Re: General Comment 1. We thank the reviewer for the helpful comments which contributed to improving our current manuscript by identifying unclear parts, and provided very useful suggestions for future research.

Re: Line 125. Dust particles are largely insoluble at the time of emission, but during atmospheric transport their solubility increases through heterogeneous uptake of trace gases, such as SO₂ and HNO₃, and the formation of soluble sulfate or nitrate coatings on the particle surface (e.g. Usher et al., 2003). For simplicity and computational efficiency, dust solubility is not explicitly calculated in our current study. Previously, Bauer and Koch (2005) found that a constant solubility of 0.5 yielded a similar global dust load compared to a model version where the formation of sulfate coatings on the dust particles was explicitly calculated. Relying on these results, we also assume a globally uniform solubility of 0.5 for all minerals in our study. We defer an explicit calculation of heterogeneous uptake, particle solubility, and its dependence upon aerosol mineral composition to a future study.

Additional References:

Bauer, S. E. and Koch, D. (2005). Impact of heterogeneous sulfate formation at mineral dust surfaces on aerosol loads and radiative forcing in the Goddard Institute for Space Studies general circulation model. J. Geophys. Res., 110, D17202.

Usher, C. R., Michel, A. E. and Grassian, V. H. (2003). Reactions on mineral dust. Chem. Rev. 2003, 103, 4883 − 4940.

Re: Line 127. Our discussion of calibration should have been more specific. There are two parts to our calibration: the global mass of emitted dust and its distribution across particle diameters. The distribution of emitted sizes is an unsettled topic of research. We specify the normalized size distribution using a combination of a theoretical relation by Kok (2011) and measurements of dust concentration during a dust outbreak by Kandler et al. (2009), assuming that the distribution is independent of wind speed, as described in Perlwitz et al. (2015a).

The coefficient linking the emitted dust mass to wind speed at a specific location can be derived empirically by measuring wind and concentration. However, the wind provided by an Earth System Model is an average over a grid box, whose dimensions are in this case on the order of 200 km. Thus, the wind at the specific location needed to calculate dust emission is not provided by the model, and another strategy needs to be used to relate the mass of dust emitted from a single grid box to the grid-box average wind. One approach is to adjust or ‘calibrate’ the relation between grid-box wind and emission using a single global factor so that the resulting global DOD lies within the observed range. In this work, we use a single global calibration factor that results in a global-mean DOD equal to roughly 0.02 (averaged over the UV-VIS band). This is at the lower end of the range 0.02 – 0.04 estimated by Ridley et al. (2016), based upon observations supplemented by model output. In general, the calibration factor is expected to depend on model resolution and the PBL parameterization, along with other physical processes represented by the model. Moreover, there is really no such thing as an ‘un-calibrated’ model. Every model must specify a factor coupling wind speed to emitted mass where both are averaged over a model grid box. For these reasons, we do not provide the calibration factor because we feel that its precise value is not meaningful to other models (including other versions of the GISS model). In this study, an important aspect is that we use the same calibration factor in all experiments, so that differences in the global mass load among the three experiments are attributable only to a different radiative feedback on emission. In the revised version of the manuscript, we will expand our explanation of calibration.

Re: Line 195-205. We appreciate the recommendation of this paper (Liu and Daum, 2008) as it contains theoretical information for calculating the effective refractive index of a composite material that is consistent with the Lorentz-Lorenz equation. The derivation of an empirical relation between the refractive index and the mass density of a material is also interesting, as it allows circumventing the lack of measurements of effective molecular weight and polarizability for atmospheric aerosols. Assuming, as a first approximation, that atmospheric aerosols can be considered as Lorentz-Lorenz mixtures (which is one of the uncertainties discussed in Section 4 of Liu and Daum, 2008), and assuming that the Lorentz-Lorenz relation holds for the complex refractive index (in that study, the authors only focus on the real index), we will definitely investigate the potential application of the Lorentz-Lorenz mixing rule for dust minerals in future studies. Regarding the host mixture in this current work, which is the specific case referred to by the reviewer, Liu and Daum (2008) also demonstrate that the Lorentz-Lorenz mixing rule reduces to the linear volume-mean mixing rule for quasi-homogeneous mixtures (Section 3.3). In a first approximation, this condition applies to our host mixture, as host minerals have relatively similar refractive indices. Therefore, even though the Lorentz-Lorenz mixing rule represents a more general method, the volume-mean mixing rule seems to be an acceptable approximation for our host mixture (as highlighted also in reviewer’s comment).

