JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 106, NO. D17, PAGES 20,255-20,273, SEPTEMBER 16, 2001
Sources
and
simulated
distributions
with
the
of dust
GOCART
aerosols
model
Paul Ginoux,
•'•'Mian Chin,•,•'Ina Tegen3 JosephM Prospero
4
BrentHolben•' OlegDubovik•,•'and Shian-JiannLine
Abstract.
The global distribution of dust aerosolis simulatedwith the Georgia
Tech/GoddardGlobalOzoneChemistryAerosolRadiationand Transport(GOCART) model. In this modelall topographiclowswith bare groundsurfaceare
assumedto have accumulated sedimentswhich are potential dust sources. The
uplifting of dust particles is expressedas a function of surfacewind speed and
wetness. The GOCART model is driven by the assimilatedmeteorologicalfields
from the Goddard Earth ObservingSystemData AssimilationSystem (GEOS
DAS) which facilitatesdirect comparisonwith observations.The model includes
sevensize classesof mineral dust ranging from 0.1-6 /•m radius. The total annual
emission
is estimatedto be between1604and 1960Tg yr-1 in a 5-yearsimulation.
The model has been evaluatedby comparingsimulation resultswith ground-based
measurementsand satellite data. The evaluationhas been performedby comparing
surfaceconcentrations,vertical distributions, depositionrates, optical thickness,
and size distributions. The comparisonsshow that the model results generally
agreewith the observationswithout the necessityof invoking any contribution from
anthropogenicdisturbancesto soils. However,the model overpredictsthe transport
of dust from the Asian sourcesto the North Pacific. This discrepancyis attributed
to an overestimate of small particle emission from the Asian sources.
1.
Introduction
where it potentially plays an important role in biogeochemical processes.
There is an increasing interest in the atmospheric
Many studies, which have characterized dust genertransport of mineral dust. Recent research suggests
ation on a micrometeorologicalscale, have shown that
that mineral dust may play an important role in clidust mobilization is sensitiveto a wide range of factors
mate forcing by altering the radiation balance in the
including the compositionof the soils,the soil moisture
atmosphere[Tegenet al., 1997]and by affectingcloud
content, the surface conditions, and the wind velocity.
nucleationand optical properties[Levin et al., 1996].
Previous global models have identified the dust sources
In addition, dust can serve as a catalyst for reactive
basedon the soil moisturecontent[Joussaume,
1990],
the
location
of
deserts
from
vegetation
data
set
[(Yenand can significantly modify photochemical processes
thon,1992a,1992b;Mahowaldet al., 1999],the location
[Dickersonet al., 1997]. Largeamountsof mineraldust
of sparselyvegetatedarea, and the soiltexture from vegaredeposited
to the oceans[Duce,1995;Prospero,1996]
gas speciesin the atmosphere[Dentenet et al., 1996]
etationandsoildatasets[TegenandFung,1994],or the
distributionof duststormfrequencies
overarid regions
[Dentenetet al., 1996]. Discrepancies
betweensimu-
1Schoolof Earth and AtmosphericSciences,Georgia Institute of Technology,Atlanta, Georgia.
2NASA Goddard SpaceFlight Center, Greenbelt, Maryland.
lated and observeddust loadinghaveled somemodelers
to invoke land surface modification as a source that con-
tributesas muchas 50% of total dustemission
[Tegen
and Fung,1995].
In this work, we determine the global distribution
many.
4RosenstielSchool of Marine and Atmospheric Science, of dust sourcesby usingthe surfacetopographicfea3Max-Planck-Institute for Biogeochemistry.Jena, Ger-
University of Miami, Miami, Florida.
5ScienceSystemsand Applications, Inc., Lanham, Mary-
tures. We assumethat the most probable sourcesare
related to the degree of depression. The dust sources
land.
estimatedhereare consistent
with the recentstudyof J.
M. Prosperoet al. (Environmentalcharacterizationof
Copyright 2001 by the American GeophysicalUnion.
Paper number 2000JD000053.
0148-0227
/ 01/ 2000JD000053509.00
global sourcesof atmosphericsoil dust derivedfrom the
NIMBUS-7 TOMS absorbingaerosolproduct, submitted to Reviewsof Geophysics,
2000, hereinafterreferred
20,255
20,256
GINOUX
ET AL.: DUST AEROSOL
MODELING
to as Prosperoet al., submittedmanuscript,2000). In of the model results with TOMS aerosol index will be
that study,usingthe long-term(1979-present)
recordof discussedin a separate paper.
the global aerosolindex retrievedfrom the Total Ozone
MappingSpectrometer(TOMS) instrument[Hermanet
al., 1997; Torreset al., 1998],Prosperoet al.(submitted 2. Model Description
manuscript,2000) haveidentifiedthe locationof all the
major dust sources. They have shown that the sources
can usually be associatedwith topographic lows which
have a deep accumulation of alluvial sedimentsformed
during the late Pleistoceneor Holocene. These sediments are composedof fine particles which are easily
eroded by winds.
To simulate the atmospheric dust distribution, emission, transport, and deposition of seven particle size
classesbetween 0.1 and 6 pm radius have been incor-
porated in the Georgia Tech/Goddard Global Ozone
Chemistry Aerosol Radiation and Trknsport (GOCART) model. This model is driven by assimilated
meteorologicalfields generated in the Goddard Earth
ObservingSystem Data AssimilationSystem (GEOS
DAS) and hasbeenappliedto simulateatmosphericsulfate aerosolsand its precursors
[Chin et al., 2000]. The
advantageof the GOCART model is that, by usingobserved meteorologicalfields, it can effectively simulate
atmospheric constituents for any specifictime period.
In this paper, we present a detailed descriptionof the
componentsincorporated in GOCART model to simulate dust distributions. The global distribution and
budget of mineral dust is discussed. Results from our
model simulationsare comparedwith ground-basedand
satellite measurements. The comparisonsinclude the
surface concentrations,vertical distribution, deposition
fluxes, optical thickness, and size distribution. We also
compare our results with those of other model studies.
This paper is focusing on the model evaluation on the
global scale and its ability to reproduce the observed
2.1.
Model Configuration
The GOCART model has been developedby Chin et
al. [2000]to simulatethe distributionof sulfurspecies
in
the atmosphere. The model solvesthe continuity equation which includes the emission, chemistry, advection,
convection, diffusion, dry deposition, and wet deposition of each species.
The meteorologicalfieldsusedto drive the GOCART
modelare assimilateddata by GEOS-DAS[Schubertet
al., 1993].The GOCART modelhasthe samehorizontal resolutionas GEOS DAS: 2ø latitude by 2.5ø longitude.
The vertical
resolution
varies with the different
GEOS DAS versions. For GEOS DAS version 1, available from February 1980 to November 1995, there are
20 vertical sigma levelsfrom the Earth's surfaceto 10
mbar. For GEOS DAS version 1.3, availablefrom April
1995 to November 1997, the data fields have been regriddedfrom the original 46 levelsinto 26 vertical levels
from the Earth's surface to 0.1 mbar, with 23 levels in
the troposphere. The GEOS DAS prognosticvariables
are instantaneousvalues savedevery 6 hours, while the
diagnosticfields are averagedvaluesover 3 or 6 hours,
as indicated
in Table
1. Values of instantaneous
fields
are linearly interpolated to every model time step (20
min). The modelis initializedwith near-zeromass,and
is spin-up for i month prior to the simulations,and the
results are saved every 6 hours.
2.2.
Particle
Sizes
Mineral particles in the atmospherehave radii rangseasonalcycle. The detailedregionaland daily analysis ing from about 0.1 to 50 pm [Duce,1995]. The uplift-
Table 1. GEOS DAS MeteorologicalFields UsedAs Model
Input
Variable
Time
Resolution
Prognostic
Surface pressure
Temperature
Wind velocity
Specific humidity
6
6
6
6
Surface
6 hours instantaneous
wetness
hours, instantaneous
hours, instantaneous
hours, instantaneous
hours, instantaneous
Diagnostic
Cloud
mass flux
6 hours, averaged
Specifichumidity change
due to moist processes
Wind speed at 10 m
Precipitation
Eddy diffusion
6
3
3
3
hours, averaged
hours, averaged
hours, averaged
hours, averaged
GINOUX
ET AL.:
DUST
AEROSOL
MODELING
ing of particles smaller than 0.1 /•m by air is limited
by the adhesiveand cohesiveforceswhich tend to form
20,257
S- ( Zmax
- zi 5
(1)
Zmax -- Zmin
largerparticlesor aggregates
[$chefferand $chatschabel, 1992]. Becauseof gravitationalsettling,particles where $ is the probability to have accumulatedsedilarger than 6/•m generally have short atmosphericlife-
mentsin the grid cell i of altitude zi, and Zmaxand Zmin
times(i.e.,lessthana fewhours)[TegenandFung,1994]
are the maximum
and limits their significanceon the global scale. The
present study is therefore restricted to particles with
a diameter larger than 0.1/•m and smallerthan 6 /•m
rounding 10ø x 10ø topography,respectively.It is not
intended
and minimum
elevations
in the sur-
here to calculate the exact amount of alluvium
but rather to define the most probable locations of sedradius. FollowingTegenand Lacis[1996],the sizedis- iment. To increasethe topographic contrast of these
tribution is modeled into sevensize bins, and the mass locations and to obtain the best fit with the sources
distribution varieslinearly with respectto radius in each identifiedby Prosperoet al. (submitted manuscript,
bins. The sevensize rangesare 0.1-0.18 /•m, 0.18-0.3
/•m, 0.3-0.6/•m, 0.6-1/•m, 1-1.8/•m, 1.8-3/•m, and 3-6
/•m, with correspondingeffectiveradii of 0.15, 0.25, 0.4,
0.8, 1.5, 2.5, and 4/•m, respectively.
The U.S. Departmentof Agriculture(USDA) defines
particles with a radius between 1/•m and 25/•m as silt,
andbelow1/•m asclay [Hillel, 1982].Mineralogicalsilt
particles are mainly composedof quartz, but they are
often coated with strongly adherent clay suchthat their
2000), the relativealtitude has been taken at the fifth
power. Only land surface with bare soil is considered
as possibledust sources.We identify the bare soil surface from the 1ø x 1ø vegetation data set derived from
the advancedveryhighresolutionradiometer(AVHRR)
data [DeFriesand Townshen&1994].
