Review of “TROPOMI Level3 tropospheric NO2 Dataset with Advanced Uncertainty Analysis from the ESA CCI+ECV Precursor Project”

The paper by Glissenar et al. presents the creation of a monthly global Level 3  dataset of TROPOMI tropospheric NO2 VCDs and uncertainties at spatial resolutions of 0.2°, 0.5°, and 1°. The spatiotemporal averaging of the NO2 data and its uncertainty is described in detail. Spatial and temporal error correlations for all sources of uncertainty in the L2 retrieval are analyzed. The total relative uncertainty in the resulting Level 3 dataset is analyzed globally for different levels of pollution. The tropospheric, stratospheric, and total vertical NO2 column is validated with ground-based measurements.

The study covers the important topic of creating L3 datasets with proper uncertainty analysis, which have received less attention from the scientific community than the L2 data product. This L3 dataset is of interest for atmospheric chemistry studies, for evaluating atmospheric models and analyzing spatiotemporal NO2 trends. The study contains important analyses of the created L3 dataset. However, in some parts the paper is hard to follow and contains some inconsistencies which could be improved by doing some major revisions and addressing the comments raised below.

General comments:

The study by Rijsdijk et al (2024) is an important study for your analysis. I think it would be necessary to introduce the main study results in the introduction before you reference it several times throughout your study.

Chapter 3 Methodology: This chapter describes how you have averaged the L2 NO2 column data and handled the uncertainties. It is the main part of your study, however, I think it is hard to follow, and would benefit from a detailed check: Does the reader know this variable already? Mention that, e.g., uncertainty x is described in more detail in subsection x. Please, check for consistency in variables and their indices. See also specific comments below.

Section 3.3.1: Isn’t this weighting creating a bias? If I understood correctly, superobservations, which are probably clear-sky observations, have a higher weight; do these have a tendency to lower/higher NO2 concentrations and create a bias? Can this be neglected in your averages, and using only observations with cloud radiance fraction < 0.5?

Inconsistent use of abbreviations SCD and N_s for the slant column density, same for the air mass factor with AMF and M^tr.

The word column is often used without defining whether it is slant or vertical.

Specific comments:

Line 8: Spatial error correlations arise not only from these two, but also from the a priori model. Add “mainly” or mention also the a priori model.

Line 9: Is it important here that the albedo climatology has a coarse grid, because it is coarse compared to the TROPOMI pixel but similar to the grid of the L3 dataset?

Line 10: You name the temporal error correlation to be 30%. Before you mention the spatial error correlation, but you are not mentioning a value.

Line 29: This sentence is incomplete, what kind of compounds do you mean?

Line 35: I think you should not only mention in-situ but also remote-sensing here.

Line 38: I think it might be a good point to introduce some satellite instruments here, especially the ones like OMI and TROPOMI you mention later, because later it might be good to know, and maybe not obvious for everyone, that OMI is the “precursor” of TROPOMI. You mention OMI in line 83 without introducing it.

Line 74: I would avoid the word observation here because the NO2 is not the observation but the resulting product. You could also mention that there are also other products from TROPOMI, besides NO2.

Line 90: This sentence is not clear: Replace “this version” with the version you mean (v2.3.1 ?) Was the qa_value bug corrected in this study? If yes, add this information and what was done. Replace (qa) value with qa_value, which was already introduced.

Line 102: What is meant by alternative processors? Please also provide references for that.

Line 105: You wrote start fields from TROPOMI? Which data exactly? TM5 has not been introduced yet.

Line 107: Is Dirksen et al. 2011 a correct reference here? It is a study on OMI stratospheric NO2. It doesn’t have any TROPOMI and model differences, which are the topic of this sentence. Is it applied in the TROPOMI or the OMI L2 algorithm?

Line 109: I can’t follow the logic here. You say that you apply a more detailed latitude- and time- dependent L2 uncertainty as derived by Rijsdijk et al. (2024). What is done in Rijsdijk, what is the connection to the sentence before and after?

Line 113: All these studies are for OMI or GOME-2; are the results applicable to TROPOMI?

Line 133: I think it would be helpful if you are more precise here: “temporal averaged estimates of the column values”. Do you mean “of the spatially averaged column values x_o,t”, which are, when temporally averaged, named x overline?

Line 135: of the retrieved column (x_i) and its uncertainty (sigma_i)

Line 136: sigma_o,t was not yet introduced, you only mentioned sigma_m,s, and sigma_r,s. I think it would be helpful if you introduce sigm_o,t together in line 131.

Figure 1: Shouldn’t it be x_o,t instead of x_o in step1? In step “0” you have x_i, sigma_i, and w_i, but in step 1 and 2 you only have x_o and x_overline, I think it might be helpful to add the uncertainties here as well.

Line 148: How were the grid resolutions of 0.2°, 0.5° and 1° selected?

Equation 2, Line 153: At this point, I was wondering if sigma_m and sigma_m,s, which you introduced in line 131 together with sigma_r,s are the same. In the following it is more clear that they are not, but it is hard to follow.

Line 180: Are you sure this is the correct reference here, it is not containing TM5 or TROPOMI?  

Line 186: I think it is important to mention here already that the a priori is in general also a large contribution to the uncertainty but is shown to become irrelevant when the averaging kernel is used (as you mention in line 259) but is still discussed later.