Re: Figure 2. This is an interesting comparison. In Figure 3 of Bain et al. (2019), we see that all the considered salts have an imaginary index close to 10^-8 for wavelengths up to 0.7 µm. These values are very much consistent with those of quartz, which is the least absorbing mineral (based on data collected by Scanza et al., 2015).

Re: Line 330-335. The observationally constrained literature estimates of the global dust mass that we use for model evaluation (e.g. Adebiyi and Kok, 2020) are independent of the measurements used to calibrate emission with one exception: the concentration measurements by Kandler et al. (2009) that we use to specify the normalized size distribution of emitted dust are one of 27 distinct measurements used by Adebiyi and Kok (2020) to estimate the normalized size distribution of global dust mass. We believe that this commonality makes only a small contribution to our model agreement with the global mass of coarse dust (with particle diameters of 5 – 20 µm) estimated by Adebiyi and Kok (2020). Note that we compare the absolute mass of coarse dust, which depends additionally on the global dust load and not just on the emitted fraction of coarse particles that we specify.

As mentioned in a previous comment, our calibration of dust emission results in a global DOD near 0.02, which is at the lower end of the range 0.02 – 0.04 estimated by Ridley et al. (2016). Despite our probable underestimate of DOD, our model produces PM₂₀ that is on the high end of a semi-observational constraint (Kok et al., 2017), which is based on the same DOD (0.03 as the central value) estimated by Ridley et al. (2016). This means that the mass extinction efficiency, linking DOD to mass, is unrealistically low in our model. Given that the dust load in the fine size range (0.1 – 5 µm in diameter) is similar in our model and in Kok et al. (2017), we suggest that this discrepancy is partly due to our representation of dust particles as spheres in the radiative calculations, which typically have a lower extinction efficiency than non-spherical particles (Huang et al., 2023; Kok et al., 2017).

Re: General Comment 1. We sincerely appreciate the valuable comments and suggestions of the reviewer, which allowed us to clarify some aspects of our methods and results and contributed to making our manuscript more robust.

Re: General Comment 2. We originally considered the reviewer’s suggestion of moving the comparison with AERONET observations before the sections presenting the modeling results and sensitivity tests. However, the current structure of the manuscript allows us to first present the main qualitative and quantitative features of the two mineral coupling schemes compared to the control case, which we then use to analyze the comparison with AERONET observations. While we recognize the advantages of the reviewer’s suggestion, we prefer to maintain our current structure of the manuscript.

Re: General Comment 3. This is a good point. Even in clear-sky conditions, the altitude of an absorbing dust layer (with fixed DOD and SSA) affects the dust DRE at the TOA, especially over lowly-reflecting surfaces, mainly because it affects the absorption of solar radiation that is scattered by atmospheric molecules: the higher the dust layer the less negative the DRE at the TOA (Meloni et al., 2005). A varying absorption profile of dust may also impact the DRE at the TOA. For example, Noh et al. (2016) calculated that a layer of scattering aerosols above a layer of absorbing aerosols makes the DRE at TOA less positive than the opposite configuration, even keeping the same column average AOD and SSA. This because incoming SW radiation is back-scattered by the upper layer before being absorbed by the bottom layer (and vice versa). This applies to dust with varying composition, whose absorption properties are expected to vary along the column. Vertical distribution also influences the climate response, even for a fixed DRE (Miller and Tegen, 1999). In general, evaluation of the vertical distribution of dust has received less attention by modelers, in part because of the limited observational constraints. In the future, we hope to evaluate against ATom and CALIOP. We will add this point to our discussion of uncertainty in our comparison with other models and their DRE, as suggested by the reviewer.

