov-Random-Field is somewhat different. For example, it includes a training stage, which is not a part of the cellular automaton. However, the MRF method (or its modification) could most probably be also used for automated stand delineation, similarly as for instance self-organizing maps or some metaheuristics. This is a good topic for future research.
Point 2: This seems to be a learning or training step. You should name it.
Response 2: This was only standardization of variables. No training aspect was included in this step. The only purpose was to remove the effect of different units of different variables in the calculation of Euclidean distance.
Point 3 How do you get these numbers (R2)? Please, explain.
Response 3: It is now explained how the R2 statistics were computed.
