Better representation of dust can improve climate models with too weak an African monsoon

The amount of short wave radiation absorbed by dust has remained uncertain. We have developed a more accurate representation of dust absorption that is based on the observed dust mineralogical composition and accounts for very large particles. We analyze the results from two fully coupled climate simulations of 100 years in terms of their simulated precipitation patterns against observations. A striking benefit of the new dust optical and physical properties is that tropical precipitation over the Sahel, tropical North Atlantic and West Indian Ocean are significantly improved compared to observations, without degrading precipitations elsewhere. This alleviates a common persistent bias in Earth system models that exhibit a summer African monsoon that does not reach far enough north. We show that the improvements documented here for the IPSL-CM61 climate model result from both a thermodynamical and dynamical response to dust absorption, which is unrelated to natural variability. Aerosol absorption induces more water vapor advection from the ocean to the Sahel region, thereby providing an added supply of moisture available for precipitation. This work, thus, provides a path towards improving precipitation patterns in these regions by accounting for both physical and optical properties of the aerosol more realistically.


Introduction
Mineral dust influences precipitation through aerosol-radiation interactions (Miller et al., 2014), changing the vertical temperature profile, and is an efficient ice nucleus in the presence of feldspar mineral (Atkinson et al., 25 2013), therefore also producing an indirect, cloud-mediated radiative perturbation. It also influences the water cycle through microphysical interactions with clouds (Nenes et al., 2014). The energy absorbed by mineral dust in the atmospheric column is defined as the difference between top-of-atmosphere and surface radiative effects.
In source regions, it represents several tens of watts per square meter (W.m -2 ) and hence is stronger than the anthropogenic aerosol-cloud interactions (Miller et al., 2014;Nenes et al., 2014). The Sahel precipitation is 30 influenced by the energy absorbed by aerosols (Miller et al., 2004;Solmon et al., 2008;Yoshioka et al., 2007), and dust absorption of energy depends on iron oxides (hematite and goethite) that are part of dust mineralogical composition (Sokolik and Toon, 1996;Claquin et al., 1999). Over the last 15 years, simulating tropical precipitation has been notoriously difficult for climate models (Fiedler et al., 2020). Improving the representation of tropical monsoons is a prerequisite to predict future changes in tropical precipitations and 35 attribute them to observed changes in greenhouse gases and to aerosol changes. We show here how a better representation of dust aerosols leads to an unequivocal improvement in the simulation of precipitation over key climatic tropical regions in the IPSL-CM6 model, namely the Sahel, tropical North Atlantic and West Indian Ocean without degrading precipitation elsewhere around the globe, and subsequently discuss the thermodynamically and dynamically-driven mechanisms at play that affect the water cycle. 40 Miller et al. (2014) showed that the increase in the Sahel precipitation in response to high dust absorption is a fairly robust result across models. The intensity and the seasonal pattern of tropical precipitations are controlled by the northward cross-equatorial transport of energy (e.g. Hwang and Frierson, 2013). CERES observations of the Earth's energy budget indicate a net northward cross-equatorial transport of energy. The atmospheric component of this cross-equatorial transport is southward whereas the oceanic component is northward (see Fig.   45 4 in Stephens et al., 2016). Haywood et al. (2016) discussed how, in the UK Met Office HadGEM2 model, tropical precipitation biases could be reduced by setting the northern and southern hemispheric albedos to be equal in agreement with multi-decadal satellite observations. Hemispheric albedo changes thus strongly influence tropical precipitations. The link between hemispheric albedo, aerosol loadings and properties in general, and dust atmospheric absorption in particular remains however poorly understood.

50
The abundancy and variation in iron oxides, as well as the presence of large particles, are known to control dust absorption (Balkanski et al., 2007;Miller et al., 2004;Ryder et al., 2018;Yoshioka et al., 2007). Although iron oxides in dust are present in minute quantities from 1 to 5% by volume (Di Biagio et al., 2019;Journet et al., 2014;Kandler et al., 2007;Liu et al., 2018;Nickovic et al., 2012;Perlwitz et al., 2015;Reid, 2003), this relatively small volume largely influences mineral dust absorption (Balkanski et al., 2007;Ryder et al., 2018; Di 55 Biagio et al., 2019). Large dust particles with diameters ranging between 10 and 100 m, are rarely accounted for in climate models. Recent measurements have shown a much stronger absorption from these large particles than from smaller ones (Ryder et al., 2013(Ryder et al., & 2018. Due to the abundancy of these large particles over source regions, the atmospheric energy balance will be altered. Furthermore, the Sahel is one of the world regions with the highest iron oxide content in soils (Journet et al., 2014;Nickovic et al., 2012). In this study, we analyse the 60 strong relationship between high iron oxide content and increased the Sahel precipitation. Here we discuss the role played by dust on tropical precipitation and describe an end-to-end physical mechanism that ties improvement in tropical precipitation to observational support for a higher level of dust absorption based on measurements of iron oxide in dust particles, measurements of the full dust particle size distribution and detailed climate simulations with interactive dust. 65 2. Methods

