General Comments:

This paper presents analysis of GOME and SCIAMACHY surface reflectance data at so-called Pseudo-Invariant Calibration Sites to evaluate cross-calibration between sensors and also temporal stability to evaluate deviations over instrument lifespan. This paper focuses on sites in the North of Africa and Arabian Peninsula, which are valuable due to the need for satellite-validation datasets at high-albedo locations. This work is within scope for AMT. While this study contributes a comprehensive overview of basic metrics for evaluating the suitability of a selection of PICS, my main comment is that the analysis may be improved by analyzing site-to-site differences and evaluating how they reflect in the stability metric.

Specific Comments:

Line 38: what does “spatial uniformity” refer to? Your text states this is based on a series of cloud-free images and seems to imply this uniformity is the peak-to-peak temporal variation in reflectance, but please clarify.

Line 47: re-word “optical sensors”, as that term may include spectrometers.

Line 54: There is only one hypothesis (explicitly) listed.

Table 1: This table summarizes the information in Sections 2.1 and 2.2. I suggest removing Secs. 2.1 and 2.2.

Line 94 and Table A1: It is not clear to me how exactly these sites were chosen. Was the number of data points a factor in this decision? It is mentioned that a subset of the chosen locations are already currently recommended for sensor cross-calibration, so it is a good idea to differentiate those and new locations.

Line 118: CF < 0.25 was used to filter the reflectance time series. Is this heuristically determined? CF 0.25 may be large enough to strongly affect the reflectance at certain wavelengths. It would be good to evaluate how sensitive this threshold is (e.g., at least report how many observations are discarded).

Table 3 and Figure 2 are useful summaries. I wonder if it may be better to shorten the description of such variables by adding another column to Table 3 with each respective calculation, as these are fairly common metrics.

Line 188: Why are these particular wavelengths selected? Is there a wavelength-dependence to this variability?

Figures 7 and 8: Is there a physical interpretation for the negative skewness in the NIR in panel (f)

Figure 12: Consider moving this to the appendix or supplementary, or condense this into a Table. Figure 11 contains the more interesting aspect, which is the wavelength dependence of the SS.

The discussion of Sections 5.2, 5.5, 6.5 are very qualitative in nature as they generally repeat the information in the respective figures. 

Line 272: Cloud contamination is mentioned again here — how did your previous CF filtering improve this influence? Are there references in the literature to support differences in cloud formation (brightness, microphysical properties?) between PICS which may explain some site-to-site differences?

Figure 11: UV vs. VIS/NIR SS very different for certain sites. Arabia3 and Algeria1 stand out. What makes Libya 3 distinct from Libya1, 2 and 4 given their geographical proximity? A brief discussion is presented in the paragraph starting Line 339, so this may be expanded. References are lacking in this discussion especially.

Conclusion: I would suggest including a final paragraph for the broader satellite community in how the information gained by analyzing high SS PICS could reflect in other satellite products, such as, as mentioned in the introduction, atmospheric gas measurements. 

Technical corrections:

There are many small technical changes, so I would suggest carefully proofreading the manuscript beyond these corrections in your revision:

Line 28: this paragraph only has one sentence - append this to the previous paragraph

Lines 35 and 38: fix in-text citation “… in Cosnefroy et al. (1996)” 

Line 64: missing indentation.

Line 70: “over 240-790” -> between 240-790

Equation 1: Dots (assuming multiplication) are not centered with the variables.

Line 126 (and subsequent lines): inconsistent italics for parameters. E.g., CV is italicized in lines 127 and 129 but not Equation 4. 

Line 156: observations -> observation

Line 157: 45 (degrees)

Line 170: “see the illustrative diagram in Fig. (2)”

Figure 2: “SS computation” overlaps a vertical line. Also force “SS_NIR” onto one line.

Line 195: “showed a more consistent pattern”

Figures 7: (1) Axis labels are too small, (2) x-axis label should be centered on all three subpanels of each panel, (3) remove some wavelengths from the x axis for the VIS/NIR spectrum to improve readability, (4) the UV subplot for each panel has horizontal lines but the VIS/NIR do not. Is the y-axis to-scale for all three spectral regions? (5) perhaps add some transparency to the lines, as it is impossible to see clearly, especially in the NIR. (6) panel e needs more significant figures.