Additional References:

Meloni, D., Di Sarra, A., Di Iorio, T. and Fiocco, G. (2005). Influence of the vertical profile of Saharan dust on the visible direct radiative forcing. J. Quant. Spectrosc. Radiat. Transf., 93, 397 – 413.

Miller, R. L. and Tegen, I. (1999). Radiative forcing of a tropical direct circulation by soil dust aerosols. J. Atmos. Sci., 56(14), 2403 – 2433.

Noh, Y. M., Lee, K., Kim, K., Shin, S.-K., Müller, D. and Shin, D. H. (2016). Influence of the vertical absorption profile of mixed Asian dust plumes on aerosol direct radiative forcing over East Asia. Atmos. Environ., 138, 191 – 204.

Re: General Comment 4. This is a good suggestion. We will add a comparison with Li et al. (2021) and Scanza et al. (2015) to the revised manuscript. The information we gathered is briefly described in the following paragraphs.

Li et al. (2021) did not report results of simulations with homogeneous dust, so we will take the baseline experiment with CAM5 using the soil map of Claquin et al. (1999) to compare their dust load, AOD, SSA and DRE with our values from the mineral experiments (EXT and INT). In general, their model calculates a higher mass extinction efficiency of dust than our model, with a global load of 25 Tg and a DOD of 0.031. This is likely due to different size distributions, as their modal approach only includes sizes up to ~10 µm in diameter while our size distribution extends to 32 µm (and the emission of coarse particles is enhanced in our model). Despite the higher DOD, they calculate a global SW DRE at TOA with smaller magnitude (-0.18 Wm^-2 compared to our values of -0.30 and -0.34 Wm^-2 in EXT and INT, respectively). Their low DRE efficiency may be partly due to their lower global SSA (0.89 compared to our 0.94 in both EXT and INT). However, this is hard to assess because their dust SSA is inferred from the diagnostic of SSA for the total aerosol, and may therefore be partly influenced by other absorbing species (e.g. black carbon).

On the other hand, Scanza et al. (2015) do compare experiments with (CAM4-m and CAM5-m) and without (CAM4-t and CAM5-t) considering mineral speciation of dust. CAM4-t is more absorbing than CAM4-m (lower SSA, consistent with our results) but it surprisingly produces more cooling at the TOA (SW DRE is -0.14 and -0.04 Wm^-2 in CAM4-t and CAM4-m, respectively). In contrast, CAM5-t is more scattering than CAM5-m (higher SSA, in contrast with our results) and as expected it produces more cooling at the TOA (SW DRE is -0.33 and -0.08 Wm^-2 in CAM5-t and CAM5-m, respectively). In both model versions, the DOD is the same in experiments with and without mineral speciation. In partial contrast with these results, we find that both the mineral experiments EXT and INT have higher SSA and produce more negative DRE at TOA than the control HOM experiment with homogeneous dust. Despite these studies using the same mineralogy map as in our model (Claquin et al., 1999), differences in the implementation of dust minerals, optical properties and radiative calculations among the models are likely the cause of the contrasting results.

