Characterization of African dust transported to Puerto ...dust.ess.uci.edu/ppr/ppr_RRM03.pdf · microphysical properties of dust particles collected on surface and airborne filter

Characterization of African dust transported to Puerto Rico by

individual particle and size segregated bulk analysis

Elizabeth A. Reid,1,2,3 Jeffrey S. Reid,3 Michael M. Meier,4 Michael R. Dunlap,4

Steven S. Cliff,4 Aaron Broumas,4 Kevin Perry,5 and Hal Maring6

Received 11 September 2002; revised 25 November 2002; accepted 24 February 2003; published 12 July 2003.

[1] As part of the Puerto Rico Dust Experiment (PRIDE), airborne and surface dustparticle samples from Africa were collected and subjected to bulk elemental and single-particle analysis. Airborne samples were collected on polycarbonate filters at variousaltitudes and underwent single-particle scanning electron microscopy with energydispersive analysis with X-rays (EDAX) to derive elemental ratios of key soil elements.Particle chemistry was related to size and morphological characteristics. At the principlesurface site, particles were collected on a Davis Rotating Drum (DRUM) cascade impactorstrips in eight stages from 0.1 to 12 mm at 4 hour time resolution. These samples weresubjected to X-ray florescence (XRF) to determine bulk elemental composition from Althrough Zn. The elemental data showed good correlation between the DRUM and theaircraft samples. Cluster analysis of single-particle data resulted in 63 statisticallysignificant clusters. Several clusters can be easily related to their parent mineralogicalspecies. However, as dust particles are to a large extent aggregates, most clusters are basedon a continuum of varied mineralogical species and cannot be easily categorized. With60,500 total particles counted from the airborne filters, a statistically significant number oflarge particles could be analyzed. Estimated mean surface area modal diameter is �5 mm,with an average aspect ratio of 1.9. An apparent change in source region is seen in themorphological data and non alumino-silicate minerals but is not seen in the aluminum tosilicon ratio. We suspect homogenization during long-range transport. INDEX TERMS:

0305 Atmospheric Composition and Structure: Aerosols and particles (0345, 4801); 0330 Atmospheric

Composition and Structure: Geochemical cycles; 0360 Atmospheric Composition and Structure: Transmission

and scattering of radiation; 0368 Atmospheric Composition and Structure: Troposphere—constituent transport

and chemistry; KEYWORDS: Sahara, dust, minerology, size distribution, particle morphology

Citation: Reid, E. A., J. S. Reid, M. M. Meier, M. R. Dunlap, S. S. Cliff, A. Broumas, K. Perry, and H. Maring, Characterization of

African dust transported to Puerto Rico by individual particle and size segregated bulk analysis, J. Geophys. Res., 108(D19), 8591,

doi:10.1029/2002JD002935, 2003.

1. Introduction

[2] Airborne mineral dust particles present special prob-lems in aerosol mechanics and radiation calculations. Dustparticles are a significant, if poorly characterized, compo-nent of atmospheric aerosols. Irrespective of the intensedifficulties in measuring dust particle properties [e.g., Reidet al., 2003b], their range of shapes, sizes and chemistriesare difficult to quantify in a consistent and usable statistical

manner. Hence their very nature defies easy light scatteringand absorption calculations. These issues have in partresulted in airborne dust having the highest uncertaintiesof any aerosol species in global direct forcing estimates[Intergovernmental Panel on Climate Change (IPCC),2001; Sokolik et al., 2001].[3] Recently, single-particle analysis from scanning and

transmission electron microscopy methods has become apopular tool for dust particle characterization. Examples ofdetailed analysis of dust are given by d’Almeida and Schutz[1983], Coude-Gaussen et al. [1987], Reid et al. [1994],Anderson et al. [1996], Gao and Anderson [2001], andKoren et al. [2001]. From these measurements have come avariety of characterizations including particle elementalratios and shape factors such as particle cross-sectionalarea, perimeter, major and minor axis. However, thesemethods have inherent shortcomings that are difficult toovercome. For the most part, analysis is based on a two-dimensional cross section of the particle. Particle chemistrycan be ambiguous due to matrix effects, electron scatter, andinternal particle in-homogeneity. Hence like so many dust-

