Quantifying the Variation in Reflectance Spectra of Metrosideros polymorpha Canopies across Environmental Gradients

Round 1

Reviewer 1 Report

This paper employs airborne imaging spectroscopy and light detection and ranging to evaluate spectral variation in Metrosideros polymorpha across six locations of varying soil substrate age and elevational gradient on the big island of Hawaii. It is a well written and interesting study that employs solid remote sensing collection and processing methodology. I’ve included minor comments below.

My one major comment is regarding the lack of physical trait data from the study sites directly to pair with the remotely sensed data for validation. The authors use calibrations generated from a previous (highly cited) paper (from the same authors), “Quantifying forest canopy traits: Imaging spectroscopy versus field survey” (Asner et al., 2015), which took place in Peru, but do not have validation samples to ensure the applied calibrations are accurate and appropriate in the context of their study in review. After reading through the referenced paper, I pose a few questions and comments to the authors:

1.      In the current paper (in review) you make a solid case for why studying intra-trait variability is important. Since the calibrations applied in this study to estimate foliar traits are not from the same region, can you comment on the influence this would have on your results? Is the species of interest in the present study one of the species that was sampled in the Asner et al., 2015 paper?

 

2.      The focus of the paper in review is on assessing intra-species traits across different locations, so I can see why an exact (highly accurate) measure of foliar traits are not necessarily needed to reach this aim, especially as many of the comparisons are made qualitatively. However, would the authors please comment on the accuracy of your estimates for these characteristics across the study sites (such as in Figure 6 and Table SI4)?

 

-          In consulting with Tables 2 and 3 in the Asner et al., 2015 paper (for which the calibrations are sourced) several traits of interest (potassium, phenols, magnesium, cellulose, lignin, and boron) have very poor validation results (≤ 30% accuracy). Though these traits are not referenced in the results/discussion much for comparisons between sites except in Figure 4 (and reported in Table SI14), it seems reporting the caveats associated with these estimations would be important as calibration models used to predict them are quite poor.  I suggest removing the poorest performing traits listed above from this analysis.

 

3.      The only foliar traits discussed in detail in the paper in review are nitrogen and LMA. Though these are among the more accurate estimations in the Asner et al., 2015 paper, a few of the study locations in the present study in review fall within the reported error associated with nitrogen (mean RMSE of 0.31%; table 3). For instance (in Table SI14) the difference between YL and YH sites nitrogen estimates are 0.3%, and OL and OH is 0.31%. With this in mind, how confident are your assessments of reported differences in sites?

Abstract

Need to add a sentence or two at the end to tie the results of the study to the broader scientific context of why these results are important (the “so what” of the study).

Line 16: italicize Latin name: Metrosideros polymorpha

Line 18: fix hyphen rendering issue

Introduction

Line 56: suggest editing to “…and can be used to estimate plant properties..”

Line 57: remove the comma after reference 17 and replace with “and” to make complete sentence

Line 59: please define “high spatial resolution” as this varies between studies and applications of remotely sensed data

Line 66” change “leaf economics spectrum” to “LES” as you introduced the acronym earlier

Lines 62-67: a very long sentence, suggest breaking it into two sentences beginning with “However, distilling the electromagnetic spectrum…”

Lines 67:68: this is a bit of an abrupt switch in topic… suggest adding more of transition between regarding imaging spectroscopy in Hawaii as it relates to climate chance and lack of intra-species studies to improve the flow.

Line 88: amend to say “full optical spectrum”

Line 89: you list the observed variations (physiological, chemical etc.) and include “spectral” here, but in the previous sentence you note that “no prior study has examined the spectral variation of M. polymorpha using the full spectrum”. Does this mean other studies have employed multispectral or other remote sensing technique? This is slightly confusing as written currently.

Line 98: what is meant by “derived trait indices”?

Line 104: add “we predict” to the front of “canopy spectra will differ” to enhance flow

Line 107: same as above comment, need something like “we predict” in front of “Derived leaf indices…”

Lines 105-106: what do you mean by “in parallel with site canopy traits”?

