Review of the study „Validation of TROPOMI and WRF-Chem NO2 across seasons using SWING+ and surface observations over Bucharest“ by Antoine Pasternak et al.

The study focuses on validating a high-resolution chemistry-transport model and satellite retrievals of tropospheric nitrogen dioxide (NO₂) over the urban region of Bucharest. The authors run the WRF‑Chem model with 1 km resolution over the Bucharest region for 17 two-day time series in 2021 and 2022 across the different seasons, and compare its output with ground-based in situ meteorological and chemical observations as well as airborne column measurements from the SWING+ instrument (17 flights between 2021–2022). They evaluate the satellite-based TROPOMI tropospheric NO₂ column product (v2.4.0) using the airborne SWING+ data as reference, and WRF-Chem as an intercomparison. The main findings are the different biases for different concentration ranges and the seasonality of the results. Results are compared to existing TROPOMI validation studies. They show that NOx emissions from the CAMS-REF inventory need to be scaled.



General comments:

The paper presents valuable and comprehensive validation work for both TROPOMI and WRF-Chem NO₂ data over the Bucharest region. It provides detailed comparisons combining satellite, model, airborne, and surface observations over an urban area with complex emission patterns. The analysis across different seasons and atmospheric conditions adds further strength.

I think it needs discussion, if not only a scaling of the CAMS-REG, but a seasonal dependent scaling is necessary. Can you also comment on how region-dependent your scaling factor is? I think the seasonal dependency in the validation results should, in general, be more highlighted in the abstract and conclusion.

I think section 4 would benefit from being less of a review and including more explicit comparisons of this study's results with previous validation studies.

Overall, this is a well-structured, technically sound paper, and its results are relevant to the air quality satellite and modelling community. I recommend publication after minor adjustments.

Specific comments:

L23: Missing reference.

L53: Add a reference for the lifetime information.

L55: Add a reference for the expected seasonality.

L101: What is this highest resolution available?

L103: Which ERA5 variables have been used?

L110: Instead of “each chemical species” maybe better “several chemical species”.  Which have you used?

L113: How do the monthly factors vary by season for the main sectors contributing to the emissions over Bucharest? Do you know how these factors are determined?

Table 2: I would suggest putting this table in the attachment. How is the NO/NO2 ratio determined for the mapping of CAMS-REG NOx into MOZART-4 NO and NO2 and how is this influencing your results?

L160: “The first data point is recorded at 01:00 LT.” Why is this different from the meteorology comparison?

Equation 1: How were the composition and ratios determined?

L216: Just to clarify, with these errors, you mean the AMF, SCDre, and DSCDs errors?

L226: You are integrating over the troposphere, but SWING+ only measured below 3km, is this correct?

L318-320: Do you mean, it is expected to see a morning, a late afternoon and an evening peak? Because you see two peaks (morning and evening), which I thought are both related to the rush hour. But if I understand your discussion correctly, you expect an late afternoon rush hour peak? What is causing the evening peak?

L331: This might be because the second day is always the flight day, so usually a clear-sky day.

L373-376: Move “It also provides statistics per season and for the entire dataset. For two dates, reported in the table, we truncate data associated with the beginning of the flight for reasons explained in Sect. 3.2.2.” to L373 after “for each separate flight.”.

Which is the other truncated day? Selecting 13:24 LT instead of what time?

L385: Be careful in the discussion, and remind the reader that some seasonal biases might compensate each other.

Table 7: Do you have an idea why the agreement is so different between the days?

Line 471: Do you have any ideas why this day you have excluded is not working well?

Line 481: …span from 2019 to 2025, cover several TROPOMI product versions and focus…

L487: and the upper bound from 10^16 to 1.5x10^16.

L558: See general comments. Please comment if seasonal scaling would be better and how region-dependent your factor might be.

L577/578 Connect this to what you have found. How much seasonality was studied in these evaluations of the TROPOMI product? Might this be something that should be investigated further?

Technical corrections:

L6/7: Split the sentence into two: …are underestimated. Satisfactory agreement with observations is achieved …

L29: like the Global Ozone …

L120: A preliminary evaluation …

Fig.3: stationary instead of staionnary for sector C

L142: …for the first 15 of the 17 SWING+ dates…

L143/144: Move the sentence “When available, …” to the end of this subsection.

Table 3: I would replace the 0 with “-“, since the meaning is not that there are 0 overpasses but just no O3 measurements at these stations

L156: including NO and NO2, and for some of them also O3.

L201: Typical and typically so close to each other doesn’t read very well.

L204: remove with

L237: DOAS was already introduced.

L245: and the offline product instead of and offline products

L252: To avoid confusions: Does “for its” refer to SWING+ or WRF-Chem?

L326: Add …modeled outputs, separated by seasons, defined as summer (June, July, August), …

L367: Thereafter, the modeled columns correlate…

L433: precision of the regression method

L476&478: For better readability you can try to add the seasons directly after each bias: -6+/-25% (summer), …

L504/505: Hard to read, check rewriting.

L537: We assess the WRF-Chem performance…

This review is for egusphere-2025-3533, titled, Validation of TROPOMI and WRF-Chem NO2 across seasons using SWING+ and surface observations over Bucharest.   The authors conduct a modeling study using WRF-Chem during specific days over Bucharest corresponding with research flights and validate them using airborne (SWING+) and ground-based data. WRF-Chem is then used as an intercomparison platform between SWING+ and TROPOMI to validate the satellite retrieval. Overall, the authors have done a thorough analysis, but this manuscript would only be suitable for publication after some more minor revisions addressing the comments and questions below but may range to a major revision depending on findings related to clouds, the SWING+ retrieval reference uncertainty, and doing a direct comparison between the SWING+ measurements and TROPOMI.