In our manuscript, we already compare our DRE calculations with Di Biagio et al. (2020). Regarding Di Biagio et al. (2019), the authors report measurements and retrievals of composition, size and optical properties of several dust aerosol samples. However, they do not perform calculations of dust DRE. Moreover, their estimates of dust SSA are based on aerosolized soil samples collected from different source locations. Therefore, a possible comparison with our model SSA at the same locations would have limited value, because the model calculates the SSA of dust in the column, which is affected by dust plumes transported at different altitudes and from different locations. The only remaining comparison would be between the mineral composition measured by Di Biagio et al. (2019) in their aerosolized soil samples and the soil mineral fractions prescribed in our model based on the soil map of Claquin et al. (1999). However, such comparison is difficult to interpret due to the confounding effects of biases in Claquin et al. (1999) along with the effect of differing sample locations and resolution. Moreover, this comparison would substantially be an evaluation of the soil map of Claquin et al. (1999), which has been recently performed by Gonçalves et al. (2023) but is out of the scope of our current study. For all these reasons, despite basing our modeling of mineral-radiation interaction on the work of Di Biagio et al. (2019), we have not included a comparison between our results and the results of Di Biagio et al. (2019) in our current work.

Re: Specific Comment 1 (Section 2.2.1). In all experiments, the model separately transports the mass of 15 mineral species (comprised of 7 pure host minerals, 7 accretions of these host minerals with a 5% mass fraction of embedded iron oxides, and finally pure iron oxides) distributed across 5 size bins (clay plus 4 silt bins), by multiplying the dust emitted flux by the mineral fractions at emission, and then transporting and depositing each mineral species during the simulation (Perlwitz et al., 2015a). Our different mixing configurations for minerals only apply to the interaction with radiation, not to the transport. Thus, in the EXT scheme mentioned by the reviewer, when the radiation routines are called at each time and location, we assign optical properties to each of the 15 mineral species according to its ‘radiative’ category: host mixture, static accretion or free iron oxides.

Re: Specific Comment 2 (Line 262). The full set of quality assurance criteria for the Level 2 inversion product in AERONET (Table 3 of Holben et al., 2006) includes the condition AOD₄₄₀ > 0.4. The AERONET Level 1.5 database provides a flag identifying the measurements that meet all the Level 2 quality assurance criteria except for the condition on the AOD₄₄₀. These data can therefore be considered ‘quasi’ Level 2 data, as they would just need to be filtered only for AOD₄₄₀ > 0.4 to become Level 2 data. We use fully Level 2 data for the comparison of SSA, because the condition on AOD₄₄₀ is mostly important for retrieving SSA with a reasonable error. In contrast, we skip this condition for the comparison of dust AOD, as AOD is directly measured and not retrieved, aiming to increase the sample of AERONET data available for comparison to the model.

Re: Specific Comment 3 (Line 292). The condition IRI_rir < 0.0042 comes from Schuster et al. (2016). In that study (see Section 3.2), the authors apply strict filters to AERONET retrievals of the IRI to separate ‘pure’ dust retrievals at West African AERONET sites (FVF below 0.05 and depolarization ratio above 0.2) from carbonaceous retrievals at South American sites during the peak season of biomass burning. They found that 95% of retrieved dust IRI at red-NIR wavelengths are below 0.0042 and 95% of biomass burning imaginary indices are above 0.0042 (see their Figure 5). In other words, this threshold allows an overlapping of only 5% of the two distributions. This result derives from the strong absorption by black carbon at long-VIS and NIR wavelengths. We apply the condition IRI_rir < 0.0042 as a third filter to select dusty scenes in AERONET, after the size filter (FVF < 0.10) and the SSA filter (SSA₆₇₅ > SSA₄₄₀). This third condition aims to reduce the presence of absorbing carbonaceous aerosols in scenes that are not filtered out by the first two conditions, for example scenes dominated by dust in which small fractions of black carbon (either in the remaining fine mode or attached to coarse dust particles) are still able to affect the absorption of the mixture. The sensitivity of our results to IRI_rir < 0.0042, therefore, depends on how many scenes like that are present in our sample. Given that the first two filters are already quite strict, we expect that removing this third condition would not have an extremely significant impact on our comparison with AERONET, also because we only consider monthly means with a minimum of 80 hourly retrievals. Using a higher threshold would admit some more retrievals into our sample, but with the risk of including more scenes contaminated by absorbing carbonaceous aerosols. Conversely, a lower threshold would potentially discard some scenes with uncontaminated dust. We have not tested the sensitivity of our results to this filtering condition. Nonetheless, we will add some sentences in the manuscript to justify more clearly our choice for the threshold for IRI_rir.

