Performance of Laser-Based Electronic Devices for Structural Analysis of Amazonian Terra-Firme Forests

Round 1

Reviewer 1 Report

The authors make the claim that 'laser-based electronic devices can be used to predict tree height in natural forests' (lines 467-468). However, this statement is not supported by the data. Here are the questions I had when reading the manuscript:

1- The reason why there are few papers on terrestrial Lidars in the tropics is because it is hard to use them due to occlusion. How representative is your study site in terms of occlusion (distance between trees and canopy cover)? 

2- Derive biomass error mathematically. For each allometric equation, we can estimate the delta biomass associated with each 1-m error in height (dAGB/dH). This can be used to estimate the error associated with different forest types (assuming different tree size distributions).

3- Accuracy vs. precision. Tree height estimates require a lot of interpretation. For range finder estimates, it depends on one's distance from the tree being measured and experience in detecting the top of the tree. For the TLS measurements, it depends on how the point cloud is processed (232-233). It would be useful to at least acknowledge that this study focused on accuracy and not precision, such that inter-individual variability is an extra source of error.

4-Make your goal clear. The introduction suggests the goal is to evaluate TLS in tropical forest inventories. However in the discussion the authors talk about other studies that were performed with single trees (lines 558-559, 490). These single-tree studies are very different studies with different objectives. Are you interested in improving tree allometric equations using single trees, or in improving biomass estimation with forest plots?

5-The error for height calculation is around 10% (Table 3) and > 25% for biomass calculation (Table 4). These results do not seem to match Figure 8. Furthermore model bias is not easy to see on Figure 7 (for example lower right plot). If the biomass error is >25%, how can we state that TLS 'provides a relatively low-cost technique for analyzing structural parameters in great detail'? By the way, there is no cost assessment being made in this paper (personnel, software, required time in the field) so I would dedicate another manuscript to describe the cost aspects.

Author Response

Response to Reviewer Comments #1

 

The authors make the claim that 'laser-based electronic devices can be used to predict tree height in natural forests' (lines 467-468). However, this statement is not supported by the data. Here are the questions I had when reading the manuscript:

 

General response: Thank you for your time and effort put into our study, in which we comparedmultiple methodological techniques for analysis of tropical vegetation structure (i.e. tree diameter, tree height and tree biomass). The aim of our study was to identify relative sources and magnitudes of measurement uncertainties associated with traditional forest inventory vs. novel laser-based techniques applied for structural analysis of tropical vegetation.

 

From our comparison, we conclude that considering the relatively good performance of the laser-based technique, i.e. systematic error was comparable for tree diameter (<11.1%) and tree height (<11.7%), while the random error was negligible (<3%), these novel techniques seem to be a suitable tool for predicting tree height in natural forests. In general, laser-based techniques were reported a technological alternative to more laborious manual measurements and strenuous tree climbing and have been proposed for local height–diameter allometries thus improving biomass estimates for tropical forests (Sullivan et al., 2018). To further clarify our statement, we have summarized our findings by concluding that laser-based electronic devices can and should be used to estimate tree height in tropical forests (lines 509-516). 

 

Based on our revised analysis we report that total error metrics among electronic devices were comparable (10.1% for TLS vs. 11.7% for electronic calliper) and that variation of estimated aboveground biomass was larger among commonly applied allometries (39.7-66.4%) than between traditional and novel measurement techniques (10.6-15.0%). Hence, we conclude that it is crucial to combine traditional and novel inventory techniques, when opting to improve local estimates of tropical aboveground biomass (line 609-625). 

 

1- The reason why there are few papers on terrestrial LiDAR in the tropics is because it is hard to use them due to occlusion. How representative is your study site in terms of occlusion (distance between trees and canopy cover)?

 

Response 1: Thank you for highlighting this important point. We agree that occlusion might be crucial when assessing terrestrial LiDAR in tropical forest ecosystems. We acknowledged the notion that occlusion of the laser beam might be problematic in tropical forests in lines 96-107and discussed this issue in lines 499-505.We now further elaborate on the relevance of occlusion in our study by stating that “strategies for sampling campaigns conducted across tropical and temperate forest plots, have been established that are capable to produce point clouds with a uniform point distribution and thus should allow inter-comparison of metrics between instruments, plots and over time (Wilkes et al., 2017).”and have included the study by Wilkes et al., 2017 in the list of references (lines 788-790). However, we want to stress that one key advantage of our study is, that the position of each tree was recorded prior to the laser scanning campaign (Fig. 1) and thus was used to identify each tree individual in the point cloud. Thereby we circumvent the problem of occlusion in our particular study. To highlight this more clearly we have added a sentence to the Methods section (lines 193-194?). 

