Configuration of the Deep Neural Network Hyperparameters for the Hypsometric Modeling of the Guazuma crinita Mart. in the Peruvian Amazon
, Duberlà Geomar Elera Gonzáles 2, Juan Rodrigo Baselly Villanueva 3, Leonardo Pereira Fardin 1 and Hélio Garcia Leite 1Round 1
Reviewer 1 Report
Journal: Forests (ISSN 1999-4907)
Manuscript ID: forests- 1681177
Title: Configuration of the deep neural network hyperparameters for the hypsometric modeling of the Guazuma crinita Mart. in the Peruvian Amazon
Overall Comments and Suggestions for Authors
Dear author,
Regarding the artificial neural networks and deep learning for height estimation in forestry field, this manuscript may be interesting to the relevant researchers who deals with similar issues such as forest biometrician, growth and yield modeler, and silviculturist. I consider the overall structure is fine and most contents were explained well. Still, some parts can be improved by adding texts or describing more. To improve the manuscript contents, I made some comments as below. I wish the developed models in this study will serve for managing forest stands of the region.
I hope that this manuscript can be improved based on peer-review’s comments. My specific comments were provided in detail as follows.
Kind regards,
Reviewer
Point 1.
In Introduction, as authors mentioned and know about this approach, ANN and deep learning approaches are relatively new compared to the regression. If possible, why don’t authors point out the previous studies in which the former method and ANN were compared in terms of forestry field? If it’s hard to find the same type of modelling, it does not have to be the height estimation, though. This could give more justification and attention on the manuscript although authors did not directly compare the models with regression in the manuscript.
Point 2.
In Materials and Method, it was not so clear to me that how authors applied the agroclimatic variables in the model type 3. Authors mentioned many agroclimatic variables in Methods section, but it’s still not so clear. In the final fitting process, did authors pick one of those? Or was there a specific formula to calculate Agroclimatology? Or did you use all of the agroclimatic variables in the model fitting?
It should be explained clearly. Also, if recommended, consider adding these variables in the summary statistics in Table 1 as shown in Age, DBH, and Height.
Point 3.
In Table 1 and Results, the decimal points of each variable were significant or meaningful? I think the expression can be followed by the instrument or dependent on the measurement. If so, four digits of decimal point may not be needed. Consider revising it in Table 1.
Point 4.
In Results and Figure 4, the bias of the developed models must be presented over predicted in residual plots. Also, the independent variables such as DBH, Age, and Agroclimatology can be checked using residual plots if it’s biased.
As a supplementary option, the dot transparency can be applied to see and compare easily between the model types.
Point 5.
In Discussion, it would be better for authors to discuss about the practicability or applicability of the developed models for forestry field. It can be about the spatial or temporal range of the input data. Or authors may be able to mention the extrapolation of applying the developed models.
Point 6.
Line 275. The subtitle “4.3. Growth and estimation of the total height of Bolaina Blanca”.
Shouldn’t it be 4.2? or did you miss 4.2 section in the manuscript?
Author Response
Sunday, April 17, 2022
Forests Journal
Dear reviewer;
We are very grateful for reviewing our manuscript entitled “Configuration of the deep neural network hyperparameters for the hypsometric modeling of the Guazuma crinita Mart. in the Peruvian Amazon” for publication in Forests journal.
The objective of this study was to build and evaluate the efficiency of the use of a deep neural network for the prediction of the total height of the Guazuma crinita Mart from a large-scale continuous forest inventory. To do this, we explore different configurations of the hidden layer hyperparameters and define the variables according to the function HT=f(x) where HT is the total height as the output variable and x is the input variable(s). We include dendrometric and climatological variables. Our contribution is to the scientific community, farmers, and companies dedicated to forest modeling, specifically located in Amazonian areas.
We appreciate your comments and questions to improve our scientific article and have answered your questions below. We have added the content in red so that the reviewer can find it in the manuscript quickly.
We appreciate your consideration.
Sincerely,
Authors
Overall Comments and Suggestions for Authors
Dear author,
Regarding the artificial neural networks and deep learning for height estimation in forestry field, this manuscript may be interesting to the relevant researchers who deals with similar issues such as forest biometrician, growth and yield modeler, and silviculturist. I consider the overall structure is fine and most contents were explained well. Still, some parts can be improved by adding texts or describing more. To improve the manuscript contents, I made some comments as below. I wish the developed models in this study will serve for managing forest stands of the region.
I hope that this manuscript can be improved based on peer-review’s comments. My specific comments were provided in detail as follows.
Kind regards,
Reviewer
Point 1.
In Introduction, as authors mentioned and know about this approach, ANN and deep learning approaches are relatively new compared to the regression. If possible, why don’t authors point out the previous studies in which the former method and ANN were compared in terms of forestry field? If it’s hard to find the same type of modelling, it does not have to be the height estimation, though. This could give more justification and attention on the manuscript although authors did not directly compare the models with regression in the manuscript.
RE: Thank you for your observation. We have included this item on line 67 of the manuscript.
Point 2.
In Materials and Method, it was not so clear to me that how authors applied the agroclimatic variables in the model type 3. Authors mentioned many agroclimatic variables in Methods section, but it’s still not so clear. In the final fitting process, did authors pick one of those? Or was there a specific formula to calculate Agroclimatology? Or did you use all of the agroclimatic variables in the model fitting?
It should be explained clearly. Also, if recommended, consider adding these variables in the summary statistics in Table 1 as shown in Age, DBH, and Height.
