The Feasibility of Leaf Reflectance-Based Taxonomic Inventories and Diversity Assessments of Species-Rich Grasslands: A Cross-Seasonal Evaluation Using Waveband Selection

Round 1

Reviewer 1 Report

The author presents a multi-temporal dataset of seventeen species. The author concludes that most leaf level spectral variance is driven by a subset of species with a series of experiments. The manuscript has many shortcomings.

Comments:

1. What is the contribution of this article? There seems to be a lot of work in the article, but there is no highlight of work.
2. The title of the article does not reflect the key content of the article, especially in the Abstract part.
3. ‘Using a sparse partial least squares discriminant analysis, we found that overall model error was low to very low on all sampling dates (0.12 - 0.02) but that complex models, comprising up to 21 independent latent variables and 42-95%’

That doesn't seem like a very important conclusion without innovation.

4. What is ‘leaf-level’? This expression seems to be irregular.
5. In line 88, what is ‘uni-temporal’? This expression seems to be irregular.
6. In line 89, what is ‘ease’? This expression seems to be irregular.
7. In line 92, what is ‘model wavelength selection’? This expression seems to be irregular.
8. There are many other irregularities in the text, such as the following.

“They also suggest that the ability of spectral reflectance at specific wavelengths to predict species classes will be unstable and that species’ relative position in spectral space may, to some extent, vary over a growing season.”

 

“To give our sampling dates a temporal context in terms of precipitation and seasonal vegetation development, we used two remotely sensed environmental variables sourced  from Copernicus Sentinel data, and regional precipitation data.”

 

“To date, species discrimination tasks using hyperspectral data have generally been focused on woody species. Despite their conservation status and importance, herbaceous species are less studied and when they are, they are mostly considered at the canopy scale where single species dominate the sward.”

 Therefore, this manuscript does not conform to the requirements of the journal.

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Please do write in the passive voice (avoid to use "we", "us", example lines: 13, 15, 22, 51,  82, 84, 100, 105, 107, 111, 138, 148, 159, 178, 214,216, 217, 221, 274, 278, 281, 456, 457, 459, 461, 463, 468, 470,480, 523, 526, 535, 564, 577, 583 and 586). Review the paper and change in passive form the text, where needed. 

Lines 97-98: I suggets to replace the coordinate with  51°19'15"N   00°20'04"E.

Line 144: typo replace "r" with "R".

Line 230-240 use ";" instead "." as done in the list presented in the lines 85-93 or try to be consistent in the whole pepar about format.

Please try to improve the Figure 1 the text for axis (Figure 1 D and E) and labels sometimes (Figure 1A, BC) is really small and not easy for the reader to understand. 

For the captions try to use the same format. Same comment for Figure 4, 6 and 7.  As example the caption of Figure 2 to be consistent with Figure 1 needs to be modified as reported: "Figure 2. A: Satellite derived time-series of surface soil moisture (Sentinel-1) at 1km resolution; B: 310 Regional daily precipitation averages; C: The site based green-up trajectory using EVI (Sentinel-2) at 10m resolution. The 13 field sampling dates are shown as red triangles."  As example the caption of Figure 7 to be consistent with Figure 1 needs to be modified as reported: Figure 7. A: the ‘scree plot’ of the models at each time point, i.e., the variance in x explained by the 443 model latent variables / components. The grey reference line represents 99% variance in x, captured by six components, irrespective of sampling time; B, species class error over time with six compo nents; C: species class error with the chosen number of components (i.e., the final model for each time point). Mean error is shown for each time over 10 model runs (the S.E. of the model runs was very small and is not shown). 

Lines 319-334 (section 3.2) are duplicated and are the same of lines 337-352. Please check and solve the issue.. 

Please move Table 1 cited in the section 3.3 after Figure 3  cited in the section 3.2 and before Figure 4. 

Figure 4 please, in the legend for Figure 4A replace the grey circles with squares. 

Please, if possible move the Figure 6 inside the section 3.3. 

Figure 8 try to centre thegraphs with the caption as done for the previous figures.

Check the size of line 483 it seems bigger. 

Supplementary Materials A:

• S1 and S2: try to use the same size for the text (label, legend, axis) as the graphs are the same type;
• S3 the text for the axis is not visible. If possible try to increase 1-2pt the size to support the reader. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear authors,
I’ve read your paper focusing on spectral class analysis and band selection for taxonomic inventories in grasslands species. I've found your paper very interesting and well writing. Methods and results are clear presented and the readability is high. I also apreciate the use (and the references) of R-codes for each kind of computation.

Only a suggestion: at line 205, the recommended approach (reference 50) should be made explicit.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors focused on the classification of 17 cross-seasonal species using the combination of hyperspectral information and sparse PLA-DA algorithms. Encouraging results were obtained and the work sounds interesting. However, there are some concerns toward the analytical methods:

(1) The results of spectral smoothing using Savitzky-Golay filtering and corresponding spectral analysis should be given,

(2) In line 148, it was unclear why the setting of 3 nm resolution was adopted while resampling the spectral data. Besides, instead of selecting the spectral band manually, usually it would be more suitable to select the informative spectral band automatically using some type of feature (band) selection algorithm, e.g., swarm intelligent optimization techniques.

(3) If possible, adding the performance comparison between the sparse and non-sparse approaches for interpreting the transferability of the classification models.

(4) Similar to (2), in lines 197-199, considering swarm optimization algorithms for the selection of the optimal number of components sounds to be more appropriate than using the observation of the error stabilization.

(5) On the description in lines 423-424, I think that the learning schemes or strategies designed for dealing with the unbalance data distribution, e.g., SMOTE or other similar re-sampling techniques, could also be introduced to decrease the error rate on the data with low number of samples, if possible.

(6) With the change of sampling date, the sPLS-DA might need to be re-built time and time again to obtain acceptable classification results of the species. So, the temporal data series collected on successive dates across the seasonal period could be considered in the training process to build a more robust multi-class classifier.

Author Response

Please see the attachment

Author Response File: Author Response.docx