Crown Structure Metrics to Generalize Aboveground Biomass Estimation Model Using Airborne Laser Scanning Data in National Park of Hainan Tropical Rainforest, China
and Huaguo Huang 2Round 1
Reviewer 1 Report
General comments:
This study compared random forest and MLR models for estimating AGB using ALS data. The authors found that the random forest model outperformed than the MLR models. In general, the paper is well written. The paper provides detail analysis on how to derive relevant variables/metrics from the ALS data. However, the regression analysis could be further improved, and further discussion on the results could be added.
Specific comments:
L129-139: Please explain why some field plots were located outside of the flight routes. Are those plots serve different purposes for the data analysis (e.g. for testing/validation of the models)?
L225: It is written as "Multivariate step..". It should be written as "Multiple stepwise..."
L232-234: It is not statistically appropriate to select a best model in MSLR based on coefficient of determination (R2), because the R2 value will always increase when you add an independent variable. Instead, you must use adjusted R2 that takes into account differences in degree of freedom (df).
Please see some statistical books on regression analysis for further explanation on R2 and R2adj; or you may read this post:
DIFFERENCE BETWEEN ADJUSTED R-SQUARED AND R-SQUARED
=> https://www.listendata.com/2014/08/adjusted-r-squared.html
Altervatively, you may also use AIC and/or BIC criterion for MLR models selection.
L268-270: It not surprising if the height metrics are strongly correlated with AGB, because you also calculated trees' AGB by using allometric models that use height variables.
L282: "Besides,...." what? It is incomplete sentence, isn't it?
L288: Does it mean that R2 would be >1 if the ntree is large? I do not think so, because statistically R2 is always <1 regardless of the number of observation (n).
L308-311: It is unfair to use R2 to compare MLRs whose different model parameters. See my comments on L232-234.
L316 (Table 8). In MLR, multicolinearity among predictor variables might exist, resulting in some problems (e.g., the MLR coefficients can swing, reduced precision of the coefficient estimates). You have to verify this essential MLR assumption, and then confirm it with relevant statistics (e.g., Durbin Watson statistic).
L337: It should be written as "...are greater than..."
L395-399: The sentence is too long and complex. You should revise it.
L410-414: "In terms of regression strategies..." seems to be redundant with L431-433 "In terms of regression...". You could merge the discussions on alternative machine learning and spatial regression methods into one paragraph.
Figure 6 and 7 showed that both the random forest and MLR models produced biases when AGB >200 t/ha. You need to discuss these results and provide some possible reasons.
L395-409: To make a balance discussion, please also discuss what would be a potential drawback of the random forest model (or other machine learning methods) compared to the classical parametric regression models. You may also discuss what would be practical issues when a user wish to use your random forest models for estimating the AGB.
L440-449: What are relevant conclussions on the results of the random forest and MLR models?
Author Response
Dear reviewers,
Thank you very much for your reviewing our manuscript entitled with “Crown Structure Metrics to Generalize Aboveground Biomass Estimation Model Using Airborne Laser Scanning Data in National Park of Hainan Tropical Rainforest, China”. We are very appreciated for your constructive advice and comments. In the revised manuscript, we have carefully considered this advice. The modifications are as follows:
- L129-139: Please explain why some field plots were located outside of the flight routes. Are those plots serve different purposes for the data analysis (e.g. for testing/validation of the models)?
Answer: We added the reason why some field plots were set outside of the study area. We are sorry that we didn't give sufficient consideration to showed the flight trajectories inside of the study area in previous version manuscript. We repainted Figure 1 and added all flight trajectories of covering whole field plots inside of the study area in the figure.
- L225: It is written as "Multivariate step..". It should be written as "Multiple stepwise..."
Answer: We corrected the expressions in the full text.
- L232-234: It is not statistically appropriate to select a best model in MSLR based on coefficient of determination (R2), because the R2value will always increase when you add an independent variable. Instead, you must use adjusted R2 that takes into account differences in degree of freedom (df). Altervatively, you may also use AIC and/or BIC criterion for MLR models selection.
