We conduct a series of experiments with GFDL AM4.0 (Zhao et al., 2018a, b) over the period 2001–2019. These experiments use the AMIP protocol, where sea surface temperature (SST) and sea ice are imposed based upon average monthly observations (see Gates, 1992, for details). Observed gridded SST and sea ice concentration boundary conditions are from the reconstructions of Taylor et al. (2000). Historical reconstructions of monthly solar spectral irradiances are from Matthes et al. (2017). For radiation calculations, global monthly mean concentrations of greenhouse gases (GHGs), including nitrous oxide (N2O), and ozone-depleting substances (ODSs, including CFC-11, CFC-12, CFC-113, and HCFC-22) are specified from Meinshausen et al. (2017). The solar irradiances and GHG databases are standard for CMIP6. Longwave (LW) scattering of aerosols is not accounted for in the model.

In AM4.0, dust emission is calculated interactively following the parameterization of Ginoux et al. (2001) with a threshold of wind erosion and global scaling factor of 3.5 ms-1 and 0.2 µgs2m-5, respectively. Dust size is represented by five bins with diameter ranging from 0.2 to 20 µm (bin 1: 0.2–2, bin 2: 2–3.6, bin 3: 3.6–6, bin 4: 6–12, bin 5: 12–20 µm). The corresponding source fractions have been updated from 0.1, 0.225, 0.225, 0.225, and 0.225 to values of 0.04, 0.14, 0.19, 0.49, and 0.14 for the five bins. These updated source functions allocate a higher fraction to coarser size bins, following the suggested brittle fragmentation theory (BFT) as proposed by Kok (2011). Dust mineral composition in the standard AM4.0 is considered uniform, with no temporal and spatial variations; in other words, the dust refractive index (RI) is temporally and spatially homogeneous (the case referred to as homogeneous dust hereafter). The dust RI in the standard AM4.0 is taken from Balkanski et al. (2007), assuming a fixed hematite content of 2.7 % by volume (HD27), which was calculated for the internal mixture of hematite and five other minerals (calcite, quartz, illite, kaolinite, and montmorillonite) using the Maxwell–Garnett mixing rule (see details in Balkanski et al., 2007). The decision to fix the hematite content for dust particles at 2.7 % was made during the development of the previous GFDL Climate Model CM3 (Donner et al., 2011). This decision was prompted by the discovery that dust absorption was unrealistically high (by a factor 3) in CM2 (Delworth et al., 2006) compared to AERONET observations (Balkanski et al., 2007). In CM3, the conjunction of a sharp decrease in black carbon (strong aerosol absorber) with a new emission inventory and the switch to more scattering dust had a negative effect on precipitation bias and late 20th century warming (see Donner et al., 2011, for details). To mitigate this bias, the selection of 2.7 % hematite was adopted in CM3, as well as in the subsequent GFDL models. The control run conducted with the homogeneous dust in the standard AM4.0 model is labeled as HD27 as described in Table .

In addition, we conduct simulations assuming homogeneous dust with hematite content of 0.9 % by volume, with RI from Balkanski et al. (2007). Similar to HD27, this experiment, labeled as HD09 in Table , does not account for the temporal and spatial variations in dust mineralogy.

