Mineral dust aerosol impacts on global climate and climate change

Mineral dust aerosols impact the energy budget of Earth through interactions with radiation, clouds, atmospheric chemistry, the cryosphere and biogeochemistry. In this Review, we summarize these interactions and assess the resulting impacts of dust, and of changes in dust, on global climate and climate change. The total effect of dust interactions on the global energy budget of Earth — the dust effective radiative effect — is −0.2 ± 0.5 W m−2 (90% confidence interval), suggesting that dust net cools the climate. Global dust mass loading has increased 55 ± 30% since pre-industrial times, driven largely by increases in dust from Asia and North Africa, leading to changes in the energy budget of Earth. Indeed, this increase in dust has produced a global mean effective radiative forcing of −0.07 ± 0.18 W m−2, somewhat counteracting greenhouse warming. Current climate models and climate assessments do not represent the historical increase in dust and thus omit the resulting radiative forcing, biasing climate change projections and assessments of climate sensitivity. Climate model simulations of future changes in dust diverge widely and are very uncertain. Further work is thus needed to constrain the radiative effects of dust on climate and to improve the representation of dust in climate models. Dust influences the global energy budget through various Earth system interactions. This Review outlines these interactions, revealing a total radiative effect of −0.2 ± 0.5 W m−2, which, alongside 55 ± 30% historical increases in dust, have contributed a radiative forcing of −0.07 ± 0.18 W m−2.


Introduction
Mineral dust aerosols are small rock-derived particles with diameter D <~100 μm that are suspended in the atmosphere 1,2 .Most dust is produced by the ballistic impacts of wind-driven sand grains on sparsely vegetated and dry soils 3 , which ejects and fragments aggregates of soil particles 1,4 .Owing to these mechanical impacts, dust is a relatively coarse aerosol, with most of its mass contained in the coarse (D > 2.5 µm) and super coarse (D > 10 µm) modes 5 .
Dust is produced in copious amounts in the world's deserts, loading the atmosphere with ~26 million tonnes of dust, which accounts for a large majority of the atmosphere's aerosol burden by mass 6,7 .The Sahara Desert and the Sahel contribute ~50% of global dust emissions (~2100 Tg/yr) and mass loading (~13 Tg), the Asian deserts ~40% (~2000 Tg/yr and ~10 Tg), and the North American and Southern Hemisphere deserts and high latitude regions another ~10% (~500 Tg/yr and ~3 Tg) (Fig. 1) 8,9 .Although much of the dust is deposited close to source regions, a substantial fraction is transported for thousands of kilometres.; emissions from high latitude regions are not included.Shading represents dryland classification based on the aridity index: hyper-arid regions (AI < 0.05; red shading), arid regions (0.05 < AI < 0.20; orange shading), semi-arid regions (0.20 < AI < 0.50; light brown shading), and dry sub-humid regions (0.50 < AI < 0.65; green shading) 228 .Most dust is emitted from drylands in North Africa and Asia, which are collectively known as the "dust belt" 229 .
The abundance and long-range transport of dust cause it to impact climate through various mechanisms.During transport, dust scatters and absorbs solar shortwave (SW) and terrestrial longwave (LW) radiation 6,11 , modifies cloud properties through seeding cloud droplets and ice crystals 12,13 , mixes with other aerosols 14 , and serves as a sink for radiatively important atmospheric trace gases [14][15][16][17] .Upon deposition, dust darkens snow and ice packs 18,19 , and stimulates ecosystem productivity and CO2 drawdown through the delivery of iron and phosphorus 20 .Because some of these mechanisms cool whereas others warm 6,14,21 , it is unclear whether dust exerts a net cooling or a net warming effect on global climate.Because measurements of dust deposition suggest that dust has increased since the pre-industrial era 22,23 , this uncertainty in the sign and magnitude of dust radiative effects means it is unknown whether dust changes have enhanced or opposed anthropogenic warming.
In this review, we examine the impacts of dust, and of changes in dust, on global climate and climate change.We first summarize the various mechanisms through which dust impacts Earth's radiation budget, and assess the radiative effect produced by each mechanism.We then constrain the increase in dust loading since pre-industrial times and assess the radiative perturbation produced by this historical increase in dust.We also discuss the radiative perturbation due to possible future changes in dust and end with recommendations for future research priorities.Radiative effects from dust arise through interactions with radiation, atmospheric chemistry, clouds, the cryosphere, and biogeochemistry (Fig. 2).Each of these mechanisms are now discussed.

Interactions with radiation
Perhaps the best understood mechanism by which dust impacts climate is through the dust direct radiative effect (DRE), the perturbation of Earth's energy balance by scattering and absorption of radiation (Fig. 2a).Since dust spans a wide range of sizes, from ~0.1 -100 μm 29 , it interacts with both SW (centered around 550 nm wavelength) and LW (centered around 10 μm wavelength) radiation 30,31 .
The sign and magnitude of the dust DRE depend on the balance between these interactions.For instance, scattering of SW radiation cools the climate while absorption of SW radiation warms, with an overall net cooling 6,32 .In contrast, scattering and absorption of LW radiation both warm the climate since both decrease the transparency of the atmosphere to terrestrial LW radiation 33 .Thus, the balance between cooling from SW scattering, and warming from SW absorption and LW scattering and absorption, dictate the dust DRE.
For SW radiation, the balance between scattering and absorption is influenced by dust particle size.Because absorption increases more strongly with particle size than scattering 34,35 , the singlescattering albedo (SSA; the ratio of scattered radiation to total extinguished radiation) decreases with particle size.Indeed, submicron dust has an SSA close to 1, whereas supermicron dust absorbs a substantial fraction of extinguished radiation, exhibiting SSAs of ~0.95 at D = 2 µm, ~0.80 at D = 10 µm, and even lower for super coarse dust 30,36 .
However, the exact SSA of dust aerosols depends on their complex refractive index, determined by particle mineralogy 31 .Absorption increases approximately linearly with iron oxide content, which is primarily provided by hematite and goethite 37 .Dust optical properties can also be affected by mixing with other aerosols, especially black carbon 38 .Observations suggest that this possible mixing has limited impact on the optical properties of most African dust 39,40 , but could substantially affect those of East Asian dust 41 .
Although dust particle size and mineralogy determine the balance between SW scattering and absorption, the efficiency with which both processes perturb the TOA radiative flux depends on the albedo of the underlying surface.Indeed, the cooling effect of SW scattering is enhanced if the dust is situated above dark (low albedo) surfaces like the ocean and forests that would otherwise absorb most of the radiation 42 .Conversely, the warming effect of SW absorption is enhanced if the dust is situated above clouds or above high albedo land surfaces like snow, ice, and deserts that would otherwise scatter most of the radiation back to space 35,43 .
Dust microphysical properties and mineralogy also influence the extinction of LW radiation.For example, because of its longer wavelengths, LW radiation is extinguished primarily by coarse dust 30,33,44 .The sensitivity of LW extinction to mineralogy is less important than for SW interactions owing to the smaller variability in LW optical properties between minerals, and because LW scattering and absorption both warm the planet 31,45,46 .
The efficiency with which dust extinction of LW radiation perturbs the TOA radiative flux also depends on the atmosphere's transparency to LW radiation and the elevation of the dust layer.Indeed, the TOA flux is only substantially impacted if the atmosphere is at least somewhat transparent to LW radiation, as is the case in the absence of clouds in the ~8 -13 μm 'atmospheric window' wavelength range 33,47 .Furthermore, because LW emission depends on temperature, the LW warming depends on the temperature difference between the dust layer and the source of the LW radiation -usually the surface or clouds below the dust layer.In addition, the atmosphere's transparency to LW radiation decreases with the concentration of water vapor and thus increases with height.As such, dust warming by LW extinction increases approximately linearly with the height of the dust layer 33,42,47,48 .
As a result of the uncertainties and opposing SW and LW DRE , it is unclear whether the dust DRE exerts a net cooling or warming effect.Combining the SW and LW DRE yields a net dust DRE of -0.15 ± 0.35 Wm -2 , consistent with other calculations 6,32,46,[49][50][51] (Fig. 3).As such, the dust DRE could either slightly warm or substantially cool the planet, or it could have little net impact.We assign medium confidence to this assessment because of the large body of research and availability of satellite-based constraints.