General comments:

Please clarify in the manuscript that this is analysis of tropospheric columns rather than total columns.  This distinction may be needed in the validation section at the end of the paper as well.

More detail is required on the SWING+ retrieval inputs and assumptions.

• What are the assumptions in albedo, a priori profiles, clouds, etc.? Is WRF-Chem used as a prior?
• It is also contradicting to state that the reference is a daily average but then to say it’s over a clean area. Both cannot be true.
• How is the reference amount estimated to add to the DSCD? It looks like some days may have reference issues in looking at the results.
• How are these airborne datasets from SWING+ validated?
• How do you come to the uncertainty estimates around line 215?
• Are these data cloud filtered? There do not appear to be gaps due to clouds in the maps. Validation over cloudy scenes would not be accurate.  

The paper uses WRF-Chem as a platform for intercomparison hitting upon understanding the challenge of vertical sensitivity and applying the averaging kernel to WRF-Chem and that the averaging kernel is different for aircraft and satellite. This is okay. However, this paper would greatly benefit with a direct comparison between SWING+ and TROPOMI rather than using the bias-corrected WRF-Chem columns as WRF-Chem columns will have a lot of uncertainty related to spatial gradients even with bias correction.  This is the biggest source of uncertainty in this analysis even with the linear bias corrections.  In addition to the analysis done already, this work would improve by adding the SWING+/TROPOMI comparison followed by discussion of the benefit of both techniques. The benefit of the direct comparison would be that the spatial information and magnitude of the column should originate from SWING+, not have an introduction of any spatial biases from WRF-Chem.

• There needs to be discussion on how the datasets are temporally matched because the linear corrections are also spanning time. This will be even more important with the direct comparisons of SWING+/TROPOMI.
• If the direct comparison is not done, the authors need to make a clearer case as to why they chose to validate TROPOMI with this technique. The model does have the benefit of filling the temporal gaps as noted in the introduction, but this benefit is not used in this analysis since its only during flight days.  

The title implies that TROPOMI is validated with SWING+ and surface observations, but this is not the case as TROPOMI is validated with bias corrected WRF-Chem.

Specific comments:

Line 40-41: The bias values noted do not appear in the report referenced.  The latest says 13% and -40% for the bias values.   Please check the references for these values.

Line 95: Please clarify, is the model span up to 20km? or above 20km? 

Line 96: It would be helpful to have the flight date table introduced here.

Line 98: The justification for the 3-hour spin-up time because Bucharest is UTC+3 is not a scientifically sound reason. Also, a 3-hour spin-up time seems incredibly short.  Is there literature to support this?

Line 105: Define WPS

Line 124-125: The sentence starting with ‘Its justification…’ needs to be moved up to the second sentence of the paragraph.

Figure 3 and Table 2 do not add much helpful information to this analysis.  Consider removing for a supplement or making clearer why it is needed within the analysis.

Line 163: Are these ground-based chemiluminescent measurements molybdenum converters or a different type? The correction factor in the literature is specific for molybdenum.

Line 203: ‘hovered’ implies flying something like a helicopter or drone.  Consider rewording to ‘flew over’ or ‘operated’.

Line 260: Need more detail on ‘insights’. 

Line 275-276: define MB and RMSE in text. 

Section 3.1.2: The NO2 in this analysis is NO2* or NO2? They appear to be used interchangeably but should be consistent throughout the text.  It was reviewed assuming NO2* throughout, so could the low bias in NO2* be due to other assumptions in the model for the NOz species which may be not represented well in the model?

Section 3.2.1: It appears that the background NO2 for SWING+ is around zero rather than a realistic background value.  Is this offset because SCDref is not added to the slant column or why is it so low? It may compensate for the offset in the peaks. It looks like the next flight has a more reasonable background value.

Line 363-366:  How are all flight objectively screened for these thermal instabilities? What metrics are used to screen the data outside it not agreeing with the model?

Line 386:

1. Justify why a factor of 1.5 was chosen objectively using the noted statistics.
2. How are the statistics different if the factor of 1.5 is not applied? The writing implies this was also done but it is not shown.

Line 399-400: Could the seasonal difference in bias be due individual flight days and issues with either the model or retrieval? Maybe boundary conditions in the model? Or perhaps the SCDref amount in the SWING+ retrieval? It appears the bias would be from 23/12/2021 in winter at least. Maybe 5/11/2021 for fall? This seems more realistic than a vertical mixing issue.

Line 424-425: While the true atmospheric vertical profile is not known, you can use the same assumption in both SWING+ and TROPOMI using WRF-Chem which would allow the two datasets to be intercompared. The text does not even share which a priori profile is used for SWING+.

Line 435: ‘Because most the uncertainty is due to the TROPOMI columns…’ this statement does not seem valid as so many factors go into the bias corrected WRF-Chem columns and the WRF-Chem columns themselves likely have a random uncertainty exceeding this value which should not decrease due to a linear bias correction.  I think this number comes from the mean bias but it doesn’t account for the random uncertainty in WRF-Chem when comparing to TROPOMI and SWING+ products.

Line 448: Does the bias correction of the WRF-Chem columns account for the random error or does it just carry it over and can be represented by the random error in LR2? I think the first sentence may need to be reworded.

Line 469: explain why 05/01/2022 is more reliable.