In this manuscript, Shi et al. use the E3SM model nudged to MERRA-2 reanalysis to determine the relative contribution of dust from six different regions to the Arctic dust load, including the local effect. They then investigate the impact of both dust from local Arctic sources (referred to as “high-latitude dust” (HLD) and dust transported from lower latitudes (referred to collectively as “low-latitude dust” (LLD) on Arctic mixed-phase clouds and the Arctic radiative budget at the surface and top of the atmosphere (TOA).  The authors find that HLD, LLD from Asia and LLD from North Africa contribute to 31%, 44% and 24% of the total dust burden in the Arctic, respectively.  The influence of HLD on Arctic mixed-phase clouds was found to be limited to the surface due to frequent stable thermodynamic conditions, while LLD particularly from Asia were found to influence mixed-phase clouds at colder isotherms at higher altitudes.  In terms of the seasonal variations, HLD exhibited more variation and peak concentrations at summer and autumn, whereas seasonal variability was minimal for LLD, although the largest concentrations were found in spring and winter.  Overall, the HLD was found to have a net cooling cloud radiative effect (CRE) at the surface to a decrease in warm liquid clouds near the surface during autumn when sunlight is relatively weak. 

HLD is currently poorly characterized yet of great importance, especially in a warming world where new sources may be emitted and is thus now becoming the focus of an increasing number of studies.  The work of Shi et al. is both interesting and insightful in this regard.  I recommend publication of the manuscript after the authors consider some additional suggestions below.

1. The limited vertical transport of HLD was claimed to be due to the existence of a stably stratified Arctic lower-troposphere. I would suggest to actually quantify this using the lower tropospheric stability (e.g. as the difference in potential temperature between 850 hPa and 2m).  Are there differences over sea-ice and open ocean surfaces when LTS is substantially different?  Does E3SM simulate LTS in reasonable agreement with observations?  Please evaluate.
2. Lines 38-42: The influence of these various cloud microphysical processes may also interact nonlinearly with one another and impact the phase partitioning of mixed-phase clouds as shown by Tan & Storelvmo (2016).
3. Source-tagging on lines 158-161: There is insufficient description of this technique in the manuscript itself. In addition to citing these references, please briefly describe the methodology and implementation of the technique.  It seems that the six different regions are set up such that in addition to tagging them, they can also be separately tuned. 
4. Is aging of aerosols and the addition of coatings of pollutants that may modify the ice-nucleation efficiency of dust INPs represented in E3SM?
5. Given the large discrepancy between model and observations in Alert, the authors should consider utilizing long-term observations of dust available in Alert as described in Sirois and Barrie (1999).
6. Are the observations and simulated AOD, dust concentrations and deposition flux directly comparable? For example, observations of cloud properties cannot be directly compared with remote sensing observations without a simulator to account for differences in the definitions of these quantities.  One would expect the same for aerosol properties as well.
7. The discrepancy (up to almost twice) between this study and previous studies in terms of the contribution of North African dust to the Arctic dust burden is quite large. Potential reasons are listed on lines 308-310: Of these processes, which process dominates?
8. Figures 12 and 13: It would be useful to compare how E3SM simulates Arctic CREs at the surface and TOA (e.g. comparing with the NASA CERES instrument), and how HLD vs LLD contributes to biases in the Arctic CREs in an additional column. Similarly for the LWP.  The MODIS simulator can be used for the sunlit months. 
9. Comparison with CALIOP: Why use observations from 2007-2009? The record extends well beyond that and the 2007 observations are partially impacted by the change in the tilt of the nadir-viewing angle.  Also, what CALIOP product was used and what was the version of the product?  Arctic aerosol layers are frequently too tenuous to be detected by CALIOP and are also furthermore impacted by the presence of clouds that can interfere with the cloud-aerosol discrimination algorithm. 

Typographical error:

• Line 137: “hour” should be “hours”

References:

Tan, Ivy, and Trude Storelvmo. "Sensitivity study on the influence of cloud microphysical parameters on mixed-phase cloud thermodynamic phase partitioning in CAM5." Journal of the Atmospheric Sciences 73.2 (2016): 709-728.