Interactions with atmospheric chemistry
Dust affects atmospheric chemistry through numerous interactions with atmospheric trace gases and aerosols.Although freshly emitted mineral dust is considered insoluble, it is reactive towards trace acidic gases derived from anthropogenic pollutants and sea salt 17,60 .Mineral dust particles collected throughout the world are notably associated with nitrate [61][62][63] .Nitric acid interacts with the non-volatile mineral cations of dust, forming salts to maintain the charge balance in the aerosol phase 64 .The uptake of such acidic vapors is very rapid due to their ability to react with carbonates and other minerals through simple acid-base chemistry 65 .Over continents, such interactions of mineral cations with anthropogenic sulfuric acid causes the accumulation of substantial amounts of sulfate on dust surfaces 66 .In contrast, over oceans, mineral cations are commonly associated with chloride derived from sea salt 67 .
Mineral dust also provides surfaces for the adsorption of inorganic (notably SO2, NO2, and O3) and organic trace gases 17 , affecting the optical properties, hygroscopicity and atmospheric residence time of both dust and anthropogenic aerosols.Therefore, dust particles provide a substantial sink for the direct removal of important atmospheric constituents like O3, affecting the oxidative capacity of the atmosphere and the ozone radiative forcing 68,69 .Dust particles also provide reaction sites for the oxidation of SO2 to sulfuric acid 70 and the formation of nitrous acid through heterogeneous reactions of NO2 71 .However, such heterogeneous formation of salts occurs at a much slower pace than through the direct uptake of acidic vapours since acid anhydrides (for example SO2) do not initially contain any acidic protons 65 .Additionally, the high pH values found on the alkaline mineral particles can promote the formation of ammonium nitrate on their surface 72,73 .
All these interactions of dust with atmospheric gases can transform the surface and even the bulk chemical composition of dust particles 74,75 .This chemical processing of dust is highly dependent on both the gas phase composition and on the dust chemical composition 64,76 , which depends on the mineralogy of the source soil 77 .
The chemical ageing of dust due to these various reactions creates a soluble coating that increases the dust particle's hydrophilicity, which in turn affects the residence time of dust and its interactions with clouds.For example, the interaction of a calcite-containing dust particle with nitric acid converts the insoluble calcium carbonate to the highly hygroscopic calcium nitrate 78 .The increased hygroscopicity of the chemically aged dust increases its water adsorption efficiency, making it grow more rapidly under humid conditions, thus causing it to form cloud droplets and extinguish radiation more efficiently.On the contrary, the increased water uptake by the large, aged dust particles can also deplete in-cloud supersaturation, thereby reducing the number of smaller anthropogenic particles that are activated and grow into cloud droplets 13 .Furthermore, chemical ageing of mineral dust can also reduce its ice nucleating ability 80 .
These heterogeneous and multiphase reactions affect the atmospheric loading of both dust and non-dust aerosols.Nitrate formation associated with the mineral cations removes nitric acid from the gas phase, decreasing the formation of ammonium nitrate aerosols.Similarly, sulfate formation on dust decreases SO2 abundance and thus the formation of sulfate aerosols.As such, dust can reduce the concentration of anthropogenic CCN both by adsorption of precursor gases and through coagulation with anthropogenic aerosols.Furthermore, the hygroscopic growth of aged dust can increase its scavenging and deposition rate, reducing its atmospheric residence time and loading 15,81 .However, modelling results suggest that these effects can enhance the total accumulation mode dust burden through a reduced loss by coagulation with coarse dust particles 15 .
The physicochemical interactions of mineral dust with atmospheric composition can thus affect the direct and indirect radiative forcing of both dust and non-dust aerosols (Fig. 2b).These effects can be both negative or positive, depending on the region and the prevailing impacts of dust on the atmospheric aerosol loading and composition 15,16 .A net cooling effect of -0.05 Wm -2 has been calculated for the effect of dust on the total aerosol DRE 15 , driven mostly by an enhanced burden of the accumulation mode dust aerosols and decreased absorption of SW radiation due to the modified aerosol composition of mineral dust.However, observations of dust during transport across the Atlantic and Mediterranean oceans indicate that the size distribution of dust with diameters less than 5 μm remains remarkably constant and that optical properties do not change appreciably 39,40,55 .Moreover, a critical effect of heterogeneous chemistry on dust surfaces is to reduce the atmospheric loading of anthropogenic aerosols, thereby decreasing their direct radiative cooling, resulting in a net warming of +0.12 to +0.20 Wm -2 16,82,83 .Overall, the impact of dust interactions with atmospheric chemistry on the aerosol DRE is highly uncertain.The resulting radiative effect is assessed at 0.10 ± 0.15 Wm -2 to encompass the possibility of the slight cooling of -0.05 Wm -2 15 as a lower bound and the larger warming found by others 16,82,83 as an upper bound.We assign very low confidence to this assessment.

Interactions with clouds
Dust particles influence clouds via multiple interactions, including changing the thermodynamic environment by absorbing SW and LW radiation and serving as cloud condensation nuclei (CCN) and ice-nucleating particles (INPs).Radiative perturbations produced by dust effects on warm clouds, mixed-phase clouds, cirrus (ice) clouds, and by semi-direct effects are discussed next.

