This manuscript presented method comparison: EnKF and 3D-Var, for assimilating surface PM2.5 observations with two settings: IC and ICBC. It is a straightforward paper. One major issue is that the prediction model CMAQ has only 15 layers up to 20km, which is too coarse. How many layers below 1km? Could this coarse vertical resolution cause artificial dilution for near-surface pollutant concentration, and result in the systematic PM2.5 underestimation? Although this manuscript focuses on data assimilation (DA), the corresponding prediction model should be reasonable, too. Otherwise, the DA methods only show their effect on correcting the systematic underestimation.

Here are the detailed comments.

Section 2. PM2.5 is not a single species in CMAQ. How do you map the PM2.5 increment to individual CMAQ aerosol species?

Page 9, line 155. What’s the vertical extent of the 50% perturbation being applied, to all layers? Considering that it is used to the assimilating surface observation, certain justification is needed.

Page 6, line 172. Do you think that the static horizontal width of 100km and vertical width of 2km fit for all scenarios, for day and night? Any discussion about it.

Line 164. Same as above. Is the 30% standard deviation of LBC perturbation applied to the all layers?

Section 2.2. The 3D-Var description in section 2.2 is too short, and needs to include more detail. What are the horizontal/vertical length scales, and model error covariance yielded by the NMC method? Could you show some plots about them?

Figure 4, it is better to include the corresponding 3D-Var increment for comparison.

Section 3.3. Does the evaluation use the same observation data as those used in DA? 

General Comments:

This manuscript describes the initialization of ground-level PM2.5 for a chemical transport model using the ensemble Kalman Filter (EnKF) method. Authors implemented EnKF in the CMAQ model and claim this method improves PM2.5 predictability compared to the 3DVar and No DA. The PM2.5 predictability of South Korea is further improved if the EnKF is applied not only to the nest domain but also mother domain with ground PM2.5 observation.

The manuscript looks like to have a reasonable structure for the paper. It will be available for publication if it is improved with a minor revision.

Specific Comments:

• In the manuscripts, ICs is an abbreviation for both initial conditions and initial concentrations. The same goes for BCs. Authors should only use the abbreviation “C” for either condition or concentration
• What are the background variables that are input to EnKF? I believe the EnKF uses background error covariance between PM2.5 and some meteorological variables. Is it correct that the experiments only update PM2.5, so other meteorological variables are the same as before DA? Adding a list of background and analysis variables is recommended.
• I am wondering about the observation operator for the PM2.5. Is PM2.5 one of the background variables? If not, the authors need to introduce the observation operator for it to calculate observation operator. Some descriptions for the observation operator would be better to be added in the manuscript.
• Line 174: What is the inflation parameter (alpha) for RTPS?
• Since you have described the parameter for the localization, it would be better to also describe the parameter for RTPS.
• Line 175: Why do you apply the RTPS to ensemble before and after DA? The RTPS compares ensemble spread before and after DA because the amount of inflation in RTPS is proportional to the ensemble spread reduced by the DA. So theoretically, RTPS can be applied only once after DA.
• When you describe DA_icbc, can you show the pm2.5 field for domain1 which contains domain 2? For example, horizontal field of PM2.5 as in figure 4, but with domain 1. The distribution of PM2.5 over domain 1 can be clearer evidence showing the effect of boundary conditions.

Technical Comments:

• Line 245: sate --> state