Figure I showsthe global distributionsof the source
function S, regriddedon the 2.5ø x 2ø GOCART grid,
the calculateddust emission(seenext section),and the
physicochemical
propertiesare similar to clay [Hillel, TOMS aerosol index. Most of the maxima of the S func1982]. The massdensityof clay (classes1-4) and silt tion are collocated with the "hot spots" of the TOMS
(classes
5-7) are 2.5 and 2.65g m-3, respectively.
aerosol index which have been identified
as dust sources
by Prosperoet al. (submittedmanuscript,2000). The
The size distributionof small particles (radius 0.1-1
/•m) is assumednot to changeduring transport, since clearest examples are the Tunisian, Libyan, Mauritathe differencesin the gravitational settling velocities nian, and Malian sources in the Sahara, the Bodele
within this size classare small. For numerical efficiency depressionin the Sahel, the Indian source along the
the first four classesare transportedas one group (0.1- Indus valley, the Taklimakan located north of the Hi1 /•m) of clay particleswith an effectiveradius of 0.75 malaya, the Lake Eyre basin in Australia, the Salton
/•m. When computing optical properties, which have Sea in southern California, the Altiplano and Patagoa strong dependency on the size distribution, the clay nia in the Andes, and the Namibian sourcein southwest
size classis redistributed into the original four classes Africa.
(0.1-0.18, 0.18-0.3, 0.3-0.6, and 0.6-1 /•m) by assuming
2.4.
Dust
Emission
a massfraction for each class:0.9, $.1,23.4, and 67.6%,
respectively[Tegenand Lacis,1996]•
Dust uplifting into the atmosphere is mainly initiated by saltation bombardment(sand blasting). Mar2.3.
Dust
sources
ticorenaand Bergametti[1995]havedevelopeda model
The origin of clay minerals at the Earth's surface is, of dust emissionbased on a complex parameterization
in the majority of cases,a processof weatheringof rocks of this process. Their emission scheme has been into form alluvium (streamdepositedsediments)[Velde, corporatedin dust transport model for regional studies
1992]. Becausethere are no data on the global distri-
[Schulzet al., 1998; Guelle et al., 2000]. The major
bution of alluvium over the land surfaces,the potential
limitation of this algorithm is that it needsdetailed in-
location
formations of soil characteristics which are not readily
available on the global scale.
of accumulated
sediments
has been determined
by comparing the elevation of any 1ø x 1ø grid point
with its surrounding hydrologicalbasin. The complex
contoursof each basin have been simplified by assuming a constant area of 10ø x 10ø. This choiceis based
on the fact that most hydrologicalbasin has a size of
roughly 10ø in the arid regions. A more precisesource
distribution
should be based on a more accurate
basin
We usean equivalentempirical formulation by Gillette
and Passi [1988]for dust uplifting which requiresthe
knowledgeof the surfacewind speedand the threshold
velocityof wind erosion.In this formulationthe flux Fp
of particle size classp is approximated by the expression:
contouringby using a hydrologicalmodel.
We assumethat a basin with pronouncedtopographic
variations containslarge amount of sedimentswhich are
otherwise
'
accumulated essentially in the valleys and depressions,
and over a relatively fiat basin the amount of alluvium is
factorequalto 1 tzgs2 m-5, S
homogeneouslydistributed. We introduce here a source whereC is a dimensional
function S, which is the fraction of alluvium available is the source function described in section 2.3, U•0m is
the horizontal wind speed at 10 m, ut is the threshold
for wind erosion, as follows:
Fp- { 0
CSSp
2
ifru•0.•
> ut
(2)
20,258
GINOUX
ET AL.:
DUST
•
•
MODELING
particle size and the soil moisture[Pye, 1989]. Belly
[1964]suggested
a relationshipfor entrainmentof sand
Source function
90N
60N
AEROSOL
•
which takes into account the particle size and soil moisture content. This relationship has been modified here
30N
to usethe surfacewetness(providedby GEOS DAS) in
determiningthe threshold wind velocity:
0
30S
60S
ttt--
90S
{AV/PP;,P'•g(I)p(1.2
) ifw<0.5
,
•x•
80
120W
I
60W
0
60E
120E
180
...........
.::
::i•iiiiiiii::ii•iiii::ii!iii•i
•iii•i:i:iii•i•i!::
!::
•i!iiiiii•::i:::•i•i
i!•iiii:.:':--':
•.'...-'
'"..---':-.....
::iiii
i:..-'•i!
ilii
i:•i:.::
•::!::ii•::i•i:::.i•i:
:..:..:.::..-'
iiii:::.•ii:'""..'•
:......2.:,':
i!:.•i
i•i•ii
:•'.'...':,../,..:...".:..'
::
i•i•'""
'--"
0
0.1
0.2
0.4
0.6
0.8
1
Dust Emission(g.m'•.yr")
•
•
(a)
where A - 6.5 is a dimensionlessparameter, w is the
surfacewetness[0.001-1],•p is the particlediameter,g
is the accelerationof gravity,pp and Pa are the particle
and air density,respectively. The typical valuesof w in
arid regionsare between0.001 and 0.1, but are higher
than 0.5 after precipitation. For w-0.1, ut is 1.5 m
90N
60N
otherwise
•
30N
s-• for the smallestclassand 3 m s-• for the largest
0
30S
class. This formulation,basedon studiesfor large sand
particles (r > 30 pm), showsan increaseof ut with
60S
increaseof particle size. However, wind tunnel and field
90S
180
120W
60W
0
60E
120E
180
0
1
5
10
50
100
500
Aerosol
Index
TOMS
for particle r < 35pm, ut decreaseswith the increaseof
radius. Therefore the threshold velocity obtained from
equation(3) will be lowerthan that from wind tunnel
experiments,and the correspondingdust emissionwill
90N
be larger. In Figure I the similarity betweenthe source
60N •
distribution(upperpanel) and the annualdustemission
averagedfrom 1987 to 1990 (middle panel) showsthe
importanceof the sourcefunction (S) relative to the
other variablesin equation(2).
30N
0
30S
60S
experiments
[Iversenand White,1982;Marticorenaand
Bergametti,1995; $hao et al., 1996]have shownthat
::::"::"'
.........
•:""':'::•:::'
.......
2.5.
Removal
Processes
90S
180
120W
60W
0
60E
120E
180
Two types of removal processesare considered,the
dry deposition which includes the turbulent transfer to
0
0.1
0.2
0.4
0.8
1.6
3.2
the surfaceand gravitationalsettling, and the wet deposition which includes rainout and washout in and below
Figure 1. Comparisonbetween(top) the dust source
function,and (middle)the dust emissionand (bottom) clouds. The turbulent transfer of particlesat the surface
is calculatedas a first-orderprocessusinga deposition
distributions of TOMS aerosol index. The dust emission
and TOMS aerosolindex are averagevaluesfrom 1987 velocity v•. The velocityv• is assumedto be equivalent
to 1990.
to the exchangevelocity for heat and moisture at the
surface. This exchangevelocityis diagnosedfrom flux
profile relationshipsand is availablein the GEOS DAS
velocity,and sp is the fraction of eachsizeclasswhich data [Takacset al., 1994].However,particleswhichare
deposited at the surfacecan be uplifted outside source
is defined below.
regions
depending on the surface wetnessw and the
The fractions of clay and silt sizes are different for
wind
velocity.
To take into accountthis fact, we define
differentsoil types at eachlocation. Owing to the unan
effective
dry
depositionvelocity6• as follows:
certainties in the available soil texture data, we choose
a simple particle size distribution, following Tegenand
Fung [1994]. The fractionof clay is basedon the assumptionthat erodibleclay represents1/10 of the total
mass of emitted silt, and that of each silt subclass is
• _{ v•(w
+(1- w)exp[-(u•0,•
- ut)])
u•0,•
>(4)ut
va
otherwise
assumedto be the same. The sp valuesare thus 0.1 In this formula, Oawill be smaller than va with strong
for the class0.1-1/•m, and 1/3 for the classes1-1.8/•m, surfacewind speedover dry surface. Over wet surfaces,
1.8-3/•m, and 3-6/•m, respectively.
The main factor affectingthe thresholdvelocityut is
the interparticle cohesionforceswhich dependson the
6a is equal to va.
For large aerosolsthe most efficientremovalprocessis
gravitational settling. The changeof concentrationby
GINOUX
ET AL.' DUST AEROSOL MODELING
Table 2. Annual Budget of Dust for the Years 19871990 and 1996 a
3.
Model
Results
The dust distribution
Year
Era,
Dry,
Wet,
Tg yr- x Tg yr- x Tg yr- x
1987
1988
1989
1990
1996
1883
1956
1804
1829
1604
1653
1704
1613
1611
1451
256
289
229
237
166
Load,
Life,
Tg
day
38
40
36
34
31
7.1
7.3
7
6.6
6.6
aThere is a 2% difference between the total emission
and total deposition due to numerical errors in the ad-
vective scheme. Emission, em; dry deposition, dry; wet
deposition,wet; atmosphericburden, load; lifetime, life.
20,259
has been simulated
over several
time periods: 1987-1990, 1996, most of 1997, and a few
months in 1994. The choice of these periods is based
on the available data sets used for comparison. In this
section we present the global budget of dust and compare the model results with observations. The observed
data sets have been selectedfor their adequate temporal and spatial coverage. These data sets can be sep-
arated into surfacemeasurements(concentrationand
depositionflux), vertically integratedground-basedobservations(sizedistributionand opticalthickness),and
satellite data (vertical distribution and optical thickness).
3.1.
Global Budget
Table 2 givesthe annual budget for 5 years, 1987-1990
gravitational settling is calculatedby an implicit scheme and 1996. The averaged global dust emission is 1814
at all vertical levels. The settling velocity vstk for a Tg yr-•. The calculatedemissionrangesfrom 1604Tg
particle of radius r is determinedusingthe Stokeslaw' yr-• in 1996to 1956Tg yr-1 in 1955,reflectingmainly
the differencesin the wind speed at 10 m. The calculated emissionsare within the range of previous model
2ppgr2
CCun
n
vstk
= 9 /•
estimateswith similar sizesrange: 1250 Tg yr-• for
particlesradiuslessthan 8/•m [Tegenand Fung,1995],
wherepp is the particle density,g is the accelerationof 1800Tg yr-• for particleslessthan 10/•m [Dentenetet
gravity,and/• isthe absolute
viscosity
ofthe air (1.5 105 al., 1996],and 3000Tg yr-• for a singleclasscentered
kg m-X S--1),andCcunnis the Cunningham
correction at 1.25 /•m [Mahowaldet al., 1999]. Dry depositionis
which takes into accountthe viscositydependencyon the main loss processwith an averaged value of 1607
Tg yr-1. Wet deposition
accounts
for 235Tg yr-1 and
air pressureand temperature[Fuchs,1964].