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. D19, 8591, doi:10.1029/2002JD002935, 2003

1Science and Technology Corporation, Hampton, Virginia, USA.2Space and Naval Warfare Systems Center, San Diego, California,

USA.3Naval Research Laboratory, Monterey, California, USA.4Material Science and Chemical Engineering Department, University of

California, Davis, California, USA.5Department of Atmospheric Sciences, University of Utah, Salt Lake

City, Utah, USA.6Rosenstiel School of Marine and Atmospheric Science, University of

Miami, Miami, Florida, USA.

Copyright 2003 by the American Geophysical Union.0148-0227/03/2002JD002935$09.00

PRD 7 - 1

measuring methods, findings from single-particle analysismust be treated only semiquantitatively.[4] The Puerto Rico Dust Experiment (PRIDE) was a

joint Office of Naval Research (ONR) and NASA fundedfield campaign to study the properties of African dustadvected into the Caribbean region [Reid et al., 2003a].Conducted at Naval Station Roosevelt Roads on the easternside of the island of Puerto Rico, participating investigatorsmonitored Saharan dust transport across Northern TropicalAtlantic Ocean between 28 June and 24 July 2000. Duringthe PRIDE campaign, the island of Puerto Rico was near thecenter of the dust transport plume from Africa. Daily dustoptical depths at 500 nm averaged 0.26, with a maximum of0.52. Dust concentrations at the surface were at timesgreater than 70 mg m�3.[5] A goal of PRIDE was to investigate the extent to

which the microphysical, chemical and optical properties ofdust particles need to be known before remote sensingsystems can accurately determine dust optical depth andradiative flux. Direct forcing column closure was not a goal,but rather an attempt was made to find constraints such thatparticle mircophysical and optical properties can be consis-tent. As part of this goal, we investigated the extent towhich particle size distributions and chemistry can even becharacterized and the uncertainties size parameterizationshave in model and satellite investigations.[6] In this manuscript we examine the chemical and

microphysical properties of dust particles collected onsurface and airborne filter samples using bulk and single-particle analysis methods. We begin with an elementalanalysis of cascade impactor data collected at the primaryPRIDE field site. This is followed with qualitative andquantitative examinations of single particles collected onthe Navajo research aircraft. Particle size, shape, andchemistry functions are presented. A cluster analysis isperformed and likely mineralogical speciation is given.We conclude by comparing findings from the bulk andsingle-particle methods and develop a consistent picture ofdust microphysics and chemistry and make recommenda-tions for further modeling studies.

2. Methods

2.1. Sample Collection

[7] Atmospheric dust samples were collected at thesurface, and by aircraft at various elevations during thePRIDE field campaign. The surface site utilized a DavisRotating Drum (DRUM) impactor to collect time resolvedaerosol data. The Davis Rotating Drum (DRUM) impactor,owned by investigators from the University of California,Davis (UCD), was deployed at the principal ground site forthe PRIDE study; Cabras Island, Puerto Rico. Cabras Island(latitude 18.21�N, longitude 65.60�W) is a small facility onNaval Station Roosevelt Roads several hundred metersoffshore of the easternmost portion of Puerto Rico. Thesampler was located on the roof of the University of Miamimobile laboratory about 10 m from the shoreline (D. L.Savoie et al., Spectrally-resolved light absorption bySaharan aerosols over the tropical North Atlantic, submittedto Geophysical Research Letters, 2003) (hereinafter referredto as Savoie et al., submitted manuscript, 2003), with asampling height of �10 m.

