Monitoring LAI, Chlorophylls, and Carotenoids Content of a Woodland Savanna Using Hyperspectral Imagery and 3D Radiative Transfer Modeling

Round 1

Reviewer 1 Report

Review of

Monitoring LAI, chlorophylls and carotenoids content of a woodland savanna using hyperspectral imagery and 3D radiative transfer modelling

by Thomas Miraglio et al., submitted to Remote Sensing.

This is a nice paper about the potential of hyperspectral data for mapping the overstory biochemical traits and leaf area index. The results are encouraging and novel and I found no majors problems with the processing and analysis steps. However, there is some ambiguity in the description of the methods. Not all steps are described with sufficient detail as I have pointed out in my specific comments. Most notably, determination of LAI, a key parameter of any vegetation canopy, remains ambiguous. The optical methods employs by the authors allow to retrieve effective LAI, which needs to be corrected for clumping to obtain the true LAI. To my knowledge, no robust method exist for inferring the correction factor (clumping index) from fisheye images. Scene construction in DART requires true LAI, but the model allows to easily compute also the effective LAI, thus the authors have the data required for the conversion as simulations were run for varying LAI.
My second major concern is the treatment of uncertainty and error. These calculations are not described in methods. Sometimes, I got the feeling that only within-image variation was included in the uncertainty estimates presented in the figures. However, it is possible that systematic between-image variation in estimates because of e.g. illumination conditions or processing artifacts dominates the uncertainty. The method to distinguish significant changes is not given, and at times the authors emphasize on explicitly insignificant differences.

There are also minor issues with terminology, e.g. confusing pigment with its content, and incorrect notation style: symbols should be in italics, acronyms and units in roman, period should not be used to indicate multiplication. Appropriate spacing should be used between quantities and their units.

Specific comments:
line 1. It's pigment content (or concentration), not the pigment itself.
line 5. It is unclear if very low refers to overstory or all LAI, and "very low" is not sufficiently quantitative.
line 7. Delete "RTM", the acronym is not used in abstract.
line 9. the expression "different criteria, either VI-based or used spectral intervals" is not understandable without reading the manuscript. Rephrase.
line 10. Car estimations -> Car content estimations
line 14. It is unclear what "inversion criteria" refers to, the inversion process is not introduced in abstract. Abstract should be self-explanatory.
line 17. Provide units for the wavelengths.
line 17 and elsewhere. Please use standard notation style, symbols in italics, acronyms and units in roman
line 19. use consistently nanometers or microns.
line 19. Car estimations -> Car content estimations.
line 67. LUT has not been defined in manuscript text (abstract is a separate document)
line 73. LAI products are available globally for different biomes. Also, woodlands have been studied with hyperspectral remote sensing for decades.
line 76. Indeed, not much attention has been paid to open-canopy ecosystems, but they are included in general vegetation studies. Some authors of the paper have published on similar topics, too, and the oldest paper my search brought up was by Zarco-Tejada, et al., Hyperspectral indices and model simulation for chlorophyll estimation in open-canopy tree crops
(2004) in Remote Sensing of Environment. One may discuss the meaning of "validation", but I would soften the statement to remain in the realm of natural sciences (and not the history of it).
line 79. "was tested": unclear, by whom.
line 81. unclear what "present" refers to.
line 100. For all data sets, please provide the total number of data points.
line 106. Unclear what is referred to as "measurement". Usually, one LAI measurement is comprised of a number of photographic measurements (hemispherical photographs), e.g. the VALERI cross.
line 109. More information on processing is required, especially on thresholding. It remains unclear what is the output of the processing: effective or true LAI? If effective LAI was retrieved from measurements, how was it later converted to model input (true) LAI?
line 11. Describe how the collection was done.
line 117. Provide a reference for the full protocol.
line 139. Provide information on the actual data used.
line 144. Give more details on the level of the correction and the reference surfaces.
line 154. Use tense consistently.
line 162. Unclear if and how the species map was used, and whether it was compared with field data. There is no need to describe processing steps which did not directly contribute to the results.
line 180. LAI: rather, canopy leaf area and its spatial distribution.
Figure 4. Add spatial scale to maps. Also, consider making the maps larger (or the area they represent smaller) as the maps are to speckled to understand.
line 190. Give details on the measurement.
lines 192-193. This needs to be opened up.
line 195. Provide more information on how the pure soil pixels were selected and what was surrounding them: multiple scattering can have a strong influence on the retrieved reflectance.
Table 3. Earlier, Car and Cab did not stand for content. Already existing definitions should not be given new meanings.
line 245. estimations -> content estimations.
line 265. close to that of pure ground - note, however, that the reflectance values were produced with TOC irradiance and hence are not exactly equal to the soil reflectance factor. The latter has to be calculated with below-canopy irradiance.
line 266. here and elsewhere: use abbreviations consistently, the abbreviation Cab was introduced for chlorophyll.
line 268. "Vegetation indices were computed over the LUT reflectances." This should be mentioned in the beginning of 3.1 to clarify that this subsection is about simulated data.
line 275. Canopies with low CC and large LAI are unrealistic and the saturation has little practical relevance. For realistic canopies, saturation starts at values larger than simulated here. Exclusion of the data is justified by the lack of realism, not only for LAI.
line 340. The decrease is smaller than model error. I do not understand the importance (significance?) criterion. On line 343, the important variation is labeled insignificant. Error estimation should be explained better.
Figure 10, legend. Explain how the standard deviations were estimated, and what are the minimum and maximum LAI estimates for a plot (related to my comment to line 106)
line 347. Unclear, what is remarkable about the variation, if it less than the RMSE.
line 348. Cab -> Cab content, Car -> Car content
line 367. Use proper citation style. Citation number cannot be used as a (pro)noun.
Figure 11 has double legends, they should be merged.
line 361. Facets do not decrease by themselves unless implemented incorrectly. I looked up the cited paper and found no support for this claim. The difference in simulated reflectance in the paper are caused by detail in the description of canopy structure.
Figure 13. Be more quantitative when describing seasons (e.g., give month numbers in the legend). Specify how the confidence intervals were estimated (in manuscript text). Which uncertainty components were included, systematic or random ones?
Figure 14. The horizontal axis contains year, not date; the placement of tick marks, labels and vertical dashed lines is unclear. If the year number refers to 1. January, this should be clarified (or are the vertical lines 01 January?).
line 409. estimations -> content estimations
lines 415-416. This is data manipulation, delete.
line 458. Here and elsewhere: discuss only significant differences, and be explicit about it.
line 463. This massive drop seems approximately 2.5xSTD (table 7). Do you consider it significant? The error in LAI estimation need not be random, there may be systematic differences between acquisitions.
line 467. In my opinion, the central finding of the paper is that 18 m hyperspectral data with a physical RT model using a simple geometric representation of the canopy allows to estimate vegetation biophysical and biochemical parameters. The exact inversion methods come second.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper addresses the problem of LUT based inversion of AVIRIS-C medium resolution hyperspectral images to retrieve LAI and biochemical properties of low-foliage open canopies in a Mediterranean-like ecoregion.

