Natural Aerosols in the Global Atmospheredust.ess.uci.edu/ppr/phd_gri04.pdf · Natural aerosols in the global atmosphere ... I have been happy to go to work and I have ... 2.5 Evaluation

Natural Aerosols in the Global Atmosphere

Alf Grini

Department of Geosciences

University of Oslo

[email protected]

Ph.D thesis defended in a public dissertation Feb. 11. 2004

1

Nar det kjem til stykketAr ut og ar inn har du site bøygd yver bøkene,du har samla deg meir kunnskapenn du treng til ni liv.Nar det kjem til stykket, er detso lite som skal til, og det veslehar hjarta alltid visst.I Egypt hadde guden for lærdomhovud som ei ape.

Olav H. Hauge

Natural aerosols in the global atmosphere

Alf Grini

Department of Geosciences, University of Oslo

This thesis is based on the following five papers:

1. Grini, A., G. Myhre., J.K. Sundet and I.S.A. Isaksen:. Modeling the annual cycle ofsea salt in the global 3D model Oslo CTM2: Concentrations, fluxes and Radiativeimpact, J. Climate, 2002, vol 15, pp 1717

2. Myhre, G., A. Grini, J.M Haywood, F. Stordal, B. Chatenet, D. Tanre, J.K. Sundetand I.S.A. Isaksen: Modeling the radiative impact of mineral dust during the SaharanDust Experiment (SHADE) campaign, J. Geophys. Res., 2003, vol 108, no. D18, pp8579

3. Grini, A., C.S. Zender and P. Colarco: Saltation Sandblasting behavior during mineraldust aerosol production, Geophys. Res. Lett., 2002, vol 29, no. 18, pp 1868

4. Grini A. and C.S. Zender: Roles of saltation, sandblasting, and wind speed variabilityon mineral dust aerosol size distribution during the Puerto Rican Dust Experiment(PRIDE), J. Geophys. Res., in press, 2004.

5. Grini, A., G. Myhre, C.S. Zender, J.K. Sundet and I.S.A. Isaksen: Model simulationsof dust sources and transport in the global troposphere. Effects of soil erodibilityand wind speed variability Institute report series, no 124, Department of Geosciences,University of Oslo, Oslo, Norway, 2003, ISBN 82-91885-26-5.

Acknowledgments

I would like to thank Ivar S.A. Isaksen for giving me the opportunity to do this thesis.Ivar has been able to combine scientific advising with administration of a large researchgroup which consists of 15 scientists.

The research group form a strong collective where it is easy to find persons with whomI can discuss scientific problems. In particular, I would like to mention discussions withGunnar Myhre and Jostein K. Sundet which have been helpful during my work. The socialaspect of the group can not be underestimated. I have been happy to go to work and I havebeen looking forward to lunch- and coffee breaks every day since I started.

During my thesis I have had the opportunity to work with Charles S. Zender. It is agreat motivation to work with Charlie. He always has good ideas and is able to formulatethem in a concise, scientific way. In addition to being a help in science, Charlie has been agreat help in practical work, teaching me how to develop software in teams using cvs. He is,of course, also a nice guy.

I am grateful to other people I have had the pleasure to work with for various reasons(David Newman, Huisheng Bian, Peter Colarco)

I would also like to thank the people who (by hazard) made me start doing atmosphericmodeling. The two persons who introduced me to atmospheric sciences were Hugo A. Jakob-sen and Leonor Tarrason.

I am grateful to my family and friends for their support.

Last (but not least) I would like to thank Cathy for being my girlfriend and for keepingme happy.

Contents

1 Aerosols and earth’s climate 11.1 General background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Importance of aerosols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2.1 What are aerosols? . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.2 Aerosol direct effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2.3 Aerosol indirect effect . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2.4 Aerosols and biogeochemistry . . . . . . . . . . . . . . . . . . . . . . 31.2.5 Aerosol/photochemistry interactions . . . . . . . . . . . . . . . . . . 3

2 Modeling studies of atmospheric aerosols 42.1 Detailed aerosol models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 Simpler and global models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 Models for atmospheric dust . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.3.1 Dust production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3.2 Global dust models . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.3.3 Anthropogenic dust? . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.4 Models for atmospheric sea salt . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.1 Sea salt production . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.2 Global sea salt models . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.5 Evaluation of models used in this thesis . . . . . . . . . . . . . . . . . . . . . 8

3 Future work 83.1 Aerosol process studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93.2 Combined use of Aerosol modeling and satellite retrievals . . . . . . . . . . . 9

4 Summary of papers 104.1 Paper 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104.2 Paper 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104.3 Paper 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104.4 Paper 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114.5 Paper 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1 AEROSOLS AND EARTH’S CLIMATE 1

1 Aerosols and earth’s climate

1.1 General background

The earth’s atmosphere is constantly heated and cooled. In a simple view, the troposphere isheated by convection with release of latent heat and cooled by long-wave radiation emission.This balance results in the temperature gradient we observe in the troposphere. This con-ceptual model was successfully implemented in simple 1-D atmospheric radiative-convectivemodels [Manabe and Strickler , 1964]. Today, these simple 1-D models are extended to ad-vanced 3-D climate models (and earth system models) which can be used to simulate theclimate [Boville and Gent , 1998]. The advanced models predict many aspects of the earth’satmosphere and the climate system.

Human influence on earth’s climate is often discussed in terms of the influence of CO2.The importance of CO2 as a climate gas was already acknowledged by Arrhenius [1896] whoconcluded that a doubling of CO2 mixing ratio would on average increase global temperatureby 5 K. More recently, advanced numerical models estimate an increase of 3 K [Hansen et al.,2002]. So one can maybe ask if changing this estimate from 5 to 3 K is the total advancementof all climate research science the last 100 years. Luckily it is not.

Throughout the 20th century, it has been acknowledged that humans can influence theatmospheric content of other greenhouse gases such as CH4 and O3. Increasing green housegases leads to an increase in surface temperatures. Humans can also influence the atmo-spheric radiative balance through changing e.g. surface albedo, or atmospheric aerosol con-

tent. The last 30 years, radiative forcing due to aerosols have received much attention inthe scientific community. A large part of the key to understanding climate change, lies inunderstanding the atmospheric radiative balance. However, climate change is not only globalaverage heating or cooling, but also e.g. changed precipitation patterns, circulation changes(both in atmosphere and ocean) and ecosystem changes [Houghton et al., 2001].

1.2 Importance of aerosols

1.2.1 What are aerosols?

Aerosols are airborne particles which are some nm to several µm in size. The smallestaerosols (smaller than 3 nm) could also be defined as thermodynamically stable clusters ofgas molecules [Kulmala et al., 2003]. The largest aerosols are cloud droplet (larger thanabout 10 µm). The limits proposed here (3 nm and 10 µm) are not absolute definitions,and the use of “molecular cluster”, “aerosol” and “cloud droplet” varies around these limitsbetween different authors.

The aerosols are lost from the atmosphere by dry deposition or wet deposition (by rain).An aerosol can also be lost from the atmosphere by coagulation (that is, it collides with, andsticks to another aerosol). In that case, aerosol mass is not lost from the atmosphere, butaerosol number is.

Aerosols can be formed in several ways. Either by formation of gaseous clusters (nu-cleation) which grow by condensation until they become atmospheric aerosols, or by directemission of particulate matter. Both these mechanisms can occur in a natural way (emissionsof dimethyl sulfoxide (DMS) from oceans, condensable biogenic gases from vegetation, sea


salt from oceans, dust from deserts) or in an anthropogenic way (e.g. emissions from cars orindustry). Aerosols are not only a phenomenon caused by man made pollution. However,humans can change the aerosol content of the atmosphere.

In the atmosphere, most aerosols are mixtures of organic and inorganic compounds. Inglobal studies, it is common to say that aerosols are sodium, chloride, nitrate, ammonium,

sulfate, organic species, pure carbon or dust. The organic species are in reality mixtures ofseveral thousands of organic compounds.

The two most important natural aerosol types (by mass) in the atmosphere are sea saltand dust aerosols. These two aerosol types are the focus of this thesis. If it is possible toestimate the atmospheric content of natural aerosols, it would help in estimating what partof the aerosol loading is due to anthropogenic impacts.

Aerosols affect the climate in several ways, some of which are explained in detail below.

1.2.2 Aerosol direct effect

The aerosol direct effect means aerosols interacting with radiation causing heating or coolingin the atmosphere. The aerosols can reflect or absorb solar radiation or absorb and re-emitlong wave radiation.

In the beginning of the 1990’s much focus was drawn to the direct (cooling) effect ofsulfate aerosols Charlson et al. [1991]. The direct effect is not only limited to sulfate. Severalstudies have addressed the aerosol direct effect of any type of aerosol.

Haywood et al. [1999] showed that when comparing fluxes of reflected short wave radiationfrom Earth Radiative Budget Experiment (ERBE) with fluxes from a General CirculationModel (GCM), the fluxes did not match each other unless realistic aerosol concentrationswere included in the GCM. This result definitively points out the importance of aerosols inthe earth’s radiative balance.

Houghton et al. [2001] estimate the anthropogenic direct effect of aerosols to be a coolingeffect of 0.4 W/m2 due to sulfate, a cooling effect of 0.2 W/m2 due to biomass burningaerosols, a cooling effect of 0.1 W/m2 due to fossil fuel organic carbon aerosols and a heatingeffect of 0.2 W/m2 due to fossil fuel black carbon aerosol. The anthropogenic forcing due tomineral dust is not estimated, but it is proposed that anthropogenic dust can contribute toa cooling or heating which in absolute value is not larger than 0.5 W/m2. The reasons forthese uncertainties are discussed in Myhre and Stordal [2001].

For comparison, the radiative forcing due to CO2 increase from 1750 to 2000 is estimatedin Houghton et al. [2001] to be 1.46 W/m2.

1.2.3 Aerosol indirect effect

Clouds consist of small water droplets. These droplets were originally aerosols which startedto grow by condensation. Aerosols able to become cloud droplets are called Cloud Conden-sation Nuclei (CCN). The lifetime of a cloud droplet ends either when it is big enough tofall out as rain, when it evaporates or when it collides and sticks to another cloud droplet.

Aerosols affect clouds in two ways:

1. More aerosols can lead to more cloud droplets. This means that the average clouddroplet will be smaller since the cloud water is shared between more drops. Smaller


cloud droplets scatter sunlight more efficiently. If humans increase the number of clouddroplets so that clouds in general scatter more sunlight, this will have a cooling effect[Warner and Twomey , 1967; Twomey , 1977]

2. More, and smaller cloud droplets will, in theory, lead to a longer lifetime of the clouddroplets. That is, it takes longer time for cloud droplets to grow to raindrop size.If humans change the cloud droplets so that the cloud lifetime goes up, this will onaverage lead to more reflected short wave radiation and a global cooling effect [Albrecht ,1989]

It is amazing to notice (today when global warming is observed) that S. Twomey in 1971submitted an abstract for a talk at the “International conference on weather modification”containing “If artificial pollution is a factor, then as pollution continues to increase, the

albedo of clouds and hence the planetary albedo must increase -i.e., the short wave energy

reaching lower levels will be reduced. If no compensating effect intervened, such a change

would be sufficient, according to the calculations of Budyko and others, to begin another ice

age.” [Twomey , 1971]In Houghton et al. [2001], indirect effects of aerosols were estimated (with large uncer-

tainty) to a cooling effect of 0 to 2 W/m2. Several estimates of the indirect effect use aerosolmass as an indication to how many CCN is formed (e.g. Menon et al. [2002]). In reality,what needs to be calculated is the number of aerosols which can form cloud droplets. Ghan

et al. [1998] and Rosenfeld et al. [2002] showed that anthropogenic size distribution change

or hygroscopic properties change are properties which need to be calculated to estimate theindirect effect.

1.2.4 Aerosols and biogeochemistry

Charlson et al. [1987] introduced the “CLAW” hypothesis speculating on links betweenaerosols, clouds and oceanic primary production. They proposed that increased temperaturecould increase primary production in the oceans leading to more DMS production. Theincreased DMS production would lead to more CCN and through the aerosol indirect effectto a cooler climate.

Although the CLAW hypothesis was mainly based on speculations which probably aretoo simple, it contained an important conclusions: Complex links can exist between aerosols,ocean chemistry and climate.

Falkowski et al. [1998] discuss increased (iron limited) phytoplankton production in theocean due to deposition of dust. The phytoplankton can take up CO2 and transport it tothe ocean interior hence influencing atmospheric CO2 levels.

Swap et al. [1992] discuss nutrient transport (phosphate) from the Sahara to the Amazonforest occurring on dust aerosols.

1.2.5 Aerosol/photochemistry interactions

Aerosols interact with tropospheric chemistry through two effects:

1. Heterogeneous uptake of gases on the aerosol surface. The uptake can be followed bychemical reactions on the aerosol surface or inside the small droplet.

2 MODELING STUDIES OF ATMOSPHERIC AEROSOLS 4

2. Changing photolysis rates (due to scattering and absorption on the aerosols).

Some works have tried to quantify the effect of dust on tropospheric chemistry: Dentener

et al. [1996] concentrated on heterogeneous uptake. He calculated that O3 would decreaseboth because O3 production decreased (N2O5 and HO2 were taken up on dust) and becausethe O3 molecules were taken up on dust themselves.

Bian and Zender [2003] quantified the effect of dust on tropospheric chemistry due toboth photolysis and heterogeneous update. They found that on a global average O3 decreaseby 0.7%, OH decrease by 11.1%, and HO2 decrease by 3.5 % when dust is added to theatmosphere.

Heterogeneous chemistry and photolysis change can have opposite effects e.g. for Ozone.In remote areas, O3 is destroyed by photolysis. Including dust, makes photochemical destruc-tion of O3 is decrease (O3 increases). Heterogeneous uptake of O3 on dust makes O3 decrease.Regional effects of dust vary because the effect depends on background photochemistry e.g.NOx content.

Sulfate is produced inside cloud droplets. (e.g. Chin et al. [1996]). Conversion of SO2 tosulfate inside sea salt aerosols has also been proposed to be a source of sulfate mass [Sievering

et al., 1992]The effect of aerosols on photochemistry is a field to explore, and it will be an interesting

research topic in the future.

2 Modeling studies of atmospheric aerosols

As described in Section 1.2.1, aerosols are complex mixtures of inorganic compounds, organiccompounds and water. Each aerosol size and composition can interact with the atmospherein it’s own way. This makes modeling of the atmospheric aerosol a complex problem. Youhave to chose your modeling tool with care depending on the problem you want to solve.

2.1 Detailed aerosol models

To accurately model size and composition of atmospheric aerosols, one should calculate mi-crophysical processes such as nucleation, condensation and coagulation [Seinfeld and Pandis ,1998]. Several approaches have been taken to model the dynamics of the aerosol population.

The model applied to solve the aerosol dynamics, often demand much computer resources.Pilinis et al. [2000]and Gaydos et al. [2003] made very detailed process studies of the atmo-spheric aerosol. The algorithm used to solve a specific problem has to reflect the amount ofthe computer resources available and the complexity of the problem itself.

Examples of ways to save computer resources are to assume chemical equilibrium (e.g.Nenes et al. [1998]) instead of calculating mass transfer explicitly and to use lognormal“modes” instead of discrete sections to describe the aerosol size distributions will also savecomputer resources [Binkowski and Shankar , 1995].

Zhang et al. [1999] discusses numerical efficiency of algorithms used to solve the aerosoldynamics. Some of the approaches are appropriate for use in 3D transport model while otherapproaches need unrealistic computer resources.


2.2 Simpler and global models

Global models of atmospheric aerosols can not take into account all the details of the at-mospheric aerosol which are calculated by advanced box models. Many global atmosphericaerosol models only focus on one aerosol component (e.g. Zender et al. [2003a] on dust,Berglen et al. [2003] on sulfate, Gong et al. [1997] on sea salt, Cooke and Wilson [1996]on black carbon, Liousse et al. [1996] on total carbon and Chung and Seinfeld [2002] orTsigaridis and Kanakidou [2003] on secondary organic carbon).

The global models often ignore aerosol microphysics and do not try to model mixing andinteraction between different types of aerosols (exceptions are e.g. Wilson et al. [2001] andAdams and Seinfeld [2002]).

2.3 Models for atmospheric dust

2.3.1 Dust production

The production of dust depends on the wind friction speed. As soon as the wind frictionspeed surpasses a threshold wind, soil is lifted from the ground [Iversen and White, 1982].Several recent studies have pointed out that dust production occurs by two related processescalled saltation and sandblasting [Gomes et al., 1990; Shao and Raupach, 1993; Marticorena

and Bergametti , 1995; Shao et al., 1996; Alfaro et al., 1997, 1998; Lu and Shao, 1999; Shao

and Lu, 2000; Shao, 2001; Alfaro and Gomes , 2001].Saltation refers to a layer of soil moving horizontally with the wind in a near surface layer.

Sandblasting refers to the release of fine dust when the saltators impact the ground. Existingdust production models predict that the smaller the aerosols are, the tighter bound to thesoil by cohesion forces they are [Shao and Raupach, 1993]. Thus the size of the dust aerosolsreleased depend on the kinetic energy of the saltators [Shao, 2001; Alfaro and Gomes , 2001].

Despite the existence of physically based dust production schemes, input data needed touse these advanced schemes are not available on the global scale.

A common way of of representing dust emissions in current models is to calculate saltationfluxes, and then fix a size distribution with mass median diameter (MMD) approximatelyequal to 2.5 µm to the dust flux (e.g. Schulz et al. [1998]; Claquin [1999]; Zender et al.

[2003a]). Others studies prefer empirical relations which relate the size distributed verticaldust flux to wind speed (e.g. Woodward [2001]; Ginoux et al. [2001])

Dust production models need realistic description of soil moisture [Fecan et al., 1999],vegetation cover and surface roughness elements [Marticorena et al., 1997].

Surface roughness is an important parameter for dust production because wind frictionspeed increases with surface roughness given a constant wind speed. Surface roughnesselements influence dust production in two ways:

1. Non erodible roughness elements absorb momentum from the air, thus making it notavailable for soil erosion. These roughness elements can be small rocks, sticks, bushesetc.

2. If the roughness elements themselves are erodible, the surface roughness will (throughincreasing wind friction speed) increase the momentum available for erosion. Gillette


et al. [1998] proposed that saltation itself will increase surface roughness which willincrease momentum transfer to the ground, thus increasing dust emissions through apositive feedback loop.

Most surface roughness elements are non erodible which means that in general, surfaceroughness will decrease dust emissions.

2.3.2 Global dust models

One of the first global 3D modeling studies of dust was done by Tegen and Fung [1994].Even though the source formulation was simple, and not consistent with the physical salta-tion/sandblasting schemes described above, the results were reasonable. During the 1990’s,several global modeling studies followed. Table 1 gives estimates of published dust produc-tion and loadings.

Prospero et al. [2002] and Ginoux et al. [2001] focused on “hot spots” for dust emissions.The “hot spots” are supposed to be areas where sediments have accumulated (for example onthe bottom of a lake which now is dry). Dust modeling studies published after 2001 containa description on soil erodibility in their source formulation [Tegen et al., 2002; Zender et al.,2003a, b].

Table 1: Some earlier estimates of yearly average dust production and loading. The numbersare dependent on the dust size distribution, and are only meant to give rough estimations

Production Mass loadingReference Tg yr−1 mg m−2

Tegen et al. [1997] 1250 36.3Claquin [1999] 2300 35.6Takemura et al. [2000] 3321 27.1Ginoux et al. [2001] 1604-1960 70.4Zender et al. [2003a] 1490 33.3This thesis 1500 37.0

2.3.3 Anthropogenic dust?

Dust is a natural aerosol. Even so, speculations of man made (anthropogenic) dust havebeen made [Tegen and Fung , 1995; Tegen and Lacis , 1996; Sokolik and Toon, 1996]. Theman made dust in these works is supposed to be due to vegetation changes.

The studies of Ginoux et al. [2001] and Prospero et al. [2002] turned focus from onlyland use change to dust “hot spots”. The “hot spots” (Section 2.3.2. Dry lakes are nowhypothesized to be the most efficient dust sources because that is where sediments (which canbecome atmospheric dust) are. Zender et al. [2003b] showed that their global model showedresults more similar to measurements when multiplying all emissions with a geomorphological“erodibility factor” accounting for downstream area.


The acknowledgment that “hot spots” exist means that earlier estimates of anthropogenicdust solely due to vegetation changes are likely to be wrong. One single dry lake canincrease dust emissions more than several dried forests or savannas. The dry lake hypothesisalso makes it difficult to distinguish between climate forcing and feedback. If humans areresponsible for drying out lakes, it can both be through pumping all the water out of it(forcing), or by changing climate locally so that the lake dries out (feedback). The estimationof human impact on dust emission is a challenging task for future research. Dust emissionsare sensitive to climate feedbacks. They have been very high in earlier climate periods likein the Last Glacial Maximum (LGM) [Mahowald et al., 1999].

2.4 Models for atmospheric sea salt

2.4.1 Sea salt production

Sea salt is assumed to be produced when bubbles burst by breaking waves on the oceansurface, or by direct tearing of the wave crests by wind. Three types of droplets are formed.

1. Film droplets are formed by bubbles bursting. These are small droplets rdry < 0.5µm.

2. Jet droplets are formed by bubbles bursting. These are droplets with 0.5µm < rdry <

4.0µm.

3. Spume droplets are formed by direct tearing of wave crests by wind. They are formedonly at high wind speeds, and are normally large (rdry > 4.0µm)

The number of droplets formed depends on the size of the bubble. Small bubbles produceonly jet drops and large droplets produce only film droplets [Martenson et al., 2003].

Models calculating sea salt concentrations, use empirical equations connecting the surfacewind speed to either the concentration in the lowest model layer or to the production fluxof sea salt.

Martenson et al. [2003] show that sea salt production depends on surface temperature.She gives no physical reason for this dependence, but speculates that “The difference in

bubble spectra, surface tension, viscosity and density of the water at different temperatures

are probably an important explanation of why the aerosol number concentration and size

distribution depend on the water temperature.” If these factors play a role, then a physicalformulation of what happens during sea salt formation is much more complicated than theparameterizations used in most models today. (Parameterizations proposed by Monahan

et al. [1986] and Smith et al. [1993] are commonly used.)

2.4.2 Global sea salt models

Atmospheric sea salt content has been calculated in several modeling studies. Table 2 showsglobal model results for sea salt. These models use empirical relations to relate surface windsto either sea salt production or sea salt concentration in the lowest model layer. The salt islost from the atmosphere by dry deposition and wet deposition.

3 FUTURE WORK 8

Table 2: Earlier estimates of yearly average sea salt production and loading. The numbersdepend on size distribution, and are only meant to give rough estimations

Production Mass loadingReference Tg yr−1 mg m−2

Petrenchuk [1980] 1000 -Erickson and Duce [1988] 10000-30000 -Tegen et al. [1997] 5900 22.4Takemura et al. [2000] 3300 11.0Guelle et al. [2001] - 25-38Haywood et al. [1999] - 7.5-36.8This thesis 6500 12.0

2.5 Evaluation of models used in this thesis

As explained above, there exists currently both detailed and simplified tools to modelaerosols. The tools used in this thesis must be regarded as rather simple compared to themost sophisticated aerosol models which exist. We do for example no attempt to describeinteraction between the aerosols and the gases in the atmosphere (except sea salt growth bycondensing water vapor).

The goal of this thesis is not to solve all questions related to atmospheric aerosols sizeand composition. The goal is to give an impression of some mechanisms controlling the massof natural aerosols in the atmosphere, and to estimate budgets of the most important (bymass) natural atmospheric aerosols. The thesis also also estimates the importance of sea saltin the global radiative balance.

The results permit to better understand the climate effect of natural aerosols. Combinedwith model studies of anthropogenic aerosols (e.g. sulfate from SO2 emissions) it will bepossible to obtain an understanding of what part of the atmospheric aerosols are man madeand what part are due to natural processes.

3 Future work

As discussed in section 1.2 aerosols influence the climate in different ways, all of whichdeserves more research. It seems to me that all processes described in section 1.2 are equallyimportant. It impossible to advise in one way or the other on what kind of research deservespriority. The priority will depend on what question a given scientist is interested in, whatcolleagues are available for cooperation, computer- or laboratory tools available etc.

I will however point to the two direction which I feel are the logical continuation of thisthesis:

3 FUTURE WORK 9

3.1 Aerosol process studies

Most global transport models today predict aerosol mass [Liousse et al., 1996; Cooke and

Wilson, 1996; Gong et al., 1997; Chung and Seinfeld , 2002; Zender et al., 2003a]. We needtools to predict aerosol number and surface area from physical and chemical principles. Someattempts have been done to calculate more than mass in global models [Wilson et al., 2001;Adams and Seinfeld , 2002, 2003]. These studies show there is not necessarily a linear relation-ship between aerosol mass and number concentration. Emissions will either change aerosolmass, or both aerosol mass and number depending on atmospheric conditions. Hygroscopicproperties can change too.

Rosenfeld et al. [2002] showed that anthropogenic aerosols might act as CCN, but theirefficiency would be reduced in presence of natural (sea salt) aerosols. The formation of acloud droplet on a particle depends on many other things than the total aerosol mass [Ghan

et al., 1998; Nenes et al., 2001]. Since size and composition of the aerosols are so important,it is important to have a process based tool to predict these properties in global models.

Main questions to answer in the future are: When will a given emission contribute toform more CCN ? In what conditions will the number of CCN stay constant even thoughpollutants are emitted? When will emissions change aerosol populations so that it interactsin a different way with radiation ? The answers depend on the pre-existing aerosol populationand atmospheric physical and chemical conditions. Good tools to model aerosol microphysics[Gelbard et al., 1980] are needed to answer these questions. Global aerosol models have toincorporate these physics within the limits of available computers.

3.2 Combined use of Aerosol modeling and satellite retrievals

3-D modeling results (like the ones presented in this thesis) are difficult to validate. Mea-surements of mass concentrations and/or optical depths are usually available, but only froma limited number of observation sites. Often, the observation sites can be influenced by localphenomenon which are not taken into account in global models.

In the last years, satellite retrievals [Husar et al., 1997; Torres et al., 1998; Nakajima and

Higurashi , 1998; Goloub et al., 1999; Mishchenko et al., 1999] have become available. Thesatellite retrievals make it possible to compare global model output to global satellite images.Satellite retrievals are not measurements. In the satellite retrievals there are assumptions onaerosol size distribution, aerosol shape, ocean surface reflectance and altitude of aerosol layer.The retrievals are only possible in cloud free areas, so the satellites apply cloud-screeningalgorithms. Aerosol layers (e.g. Saharan dust storms) can be optically thick and thereforedifficult to distinguish from a thin cloud. Myhre et al. [2003] shows that satellite retrievalsvary from one to another.

When using satellite data, one should be aware of limitations. If done with care, com-bining model results and satellite retrievals will be a powerful tool in the future. It is logicalto use satellite data to better evaluate the aerosols models used in this thesis. In partic-ular it would be valuable to compare retrievals and model results for specific mineral dustoutbreaks, both with respect to aerosol size distribution and total dust mass emitted to theatmosphere. Hopefully, detailed comparisons between aerosol properties from model andsatellite can give an better understanding of natural aerosol production and transport.

4 SUMMARY OF PAPERS 10

4 Summary of papers

4.1 Paper 1

Grini, A., Myhre, G., Sundet, J.K. and Isaksen, I.S.A. Modeling the annual cycle of seasalt in the global 3D model Oslo CTM2: Concentrations, fluxes and Radiative impact, J.

Climate, 2002, vol 15, pp 1717

This paper estimates the concentrations, column burden and radiative impact of sea saltaerosols in the global atmosphere using a 3D tracer transport model. The main conclusionsof the paper is that the model is capable of simulating seasonal patterns of sea salt observedat several sites around the world. The aerosols scatter incoming sunlight, and the climateeffect of the aerosols through this process is estimated to -1.1 W m−2 on the top of theatmosphere. The effect is stronger if clouds are not taken into account. Much of the salt areproduced in ares with high winds and over sea which often correlates with areas with highcloud cover. If the aerosols are under clouds, sunlight is reflected by the overlying clouds, andthe effect of the aerosols is decreased. If clouds are not taken into account in the radiativetransfer calculations, the climate effect is calculated to be -2.2 W m−2.

4.2 Paper 2

Myhre, G., A. Grini, J.M Haywood, F. Stordal, B. Chatenet, D. Tanre, J.K. Sundet andI.S.A. Isaksen: Modeling the radiative impact of mineral dust during the Saharan DustExperiment (SHADE) campaign, J. Geophys. Res., 2003, vol 108, no. D18, pp 8579

This paper describes modeling of a dust outbreak which took place during the SHADEaerosol measurement campaign in September 2000. A simple dust production mechanism wasused to generate dust concentrations in the 3D transport model Oslo CTM2. The modeleddust concentrations were similar to the measured ones. The radiative impact locally verylarge, up to -115 W/m2. Modeled scattering, single scattering albedo and asymmetry factorscompared well with measured ones. Vertical profiles of dust were well modeled. The modelunderestimated the production of large aerosols leading to an underestimation of the longwave radiative effect of the minerals. For an square between 30oW, 0oN to 40oE,40oN thediurnal mean solar radiative impact of dust was estimated to -8- -10W m−2 excluding clouds.

4.3 Paper 3

Grini, A., C.S. Zender and P. Colarco: Saltation Sandblasting behavior during mineral dustaerosol production, Geophys. Res. Lett., 2002, vol 29, no. 18, pp 1868

This paper discusses a dust production model earlier published by Alfaro and Gomes

[2001]. Before using the model in a large scale model, it was necessary to refine the numericalimplementation of it. As discussed earlier, dust production is a result of both saltation andsandblasting. To calculate the size distributed (vertical) dust flux, it is necessary to firstcalculate the size distributed saltation (horizontal) flux.


We show that this can only be done with a very high resolution in saltator size whichwill be expensive calculations to do in large scale models. We discuss the sensitivity of thedust production model to the size resolution both in terms of number and mass flux. Weshow that although the dust flux can be determined accurately in terms of number, theseaerosol can have wrong size distribution, and thus the mass flux of dust can be wrong byseveral orders of magnitude.

4.4 Paper 4

Grini A. and C.S. Zender: Roles of saltation, sandblasting, and wind speed variability onmineral dust aerosol size distribution during the Puerto Rican Dust Experiment (PRIDE),J. Geophys. Res., in press, 2004.

This paper treats dust production using the saltation/sandblasting model discussed inPaper 3. The numerical problems were solved pre-calculating several parameters and storingthem in look up tables during the calculations. Computing in this manner, the size dis-tributed dust flux is a function only of wind friction speed and soil size distribution. Eachdust emission event will have it’s own size distribution depending on the wind friction speedand the soil size distribution.

The 3-D tracer transport model MATCH was used to transport dust. After transport,the size distributed dust concentrations were compared to measured size distributions atPuerto Rico during the PRIDE experiment. We show that a good knowledge of the soilsize distribution and wind speed variability in source areas is essential for modeling sizedistributed dust concentrations at remote sites.

4.5 Paper 5

Grini, A., G. Myhre, C.S. Zender, J.K. Sundet and I.S.A. Isaksen: Model simulations of dustsources and transport in the global troposphere. Effects of soil erodibility and wind speedvariability Institute report series, no 124, Department of Geosciences, University of Oslo,Oslo, Norway, 2003, ISBN 82-91885-26-5.

This paper treats the annual, global atmospheric dust cycle. We show that the 3-D tracertransport model Oslo CTM2 is capable of reproducing most aspects of the dust cycle. Wedo several model runs calculating global dust production, transport and deposition for theyear 1996. We discuss two important aspects of global dust modeling

1. The role of soil erodibility which is the ability of a soil to yield dust emission uponerosion. Several erodibility factors have been proposed earlier. We show that usingsurface reflectivity as an indication of soil erodibility can be a good approximation inglobal models.

2. Wind variability which means calculating a sub-grid wind distribution. As dust emis-sions only takes place when winds exceed a certain wind threshold, mean winds overa large area, are not appropriate to calculate dust emissions. Ignoring wind speedvariability over estimates dust emissions from areas dominated by high wind speeds.


Assuming the annual, global emissions are 1500 Tg, we estimate the global, annual burdento be approximately 18 Tg. Approximately 400 Tg of dust are deposited in the global ocean.

REFERENCES 13

References

Adams, P., and J. Seinfeld, Predicting global aerosol size distributions in general circulationmodels, J. Geophys. Res., 107 , 4370, 2002.

Adams, P., and J. Seinfeld, Disproportionate impact of particulate emissions on globalcould condensation nuclei concentrations, Geophys. Res. Lett., 30 , 1239–1243, 2003,doi:1.1029/2002GL016303.

Albrecht, B., Aerosols, cloud microphysics, and fractional cloudiness, Science, 245 , 1227–1230, 1989.

Alfaro, S. C., and L. Gomes, Modeling mineral aerosol production by wind erosion: Emissionintensities and aerosol size distributions in source areas, J. Geophys. Res., 106 , 18,075–18,084, 2001.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Modeling the size distribution of asoil aerosol produced by sandblasting, J. Geophys. Res., 102 , 11,239–11,249, 1997.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Mineral aerosol production by winderosion: aerosol particle sizes and binding energies, Geophys. Res. Letters , 25 , 991–994,1998.