Wet scavenging
in the model includesrainout (in- representsonly 10% of the total loss. The atmospheric
cloudprecipitation)and washout(belowcloudprecipi- dust load is 36 Tg with an averaged lifetime of about
tation) in large-scale
precipitationand in deepconvec- 1 week. Table 3 lists the budget and lifetimes for each
tive cloud updraft. The parameterizationof thesepro- size class. The highest emissionis obtained for particles
cesses for sulfate aerosols has been described in detail
by Chin et al. [2000]. Unfortunately,there are few
data concerningthe scavenging
efficiencyof dust particles. Here we use the samewet scavengingparameters
for dust as those used for sulfate, which is probably
reasonable,sincethe scavengingefficiencyreported by
Gallowayet al. [1993]for non-sea-saltsulfateand nitrate are roughlythe sameas the valuesusedby Duce
et al. [1991]for dustoverthe North Atlantic Ocean.
2.6.
Transport
Dust particlesare transportedin the atmosphereby
advection,convection,and turbulent mixing. The different methods use to solve these processeshave been
describedelsewhere[Allen et al., 1996; Chin et al.
2000]. Here, briefly,advectionis computedby a flux-
form semi-Lagrangian
method [Lin and Rood, 1996].
Moist convectionis parameterizedusing archivedcloud
massflux fieldsfrom the GEOS DAS. Turbulentmixing
due to mechanicalshear and buoyancyis approximated
by eddy diffusion. The eddy diffusion coefficientis calculated using a 2.5 order local closureparameterization
[Helflandand Labraga,1988].
radii ranging from 1 to 2 /•m. The lifetime of the two
smallest classesis 2 weeks and drops to 1 day for the
largest class(3-6 /•m). Our estimatedlifetime for the
smallestparticlesis 50% longerand for the largestparticles is 50% shorter than the valuesreported by Tegen
and Lacis[1996].The differences
canbe explainedby a
stronger dry deposition rate and a weaker wet deposition rate in our case. •Vhile the lifetime of particles size
1-2 /•m agreeswith Tegenand Lacis[1996],it is twice
Table 3. Annual Budget, Averaged Over 5 Years
(1987-1990and 1996) for Each Size Classes
a
Size,
Emis,
Dry,
Wet,
Load,
Life,
/•m
Tg yr- •
Tg yr- •
Tg yr- •
Tg
day
0.1-1
1-2
335
571
265
484
79
97
14
14.5
14
9
2-3
3-6
0.1-6
504
404
1814
461
396
1606
49
l0
235
6.1
1.3
35.9
4.3
1.1
7.1
aEmission, em; dry deposition, dry; wet deposition,
wet; atmospheric burden, load; and lifetime, life.
20,260
GINOUX ET AL.: DUST AEROSOL MODELING
Table 4. Total Simulated Deposition to Various Ocean Regions
ComparedWith EstimatedValuesby Duceet al. [1991]and Prospero[1996]•
Ocean
North
South
North
South
North
South
Global
Duceet al. [1991]
Pacific
Pacific
Atlantic
Atlantic
Indian
Indian
Prospero[1996]
480
39
220
24
100
44
910
96
8
220
5
20
9
358
GOCART
92
28
184
20
138
16
478
aUnitsare in T g yr-•.
the valuereportedby Mahowaldet al. [1999]. How- lated to the movementsof the Intertropical Convergence
ever, Mahowaldet al. [1999]considered
only onedust Zone (ITCZ) whichoccupiesits southernmostposition
size classthat may result in a shorter lifetime.
in winter (•5øN) and northernmostpositionin summer
The simulated total annual deposition to various
(•20øN). The seasonal
shift in the modelmatchesthat
ocean regions are compared in Table 4 with Duce et
observedby satelliteinstruments,suchas the advanced
al. [1991]and Prospero[1996]estimations
fromin situ veryhighresolution
radiometer(AVHRR) [Husaret al.,
measurements.
Duceet al. [1991]usea scavenging
ratio 1997].
for dust of 200 for the North
Atlantic
Ocean and 1000
The
surface concentrations
are similar
to the val-
overthe remainderof the worldOcean.Prospero
[1996] ues calculatedby Dentenet et al. [1996]over North
has suggestedto use a scavengingratio of 200 for all Africa with maxima over the Sahel and Sahara, but
ocean regions. Our calculated total annual deposit.
ion they differ significantlyelsewhere. The most striking
over the North Pacific and North Atlantic are compara- difference is over the Taklimakan desert which is a re-
ble to that estimatedby Prospero[1996].However,over gionalminimumin the work of Dentenet et al. [1996]
the other oceans,the model results are closerto values
but a maximum in Figure 2. Another important dif-
estimatedby Duceet al. [1991].The majordifference
is ferenceis near the sourcesin the Southern Hemisphere
over the North
Indian
Ocean where the simulated
total
depositionis 8 times higher than the value estimated by
Prospero[1996].As mentionedby the author,thereis a
large uncertainty in the estimated value becausethere
were very few data over the Indian Ocean.
3.2. Dust Concentrations
Layer
in the Boundary
where the GOCART
model surface concentrations
are
significantlyhigherthan Dentenetet al. [1996],except
over Australia. The comparisonwith Tegen and Fung
[1994]indicatessomedifferencesin Africa where the
Sahel source was not included in their study and in
Australia
where the surface concentrations
from their
model were as high as over the Sahara.
The monthly dust concentration has been compared
The calculated seasonal dust concentrations
in the
with the observationsat 18 sites indicated in Figure 2
planetaryboundarylayer (0 to 1 km altitude) areshown with their number. These sites were operated by the
in Figure 2. These values have been averagedover 5 Universityof Miami [Prospero,1996]. Somesitesare
years (1987-1990 and 1996). The sourceregionsap- located closeto the sourceregions: Izana (station 4)
pear clearly in this figure as well as the patterns due near North Africa, Kaashidhoo(8) near India, Cheju
to long-range transport. The highest concentrations (9) and Hedo (10) in the Asian east coast,Cape Grim
(above250/•g m-3) are locatedin the NorthernHemi- (17) nearAustralia,and CapePoint (7) in SouthAfrica.
sphere: over the western Sahara and the Sahel region, Other sitesare located downwindof the sourceregions:
in part of the Arabic Peninsula and Iran, and in the Barbados(1), Miami (2), and Bermuda (3) alongthe
Asian sourceregions(Taklimakanand Gobi deserts). Saharandust plumes;Enewetak(11), Midway (12) and
Over North America there is a lesspronouncedmaxi- Oahu (13) in the corridorof Asian and North Amermum which is centered on the Salton Sea in southern
ican plumes;Norfolk Island (18) and New Caledonia
California. In the Southern Hemispherethe peak values (16) downwindof the Australiandust sources.The reare locatedin the regionsof the Lake Eyre basin (Aus- maining sitesare located in remote regionsof the North
tralia), Altiplano (Bolivia), Patagonia(Argentina),and Atlantic, Mace Head (5); South Atlantic, King George
Namibia (South Africa). Thesemaxima are quite per- (6); and Pacific,Nauru (14) and Funafuti (15). Most
sistent through the seasons. The model showsa clear data have been collected in the 1980s and 1990s, but
latitudinal shift of the dust plume over the North At- the period of measurementsvaries betwen sites. The
lantic from winter to summer. This seasonal shift is relocationsof the 18 sites are given in Table 5.
GINOUX
ET AL-
DUST AEROSOL
Figure 3 showsthe comparisonof the annually averageddust concentrationsmeasuredand simulatedat the
18 siteswherethe modelresultsare 5 years(1987-1990,
1996) averagedvalues.The annualconcentrations
vary
by more than 3 ordersof magnitude from more than 30
MODELING
20,261
siteswithvalues
greaterthan2/•g m-• andtheremote
siteswithvalueslessthan2/•g m-•. Themodelconcentrations agreewith the observationsto within a factor
of 2 in the dusty regions,but the model overestimates
the concentrationin the remote regions,particularly in
/•g m-a at Izanato lessthan 0.1/•g m-1 at NauruIs- the North Pacific. For example, dust concentration at
land. The sitescan be divided into two groupsdepend- Midway (12) is overestimated
by as muchas a factor of
ing on the range of surfaceconcentrations:the dusty 5. Part of this discrepancyis due to the fact that this
Dec-don-Feb
::=:•=:•:2='"•
=
............
sos
90SI
": -:=::==
..,.=
.........
:•i•/•s•s/•:/,::•:•i•,•:•:•:?:•s•*•-:C'
'.•'.:•.
-'-•"•....
:'•=•"-'-ß
- '.•.•i•..-'•½½i'•
iiii!.?,•i.-"..-".-"-•-"--•'....-•',.-'..'iiii.:--::-}:½/,:/:ii•8•
-'---'
...........
Mor-Apr-Moy
i
90N
_
60N
............................................................
50N
_
0_
5OS _
60S 90S
Jun -Jul-Aug
90N
_,..::•
::•.-.. ......
,?:-::::s::. ' "•
60N
-'......
•
..•..•
. ..
50N
-.'=':==
==
o 5
5OS
••
-
•
90S
Sep-Oct-Nov
90N
;,..;
!•ii•.-'..-i:::½.':::..:..}:
....!,
.•.,;
::;i.
.............
.:..:.
?--'....?.::.i!•./
......
-:-':•:-"C.!
....... - '"'
"•"•'"::":'
......
-:"*":'•-:
...............
:'::'
:-- '......
-::-::'::::•:•::-::::
............
:•:?'•:•':'
............
5?%?:/%:
60N i}•i:•;•;•!•/•??•?•::??:•:..•.•;.s::.;:...:•.:.....•:!:•..`....•...:.`..•;•:i•:•;•:•
-'-!i•--:
..................
-:.--:•::--..•:
'"'•'
i ....
o
3os
6os
9os
180
120W
60W
2.5
0
5
60E
25
120E
125
250
180
500
Figure 2. Seasonalvariationof dust concentration,
averagedover5 years(1987-1990and 1996),
in unitsof/•g m-3.