[8] The DRUM sampler collects particles in eight stageswith 50% cut points at: 5 mm, 2.5 mm, 1.1 mm, 0.74 mm,0.56 mm, 0.34 mm, 0.24 mm, 0.09 mm in diameter, respec-tively. This sampler is a modified version of the originalinstrument described by Cahill et al. [1985]. It was recentlyaltered by using a slit jet (instead of a circular jet) for eachof the eight stages, and increasing the flow rate from 5 to10 L min�1. Samples were collected on Apiezon greasecoated strips on rotating drums moving at �1 mm each 4hours, giving 4 hour resolution. Sampling was performedcontinuously from 3 July through 24 July 2000, with minorpower outages on 9 and 15 July.[9] Samples for single-particle analysis were collected

with a twin-engine, 8-seat Piper Navajo owned and operatedby Gibbs Flite Center and contracted by SSC San Diego. Asmall inlet mounted on top of the Piper Navajo researchaircraft collected aerosol samples in flight. The samplingsystem pulled a flow rate of 5 L min�1 through aknife-edge inlet for particle collection on a 37 mm,0.8 micron pore size polycarbonate filter. The samplingregime was nearly isokinetic, with a rough 50% inlet sizecut of approximately 6 mm for sea salt and 10 mm for dry dust.We expected that some large particles would be collected,especially at elevation in the low humidity regimes, andindeed we do see a very few particles with sizes�20microns.[10] Forty aircraft samples were collected during the

PRIDE study over fourteen flight days. Of this set ninehave been fully analyzed and the data are summarized inTable 1. Sample duration varied from 19 to 72 minutes,with sample elevation ranging from the surface to 5000 m.All samples were collected in cloud free areas. The aircraftelevation during sample collection typically was main-tained within the layer being studied: anywhere from1000 to 4000 m thick. For this analysis we groupedparticle collection regimes into three altitude ranges. Par-ticles collected above the trade inversion in the Saharan Airlayer are given the SAL designation. As they are above thetrade inversion, these particles are almost exclusively dustmostly free of marine influence. Second, particles collectedin the marine boundary layer (altitudes <1000 m) are giventhe MBL designation. These particles are a combination ofdust and sea salt. Finally, on one occasion, an integratedsample, designated ‘‘integrated’’ (INT), was collected

Table 1. Summary of Airborne Filter Samples Analyzed by

Single-Particle Analysisa

DateDuration,

min Region

Number ofParticlesAnalyzed Hi-Magnification

5 July 2000 54 Integrated 4504 Y16 July 2000 33 SAL 17043 Y16 July 2000 31 MBL 2583 N20 July 2000 72 SAL 9941 N21 July 2000 33 SAL 9348 Y22 July 2000 19 SAL 8564 N22 July 2000 44 SAL 6038 N24 July 2000 59 MBL 2928 N

aAll samples were analyzed for particles >1.5 mm in diameter. Selectedsamples were also analyzed at higher magnification to examine submicronparticles. Sampling region is subcategorized to samples taken in the marineboundary layer (MBL), typically <500 m in altitude; Saharan Air layer(SAL), above the trade inversion; and integrated taken from the surface tothe top of the dust layer.

PRD 7 - 2 REID ET. AL.: CHARACTERIZATION OF AFRICAN DUST IN PRIDE

where sampling occurred from the surface to the top of thedust layer.

2.2. DRUM Analyses

[11] After collection, DRUM sample strips were sub-jected to X-Ray Fluorescence (XRF) analysis in beamline10.3.1 from the synchrotron source at the Advanced LightSource of Lawrence Berkeley National Laboratory. Bulkelemental concentrations of Na through Cu were mea-sured. However, as light elements such as Na emit lowenergy X-rays, attenuation and interference make detectiondifficult, and significantly increase the measurement uncer-tainty. Na, being the lightest element analyzed, had totaluncertainties (analytical and sampling) of roughly 45 to 50%for typical DRUM sample loadings. This uncertainty wassignificantly lower for the heavier elements, such as Si,which had uncertainties in the range of 15–20%. Thesampling uncertainty remained stable at roughly 5%, whilethe analytical uncertainty depended on the sample loading,the particle sizes, the calibration curve fit, and the X-rayspectra peak fitting. For example, the uncertainty in thechlorine measurements increased when sulfur was present,due to difficulties in deconvoluting the spectral peak overlap.While sodium and chlorine correlated very well (r = 0.9) thechlorine to sodium mass ratio was �1.9, significantly higherthan the nominal value of 1.5 which is more typical for agedsea-salt particles.[12] As discussed by Reid et al. [2003b], on the basis of