I have found the paper both interesting and complete, therefore I recommend that the paper should be published.

Here a few comments and remarks to improve the paper.

The methodology can be applied to clear-sky, what about cloud detection in case of operational uses? The paper mostly limits to QUDO species, however, the coastal ecosystems in the Mediterranean basin, e.g., are mainly populated by Pines. The authors should better clarify the generality of their approach.

Minor

Introduction, first line: I would change “Mediterranean…” to “Mediterranean-like”…to avoid confusion with the Mediterranean basin. After the full stop, “Theses” should read “These”.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript explored a new method for the LAI, chlorophylls and carotenoids estimation in heterogeneous tree-grass, which is a relatively difficult topic in vegetation parameter retrieval field caused by the complex vegetation structure of savanna. The topic has significant meaning and suitable for the Journal of Remote sensing. However, there are several problems in this paper. The manuscript needs to address the following issue before publishing.

 

Major:

Too few field measurements to present the LAI of the plot. L199-202. The hyperspectral images have been atmospheric corrected, corresponding to the simulation without the atmosphere. The author should use no atmosphere during the DART simulation. Direct comparison has been commonly used in high spatial resolution parameter image validation. Suggest validating using the average of 3*3 neighboring pixels rather than spatially smoothed LAI images. L253-259 When you estimate parameters using the look-up table, you should make sure all the remote sensing observations are within the simulated lookup table. Otherwise, you need to adjust the variation range of the parameter (table3) as well as the fixed parameters, such as tree cover, crown shape/height, inter-trees distance, and clumping of the tree. The fixed parameters should also be listed somewhere in this manuscript. L274 If the author remote the 10% Cab category, it means this algorithm does not suitable for low chlorophyll trees. Suggest exploring other hyperspectral indexes that sensitive to low chlorophylls. Otherwise, the author needs to indicate this limitation in the conclusion.

 

Minor

UV-VIS spectrophotometer? Direction effects cannot be avoided unless both of the solar zenith angle and view azimuth angle are in nadir direction. L192-L194. What do you mean by “The trees were placed so that sun azimuth angle influence on the simulated TOC reflectances is minimal (proportion of shadows in the scene do not vary much with sun azimuth) and the scene BRF is the closest to a forest BRF”? L198 “Sun and atmosphere were the only radiation sources” Did you set the diffuse ratio? L369: You mean Figure6?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

 

I insist the DART simulation should use no atmosphere mode. The reason is the hyperspectral images have been gone through atmospheric correction. The indirect radiation sources could be defined in DART no atmospheric mode. I insist the look-up table should cover the possible condition in your study area. This is a basic law for vegetation parameter inversion using the LUT method.

 

I would suggest the author contact the DART group to consult the simulation parameters setting which is important for the inversion accuracy.