Dust indirect effects on warm clouds
There are three main pathways through which dust particles can affect warm clouds: first, by increasing the concentration of CCN, as laboratory studies have shown that various types of (unprocessed) mineral dusts possess a modest ability to act as CCN 84,85 , which is further enhanced by atmospheric processing (ageing) of dust 86 ; , second, by reducing the concentration of non-dust CCN through coagulation and adsorption of precursor gasses; and third by acting as giant CCN, which can form cloud droplets at relatively low supersaturation and thus deplete water vapor to such an extent that overall cloud droplet formation is suppressed.The second and third pathways are both thought to reduce cloud droplet number concentrations (CDNC), and thus reduce cloud albedo and shorten cloud lifetimes.
Although some modelling results found that the pathways that decrease CDNC dominate (Fig. 2c, lower branch), amounting to a decrease of as much as -11% in the global mean CDNC 13 , most other modelling has found that dust slightly increases the global mean CDNC abundance.These latter results thus suggest that the effect of dust acting as CCN dominates (Fig. 2c, upper branch), albeit with large differences in the magnitude of the dust-induced CDNC contribution 21,87,88 .As a CDNC increase is expected to increase cloud albedo and extend cloud lifetime through well-established mechanisms 28 , a dust-induced increase in CDNC would have a net cooling effect (Fig. 2c, upper branch).For example, global simulations with the CAM5 model 21 resulted in a ~1% increase in CDNC for a 3-fold increase in dust emissions, and this CDNC increase in turn produced a negative forcing of -0.01 Wm -2 .Such an effect is indeed supported by estimates based on satellite observations 89 .
Although past work thus reached contradictory conclusions regarding the net global effects of dust on warm clouds, there is broad agreement that the sign and magnitude of the dust contribution to CDNC is highly heterogeneous in both space and time 13,21 .Given the relatively sparse research and disagreement on the sign of the global mean CDNC contribution from dust, we assess the corresponding perturbation to Earth's TOA radiation budget through changes to liquid clouds to likely be negative but close to zero, with an uncertainty range of -0.10 to +0.10 Wm -2 .This assessment is based on scaling the estimates of CCN/CDNC changes 13,21,87,88 with the forcing estimate per change in CCN/CDNCcited above 21 , and has low confidence.

Dust indirect effects on mixed-phase clouds
Although the ability of dust to act as CCN is somewhat ambiguous, their ice-nucleating ability is undisputed 90,91 .A wide variety of dust particles have been investigated in the laboratory and found to be efficient INPs both in the immersion mode (freezing cloud droplets from within) and in the deposition mode (nucleating ice through vapor depositing onto them, possibly triggered by freezing of condensed water in particle pores 92 ).The former is the ice formation mechanism thought to be of greatest relevance for mixed-phase clouds (MPCs).These are clouds with temperatures between approximately -38°C and 0°C that can consist of either supercooled liquid droplets, ice crystals, or a mixture.MPCs are generally optically thick and efficiently reflect incoming SW radiation (a cooling effect).Their optical thickness also allows them to absorb virtually all outgoing LW radiation, reducing the amount of LW radiation emitted to space (a warming effect).The former (SW) effect has been found to dominate in the global mean 93 .In an INP-limited ("pristine") environment, MPCs will be optically thick and usually have liquid cloud tops 94 , with only small amounts of ice residing in the cloud interior or below cloud base as ice crystals rapidly grow and sediment out (Fig. 2d, left schematic).In a dust-enriched environment, MPCs will be partly or completely glaciated, depending on the dust abundance and INP efficiency.This cloud glaciation results in an overall reduction of cloud albedo and thus a positive (warming) radiative effect (Fig. 2d, right schematic).An increase in dust loading, and thus INPs, therefore likely produce a warming effect on climate by reducing the cooling effect of MPCs (Fig. 2d).
Modelling results on the effects of dust on MPCs generally agree qualitatively, but differ quantitatively.Global simulations with the E3SM model 95 found that dust effects on mixedphase clouds perturb the TOA radiation budget by +0.05 to +0.26 Wm -2 .This perturbation arises from a reduction in cloud liquid water and a corresponding increase in cloud ice (Fig. 2d).For comparison, global simulations with the CAM5 model estimated that going from a very pristine state with only 10% of current dust emissions to present-day dust emissions induced a perturbation of only 0.01 to 0.10 Wm -2 through dust-INP effects on mixed-phase clouds, again by shifting cloud phase in favor of more ice 21 .However, as the atmospheric dust loading change in the latter study is smaller than in the former, these estimates are broadly consistent with each other.These modelling results are further supported by satellite observations that found that dustenriched environments tend to have MPCs with a larger proportion of ice than their counterparts in largely dust-free environments 96,97 .Thus, a perturbation to the TOA radiation budget of approximately 0.10 Wm -2 due to dust effects on MPCs is supported, but with a relatively large assessed uncertainty range of 0 to 0.20 Wm -2 and low confidence, owing to the limited body of research.

Dust indirect effects on cirrus clouds
The dominant role of dust particles in cirrus cloud formation worldwide is supported by in situ measurements, satellite observations, and numerical modelling 98,99 .Cirrus clouds are pure ice clouds residing in the upper troposphere at temperatures below approximately -38°C.These clouds have a net warming effect on climate by reducing emission of LW radiation to space more effectively than they reflect SW radiation 100 .Cirrus clouds can form by two different mechanisms: homogeneous freezing, in which small solution droplets freeze spontaneously, and heterogeneous freezing, in which ice crystals form on INPs 91 .The latter mechanism requires only modest supersaturation but can only occur when sufficient INPs are present and typically results in low concentrations of large ice crystals.The former mechanism requires much higher supersaturation but does not rely on the presence of INPs and typically results in high concentrations of small ice crystals 12 .The transition from homogeneous to heterogeneous freezing has been estimated to occur for INP concentrations between 10 and 100 L -1 . 101e impact of dust on cirrus clouds is thus highly dependent on whether non-dust INPs are present (Fig. 2e).In conditions that favor heterogeneous freezing (high INP concentration), additional dust INPs would add ice crystals and reduce their size, while in conditions that favor homogeneous freezing (low INP concentration), additional dust could reduce the number of ice crystals and increase their size by shifting nucleation from occurring homogeneously to occurring heterogeneously.The former scenario would make cirrus clouds optically thicker and extend their lifetimes, while the latter scenario would do the opposite.
The perturbation of the TOA radiation budget would naturally be opposite in the two scenarios, and at present it is unclear which one dominates globally.Thus, although global modelling results of dust impacts on cirrus clouds have in the past produced net radiative perturbations of opposite sign 12 , this difference does not signify a complete lack of process understanding, but rather indicates different assessments of which cirrus formation mechanism dominates in the absence of dust.
Research that incorporated up-to-date laboratory results of ice nucleation on dust particles 92,102,103 generally find an optical thinning of cirrus clouds due to dust (Fig. 2e, top schematic).This thinning yields large opposing perturbations to both LW and SW radiation at the TOA, but the LW effect tends to dominate, producing a net negative (cooling) perturbation 21,104 .The corresponding overall radiative effect was estimated at -0.4 Wm -2 using global simulations with the CAM5 model 104 , whereas simulations for a more moderate dust change (going from 10% to 100% of present emissions) with a modified version of the same model 21 found a range from -0.32 to +0.05 Wm -2 .We therefore assess the perturbation of the TOA radiation budget due to dust effects on cirrus clouds to -0.20 Wm -2 , with a 90% confidence interval of -0.40 to +0.10 Wm -2 .This range encompasses the strongest reported cooling effects 104 as a lower bound and the possibility of a slight warming as an upper bound.We assign low confidence to this assessment due to the limited body of research.

