Study on Surface Reflectance Sampling Method and Uncertainty Based on Airborne Hyperspectral Images

Round 1

Reviewer 1 Report

This manuscript focused on the uncertainty analysis of sampling method in validation of surface reflectance product. Airborne hyperspectral images were used in this study. The research topic is meaningful because scale conversion is one of the important steps in validation study. Uncertainty analysis is an important guarantee for the reliability of validation results. However, the manuscript has some shortcomings that need to be revised and corrected by the authors.

(1) The most important point is that there are some issues with the conclusion of the manuscript. The authors believe that the 9-point method can ensure an uncertainty of less than 1%. This conclusion is not very rigorous, and the conditions for its validity need to be in specific ground scenarios. For ground scenarios with poor heterogeneity, this conclusion may not necessarily hold true. Therefore, for rigorous papers, it is necessary to consider ground scenarios with different heterogeneity, conduct sufficient analysis, and obtain reliable conclusions.

(2) This manuscript focuses on uncertainty analysis. Formula (2) provides a formula for calculating uncertainty. However, the uncertainty analysis results shown in Figure 13 have negative values, which makes me doubt the reliability of the paper's results.

(3) The manuscript does not provide an evaluation of the non-uniformity of each ground scene, which can be supplemented by the necessity of studying sampling strategies.

(4) “If the site has a variety of ground object types or the measurement is difficult, the 4-point or 5-point method can be used, but the uncertainty may increase to 1-2%.” The condition for this conclusion to hold in the article is also scenario dependent, which means that the non-uniformity of the scene needs to be limited.

(5) Optimizing the experimental design of the article can ensure more reliable analysis results and conclusions. Furthermore, the conclusion must be improved. Please write the conclusion in the form of important points.

(6) Please revise Subsection 5.3. On the one hand, the caption of Table 6 is inappropriate; On the other hand, the following text explains cannot accurately explain the conclusions, especially the line 545.

(7) Please carefully check Formula 11 and the explanatory text below.

 

(1) Abbreviation is not given full names when they are used for the first time, which appeared in line 57, p2.

(2) Chinese characters appeared in Fig. 1.

(3) Line 136, p4, “drones” should be “UAV”.

(4) Replace “This paper” or “This article” with “This study”, would be better.

(5) There are also existing few language, spelling, and formatting issues in the manuscript, e.g., line 106 & 108, p3; line 293, p9; line 362, p12; the authors should address all errors thoroughly. Chinese characters appeared in Fig. 1.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposes a method for estimating the uncertainty of ground-based surface reflectance spectral measurements with varying the number of measurement points for use in validating satellite measurements at the satellite sensor pixel scale. Spectra inverted from airborne hyperspectral images using an atmospheric correction procedure are used as the truth ground-based reflectance spectra.

The work is of interest to specialists involved in the in-flight calibration of satellite imaging systems, validation of satellite images, their processing and object classification.

 

Notes:

 

1. The authors use the term “pixel”, in particular, in Abstract, without indicating whether it refers to a pixel of a hyperspectral image from a UAV, or in relation to a pixel of a space sensor, which sometimes leads to confusion.

2. In the overview part of the Introduction there is no information on the important project and tool for calibrating and validating satellite images as the worldwide network RadCalNet.

3. In Fig. 1, the legend captions are in Chinese.

4. Line 171: “…hyperspectral images from apparent irradiance to surface reflectance”

It should obviously be “apparent radiance” instead of “apparent irradiance”.

If the surface reflectance spectra obtained as a result of atmospheric correction (solution of the inverse problem) are subsequently used as the ground truth spectra for ground-based measurements (the uncertainties of various sampling methods are estimated relative to them), then the uncertainties of the retrieved spectra (by atmospheric correction) must be estimated and given.

5. Line -192: “As shown in Figure 5, by stitching surface reflectance products of different flight strips that have been processed by atmospheric correction, a larger area of hyperspectral image can be obtained, and the image has no chromatic aberration.”

Color images are shown on Figure 5: how were they obtained, using which spectral channels? Should we understand that before stitching (on the left on Fig. 5) the spectral radiance images are shown, and after stitching (on the right) - reflectance? And the authors claim that chromatic aberration is removed by atmospheric correction?

6. Line -225: “20, 25, 45, 90, and 125, respectively. In order to use hyperspectral images to quantitatively"

Obviously it should be “20, 25, 45, 80, and 125…”

7. Lines -239-243: “According to the position of the sampling system and the observation field of view, the hyperspectral reflectance mask template is established. In this template, the white areas represent airborne hyperspectral pixels that have undergone spectral measurements, while the black areas represent airborne hyperspectral pixels that have not been spectrally measured."

Inaccurate language construction: instead of “the white areas represent airborne hyperspectral pixels that have undergone spectral measurements,” it would be more accurate to say “the white areas represent airborne hyperspectral pixels for which there are ground-based spectral measurements.” And similarly for the black areas.

In article there is no information on the method of spatial reference (visually or by geographic coordinates (low accuracy)?). The uncertainties of the spatial alignment (reference) of ground-based spectra measurement points and the corresponding pixels of the hyperspectral image are not given.

8. Lines -296-298: “...where M represents the number of pixels in the airborne hyperspectral image within the satellite pixel scale, and the value is 1111, that is, the row and column of the image are both 1111; ?_?,? represents the reflectance value of the hyperspectral image at the i-th row and j-th column".