Claquin et al. (1999) (C1999) is the earliest study providing a soil mineralogy map oriented toward atmospheric and climate modeling. The soil map provides the mineral mass fractions present in the clay and silt size ranges for eight different minerals, namely illite, smectite, kaolinite, calcite, quartz, feldspars, gypsum, and hematite. In this study, we implement the eight minerals from the soil mineralogy map provided by C1999 in GFDL AM4.0 to resolve dust mineralogy. The soil map is based on soil analyses that are usually done after wet sieving, which disperse mineral aggregates into small particles. This dispersal is particularly relevant for the phyllosilicates, typically found in the form of aggregates in soils. They are detected in the atmosphere with higher proportions at coarser (silt) sizes than those reported in the soil maps (Perlwitz et al., 2015b; Perez Garcia-Pando et al., 2016). These recent studies also show that the brittle fragmentation theory (BFT; Kok, 2011) represents a practical framework to generate the emitted particle size distribution based on the dispersed soil PSD, which facilitates the utilization of soil mineralogy maps. In our simulations, we employ BFT to reconstruct the mineral aggregates emitted from the original undispersed soils, following the methods described in Gonçalves Ageitos et al. (2023). The mass densities of the eight minerals, along with a brief description of their importance to Earth's climate, are listed in Table . The density of minerals impacts their settling velocity, which is relevant to the removal of particles in the atmosphere. Goethite and hematite are the two major types of iron oxides present in soils. Goethite is less absorptive than hematite and is not resolved in C1999. So, iron oxides are represented by hematite in this study. Hematite has larger density than other minerals so that hematite deposits more quickly and is not able to be transported to remote regions when not aggregated or internally mixed with lighter clay minerals. Moreover, among the minerals considered here, hematite is the strongest absorber at ultraviolet (UV) and visible wavelengths, while it does not have noticeable absorption at infrared (IR) wavelengths compared to other minerals (Sokolik and Toon, 1999). As such, the correct representation of hematite content in dust aerosols is critical in improving the representation of dust interaction with SW radiation in climate models. All minerals are considered to be externally mixed, except for iron oxides. A large part of the emitted flux of iron oxides is considered to be internally mixed with other minerals, e.g., in the form of accretions in phyllosilicates, in line with observational evidence and previous modeling studies (Kandler et al., 2009; Perlwitz et al., 2015a; Zhang et al., 2015; Panta et al., 2023). As suggested by Gonçalves Ageitos et al. (2023), we define two different types of tracers for the iron oxides: one set of tracers carries the mass of the hematite that constitutes small accretions in clay minerals (i.e., internally mixed with clay minerals), which are allowed to be up to 5 % of the masses of their host minerals at emission (Perlwitz et al., 2015a; Gonçalves Ageitos et al., 2023). Given the low fractional mass of hematite compared to their host minerals, we assume that these accretions do not change the density of their host particles. These internally mixed accretions form the largest fraction of the emitted hematite. Another smaller set of tracers carries the mass of the remaining fraction of hematite, which is considered to be externally mixed with the other minerals, including the internal mixtures of hematite with clay.

In addition to the similar roles of clay minerals in carrying iron oxides, the optical properties of the three clay minerals are very similar, and the optical properties of their external mixture are found to be almost identical to their internal mixture (see Sect. S1 in the Supplement). This finding suggests the use of a single mineral species to represent all three clay minerals in their interaction with radiation to reduce computational cost. Therefore, the optical properties of one single mineral (clay433) are used to represent the optical properties of all three clay minerals. The clay433 represents a mixed mineral comprising three clay minerals: illite, kaolinite, and smectite, with mass fractions of 40 %, 30 %, and 30 %, respectively (see detailed descriptions in Sect. S1). This simplification streamlines the calculations of optical properties for internal mixtures of hematite and the three clay minerals (illite, kaolinite, and smectite), reducing it from an internal mixture of four minerals (hematite, illite, kaolinite, and smectite) to an internal mixture of two minerals (hematite and clay433).

The optical properties of the internal mixture of hematite and clay433 are calculated using three mixing rules: volume-weighted average (VOL-mixing), Maxwell–Garnett mixing rule (MG-mixing), and Bruggeman mixing rule (BM-mixing). Generally, VOL-mixing is used for a quasi-homogeneous mixture: that is, when the components have a similar refractive index. For cases involving a dominant homogeneous host with small inclusions of contrasting composition, MG-mixing is appropriate. BM-mixing is suitable for mixtures in which the inclusions virtually occupy the entire volume of the particle, and the host disappears. A detailed discussion regarding the three mixing rules and their applications can be found in Liu and Daum (2008) and Markel (2016). The appropriate selection of mixing rules is important for the determination of the optical properties of the mixtures. Therefore, we incorporate all three mixing rules in this study. These calculations are performed for various volume mixing fractions of hematite with respect to clay433 to construct a lookup table (LUT) for each mixing rule. The optical properties of each mineral as well as the internal mixtures of hematite and clay433 are calculated offline using Mie code with a spherical shape assumption. As all other minerals have similar SW absorption, internal or external mixing does not change their absorption properties. So, we assume all other minerals to be externally mixed. More details about optical properties of minerals will be discussed in Sect. .