 

We would like to add that our study site is representative ofAmazon terra-firme forest with a site mean canopy cover of 87-89% (dry/wet season), leaf area index of 5.3-6.2 m2/m2 (dry/wet season) and plot mean maximum tree height of 39.4 m and mean tree density (>10 cm DBH) of 708 individuals ha⁻¹. For comparison, foregoing studies conducted in the same region reported that stem density and aboveground biomass was representative of Amazon terra-firme forest (Higuchi 1998; Chambers et al., 2001; Vieira et al., 2004), witha closed canopyaveraging 35 m in height, and maximum tree height up to 55 m (Laurance et al., 2011). Tree density (>10 cm DBH) was 626 individuals ha−1, and 14.5% of the trees were found in the medium (30–49.9 cm DBH) size class, whereas 70.8% of aboveground biomass (AGB) were represented in small and medium size classes, and trees >60 cm DBH accounted for 16.7% of total AGB (Vieira et al., 2004), which accounted for 204.5 Mg C ha−1 based on allometry established by Silva et al. 2007 AGB(kgC)=2.7179*DBH^1.8774*0.584*0.485 (presented in Lima et al. 2012).

 

The study site situated in the Brazilian Amazon rainforest (ca. 70 km north of Manaus, Figure 1) is maintained by Brazil’s National Institute for Amazon Research (INPA in Manaus, Brazil) and represents one of the best-studied regions in the Central-Eastern Amazon (Lapola & Norby 2014). For instance, forest inventory data have been collected since the mid-1980s by the INPA Tropical Forestry group from three plots totaling 3 ha. Within each plot, all stems ≥10 cm DBH were tagged, mapped, and diameters were re-measured annually since the mid- 1980s. Furthermore, two transect plots (20m ×2,500 m (5 ha) each), were established in 1996 within the Jacaranda Project (a collaboration between INPA and Japan International Cooperation Agency, JICA). Hence, given the fact that the Amazon forest represents the largest continuous patch of rainforest on the planet, we believe that results presented in this study should be representative for at least central-eastern Amazonian tropical forest ecosystems. To further highlight this, we have revised the first part of the Methods section (lines 124-147).

 

2- Derive biomass error mathematically. For each allometric equation, we can estimate the delta biomass associated with each 1-m error in height (dAGB/dH). This can be used to estimate the error associated with different forest types (assuming different tree size distributions).

 

Response 2: We agree that it is useful to calculate this metric (biomass error per 1-m error in height) when comparing different forest types with different vegetation structure and tree size distributions. However, in our study we compare different measurement techniques conducted in the same forest plot and even based on the same tree individuals. To that end, we first compare error metrics among vertical and horizontal vegetation structure and then apply different allometric models based on these parameters, thus resulting in different estimates when based on different input variables (i.e. tree diameter, tree height, wood density). As a result, for some parameters investigated in this study, such as those resulting from biomass estimates based on allometric equations excluding height (i.e. Higuchi 1998 D) we would be unable to assess respective error metrics using the proposed method and thus prefer to keep the methodology and results presented in the analysis. Nonetheless, to further clarify which error metrics have been assessed (based on the recommendation of reviewer #2) we added some additional text to the data analysis section 2.4 (lines 297-304). 

 

3- Accuracy vs. precision. Tree height estimates require a lot of interpretation. For range finder estimates, it depends on one's distance from the tree being measured and experience in detecting the top of the tree. For the TLS measurements, it depends on how the point cloud is processed (232-233). It would be useful to at least acknowledge that this study focused on accuracy and not precision, such that inter-individual variability is an extra source of error.

