"It seems that the MOSAIC scheme with the EMR dry deposition produces overall the best set of results, albeit with a computational cost 3x the CAM run. Some improvements to the CAM configuration seem to be possible. A difference in the thermodynamics (no treatment of cations in CAM while they are treated in MOSAIC) are important to the nitrate simulation. There does not seem to be sufficient information given to judge the “best” approach. For example, if the goal is 5 day forecast of PM and it takes 1 hour to produce with CAM and 3 hours to produce with MOSAIC that seems insignificant. If it is 1 day versus 3 days then it matters. I can’t judge this given what was in the paper."

We have revised the Introduction and Conclusion sections to clarify the point raised by the reviewer.   The purpose of MOSAIC in GEM-MACH is to improve the scenario simulation capability as opposed to operational forecasts.   There are numerous missing process representations in GEM-MACH such as organic thermodynamics and heterogeneous chemistry.    It is not possible to include these important processes in the operational model due to computational cost and time penalty.  However, GEM-MACH is used for policy scenario simulations in a research mode.   Here the time penalty is not critical, and the computational cost of better process representation is justified. 

"Line 70 - what is the implication here of chemical and aerosol tracers not being transported by convection? This seems like it would be important to determining the vertical profile. Can you add anything about why this is not a first order issue to be addressed?"

This is indeed an important limitation of GEM-MACH but it is mitigated to a great degree by resolved transport by synoptic and mesoscale systems (e.g. Polvani and Esler, 2007; Lyons et al., 1995).  However, ventilation of the atmospheric boundary layer by shallow convection is not properly captured and is a model bias (Polavarapu et al., 2016).  We have added more discussion of this aspect in the text and included pertinent references.

"Line 213 - I am confused about what is being discussed here, noting particularly the claim to be using GEOS-5 driving data. Later on line 264 you say you are getting dynamical information from GDPS. Is something in error here?"

The boundary conditions were produced by a MOZART-4 simulation which is a CTM distinct from GEM-MACH.    MOZART-4 is driven by GEOS-5 data, GEM-MACH is driven by GDPS and other ECCC produced inputs based on the operational forecast model GEM.   We have revised the text to make this distinction clearer.

"Captions to figures 2b and 2c need to be corrected to refer to figure 2a (not 3a)"

This has been corrected.

"The numerical values in the labeling on Figure 5 are impossible to read even blown up on my screen. I suggest that you tabulate these either in the main text of the supplement. This would also give a more clear presentation of “best” values. Same comment applies to figures 6-9."

We have increased the font size in the figures.   We have also added section S3 in the supplement which tabulates the similarity metrics in Figure 5.    

"Line 625 - I find the appeal to non-sphericity to be unconvincing (although it could even be right). Why not consider that you just don’t have a great simulation of coarse aerosol amounts? You don’t seem to be evaluating it particularly here and so maybe things like dust and sea salt that gets into your domain from remote sources isn't done well (or local dust size distribution)."

We have changed the text to de-emphasize this explanation and point out the underestimation aspect.

"Line 711 - type - “presented in Sections 4.1 and 4.2”

This has been revised.

"Figure 14 - State explicitly here that the difference is EMR-REF"

This correction has been added.

As part of the reply to RC2 we discovered an error in the processing of AERONET VSD.   We had Level 1.5 data plotted in Figure 10 instead of Level 2.0.   The figure has been revised and the coarse mode peak is now at about 6 um instead of 8 um.    The discussion of the disagreement between the model and the observationally inferred VSD has been changed accordingly.  

"The paper is very detailed and provides a lot of useful information. However, it lacks a clear message and a prioritisation of the importance of the findings for the reader, who is not directly working with the particular version of the GEM-MACH model.  Overall, the paper is too long, and I suggest omitting certain evaluation aspects for the sake of clarity."

We have attempted to address the clarity aspect raised by the reviewer.   The additional content required to answer the reviewer's comments as discussed below has been put in the Supplement to reduce the size of the paper.   However, we believe that pruning the existing text and analysis for brevity undermines our article.   In particular, the comparison between the REF and EMR runs establishes the need for a new dry deposition scheme for aerosols.   

"On the other hand, the  discussion of the reasons for the differences in the results and their attribution to the differences in the modelling approach can be strengthened.  I strongly recommend to add a more concise overview (e.g. table) of the key differences and communalities in the aerosol modelling (order of the scheme, hydration, thermodynamic system solution regimes, role of cations etc )"

We Table 1 in Section 3.1 which describes the differences between CAM and MOSAIC.

