Leaf Trait Hyperspectral Characterization of Castanea sativa Miller Affected by Dryocosmus kuriphilus Yasumatsu

Round 1

Reviewer 1 Report

The paper is dedicated to the elaboration of hyperspectral-reflectance-based method for detection of chestnut plants infected by Asian chestnut gall wasp. This topic is of vast practical importance for horticulture. However, I should notice that the method suggested by the authors can hardly by useful for practical use, as it is based on contact measurements performed by handheld spectrometer. A method based on remote sensing (using airborne and spaceborne instruments) should be more valuable for horticulture applications. However, the current study shows that accuracy of hyperspectral detection of chestnut infection is rather low, so elaboration of such remote detection methods might be challenging. So the paper may be recommended for publication, if obvious issues with methods, results and conclusions are fixed.

First of all it should be noted that the relation Absorbance = log(1/reflectance) is nonsense. log(1/reflectance) = -log(reflectance) - the negative logarithm of reflection, and nothing more. This feature may be useful, as it amplifies reflectance differences in spectral regions where reflection is low - for example, near chlorophyll absorption maxima - but it is definitely not absorbance. To calculate absorbance, you need to measure both reflectance (R) and transmittance (T) spectra, and absorbance should be calculated as A=1-R-T. Of course, taking logarithm of this value may help in above-mentioned manner.

Spectral differences shown in Figures 4-7 are very small. It looks like they are much smaller than the variance within each sample. Corresponding standard deviation (standard deviations of the samples, rather than standard deviations of sample means) should be shown along with the spectra.

Confusion matrix (Table 1) seems strange. It is noted in the Materials and Methods section, 'To ensure a health status that was indicative of the whole sample, the same number of leaves (n=10) from each analysed branch was considered for each chestnut gall wasp damage level, resulting in 100 leaves for spectral measurement' (lines 139-141). However, in the confusion matrix we see 20 samples in N group, 6 - in L group, 5 - in M, 4 - in H, and 18 - in VH. Is there a mismatch between reference and predicted values?

'In sum, our findings suggest that field spectroscopy could be a useful non-destructive tool for monitoring Asian chestnut gall wasp infestations' (lines 304-305) conclusion seems to be too optimistic. The authors should either provide a usecase for their method, or provide a more sound conclusion. For my opinion, the paper shows that this method is not useful for such applications, but this doesn't deride the value of the paper.

Typo - 'ausence' (Fig. 4, 5).

Author Response

Leaf trait hyperspectral characterization of Castanea sativa Miller affected by Dryocosmus kuriphilus Yasumatsu

Ref.: agronomy-2239489

Revision #1

Response to Reviewer #1‘s comments

REVIEWER #1

Before responding to your comments and suggestions, we’d like to thank you for your review of this work, which has contributed greatly to improving this article and to the focus of future work.

Response to specific comments

Point 1. First of all it should be noted that the relation Absorbance = log(1/reflectance) is nonsense. log(1/reflectance) = -log(reflectance) - the negative logarithm of reflection, and nothing more. This feature may be useful, as it amplifies reflectance differences in spectral regions where reflection is low - for example, near chlorophyll absorption maxima - but it is definitely not absorbance. To calculate absorbance, you need to measure both reflectance (R) and transmittance (T) spectra, and absorbance should be calculated as A=1-R-T. Of course, taking logarithm of this value may help in above-mentioned manner.

Response 1. The spectroscopic transformation from reflectance to absorbance was carried out in unscramble (CAMO Analytics, Montclair, NJ, USA), this software assumes that the absorbance is simply logarithm (1/ reflectance). In addition, the use of this equation log = (1/ reflectance) is widespread to obtain absorbance values while working with spectroscopy and specially with diffuse reflectance, the following works [1–4] prove this claim. To sum it up, we have decided not to use transmittance values for the time it takes to collect this information trying to propose an efficient methodology in data collection.

