Zou et al. corrected the degradation trend of GOME-2A SIF with a pseudo-invariant method, based on a 1 degree x 1 degree calibration region in the Sahara Desert. Unlike previous studies, the correction in this study was conducted at radiance level and daily basis, using a fitted quadratic polynomial function. Then the SIF signal was retrieved using a data-driven algorithm, which was further scaled to monthly average with a PAR-based upscaling model. The global trend of the corrected SIF data was also compared with several existing SIF/GPP/NDVI products. The study is overall nicely conducted. Please find my comments below:

Major comments:

1. Benchmarks used for trend evaluation: The authors employed MODIS NDVI and a number of GPP datasets from MODIS and model simulations, to evaluate the trend of TCSIF. However, MODIS NDVI has a known saturation effect, which would especially influence the evaluation of annual maximum (Fig. 11). MODIS GPP and TRENDY GPP are also known to have issues with their trends (Anav et al., 2015, https://doi.org/10.1002/2015RG000483). I would not trust the evaluation against these datasets.
2. Pseudo-invariant method and site: The authors employed a pseudo-invariant method and selected a 1 degree x 1 degree non-vegetated region which trend was taken as the temporal degradation in the GOME-2A instrument. It has a strong underlying assumption that the degradation trend is the same for different locations and different radiance levels. It remained to be tested/discussed to what extent this is true. For example, in Fig. 9, the authors showed the difference in the temporal trend with and without the correction, which is basically the degradation trend. There are large spatial variabilities in the degradation trend.

Minor comments:

Line 13: This is not accurate. As the authors cited in Table 2, there have been several efforts for correcting the temporal degradation of GOME-2A SIF.

Line 19: By “weather conditions”, I assume it referred to only light conditions?

Line 88: MCD43C4 is provided at daily resolution.

Line 112: The authors did not provide the links for the datasets used in this study. For NASA SIF, there is a recent updated version here: https://daac.ornl.gov/SIF-ESDR/guides/MetOpA_GOME2_SIF.html. It is generated with updated GOME-2 Level 1B radiances and irradiances, and I heard that the degradation trend has been largely corrected. If the authors used an older version, I’d suggest comparing with this updated dataset too.

Line 159: The function f should be introduced here.

Line 217: “upscale the daily observations to monthly values” – This is confusing. The previous sentence emphasized the potential bias of upscaling instantaneous measurements to daily values.

Line 233: Could you clarify what the “normalized coefficients” stand for?

Fig. 10: The two green colors are not very distinguishable visually. Also, comparison with other datasets in terms of the trend maps may be useful, e.g., which regions have significant increase/decrease, are they consistent among different datasets?

Line 311: Please justify why the trend of annual maximum was selected for evaluation, not annual mean or annual minimum.

Fig. 12: it is also worth noting that the year-to-year variations are very different among different SIF products. Any idea why? Could you also add NASA SIF here to see if TCSIF has consistent year-to-year variations with NASA SIF?

Some editorial suggestions:

Line 41: “Given” -> “Given that”

Line 52: “Forest” -> “forests”

Line 116: “expansion” -> “extrapolation”

Zou et al. explored a pseudo-invariant method to resolve the temporal degradation issue with GOME-2A data and tested how the corrected SIF product improved. The study is of great significance in providing more reliable long-term SIF data. The manuscript was well written and the messages were well delivered. See my detailed comments below.

Major comments:

1. Not enough information has been given regarding why the GOME-2A is subject to degradation and where it occurred except in lines 43-44 that "GOME-2A is an optical spectrometer that measures reflected sunlight and is therefore sensitive to instrument degradation". If degradation is a problem for all the optical spectrometers, it should also be a problem for MODIS VIS/NIR bands. Identifying the reason for the degradation is crucial for determining the correction method to apply. The correction method applied in the present study assumes the GOME-2A radiance is uniformly downscaled at all wavelengths. This would not be correct if the degradation is caused by a dirty lens. Some results and discussions are required to validate this assumption.
2. A more systematic result session (data validation) is required. The pseudo-invariant method is actually a 2-stage correction to SIF: radiance correction and SIF retrieval. Thus, the method needs to be validated at both stages: comparison of radiance/reflectance to other products such as MODIS, and comparison to other products such as OCO-2 and TROPOMI. The data validation of the radiance/reflectance was missing from the present study.
3. Following my comment 1, the GOME-2A TOA reflectance at the calibration site (before calibration) has a very clear seasonality in it. Similar seasonality is also found in MODIS data as shown in Fig. 2. I believe the seasonality is due to the BRF effect caused by sun-sensor geometry. But the variation of GOME-2A reflectance seems to be higher than MODIS, why? Also, the GOME-2A reflectance variation seems to increase with time (Fig. 3a). If the degradation is due to the dirty lens, the radiance/reflectance variation should also scale with the Dfactor, right?

Minor comments:

• Line 10. SIF cannot provide a "direct way" to monitor photosynthesis. SIF and GPP are not linearly correlated.
• Line 11. I thought the TROPOMI and OCO SIF datasets were more popular. It is not necessary to say it is the most popular.
• Line 28. of monitoring -> to proxy.
• Line 45. The contamination assumption does not seem to be able to explain the variations in TOA reflectance.
• Line 91-95. The use of external PAR to rescale SIF makes the SIF data more prone to errors in external data. Also, since the GOME-2A radiance is corrected using external NIR data from MODIS, should it be considered an L3 product? My understanding is that the L2 product is purely inferred from the L1 product without any external data correction.
• Line 96. Should it be better to use MODIS data to validate the corrected GOME-2A radiance/reflectance and other SIF data to validate the SIF product? The NDVI/GPP can only be used as indirectly supporting results.
• Fig. 1. How about other bands? Are they also very stable?
• Fig. 2. Since there is no vegetation in the calibration site, the spectral curves should align to the same standard curve when rescaled properly? If you rescale them, is it the case? If so, the Dfactor method is fine; if not, the Dfactor needs to be a wavelength-dependent function!
• Line 221. Is the EVI used as a f_APAR here?
• Fig. 3. The degradation looks to be an exponential curve, why use a polynomial function to fit it?
• Fig. 6. Combine it with Fig. 5.
• Fig. 7. You also need to show examples of how the TCSIF and NASASIF differ in the year 2021 (or more recently) to show the degradation effects.
• Line 307. Consider moving these indirect results to a separate section to the very end or discussion.
• More direct validation of SIF to OCO/TROPOMI and reflectance to MODIS are required.