General comments:

Overall, this is a well-written paper evaluating an interesting aspect like the effect of the columnar NO₂ correction in the accuracy of the Aerosol Optical Depth (AOD), the Angström Exponent (AE) or the Single Scattering Albedo (SSA) using multiannual records (5 years). This paper addresses some important aspects for the scientific community, such as the investigation of the effects of using different NO₂ data (no correction, direct retrievals or climatological values), which can impact the aerosol products retrieved with different global aerosol networks (NASA-AERONET, GAW-PFR or SKYNET-Prede). Moreover, NO₂ satellite data (TROPOMI) and spectral ground-based data (from Pandonia Global Network, PNG) were used to investigate the possible improvement in aerosol properties retrieved from these three largest ground-based aerosol networks. Trend analysis has been included to understand the impact of the NO₂ correction on the derived aerosol products, although the authors make it clear throughout the article that the number of data in the database is insufficient to carry out this type of study.

I consider that this manuscript fits perfectly into the scope of AMT and that the results presented here are relevant. There are only a few minor remarks.

Minor comments:

Abstract: Is uncertainty estimation a goal of this paper? I consider that this work deals with the impact of the columnar NO₂ effect rather than evaluating/investigating the uncertainty introduced by this term. Please state.

Page 2, lines 34-37: The authors are introducing the direct and indirect effects of aerosols. Don't the authors believe that there are more adequate references to introduce these effects? At least one more recent version of the IPCC exists than the one included in this article.

Page 4, line 101: Is “specifically” right in this sentence?

Page 4, line 112: The authors write “Cimel” in this sentence but “CIMEL” later in the text. Please homogenize.

Page 4, line 118: Please note that Version 3 Level 1.5 includes data with near-real-time automatic cloud screening and automatic instrument anomaly quality controls while Level 2.0 additionally applies pre-field and post-field calibrations. This means that the 1.5 level does not in any way apply the final calibration, so the lack of certainty in the verb “may” does not seem correct.

Page 4, Fig. 1: The authors present here a time evolution of the AOD and AE observations at APL-SAP and CNR-ISAC. I don’t see the point of including such a figure, because these data are not exploited here nor are they mentioned throughout the text.

Page 4, last paragraph: The authors have used level 1.5 SSA information from AERONET. However, as stated by Sinyuk et al. (2020), quality-controlled SSA data (level 2.0) should be retrieved for AOD larger than 0.4 and SZA larger than 50°. How the authors have ensured the quality of the SSA information included in this paper? Why the authors have not included AERONET Level 2.0 data in this study?

Page 6, lines 183-185: The authors introduce here a past comparison between Pandora and Brewer without giving any result of this comparison. This sentence seems dispensable if it does not provide more information about the validity of Pandora NO₂ data.

Page 6, last paragraph: This paragraph explains the NO₂ deviation Pandora versus OMI (AERONET) as is displayed in Fig. 4. It is written that, according to Fig. 4 (lower panel), biases (Pandora-OMI, I guess) of 89% and 87% are found. I’m not able to see these results in the lower panel. Later, the authors give another result: Pandora-OMI average differences of 61.5% at both stations. Could you please explain more in detail these different results and where do they come from?

Page 9, line 281: As mentioned before, the authors acknowledge throughout the text that this database is too short to perform statistically meaningful trend analysis. The question is obvious: why then carry out this analysis?

Page 11, lines 311-314: A reference to previous studies in Rome including some climatological data and aerosol types would be useful in this context.

Page 11, line 332: The values of 1.1% and 1.9% included in this line (as well as in the following lines) don’t correspond to the values in the table. Are the authors reducing the floating points in the text? The use of these values can cause confusion in the reader.

Page 12, lines 350-352: The authors stated that, according to Table 2, SKYNET retrievals are quite similar irrespective of the TROPOMI data used. However, similar results (low difference with the PNG product) were retrieved also in Table 1 for Pandora. Furthermore, mean deviations of AERONET products also displayed very low values…

Page 12, Figs. 8 and 9: Why not merge these two figures into one?

