Review of Li et al.

Li et al. present a comprehensive evaluation of various processes affecting the simulation of diel variation of PM2.5 in the United States in 2016 using the GEOS-Chem in a chemical transport model (CTM), 0.25-degree configuration. The base, unmodified model presented a 105% high bias compared to observations; the authors investigate the effects of temporal resolution, hourly vs. monthly averaged emission inventory temporal resolution, resolution of vertical gradient in the lowest model level, revised dry deposition parameterization, and adjustments to boundary layer height (PBLH) on improving this bias. The work is a useful reference for the effects of these factors and well fit for publication in Atmospheric Chemistry and Physics. I have minor comments prior to recommending the manuscript for publication.

Major comments:

1. The authors use the term "vertical representativeness" in the abstract (L8) and later on in the text to represent the correction of PM2.5 masses from model level center (which is the conventional way we interpret mass in a model vertical level) to the height of surface measurements, correcting for aerodynamic resistance. I understand this is a complex concept to explain but I wished that it could be explained first in the abstract then defined as the term "vertical representativeness". Or maybe clarify this as the impact of modeling the subgrid vertical gradient (this would be more specific and easier to understand). This would make the text easier to understand.

2. The abstract and conclusion say that the PM2.5 diel variation is "driven by ... 1) to 4)". Perhaps I am missing something here, but these conclusions kind of reflect what we already know about pollutants and their interaction with boundary layer dynamics, perhaps a little too close to the textbook. Could the authors, given the specific conclusions about modeling processes affecting diel PM2.5 variation in their work, elaborate and provide more insight on model representation that could be derived from this work, and how we could improve diel PM2.5 variation simulation in general in models or GEOS-Chem in particular?

3. The use of hourly temporal resolution inventory vs. a monthly mean inventory in NEI presents interesting implications. The authors mention that monthly mean inventories are usually all we get, and that's true for most of the world. Does GEOS-Chem / HEMCO not have an anthropogenic diurnal profile for emissions applied over the monthly mean data? It would be strange not to.

If HEMCO applies a diurnal profile to the monthly mean data in NEI by default, then the work here is considering the impact of a more accurate diurnal profile from the actual inventory hourly data versus the default "prescribed" profile that is applied constantly, and it would be quite surprising to find such an improvement in the diel amplitide bias by simply improving the diurnal profile.

If HEMCO does not apply such profile, it seems to me it is an obvious oversight in the model. What if the normalized profile shown in Figure 3 was applied to every day in the simulation? How much improvement would it yield, and how much compared to a "true" hourly emissions input? There are other implications here if simply applying the normalized profile could get us most of the benefits, because reading hourly data is computationally expensive, especially as we move to higher resolution.

Specific (minor) comments:

1. L31: WRF-Chem is not a CTM (which usually implies offline meteorology), it is driven by online meteorology from WRF. Also it would be useful to state the mechanism used in WRF-Chem here as it has a wide range of configurations and it helps to be specific.

2. L36: Another minor comment, but I suggest "lowest model level" instead of "first model level". First level can be ambiguous; some models (such as CESM or GEOS-5) have first level as top of atmosphere.

3. L56: Was GEOS-FP or MERRA2 used in this work? This has some implications as the PBLH used could be different. The authors don't include the PBLH in the final result and only point to its potential importance, but I think that it is fine. Fixing "PBLH" is only a band-aid because PBLH is a diagnostic from the GEOS output; the PBL mixing scheme in GEOS-Chem takes that PBLH diagnostic value to do the mixing, but it would be inconsistent with GEOS dynamics. But improvements to the PBLH value itself passed to GEOS-Chem can provide improvements to model PM2.5 simulation (e.g., as demonstrated by improved PBLH driving improved PM2.5 in the WRF-GC model, which uses the GEOS-Chem PBL mixing routines as well) and this work further confirms this conclusion.

4. L75 / Table 1 & L150 / Intro of Section 5: I suggest adjusting the table columns to use similar terminology and the same order as they're mentioned in the text.

5. L134: To confirm, the observations in one GEOS-Chem 0.25x0.3125 grid box are averaged for the purpose of comparing to the model, or the closest site to grid box center are used?

6. Figure 4: Please use consistent unit labeling (cm s-1 in (a) and cm/s in (b)). "constrains" -> "constraints" in the figure legend.

7. L252-L254: It's good to know that diel PM2.5 variation is shown to be insensitive to updates in dry deposition parameterization. Out of curiosity, have the dry deposition updates affected other aspects of the GEOS-Chem simulation or certain aerosol species in particular?

8. L423: The authors briefly mention the effect of horizontal resolution in improved representation of PM2.5 diel amplitude and timings of the min/maxima - this is consistent with recent model sensitivity experiments (e.g., in MUSICA/CAM-chem) where improved horizontal resolution improves model representation despite the same underlying physics/dynamics/chemistry. Considering the authors' development of improved representation of vertical gradient in the lowest model layer, do the authors also think that improved vertical resolution in the lowest model levels would help as well, instead of applying corrections?

Please see the attached review.