This paper could have more impact if the remarks from the reviewers were more fully taken into consideration. I would like the authors to go back to the following three comments from reviewers and insert in the text several sentences to make the best attempt to account for them:

1- Figures 2 and 6 show a very poor behavior of the model (regardless of explicit convection) when compared to observations. For example, there is a factor 3 to 4 difference in the AOD and an uncorrelated seasonality when comparing the model to the MODIS observations. I have several questions in this respect:
Have the model outputs been spatially and temporally collocated with the MODIS data in order to perform the comparison (i.e. did you select the modeled times and places corresponding to the availability of the MODIS data?) If not, to what extent your comparison can be affected? My experience is that it matters a lot. Would this make sense with your current model set up, i.e. a regional climate run only fed by the analysis data though the boundaries?

2- On the following comment, you should acknowledge that a comparison to local sites and not only for AOD but also for surface concentrations and deposition would help you understand the source of the differences between model and observations:
"Why AERONET stations were not used? There are quite a few stations in the domain for 2011. AERONET is reliable and is the main tool used to evaluate model performance. Without the AERONET evaluation is difficult to judge the performance of this model compared to other models. Nowadays
many regional models represent reasonably well the seasonality of dust in AERONET stations (daily correlations between 0.6 and 0.8 when reinitialized daily and without dust data assimilation). There are also available high resolution PM10 surface observation concentrations for the Sahelian Transect (Marticorena et al 2010) that would really help evaluating the model."

3- It is not straightforward to me why you rule out the role soil humidity has on modulating emissions, try to better address the following comment:
"Concerning the general decrease of dust in the model from May to September (compared to the 30 observations showing a peak around July): Given that the model is not reinitialized every, has the humidity in May 1 been warmed up for at least 1 year? If not, this could be a reason for such behavior (the model could be showing a trend in dust because of a drift in the soil humidity). Has the model been evaluated for the same time period reinitializing the atmosphere and the soil every day from the parent domains?"

- Finally, understanding why peak winds are not as strong in the 4km- model resolution than in the 12-km model resolution should be key for future research.

Sincerely,

Yves Balkanski

I am satisfied of the efforts made by the first author to address the three important points raised by the reviewers. It also reflects well on the paper with additional information that can guide the readers on the choices made in the analysis of the results , and on the model setup.
These improvements lift my initial reserve to see this paper in ACP. It can now be published.

Sincerely,

Yves Balkanski