20,262
GINOUX
ET AL.: DUST AEROSOL
Table 5. List of Stations Managed by the Rosenstiel
trations
MODELING
are within
the observed standard
deviations
ex-
Schoolof Marine and AtmosphericScience(Univer- cept in December and January. However, the simulated
sity of Miami) for the Measurementof Atmospheric standard deviation is particularly large during these 2
Dust
Concentration
Site
Name
Latitude
Longitude
1
2
3
4
5
Barbados
Miami
Bermuda
Izana
Mace Head
13.17øN
25.75øN
32.27øN
28.3øN
53.32øN
59.43øW
80.25øW
64.87øW
16.5øW
9.85øW
6
7
King George
Cape Point
62.18øS
34.35øS
58.3øW
18.48øE
8
K aashido o
4.95 oN
73.45 o E
9
Cheju
33.52øN
126.48øE
10
Hedo
26.92øN
128.25øE
11
Enewetak
11.33øN
162.3øE
12
Midway
28.22øN
177.35øW
13
14
15
16
Oahu
Nauru
Funafuti
New Caledonia
21.33ø51
0.53 øS
8.5ø S
22.15øS
157.7øW
166.95 o E
179.2 øW
167øE
17
Cape Grim
40.68øS
144.68øE
18
Norfolk
29.08øS
167.98øE
Island
months. This comparison shows no indication of overprediction of dust concentrationaloft from the African
sources.The situationin Asia is generallydifferentfrom
that in Africa. The Taklimakan desert, which is the ma-
jor Asiandustsource[Zhanget al., 1998],is locatedbetween 1 and 2 km above sea level. There is evidence that
Asiandustis transportedaround8 km altitude [Merrill
et al., 1989]. The long-rangetransportat sucha high
altitude is much more rapid than commonly found for
Africandustplumes(•3-5 km altitude) [Karyarnpudiet
al., 1999]. An overprediction
of dust emissionfrom the
Taklimakan will have a longer and stronger impact on
the remote atmospherethan from African dust sources.
To better understand the origin of the discrepancyover
the North Pacific,it is necessaryto validate the emission
rates and dust loads close to the Asian sources and the
altitude of the traveling plumes. This will be discussed
in the next
sections.
3.2.2. Southern Hemisphere.
In the Southern
Hemisphere the model simulates the seasonalcycle at
the different
sites downwind
of the dust sources reason-
site was operated with sector control sothat the samples ably well (sites6, 7, 14, 15, 16, 17, and 18 in Figure
were taken only when the winds came from the ocean. 4). At Cape Grim, southeastof the Australian dust
However, such sampling bias may not explain a factor source,the model reproducesthe seasonalcycleof dust
5 differencebetweenthe model and the data. The good concentration,although the austral summer maximum
agreementin the dusty regionsand the large discrepancy in the remotePacific suggesta somewhatdeficient
representationof the transport, the removal processes,
Annuolconcentrotion
m
or the size distribution in the model. Figure 4 shows
the seasonal variation
of dust concentration
at each of
the 18 sites,with 11 sites(1, 2, 3, 4, 5, 8, 9, 10, 11, 12,
and 13) in the NorthernHemisphereand 7 sites(6, 7,
14, 15, 16, 17, and 18) in the SouthernHemisphere.
3.2.1. Northern Hemisphere.
Over the dusty
regionsthe model reproducesthe seasonalvariation although sometimesthe observed values are underestimated. For example, the June maximum at Barbados is
underestimatedby 50•. Farther away from the African
dust source, at Miami and Bermuda, the model results
are within
ues.
the standard
deviation
The differences between
12
'
.....
.7'
of the observed val-
these three sites can be
due to subgrid scale local meteorologyor to the fact
that the measurementshave been averagedover different periods. Another example is at Cheju: the spring
maximum is well reproducedbut is underestimatedby
50%. Similarly,the modelfails to capturethe dust peak
in April at Hedo.
A difficulty in assessing
the model resultsat the surface is that it is not necessarilyrepresentativeof dust
concentration aloft where most dust plumes are traveling. Transport aloft is tested with data from Izana
(site 4 in Figure 2) whichis locatedabout 2.5 km above
sea level. Measurements
lO.O
at this location
are considered
1.0 .,t::•i•;61.
'
0.1
o.1
1.o
lO.O
Meosurements
Figure 3. Comparison of annual mean concentration
simulated(bold line) and observed(dots) at 18 sites;
to be representative
of mostAfricandustplumes[Chia- the standard deviation of the simulated and observed
pelloet al., 1997].The observedconcentrations
at Izana valuesare representedby shadingand vertical segments,
are the largest of the 18 sites. The simulated concen- respectively.
GINOUX ET AL.' DUST AEROSOL MODELING
, . ,(1.).B,or.bo,
do,
s, , .
•'• 50
E 40
o, 30
3 2o
20,263
(3) Bermuda
(2) Miami
? 25
E 20
? 20
E 15
•.•.1,5
•o
•• •o
r2
......
r-,
J FMAM
J FMAMJJASOND
E
8
o-,
6
0
J FMAMJ
D
.. (s) .u0ce
?-1o
3
J JASON
JASOND
2.0
1.5
•.o
4
0.5
0.0
J FMAMJ
•-;'-'-10
E
8
or,
6
,. (7! .Co.
pe,
,
J
JASOND
FMAMJJASOND
(9) Cheju
(8) Kooshidoo
•'
-'-" 25
5O
E 20
...........
.......
• so
• 2o
•15
'-'10
J
FMAMJ
J
JASOND
FMAMJ
J FMAMJ
JASOND
(12) Midway
(1 1) Enewetok
(1 0) Hedo
?
• 50
40
C'1o
2.0
E
E 1.5
o,
• zo
2O
3
::::
JASOND
1.0
3
8
6
4
% 0.5
g o.o
J FMAMJ
J FMAMJ
JASOND
(15) Oohu
•
J
JASOND
(14) Nouru
10
½ 2.0
8
E 1.5
E 1.5
o', 6
3 4
• 1.0
'•
• 1.0
'-"
%
= •
•
C'5
• 0.5
• 0.5
g o.o
...................................
d FMAMd
d ASON
D
(16) NewCaledonia
E 4
A o.o
d FMAM
d d ASOND
d FMAM
(17) CapeGrim
.,---5 ............
E 4
J FMAMJ
d ASOND
J J ASOND
(15) Funofuti
2.0
E
•
FMAM
J FMAM
C'5
E 4
J J ASOND
d d ASON
(18) NorfolkIsl
........
J FMAM
©'
D
' '
J J ASOND
Figure 4. Comparisonof •nonthlty dust concentration,simulatedand observedat 18 sites,in
unitsof pg m-3.
is underpredictedby a factor of 2. Farther east, at Norfolk Island, the austral winter minimum is overpredicted
by a factor of 2. At Cape Point, near the Namibian
dust source, the model simulates correctly, within the
standard deviation, the increase of concentration from
March to July. Finally, at King George which is influenced by dust emissionfrom Patagonia, the model is
able to reproduce the weak seasonalcycle with a slight
increase in January and December. The two observed
tropical Pacificby a factor of 2 to 5. This overestimation probablyappliesto the higherlatitudesof North
Pacific as well. This discrepancy could be related to
a misrepresentation
of the altitude of the Asian dust
plumes,dust emissions
from the Asiansources,or the
removal
3.3.
rates.
Vertical
Distribution
Figure 5 showsthe 4-year (1987-1990)zonally avhigh peaksin Januaryand April are due to dust events
eraged
dust concentrationsin the model. Dust congeneratedduring the limited period of measurements.
rapidly with height. In
To summarize these comparisons, we conclude that centrationgenerallydecreases
the model reproduces the dust concentration and seasonal variations at most sites, but overpredicts the dust
levelsfrom Asian dust sourcesover the tropical and sub-
the free troposphere, zonal mean concentrations range
from near 0 to 5 pg m-3. Figure5 presentssomedifferences with the model results reported by De•te•er
20,264
GINOUX
ET AL-
DUST
AEROSOL
MODELING
and 125 (September17, 1600UT) and the corresponding simulateddust concentrationsfrom the sealevel to
10 km altitude. The LITE orbit 115 signal showsthe
African plume which rises over the African west coast
and extendsto 5 km altitude (first panel). The strong
signalnear the oceansurfaceis due primarily to other
..........................................................................
typesof aerosols(e.g., seasalt). As seenin the second
panel of Plate 1, the model reproducesthe vertical and
Zonal concentrotibn(/xg m-3)
200
] ...........
•
400
3 600
c:/...i...••i
horizontal extensionof the dust plume, and the location
• 1000
800
:iiii',i:i','"'"'•:':':"•'":•..-:
•:•i':•:'":""•:
'a:i!i•:.--•i•i•.-"---:•i•i
•
'•'-"•"--•""••••'•••of the dust source in the model is located correctly at
90S
60S
30S
0
30N
60N
90N
Latitude
between 18.5ø and 20øN. The last two panels in Plate
I are the vertical distributions along the orbit 125 over
China. The LITE data (third panel) showa dust plume
Figure 5. Zonal mean dust concentration with isolines extending to 7-8 km altitude as well as dust mixed in
at 0.5, 1, 5, 10, and 50/•g m-3.
the boundary layer. The model can reproducethis extension, but does not capture the pronouncedfeature
et al. [1996]. We obtain a strongernorthwardmeridional transport to high latitudes at around 800 mbar,
particularly in the Northern Hemisphere. Tegen and
Fung [1994]obtainedalso a meridionaltransportbut
at higheraltitude (around100 mbar). Althoughthere
is considerableevidenceof long-rangetransport to high
northernlatitudes,for example,in Norway[Franz•net
al., 1994]and Greenland[Davidsonet al., 1993],there
are few direct
measurements
of dust vertical
distribu-
tion.
The most useful data with which to compare the
model vertical
distribution
are the aerosol backscatter-
at 3-4 km (last panel). Theseresultsindicatethat dust
particles emitted from the Asian sourcesare uplifted
more efficientlythan over the Sahara to high altitude in
the jets where they are subjectedto efficientlong-range
transport. Such results have been previouslysuggested
[e.g.,Merrill et al., 1989].
3.4.
Dust Deposition
Figure 6 showsthe annualtotal deposition(dry and
wet deposition)averagedover 5 years (1987-1990and
1996). In this figure a latitudinal gradient appears
clearly, with a maximum between 0ø and 45øN, and
a minimum below 45øS. The highest deposition is over
ing measurementsmade during the Lidar in SpaceTech-
westernChinawith a maximumof 162g m-2 yr-1.
nologyExperiment(LITE) [Winker et al., 1996]. The
Figure 7 showsthe comparisonbetweenthe annual
observed and simulated
removal
rate at 16 sites.
The
data cover 10 days in September 1994. The lidar return signal was sampled at 10MHz, correspondingto
a spatial resolution of 15 m between the Earth's surface to 20 km altitude. The data used for comparison
site location, the period of measurements,and the observed and simulated values of the total annual deposition flux are given in Table 6. The 16 sites are also
are the backscatteringratio at 532 nm (http://wwwlite.larc.nasa.gov).The orbits passingover westernSahara (orbit 115) and the Taklimakandesert(orbit 125)
observedvaluesvariesfrom 450 g m-2 yr-• overthe
Taklimakanto 0.08 g m2 yr-• in equatorialPacific.
have been selectedfor comparisonwith model simulations. Plate I showsthe backscatteringratio measured
along the orbits 115 (September17, 1994, 0100 UT)
indicatedin Figure 6 with their number. The range of
Figure 7 showsthat the model successfully
predicts
high versuslow depositionrangingover4 ordersof magnitude, although it tends to underpredict the highest
Annual Deposition(g m'2yr")
90N
.:• -
'..............