Dust semi-direct effects on clouds
Absorption of radiation by mineral dust can modify the temperature profile 105 , which can change atmospheric stability, the moisture profile, and secondary circulations, all of which can alter cloud distributions [106][107][108] .These processes, known as aerosol semi-direct effects (SDEs) 109,110 , were broadly described in the IPCC's Sixth Assessment Report as atmospheric adjustments to instantaneous aerosol direct radiative effects without considering effects due to changes in surface temperature 25,28 .Because dust accounts for about a third of shortwave absorption by all aerosols, the contribution of dust to SDEe is crucial to accurately quantifying the overall dust effective radiative forcing 111,112 .
The magnitude of the dust SDE, and whether it results in a positive (warming) or a negative (cooling) radiative effect, depends primarily on two factors: the relative position of the dust and cloud layers within the atmospheric column and the amount of radiation absorbed by the dust layer 106,107 .In turn, radiation absorption by dust depends on dust loading and microphysical properties, including dust mineralogical composition, shape, and size distribution 42,43,45,113 .
Understanding of the pathways through which dust semi-directly impacts different cloud regimes follows that of SDEs produced by other absorbing aerosols, like black carbon 106,114 .For lowaltitude clouds, the pathways for dust SDEs can be categorized into cases where the dust layer is above, within or near, and below the cloud layer (Fig. 2f).When dust is located above boundarylayer clouds, local heating by the above-cloud dust can stabilize the boundary layer by strengthening its capping inversion, causing an increased build-up of moisture in the boundary layer.This increased moisture increases the cloud cover, which results in a negative SDE (left schematic of Fig. 2f) 107,115 .Conversely, when dust is located within or near boundary-layer clouds, the local heating could result in reduction of relative humidity, which could evaporate the cloud and result in a positive SDE (middle schematic of Fig. 2f) 107,116 .Finally, when dust is located below boundary-layer clouds, the local heating may enhance convergence and available moisture, increasing cloud cover and resulting in a negative SDE (right schematic of Fig. 2f) 117,118 .
Radiation absorption by dust can also generate SDEs for mid and high-altitude clouds.These SDEs involve the compensation between the warming effect produced by dust absorption, which tends to decrease cloud cover, and an increase in moisture convergence, which tends to increase cloud cover 106 .Although the effect of the enhanced moisture convergence can overwhelm the warming effect, resulting in increased globally averaged high-altitude cloud cover during the summer, the overall annual-mean dust SDE is to decrease the high cloud cover 106,119,120 .This understanding of dust SDEs assumes that dust, like other absorbing aerosols such as black carbon, warms the atmospheric layer in which they are present 121 .This assumption is based on evidence that dust radiative warming due to SW absorption dominates over dust radiative cooling due to LW emission 122,123 .However, past research likely underestimated the amount of coarse dust, which emits LW radiation more strongly than fine dust 49,56 .Because accounting for the observed abundance coarse dust particles could produce substantial LW radiative cooling of the atmosphere 32,36,124 , the understanding of the different pathways through which dust can semidirectly impact clouds remains incomplete.
Because of the uncertainties in the various pathways by which dust absorption semi-directly influences cloud cover (Fig. 2f), a global observational estimate of dust SDE is not currently available.Instead, observationally based assessments have focused on dust-dominated regions 107,116,125 .For example, satellite observations show that annual dust SDE is negative (-1.2 ± 1.4 Wm -2 ) over the North Atlantic Ocean 107 .Since estimates of dust SDE show strong spatial variability and because dust SDE is driven by different dominant mechanisms for different cloud regimes over the ocean than over land 106 , scaling such observationally based regional dust SDE estimates to global values is difficult.In addition, accurate retrievals of dust microphysical properties, including dust optical properties and size distribution, are lacking from global-scale satellite and ground-based platforms 112 , making it difficult to obtain global estimates of dust SDE.
In the absence of global observational estimates, climate models simulations have reported a net positive global annual mean dust SDE 126 .These estimates vary by over an order of magnitude, between 0.01 and 0.16 Wm -2 , and depend on the climate model used 21,126,127 .These positive SDE estimates are consistent with an overall decrease in cloud cover in these simulations.Although model estimates of dust SDE and cloud changes are thus relatively consistent with each other, they could be biased because of unaccounted for uncertainties in dust absorption properties, the vertical distributions of dust and clouds, an underestimate of LW radiative cooling by coarse dust, and the parameterization of cloud processes 56,58,111 .Therefore, based on the above model simulations, the dust SDE is estimated at 0.07 ± 0.07 Wm -2 , but with low confidence due to these possible biases and limited research.

Interactions with the cryosphere
Dust interactions with the cryosphere impact climate by altering cryospheric conditions via dust direct and indirect radiative effects (Figs.2a-f) and by darkening snow and ice surfaces after deposition (Fig. 2g), which leads to a positive surface radiative effect ( Fig. 2g).This dustinduced snow albedo effect accelerates snow and glacier melting 18,128,129 , which triggers a strong, positive surface albedo feedback on the climate system 130 .The dust-induced snow albedo effect is influenced by many factors, including dust concentration in snow 131,132 , dust optical properties as determined by its size distribution and chemical composition 132,133 , dust-snow mixing state 134,135 , snow grain size and shape 134 , snowpack properties 136,137 , and illumination conditions 132,138 .
Observations indicate strong heterogeneity in dust concentrations in snow/ice.Along with different snowpack and atmospheric conditions, this variability in dust concentrations leads to large variations in the dust-induced snow albedo reduction and the associated surface radiative effects.For instance, the springtime dust-induced snow albedo effect is estimated to be less than 0.5 Wm -2 for the Arctic 139,140 , up to 5 Wm -2 for remote mid-latitude snowpacks (such as the Tibetan Plateau) 139,141 , and about 10-50 Wm -2 over polluted mid-latitude snowpacks (such as the U. S. Rocky Mountains) 18,128 .In some extremely polluted mid-latitude mountains, the local instantaneous snow albedo effect can be as high as 100-300 Wm -2 142,143 .The dust-induced snow albedo effect is typically larger in aged snow than in fresh snow 18 , because of the stronger light penetration and hence larger light absorption by dust in aged snow.Most research has focused on a few cryospheric hotspots in the Northern Hemisphere (the Rocky Mountains, Tibetan Plateau, European Alps, and the Arctic) during spring, when the dust-induced snow albedo effect is more prominent and often reaches its annual maximum.
There are only limited estimates of the global annual mean dust-induced snow albedo effect, with a central estimate of +0.013 Wm -2 and a 90% confidence interval of 0.007-0.03Wm -2 19,131,144 .Although the snow albedo radiative effect is smaller than most other dust radiative effects, it can still be more substantial regionally, particularly over polluted mid-latitude snowpacks 18 .
Estimates of the dust-induced snow albedo effect are still associated with large uncertainties due to complicated and poorly constrained dust-snowpack-radiation interactions.Variations in the poorly constrained dust-snow mixing state, snow grain shape, dust size distribution and dust chemical composition can cause up to a factor of two uncertainty in the dust-induced snow albedo effect 134,135 .Moreover, the limited knowledge of dust evolution within the snowpack -for instance due to dust scavenging by melting water and dust enrichment at the snowpack surfacealso adds to the uncertainty of the estimated snow albedo effect.Owing to the potential nonlinearity in dust-snow-radiation interactions and dust wet deposition, the dust-induced snow albedo effect may not increase linearly with dust concentration in the atmosphere or snowpack.Considering these uncertainties and the limited research, we assign low confidence to our estimate of the dust-induced snow albedo effect.