Sirois, Alain, and Leonard A. Barrie. "Arctic lower tropospheric aerosol trends and composition at Alert, Canada: 1980–1995." Journal of Geophysical Research: Atmospheres 104.D9 (1999): 11599-11618.

Shi et al. present a global modeling study to estimate the contribution of dust emitted at high latitudes as source of ice nucleating particles in the Arctic. They added a source tagging technique for dust from different regions to accomplish this. The role of dust in the climate system is an important topic, especially in high latitudes, which are experiencing a rapid change in climate. The paper makes an important contribution to this topic, since high-latitude dust contributions are largely unknown. At the same time the paper highlights several challenges -- simulating the global distribution of dust itself, but also estimating the concentration of ice nucleating particles based on the simulated dust distribution. The paper fits well within the scope of ACP, and I recommend the paper to be published after the following comments are addressed:

1. Line 124: More detail is needed to describe the ice-nucleation parameterization that is used in the simulations (I believe this is the soccer-ball model?). How is dust represented using this parameterization? Are the same model parameters applied not matter from which region the dust comes from (i.e. different mineralogical composition is ignored)?
2. Related to this, in section 3.3, different ice-nucleation parameterizations are used for the comparison with measurements. It seems that the default parameterization should be part of this comparison. I suggest adding this to this section.
3. Line 134: What was the rational for choosing the time period 2006 to 2011? (and not for example a more recent time period for example)
4. Line 150: Similar to comment 1, I recommend explaining more detail (i.e. equations) about the dust emission parameterization as well as the source tagging procedure. INP parameterization, dust emission parameterization and source tagging are central to this paper, so even though they are described in other papers, it will be helpful for the reader to have the information easily available. This could go in an Appendix or even the SI.
5. Line 155-157: Are the differences between the mass assignments to modes in Z03 and Kok (2011) significant?
6. Explanation of quantities shown in the figures: Please add information in the caption over what time period the model results were averaged. For example, in Fig 3, is this the average over the entire simulated period (2006-2011)? If so, what is the year-to-year variability? And is it the same time period for the observations? What does the grey band represent?
7. Fig 4: The caption mentions that the model results were averaged w.r.t time (2007-2011) and space. I suggest showing some measure of variability, with respect to time and space, to be better able to judge the agreement with the observations.
8. Figs 5, 6, 7, 9, 10, 11, 12, 13: Include the time averaging interval in the captions.
9. Line 257: I would argue that also the MAM case is consistently underestimated by the model and that DJF is underestimated near the surface. However, this is difficult to judge because neither observations nor model results contain any measure of uncertainty. It may well be that the two actually agree within the level of uncertainty. Please include some discussion about this.
10. Figure 6: It would be instructive to additionally represent this figure with percentage contributions rather with absolute values for the column burden. This would make it easier to convey the information how much each region contributes to the burden in a given location.
11. Figure 8: Suggest adding quantity and unit (INP conc. / L^-1) as axes labels. How are the temperatures chosen – are they determined by what was used in the observations? How exactly were the time intervals of observations and measurements matched up? Line 346 mentions “monthly averaged aerosol populations” while Table 3 only lists Spring/Summer/Autumn of various years.
12. Section 3.3: there are various INP parameterizations out there for various kinds of dust. While I don’t expect the authors to compare all of them, I suggest discussing the reason for the choices made for this paper since the choice of the parameterization can have a big impact on the result. (However, I do recommend including the default parameterization, see comment 2 above).
13. Code availability: The github link only points to the general E3SM repo. The authors should include the code that was actually used to run the simulations (including the tagging), for example using zenodo archiving: https://guides.github.com/activities/citable-code/

Typographical errors:

Line 126: Explain WBF

Line 133: Should read “are shown”

General comment: Axes labels: I recommend to not use the format 1E-6 etc. to represent numbers, but use 10^-6 etc. instead.