Arrhenius, S., On the influence of carbonic acid in the air on the temperature on the ground,Philos , 41 , 237–276, 1896.

Berglen, T., T. Berntsen, I. Isaksen, and J. Sundet, A global model of the coupled sul-fur/oxidant chemistry in the troposphere: The sulfur cycle, J. Geophys. Res., submitted ,2003.

Bian, H., and C. Zender, Mineral dust and global tropospheric chemistry: Relative roles ofphotolysis and heterogeneous uptake, J. Geophys. Res., In press , 2003.

Binkowski, F., and U. Shankar, The regional particulate matter model .1. model descriptionand preliminary results, J. Geophys. Res., 100 , 26,191–26,209, 1995.

Boville, B., and P. Gent, The NCAR climate system model, version one, J. Climate, 11 ,1115–1130, 1998.

Charlson, R., J. Lagner, H. Rodhe, C. Leovy, and S. Warren, Perturbation of the northern-hemisphere radiative balance by backscattering from anthropogenic sulfate aerosols, Tel-

lus , 43 , 152–163, 1991.

Charlson, R. J., J. E. Lovelock, M. O. Andreae, and S. G. Warren, Oceanic phytoplankton,atmospheric sulphur, cloud albedo and climate, Nature, 326 , 655–661, 1987.

Chin, M., D. Jacob, G. Gardner, P. Spiro, M. Foreman-Fowler, and D. L. Savoie, A globalthree dimensional model of tropospheric sulfate, Journal of geophysical research, 101 ,18,667–18,990, 1996.

REFERENCES 14

Chung, S., and J. Seinfeld, Global distribution and climate forcing of carbonaceous aerosols,J. Geophys. Res., 107 , 2002, doi:10.1029/2001JD001397.

Claquin, T., Modelisation de la mineralogie et du forcage radiatif des poussieres desertiques,Ph.D. thesis, University of Hamburg, 1999.

Cooke, W., and J. Wilson, A global black carbon model, J. Geophys. Res., 101 , 19,395–19,409, 1996.

Dentener, F. J., G. R. Carmichael, Y. Zhang, J. Lelieveld, and P. J. Crutzen, Role ofmineral aerosol as a reactive surface in the global troposphere, J. Geophys. Res., 101 ,22,869–22,889, 1996.

Erickson, D. J., and R. A. Duce, On global flux of atmospheric sea salt, J. Geophys. Res.,93 , 14,079–14,088, 1988.

Falkowski, P., R. Barber, and V. Smetacek, Biogeochemical controls and feedbacks on oceanprimary production, Science, 201 , 200–206, 1998.

Fecan, F., B. Marticorena, and G. Bergametti, Parameterization due to the increase of theaeolian erosion threshold wind friction velocity due to soil moisture for arid and semi-aridareas, Annales Geophysicae, 17 , 149–157, 1999.

Gaydos, T., B. Koo, S. Pandis, and D. Chock, Development and application of an effi-cient moving sectional approach for the solution of the atmospheric aerosol condensa-tion/evaporation equations, Atm. Env., 37 , 3303–3316, 2003.

Gelbard, F., Y. Tambour, and J. Seinfeld, Sectional representations for simulating aerosoldynamics, Journal of colloid and interface science, 76 , 541–556, 1980.

Ghan, S., G. Guzman, and H. Abdul-Razzak, Competition between sea salt and sulfateparticles as cloud condensation nuclei, J. Atm. Sci., 55 , 3340–3347, 1998.

Gillette, D., B. Marticorena, and G. Bergametti, Change in the aerodynamic roughnessheight by saltating grains: Experimental assessment, test of theory and operational pa-rameterization, J. Geophys. Res., 103 , 6203–6209, 1998.

Ginoux, P., M. Chin, I. Tegen, J. Prospero, B. Holben, O. Dubovik, and S.-J. Lin, Sourcesand distributions of dust aerosols simulated with the GOCART model, J. Geophys. Res.,106 , 20,255, 2001.

Goloub, P., D. Tanre, J. Deuze, M. Herman, A. Marchand, and F. Breon, Validation ofthe first algorithm applied for deriving the aerosol properties over the ocean using thePOLDER ADEOS measurements, IEEE Transactions of Geoscience and Remote Sensing ,37 , 1586–1596, 1999.

Gomes, L., G. Bergametti, G. Coude-Gaussen, and P. Rognon, Submicron desert dusts: Asandblasting process, J. Geophys. Res., 95 , 13,927–13,935, 1990.

REFERENCES 15

Gong, S. L., L. A. Barrie, and J.-P. Blanchet, Modeling sea-salt aerosols in the atmosphere1. model development, J. Geophys. Res., 102 , 3805–3818, 1997.

Guelle, W., M. Schulz, Y. Balkanski, and F. Dentener, Influence of the source formulationon modeling the atmospheric global distribution of sea salt aerosol, J. Geophys. Res., 106 ,27,509–27,524, 2001.

Hansen, J., et al., Climate forcings in Goddard Institute for Space Studies SI2000 simulations,J. Geophys. Res., 107 , 4347, 2002.

Haywood, J., V. Ramaswamy, and B. Soden, Tropospheric aerosol climate forcing in clearsky satellite observation over the oceans, Science, 283 , 1299–1303, 1999.

Houghton, J., Y. Ding, D. Griggs, M. Noguer, P. Linden, X. Dai, K. Maskell, and C. Johnson(Eds.), Climate change 2001,: The scientific basis , Camebridge University Press, 2001.

Husar, R., J. Prospero, and L. Stowe, Characterization of tropospheric aerosols over theoceans with the NOAA advanced very high resolution radiometer optical thickness prod-uct, J. Geophys. Res., 102 , 16,889–16,909, 1997.

Iversen, J., and B. White, Saltation threshold on Earth, Mars and Venus, Sedimentology ,29 , 111–119, 1982.

Kulmala, M., V. Kerminen, T. Antilla, A. Laaksonen, and C. O’Dowd, Organic aerosolformation via sulphate cluster activation, J. Geophys. Res., submitted , 2003.

Liousse, C., J. Penner, C. Chuang, J. Walton, H. Eddleman, and H. Cachier, A global three-dimensional model study of carbonaceous aerosols, J. Geophys. Res., 101 , 19,411–19,432,1996.

Lu, H., and Y. Shao, A new model for dust emission by saltation bombardment, J. Geophys.

Res., 104 , 16,827–16,842, 1999.

Mahowald, N., K. Kohfeld, M. Hansson, Y. Balkanski, S. Harrison, I. Prentice, M. Schulz,and H. Rodhe, Dust sources and deposition during the last glacial maximum and cur-rent climate: A comparison of model results with paleodata from ice cores and marinesediments, J. Geophys. Res., 104 , 15,895–15,916, 1999.

Manabe, S., and R. F. Strickler, Thermal equilibrium of the atmosphere with a convectiveadjustment, J. Atmos. Sci., 21 , 361–385, 1964.

Martenson, E., E. Nilsson, G. deLeeuw, L. Cohen, and H. Hansson, Laboratory simulationsand parameterization of the primary marine aerosol production, J. Geophys. Res., 108 ,2003, doi:1029/2002JD002263.

Marticorena, B., and G. Bergametti, Modeling of the atmospheric dust cycle: 1. design of asoil derived dust emission scheme, J. Geophys. Res., 100 , 16,415–16,429, 1995.

REFERENCES 16

Marticorena, B., G. Bergametti, B. Aumont, Y. Callot, C. N’Doume, and M. Legrand,Modeling the Saharan dust cycle: 2. Simulation of Saharan dust sources, J. Geophys.

Res., 102 , 4387–4404, 1997.

Menon, S., A. DelGenio, D. Koch, and G. Tselioudis, GCM simulations of the aerosol in-direct effect: Sensitivity to cloud parameterization and aerosol burden, Journal of the

atmospheric sciences , 59 , 692–713, 2002.

Mishchenko, M., B. Geogdzhayev, B. Cairns, W. Rossow, and A. Lacis, Aerosol retrievalsover the ocean by use of channels 1 and 2 AVHRR dat: Sensitivity analysis and preliminaryresults, Appl. Optics , 38 , 16,831–16,847, 1999.

Monahan, E., D. Spiel, and K. Spiel, Oceanic whitecaps , Reidel, 1986.

Myhre, G., and F. Stordal, Global sensitivity experiments of the radiative forcing due tomineral aerosols, J. Geophys. Res., 106 , 18,193–18,204, 2001.

Myhre, G., et al., Intercomparison of satellite retrieved aerosol optical depth over ocean, J.

Atm. Sci., In press , 2003.

Nakajima, T., and A. Higurashi, A use of two-channel radiances for an aerosol characteriza-tion from space, Geophys. Res. Lett., 25 , 3815–3818, 1998.

Nenes, A., S. Pandis, and C. Pilinis, ISORROPIA: A new thermodynamic equilibrium modelfor multiphase multicomponent inorganic aerosols, Aquat. Geoch., pp. 123–152, 1998.

Nenes, A., S. Ghan, H. Abdul-Razzak, P. Chuang, and J. Seinfeld, Kinetic limitations oncloud droplet formation and impact on cloud albedo, Tellus , 53B , 133–149, 2001.

Petrenchuk, O. P., On the budget of sea salts and sulphur in the atmosphere, J. Geophys.

Res., 85 , 7439–7444, 1980.

Pilinis, C., K. Capaldo, A. Nenes, and S. Pandis, MADM-a new multicomponent aerosoldynamics model, Aerosol Science and Technology , 32 , 482–502, 2000.

Prospero, J., P. Ginoux, O. Torres, S. Nicholson, and T. Gill, Environmental characteriza-tion of global sources of atmospheric soil dust identified with the NIMBUS 7 total ozonemapping spectrometer (TOMS) absorbing aerosol product, Reviews of Geophysics , 40 ,2002, doi:10.1029/2000RG000095.

Rosenfeld, D., R. Lahav, A. Khain, and M. Pinsky, The role of sea spray in cleansing airpollution over ocean via cloud processes, Science, 297 , 1667–1670, 2002.

Schulz, M., Y. Balkanski, W. Guelle, and F. Dulac, Role of aerosol size distribution andsource location in a three dimensional simulation of a saharan dust episode tested againstsatellite derived optical thickness, J. Geophys. Res., 103 , 10,579–10,592, 1998.

Seinfeld, J. H., and S. N. Pandis, Atmospheric chemistry and physics, From air pollution to

Climate change, John Wiley and Sons, 1998.

REFERENCES 17

Shao, Y., A model for mineral dust erosion, J. Geophys. Res., 106 , 20,239–20,254, 2001.

Shao, Y., and I. Lu, A simple expression for wind erosion threshold friction velocity, J.

Geophys. Res., 105 , 22,437–22,443, 2000.

Shao, Y., and M. Raupach, Effect of saltation bombardment by wind, J. Geophys. Res , 98 ,12,719–12,726, 1993.

Shao, Y., M. R. Raupach, and J. F. Leys, A model for predicting aeolian sand drift and dustentrainment on scales from paddock to region, Aust. J. Soil Res., 34 , 309–342, 1996.

Sievering, H., J. Boatman, E. Gorman, Y. Kim, L. Anderson, G. Ennis, M. Luria, andS. Pandis, Removal of sulfur from the marine boundary-layer by ozone oxidation in sea-salt aerosols, Nature, 360 , 571–573, 1992.

Smith, M. H., P. M. Park, and I. E. Consterdine, Marine aerosol concentration and estimatedfluxes over sea, Q. J. R. Meteorol. Soc., 119 , 809–824, 1993.

Sokolik, I., and O. Toon, Direct radiative forcing by anthropogenic airborne mineral aerosols,Nature, 381 , 681–683, 1996.

Swap, R., M. Garstang, and S. Greco, Saharan dust in the Amazon basin, Tellus , 44B ,133–149, 1992.

Takemura, T., H. Okamoto, Y. Maruyama, A. Numaguti, A. Higurashi, and T. Nakajima,Global three dimensional simulation of aerosol optical thickness distribution of variousorigins, J. Geophys. Res., 105 , 17,853, 2000.

Tegen, I., and I. Fung, Modeling of mineral dust in the atmosphere: Sources, transport andoptical thickness, J. Geophys. Res., 99 , 22,897–22,914, 1994.

Tegen, I., and I. Fung, Contribution to the atmospheric mineral aerosol load from landsurface modification, J. Geophys. Res., 100 , 18,707–18,726, 1995.

Tegen, I., and A. A. Lacis, Modeling of particle size distribution and its influence on theradiative properties of mineral dust aerosol, J. Geophys. Res., 101 , 19,237–19,244, 1996.

Tegen, I., P. Hollrig, M. Chin, I. Fung, D. Jacob, and J. Penner, Contribution of differentaerosol species to the global aerosol extinction optical thickness: Estimates from modelresults, J. Geophys. Res., 102 , 23,895–23,915, 1997.

Tegen, I., S. Harrison, K. Kohfeld, I. Prentice, M. Coe, and M. Heimann, Impact of vegetationand preferential source areas on global dust aerosol: Results from a model study, J.

Geophys. Res., 107 , 4576, 2002, doi:10.1029/2001JD00096.

Torres, O., P. Bhartia, J. Herman, Z. Ahmad, and J. Gleason, Derivation of aerosol propertiesfrom satellite measurements of backscattered ultraviolet radiation: Theoretical basis, J.

Geophys. Res., pp. 17,099–17,110, 1998.

REFERENCES 18

Tsigaridis, K., and M. Kanakidou, Global modelling of secondary organic aerosol in thetroposphere: a sensitivity analysis, Atmos. Chem. Phys., 3 , 1849–1869, 2003, www.atmos-chem-phys.org/acp/3/1849/.

Twomey, S., The influence of atmospheric particulates on cloud and planetary albedo, Bul-

letin of the american meteorological society , 52 , 640, 1971.

Twomey, S., The influence of pollution on the shortwave albedo of clouds, Journal of the

atmospheric sciences , 34 , 1149–1152, 1977.

Warner, J., and S. Twomey, The production of cloud nuclei by cane fires and the effect oncloud droplet concentration, Journal of the atmospheric sciences , 24 , 704–706, 1967.

Wilson, J., C. Cuvelier, and F. Raes, A modeling study of global mixes aerosol fields, J.

Geophys. Res., 106 , 34,081–34,108, 2001.

Woodward, S., Modeling the atmospheric life-cycle and radiative impact of mineral dust inthe Hadley center climate model, J. Geophys. Res., 106 , 2001.

Zender, C., H. Bian, and D. Newman, The mineral dust entrainment and deposition (dead)model: Description and global dust distribution, J. Geophys. Res., 108 , 4416, 2003a.

Zender, C., D. Newman, and O. Torres, Spatial heterogeneity in aeolian erodibility: Uniform,topographic, geomorphologic, and hydrologic hypotheses, J. Geophys. Res., 108 , 4543,2003b, doi:10.1029/2002JD00303.

Zhang, Y., C. Seigneur, J. Seinfeld, M. Jacobson, and F. Binkowski, Simulation of aerosoldynamics: A comparative review of algorithms used in air quality models, Aerosol Science

and Technology , 31 , 487–514, 1999.

1 JULY 2002 1717G R I N I E T A L .

q 2002 American Meteorological Society

Modeling the Annual Cycle of Sea Salt in the Global 3D Model Oslo CTM2:Concentrations, Fluxes, and Radiative Impact

ALF GRINI, GUNNAR MYHRE, JOSTEIN K. SUNDET, AND IVAR S. A. ISAKSEN

Department of Geophysics, University of Oslo, Oslo, Norway

(Manuscript received 30 April 2001, in final form 28 December 2001)

ABSTRACT

A global three-dimensional chemical transport model (CTM) is used to model the yearly cycle of sea salt.Sea salt particles are produced by wind acting on the sea surface, and they are removed by wet and dry deposition.In this study, forecast meteorological data are taken from the ECMWF. The modeled concentrations are comparedto measured concentrations at sea level, and both absolute values and monthly variations compare well withmeasurements. Radiation calculations have been performed using the same meteorological input data as theCTM calculations. The global, yearly average burden of sea salt is found to be 12 mg m22. This is within therange of earlier estimates that vary between 11 and 22 mg m22. The radiative impact of sea salt is calculatedto be 21.1 W m22. The total, yearly flux of sea salt is estimated to be 6500 Tg yr21.

1. IntroductionThere has been an increased focus on the climate

effect of tropospheric aerosols during the last few years.These aerosols consist of natural and anthropogenic spe-cies in particle form with radii from several nm to mm.Typical anthropogenic sources can be emissions of sul-phuric gases leading to the formation of sulphate par-ticles, or combustion to give carbon particles. The mostimportant natural particles are sea salt and mineral dustthat are emitted into the atmosphere because of windstress at the ocean surface and arid land areas, respec-tively.

Several studies have quantified the flux and concen-tration of sea salt particles. Estimates of the total, globalflux range from 1000 to 10 000 Tg yr21 (Blanchard1985; Seinfeld and Pandis 1998). These estimates haveoften used empirical relations for the surface concen-tration, assumed a relationship between dry and wetdeposition, and calculated the dry deposition. Tegen etal. (1997) found the global average burden of sea saltto be 22 mg m22 (with a source strength of 5900 Tgyr21) using a 3D tracer model whereas, Takemura et al.(2000) calculated the source strength to be 3321 Tg yr21

and the global burden to be 11 mg m22 using a generalcirculation model. Both studies calculate the sea saltconcentration directly from an empirical dependence onsurface wind.

Sea salt particles are important for the radiative bal-

Corresponding author address: Dr. Alf Grini, Department of Geo-physics, University of Oslo, P.O. Box 1022, Blindern, Oslo 0315,Norway.E-mail: [email protected]

ance, both directly, as a reflector of radiation, and in-directly as cloud condensation nuclei. Large uncertain-ties are associated with estimates of climate effect ofnatural and anthropogenic aerosol.

Earlier estimates of direct radiative effects of variousaerosols have mainly been derived from models (Hay-wood and Boucher 2000). Satellite retrievals for aerosolpurposes have advanced considerably over the last fewyears and will in the near future be an important toolfor constraining the radiative forcing due to the directaerosol effect, in particular, the anthropogenic part(King et al. 1999). Combinations of satellite observa-tions and radiative transfer models have been used toestimate the clear-sky direct aerosol effect (Bergstromand Russel 1999; Boucher and Tanre 2000; Haywoodet al. 1999). Haywood et al. (1999) found a substantialdifference between the top of the atmosphere clear-skyflux from the Earth Radiation Budget Experiment(ERBE) and a GCM. Including several aerosol com-ponents resulted in a much better agreement betweenthe model and observation. Boucher and Tanre (2000)used retrieved aerosol optical properties and a radiativetransfer model to estimate the direct radiative effect ofaerosols. Both studies indicate a direct clear-sky radi-ative effect of aerosols of several W m22. The studiesby Haywood et al. (1999) and Boucher and Tanre (2000)were limited to over-ocean and mostly clear-sky con-ditions. Uncertainties are large regarding natural versusanthropogenic aerosols contributions. Haywood et al.(1999) estimated a range for the global mean clear-skyradiative impact of sea salt from 21.0 to 23.5 W m22

(21.5 to 25.0 W m22 over the ocean).In this study, a 3D chemical transport model (CTM)

1718 VOLUME 15J O U R N A L O F C L I M A T E

has been used to calculate the global concentration ofsea salt particles. A radiative transfer model is used toestimate the radiative impact of sea salt aerosols in theatmosphere. The CTM is driven by forecast meteoro-logical data from the European Centre for Medium-Range Weather Forecasts (ECMWF). The meteorolog-ical data are used in a consistent way so that both theradiation study and the transport study use the samedata regarding wind-driven sources, transport, and ra-diative effects. Previous studies have either used less-sophisticated models or calculated boundary layer con-centrations directly from wind speed correlations.

2. Modelinga. CTM modeling

The processes taken into account in the model areproduction, transport by advection and convection, par-ticle growth (by absorption of humidity from the air),dry deposition, and wet deposition. Each of these pro-cesses are described in detail below. Coagulation of par-ticles has not been taken into account in this study.

b. Oslo CTM2

The model used in the calculations is the 3D OsloCTM2 model described in Sundet (1997). It is an offlineCTM that uses precalculated transport and physicalfields to simulate chemical turnover and distribution inthe atmosphere. The model is valid for the global tro-posphere with 19 vertical layers and the model domainreaching from the ground up to 10 hPa. Horizontal res-olution is 5.6258 (T21). The model uses data fromECMWF. Advection is done conserving second-ordermoments (Prather 1986) and convection is based on theTiedkte mass flux scheme (Tiedtke 1989), where verticaltransport of species is determined by the surplus/deficitof mass flux in a column. Meteorological data for 1996are used.

1) SEA SALT PRODUCTION

Sea spray is generated by the wind stress on the oceansurface. Air bubbles, which constitute the whitecaps re-sulting from breaking waves, burst at the water surfaceand produce small droplets by means of two mecha-nisms. Film drops are produced when the thin liquidfilm that separates the air within a bubble from the at-mosphere ruptures. The remaining surface energy of thebubble, after bursting, results in a liquid jet that becomesunstable and breaks into a number of jet drops (Smithet al. 1993). The formation of film and jet drops can becalled indirect mechanism. At wind speeds greater than10–12 m s21, spume drops torn directly from the wavecrests by the strong turbulence make an increasing con-tribution to the sea salt and dominate the concentrationat larger particle sizes. The formation of spume dropsis called the direct mechanism.

Production is described empirically by Monahan etal. (1986) from laboratory experiments:

]F 22B3.41 23 1.05 1.19e5 1.373U r (1 1 0.057r )10 , (1)10]r

where B is

0.380 2logrB 5 ,

0.650

for the indirect mechanism (bubbles bursting) and

]F26 2.08U 22105 8.60 3 10 e r (2)

]r

for direct mechanism (spume).Here, F is flux in particles m22 s21, U10 is wind speed

at 10-m height in m s21, and r is particle radius in mm.A similar production mechanism has been used in

other works (Gong et al. 1997a; Pryor and Sorensen2000). Equation (1) is strictly for particles bigger thanradius of 0.8 mm at 80% RH, but since we do not haveany better expression for smaller particles, Eq. (1) isused generally in this study, as was done by Gong etal. (1997a).

Smith et al. (1993), Gong et al. (1997a), and Pryorand Sorensen (2000) point out that the fluxes given byMonahan et al. (1986) were too big for the spume mech-anism. Based on measurements, Smith et al. (1993) pro-posed another expression that gives a more correct rep-resentation of the fluxes for bigger particles. The ex-pression given by Smith et al. (1993) is used in thiswork for production of sea salt by the spume mecha-nism.

The expression proposed by Smith et al. (1993) forthe sea salt flux, is

2]F 22 f [ln(R /R )]i i 0i5 A e , (3)O i]r i51

where f 1, f 2, R01, R02, have the values 3.1, 3.3, 2.1,and 9.2 mm, and

logA 5 0.0676U 1 2.431

1/2logA 5 0.959U 2 1.476,2

where index 1 means the indirect mechanism and index2 means the direct mechanism.

Since the equations given by Monahan et al. (1986)seem to give reasonable fluxes for small particles, butnot for large ones, the production of particles smallerthan 7 mm was calculated using these equations. Largerparticles were calculated according to Smith et al.(1993). At a radius of 7 mm these two schemes giveapproximately the same flux for different wind speeds.

Sea salt aerosols up to about 20 mm are found in theatmosphere (Erickson and Duce 1988). We assume thatthe 16 size bins given in Table 1 are sufficient to estimatethe sea salt mass balance.

For dry particles, we assume a density of 2200 kg m23,

1 JULY 2002 1719G R I N I E T A L .

the same density as pure and dry NaCl (Hess et al.1998). Sea salt production is calculated assuming a rel-ative humidity of 80%. At this relative humidity, theparticle radius will be twice the dry radius (Fitzgerald1975), and the density used in the production is thus1150 kg m23.

The number of particles produced is converted tomass of particles according to

43M 5 N 3 r 3 pr , (4)p 3

where M is total mass produced in one grid cell (kg),N is total number produced, rp is particle density, andr is radius.

The sea salt mass and radius are corrected for waterusing the formulas described in section 2b(2). Only thedry mass is added to the bin and transported. Thus wedo not transport any water. Mass is not transferred be-tween the bins because of growth. However, we do cal-culate wet density and radius for all the size bins as thiswill influence the dry deposition.

2) PARTICLE GROWTH

The particles will absorb water and grow as a functionof relative humidity in the air. Ideally, a chemical equi-librium approach should be used. A chemical equilib-rium model would calculate composition of aerosolswith respect to water, and aerosol species based on ther-modynamics (Zhang et al. 2000).

Since chemical equilibrium is not taken into account,the composition of the particles is constant during trans-port. The chemical composition of the dry particles isassumed to be 30.6% sodium, 55.04% chlorine, and14.4% of other inorganic components as given by, forexample, Seinfeld and Pandis (1998).

A simplified approach, described by Fitzgerald(1975), was used here. The radius of the aerosol afterabsorbing water vapor is given as

br 5 ar ,d (5)

where rd is dry radius, and a and b are coefficientsdepending on chemical composition of aerosol and onrelative humidity.

When calculating the density of the particles, it isassumed that the volume of water and the volume ofdry particle can be added together to give the total vol-ume. Thus it is assumed that the dry particle still oc-cupies the same volume in the wet particle as when dry.

The equations from Fitzgerald (1975) are valid forparticles with dry radius up to 3 mm. However, they areused generally for all particle sizes in this work and willprobably introduce some errors for the larger particles.

3) DRY DEPOSITION

The dry deposition velocity is the velocity at whichthe particles are transported to the ground. The fluxtoward the ground is given as

flux 5 [C(z) 2 C(0)] y ,dep (6)

where flux is given as mg (m2s)21, z is the height (m),C is concentration (mg m23), and ydep is the dry depo-sition velocity (m s21). The concentration on the groundis assumed to be 0. (That is, the particles are taken upcompletely if they reach the ground.)

The dry deposition velocity is calculated from thescheme given by Seinfeld and Pandis (1998). The drydeposition velocity is given by

1y 5 1 y , (7)dep str 1 r 1 r r ya b a b st

where ra is aerodynamic resistance, rb is resistance inthe quasi-laminar sublayer, and y st is the falling velocity.The expressions used to find these parameters are shownbelow.

The aerodynamic resistance is given by

1 zln 1 4.7(j 2 j ) (stable)0[ ]ku z0* 1 z

r 5 ln (neutral)a [ ]ku z0*2 2z (h 1 1)(h 1 1)0 0ln 1 ln 1 2(arctanh 2 arctanh ) (unstable),r 02 21 2 [ ]z (h 1 1)(h 1 1)0 r r

where j is z/L and h 5 (1 2 15j)1/4. Here L is theMonin–Obukhov length (the height above the groundwhere the production of turbulence by mechanical andbuoyancy forces are equal). The subscript 0 denotes the

roughness length and r denotes the reference height(which is the middle of the grid box). The equationsapply to different states of the atmosphere (stable, un-stable, and neutral).


TABLE 1. The radii of the different size bins used in thecalculations. The radii are given at 80% relative humidity.

BinRadius (from)

(mm)Radius (to)

(mm)

123456789

10111213141516

0.030.0470.0700.110.160.250.370.570.871.32.03.14.77.1

10.816.4

0.0470.0700.110.160.250.370.570.871.32.03.14.77.1

10.816.425.0

The sublayer resistance is

1r 5 ,b 2/3 23/Stu (Sc 1 10 )*

where u* is friction velocity, Sc is Schmidts number,and St is Stokes number of particle.

The falling velocity is2D r g1 p p

y 5 ,st 18 m

where Dp is particle diameter, rp is particle density, gis the acceleration of gravity, and m is viscosity of air.All parameters needed to determine the dry depositionvelocity are available in the Oslo CTM 2 model, forexample heat flux, friction wind, and temperature in thelowest model layer.

The density of the particles is calculated as a functionof relative humidity in the box as described in section2b(2). The dry-deposition velocity and the particle den-sity is calculated for each bin. Thus, each grid box (andeach bin in that box) will have its own density and itsown radius. These densities and radii are used in thedry-deposition calculations.

The fall velocity (y st) is used as a loss term for thelayers above the lowest. To conserve mass balance, theparticles removed by falling from one layer are countedas a production term in the layer below.

4) WET DEPOSITION

The amount of sea salt dissolved is assumed to beproportional with the product of cloud liquid water con-tent and cloud volume. Rain out of the dissolved seasalt is proportional to the rainfall limited by the availablecloud liquid water content. Reevaporation is not takeninto account.

Cloud fraction and cloud liquid water are diagnosed

in the ECMWF model as standard fields available fromthe prognostic cloud scheme. Rainfall is not a standardparameter and is diagnosed as the sum of all precipi-tation processes in the ECMWF model; thus, rainfall isthe sum of large-scale precipitation and convective pre-cipitation. These fields provide an accurate three-di-mensional description of the rainfall as it is estimatedin the ECMWF model.

c. Radiation transfer model and optical properties ofsea salt

In the radiative transfer calculations a multistreammodel using the discrete ordinate method is used (Stam-nes et al. 1988). In this study, eight streams are used.Rayleigh scattering, scattering and absorption by aero-sols and clouds, and absorption by ozone and watervapor are included in the radiative transfer scheme. Ab-sorption by ozone and water vapor are included usingthe exponential sum fitting method (Wiscombe andEvans 1977). Four spectral regions are used [see Myhreand Stordal (2001) for further details]. In the radiativetransfer scheme meteorological fields (temperature, wa-ter vapor, clouds, and surface albedo) from ECMWFdata are used, so that it is consistent with the sea saltcalculations from the Oslo CTM2.

Clouds are included on the basis of cloud liquid watercontent and cloud amount from the ECMWF data. Basedon radar observations (Hogan and Illingsworth 2000),the random cloud overlap assumption is used. The op-tical properties of the clouds are based on Slingo (1989)with an effective radius of 10 mm for clouds at pressurelevels higher than about 300 hPa and 18 mm for cloudsat lower pressures (Stephens 1978; Stephens and Platt1987). Radiative transfer calculations are performed ev-ery 3 h, taking into account the zenith angle variations,updated meteorological fields, and sea salt concentra-tions.

Optical properties (specific extinction coefficient, sin-gle-scattering albedo, and asymmetry factor) of the seasalt aerosols are calculated using Mie theory (Wiscombe1980), assuming all particles to be spherical. Mixingwith other particles is not taken into account.

The complex refractive index for dry sea salt particlesis taken from Shettle and Fenn (1979). The density ofthe dry sea salt aerosols is taken to be 2200 kg m23

(Hess et al. 1998). The size of aerosols in the calcula-tions of the optical properties is that modeled in theOslo CTM 2. Hygroscopic growth of the sea salt aero-sols is based on Fitzgerald (1975) consistent with theCTM modeling of this effect [see section 2b(2)]. Therefractive index and density for the sea salt aerosol withwater uptake are calculated using volume weighting ofsea salt and water.

1 JULY 2002 1721G R I N I E T A L .

TABLE 2. Some earlier estimates for the global flux.

Reference Total flux estimate (Tg yr21) Comment

Petrenchuk (1980) 1000 Assumes that the sea salt is homogeneously distributed and that drydeposition accounts for 10% of the deposition.

Erickson and Duce (1988) 10 000–30 000 Calculates concentrations from empirical equations. Uses one scav-enging coefficient for the whole world.

Gong et al. (1997b) 10 000 Does not take into account spume production. Averages local pro-duction rates to obtain global production rate by using a 1D modelat different sites of the world.

Tegen et al. (1997) 5900 Calculates surface layer concentrations from empirical functions,and not from production loss terms. Neglecting growth of particlesfrom absorption of water vapor in the air.

Takemura et al. (2000) 3300 Calculating first layer concentrations from empirical functions, andnot from production loss terms.

This study 6500 Uses global wind data from ECMWF together with equations fromMonahan et al. (1986) and Smith et al. (1993) to give globalproduction.

3. Resultsa. Global results/budgets

1) TOTAL FLUX

Most estimates of sea salt fluxes are between 1000and 10000 Tg yr21 (Blanchard 1985; Seinfeld and Pan-dis 1998). In Table 2, some earlier studies in are listed.The obtained flux of 6500 Tg yr21 is within the expectedrange.

Fields for the total production are given in Figs. 1aand 1b. It can be seen that the production is largestwhen the wind is high, for instance at midlatitudes.

2) REMOVAL MECHANISMS

Removal occurs by wet deposition and dry depositionas described above.

In this study, we find that dry deposition is the dom-inant removal mechanism (see Table 3).

Erickson and Duce (1988) used a simplified equationto calculate sea salt concentration as a function of windspeed only. Their calculation did not take into accountother parameters such as cloud cover, rainfall, and rel-ative humidity (since relative humidity modifies the ra-dius and thus the dry deposition). To calculate wet de-position, Erickson and Duce (1988) used a constantscavenging coefficient globally. This coefficient givesCrain/Cair, and is known to vary globally. This coefficientwas used together with global rainfall data.

Gong et al. (1997b) used a one-dimensional modelto calculate the production and loss mechanisms at dif-ferent locations around the world. They averaged andestimated the global fluxes. As can be seen from Fig.1, the production flux is not homogeneous and an av-eraging based on estimates from runs with a 1D modelis uncertain.