It would be logical to further investigate the uncertainty of ground-based values of ?_? in relation to the measured hyperspectral values at the same points (white dots in Fig. 8), and, perhaps, use the average value of hyperspectral pixels over the simulated ground as ρ_T in formula (4) spectral measurement mask (white dots in Fig. 8)

9. Lines -328-330:“In order to quantitatively analyze the consistency of the ground measurement surface reflectance and the surface reflectance retrieved by the UAV, we selected two sets of endmember spectra acquired by the UAV and the ground in the soybean area on June 29, 330 2023.”

The article does not describe the procedure by which the ground-based reflectance spectra were obtained. The first mention of the used ground-based spectrometer (SR8800) appears only in line 379, (although the spectra are shown earlier in Fig. 10) without a description of the spectrometer parameters (spectral sampling 1 nm is not equal to spectral resolution, as a rule).

The spectrometer does not measure reflectance spectra; it measures spectral radiance, provided it has undergone appropriate radiometric calibration. Reflectance is usually obtained by measuring a reference diffuse screen in parallel. In principle, the atmospheric correction procedure (with a zero sensor height) can also be applied to ground-based measurements, but this is not mentioned in the article. The lack of description of the method for obtaining ground-based reflectance spectra and their uncertainties is a major methodological shortcoming of the article.

Why is there a wavelength axis in Fig. 10e,f is displayed starting from 0 nm but not from 400 nm?

10. Lines -344-347:“In order to quantitatively compare the consistency of the ground measurement spectra and the UAV hyperspectral image retrieval spectra, it is necessary to convolve the ground measurement spectra and the spectral response function of Hyspex to simulate the ground equivalent spectra of Hyspex."

If the ground-based reflectance spectra are obtained by the correct procedure (which is not described in the article, see note above) and are reflectance in dimensionless units (0 to 1) with high spectral resolution (1 nm), and if the reflectance spectra measured by UAV are obtained as a result of atmospheric correction (also dimensionless in the range 0-1) and also with a sufficiently high spectral resolution (3 nm), there is no need to perform convolution as in formula (6), a direct comparison of reflectance spectra can be made. Moreover, it is methodologically incorrect to use the spectral response function for convolution with reflection spectra; this procedure can only be performed with spectral radiance (because the spectral response function of the device is determined in relation to the absolute value of the input spectral radiance), the relationship between reflectance and spectral radiance is essentially non-linear. The fact that the authors obtain a good match of the spectra only indicates that the procedure according to formula (6) is not significant (its influence is insignificant). This may be if the integration range is small, which the authors do not give (what is FWHM? 2-3 nm?). But its application is methodically erroneous!

11. Why is the letter x used instead of ? for the wavelength in formula (7)? Why is formula (9) needed, and in a such terrible form?

12. Lines -344-347: “Nevertheless, the absolute accuracy has no determinate role in the results of this paper, because the pixel scale uncertainty is measured by the difference between the average value of the hyperspectral image mask area and the average value of all pixels, both of which are calculated from the same image"

The validity of this statement is doubtful, and the explanations given are written very unclearly; it needs to be reformulated more clearly.

13. What spectra (5) are shown in Fig. 12a, corresponding to five ground-based measurement points on an area of 30x30m? And in Fig. 12b, the scale conversion spectrum is the average spectrum over these five?

Conclusion:

The work contains elements of methodological novelty, but there is an incomplete description of the data (how ground-based reflectance spectra were obtained), and there are also methodological errors (see notes 9, 10, etc.).

 

Recommendation: With hyperspectral images of high spatial resolution, the authors could, within each image, accurately determine the percentage of open soil and green vegetation present (by classifying into two classes), calculate mixed site spectra (using endmember spectra), and analyze the relationship between the required number of ground points measurements and the degree of heterogeneity of the scene, and give the uncertainty values of the 5-point method depending on the degree of surface heterogeneity (instead of specific five sample areas of bare soil, wheat, soybean, corn 1, and corn 2), which would be useful when conducting polygon ground measurements.

 The article needs to be significantly revised.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper proposes to measure and discuss the uncertainty in the process of converting point-scale spectra from ground sampling to pixel-scale spectra, a phenomenon known as 'point-to-pixel scale conversion uncertainty.' I have several concerns about this paper:

1. What contributions can be gained from this paper? It is often the case that the ground-measured spectrum can differ significantly from the spectrum measured by airborne HSI sensors due to surface or other environmental issues. Could the authors provide some examples of how the uncertainty results measured by this paper can be used to enhance hyperspectral image sampling and processing? Can the authors also discuss how to calibrate these point-to-pixel scale errors?

2. Most of the HSI pixels are mixed pixels. Since a pixel's spectrum is generally a mixture of the spectra of many different substances, the ground sampling method used in this paper might not adequately capture a representative spectrum when the number of sampling points is small.

3. Please also describe how to use the proposed method to improve the practical hyperspectral image applications.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

（1）The revised manuscript evaluated the non-uniformity of the ground scene and used CV values to characterize it. However, there is some confusion about the calculation of this CV. The coordinate axis in the figure is named CV (%), with the unit being %. Therefore, the CV values are somewhat abnormal, mostly within 1%. However, the formula for CV is based on the standard deviation over the mean, without the process of converting to a percentage. I suspect there may be some issues with the modification content, and the CV values for surface scenarios should not be so small, such as some vegetation scenes.

（2）Some expressions may need to be modified, such as single point method, three-point method, etc., which should actually be single ESU method or three ESU method. Because one ESU in Figure 7 corresponds to five measurement points, the “single point” is actually five measurement points. So, it is recommended to make modifications to eliminate ambiguity.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you , I am mostly satisfied with the edits made by the authors in the revised version of the article

Author Response

We appreciate your valuable time and efforts in reviewing this manuscript, and we are very pleased with your approval.