A Wide-Angle Hyperspectral Top-of-Atmosphere Reflectance Model for the Libyan Desert

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please find the document attached.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

thank you very much for your manuscript. The article "A Wide-Angle Hyperspectral Top-of-Atmosphere Reflectance Model for the Libyan Desert" describes how the pseudo-invariant calibration site (PICS) Libya 4 - located in the Libyan Desert - can be used for absolute radiometric calibration of hyperspectral satellite sensors. A wide-angle hyperspectral reflectance model for Top-of-Atmosphere was developed using Terra/Aqua and EO-1 data from 2003 to 2012. The model's accuracy has been tested using data from Landsat 8 OLI and Suomi NPP VIIRS and shows a very good agreement.

The introduction is very well-written, complete, and contains sufficient references. It provides a good up-to-date review on the topics of PICS, cross-sensor validation, and the on-orbit calibration. It reveals, that existing models to derive TOA reflectance from PICS are limited to viewing angles of up to 20°, for hyperspectral imagers it is even less and requires simultaneous observations of MODIS. It concludes with the here presented model, that uses the targets and atmospheric stability to derive correct TOA reflectance predictions for VZA up to 65°.

In section 2, the PICS site Libya 4 is presented, as well as the used datasets. Its stability - both the target and the atmospheric parameters (AOD, water vapor, and ozone) are displayed in the text and several figures in a good and comprehensive way.

In the method chapter (section 3), the influences of different viewing angles (zenith and azimuth) and the reflectance are very well visualized, and grouped into 16 classes for the following model. The model itself has described well also the spectral extension (MODIS / Hyperion) and the measures used for evaluation.

In the Results chapter (section 4) first, the derived coefficients for the model are presented, then the validation with MODIS data from 2013 to 2020, and afterward the cross-validation with Landsat 8 OLI. Here, it is a bit confusing what figure shows which sensor. It would be helpful to add the sensor names to Figures 14, 15, and 16 (like "MODIS Band 1 (659 nm)" and "OLI Band 1 (443 nm)"). Besides this confusion, everything is well-written and comprehensive. A Discussion chapter is missing, which is okay in my eyes because the discussion points are included in section 4 - which makes it easier for the reader to read.

The Paper concludes with a summary, of why it is beneficial to use the spectral expansion method to get a hyperspectral signal from MODIS measurements, which can then be used to be resampled to a wider range of multispectral sensors.

For reproducibility, I strongly recommend naming the computing environment that was used for this study. The reader has to read to page 24 where it is stated in the acknowledgments that Python was used. But it is nowhere mentioned which packages are used to read the EO data, to get the fitted parameters of the model, and so on. So, I recommend inserting on paragraph where this information is given.

The manuscript is already very well written and there are only a few minor things to suggest:

Line 78: You are naming here the sources of satellite image data - but the URLs are missing.
Line 95: "Figure 1 is a false-color image generated using the near-infrared band (865 nm) of Landsat 8’s TOA reflectance": a false-color image is a combination of spectral bands other than the red-green-blue spectral ranges. Figure 1 is only showing a greyscaled single-band image.
Figure 1: I would suggest adding a small overview map (e.g. a shapefile) to show where the site is located.
Figure 2: Is the offset (of approximately 2 %) between Landsat 4/5 and Landsat 7/8 due to different spectral response functions? Maybe you could explain that in one sentence. And also add the Sensor names, not only the platforms (L4: MSS, L5: TM, L7: ETM+, L8: OLI).
Line 106: Please replace "remote sensor" with "remote sensing sensor".
Lines 113 to 116: Please name the URL (or DOI-URL) from where the data was derived, line "MOD021KM (http://dx.doi.org/10.5067/MODIS/MOD021KM.061)".
Line 286: "The optimal coefficients are obtained by minimizing the fitting errors." Consider already naming the fitting method here, it is explained later in line 366.
Figure 10: what are the different colors of the lines ranging from purple to orange? You only say that blue is the mean.
Figure 16: The dot color of the legend does not match the dots within the canvas. The highest value of VAA 80° is here orange and red in the figure.
Figure 17: Please correct the spelling error on the y-axis "Relative Diffrence" to "Relative Difference".
Line 467: "The overall predicted values of VIIRS are lower than observations but higher at 490
nm." Can you explain this? Since there is no discussion chapter, it should be mentioned here what could be the cause. Also, Landsat 8 OLI has a higher positive deviation here similar to VIIRS.
Figure 18: Why are the circles of VIIRS larger than the others? Please keep this uniform.