1. Vašát, R.; Kodešová, R.; Klement, A.; Borůvka, L. Simple but Efficient Signal Pre-Processing in Soil Organic Carbon Spectroscopic Estimation. Geoderma 2017, 298, 46–53, doi:10.1016/j.geoderma.2017.03.012.
2. Stenberg, B.; Viscarra Rossel, R.A.; Mouazen, A.M.; Wetterlind, J. Visible and Near Infrared Spectroscopy in Soil Science. Adv. Agron. 2010, 107, 163–215, doi:10.1016/S0065-2113(10)07005-7.
3. Dharumarajan, S.; Gomez, C.; Lalitha, M.; Kalaiselvi, B.; Vasundhara, R.; Hegde, R. Soil Order Knowledge as a Driver in Soil Properties Estimation from Vis-NIR Spectral Data – Case Study from Northern Karnataka (India). Geoderma Reg. 2023, 32, e00596, doi:10.1016/j.geodrs.2022.e00596.
4. Tavakoli, H.; Correa, J.; Sabetizade, M.; Vogel, S. Predicting Key Soil Properties from Vis-NIR Spectra by Applying Dual-Wavelength Indices Transformations and Stacking Machine Learning Approaches. Soil Tillage Res. 2023, 229, 105684, doi:10.1016/j.still.2023.105684.

Point 2. Spectral differences shown in Figures 4-7 are very small. It looks like they are much smaller than the variance within each sample. Corresponding standard deviation (standard deviations of the samples, rather than standard deviations of sample means) should be shown along with the spectra.

Response 2. We have included the standard deviation in the reflectance and absorbance charts. But as it is possible to see in Figure 1 the result is not clean, and it is difficult to observe and analysis the spectral curves.

May be a good option while plotting absence and presence data but not with all the infestation levels, trying to obtain trying to obtain a clear chart. However, in this case we want to keep the original version but, in any case, if you think it is necessary to change it, we are willing to do so.

Figure 1. Spectral characterization using reflectance values.

Point 3. Confusion matrix (Table 1) seems strange. It is noted in the Materials and Methods section, 'To ensure a health status that was indicative of the whole sample, the same number of leaves (n=10) from each analysed branch was considered for each chestnut gall wasp damage level, resulting in 100 leaves for spectral measurement' (lines 139-141). However, in the confusion matrix we see 20 samples in N group, 6 - in L group, 5 - in M, 4 - in H, and 18 - in VH. Is there a mismatch between reference and predicted values?

Response 3. We have done a correction in lines 136-139, the resulting text is as follows: ‘To ensure a health status that reflected the whole sample, selected from all the analysed branches for each chestnut gall wasp infestation level were at least 10 leaves, resulting in 102 samples for spectral measurement.’. Moreover, we have also corrected the confusion matrix, it is necessary to note that validation data is made up for 52 samples all of them represented in the matrix.

Point 4. 'In sum, our findings suggest that field spectroscopy could be a useful non-destructive tool for monitoring Asian chestnut gall wasp infestations' (lines 304-305) conclusion seems to be too optimistic. The authors should either provide a usecase for their method,or provide a more sound conclusion. For my opinion, the paper shows that this method is not useful for such applications, but this doesn't deride the value of the paper.

Response 4. Trying to show a sounder claim we have added to the conclusion (lines 305-307) the following text: ‘Even if differences are not very great, they could be useful when choosing remote sensors based on their spectral resolution and in upscaling this health monitoring technique.’

Point 4. Typo - 'ausence' (Fig. 4, 5).

Response 4. Figs. 4 and 5 have been corrected in the manuscript.

 

Reviewer 2 Report

MS: Leaf trait hyperspectral characterization of Castanea sativa Miller affected by Dryocosmus kuriphilus Yasumatsu

This study detects the gall wasp infestation on the European chestnut using hyperspectral data with moderate accuracy. Two models were used to distinguish healthy and affected leaves under five different infestation levels. Despite the lower accuracy of detecting different infestation levels, the study is useful to detect gall absence and presence. Spectroradiometer is used to collect in-situ data, however for a large scale detection it need drone survey with a hyperspectral sensor. Then this spectral signature results are useful to detect healthy vs unhealthy trees. Also it is worth to discuss how moderately resolution data (i.e. Landsat data) can be used to detect healthy and unhealthy trees when we know the specific range of spectral variability.  I’m wondering whether authors can use different vegetation indices to detect different infestation levels since the hyperspectral data cover full Electromagnetic radiation of the spectrum.

Here are some minor comments.

Long keywords: partial least square discriminant analysis

L 63: etc.

L 95: Is that UAV imagery for study plot, if so mention in in the Fig. caption.

L 96: Give descriptive Figure caption

L 106: Give descriptive Figure caption

L118-L  L122: Combine Table 2 and 3 to a single table. In Table 3, the first column label as Infestation levels & affected bud %. Then the value can be given in the parenthesis for infestation levels i.e. None (0%), Low (≤ 30) etc.

