• My primary concern is related to WRF performance in reproducing dynamics and transport. As noted by the authors, wind speed and direction are both crucial components in overall model performance due to their role in determining the transport of pollutants and precursors. The failure of the nested WRF model to adequately represent these features is therefore a pretty big issue in my mind: it's hard to tell how much of the subsequent analyses of chemical composition are impacted by problems in dynamics, casting many of the subsequent conclusions in doubt. I understand that this is a very thorny modeling problem, but I would very much like to see the authors attempt to address this problem in some form, at the very least to try and better understand when and where their WRF simulations are failing, and to quantify the sensitivity of their final results to transport issues.
  • On a related note, I was curious about the chosen WRF nesting scheme used here, which employs a 16:4:1 km grid size scheme. My understanding is that WRF best practice is to use an odd ratio less than 7 (typically 3:1 or 5:1) for spatial dimensions of nested grids to minimize interpolation errors. I'd be very curious to see whether a shift to 15:3:1 km grid sizes might help to improve WRF output here.
  • Since the intended focus of this paper is not on WRF model performance itself, I also wonder whether it would be better to use meteorological data assimilation techniques (or even established high-resolution meteorology data products such as HRRR) to further improve modeled wind speed and direction, providing a clearer focus on emissions and chemistry.
  • Regardless of how well these (or other) techniques might improve model performance, their use would also allow for a kind of sensitivity study to assess the impact of transport on modeled chemistry. It would be helpful and informative to assess CMAQ predictions under alternative (hopefully improved!) meteorological methods such as those described above, and to compare that performance against predictions made using original meteorology, thereby estimating sensitivities.
• Why were lightning NOx and windblown dust sources omitted? While they may not dominate in this urban region, they still seem like sources better included than excluded. (If the omission was due to a lack of trust in modeled wind speeds and convection, this seems like all the more reason to improve these elements.)
• This is a more subjective and minor request, but I also would like to suggest that the authors consider an alternative and consistent color scheme for the panel maps of Fig. 6. Rainbow schemes such as this one produce artificial bands across specific color ranges, making it harder to consistently evaluate differences in the maps across concentration thresholds. It also looks like some of the maps use discretized color bins while others are continuous. Unless there is a specific reason for this, please make this a consistent decision one way or the other.

This work describes a model framework for investigating air quality and sources of pollution in the Los Angeles region. As a region with complex emissions, understanding the sources of air pollution is critical to improving air quality in the area. The model framework and validation against observations is exceptionally well-described and the work fits well within the scope of ACP. While I think the article could essentially be published as is, additional discussion on the limitations of this model framework based on the analysis of model error would provide more context on how to interpret future results. Therefore I would recommend publication and would encourage the authors to consider addressing the following things. I look forward to part 2!

1. While the evaluation of the model against available measurements is quite thorough and well-described in the text, there is little discussion of how this evaluation impacts the interpretation of the results as well as the strengths and weaknesses of the model. A comprehensive assessment of how all the biases in specific pollutants could affect results would be largely speculative and thus is unnecessary, but I believe a brief discussion on how the model’s performance against measurements could affect big picture results would be appropriate. Below I provide a few questions that struck me as meaningful to address, however they need not all be.

• Does the good representation of POA, but poor representation of SOA mean that generally, this model will predict OA better near sources and diminish in its effectiveness further away? Could the poor NOx prediction impact the conclusions surrounding the NOx vs VOC-limited ozone regimes? Because of the large biases in certain species does this model’s strength lie in predictions of relative changes in species (as the results shown in this work are) rather than predicting absolute values or are there certain species the author’s feel confident could be predicted? Do the large errors in wind speed and direction indicate a systematic bias of air from different areas into the domain region (e.g. higher sea spray aerosol from the ocean vs more agricultural or road sources from in-land) or is the spread too large?

2. Are wildfire emissions included? If they are included in one of the emission inventories then that should be made clear.

3. Why were lightning NOx and dust sources not included? Are these just not large emission sources in the area?

Technical comments:

1. Caption Fig 2: The caption refers to 2 different resolution scenarios as “d01.” I assume this is a typo, but as I don’t think this notation is used again, it may be unnecessary.