Same comments for Figure 8, as applicable. 

Section 5.3 header: Inconsistent capitalization

Figure 12: The +/- labels for many sites are overlapping, and the two Mauretania labels on the x-axis are mis-aligned

Line 309: Missing period for middle name in Lieuwe G. Telstra?

Line 324, 325: Fix in-text citation for Wang et al. (2022) and Kim et al. (2023)

Line 337: Sudan1 and Arabia2 exhibited a lower score than Libya4

Section 7: rename to Conclusions (because you have more than one conclusion)

Summary of the paper

This paper addresses the (top-of-atmosphere) reflectance stability over 20 pseudo-invariant calibration sites, based on the GOME and SCIAMACHY instruments. The purpose is to extend the use of PICS from imagers to spectrometers by developing a statistical framework for comparison and application of this to the selected PICS.

The method used to obtain these objectives is statistical analysis of the extensive nearly decade-long overlapping timeseries of GOME and SCIAMACHY measurements, after correction for solar zenith angle and viewing zenith angle variations for the selected PICS, and filtering on proximity to the pics and cloud fraction. The corrected reflectance for selected wavelength ranges in the UV, visible and NIR is then statistically analysed resulting in a range of metrics.

The key results and conclusions are the definition of a robust framework for sensor performance evaluation, radiometer drift monitoring, and identification of the constrained and reference sensors for cross-calibration. The stability of the 20 PICS was assessed using a Stability Score, resulting in a ranking of the PICS and identification of 3 PICS that were observed to be most stable.

Major comments

ToA versus surface reflectance. From the description of the data processing, it is clear that the top-of-atmosphere (ToA) reflectance is used for the analysis. For PICS use with imagers on the other hand, it is quite common to use surface reflectance or -BRDF, with a corresponding atmospheric correction applied to remove the impact of the atmosphere.

The main concern with the approach taken in the paper is that distinction between atmospheric variability and surface reflectance variability is difficult to make, if at all possible. From the perspective of a user of PICS for imaging applications the coefficient of variation may seem excessively high. From the perspective of a user focused on the Earth atmosphere the numbers are not at all surprising.

I suggest addressing the difference between ToA and surface reflectance quite early in the paper to make sure that reader is aware of this difference. Likewise, in the discussion on the stability of PICS this difference will need to be properly addressed; direct comparison of the ToA reflectance stability with surface reflectance stability can not be done without carefully addressing the atmosphere and/or atmospheric corrections.

SZA and VZA corrections and the slope parameter. Solar Zenith Angle and Viewing Zenith Angle corrections are considered during data processing. Figures 10a and 10b show the mean reflectance as function of these parameters for all selected PICS observations. Based on these plots the conclusion is drawn that only SZA dependence is corrected.

Do I interpret this correctly?

From basic principles, I would expect a Lambert-Beer-like behaviour for each wavelength as function of the airmass of the combined path from sun to surface to satellite, assuming that the majority of the light reaches the surface and the atmosphere is homogenous. In the case of strong absorption (e.g. ozone, oxygen, high aerosol loads, Rayleigh scattering at short wavelengths) this assumption may not be valid. Likewise, variability of atmospheric constituents will increase the scatter. A Langley plot (or the equivalent thereof for the observation of the Earth surface from orbit) should indicate whether correction is need or can be neglected for the wavelength in question.

Taking this one step further, the metrics would probably be able to better distinguish between atmospheric variability, PICS surface variability, and instrument variability if SZA and VZA airmass dependence were taken into account with an additional wavelength-dependent atmospheric absorption coefficient, as well as the instrument-dependent slope parameter. Note that from the Level 2 products the atmospheric composition is known for each observation, in case that ozone variability is to be taken into account.

Instrument degradation and scan angle dependence. Both GOME and SCIAMACHY suffer from instrument degradation, with a non-negligible component of that being scan angle dependent degradation. Both instruments have monitoring capabilities and correction for degradation during level 0 to level 1 processing, but this correction is not perfect. Especially for SCIAMACHY, the quality of the degradation correction depends on the version of the processor and corresponding calibration key data.