Answer: We added equation of adjusted R2 and calculated adjusted R2 for MSLR in Results.
- L268-270: It not surprising if the height metrics are strongly correlated with AGB, because you also calculated trees' AGB by using allometric models that use height variables.
Answer: We revised the descriptions.
- L282: "Besides,...." what? It is incomplete sentence, isn't it?
Answer: We are sorry that we didn't examine carefully. We deleted “Besides,”.
- L288: Does it mean that R2would be >1 if the ntree is large? I do not think so, because statistically R2 is always <1 regardless of the number of observation (n).
Answer: We are sorry that we wrote it wrong in previous text. We corrected “R2 < -1”.
- L308-311: It is unfair to use R2to compare MLRs whose different model parameters. See my comments on L232-234.
Answer: We added the value of adjusted R2 in Results.
- L316 (Table 8). In MLR, multicolinearity among predictor variables might exist, resulting in some problems (e.g., the MLR coefficients can swing, reduced precision of the coefficient estimates). You have to verify this essential MLR assumption, and then confirm it with relevant statistics (e.g., Durbin Watson statistic).
Answer: We recalculated MSLR model, and used Durbin-Watson test and adjusted R2 to evaluated the MSLR performance.
- L337: It should be written as "...are greater than..."
Answer: We corrected it.
- L395-399: The sentence is too long and complex. You should revise it.
Answer: We revised this sentence.
- L410-414: "In terms of regression strategies..." seems to be redundant with L431-433 "In terms of regression...". You could merge the discussions on alternative machine learning and spatial regression methods into one paragraph.
Answer: We merged two paragraphs.
- Figure 6 and 7 showed that both the random forest and MLR models produced biases when AGB >200 t/ha. You need to discuss these results and provide some possible reasons.
Answer: We added analysis about it.
- L395-409: To make a balance discussion, please also discuss what would be a potential drawback of the random forest model(or other machine learning methods) compared to the classical parametric regression models. You may also discuss what would be practical issues when a user wish to use your random forest models for estimating the AGB.
Answer: We added analysis and discussions of random forest model.
- L440-449: What are relevant conclusions on the results of the random forest and MLR models?
Answer: We added the results about random forest and MLR in Conclusions.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Review manuscript#1730797
In the current article, an attempt is made to estimate AGB in southern China, using random forest and multivariate stepwise linear regression along with different variables extracted from ALS point cloud. Although interesting, there are some serious problems that need to be addressed. Those are referred below.
General comments
-The structure of the manuscript is complex for a reader to follow. There are many repetitions, unfinished sentences, and unrelated parts that need to be transferred.
-The language style needs to be corrected.
-Statistical issues. Some parts need further analysis or changes in order to support the conclusions.
Specific comments
L14-16 It is unclear what the authors mean. Please revise.
L26 The aboveground Biomass (AGB) is already mentioned in the title.
L32 Please replace (,) with “and the” before greenhouse.
L37-43 This part is not relevant to the upper part and it better suits to the Study Area section.
L45-47 It is unclear what the authors mean. Please revise.
L59 “In Ghana” looks like a separate part. Please revise.
L57-79 This part belongs to the discussion, due to the high level of detail.
L87-92. In this last paragraph, the aims of the study must be mentioned. I suggest the authors revise accordingly.
L96 It is already mentioned in L37.
L101-103 It is unclear what the authors mean. Please revise.
Figure 1. The color inside the study area’s polygon is not solid. Please correct it.
L109-110 This information is not relevant. Please remove it.
L127 Li et al. must be numbered.
L130 The “investigation” word is not correct. Please revise.
L131-133 It is unclear what the authors mean. Please revise.
L135 Which laser altimeter?
L136 Is it caliper instead?
L153 Who “they”?
L154 Crown width is better suited.
L172 “They can be used…[]”
L179. Please delete the dot before “but”.
L180-181 It is unclear what the authors mean. Please revise.
L220 It is unclear what the authors mean. Please revise.