Interactions with biogeochemistry
Dust can influence ocean and land biogeochemistry, both directly through the addition of nutrients and pollutants to ecosystems, as well as indirectly through modifying precipitation, temperature, and radiation 20 .Atmospheric deposition of dust onto oceans provides iron, a limiting nutrient in high nutrient low chlorophyll (HNLC) regions 145,146 .In addition, nitrogen fixing organisms in the ocean have higher iron requirements, thereby linking iron deposition to the oceanic nitrogen cycle 147,148 .Although initial research suggested that atmospheric deposition was the dominant source of new iron 145,149 , other ocean sources also have a substantial role in the iron cycle [150][151][152] .Overall, atmospheric inputs of iron to the ocean modulate ecosystem productivity and carbon sequestration on the timescale of decades 146,153 .
, The soluble fraction of the iron is the most important for dust particles sinking through the ocean mixed layer.The deposition of soluble iron has increased since pre-industrial times, both because of the historical increase of dust over this time period and because of an increase in iron solubilization during transport due to increased anthropogenic pollution 22,154,155 .Additionally, some other important sources of soluble iron have also increased, including from wildfires and anthropogenic combustion 156,157 .The resulting alleviation of iron limitation has increased ecosystem productivity, which in turn has reduced the atmospheric concentration of carbon dioxide and its radiative forcing (Fig. 2h).
Several ocean biogeochemical models include iron and its coupling to the nitrogen cycle and can therefore estimate the reduction of CO2 concentrations due to the alleviation of iron limitation 158,159 .These models suggest that the increased deposition of soluble iron over the 20th century resulted in the uptake of ~4 ppm of CO2, producing a radiative perturbation of -0.07 ± 0.07 Wm -2 20,160 .Because approximately half of this increase in soluble iron was estimated to be due to a simulated ~40% increase in dust over the 20th century, these results imply a radiative effect due to dust-biogeochemistry interactions of -0.12 ± 0.12 Wm -2 (Eq.1).Confidence in this assessment is very low, as it is based on only one study.Note that the radiative effect due to dust-biogeochemistry interactions differ from that due to other interactions in that its effect increases over time.Consequently, the radiative perturbation that it produces depends on the timescale.
Dust also contains phosphorus, a limiting nutrient in some tropical forests and grasslands 161,162 , as well as in some ocean ecosystems 146,163 .For example, phosphorus from long-range transported North African dust may help maintain the productivity of the Amazon rainforest 164 .However, because inputs from atmospheric deposition of desert dust are thought to be important in the Amazon on millennial time scales 165 any contribution of changes in this phosphorus input probably produces a negligible contribution to dust radiative forcing since pre-industrial times.Dust also serves as a ballast, enhancing the downward transfer of organic material within the ocean, but there is not yet a quantitative estimate of the impacts in terms of productivity or carbon uptake feedback from this process 170,171 .In addition, desert dust could include elements that can be toxic to ocean or land ecosystems, such as Cu, although current estimates suggest that this effect is not important to Earth's radiation budget 172 .

The dust effective radiative effect
To determine the climatic impact of past and future changes in atmospheric dust, it is critical to assess the dust effective radiative effect R (Eq. 2), which equals the sum of the various radiative effects generated by dust (Fig. 3).Many of these radiative effects oppose one another, resulting in a median estimate of  = -0.2Wm -2 , with a wide 90% confidence interval of -0.7 to +0.3 Wm - 2 .(Note that we neglected some rapid adjustments in assessing , such as responses by water vapor and the lapse rate to dust direct radiative effects, but these adjustments are likely small 173 .)As such, the net effect of dust on Earth's global radiation budget could be negligible, a substantial net cooling, or a small net warming.
On regional scales and for different seasons, the dust effective radiative effect can differ substantially from its global and annual mean in Figure 3.This regional and seasonal variability occurs because the various radiative effects are sensitive to the spatiotemporal variability in dust concentration, microphysical properties (mineralogy and size distribution), and environmental conditions (surface albedo and cloud cover).For instance, dust over reflective deserts likely produces substantial warming because of the high dust concentration, coarse size distribution 124 , and because reflective surfaces reduce cooling produced by SW scattering and enhance warming produced by SW absorption 35,174 .Similarly, dust likely produces net warming over snow and icecovered regions because the high surface albedo enhances warming produced by dust absorption of SW radiation and because dust deposition decreases the surface albedo 42,131 .In contrast, dust over oceans usually produces cooling because dust is finer further from source regions and because the ocean albedo is only ~0.1 in the visible spectrum 175 .To determine the climate impacts of dust, it is thus critical not only to constrain the global mean dust effective radiative effect but also to constrain its spatiotemporal pattern.The sum of all radiative effects equals the dust effective radiative effect R (Eq. 2) and the portion of that dust effective radiative effect that is due to the increase in dust since pre-industrial times is the effective radiative forcing Δ p→m (Eq.4).Error bars denote the 90% confidence range.The column on the right denotes the level of scientific understanding (LOSU), or confidence in the assessment of each radiative effect, following past practice 27 .The global mean dust effective radiative effect and radiative forcing of dust are uncertain in sign and magnitude, but are more likely to cool than to warm the climate.

Dust radiative forcing
Because dust produces a potentially large effective radiative effect, a change in atmospheric dust loading since pre-industrial times could have produced a substantial effective radiative forcing.Dust loading could have changed due to both climate change and widespread human land use changes (Box 1).Knowledge of the change in dust loading from pre-industrial to modern times depends largely on dust deposition records that resolve both the modern and the pre-industrial climate.Many of these deposition records show increases in dust deposition between modern and pre-industrial times, sometimes by a factor of ~423,169,176,177 .