Overall, we implement nine types of mineral tracers: seven non-hematite minerals along with distinguished internal and external hematite, as listed in Table . Each type is distributed across five size bins. As a result, 45 mineral tracers have been incorporated in AM4.0 to account for dust mineralogy.

The AERONET Version 3 Level 2.0 Almucantar inversion retrievals (Giles et al., 2019; Sinyuk et al., 2020) from 2000 to 2020 are screened for dust events following the methodology in Gonçalves Ageitos et al. (2023) and Obiso et al. (2023). This screening process aims to select dust-dominated events and filter out the AERONET scenes contaminated by other absorbing aerosols. The criteria that are applied to AERONET retrievals to screen dust events are as follows: (1) hourly retrievals from AERONET are considered to represent dust when the fine volume fraction is small (below 15 %), (2) the SSA increases from 440 to 675 nm (a feature that distinguishes dust from other species, see Dubovik et al., 2002), and (3) the mean of the imaginary index at red and near-infrared wavelengths (675, 870, and 1020 nm) is lower than 0.0042 (as higher values would indicate the presence of absorbing black and brown carbon, following Schuster et al., 2016). We calculate AERONET SSA in the visible by averaging AERONET-retrieved SSA at two visible wavelengths (0.44 and 0.67 µm) weighted by the solar spectrum following Eq. (2).

The lab-measured dust SSA at 550 nm is obtained from Di Biagio et al. (2019) (DB-2019 hereafter), in which dust SSA was directly retrieved from scattering and absorption measurements. We acknowledge the limits of laboratory measurements, where the dust samples are not aerosols present in the atmosphere, but instead are reemitted in the lab from soil samples collected from various source regions. Consequently, the laboratory measurements in DB-2019 do not account for dust aerosols transported from other regions to the regions of interest. In addition, in contrast to the modeled dust diameter range of 0.2 to 20 µm, DB-2019 measures dust particles with a diameter ranging from 0.2 to 10 µm.

To compare modeled fluxes at TOA with observations, we use the Clouds and the Earth's Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) Edition-4.2 data (Loeb et al., 2018). The standard CERES level-3 products provide clear-sky fluxes by averaging all CERES footprints within a region that are completely free of clouds. Therefore, there are many missing regions in monthly mean clear-sky TOA flux maps because completely cloud-free conditions are not always observed at the CERES footprint scale (∼ 20 km at nadir). In contrast to the standard CERES level-3 products, the CERES_EBAF product infers clear-sky fluxes from clear portions of partly cloudy CERES footprints, thereby producing a clear-sky TOA flux climatology free of any missing regions (details in Loeb et al., 2018). Starting from CERES_EBAF_Ed4.1, the product also provides clear-sky flux estimates for the total region (i.e., the total CERES footprints) by combining CERES observations and radiative transfer calculations, which represents clear-sky flux with clouds removed from the entire atmospheric column of CERES footprints. These clear-sky fluxes for the total region are defined in a way that is more consistent with how clear-sky fluxes are represented in climate models (for details see the CERES_EBAF_Ed4.1 Data Quality Summary). In this study, the monthly mean TOA “Clear-Sky Flux Estimate for Total Region” variables in the CERES_EBAF_Ed4.2 product, the most recent version of the product, are used to compare with modeled monthly mean clear-sky flux at TOA. The comparisons allow us to examine the agreement of modeled clear-sky fluxes from different experiments with observations. The comparison results are shown in Sect. .

