This manuscript presents WRF-Chem model simulations of tropospheric NO2 over Europe. The model is evaluated against network NO2 surface measurements and remote sensing columns from TROPOMI and from MAX-DOAS instruments at 5 stations. The main point of the paper is that the modulation of sector-specific diurnal profiles of NOx emissions brings a significant improvement in the model agreement with NO2 data. The optimization of the diurnal profile is performed by trial-and-error, using the diurnal profiles from a previous modelling study (Kumar et al., 2021) as a priori. The adjustment decreases the nighttime emissions and increases the emissions during the day (see Figure 2), which effectively brings down the mean relative bias between the model and network data (Figure 2(i)). The adjusted emissions are claimed to improve also the model agreement against TROPOMI and MAX-DOAS data, which leads the authors to consider their updated model as significantly more realistic than the original simulation.  

Major comment

The diurnal cycle adjustment does reduce the model bias, but there are several reasons to believe that the model bias is not (only) due to those temporal factors: 

1) Several features of the adjustment seem suspicious, such as the disappearance of the late afternoon traffic peak (the 17-19h emissions are decreased by about a factor of 2!), the disappearance of the early evening "energy for building" peak, the pronounced early afternoon enhancement for the industrial and energy sectors, etc. The authors do not provide any validation (not even a discussion) for these changes. The changes are claimed to be reasonable, but this needs to be demonstrated. 

2) Using another model (MECO(n)), Kumar et al. (2021) succeeded in reproducing quite well the NO2 diurnal cycle from MAX-DOAS stations (also in situ data, to a lesser extent). No adjustment of the emission temporal profile was found necessary to achieve this result.

3) Many WRF-Chem studies have shown that the simulation of vertical mixing (especially during the night) plays an important role in contributing to the model bias (Kuik et al. 2016, 2018; Visser et al. 2019; Du et al., 2020; Poraicu et al. 2023; etc.). 

4) Studies have also shown the importance of the chemical mechanism, e.g. Knote et al. 2015. Furthermore, the VOC emissions in the model are not well validated and could have a large impact on the NOx, due to their effects on OH radical levels and organic (peroxy)nitrate formation. 

5) Most importantly, the NO2 measurements used at 92% of the German sites are made using chemiluminescence instruments, which are characterized by sometimes large positive biases due to interferences of other NOy compounds such as PAN, HNO3 and HONO (Lamsal et al., 2008; Villena et al., 2012). As shown by Villena et al. (2012), the bias can reach up to a factor 4 and correlates with ozone, presumably because ozone correlates with photochemical activity. As discussed above, the emission update proposed by Kuhn et al. consists essentially in a significant increase during daytime, especially between 8 and 17h (Figure 2), precisely when ozone is most abundant (Figure 3). It is very unlikely that this would be a coincidence.

Furthermore, the claimed model improvement against TROPOMI is far from being evident, judging from Figures 5g and 6h (also the slope and RMSE, see Table 4). The NO2 model overestimations over the Rhine/Ruhr and other areas are lower with the original temporal profile from Kumar et al. (2021). Finally, the model with updated profiles overestimates the MAX-DOAS NO2 data at most sites, and by as much as 30, 50 and 79% at Heidelberg, Mainz and Uccle. Evaluation of the diurnal cycle of modelled NO2 against the MAX-DOAS data is missing, as well as the assessment of the effect of the new profiles on the model agreement against MAX-DOAS data.

In conclusion, I cannot recommend publication of the manuscript in its present form. The WRF-Chem comparisons with in situ, MAX-DOAS and TROPOMI data are interesting, but the emission adjustment (which is the main point of the paper) has no added value. I suggest to show results with the original diurnal profiles of Kumar et al., and to explore the various possible causes for the biases, possibly through additional sensitivity calculations.  

Minor comments

l. 8 "crucial aspects of the retrieval etc.": the application of averaging kernels has become quite standard in model evaluations against UV-Vi satellite data. As I understand, you only follow the standard recommendations. Please re-phrase. 

l. 46 Mar et al. (2016): "the choice of mechanism barely impacts the NOx underestimation": this statement is based on a comparison of only 2 mechanisms! This cannot begin to describe the uncertainties related to the chemical mechanism. For example, both mechanisms considered might have similar flaws, such as the absence of HONO chemistry (and heterogeneous production), outdated organic nitrate chemistry, overestimated rate constant for NO2+OH (see e.g. Mollner, 2010), etc. Please re-phrase or drop the sentence. 

l. 61 "a systematic investigation to prove a general bias of the NO2 in-situ measurements in Europe is still missing": however, Villena et al (2012) showed an overestimation of up to a  factor 4 compared to DOAS NO2 measurements. The overestimation peaks in the afternoon when ozone is at its highest levels, which is also when the WRF-Chem model NO2 underestimation is highest compared to the network measurements is highest. 

l. 62 It is not because the instruments "follow strict regulations" that they are free of artefacts. The technique itself leads to the presence of interference. 

The overestimation of 7% in test measurements mentioned in the manuscript does not validate the technique, because the interference is highly variable, as shown by Villena et al. (2012). A "personal communication" is impossible to verify and place in proper context. The variability in the bias 

due to interferences appears related to photochemical activity. 

l. 76-77 The "validation" of the temporal profiles of emissions using TROPOMI and MAX-DOAS is of very limited value since TROPOMI measures only in early afternoon, and only midday MAX-DOAS data are used in this manuscript. 

l. 204 scaling of the O3 BCs : a uniform scaling? Is this based on ozonesonde data or only surface data?  How about the vertical dimension? Why not use CAMS profiles ? Those are likely more realistic than CAM-chem. 

l. 224 Why would the uncertainty on simulated surface concentration be equal to the maximum hour-to-hour variation in concentration? This does not make sense. 

l. 245-246 'the temporal profiles (...) for the traffic sector (Fig 2a) have a shape similar to the NOx time series of the traffic stations (...) This can be seen as further validation of our results." : certainly not, since the NOx concentrations temporal profile reflects not only the emission profile but also

the temporal cycle of boundary layer mixing.