Re: Specific Comment 4 (Line 300). This represents an intrinsic limitation of the comparison between monthly model output with AERONET retrievals. We tried different thresholds for the minimum of hourly measurements required to consider a month for a given station. Selecting dust scenes with our strict conditions already filters out a considerable amount of hourly data. Increasing the minimum threshold would further decrease the number of data. Thus, the used threshold of 80 hourly data per month is a compromise between the regional-seasonal representativeness and that of the monthly means. However, the reviewer is correct that using as few as 80 hourly data for calculating a monthly mean represents an uncertainty in our comparison, that we will mention in our discussion.

Re: Specific Comment 5 (Section 2.3). The reviewer is correct that AE has been used widely in studies comparing model results with AERONET to distinguish large and small particles. Values of AE below 1 roughly indicate size distributions dominated by coarse mode aerosols, while AE above 2 are typically associated with fine mode aerosols. However, AE is not a measurement of size: it only describes the spectral dependence of the measured extinction that is approximated as linear in the log scale. The relationship between AE and size distribution is only qualitative and is not unique. Thus, any given AE can correspond to a large range of FVF and fine mode effective radii (e.g. Figure 11 of Schuster et al., 2006). On the other hand, the FVF retrieved in AERONET is superior to AE because it is constrained by the almucantar scan in addition to being constrained by the spectral dependence of extinction. For this reason, we use FVF in our filtering method as an indicator of size distribution that is more robust and precise than AE.

Additional References:

Schuster, G. L., Dubovik, O. and Holben, B. N. (2006). Angstrom exponent and bimodal aerosol size distributions. J. Geophys. Res., 111, D07207.

Re: Specific Comment 6 (Lines 341-342). We considered this question, in part to understand why our global mass is at the high end of an observationally constrained range (Kok et al., 2017), while our total DOD is at the low end (Ridley et al., 2016). The literature has conflicting results about the importance of non-sphericity to optical properties typically used in models (e.g. Mishchenko et al., 1997; Räisänen et al., 2013; Kok et al., 2017; Huang et al., 2023). In a separate project with compositionally homogeneous particles, we find an enhancement of dust extinction by triaxial ellipsoids that is comparable to that found by Kok et al. (2017). We expect that ellipsoidal effects in the present study would significantly reduce the low bias of our DOD while keeping the good agreement of our size-distributed mass load with the estimate by Adebiyi and Kok (2020). However, one complication that prevents us from making any quantitative estimate is that, for linking dust absorption to varying composition, we use an empirical relation between the dust IRI and fractional iron oxide mass that is retrieved from measurements by Di Biagio et al. (2019) assuming spherical particles. Inferring non-spherical radiative impacts has to be done carefully to be consistent with the assumptions underlying the empirical relations, a topic that is beyond the current study.

Additional References:

Mishchenko, M. I., Travis, L. D., Kahn, R. A. and West, R. A. (1997). Modeling phase functions for dustlike tropospheric aerosols using a shape mixture of randomly oriented polydispersed spheroids. J. Geophys. Res., 102(D14), 16831 – 16847.

Räisänen, P., Haapanala, P., Chung, C. E., Kahnert, M., Makkonen, R., Tonttila, J., Nousiainen, T. (2013). Impact of dust particle non-sphericity on climate simulations. Q. J. R. Meteorol. Soc., 139, 2222 – 2232.

Re: Specific Comment 7 (Line 350). We thank the reviewer for this suggestion. In addition to the reported range of DRE values constrained by Kok et al. (2017), we will also report the central value.