The CRU TS (Climatic Research Unit gridded Time Series) provides a high-resolution (0.5° latitude × 0.5° longitude) climate dataset over land except Antarctica. The dataset is based on extensive networks of weather stations going back to 1901 (Harris et al., 2020). This dataset has been widely used in various research areas since its first release in 2000. The mean 2 m temperature (TMP) and precipitation rate (PRE) variables from CRU TS v4.07 are used to evaluate our model simulations. The results are shown in Sect.  and .

We use the refractive indices (RIs) of each mineral from Scanza et al. (2015) and the RIs of HD27 and HD09 from Balkanski et al. (2007) to calculate the LUT of optical properties. The spectral RIs of each mineral and homogeneous dust (e.g., HD27 and HD09) are shown in Fig. . The HD09 dust has a lower imaginary part of RI at 550 nm than HD27 dust, indicating its lower absorption in the visible band due to a reduced content of hematite.

After calculating the LUT of optical properties (details in Sect. ), we incorporate the interaction of minerals with radiation into GFDL AM4.0. The modeled emission, load, deposition, and lifetime for each mineral are provided in Table S1 in the Supplement. Table  provides global total dust emission and load, globally averaged dust aerosol optical depth (DAOD), and SSA for each experiment listed in Table  and their comparisons with previous studies. DAOD and SSA from AM4.0 simulations are averaged in the visible band (0.44–0.625 µm) of GFDL AM4.0. Unless otherwise specified, DAOD and SSA in this study refer to the average in the visible band. Note that in our calculations, the domain-averaged DAOD is always weighted by the area of each grid cell. The domain-averaged SSA is always weighted by the area and DAOD of each grid cell. Additionally, the spectrally averaged DAOD and SSA are always weighted by the TOA solar radiation intensity at the corresponding wavelengths, peaking around 0.50 µm, as shown in Eqs. (1) and (2). 1DAOD‾=∫λ1λ2DAOD(λ)B(λ)dλ∫λ1λ2B(λ)dλ2SSA‾=∫λ1λ2SSA(λ)DAOD(λ)B(λ)dλ∫λ1λ2B(λ)DAOD(λ)dλ B(λ) describes the solar radiation energy intensity, which can be calculated by means of the Planck's function B(T,λ), using the temperature of the Sun (T = 5800 K).

The lowest SSA of HD27 in Table  suggests that HD27 dust, which has been used in the standard AM4.0 model, is the most absorptive among all experiments. The HD09 dust is much less absorptive, attributed to its smaller hematite content, as indicated by the lower imaginary part of RI in the visible range (Fig. ). For the three mineral-resolved experiments, the lower global mean SSA (SSA‾) in VOL suggests that the VOL-mixing dust is more absorptive than MG-mixing and BM-mixing dust. This finding is consistent with previous studies that have suggested that VOL-mixing method, when applied to minerals to compute the bulk aerosol optical properties, may artificially enhance absorption relative to scattering and lead to a lower SSA for bulk dust aerosol (Zhang et al., 2015). We can see that the global mean SSA (SSA‾) of HD09 dust is comparable to the values obtained in cases where minerals are resolved (e.g., MG and BM). This implies that, from a global perspective, HD09 dust is as absorptive as mineral-resolved dust (e.g., MG and BM).

In addition to the globally averaged dust properties listed in Table , we illustrate the distribution of global dust mass across five size bins (Fig. S3 in the Supplement) and the global distribution of DAOD (Fig. ) for the three experiments: before (e.g., HM27 and HD09) and after (e.g., BM) resolving mineralogy. The global dust mass distribution across the five size bins remains largely unchanged across experiments. Besides the subtle difference (∼ 10 %) in global mean DAOD across the three experiments as listed in Table , the global distribution of DAOD responds differently in HD09 and BM. Compared to HD27, reducing hematite content to HD09 generally decreases DAOD, except over the Sahel region. In contrast, resolving mineralogy as in BM decreases DAOD over the Sahara region while increasing DAOD over the Sahel and Asia regions. The reduction in DAOD over the Sahara region further contributes to the decrease in dust absorption over the region, primarily attributed to the change in dust optical properties, such as the enhancement in dust SSA. The indistinct variation in DAOD across different experiments results from the feedback of dust interactions with radiation (Miller et al., 2004a, 2014; Pérez et al., 2006), which is influenced by the distinct scattering properties of dust aerosols in each experiment as shown in Table .