"The comparison of the two dry deposition configurations (i.e. four model runs have be compared rather than two) does not add a lot of useful information because the the science of the dry deposition scheme is not sufficiently discussed.  I suggest to present only the runs with the EMR update of the Zhang scheme because it seems to lead to better results."

We have added additional discussion and references of the improvements of the Emerson et al. (2020) scheme over the Zhang et al. (2001) scheme used in the reference GEM-MACH.   The Zhang scheme has a high bias in the scavenging of the accumulation and Aitken modes.  In fact, it is an outlier in having the scavenging minimum occurring above 1 µm.  A detailed discussion of the science of dry deposition is beyond the scope of our paper.   The Emerson paper makes sufficient justifications, based on extensive new observational data, for the updated parameters it provides.  In this it is supported by Pleim et al. (2022) (doi:10.1029/2022MS003050).   Our results also support the Emerson et al. (2020) scheme since the Zhang et al. (2001) scheme is negatively impacting the more accurate MOSAIC size distribution while working much better with the pathological CAM size distribution in GEM-MACH.  This undermines the performance of MOSAIC compared to CAM.   We consider an evaluation of the impact of the Zhang scheme to be relevant information that serves as justification for using the Emerson scheme.   The inclusion of the comparison with both REF and EMR runs serves to justify the use of a new dry deposition scheme and we consider this to be a significant result of our study.

"The comparison of the seasonal maps of surface nitrate, ammonium and sulphate with a model-observation composite product (Donkellaar et al., 2019) should not be called “observations” because it is based on GEOS-Chem, surface observations and uncertain satellite-based PM fields.  The usefulness of that composite product for the evaluation should be properly discussed before it used. The observation included in the product are more or less the same observations used by the authors themselves in the next section, which is to some extent a duplication.  However, it is good to compare the simulated  fields as map to show the differences also in areas without in-situ observations (e.g. over oceans)"

We have replaced the term “observation” with “observation product” to refer to the Donkelaar et al. synthesis product in Section 4.1 of the text.   But the current text already makes it clear that this is not from direct observations.   We believe that justifying the use of this product is unnecessary content.  There are no direct observational maps of surface aerosol constituents and objective analysis products have value and are used routinely.   The station data is point-wise analysis which we do not consider a duplication.  

"The comparison with the size resolved AERONET VSD product offers interesting insight but also the reliability of that data set needs to be further addressed. I also suggest including in the paper a more basic evaluation with AERONET AOD or AE observations. The same holds for a verification with standard PM2.5 observations, which should be reported on."

We have included an evaluation for measurement uncertainty in new Section S3 of the Supplement for surface network stations where possible and refer to these values in the text.  We have chosen not to revise Figures 5-9 to include these measurement uncertainties.  Networks do not supply measurement error data (precision and bias) in the files they provide, or it is missing.  AERONET has uncertainty analysis data which can be downloaded on a station-by-station basis which we have done for the 52 stations used in our paper.  For some reason this uncertainty data is not included in the regular aggregated station data files.   New Section S4 in the Supplement includes the VSD size distribution uncertainty and provides seasonal average plots with the corresponding error bars in new Figure S15.

 We have added content to compare AERONET AOD, single scattering albedo and AE fields with GEM-MACH output in the article (new Section 4.4) and new Section S5 in the Supplement.  This analysis is limited by our offline aerosol optical properties code which is based on Bohren and Huffman (1983) and does not consider aerosol mixing state and chemical composition.   However, it serves to underscore the difference in the size distribution between CAM and MOSAIC which leads to better agreement overall with observations for MOSAIC.  As with the VSD data, this is a limited sampling that is more qualitative in nature.   

In the course of carrying out this additional analysis we found a mistake in our original VSD processing where we used Level 1.5 instead of 2.0 products.   The error data is only available for Level 2.0 and we have updated Figure 10 accordingly.  The coarse mode peak is now at about 6 µm instead of 8 µm which agrees better with the model.

"It would be very welcome if the authors could identify more clearly in the conclusions, which aspects of the MOSAIC module seems to be most important for achieving the reported improvements in model performance."

We have added a short paragraph at the beginning of the Conclusions section to summarize the discussion currently present in the rest of the section.