Line 199: What is the meaning of a correlation length?

Line 226: How is g determined?

Line 226: How do you define superobservations?

Line 233: Is it correct that T is the total number of valid superobservations? I thought that L3 column is averaged over all valid observations, with giving a higher weight to the superobservations.

Line 241/244: You mention the spatial representativeness uncertainty twice but not the a priori uncertainty.

Line 299: Only to avoid misunderstanding, you say “carried out for the NH summer period 1 June -31 August” but the correlation coefficient is determined globally, is this correct? Have you added the information NH to clarify that it is summer on the NH but winter on the SH? Do you expect differences, should you differentiate between NH summer/winter and SH summer/winter?

Line 321: This is not really clear to me. To correct for what, the temporal representativeness?

Line 360: You write “this method” but I think the reference is not clear. Do you mean the sampling test/Wald-Wolfowitz test?

Line 384: Probably a combination of longer lifetimes and higher emissions due to heating. Add a reference.

Line 389/390: Do you have an idea why the difference in the average L2 relative uncertainty is so large between Amsterdam (52%) and Beijing (28%)?

Line 429: I think this reference doesn't contain the 10 mentioned sites, but if it's for SAOZ in general, then mention it directly after SAOZ.

Line 436/Fig 8: You have written an excellent agreement. Do you have an idea why there is perfect agreement from July to October but actually quite some deviations always in spring?

Figure 9: You mention an explanation for the deviation of the Paris site. Do you have an idea for the large deviation of the Dome Concorde site in Antarctica? Is it also visible in the L2 Verhoelst et al. study?

Figure 10/Line 450: The comparison for Xianghe shows a good agreement in summer but is low biased in the much more polluted winter months. The polluted winter months are the months with higher spatial and temporal variability, which are smeared out in the L3 data. I think this could be discussed in the text.

Figure 9/10/13: Please add data period (2018-2021?).

Line 493: “clear-sky or low-cloud conditions”, I think low-cloud conditions is misleading, please be more precise.

Line 532/Figure A1: Do you have an idea why the difference between the ascending and descending part of the orbits is much less obvious in the Antarctic region?

Technical corrections:

Line 4: Remove brackets around NO2

Line 6: Introduce Level 2 (L2) like you have done in line 3 for L3 and use L3 and L2 in lines 13 and 14.

Line 14: Replace separate with individual.

Line 14: GCOS is not a commonly known abbreviation, please introduce it.

Line 15: Replace (sub-)columns, I think it is not clear what is meant by that. “Validation of the tropospheric, stratospheric, and total columns”

Line 29: tropospheric NO2 columns instead of tropospheric columns NO2

Line 36: Change measurement techniques to measurements.

Line 38: Change “make them fit for purpose for climate monitoring”

Line 42: This is a long sentence, I would suggest splitting it into two sentences: …by the scientific community. However, L3 data are relevant for model evaluation…

Line 59: A long sentence, you could split it after the reference to Labzovskii.

Line 65: into instead of in to

Line 67: “of with” delete of

Line 68: You mention ESA CCI+ here for the first time; it is not clear what it is. Remove it, or maybe even better, introduce it in your introduction.

Line 80: Change to “are the L2 TROPOMI NO2 tropospheric vertical columns on an orbital basis”

Line 82: Remove brackets  to have it like this OMI QA4ECV v1.1 product (Boersma et al., 2018)

Line 82: Add TROPOMI and RPRO or PAL information in front of the v2.3.

Line 83: Split into two sentences: …(Boersma et al., 2018). These OMI and TROPOMI data products…

Line 84: Add information about which period of data is used. …which allows for better merging of the datasets, and allows using data from year x to year x.

Line 93: First time use of slant column density, introduce abbreviation SCD or N_s. Remove the introduction of SCD in line 100 and use the abbreviation in the following, e.g., line 103.

Line 113: You have used OMI abbreviation already before.

Line 113: Please introduce QA4ECV.

Line 120, 123: Please be precise: stratospheric SCD instead of stratospheric column. Tropospheric vertical column instead of tropospheric column. Abbreviation Nvtrop was not introduced yet.

Line 132: sigma_rs comma is missing between r and s

Line 148: Please correct “we provide spatial the here created dataset at resolution…”

Line 181: I would suggest to mention the used grid sizes earlier: This is coarser than the spatial average grid sizes of 0.2 to 1.0° considered here. Therefore, it is assumed that the error in the stratospheric column is fully correlated in space with the L3 grid resolution.

Line 218: Change to “the L3 grid resolution”

Figure 2: “Larger black points” instead of “Larger points”

Line 269/270: an/the albedo

Line 299: Add: for the NH winter period 1 January to 31 March

Line 399/Figure 5/Figure 6: Add that it is the vertical column.

Line420: remote sensing measurements

Line 425: L3 qa_value

Line 432: Replace S5P with TROPOMI, you have always used TROPOMI.

Line 433: Delete operator.

Line 443: Various instruments/operators instead of sources?

Line 449: The location of the given reference doesn’t make sense here.

Line 451: Please check this sentence for typos.

Figure 12: Please add the L2 version, similar as in Figure 10.

Line 477: I think it is -39% instead of -35%

Line 480: from nearly 20%