comparisons with bulk samples collected by the Universityof Miami, both the DRUM and a colocated MOUDI under-estimated dust concentrations by a third. Preliminarily wesuspect inlet losses. Further, Reid et al. [2003b] found masssize distributions from the DRUM sampler were smallerthan any other sizing instrument from PRIDE and suspectparticle bounce-off and break up to be the cause. In thismanuscript we combine stages 1–4 and 5–8 to comparetotal fine- and coarse-mode particle mass. The stage 4–5split was chosen as stages 1–4 contain 90% of dust mass.These issues are discussed in section 3 and by Reid et al.[2003b].

2.3. Aircraft Sample Single-Particle Analysis

[13] The aircraft samples underwent individual particleanalysis at the University of California, Davis MaterialsScience Department microscopy lab using a FEI XL-30sFEG scanning electron microscope (SEM). An EDAXPhoenix Energy Dispersive Spectrometer (EDS) systemcollected the X-ray spectra. The samples were prepared byremoving a portion of each total filter and mounting themon aluminum stubs. The mounted stubs were then carboncoated with 30 nm of carbon and previewed to determineoptimal settings for automated analysis.[14] The samples next underwent automated analysis in

an FEI SL-30sFEG SEM with an EDAX Phoenix EDSdetector with an ultrathin window to derive semiquantitativeelemental concentrations for Na through Cu. The micro-scope operated at 20 keV spot 5 in the EDS imaging modewith a beam size of 5 nm and a beam current of 2.39 nA. Ananalysis grid ranging from 10 � 10 to 20 � 20 was set upfor each sample to ensure that at least 100 separate fieldswould be analyzed and imaged in backscatter mode at 500Xmagnification during automated analysis. All nine samples

were analyzed at 500X magnification, with a minimum sizethreshold for analyzed particles of 1.5 mm average diameter.Three selected filters were also analyzed at 2000X toexamine characteristics of particles with �0.15–3.0 mmaverage diameters. The volume analyzed in the 500Xanalysis was roughly 2–3 cubic microns. The EDAXEDS software used gray scale thresholding of the backscat-ter image to identify particles. Once identified, each particlewas characterized by area, longest axis, perimeter, averagediameter (diameter of a circle having equivalent area),particle centroid, orientation, aspect ratio (longest projec-tion/average diameter), and shape or roundness (area ofcircle with equivalent perimeter/measured area). EDS spotanalysis was performed on the calculated centroid of eachparticle with a dwell time of 10 sec to give adequate signal-to-noise ratios.[15] EDAX Remote Particle/Phase Analysis software was

used to fit reference elemental spectra to the particle spectra.The peak to background counts ratio threshold was set to0.4 for peak identification. Spectral peak values werecorrected using the ZAF matrix correction to generateweight percents of the elements for each analyzed particle.From the weight percents, mole fractions of elements werecalculated.[16] Single-particle analysis by nature has large uncer-

tainties. System errors include uncertainties introduced bythe three-dimensional analysis surface; as with all standard-less quantitation programs, the EDAX program assumes thesamples are flat and that the take-off angle is well described.This is not the case for particles deposited on a substrate, sothere is some systematic error introduced in the measuredelemental quantities, though not in the qualitative presenceof the elements. Further error arises for particles havingsmall diameter or thickness (less than 2 microns), and forloose aggregate particles through which the polycarbonatesubstrate can be imaged.[17] Functional difficulties include biases introduced by

particles nearly touching being counted as single-particles,and by the small beam size of the electron microscopeanalyzing one side of particle in favor of another for largeaggregates (although we do not feel these to be significantuncertainties in this analysis).

