General Comments:

This paper presents NO2 results from the GEMS instrument for June-August 2021 using the POMINO algorithm from Peking University. POMINO involves detailed improvements to cloud retrievals, surface reflectance and profiles and has previously been applied to measurements over Asia from TROPOMI and OMI. This is the first NO2 retrieval I have seen from GEMS in the refereed literature, and it is exciting to see this first attempt at NO2 retrievals.

Unfortunately the early operational version of GEMS NO2 slant column retrievals have shown significant biases, and so the authors apply a scaling to the slant columns using TROPOMI and GEOS-Chem. This is not ideal, but at least allows the authors to proceed with a comprehensive NO2 retrieval while the NO2 slant column retrievals are being improved.

The comparisons with MAX-DOAS and an extensive set of surface monitors show some biases but are actually pretty promising for a first attempt of NO2 retrievals from geostationary orbit. It would be nice to see a more detailed discussion of possible uncertainties in the product, but on the other hand, this is a first attempt and there will likely be many campaigns and retrieval improvements to come that will help to isolate error sources, and perhaps those discussions can be saved for future work.

As this is the first geostationary mission able to measure NO2, it would be nice to see more discussion about sources of diurnal uncertainties (even qualitative discussion). The MAX-DOAS and GEMS NO2 seem to have different trends in the afternoon measurements at many sites. What could cause this? How accurate are the GEOS-Chem profiles over a day? Do they look like the MAX-DOAS profiles? Are any errors expected from the application of a LEO BRDF to an AMF calculation? Also, there is no discussion of MAX-DOAS uncertainties themselves.

Overall, I think this is a well-written and clear paper, and a careful first analysis of GEMS data. I recommend it be published after addressing a few minor comments.

Specific Comments:

Line 75: This is a specific technique that is used for many missions and trace gases, but not all (for instance, direct fitting of radiances can also be used). Suggest change to more general “using spectral fitting” or similar.

Line 88: Are they using an online calculation or look up tables based on VLIDORT?

Line 154: “daily NO2, pressure, temperature and aerosol vertical profiles”. These haven’t been introduced yet. Are they coming from the GEOS-Chem model or TROPOMI?

Figure 2 caption: Is GEMS product only at TROPOMI overpass time or all hours? Described in text but should also be mentioned in caption.

Section 2.1.3: There must be several assumptions made to use this method of scaling GEMS to TROPOMI. Can you mention them? For instance, geometric AMFs won’t account for GEO vs LEO issues like relative azimuth angle. Do these make any difference?

Page 232: What do you use over water where BRDF is not available (open ocean) or is inaccurate (for example in coastal regions)?

Page 262: Since these data are being used for validation, it would be good to further justify “multiplied by a factor of 2 to roughly account”. Does NO2 necessarily change linearly in those bottom 130 m?

Section 2.4: What are uncertainties in MEE measurements and what are the details of the observations? Are they chemiluminescence measurements that suffer from biases in NO2? This is mentioned later but I think is appropriate to include it in this section.

Figure 7 and Line 342-354: The bias between GEMS and TROPOMI is different between ocean and land. Several reasons are given but I don’t understand why these would produce different biases over land and water – is it just that the biases are actually following locations of no aerosols vs high aerosols and not necessarily associated with water/land? Is there any way that the surface itself can influence this bias?

Line 455: “assume no error contributions from the GEOS-Chem-based scaling”: Wondering here on what this assumption is based? Are there any references describing accuracy of diurnal column variation of NO2 from GEOS-Chem?

Line 460: Related to previous comment, how good are NO2 a priori profiles from the model at various times of day? Does uncertainty vary over the day? Also, there is a free troposphere NO2 bias in GEOS-Chem which can give large errors in NO2 measurements over remote regions – maybe mention this as a source of uncertainty.

Technical Comments:

Please define POMINO acronym early on. I’m not sure what it stands for.

Figure 6: Consider adding another set of lat/lon values on the axes. For someone not very familiar with the shape of Chinese provinces, it’s hard to figure out the region being examined.

Figure S5: I find this figure very hard to read, even when zooming. Perhaps increasing the resolution would help (or maybe color palette and/or symbol size?). The sub-figures are even harder to decipher. What are these – they are lacking circles and it’s not clear if they are a measurement like the others?

Line 258: Write out “molecules” instead of using “molec”

General comments

This paper presents NO2 results from the GEMS instrument for June-August 2021. As the GEMS NO2 slant columns are biased, the authors present a correction at S5P overpass time based on the TROPOMI NO2 SCD. The stratospheric correction is also based on the TROPOMI NO2 product, and on the GEOSCHEM model, including its stratospheric diurnal variation. The POMINO algorithm is then applied to derive the AMF and the final NO2 tropospheric columns. The POMINO GEMS NO2 columns are finally compared with the POMINO TROPOMI product, as well as with MAX-DOAS columns and NO2 surface concentrations.

