Please see attached pdf.

The manuscript does a good job of communicating a complex topic with clarity and logical progression. WRF-Chem 3.9.1 is a classical atmospheric chemical transport model, and the manuscript combines this model and assimilation method to analyse the air pollutant concentration well. On this basis, the influences of parameter selection about ensemble member size and assimilation windows length on simulation under heavy pollution conditions are explored. The manuscript also summarizes the general experience of parameter selection for PM_2.5 and PM₁₀ simulation in initial condition assimilation. Finally, in order to study conclusions applicable range, the manuscript discusses the case in moderate pollution conditions.

I think it is an important work that advances our understanding about the selection of basic parameters in the 4D-LETKF assimilation system and the performance of ensemble simulation in a particulate matter polluted environment. In my opinion, the manuscript is well written, the investigation is an interesting topic that fits well within the scope of journal, so it is suitable to be published in GMD. Before that, there are some points that should be better considered for discussion. I list the general and additional minor comments. I hope this revision will be of help to improve the manuscript.

General comments:

1 As shown in the line 182, this study generates different emission samples by adding perturbations to the emission source inventory. This perturbation is spatially and temporally related, but this perturbation seems to be relatively strict. What is the reason for using this disturbance? Are there other disturbances that can be added to the emission inventory?

2 Picture 2 is the Flow chart of the WRF-Chem/4D-LETKF assimilation system for particulate matter. In the picture, the emission adds a disturbance to produce first-guess, and then combined with the observation and 4D-LETKF to produce the analysis result. The analysis results are used as input for the next assimilation cycle. No perturbations are added to emissions in the second cycle. Why manuscript only adds one times of perturbations into emissions at the first cycle of assimilation? Continuously adding perturbations to emissions during each assimilation cycle will increase the standard deviation in the first-guess field to produce more accurate simulation results?

3 In Table 3, ΔRMSE is basically negative, ΔCORR is positive, and the larger AE indicates the higher assimilation efficiency. It shows that the neighboring provinces outside the Beijing-Tianjin-Hebei region also gained greater assimilation gains. The main source of these gains is local initial field assimilation? or they come from the assimilation of the Beijing-Tianjin-Hebei region?

4 In Figure8, the sensitivity experiments are conducted by the selection of ensemble member size and the length of assimilation window. What are the main reasons for the selection of 20, 40, 60 for ensemble member size and 24, 48 72 hours for the length of assimilation window?

5 PM_2.5 and PM₁₀ have similar aerosol components, such as sea salt, and dust in GOCART scheme. And the state variables include these matters. However, As shown in Figure4 and Figure5, the RMSE of PM_2.5 in Severe-40m-48h is 31.19 and 76.83 µg m^-3 for PM₁₀. Why the performance of PM₁₀ prediction in Severe-40m-48h shows more uncertainty than those of PM_2.5?

A handful of additional minor comments:

1 Write the formulation about correlation coefficient and root mean square error.

2 Replace the “peak” in line 308 by the “peak value of”.

3 Add the units in Figure 5.

4 Replace “air quality index” in line 307, 319, 611 by “AQI”.

5 Add the units in Figure6.

6 Add the units in table 3.

7 Replace “larger” in line 314 by “all larger”, replace “below” in line 315 by “all below”.

8 Replace “adjustment” in line 324 and 600 by “initial condition adjustment”.

9 Replace “Beijing-Tianjin Hebei” in line 408 by “BTH”.

10 Replace “provide” in line 105 by “providing”.

11 Replace “event” in line 106 by “events”.

12 Replace “dimensions” in line 426 by “dimensions”.

13 Replace “investigated” in line 628 by “the investigated”.