After these minor changes are made, the manuscript can be published.
All the best!

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper presents a wide-angle range hyperspectral model for predicting TOA reflectance over the Libyan desert. The model was developed using 10-year (2003-2012) of collection 6.1 Terra and Aqua MODIS level 1b and atmospheric parameter products, using Hyperion reflectance as the reference.  It was evaluated using MODIS, Landsat 8 OLI, and Suomi NPP VIIRS data from 2013-2020. The TOA reflectance model proposed in this study is intent to be used for remote sensing instrument calibration, however, the accuracy/stability of input data and the error budget of the model developed are not sufficiently provided. The overall uncertainties of this TOA reflectance model need to be provided quantitatively, preferably as functions of solar and satellite view angles.

1.       According to Section 3.3, Hyperion reflectance was used as the absolute radiometric reference in this study. What is the in-flight radiometric calibration method for Hyperion? What is absolute radiometric calibration accuracy of Hyperion L1B product? The spectral resolution of Hyperion is relatively coarse (10 nm only). The RSRs for some MODIS/Landsat/VIIRS may not be sufficiently covered. The impacts of spectral mis-match also need to be discussed.

2.       What is the radiometric calibration accuracy of the Collection 6.1 MODIS L1B data used in this study?

3.       Section 2.4: (1) Figure 6:  what data were used to generate the 3 histograms on the left panel of Figure 6? Based on lines 204-206 and Table 5, it seems only one day of atmospheric parameters were used in the sensitivity analysis. (2) What are the typical AOD, water vapor, and ozone values over the Libyan desert? (3) please clarify how to use the sensitivity study results to understand the validation results in Section 4.

4.       Please clarify which AOD, water vapor, and ozone products were used for predicting Landsat and NPP TOA reflectance.

5.       Figures 13&15: It seems the patterns during 2013-2014 (such as bands 1-2) are different from other years.  What are the potential factors that cause the differences?

6.       Besides absolute calibration accuracy, calibration stability of level 1b data is also critical for long-term environmental and climate studies.  What are the differences in long-term trends between model-predicted and observed TOA reflectance for MODIS, Landsat 8 OLI, and Suomi NPP VIIRS?

7.       Please discuss how to interpret the differences between the model predicted and the MODIS/Landsat/VIIRS observed TOA reflectance.  In other words, are the differences caused by calibration errors or caused by the uncertainty of the model developed in this study.

8.       Both MODIS and Suomi NPP science teams produce daily surface BRDF model parameters. It will be useful to compare the surface BRDF parameters (fiso, fvol,fgeo in Equation 3) derived from this study with the MODIS and VIIRS surface BRDF model parameter products. https://lpdaac.usgs.gov/products/mcd43a1v061/

https://e4ftl01.cr.usgs.gov/.

9.       To my best understanding, Suomi VIIRS L1B is calibrated using solar diffusor and lunar cal. But desert sites (such as Libyan-4) were used to calibrate both MODIS and Landsat L1B data. Please discuss how this fact affects the results.

10.   Line 140-141: this sentence is not accurate. VIIRS actually has a larger coverage area and higher temporal resolution than MODIS. The swath widths of VIIRS and MODIS are ~3000 km and 2330 km, respectively. The repeat cycles of MODIS and VIIRS are both 16 days.

11.   Suggest removing Figure 7.  Figure 8 alone is sufficient.

12.   Figures 13 & 14: “Aqua MODIS” should be “Terra MODIS”.

13.   Lines 399-400: Were Terra MODIS bands 5-7 data after 2016 used for model-fitting?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks to the authors for their careful reply. According to the author's latest version, the BRDF phenomenon of relative difference is indeed insignificant from the author's results, showing that the author's model is valid. My question is, the model used by the authors is an update on the Ross-Li BRDF model, and according to the f-coefficients fitted results from table1, the values of several f-coefficients corresponding to the atmospheric factors are very small, may I ask how much the results obtained are different from the existing results when compared to the Ross-Li model or when the authors use only the f-coefficients related to the observation angle (i.e., the first three terms on the right-hand side of Eq. 3)? How necessary is the presence of the three atmosphere-dependent terms? Also:

 

Figure 2, I suggest using the reflectance time series after RSR matching.

