Comments to “Constraint of non-methane volatile organic compound emissions with TROPOMI HCHO observations and its impact on summertime surface ozone simulation over China” by Feng et al.

As the key precursors of Ozone (O₃), Non-methane volatile organic compounds (NMVOC)  have an important influence on the formation of photochemical, secondary organic aerosols and organic acids, harming human health. It is important and challenge to accurate estimate the spatiotemporal distribution of NMVOC emissions. This study presents the NMVOC emissions over China based on EnKF method by assimilating TROPOMI HCHO retrievals. Authors also optimize NOx emissions to reduce the influence of VOC-NOx-O₃ chemical feedback. The results showed that the forecast experiment with posterior NMVOC emissions reduced the uncertainty of HCHO and concentrations simulation. And the impact on surface O₃ simulation with prior and posterior NMVOC emissions was analyzed. The results will help to improve model forecasts of HCHO, NOx, and O₃ concentrations and contribute to design suitable emission reduction policies. 

However, the structure of the article should be revised. Authors conducted four set of DA experiments and five set of forecast experiments. They discuss the influence of background error and observation error on the effect of optimizing HCHO emissions. And They also analyzed the impact on surface O₃ simulation with prior and posterior NMVOC emissions. Thus, there are too many goals in the study, and it is difficult for readers to remember the setting of these nine experiments. I suggested to delete the discussion about the influence of background error (B) and observation error (R) on the effect of optimizing HCHO emissions in the section 4.4. It would be nice to discuss the influence of the B and R when introducing the EnKF method and explain why authors design the B and R to optimize NMVOC emissions in this study.

There are several issues that need to be addressed.  

specific comments：

1. Line 40: It should be “Compared with the forecast experiment with prior emission, the forecast with posterior ...”. The statement should be revised.
2. Line 42: “Moreover”should be deleted. And the statement also should be revised
3. Line 176: What did you consider about the boundary condition of NMVOC and NOx?
4. Line 204~207: Did author consider about the correction of NOx and NMVOCs in the DA system?
5. Line 209~210: As NO₂is a kind of short lifetime gas, the concentration of surface NO₂ measurements not only present NO₂, but also may include NOx. What did you consider about the influence of NO₂ observation uncertain on optimizing NOx emissions?
6. Line 265: It would be better to use mosaic diagram to present the data amount of TROPOMI HCHO.
7. Line 299: Please added the year of the study period.
8. Line 307~314: The background error covariance is implicitly expressed in the EnKF method. How did author implement EMS1 experiment in the DA system? And itwould be better to introduce EMS1-3 experiment follow the EMDA, making the text description consistent with the Table1.
9. Line 324 and 351: “prior and posterior emissions” should be “prior and posterior NMVOC emissions”, and “EMGAN”should be “MEGAN”.
10. Line 440-441, Figure 5: It is difficult for readers to remember the setting of experiments. And I think that “CEP3”should be “CEP1” in the Fig. 5a?
11. Line 515-518: The background errors and observation errors play an important role in the DA system. It would be better to give a detailed explanation of why the difference in two posterior NMVOC emissions was small by using ‘two-step’inversion strategy in the DA system.

Accurate NMVOC emissions are essential for predicting air quality. Currently large uncertainties exist in NMVOC emissions, both in the anthropogenic and biogenic sources, as compared to other pollutants, such as SO2 and PM.  In this study, the authors use the RAPAS assimilation system incorporated with the EnKF assimilation algorithm to optimize NMVOC emissions using TROPOMI HCHO retrievals. They use MEIC 2020 for anthropogenic emissions and MEGANv2.1 for biogenic sources as the priori NMVOC emissions. They find that NMVOC emissions are largely overestimated, especially biogenic NMVOC emissions. They also find O3 predictions would be lowered using the posterior NMVOC emissions.

The study seems interesting, however, I have a few concerns about some of the key results in this stuy.

First, the CMAQ model overpredicts O3 in China largely (over 20 ug/m3) in most sites of China (Figure 4a) with the WRF-MEIC-MEGAN setups. The overpredictions are consistent over the whole month (Figure 5). Such ‘large’ overprediction problem of O3 in CMAQ in China (or any other countries/regions) has not been reported. The overprediction seems very consistent in space and time. The spatial distribution of VOC emissions (in Figure 2a) also looks very uniform. More evaluation and check on the model setups and results should be performed and provided to fully understand this problem. Honestly, attributing such large O3 predictions to VOCs emissions is somewhat dangerous. How are the predictions on CO/ SO2/EC (the species that are less chemically reactive)? What about the meteorology predictions?

Figure 4a and Figure 6a seem not consistent, the difference in south China in Figure 6a looks not as significant as the north. Also the observations look no significant spatial variation  on Figure 6a, and MDA8 O3 in August is most in blue-green color (~110 ug/m3).

What do different symbols/colors in Figure 1 mean?

Why not choose 2020 as the study year if you have 2020 MEIC emissions?

BVOCs is greatly overestimation by MEGAN (over 50%). Have any other studies reported similar findings with MEGAN in any regions? If no, please explain why such problem occurs in China? Why previous modeling studies in China with CMAQ have not encountered such problems (also the O3 overprediction problem)?