L226 It is unclear what the authors mean. Please revise.
L231 The alpha 2, alpha 3 …. are the model’s parameters and the alpha 1 is the intercept.
L273 It is unclear what the authors mean. Please revise.
L282 Besides…..[] Unfinished phrase. Please revise.
L286 200 of what? Iterations perhaps or something else?
L288 R2 is always smaller than 1. Isn’t it evident?
L297 Please, avoid any reference in the results section.
Table 8 In order to retain a variable in the model, the associate parameter must be significant according to the t-statistic. Otherwise, it must be dropped from the model. This is not clear and it must be clarified. The discussion must be changed accordingly.
Figure 6. Despite the fact that it is common to be used in this kind of comparison, the R2 is not the correct term, simply because it is not a linear regression rather than a line of a perfect fit. Instead, the modeling efficiency (EF) index can be used (please see 10.1016/0304-3800(93)90105-2).
Author Response
Dear reviewers,
Thank you very much for your reviewing our manuscript entitled with “Crown Structure Metrics to Generalize Aboveground Biomass Estimation Model Using Airborne Laser Scanning Data in National Park of Hainan Tropical Rainforest, China”. We are very appreciated for your constructive advice and comments. In the revised manuscript, we have carefully considered this advice. The modifications are as follows:
- L14-16 It is unclear what the authors mean. Please revise.
Answer: We revised this sentence. Maybe the comma before “therefore, ……” caused ambiguity.
- L26 The aboveground Biomass (AGB) is already mentioned in the
Answer: We replaced a keyword.
- L32 Please replace (,) with “and the” before greenhouse.
Answer: We corrected it.
- L37-43 This part is not relevant to the upper part and it better suitsto the Study Area section.
Answer: We revised this part and rearranged the part of study area introduction.
- L45-47 It is unclear what the authors mean. Please revise.
Answer: We revised the sentence, it repeated with previous one. We deleted the previous one.
- L59 “In Ghana” looks like a separate part. Please revise.
Answer: We are sorry that we wrote it wrong in previous text. We corrected “In Ghana” to “in Ghana”.
- L57-79 This part belongs to the discussion, due to the high level of
Answer: We didn't correct this part, the paragraph only introduced some typical previous researches of AGB estimation with LiDAR, instead of using it to compare our study results.
- L87-92. In this last paragraph, the aims of the study must be I suggest the authors revise accordingly.
Answer: We added description of the main aim.
- L96 It is already mentioned in L37.
Answer: We rearranged the overview of the study area.
- L101-103 It is unclear what the authors mean. Please revise.
Answer: The sentence used to describe dominant tree species in the study area, we revised this sentence.
- Figure 1. The color inside the study area’s polygon is not solid.Please correct it.
Answer: We redrew the Figure 1.
- L109-110 This information is not relevant. Please remove it.
Answer: We removed the sentence.
- L127 Li et al. must be numbered.
Answer: We moved the citation number after the author name.
- L130 The “investigation” word is not correct. Please revise.
Answer: We corrected the word.
- L131-133 It is unclear what the authors mean. Please revise.
Answer: We revised the sentence.
- L135 Which laser altimeter?
Answer: We added the model of laser altimeter.
- L136 Is it caliper instead?
Answer: We revised the sentence. We used diameter tape.
- L153 Who “they”?
Answer: We revised this paragraph and moved to the Introduction part.
- L154 Crown width is better suited.
Answer: We corrected it to “crown width”.
- L172 “They can be used…[]”
Answer: We corrected to “They can be used ……”.
- Please delete the dot before “but”.
Answer: We deleted the dot.
- L180-181 It is unclear what the authors mean. Please revise.
Answer: We deleted the sentence.
- L220 It is unclear what the authors mean. Please revise.
Answer: We revised.
- L226 It is unclear what the authors mean. Please revise.
Answer: We revised.
- L231 The alpha 2, alpha 3 …. are the model’s parameters and thealpha 1 is the intercept.
Answer: We are sorry that we missed the error term description in previous text, rather than the reviewer’s advice.