My main concern is the strong correction applied to the GEMS observations. At S5P overpass time, the GEMS NO2 SCD are basically replaced by the TROPOMI NO2 SCD, on a grid cell basis. To my understanding, the only true GEMS NO2 information remaining is the diurnal variation relative to the mid-morning values. Unfortunately, the MAX-DOAS validation results are poor when it comes to diurnal variations. This is a serious limitation. This should be further discussed in the paper.

On the same idea, the authors present a comparison between POMINO GEMS and PROMINO TROPOMI. The comparison results are obviously very good, but the study is biased. I strongly recommend to use independent satellite NO2 products; such as OMI or GOME-2 products. The addition of OMI and GOME-2 would allow to compare with the GEMS observed diurnal variation. 

Since this is the first study about GEMS NO2 measurements, the paper should provide a section where the GEMS operational VCDs are compared to the presented product and provide some conclusions on the regions and periods where the GEMS NO2 tropospheric VCD are performing good or bad.

In the diurnal variation plots (figure 9 and figure 11), the uncorrected GEMS NO2 VCD should also be plotted. (uncorrected GEMS NO2 VCD = uncorrected GEMS NO2 SCD – NO2 stratospheric columns)/POMINO GEMS AMFs.

The paper is a well-written and clear. I recommend publication after addressing the above major comments.

Specific Comments

Abstract

Line 33: I suggest to remove the very first sentence, that sounds a bit obvious and is already in the introduction: Nitrogen dioxide (NO2) is a major air pollutant.

Line35: LEO NO2 retrievals are not limited only by insufficient temporal sampling, but also by retrieval uncertainties and spatial resolution. The two last limitations exist also for GEMS.

Line 37: at an unprecedented hourly resolution during the daytime.

Line 41: “We then derive tropospheric NO2 air mass factors (AMFs) with explicit corrections for the anisotropy of surface reflectance and aerosol optical effects, through pixel-by-pixel radiative transfer calculations.” The authors do not present the impact of those two corrections in the rest of the paper. It should either be presented in the manuscript (see my AMF comments later) or removed from the abstract.

Line 44: The term “reveals” is overused, since the NO2 hotspots signals are well known from LEO observations.

Line 45: As intended by the presented method, POMINO-GEMS NO2 VCDs agree well with POMINO TROPOMI v1.2.2 product. Please indicate in the abstract that the remaining differences are coming from AMF differences.

Introduction

Line 66: the provided references are for NO2 datasets rather than LEO missions themselves. Please add more appropriate references for GOME, OMI, GOME-2, TROPOMI.

Line 90: Validation results have shown the overall capability of the official GEMS NO2 algorithm. I’m not sure this is true. You should provide reference to support this affirmation.

Method and data

Line 128: Please explain briefly what is meant by “continuum reflectances”.

Line 140: Please explain briefly what is meant by “area-weighted oversampling technique”.

Please provide basic information on the slant columns retrieval settings for GEMS and TROPOMI operational products: wavelength interval, cross-sections, reference spectrum.

Total NO2 SCDS

The correction based on the  TROPOMI SCDs is somehow radical, since is it calculated for every grid cell. Have you tested more softer corrections, for example based on much larger grid cells, or based on meridionally averaged grids?

More examples of figure 2b and d could be shown for other GEMS hours (maybe in the supplement).

Line 199: Please comment on the diurnal variation of the GEMS systematic problems. For example, is the high bias over northern and northwestern part of GEMS FOV constant during the day or does it increase?

AMFs

A figure presenting the POMINO GEMS amfs should be added, as well as a comparison with the POMINO TROPOMI AMFs.

Estimation of surface NO2 concentrations

Please specify if the Rgc GEOS-Chem simulated ration is time dependent or constant. In other words, is there a diurnal variation of the model introduced with this correction? If yes, what is the observed GEMS diurnal variations if you use a constant ratio?

Line 277: Please explain briefly what is the grubbs statistical test and provide a reference.

Results and discussion

As the NO2 total and stratospheric SCDs are almost the same by definition of the presented “fusion” technique between GEMS and TROPOMI, I suggest to skip section 3.2 and to replace it by a comparison of AMFs from POMINO GEMS and TROPOMI.

Figure 8: the regression line values are exactly the same between plots a and b. This seems strange, please check.

Figure 9: please use a fixed scale, or at least only two different scales for high and background NO2 levels.