Materials and Methods

Consulting Table 1 do you think there is bias introduced in your comparisons related to either physical area sampled for a given soil/elevation site or the corresponding number of pixels associated with each site after the filtering steps detailed in Section 2.2?

Line 154: delete extra semicolon after Boulder, CO. “Miller 2002 should be cited in the same format as the journal citation style

Figure 2: may be more appropriate in results section

Table 3: this is a useful table, but I suggest adjusting the “Purpose” column to be left justified for a cleaner looking table. Also why only compare sites and intra-site variation qualitatively?

In reading through the methods, I’m not certain how the imagery was subset for a given site in order to conduct an intra-species comparison within a site. In the results (lines 257-258) it states “intra-site spectral variation was primarily driven by soil substrate age.” What was your unit of analysis within a given site for which this comparison is made? There is a nice description of the filtering of pixels that took place in section 2.2, but it does not describe whether pixels (as the unit of analysis) are used to compare reflectance characteristics across a given sampling location or whether individual trees were identified for this intra-site comparison.

Outside the scope of this paper, but I do wonder how other canopy structural characteristics vary between sites that may influence your trait estimations (leaf density, angle distribution, etc.) which could also be potentially assess with lidar.

Results

Lines 283-284: despite being the “lowest accuracy” these are still quite accurate. Suggest revising this description.

Lines 284-286: I don’t understand this sentence, when you say “Sites on the medium substrate age were confused”, are you referring to the 0.2% of pixels misclassified?

Lines 295-302: Would the differences in area between sites (as list in Table 1) influence these Euclidian distances in PC space? I ask out of genuine curiosity as I don’t use PC for such applications.

Figure 4: which leaf traits are going into this analysis (all listed in Table 2)? See my previous comment about reducing leaf trait comparisons to those that have the highest accuracy from Asner et al., 2015 (Table 3)

Figure 5: please update methods section to include description of the way the “areal extent of each site was controlled”

Discussion

A solid discussion of the findings and their relationship to the literature

Lines 414: and 447 change to “Hawai’i” to be consistent with the other references to the island in the paper.

 

Author Response

In response to the reviewer’s comments, we heavily edited the methods to make our process clearer and more straightforward. We hope that this addresses many of the concerns raised by the reviewers. In doing so, we reassessed the methods used in one of our analyses – calculating the intrinsic spectral dimensionality for all sites combined. We adjusted our methods to achieve results that were more robust, as is described in the current version of the manuscript. As our subsequent analyses relied on these results, we reran them with 16 PC rather than 20. The Euclidean distances were unaffected by the change, and there will minimal changes to the SVM results. 

 

Reviewer 1

This paper employs airborne imaging spectroscopy and light detection and ranging to evaluate spectral variation in Metrosideros polymorpha across six locations of varying soil substrate age and elevational gradient on the big island of Hawaii. It is a well written and interesting study that employs solid remote sensing collection and processing methodology. I’ve included minor comments below.

My one major comment is regarding the lack of physical trait data from the study sites directly to pair with the remotely sensed data for validation. The authors use calibrations generated from a previous (highly cited) paper (from the same authors), “Quantifying forest canopy traits: Imaging spectroscopy versus field survey” (Asner et al., 2015), which took place in Peru, but do not have validation samples to ensure the applied calibrations are accurate and appropriate in the context of their study in review. After reading through the referenced paper, I pose a few questions and comments to the authors:

1. In the current paper (in review) you make a solid case for why studying intra-trait variability is important. Since the calibrations applied in this study to estimate foliar traits are not from the same region, can you comment on the influence this would have on your results? Is the species of interest in the present study one of the species that was sampled in the Asner et al., 2015 paper?