'
..:::•i::::::iii:•iiii:::'•"•::...•:•:'
..............
•::•:'.,•"""5
•....`....•i..•:•:J•t..`...::i!...`...ii!::!::•.•...•...•..•ii...•;i•..•L::
•."':"
;.:•
.::...%:-:.:g'
.:•
'..
:.::"
.:.
;.).•...'...'.'
::.'.,'•
:;:::"'-.•r.x,'"•,
"•'••;..'•iiii•
......
•'":.
'"*•'
"':""
*:':'
'"'
-.
'--'
30N [':::;:'"'"'"'"'"'"'-•'""•
•iiii
iiiig?'"•JJJi•;-"liiiiiiiii!!i:-•:,-:•'•?';!
.::-'
.....:-'::::*!•:•:.'7-...-::/'.,'0,,•
:":::"-:.'•;':'.':'?•::::"
'"..':::.::'
......
•.•:.•..•:/•:•.`:•:J:....•.;•.ii;•i;.:•:...•..•;`.•.*i•4:.:..•`.•:.•;•:;•:.:...:•.•.:J•:...J•..;•i•...w:•.•ii:jijiiiii.•J5
,.,J•::'"'"'•:'
:•;;::•`•:•::•!•:•!::•!;•4•;!•::•.•:•;•;.•:•i•::;•5•;•iii•ii•`•
'"•':.'9--'""
"• '•:•/•i::•:;!•,•,::,::•,!!
.....................................
,I,'" 1
..................
--"-;:.--'i•!
.....
:.........................................
i:,:•:,iliji:•:•'"•i•:11,/ji:il;•
.......
..-..:iie..-'ji'":"'""•m•?
...........
ß
....'
.
90S[
180
120W
60W
0
60E
120E
180
[
o
o.1
0.5
1
5
10
50
1O0
Figure 6. Global distributionof the annualtotal depositionflux (g m-2 yr- 1), averaged
over5
yearssimulation(1987-1990and 1996).
GINOUX ET AL.: DUST AEROSOL MODELING
m ø.
o
000o
o
o
o
ß
0
0
o
o
a0
tun'ePn:l!11¾'•,•
20,265
20,266
GINOUX ET AL.- DUST AEROSOL MODELING
differences. This indicates that there is a large local
variability in depositionrates that is not resolvedby
the model. Despite this fact, we concludethat the depositionfluxesare in generalsimulatedreasonablywell.
Annual Deposition(g.m'.yr
1000.0
3.5.
100.0
Optical Thickness
The optical thicknessr is calculatedfrom the dust
massload by the relation
n'
10.0
-
O
......
5"'
O
.g
..':%
..i
1.0
0.1
0.1
1.0
10.0
Measurements
100.0
1000.0
Figure ?. Comparisonof annual total depositionflux
simulated
and observed at 16 sites.
3Qext(A,ri)Mi
i
'
(6)
where Qext(A,ri) is the extinction efficiencyfactor at
wavelengthA and effectiveradius ri, Mi is the column
massloading for the size classi, and pi is the massdensity of the size classi. The valuesof Qext(A, ri) are
calculated using Mie theory and assuminglognormal
sizedistributions(M. Chin et al., Troposphericaerosol
optical thickness from the GOCART model and comparisons with satellite and Sun photometer measurements, submitted to Journal of Atmospheric Science,
2001, hereinafter referred to as Chin et al., submitted
manuscript,2001).
3.5.1. Comparison with AVHRR
data.
The
global distribution of the aerosoloptical thicknesshas
valuesand overpredict the lowestvalues. The underpre-
been retrieved
dicted valuesover the Taklimakan(15), Tel Aviv (16),
and Spain (3) are most likely due to the contribution
of particleslarger than 6/•m (largestsimulatedparticle
radius). The closerthe measurements
to the sourcethe
higher the fraction of large particles. Observeddeposi-
[Husar et al., 1997]. Owingto the large variabilityof
tion fluxes at three closelylocated sites 12, 13, and 14
vary by a factor of 5, while the model showsonly 20%
at 630 nm from the AVHRR
instrument
land surfacereflectancein the visible range, only information over the ocean can be retrieved. The optical
thicknesshas been derived assuminga singlescattering
albedo equal to I which leadsto underestimatethe true
value. Although the AVHRR measurestotal aerosoloptical thickness,dust is the major aerosolcomponentat
Table 6. Model Versus ObservedTotal Dust Deposition Fluxes at 16 Sitesa
Site
Location
I
2
3
Shemya
French Alps
Spain
4
Midway
5
Miami
6
Oahu
7
Enewetak
8
9
52.92øN
45.5øN
41.8øN
28.2øN
174.06øE 1981-1987
6.5øE
1955-1985
2.3øE
1987-1990
177.35øW 1981-1987
25.75øN
80.25øW
1982-1983
21.3øN
157.6øW
1981-1987
11.3øN
162.3øE
1981-1987
Fanning
Nauru
10
11
Samoa
Rarotonga
12
13
New Caledonia
Norfolk Island
14
15
16
Latitude Longitude Years
New Zealand
Taklimakan
Tel Aviv
Measurement
b GOCART
c
0.6
2.1
5.3 :t: 2.6
0.6 (0.3-1.1)
0.9 :t: 0.1
2.6 :t: 0.5
3.3 :t: 0.6
1.8 :t: 0.1
1.62
3.3 :t: 0.5
0.42 (0.4-0.5)
1.3 :t: 0.2
0.44
0.9 :t: 0.3
3.9øN
159.3øW
1981-1987 0.09 (0.05-0.22)
0.53øS
166.95øE
1981-1987
14.25øS
21.25øS
170.6øW
159.75øW
1981
0.15 (0.02-0.22)
1981-1987
0.21
0.35 + 0.07
0.23 :t: 0.04
22.15øS
29.08øS
167øE
167.98øE
1983-1985
1983-1987
0.25 :t: 0.05
0.3 :t: 0.06
34.5øS
40øN
32øN
172.75øE
85øE
34.5øE
1983
1994
1972
0.23
0.37
0.64
0.14 (0.1-1)
450 (110-1900)
30 (20-40)
0.4 :t: 0.1
0.35 :t: 0.1
0.25 :t: 0.05
133 :t: 7
11 :t: 0.7
aUnitsarein g m-1 yr-1
bReferences:Prosperoet al. [1989]for sites1, 4, 6, 7, 8, 9, 11, 12, and 13; De Angelisand
Gaudichet
[1991]for site2; Avila et al. [1997]for site3; Prospero
et al. [1987]for site5; Arimoto
et al. [1987]for site 10; Arimotoet al. [1990]for site 14; Zhanget al. [1998]for site15; Ganor
and Mamane [1982]for site 16.
CModelresultsare 5 years(1987-1990and 1996) average.
GINOUX
ET AL.: DUST
AEROSOL
MODELING
20,267
GOCART AOT (650 nm) 1987-1990
90N
60N
....
.
....
............
:.. ...:
......
:--'
..........
30N
30S
60S
90S
AVHRR AOT (630 nm) 1987-1990
90N
60N
30N
30S
60S 90S
120W
180
60W
0
0.1
0.05
60E
0.2
0.3
120E
0.4
180
0.5
0.6
Figure 8. Comparison
of the meanopticalthicknesses
(top) simulatedat 650 nm and (bottom)
observedat 630 nm by AVHRR from 1987 to 1990.
several regions such as the tropical-subtropical North
Atlantic, western Pacific, and Arabian Sea.
Figure 8 showsthe mean optical thicknessat 650 nm
in comparisonwith the mean AVHRR optical thickness
at 630 nm from 1987 to 1990. The comparison shows
that the calculated optical thicknesscorrespondswell to
the observed values over the dust dominated
oceanic re-
Hemisphere. The observed high values in the tropical
regionsis due to biomassburning aerosolswhich are not
simulated.
It has been reported from the AVHRR data that the
location of the aerosolplume in the tropical North At-
lantic shiftswith the seasons[e.g.,Rao et al., 1988]. In
their modelingstudy, Tegenand Fung [1995]suggested
gions. The model reproducescorrectly three major dust
plumesextending over the North Atlantic, the Arabian
that emissions from human
disturbed
soils had to be in-
Sea, and the North Pacific. The simulatedbackground
valuesof optical thicknessin the Northern Hemisphere
are between 0.05 and 0.1 which is comparable to the
observedvalues. Becauseof the significantcontribution
of other aerosoltypes which have not been included in
this work, it is possiblethat the model overestimatesthe
backgroundvaluein the remoteregionsof the Northern
seasonal
shift. On the otherhand,recentstudieshave
cluded in the model in order to reproduce the observed
shown that it is the combination of biomass burning
and dust aerosolsthat causesthe seasonalshift [Takamura et al., 2000; Chin et al., submitted manuscript,
2001]. To understandthe dust contributions,Figure 9
showsthe monthly variation of the latitudinal position
of the maximum dust aerosol optical thickness along
25
20-
1510-
I
4
7 10
1987
I
4
7 10
1988
I
4
7 10
1989
I
4
7 10
1990
Month
Figure 9. Latitudinal variation of the maximum optical thicknessalong the 25øW meridian de-
ducedfrom AVHRR data (continuous
line) betweenJanuary1987and December1990 (November
1988 is missing),and from the modelsimulations(dashedline)
20,268
GINOUX
ET AL.: DUST AEROSOL MODELING
(1) Barbados
(2) Cape Verde
0.50
0.40
0.30
0.20
1.4
1.2
1.2
1.0
1.0
o.8
0.6
0.8
0.6
0.4
O. lO
0.00
(,5..).Ban.
izo.
um.
bo.
u
1.4
0.2
0.0
......
1
0.2
0.0
.............................
•
o
J
1.0
F
M
A
M
J
J
A
S
0
N
D
ß . . [4) Sde Baker
0.8
1
o
J
1.0
...........
F
M A
....
M
J
J
A
S
0
N
D
(5.) Bahrain
J
1.0
0.8
F
M
A
M
J
J
A
S
0
N
D
ß . . (.6) .Do[on.zod.
go.d. . .
0.8
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0.0
0.0
0.6
ß
0.0
,
.