3. Results: DRUM Bulk Analysis

[18] Figure 1 shows the time series of key elements fromthe DRUM sampler for 3 July through 24 July. Forcomparison, data are segregated into the upper and lowerstages (stages 1–4, �0.74 mm < dae < 11 mm and stages5–8, �0.09 mm < dae < 0.74 mm, respectively). Reid et al.[2003a] reported that during the PRIDE campaign sixsignificant dust events with aerosol optical depths in excessof 0.3 impacted Puerto Rico. Maximum AOTs for theseevents were roughly centered on 28–30 June and 5, 9, 15,21, and 23 July. The impact of these dust events at thesurface is clearly visible in the coarse-mode aluminum andsilicon data, the two strongest tracers for dust (Figure 1). Inaddition to the significant dust events, additional peaks indust not associated with high optical depth days were alsofound. For example, 13 July showed aluminum and siliconconcentrations in excess of those from the 21 and 23 Julydust events, even though optical depths were less than 0.2.

REID ET. AL.: CHARACTERIZATION OF AFRICAN DUST IN PRIDE PRD 7 - 3

[19] Other key crustal elements in the coarse-mode asso-ciated with dust events are shown in Figure 1b. Peaks insulfur, potassium, calcium and iron were coincident with thealuminum and silicon spikes from these dust events. Whileiron and potassium closely tracked silicon, it is clear that therelative concentration of sulfur and calcium (probablygypsum or anhydrite) is more variable. The 5 and 13 Julyevents showed unusually strong peaks in these elements.[20] As expected, particles in the accumulation mode

(Figure 1c, dae < 0.74 mm) did not significantly correlate

with the passing of the dust events with the exception ofsome enhancement in silicon. Mass concentrations fromsilicon and other key crustal material were roughly 10%of that from the first four stages. Submicron potassium,a strong indicator for biomass burning, was typically<0.1 mg m�3 and did not vary during the study. The behaviorof sulfur, however, was more complicated. Submicron sulfurconcentration exceeded values in the coarse mode. Submi-cron sulfur also tracked the submicron silicon from individ-ual dust events, such as before 10 and 23–24 July. However,additional strong peaks not associated with dust are clearlyevident on 14 July and after the dust maximum on 21 July.Much of this anomalous sulfur is on the smallest stages, andis not seen by EDS single-particle analysis of the aircraftfilter samples due to the 0.15 mm minimum particle sizethreshold for analysis.[21] We can examine the relationship between the key

crustal elements through a series of regressions presented inFigure 2. Here we plot the five elements with the highestconcentrations versus silicon. Silicon was selected as ourprime tracer for dust as the African dust primarily consistsof silicates, and because the DRUM data analytical uncer-tainties are lower for silicon than for lighter elements such asaluminum. Data points only include data for dae > 0.74 mm.As one would expect, after examining Figure 1, aluminum,potassium, and iron had very strong correlations withsilicon, with r2 values in excess of 0.9. Calcium had onlymoderately strong association with silicon with an r2 valueof 0.53. Sulfur had the least association with silicon with anr2 value of only 0.25. However, sulfur was much bettercorrelated with calcium (Figure 2f; r2 = 0.75), againsuggesting an independent gypsum or anhydrite (CaSO4)component to the dust in addition to the silicates and clays.Strong correlations also existed between silicon and severaltrace elements such as magnesium, titanium, chromium,manganese, and cobalt (Figure 3).[22] Regression statistics for Figures 2 and 3 can be found

in Table 2. Included are the mass ratios (i.e., the slopes fromthe regressions), and computed molar ratios relative tosilicon. These regressions indicated that for the most part,the chemical composition of the dust was static during thefield campaign. The exception of the enrichment of gypsumor anhydrite did not appear to affect the elemental ratios ofother species. Stoichiometry is reasonable for clay mineralsfrom Africa. The roughly 2:1 molar ratio between siliconand aluminum is somewhat consistent with the observationthat African dust is primarily illite [K0.6(H3O)0.4Al1.3Mg0.3Fe0.1