Metrosideros polymorpha is endemic to Hawaii and therefore is not represented in the Asner et al., 2015 paper. However, the data used to develop the chemometric equations included species spanning tropical lowlands to montane forests, so the biomes represented by our study sites were present in the training data. As the data were calibrated in Peru and applied to Hawaii, we note that the leaf traits derived from the spectra are estimations in the paper as their absolute value is less important than their relative values. Further, these chemometric equations have been applied to forests in Costa Rica with similar success to those reported in the 2015 paper (Balzotti et al., 2016), and similar equations emerged from separate work in Borneo (Martin et al., 2018). These methods were applied to sites in Hawaii to map foliar nitrogen (Balzotti & Asner, 2018).

Balzotti CS, Asner GP, Taylor PG, Cleveland CC, Cole R, Martin RE, Nasto M, Osborne BB, Porder S, Townsend AR. 2016. Environmental controls on canopy foliar nitrogen distributions in a Neotropical lowland forest. Ecol Appl 26:2451–64.

Balzotti, C. S., & Asner, G. P. (2018). Biotic and abiotic controls over canopy function and structure in humid Hawaiian forests. Ecosystems, 21, 331-348.

Martin, R. E., Chadwick, K. D., Brodrick, P. G., Carranza-Jimenez, L., Vaughn, N. R., & Asner, G. P. (2018). An approach for foliar trait retrieval from airborne imaging spectroscopy of tropical forests. Remote Sensing, 10(2), 199.

 

2. The focus of the paper in review is on assessing intra-species traits across different locations, so I can see why an exact (highly accurate) measure of foliar traits are not necessarily needed to reach this aim, especially as many of the comparisons are made qualitatively. However, would the authors please comment on the accuracy of your estimates for these characteristics across the study sites (such as in Figure 6 and Table SI4)?

In situ estimates of foliar nitrogen range from 0.2 to 1.3 (Martin & Asner, 2009; Treseder & Vitousek, 2001). Most of our estimates fall within this range, though the ML site tends to have higher percent foliar N (mean 1.79). Some of this error may be attributed to the accuracy of the chemometric equations, as you noted above, or other factors such as noise or understory vegetation. Other possible explanations may relate to the method of sampling in that our data includes a greater spatial extent in addition to representing a larger portion of the canopy (which can have within-canopy variation of N; Kamoske et al., 2021). While some of our estimates fall outside the range of N noted in the literature, we assume that biases in the model are systematic and are therefore do not affect our conclusions in this comparative analysis.

LMA measurements of M. polymorpha range from 140 to nearly 500 g/m2 (Seeley et al., in review; Tsujii et al., 2016).

Martin, R. E., & Asner, G. P. (2009). Leaf chemical and optical properties of Metrosideros polymorpha across environmental gradients in Hawaii. Biotropica, 41(3), 292-301.

Kamoske, A. G., Dahlin, K. M., Serbin, S. P., & Stark, S. C. (2021). Leaf traits and canopy structure together explain canopy functional diversity: an airborne remote sensing approach. Ecological Applications, 31(2), e02230.

Seeley, M., Stacy, E.A., Martin, R.E., Asner, G.P. (in review). Foliar functional and genetic variation in a keystone Hawaiian tree species estimated through spectroscopy

Treseder, K. K., & Vitousek, P. M. (2001). Potential ecosystem-level effects of genetic variation among populations of Metrosideros polymorpha from a soil fertility gradient in Hawaii. Oecologia, 126, 266-275.

Tsujii, Y., Onoda, Y., Izuno, A., Isagi, Y., & Kitayama, K. (2016). A quantitative analysis of phenotypic variations of Metrosideros polymorpha within and across populations along environmental gradients on Mauna Loa, Hawaii. Oecologia, 180, 1049-1059.

 

-          In consulting with Tables 2 and 3 in the Asner et al., 2015 paper (for which the calibrations are sourced) several traits of interest (potassium, phenols, magnesium, cellulose, lignin, and boron) have very poor validation results (≤ 30% accuracy). Though these traits are not referenced in the results/discussion much for comparisons between sites except in Figure 4 (and reported in Table SI14), it seems reporting the caveats associated with these estimations would be important as calibration models used to predict them are quite poor.  I suggest removing the poorest performing traits listed above from this analysis.