The caption for Figure 12 should be reflectance?

 

Author Response

Dear Reviewer,

Thank you for your positive acknowledgment of our model's validity and your insightful comments. We appreciate the opportunity to further clarify and enhance our manuscript based on your feedback.

Below are the responses to the questions you asked:

 

1. Thank you for pointing this out. Initially, we designed the model without the atmospheric term and found that the pure angular model had a large seasonal fluctuation. This was affected by the atmospheric parameter after our troubleshooting and analysis. Using Aqua's prediction as an example, the model's accuracy in the 2130 nm band improved from 2.32% to 0.74% after adding the atmospheric parameter, as compared to the Aqua observations. The atmospheric parameter significantly reduces seasonal variations in the model, particularly in the atmospheric absorption band.

 

2. Thank you for your suggestions. We have replaced this graph with a more consistent red band data and corrected the reflectance to Landsat 8 OLI band 4 by RSR matching according to Villaescusa-Nadal (2019). The reflectance now shows high consistency throughout the Landsat series of satellites.

"Figure 2 shows the stability of TOA reflectance in the Libyan Desert from 1985 to 2021. The TOA reflectance data in the red band are extracted from the Landsat series of satellites. There are minor variations in the relative spectral response profiles among corresponding bands from different Landsat sensors [16]. The data from Landsat 4, 5, and 7 have been spectrally adjusted to match Landsat 8 OLI band 4 [17]. The TOA reflectance does not exhibit any significant trend."

"Figure 2. Long-term clear-sky TOA reflectance in the red band of Landsat satellites."

[Reference] J. L. Villaescusa-Nadal, B. Franch, J. -C. Roger, E. F. Vermote, S. Skakun and C. Justice, "Spectral Adjustment Model's Analysis and Application to Remote Sensing Data," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 12, no. 3, pp. 961-972, March 2019, doi: 10.1109/JSTARS.2018.2890068.

 

3. We apologize for the error and have now corrected the caption to

"Figure 12. Observed and predicted TOA reflectance data for Terra MODIS."

 

We believe these responses and planned revisions address your concerns and improve the manuscript. We look forward to any further guidance you may provide.

Sincerely,

Authors

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript was significant improved after the revision, and thanks the authors for adding the uncertainty analyses.  Majority of my comments have been addressed.

 1.       Section 4.6 and Table 12: Is the uncertainty of the model also a function of VZA? Uncertainty usually increases with VZAs. Similar observations were also reported in Fig. 16 and lines 477-482. The model coefficients were developed for each VZA/VAA group. Moreover, SZA varies with day of year. It is more practical for the calibration community to select data based on VZA instead of SZA to validate the calibration performance of a satellite radiometer using the model proposed in this study.

 

2.       In Table 7, please also add fitting coefficient values for the groups 2 and 9-10 defined in Table 6 (VZA <~30 degree). The coefficients for the near nadir groups are very useful for applying the model developed in this study to validate various satellite radiometers. Sorry for not providing this comment during the first review.

Author Response

Dear Reviewer,

We are grateful for your acknowledgment of the improvements made to our manuscript, including the addition of the uncertainty analyses. Your guidance is invaluable to us. Here’s how we’ve addressed your latest comments:

1. Thank you for pointing this out. We agree that using VZA is reasonable as it better reflects the patterns of change in calibration performance, while SZA varies throughout the year. This is more practical for readers in the calibration community. We have modified Table 12 as shown below.

2. Thank you for your suggestions. Table 7 has now been expanded to include information on three additional groups.

We believe these updates have further strengthened our manuscript and made our model more accessible and practical for the calibration community. If you have any more suggestions or require additional clarification, we are more than willing to discuss them.

Thank you once again for your constructive feedback.

Sincerely,

Authors