Page 12, line 363: Please define what “modified AOD” is.

Page 12, lines 365-370: I find relevant the lack of information (numbers) to quantify these results.

Page 14, section 3.6: The authors stated in section 2.2.1 that level 1.5 SSA AERONET data were used in this paper. However, in this section, it is not clear to me what SSA product was used. If I understand well, a mimic of the AERONET product retrieved by GRASP was used as a reference, instead of the AERONET SSA standard product. If so:

• Please correct the information provided in section 2.2.1 including a suitable explanation of GRASP algorithm and products used in this paper.
• Why not use the real product instead a “mimic” product? At least these two SSA should be compared…

Could you please clarify it?

Page 15, Fig. 2: There is no information about the lower panel plot. Is the SSA difference?

Page 15, Fig. 2: Y-axis of the upper plot should be SSA and not NO₂.

Page 15, Fig. 2: Information about correlation is written in the text in terms of r-squared while in this figure is expressed as correlation coefficient “R” (in capital letters). Please homogenize.

Page 15, line 454: Again, the numbers provided in the text do not correspond to the ones provided in the plot. It is a matter of rounding correctly to the appropriate significant digit. For example: with RMSE values of 0.035 and 0.031 I don’t consider it appropriate to conclude that RMSE is < 0.035. The same for R squared.

Page 15, line 452: Why the threshold of 0.9 DU?

Page 15, line 453-454: The authors stated that a positive bias of 0.02 was found in conditions of high NO₂ concentrations. Are they talking about SSA or NO₂? From what figure (upper or lower panel) this result comes from? I see in the lower panel an average difference of 0.01 for NO₂ > 0.9 (high NO₂ conditions) but 0.02 for all conditions. From where did the authors find this result? I feel lost with this section.

Page 15, lines 455-458: This sentence seems confusing to the reader. Please rephrase. It has also some typos, like the comma after the word “studies”.

Page 16, line 477: The general result stated here (AOD differences below 0.01 because of this NO₂ correction) seems really relevant. In fact, this is the main result a reader is expecting. However, is this general result written somewhere in the text?

Reviewers' comments:

The authors use extensive measurements to investigate the impact of NO2 concentrations on AOD and AE retrievals. This paper contributes to better understanding that considering NO2, which is highly diurnal-variable, is important to improve aerosol properties in the spectral range where NO2 absorption is strong. Since the manuscript is well-written, I think readers may understand your approach and result well. I believe the paper can be published for AMT after addressing the concerned expressed below.

Minor Issues and specific comments:

P4 L104:

In AERONET site information (https://aeronet.gsfc.nasa.gov/new_web/photo_db_v3/Rome_Tor_Vergata.html),

Rome-Tor Vergata site is located at elevation=130 m but your description is shown as 117 m.

Which one is correct?

P6 L166

Do you use NO2 VCD (vertical column density) or SCD (slant column density) from Pandora product? For clarification, it might be better to mention you use NO2 VCD in Section 2.3.1

P6 L183:

Do Brewer NO2 and Pandora NO2 show good agreement? You need to mention how good quality in your Pandora NO2 measurement since you use Pandora NO2 to correct AERONET and SKYNET operational AOD, AE, and SSA product. More reliable NO2 measurements make your study more meaningful. So, add one or two sentences to show how Pandora NO2 agrees well with NO2 from other instruments.

P6 L191: The Pandora data -> The Pandora NO2 data

P7 L198: You did not show the absolute NO2 difference. However, I think Pandora NO2 is one of the most essential parts in your method. So, it had better to create this plot in the main or the supplement to show how much absolute difference between Pandora NO2 and climatology OMI. If so, readers will understand your approach better.

P7 P225: Are there any specific reasons to exclude the COVID-19 lockdown period?

If so, please mention briefly.