RE: Thank you for your questions. We have specified this term on line 119 (Table 1) and 127 of the manuscript. For more information on the method of acquisition of agroclimatic variables, please visit the following web access: https://power.larc.nasa.gov/docs/methodology/
Point 3.
In Table 1 and Results, the decimal points of each variable were significant or meaningful? I think the expression can be followed by the instrument or dependent on the measurement. If so, four digits of decimal point may not be needed. Consider revising it in Table 1.
RE: Thank you for your suggestion. We have corrected it.
Point 4.
In Results and Figure 4, the bias of the developed models must be presented over predicted in residual plots. Also, the independent variables such as DBH, Age, and Agroclimatology can be checked using residual plots if it’s biased.
As a supplementary option, the dot transparency can be applied to see and compare easily between the model types.
RE: Thank you for your comment. The graph was modified on line 291 (Figure 4) of the manuscript. We have decided to show the residuals separately. We consider this way better because it can be seen in more detail and the reader can evaluate the RE% according to the functions and models.
Point 5.
In Discussion, it would be better for authors to discuss about the practicability or applicability of the developed models for forestry field. It can be about the spatial or temporal range of the input data. Or authors may be able to mention the extrapolation of applying the developed models.
RE: Thank you for your observation. We have included this item on line 328 of the manuscript.
Point 6.
Line 275. The subtitle “4.3. Growth and estimation of the total height of Bolaina Blanca”.
Shouldn’t it be 4.2? or did you miss 4.2 section in the manuscript?
RE: Thank you for your observation. We have corrected it.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This study aims to build and evaluate the efficiency of the use of a deep neural network for the prediction of the total height of Guazuma crinita Mart from a large-scale continuous forest inventory. The different configurations of the hidden layer hyperparameters are explored. This research is relevant and interesting. However, some revisions may be needed before further processing. There are some shortcomings in the writing.
DETAILED COMMENTS:
(1) In abstract: numbers need to be accurate to two decimal places, and the model with best performance should be highlighted.
(2) The objective of the study is too broad, and specific research objectives should be added.
(3) Detailed description of the DBH and tree height measurement should be added to the method. Moreover, what's the size of the sample plots? Is it a circle or square? And what's the distribution of them? Please add their distribution to the figure 1.
(4) What is the time period of the agroclimatic variables? Is it consistent with the field measured data? After all, the time span of ground survey is so long.
(5) Some contents of method are placed in the result, such as the training status. Please modify them.
(6) In results: a deeper analysis should be added, not only the accuracy performance of models. How efficient are the models? How long is the calculation time? Moreover, a sensitivity analysis should be conducted to find the key variables and the influence of the layers to model performance.
Author Response
Sunday, April 17, 2022
Forests Journal
Dear reviewer;
We are very grateful for reviewing our manuscript entitled “Configuration of the deep neural network hyperparameters for the hypsometric modeling of the Guazuma crinita Mart. in the Peruvian Amazon” for publication in Forests journal.
The objective of this study was to build and evaluate the efficiency of the use of a deep neural network for the prediction of the total height of the Guazuma crinita Mart from a large-scale continuous forest inventory. To do this, we explore different configurations of the hidden layer hyperparameters and define the variables according to the function HT=f(x) where HT is the total height as the output variable and x is the input variable(s). We include dendrometric and climatological variables. Our contribution is to the scientific community, farmers, and companies dedicated to forest modeling, specifically located in Amazonian areas.
We appreciate your comments and questions to improve our scientific article and have answered your questions below. We have added the content in red so that the reviewer can find it in the manuscript quickly.
We appreciate your consideration.
Sincerely,
Authors
This study aims to build and evaluate the efficiency of the use of a deep neural network for the prediction of the total height of Guazuma crinita Mart from a large-scale continuous forest inventory. The different configurations of the hidden layer hyperparameters are explored. This research is relevant and interesting. However, some revisions may be needed before further processing. There are some shortcomings in the writing.
DETAILED COMMENTS:
(1) In abstract: numbers need to be accurate to two decimal places, and the model with best performance should be highlighted.
RE: Thank you for your comment. We have corrected the remarks in the manuscript on line 26.
(2) The objective of the study is too broad, and specific research objectives should be added.
RE: Thank you for your observation. We have corrected the remarks in the manuscript on line 78.
(3) Detailed description of the DBH and tree height measurement should be added to the method. Moreover, what's the size of the sample plots? Is it a circle or square? And what's the distribution of them? Please add their distribution to the figure 1.
RE: Thank you for your comment. We have corrected the remarks in the manuscript on line 90 and 94. The distribution of the plots was added in Figure 1, line 97.
(4) What is the time period of the agroclimatic variables? Is it consistent with the field measured data? After all, the time span of ground survey is so long.
RE: Thank you for your comment. We have corrected the remarks in the manuscript on line 108.
(5) Some contents of method are placed in the result, such as the training status. Please modify them.
RE: Thank you for your comment. We have corrected the comments in the manuscript, the methodology was corrected in line 130 and the results were detailed in line 216.
(6) In results: a deeper analysis should be added, not only the accuracy performance of models. How efficient are the models? How long is the calculation time? Moreover, a sensitivity analysis should be conducted to find the key variables and the influence of the layers to model performance.
RE: Thank you for your suggestion. We have corrected the remarks in the manuscript on line 200, 214 and 286. The authors consider that it is not necessary to perform a sensitivity analysis, since the agroclimatic variables are few and easy to obtain. Besides several authors have used agroclimatic variables influencing the modeling, we specify in line 327 of the manuscript.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors have incorporated most of the modifications proposed, to improve their manuscript, which is highly appreciated. I recommend accepting and publishing in the Forests.