L 228: Table 2 - Why ±10 nm intervals and 4 wavelengths values are same.

Author Response

 

Leaf trait hyperspectral characterization of Castanea sativa Miller affected by Dryocosmus kuriphilus Yasumatsu

Ref.: agronomy-2239489

Revision #1

Response to Reviewer #2‘s comments

REVIEWER #2

Before responding to your comments and suggestions, we’d like to thank you for your review of this work, which has contributed greatly to improving this article and to the focus of future work.

Response to specific comments

Point 1. This study detects the gall wasp infestation on the European chestnut using hyperspectral data with moderate accuracy. Two models were used to distinguish healthy and affected leaves under five different infestation levels. Despite the lower accuracy of detecting different infestation levels, the study is useful to detect gall absence and presence. Spectroradiometer is used to collect in-situ data, however for a large scale detection it need drone survey with a hyperspectral sensor. Then this spectral signature results are useful to detect healthy vs unhealthy trees. Also it is worth to discuss how moderately resolution data (i.e. Landsat data) can be used to detect healthy and unhealthy trees when we know the specific range of spectral variability.  I’m wondering whether authors can use different vegetation indices to detect different infestation levels since the hyperspectral data cover full Electromagnetic radiation of the spectrum.

Response 1. In response to the proposal to include in the discussion how satellite sensors can be used to detect healthy and diseased trees. We should mention that we are working in the upscaling of this work, and include remote sensors is a future objective.

Since we have not included spectral indices, we have to say that given that the hyperspectral data covers the entire electromagnetic spectrum we have tried to use different vegetation indices. However, no significant differences were observed between the different levels of infestation when using them.

Point 2. Long keywords: partial least square discriminant analysis.  

Response 2. We have decided to use PLS-DA this instead.

Point 3. L 63: etc.

Response 3. We have removed that.

Point 4. L 95: Is that UAV imagery for study plot, if so mention in in the Fig. caption.

Response 4. We have added a caption including the study area and the analysed trees on a UAV orthoimage.

Point 5. L 96: Give descriptive Figure caption

Response 5.  We have added a descriptive figure caption in lines 95-96, the text is the following ‘Study site in Robledo de las Traviesas (42° 42' 27.65'' N, 6° 26' 13.54'' W). For both plots, the analysed trees are depicted in an UAV orthoimage.’

Point 6. L 106: Give descriptive Figure caption

Response 6. For both plots the workflow followed were the same, so we have described it as follows (lines 107-109): ‘Workflow, as follows: 1. Infestation assessment; 2. Spectral measurements; 3. Spectral data pre-processing; 4. Classification model fittings; 5. Classification model validations; 6. Best classifier se-lection.’.

Point 7. L118-L  L122: Combine Table 2 and 3 to a single table. In Table 3, the first column label as Infestation levels & affected bud %. Then the value can be given in the parenthesis for infestation levels i.e. None (0%), Low (≤ 30) etc.

Response 7. Attending to this advice we have join both in a single table (table 2) at line 118.

Point 8. L 228: Table 2 - Why ±10 nm intervals and 4 wavelengths values are same.

Response 8. We have chosen these values because there are significant differences while distinguishing by presence and absence. Then around these wavelength ±10 nm intervals were defined trying to take advantage of differences between spectral features, if any. This is also due to the fact that the resolution of the spectrodiometer used is 10 nm between 700 and 2500 nm of the electromagnetic spectrum

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

1. I cannot accept the responce to point 1. Indeed, absorbance is 1-reflectance for non-translucent materials, and it is definitely OK to use mentioned relations in soil studies as soil is not translucent. But plant leaf is translucent, and spectral properties of reflected and transmitted light may differ greatly. This is discussed in ref 34 you cite in your paper. I don't insist that you should use both reflectance and transmittance spectra in your study, but designation of -log(reflactance) as absorbance is simply incorrect for plant leaves.

2. Unfortunately, I cannot find any chart with standard deviations. You refer to Figure 1, but there is no Figure attached to the responce - or maybe the MDPI online reviewing system does not show it to me.

3. To make it clear, I suggest you to provide full information: how many leaves were in each category in training set and in validation set. And - all the samples were used for calculation of average spectra, am I right?

4, 5. OK

Reviewer 2 Report

Thanks for making the inputs for improving the quality of the manuscript. Authors addressed the comments suggested.