Dust reconstruction
We reconstructed the evolution of the global dust mass loading since pre-industrial times by combining 19 dust deposition records 23,169,176,177 with constraints on the source regions providing the deposition flux to each deposition core 8,44 (see Supplementary Information).This dust reconstruction used a bootstrap resampling method to propagate uncertainties in both the experimental deposition records and the constraints on source region-resolved deposition fluxes to each deposition site; nonetheless, errors should be interpreted as a lower bound.The atmospheric loading of dust with a volume-equivalent diameter less than 20 μm has increased from 19 ± 6 Tg in the pre-industrial period (defined here as 1841-1860) to 29 ± 8 Tg in the modern climate (1981-2000).As such, global dust mass loading has increased by 55 ± 30 % (Fig. 4a).Although substantial, this increase is less than the doubling of dust suggested by previous research 22,23 .A large contributor to this increase has been Asian dust, which has increased by 74 ± 37 % from 8 ± 3 Tg in pre-industrial times to 13 ± 5 Tg in modern times (Fig. 4c).North African dust has increased less, from 10 ± 4 Tg in the pre-industrial period to 14 ± 4 Tg in the modern climate, representing a 46 (2 -102) % increase.Both African and Asian dust mass loading peaked in the 1980s and then decreased substantially, consistent with changes observed from long-term dust concentration measurements, visibility records, and satellite observations [178][179][180][181][182][183][184] .Dust has likely also increased in the Southern Hemisphere, from 1.2 (0.6-2.2) Tg to 1.6 (0.8-2.2) Tg, representing a 27 (-14 to 88) % increase (Fig. 4d).Satellite observations suggest that global dust mass loading has been relatively stable since the year 2000, the end point of the analysis, with some notable regional trends, such as in Central and East Asia 185 .This large historical increase in dust mass loading is inadequately accounted for in current climate models and climate assessments.In fact, twelve climate models with prognostic dust cycles in the Coupled Model Intercomparison Project phase 6 (CMIP6) model ensemble 186,187 show little change (2 ± 11%) in dust mass loading since pre-industrial times (Fig. 5).This failure of models to reproduce the historical dust increase could be due to several reasons (Box 1).If the dust increase has been largely driven by natural and anthropogenic climate changes, then the model failure could be either due to an inaccurate representation of these changes in models or because modelled dust emissions are not sufficiently sensitive to changes in climate.This latter possibility is suggested by the common use in climate models of empirical dust source functions to parameterize the spatial distribution of dust emissions 188,189 .Because dust source functions are static, they mask physical links between changeable surface properties and dust emissions.As such, their use can cause models to underestimate the sensitivity of dust emissions to changes in climate 190 .Conversely, if the dust increase has been largely driven by human land use changes (Box 1), as suggested by research indicating that approximately a quarter of current dust emissions originate from regions heavily impacted by human land use 191 (Box 1), then the model failure to reproduce the dust increase could be caused by an underestimation of land use and land cover changes in drylands and the resulting increases in dust emissions.(thin colored lines).CMIP6 data are 10-year running means from historical runs 186 .Grey shading denotes the 90% confidence interval for the dust reconstruction.All models and the ensemble mean (dashed black line) fail to reproduce the large historical increase in dust loading.
Box 1. Drivers of the historical increase in dust loading.The large historical increase in dust observed in deposition records and the reconstruction of dust mass loading (Fig. 4) can be either due to human land use changes or due to natural and anthropogenic changes in climate 230 .
The observational record shows that dust is highly sensitive to climate.Indeed, dust records in some regions show a variation of a factor of ~2-4 due to climate variability over the 20th century 178,180,183 and dust has increased by a factor of ~2-4 in transitions between interglacial and glacial periods 231,232 .As such, changes in aridity, vegetation cover, and wind speed due to natural climate variability could have driven (part of) the long-term increase in dust loading, as has been suggested for North Africa 178,180,233 .In addition, anthropogenic changes to climate and atmospheric composition could also have affected dust loading, both by increasing aridity and by higher CO2 concentrations fertilizing plants at desert margins 234 , with the net effect on desert extent and dust emissions still unclear 235 .
Human land use changes could also have increased dust emissions.The Industrial Revolution and the rise of industrialized agriculture have resulted in a dramatic increase in the area of land used by humans: the fraction of the ice-free land area used for agriculture has quadrupled from ~9% in 1850 to ~35% in 2000 236 .This large-scale conversion of wildlands to agricultural land has included many semi-arid and arid regions (Fig. 1), for which human land use changes can result in dramatic increases in dust emission [237][238][239] .Additionally, anthropogenic changes in water management that result in the drying of inland bodies of water might also have substantially increased dust emissions, such as has occurred for Owen's Lake in California in the early 20th century 240 and more recently for the Aral Sea in Central Asia 241,242 .
Modelling has been unable to determine whether the historical increase in dust, which models have been unable to reproduce 22 (Fig. 5), has been primarily driven by climate or land use changes.Indeed, past research has diverged on the fraction of the global dust burden in the current climate emitted from anthropogenically disturbed sources, with results ranging from as little as 0% to as much as 50% 199,207,230,[243][244][245] .Similarly, modelling results on effects of changes in climate and CO2 concentrations on dust loading also differ, with results varying between a decrease of -20% and an increase of +60% in dust loading 199,230,235 .
Although large uncertainties thus remain in how climate and land use changes have contributed to the historical increase in dust loading (Fig. 4), two observational findings suggest that anthropogenic land use change has been a key driver of the long-term increase in dust loading (Fig. 4).First, the timing of increases in dust deposition in various deposition records appears to coincide with the rise of industrialized agriculture in source regions 23 .And second, satellite observations suggest that ~25% of modern dust emissions originate from regions heavily impacted by human land use 191 .This finding implies that human land use changes have increased dust mass loading by ~33% since pre-industrial times, which accounts for the majority of the 56 ± 29% increase in dust mass loading since pre-industrial times (Fig. 4).Moreover, satellite observations indicate that the fraction of dust emitted from anthropogenically disturbed surfaces is substantially higher for Asian than for North African source regions, which is qualitatively consistent with the finding of a larger historical increase of Asian than of North African dust (Figs.4b-d).Nonetheless, substantial additional work is needed to determine the exact causes of the historical increase in dust for each of the world's main dust source regions.