•0
1 •--•-•-•.ee_.e
ßß • 1 .....e-e-e-l-le-i-e-e• ....e... • 1e.e_
e-e-e-•-e
--e..........
o
J
F
M A
M
J
J
A
S
0
N
D
o
J
F
M
A
M
J
J
A
S
0
N
D
J
F
M
A
M
J
J
A
S
0
N
D
Figure 10. Comparisonof opticalthicknesses
simulated(bold line) and measuredby AERONET
(dots) at six sites;the standarddeviationof the simulatedand observedvaluesare represented
by shading and vertical segments,respectively. The lower panels give the monthly Angstrom
parameter a, calculatedfrom the AERONET optical thicknessesat 440 and 670 nm.
without the necessityof invokingany contributionfront
the 25øW meridian. From this figure it appears that
the model reproducesthe observedlatitudinal shift in
1989 and 1990. However, in 1987 and 1988 from late fall
to early spring,the simulatedlatitude of the dust plume
is located farther north. The difference is particularly
significativein March 1988. To understandthis difference, we have analyzed the daily distributions. From
this analysisit appearsthat the GOCART model simulates an intense dust plume originating from the west
Sahara and moving to Europe over the last week of
March 1988, in agreementwith the TOMS aerosolindex. For the same period the TOMS aerosolindex has
maxima over Nigeria and Ghana, south of 10øN. These
maxima indicate the presenceof biomassburning activ-
ters that
are monitored
and maintained
at the NASA
ities.
extinctionbehavior[Eck et al., 1999]. The empirical
From this comparisonit seemsthat the contribution
of dust load to the aerosol optical thicknessis generally not sufficientto explain the interannual variability
of the latitudinal shift of the aerosol plume over the
North Atlantic. The emissionof other aerosoltypes by
biomassburning is the major factor to explain maximum optical thicknesssouth of 10øN over the Atlantic
in winter. However,due to the interannual variability of
fire activities and surface wind conditions, the maxima
of dust load and aerosoloptical thicknesscan occasionally correspondin winter, as it was for the years 1989
and 1990. The low latitude of dust plumes is simulated
anthropogenicdisturbancesto soils.
3.5.2. Comparison with AERONET
data. The
simulated optical thicknesshas been comparedwith the
expressionof ct is given as
Goddard SpaceFlight Center [Holbenet al., 1998].
Data have been collectedsince 1993, but most sites have
been operational after 1996. All the AERONET values
used in this comparisonhave been screenedfor cloud-
free conditions[Smirnovet al., 2000a].This screening
have rejected a substantial number of data. To avoid
bias, the comparisonis basedon model resultsextracted
for the dayswhen cloud-screened
data are available. We
select six sites where dust is considered to be the dom-
inant aerosoltype, although the contribution of other
aerosoltypes can be significant. The presenceof other
aerosolscan be detected on the basisof the wavelength
dependencyof optical depth. The Angstrom parameter a can be used as a first-order indicator of spectral
c• - - •
in(,x3-}•
)'
(7)
where r2 and rz are the optical thicknessat wavelength
•z (440 nm) and •2 (670 nm), respectively.In general,
the smallerthe particle size, •he strongerthe wavelength
dependence,thus the larger the ct values. Typical values
of c• range from >2 for smokeparticles and pollution
aerosolsto nearly zero for high optical thicknessdust
particles[Holbenet al., 1991].
Figure 10 showsthe comparisonof opticalthicknesses
calculated
at 450 nm from the model results and ob-
served at 440 nm at six AERONET
sites. The c• values
AerosolRobotic Network (AERoNET) data. AERONET is a federatedworldwidenetwork of Sun photome- retrieved at these sites are also shownin Figure 10 for
GINOUX ET AL.: DUST AEROSOL MODELING
20,269
Thesecomparisons
suggestthat the modelreproduces
indicationof particle sizes.Table 7 showsthe location,
associated
with majordustplumes
the selectedperiodof measurements
for eachsite, and the opticalthickness
over the oceans. The simulated backgroundvalues in
the correspondingsimulationperiod.
At Barbados the model simulates the observed sea- the Northern Hemisphereat 650 nm is 0.05-0.1. Near
sonalvariation of optical thicknesswithin the standard the dust sources the simulated values are within the
deviation. The observed June 1997 maximum is shifted
standard deviation of the AERONET
data, with the
of the Asiansite. Fromthesecomparisons
it
to July by the model,but the simulatedvaluesare well exception
within the observed standard deviation.
$mirnov
et
al. [2000b]havestudiedin detailthe AERONET data
does not seem that the model overestimates dust emissions.
overBarbadosand founda high correlationbetweenthe
monthlymeanopticalthicknessand dust concentration 3.6.
at the surface.Figure 4 showsthat the model captures
the observed maximum
surface concentration
Aerosol volume size distribution has been retrieved
in June
(1987-1990and 1996). The discrepancybetweenthe
model and observationsin June 1997 is thus not sys-
tematic but is most likely related to the variation of
calculated emission rates in 1997.
The Cape Verde site is closeto the Saharandust
source. At this site the model reproduces,within the
standard deviation, the observedoptical thickness,at
the exceptionof June 1997. For this month the c• pa-
Size Distribution
from the AERONET data by an iterative inversion
algorithm[Dubovikand King, 2000]. The inversion
algorithmcalculates
the effectivesizedistributionfor
the total atmosphericcolumn,assumingaerosolparticlesas polydisperse
homogeneous
spheres.Duboviket
al. [2000]showed
that the sizedistribution
canbe re-
trieved reasonablywell. For dust particlesradius less
than 0.3/•m the retrievalproblemassociated
with nonrameter is around 0.1 which is characteristic for dust, sphericityis avoidedwhen the angularrange of sky
and the discrepancycannotbe attributed to the pres- radiancesis limited to scattering angles smaller than
ence of other aerosols. It is most likely due to strong
30ø-40
ø (aroundnoon). For largerdust particlesthe
perturbationsoverthe dust sourcewhichwerenot re-
limitation to small scattering angles does not apply.
The size distribution has been retrieved at only a few
solved in the model.
The Banizoumbou site is southwest of the Bodele
sites(http://aeronet.gsfc.nasa.gov).
Figure 11 shows
depression
and is representative
of the Saheliandust the seasonalvariation of the vertically integrated volsources.The modelreproduces
the seasonalcyclewithin Umesizedistribution at Cape Verde, Banizoumbou,and
the standard deviation except in April where it under- Dalanzadgad,wherethe AERONET retrievedsizedisestimatesthe optical thicknessby a factor of 3. The tributions have been averagedover 3 months. The obexplanationgivenfor the discrepancy
at Cape Verde, serveddistributiongenerallyshowsa bimodaldistribution with a fine (sulfateor carbonaceous
aerosols)
and
June 1997, can also be given for this case.
a
coarse
mode
with
particles
larger
than
0.6/•m.
Only
At SdeBoker (Israel) and Barhainthe modelreproduces the seasonal variation within the standard devi-
the coarse mode is of interest for this analysis.
At Cape Verdethe modelreproduces
the volumesize
distribution
for
each
season,
although
it
systematically
allyhigh,indicatingthepresence
of submicron
particles.
underestimates
the
larger
size
fraction.
The contribuThe dust opticalthicknessis probablyoverestimated
in
the model.
tion of the largestfractionto the total massis 5 times
ation. However, the c• values at both sites are gener-
At Dalanzadgad(Mongolia)the simulatedoptical lower than the fraction centered at 1.5 /•m. It thus
thicknessis higherthan observedby as muchas a fac- seemsthat the lower than observed simulated optical
tor of 2 in July, but is within the observedstandard thicknessin June 1997 at Cape Verde is due to subscale
deviation for the other months. As at the Sde Boker perturbations
unresolved
by the model.At Banizoumand Bahrain sites,the c• valuesare generallyaround 1 bou the model reproducesthe maximum volumedistrithe fraction
which indicatesthat the dust optical thicknessis prob- butionat 1.5/•m radiusbut overestimates
below 1.5 /•m and underestimates
the fraction above.
ably overestimatedin the model.
Table 7. Site Name, Latitude, Longitude,SelectedYear of Measurements,and
Corresponding
Year of Simulationfor Each AERONET Site Usedfor Comparison
Site
Name
Latitude
Longitude
Year Observed
Year Simulated
1
Barbados
13.2øN
59.5øW
1997
1997
2
Cape Verde
16.7øN
22.9øW
1997
1997
3
4
5
Banizoumbou
Sde Boker
B ahr ain
13.5øN
30.5øN
26.3 øN
2.65øE
34.5øE
50.5 oE
1996
1996
1998-1999
1996
1996
1996-1997
6
Dalanzadgad
43.6øN
104.4øE
1997
1997
20,270
GINOUX ET AL.' DUST AEROSOL MODELING
Cape Verde
Dec-Feb
,
1.000
Banizoumbou
97
...........
Dec-Feb
,
Dalanzadgad
96
Dec-Feb
•,,,,
97
.000
O. lOO
.100
O.OLO
.010
o.ool
.001
0.1
1
10
0.1
1
Mar-May 97
10
1
Mar-May 96
1.000
,
1.00C
O.lOO
Mar-May
,
..........
10
97
i
,,,,,
0.10C
O.OLO
0.01C
o.ool
0.001
0.1
1
10
0.1
1
Jun-Aug 97
10
0.1
Jun-Aug 96
1.000
.000
0.100
.100
.100
0.010
.010
.O10
0.001
.001
.............
'
1
10
Jun-Aug 97
.000
o
0.1
1
Sep-Nov
10
"
.001
0.1
97
1
Sap-Nov
1 .ooo
.000
O. lOO
.100
O.OLO
.010
o.ool
10
0.1
96
................
1
Sep-Nov
10
97
i
,,,,,,
,001
0.1
1
Radius (,u,m)
10
0.1
1
Radius(•m)
10
1
10
Radius (•m)
Figure11. Comparison
ofvolume
sizedistribution
(/•m
a /•m-2) simulated
(dashed
line)and
retrieved(continuous
line) fromAERONET data.
Also,the modelunderestimates
by 50%the sizedistri- opticalthickness(Figure10) and surfaceconcentration
butionat 1.5 /•m in winterandspring.Thiscouldex- (Figure4) neartheAsiansources
arenotlargelyoverplain the lowerthan observedopticalthickness
in 1997 estimated.
However,
because
thelifetimeof 1.5/•msize
at Banizoumbou
(Figure10).
particlesis muchlongerthan that of particleslarger
At Dalanzadgad
(Mongolia)the observed
sizedistri- than 6 /•m, the overprediction
of smallparticleemisbutionis quitedifferentfromthat at CapeVerdeand sionwill producea modelbiaswhichis stronger
far
Banizoumbou.