The iron oxide content of dust determines shortwave radiation absorption by dust: the higher the quantity of iron oxides, the lower the SSA. Following our calculations of dust optical properties in Sect. , we compare GFDL AM4.0 modeled dust SSA (averaged in the visible band 0.44–0.625 µm) against AERONET SSA retrievals (averaged at two visible wavelength: 0.44 and 0.67 µm) in Sect.  and laboratory measurements of SSA (at 0.55 µm) in Sect. . The modeled dust SSA is evaluated against observation-based results utilizing the following evaluation metrics: the mean SSA (mSSA) is calculated based on SSA for all locations displayed in Fig. . The standard deviation (σ), derived from SSA for all locations displayed in Fig. , is used as an indicator of dust SSA spatial variation. The normalized mean bias (nMB) and normalized root mean square error (nRMSE) are utilized to assess the mean bias and root mean square error, respectively, of modeled SSA in comparison to observed SSA. Definitions of nMB and nRMSE are provided in Sect. S3.

We start our analysis by examining the impacts of resolving dust mineralogy on clear-sky radiative fluxes. By “clear sky”, we mean that our results do not consider the radiative effects of clouds. We use anomalies (ΔF) to evaluate the impacts of resolving mineralogy on clear-sky radiative fluxes, which is defined as ΔF=F1↑-F2↑, where F1↑ represents the 19-year mean clear-sky upward radiative fluxes with resolved mineralogy, and F2↑ represents the 19-year mean clear-sky upward radiative fluxes for homogeneous dust. Section S5 provides the clear-sky radiative fluxes anomalies at TOA and the surface (SFC) induced by resolving dust mineralogy over the global scale, we see much more significant anomalies over North Africa than other regions, which makes senses because that dust aerosol is the most dominant aerosol species in this area. The changes in dust aerosol optical properties have a greater potential to lead to significant impacts on radiation and climate over the region than in the others. Therefore, this section focuses on the North Africa region, where the Sahara, the largest dust source in the world, is located. The Sahara (20–30° N, 10° W–35° E) and the Sahel (10–20° N, 10° W–35° E) regions are studied separately. We specifically analyze the results for the June–July–August (JJA) season when dust loading is at its highest and the West African monsoon is the strongest.

Here we explore the impacts on the temperature vertical profile and near-land surface temperature relative to HD27 and HD09, respectively. Compared to the HD27 control run, lower absorption of radiation in the atmosphere by mineral-resolved dust aerosols results in statistically significant negative temperature anomalies in the atmosphere ranging from 800 up to 500 mb where dust aerosols are mainly located (Fig. ). In contrast, there is no statistically significant temperature anomaly for mineral-resolved dust cases compared to HD09, as illustrated by the red curves in Fig. . This finding aligns with the insubstantial anomalies (< 5 %) in clear-sky NET radiative fluxes discussed in Sect. . In the subsequent part of the section, we will delve into comparing a mineral-resolved experiment (using BM as an example) with the HD27 control run. This comparison will help us further understand the impact of dust aerosols with distinct absorption on land temperature.

Figure a shows air temperature at 2 m from the HD27 control run over North Africa. Near the land surface, more scattering mineral-resolved dust induces a temperature decrease (i.e., negative temperature anomaly -0.66 K) over the Sahara and a temperature increase (i.e., positive temperature anomaly 0.70 K) over the Sahel as shown Fig. b. To understand this phenomenon, we further analyze the surface energy budget in Fig. c–f.