Fluxes for dry deposition and wet deposition foundin this study are given in Figs. 1c–1f. The figures forproduction and loss all show the same seasonal depen-dence. The loss is high in the area where production ishigh.

Erickson and Duce (1988) have given flux fields inthe same way as above for dry and wet deposition. Thefields from this study and the fields from Erickson andDuce (1988) do not differ much [even though Ericksonand Duce (1988) get a yearly flux of 10 000–30 000Tg yr21 as opposed to 6500 Tg yr21 found in this study].

3) CONCENTRATION FIELDS

The distribution of sea salt particles near the groundis shown in Figs. 1g and 1h. It can be seen that con-centrations over land are very small. It can be seen fromFigs. 1g and 1h that the concentrations are largest wherethe winds are large, for instance at midlatitudes. Theareas with high concentrations are the same as the areaswith high production (see Fig. 1).

4) BURDEN

At the surface the geographical distribution of theannual mean sea salt concentration is compared to acompilation by Koepke et al. (1997). The global aerosoldataset (GADS) in Koepke et al. (1997) is a climato-logical compilation based on observations and models.GADS is often used in satellite retrievals. In the mod-eled distribution in Fig. 2a, the highest concentrationsare found at midlatitudes, particularly in the SouthernHemisphere. High concentrations are also found nearthe equator.

In Fig. 2b, the annual mean sea salt concentration inthe lowest layer from Koepke et al. (1997) is shown.The geographical pattern is generally very similar in thetwo distributions, with the largest difference near theequator associated with higher values in the modeleddistribution. At midlatitudes the modeled concentrationsare substantially higher in the Southern Hemisphere thanin the Northern Hemisphere, whereas in Koepke et al.(1997), the maximum concentrations in the NorthernHemisphere are slightly higher than in the SouthernHemisphere.

The geographical distribution of the annual mean at-


FIG. 1. Global fields: Production in mg m22 day21 for (a) Jan and (b) Jul, dry deposition in mg m22

day21 for (c) Jan and (d) Jul, wet deposition in mg m22 day21 for (e) Jan and (f ) Jul, and concentration[mg(Na) m23] for (g) Jan and (h) Jul.

mospheric burden of sea salt particles from our modeland from Koepke et al. (1997) is shown in Figs. 2c and2d. In Koepke et al. (1997), the surface concentrationis given and the vertical distribution is calculated usingexponential profiles with a scale height (Hess et al.1998), which is the same over all maritime regions.

Surface concentration depicted in Fig. 2 show goodagreement between GADS and our model. However thegeographical distribution of the total sea salt burden

differ substantially between the model and GADS. Inthe model, burden maximum is at low latitudes, whereasthe burden from the Koepke et al. (1997) dataset has amaximum at midlatitudes. In the latter pattern, the dis-tribution of the burden is actually reflecting the surfaceconcentration since the scale height is the same overocean. At midlatitudes in the Southern Hemisphere thesea salt concentration is about 50% higher in the modelthan in the Koepke dataset at the surface level. On the

1 JULY 2002 1723G R I N I E T A L .

FIG. 2. Yearly average sea salt concentration in surface layer [given in mg(sea salt) m23] for (a) the Oslo CTM2data and (b) the Koekpe et al. (1997) dataset, and yearly average burden given in g m 22 for (c) the CTM2 data and(d) Koepke data.

TABLE 3. Budget for removal mechanisms (percent of mass).

Removal mechanism

Gong et al.(1997b)

(%)

Erickson andDuce (1998)

(%)This study

(%)

Dry depositionSubcloud wet depositionIn cloud wet depositionTotal wet deposition

6633

134

70

30

80

20

other hand the burden of sea salt is about 50% higherin the Koepke dataset than in the model at midlatitudesin the Southern Hemisphere.

Transport and deposition processes are important forthe calculated vertical profiles. In particular wet depo-sition is important at midlatitudes. This shows that theuse of scale heights to estimate the vertical distributionof sea salt particles is a strong simplification.

Global and annual mean burden of sea salt is 12 and15 mg m22 in our model and in the Koepke et al. (1997)dataset, respectively. In Tegen et al. (1997) the sea saltburden was calculated to 22.4 mg m22, whereas in Hay-wood et al. (1999), the burden in the high case was 36.8and in the low case 7.5 mg m22. Takemura et al. (2000)calculated a sea salt burden of 11.0 mg m22.

In comparison, global average burden of anthropo-genic sulfate aerosols range from 1.7 to 3.2 mg m22

(Myhre et al. 1998) and total burden of mineral dust fortwo datasets were 35 mg m22 and 110 mg m22, re-spectively (Myhre and Stordal 2001).

5) RIVER RUNOFF

Petrenchuk (1980) estimates that the river runoff ofsea salt is 300–400 Tg yr21. This should equal what isdeposited over land (as nothing is assumed accumulatedon land). Using yearly river data of 35.6 3 103 km3

and an average chlorine concentration in rivers of 6.4–7.8 mg L21, Petrenchuk (1980) finds 230–280 Tg yr21

of chlorine. This should be approximately the same asthe chlorine deposited over land from sea salt aerosols.Our model deposits 161 Tg yr21 of sea salt over land,corresponding to 88 Tg yr21 of chlorine. Given the un-certainties and assumptions made in such a comparisonthe estimates compare reasonably well. The comparisonassumes that river concentration of chlorine is constant,that we have a steady state in chlorine at land, that nochlorine evaporates from the aerosols, and that the onlyway the chlorine can escape from land to ocean is byrivers.

b. Local concentrations and distributions

In this section, local concentrations and size distri-butions are shown. Some concentrations have been com-pared to measurements, and the size distributions arediscussed based on data found in literature. The con-centrations [in mg(Na) m23] are calculated by assumingthat the weight percent of Na in sea salt is 0.3061 (Sein-feld and Pandis 1998).


FIG. 3. Concentrations in mg(Na) m23 at (left) Barbados and (right) Bermuda for 1996.

FIG. 4. Concentrations in mg(Na) m23 at Heimaey, Iceland, for1996 and concentrations in grid box to the east of Heimaey. The gridbox in the Oslo CTM2 containing the Heimaey station has a largepercentage of land.

1) TIME SERIES 1996

Monthly average measurements for 1996, were ob-tained from D. Savoie at the University of Miami Ro-senstiel School of Marine and Atmospheric Science(2000, personal communication). The CTM is run forthe same year with 2-month spinup.

Figures 3 and 4 show the concentrations at severalsites of the world.

The measurements for Bermuda and Barbados fit verywell with measured data for the mass of sea salt aerosols.Both of these stations lie in areas where neither theproduction or any of the removal processes have largegradients (see Fig. 1). Both Bermuda and Barbados havean ‘‘average’’ production, dry and wet deposition, whichmeans that the gridbox mean would not be an unrealisticapproximation for the actual concentration at the station.

The fit is not so good for Heimaey, Iceland. Therecan be several reasons for the discrepancy. First, thegrid containing Heimaey consists of 50% land in themodel. This means that if the measurements are doneclose to the sea, the modeled concentration is an average

of sea–land concentration that will be too low for a goodfit. The grid cell will have emissions only half of whichit would have had if the whole cell was covered by sea.Local conditions that will not be captured by the modelcan also play a part. It can be seen that the argumentof 50% of the grid box being land plays a part as con-centrations are significantly higher in the grid box tothe east of Heimaey with 100% sea cover (see Fig. 4).In both cases, the model fails to reproduce the maximummeasured in December. This maximum could be due tolocal conditions at Heimaey.

2) TIME SERIES 5-YR AVERAGE

Gong et al. (1997b) give time series at some morestations as 5-yr averages. It is useful to compare timeseries modeled for 1996 with these measurements toverify that the model can predict the right order of mag-nitude and the right seasonal variations for the concen-trations. Figure 5 shows the modeled concentrationcompared to measured 5-yr averages.

The size order and seasonal variations are reproducedwell for these two stations. It is important to notice thatthe Oslo CTM2 model manages to reproduce the min-imum for the Northern Hemisphere summer at Hawaii.Gong et al. (1997b) point out that since they have awind maximum (high production) and a precipitationminimum (low scavenging) for this period, their modelshould have problems reproducing the concentrationminimum. Hawaii is situated in the Pacific in an areawith large gradients in wind speed. Using winds fromECMWF, we do not get a production minimum in sum-mer for this station, but a rather constant productionthroughout the year. This underlines the need for real-istic meteorological data in combination with realisticproduction, loss, and transport in the study of sea saltparticles.

The Oslo CTM2 fails to reproduce the winter max-imum over Mace Head, Ireland. The model underesti-mates the concentrations by 24% in January, 44% in

1 JULY 2002 1725G R I N I E T A L .

FIG. 5. Concentrations in mg(Na) m23 (5-yr averages) for (left) Hawaii, United States, and (right)Mace Head, Ireland, compared to model results for 1996.

TABLE 4. Yearly averaged standard deviation from measuredvalues for different stations.

Station Std dev (%)

BarbadosBermudaHawaiiMace Head

15.915.715.627.0

February, and 48% in March. This could be a result oflocal conditions similar to those discussed above forHeimaey. It can also be seen that the modeled results(1996) show more variation than the 5-yr mean curvethat is smoother. It is normal that calculations for oneparticular year deviate from a 5-yr average.

In Table 4, some average standard deviations for themonthly mean time series are shown. The deviation iscalculated for each month and then averaged for thewhole year. Heimaey is not included for reasons dis-cussed above.

3) WEEKLY AVERAGES

At some stations, we have compared weekly averagesmeasured from 1996 to weekly averages from the mod-el. This should give an indication if the dependence onmeteorology (e.g., wind, rainfall) is correct. The vari-ations should correspond not only seasonally, but alsoon shorter timescales.

Figure 6 shows weekly averages for two stations dur-ing the first few weeks of 1996. The modeled variationsseem to follow the observations nicely. The model con-centrations covary with the measured concentrations.

At the Reunion Island site we have quite large weeklyvariations, something that is reproduced by the model.The model generally overestimates the concentrations atthis site with a minimum overestimation of 18% in week6 and a maximum overestimation of 470% in week 5.

At the Chatman Island site where the measurements

show low variance, the model shows low variance aswell. At this site, the model generally overestimates theconcentrations. The model overestimates by 52% inweek 1, 60% in week 4, and 76% in week 8, but forsome weeks, the model underestimates the concentra-tions, such as in week 3 (12%) and week 5 (31%).

4) SIZE DISTRIBUTIONS

The mass size distributions at selected stations areshown in Fig. 7.

The size distribution is expected to be lognormal. Inthis case, we get a bimodal distribution with anothermaximum at about r 5 10 mm because of the contri-bution of spume particles to the production.

Erickson and Duce (1988) propose that the distri-bution fitted to lognormality should be distributed aboutthe radius r 5 0.422u 1 2.12 where u is surface windspeed. This would mean that for a surface wind speedof 10 m s21, the distribution would be lognormally dis-tributed around r 5 6 mm. The actual measured massmedian radius from the work of Erickson and Duce(1988) varied between 3.5 and 7.5 mm. They do notgive any standard variation. Seinfeld and Pandis (1998)also propose that marine background aerosols are log-normally distributed around r 5 6 mm. The Oslo CTM2approximately reproduces this size distribution.

5) VARIATION WITH HEIGHT

Mass concentrations decrease with height. The lossterms that lead to this are the falling velocity and thescavenging. The variation of total mass with height isshown in Fig. 8.

We do not get significant transport of sea salt massto higher altitudes in the troposphere. The zonal meanor aerosol mass is high at latitudes with high wind andhigh ocean fraction. The washout process removes anysea salt that is transported to higher altitudes, therefore


FIG. 6. Weekly averages for surface level concentrations in mg(Na) m23 at (left) Chatman Islandand (right) Reunion Island for 1996.

FIG. 7. Modeled size distribution for (left) Barbados and (right) Mace Head, Ireland. Radius is at80% relative humidity.

no significant amount of sea salt mass is found over 750hPa.

The high concentrations at 608S is a consequence ofhigh wind speeds, and small land fraction at this latitude.

The mass of different size bins vary differently withheight. The larger particles have a higher falling veloc-ity, and the concentration of these particles decreasemore rapidly with altitude than the mass of the smallerparticles. Figure 9 shows the concentration as a functionof height at two stations.

At Heimaey in January, where washout is expectedto be efficient, we see a more rapid decrease in con-centrations with height than at Hawaii where washoutis expected to be less efficient. The vertical profile isalso influenced by different production (of different siz-es of particles) or differences in transport.

We have made a global average column of the mod-eled results. The concentrations have been approximatedto a formula of the form C 5 C0e2Z/H where C is con-centration in mg m23, Z is altitude, and H is scale height(both measured in km). There are two distinguishedregimes with H 5 1 km below approximately 2 km and

H 5 5.5 km above approximately 2 km. For comparison,Koepke et al. (1997) used scale heights of 1 and 8 kmfor the two regions. This means that our model containsless sea salt mass in the free troposphere.

c. Radiative impact

In this section we will use the term radiative forcingof the radiative impact of sea salt. Usually radiativeforcing is used for an external perturbation (anthropo-genic or natural perturbation such as solar irradiationvariation or aerosols from volcanic eruptions; Houghtonet al. 1996); however, here we use the term radiativeforcing for a natural component.

Results are performed with clouds included and forclear sky (clouds excluded in the calculations). Resultsfor clear sky are presented first as these are most com-parable to the results from Haywood et al. (1999) andBoucher and Tanre (2000), which are combinations ofsatellite observation and modeling. However, note thatsatellite observations are only performed for clear sky

1 JULY 2002 1727G R I N I E T A L .

FIG. 8. Zonal averages in mg(Na) m23 of total sea salt mass for (left) Jan and (right) Jul.

FIG. 9. Variation of concentrations with height.

when clouds are not present and therefore not identicalto our clear-sky assumption.

Figure 10a shows the annual mean geographical dis-tribution of the clear-sky radiative forcing due to seasalt. In general the pattern of the clear-sky radiativeforcing is very similar to the aerosol optical depth at500 nm (not shown). The maximum forcing is around608S, with a secondary maximum at 208 in each hemi-sphere. The aerosol optical depth shows a somewhatsmaller difference between these maxima than the forc-ing. This can be explained by noting that the radiativeforcing due to particles is largest at high solar zenithangles (Haywood and Shine 1997).

A similar pattern as that shown in Fig. 10a is foundin Haywood et al. (1999) for sea salt; however, in theiranalysis, few observations around 608S exist. In Bou-cher and Tanre (2000) all aerosol components are in-cluded; however over ocean in nonmineral regions aswell as far from land, similar radiative impact of aero-sols can be found. The radiative effect of sea salt overland is very small in accordance with the low concen-trations estimated.

In Fig. 10b, clouds are included. This reduces themagnitude of the forcing as for other scattering aerosols.At midlatitudes the radiative forcing due to sea salt issubstantially weakened as the cloud cover in these re-gions is very high. In the more cloud-free regionsaround 208 the impact of clouds is smaller, therefore thestrongest radiative forcing is in these regions. Cloudsat midlatitudes therefore strongly reduce the radiativeimpact of sea salt both by wet deposition and scatteringof sunlight.

The global mean radiative forcing due to sea salt is21.1 W m22 for a cloudy atmosphere, and 22.2 W m22

for clear sky. The clear-sky estimate is between the highand low estimates of sea salt in Haywood et al. (1999).Boucher and Tanre (2000) estimated a global meanclear-sky radiative forcing over ocean of about 25.5 Wm22 including natural and anthropogenic aerosols. Toinvestigate the forcing due to sea salt within the cloudswe used the procedure described in Haywood and Shine(1997). Neglecting sea salt aerosols in regions withclouds (in the whole column) gives a forcing of 20.7W m22. This estimate is one-third of the clear-sky ra-


FIG. 10. Yearly averaged direct radiative forcing due to sea salt in (a) clear-sky conditions and(b) taking clouds into account.

diative forcing, a result that shows that clouds are veryoften present in regions with sea salt. We find a forcingof 20.4 W m22 inside cloudy regions or 35% of theradiative forcing.

Normalized forcing (global mean radiative forcingdivided by the global mean burden) has been calculatedby many investigators for sulfate aerosols (see Myhreet al. 1998, and references therein). It ranges from about2550 to 2125 W g21. For mineral dust the normalizedforcing range from 211 to 14 W g21 in Myhre andStordal (2001). In the calculations performed in thisstudy the normalized forcing for sea salt is 288 W g21.The lower normalized forcing for sea salt aerosols thanfor sulfate is mainly due to larger sizes of the sea saltaerosols. Mineral aerosols also have large sizes like thesea salt particles. In addition they absorb solar radiationand have a significant thermal infrared radiative forcingleading to small normalized forcing for mineral aero-sols.

We have performed two additional calculations inwhich we have reduced the number of size bins from16 to 8 and 4. Differences in results when the numberof size bins are reduced may have several origins. This

will influence the dry deposition as the dry depositionshows a strong, nonlinear variation with the radius(Seinfeld and Pandis 1998), and the total burden ischanged. Further, the optical properties may change ei-ther as a result of change in the calculated size distri-bution or simply in the averaging procedure in the cal-culation of the optical properties. Reducing the numberof size bins from 16 to 8 decreases the radiative forcingonly by 4%. However, as the total burden is slightlyincreased (2%), the normalized forcing is reduced by6%. When four size bins are used, larger changes occur.The radiative forcing is 7% stronger, whereas the totalburden is 23% higher, giving a normalized forcing thatis 13% lower. The scattering efficiency for aerosols isstrongly dependent on the size of the aerosols. In thevisible region the scattering efficiency has a strong max-imum for aerosols with radius around 0.5 mm (see, e.g.,Seinfeld and Pandis 1998). This nonlinear variation inthe scattering efficiency leads to an underestimation ofthe radiative forcing for 4 and 8 size classes comparedto 16 size classes. However, larger differences in theburden is found, mainly because of averaging the drydeposition over larger size intervals.

1 JULY 2002 1729G R I N I E T A L .

4. Summary

A global 3D CTM (the Oslo CTM2) has been usedto simulate the concentration of sea salt particles in theatmosphere. The distribution is determined by produc-tion (generation by wind), transport (advection and con-vection), wet and dry deposition, and growth by con-densation. This study calculates the fluxes of sea salt asa function of wind speed. Earlier studies have eithercalculated sea salt concentrations in the boundary layerdirectly from empirical wind speed correlations or usedless-sophisticated models (e.g., 1D models).

Monthly averaged distributions have been comparedwith measurements from Hawaii, Barbados, Bermuda,Ireland, and Iceland, and with weekly averages fromChatman Island and the Reunion Island. The calculatedsurface concentrations vary between 0 and 16 mg(Na)m23 and fit well with the measured values.

The global flux of sea salt is calculated to be 6500Tg yr21. Earlier estimates lie between 1000 and 10 000Tg yr21. Average burden of sea salt is calculated to 12mg m22, which is within the range of earlier estimatesthat are between 7.5 and 36 mg m22. Global averageradiative forcing of sea salt is estimated to 21.1 W m22

when the effect of clouds is taken into account and 22.2W m22 when the effect of clouds is ignored. It is im-portant to take the effect of clouds into account in aconsistent way using the same meteorology to calculateemissions, transport, and radiative forcing, since the seasalt concentrations often are high in areas with extensivecloud cover.

Acknowledgments. We are grateful to Prof. David Sa-voie for providing his measurements of sea salt particles.They were important for validating the model results.This work has been sponsored by the Norwegian Re-search Council through Grant 139810/720 (CHEM-CLIM).

REFERENCES

Bergstrom, R., and P. Russel, 1999: Estimation of aerosol directradiative effects over the mid-latitude North Atlantic fromsatellite and in situ measurements. Geophys. Res. Lett., 26,1731–1734.

Blanchard, D., 1985: The oceanic production of atmospheric sea salt.J. Geophys. Res., 90, 961–963.

Boucher, P., and D. Tanre, 2000: Estimation of the aerosol perturbationto the earth’s radiative budget over oceans using polder satelliteaerosol retrievals. Geophys. Res. Lett., 27, 1103–1106.

Erickson, D. J., and R. A. Duce, 1988: On global flux of atmosphericsea salt. J. Geophys. Res., 93, 14 079–14 088.

Fitzgerald, J. W., 1975: Approximation formula for the equilibriumsize of an aerosol particle as a function of its dry size andcomposition and the ambient relative humidity. J. Appl. Meteor.,14, 1044–1049.

Gong, S. L., L. A. Barrie, and J.-P. Blanchet, 1997a: Modeling sea-salt aerosols in the atmosphere. 1: Model development. J. Geo-phys. Res., 102, 3805–3818.

——, ——, J. M. Prospero, D. L. Savoie, G. P. Ayers, J.-P. Blanchet,and L. Spacek, 1997b: Modeling sea-salt aerosols in the atmo-

sphere. 2: Atmospheric concentration and fluxes. J. Geophys.Res., 102, 3819–3830.

Haywood, J., and K. Shine, 1997: Multi-spectral calculations of thedirect radiative forcing of tropospheric sulphate and soot aerosolsusing a column model. Quart. J. Roy. Meteor. Soc., 123, 1907–1930.

——, and O. Boucher, 2000: Estimates of the direct and indirectradiative forcing due to tropospheric aerosols: A review. Rev.Geophys., 38, 513–543.

——, V. Ramaswamy, and B. Soden, 1999: Tropospheric aerosolclimate forcing in clear sky satellite observation over the oceans.Science, 283, 1299–1303.

Hess, M., P. Koepke, and I. Schult, 1998: Optical properties of aero-sols and clouds: The software package OPAC. Bull. Amer. Me-teor. Soc., 79, 831–844.

Hogan, R. J., and A. J. Illingsworth, 2000: Deriving cloud overlapstatistics from radar. Quart. J. Roy. Meteor. Soc., 126, 2903–2909.

Houghton, J. T., L. G. Meira Filho, B. A. Callander, N. Harris, A.Kattenberg, and K. Maskell, Eds.,1996: Climate Change 1995:The Science of Climate Change. Cambridge University Press,572 pp.

King, M., Y. Kaufman, D. Tanre, and T. Nakajima, 1999: Remotesensing of tropospheric aerosols from space: Past, present, andfuture. Bull. Amer. Meteor. Soc., 80, 2229–2259.

Koepke, P., M. Hess, I. Schult, and E. Shettle, 1997: Global aerosoldata set. Max Planck Institute for Meteorology Tech. Rep. 243,Hamburg, Germany, 50 pp.

Monahan, E., D. Spiel, and K. Spiel, 1986: Oceanic Whitecaps. Rei-del.

Myhre, G., and F. Stordal, 2001: Global sensitivity experiments ofthe radiative forcing due to mineral aerosols. J. Geophys. Res.,106 (D16), 18 193–18 204.

——, ——, K. Restad, and I. Isaksen, 1998: Estimates of the directradiative forcing due to sulfate and soot aerosols. Tellus, 50B,463–477.

Petrenchuk, O. P., 1980: On the budget of sea salts and sulphur inthe atmosphere. J. Geophys. Res., 85, 7439–7444.

Prather, M. J., 1986: Numerical advection by conservation of second-order moments. J. Geophys. Res., 91, 6671–6681.

Pryor, S. C., and L. L. Sorensen, 2000: Nitric acid–sea salt reactions:Implications for nitrogen deposition to water surfaces. J. Appl.Meteor., 39, 725–731.

Seinfeld, J. H., and S. N. Pandis, 1998: Atmospheric Chemistry andPhysics, From Air Pollution to Climate Change. John Wiley andSons, 1326 pp.

Shettle, E., and R. Fenn, 1979: Models for the aerosols of the loweratmosphere and the effects of humidity variations on their opticalproperties. AFGL Tech. Rep. TR-79-0214.

Slingo, A., 1989: A GCM parameterization for the shortwave radi-ative properties of water clouds. J. Atmos. Sci., 46, 1419–1427.

Smith, M. H., P. M. Park, and I. E. Consterdine, 1993: Marine aerosolconcentration and estimated fluxes over sea. Quart. J. Roy. Me-teor. Soc., 119, 809–824.

Stamnes, K., S. Tsay, W. Wiscombe, and K. Jayaweera, 1988: Anumerically stable algorithm for discrete-ordinate method ra-diative transfer in multiple scattering and emitting layered media.Appl. Opt., 27, 2502–2509.

Stephens, G., 1978: Radiation profiles in extended water clouds. I:Theory. J. Atmos. Sci., 35, 2111–2122.

——, and C. Platt, 1987: Aircraft observations of the radiative andmicrophysical properties of stratocumulus and cumulus cloudfields. J. Climate Appl. Meteor., 26, 1243–1269.

Sundet, J. K., 1997: Model studies with a 3-d global CTM usingECMWF data. Ph.D. thesis, University of Oslo, Oslo, Norway,102 pp.

Takemura, T., H. Okamoto, Y. Maruyama, A. Numaguti, A. Higurashi,and T. Nakajima, 2000: Global three-dimensional simulation of


aerosol optical thickness distribution of various origins. J. Geo-phys. Res., 105, 17 853–17 873.

Tegen, I., P. Hollrig, M. Chin, I. Fung, D. Jacob, and J. Penner, 1997:Contribution of different aerosol species to the global aerosolextinction optical thickness: Estimates from model results. J.Geophys. Res., 102, 23 895–23 915.

Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulusparameterization on large scale models. Mon. Wea. Rev., 117,1779–1800.

Wiscombe, W., 1980: Improved mie scattering algorithms. Appl. Opt.,19, 1505–1509.

——, and J. Evans, 1977: Exponential sum fitting of radiative trans-mission functions. J. Comput. Physiol., 24, 416–444.

Zhang, Y., C. Seigneur, J. Seinfeld, M. Jacobson, S. Clegg, and F.Binkowski, 2000: A comparative review of inorganic aerosolthermodynamic equilibrium modules: Similarities, differences,and their likely causes. Atmos. Environ., 34, 117–137.

Modeling the radiative impact of mineral dust during

the Saharan Dust Experiment (SHADE) campaign

Gunnar Myhre,1,2 Alf Grini,1 James M. Haywood,3 Frode Stordal,1,2 Bernadette Chatenet,4

Didier Tanre,5 Jostein K. Sundet,1 and Ivar S. A. Isaksen1

Received 24 May 2002; revised 18 September 2002; accepted 20 January 2003; published 30 July 2003.

[1] The Oslo chemical transport model (Oslo CTM2) is driven by meteorological data tomodel mineral dust during the Saharan Dust Experiment (SHADE) campaign in September2000. Model calculations of the optical properties and radiative transfer codes are used toassess the direct radiative impact in the solar and terrestrial regions of the spectrum. Themodel calculations are compared to a wide range of measurements (satellite, ground-based,and aircraft) during the campaign. The model reproduces the main features during theSHADE campaign, including a large mineral dust storm. The optical properties and thevertical profiles are in reasonable agreement with the measurements. There is a very goodagreement between the modeled radiative impact and observations. The strongest localsolar radiative impact wemodel is around�115Wm�2. On a global scale the radiative effectof mineral dust from Sahara exerts a significant negative net radiative effect. INDEXTERMS:

0305 Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0360 Atmospheric

Composition and Structure: Transmission and scattering of radiation; 3337 Meteorology and Atmospheric

Dynamics: Numerical modeling and data assimilation; 3359 Meteorology and Atmospheric Dynamics:

Radiative processes; KEYWORDS: aerosols, single scattering albedo, transport model, aircraft measurements

Citation: Myhre, G., A. Grini, J. M. Haywood, F. Stordal, B. Chatenet, D. Tanre, J. K. Sundet, and I. S. A. Isaksen, Modeling the

radiative impact of mineral dust during the Saharan Dust Experiment (SHADE) campaign, J. Geophys. Res., 108(D18), 8579,

doi:10.1029/2002JD002566, 2003.

1. Introduction

[2] Among the major components contributing to thedirect aerosol effect mineral dust is the most uncertain.Even the sign of the radiative forcing due to mineral dust isunresolved [Intergovernmental Panel on Climate Change(IPCC), 2001]. The main uncertain factors are the degree ofabsorption of solar radiation (an important part is therefractive index), the influence of mineral dust on longwaveradiation, the abundance of mineral dust in the atmosphere,and in particular the human influence on this abundance.[3] Mineral dust has a cooling effect over ocean in the

absence of clouds. Haywood et al. [2001] measured a strongradiative impact (or radiative effect) during a dust event.However, over brighter surfaces (as over desert and aboveclouds) the radiative impactwill be strongly altered.Haywoodand Shine [1995] andHansen et al. [1997] showed that singlescattering albedo is a critical factor in determining the signof the solar radiative impact for various surface albedos.The longwave radiative impact of mineral dust will always

be positive, but the magnitude depends strongly on the sizeof the particles, the refractive indices and the altitude. There-fore the net radiative impact (sum of solar and longwave)exhibits large regional variations, and this explains whythe global mean is difficult to estimate. Several studieshave investigated the radiative impact of mineral dust [Tegenet al., 1996; Sokolik and Toon, 1996; Myhre and Stordal,2001; Woodward, 2001; Weaver et al., 2002] with a widerange in the results.[4] Tegen et al. [1996] and Sokolik and Toon [1996]

estimated that the human contribution to the mineral dustabundance can be as large as 30–50%. These numbers arehighly uncertain and it is an unresolved question whetherthis is a direct climate forcing mechanism or a climatefeedback (see discussion by Myhre and Stordal [2001]).[5] The Saharan Dust Experiment (SHADE) took place in

September 2000 on the west coast of Africa to improveknowledge about the radiative impact of mineral dust using awide range of measurements [Tanre et al., 2003]. This studyuses meteorological data to model the production and trans-port of mineral dust during the campaign and compares theresults against various in-situ and remotely sensed airborneand surface-based measurements and calculates the directradiative impact for a larger region surrounding the Sahara.

2. Models

2.1. Transport Model

[6] Oslo CTM2 is an off-line chemical transport/tracermodel (CTM) that uses precalculated transport and physical

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. D18, 8579, doi:10.1029/2002JD002566, 2003

1Department of Geophysics, University of Oslo, Oslo, Norway.2Norwegian Institute for Air Research, Kjeller, Norway.3Met Office, Bracknell, UK.4Laboratoire Interuniversitaire des Systemes Atmospheriques, Univer-

sites Paris VII et Paris XII, Centre Nationale de Recherche Scientifique,Creteil, France.

5Laboratoire d’Optique Atmospheriques, Centre Nationale de Re-cherche Scientifique, U.S.T. de Lille, Villeneuve d’Ascq, France.

Copyright 2003 by the American Geophysical Union.0148-0227/03/2002JD002566$09.00

SAH 6 - 1

fields to simulate tracer distributions in the atmosphere. Themodel represents the global troposphere and is three-dimen-sional with the model domain reaching from the ground upto 10 hPa for the current data sets. In the horizontal andvertical the model has a variable resolution that is deter-mined by the input data provided. In this study we useforecast data from the European Center for Medium-RangeWeather forecast (ECMWF) with 1.875� � 1.875� degreeshorizontal resolution and with 40 levels in the vertical witha resolution of about 600 meters in most of the troposphere.[7] The second-order moment method is used for advec-

tion [Prather, 1986]. Convection is based on the Tiedktemass flux scheme [Tiedkte, 1989; Sundet, 1997], wherevertical transport of species is determined by the surplus/deficit of mass flux in a column. Dry and wet deposition aretreated as in the work of Grini et al. [2002a].[8] The forecast data are constructed from merged fore-

casts where each forecast is run for 36 hours with the first12 hours discarded as ‘‘spin-up.’’ The data are stored atthree hours interval for August and September 2000. Mostof the data that are diagnosed from the ECMWF model arestandard model output (cloud cover, surface properties etc.)and is thus the state of the art parameterizations of therelevant processes. Alas, since the CTM is off-line there willbe no feedback from the CTM to the data that are diagnosedin the ECMWF model. Data that are not part of theECMWF diagnostic (convective fluxes, rain fall) have beendiagnosed explicitly with the built in diagnostic system inthe ECMWF model, and the physical properties of the fieldsare described by Tiedkte [1987].