Thecoarse
fractionis characterized
by away from the sourcethan near the source. Further-
twomodes:onewith a meanradiusat 1.5/•m andthe more,we haveseen(Plate 1) that dust plumesfrom
second
witha meanradiusat 7 /•m (Figure11). The Asiansources
travelat highaltitudeswherethe winds'
contribution of the secondmode to the volumesizedisarestronger.We suspect
that thisefficient
long-range
tribution is dominant. This mode is outsidethe simu- transport,especially
for smallparticles,
is at theorigin
lated particlesizerange. The modeloverestimates
the of the too highbackground
valuesof opticalthickness
firstmodebya factorof3 in springandbya factorof 5 (0.05-0.1)andsurface
concentration
(5 timesthe obin summer.The overestimation
of smalldustparticles servedvalue) estimatedby the model over the North
compensates
for the absence
of particleslargerthan6 Pacific.
/•m suchthat the simulatedatmospheric
dust load is
Finally,Figure11 highlights
the largedifference
in
comparable
to the observations.
Thisexplains
whythe the AERONET retrievedsizedistributionsbetweenthe
GINOUX
ET AL.: DUST
AEROSOL
MODELING
20,271
African and Asian stationswhichdoesnot appearin the Fung [1995]attributed to humanactivity. Also, other
modelresults. In equation(2) the two variableswhich models which identify deserts as dust sourcestend to
depend on the size distribution are the soil size fraction greatly overestimatedust emissionsin Australia. Our
Spand the thresholdvelocityut. In our formulation,ut results did not show such a discrepancy because the
is proportional to the squareroot of the particle radius. model limits the sourcelocation to the Lake Eyre basin
As we discussedin section 2.4, there is some evidence which is consistent with the analysis by Prospero et al.
that below 30/•m the threshold velocity is instead de- (submitted manuscript,2000). The comparisonwith
creasingwith the increaseof radius. This could explain LITE data showsthat dust uplifting is simulated at the
the lower than observedfraction of particles between 3 correct place and time, although the model does not
and 6/•m. The use of a more realistic parameterization reproduce all observed features in the vertical aerosol
[e.g.,MarticorenaandBergametti,1995]mightimprove
distributions.
our results by increasingthe fraction of large particles.
However, it will not help reproducethe large variability in size distribution between sites. This variability
is most likely due to differencein soil properties. The
Comparisons with observations have shown that the
model is generally able to reproduce the seasonalvariation of the surface concentration and optical thickness
valueof Spshouldbe regionspecific,and in the caseof
Asia it should have a higher percentageof large parti-
mote regionswhere concentrationsare <2/•m the model
tends to overestimate the concentrations. Comparison
with surfacedata is difficult becausethere is a large local variability which is not resolvedby the model. Nevertheless, the model overestimates the surface concentration in the North Pacific by a factor of 3 to 5 which
is much larger than the local variability. To understand
the origin of this discrepancy,the dust concentration,
vertical distribution, optical thickness, and size distribution have been compared with the observations in
the vicinity of the Asian dust sources.The comparison
did not show any significantdiscrepancyexcept for the
size distribution. The model overestimates by a factor of 3 to 5 the size distribution at 1.5 ttm while, in
contrast, the observationsshow a peak at 7 ttm. Near
the sourcethe overestimation of small particles compensates the absenceof particles larger than 6 /•m radius
and produces a simulated atmospheric dust load which
is comparable to the observations. Far away from the
sourcesthe atmospheric dust load is largely overestimated because small particles have a longer lifetime.
Also, our comparisonwith LITE data showsthat Asian
dust particles are uplifted efficiently to 7-8 km altitude
where they are subject to long-range transport by the
jet streams. Therefore the atmospheric dust loading
over the North Pacific is sensitive to the parameterization of dust uplifting and to the soil size distribution in Asian desert. From the comparison of the size
distribution near the dust sources, it clearly appears
that the size distribution differs significantly between
cles.
More
measurements
are needed
to constrain
the
soil size distribution.
4.
Conclusions
We have developed a new global dust source function
which
identifies
the dust
source locations
at the
topographically low regionswith bare soil surface. This
sourcefunction is a generalizationof the finding in satel-
lite data analysis(Prosperoet al., submittedmanuscript,
2000) that the dust sourcesare associatedwith topographic depressions. The accumulation of sediments
in these depressionsduring pluvial periods servesas a
sourceof loose particles which can be easily uplifted
into the atmosphere. We have generalizedthis finding
and assumethat all topographic lows have accumulated
sediments which serve as potential dust sources.
With
this distribution
of dust sources we have sim-
ulated the atmospheric dust distribution with the GO-
CART modelwhichis drivenby assimilatedmeteorological fields from the GEOS DAS. We have incorporated
the emission, transport, and removal processesof dust
sizesranging from 0.1 to 6 /•m radius. The model has
been evaluated globally over a period of severalyears
by comparing simulation results with observationsof
dust concentration at the surface,vertical distribution,
surface deposition flux, optical thickness, and size distribution.
in the "dusty" regions(concentrations>2 ttm). In re-
We have been successfulin reproducing observed the African and Asian soils. Past studies have been
global scale dust distribution patterns by considering concentrated on dust characterization and mobilization
dust emissions exclusively from these so-called "hot over the Sahara, and there ares little data available on
spots" regions. The total emissionis estimated at 1604- Asian dust sources. This study showsthat such data
1960Tg yr-• in a 5-yearsimulation.Thisresultis simi- are needed to simulate correctly dust distribution over
lar to other modelresultsalthoughdistribution patterns East Asia and the North Pacific.
differ substantially. The model source distribution allowsto simulatethe latitudinal shift of dust plumesover
Acknowledgments.
We are very grateful to Andrea
the North Atlantic but doesnot always reproducethe Molod for her continuous help with the DAO data and
low latitude of the aerosol plumes in winter. The con- Richard Rood for providing the computational resources.
We would like to thank Dennis Savoie, University of Miami,
tribution of aerosolsfrom biomassburning is the most for providing us with the measured dust concentrations at
likely reason for the discrepancyin winter, rather than the surface. We are grateful to Didier Tanr• for giving us
the contribution from disturbed sourcesthat Tegenand accessto the Sun photometers data at Cape Verde and Ban-
20,272
GINOUX
ET AL.' DUST
izoumbou, and to Charles McClain for the Sun photometer
data at Bahrain. Finally, we would like to thank the three
anonymous reviewers for their fruitful comments. This work
has been supported by the NASA Global Aerosol Climatology Program.
References
Allen, D. J., P. Kasibhatla, A.M. Thompson,R. B. Rood,
B. G. Doddridge,K. E. Pickering,R. D. Hudson,and S.J. Lin, Transport-inducedinterannual variability of carbon monoxide determined using a chemistry and transport model, J. Geophys.Res., 101, 28,655-28,669,1996.
Arimoto, R., R. A. Duce, B. J. Ray, A.D. Hewitt, and
J. Williams, Trace elementsin the atmosphereof American Samoa: Concentrationsand depositionto the tropical
South Pacific, J. Geophys.Res., 92, 8465-8479, 1987.
Arimoto, R., B. J. Ray, R. A. Duce, A.D. Hewitt, R. Boldi,
and A. Hudson, Concentrations, sources,and fluxes of
trace elementsin the remote marine atmosphereof New
Zealand, J. Geophys.Res., 95, 22,389-22,405, 1990.
Avila, A., I. Queralt-Mitjans,and M. Alarc6n,Mineralogical
compositionof African dust delvered by red rains over
northeasternSpain, J. Geophys.Res., 102, 21,977-21,996,
AEROSOL
MODELING
Eck, T. F., B. N. Holben, J. $. Reid, O. Dubovik, A.
Smironov, N. T. O'Neill, I. Slutsker, and S. Kinne, Wavelength dependenceof the optical depth of biomassburning, urban, and desert dust aerosol, J. Geophys. Res.,
10d, 31,333-31,350, 1999.
Franz•n, L. G., M. Hjelmroos, P. Kallberg, E. BrorstrSmLundin, S. Juntto, and A.-L. Savolainen,The yellow snow
episode of northern Fennoscandia, March 1991- A case
study of long-distance transport of soil, pollen and stable organic compounds,Atmos. Environ., 28, 3587-3604,
1994.
Fuchs, N. A., The Mechanics of Aerosols,Pergamon, New
York, 1964.
Galloway, J. N., D. L. Savoie, W. C. Keene, and J. M.
Prospero, The temporal and spatial variability of scavenging ratios for nss sulfate, nitrate, methanesulfonate
and sodium in the atmosphere over the North Atlantic
Ocean, Atmos. Environ., Part A, 27, 235-250, 1993.
Ganor, E., and Y. Mamane, Transport of Saharan dust
acrossthe eastern Mediterranean, Atmos. Environ., 16,
581-587, 1982.
Genthon, C., Simulations of the long range transport of
desert dust and sea-salt in a general circulation model,
in Precipitation, Scavengingand AtmosphericSurfaceEx1997.
change,edited by S. E. Schwartz and W. G. N. Slinn, pp.
1783-1794, Taylor and Francis, Philadelphia, Pa., 1992a.
Belly, P. Y., Sand movementby wind, U.S. Army Coastal
Genthon, C., Simulations of desert dust and •a-salt aerosols
Eng. Res. Tech. Memo., 1, 38 pp., 1964.
in Antarctica with a general circulation model of the atChiapello, I., G. Bergametti, B. Chatenet, P. Bousquet,
mosphere, Tellus, Set. B, dd, 371-389, 1992b.
F. Dulac, and E. SantosSoares,Origins of African dust
transported over the northeastern tropical Atlantic, J. Gillette, D. A., and R. Passi,Modeling dust emissioncaused
by wind erosion, J. Geophys. Res., 93, 14,233-14,242,
Geophys.Res., 102, 13,701-13,709, 1997.
Chin, M., R. B. Rood, S.-J. Lin, J.-F. Milllet, and A.
Thompson, Atmospheric sulfur cycle simulated in the
global model GOCART: Model descriptionand global
properties, J. Geophys.Res., 105, 24,671-24,687,2000.
Davidson C. I., et al., Chemical constituents in the air and
snow at Dye 3, Greenland, I, Seasonalvariations, Atmos.
Environ., Part A, 27, 2709-2722, 1993.
1988.
Guelle, W., Y. Balkanski, M. Schulz, B. Marticorena, G.
Bergametti, C. Moulin, R. Arimoto, and K. D. Perry,
Modeling the atmospheric distribution of mineral aerosol:
Comparison with ground measurementsand satellite observations for yearly and synoptic timescales over the
North Atlantic, J. Geophys.Res., 105, 1997-2012, 2000.
