Airborne Tree Crown Detection for Predicting Spatial Heterogeneity of Canopy Transpiration in a Tropical Rainforest

Round 1

Reviewer 1 Report

Considering the dense tree canopies in the study area, what is the accuracy of the DEM generated from the photogrammetric point cloud data? What’s the accuracy for GPS measurements? Considering the very high image resolution (7.5cm), how to use a relatively low GPS coordinates to geo-reference the UAV images? Some details should be included in the manuscript. In Page 5, line 154, the point cloud was reduced down from 198 to 58 square meter, but in line 165, some tree crowns were eliminated because of low density of points (<40). If I could say, you selected an improper or wrong soft application or made wrong processing of the original point data? Explanation should be added. The segmentation process of the tree crowns should be clarified more in detail in Section 3.2 due to the complexity of the dense crowns and vertically multiple layers in the tropical forest.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

No doubt, tree crown metrics are related with tree water use and a method that allows deriving transpiration from UAV-based data would be a helpful tool in environmental sciences.

However, the submitted manuscript does not bring any innovation in comparison to the journal article of the same authors, which is cited here: Ahongshangbam, J.; Khokthong, W.; Ellsäßer, F.; Hendrayanto, H.; Hölscher, D.; Röll, A. Drone-based 389 photogrammetry-derived crown metrics for predicting tree and oil palm water use. Ecohydrology 2019, 12; 390 DOI:10.1002/eco.2115.

Moreover, this manuscript contains number of serious errors and failings in experimental design, data processing and interpretation. Here I summarize the biggest errors and mystifications.

-----

Line 108: Explain why have you established four upland plots and utilized only one for sap flux measurement? The other three are completely useless, under such circumstances?

Moreover, you kept no plots for validation. Therefore accuracy of all the results is unknown. For example the maps of spatial heterogeneity (Figure 4) is pure expectation based on a model build on data of completely unknown accuracy.

Line 141: You mention here both manual and automatic crown detection. Further in the text (results or water use vs. crown metrics relationship development) you do not automatic and manual delineation again. What delineation was used? Why both delineations were carried out? What was the accuracy of automatic delineation in comparison with manual?

Line 171: What was the threshold distance?

Figure 3: you indicate that there is a difference in sap flux between riparian and upland plots. I suppose you should distinguish between the two categories in the plots (eventually create distinct plots for each category) and build a distinct model for each category. If there is a significant difference, the regressions are different as well.

You should also show how many trees were used in each zone to build the water use vs. crown surface relationship. If 42 trees from 72 were identified (58% average), then for upland plot from 15 might be identified 8-9 trees (?), what is simply to few to build a model.

Line 257: Three quarters of the not clearly identified trees belonged to the lower 75-percentile in terms of DBH -- this is exactly the same like: the not clearly identified trees were uniformly distributed in terms of DBH; therefore the probability of a tree not being identified is equal for dominant and suppressed trees. Please, do not manipulate and mystify readers.

Line 268: “This failure of detecting small statured trees is probably of  minor importance because such trees usually do not contribute much to the water exchange between the forest canopy and the atmosphere”. This is not true. Your model implies, that trees with smaller crowns have relative higher water use (water use per square meter) than trees with larger crowns. Therefore unidentified small trees have big impact on error.

Line 287: “the sum of individual over- or under-segmented crowns within the plot boundaries will inevitably equal stand crown surface area and thus the predictedstand transpiration value.” This is again a mystification. Since the model is tree-based, giving different water use per area unit for small and large trees, the size individual trees matters. If your statement “individual segmentation accuracy is not a constraint when assessing stand Et” is true, then there is no need to identify trees in aerial data and no need for the data.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

 

The manuscript:

 

Airborne tree crown detection for predicting spatial heterogeneity of canopy transpiration in a tropical rainforest

 

is an interesting material, well designed, with well presented results and described with good language (both in terms of understanding and quality of English). Congratulations!

There are only few comments in the main file to be considered before publication. Mainly, abstract and the objective should be corrected and some explanations given in the methods. Deeper explanation for undertaking the research would make the paper more solid (marked in the main file in Introduction).

Good luck with your further work!

 

With best wishes,

 

Reviewer

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

This manuscript is appropriate and makes a positive contribution to the "3D Forest Structure Observation" edition of Remote Sensing.  I only have minor edits and questions.

 

Line 99: More information should be provided on the study sites.  Particularly, distance between each site for statistical concerns. 

Line 109: I'm confused on how where 72 sap flux trees come from on line 143 when it shows 15 trees were selected per plot with the exception of HFr1 (n=12).  Please provide more clarification on how this number of sap flux trees was achieved.  It would also be beneficial to provide more information on why HFr1 only had 12 sap flux trees.  

Line 156: please provide more detail on pixel resolution. 

Line 265: Other than using balloons, were there any other methods attempted to better identify target crowns?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors provided extensive explanation of their method regarding the comments. In some cases, the changes in the text changed the meaning of the utilized methods. I try to believe that it was due to previous misformulation. But still there remain methodological issues that should not be present in a scientific work. I am not confident that this study should be published in Remote Sensing. Anyway, I suggest some points to improve the manuscript:

The main problem of the manuscript is as follows: As the authors state, their aim is to verify a possibility to link automatically derived canopy metrics with water-use data. Still, in the revised manuscript, there is not any clue about “testing” the relationship. There is a regression model linking real water-use measurement with metrics of manually delineated crowns. But the authors apply the model to automatically delineated crowns. Without any validation of either 1) metrics of automatically crowns or 2) Et derived from automatically delineated crowns. Any of these two ways would bring at least a little light into the current darkness completely hiding the accuracy of the method. So, at the current situation, the accuracy is completely unknown, it might be perfect as well as totally wrong. As I believe, a study CANNOT be published as a scientific work without ANY clue of validation or accuracy estimation. But there is a very simple way to do that, as I suggest below.

A few suggestions:

Figure 1: can you please provide the map with more distinct markers, preferably different marker shapes, for riparian vs. upland plots?

Line 170: As I understood, you used the manually delineated crowns of sap-flux-measured trees to build the model and the automatically delineated crowns of all trees for water-use and Et predictions. Because your study does not contain any validation (as authors confirm in their responses to the comments), you definitely should take this opportunity to validate (show the accuracy of) automatic crown delineation by comparing the 42 clearly visible crowns in automatic and manual delineation and provide the comparison (how the crown metrics differ, how the derived Et differs from manually vs automatically derived crowns).

If you admit that due to the low number of observation you join all the observations in one model, then you cannot state that there was a difference between the two forest types (abstract, line 26; discussion, line 282). In that case, the difference found in Et between riparian and upland plots (line 260) is due to different crown metrics in the two forest types (because you predict Et using one pooled regression model in both forest types)? You should mention it here: there was significant difference in crown metrics and THEREFORE there are differences in Et estimations based on the model.

Line 354. Sorry, I might misunderstand, but your newly added sentence does not really explain the indicated congruence with the previous sap flux measurement. Table A3, to which this sentence refers, does not show anything about the congruence, it compares Et estimations from drone-based and ground-based crown metrics. Please, provide a quantitative expression of the indicated congruence.

Author Response

Please see the attachment

Author Response File: Author Response.docx