2+ Si3.5O10(OH)2 � (H2O)] which has a molar ratio onaverage of �2.7:1. Magnesium and iron are also consistentwith the standard model. Other lesser species have molarratios that are less consistent with illite, suggesting lowconcentration enrichment by other individual minerals inthe dust. Potassium with a molar ratio to silicon of 1:17, issignificantly depleted from the standard model with 1:6.Such deviations are indeed expected in bulk analysis, asdust is a heterogeneous mix of hundreds of individualspecies and variations.[23] After dust, sea salt was the second largest component

to aerosol particle mass at Cabras Island during PRIDE.Regressions between sodium, chlorine, and silicon (as adust tracer) are presented in Figure 4. Also, in Figure 4 is atime series of sodium and chlorine similar to Figure 1 for

Figure 1. Time series of elemental concentrations fromthe DRUM sampler at Cabras Island during the PRIDEcampaign. (a) Aluminum and silicon concentrations forDRUM stages 1–4 (dae > 0.74 mm). (b) Sulfur, potassium,calcium, and iron concentrations for DRUM stages 1–4(dae > 0.74 mm). (c) Silicon, sulfur, and potassium, forDRUM stages 5–8 (dae < 0.74 mm). Periods when the DRUMsampler was not operating are blacked out.


the PRIDE study period. Sodium and chlorine values wererelatively constant throughout the mission with mean valuesof 0.9 and 5.2 mg m�3, respectively. Spikes in these valueswere occasionally seen but were not associated with highwinds (perhaps instead these are due to a local breakingwave event).[24] As expected, sodium and chloride are highly corre-

lated, and sea salt was an independent species from dust. Nocorrelations were found between salt and any dust or sulfurrelated species. Because of the overwhelming presence ofsea salt, any trace values of sodium or chloride associatedwith dust would not be detectable. On the basis of all datapoints, the sodium to chlorine regression suggests a NaClmass ratio of 1:1.3 (Figure 4c). This would suggest a molarratio of 1 to 0.84, a reasonable value for sea salt (chlorine isnominally depleted due to photochemical reactions [Keeneet al., 1998, 1999]. However, this regression is stronglyaffected by a single data point. Removal of this data pointincreases the mass slope to 1:1.9, or a molar ratio of 1 to1.24 (Figure 4d). This curvature in the NaCl regressionsuggests that the XRF analysis systematically underesti-mated sodium at lower concentrations. As sodium is thelowest energy peak of the spectrum to be analyzed, highminimum detectable limits and uncertainties are not entirelyunexpected.

[25] As gravimetric analysis cannot be performed onDRUM strips, the total dust mass concentration must beestimated based on the available elemental data. The esti-mated mass percentage of the elements in bulk dust ispresented in Table 2 by assuming aluminum is 8% of totalbulk dust mass (this aluminum to mass ratio is commonlyused in analytical soil studies [Taylor and McLennan, 1995;Maring et al., 2000]). Figure 5 presents estimated massconcentrations from dust, sea salt, and anthropogenics.These are based on the assumption that aluminum is 8%of total dust mass and that chlorine is 58% of the total sea-salt mass. We assume that submicron sulfur is in the form ofammonium sulfate. A �15% correction based on a siliconrelation is included to account for any submicron sulfurassociated with dust (although it is very possible that someof the correlation between submicron sulfur and dust isbecause of anthropogenic sulfate being advected at the sametime as the dust. However, with only a �15% correction,this error is not large).[26] From Figure 5a we find that dust is for the most part

the dominant aerosol species at Cabras Island. Submicronsulfate concentrations were typically under 2 mg m�3.However, reconstructed mass from the DRUM samplersignificantly underestimated the total aerosol particle massconcentration as determined by the University of MiamiTapered Element Oscillating Microbalance (TEOM) colo-cated at Cabras Island. While the DRUM follows theTEOM for the episodic dust events, on a whole it under-estimates total mass concentrations from 10–50%. Much of

Figure 2. Elemental mass ratio regressions for the sum ofDRUM stages 1–4 (dae > 0.74 mm) for significant elementalspecies. (a) Aluminum versus silicon. (b) Sulfur versussilicon, (c) Potassium versus silicon. (d) Calcium versussilicon. (e) Iron versus silicon. (f) Calcium versus sulfur.