- L273 It is unclear what the authors mean. Please revise.
Answer: We revised.
- L282 Besides…..[] Unfinished phrase. Please revise.
Answer: We are sorry that we didn't examine carefully. We deleted “Besides,”.
- L286 200 of what? Iterations perhaps or something else?
Answer: We revised.
- L288 R2 is always smaller than 1. Isn’t it evident?
Answer: We are sorry that we wrote it wrong in previous text. We corrected to “R2 < -1”.
- L297 Please, avoid any reference in the results section.
Answer: We moved the citations.
- Table 8 In order to retain a variable in the model, the associateparameter must be significant according to the t-statistic. Otherwise, it must be dropped from the model. This is not clear and it must be clarified. The discussion must be changed
Answer: We recalculated MSLR model, and used Durbin-Watson test and adjusted R2 to evaluated the MSLR performance.
- Figure 6. Despite the fact that it is common to be used in this kindof comparison, the R2 is not the correct term, simply because it is not a linear regression rather than a line of a perfect fit. Instead, the modeling efficiency (EF) index can be used.
Answer: We added adjusted R2 to access performance of the AGB regression models.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors have revised the manuscript following my previous comments. However, I have further comments and suggestions as follows:
L41-43: "Although ...with..." seems does not fit into this sentence. What the traditional survey do you mean? How it can destruct the forests? Please revise this sentence to make a clear meaning! The forest biomass estimation is commonly conducted by using allometric biomass equations, hence it will not destroy the forests. Of course, when you develop an allometric biomass equation then you need to cut some sample trees.
L244-245: The variables explanation should be put directly underneath the Eq. (3) - (6).
Table 8: Why do the MLR model of the alpha-shape metrics is the same as that of the mixed metrics, including the DW and adjusted R2 values but with different F.Statistics?
Anyway, the F.Statistics is difficult to interpret when readers do not hold the F-distribution table. Instead, I suggest the authors to use P-value to facilitate easy interpretation whether the models are statistically significant or not.
Author Response
Dear reviewers,
Thank you very much for your reviewing our manuscript entitled with “Crown Structure Metrics to Generalize Aboveground Biomass Estimation Model Using Airborne Laser Scanning Data in National Park of Hainan Tropical Rainforest, China”. We are very appreciated for your constructive advice and comments, and we are sorry to delay some time. In the revised manuscript, we have carefully considered this advice. The modifications are as follows:
- L41-43: "Although ...with..." seems does not fit into this sentence. What the traditional survey do you mean? How it can destruct the forests? Please revise this sentence to make a clear meaning! The forest biomass estimation is commonly conducted by using allometric biomass equations, hence it will not destroy the forests. Of course, when you develop an allometric biomass equation then you need to cut some sample trees.
Answer: We corrected the problematic statements in this sentence.
- L244-245: The variables explanation should be put directly underneath the Eq. (3) - (6).
Answer: We are sorry about the layout fault in previous version. We replenished the variables explanation underneath the equations.
- Table 8: Why do the MLR model of the alpha-shape metrics is the same as that of the mixed metrics, including the DW and adjusted R values but with different F.Statistics? Anyway, the F.Statistics is difficult to interpret when readers do not hold the F-distribution table. Instead, I suggest the authors to use P-value to facilitate easy interpretation whether the models are statistically significant or not.
Answer: We recalculated and examined the statistical results, and we removed the F.Statistics and replenished P-value in Table 8.
Author Response File: Author Response.docx
Reviewer 2 Report
The authors followed all the recommended changes and the manuscript improved significantly.
Author Response
Dear reviewers,
Thank you very much for your reviewing our manuscript entitled with “Crown Structure Metrics to Generalize Aboveground Biomass Estimation Model Using Airborne Laser Scanning Data in National Park of Hainan Tropical Rainforest, China”. We are very appreciated for your constructive advice and comments in previous version, and we are sorry to delay some time to reply. Thank you again for your comments on our modifications of previous version manuscript.
Author Response File: Author Response.docx