 

Response 3: Thank you for pointing this out. We are aware that measuring tree height in tropical forests strongly depends on measurement method, i.e. tangent vs. sine method and experience among different observers (Larjavaara & Muller-Landau, 2013). We now elaborate on that issue in more detail in the revised manuscript (lines 499-516). The aim of our study was clearly to investigate systematic and random errors of laser-based electronic devices and not individuals operating these. We thus added some clarifying remarks to the main text (lines 291-295) highlighting the aspects of precision vs. accuracy by stating: “random errors are typically associated with statistical variability due to reproducibility of the measurement (i.e. precision); systematic errors typically refer to a statistical bias between a result and a “true” value (i.e. accuracy).”

 

4-Make your goal clear. The introduction suggests the goal is to evaluate TLS in tropical forest inventories. However in the discussion the authors talk about other studies that were performed with single trees (lines 558-559, 490). These single-tree studies are very different studies with different objectives. Are you interested in improving tree allometric equations using single trees, or in improving biomass estimation with forest plots?

 

Response 4: We apologize for not being clear on the goal of our study. We have revised the description of the study goals at the end of the introduction (l.120-126). We refer to single-tree studies only to clarify respective uncertainties associated with different structural vegetation parameters, i.e. horizontal plot structure (i.e. tree diameter vs. vertical plot structure (i.e. tree height). We added a clarifying statement to the first paragraph of the discussion section (lines 450-453), which now states: “In this study, we evaluate relative sources and magnitudes of uncertainty associated with measurements obtained from traditional forest inventory and remote sensing techniques and report respective error metrics for commonly surveyed vegetation parameters in order to derive tropical biomass estimates.”

 

5-The error for height calculation is around 10% (Table 3) and > 25% for biomass calculation (Table 4). These results do not seem to match Figure 8. Furthermore model bias is not easy to see on Figure 7 (for example lower right plot). If the biomass error is >25%, how can we state that TLS 'provides a relatively low-cost technique for analyzing structural parameters in great detail'? By the way, there is no cost assessment being made in this paper (personnel, software, required time in the field) so I would dedicate another manuscript to describe the cost aspects.

 

Response 5: Thank you for this valuable comment. It is correct that respective error metrics differ for tree diameter (Table 1), tree height (Table 2) and aboveground biomass (Table 3). As discussed in the main text this results from the fact that additive errors can either amplify or cancel out, such that negative deviations cancel out positive deviations and vice versa (lines 318-320). As a result, systematic errors within respective measurements can be close to zero, while total errors increase due to additive effects of random errors. For instance, we found that laser-based techniques on average overestimated AGB by 16.5% and 18.2% for FM and TLS due to a systematic overestimation of tree height compared to conventional tape measurements. On the other hand, AGB was underestimated by 4.3% and 2.1% for FM and TLS because of a systematic underestimation of tree diameters with increasing tree size class(lines 581-588). To further clarify this, we referred to the study by Chave et al., (2004), which highlighted that a 10% error in Ht and 5% in DBH can lead to a 21.6% error in AGB (lines 68-70). Based on the recommendation of reviewer #2, we have revised Figure 5andFigure 7in order to better represent model bias by switching x-axis and y-axis (Piñeiro et al., 2008) and further highlight differences among biomass estimates based on differences in parameterization considering tree height and/or wood density (Figure 8).

 

Eventually, although it was one of the initial ideas when drafting the manuscript it was not feasible to perform an adequate cost assessment among the investigated methods in this study and thus we have removed this aspect from the revised manuscript by omitting the term “low-cost” (lines 444-445), which now reads “At the local scale, terrestrial laser scanning (TLS) provides highly detailed information of structural vegetation parameters”. Nonetheless, we would be happy to dedicate another manuscript to describe the cost aspects, but in this study focus on the assessment of error metrics among inventory techniques. 

Author Response File: Author Response.docx

Reviewer 2 Report

The  manuscript entitled “Performance of laser-based electronic devices for structural analysis of Amazonian terra-firme forests” is a well-structured and coherent work aiming at assessment of the total error, bias and random error in the estimation of main structural variables (DBH, tree height) used in forest mensuration and subsequent estimates of above-ground biomass. The Authors have conducted a field survey using traditional forestry methods (tape measurement), “Field-map bundle” (electronic callipers and laser rangefinder), and terrestrial laser based scanning from multiple points within the survey plot. The results are presented in a clear, albeit somewhat lengthy, way. The results are valuable contribution to the assessment of the performance of TLS in forestry; with additional value when taking into account that the survey has been made in forest type where similar surveys have been, so far, rarely made (Amazonian terra-firme forest). The manuscript has no major flaws, although some technical improvements should be made, and one of them I find particularly important.