Figure 11: I suggest to detail the comparison with MEE diurnal variations for different groups of sites (urban, rural, northeast, southwest China). This could provide more information on the regions where the GEMS diurnal variation is valid or not.

Since the uncertainties on the measured diurnal variations appear to be large, I suggest to applied to the MAX-DOAS measurements a similar “column to surface column transformation” as for the satellite columns, and to compare directly MEE and MAX-DOAS diurnal variations.

Error estimates

10% error on the GEMS NO2 SCD (or we could say on the TROPOMI NO2 SCDs) seems to be underestimated. Furthermore, the diurnal variations of the error o     n the GEMS fits is not taken into account.

Conclusions

The observed  added-value of GEMS should be discussed in a more balanced way in the conclusions, as well as the current limitations.

Review of “POMINO-GEMS：A Research Product for Tropospheric NO₂ Columns from Geostationary Environment Monitoring Spectrometer” by Zhang et al.

The Korean GEMS satellite is the first of a series of geostationary satellite instruments providing hourly observations of key air pollution species, including NO₂. These data are of large interest for air quality studies. As the current operational GEMS tropospheric NO₂ product still has some deficiencies, there is the need for improvements, and this manuscript is aiming at improving on that.

The approach taken in this study is to use the reprocessed PAL version of the TROPOMI NO₂ product to determine a pixel specific slant column offset of the GEMS data at TROPOMI overpass time, and to apply it to all GEMS measurements. The stratospheric correction is based on TM5 stratospheric VC data, again from the TROPOMI product, together with the diurnal variation taken from a GEOS-Chem run. Cloud correction and AMFs are computed using an updated version of the POMINO retrieval framework. The algorithm is applied to three months of data and the resulting columns compared to TROPOMI data, MAX-DOAS observations and in-situ surface measurements.

The manuscript is clearly written, covers a topic of interest to the AMT readership and reports on a relevant study. However, I have some concerns that the authors need to address before the manuscript can be accepted for publication.

Major comments

My main criticism about the paper is that the approach taken (correction of GEMS SCD data using TROPOMI retrievals) is a temporary solution at best. Clearly, problems in the GEMS SCD retrievals need to be solved in the spectral fit and not using an ad-hoc correction linking it to data from another satellite instrument. Also, the assumption that all the problems in GEMS data can be described by a slant column offset determined at the time of TROPOMI overpass is probably not correct, as solar zenith angle and relative azimuth change over the day. Therefore, the most important measurement quantity of GEMS, the diurnal variation of NO₂, could be affected by the applied method.

It is also important to realize, that GEMS and TROPOMI data taken at the same time of the day do not have the same scattering geometry, and thus not the same AMF. The slant columns can therefore be different, even after geometric correction. These problems of the current approach need to be discussed in the manuscript.

My second concern is about the comparison of GEMS and TROPOMI data shown in the manuscript. As GEMS slant columns are forced to agree with TROPOMI data, this comparison makes little sense and only shows that no technical mistake was made. The only comparisons providing additional information are those to external data.

My third point is, that the uncertainty discussion is very superficial and in my opinion not correct. The SC uncertainty should be driven by shot noise and therefore be described as an absolute, not a relative uncertainty. The overall uncertainty of 0.2E15 molec cm2 derived for the tropospheric SCDs appears very low, but it is anyway not clear if this is the uncertainty for an individual GEMS measurement, a monthly average, or the three monthly average discussed here. This discussion needs to be improved.

In the data availability section, it is stated that the data is available through http://www.pku-atmos-acm.org/acmProduct.php/. This does not appear to be the case and data could therefore not be checked for this review.

Minor comments

Line 118: Isn’t the current GEMS NO₂ product provided at 3.5 x 8 km2?

Line 128: How does the known GEMS uncertainty in irradiances affect the reflectances and thereby cloud retrievals?

Line 262: This ad hoc factor needs to be mentioned again when later comparing the retrievals with the in-situ observations

Line 277: Please provide a bit more information on this – how many data points were excluded? What exactly were the criteria?

Figure 4: What are the regions shown in grey in the figure? Are these negative values or missing data?

Figure 6 / Line 323: I do not find the discussion of the observed increase in NO₂ convincing. The observed changes are large and have clear patterns, and I suspect they are retrieval artefacts.

Figure 7: As discussed above, the only surprise with this figure is that the agreement is not even better.

Line 433: Maybe mention that the main difference between column and surface concentrations is that the column is insensitive to boundary layer height changes.

Line 443: Please provide information on for what the uncertainty calculations are made – individual measurements or averages?

Line 448. Why should SCD have a relative uncertainty?