Radiative flux perturbation over land is quickly equilibrated by balancing surface radiative fluxes with sensible heat flux, latent heat flux, and ground flux (i.e., downward heat flux into the ground), which results in nearly zero net energy flux at the land surface as shown in Fig. f. Specifically, the radiative flux anomaly comprises two contributions: one is the instantaneous radiative forcing (IRF) caused by the change in dust mineralogy in the atmosphere, and the other one is the associated radiative feedbacks. For simplicity, we will not partition the radiative flux anomaly in our discussion here. Over the Sahel region, the positive net radiative flux anomaly at the land surface is balanced out by the increased sensible heat flux and the decreased latent heat flux as well as ground flux. Note that the ground flux is generally small in magnitude and not shown in Fig. , but we include it in calculating net surface energy flux in Fig. f. The decrease in latent heat flux over the Sahel in the BM case (Fig. d) is due to the depletion of soil moisture (and therefore evaporation) in the region as shown in Fig. . The depletion of soil moisture is caused by the decrease in moisture carried by onshore winds over the Sahel and the decrease in precipitation over the same region, as discussed in Sect. . Therefore, a large enhancement of sensible heat flux (∼ 7.6 Wm-2) is needed (Fig. c) not only to compensate for the depletion in latent heat flux (∼ 5.7 Wm-2 in d), but also to balance out the increased net radiative flux (∼ 2.3 Wm-2 in Fig. e). As a result, higher land surface temperature with an anomaly around 0.7 K is needed in the region to achieve the required sensible heat flux enhancement.

Over the Sahara region, latent heat flux does not change from the HD27 case to BM case; therefore, the increased net radiative flux (∼ 3.69 Wm-2) in BM compared to HD27 is mainly balanced out by the enhanced sensible heat flux (∼ 3.98 Wm-2), which requires a larger temperature gradient between the surface and atmosphere. However, there is a very strong negative temperature anomaly (around -1 K) in the atmosphere near 700 hPa due to less dust absorption in BM as we discuss in Fig. . The strong negative temperature anomaly in the lower atmosphere effectively increases the vertical temperature gradient. As such, it is not necessary for the land surface temperature to increase; in fact, it may need to decrease by approximately 0.66 K to achieve the desired enhancement in sensible heat flux and reach equilibrium.

Additionally, to assess the effectiveness of various dust scattering properties (e.g., HD27, HD09, and BM) in matching observations of near-surface temperature, we compare the modeled near-surface temperature (T2m) with CRU TS observations, which is described in Sect. , over the Sahara and Sahel regions (Table ). Considering the relatively large inter-model spread of regional surface air temperature, we compare the Sahara–Sahel regional contrast in surface air temperature to the CRU rather than comparing their absolute values. Table  shows that HD09 and BM improve the agreement with CRU in Sahara–Sahel temperature contrast compared to HD27, and BM exhibits the closest agreement with CRU.

To understand the fast circulation and hydrological response resulting from resolving dust mineralogy, we examine surface wind speed anomalies (Fig. ) and precipitation anomalies (Fig. ) induced by mineral-resolved dust. We compare mineral-resolved experiments (using BM as an example) with HD27 and HD09, respectively, to investigate the effects of resolving dust mineralogy.

Global precipitation is higher by 0.017 mmd-1 for the BM mineral-speciated case compared to the HD27 experiment. This is consistent with the lower SW absorption in the former, given the global compensation between latent heating associated with precipitation and net radiative cooling (e.g., Allen and Ingram, 2002; Samset, 2022). (Global radiative cooling is also compensated for, although to a smaller extent, by the sensible heat flux.)

However, within the Sahel, precipitation for BM is reduced compared to HD27 (Fig. b), with weaker onshore flow during the summer monsoon (Fig. b), displacing West African precipitation toward the Guinea coast. This reduction is consistent with several previous calculations of the fast response calculated with fixed SST (e.g., Stephens et al., 2004; Miller et al., 2004b; Lau et al., 2009; Jin et al., 2016; Jordan et al., 2018).