Namely, since the manuscript deals with the assessment of errors, the proper comparison of the observed vs. predicted is of principal importance. Therefore, I would strongly recommend to the Authors to revise their representation of the “Observed” vs. “Predicted” figures. Although it is not uncommon to find articles where the representation, as made by the Authors, is also used (observed on the x-axis and predicted in the y-axis), it has been demonstrated that is not the best way. According to Piñeiro et al. (2008), the Observed (in the y-axis) vs. Predicted (in the x-axis) representation should always be used.

 

Other than that, there are several minor issues which would need addressing (see specific comments).

 

Reference

Piñeiro, G.; Perelman, S.; Guerschman, J.P.; Paruelo, J.M. How to evaluate models: Observed vs. Predicted or Predicted vs. Observed? Ecol Model 2008, 216, 316-322.)

 

 

Specific comments

 

L145: “boarder” (Misspelled word?) “border”?

 

L174-L179: It seems that the notation of figures (a, b, c, d) has been mixed up. Please check it.

 

L191 “Figure 2a” It seems it should be “Figure 2d”.

 

L299 “11.6% and 11%” -> “11.6% and 11.0%”, or “12% and 11%”. Please check throughout the text and use number of decimals consistently.

 

L304-307: “observed (actual) DBH” I would suggest to use “observed (tape measured) DBH”, as it is more precise.

The use of “≥” in “10 ≥ DBH < 30 cm” is incorrect. It should be “10 cm ≤ DBH < 30 cm”. Also, please see my comment to L299 (if you use one decimal precision, 0.02 should be written as “0.0” or better “<0.1” Alternatively, all other values should have two decimal places.

 

L311-312: Taking into account my general comment on the presentation of the results and the recommendation to use “Observed (y-axis) vs. Predicted (x-axis)”, the regression coefficients discussed in this section may or may not be statistically different from 1. If the slopes are not significantly different from 1, there is no justification for their comparison. Furthermore, if slopes are not statistically different from 1 and intercepts from 0, there is no justification for the statement “indicating that the two TLS methods tended to overestimate larger diameters”. The observed deviation from slope=1 and intercept=0 may be the consequence of random error due to the low number of trees in the sample. Furthermore, it is not clear how the irregularity of stem was addressed with TSL, because it was addressed with TM (L152-154). If it was not addressed in data processing of TSL, one might suspect that the overestimation of DBH might be the result of the fact the average height of DBH measurement with TM is higher (due to trees with buttresses, L152-135) than the height of DBH measurement for TSL (L246).

 

L315: Consider swapping x and y axes (see general comment and “Observed “(y-axis) v.s. “Predicted” (x-axis) suggestion). Also, consider adding confidence interval around the regression lines.

 

L337: “diameter size” -> I guess it should be “height” instead of “diameter size”.

 

L347: Seem my comment to L315.

 

L360: “...; Table2).” -> I guess it should be “...; Figure 7).” instead.

 

L367: Please revise the use of “≥”. I would recommend “Height 10 – 20 m” and “Height 20 – 35 m”

 

L393: Please specify which TSL method was used for the estimation of DBH and whether TSL_Height or TSL_Length was used in biomass equations.

 

L435-490. This part refers to DBH, while L:491-492 seem to refer to height measurements! But there is no mention of the “height measurements” and it can only be deduced from the size of the error (in meters) and the “LR” in parenthesis. Please revise wording of the sentence by mentioning “height”.

 

L501: I wander how do you “derive” photon flux density (and at which level/strata) from TSL measurements? I would suggest dropping it.

 

L609-610: Please revise wording of the sentence (“... Higuchi et al 1998 no estimate directly the”). Also please explain the “correction factor” and how did you use it? Is this correction factor also present in the Higuchi et al DH equation? If yes, please elaborate in the main text as this is the reference equation.