2.2. Dust Production

[9] Production of dust is parameterized using the formula

production ¼ const � u2* u* � u*;tr

� �

; ð1Þ

where production is in kgm�2s�1 and u* is friction velocity(see definition in the work of Stull [1988]) in m/s. Theconstant used is 1.7 10�5 kg s2m�5. The threshold frictionwind speed (u*,tr) is the minimum friction wind assumed togive dust emission.[10] Different landscapes have different threshold winds

for dust emissions, normally between 6 m/s and 20 m/s [see,e.g., Chomette et al., 1999]. We have used global data setsto describe the vegetation. An average procedure wasdeveloped using the data set of Olson et al. [1983],Matthews [1983], Wilson and Henderson-Sellers [1985],Ramankutty and Foley [1999], and SARB http://www-surf.larc.nasa.gov/surf/ ). The lowest threshold wind of6 m/s was used where all data sets agreed on pure sanddeserts. The highest threshold wind of 20 m/s was used inareas where all data sets agreed on shrub. To convertthreshold wind (Utr) speeds to threshold friction windspeeds, we used the formula u*,tr = 0.035 � Utr (based onthe work of Nickling and Gillies [1987, Table 3]). Theconstant used in equation (1) gives the same production asin the work of Tegen and Fung [1994] for a wind speed of10 m/s and a threshold wind speed of 6.5 m/s.[11] For areas with large roughness elements, the friction

wind velocity is very high as the friction velocity increaseswith roughness length. In arid areas with large roughness,for instance over the Moroccan Atlas mountains, the

ECMWF data have a very large friction wind speed. Insuch areas, we use the formula production = CU2 (U � Utr)with C equal to 0.7 mg s2m�5 as proposed by Tegen andFung [1994]. For areas with large roughness elements, largewind friction speeds reflect large momentum flux to thesurface because of these elements. It will not reflect themomentum appropriate for dust production (we assume thatthe dust production takes place at smooth surfaces, and itshould thus not be influenced by large scale roughnesselements). In our model, the threshold wind speeds do notchange with surface roughness, but only with vegetationcover. For a further discussion on use of wind or frictionwind in dust modeling, see Liu and Westphal [2001].[12] The dust flux is given a size distribution which is

lognormal with mass median diameter Dg,mass = 2.5 mmand standard deviation s = 3.2 as given by Shettle [1984].Note that this size distribution is given in the sourceregions of the mineral aerosols and thereafter atmosphericprocesses alter the size distribution. The calculations areperformed with eight size bins. Several studies [Guelle etal., 2000; Schultz et al., 1998] have argued that s should becloser to 2. However, we chose to keep the original,broader size distribution and leave out the small and largemodes given by Shettle [1984], giving better agreementwith observations, particularly the spectral variation in theaerosol optical depth. The size distribution at the source isdifficult to establish due to the fact that only limitedobservations have been made. Calculating the size distrib-uted flux of dust aerosols is a complicated task, e.g., sincesoil characteristics are not available for large scale model-ing and since wind speeds interact to give the sizedistributed flux. Some studies have made approaches toparameterize these processes [Lu and Shao, 1999; Shao,2001; Alfaro and Gomes, 2001, Grini et al., 2002b], butmuch work remains before the size distributed fluxes ofdust can be modeled realistically.[13] The soil moisture makes the soil more difficult to

erode as the water increases the interparticle capillary forcesin the soil. Fecan et al. [1999] proposed relationships todeal with the increase in threshold wind speed with increas-ing soil water content. The soil moisture from the ECMWFmodel is not believed to be realistic enough to be used tocalculate the increase in threshold wind because in theECMWF model there is no evaporation from soils withhumidity lower than a prescribed, globally uniform perma-nent wilting point of 0.171 m3/m3. This value is much largerthan the values discussed by Fecan et al. [1999] (P. Viterbo,personal communication, 2001). To deal with the problem,we chose to set dust emission to zero for the next two dayswhenever rainfall is higher than 2 mm in 24 hours in a gridsquare. similar to the approach of Claquin [1999]. Oursimple dust production formulation is a source for uncer-tainty in the model, but this formulation will capture thelargest dust outbreaks.

2.3. Dust Optical Properties

[14] The specific extinction coefficient, the single scat-tering albedo, and the asymmetry factor are modeled usingMie scattering theory. In Mie scattering theory sphericalparticles are assumed and particle size and refractive indexneed to be specified. Haywood et al. [2003] show that themineral dust aerosols are not spherical, butMishchenko et al.

SAH 6 - 2 MYHRE ET AL.: RADIATIVE IMPACT OF MINERAL DUST

[1997] show that this simplification introduces only a modestuncertainty in calculation of radiative fluxes. The size distri-bution is explicitly modeled in the transport model andadopted in the optical and radiative calculations. The refrac-tive indices for mineral dust are highlighted as the single mostimportant factor for the large uncertainty in the radiativeimpact of mineral dust [Myhre and Stordal, 2001]. Kaufmanet al. [2001] show that the single scattering albedo at solarwavelengths is much higher and that the imaginary part of therefractive index is substantially lower than in previousstudies. We have chosen to use refractive indices in the solarspectrum from the Dubovik retrieval from the Sun photom-eter at Cape Verde. Real refractive index of 1.48, 1.47, 1.43,at 1.42 at 440, 670, 870, and 1020 nm, respectively andimaginary refractive index of 0.0014, 0.0011, 0.0011, and0.0011 at 440, 670, 870, and 1020 nm, respectively areadopted. In the thermal infrared spectrum we use refractiveindexes from Fouquart et al. [1987] as these compare mostfavorably with the detailed measurements and modelingpresented by Highwood et al. [2003]. The density of themineral aerosols, used in the calculations of the specificextinction coefficient, are taken to be 2600 kgm�3 [Hess etal., 1998].

2.4. Radiative Transfer Model

[15] We apply separate longwave and shortwave multi-stream models in the radiative transfer calculations usingthe discrete-ordinate method [Stamnes et al., 1988]. In thecalculations in this study eight streams are used. The solarradiative transfer model includes radiative effects of aero-sols, clouds, Rayleigh scattering, and the exponential sumfitting method [Wiscombe and Evans, 1977] is used toaccount for absorption by gases. There are 4 bands in thesolar spectrum (see Myhre et al. [2003] for further details).The longwave scheme takes into account absorption bygases and clouds in addition to scattering and absorption byaerosols. The exponential sum fitting method [Wiscombeand Evans, 1977] is used to account for absorption by watervapor, carbon dioxide, and ozone. Five spectral regions areused in the longwave spectrum (see Myhre and Stordal[2001] for further details).[16] Our calculations are performed in T63 spatial reso-

lution and 40 vertical layers. The meteorological data oftemperature, water vapor, clouds used in the radiativetransfer calculations during the campaign period are de-scribed in section 2.1. Within each model grid column weperform separate radiative transfer calculations in clear skyand cloudy regions to establish an overall radiative flux forthe column. Hogan and Illingworth [2000] use radar obser-vations to show that the random cloud overlap is a satis-factory assumption to estimate the total cloud cover. In thismodel version with 40 vertical layers we combine therandom cloud overlap (to describe effect of clouds indifferent height regions) and the maximum cloud overlapassumption (for nearby levels) to estimate a realistic totalcloud cover. The optical properties of clouds in the solarregion are calculated using the procedure described in thework of Slingo [1989] with effective radius of 10 mm forlow clouds and 18 mm for high clouds [Stephens, 1978;Stephens and Platt, 1987]. In the thermal infrared spectrumnonscattering clouds are assumed with absorption coeffi-cients dependent on the altitude of the clouds [see Myhre

and Stordal, 2001]. There is no interaction between themineral dust aerosols and the cloud microphysics in themodel.[17] The surface albedo is modeled as a function of solar

zenith angle and solar spectral region. Over ocean the surfacealbedo dependence on solar zenith angle is modeled accord-ing to Glew et al. [2003], whereas over land as in the work ofBriegleb et al. [1986]. Surface albedo over land is based onvegetation data from Ramankutty and Foley [1999] withspectral albedo values from Briegleb et al. [1986].[18] We have performed radiative transfer calculations

including and excluding mineral dust. The difference be-tween these two simulations at the top of the atmosphere istaken as the radiative impact of the aerosols. This is similarto the radiative forcing concept, except that we include bothanthropogenic and natural abundance of the aerosols.

3. Results

[19] The purpose of our calculations is to compare ourresults with intensive observations during the SHADE period19 to 28 September 2000. We first calculate the dust distri-bution using the CTM to simulate production of dust, thetransport, and deposit of the dust particles. These calculationsare started 1 August, in order to spin up the model. Thereafterwe calculate optical properties and radiative impact of thedust during the campaign period.

3.1. Production

[20] Figure 1 shows the modeled production of mineraldust during the SHADE campaign. A substantial diurnalvariation can be seen, with much larger production during theday. This indicates that solar heating is an important factor inthe generation of the surface winds. Further, the modelindicates an important shift in the source regions during thecampaign. During the first part of the campaign the sourceregions seem to be closer to the coast compared to the maindust storm during the period 22–24 September 2000. This isdue to a shift in the region with friction velocity above thethreshold value. The threshold friction velocity is somewhathigher at the coast than in the main source region during thecampaign, affecting the magnitude of the production. Basedon the model, the mineral dust transported over the campaignarea mainly has its origin in the western part of Sahara.

3.2. Modeled Aerosol Optical Depth

[21] The modeled aerosol optical depth at 550 nm duringthe campaign is shown in Figure 2a. It clearly reveals thelarge dust storm starting in western Sahara the 22th anddeveloping over the next days and with a westward move-ment of the mineral dust plume. Over the main campaignarea near Sal Island and Dakar it shows that AOD ismoderate during the 2 or 3 first days. Thereafter there aresome days with low AOD before the dust starts to moveover ocean the 24th. A maximum AOD over the campaignarea can be seen the 26th. The AOD weakened as themineral dust move westward, but is still not insignificant asthe dust plumes cross the Atlantic ocean in accordance withearlier measurements.

3.3. Comparison With MODIS Aerosol Optical Depth

[22] In Figure 2b the AOD from the Moderate ResolutionImaging Spectroradiometer (MODIS) [Kaufman et al.,

MYHRE ET AL.: RADIATIVE IMPACT OF MINERAL DUST SAH 6 - 3

1997; Tanre et al., 1997] aboard the Terra satellite is shownfor 25–26 September 2000. The pattern with high AODalong the coast 25 September and the mineral dust trans-ported further out over the ocean is clearly evident andrather similar to the model. Note in particular the similarshape of the dust plume for 26 September in the satelliteretrieval and the model results. However, it seems that theplume is slightly displaced eastward in the model comparedto the satellite retrievals. Further the maximum AOD issomewhat higher in the satellite retrievals than in the model,which can partly be explained by the much higher horizon-tal resolution in the satellite data. The AOD values from

MODIS compare well with AERONET observations overocean [Remer et al., 2002]

3.4. Comparison With AERONET AerosolOptical Depth

[23] Aerosol optical depths from the model are comparedto surface Sun photometers at Cape Verde and Dakar duringthe campaign period and shown in Figure 3. These instru-ments are part of the AERONET which is a global ground-based network of Sun photometers (see Holben et al. [1998]for a description). The general pattern which can be dividedin four periods is in reasonable agreement given the large

Figure 1. Production of mineral dust (10�9 kgm�2s�1) during the campaign period 19–28 September2000 at 0 and 12 UTC.


Figure 2. AOD at 550 nm. (a) Modeled during the period 19–28 September 2000 at 0 and 12 UTC.(b) MODIS data for 25–26 September between 9 and 14 UTC. Note that both panels have the same colorscale. Cape Verde is located at 16.73N and 22.94W, and Dakar is at 14.39N and 16.96W.


uncertainties in the source function of mineral dust. In thebeginning of the campaign there was moderate AODs,thereafter much lower values, followed by large AOD withmaximum values above 1, and finally more moderatevalues. The modeled AOD is lower than the maximumobserved values, in particular at Cape Verde. Further highermodeled AOD is indicated during the period from 21st to23st with very low observed values.[24] Another feature is the stronger wavelength depen-

dence of AOD in the model compared to the observations.This is most pronounced during high AOD conditions. Thisdifference most likely arises from a higher number of smallparticles in the model compared what was observed (seesection 3.5.1).

3.5. Comparison With C-130 Aircraft Data

[25] Haywood et al. [2003] describes the details of theC-130 aircraft measurements. The C-130 was equippedwith instrumentation to measure aerosol size distribution,chemical composition, optical properties, and radiative

fluxes. We concentrate on measurements made betweenCape Verde and Dakar, and mainly from the flight a797 on25 September, see the C-130 flight pattern in the work ofTanre et al. [2003].3.5.1. Size Distribution[26] Figure 4 shows a comparison of the modeled size

distribution and several lognormal size distributions fitted tothe observations. The modeled size distributions are shownfor 25 September (heavy dust loading) and 19 September(moderate dust loading). The lognormal distribution basedon the observations is grouped into flights with heavy dustloading (a796, a797, and a798) and moderate dust loading(a778, a783, and a794) [see Haywood et al., 2003]. Theobservations show higher abundance of large particles in thecases with high AOD, whereas in the model this is viceversa. The difference in the measured size distribution withdust loading can arise from differences in the source of themineral aerosols, but also during the transport of theaerosols. The model indicates that in the beginning ofthe campaign period there was a different main sourceregion compared to in the period with heaviest dust loading,which may be the reason why the observed size distributionchanges during the campaign.[27] The modeled size distributions compare relatively

well with the observed distributions for particles below1.0 mm, but the model clearly underestimates the fractionof larger aerosols. We have tried replacing the size distri-bution we have adopted in the source region with analternative distribution [Schultz et al., 1998] (a three modelognormal distribution Dg,mass = 0.011, 2.52, and 42.3 mm,respectively and s = 2.13, 2.00, and 1.89, respectively)without significant change in the number of large particles.3.5.2. Vertical Profiles[28] Figure 5 shows the vertical profile of scattering,

single scattering albedo, and asymmetry factor. A compari-son is made between the C-130 aircraft observations in themain dust layer and the model for two locations during thea797 flight (see Haywood et al. [2003] for a description ofthe flight). Figures 5a and 5d show the vertical profiles ofaerosol scattering at 550 nm from the model compared

Figure 3. Modeled AOD compared with Sun photometers(level 2) data at (a) Cape Verde and (b) Dakar. Included aretwo observations from level 1.5 data, 24 September forCape Verde and 23 September for Dakar. The observationsare instantaneous values. The model has a resolution of3 hours.

Figure 4. A comparison of observed and modeled sizedistributions. The observations are shown when the mineraldust loading is heavy and moderate, and the model resultsare shown for 1 day (25 September) with heavy and one daywith moderate (19 September) mineral dust loading.


against those from the nephelometer on board the C-130aircraft. The comparisons are made near Cape Verde and nearDakar during the a797 flight on the 25 September 2000 (seeHaywood et al. [2003] for a full description of this flight).The C-130 derived scattering coefficient includes correctionsfor variations from STP, truncation of the scattered radiation,and for deficiencies in the illumination source [Anderson andOgren, 1998]. An additional correction factor of 1.5 isapplied to account for the significant contribution to theaerosol scattering from supermicron particles that are notsampled efficiently by the inlet system [Haywood et al.,2003]. Figures 5b and 5e show comparisons of the singlescattering albedo at 550 nm from the model against thosederived from the C-130. In deriving the single scatteringalbedo from the C-130, vertically resolved 1-min averages ofthe aerosol size distribution from the PCASP and identicalrefractive indices to those used in the model are assumed inMie scattering calculations. This method is preferred todetermining the single scattering albedo from measurementsof particle scattering by the nephelometer and particle scat-tering by the Particle Absorption Soot Photometer (PSAP),because it is difficult to obtain representative values from twovertical profiles owing to variability. The refractive indexused in this study is derived from campaign averages ofscattering and absorption. An additional benefit from usingthe PCASP size distribution to determine the optical param-eters is that it provides the asymmetry factor which iscompared in Figures 5c and 5f.[29] The profiles of the scattering show that the mineral

dust is between 1 and 5 km, but with somewhat different

distribution in the model and the observations. The obser-vations show more distinct peaks, which the model withmuch coarser spatial resolution does not reproduce. Themodeled single scattering albedo and asymmetry factor bothshow substantial vertical variations. The modeled singlescattering albedo increases with altitude and the asymmetryfactor decreases with altitude, which both indicate smallerparticles in the upper part of the dust plume. For theobserved single scattering albedo and asymmetry factor nosuch systematic changes with altitude can be found. Incomparison with the observations the modeled single scat-tering albedo and asymmetry factor are higher and lower,respectively, in both cases mainly because the modeledparticles are slightly smaller than the observed ones. Thedifferences between the modeled and measured values arerelatively small, and the modeled single scattering albedoshown in this figure is within the range of averagedmeasured single scattering albedo for the C-130 flightsduring SHADE (range from 0.95 to 0.98 and in the transitfrom Ascension Island it was 0.99). Actually, the measuredsingle scattering albedo during this particular flight (a797)was the lowest of the 6 flights [Haywood et al., 2003].[30] In Figure 6 vertical profiles of scattering for the

entire flight a797 are shown. The observed averaged profileshows a rather homogeneous scattering from 1 to 4–5 km,whereas in the model there is a more clear maximum around3 km. The observed standard deviations are largest in thelower part of the dust plume, in particular around 1 km.However, in the model the largest standard deviations are inthe middle and upper part of the dust plume. Overall the

Figure 5. Comparison of vertical profiles between the C-130 observations and the model at twodifferent locations: (a–c) near Cape Verde (P1 from flight a797, see Haywood et al. [2003] for adescription) and (d–f ) from P14 near Dakar. Figures 5a and 5d show profiles of scattering (km�1),Figures 5b and 5e single scattering albedo, and Figures 5c and 5f the asymmetry factor. All values are for550 nm. The measured single scattering albedo and asymmetry factor values are 1 min averages.


agreement in the vertical profile is good, with the verticalposition of the top and the bottom of the dust plume beingwell represented as well as the absolute magnitude of thescattering when an entire flight of scattering data isincluded.3.5.3. Radiative Measurements[31] To allow for an easier comparison of the radiative

impact of mineral dust between the measurements and themodel we introduce a normalized radiative impact (radiativeimpact divided by AOD). Clouds are excluded in theradiative transfer calculations also for a better comparisonwith the measurements which also are performed for clearsky. The model results are shown for the region 16.9–22.5�W and 13.1–16.9�N.[32] Figure 7 shows that the modeled normalized radia-

tive impact is between around �80 to �120 Wm�2 duringthe day. The normalized radiative impact is stronger at 9 and18 UTC than closer to noon (local noon deviates from UTCnoon by 1–2 hours in the campaign region) as the magni-tude of the radiative impact of non-absorbing particles ishighest for large solar zenith angles ([see Haywood andShine, 1997]. This effect is almost compensated by surfacealbedo of the ocean being higher for large solar zenithangles. The day-to-day variation in the normalized radiativeimpact is small at 12 and 15 UTC; however, at 9 and 18 UTCthe day-to-day variation is significant. This day-to-dayvariation is mostly caused by changes in AOD with weakestnormalized radiative impact for high AOD. Furthermore, thevariation is also slightly related to the size distribution.[33] The modeled and measured normalized radiative

impact are relatively similar during the middle of the dayand this is an important result. In the model the values arearound �80 Wm�2, whereas around �90 Wm�2 in themeasurements. The normalized radiative impact dependsslightly on the AOD, with decreasing normalized radiativeimpact with increasing AOD. Results from the measure-ments are only shown for small solar zenith angles as theuncertainties in the measurements increase significantly forlarger solar zenith angles owing to uncertainties in thesurface albedo of ocean for large solar zenith angles. The

aircraft measurements of the radiative impact compare wellwith satellite derived radiative impact from the CERESinstrument [Haywood et al., 2003].

3.6. Radiative Impact

[34] Above normalized radiative impact is discussed,below solar and longwave radiative impact without normal-ization is presented.3.6.1. Solar[35] The solar radiative impact of mineral dust in the

period 24–26 September is shown for clear sky and whenclouds are included in Figure 8. The radiative impactfollows mainly the pattern of AOD. Clouds reduce theradiative impact of mineral aerosols. This is most pro-nounced west of 30�W. The radiative impact shows strongnegative values, with some small positive values associatedwith very limited cloudy regions. The maximum radiativeimpact is 6 Wm�2. Mineral dust influences the solarradiation strongly not only near the source regions and itsassociate regions, but we can see a significant influenceeven in the western part of the Atlantic ocean. The strongestvalues reach almost �115 Wm�2. This is in accordancewith measurements in the work of Haywood et al. [2003],who reported values ranging down to nearly �130 Wm�2.The strongest radiative impact is rather similar over land(24 September) and over ocean (26 September), despite theAOD is stronger 24 September. This is due to the fact thatthe normalized radiative impact is at least a factor twostronger over ocean than over land owing to the differencein surface albedo.[36] In the work of Myhre et al. [2003] it was found that

neglect of spectral and solar zenith angle variations in thesurface albedo weakened the radiative impact of aerosolsfrom biomass burning by 25–30% over the region which is

Figure 6. Vertical profiles of scattering for the C-130observation and the model for flight a797. Standarddeviations are shown as horizontal lines.

Figure 7. Modeled and observed normalized clear skyradiative impact (given as Wm�2 per unit AOD) for varioustimes during the day. Results from the model are shown inthe period 19–28 September 2000. The measurements areshown from a797 (R6; see Haywood et al. [2003]) on 25September with about 5 min averages. Measurements weremade at the top of the dust layer (about 5 km) and the modelresults at the top of the atmosphere (consistent with the restof the model results). In the model the difference inradiative impact of mineral dust is negligible between 5 kmand the top of the atmosphere.


influenced by aerosols from biomass burning in southernAfrica. For mineral dust the influence of spectral and solarzenith angle variation is much smaller (5–10%). Thereduction due to neglect of such variations is slightly largerover land than over ocean. The weaker response to surfacealbedo parameterization for mineral dust compared to forbiomass burning aerosols are due to the much strongerspectral dependence of specific extinction coefficient andsingle scattering albedo than for biomass aerosols.[37] In the work of Myhre and Stordal [2001] a large

number of sensitivity calculations of the radiative forcingdue to mineral dust were performed, pointing to refractiveindices as a major source to the large uncertainty. Inaddition to the calculations performed using the refractiveindices derived from Sun photometers, we also performedcalculations using the refractive indices of d’Almeida et al.[1991] which is the base refractive index used in thecalculations of Myhre and Stordal [2001]. Due to the muchhigher imaginary part of these data for the refractive indexcompare to the one based on the Sun photometers this yieldsa reduction in the single scattering albedo at solar wave-lengths; see Haywood et al. [2003] for discussion. Theradiative impact over the region of interest decreases byabout 25% for clear sky and as much as 50% when cloudsare included. The strongest radiative impact is weakenedfrom around �115 Wm�2 to around �80 Wm�2. Positiveradiative impact of up to 50 Wm�2 was then estimated, asopposed to 6 Wm�2 in the original calculation. The increasein the radiative impact was much larger over land than overocean owing to the brighter surface.[38] Figure 8 shows modeled diurnal mean solar radiative

impact when clouds are excluded and included in the radia-tive transfer calculations over the region shown in Figure 9.The radiative impact is somewhat stronger during the end ofthe campaign period than in the beginning. The figure showsthat the radiative impact for clear sky is 40–50% strongerthan when clouds are taken into account. When clouds areincluded in the simulations the solar radiative impact isaround �6 Wm�2 over the region, corresponding to aglobal mean radiative impact of �0.4 Wm�2 (scaled accord-

ing to the fraction of the area of this region relative toEarth’s surface). Note that these numbers take into ac-count almost the entire radiative effect of mineral dustparticles originating from Sahara, but not from othersources.3.6.2. Longwave[39] The diurnal mean longwave radiative impact is

shown in Figure 10a. The longwave radiative impact ismuch smaller in magnitude than the solar part in oursimulations. The radiative impact is strongest over landnear the source regions where the AOD is highest, but inaddition the surface temperature is higher and the tropo-sphere is less humid compared to over ocean contributing toan even larger land/ocean contrast in the results. Themaximum diurnal mean longwave radiative impact is closeto 8 Wm�2. Figure 4 indicates that the model underesti-mates the number of larger mineral particles. This causesthat the longwave radiative impact is underestimated in ourcalculations. In a calculation using the lognormal sizedistribution based on the observations shown in Figure 4resulted in 10–20% stronger longwave radiative impact. Asensitivity calculation with the refractive index fromd’Almeida et al. [1991] instead of from Fouquart et al.[1987] is performed. This resulted in a weakening of thelongwave radiative impact by a factor more than 2. Theoptical properties changes substantially with the change inrefractive indices [see Highwood et al., 2003].[40] Figure 11 shows the regional mean longwave radia-

tive impact during the campaign (region as shown inFigures 9 and 10. As in Figure 8 we see that the radiativeimpact is stronger during the end of the campaign. Cloudsweakened the radiative impact by about 20%. The regionalmean longwave radiative impact varies from around 0.8 toslightly more than 1.0 Wm�2 and is in magnitude about afactor 6–7 weaker than the solar radiative impact.3.6.3. Net Radiative Impact[41] Figure 10b shows the diurnal mean net radiative

impact of mineral aerosols during the SHADE campaignperiod. The radiative impact is strongest on 26 Septemberwith diurnal mean values close to �50 Wm�2 off the coastof western Africa. The small positive values of the netradiative impact is associated with clouds. The ratio of themagnitude of the longwave and solar radiative impact isgenerally a factor of 2 higher over land than ocean. Thediurnal and regional mean net radiative impact of mineralaerosols is slightly stronger than �5 Wm�2 when clouds areincluded in the simulations.

4. Summary

[42] In an attempt to reduce the large uncertainties linkedto the direct effect of the mineral dust, we model theradiative impact of the mineral dust with a comprehensivecomparison with measurements obtained during theSHADE campaign. We use meteorological data for thecampaign period, which enables us to reproduce the mainpattern of mineral dust during SHADE. The general agree-ment with observations is very good. Satellite retrievalsshow that the general pattern of mineral dust during thecampaign is reproduced. MODIS data is used during themajor dust storm, and also TOMS aerosol index (notshown) reveals a very similar feature to the model during

Figure 8. Diurnal mean solar radiative impact during theperiod 19–28 September 2000 (Wm�2), excluding andincluding clouds. The results are for the region shown inFigure 9.


the entire campaign, even the transport of mineral dust tothe north of Sahara. Comparison with ground-based Sunphotometers strengthens the agreement of the main features.In addition the possibility to compare the model against insitu aircraft data is very valuable. Most important in thisrespect is the radiative effect of aerosols, which is of largescientific value. Given the fact that the model is completelyindependent of the measurements, a radiative effect in themodel and observations within about 10% is very encour-

aging, demonstrating the usefulness of larger aerosol cam-paigns. Both in the model and observations a weakdependence of the normalized radiative impact on AOD isfound close to noon. The modeled size and vertical profilesare compared with the aircraft measurements with reason-able agreement, except for the larger sized mineral particleswhich are underestimated in the model. This affects moststrongly the longwave radiative impact results, and only to asmall extent the shortwave results.

Figure 9. Radiative impact of mineral dust in the period 24–26 September for (Figures 9a–9d and 9i–9l)four times each day, and (Figures 9q–9t) clouds are included in the calculations, whereas in Figures 9e–9h,9m–9p, and 9u–9x, clouds are excluded.


[43] The problem with the size distribution is a result of thecurrently limited knowledge of the physical processes in thesource of mineral dust. Further, few observations in theseareas are available. The physical uncertainties of the sourceare not only linked to dependence on friction velocity andsurface composition, but also on data characterizing thesurface. A problem for a global model is also to representphysical processes that occur on a small scale. A major resultof the SHADE campaign is that it confirms the recent findingsin the work of Kaufman et al. [2001], who reported a muchhigher single scattering albedo than earlier measurements

indicate [Haywood et al., 2003]. This can either be due to thefact that in earlier measurements mineral dust has beenmixedwith aerosols from biomass burning or that the source regionsare located in other regions than during SHADE.Clarke et al.[2001] find a similar high single scattering albedo for dustover China, but on the other hand Claquin et al. [1999] showthat the mineral composition varies substantially in Saharaand the absorbing hematite has a much higher abundance inthe Sahel region than in the rest of Sahara.[44] We estimate a much stronger solar radiative impact

than the longwave radiative impact, and thus a strong net

Figure 10. Diurnal mean radiative impact of mineral dust during the campaign period 19–28September. (a) Longwave. (b) Net.


negative radiative impact or cooling effect of the mineraldust from Sahara during the SHADE campaign. The stron-gest modeled net radiative impact is around �110 Wm�2,occurring around local noon on 26 September. Over aregion (latitude 0�N–30�N, longitude 60�W–40�E) aroundSahara the diurnal mean radiative impact during theSHADE campaign is between �5 and �6 Wm�2. Thistranslates to a global mean net radiative impact of Saharandust aerosol of around �0.4 Wm�2 when averaged over theglobe. The solar and longwave radiative impact results arebased on refractive indices which are from observations orcompared closely to measurements. In accordance withearlier studies we find a strong sensitivity to the refractiveindices and the use of refractive indices from commonlyused sources resulted in solar and longwave radiativeimpact a factor of 2 weaker than our best estimates.[45] The diurnal and regional mean of the solar radiative

impact is between �8 and �10 Wm�2 when clouds areexcluded. This is somewhat stronger than the global meandue to all kinds of aerosols, as found in several studiesbased on satellite data [Haywood et al., 1999; Boucher andTanre, 2000; Loeb and Kato, 2002]. Our results yieldstronger radiative impact than the ones derived globallyfrom observations, as we are focusing on a region with largecontributions from strongly reflecting dust aerosols. How-ever, also large areas in the regions considered in our studyare only weakly influenced by dust aerosols.

[46] Acknowledgments. This work has been supported by the EUproject AIRARF funded through the LSF program CAATER and from theResearch Council of Norway, through the ChemClim project.

ReferencesAlfaro, S. C., and L. Gomes, Modeling mineral aerosol production by winderosion: Emission intensities and aerosol size distributions in sourceareas, J. Geophys. Res., 106, 18,075–18,084, 2001.

Anderson, T. L., and J. A. Ogren, Determining aerosol radiative propertiesusing the TSI 3563 integrating nephelometer, Aerosol Sci. Technol., 29,57–69, 1998.

Boucher, O., and D. Tanre, Estimation of the aerosol perturbation to theEarth’s radiative budget over oceans using POLDER satellite aerosolretrievals, Geophys. Res. Lett., 27, 1103–1106, 2000.

Briegleb, B. P., P. Minnis, V. Ramanathan, and E. Harrison, Comparison ofregional clear-sky albedos inferred from satellite observations and modelscomparisons, J. Clim. Appl. Meteorol., 25, 214–226, 1986.

Chomette, O., M. Legrand, and M. Marticorena, Determination of the windspeed threshold for the emission of desert dust using satellite remotesensing in the thermal infrared, J. Geophys. Res., 104, 31,207–31,215,1999.

Claquin, T., Modelisation de la mineralogie et du forcage radiatif des pous-sieres desertiques, Ph.D. thesis, Univ. of Hamburg, Hamburg, 1999.

Claquin, T., M. Schulz, and Y. J. Balkanski, Modeling the mineralogy ofatmospheric dust sources, 104, 22,243–22,256, 1999.

Clarke, A. D., W. G. Collins, P. J. Rasch, V. N. Kapustin, K. Moore,S. Howell, and H. E. Fuelberg, Dust and pollution transport on globalscales: Aerosol measurements and model predictions, J. Geophys. Res.,106, 32,555–32,569, 2001.

d’Almeida, G. A., P. Koepke, and E. P. Shettle, Atmospheric aerosols, inGlobal Climatology and Radiation Characteristics, A. Deepak, Hampton,Va., USA, 1991.

Fecan, F., B. Marticorena, and G. Bergametti, Parameterization due to theincrease of the aeolian erosion threshold wind friction velocity due to soilmoisture for arid and semi-arid areas, Ann. Geophys., 17, 149–157,1999.

Fouquart, Y., B. Bonnel, G. Brogniez, J. C. Buriez, L. Smith, and J. J.Morcrette, Observations of Saharan aerosols: Results of ECLATS fieldexperiment, part II: Broadband radiative divergence, J. Clim. Appl. Me-teorol., 26, 38–52, 1987.

Glew, M. D., P. Hignett, and J. P. Taylor, Aircraft measurements of seasurface albedo, J. Atmos. Sci, in press, 2003.

Grini, A., G. Myhre, J. K. Sundet, and I. S. A. Isaksen, Modeling the annualcycle of sea salt in the global 3-D model OSLO CTM-2: Concentrations,fluxes and radiative impact, J. Clim., 15, 1717–1730, 2002a.

Grini, A., C. S. Zender, and P. R. Colarco, Saltation Sandblasting behaviorduring mineral dust aerosol production, Geophys. Res. Lett., 29(18),1868, doi:10.1029/2002GL015248, 2002b.

Guelle, W., Y. J. Balkanski, M. Sbulz, B. Marticorena, G. Bergametti,C. Moulin, R. Arimoto, and K. D. Perry, Modeling the atmosphericdistribution of mineral aerosol: Comparison with ground measurementsand satellite observations for yearly and synoptic timescales over theNorth Atlantic, J. Geophys. Res., 105, 1997–2012, 2000.

Hansen, J., M. Sato, and R. Ruedy, Radiative forcing and climate response,J. Geophys. Res., 102, 6831–6864, 1997.

Haywood, J. M., and K. P. Shine, The effect of anthropogenic sulfate andsoot aerosol on the clear sky planetary radiation budget, Geophys. Res.Lett., 22, 603–606, 1995.

Haywood, J. M., and K. P. Shine, Multi-spectral calculations of the directradiative forcing of tropospheric sulphate and soot aerosols using a col-umn model, Q. J. R. Meteorol. Soc., 123, 1907–1930, 1997.

Haywood, J. M., V. Ramaswamy, and B. J. Soden, Tropospheric aerosolclimate forcing in clear-sky satellite observations over the oceans,Science, 283, 1299–1303, 1999.

Haywood, J. M., P. N. Francis, M. D. Glew, and J. P. Taylor, Opticalproperties and direct radiative effect of Saharan dust: A case study oftwo Saharan dust outbreaks using aircraft data, J. Geophys. Res., 106,18,417–18,430, 2001.

Haywood, J. M., et al., Radiative properties and direct radiative effect ofSaharan dust measured by the C-130 aircraft during Saharan Dust Ex-periment (SHADE). 1: Solar spectrum, J. Geophys. Res., 108(D18),doi:10.1029/2002JD002687, in press, 2003.