Dust absorbs radiation and redistributes heating from the surface to within the dust layer (Miller and Tegen, 1999; Strong et al., 2015). The heating of the air warms the lower to middle troposphere, thereby enhancing upward motion. The rising warm air spawns a large-scale onshore flow, carrying the low-level moist air from the Atlantic to the Sahel, thus enhancing precipitation over this region (Balkanski et al., 2021). The more scattering mineral-resolved BM and MG dust absorbs less radiation and causes less warming of the atmosphere (Sect. ), reducing adiabatic cooling through ascent and Sahel precipitation. The suppressed ascent in BM compared to HD27 is associated with a reduction in both the wind divergence aloft and convergence at the surface (Fig. b). The reduction in convergence results in northeast wind anomalies at 10 m over the Sahel (Fig. b), which are opposite in direction to the southwest onshore winds of the West African monsoon (Fig. a). The inhibition of onshore winds, bringing less moisture to the Sahel, is consistent with the reduction of ascent and precipitation over this region (Fig. a and b).

Besides the Sahel, there is a statistically significant positive anomaly (0.92 mmd-1) of precipitation over the region to the south of the Sahel in BM relative to HD27 (Fig. b). We will call this region the Guinea coast (GC: 0–10° N, 10° W–15° E). One possible reason for the increase in precipitation over the GC is that the region is located to the south of the Saharan dust layer, where the suppression of ascent over the Sahel in BM suppresses the subsidence in the GC region and, therefore, enhances the precipitation (Guo et al., 2021). Alternatively, the moist onshore flow that is weakened in BM is subject to greater dilution by the dry desert air, resulting in reduced moist static energy and buoyancy, limiting convection to the coastal region.

These changes in precipitation have non-negligible effects on soil moisture content in North Africa due to its moisture-starved environment. The decrease in precipitation over the Sahel in BM leads to a reduction in soil moisture content. Conversely, the increase in precipitation over the GC leads to increases in soil moisture (Fig. ). The change in soil moisture content further affects the partitioning of surface energy fluxes and the efficiency of the latent heat flux, thereby affecting land surface temperature, as illustrated by Fig. .

So far, we have been focusing on discussing the impacts of resolving dust mineralogy on winds and precipitation relative to the HD27 control run. The large discrepancy in optical properties between HD27 and mineral-resolved dust allows us to better understand how distinct dust absorption impacts our climate through its distinct radiative effects.

As discussed in Sect. , HD09 dust is nearly as scattering as mineral-resolved dust but exhibits smaller regional variability. Section  shows that resolving dust mineralogy does not lead to statistically significant anomalies of radiation relative to HD09. Consistently, there are no further statistically significant impacts on winds (Fig. c and d) and precipitation (Fig. c and d).

To investigate the effectiveness of various dust scattering properties (e.g., HD27, HD09, BM) in matching observations of precipitation rate, we compare the modeled precipitation with CRU TS observations over the Sahara, Sahel, and GC regions (Table ). The greater difference between HD09, BM, and CRU (i.e., HD09–CRU and BM–CRU) indicates that more scattering HD09 and BM lead to a larger discrepancy between the modeled precipitation and CRU observations. In contrast, Balkanski et al. (2021) describe the same balance of increased dust absorption and Sahel precipitation but find improved agreement with Global Precipitation Climatology Project (GPCP) data by assuming homogeneous dust containing 3 % iron oxides by volume. Contrasts between that study and ours result from differences between the GPCP and CRU datasets, contrasts in dust absorption (related to contrasts in the dust size distribution or assumed index of refraction), non-dust model biases in precipitation, or differences between the slow response computed by Balkanski et al. (2021) and the fast response that we calculate. The fast and slow responses even exhibit differences in the sign of the calculated precipitation anomaly within some regions of the WAM (Miller and Tegen, 1998; Jordan et al., 2018).