      In response to your suggestion, we have removed the six traits with poor validation results from the table. We reran the Euclidean distance and SVM without these traits and included these results in the table instead.

 

3. The only foliar traits discussed in detail in the paper in review are nitrogen and LMA. Though these are among the more accurate estimations in the Asner et al., 2015 paper, a few of the study locations in the present study in review fall within the reported error associated with nitrogen (mean RMSE of 0.31%; table 3). For instance (in Table SI14) the difference between YL and YH sites nitrogen estimates are 0.3%, and OL and OH is 0.31%. With this in mind, how confident are your assessments of reported differences in sites?

While you point out an important source of error, we note that the potential negligible differences between YL/YH and OL/OH do not refute our argument that nitrogen availability is largely driven by soil substrate age, as is supported by many Hawaii-based studies. We included noted the potential for a negligible difference between these sites based on the RMSE in the results in response to this question.

Abstract

Need to add a sentence or two at the end to tie the results of the study to the broader scientific context of why these results are important (the “so what” of the study).

We changed the last sentence to discuss broader implications of the study.

 

Line 16: italicize Latin name: Metrosideros polymorpha

Italicized all latin names, though this seems to be a rendering issue. Will try to fix in the abstract submission form.

 

Line 18: fix hyphen rendering issue

Changed hyphen

 

Introduction

Line 56: suggest editing to “…and can be used to estimate plant properties..”

Changed

 

Line 57: remove the comma after reference 17 and replace with “and” to make complete sentence

We added ‘thereby’ to enhance the sentence flow as using ‘and’ changes the meaning of the sentence.

 

Line 59: please define “high spatial resolution” as this varies between studies and applications of remotely sensed data

Added ‘(2 m x 2 m)’ to define the resolution.

 

Line 66” change “leaf economics spectrum” to “LES” as you introduced the acronym earlier

Changed.

 

Lines 62-67: a very long sentence, suggest breaking it into two sentences beginning with “However, distilling the electromagnetic spectrum…”

We broke the sentence into two.

 

Lines 67:68: this is a bit of an abrupt switch in topic… suggest adding more of transition between regarding imaging spectroscopy in Hawaii as it relates to climate chance and lack of intra-species studies to improve the flow.

We flipped the topic sentence to improve the flow.

 

Line 88: amend to say “full optical spectrum”

As the optical spectrum refers to wavelengths between 400 and 750 nm, this would not be accurate as we are including wavelengths beyond the visible.

 

Line 89: you list the observed variations (physiological, chemical etc.) and include “spectral” here, but in the previous sentence you note that “no prior study has examined the spectral variation of M. polymorpha using the full spectrum”. Does this mean other studies have employed multispectral or other remote sensing technique? This is slightly confusing as written currently.

Here we are referring to prior studies that either used part of the spectra (e.g. 400-800 nm in Martin & Asner, 2009) or band calculations (e.g. green-red ratio, simple ratio in Martin et al., 2007). We included “spectral traits across select wavelengths” to clarify the difference.

 

Line 98: what is meant by “derived trait indices”?

We changed this sentence to read: … versus trait indices estimated from reflectance data

 

Line 104: add “we predict” to the front of “canopy spectra will differ” to enhance flow

Added.

 

Line 107: same as above comment, need something like “we predict” in front of “Derived leaf indices…”

We added ‘we predict’ to the sentence

 

Lines 105-106: what do you mean by “in parallel with site canopy traits”?

We simplified the sentence to read: … ‘spectral characteristics and canopy traits”

 

Materials and Methods

Consulting Table 1 do you think there is bias introduced in your comparisons related to either physical area sampled for a given soil/elevation site or the corresponding number of pixels associated with each site after the filtering steps detailed in Section 2.2?

This is a valid point. We had randomly selected 12997 pixels from each site to control for the number of pixels. As this was not expressed in our methods previously, we added a section describing this in the methods. While we did not attempt to control for site size when controlling for the number of pixels at each, the sites were chosen based on the homogeneity of the canopy. We did control for site size when calculating the intrinsic spectral dimensionality and obtained similar results to those of the CV which only controlled for number of pixels, indicating that site size (in total ha) had a minimal effect on the results.  

