This paper describes a chemistry-transport model that is implemented with a data assimilation compartment 

using the Parallel Data Assimilation Framework. First of all, I would admit that I don't expertise in the

 atmosphere chemistry area, all my comments are from the data assimilation with PDAF. As far as I know,

 the online data assimilation approach with PDAF is first thoroughly described in Nerger et al., (2019) GMD paper,

 where the structure, the algorithm, the implementation are shown based on a climate model AWI-CM. Actually,

  such implementation has been widely used also for other ocean models such as FESOM, MITgcm etc. Another group 

  like C-Coupler (Liu et al., 2020 GMD) provides similar data assimilation functions as well.

 I found this research is comprehensive with both technical and experimental perspectives. The results 

 are clearly present and well organized. I would recommend acceptance after minor revisions.

1. As for the paper structure, I would suggest the authors rephrase 2.3.1 and 2.3.2, i.e., the technical parts.

 Most of these implementation details are already well described in Nerger et al 2019. I didn't see too many 

 differences compared to Nerger's. Currently, it's rather a repetition. Please cite this paper directly and show your

 differences to condense the context.

 For 2.3.2, again, not necessarily a repetition of these algorithm details, which are well-known in amount of studies. 

 The authors should concentrate on things that are distinguishable from others' work. For example, localization radius,

  whether it is sensitive to your configuration (I see the authors cite two other researches, but actaully these are 

  different stories if the configuration changes based on my practices), are you using the same localization radius

   for different observations; forgetting factor, which value is set, why is that; things like that.

2. Please use a larger fontsize for Figures 10 and 11. The caption tells "the increments in the RMSEs of the surface 

PM2.5 forecast (g, h, i) and the increments in the RMSEs of the surface PM2.5 forecast (j, k, i)". Something goes wrong

there?

3. Taking Figures 14 and 15, the authors found that the system seems not well constrained by DA in high level. Could the 

authors add some discussions about the physical reason or other aspects behind this problem?

In general, I think the paper is well written with adequate evidence for their results.

General Comment:

Aerosol vertical structure is important for investigating global climate change, air pollution transport and control. The authors developed an online data assimilation system for vertical observation by coupling NAQPMS with PDAF, which offers spatiotemporally continuous aerosol vertical profiles. The system can make efficient use of parallel computational resources and produce great improvement in the aerosol vertical structure and surface PM_2.5 concentration. Overall, the whole manuscript is within the scope of GMD and well written. I think the research is novelty to impact on other one’s research and could be reconsidered after major revisions.

Major comment:

• In previous assimilation studies of satellite products, 6 hr or 12 hr has been chosen as the assimilation window. The whole manuscript is based the analysis and subsequent 1-hr forecast. So why do authors choose one hour as the DA window？What is the difference between 1 hr and 6hr or 12 hr in terms of assimilation effect?

• The paper chooses ESTKF as the assimilation algorithm. What are the improvements or advantages of this Kalman filter algorithms compared to the previous KF algorithms?

Specific comments:

• L535-536: What does “Although the orbits are slightly covered by the model domain, the only difference between the FR and NP-LIDAR experiment is whether ground-based lidar measurements are assimilated (Fig. 12b)” mean? I do not understand this very well. What is the connection between these two sentences?
• L605: Adding “measured by lidar” after “The aerosol vertical profile” for clarity.
• L675: authors listed several reasons to explain that only assimilating lidar measurements has a weaker performance than only assimilating surface PM₅ measurements. However, these reasons are just a guess without any detailed analysis. So, these reasons should not be listed in conclusion.
• L685: “a systematic data quality control of lidar measurements is urgently needed to solve this problem in future research” should be deleted. The reason is same as the above comment.
• 7: The description of Fig. 7d is missing.
• 12: “(e)” is missing.
• 15: “2021” should be “2019”.