Hess, M., P. Koepke, and I. Schult, Optical properties of aerosols andclouds: The software package OPAC, Bull. Am. Meteorol. Soc., 79,831–844, 1998.

Highwood, E. J., J. M. Haywood, M. Silverstone, S. M. Newman, and J. P.Taylor, Radiative properties and direct effect of Saharan dust measured bythe C-130 aircraft during Saharan Dust Experiment (SHADE). 2: Terres-trial spectrum, J. Geophys. Res., 108(D18), doi:10.1029/2002JD002552,in press, 2003.

Hogan, R. J., and A. J. Illingworth, Deriving cloud overlap statistics fromradar, Q. J. R. Meteorol. Soc., 126, 2903–2909, 2000.

Holben, B. N., et al., AERONET: A federated instrument network and dataarchive for aerosol characterization, Remote Sens. Environ., 66, 1–16,1998.

Intergovernmental Panel on Climate Change (IPCC), Climate Change2001, The Scientific Basis, edited by J. T. Houghton et al., pp. 349–416, Cambridge Univ. Press, New York, 2001.

Kaufman, Y. J., D. Tanre, L. Remer, E. Vermote, and B. N. Holben, Opera-tional Remote Sensing of Tropospheric Aerosols over the Land fromEOS-MODIS, J. Geophys. Res., 102, 17,051–17,068, 1997.

Kaufman, Y. J., D. Tanre, O. Dubovik, A. Karnieli, and L. A. Remer,Absorption of sunlight by dust as inferred from satellite and ground-based remote sensing, Geophys. Res. Lett., 28, 1479–1482, 2001.

Figure 11. Longwave radiative impact during the period19–28 September 2000 (Wm�2), excluding and includingclouds. The results are for the region shown in Figures 9and 10.


Liu, M., and D. Westphal, A study of the sensitivity of simulated mineraldust production to model resolution, J. Geophys. Res., 106, 18,099–18,112, 2001.

Loeb, N. G., and S. Kato, Top-of-atmosphere direct radiative effect ofaerosols over the Tropical oceans from the Clouds and the Earth’s Ra-diant Energy System (CERES) satellite instrument, J. Clim., 15, 1474–1484, 2002.

Lu, H., and Y. Shao, A new model for dust emission by saltation bombard-ment, J. Geophys. Res., 104, 16,827–16,842, 1999.

Matthews, E., Global vegetation and land use: New high-resolution databases climate studies, J. Clim. Appl. Meteorol., 22, 474–487, 1983.

Mishchenko, M. I., L. D. Travis, R. A. Kahn, and R. A. West, Modelingphase functions for dustlike tropospheric aerosols using a shape mixtureof randomly oriented polydisperse spheroids, J. Geophys. Res., 102,16,831–16,847, 1997.

Myhre, G., and F. Stordal, Global sensitivity experiments of the radiativeforcing due to mineral aerosols, J. Geophys. Res., 106, 18,193–18,204,2001.

Myhre, G., T. K. Berntsen, J. M. Haywood, J. K. Sundet, B. N. Holben,M. Johnsrud, and F. Stordal, Modeling the solar radiative impact ofaerosols from biomass burning during the Southern African RegionalScience Initiative (SAFARI-2000) experiment, J. Geophys. Res.,108(D13), 8501, doi:10.1029/2002JD002313, 2003.

Nickling, W. G., and J. A. Gillies, Emissions of fine-grained particulatesform desert soils, in Paleoclimatology and Paleometeorology: Modernand Past Patterns of Global Atmospheric Transport, NATO ASI Ser.,edited by M. Leinen and M. Sarnthein, pp. 133–167, Kluwer Acad.,Norwell, Mass., 1987.

Olson, J. S., J. A. Watts, and L. J. Allison, Carbon in live vegetation ofmajor world ecosystems, Rep. ORNL_5862, Oak Ridge Natl. Lab., OakRidge, Tenn., 1983.

Prather, M. J., Numerical advection by conservation of second-order mo-ments, J. Geophys. Res., 91, 6671–6681, 1986.

Ramankutty, N., and J. Foley, Estimating historical changes in global landcover: Croplands from 1700 to 1992, Global Biogeochem. Cycles, 13,997–1027, 1999.

Remer, L. A., et al., Validation of MODIS aerosol retrieval over ocean,Geophys. Res. Lett., 29(12), 8008, doi:10.1029/2001GL013204, 2002.

Schulz, M., Y. Balkanski, W. Guelle, and F. Dulac, Role of aerosol sizedistribution and source location in a three dimensional simulation of aSaharan dust episode tested against satellite derived optical thickness,J. Geophys. Res., 103, 10,579–10,592, 1998.

Shao, Y., A model for mineral dust erosion, J. Geophys. Res., 106, 20,239–20,254, 2001.

Shettle, E. P., Optical and radiative properties of a desert aerosol model, inProceedings of the Symposium on Radiation in the Atmosphere, edited byG. Fiocco, pp. 74–77, A. Deepak, Hampton, Va., 1984.

Slingo, A., A GCM parameterization for the shortwave radiative propertiesof water clouds, J. Atmos. Sci., 46, 1419–1427, 1989.

Sokolik, I. N., and O. B. Toon, Direct radiative forcing by anthropogenicairborne mineral aerosols, Nature, 381, 681–683, 1996.

Stamnes, K., S. C. Tsay, W. Wiscombe, and K. Jayaweera, A numericallystable algorithm for discrete-ordinate-method radiative transfer in multi-

ple scattering and emitting layered media, Appl. Opt., 27, 2502–2509,1988.

Stephens, G. L., Radiation profiles in extended water clouds, I, Theory,J. Atmos. Sci., 35, 2111–2122, 1978.

Stephens, G. L., and C. M. R. Platt, Aircraft observations of the radiativeand microphysical properties of stratocumulus and cumulus cloud fields,J. Clim. Appl. Meteorol., 26, 1243–1269, 1987.

Stull, R. B., An Introduction to Boundary Layer Meteorology, 666 pp.,Kluwer Acad., Norwell, Mass., 1988.

Sundet, J. K., Model studies with a 3-D global CTM using ECMWF data,Ph.D. thesis, Dept. of Geophys., Univ. of Oslo, Oslo, 1997.

Tanre, D., Y. J. Kaufman, M. Herman, and S. Mattoo, Remote sensing ofaerosols properties over oceans using the MODIS/EOS spectral ra-diances, J. Geophys. Res., 102, 16,971–16,988, 1997.

Tanre, D., et al., Measurement and modeling of the Saharan dust radiativeimpact: Overview of the Saharan Dust Experiment (SHADE), J. Geo-phys. Res., 108(D18), doi:10.1029/2002JD003273, in press, 2003.

Tegen, I., and I. Fung, Modeling of mineral dust in the atmosphere:Sources, transport and optical thickness, J. Geophys. Res., 99, 22,897–22,914, 1994.

Tegen, I., A. A. Lacis, and I. Fung, The influence of mineral aerosols fromdisturbed soils on climate forcing, Nature, 380, 419–422, 1996.

Tiedkte, M., A comprehensive mass flux scheme for cumulus parameter-isation on large scale models, Mon. Weather, Rev., 117, 1779–1800,1989.

Weaver, C. J., P. Ginoux, N. C. Hsu, M.-D. Chou, and J. Joiner, Radiativeforcing of Saharan dust: GOCART model simulations compares withERBE data, J. Atmos. Sci., 59, 736–747, 2002.

Wilson, M. F., and A. Henderson-Sellers, A global archive of land coverand soils data for use in general circulation climate models, J. Climatol.,5, 119–143, 1985.

Wiscombe, W., and J. Evans, Exponential-sum fitting of radiative transmis-sion functions, J. Comput. Phys., 24, 416–444, 1977.

Woodward, S., Modeling the atmospheric life cycle and radiative impact ofmineral dust in the Hadley Centre climate model, J. Geophys. Res., 106,18,155–18,166, 2001.

��

B. Chatenet, Laboratoire Inter-universitaire des Systemes, Atmospher-iques, CNRS, Universite Paris 7 12, 61, avenue du General de Gaulle,F-94010 Creteil, France. ([email protected])A. Grini, I. S. A. Isaksen, G. Myhre, and J. K. Sundet, Department of

Geophysics, University of Oslo, P.O. Box 1022 Blindern, N-0315 Oslo,Norway. ([email protected]; [email protected];[email protected]; [email protected])J. M. Haywood, Met Office, Y46 Bldg, DERA, Farnborough, Hants

GU14 0LX, UK. ([email protected])F. Stordal, Norwegian Institute for Air Research (NILU), P.O. Box 100,

N-2027 Kjeller, Norway. ([email protected])D. Tanre, Laboratoire d’Optique Atmospherique, CNRS, Universite de

Lille 1, F-59655, Villeneuve-d’Ascq CEDEX, France. ([email protected])


Saltation Sandblasting behavior during mineral dust

aerosol production

Alf GriniDepartment of Geophysics, University of Oslo, Oslo, Norway

Charles S. ZenderDepartment of Earth System Science, University of California, Irvine, CA, USA

Peter R. ColarcoLaboratory for Atmospheric and Space Physics, Program in Atmospheric and Oceanic Sciences, University of Colorado,Boulder, CO, USA

Received 1 April 2002; revised 11 June 2002; accepted 12 June 2002; published 20 September 2002.

[1] The dominant process in producing fine dust aerosolsduring saltation is thought to be sandblasting. Recent studiesclaim that due to competing physical processes, emissionefficiencies of dust aerosols oscillate with increasing windfriction speed. These oscillations can result in order ofmagnitude changes in dust mass emissions. Our work showsthat emission efficiencies, and hence emissions of dustaerosols are smooth functions of the wind friction speed fornatural soil size distributions. This rules out oscillations asan explanation for scatter in experimental data. We show andexplain the reasons for the oscillations. INDEX TERMS:

0305 Atmospheric Composition and Structure: Aerosols and

particles (0345, 4801). Citation: Grini, A., C. S. Zender, and P.

R. Colarco, Saltation Sandblasting behavior during mineral dust

aerosol production, Geophys. Res. Lett., 29(18), 1868, doi:10.1029/

2002GL015248, 2002.

1. Introduction

[2] Several modeling studies have shown the importanceof mineral dust for the atmosphere’s radiative balance[Tegen and Fung, 1994] and for its chemistry [Denteneret al., 1996]. The size resolved emissions of dust is a keyuncertainty in modeling dust transport [Schulz et al., 1998]and radiative forcing [Myhre and Stordal, 2001]. A betterunderstanding of dust emissions, both with respect to dustsize distributions and total magnitude, is important toimprove estimates of the climate impact of atmosphericdust.[3] Saltation and sandblasting has been recognized as the

most important mechanism for producing small dust aero-sols [Shao and Raupach, 1993]. Saltation refers to a layer ofsoil moving with the wind just above the surface. Sand-blasting refers to the release of dust aerosol during impactsby saltating particles. The dust aerosol may be remnants ofdisintegrating aggregate saltators or surface particles ejectedby the saltator impact. Early dust production models [Mar-ticorena and Bergametti, 1995; Marticorena et al., 1997]have used this approach to model the flux of dust aerosolsemitted into the atmosphere. These models did not model thesize distribution of the emitted dust aerosols. Recent dust

production models [Shao and Raupach, 1993; Shao et al.,1996; Alfaro et al., 1997, 1998; Lu and Shao, 1999; Shaoand Lu, 2000; Alfaro and Gomes, 2001; Shao, 2001] usevarious physical approaches to model the sandblastingprocess to give equations for the size distributed flux of dustaerosols.[4] The ratio of the vertical mass flux of dust aerosols to

the saltating mass flux is called the sandblasting efficiencya [Gillette, 1979]. Defined in this manner, a largely reflectsthe size of particles emitted rather than their number or theenergy consumed in the sandblasting process. We thereforeuse the term mass sandblasting efficiency to describe a. Theearly dust production models used an empirical relation toestimate a based on soil clay content [Marticorena andBergametti, 1995]. Alfaro and Gomes [2001] use a physicalmodel of the binding energy of dust and the kinetic energyof the saltation layer to calculate a. They argue that, evenfor a continuous soil size distribution, a combination ofseveral competing effects results in a strongly oscillatorybehavior of the mass sandblasting efficiency with increasingwind friction speed.[5] Our results lead to a new interpretation of the physical

processes governing saltation and sandblasting. Thereported oscillatory behavior of a is an artifact caused byinadequacies in the numerical procedure used to evaluate a.The actual mass sandblasting efficiency is a smooth func-tion of the wind friction speed for natural (i.e., continuous)soil size distributions.

2. Current Understanding

[6] Wind friction speeds of about 0.20 m s�1 can directlyentrain soil grains of about 75 mm into the saltation layer.The threshold wind friction speed is larger both for smallerand larger soil grain sizes [Iversen and White, 1982]. Due tolarge binding energies, small dust aerosols need large windfriction speeds to be mobilized by direct entrainment[Iversen and White, 1982].[7] Several works have pointed out sandblasting as the

most important mechanism for releasing dust aerosols from asoil [Shao and Raupach, 1993; Shao et al., 1996; Marticor-ena and Bergametti, 1995]. Shao and Raupach [1993] andShao et al. [1996] proposed that the number of of dustparticles dislodged from the surface per saltation impact is

GEOPHYSICAL RESEARCH LETTERS, VOL. 29, NO. 18, 1868, doi:10.1029/2002GL015248, 2002

Copyright 2002 by the American Geophysical Union.0094-8276/02/2002GL015248$05.00

15 - 1

proportional to the ratio of the kinetic energy loss during asaltation impact to the typical binding potential energyholding a dust particle to the surface, C, which is higherfor small dust particles as they are bound by strongercohesion forces. Thus the vertical, size distributed dust fluxis sensitive to the mass and the speed of the saltating soilgrains.[8] Each soil size has its own mass sandblasting effi-

ciency. Soil particles with low energy will release largeparticles, and those with high energy will be able to releasesmall particles. It is thus important to know the sizedistribution of the particles in the saltation layer at anywind friction speed and to carefully integrate the masssandblasting efficiency over this size distribution. The total,integrated mass sandblasting efficiency for a soil, is highlysensitive to its size distribution [Shao and Raupach, 1993;Shao et al., 1996; Shao and Lu, 2000; Alfaro and Gomes,2001]. Not all of the kinetic energy of the saltating particleis used to release fine dust. Lu and Shao [1999] and Shao[2001] emphasize that a large part of the kinetic energy ofthe saltating particle is actually used for plastic deformationof the soil, and that the emissions from a soil largely dependon the plastic pressure of the soil surface.[9] Alfaro and Gomes [2001] developed a dust produc-

tion model where the wind friction speed (u*) together withthe size distribution of the saltating soil particles determinesthe size distribution of the emitted dust aerosols. Theypropose that the emitted dust aerosols will be a compositionof three lognormal modes with mass median diameters ofd1 = 1.5 mm, d2 = 6.7 mm and d3 = 14.2 mm. Each of themodes has a characteristic binding energy which must beexceeded before it is released. The binding energies wereestimated in wind tunnel experiments [Alfaro et al., 1998].[10] Adopting the notation of Alfaro and Gomes [2001],

the vertical number flux of aerosols from mode i fromsandblasting by a given soil particle size is:

dNi Dp

� �

¼ bdFh Dp

� � pi Dp

� �

eið1Þ

where N is upward, vertical number flux of aerosols inmode i (m�2 s�1), b is a constant (163 m s�2), Fh ishorizontal flux of aerosols (kg m�1 s�1), pi is fraction ofenergy used to release aerosols from mode i, ei is bindingenergy of mode i (J) and Dp is diameter of saltating soilgrains (m). The net vertical dust flux is the upward flux (1)minus the depositional fluxes (gravitation, turbulent mix-out, scavenging) with which this paper is not concerned.[11] The vertical mass flux of aerosols from mode i is:

dFaeros;i Dp

� �

¼p

6rpbdFh Dp

� � pi Dp

� �

�d3i

eið2Þ

where Faeros is the upward, vertical mass flux of aerosols inmode i (kg m�2 s�1), rp is the density of aerosols (kg m�3)and di is the mass mean diameter of the aerosol mode (m).Alfaro and Gomes, 2001 use mass median diameter insteadof mass mean diameter in (2). Doing so overestimates massflux by a factor of about 2.5.[12] As noted by Alfaro and Gomes [2001], several

effects compete in determining the vertical mass flux ofdust aerosols from a soil with a lognormal size distribution.The change in mass sandblasting efficiency with increasingfriction wind speed is influenced by the following factors:1) Already saltating aggregates release finer and finerparticles, decreasing a 2) A larger amount of coarseaggregates of poor mass sandblasting efficiency enter sal-tation, decreasing a and 3) Smaller aggregates alreadysaltating but previously inefficient for sandblasting becomeproductive, increasing a.[13] During the interplay of complicated processes, sev-

eral of which include threshold values, one might expectoscillations. Alfaro and Gomes [2001] explain that aoscillates with increasing wind friction speed because ofthis interplay. They compare the mass sandblasting efficien-cies from their dust production model to measured masssandblasting efficiencies of between 10�6 and 10�3 m�1 forvarying soils and wind friction speeds. The oscillatingnature of the calculated mass sandblasting efficiency seems

Figure 1. Size distribution of the saltation layer over finesand (FS) at different wind friction speeds. The sizedistribution is given in fraction of mass at a given size perunit size.

Figure 2. Mass sandblasting efficiencies (m�1) calculatedby using 100 points (solid line), 1000 points (circles) and10000 points (crosses) along the soil size distribution.

15 - 2 GRINI ET AL.: SALTATION SANDBLASTING BEHAVIOR

consistent with the scatter in these observations. We findthat, when using continuous soil size distributions, a is non-oscillatory and that the observed scatter in a is due to othercauses.

3. Proposed New Interpretation

[14] We now show that the mass sandblasting efficienciesfor a soil size distribution depend strongly on the numericalprocedure employed to calculate them. Accurate integration,using small steps along the soil-diameter axis is importantdue to shift in size distributions.[15] We calculated a using (2) for the soil type called fine

sand (FS) by Chatenet et al. [1996]. This soil is lognormallydistributed, it has a mass median diameter of 210 mm and astandard deviation (s) of 1.6. In the following, we assume asmooth surface with no drag partitioning effects. Figure 1shows how the size distribution of the saltating flux changeswith the wind friction speed [Iversen and White, 1982]. Atlow wind friction speeds, only the 75 mm particles areavailable. At high wind friction speeds, all sizes in the soilsize distribution are available for saltation.[16] Figure 2 shows the mass sandblasting efficiency (a)

calculated with the same numerical procedure using differ-ent size resolutions (N = 100, 1000, 10000 points) logarith-mically evenly spaced along the soil-diameter axis. Theoscillations occur in the low resolution (N = 100) calcu-lations. Increasing the size resolution damps the oscillations(N = 1000) until they eventually disappear (N = 10000). Itshould be noted that the oscillations for the lowest reso-lution calculation are of an order of magnitude, consistentwith Alfaro and Gomes [2001]. The mass sandblastingefficiency goes through a maximum at 0.54 m s�1 and thendecreases. This maximum is not very distinct for FS, but itis more distinct for soils with larger mass median diameter(not shown), such as coarse sand (CS) and salts (SS)[Chatenet et al., 1996]. We performed a large range ofsensitivity studies to soil size distributions (using soil sizedistributions from Chatenet et al. [1996]) and friction speed(varying from 0 to 1 m s�1) to verify that a does notoscillate under any conditions.[17] In Figure 3, we show total emissions calculated from

the dust production model calculated at different locations

in Africa. The locations contain blends of four natural aridsoil types which can be found in nature [Chatenet et al.,1996]. The predicted mass fluxes at each location showspurious, order of magnitude oscillations unless adequateresolution is used to compute them. The noise is largest forthe coarse mode aerosols which is consistent with the reasonfor the oscillations (see below).[18] The oscillations occur because the parameter pi in

(1) and (2), and hence the number fraction of dust in eachmode, is very sensitive to the saltator soil size. This isillustrated for a wind friction speed of 0.50 m s�1 inFigure 4. The fraction of coarse mode aerosols increaseand decreases very rapidly with increasing saltator soil sizeindicating that a fine size resolution is needed to capturethis behavior. The smallest soil sizes which contribute tosandblasting determine all flux from the coarsest mode.[19] At saltator size resolutions which do not resolve the

narrow emission peak of coarse mode aerosols, the pre-

Figure 3. Total fluxes calculated at different locations in Africa. CS, FS, CMS, and SMS are different soil sizedistributions (see Chatenet et al. [1996] andMarticorena et al. [1997]). The lines show the flux of fine (light grey), medium(medium grey), and coarse (black) mode aerosols calculated with the dust production model. Solid lines use a coarseintegration over the soil sizes (400 bins), and the dashed lines use a fine integration over soil sizes (4000 bins). The 10mwind speed is calculated from wind friction speed assuming neutral boundary layer.

Figure 4. Fraction of number flux coming from each ofthe fine (crosses), medium (circles) and coarse (solid) modeaerosols for wind friction speed of 0.50 m s�1. The fractionof coarse mode number flux increases and decreases rapidlywith saltator size at constant wind friction speed.

GRINI ET AL.: SALTATION SANDBLASTING BEHAVIOR 15 - 3

dicted contribution from the coarse mode aerosols will betoo high or too low. This random bias shows up asoscillations when calculating the mass sandblasting effi-ciency at low resolutions. The total mass flux is mostlydependent on the coarse mode aerosols due to weighting ofthe number flux from mode 3 by �di

3 in (2). It is thereforedetermined by the smallest sandblasting soil sizes. A similarbehavior (not shown) occurs at other wind friction speed,but the peak for the coarse dust particles occurs at lower(higher) soil sizes for higher (lower) wind friction speed.[20] Interestingly, as this phenomenon leads to oscilla-

tions in total mass fluxes at low resolutions (because ofchanges in the size distribution of the dust flux), the totalnumber flux can be calculated accurately with low resolu-tion. The total number flux of dust is quite insensitive topi (1). pi being inaccurate only means that the number fluxwill come from another mode. The total number flux canthus be approximately correct even though the pi (and hencethe size distribution) is wrong.[21] Our results rule out oscillations as explanation to

scatter in observed mass sandblasting efficiencies. Thescatter is more likely due to differences in soil propertiesor measurement techniques.[22] Our results alter the conceptual understanding that

activation of progressively finer aerosol modes by progres-sively higher kinetic energy leads to oscillatory behavior ina. This is important to keep in mind if the dust productionmodel is to be used in transport models of mineral dustaerosols since total emissions is a key output from suchmodels.

4. Summary

[23] It has been argued that, even for continuous soil sizedistributions, competing threshold processes interact toyield an oscillatory behavior in the mass sandblastingefficiency of a soil. We find that these oscillations onlyoccur if one does not use high enough resolution whenintegrating the mass sandblasting efficiency over the soilsize distribution.[24] Our results change the conceptual understanding

proposed earlier that activation of progressively finer aero-sol modes by progressively higher kinetic energy leads tooscillatory behavior in the mass sandblasting efficiency, a.

[25] Acknowledgments. AG acknowledges support from the Norwe-gian research council grant 139810/720 (CHEMCLIM). CSZ gratefullyacknowledges support from NASA grants NAG5-10147 (IDS) and NAG5-

10546 (NIP). PRC acknowledges support from NASA ESS fellowshipNGT-30155 and NASA TOMS NAG5-11069.

ReferencesAlfaro, S. C., and L. Gomes, Modeling mineral aerosol production by winderosion: Emission intensities and aerosol size distributions in sourceareas, J. Geophys. Res., 106, 18,075–18,084, 2001.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Modeling the sizedistribution of a soil aerosol produced by sandblasting, J. Geophys. Res.,102, 11,239–11,249, 1997.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Mineral aerosolproduction by wind erosion: Aerosol particle sizes and binding energies,Geophys. Res. Letters, 25, 991–994, 1998.

Chatenet, B., M. Marticorena, L. Gomes, and G. Bergametti, Assessing themicroped size distributions of desert soils erodible by wind, Sedimentol-ogy, 43, 901–911, 1996.

Dentener, F. J., G. R. Carmichael, Y. Zhang, J. Lelieveld, and P. J. Crutzen,Role of mineral aerosol as a reactive surface in the global troposphere,J. Geophys. Res., 101, 22,869–22,889, 1996.

Gillette, D., Environmental factors affecting dust emission by wind erosion,in Saharan dust, edited by C. Morales, pp. 71–94, John Wiley, 1979.

Iversen, J. D., and B. White, Saltation threshold on Earth, Mars and Venus,Sedimentology, 29, 111–119, 1982.

Lu, H., and Y. Shao, A new model for dust emission by saltation bombard-ment, J. Geophys. Res., 104, 16,827–16,842, 1999.

Marticorena, B., and G. Bergametti, Modeling of the atmospheric dustcycle: 1. Design of a soil derived dust emission scheme, J. Geophys.Res., 100, 16,415–16,429, 1995.

Marticorena, B., G. Bergametti, B. Aumont, Y. Callot, C. N’Doume, andM. Legrand, Modeling the Saharan dust cycle: 2. Simulation of Saharandust sources, J. Geophys. Res., 102, 4387–4404, 1997.

Myhre, G., and F. Stordal, Global sensitivity experiments of the radiativeforcing due to mineral aerosols, J. Geophys. Res., 106, 18,193–18,204,2001.

Schulz, M., Y. Balkanski, W. Guelle, and F. Dulac, Role of aerosol sizedistribution and source location in a three dimensional simulation ofa saharan dust episode tested against stellite derived optical thickness,J. Geophys. Res., 103, 10,579–10,592, 1998.

Shao, Y., A model for mineral dust erosion, J. Geophys. Res., 106, 20,239–20,254, 2001.

Shao, Y., and I. Lu, A simple expression for wind erosion threshold frictionvelocity, J. Geophys. Res., 105, 22,437–22,443, 2000.

Shao, Y., and M. Raupach, Effect of saltation bombardment by wind,J. Geophys. Res, 98, 12,719–12,726, 1993.

Shao, Y., M. R. Raupach, and J. F. Leys, A model for predicting aeoliansand drift and dust entrainment on scales from paddock to region, Aust. J.Soil Res., 34, 309–342, 1996.

Tegen, I., and I. Fung, Modeling of mineral dust in the atmosphere:Sources, transport and optical thickness, J. Geophys. Res., 99, 22,897–22,914, 1994.

��

A. Grini, Department of Geophysics, University of Oslo, P.O. Box 1022Blindern, 0315 Oslo, Norway. ([email protected])C. S. Zender, Department of Earth System Science, University of

California, Irvine, CA, 92697-3100, USA. ([email protected])P. R. Colarco, Laboratory for Atmospheric and Space Physics, Program

in Atmospheric and Oceanic Sciences, University of Colorado, Boulder,CO 80309-0392, USA. ([email protected])

15 - 4 GRINI ET AL.: SALTATION SANDBLASTING BEHAVIOR

Submitted to J. Geophys. Res. October 9, 2003. Revised December 15, 2003. Accepted December 19th 2003

Roles of saltation, sandblasting, and wind speed variability onmineral dust aerosol size distribution during the Puerto RicanDust Experiment (PRIDE)

Alf GriniDepartment of Geosciences, University of Oslo, P.O. Box 1022 Blindern, 0315 Oslo, Norway

Charles S. ZenderDepartment of Earth System Science, University of California, Irvine, CA, 92697-3100, USA

Abstract. Recent field observations demonstrate that a significant discrepancyexists between models and measurements of large dust aerosol particles at remotesites. We assess the fraction of this bias explained by assumptions involving fourdifferent dust production processes. These include dust source size distribution(constant or dynamically changing according to saltation and sandblasting theory),wind speed distributions (using mean wind or a probability density function(PDF)), parent soil aggregate size distribution, and the discretization (number ofbins) in the dust size distribution. The Dust Entrainment and Deposition (DEAD)global model is used to simulate the measurements from the Puerto Rican DustExperiment, PRIDE (2000). Using wind speed PDFs from observed (NCEP)winds results in small changes in downwind size distribution for the productionwhich neglects sandblasting, but significant changes when production includessandblasting. Saltation-sandblasting generally produces more large dust particlesthan schemes which neglect sandblasting. Parent soil aggregate size distributionis an important factor when calculating size distributed dust emissions. Changingfrom a soil with large grains to a soil with smaller grains increases by 50 % thefraction of large aerosols (D > 5 µm) modeled at Puerto Rico. Assuming that thecoarse medium sand typical of West Africa dominates all source regions producesthe best agreement with PRIDE observations.

1. Introduction

Accurate representation of generation and long-range at-mospheric transport of large particles is important for manyaerosol species including mineral dust, soot, sea salt, and icecrystals. Large particles (e.g., D > 5 µm) usually accountfor a significant fraction of the mass distribution of mostaerosols [e.g., Zender and Kiehl, 1994; Seinfeld and Pandis,1997]. If dust mass concentration and deposition measure-ments are sensitive to relatively small numbers of relativelylarge particles [Arimoto, 2001], then so too are biogeochem-ical cycles in regions, such as the Southern ocean [Mooreet al., 2002], influenced by dust-borne nutrients such as iron[Martin, 1990]. Global models of distribution of mineraldust distribution reproduce many observed features of themass concentration and optical depth [e.g., Tegen and Fung,1994; Woodward, 2001; Ginoux et al., 2001; Zender et al.,2003a]. However, there are not many detailed studies of

the dust size distribution. Recent studies underpredict theobserved size distributions of large dust particles after longrange transport [Colarco et al., 2002; Reid et al., 2003; Gi-noux, 2003].

Dust aerosols are produced by two related processes calledsaltation and sandblasting. Saltation is the net horizontalmotion of large particles moving in a turbulent near-surfacelayer. Sandblasting is the release of dust and larger parti-cles caused by saltators as they impact the surface. Duringimpact, saltators may disintegrate or eject fine dust which isotherwise tightly bound to the soil or to the saltating soil ag-gregates themselves. These two processes are documentedand discussed in several works [e.g., Gomes et al., 1990;Shao and Raupach, 1993; Alfaro et al., 1998; Shao and Lu,2000; Alfaro and Gomes, 2001]. Models can best predict thesize distribution of the dust (vertical) flux by first predictingthe size distribution of the saltation (horizontal) flux.

Large scale transport models have trouble reproducing

1

2 GRINI AND ZENDER

the transport of large dust particles measured in remote sta-tions [Colarco et al., 2002; Reid et al., 2003]. Asian mineraldust particles D > 20 µm have been measured at MaunaLoa [Perry and Cahill, 1999]. During ITCT-2k2, the MassMedian Diameter (MMD) of springtime East Asian dustreaching the west coast of North America was greater than5 µm (K. Perry, Personal communication, 2003). Reid et al.[2003] compared several measurement techniques during thePuerto Rico Dust Experiment (PRIDE) campaign (2000).Large dust particles were measured throughout the cam-paign. After evaluating optical and mass-based techniques,their best estimate of dust MMD in Barbados is 3.5 µm. Us-ing only optical measurement techniques increases this es-timate. Using only aerodynamic measurement techniquesdecreases it slightly.

Recent research has partially addressed the problem ofwhy models do not predict enough large particles at remotesites. Colarco et al. [2002] evaluated modeled size distribu-tions and compared them to measured AERONET size dis-tributions. They point to several factors which impact thedust size distribution at Barbados: 1) The size distributionof the emissions. In the cases where more large particlesare emitted, more large particles are also observed at Barba-dos. 2) Particle fall speed: Aerosols which are not spheresfall slower than spheres. Using fall speed of disks ratherthan spheres, large aerosol concentration at Barbados in-creases. They found that only the source size distribution ofGinoux et al. [2001] combined with disk fall speed gave rea-sonable size distributions. However, the production schemeof Ginoux et al. [2001] does not produce dust by saltationand sandblasting. Ginoux [2003] showed that a moment-conserving advection scheme [Prather, 1986] reduces lossesdue to gravitational settling. This explains some of the too-large loss of large particles seen in models.

Three approaches for modeling the size distributed dustproduction appear in the literature.

1. Empirical equations relate the size distributed verticaldust flux to the wind speed [ Tegen and Fung, 1994;Ginoux et al., 2001; Woodward, 2001]. This approachincludes a variety of ad hoc prescriptions and predic-tions of size distributions but does not account for thesaltation-sandblasting process. Ginoux et al. [2001]use a threshold wind friction speed u∗t that increaseswith dust aerosol diameter. Woodward [2001] calcu-lates a saltation fluxes of all soil sizes and convertsthem to dust fluxes of the same size assuming dustfluxes of a certain size are proportional to soil fluxes.

2. Saltation flux is explicitly calculated and used to pre-scribe a vertical dust flux through empirical relation-ships. This “half-way” approach explicitly representssaltation but not sandblasting. Studies which employthis approach often distribute the vertical dust flux intothree lognormal modes with different shapes and massfractions [Marticorena et al., 1997; Schulz et al., 1998;Zender et al., 2003a].