 

Line 154: delete extra semicolon after Boulder, CO. “Miller 2002 should be cited in the same format as the journal citation style

Thank you for catching this. The citation format was changed.

 

Figure 2: may be more appropriate in results section

While we agree that Fig. 2 could be considered a result, we believe it would be out of place in the results as we used it as a means of describing the structure of the sites to compliment Fig. 1. Further, we do not directly discuss canopy height in the results.

 

Table 3: this is a useful table, but I suggest adjusting the “Purpose” column to be left justified for a cleaner looking table. Also why only compare sites and intra-site variation qualitatively?

Agreed. The table was justified. We used both qualitative and quantitate methods to compare the sites. We qualitatively compared the CV as well as calculated the intrinsic spectral dimensionality to better understand intra-site variation. We also used the brightness-normalized reflectance to compare site reflectance qualitatively, but the Euclidean distance and SVM were our methods of comparing sites quantitatively.

 

In reading through the methods, I’m not certain how the imagery was subset for a given site in order to conduct an intra-species comparison within a site.

Since our sites were chosen because the canopies were entirely M. polymorpha, our intra-site comparisons were inherently intra-specific.

In the results (lines 257-258) it states “intra-site spectral variation was primarily driven by soil substrate age.” What was your unit of analysis within a given site for which this comparison is made? The lines that you point to were discussed further below in section 3.2. We used the CV and intrinsic spectral dimensionality results to draw this conclusion as intra-site variation, as approximated via these methods, had a stronger response to soil substrate age than elevation.

There is a nice description of the filtering of pixels that took place in section 2.2, but it does not describe whether pixels (as the unit of analysis) are used to compare reflectance characteristics across a given sampling location or whether individual trees were identified for this intra-site comparison.

We added a section regarding the subsampling of pixels in the methods to control for number of pixels included in the analyses. As M. polymorpha canopies grow together, canopy-based analyses are very challenging. Therefore all analyses were pixel-based.

 

Outside the scope of this paper, but I do wonder how other canopy structural characteristics vary between sites that may influence your trait estimations (leaf density, angle distribution, etc.) which could also be potentially assess with lidar.

The canopy structural characteristics you listed do affect canopy reflectance. However, most (although not all) of these effects are removed during brightness normalization.

 

Feilhauer, H., Asner, G. P., Martin, R. E., & Schmidtlein, S. (2010). Brightness-normalized partial least squares regression for hyperspectral data. Journal of Quantitative Spectroscopy and Radiative Transfer, 111(12-13), 1947-1957.

 

Results

Lines 283-284: despite being the “lowest accuracy” these are still quite accurate. Suggest revising this description.

We adjusted the sentence to reflect this.

 

Lines 284-286: I don’t understand this sentence, when you say “Sites on the medium substrate age were confused”, are you referring to the 0.2% of pixels misclassified?

Yes, this is what we were referring to. We adjusted the sentence to make this clearer.

 

Lines 295-302: Would the differences in area between sites (as list in Table 1) influence these Euclidian distances in PC space? I ask out of genuine curiosity as I don’t use PC for such applications.

The number of pixels included in the analysis would affect the PCA if they differed between the sites. While we controlled for number of pixels, we rely on the homogeneity of canopies at each site to control for site size.

Figure 4: which leaf traits are going into this analysis (all listed in Table 2)? See my previous comment about reducing leaf trait comparisons to those that have the highest accuracy from Asner et al., 2015 (Table 3)

Yes, all leaf traits from Table 2 were included. We noted this in the figure caption. As per your comment above, we reduced the traits included in the analysis.

 

Figure 5: please update methods section to include description of the way the “areal extent of each site was controlled”

While we included a brief mention of this in the methods, we expanded our description to make it clearer.

 

Discussion

A solid discussion of the findings and their relationship to the literature

Lines 414: and 447 change to “Hawai’i” to be consistent with the other references to the island in the paper.

Thank you for catching this. The changes have been made.