IPSL-CM6 description
The climate model used here is the low resolution model from the Institut Pierre-Simon Laplace Climate Modelling Centre described by Boucher et al. (2020). The horizontal resolution is 2.5° in longitude and 1.28° in latitude with a discretization of the vertical into 79 layers that extends to about 80 km. For the ocean model, 70 NEMO, that includes sea-ice and biogeochemistry, the horizontal resolution is 1° and the model is discretized using 75 vertical levels. The aerosols are run interactively in the simulations presented here.

Dust modeling
Dust emission fluxes are calculated in two steps: in a first step, we derive the horizontal flux of dust that is mobilized based upon three criteria: a threshold velocity that depends on the nature of the upper soil, the wind 75 speed at 10-meters and an erodibility factor that takes into account the effect of soil moisture. These erodibility factors were tuned following the procedure described in Balkanski et al. (2004). Total emissions and loads compare well with the constraints given by Ridley et al. (2016) and by Kok et al. (2017). The dust particle size distribution are emitted with a constant shape following the Brittle theory described by Kok (2011). The size distribution is represented by one or by four modes, each one represented with a log-normal distribution 80 consisting of a mass median diameter which varies in response to the sink processes that affect the dust cycle.
When using one mode for the simulation, this mode is centered at 2.5 m with a width of 2.0 and represents both the accumulation and coarse modes (Schulz et al., 1998;Denjean et al., 2016). The four modes size distribution allows to account for large particles up to 100 m (Di Biagio et al., 2020). These four modes have mass median diameters of 1.0, 2.5, 7.0 and 22.0 m, respectively and are fitted to the size distribution predicted by the brittle 85 theory for particles of diameter below 20 m and to observations made during the campaigns Fennec and AER-D (Ryder et al., 2013(Ryder et al., & 2018 for diameters above 20 m as described in Di Biagio et al. (2020).
Dust absorption is influenced mainly by the iron oxide embedded in the dust aggregates (Lafon et al., 2006;Ryder et al., 2018;Di Biagio et al., 2019). Measurements of iron oxides on soils from around the world have been reported for two particle size classes: the clays with diameters of less than 2 microns and the silts with 90 diameters between 2 and 64 m. The iron oxide varies drastically depending on the soil types and most measurements indicate a weight content of 1 to 7% equivalent to 0.5 to 3.5% (Lafon et al., 2006;Di Biagio et al., 2019;Moosmüller et al., 2012;Engelbrecht et al., 2016;Journet et al., 2014) by volume. To determine the amount of iron oxides over the Sahel, we used the high-resolution database published by Nickovic et al. (2012) and determined the amount of hematite over the Sahel region (16°W-36°E; 10°N-20°N) as shown in Fig. 1. With 95 the 30'' grid high-resolution of the database, the mineral content of hematite could be retrieved for 6,026,016 points. For the clay fraction (diameter ≤ 2 m), 50% of these points had an hematite content of more than 2% by weight (equivalent to 1% by volume since density of hematite is twice that of all other minerals except goethite); 30% (respectively 17%) of the points had an hematite content of more than 3% (resp. 4%) by weight. For the silt fraction (diameter > 2 m), 49% of these points had and goethite content of more than 2% by weight (equivalent 100 to 1% by volume since density of hematite is twice that of all other minerals except goethite); 30% (respectively 12%) of the points had an hematite content of more than 4% (resp. 5%) by weight. Assuming that hematite and goethite contents are the same for these soils and accounting for the density of hematite (5300 kg m -3 ) and goethite (3800 kg m -3 ), we estimate that iron-oxides (hematite+goethite) represent 5.3% by weight and 3.0% by volume of mineral dust that has a density of 2650 kg m -3 . Hence, in the simulation, we took the optical properties 105 of mineral dust with a volume of 3.0% made of iron oxides.
The difference in optical properties between goethite and hematite are not considered in this study as we focus on the mechanisms by which dust absorption causes an increase of precipitation over the Sahel and not into 110 having a very precise estimate of this absorption. We keep a constant volume content of hematite of 3.0% representative of the Sahel region. The first step of the computation is to compute the refractive index of each of the above cited minerals with hematite, for example for 3% hematite, all mixtures are composed of 97-3%.
Using the Maxwell Garnett mixing rule (Chylek et al., 1988), we compute the equivalent refractive index (me) of a particle that consists of the assemblage of 2 minerals is computed in the following way:

115
(1) where m0 is the refractive index of mineral around the inclusion, and mA is the refractive index of the inclusion in the internal mixture; and vA is the volume fraction of inclusion of the particle.
The second step is to compute the refractive index starting first with the most abundant constituents of these clay mixtures illi-hema and kaol-hema. The third step takes the resulting mixture illi-kaol-hema and mixes it with 120 montmorillonite. The resulting mixture illi-kaol-montmo-hema is then mixed with quartz. And finally the mixture illi-kaol-montmo-quartz is mixed with calcite, the least abundant of those minerals. We refer the reader to Table 1 of Balkanski et al. (2007) that explains the abundancies of the different assemblages and minerals.
We ran 100-year simulation of the fully-coupled IPSLCM6 model with all interactive components of the aerosol including dust for the 1915-2014 period, as well as another 100-year simulation without the dust. We analyzed 125 the last 30 years of the coupled simulations  for the period when precipitation is most abundant over the Sahel from June to September referred to as JJAS in the rest of the text. We checked for all variables the consistency of the results compared to the previous 30-year period from 1955 to 1984. We compare the SW and LW radiative effects of two simulations that have the same absorption: 4 modes (including large particles, 10 µm < D < 100 µm) with 3% iron oxides, to the simulation with 1 mode (without large particle, 10 µm < D < 100 130 µm), the results are shown in Table 1. We ran for the full 100 -year period only the simulation with 1 mode and 5% iron oxide content equivalent to the full size distribution (4 modes) and 3% iron oxide.
In order to compute of the Direct Radiative Perturbation (DRP) from dust, we compare the fluxes both at top-of-  (1980)) and the Rapid Radiative Transfer Model with sixteen bands between 3.33 and 1,000 μm (Hogan and Bozzo, 2016). The model accounts for the diurnal cycle of solar 140 radiation and allows fractional cloudiness to form in a grid box. The reflectivity and transmissivity of a layer are computed using the delta Eddington approximation (Joseph et al., 1976) in the case of a maximum random overlap (Morcrette and Fouquart, 1986) by averaging the clear and cloudy sky fluxes weighted linearly by their respective fractions in the layer. The radiative fluxes are computed every ninety minutes, with and without the presence of clouds, and with and without the presence of aerosols; they are sampled at the top-of-atmosphere and 145 at the surface. The clear-sky and all-sky aerosol radiative effects can then be estimated as the differences in radiative fluxes with and without aerosols.

Deriving the Water Budget along the Sahel region (10°N-20°N; 15°W-35°E)
Following Sheen et al. (Sheen et al., 2017), we seek to determine the total flux of moisture depth across each of the airshed boundaries. Hence we compute the integrated moisture flux through the following integral: where is the density of water, g is the acceleration due to gravity, ps is the surface pressure, 〈 ⃗⃗ 〉 represents the monthly mean of the product of the water content with the wind speed. The fluxes are expressed in unit of kg.m -1 s -1 . The integrated fluxes across each side of the airshed are first averaged, scaled by the length along the flux trajectory (Trenberth, 1999), and then divided by the airshed area to obtain quantities in units of mm per day 155 that can be compared to the precipitation and the evaporation fluxes over the airshed.