Author Response

Response to Reviewer Comments #2

 

The manuscript entitled “Performance of laser-based electronic devices for structural analysis of Amazonian terra-firme forests” is a well-structured and coherent work aiming at assessment of the total error, bias and random error in the estimation of main structural variables (DBH, tree height) used in forest mensuration and subsequent estimates of above-ground biomass. The Authors have conducted a field survey using traditional forestry methods (tape measurement), “Field-map bundle” (electronic callipers and laser rangefinder), and terrestrial laser based scanning from multiple points within the survey plot. The results are presented in a clear, albeit somewhat lengthy, way. The results are valuable contribution to the assessment of the performance of TLS in forestry; with additional value when taking into account that the survey has been made in forest type where similar surveys have been, so far, rarely made (Amazonian terra-firme forest). The manuscript has no major flaws, although some technical improvements should be made, and one of them I find particularly important.

 

Namely, since the manuscript deals with the assessment of errors, the proper comparison of the observed vs. predicted is of principal importance. Therefore, I would strongly recommend to the Authors to revise their representation of the “Observed” vs. “Predicted” figures. Although it is not uncommon to find articles where the representation, as made by the Authors, is also used (observed on the x-axis and predicted in the y-axis), it has been demonstrated that is not the best way. According to Piñeiro et al. (2008), the Observed (in the y-axis) vs. Predicted (in the x-axis) representation should always be used.

 

Response 6: Thank you for the positive evaluation of our manuscript and for the valuable comments. According to your suggestions, we have changed the presentation of Figure 5and Figure 7, which are now depictingObserved (in the y-axis) vs. Predicted (in the x-axis) following Piñeiro et al. (2008). Furthermore, we have added an analysis based on reduced major axis regression (RMA),which should be applied when x and y are assumed to be symmetrical (cf Bohonak, 2004 http://www.bio.sdsu.edu/pub/andy/RMA.html).

 

Other than that, there are several minor issues, which would need addressing (see specific comments).

 

Specific comments

 

L145: “boarder” (Misspelled word?) “border”?

 Corrected.

 

L174-L179: It seems that the notation of figures (a, b, c, d) has been mixed up. Please check it.

 Corrected.

 

L191 “Figure 2a” It seems it should be “Figure 2d”.

 Corrected.

 

L299 “11.6% and 11%” -> “11.6% and 11.0%”, or “12% and 11%”. Please check throughout the text and use number of decimals consistently.

 Corrected.

 

L304-307: “observed (actual) DBH” I would suggest to use “observed (tape measured) DBH”, as it is more precise.

The use of “≥” in “10 ≥ DBH < 30 cm” is incorrect. It should be “10 cm ≤ DBH < 30 cm”. Also, please see my comment to L299 (if you use one decimal precision, 0.02 should be written as “0.0” or better “<0.1” Alternatively, all other values should have two decimal places.

Corrected.

 

L311-312: Taking into account my general comment on the presentation of the results and the recommendation to use “Observed (y-axis) vs. Predicted (x-axis)”, the regression coefficients discussed in this section may or may not be statistically different from 1. If the slopes are not significantly different from 1, there is no justification for their comparison. Furthermore, if slopes are not statistically different from 1 and intercepts from 0, there is no justification for the statement “indicating that the two TLS methods tended to overestimate larger diameters”. The observed deviation from slope=1 and intercept=0 may be the consequence of random error due to the low number of trees in the sample. Furthermore, it is not clear how the irregularity of stem was addressed with TSL, because it was addressed with TM (L152-154). If it was not addressed in data processing of TSL, one might suspect that the overestimation of DBH might be the result of the fact the average height of DBH measurement with TM is higher (due to trees with buttresses, L152-135) than the height of DBH measurement for TSL (L246).

First of all, thank you again for this valuable recommendation, which we have considered in the revised manuscript (by adapting the methods section and display of Fig. 5and Fig. 7). We agree that differences among methods due to irregularity of tree stems have to be addressed by respective methods, which is why we chose to present respective results for different size classes (i.e. 10 cm ≤ DBH < 30and all sampled trees), showing that error metrics among measurement techniques are larger when considering larger diameter trees (due to trees with buttresses). However, we further want to stress that one key advantage of our study was that the position of each tree was recorded prior to the laser scanning campaign (Fig. 1) and thus could be used to identify each tree individual in the point cloud. Thereby we circumvent the problem of occlusion in our particular study. To highlight this more clearly we have added a sentence to the Methods section (lines 193-194).Furthermore, we have added a description of the so-called “stem curve command”, which was used to extract diameters from the point cloud at a specific height, i.e. at 0.65 m, 1.3 m, 2 m from the ground (lines 254-260) in order to assess difference among traditional inventory and novel laser-based techniques. 