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, the impact of environmental filtering on the function of a single species was evaluated by comparing the spectra of a forest of closed-canopy polytypic woody plants. The results showed that the spectra of the six forest sites varied with the age of the soil substrate and the elevation gradient.

The separability of each point was verified by calculating the Euclidean distance (Euclidean dis-407) between each point and canopy features in the PC space of vegetation reflectance data using Support Vector Machine (SVM) method. The higher classification accuracy of the SVM on the reflectance data indicates that there is a large difference in canopy reflectance among the sites. Differentiation of site reflectance data may be due to environmental filtering to select different genotypes.

SVM results show that canopy reflectance data can distinguish polymorphic m.a from six stands representing environmental gradients that select for different polymorphic m.a traits. SVM was used to determine the separability of the six sites, but evaluating the performance of the model on spatially independent sites still needs to be added to the study. In addition, factors such as the environment, light, or atmosphere may also affect the performance of the model, so they also need to be reflected in the experiment. The application of imaging spectral data to the model system demonstrates the potential of canopy reflection data to address forest community assembly at large spatial scales.

The study in this paper is limited to six sites, each representing a unique soil matrix age-elevation combination, and the lack of reproducible combinations makes the experimental results lack generality. These combinations can be extended experimentally for a more rigorous statistical analysis of the environmental factors driving canopy reflection patterns. In addition, many algorithms, e.g., SVM, random forest and Naive Bayesian Classifier, could be employed to test the consistency of the quantitative conclusion.

Finally, a similar work named “Shortwave Radiation Calculation for Forest Plots Using Airborne LiDAR Data and Computer Graphics, Plant phenomics” also elaborates on the forest reflectance of the shortwave radiation and accounts for the occlusion among the forest canopy, which could be mentioned in the Introduction section.

Author Response

In response to the reviewer’s comments, we heavily edited the methods to make our process clearer and more straightforward. We hope that this addresses many of the concerns raised by the reviewers. In doing so, we reassessed the methods used in one of our analyses – calculating the intrinsic spectral dimensionality for all sites combined. We adjusted our methods to achieve results that were more robust, as is described in the current version of the manuscript. As our subsequent analyses relied on these results, we reran them with 16 PC rather than 20. The Euclidean distances were unaffected by the change, and there will minimal changes to the SVM results. 

Reviewer 2

In this paper, the impact of environmental filtering on the function of a single species was evaluated by comparing the spectra of a forest of closed-canopy polytypic woody plants. The results showed that the spectra of the six forest sites varied with the age of the soil substrate and the elevation gradient.

The separability of each point was verified by calculating the Euclidean distance (Euclidean dis-407) between each point and canopy features in the PC space of vegetation reflectance data using Support Vector Machine (SVM) method. The higher classification accuracy of the SVM on the reflectance data indicates that there is a large difference in canopy reflectance among the sites. Differentiation of site reflectance data may be due to environmental filtering to select different genotypes.

Thank you for your comments. In response to your suggestions, we heavily revised the methods section to make our objectives and methods clearer.

SVM results show that canopy reflectance data can distinguish polymorphic m.a from six stands representing environmental gradients that select for different polymorphic m.a traits. SVM was used to determine the separability of the six sites, but evaluating the performance of the model on spatially independent sites still needs to be added to the study.

While we agree that adding spatially independent sites would be important for evaluating most models, the purpose of the SVM here was what you had stated – to determine the separability of the six sites. We included text in the methods to make this distinction clearer. We believe that applying the model to other sites would not support this conclusion and would likely obfuscate the results.

 

In addition, factors such as the environment, light, or atmosphere may also affect the performance of the model, so they also need to be reflected in the experiment. 

These factors are important considerations, and to address them we used the ACORN atmospheric correction model as well as brightness normalization (Fig. 3). While these methods do not remove all the effects of atmosphere and light angles, we assume, like other studies in this field, that the effect of these factors is minimal. Addressing these issues is outside the scope of this paper as it would be a whole paper unto itself.