Results
We first present how dust absorption is calculated and then present the results from a fully-coupled climate simulation of 100 years (from which we analyse the last 30 years) and compare simulated precipitations with observations. Iron oxide content is based on the observations of dust mineralogical composition over the Sahel 160 region (Lafon et al., 2006;Di Biagio et al., 2019) and displayed for soils in Fig. 1. We infer the refractive index of the mineral dust using an optical model. Figure 2 illustrates the influence of the iron oxide content and the size of a particle on its radiative absorption. In this Figure, the aerosol absorption increases along the x-axis with the aerosol co-single-scattering-albedo (coSSA) calculated at 550 nm and defined as: The solid blue line illustrates the absorption calculated when dust particles with diameter less than 10 m are considered. The shift from the solid blue to the solid orange line indicates the increase in dust absorption when particles with a diameter larger than 10 m are accounted for. This graph also indicates that particles of diameter less than 10 m with an iron oxide content of 5.0% absorb the same amount of radiation than particles with 3.0% iron oxide for which we consider also the diameters greater than 10 m (see Table 1). With these mineralogical 170 compositions we obtain a coSSA of 0.09. For dust particles with an iron oxide volume content of 1.5%, thought to be the global median value as discussed in Balkanski et al. (2007), the coSSA is 0.032 and it almost doubles to 0.058 when large particles are also considered. Hence, the substantial increase in aerosol absorption caused by large particles will be particularly marked over dust source regions. This has yet to be taken into account in many models that generally do not include particles sizes above 10 m.

175
We now discuss the changes in aerosol direct radiative effect for dust containing 3.0% iron oxide which corresponds to the amount of iron oxides measured over the Sahel (Fig. 1). Figure 3 illustrates the June, July, August and September (JJAS) mean radiative perturbation due to the presence of dust. Over the bright surfaces of the Sahel region, the top-of-atmosphere shortwave radiative perturbation is positive, i.e., an atmosphere with dust is less reflective than an atmosphere without dust. At the surface, the radiative effect from dust is strongly 180 negative (-16 W.m -2 over the Sahel) as the dominant term is the reduction of shortwave radiation due to either back-scatter or absorption of radiation by dust. The difference between TOA and surface determines the dust JJAS mean atmospheric absorption due to dust, estimated at +17 W m -2 over the Sahel). Since dust is highly  Figure 4 shows that aerosol absorption induces more water vapor advection from the ocean to the Sahel, thereby providing an added supply of moisture that is available for precipitation. When sufficient water vapor is advected towards the African continent, aerosol absorption changes the regional energy balance and enhances convection through the air column (Miller et al., 2014;Solmon et al., 2008). The change in 205 water flux into the Sahel region amounts to an increase of 0.365 mm day -1 for the JJAS period. Precipitation and evaporation over the Sahel region increase by 0.40 and 0.09 mm day -1 , respectively, for the same months over the 30-year period . Hence there is a small residual in the water budget which we attribute to imperfections in the way the advected water fluxes are diagnosed in the model.