3. Saltation and sandblasting are both explicitly repre-

Table 1. Lognormal size distributions of dust produced bysaltation and sandblasting

Lognormal Mode 1 2 3

D’Almeida [1987]MMD [µm] 0.83 4.82 19.36Standard deviation 2.10 1.90 1.60Mass fraction 0.036 0.957 0.007Claquin [1999]MMD [µm] 0.011 2.54 42.10Standard deviation 1.89 2.0 2.13Mass fraction 2.6 ×10−6 0.78 0.22Alfaro and Gomes [2001]MMD [µm] 1.5 6.7 14.2Standard deviation 1.7 1.6 1.5Mass fraction varies varies varies

sented and interact in a physical way to produce theemitted dust size distribution [Shao et al., 1996; Shaoand Leslie, 1997; Gong et al., 2003]. To our knowl-edge only two distinct methods which take this ap-proach have been proposed [Shao and Raupach, 1993;Shao, 2001; Alfaro and Gomes, 2001].

Table 1 shows size distributions representative of the lat-ter two approaches. The mode with MMD ∼ 2.5 µm iswidely cited as important in dust emissions [Shettle, 1984;Schulz et al., 1998; Guelle et al., 2000; Myhre et al., 2003].However there is no physical reason for a constant mode ofthis size. Saltation sandblasting theory predicts time-varyingdust size distributions dependent on soil properties and me-teorology.

Previous studies show that Approaches 1 and 2 ade-quately simulate observed dust distributions on regional [Mar-ticorena et al., 1997] and global [Ginoux et al., 2001; Zen-der et al., 2003a] scales. Regional simulations using Ap-proach 3 [Shao and Leslie, 1997; Gong et al., 2003] showthat saltation-sandblasting well explains continental-scaledust emissions. Until now, global simulations using saltation-sandblasting microphysics have not been conducted. One ofour objectives is to show whether saltation-sandblasting pro-duction explains the observed size distribution of long-rangedust than better than other approaches and mechanisms.We find that it does. Moreover, saltation-sandblasting pro-duces dust size distributions very sensitive to surface winds.Thus our second objective is to show whether saltation-sandblasting forced by realistic wind speed distributions ex-plains the observed size distribution of long-range dust betterthan forcing with mean winds.

The paper is organized as follows. Section 2 describesthe dust production simulations and the measurements usedto evaluate them. Section 3 presents the PRIDE simulations.Section 4 summarizes the conclusions of the study.

DUST SIZE DISTRIBUTION DURING PRIDE 3

2. Theory and Methods

2.1. Dust production

This study is, to our knowledge, the first inter-continentalsimulations of dust production and transport using saltation-sandblasting theory. Our PRIDE simulations employ a se-ries of increasingly complex (and, we believe, more realistic)representations of dust production mechanisms:

1. Saltation Production: Saltation is explicitly representedthough sandblasting is not. Size distribution of sourceemissions is prescribed from observations.

2. Dynamical Dust Production: Saltation and sandblast-ing are explicitly represented using the method of Al-faro and Gomes [2001]. Large scale (grid-cell mean)winds determine dust production.

3. Wind Speed Distribution: A probability density func-tion of wind speeds [Justus et al., 1978] drives the dustproduction mechanism.

The details of each production mechanism are presented be-low.

2.1.1. Saltation-Only Production The control (here-after CTL) production method is documented and evaluatedin Zender et al. [2003a] and Zender et al. [2003b]. The salta-tion mass flux depends on the wind friction speed excessover the threshold speed for entrainment of optimally sized(D = 75 µm) particles [White, 1979; Iversen and White,1982]. A globally uniform sandblasting mass efficiency [Al-faro et al., 1997; Marticorena and Bergametti, 1995] con-verts the predicted saltation into a vertical dust flux. Thedust flux has a (globally uniform) prescribed size distribution[D’Almeida, 1987] when entrained from the surface (see Ta-ble 1). Thus the CTL mechanism accounts for saltation andneglects sandblasting. The size distribution changes duringtransport due to size-dependent wet and dry deposition.

2.1.2. Saltation-Sandblasting Production Saltation-sandblasting(hereafter SS) produces dynamic dust size distributions us-ing the formulation of Alfaro and Gomes [2001]. The sizedistributed saltation flux changes with wind friction speed[Iversen and White, 1982; Grini et al., 2002]. The dustaerosol size released by sandblasting depends on saltatorkinetic energy. Thus the dust size distribution depends onthe parent soil aggregate size distribution and the wind fric-tion speed. High kinetic energy saltators release small dustaerosols and low kinetic energy saltators release large dustaerosols [Shao and Raupach, 1993]. The soil aggregate sizedistribution is thus an important factor [Alfaro and Gomes,2001]. The kinetic energy is proportional to the mass ofthe saltator (and the square of the friction speed). There-fore coarse-textured soils produce more saltators with highkinetic energy than fine-textured soils. The computationalefficiency problems with this method noted by Grini et al.[2002] were solved storing all parameters needed in pre-computed lookup tables. The dust size distribution at thesource is then a simple function of wind friction speed andsoil size distribution. It is important to distinguish between

two types of size distributions often confounded in dust mod-eling: soil texture is the fraction of sand, silt, and clay-sizedparticles measured by soil scientists after dis-aggregating thesoil (e.g., ultrasonically in water). This study uses soil ag-gregate size distribution, the dry size distribution includingaggregates. It is most appropriately measured by dry-sievingthe soil. We assume globally uniform lognormal distribu-tions [Chatenet et al., 1996] because better data are not avail-able at the global scale.

In the CTL formulation, dust emissions commence atthe saltation threshold friction velocity, u∗t ≈ 0.20 m s−1,which is also a widely cited threshold for dust emissions[Marticorena and Bergametti, 1995; Marticorena et al., 1997].The SS formulation [Alfaro and Gomes, 2001], in contrast,produces dust only once u∗ ∼

> 0.3 m s−1, (the sandblast-ing threshold). This difference in saltation and sandblastingthresholds causes a difference in the frequency of emissionssince the CTL mechanism emits more readily than the SSmechanism. Section 3.2 discusses how this difference af-fects our results.

Emission efficiencies also differ between experiments.Zender et al. [2003a] use a fixed fixed mass sandblastingefficiency of α = 10−4 m−1 which corresponds to a soilclay content of 20% [Gillette, 1979]. The Alfaro and Gomes[2001] mass sandblasting efficiencies are generally less, andthey depend on the instantaneous saltation-sandblasting pro-cess.

2.1.3. Probability Density Function Wind Speeds TheProbability Density Function (hereafter PDF) productionmechanism uses a wind speed PDF rather than the grid-cell mean wind speed to initiate dust entrainment. Follow-ing Gillette and Passi [1988], we assume average variabilitywinds and construct a Weibull distribution with the NCEP-estimated mean wind speed U [Justus et al., 1978]. The freeparameters in the PDF are a shape factor k and a scale factorc m s−1:

kr = 0.94√

Ur (1)cr = Ur[Γ(1 + 1/k)]−1 (2)

where the subscript r denotes quantities at reference height(10 m) and Γ is the Gamma function.

Justus et al. [1978] provides the method to transfer theshape parameter from the reference height zr (10 m) to themidpoint height zatm

katm = kr[1 − 0.088 ln(zr/10)]/[1 − 0.088 ln(zatm/10)] (3)

catm comes from (2) with values for midpoint height insteadof reference height.

We discretize the central 90% of the wind PDF. The min-imum wind speed represented is slower than 95% of thewinds, and the maximum wind speed is faster than 95% thewinds. The cutoff wind speeds come from the cumulativeform of the Weibull distribution

pW (U < Ut) = 1 − exp

[

−

(

Ut

c

)k]

(4)

4 GRINI AND ZENDER

0 5 10 15 20 25 300

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Umean

Umedian

σ σ/Umean

2 1.65 1.52 0.76

MEAN WIND SPEED AND WEIBULL PDF

WIND SPEED [m/s]

PR

OB

AB

ILIT

Y

5 4.74 2.50 0.50

10 9.90 3.66 0.37

15 15.04 4.58 0.31

20 20.17 5.36 0.27

Figure 1. Weibull probability density function of windspeeds pW (U) for U = 2, 5, 10, 15, and 20 m s−1. Meanand median winds are indicated on the curves with stars andcircles, respectively.

where pW (U < Ut) is the probability of wind speed U be-ing slower than the threshold Ut. We discretize U into fivebins, calculate dust production for each bin, and weight theseproduction terms by the Weibull PDF.

Figure 1 shows the Weibull wind speed PDF pW (U). Thestandard deviation of pW (U) increases faster than U . Thus,neglect of the wind speed PDF is a worse approximation forfaster winds, when dust emission is most important.

2.2. Numerical tools for dust emission and transport

2.2.1. Dust Entrainment and Deposition Model Weuse the Dust Entrainment and Deposition model (DEAD)[Zender et al., 2003a] to calculate dust production and de-position. DEAD produces dust as a function of wind speedand stability in the boundary layer. A geomorphic erodibil-ity factor accounts for spatial heterogeneity of dust sources[Zender et al., 2003b]. Dust is transported using four sizebins with boundaries at [0.1, 1.0, 2.5, 5.0, 10] µm. Simula-tions with more than four transport bins have boundarieslogarithmically distributed between 0.1–10 µm. The dis-tribution of mass in each bin is lognormal with mass me-dian diameter of 2.5 µm and geometric standard deviationσg = 2.0. All properties specific to each bin (e.g., dry depo-sition velocity) are weighted according to this sub grid sizedistribution. In this study, DEAD has been expanded with asaltation/sandblasting module and a wind PDF module.

2.2.2. Transport model The dust is transported in theMATCH chemical transport model [Rasch et al., 1997] usingtime-interpolated six hourly meteorology from NCEP/NCARreanalyses [Kalnay, 1996]. Wet and dry deposition processesremove more clay and silt-size particles, respectively. Nucle-ation and sub-cloud scavenging are treated separately withsize-dependent scavenging coefficients which vary based on

the precipitation size distribution (stratiform or convective).Dry deposition uses a resistance method that incorporatesgravitational, turbulent, and quasi-laminar terms [Zenderet al., 2003a].

2.2.3. Box model simulations Simulations with DEADin box model mode demonstrate the effect of saltation andsandblasting on dust production. Table 2 shows the fractionof transported dust in each of the standard four size bins asa function of soil size distribution and wind speed. The soiltypes are described in Chatenet et al. [1996]. It can be seenthat increasing the parent soil size increases the fraction ofsmall dust produced (for a given wind speed). However thisis not a general conclusion, and it is dependent on the soilsize distribution [Grini et al., 2002].

2.3. Model-constraining measurements performedduring PRIDE

2.3.1. AERONET Sun-Photometers AERONET sun-photometers estimate the column aerosol phase functionand spectral optical depth from surface radiance measure-ments [Dubovik and King, 2000]. Column volume pathV [m3 m−2] and Aerosol Optical Depth (AOD) are retrievedfrom these measurements. The inversion procedure assumesthe aerosol size distribution does not change with height.The retrieval error for the size distribution is 10–35% for in-termediate size particles (0.2 < D < 14 µm), and increasesto 80–100% outside this range [Dubovik et al., 2000]. Theabsolute uncertainty in retrieved AOD is ±0.02. Thus thefractional error is large for low AOD and small for highAOD. The sun-photometers only measure during daylight.Plumes arriving at Puerto Rico during the night would not bemeasured unless they were still present the next non-cloudyday.

2.3.2. Near-surface Mass Concentration Observed dustmass concentrations are from University of Miami aerosolnetwork observations. The instruments collect aerosol ontofilters which are weighed before and after ashing. Dustconcentration is estimated as the residuals ash times 1.3,a factor that accounts for organic and volatile soil compo-nents. This technique agrees with independent estimatesfrom Al measurements [Arimoto et al., 1995; Maring et al.,2003]. The upper size limit of U. Miami measurements isabout D = 30–40 µm.

During PRIDE, concentration measurements were madebetween between local noon (16h00 UTC) and the followingnoon. Our model data are averages from 00h00–24h00 UTC.For U. Miami data, the time of mid-sample determines themeasurement date. For most measured days, mid-sample oc-curred at about 04h00 UTC. Our comparison may thereforebe inconsistent occasionally, especially on days when dustarrived late in the evening.

2.3.3. Near-surface Size Distribution The observednear-surface size distribution measurements are from theUniversity of Utah (K. Perry, Personal communication, 2002).These size distributions were retrieved using an eight-stage


Table 2. Soil type, Wind Speed, Fraction of Transported Dust a

Soil Typeb Wind Speedc 0.1–1.0 µm 1.0–2.5 µm 2.5–5.0 µm 5.0–10 µmm s−1 Fraction Fraction Fraction Fraction

Dal87 All 0.03 0.17 0.40 0.38ASS 6 0.0044 0.020 0.13 0.85ASS 12 0.0083 0.032 0.14 0.82ASS 18 0.0011 0.037 0.15 0.80FS 6 0.0052 0.022 0.14 0.84FS 12 0.012 0.042 0.15 0.79FS 18 0.017 0.056 0.16 0.77SS 6 0.011 0.038 0.15 0.80SS 12 0.051 0.15 0.18 0.62SS 18 0.093 0.26 0.19 0.46CS 6 0.014 0.048 0.16 0.77CS 12 0.080 0.23 0.19 0.50CS 18 0.14 0.39 0.19 0.28

aFraction of total 0.1 < D < 10 µm dust produced in each bin.bSources: Dal87 is “Background Mode” is from D’Almeida [1987]. All other types are from Chatenet et al. [1996]: ASS = Aluminosilicated Silt, FS =

Fine Sand, SS = Salts and clay, CS = Coarse sand.c10 m wind speed (mean of Weibull PDF).

DRUM (Davis Rotating Unit Monitoring) impactor. DRUMand MOUDI impactors had trouble measuring all large parti-cles during PRIDE [Reid et al., 2003]. The inlet aerodynam-ics of DRUM impactors have subsequently been changedto alleviate this problem (K. Perry, Personal communica-tion, 2003). However, DRUM aerodynamic measurementsprobably underestimate the mass fraction of particles withD > 5 µm during PRIDE. For these reasons, we use DRUMmeasurements primarily for size distribution estimates. Wepresent the DRUM total concentration measurements for in-formational purposes only, and as an indication of samplinguncertainty. We convert the DRUM size-distributed Si con-centration measurements to dust mass size distribution as-suming the measured dust has a constant, size-independentSi mass fraction of 0.33 [Seinfeld and Pandis, 1997].

The DRUM data do not measure the same thing as theU. Miami filters. Our results show that three times measuredSi (DRUM dust estimate) differs from 1.3 times measuredAsh (U. Miami dust estimate), The U. Miami concentrationsare about three times higher, although the peak timings arevery similar. This may be due to to measuring differentchemical species and/or to problems with the DRUM inletsize cut-off.

2.3.4. Evaluation Metrics There is a systematic biasbetween size distributions of dust retrieved from aerody-namic and from optical methods [Reid et al., 2003]. Op-tical methods in general retrieve more aerosols of largersizes. Thus simulation size distributions should be com-pared to both types of retrievals. We introduce two metrics,

m(D > 5) and V (D > 5), for these comparisons.

m(D > 5 µm) ≡

∫ D=∞

D=5D3nn(D) dD

∫ D=∞

D=0D3nn(D) dD

=m(D > 5)

m(5)

where nn(D) [# m−3 m−1] is the number distribution ofdust, m [kg m−3] is the total mass concentration of dust inthe lowest model layer, and the metric m(D > 5) is thefractional mass concentration of particles larger than five mi-crons at measurement height (or in the lowest model layer).Aerodynamic methods, e.g., DRUM impactors, allow directestimation of m(D > 5).

Similarly, surface remote sensing methods allow directestimation of column-integrated properties such as

V (D > 5 µm) ≡

∫ z=∞

z=0

∫ D=∞

D=5D3nn(D, z) dD dz

∫ z=∞

z=0

∫ D=∞

D=0D3nn(D, z) dD dz

(6)

where z [m] is height and V (D > 5) is the fractional vol-ume path of dust particles larger than five microns in theatmospheric column. AERONET retrievals allow direct es-timation of V (D > 5) [Dubovik and King, 2000]. Together,m(D > 5) and V (D > 5) provide convenient metrics toquantify the impact of dust production mechanism in eachexperiment on the large particles in the downwind size dis-tribution.

2.4. Numerical Experiments

Table 3 summarizes our numerical simulations. Modelemissions were adjusted a posteriori to yield exact agree-ment between modeled and measured (by U. Miami filters)

6 GRINI AND ZENDER

PRIDE-mean station concentrations at Roosevelt Road. Thismakes the simulated size distributions easier to intercompareby eliminating the degree of freedom associated with abso-lute mass concentration.

2.4.1. Simulations with Saltation dust source Runslabeled CTL use the saltation size distribution for emissions.Two sensitivity tests were done with the CTL production for-mulation. The CTL10 sensitivity test used ten transport binsinstead of four. The size distribution of particles within eachbin is unchanged, only the number of bins changes [Zenderet al., 2003a]. Differences between CTL10 and CTL resultsindicate the sensitivity of the transported size distribution tothe number of transport bins. These differences arise fromdiscretization of, e.g., dry deposition velocity and scaveng-ing coefficients within a bin. The CTLPDF sensitivity testuses a prescribed sub-gridscale distribution of wind speedsto determine emissions, rather than the gridcell mean windspeed. Dust emissions require wind speeds in excess of athreshold and representing the wind speed PDF allows mo-bilization to occur in areas where the gridcell-mean windspeed does not reach the threshold. Thus differences be-tween CTLPDF and CTL results indicate the sensitivity ofthe size distribution to the wind speed distribution in sourceregions.

2.4.2. Simulations with dynamical dust source In theSS and SSPDF simulations, all soils are assumed to be“Coarse Medium Sand” (CMS) [Chatenet et al., 1996; Mar-ticorena et al., 1997]. CMS is a blend of Coarse Sand (CS,90% mass fraction, MMD = 690 µm, σg = 1.6) with FineSand (FS, 10% mass fraction, MMD = 210 µm, σg = 1.6).About 50% of the surface area of the Western Sahara is CMS[Marticorena et al., 1997]. The SS experiment uses the gridmean wind speed to predict emissions, while the SSPDFexperiment uses Weibull PDF-distributed winds to predictemissions.

The SSPDFSFS experiment assumes that the soil is “SiltyFine Sand” (SFS) and that winds are Weibull-distributed.SFS is a blend of Fine Sand (FS, 62.5% mass fraction)with Aluminosilicated Soils (ASS, 37.5% mass fraction,MMD = 125 µm, σg = 1.8). SFS soils account for about10% of the Western Sahara desert [Marticorena et al., 1997].The SSPDFSFS experiment shows the sensitivity of down-wind dust distribution to soil size distribution in the sourceregion. This sensitivity is largely determined by the effi-ciency of the sandblasting process.

The mass fraction of each soil type is converted to a sur-face fraction using relations appropriate the lognormal sizedistribution. We calculate a horizontal flux of each soil typefor a given wind speed, and weight this flux by the surfacefraction covered by this soil type. The mass fraction of largesoil aggregates is smaller than the surface fraction of largeaggregates since the larger aggregates have less surface perunit mass.

Table 3. Mobilization Processes in Each SimulationSimulation Production Mechanism #a Windb Soilc

CTL Saltation only 4 Mean CMSCTL10 Saltation only 10 Mean CMS

CTLPDF Saltation only 4 PDF CMSSS Saltation-Sandblasting 4 Mean CMS

SSPDF Saltation-Sandblasting 4 PDF CMSSSPDFSFS Saltation-Sandblasting 4 PDF SFS

aNumber of transport bins.bMean = time-interpolated NCEP wind; PDF = time-interpolated NCEP

wind is mean of diagnosed Weibull PDF [Justus et al., 1978].cSoil size distribution: CMS = Coarse Medium Sand; SFS = Silty Fine

Sand [Marticorena et al., 1997].

184 186 188 190 192 194 196 198 200 202 204 2060

20

40

60

80

100

120

140

DAY IN 2000

DU

ST

CO

NC

EN

TRA

TIO

N (u

g/m

3)

Concentrations at Puerto Rico

RUN MEAN CORR(Miami) CORR(Utah)

OBS (Miami) 22.51 1.00 0.84

OBS (Utah) 7.15 0.84 1.00

CTL 22.15 0.47 0.36

CTL10 22.06 0.44 0.34

CTLPDF 22.62 0.52 0.40

SS 21.68 0.38 0.29

SSPDF 23.19 0.41 0.30

SSPDFFS 24.94 0.36 0.25

Figure 2. Near-surface dust mass concentration m [kg m−3]at Roosevelt Roads, July 2–24, 2000. U. Miami measure-ments (black solid), U. Utah measurements (black dashes),CTL (yellow solid), CTL10 (magenta circles), CTLPDF(cyan x’s), SS (red plus), SSPDF (green stars), SSPDFSFS(blue dashes).

3. Results

3.1. Near-Surface Mass Concentration

Figure 2 compares the measured mass concentration ofdust at Roosevelt Roads in Puerto Rico to all simulations (cf.Table 3). The model predicts the correct number of peaksduring PRIDE. The agreement is very good between modeland measurements. Peaks on days 189, 192, 195 and 198 arewell predicted by the model. The day 203 peak occurs twoday early in the model. Figure 2 also shows total mass con-centration based on Si measurements from the DRUM im-pactors. DRUM impactors measure less total concentrationthan the U. Miami filters. DRUM and U. Miami observationsagree in timing but not magnitude of dust events. This can beboth because of different size cut-off and different chemical


180 185 190 195 200 2050

0.2

0.4

0.6

0.8

1

1.2

DAY IN 2000

OP

TIC

AL

DE

PTH

Aerosol optical depth at Puerto Rico

RUN MEAN CORRCOEFF

OBS: 0.27 1.00

CTL: 0.22 0.32

CTL10 0.22 0.35

CTLPDF 0.21 0.32

SS 0.22 0.32

SSPDF 0.23 0.32

SSPDFSFS 0.18 0.33

Figure 3. Aerosol Optical Depth at Roosevelt Roads,June 27–July 23, 2000. AERONET measurements (blackstars), CTL (yellow solid), CTL10 (magenta circles),CTLPDF (cyan x’s), SS (red plus), SSPDF (green stars),SSPDFSFS (blue dashes).

analysis (see Sections 2.3.2 and 2.3.3).

3.2. Aerosol Optical Depth

AERONET observations are available on nearly all dayswith significant cloud-free periods. We compute mean AODand correlation coefficients to AERONET using only dayswhen measured AOD is available. Figure 3 compares AERONETobservations to the simulated AOD at Puerto Rico. Themodel under-predicts AOD during most dust events. Thepredicted timing is accurate for days 180 and 202, and toolate on days 186 and 190. The model predicts two indepen-dent peaks on days 195 and 197 instead of the single peakobserved on day 196.

The observed optical depths are generally higher thanthe modeled ones. This is partially due to other naturalaerosols (e.g., sea salt) or aerosols from industrial pollutionor biomass burning. The 340/380 nm Angstrøm exponentα for AOD measurements at Puerto Rico is about 0.5 forthe periods of high optical depth during PRIDE. The lowAngstrøm exponent indicates that the aerosol is dust. α in-creases to 1.5–2.5 when optical depth is low (near 0.1). Thusabout 0.1 of AOD may be explained by small aerosols in-cluding industrial pollution.

Figure 4 compares the simulated AOD to AERONET ob-servations at Cape Verde, about 700 km from the Africancontinent. The SS and SSPDF production mechanisms yieldunreasonably large optical depths at Cape Verde. Figure 4serves two purposes. First, it verifies that model and mea-sured values reasonably agree for dust crossing Cape Verde.Much of the dust reaching Puerto Rico first crosses CapeVerde. The transit time from Sahara to Puerto Rico is aboutfive days [Mahowald et al., 2002; Colarco et al., 2002]. To

170 175 180 185 190 1950

0.5

1

1.5

2

2.5

3

3.5

4

4.5

DAY IN 2000

OP

TIC

AL

DE

PTH

Aerosol optical depth at Cape Verde

RUN MEAN CORRCOEFF

OBS: 0.53 1.00

CTL: 0.87 0.66

CTL10 0.67 0.66

CTLPDF 0.82 0.67

SS 0.94 0.66

SSPDF 0.96 0.65

SSPDFSFS 0.86 0.63

Figure 4. Aerosol Optical Depth at Cape Verde, June 16 toJuly 16, 2000. AERONET measurements (black stars), CTL(yellow solid), CTL10 (magenta circles), CTLPDF (cyanx’s), SS (red plus), SSPDF (green stars), SSPDFSFS (bluedashes).

compare the same dust, the Cape Verde results are shown foran earlier period than results for Puerto Rico. A high corre-lation at Cape Verde five days before makes us confident thata high correlation at Puerto Rico is due to correct timing ofdust production events.

Secondly, Figure 4 verifies that emission frequency dif-ferences do not propagate far from the source region dur-ing the simulation period. As mentioned in Section 2.1.2,the saltation-sandblasting formulations only emit dust whenwind friction speeds exceed 0.3 m s−1 whereas the saltation-only formulations produce dust when the wind friction speedsexceed 0.2 m s−1. The simulated frequency of dust eventscrossing Cape Verde shows little sensitivity to the dust pro-duction formulation. Thus our simulated differences in dustproduction frequency are either small, or they do not influ-ence dust concentrations close to the source. This is consis-tent with Colarco et al. [2002], who found that increasingwind speed threshold from the formulation of Marticorenaand Bergametti [1995] (with u∗t = 0.2 m s−1) to a fixed10 m wind speed threshold (U10,t = 6.5 m s−1) does notchange the frequency, but does change the magnitude of dustevents.

Figure 5 shows the covariation of observed mass concen-tration and AOD at Puerto Rico. Near-surface mass concen-tration peaks generally occur later than AOD maxima. Webelieve that the tropical easterly winds transport dust aloftmore rapidly than near-surface dust. The lag causes the verylow correlation coefficient (r = −0.17) even though bothmeasurements clearly show the same dust events. The dif-ference may also be due to U. Miami sampling local noon tolocal noon whereas AERONET measures during local day-light only.

8 GRINI AND ZENDER

10−1 100 101

10−5

10−4

10−3

10−2

10−1

Aerosol size (um)

volu

me

path

dV

/dln

r (um

3/um

2)

187

10−1 100 101

10−5

10−4

10−3

10−2

10−1

Aerosol size (um)vo

lum

e pa

th d

V/d

lnr (

um3/

um2)

195

10−1 100 101

10−5

10−4

10−3

10−2

10−1

Aerosol size (um)

volu

me

path

dV

/dln

r (um

3/um

2)

197

10−1 100 101

10−5

10−4

10−3

10−2

10−1

Aerosol size (um)

volu

me

path

dV

/dln

r (um

3/um

2)

202

Figure 6. Columnar volume distributions nv(D) [m3 m−2] at Puerto Rico in 2000 for July (day of month)/(day of year)(a) 5/187, (b) 13/195, (c) 15/197, and (d) 20/202. Measurements (black), CTL (yellow), CTL10 (magenta), CTLPDF (cyan),SS (red), SSPDF (green), SSPDFSFS (blue).

3.3. Size distributions from Sun-Photometry

Figure 6 shows the columnar volume distribution fromAERONET sun-photometry and simulations. Size distribu-tions are shown for days 187, 195, 197, and 202. Bothmodel and observations show high optical depths on thesedays (Figure 3). The low bias in simulated silt-sized parti-cles (D > 2.5 µm) approaches and often exceeds an order ofmagnitude for all production mechanisms. Locally producedsea-salt and crustal aerosol may contribute to the observedsilt-sized particles. The difference in size distribution is notonly attributable to dust production. Errors in dry or wet de-position rates also influence the results. Since no productionmechanism reproduces the flatness of the AERONET sizedistributions, our deposition rates of large aerosols may betoo high.

The fixed production formulations (CTL and CTL10)yield higher volume paths in the 1–5 µm size range. CTL10

yields more clay (D < 2.5 µm) particles than CTL becausethe wet scavenging coefficients for clay in CTL10 are nottuned as they are in the other simulations [Zender et al.,2003a]. CTL10 averages the sedimentation velocity of eachsize over a narrower bin, thus large particles fall at speedscloser to reality. This explains the reduced volume in thelargest sizes in CTL10. As we show later, these differencesdue to bin discretization are small compared to changingfrom fixed to dynamical source schemes.

Among dynamical schemes, SSPDF produces the mostsmall aerosols and SSPDFSFS produces the most large aerosols.Column volume paths of large sizes in the sandblasting sim-ulations vary with the meteorological conditions during dustproduction. Figure 6 shows that SSPDFSFS consistentlyproduces the most large particles. Small saltators dominatethe SFS soil blend (cf. Section 2.4.2) and preferentially sand-blast large dust particles with low binding energies [Alfaro


175 180 185 190 195 200 205 2100

10

20

30

40

50

60

70

DAY IN 2000

100x

AO

T(44

0nm

), C

ON

CE

NTR

ATI

ON

(ug/

m3)

PUERTO RICO: 100xAOT (440nm) AND GROUND CONCENTRATION (ug/m3)

Correlation coefficient: −0.17

Figure 5. Co-variation of observed optical depth and con-centration at Puerto Rico. AOD is multiplied by 100 and isrepresented by stars/dashed line. The mass concentrationsare represented by circles/solid line.

and Gomes, 2001]. Thus sandblasting theory demonstratesthe sensitivity of the downwind size distribution to the soilsize distribution in source regions, which is usually poorlyconstrained.

3.4. Near-Surface Size Distribution

Next we contrast the observed (Section 2.3.3) and mod-eled near-surface mass distributions. Figure 7 shows massdistributions for days 188, 189, 197, and 198. Both modeland observations show relatively high concentration on thesedays (Figure 2). Saltation-sandblasting produces larger vari-ations among events than saltation alone. The model predictsmore dust than DRUM measurements because it is tuned tomatch U. Miami dust concentration measurements which, ingeneral, exceed the aggregated DRUM measurements.

Note that SSPDFSFS consistently predicts the most largeaerosols while the differences between SS and SSPDF pre-dictions depend on day and event. We also note an impor-tant aspect of using soil blends: Blending two soil size dis-tribution does not result in a linear contribution from eachsoil proportional to their mass fractions. The effects of asmall fraction of fine-textured soil may dominate the massflux produced by a coarse-textured soil for two reasons.

1. Fine soils have larger influences than indicated bytheir mass fractions. We model the saltation flux froma soil as proportional to the surface covered by thatsoil. Even for small mass fractions of fine soil, thesurface fraction can be significant.

2. Fine soils sandblast more large aerosols and coarsesoils sandblast more small aerosols. Thus fine soilshave much larger mass sandblasting efficiencies. Usu-ally the sandblasting performed by the fine soil frac-

OBS CTL CTL10 CTLPDF SS SSPDF SSPDFSFS0

0.05

0.1

0.15

0.2

0.25

0.3

0.35m(D>5um)

Figure 8. Observed and simulated m(D > 5), the near-surface fractional mass concentration of particles larger than5 µm.

OBS CTL CTL10 CTLPDF SS SSPDF SSPDFSFS0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45V(D>5um)

Figure 9. Observed and simulated V (D > 5), the fractionalvolume path of dust particles larger than 5 µm.

tion of a soil blend dominates the dust mass flux.

3.5. Fraction of Large Aerosols

Figures 8 and 9 compare the mean observed and modeledm(D > 5) and V (D > 5) (Equations 5 and 6), respec-tively, for the PRIDE period. Saltation-sandblasting (SS)physics consistently produces more large aerosols than thesaltation-only (CTL) mechanism. The SS source distribu-tion contains modes with MMDs at 6.7 and 14.2 µm [Al-faro and Gomes, 2001]. These modes activate first as slowerwinds are required to exceed their threshold velocity. Thesemodes are significantly larger than the4.82 µm mass mediandust diameter [D’Almeida, 1987] prescribed at the source inthe CTL simulations. Prescribing a larger diameter at the

10 GRINI AND ZENDER

10−1 100 1010

10

20

30

40

50

60

70

80 188

Diameter (um)

dM/d

lnD

(ug/

m3)

10−1 100 1010

10

20

30

40

50

60

70 189

Diameter (um)

dM/d

lnD

(ug/

m3)

10−1 100 1010

2

4

6

8

10

12

14

16

18 197

Diameter (um)

dM/d

lnD

(ug/

m3)

10−1 100 1010

5

10

15

20

25 198

Diameter (um)

dM/d

lnD

(ug/

m3)

Figure 7. Near-surface mass distributions nm(D) [µg m−3] at Puerto Rico in 2000 for July (day of month)/(day of year)(a) 6/188, (b) 7/189, (c) 15/197, and (d) 16/198. Measurements (black), CTL (yellow), CTL10 (magenta), CTLPDF (cyan),SS (red), SSPDF (green), SSPDFSFS (blue). Note difference in scales between the four figures.

source would produce more large aerosols at the source in-dependent of wind speed. As mentioned in Section 2.3.3,the DRUM size-resolved aerodynamic measurements mayunderestimate the mass fraction of particles with D > 5 µm.This would bring the m(D > 5) (5) simulated by saltation-sandblasting processes into even better agreement with ob-servations.

Both the SS and the SSPDF experiments use the same soilsize distribution yet SSPDF transports fewer large aerosols.Sandblasting physics explain this outcome. In SSPDF, asmall but significant fraction of the winds are fairly fast (cf.Figure 1). Faster winds sandblast more small aerosols thanslower winds provided that the saltators are large enough toexceed the binding energy of the finest dust. Sufficientlylarge saltators are required to sandblast fine dust. Increas-ing mean winds produces more small particles for the soilsize distributions that we employ (Table 2), but this is not ageneral rule.