Re: Specific Comment 8 (Line 341). We assume this comment refers to line 351, where we compare our DRE with Kok et al. (2017). We find our DRE (-0.25 Wm^-2) to be closer to the less cooling end of the range constrained by Kok et al. (2017) (roughly from -0.80 to -0.15 Wm^-2), even after adjusting our DOD to their higher value of 0.03 (-0.38 Wm^-2). Typically, at solar wavelengths, with a non-zero imaginary index, coarse dust particles absorb more than fine particles (resulting in a lower SSA) because of their larger volume. In contrast, fine particles are more efficient scatterers because scattering efficiency peaks for sizes similar to the wavelength (Hansen and Travis, 1974). Our size distribution extends to 32 µm in diameter (with enhanced emission of coarse particles), while the DRE estimates of Kok et al. (2017) refer to diameters below 20 µm. This means that our calculation of dust DRE accounts for warming by particles with diameters above 20 µm that were omitted by Kok et al. (2017). Thus, we partly attribute to this size discrepancy our less negative adjusted DRE (-0.38 Wm^-2) compared to the central value (~ -0.50 Wm^-2 in the SW) from Kok et al. (2017).

Re: Specific Comment 9 (Line 356). In this case, we compare with Di Biagio et al. (2020) who considered giant particles with diameters up to 150 µm (indicated by aircraft and field measurements), which are not included in our calculation. That may partly explain their less negative DRE efficiency (-9.62 Wm^-2 per unit DOD at 0.550 µm) compared to ours (-12.75 Wm^-2 per unit DOD in the UV-VIS band) for the same reason explained in the previous comment (i.e. large particles warm more). We will better clarify this mechanism in the revised version of the manuscript.

Re: Specific Comment 10 (Line 394). We thank the reviewer for identifying this unclear sentence. The meaning of this specific sentence is that we find a general increase of dust scattering due to varying composition if we take the IRI of Sinyuk et al. (2003) as the reference (HOM experiment). If we had used a lower imaginary index in the HOM experiment, likely we would have obtained scattering increases in some areas and decreases in other areas in the mineral experiments. We agree that the sentence the comment refers to can be rephrased in a clearer way, for example: “It is important to point out that, while the emphasized regional differences in SSA are a direct consequence of the varying mineral composition, the general increase of dust scattering compared to the control run is a result that depends on the IRI that we use for unspeciated dust in our control experiment (HOM). The values retrieved by Sinyuk et al. (2003) are just one of the possible IRI sets for dust available in literature, … “.

Re: Specific Comment 11 (Line 420). This is a useful suggestion that we accept. We will reverse the color scale for SSA variations in the revised version of the manuscript.

Re: Specific Comment 12 (Line 454). We thank the reviewer for this and the following comments about this whole paragraph (lines 451 – 461) that we plan to rewrite in a clearer way. Regarding this first comment, with “very large particles” we do not refer only to diameters between 30 and 32 µm. One of the assumptions in retrieving the size distribution in AERONET is that the volume distribution must go to zero for diameters approaching to 30 µm. In principle, if particles near this size or larger are present in the column (more likely close to sources), and if those particles remain suspended for a time long enough to affect the measurements of radiance in AERONET, the above-mentioned retrieval assumption for the size distribution in AERONET could lead not only to missing diameters larger than 30 µm but also to underrepresenting slightly smaller diameters close to the threshold of 30 µm. This potential limitation of the AERONET size retrieval for detecting larger particles is under discussion by researchers but its practical importance has not been quantified. Thus, we will rephrase the paragraph to better explain this issue as an uncertainty in assessing our dust SSA against AERONET values (see next comment).