P9 L262: In AERONET (Eck et al., 1999), AE is ->

The AERONET AE product (Eck et al., 1999) is

P9 L279: the impact of AOD and AE modified retrievals -> the impact of modified AOD and AE retrievals

P9 L280: to investigate the possible effect on the AOD and AE trends ->

to investigate the possible effect of NO2 absorption on the AOD and AE trends

P10 L299: to investigate the impact of AOD and AE modified calculations on the derived temporal trends ->

to investigate the impact of modified AOD and AE calculations on the derived temporal trends

P11 L311: Any references? Or is this your finding in this research? Then, plot it to explain or direct the figure you show this. You can show the correlation between NO2 and AOD.

P11 L336: Do you have any reason to use SKYNET AE for 400-1020 nm?

You use AERONET AE for 440-870 nm. Then, is it more consistent to use similar wavelength pair like SKYNET AE for 400-870 nm?

P11 L338: You show how modified AOD and AE by considering Pandora NO2 and then show modified AOD and AE by implementing TROPOMI NO2. Reader can ask how Pandora NO2 and TROPOMI NO2 are consistent. It had better to add one or two sentences to show how both NO2 measurements are in good agreement. You can refer previous studies about this.

P13 L381: The results -> The results in Table 3

P13 L381-388: The description is the analysis in Table 3. Readers may also be curious about the trend itself. AE trends in CNR-ISAC and APL-SAP shows positive and negative, respectively. Do you have any interpretation for this? Is it because inhomogeneous local emission patterns and photochemical destruction you mentioned in P15 L465? Or during your trend analysis period, were there more frequent transports of dust from Africa and caused it negative AE trend in APL-SAP?

P13 L402: Font type looks different.

P14 L432: You used not standard AERONET aerosol retrieval but GRASP algorithm.

If both are the same condition, retrieved SSA from GRASP algorithm is the same with that from standard AERONET retrieval? If not, how much difference of SSA is apparent?

Also, when you use SSA from AERONET, there are quality assurance criteria (Mok et al., 2018). In Figure 12, do you plot SSA when AOD > 0.4? SSA when AOD is small shows large error.

In addition, for SSA calculation, I am wondering you use the consistent surface albedo for SSA retrievals. Incorrect surface albedo makes a systematic bias in SSA retrievals (Mok et al., 2018).

Mok, J., Krotkov, N. A., Torres, O., Jethva, H., Li, Z., Kim, J., Koo, J.-H., Go, S., Irie, H., Labow, G., Eck, T. F., Holben, B. N., Herman, J., Loughman, R. P., Spinei, E., Lee, S. S., Khatri, P., and Campanelli, M.: Comparisons of spectral aerosol single scattering albedo in Seoul, South Korea, Atmos. Meas. Tech., 11, 2295–2311, https://doi.org/10.5194/amt-11-2295-2018, 2018.

Lastly, overestimation in AOD lead to the underestimation in SSA. When you compare SSA from GRASP/Standard AERONET with that from GRASP/Pandora NO2, do you use the same AOD?

For this, in figure 12, you should add a plot of difference of SSA as a function of difference of AOD.

P15 L458 or in conclusion:

You may add one or two sentences about the importance of your research to estimate the effect of NO2 on the spectral dependence of SSA (i.e., absorption Ångström exponent (AAE)) as a future study.

P27 L800 (Table 1)

Why NO2 values in Table 1 is different in different wavelengths? Is this because the number of data you used for 380 and 440 is different? Why don’t you use the same number of data at all wavelengths? Since we look at AE which is the relationship of AOD between wavelengths, I think you should match the data for all wavelengths. In case one event has some information at one wavelength is missing, it is caused by some issues like small fraction of cloud is passing etc.

P41 L890 (Figure 12)

In upper left figure, the number of data shown in the figure is not the same with the legend (N=32). Also, there is no explanations for different color (e.g., green and red dots). It is hard to recognize the dots in the plot. Please modify them with increasing size.