The difference between SSPDF and SSPDFSFS is strik-ing in Figures 8 and 9. Accurate predictions of sandblasting,

and thus dust emissions and downwind size distributions,requires detailed source soil size distribution data. Thesedata must include the soil size distribution of all potentialsaltators—roughly 40 < D < 700 µm [Iversen and White,1982]. Remote sensing is one promising avenue for produc-ing these data [Okin and Painter, 2003].

Our sub-gridscale wind distribution, the Weibull distribu-tion, is based on empirical parameterizations [Justus et al.,1978] previously used in other wind erosion models [Gilletteand Passi, 1988; Shao et al., 1996]. In this formulation, windspeed standard deviation (and thus variability or gustiness)increases with the mean wind speed. Figure 1 shows that thestandard deviation increases with the wind speed. Howeverthe standard deviation decreases relative to the mean wind(from 76% for 2 m s−1 to 27% for 20 m s−1). In reality, windspeed variability is also linked to boundary layer turbulenceand to both dry and wet convection. Since these processesare simulated in 3D meteorological models, they could belinked into the variability of the wind PDF (Equation 2).

In 2002, the MMD of springtime East Asian dust did


Table 4. Change in Size Distribution During Trans-Atlantic TransportRegiona Day of Year Property 0.1–1.0 µm 1.0–2.5 µm 2.5–5.0 µm 5.0–10 µm

Fraction Fraction Fraction Fraction

Sahara 170-200 Production flux 0.021 0.069 0.16 0.75Sahara 170-200 Dust concentration 0.057 0.17 0.24 0.53

Puerto Rico 175-205 Dust concentration 0.20 0.40 0.25 0.14

aThe Sahara is defined as−20◦ < longitude < 35

◦E, 10◦ < latitude < 35

◦N, Puerto Rico as −90◦ < longitude < −60

◦E, 15◦ < latitude < 25◦N

not change much between measurements at Gosan, Sea ofJapan, and Crater Lake, Oregon, where it exceeded 5 µm af-ter crossing the Pacific! (K. Perry, Personal communication,2003). The trans-Atlantic crossing time of African dust isabout five days [Mahowald et al., 2002; Colarco et al., 2002]similar to the trans-Pacific crossing time of Asian dust whichis embedded in a stronger westerly flow. Table 4 shows thesimulated change in monthly size distribution from the Sa-haran source region to Puerto Rico in the SSPDF simula-tion. The simulated fraction of aerosols larger than 5 µmdecreases from 0.75 in emission, to 0.53 in near-Sahara con-centration, to 0.14 in near-Puerto Rico concentration. Thussilty dust appears to settle or wash out much more rapidly inour model than in observations.

4. Conclusions

Three different source formulations for dust productionwere used in a chemical transport model to simulate emis-sions and long range transport of mineral dust during thePRIDE campaign. The model results were compared to mea-surements of aerosol size distribution, mass concentration,and optical depth from optical, aerodynamic, and filter in-struments. Physically based sensitivity studies addressed themechanisms which could explain why mineral dust emis-sion and all known transport models underpredict long rangetransport of the large particles [Colarco et al., 2002; Ginoux,2003] seen in observations [Perry and Cahill, 1999; Reidet al., 2003].

Previous model attempts to reproduce the large particlesize distribution measured downwind of dust sources fo-cused on aspherical particle shape effects and on improv-ing advection schemes [Colarco et al., 2002; Ginoux, 2003].These studies left a significant fraction of the large parti-cles unexplained. We showed that a production mechanismwhich accounts for saltation but neglects sandblasting un-derpredicts the fraction of particles larger than five micronsobserved after long range transport. A production mecha-nism which accounts for saltation and sandblasting producesgreater fractions of large particles, in closer agreement withobservations. Contingent on the sandblasting process andsaltator population, realistic wind speed variability tends toreduce the emitted and transported fraction of large particles.Saltation sandblasting, in conjunction with wind speed vari-ability, significantly reduces, and, in some cases, eliminates,

model biases in production and long range transport of largedust particles.

The observed fraction of long range transported largeaerosol (D > 5 µm) is the net result of a complex interac-tion of non-linear processes where saltation sandblasting andwind variability interact. Whether each process increases ordecreases the mobilization of large particles depends on thevalues of the other, and on the boundary conditions (par-ent soil size distribution). Thus saltation-sandblasting andwind variability ought to be considered together. We con-clude that a significant fraction of the remaining discrepancyis explained by interactions of saltation sandblasting, windvariability, and soil size distribution in source regions.

The downwind size distribution is very sensitive to thesoil size distribution in source regions, which is usuallypoorly constrained. Assuming most source soil is the coarsemedium sand typical of West Africa produces the best agree-ment with observations during PRIDE. This study points outthe need for improved global soil size distribution datasetthat goes beyond the sand, silt, and clay texture classifica-tion traditionally used by soil scientists. The dataset mustinclude as much information as possible on the soil sizedistribution of all potential saltators—roughly 40 < D <700 µm [Iversen and White, 1982]. To reproduce observa-tions, global models must predict the correct emitted sizedistribution and size-dependent deposition processes duringtransport. Since reliable soil size distribution and wind vari-ability data are unavailable globally, it is difficult to assesswhether global dust transport models err more in productionor in transport. The ideal experiment to isolate these pro-cesses would be Lagrangian monitoring of dust in a trans-oceanic plume.

Acknowledgments. AG acknowledges support from the Nor-wegian research council grant 139810/720 (CHEMCLIM). CSZgratefully acknowledges support from NASA grants NAG5-10147(IDS) and NAG5-10546 (NIP). We thank Kevin Perry (U. Utah)and Dennis Savoie and Hal Maring (U. Miami) for sharing theirdata. We thank Jeff Reid and all the organizers and contributors toPRIDE for coordinating this useful experiment. Comments fromtwo anonymous reviewers helped improve this paper.

ReferencesAlfaro, S. C., and L. Gomes, Modeling mineral aerosol production

by wind erosion: Emission intensities and aerosol size distri-

12 GRINI AND ZENDER

butions in source areas, J. Geophys. Res., 106, 18,075–18,084,2001.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Modelingthe size distribution of a soil aerosol produced by sandblasting,J. Geophys. Res., 102, 11,239–11,249, 1997.

Alfaro, S. C., A. Gaudichet, L. Gomes, and M. Maille, Mineralaerosol production by wind erosion: aerosol particle sizes andbinding energies, Geophys. Res. Letters, 25, 991–994, 1998.

Arimoto, R., Eolian dust and climate: relationships to sources, tro-pospheric chemistry, transport and deposition, Earth Sci. Revs.,54, 29–42, 2001.

Arimoto, R., R. A. Duce, B. J. Ray, W. G. Ellis Jr., J. D. Cullen, andJ. T. Merrill, Trace elements in the atmosphere over the NorthAtlantic, J. Geophys. Res., , 100, 1199–1213, 1995.

Chatenet, B., M. Marticorena, L. Gomes, and G. Bergametti, As-sessing the microped size distributions of desert soils erodibleby wind, Sedimentology, 43, 901–911, 1996.

Claquin, T., Modelisation de la mineralogie et du forcage radiatifdes poussieres desertiques, Ph.D. thesis, University of Hamburg,1999.

Colarco, P., O. Toon, and B. Holben, Saharan dust transport to thecaribbean during PRIDE: Part 1. Influence of dust sources andremoval mechanisms on the timing and magnitude of downwindAOD events from simulations of and remote sensing observa-tions, J. Geophys. Res., 108, 8589, doi:10.1029/2002JD002,658,2002.

D’Almeida, G. A., On the variablility of desert aerosol radiativecharacteristics, J. Geophys. Res., , 92, 3017–3026, 1987.

Dubovik, O., and M. King, A flexible inversion algorithm forrtrievel of aerosol optical properties from sun and sky radiancemeasurements, J. Geophys. Res., 105, 20,673–20,696, 2000.

Dubovik, O., A. Smirnov, B. Holben, M. King, Y. Kaufman, T. Eck,and I. Slutsker, Accuracy assessment of aerosol optical proper-ties retrieval from aeronet sun and sky radiance measurements,J. Geophys. Res, 105, 9791–9806, 2000.

Gillette, D., Environmental factors affecting dust emission by winderosion, in Saharan dust, edited by C. Morales, pp. 71–94, JohnWiley, 1979.

Gillette, D. A., and R. Passi, Modeling dust emission caused bywind erosion, J. Geophys. Res., 93, 14,233–14,242, 1988.

Ginoux, P., Effects of non-sphericity on mineral dust modeling,J. Geophys. Res., 108, 4052, doi:10.1029/2002JD002,516, 2003.

Ginoux, P., M. Chin, I. Tegen, J. Prospero, B. Holben, O. Dubovik,and S.-J. Lin, Sources and distributions of dust aerosols simu-lated with the GOCART model, J. Geophys. Res., 106, 20,555–20,273, 2001.

Gomes, L., G. Bergametti, G. Coude-Gaussen, and P. Rognon, Sub-micron desert dusts: A sandblasting process, J. Geophys. Res.,95, 13,927–13,935, 1990.

Gong, S. L., X. Y. Zhang, T. L. Zhao, I. G. McKendry, D. A.Jaffe, and N. M. Lu, Characterization of soil dust aerosol inChina and its transport and distribution during 2001 ACE-Asia:2. Model simulation and validation, J. Geophys. Res., 108, 4262doi:10.1029/2002JD002,633, 2003.

Grini, A., C. S. Zender, and P. Colarco, Saltation sandblasting be-havior during mineral dust aerosol production, Geophys. Res.Lett., 29, 1868, doi:10.1029/2002GL015,248, 2002.

Guelle, W., Y. Balkanski, M. Shulz, B. Marticorena, G. Bergametti,C. Moulin, R. Arimoto, and K. D. Perry, Modeling the atmo-spheric distribution of mineral aerosol: Comparison with groundmeasurements and satellite observations for yearly and synoptictimescales over the north atlantic, J. Geophys. Res., , 105, 1997–2012, 2000.

Iversen, J. D., and B. R. White, Saltation threshold on Earth, Mars,and Venus, Sedimentology, 29, 111–119, 1982.

Justus, C., W. Hargraves, A. Mikhail, and D. Graber, Methods forestimating wind speed frequency distributions, J. App. Met., 17,350–353, 1978.

Kalnay, E., The NCEP/NCAR 40-year reanalysis project, Bull. Am.Meteorol. Soc., , 77, 437–471, 1996.

Mahowald, N. M., C. S. Zender, C. Luo, D. Savoie, O. Tor-res, and J. del Corral, Understanding the 30 year Bar-bados desert dust record, J. Geophys. Res., 107, 4561,doi:10.1029/2002JD002,097, 2002.

Maring, H., D. L. Savoie, M. A. Izaguirre, L. Custals, and J. S.Reid, Mineral dust aerosol size distribution change during at-mospheric transport, In Press in J. Geophys. Res., , 2003.

Marticorena, B., and G. Bergametti, Modeling the atmospheric dustcycle: 1. Design of a soil-derived dust emission scheme, J. Geo-phys. Res., 100, 16,415–16,430, 1995.

Marticorena, B., G. Bergametti, B. Aumont, Y. Callot,C. N’Doume, and M. Legrand, Modeling the atmospheric dustcycle: 2. Simulation of Saharan dust sources, J. Geophys. Res.,102, 4387–4404, 1997.

Martin, J. H., Glacial-interglacial CO2 change: The iron hypothe-sis, Paleoceanography, 5, 1–13, 1990.

Moore, J. K., S. C. Doney, D. M. Glover, and I. Y. Fung, Ironcycling and nutrient-limitation patterns in surface waters of theWorld Ocean, Deep-Sea Res. II, 49, 463–507, 2002.

Myhre, G., A. Grini, J. Haywood, F. Stordal, B. Chatenet, D. Tanre,J. Sundet, and I. Isaksen, Modelling the radiative impact of min-eral dust during the saharan dust experiment (SHADE) cam-paign, J. Geophys. Res, 108, 8579, 2003.

Okin, G. S., and T. H. Painter, Effect of grain size on spectral re-flectance of sandy desert surfaces, Submitted to J. Geophys. Res.,2003.

Perry, K. D., and T. A. Cahill, Long-range transport of anthro-pogenic aerosols to the National Oceanic and Atmospheric Ad-ministration baseline station at Mauna Loa Observatory, Hawaii,J. Geophys. Res., 104, 18,521–18,533, 1999.

Prather, M. J., Numerical advection by conservation of second-order moments, J. Geophys. Res., 91, 6671–6681, 1986.

Rasch, P., N. Mahowald, and B. Eaton, Representation of trans-port, convection and the hydrological cycle in chemical trans-port models: Implications for the modeling of short lived andsoluble species, J. Geophys. Res., 102, 28,127–28,138, 1997.

Reid, J. S., et al., Comparison of size and morphological measure-ments of coarse mode dust particles from Africa, J. Geophys.Res., 108, 8593, doi:10.1029/2002JD002,485, 2003.

Schulz, M., Y. Balkanski, W. Guelle, and F. Dulac, Role of aerosolsize distribution and source location in a three dimensional sim-ulation of a saharan dust episode tested against stellite derivedoptical thickness, J. Geophys. Res., 103, 10,579–10,592, 1998.

Seinfeld, J. H., and S. N. Pandis, Atmospheric Chemistry andPhysics, John Wiley & Sons, New York, NY, 1997.

Shao, Y., A model for mineral dust erosion, J. Geophys. Res., 106,20,239–20,254, 2001.

Shao, Y., and L. M. Leslie, Wind erosion prediction over the Aus-tralian continent, J. Geophys. Res., 102, 30,091–30,105, 1997.

Shao, Y., and I. Lu, A simple expression for wind erosion thresholdfriction velocity, J. Geophys. Res., 105, 22,437–22,443, 2000.

Shao, Y., and M. Raupach, Effect of saltation bombardment bywind, J. Geophys. Res, 98, 12,719–12,726, 1993.

Shao, Y., M. R. Raupach, and J. F. Leys, A model for predictingaeolian sand drift and dust entrainment on scales from paddockto region, Aust. J. Soil Res., 34, 309–342, 1996.


Shettle, E. P., Optical and radiative properties of a desert aerosolmodel, in Proceedings of the symposium on radiation in the at-mosphere, edited by G. Fiocco, pp. 74–77, 1984.

Tegen, I., and I. Fung, Modeling of mineral dust in the atmosphere:Sources, transport and optical thickness, J. Geophys. Res., 99,22,897–22,914, 1994.

White, B. R., Soil transport by winds on Mars, J. Geophys. Res.,84, 4643–4651, 1979.

Woodward, S., Modeling the atmospheric life-cycle and radiativeimpact of mineral dust in the Hadley center climate model, J.Geophys. Res., 106, 2001.

Zender, C. S., and J. T. Kiehl, Radiative sensitivities of tropicalanvils to small ice crystals, J. Geophys. Res., 99, 25,869–25,880,1994.

Zender, C. S., H. Bian, and D. Newman, Mineral DustEntrainment And Deposition (DEAD) model: Descriptionand 1990s dust climatology, J. Geophys. Res., 108, 4416,doi:10.1029/2002JD002,775, 2003a.

Zender, C. S., D. J. Newman, and O. Torres, Spatial hetero-geneity in aeolian erodibility: Uniform, topographic, geomor-phic, and hydrologic hypotheses, J. Geophys. Res., 108, 4543,doi:10.1029/2002JD003,039, 2003b.

Alf Grini, Department of Geosciences, University ofOslo, P.O. Box 1022 Blindern, 0315 Oslo, Norway. (e-mail:[email protected])

Charles S. Zender, Department of Earth System Science,University of California, Irvine, CA, 92697-3100, USA (e-mail: [email protected])October 9, 2003; revised December 15, 2003; accepted Decem-ber 19, 2003.

This preprint was prepared with AGU’s LATEX macros v4, with the ex-tension package ‘AGU++’ by P. W. Daly, version 1.6b from 1999/08/19.

Institute Report Series, Department of Geosciences, University of Oslo, 124, 2003, ISBN 82-91885-26-5

Model simulations of dust sources and transport in the globaltroposphere. Effects of soil erodibility and wind speed variability

Alf Grini, Gunnar Myhre

Department of Geosciences, University of Oslo, P.O. Box 1022 Blindern, 0315 Oslo, Norway

Charles S. ZenderDepartment of Earth System Science, University of California, Irvine, CA, 92697-3100, USA

Jostein K. Sundet and Ivar S.A. IsaksenDepartment of Geosciences, University of Oslo, P.O. Box 1022 Blindern, 0315 Oslo, Norway

Abstract. Global atmospheric dust is simulated using DEAD (Dust Entrainmentand Deposition model) in combination with the chemical transport model OsloCTM2 with meteorological data for 1996. Dust sources are calculated bothusing wind speeds in model resolution (1.9 x 1.9 degrees) and using differentassumptions on soil erodibility and on wind speeds. Some aspects of the annualdust cycle (such as the east Asian dust emissions) are largely dependent on thedata used to determine soil erodibility. Other aspects (such as the timing of themaximum in the African plume at Northern hemisphere summer) is well modeledwith all datasets applied here. We show that the daily variation in optical depthat Cap Verde at the west coast of Africa is very well simulated assuming thaterodibility is correlated with surface reflectivity from MODIS satellite data. Usinga sub-grid probability density function of wind speed to drive the dust sourcesfacilitates dust emissions in areas with low wind speeds. Dust concentrations inremote areas are sensitive to the parameterization of wet deposition. Our resultspoint out the need for a detailed soil erodibility dataset for global dust modeling,and they suggest that MODIS surface reflectivity data is potentially valuable forevaluating such datasets.

1. Introduction

Atmospheric mineral dust plays a role in regulation of theearth’s climate.

The dust aerosols modify the atmospheric radiation bal-ance by scattering and absorbing radiation and in this way al-ter the Earth’s energy budget [Tegen and Lacis, 1996; Myhreand Stordal, 2001]. Haywood et al. [1999] showed the im-portance of aerosols for the global radiative balance whencalculated fluxes of reflected sunlight from a General Circu-lation Model (GCM) and found that they could not matchEarth radiative budget Experiment (ERBE) measurementsunless aerosols were taken into account. The minerals caninteract both with solar and terrestrial radiation.

Tegen et al. [1996] proposed that mineral emissions fromdisturbed soils could cause a local top of the atmosphereforcing of -2.1 to +5.5 W m−2 giving a globally averagedforcing of +0.09 W m−2 for anthropogenic dust and +0.14

W m−2 for total dust. Later studies have proposed that Sa-haran dust has a rather high Single Scattering Albedo (SSA)around 0.95 or higher at 550 nm [Kaufman et al., 2001]leading mostly to a cooling effect of dust aerosols. Measure-ments during the SHADE experiments confirmed this highvalue for [Tanre et al., 2003].

The minerals can also interact with atmospheric photo-chemistry. Dentener et al. [1996] proposed an important de-pletion of NOx on dust. Bian and Zender [2003] showed thatdust can suppress atmospheric OH significantly in sourceareas due to lower incoming UV radiation. Heterogeneouschemistry and photolysis change can have opposite effects.For example will dust in remote areas lead to less photo-chemical destruction of Ozone. On the same time it willlead to loss of Ozone due to heterogeneous uptake on thedust aerosols.

Aerosols can serve as cloud condensation nuclei [Prup-pacher and Klett, 1997]. Mineral aerosols are also supposed

1

2 GRINI ET AL.

to be important as ice nuclei, and they are therefore impor-tant for ice cloud formation [Lohmann, 2002]. Sherwood[2002] argued that dust or biomass can influence strato-spheric water vapor because a decrease in the size of ice nu-clei (and an increase in their number concentration) wouldincrease evaporation near the tropical tropopause.

Dust contribute to transport of nutrients to ocean areas[Prospero, 1996]. Iron can regulate production of phyto-plankton in the oceans. The phytoplankton can be one of thefactors regulating the atmospheric CO2 flux to the oceans.Falkowski et al. [1998] proposed that in the contemporaryocean, photosynthetic carbon fixation by marine phytoplank-ton leads to formation of 45 gigatons of organic carbon perannum of which 16 gigatons are exported to the ocean inte-rior. It has also been proposed that dust can bring nutrientssuch as phosphate to the Amazon rain forest [Swap et al.,1992]

All the factors described above makes it important to un-derstand what factors regulate the production, transport andloss of mineral dust in the global atmosphere.

Several studies have tried to quantify the global produc-tion and transport of atmospheric mineral dust [Tegen andFung, 1994; Claquin, 1999; Ginoux et al., 2001; Woodward,2001; Zender et al., 2003a]. Early estimates for global pro-duction lies in the range 500 - 5000 Tg/yr. This number islargely dependent on the size distribution of the dust whichis produced. Prospero [2003] evaluated dust concentrationdata at Barbados together with Saharan rainfall data and pro-posed that dust production is closely coupled to rainfall inthe prior year. The uncertainties are large with respect toproduction, loss and transport of dust, and with respect tophysical and chemical characteristics of dust (e.g. Myhreand Stordal [2001]).

Early modeling studies (e.g. Tegen and Fung [1994] cal-culated dust production and transport using as hypothesisthat all desert was equally available for soil erosion. Sup-posing that all deserts are erodible and changing vegeta-tion data results in large change in dust production. Thosechanges were attributed to human influence by Tegen andLacis [1996] and Sokolik and Toon [1996]. The weaknessof those studies is that they give a large human influence ondust production without taking into account that soils mightbe non erodible.

Later studies [Ginoux et al., 2001; Prospero et al., 2002]have focused on the existence of dust “hot spots” which aredry lakes where sediments earlier have accumulated and arenow released. The main conclusion from these works isthat dust production is closely tied to the water cycle andriver flow. If there is a human influence on dust production,this influence must be examined in connection with dryingof rivers and lakes, and not only with changed vegetation.Applying these theories, Tegen et al. [2002] simulated soilerodibility using a water routing and storage model. Zenderet al. [2003b] calculated the erodibility by assuming it wasproportional to the upstream area from which sediments mayhave accumulated locally through all climate regimes.

Schaaf et al. [2002] showed how land types can be ob-served retrieving surface reflection with the MODerate Imag-ing Spetroradiometer (MODIS) satellite. Even though sur-face reflection is largely dependent on soil types, Tsvetsin-skaya et al. [2002] point out that sand dunes have the highestsurface reflectivity of all surface types.

We suppose that sand dunes are easily erodible. Thereforewe would like to find out whether areas with high reflectivityalso are areas with high erodibility. To examine the validityof this hypothesis, we run a 3D global dust production andtransport model drive by meteorological data from 1996 tocalculate atmospheric dust production and transport.

Since our goal is to explore the connection between soilreflectivity and soil erodibility, we introduce two new erodi-bility factors based on MODIS surface reflection and com-pare them to model results using the earlier published erodi-bility factors.

We also explore the importance of using a probabilitydensity function of wind speeds [Justus et al., 1978] to drivedust production rather than using mean winds at model res-olution.

We confirm that global dust budgets and concentrationsfrom the DEAD/Oslo CTM2 model are within reasonablerange. We compare our results both to earlier published re-sults and to measurements.

2. Modeling

2.1. Dust production

Dust emissions are modeled using the Dust Entrainmentand Deposition model (DEAD) [Zender et al., 2003a]. Thismodel is based on the work of Marticorena and Bergametti[1995]. Emissions start when wind friction speeds reach athreshold wind friction speed of approximately 0.2 m s−1[Iversen and White, 1982]. A horizontal saltation (soil) fluxand a vertical (dust) flux are calculated. The size distributionof the vertical dust flux is distributed to three modes accord-ing to [D’Almeida, 1987] The most important mode (96 % ofthe mass) has a mass median diameter (MMD) of 4.82 µm.

We do not explicitly take into account production of dustby saltation and sandblasting (e.g. Shao [2001]). The sizedistributed production of dust is a complex interplay betweenwind speed and soil properties. This interplay can cause dif-ferences in dust fluxes of several orders of magnitude [ Alfaroand Gomes, 2001; Grini et al., 2002b]. Grini and Zender[2004] conclude that input data needed to drive dust produc-tion with these mechanisms are not available on global scale.

The dust production calculated by the physical equationsare modified by two factors:

1. A global Tuning factor, T: This factor is determined“a posteriori” and ensures that global emissions in allsimulations are the same. T is globally constant.

2. An erodibility factor, RDBFCT: This factor is de-scribed in detail below. The factor is meant to take into

Simulations of dust sources and transport 3

account that some desert surfaces are easier to erodethan others [Prospero et al., 2002].

The total emissions in a given grid, is thus

EM = EMphys × RDBFCT × T (1)

where EMphys corresponds to the emissions modeled byphysical equations and EM to total emissions.

2.2. Soil moisture

Soil moisture inhibits dust production [Fecan et al., 1999].Soil moisture in desert areas is very low. The parameteriza-tion of evaporation used in the ECMWF model [Viterbo andBeljaars, 1995] does not capture variations in soil moistureat the scale described by Fecan et al. [1999]. Evaporationfrom soils in the ECMWF model stops at a globally constantpermanent wilting point of 0.171 m3 m−3.

Instead of tuning the Fecan et al. [1999] parameteriza-tions to fit the ECMWF soil moisture, we have chosen to de-velop a simplified approach based on rainfall. Our approachtakes into account two factors:

1. After a rainfall, the soil has to dry before it can startproducing dust.

2. The soil needs longer time to dry after a large rainfallthan after a small rainfall.

The time required by the soil to dry depends on air tempera-ture and humidity, surface winds and soil texture. To imple-ment these processes, we would need to build a whole newsoil moisture model which is beyond the scope of this study.To get around the problem, we made the following simpleassumptions:

1. the production of dust is stopped if precipitation dur-ing the last 24 hours is larger than 0.50 mm.

2. The length of the period without emissions (in days)is equal to the amount of rain during the last 24 hours(in mm)

3. If no rain has fallen the last 5 days, the soil is assumedto be dry no matter the size of the last rainfall

A similar approach has been used by Claquin [1999] andMyhre et al. [2003]. Claquin [1999] used a more sophisti-cated way to calculated the time needed for the soil to dry(taking into account soil temperature).

2.3. Soil erodibility

In this study, we use 4 erodbility factors to simulate dustproduction. Two of them are already published ([Ginouxet al., 2001]and Zender et al. [2003b]). Ginoux et al. [2001]applied the idea that wherever there are large basins in theworld, rivers and lakes would have accumulated loess andsand to give large erodibility. This reasoning led to the useof the simple formula.

TOPO = (z − zmin

zmax − zmin

)5 (2)

where TOPO is the erodibility factor with which all dustemissions are multiplied.

Zender et al. [2003b] calculated the erodibility assumingit was proportional to the upstream area from which sedi-ments may have accumulated locally during different cli-mate regimes. (The details of the algorithms used, can befound in Jenson and Domingue [1988]). Using a globaltransport model, they found that correlation with measure-ments improved when changing the Ginoux et al. [2001]erodibility factor with the more advanced one. The erodibil-ity factor proposed by Zender et al. [2003b] is called GEOin the rest of this work.

Two new erodibility factors are based on the assumptionthat erodibility is correlated with surface reflectance. Weused the data set MOD09 [Schaaf et al., 2002] from theMODIS satellite and produced two new erodibility factorswhich are used in runs labeled MDSLNR and MDSSQR.The factor MDSLNR is calculated according to Equation 3and MDSSQR is calculated according to Equation 4. Theerodibility described by the MDSLNR will be our base casein this work.

MDSLNR(i, j) =SR(i, j)

SRmax

(3)

MDSSQR(i, j) =SR(i, j)2

SR2max

(4)

where SR means surface reflectance.We compare the erodibility factors in Figure 1.Figure 1 shows that in the Sahara, all three datasets pro-

pose high erodibility in the West (Mali/Mauritania/Algeria)area, south-east (Lake Chad) and east (Egypt/Libya/Sudan/Chad).Even though the placements show similarities, both Ginouxet al. [2001] and Zender et al. [2003b] propose high erodibil-ity in Mauritania further west than MODIS which has high-est reflectance in a square between (12 oW 17oN) and (3oW22 oN). In eastern Sahara, Ginoux et al. [2001] does not pre-dict the same area as MODIS and Zender et al. [2003b]. Gi-noux et al. [2001] predicts high erodibility in a small area inNorth-Eastern Libya.

In East Asia, MDSLNR and MDSSQR give low erodibil-ity in both Taklamakan and Gobi deserts compared to Zen-der et al. [2003b] and Ginoux et al. [2001]. However, thegeographical placements of the maximum is approximatelyequal in all datasets.

In Arabia, all datasets agree on high erodibility in South-ern Saudi Arabia. MODIS gives high reflectivity in north-ern Saudi Arabia too, whereas both Ginoux et al. [2001] andZender et al. [2003a] gives a maximum in Iraq along the Eu-rfrat and Tigris rivers which is not indicated by MODIS.

In Australia, all datasets propose higher erodibility in theLake Eyre basin (South East) compared to the great Sandydesert (North west).

4 GRINI ET AL.

Figure 1. Erodibility factors obtained by four different methods. Upper left is the method of Ginoux et al. [2001], upperright is the method of Zender et al. [2003b], lower left is the factor described in Equation 3 and lower right is the factordescribed in Equation 4. Note that the images have different color scale.


2.4. Probability density function of wind speeds

We use a probability density function (PDF) of windsspeeds proposed by Justus et al. [1978] to drive dust emis-sions. The Weibull distribution is described by a shape fac-tor (k) and a scale factor (c). The factors are calculated fromwind speed at reference height (10 m) from Equation 5 and6.

kref = 0.94√

Uref (5)

where kref is the Weibull distribution shape factor, and Uref

is the wind speed at reference height

cref = Uref/Γ(1 +1

kref

) (6)

where cref is the Weibull distribution scale factor and Γ isthe gamma function. Further discussion on the Weibull dis-tribution or the gamma function can be found in any textbookon probability, e.g. Dougherty [1990].

To get the shape of the probability density function at themidpoint height, we use the formula proposed by Justus et al.[1978]:

kmdp = kref [1−0.088 ln(zrfr/10)]/[1−0.088 ln(zmdp/10)](7)

cmdp is found using (6) with values for midpoint height in-stead of reference height.

We use a 95 % percentile in the distribution, meaning thatthe lowest wind speed taken into account is the one where95 % of the winds are higher, and the highest wind speedtaken into account, is the one where 95 % of the winds arelower. This is given using the cumulative form of the Weibulldistribution 8:

p(U < Ux) = 1 − exp (−(Ux/c)k) (8)

where p is the probability of wind being lower than the windUx.

2.5. Atmospheric transport

We use the Oslo CTM2 for atmospheric transport. Themodel is a chemical transport model which is described inSundet [1997]. It has been used for several chemistry andtransport studies [Grini et al., 2002a; Myhre et al., 2003;Berglen et al., 2003; Endresen et al., 2003; Gauss et al.,2003]. It is an off line model driven by ECMWF fore-cast data. Advection is done with second order momentmethod [Prather, 1986], convection is done with the massflux scheme of Tiedtke [1989].

2.6. Wet deposition

Wet deposition is a very difficult process to model. Thehygroscopical properties of dust are not well known. We ex-pect the hygroscopical properties of dust to change duringtransport. The dust can for example get coated with watersoluble organics or sulfate which will make it easier to lose

during rain events. To calculate the right wet deposition, onewould need a detailed description of cloud and aerosol mi-crophysics (see e.g. Ghan et al. [1998]; Nenes et al. [2001]).

Because of the complexity of wet deposition, we chose toinclude a simple non size dependent wet deposition scheme.To test sensitivity to different washout efficiencies, we usea parameter ε to describe the efficiency with which dust re-moved by rain (Equation 9) and we test sensitivity to ε usingvalues equal to 1.0 and 0.3.

Our model includes two different types of wet deposition:2.6.1. Large scale Wet deposition is done using three-

dimensional rainfall data and assuming that dust washout isproportional to rainout, cloud liquid water and cloud frac-tion. Re-evaporation is taken into account only if all the rainevaporates.

loss = ε ∗ Cdust ∗ CLDFRC ∗

RAIN

CLDLWC(9)

where loss is the loss in kg, ε is a factor between 0 and 1,Cdust is dust concentration in kg, CLDFRC is the fractionof the grid cell covered by cloud, RAIN is rainfall in thetimestep (kg), and CLDLWC is the cloud liquid water (kg).

2.6.2. Convective This process removes dust when-ever the air rising in a convective tower gets super saturated.It is important to couple removal directly to the convectivetransport so that dust is not first transported to high altitudesby convection before any removal algorithm is applied. Af-ter transport, the dust would no longer be available in theclouds to be removed by rain. The convective rain removesdust with an efficiencyε equal to the one described for largescale wet deposition. The convective removal is described indetail by Berglen et al. [2003].

2.7. Dry deposition

The dry deposition uses a resistance method as describedin Zender et al. [2003a] where dry deposition velocity is de-pendent on wind friction speed and surface stability. Drydeposition is largest for the large aerosols and the areas withhigh wind friction speed [Seinfeld and Pandis, 1998].

3. Description of model runs

We did 6 model runs. The runs where normalized usingdifferent tuning factors (T) so that all gave the same annualemissions of 1500 Tg yr−1. The runs are described in Table1.