Re: Specific Comment 13 (Lines 455-456). This comment follows from the previous one. The potential underrepresentation of coarse particles in AERONET, due to the cutoff at 30 µm in diameter for the particle volume distribution, would result in a larger imaginary index, compensating for the missing absorption by large particles, but would leave unaffected the retrieved SSA. This would happen because the size distribution and imaginary index are co-retrieved in AERONET, under the constraint that the resulting SSA (in conjunction with the measured DOD) must allow the best possible match of the radiances measured during the almucantar scan. As mentioned in the previous comment, this potential issue in the AERONET representation of size distribution has not been quantitatively demonstrated yet. Thus, we cannot assess to what extent our underestimation of AERONET SSA in the HOM case (Figure 12-A) results from our overestimation of the particle volume fractions retrieved by AERONET in the 8 – 32 µm diameter range (Figure S3). However, our fine and coarse dust mass fractions are in line with the semi-observational constraint by Adebiyi and Kok (2020). This evidence suggests that the IRI by Sinyuk et al. (2003) may be excessively absorbing and could contribute, combined with our coarse size distribution, to our underestimation of AERONET SSA in the HOM case (see next comment).

Re: Specific Comment 14 (Line 460). The reviewer rightly expects a stronger match between our model SSA and AERONET data in regions associated with the IRI used in our HOM case. Sinyuk et al. (2003) used observations at 4 locations to obtain the dust IRI, including Cape Verde and Dakar, both affected by Saharan dust transported over the Atlantic Ocean, along with Bidibahn and Bondoukoui in Burkina Faso. According to our designated regions (Table 2, Figure 5), 3 out of these 4 stations fall within the Sahel, marked with down-triangle symbols in our SSA scatter plot (Figure 12). We do point out that some monthly means at Sahel locations in our control HOM case (Figure 12-A) closely align with AERONET values (for SSA ~ 0.92). However, it is also true that the model underestimates AERONET SSA in many other locations within the Sahel.

In general, we may indicate two potential uncertainties associated with the IRI from Sinyuk et al. (2003) that may impact our comparison. First, their IRI data cover only two UV wavelengths, and they inferred values from 0.331 to 0.670 µm by linear interpolation (in the log scale) using AERONET retrievals at 0.440 and 0.670 µm. We further extend these IRI values to encompass the entire VIS band in our model (0.30 – 0.77 µm). This means that our calculation of SSA may be influenced by potential inaccuracies originating from this extrapolation. For instance, linear interpolation may not accurately capture IRI variations across VIS wavelengths, potentially overemphasizing absorption.

The second issue closely relates to the previous comment. Sinyuk et al. (2003) incorporated AERONET size distributions into their inversion procedure to obtain their IRI. As already mentioned, our size distribution represents higher fractions of coarse dust compared to AERONET (Figure S3), in reasonable agreement with Adebiyi and Kok (2020). Thus, our use of the IRI from Sinyuk et al. (2003) combined with our coarse size distribution may lead to underestimating the AERONET SSA in our control HOM case, even in those areas that are used by Sinyuk et al. (2003) to retrieve the IRI.

Re: Specific Comment 15 (Line 486). This is an interesting question, but beyond the scope of this study. The low correlation could be due to a poor representation of the spatially varying source mineralogy, or to errors in transport that determines which source, in a certain moment, supplies the dust arriving at an AERONET station. It is difficult to distinguish the two types of error, because our model winds are calculated and therefore model transport differs from observed wind events. We would be able to attribute model errors (with respect to AERONET) to uncertainties in source composition or transport only in a statistical sense.

Re: Specific Comment 16 (Line 509). We thank the reviewer for highlighting this detail that may be confusing for the reader. In this study, with AERONET DOD, we always refer to the AOD measured at the dusty scenes identified by our multiple filtering conditions on FVF, SSA and IRI (the same applies to AERONET SSA). We will clarify this in our revised version of the manuscript.

Re: Specific Comment 17 (Line 630). As discussed in the previous comment, the observations used by Sinyuk et al. (2003) to obtain the IRI were taken at locations that are affected by Saharan dust but are located within the Sahel region. Our comparison in Figure 12-A shows that some monthly means in the Sahel (as defined in our study) are in agreement with AERONET SSA. Nonetheless, this is not true for several other stations and months within the Sahel. In the revised version of the manuscript, we will add some sentences touching upon the possible uncertainties (discussed in our previous comment) in our use of the IRI from Sinyuk et al. (2003) for unspeciated dust in the HOM case.