Please report briefly in section 2.3 which version of the data was used.

In view of the scan angle dependent degradation, have you looked at the PICS reflectance time series as function of the scan angle? Figure 5 suggests three “lines” for the 330 nm and much of the 450 nm time series. Do these happen to correspond to the East, Nadir, and West pixels? If so, does this affect your conclusions?

Seasonal dependence. Section 5.4 mentions clear seasonal dependence with the largest correction values during the summer months. This happens to be the months with highest aerosol load at many desert sites. Have you looked at the correlation between the absorbing aerosol index product or aerosol optical depth and the observed seasonal dependence?

Looking at figure A1, the heatmap of the difference in reflectance, clear seasonal variability of ozone can be seen in the typical Huggins band spectral features in the UV. Likewise, the oxygen A-band shows up clearly in the spectral behaviour of the time series. This clearly points at atmospheric effects being a large contributor to the observed variability.

You state that the required adjustments were minor, on the order of 0.05. With a typical mean reflectance of 0.10 to 0.30 in the UV to visible, the correction of 0.05 however corresponds to 15% to 50% of the typical reflectance. This is not directly minor. Please elaborate. This is also reflected in the coefficient of variation, which varies between 0.10 and 0.25 for wavelengths in the UV-vis.

Minor comments and typos

Line 48: “spatial footprint […] of PICS”. Do you mean spatial extent? Please rephrase or clarify.

Equation 13, line 172: What is the index j? And what happened to the wavelength lambda? Please clarify/correct.

Line 185-186: Since the degradation of both GOME and SCIAMACHY is corrected to some degree, “less degradation” could be better formulated as “better degradation correction”.

Line 188: Should 650 nm be 450 nm? Please correct.

Line 156-157, Table 3, and line 210-211: is the slope parameter m the same in these cases? If so, do you have a fit parameter for time dependence? If not, please clarify and update.

Line 214: Could aerosols have an influence as well as clouds?

Line 224: The “slight changes in the spectral signature of the sites” of the slope parameter is a remarkable feature that I do not directly expect or understand. The spectral behaviour of the slope parameter in figure 8e remind me of the detector etalon of SCIAMACHY (sinusoidal pattern) and around 350 nm of a detector feature that was already present during the on-ground calibration phase and which increased slowly over time in size on the detector. Apart from polarisation dependence and perhaps spatial non-uniformity on a scale of the projected length of the slit I would not expect site-dependent variation.

Line 225: The “long-term trends” cover atmosphere, instrument, as well as site behaviour. Better correction or modelling of atmospheric effects and instrumental effects should highlight any remaining site-dependent effect. As it is now, those effects seem to be drowned in the instrumental effects for the slope parameter, and in the atmospheric effects for the standard deviation and interquartile range parameters.

Line 233: In addition to stable sensor data, the effects of illumination and viewing geometry are also of high importance. Think also about the Local Time of Ascending Node (LTAN) crossing (overpass time) when comparing results between sensors. In this case, ERS-2 and Envisat have quite similar and stable LTAN.

Figure 7e and 8e: add more decimals to the vertical axis, they show only zeroes now.

Line 254: What is the definition of a spectral channel? GOME and SCIAMACHY naming suggest one of 4 (GOME) or 8 (SCIAMACHY) spectrometer channels, each consisting of 1024 spectral pixels. Other more modern definitions denote individual detector pixels with the term “spectral channel”. Please define or clarify.

Line 258: Typo: “sereis” -> “series”.

Line 259: The aerosols in the summer months (May-September) may exhibit in addition to radiometric effects also polarisation effects. The polarisation effects of GOME and SCIAMACHY have distinct spectral structures and may slowly change over time due to instrument degradation. I expect this to be a small effect, however.

Figure 10b: Is this the viewing zenith angle dependence, or does it show something else, e.g. the viewing azimuth angle? The range from approximately 100 to 300 degrees is not consistent with zenith angles. Please update.

Figure 12: alignment of the Mauretania1 and Mauretania2 names is shifted, please update.