Radiative forcing due to dust increase
The historical increase in dust loading could have produced a substantial radiative forcing.Combining  = -0.2± 0.5 Wm -2 with the 55 ± 30% historical dust loading increase yields a dust effective radiative forcing from 1841 to 2000 of Δ p→m = -0.07± 0.18 Wm -2 .Dust radiative forcing could thus either have substantially contributed to, or slightly opposed, the total aerosol effective radiative forcing of -1.1 (-1.7 to -0.4) Wm -2 for the period of 1750 to 2019 28 .
Note that the calculations of R and Δ p→m are subject to important limitations.First, these calculations assume that radiative effects increase linearly with aerosol loading 6,192 (Eqs.3 and 4).However, the increase of radiative effects with aerosol loading is usually less-than-linear, especially for interactions with clouds and biogeochemistry 20,21,26 .Moreover, the radiative effects of dust vary in space, such that Δ p→m depends on the spatial pattern of dust increases, which the simple calculation here does not account for.For instance, Asian dust likely has an outsize impact on Northern Hemisphere cirrus clouds 98 and high latitude dust emissions are likely important in controlling the glaciation of mixed-phase clouds 193,194 but are not explicitly included in the dust reconstruction.Careful simulations with coupled climate models that reproduce the historical dust increase are thus needed to better constrain dust radiative forcing.
Because current climate models do not reproduce the historical dust increase, these models omit the potentially important radiative forcing due to increased dust interactions with radiation, clouds, atmospheric chemistry and the cryosphere.(Note that changes in CO2 and other greenhouse gases due to dust interactions with biogeochemistry are inherently included in climate model runs forced by observed greenhouse gas concentrations.)Dust radiative forcing was thus not accounted for in constraints on the total aerosol effective radiative forcing in the IPCC Sixth Assessment Report 28 .Because constraints on climate sensitivity depend strongly on the aerosol radiative forcing since pre-industrial times 195 , the failure by models and climate assessments to account for the historical increase in dust could thus have biased constraints on climate sensitivity and projections of future climate changes 196 .

Future changes in dust radiative forcing
Future changes in dust radiative forcing are likely to be dominated by changes in atmospheric dust loading, which in turn will be determined by several factors.One important factor will be future changes in soil moisture since drier soils are more susceptible to aeolian erosion because of reduced soil cohesive forces and less vegetation 1,2 .In CMIP5 and CMIP6 models, changes in precipitation are the main driver of soil moisture changes, yet there is a wide degree of divergence in model projections of precipitation 197 .Models do consistently show that as the planet warms the evaporative demand over land increases 198 , which by itself would reduce soil moisture.However, the effects of reduced soil moisture may be countered by CO2 fertilization, which reduces plant water losses.This could reduce dust emissions by driving an expansion of vegetation into arid regions 199 , although the magnitude of this effect is uncertain 200 .Terrestrial stilling, the observed downward trend in surface wind speeds over land surfaces 201 , could also affect dust emissions, with models suggesting a future reduction in Northern Hemisphere winds .However, changes in atmospheric circulation patterns thought to impact surface wind speeds over dust producing regions may be more important 180 .Another consequence of planetary warming is an increase in precipitation variability 203 , and thus extreme rainfall events 204 , potentially increasing future sediment supply-and aeolian erosion-via alluvial and fluvial recharge 205 .Finally, future climate and land use changes could drive a decline in biological soil crusts that reduce dust emissions, which is a mechanism for increasing emissions that is not accounted for in current models 206 .
Model estimates of future changes in dust are sensitive to methodology 207 and span the range of an increase in dust due to increasing aridity 208 to a decrease due to CO2 fertilization driving an expansion of vegetation into arid regions 199,209 .Starting with CMIP5, simulations from some models included either prescribed dust emissions or fully interactive dust.However, both regional 189,[210][211][212] and global 213 analyses of these models found that the dust mean state had substantial biases, that CMIP5 models did not reproduce historical dust variability, and that modelled dust emissions were insufficiently sensitive to changes in surface conditions.An analysis of dust changes over land in RCP 8.5 simulations, for which CO2 emissions continue unabated throughout the 21 st century, showed no secular trends in global dust 213 .An analysis of CMIP6 simulations demonstrated that many of these previously identified model deficiencies also exist in these newer climate model simulations (Fig. 5) and that the inter-model differences in dust are also growing relative to earlier CMIP efforts, suggesting that as model complexity increases so does model divergence in future projections of dust 214 .
Given the inability of models to reproduce historical dust changes and the large spread in model projections of future dust change, it is not surprising that estimates of the change in dust radiative forcing per degree planetary warming, the so-called dust-climate feedback (units Wm −2 K −1 ), is similarly uncertain.An analysis of the output from 6 CMIP6 models that participated in an aerosol intercomparison project found that these models differed in the sign of the dust-climate feedback 187 , with a multimodel mean feedback of 0.0026 ± 0.0048 Wm −2 K −1 .Other research has speculated that a key driver of the model inconsistencies was the simulation of surface winds 187 , which in turn may be related to the relatively coarse resolution of a typical climate model 215 .These results from CMIP6 are consistent with earlier research that estimated a multimodel mean feedback for CMIP5 models that was not statistically different from zero 216 .However, using a dust emission scheme that responded more realistically to changes in climate 217 enhanced the dust climate feedback due to changes in the dust direct radiative effect by an order of magnitude, yielding a range of -0.04 to +0.02 Wm −2 K −1 .On a regional scale, the dust climate feedback close to source regions is likely an additional order of magnitude larger 216 .
Given the lack of confidence in model projections of future changes in the dust burden, and the substantial uncertainties in dust direct and indirect radiative effects, there is a low degree of confidence in the ability of models to predict future changes in the dust radiative forcing.