Figure 3. Same as Figure 2 but for trace elements versussilicon. (a) Magnesium, (b) titanium, (c) chromium, (d)manganese, (e) cobalt, and (f) zinc.


the noise in the DRUM reconstructed mass lies in the seasalt. Typically, chlorine is a poor choice as an indicatorspecies for sea salt as its mass ratio varies considerably withage and non-sea-salt sulfate concentration [Keene et al.,1999]. However, as we know that sodium is underestimatedin the DRUM analysis, we are left with little choice in thiscomparison but to use chlorine. However, we know fromReid et al. [2002] and Savoie et al. (submitted manuscript,2003) that the daily average dry sea-salt concentration inPRIDE varied between 14–22 mg m�3, with a mean valueof �18 mg m�3. In Figure 5b we compared the DRUMreconstructed dust mass to the TEOM data minus anaverage value of sea salt of 18 mg m�3. In this case theDRUM and TEOM track considerably better. However,peak dust concentrations are still underestimated in theDRUM sampler. Integrating through each event, on averagethe DRUM is only reconstructing 60% of the dust mass.[27] There are several possible reasons why the DRUM is

underestimating dust mass. First, we can question whetheraluminum is not 8% of dust particle mass, but rather 5%.Such a decrease would more than make up the difference inmass. However, as will be shown in the discussion section,

Table 2. Bulk Elemental Mass and Molar Ratios Versus Silicon

for DRUM Data Where the Aerodynamic Diameter is >0.74a

Elements Mass Ratio Molar Ratio r2 Percent of Mass

Al 0.47 0.49 0.90 h8iMg 0.057 0.066 0.47 1.0Si 1 1 – 17.1S 0.1 0.088 0.25 1.7K 0.082 0.059 0.92 1.4Ca 0.18 0.13 0.53 3.1Ti 0.021 0.012 0.81 0.36Cr 0.0004 0.0002 0.59 0.007Mn 0.0027 0.0013 0.74 0.046Fe 0.14 0.070 0.92 2.4Co 0.0005 0.0002 0.67 0.008Zn 0.0004 0.0002 0.14 0.007Total 35.1%aThese are derived from the scatterplots in Figures 2 and 3. Also shown is

the r2 for the regression and the estimated percentage of total dust massattributed to each element assuming aluminum is 8% of dust total mass.

Figure 4. Regression of (a) sodium to silicon, (b) chlorineto silicon, (c) chlorine to sodium, (d) same as Figure 4c butwith a scale change, and (e) reconstructed mass estimatesfor chlorine and sodium.

Figure 5. Reconstructed mass estimates for the DRUMsampler. For comparison, measurements from the Univer-sity of Miami TEOM are also presented. (a) Reconstructedmass for dust, sea salt, and submicron sulfate and massconcentration from the University of Miami TEOM.(b) Reconstructed mass for dust. TEOM mass concentrationestimates for dust where the mean 18 mg m�3 sea-salt massconcentration was subtracted from the total value.


stoichiometry prohibits such a drastic change. Second, wecan assume that there is some error in the chemical analysisand that the aluminum mass concentration is underesti-mated. We know sodium is heavily underestimated, andaluminum is the next element in the X-ray spectrum. Butagain, such a shift in the silicon-aluminum-iron ratio isunreasonable. After the analysis of Reid et al. [2003a], wesuggest that inlet issues may be to blame. With a flow rateof 10 L min�1, it is quite possible that at the 7 m s�1 meanwind speeds, large dust and salt particles could not make thebend in the simple cap inlet. The colocated MOUDI samplerappears to have suffered a similar problem. However, forthe purposes of determining bulk chemical ratios in dust,this should not be a significant artifact.