 

The application of imaging spectral data to the model system demonstrates the potential of canopy reflection data to address forest community assembly at large spatial scales.

 

The study in this paper is limited to six sites, each representing a unique soil matrix age-elevation combination, and the lack of reproducible combinations makes the experimental results lack generality. These combinations can be extended experimentally for a more rigorous statistical analysis of the environmental factors driving canopy reflection patterns.

While we agree that this case study has limitations relating to the lack of replication of each soil substrate age-elevation combination, as was noted in the conclusions, the patterns in our data reflected known ecological principles. In this study, we are not attempting to describe a unique phenomenon; rather, we demonstrated how general ecological principles to canopy reflectance of a single species across these environmental gradients.

 

In addition, many algorithms, e.g., SVM, random forest and Naive Bayesian Classifier, could be employed to test the consistency of the quantitative conclusion.

We employed a few different quantitative tests (e.g. intrinsic spectral dimensionality and the Euclidean distance) to address different aspects of the conclusion. SVM was chosen because their superiority over other algorithms like random forest in classifying species in imaging spectroscopy has been well documented. To demonstrate this, we ran a random forest and obtained an accuracy and precision of 0.988 (compared to the SVM, which was 0.998 for both precision and accuracy). Since this paper already has many tests, we feel that including these results would complicate the results, especially as the classification was used to simply demonstrate that the sites differ in the canopy spectra.

 

Finally, a similar work named “Shortwave Radiation Calculation for Forest Plots Using Airborne LiDAR Data and Computer Graphics, Plant phenomics” also elaborates on the forest reflectance of the shortwave radiation and accounts for the occlusion among the forest canopy, which could be mentioned in the Introduction section.

While this paper was interesting to read and provided valuable insight into the interaction of light in the canopy, it is difficult to place in the introduction as it is currently written. The paper does not directly relate to our study as the focus was using a combination of LiDAR and ground-based measurements to model the interaction of light with forest canopies. While this paper would be useful for an in-depth description of imaging spectroscopy, we did not delve deep into this topic in our introduction.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript focuses on the application of imaging spectrometry to assess trait variation at different environmental gradients. This is a relevant topic which is well addressed in the manuscript. The methods are accurate, replicable and adequately described. The results are clear and all the important questions are addressed in the discussion section, with adequate references.

My only minor comment is related to the NDVI threshold used. In section 2.2 (page 5) the authors state that they use the value of 0.7 for filtering non-photosynthetic vegetation. As this is a quite high threshold value, authors should explain it, maybe because the specific characteristics of this environment.

•  

Author Response

In response to the reviewer’s comments, we heavily edited the methods to make our process clearer and more straightforward. We hope that this addresses many of the concerns raised by the reviewers. In doing so, we reassessed the methods used in one of our analyses – calculating the intrinsic spectral dimensionality for all sites combined. We adjusted our methods to achieve results that were more robust, as is described in the current version of the manuscript. As our subsequent analyses relied on these results, we reran them with 16 PC rather than 20. The Euclidean distances were unaffected by the change, and there will minimal changes to the SVM results. 

 

Reviewer 3

The manuscript focuses on the application of imaging spectrometry to assess trait variation at different environmental gradients. This is a relevant topic which is well addressed in the manuscript. The methods are accurate, replicable and adequately described. The results are clear and all the important questions are addressed in the discussion section, with adequate references.

My only minor comment is related to the NDVI threshold used. In section 2.2 (page 5) the authors state that they use the value of 0.7 for filtering non-photosynthetic vegetation. As this is a quite high threshold value, authors should explain it, maybe because the specific characteristics of this environment.

Thank you for your comments. We chose a high NDVI to reduce error/bias in our dataset by selecting only pixels with the healthiest canopies. We tested a few different NDVI thresholds and based on these initial investigations and a priori knowledge of the study system, we settled on 0.7. Further, Rapid Ohia Death is very prevalent on Hawaii Island, and one of the documented symptoms is a reduction in NDVI. We included our reasons in the methods based on your suggestion.

 

Author Response File: Author Response.pdf