Summary & Future Perspectives
We assessed the global mean effective radiative effect of dust in the modern climate at  = -0.2± 0.5 Wm -2 (Fig. 3).Despite the considerable uncertainty in the sign and magnitude of , which arises from the numerous uncertain and sometimes opposing mechanisms through which dust impacts climate, it is more likely that dust cools than that it warms global climate.We further found that global dust loading in the modern climate is 55 ± 30% higher than it was in preindustrial times (Fig. 4), which has exerted a global mean effective radiative forcing of Δ p→m = -0.07± 0.18 Wm -2 .The historical increase in dust has thus likely somewhat counteracted greenhouse warming.
Current climate models fail to capture the historical increase in dust loading (Fig. 5) and thus inadequately account for dust radiative forcing, which could have caused biases in assessments of climate sensitivity and projections of future climate changes 195,196 .Substantial additional research is thus needed both to better constrain  and Δ p→m and to enable climate models to reproduce the historical increase in dust.
The dust direct radiative effect (DRE) contributes most to the uncertainty in  and Δ p→m (Fig. 3).Future research should focus on reducing its uncertainty by better constraining dust optical properties through in situ and remote sensing observations.For instance, the information on soil mineralogy to be provided by NASA's 2022 Earth Surface Mineral Dust Source Investigation (EMIT) mission could help constrain dust optical properties 218 .Additionally, models likely greatly underestimate the atmospheric concentration of super coarse dust 49,55,56,124 , which warms by absorbing SW and LW radiation.This should be addressed by obtaining more measurements of emitted and transported dust that extend to the difficult-to-measure super coarse dust size range 29,36,54,55 , and by developing improved parameterizations of super coarse dust emission 220 and deposition and implementing those in climate models.
Another priority for future research should be better constraining the radiative effects of dust due to interactions with clouds, anthropogenic aerosols, and biogeochemistry, which together contribute the remaining uncertainty in  (Fig. 3).Because of a dearth of observational constraints, our assessment of these radiative effects was mostly based on modelling studies.However, models struggle to correctly account for interactions of dust with clouds and anthropogenic aerosols, in part because of the mismatch in scales between the small scales at which the relevant processes occur and the large scales of climate model grid boxes 38,219,221 .As such, there is an urgent need for comprehensive in situ and satellite observations to constrain these interactions 12,193 .For instance, more satellite and in situ observations of cirrus interactions with dust and other INPs 98,99 could elucidate the relative importance of homogeneous and heterogeneous nucleation of ice crystals, which determines the sign of the radiative effect of dust interactions with cirrus (Fig. 2e) 12 .Furthermore, dust radiative effects due to interactions with clouds could be better constrained with future model simulations at a sufficiently high (kilometre-scale 221 ) resolution to resolve the critical sub-grid scale turbulence and cloud processes that currently must be parameterized in models 219 .Finally, constraining radiative effects due to dust interactions with biogeochemistry requires an improved characterization of dust composition and how this evolves during transport, as well as accurate knowledge of which land and ocean regions are nutrient limited 153 .
We also recommend that the community conducts multi-model experiments to obtain more robust estimates of the various dust radiative effects and of  and Δ p→m .These experiments should also investigate the uncertainty in radiative effects that result from model differences in dust optical properties, size distribution, model resolution, meteorology, the spatiotemporal distribution of dust emission fluxes, and parameterizations for dust deposition and dust interactions with clouds, radiation, atmospheric chemistry, the cryosphere, biogeochemistry, and other aerosols.Such multi-model experiments could be done in the context of the Aerosol Model Intercomparison project (AeroCom), which has previously performed multi-model experiments for anthropogenic aerosols 25,222 .
Future research should also prioritize addressing the failure of models to reproduce the historical increase in dust (Fig. 5).Doing so requires an improved understanding of the factors driving changes in the atmospheric dust loading since pre-industrial times, including the relative roles of changes in land use, wind speed, soil properties, sediment supply, and vegetation cover 180,223 .Additionally, new observations and modelling are needed to clarify the meteorological processes that generate the high wind speeds that produce dust, such as cold pool outflows from moist convection 215,224,225 , and to improve the representation of those processes in climate models.Finally, more physically based dust emission schemes need to be developed and implemented into climate models.These schemes should explicitly account for dust from high latitudes, which have an outsize effect on climate through interactions with clouds 193,194 .Furthermore, dust emission schemes should avoid using empirical dust source functions as these do not respond to changes in climate; instead, emission schemes should use process understanding to account for the dependence of the spatiotemporal pattern and mineralogical composition of dust emissions on wind, soil properties, sediment supply, and vegetation coverage 190,226,227 .A challenge will be to achieve this without making these schemes too sensitive to parameters such as soil moisture that non-linearly increase dust emissions 1,2 and that have considerable variability in climate models 197 .These fundamental improvements in dust emission schemes are also needed for meaningful predictions of future changes in dust and for more accurate predictions of dust impacts on regional climate.

Figure 1 .
Figure 1.Main sources and sinks of dust in the global dust cycle.Emission fluxes (blue bars) from the world's main dust source regions and deposition fluxes (orange arrows) to regions where dust can impact surface albedo or biogeochemistry.Fluxes are for dust with geometric (volume-equivalent) diameter up to 20 µm and are based on constraints for 2004-200844 ; emissions from high latitude regions are not included.Shading represents dryland classification based on the aridity index: hyper-arid regions (AI < 0.05; red shading), arid regions (0.05 < AI < 0.20; orange shading), semi-arid regions (0.20 < AI < 0.50; light brown shading), and dry sub-humid regions (0.50 < AI < 0.65; green shading)228 .Most dust is emitted from drylands in North Africa and Asia, which are collectively known as the "dust belt"229 .

Figure 2 .
Figure 2. Mechanisms through which dust impacts climate.a | dust direct interactions with shortwave (SW) and longwave (LW) radiation.b | dust interactions with atmospheric chemistry and the induced perturbations to the radiative fluxes at the top-of-atmosphere exerting a warming (left) or cooling (right) effect on global climate.The brown core represents the freshly emitted insoluble dust particle and the surrounding blue area represents the acquired soluble coating through interactions with atmospheric chemistry.c | dust indirect effects on warm clouds occur by dust increasing cloud albedo through adding to CCN and increasing CDNC (upper branch) and by dust decreasing cloud albedo by reducing non-dust CCN through enhanced particle coagulation and adsorption of precursor gases and by dust giant CCN reducing in-cloud supersaturation (lower branch).d | dust indirect effects on mixed-phase clouds (MPCs), illustrated by MPC formation in pristine (INP limited; left) and dust-enriched (right) environments.e | dust indirect effects on cirrus clouds, separated by the dominant ice crystal formation mechanism in the absence of dust.f | dust semi-direct effects on low clouds due to local heating generated by dust absorption, separated by location of dust relative to clouds.g | radiative effects of dust deposited on snow and ice, illustrated by snow reflectivity without (left) and with (right) dust deposited onto the snowpack.h | effect of dust on CO2 concentrations due to interactions with ocean biogeochemistry.Yellow arrows represent SW radiation and red arrows represent LW radiation.

Figure 3 .
Figure 3.The global mean effective radiative effect and radiative forcing of dust at the top-ofatmosphere.Perturbations to Earth's radiation budget by dust through direct radiative effects, dustmediated cloud radiative effects, and various other radiative effects.The sum of all radiative effects equals the dust effective radiative effect R (Eq. 2) and the portion of that dust effective radiative effect

Figure 4 .
Figure 4. Atmospheric dust mass loading changes since pre-industrial times.a| Reconstructed globally integrated dust mass loading.b| as in a, but for loading contributed by dust from North Africa.c| as in a, but for loading contributed by dust from Asia.d| as in a, but for loading contributed by dust from the Southern Hemisphere.The solid line denotes the median dust loading estimate, the shading the 90% confidence range, and the dotted line the average pre-industrial (1841-1860) dust loading.Dust loadings were obtained by combining 19 records of dust deposition with constraints on the spatially resolved dust deposition fluxes produced by the world's main dust source regions8,44 ; see Supplement for details.Dust has increased in all three regions, translating to a 55 ± 30 % rise in global dust mass loading in modern times(1981-2000) compared to pre-industrial.

Figure 5 .
Figure 5. Climate model representations of historical changes in dust loading.Changes in global dust loading relative to the period 1841-1860 obtained from the dust reconstruction (solid black line) and simulated by 12 CMIP6 climate models with prognostic dust aerosol cycles 214