The section 3.1 could be rewritten by making more rapidly your point that the new simulation performs better, since the optimization of profiles was designed to do that, and it is not surprising at all that the results reflect that. On the other hand, you should explain why your new emission profiles can be considered more realistic. 

L. 285 The increase in NO2 above 10 km altitude is not necessarily a result of stratosphere-troposphere exchange. Higher NO2 in the UT (compared to the mid-troposphere) could result from higher NOx lifetime, aircraft emissions or lightning emissions. I recommend to use the best estimate of the tropopause level. What is the impact of removing 3 layers from the calculated tropospheric column? 

l. 300-308 This discussion could be shortened, since you are simply applying the standard recommendation issued for TROPOMI NO2. 

The panels (k)-(o) of Figure 6 would be better placed beneath the panels (f)-(j) of Figure 5. In this way, the reader can better judge the impact of the new temporal profiles on the comparisons with TROPOMI, which is the purpose of this section. The improvement of results due to the application of averaging kernels is a pretty standard result. 

l. 331 Based on the figures 6(m) and (5)h, the simulation using the Kumar profiles performs better. Is the low bias of -15.7% the mean relative bias, or the relative bias of the mean column? The results shown in Table 4 suggest a better slope and RMSE with the old profiles. The slightly larger bias could be due to an emission underestimation.

l. 339 "In comparison to monthly means, the modelled NO2 VCDs are smeared out": is this what you really mean? Please re-phrase.

The paragraph 336-345 brings very little to the discussion. The stronger noise is expected. Consider removing that part (and the figure)

l. 389 "Elevated layers are typically caused by elevated emissions, e.g. from a power plant stack at a few hundred meters height": Is there really a tall power plant stack in the vicinity of every MAX-DOAS station ? 

l. 446 Overestimation of NO2/NO: might reflect the fact that interferences affect only NO2, not NO. 

l. 450 Evaluation of the VOC emissions could be done using TROPOMI HCHO columns.

l. 453-465 This discussion ignores the fact that TROPOMI NO2 data also present biases, as shown in numerous validation studies (e.g. Judd et al., 2020; van Geffen et al., 2022). Unfortunately, validation is still lacking for background conditions, but in any case, we can expect deviations of TROPOMI wrt the truth. 

l. 468-469 "General agreement in the overall shape and magnitude": not really! There is a clear overestimation at most sites, suggesting overestimation of midday emissions. It would be enlightening to check if there is a diurnal variation of the difference between WRF-Chem and MAX-DOAS columns (or near-surface concentrations). 

l. 478-479 "the model accuracy (...) in RCT simulations can be strongly improved by optimizing the (...) emission profiles": sure, but probably not for the good reasons. If biases are due to other causes (measurement biases, mixing issues etc.), the updated emission profiles are worthless. 

References

Du et al. 2020  : https://doi.org/10.5194/acp-20-2839-2020

Judd et al. 2020 : https://doi.org/10.5194/amt-13-6113-2020

Knote et al. 2015  : http://dx.doi.org/10.1016/j.atmosenv.2014.11.066

Kuik et al. 2016  : www.geosci-model-dev.net/9/4339/2016/

Kuik et al. 2018  : https://doi.org/10.5194/acp-18-8203-2018

Kumar et al. 2021 : https://doi.org/10.5194/amt-14-5241-2021

Lamsal et al. 2008 : https://doi.org/10.1029/2007JD009235

Mar et al. 2016 : www.geosci-model-dev.net/9/3699/2016/

Mollner et al. 2010 : https://doi.org/10.1126/science.1193030

Poraicu et al. 2023 : https://doi.org/10.5194/gmd-16-479-2023

van Geffen et al. 2022 : https://doi.org/10.5194/amt-15-2037-2022

Villena et al. 2012 :  www.atmos-meas-tech.net/5/149/2012/

Visser et al. 2019 : https://doi.org/10.5194/acp-2019-295

In the present manuscript, Kuhn et al designed and tested a method to improve the air quality simulations by incorporating region-specific diurnal variability in the NOx emissions. This is done by varying temporal profiles iteratively and minimizing bias against ground-based observations. Detailed evaluations of the best simulation are conducted further against the space (TROPOMI) and ground-based (MAX-DOAS) observations.

Besides tuning the NOx emission profiles, other important refinements: (1) adjusting boundary conditions for O3, and 2) implementing high-resolution WRF-Chem simulation as a priori in the TROPOMI retrievals have also been made for the investigation. Through a comprehensive analysis, it is suggested that optimization of emissions over diurnal scale can significantly improve simulations of NOx and O3 over the central Europe. It is argued that similar approach can be applied to develop region-specific profiles for other chemical tracers and other spatial regions.

Overall manuscript is well-written and the methodology, analysis is detailed, and results are described clearly. Manuscript can be published in Atmospheric Chemistry and Physics, however,  more clarity or further analysis is needed on 1-2 points.

While most of the NOx emission is set to be in form of NO, why the model simulated NO does not show significant changes and the model still misses its peaks systematically? The overestimation of NO2/NO ratio suggests (authors have also pointed this out) that model is having a limitation in capturing the chemistry accurately. This needs investigation why the NO is getting converted to NO2 more rapidly than reflected in the observational data. Sensitivity simulations should be conducted with say +- 10% VOC emissions to gain insight into the likely causes and probably to reach better estimation of the diurnal variability in NOx emissions over this region.