 

L315: Consider swapping x and y axes (see general comment and “Observed “(y-axis) v.s. “Predicted” (x-axis) suggestion). Also, consider adding confidence interval around the regression lines.

Corrected.

 

L337: “diameter size” -> I guess it should be “height” instead of “diameter size”.

Corrected.

 

L347: Seem my comment to L315.

Corrected.

 

L360: “...; Table2).” -> I guess it should be “...; Figure 7).” instead.

Corrected.

 

L367: Please revise the use of “≥”. I would recommend “Height 10 – 20 m” and “Height 20 – 35 m”

Corrected.

 

L393: Please specify which TSL method was used for the estimation of DBH and whether TSL_Height or TSL_Length was used in biomass equations.

We used DBH_RHT and TLS_Heightto estimate biomass and adapted the description of the figure legend accordingly (line 407-411): “Figure 8. Bar chart showing estimates of aboveground biomass based on different methods: tape measurement (TM), Field-Map bundle (FM), Terrestrial LiDAR (TLS), to calculate estimates of aboveground biomass (using DBH_RHTand TLS_Height) and based on different allometric equations (listed in Table A1), and assessed vegetation parameters such as diameter (D in cm), height (H in m) and wood density (ρ in g cm-3).”

 

L435-490. This part refers to DBH, while L:491-492 seem to refer to height measurements! But there is no mention of the “height measurements” and it can only be deduced from the size of the error (in meters) and the “LR” in parenthesis. Please revise wording of the sentence by mentioning “height”.

As indicated by respective sub-header titles the paragraph starting line 306refers to “horizontal vegetation structure” (tree diameter), while the paragraph starting line 349refers to “vertical vegetation structure” (tree height). To make this more clear, we added this information to lines 522-525, which now reads “A more robust analysis of natural eucalyptus forest by Calders et al. [45]compared laser based estimates of Ht to measurements from trees felled after completing the scan, and associated error metrics showed that TLS estimates were actually more accurate than conventional methods (RMSE 0.55, 1.28 m, for TLS and LR respectively).”

 

L501: I wonder how do you “derive” photon flux density (and at which level/strata) from TSL measurements? I would suggest dropping it.

Wang et al. 2016 proposed an approach to estimate photon flux density in Populus sp. based on TLS measurements. The authors use a portable photosynthesis system and specific software to extract characteristics, such as crown structure, branch traits, and crown light distribution from TLS point clouds. (Wang et al., 2016). However, given the fact that we only wanted to exemplify additional parameters potentially available from TLS we removed this part from the text (lines 531-534), according to your suggestion.

 

L609-610: Please revise wording of the sentence (“... Higuchi et al 1998 no estimate directly the”). Also please explain the “correction factor” and how did you use it? Is this correction factor also present in the Higuchi et al DH equation? If yes, please elaborate in the main text as this is the reference equation.

Higuchi et al use a correction factor to estimate dry biomass from destructively harvested trees. The authors state that approximately 49% of the fresh weight of a trunk is water. Therefore, Goodman et al. (2014), suggest to multiply estimates from Higuchi et al 1998 by 0.6028 (1 - mean moisture content). We now added this study to the references (lines 864-865) as well as legend of Table A1 (lines 647-650) stating “Equations from Higuchi et al. 1998 directly estimate dry biomass based on field measurements of destructively harvested “fresh biomass” (W) and applying a correction factor of 0.6028[70]” and furthermore elaborate this finding in the methods section (lines 301-304), which now states “The local equation proposed by Higuchi et al. [69]was used as reference for calculating errors metrics among biomass estimates resulting from different allometric relationships evaluated in this study. This equation is based on fresh weight from destructively sampled trees and applying a correction factor of 0.6028 (1 - mean moisture content) as has been presented by Goodman et al. [70]”.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The Authors have amended the manuscript taking into account and correctly addressing all the points that I have raised with the first review. In my opinion the manuscript, as is, brings a valuable contribution to the emerging segment of application of the terrestrial LiDAR in forestry. I have no further objections to the publication of the manuscript.