De Angelis,M., and A. Gaudichet,Saharandust deposition Helfiand, H. M., and J. C. Labraga,Designof a nonsingular
level 2.5 second order closure model for prediction of atover Mont Blanc (FrenchAlps) during the last 30 years,
mosphericturbulence,J. Atmos. Sci., d5, 113-132,1988.
Tellus, Set. B, 43, 61-75, 1991.
DeFries, R. S., and J. R. G. Townshend,NDVI-derived land Herman, J. R., P. K. Bhartia, O. Torres, C. Hsu, C. Seftor,
and E. Celarier, Global distributionof UV-absorbingaerocoverclassificationat a globalscale,Int. J. RemoteSens.,
solsfrom Nimbus 7/TOMS data, J. Geophys.Res., 102,
15, 3567-3586, 1994.
16,911-16,922, 1997.
Dentener, F. J., G. R. Carmichael, Y. Zhang, J. Lelieveld,
and P. J. Crutzen, Role of mineral aerosol as a reactive Hillel, D., Introductionto Soil Physics,Academic,SanDiego,
Calif., 1982.
surfacein the global troposphere,J. Geophys.Res, 101,
22,869-22889, 1996.
Dickerson, R. R., S. Kondragunta, G. Stenchikov, K. L.
Civerolo, B. G. Doddrige, and B. N. Holben, The impact
of aerosolson solar ultraviolet radiation and photochemical smog, Science,278, 827-830, 1997.
Dubovik, O., and M.D. King, A flexibleinversionalgorithm
for retrieval of aerosolopticalpropertiesfrom Sun and sky
radiance measurements,J. Geophys. Res., 105, 20,67320,696, 2000.
Holben, B. N., T. F. Eck, and R. S. Fraser, Temporal and
spatial variability of aerosol optical depth in the Sahel
region in relation to vegetation remote sensing, Int. J.
Remote $ens., 12, 1147-1163, 1991.
Holben, B. N., et al., AERONET- A federated instrument
network and data axehivefor aerosolcharacterization, Re-
mote. Sens. Environ., 66(1), 1-16, 1998.
Husar, R. B., J. M. Prospero, and L. L. Stowe, Characterization of tropospheric aerosols over the oceans with
the NOAA advancedvery high resolutionradiometeroptiDubovik, O., B. N. Holben, M.D. King, Y. J. Kaufman, A.
Smirnov,T. F. Eck, and I. Slutsker,Accuracyassessments cal thicknessoperational product, J. Geophys. Res., 102,
16,889-16,909, 1997.
of aerosol optical properties retrieved from AERONET
Iversen,
J. D., and B. R. White, Saltation threshold on
Sun and sky-radiancemeasurements,J. Geophys.Res.,
Earth, Mars and Venus, Sedimentology,29, 111-119, 1982.
105, 9791-9806, 2000.
Duce, R. A., Sources, distributions, and fluxes of min- Joussaume, $., Three dimensional simulations of the atmosphericcycle of desertdust particlesusing a generalcireral aerosolsand their relationship to climate, in Dalhem
culation model, J. Geophys.Res., 95, 1909-1941, 1990.
Workshopon Aerosol Forcing of Climate, edited by R.
J. Charlsonand J. Heintzenberg,pp. 43-72, John Wiley, Karyampudi, M. V., et al., Validation of the Saharan dust
New York, 1995.
plume conceptualmodel asinglidar, Meteosat, and
Duce, R. A., et al., The atmosphericinput of trace speciesto
ECMWF data, Bull. Am. Meteorol. Soc., 80, 1045-1076,
1999.
the world ocean, GlobalBiogeochem.Cycles,5, 193-259,,
1991.
Levin, Z., E. Ganor, and V. Gladstein, The effectsof desert
GINOUX ET AL.: DUST AEROSOL MODELING
particles coated with sulfate on rain formation in the eastern Mediterranean, J. Appl. Meteorol., $5, 1511-1523,
20,273
Takacs,L. L., A. Molod, and T. Wang, Documentation
of the Goddard Earth ObservingSystem(GEOS) General Circulation Model-Version 1, Technical report series
1996.
on globalmodelinganddata assimilation,
NASA Tech.
Lin, S.-J., and R. B. Rood, Multidimensional flux-form semiLagrangian transport schemes,Mon. Weather Rev., 1
2046-2070, 1996.
Mahowald, N., K. Kohfeld, M. Hansson, Y. Balkanski, S. P.
Harrison, I. C. Prentice, M. Schulz, and H. Rodhe, Dust
sourcesand deposition during the last glacial maximum
and current climate: A comparison of model results with
paleodata from ice cores and marine sediments, J. Geophys. Res., 10•, 15,895-15,916, 1999.
Marticorena, B., and G. Bergametti, Modeling the atmospheric dust cycle, l, Design of a soil-derived dust emission scheme, J. Geophys. Res., 100, 16,415-16,430, 1995.
Merrill, J. T., M. Uematsu, and R. Bleck, Meteorological
analysisof long range transport of mineral aerosolsover
the North Pacific, J. Geophys.Res., 9•, 8584-8598, 1989.
Prospero, J. M., The atmospheric transport of particles to
the ocean, in Particle Flux in the Ocean, edited by V.
Ittekkot, P. Sch•ffer, S. Honjo, and P. J. Depetris, John
Wiley, New York, 1996.
Memo., 10,i606, 1, 100 pp., 1994.
Takamura,T., H. Okamoto,Y. Maruyama,A. Numaguti,
A. Higurashi,
andT. Nakajima,
Globklthree-dimensional
simulationof aerosolopticalthickness
distributionof var-
iousorigins,J. Geophys.
Res.,105,17,853-17,873,
2000.
Tegen,I., andI. Fung,Modeling
of mineraldustin the atmosphere:Sources,
transport,and opticalthickness,
J.
Geophys.Res., 99, 22,897-22,914,1994.
Tegen,I., andI. Fung,Contribution
to theatmospheric
mineral aerosolload from land surface modification, J. Geo-
phys. Res., 100, 18,707-18,726,1995.
Tegen,I., andA. A. Lacis,Modeling
ofparticlesizedistribution and its influenceon the radiative properties of min-
eral dust aerosol,J. Geophys.Res., 101, 19,237-19,244,
1996.
Tegen,I., A. Lacis,andI. Fung,The influence
of mineral
aerosolfrom disturbedsoilson the global radiation budget, Nature, 380, 419-422,1996.
Prospero,J. M., R. T. Nees, and M. Uematsu, Deposition Tegen,I., P. Hollrig, M. Chin, I. Fung, D. Jacob,and
J. Penner, Contribution of different aerosolspeciesto
rate of particulate and dissolvedaluminiumderivedfrom
Saharandust in precipitationat Miami, Florida, J. Geothe global aerosolextincti(,n optical thickness:Estimates
from model results, J. Geophys.Res., 102, 23,895-23,915,
phys. Res., 92, 14,723-14,731,1987.
1997.
Prospero, J. M., M. Uematsu, and D. L. Savoie, Mineral aerosoltransport to the Pacific Ocean, in Chemical Torres, O., P. K. Bhartia, J. R. Herman, Z. Ahmad, and
J. Gleason, Derivation of aerosol properties from satelOceanography,
pp. 187-218,Academic,San Diego,Calif.,
lite measurements
1989.
Pye,K., Aeolian
DustandDustDeposits,
2nded.,Academic,
of backscattered
ultraviolet
radiation:
Theoretical basis, J. Geophys. Res., 103, 17,099-17,110,
1998.
San Diego, Calif., 1989.
Rao, C. R. N., L. L. Stowe, E. P. McClain, and J. Sapper, Velde, B., Introduction to Clay Minerals, Chapman and
Hall, New York, 1992.
Development and application of aerosolsremote sensing
with AVHRR data from the NOAA satellites, in Aerosol Winker, D. M., R. H. Couch, and M.P. McCormick, .An
and Climate, editedby P. V. Hobbsand M.P. McCormick,
overview of LITE: NASA's Lidar In-space TechnologyExpp. 69-79, A. Deepak, Hampton, Va., 1988.
periment, Proc. IEEE, 8•, 164-180, 1996.
Scheffer,F., and P. Schachtschabel,Lehrbuchder Boden- Zhang,X. Y., R. Arimoto, G. H. Zhu, T. Chen, and G. Y.
kunde,13th ed.,F. Enke,Frankfurt,Germany,1992.
Zhang,Concentration,size-distributionand depositionof
Schubert,S. D., R. B. Rood, and J. Pfaendtner, An assimmineral aerosolover Chinese desert regions, Tellus, Set.
ilated data set for Earth scienceapplications, Bull. Am.
B, 50, 317-330, 1998.
Meteorok Soc., 7•, 2331-2342, 1993.
Schulz, M., Y. J. Balkanski, W. Guelle, and F. Dulac,
M. Chin, O. Dubovik, B. Holben, P. Ginoux ,
and S.-J. Lin, Laboratory for Atmospheres, NASA Goa three-dimensional simulation of a Saharan dust episode dard Space Flight Center, Greenbelt, MD 20771, USA.
testedagainstsatellite-derived
opticalthickness,J. Geo- (chin@rondo.
gsfc.nasa.
gov; [email protected];
phys. Res., 103, 10,579-10,592,1998.
[email protected];[email protected].
gov;
Shao,Y., R. Raupach,and J. F. Leys,A modelfor predicting [email protected])
aeolian sand drift and dust entrainment on scales from
J. M. Prospero, Rosenstiel School of Marine and
paddockto region,Aust. J. Soil. Res., 3•, 309-342,1996. Atmospheric Science, University of Miami, 4600 RickSmirnov, A., B. N. Helben, T. F. Eck, O. Dubovik, and I.
enbacker Causeway, Miami, FL 33149, USA. (jprosSlutsker,Cloud-screeningand qualtity control algorithms [email protected])
for the AERONET database, Remote Sens. Environ., 73,
I. Tegen, Max-Planck-Institute for Biogeochemistry,
337-349, 2000a.
P.O. Box 100164, 07701 Jena, Germany. (itegen@bgcSmirnov, A., B. N. Holben, D. Savoie,J. M. Prospero,Y. jena.mpg.de)
J. Kaufman, D. Tanre, T. F. Eck, and I. Slutsker, Relationshipbetweencolumn aerosoloptical thicknessand (ReceivedOctober 17, 2000; revisedMarch 27, 2001;
in situ groundbaseddust concentrations
over Barbados, acceptedApril 16, 2001.)
Geophys.Res. Left., 27, 1643-1646,2000b.
Role of aerosol size distribution
and source location in