We use four different descriptions of soil erodibility: Thesoil erodibility (Equation 1) is varied in simulation 1-4. InTable 1, the different erodibility data used are labeled MD-SLNR (Equation 3), MDSSQR (Equation 4), GEO [Zenderet al., 2003a] and TOPO (Equation 2).

All simulations except simulation 5 use the Weibull dis-tribution to describe the winds used to drive dust production.Simulation 5 use mean wind speeds at model resolution. Thedifferences between simulation 5 and 1 show sensitivity to

6 GRINI ET AL.

ignoring wind speed variability in dust production. The sim-ulations using Weibull winds are labeled “Yes” in Table 1,and the simulation using mean wind is labeled with “No” inTable 1.

The efficiency of wet deposition is changed in simulation6. The parameter ε applied in Equation 9 was varied be-tween 0.3 (simulation 6) and 1.0 (simulation 1). Differencesin simulation 1 and 6 show sensitivity to washout efficiency.(Section 2.6).

4. Results and sensitivity tests

4.1. Global Budgets and fluxes

Table 2 shows the global fluxes in the different runs. Allruns are prescribed a total of 1500 Tg yr−1 emissions. Ap-proximately 50-60 % of the dust mass is lost by dry depo-sition in all simulations. 30-40 % is lost by large scale wetdeposition, and 10 % is lost by convective rainout. In thesimulation where rainout is less effective, the dry depositionfraction is highest.

Decreased rainout gives increased lifetime of dust. Theglobal and yearly averaged burden increases by 55 % whenonly 30% of the dust is assumed to fall out as rain. Figure 2shows where the different loss processes are important. Dustclose to the sources is mainly lost by dry deposition. Wet de-position is more important in remote ocean regions. This dif-ference reflects change in dust size distribution during trans-port. The largest aerosols (which contain most mass) fallout close the source areas. Another reason is that close tothe sources (in the deserts) there is not much rain which cangive wet deposition. The convective washout is (as expected)most efficient close to the equator, in the Inter Tropical Con-vergence Zone (ITCZ).

4.2. Nutrient budgets

As mentioned in Section 1, dust can participate in bio-logical processes. For example, dust is an important nutrientfor phytoplankton in the oceans. Modifying oceanic primaryproduction can influence the ocean carbon cycle.

Table 3 shows the yearly deposition to different ocean andforest regions. Biological processes in the ocean and forestsare limited by trace metals such as iron and phosphate. Allthe simulations yield approximately 400 Tg yr−1 depositionto the oceans. Zender et al. [2003a] calculated a total flux of315 Tg yr−1 and Prospero [1996] gives values between 358Tg yr−1 and 910 Tg yr−1.

The amazon forest receives 3-4 Tg yr−1. Simulation 6(with reduced rainout) yields 9 Tg yr−1 deposition to theamazon forest. The increased lifetime of dust in simulation6 allows more dust to be transported to the Amazon forestbefore being deposited. For the same reason, simulation 6gives higher deposition to the Northern Pacific. When re-ducing wet deposition rate, Asian dust is transported longerdistances before it is lost by rainout. Swap et al. [1992] pro-pose that 14 Tg yr−1 of dust are deposited in the Amazon

forest.

4.3. Global production

4.3.1. Effect of erodibility In Figure 3, we comparethe yearly averaged dust production for simulation 1-4. Thesesimulations all use different erodibility factors. We show to-tal production in kg m2 s−1 for simulation 1 and the absolutedifference from simulation 1 to simulation 2-4. The resultsshow that production is extremely dependent on erodibility.Figure 3 leads to the following main conclusions:

1. Simulation 3 and 4 yield higher emissions than sim-ulation 1 and 2 in East Asia (in both the Taklamakanand Gobi deserts).

2. In the Western Sahara, simulation 3 and 4 give emis-sions further west than simulation 1 and 2. Simulation3 and 4 also gives high emissions in an area close to(0E,25N) which is slightly further north than the mostactive areas in simulation 1 or 2.

3. Simulation 4 gives higher emissions in Somalia thanthe other simulations.

4. Simulation 3 and 4 give higher emissions in India thansimulation 1 and 2.

5. Simulation 3 and 4 give higher emissions around theCaspian sea than simulation 1 and 2.

6. Simulation 1 and 2 have high emissions in NorthernSaudi Arabia instead of in Iraq (simulation 3 and 4).

7. In Australia, simulation 3 and 4 give higher emissionsin the lake Eyre basin rather than in the Great Sandydesert which is the opposite of simulation 1 and 2.Simulation 3 and 4 also give higher total Australianproduction.

8. Simulation 3 and 4 give higher emissions in NorthAmerica than simulation 1 and 2.

4.3.2. Effect of wind speed variability Figure 4 showsthe effect of ignoring wind speed variability. The figure isdifficult to interpret since all emissions are fixed to 1500 Tg.

Ignoring wind speed variability reduces total emissions.We need a tuning factor twice as high for the emissions whenwind speed variability is ignored. Figure 4 actually showsthe areas where the increase in dust production due to in-creasing the tuning factor (Equation 1) is larger than the“natural” increase when including wind speed variability.

In our simulations, ignoring wind speed variability (andkeeping total emissions fixed) increases yearly Australianemissions. It increases emissions around the Caspian seaand in the Gobi desert and decreases emissions from Takla-makan in Asia. In Western Sahara, the increase in emissionswhen ignoring wind speed variability is co-located with pref-erential source areas in Mali, Mauritania and Algeria. Ineastern Sahara, the Egyptian and Libyan emissions are in-creased. Ignoring wind speed variability increases Somalianemissions.

Areas with high wind (e.g. coastal areas) produce moredust when ignoring wind speed variability. This is explain-


Table 1. Description of model runs. Dust production is changed using different assumptions on soil erodibility and windvariability. Wet deposition loss of dust is changed in simulation 6

Run Number Erodibility PDF Winds Washout1 MDSLNR Yes 100 %2 MDSSQR Yes 100 %3 GEO Yes 100 %4 TOPO Yes 100 %5 MDSLNR No 100 %6 MDSLNR Yes 30 %

Table 2. Global budget for loss fluxes and burden. The fluxes are expressed as percentage of mass. The annual totalproduction flux is 1500 Tg. The burden is given in Tg. DRYDEP is dry deposition flux (Section 2.7), LS WETDEP is large

scale wet deposition flux (Section 2.6.1), CNV WETDEP is convective wet deposition flux (Section 2.6.2)Run number DRYDEP LS WETDEP CNV WETDEP BURDEN

[%] [%] [%] [Tg]1 52.9 36.3 10.8 18.92 53.2 26.1 10.7 19.53 54.1 35.1 10.8 19.24 53.3 35.2 11.5 16.35 52.0 37.2 10.8 17.46 57.0 35.1 7.9 29.3

Figure 2. Importance of the different loss processes for simulation 1. The figures show annual mean production, drydeposition, large scale rainout and convective rainout. All fluxes are in kg m−2 s−1

8 GRINI ET AL.

Table 3. Yearly deposition to oceans and forests (Tg). The following abbreviations are used: GOC: Global ocean, NATL:Northern Atlantic ocean, SATL: Southern Atlantic ocean, NPAC: Northern Pacific, SPAC: Southern Pacific, IND: Indian

ocean, Persian Gulf and the Red sea, MED: Mediterranean and AMZ: Amazon forestRun number GOC NATL SATL NPAC SPAC IND MED AMZ

1 382 103 78 18 8 106 53 42 358 105 75 17 4 86 54 43 359 95 85 23 9 84 42 44 457 94 68 23 6 198 53 35 407 102 73 19 9 132 58 36 416 102 77 58 14 104 41 9

Figure 3. Global and yearly averaged dust production simulated by Oslo CTM2. Upper left shows production in simulation1. The other three figures are differences.


able: Using wind speed PDF facilitates emission from ar-eas with low winds. Keeping the total emissions constants,the emission from high wind areas must decrease when windPDF is taken into account.

4.4. Global loading

4.4.1. Yearly variations Figure 5 shows global masscolumn burdens. There is a maximum in the northern hemi-sphere summer over Sahara, and a maximum in the north-ern hemisphere spring in Asia. The Asian “spring dust” is awell known phenomenon. The plume out of Western Saharapeaks in July giving most transport across the Atlantic oceanin the northern hemisphere summer. Loadings over Australiaare low, but peak in the southern hemisphere summer.

In figure 6 we give zonal mean mass mixing ratios of dust.The level which the dust is lifted to, is higher in the north-ern hemisphere summer. During the northern hemispheresummer, the deserts at 20o north are heated making air risehigher. In July, significant amounts of dust are lifted abovethe 200 hPa level, whereas in January, the dust is only liftedto approximately 600 hPa. The higher lifting is important forthe interaction with terrestrial radiation. When dust is liftedhigh, it absorbs and re-emits long wave radiation at lowertemperatures, giving a stronger long wave radiative effectthan it would have at low levels.

4.4.2. Effect of erodibility In Figure 7 we show the ef-fect of changing the soil erodibility on the annual burden.The results are summarized below. Not all of these observa-tions can be evaluated because we do not have measurementdata for all the regions, but we try to evaluate some of theobservations in Section 4.6:

1. Both simulation 3 and 4 give higher dust loading overAsia (Taklamakan/Gobi and over the Caspian sea)than simulation 1 and 2.

2. Simulation 3 gives higher burdens in Southern Sa-hara due to very strong emissions close to Lake Chad.These emissions are transported south-west in ourmodel.

3. In simulation 3 and 4, both the Lake Chad area and theMali/Mauritania/Algeria have very high loadings.

4. Simulation 4 generally gives low burdens over Sahara.5. Simulation 3 gives low Saudi-Arabian burdens6. Simulation 4 gives high Arabian burdens, mainly orig-

inating from Somalia and South-western Saudi Ara-bia.

7. In Australia, simulation 4 gives more dust transportedEast of Australia instead of North-West as in simula-tion 1 and 2. This is due to higher emissions in theLake Eyre basin than in the Great Sandy Desert area.

4.4.3. Effect of wind speed variability Figure 8 showsthe effect of ignoring wind speed variability on the annualburden. This figure must be seen in connection with Figure4. When ignoring wind speed variability, the increased emis-sions in Somalia and Australia increase burden over all the

Indian ocean. Both these areas contribute to the increaseddeposition to the Indian ocean in simulation 5 (Table 3).

In the Saharan desert, the burden increases over coastalareas when ignoring wind speed variability. This is due towinds being high near the coast (section 4.3.2).

4.5. Effect of reducing washout

In figure 9 we show the annual mean absolute differencedifference in burden for the runs 1 and 6. (In simulation6 washout efficiency has been reduced). The results showthat only washing out 30 % of the dust increases dust burdensignificantly. As shown in Table 2, the annual mass load-ing increases by 55 % (from 18.9 Tg to 29.3 Tg) when only30 % of the dust is washed out. Dust washout is hard tomodel since the dust can change hygroscopic properties dur-ing transport (Section 2.6).

Figure 9 shows that the effect of reducing washout islargest in the Atlantic ocean in the ITCZ. It can also be seenthat the dust emitted from the Taklamakan Chinese desert isincreased when washout is less efficient. Table 3 shows thatdeposition to the Pacific goes up when rainout is less effec-tive. This higher deposition in the Pacific is due to increasedtransport from Asia to the Pacific due to the increased dustlifetime in simulation 6.

4.6. Observed Aerosol Index

Figure 10 shows the annual average Aerosol Index fromthe Total Ozone Mapping Spectrometer (TOMS) satellite[Torres et al., 1998] for the year 1980. Even though 1980is not the year simulated, the picture reproduces many of thefeatures from our different simulations.

TOMS data should be interpreted with care, because itdoes not only observe dust. TOMS observe all absorbingaerosols, and it detects black carbon and biomass aerosols inaddition to dust aerosols.

Several observations can be made when comparing Fig-ure 10 with Figure 7 and Figure 8:

1. TOMS indicates very high aerosol index in the LakeChad area consistent with simulation 3.

2. TOMS indicates very high aerosol index in the Chi-nese Taklamakan desert, consistent with simulation 3and 4. The Gobi desert seems like a much weaker dustsource (inconsistent with simulation 3 and 4).

3. TOMS indicates low Somalian emission, consistentwith simulation 1,2 and 3.

4. TOMS indicates that Australian emission should mainlycome from the Lake Eyre area, and not from the GreatSandy desert. This is consistent with simulation 3 and4.

5. The high Caspian sea dust loading is found in TOMSand in all simulations.

10 GRINI ET AL.

Figure 4. Global and yearly averaged dust production simulated by Oslo CTM2. Left shows production in simulation 1.Right shows difference in production between simulation 1 and 5. The difference is due to neglecting wind speed variability.

Figure 5. Column burden mass in January (Upper left), April (Upper right), July (Lower left) and October (lower right)simulated by Oslo CTM2


Figure 6. Zonal mean mass mixing ratio in January (Upper left), April (Upper right), Julys (Lower left) and October (lowerright) simulated by Oslo CTM2

Figure 7. Effect of changing soil erodibility on column burden. Upper left is reference simulation (1). The other three areabsolute difference in simulation 2-4

12 GRINI ET AL.

Figure 8. Effect of neglecting wind speed variability on column burden. Left is reference simulation. Right is difference insimulation 5

Figure 9. Absolute difference in dust burden (kg/m2) for 1996. Left is Yearly averaged burden in simulation one, and rightis increase in burden when washout is decreased

Figure 10. Yearly average TOMS aerosol index for the year 1980


4.7. Station comparisons

4.7.1. Monthly Mass concentrations Model output fromCTM2 has been compared to mean mass concentrations atseveral stations. For some stations, we have concentrationsfrom 1996 (Figure 11), and for the others, we only have cli-matological means (Figure 12).

Figure 11 shows that all the stations in the Atlantic plumeare simulated reasonably. For the remote stations (Barbados,Bermuda and Miami), all the simulation with 100% rainoutshow similar results. When rainout is decreased, simulatedconcentrations in these stations are too high compared tomeasurements.

For the stations close to the source (Izana, Sal Island),the parameterization of rainout does not change the resultsmuch. This is logical since there is not much rainfall closeto the desert sources. Dry deposition is the most importantloss mechanism close to the deserts. Therefore, the differentwet loss mechanisms are not so important when evaluatingstations near to sources.

The only Asian station we have is Cheju. All the simu-lations show too low mixing ratios at this station. Simula-tions 3 and 4 is closest to reproducing the observed concen-trations. Being situated south of Korea it should be dom-inated by dust from Gobi (and Taklamakan). This resultssuggests that our Asian sources are too weak in all simula-tions. The simulation with decreased washout (6) gives sig-nificantly higher concentrations than the simulation (1) withfull washout, but not high enough according to the measure-ments.

For the stations where we only have climatological means,all simulations (except simulation 6) represent well the con-centration in the remote Mace Head stations. For the remoteOahu station, the measurements are in between the modelresults with full washout (1-5) and the simulation with de-creased washout (6). We suspect that the washout in simu-lation 6 is too low. Too low washout compensates for toolow Asian emissions to give reasonable values for Oahu insimulation 6.

The Kaasidogo station is situated in the Ocean to thesouth of India. All simulations give reasonable concentra-tion levels at this station. The fact that simulation 6 does notdiverge much from simulation 1-5 indicates that this stationis not far from a source area (the deserts in North westernIndia).

4.7.2. Optical depths in the Western Saharan plumeTo check if the choice of soil erodibility had a large effect onthe simulation of specific dust episodes, the optical depthsobserved at Cape Verde (in the western Saharan plume) havebeen compared to model derived optical depth on a dailybasis. The choice of erodibility factors have a large effect onthe output. Cape Verde was chosen because it is situated inthe middle of the dust plume leaving the Saharan desert.

Table 4 gives correlation coefficients for the differentmodel runs and the daily aerosol optical depth at Cape Verde.All the MODIS based erodibility datasets yield high correla-

Table 4. Correlation coefficients for aerosol optical depth atCap Verde

Run Number Correlation coefficient1 0.742 0.733 0.604 0.705 0.736 0.79

tion coefficients (larger than 0.70).Figure 13 (left) shows the aerosol optical depth at Cape

Verde for model run 1 and 6 which have the highest corre-lation coefficients at this location. It can be seen that boththe magnitude and the timing of the events is good. It is sur-prising that simulation 6 gives the best correlation with mea-surements. Globally, this simulation gives a rather high dustloading compared to loadings generally proposed by otherauthors. It is possible that for a station so close to the sourcearea, the low wet deposition efficiency is realistic becausethe dust has not had time to get coated with materials mak-ing it more hydrophilic (Section 2.6).

Figure 13 (right) shows the same plot with simulation 3and 4. It can be seen that whenever there are peaks in sim-ulation 3, the peaks are too large. In particular simulation 3gives peaks which are much higher than the measured ones.Simulation 4 generally gives too low optical depth in periodswhen the optical depth is not peaking. Note that the scale inthe right panel of Figure 13 is different from the scale in theleft panel due to the large values of the peaks in simulation3.

Simulation 3 shows significantly lower correlation thanthe other simulations at Cape Verde. This indicates thatthe method applied by Zender et al. [2003a] does not givethe most appropriate erodibility dataset for western Sahara.The amplitude of the peaks in simulation 3, indicate that theerodibility datasets proposed by Zender et al. [2003b] is tooheterogeneous, and that the differences in soil erodibility be-tween the desert areas in the world are not as large as pro-posed by the GEO dataset.

In the different data sets, the erodibility factor can varyby an order of magnitude for a given area. We have com-pared several Saharan dust outbreaks to TOMS aerosol index(not shown). The results show that using different erodibil-ity factors can replace the outbreaks. (That is: The outbreakswill come from different areas depending on the erodibilitydataset applied). This means that even though on average,the plume will originate from approximately the right ares(e.g. Western Sahara), each individual outbreak can origi-nate from an area only close to the correct source area. Cam-paign studying individual Saharan dust outbreaks can hope-fully help determine the exact areas with the most erodiblesoils.

The good correlations obtained by the MDSLNR and

14 GRINI ET AL.

0 2 4 6 8 10 120

0.2

0.4

0.6

0.8

1

1.2

1.4x 10−7 Concentration Brb lon= 301 lat= 13

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5

4

4.5x 10−8 Concentration Brm lon= 295 lat= 32

monthdu

st m

ixin

g ra

tio (k

g/kg

)

0 2 4 6 8 10 120

1

2

3

4

5

6

7x 10−8 Concentration Mmi lon= 280 lat= 25

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

0.2

0.4

0.6

0.8

1

1.2

1.4x 10−7 Concentration Izn lon= 344 lat= 28

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

1

2

3

4

5

6

7x 10−7 Concentration SlI lon= 335 lat= 15

month

dust

mix

ing

ratio

(kg/

kg)

Figure 11. Mass concentrations form U. Miami network (for 1996) compared to CTM2 concentrations (for 1996). The fig-ures represent Barbados (301E,13N), Bermuda (295E,32N), Miami (280N,25E), Izana (17W,28N) and Sal Island (25W,15N).Black solid line is measurements. Red dashed is simulation 1, blue dashed is simulation 2, yellow dashed is simulation 3,cyan dashed i simulation 4, green stars is simulation 5, and blue stars is simulation 6


0 2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5

4x 10−8 Climatolgy Jej lon= 127 lat= 34

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

1

2

3

4

5

6

7x 10−8 Climatolgy Kaa lon= 74 lat= 5

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

1

2

3

4

5

6

7

8x 10−9 Climatolgy McH lon= 10 lat= 53

month

dust

mix

ing

ratio

(kg/

kg)

0 2 4 6 8 10 120

0.5

1

1.5

2

2.5

3

3.5x 10−9 Climatolgy Oah lon= 202 lat= 21

month

dust

mix

ing

ratio

(kg/

kg)

Figure 12. Mass concentrations form U. Miami network (climatological means) compared to CTM2 concentrations (for1996). The figures represent Cheju (126E, 33N), Kaasidogo(73E,5N), Mace Head (10E,53N) and Oahu (202E,21N). Blacksolid line is measurements. Red dashed is simulation 1, blue dashed is simulation 2, yellow dashed is simulation 3, cyandashed i simulation 4, green stars is simulation 5, and blue stars is simulation 6

Figure 13. Aerosol optical depth at Cape Verde. Left figure is simulation 1 and 6 compared to measurements. Right figureis simulation 3 and 4 compared to measurements. Note the difference in scale between left and right panel.

16 GRINI ET AL.

MDSSQR are encouraging given the simplicity of thesedatasets. We believe there is connection between soil erodi-bility and soil reflectivity. Both MDSSQR and MDSLNRare only simple concepts which can probably be refined togive better datasets for soil erodibility.

5. Summary

Several different parameterizations of dust productionand transport have been tested and evaluated. We have fo-cused on the parameters “soil erodibility” and wind speedvariability. Both these influence the magnitude and geo-graphical location of the dust emissions.

We show that the geographical location and magnitude ofatmospheric dust can vary significantly when changing soilerodibility datasets. The emissions are highest in the areaswhere the erodibility is highest.

When ignoring wind speed variability, the geographicalpattern of dust emissions is changed. Ignoring wind speedvariability (and fixing total dust emissions) gives high pro-duction in areas which have high wind (e.g. areas close tothe coast)

The datasets used here to describe soil erodibility havedifferent weaknesses and strengths. We show that using thesoil erodibility dataset of Zender et al. [2003b] reproducesseveral typical aspects of the global dust cycle (high Chinesedust, high loading in Lake Chad area, higher emissions inTaklamakan than in Gobi). However it fails reproduce otheraspects such as Lake Eyre (and not the great sandy desert)being the dominant Australian source. The dataset proposedby Ginoux et al. [2001] gives high Somalian emissions. Thisis not reproduced by any other model run.

The reflectivity based erodibility factors give low emis-sions in the Chinese deserts, however they capture the mainSaharan regions such as north eastern Sahara (Egypt/Libya),western Sahara (Mali/Mauriania/Algeria) and Lake Chad.

Some aspects of the annual mineral dust cycle, such asmaximum dust transport from Sahara to the Caribbean in thenorthern hemisphere summer, maximum dust production inAustralia in the southern hemisphere summer and maximumdust production in the Chinese deserts in spring is repro-duced by all model simulations. Monthly average concentra-tions at remote stations in the Saharan plume are similar inall simulations except when wet loss of dust is significantlyreduced.

When comparing aerosol optical depths at Cape Verdeday by day we show that assuming that soil reflectivity canrepresent soil erodibility is a good assumption which repre-sents the measured optical depth with a correlation coeffi-cient higher than 0.70. The other erodibility give lower cor-relation coefficients, and the optical depths are too high inmany of the dust episodes.

The good correlations obtained by the MDSLNR andMDSSQR are encouraging given the simplicity of thesedatasets. We believe there is connection between soil erodi-

bility and soil reflectivity. Both MDSSQR and MDSLNRare only simple concepts which can probably be refined togive better datasets for soil erodibility.

Acknowledgments. AG, GM, JKS and ISAI acknowledgesupport from the Norwegian research council grant 139810/720(CHEMCLIM). CSZ acknowledges support from NASA grantsNAG5-10147 (IDS) and NAG5-10546 (NIP)


References

Alfaro, S. C., and L. Gomes, Modeling mineral aerosol productionby wind erosion: Emission intensities and aerosol size distribu-tions in source areas, J. Geophys. Res., , 106, 18,075–18,084,2001.

Berglen, T., T. Berntsen, I. Isaksen, and J. Sundet, A global modelof the coupled sulfur/oxidant chemistry in the troposphere: Thesulfur cycle, J. Geophys. Res., , submitted, 2003.

Bian, H., and C. Zender, Mineral dust and global troposphericchemistry: Relative roles of photolysis and heterogeneous up-take, J. Geophys. Res., , In press, 2003.

Claquin, T., Modelisation de la mineralogie et du forcage radiatifdes poussieres desertiques, Ph.D. thesis, University of Hamburg,1999.

D’Almeida, G. A., On the variability of desert aerosol radiativecharacteristics, J. Geophys. Res., , 92, 3017–3026, 1987.

Dentener, F. J., G. R. Carmichael, Y. Zhang, J. Lelieveld, andP. J. Crutzen, Role of mineral aerosol as a reactive surface inthe global troposphere, J. Geophys. Res., , 101, 22,869–22,889,1996.

Dougherty, E., Probability and statistics for scientists and engi-neering, computing and physical sciences, Prentice Hall, Inc.,1990.

Endresen, Ø., E. Søgard, J. Sundet, S. Dalsøren, I. Isaksen,T. Berglen, and G. Gravir, Emission from international sea trans-portation and environmental impact, J. Geophys. Res., , 108,2003, doi:10.1029/2002JD00289.

Falkowski, P., R. Barber, and V. Smetacek, Biogeochemical con-trols and feedbacks on ocean primary production, Science, 201,200–206, 1998.

Fecan, F., B. Marticorena, and G. Bergametti, Parameterization dueto the increase of the aeolian erosion threshold wind friction ve-locity due to soil moisture for arid and semi-arid areas, AnnalesGeophysicae, 17, 149–157, 1999.

Gauss, M., I. Isaksen, S. Wong, and W. Wang, Impactof H2O emissions from cryoplanes and kerosene air-craft on the atmosphere, J. Geophys. Res., , 108, 2003,doi:101029/2002JD00262.

Ghan, S., G. Guzman, and H. Abdul-Razzak, Competition betweensea salt and sulfate particles as cloud condensation nuclei, J.Atm. Sci., 55, 3340–3347, 1998.

Ginoux, P., M. Chin, I. Tegen, J. Prospero, B. Holben, O. Dubovik,and S.-J. Lin, Sources and distributions of dust aerosols simu-lated with the GOCART model, J. Geophys. Res., , 106, 20,255,2001.

Grini, A., and C. Zender, Roles of saltation, sandblasting, and windspeed variability on mineral dust aerosol size distribution duringthe puerto rican dust experiment (pride), J. Geophys. Res., , inpress, 2004.

Grini, A., G. Myhre, J. Sundet, and I. Isaksen, Modeling the annualcycle of sea salt in the global 3D model Oslo CTM2; concen-trations, fluxes and radiative impact, J. Climate, 15, 1717–1730,2002a.

Grini, A., C. Zender, and P. Colarco, Saltation sandblasting behav-ior during mineral dust aerosol production, Geophys. Res. Lett.,29, 1868–1871, 2002b.

Haywood, J., V. Ramaswamy, and B. Soden, Tropospheric aerosolclimate forcing in clear sky satellite observation over the oceans,Science, 283, 1299–1303, 1999.

Iversen, J., and B. White, Saltation threshold on Earth, Mars andVenus, Sedimentology, 29, 111–119, 1982.

Jenson, S. K., and J. O. Domingue, Extracting topographic struc-ture from digital elevation data for geographic information sys-tem analysis, Photogrammetric Engineering and Remote Sens-ing, 54, 1593–1600, 1988.

Justus, C., W. Hargraves, A. Mikhail, and D. Graber, Methods forestimating wind speed frequency distributions, J. App. Met., 17,350–353, 1978.

Kaufman, Y., D. Tanre, O. Dubovik, A. Karnieli, and L. Remer,Absorption of sunlight by dust as inferred from satellite andground-based remote sensing, Geophys. Res. Lett., , 28, 1479–1482, 2001.

Lohmann, U., Possible aerosol effects on ice clouds via contactnucleation, J. Atm. Sci., 59, 647–656, 2002.

Marticorena, B., and G. Bergametti, Modeling of the atmosphericdust cycle: 1. design of a soil derived dust emission scheme, J.Geophys. Res., , 100, 16,415–16,429, 1995.

Myhre, G., and F. Stordal, Global sensitivity experiments of theradiative forcing due to mineral aerosols, J. Geophys. Res., , 106,18,193–18,204, 2001.

Myhre, G., A. Grini, J. Haywood, F. Stordal, B. Chatenet, D. Tanre,J. Sundet, and I. Isaksen, Modeling the radiative impact of min-eral dust during the saharan dust experiment (SHADE) cam-paign, J. Geophys. Res, 108, 8579, 2003.

Nenes, A., S. Ghan, H. Abdul-Razzak, P. Chuang, and J. Seinfeld,Kinetic limitations on cloud droplet formation and impact oncloud albedo, Tellus, 53B, 133–149, 2001.

Prather, M. J., Numerical advection by conservation of second-order moments, J. Geophys. Res., , 91, 6671–6681, 1986.

Prospero, J., The atmospheric transport of particles to the ocean,in Particle flux in the ocean, edited by V. Ittekot, P. Shefer,S. Honjo, and P. Depetris, John Wiley & Sons Ltd., 1996.

Prospero, J., African droughts and dust transport to the Caribbean:Climate change implications, Science, pp. 1024–1027, 2003.

Prospero, J., P. Ginoux, O. Torres, S. Nicholson, and T. Gill, Envi-ronmental characterization of global sources of atmospheric soildust identified with the NIMBUS 7 total ozone mapping spec-trometer (TOMS) absorbing aerosol product, Reviews of Geo-physics, 40, 2002, doi:10.1029/2000RG000095.

Pruppacher, H., and J. Klett, Microphysics of clouds and precipita-tion, Kluwer academic publishers, 1997.

Schaaf, C., et al., First operational BRDF, albedo nadir reflectanceproducts from MODIS, Remote sensing of Environment, 82,135–148, 2002.

Seinfeld, J. H., and S. N. Pandis, Atmospheric chemistry andphysics, From air pollution to Climate change, John Wiley andSons, 1998.

Shao, Y., A model for mineral dust erosion, J. Geophys. Res., , 106,20,239–20,254, 2001.

Sherwood, S., A microphysical connection among biomass burn-ing, cumulus clouds and stratospheric moisture, Science, 295,1272–1275, 2002.

Sokolik, I., and O. Toon, Direct radiative forcing by anthropogenicairborne mineral aerosols, Nature, 381, 681–683, 1996.

Sundet, J. K., Model studies with a 3-D global CTM using ECMWFdata, Ph.D. thesis, University of Oslo, 1997.

Swap, R., M. Garstang, and S. Greco, Saharan dust in the Amazonbasin, Tellus, 44B, 133–149, 1992.

Tanre, D., et al., Measurement and modeling of the saharandust radiative impact: Overview of the Saharan Dust Ex-periment (SHADE), J. Geophys. Res., , 108, 8574, 2003,doi:10.1029/2002JD003273.

Tegen, I., and I. Fung, Modeling of mineral dust in the atmosphere:Sources, transport and optical thickness, J. Geophys. Res., , 99,22,897–22,914, 1994.

18 GRINI ET AL.

Tegen, I., and A. A. Lacis, Modeling of particle size distribu-tion and its influence on the radiative properties of mineral dustaerosol, J. Geophys. Res., , 101, 19,237–19,244, 1996.

Tegen, I., A. A. Lacis, and I. Fung, The influence of mineralaerosols from disturbed soils on climate forcing, Nature, 380,419–422, 1996.

Tegen, I., S. Harrison, K. Kohfeld, I. Prentice, M. Coe, andM. Heimann, Impact of vegetation and preferential source areason global dust aerosol: Results from a model study, J. Geophys.Res., , 107, 4576, 2002, doi:10.1029/2001JD00096.

Tiedtke, M., A comprehensive mass flux scheme for cumulus pa-rameterization on large scale models, Mon. Wea. Rev., 117,1779–1800, 1989.

Torres, O., P. Bhartia, J. Herman, Z. Ahmad, and J. Gleason,Derivation of aerosol properties from satellite measurements ofbackscattered ultraviolet radiation: Theoretical basis, J. Geo-phys. Res., , pp. 17,099–17,110, 1998.

Tsvetsinskaya, E., C. Schaaf, F. Gao, A. Strahler, R. Dickinson,X. Zeng, and W. Lucht, Relating MODIS-derived surface albedoto soils and rock types over Northern Africa and the Arabianpeninsula, Geophys. Res. Lett., , 29, 67–71, 2002.

Viterbo, P., and A. C. Beljaars, An improved land surface param-eterization scheme in the ECMWF model and its validation, J.Climate, 8, 2716–2748, 1995.

Woodward, S., Modeling the atmospheric life-cycle and radiativeimpact of mineral dust in the Hadley center climate model, J.Geophys. Res., , 106, 2001.

Zender, C., H. Bian, and D. Newman, The mineral dust entrain-ment and deposition (dead) model: Description and global dustdistribution, J. Geophys. Res., , 108, 4416, 2003a.

Zender, C., D. Newman, and O. Torres, Spatial heterogeneity inaeolian erodibility: Uniform, topographic, geomorphologic, andhydrologic hypotheses, J. Geophys. Res., , 108, 4543, 2003b,doi:10.1029/2002JD00303.Alf Grini, Gunnar Myhre, Jostein K. Sundet and Ivar

S.A. Isaksen Department of Geosciences, University ofOslo, P.O. Box 1022 Blindern, 0315 Oslo, Norway. (e-mail:[email protected])

Charles S. Zender, Department of Earth System Science,University of California, Irvine, CA, 92697-3100, USA (e-mail: [email protected])

This preprint was prepared with AGU’s LATEX macros v4, with the ex-tension package ‘AGU++’ by P. W. Daly, version 1.6b from 1999/08/19.

Documents

Natural Aerosols in the Global Atmospheredust.ess.uci.edu/ppr/phd_gri04.pdf · Natural aerosols in the global atmosphere ... I have been happy to go to work and I have ... 2.5 Evaluation