General Comments

This study aims to evaluate the impact of the NO₂ correction for the aerosol retrievals based on ground-based instruments (i.e., AERONET and SKYNET). They utilized multi-annual data collected at urban and suburban sites in Rome, Italy. For the NO₂ correction, they used ground-based Pandora instruments as well as the TROPOMI data. The NO₂-corrected aerosol retrievals are compared with the operational methods to assess their effects.

This manuscript analyzed valuable collocated data from the AERONET, Pandora, and SKYNET, and presented various results using the data. However, I do not fully agree with their main conclusions, which insist significance of the NO₂ corrections for the AERONET and SKYNET products. In the major part of the results, the effects of NO₂ correction seem to be negligible to me, which is the reason why the previous algorithms neglected NO₂ effects or utilized climatology. I believe the authors need to demonstrate their conclusions based on the statistical test to assess the impacts of the NO₂ corrections on aerosol retrievals. Therefore, I would recommend considering the publication of this manuscript after clarifying the below comments.

Major comments

Lines 316-317: This overestimation should be quantified by suggesting statistical values in the main script although the values are listed in the tables. The values should be compared with the reported uncertainties of the AERONET (i.e., 0.01 in the visible and NIR and 0.02 in the UV) and SKYNET.

Section 3.4: The authors summarized the trend analysis in the abstract and conclusion sections. However, this section suggests that the impact of NO₂ absorption on the aerosol retrievals is insignificant for their measurements, but suggested “possible importance”. I think this can mislead the readers unless they show other cases showing the significance of the NO₂ absorption on the aerosol trend analysis.

Lines 415-418: As this result is one of the main conclusions, the authors should report the statistical significance of differences between original and modified data.

Section 3.6: I believe this section is one of the most meaningful results to me. If the impact of the NO₂ corrections on the AOD and trend analysis is not statistically significant, I recommend elaborating on this section (e.g., adding more cases or locations, etc.).

Lines 463-464: I don’t agree that the difference (i.e., lower than 0.003 in table 1) is “quite significant errors” as the errors are typically smaller than the reported uncertainties of the AERONET and/or SKYNET.

Lines 477-479: Again, according to table 2, it is lower than 0.0011 for AERONET, and 0.0051 for SKYNET, which is much lower than 0.01. I don’t believe it is significant given that the AEORNET uncertainty is higher than 0.01.

Lines 505-506: I don’t agree that NO₂ absorption is very important for the AE, AOD, and SSA retrievals.

Minor comments

Lines 61-62: Please add references of the SKYNET, GAW-PFR, AERONET regarding the NO₂ corrections for the aerosol retrievals.

Figure 4: I’m not quite sure if the upper panels of figure 4 are meaningful. I would recommend adding temporal plots of the biases (Pandora - OMI) vs. time over whole measurement periods. I believe that chart can show how the simple assumption of the AERONET can affect the temporal analysis of the AOD over a few years.

Lines 194-197: Underestimation of satellite NO₂ retrievals (e.g., OMI, TROPOMI) compared to ground-based retrievals (e.g., MAX-DOAS, Pandora, etc) is quite a well-known phenomenon and it is attributable to the different field of view (FOV). I think it is worth noting that NO₂ correction using the Pandora is more accurate than the satellite retrievals since the FOV of the Pandora is similar to that of the AERONET in the main script.

Lines 305-306: Is there any reason for the opposite definition between 𝛥𝜏 and 𝛥α?

Lines 311-312: This sentence is not clear to me. Typical “pollution events” do not always accompany high loadings of NO₂, which depends on emissions sources and environmental conditions. Also, Figure 4-6 does not directly demonstrate the relationship between the AOD and NO₂. Scatter plots between AOD and NO₂ might be helpful for this statement.

Lines 342-343: As spatiotemporal variabilities of the NO₂ are significantly high, the authors should state the spatial and temporal window of this collocation.

Line 403: font of “Wei et al., 2019” needs to be “times new roman”?

Lines 407-408: Which data were used to calculate the NO₂-modified AERONET? (Pandora or TROPOMI?)