210
We now seek to analyze if the changes in precipitations, in SW and LW upward radiative fluxes brought about by the presence of dust improve or degrade these variables when compared to observations. We selected observations from the Global Precipitation Climatology Project (GPCP) from June to September over the period 1985 to 2014, and the radiative fluxes measured by CERES (Clouds and the Earth's Radiant energy System).
Both, the precipitation and the SW and LW upward radiative fluxes statistics based upon the 215 model/measurements comparison show significant improvements over the Sahel, North Africa and the North Atlantic Ocean. We quantify these improvements in Tables 2, 3 & 4 through a comparison of the bias, root mean square error (RMSE) and spatial correlation between the simulated and observed fields. For precipitation, the largest changes occur over the Sahel where precipitation increases by 21% over the period and the negative bias and the RMSE are reduced by 34 and 29%, respectively. The spatial correlation of the precipitation is also 220 improved from 0.951 to 0.965. Other regions where all of the mentioned statistics are improved are the North Atlantic and the West Indian Ocean. Over Northern Africa, the strong dust absorption causes an excess precipitation but there are improvements in the RMSE and the spatial correlation.
The Sahel region is prone to substantial atmospheric dust absorption as it is an important mineral dust source region. Being an active source region, the lower troposphere above the Sahel experiences high dust loads and a 225 large mass fraction comprises large-size particles of diameters above 10 m (larger particles being more absorbing than smaller ones, see Figs. 8a,b in Ryder et al., 2013). Compounding these effects, soils from the Sahel have a higher iron oxide content than other soils from North Africa (Kandler et al., 2007;Liu et al., 2018;Di Biagio et al., 2019;Formenti et al., 2014). The high atmospheric absorption that exists over the Sahel and illustrated by Fig. 3, induces water moisture advection from the North Atlantic in regions off the West Coast of 230 Africa during the period of the African summer monsoon (Fig. 4). The action of mineral dust aerosol on precipitation is felt through the modifications of the diabatic heating and how the aerosol affects evaporation.
The terms in the energy balance that influence evaporation are the dimming caused by the aerosol layer of -18 W.m -2 over the Sahel (see Fig. 3) which needs to be compensated by the decrease of net latent heat flux that leaves the surface. Table 5 shows the different terms of the change in surface energy budget. Ignoring the 235 changes in surface heat storage, we have: where   represent the net change in respectively the SW and LW radiative fluxes at the surface, and the two other terms,  and  , indicate the changes in turbulent fluxes of latent and sensible heat due to the presence of dust. Over the oceans, the change in latent heat flux dominates and the evaporation is 240 reduced hence diminishing precipitation over these regions (see last column of Table 1). Figure S2 indicates the vertical amount of transported Moist Static Energy (MSE). Two regions can be distinguished, one over the Sahel where the transport is upward as a result of the absorbed energy by dust, and the surrounding regions with a downward motion. These are the main perturbations that accompany the precipitation anomaly caused by the presence of dust. Other variables that increase are: evaporation, low-level cloud cover, liquid water path. As 245 noticed by Miller et al. (2004), and reinforced by the results presented here, evaporation over the Sahel increases with dust absorption (Fig. S4), while precipitation increases over the same region. Part of the evaporation is supplied by the moisture brought through precipitation over the same region. One mechanism for supplying the moisture is the advection that is evidenced in Fig. S1 and in the water budget presented on Fig. 6.
To complete the analysis of the possible mechanisms that may have explained this improvement in the Sahel 250 precipitation, we examined whether a different phasing of the Atlantic Multidecadal Variability (Enfield et al., 2001) (AMV), a basin-wide low-frequency variations of the sea surface temperature over the North Atlantic, could be responsible for better reproducing observed precipitation fields. Figure S5 compares the observed and simulated AMV Index (Enfield et al., 2001;Trenberth and Shea, 2006). The AMV index is more in phase in the simulation without dust than in the simulation with dust. We conclude from this analysis that the precipitation 255 improvements brought about by dust absorption, which are also evident for earlier portions of the 100-year long simulations, are not due to a better phasing of natural variability in the dust simulation. Instead they correspond to dynamical effects that respond to a thermodynamically driven forcing.

Conclusion
This study was designed to realistically represent dust absorption over the Sahel region and describe the 260 mechanisms by which dust stimulates summer precipitation (JJAS) over the region. Our modelling includes two aspects not accounted for prior to it: first it accounts for the high iron oxide content of the region in the optical properties, and second it takes into consideration the extra-absorption from very large particles, with diameters greater than 10 m, generally not represented in climate models. The striking benefit of these refinements in the dust physics is that, at least in the IPSL-CM6 model 1 considered here, tropical precipitations and top-of-265 atmosphere LW and SW fluxes are significantly improved compared to observations. This important result came almost serendipitously when checking if the comparison between simulated and observed precipitation over 30 years showed improvement. In key regions of tropical precipitations, namely the Sahel, tropical North Atlantic and West Indian Ocean, the precipitation in the IPSL-CM6 climate model is significantly improved without degradation elsewhere. We speculate that this improvement may not be restricted to this climate model 270 as other models participating in the CMIP exercises have low dust absorption as well as a similar bias over the Sahel, which is a cause for too little advection of water vapor northward into the region during the Northern Hemisphere summer months. We thus provide a path towards improving precipitation patterns in these regions by more realistically accounting for both physical (size-based) and optical (absorption) properties of the aerosol.
Our results also offer a strong physical basis for a stabilizing feedback loop involving dust emission, 275 atmospheric absorption, the Sahel precipitation, and vegetation, as hypothesized by Carslaw et al. (2013), which could create multiannual or multidecadal oscillations at these latitudes that could also interact with natural models of variability in the Atlantic region. Future studies should therefore account for the role of water recycling from semi-arid vegetation (Yu et al., 2017)    is for dust particles with diameters of less than 10 m. The shift from the blue to the orange line shows the increase in dust absorption when the assumed size distribution is taken from observations made during the Fennec campaign by Ryder et al. (2013) as shown in Fig.1 of Di Biagio et al. (2020). The optical model used is described in the Methods. The ranges of absorptions measured during the AER-D campaign (Ryder et al., 2018) over Sahara and the near Atlantic are 460 indicated by the blue shading (co-SSA at 550 nm ranging from 0.01 to 0.09), the range measured during the FENNEC campaign (Ryder et al., 2013) with the red shading (coSSA at 550nm ranging from 0.02 to 0.1.