4. Results: Single-Particle Analysis Results

[28] While the DRUM sampler analysis can give usreasonable results for bulk composition, we need to go tosingle-particle analysis to interpret the results with respectto mineralogy and optical properties. For example, howmany mineralogical species are influencing the found alu-minum to silicon ratio, and is illite a reasonable model? Are

the dust particles aggregates or single particles? Doesparticle chemistry covary with size or morphology? In thefollowing sections we explore these questions.

4.1. Single-Particle Morphology and Size

[29] As dust is heterogeneous mix of particles of variousmorphologies, no standard set of size statistics can be easilyapplied. Examples of secondary electron micrographs fordust collected in the SAL on 21 July 2000 are presented inFigure 6. Four magnifications are shown, depicting particlesin 10–20, 5–10, and <5 mm size ranges.[30] Data from this flight was typical for the study. Dust

is predominately in the form of large, amorphous alumino-silicate clay particles. These can be relatively large, withparticles as large as 30 microns being detected on the filters.Particles can either be free or included in some form ofaggregate. Evidence of aggregation can be found by thepresence of composite particles that that broke apart onimpact.[31] More detailed images of single particles can be found

in Figure 7 (images of large aggregate particles are given byReid et al. [2003a]). Particles generally fit into three broadcategories. Most prevalent were layered silicates (Al-Si clay

Figure 6. Secondary electron images of dust particles collected in the Saharan Air Layer near PuertoRico on 21 July 2000. Because some size segregation occurs on the filter substrate, size and shapedistributions from individual images are not representative of the dust as a whole. (a) 1000X, (b) 2000X,(c) 4000X, and (d) 12000X.


minerals or feldspars), which qualitatively account for�70% of all particles. Dominant species are consistent withillite, kaolinite, and montmorillonite. Examples includeFigures 7a (left-hand particle), 7b, 7f, 7g, 7h, and 7j.The second most prevalent (�20%) are the amorphoussilicates, some of which appear to be agglomerates of clayparticles and are depicted in Figures 7d, 7e and 7i. Finallythere are non silicon trace species such as gypsum andcalcium carbonate which account for <10% of all particles(Figures 7a, right-hand side, and 7c). In general, particleswere found in some form of aggregate (�50% of allparticles, 70% of particles >3 mm in diameter). Larger clayminerals were usually found to be carrying smaller particles(e.g., Figures 7a, left side, and 7h). More typically, dustparticles less than 10 mm in diameter were aggregated inclusters of 2 to 5 components.[32] As is discussed later in this paper, relating these size

statistics to quantities usable in particle models is notstraightforward. In this study three principle particle meas-urements were recorded for each particle from the electronbackscatter image. These are cross-sectional area (A),particle perimeter (P), and longest projection (LProj). Fromthese measurements, statistics relating to particle diameterand area can be generated. These include computed averagediameter (diameter of a circle with equivalent area), andminor axis (using an ellipse model; assigning the longestprojection as the major axis and preserving the particlecross-sectional area to derive a normal minor axis).[33] Computing particle volume is more uncertain as it

requires some estimate of particle thickness (a quantity thatusually is not measured). Previously, researchers have usedthe minimum diameter [e.g., Anderson et al., 1996] or afraction of the minor projection [Okada et al., 2001] as thebest estimate of particle height, and modeled the particles asoblate spheroids. As particles tend to lie flat on thesubstrate, these values should be considered maximumheights. For example, Okada et al. [2001] measured particleheight for Asian dust particles, and determined 1/2 to 1/4the orthogonal diameter to be a better estimate for particle

height for mineral aerosols in the 0.1 to 6 micron diameterrange. However, if we are interested in the volume distri-bution normalized to total volume, as long as there is aconsistent trend in the height-width aspect ratio with sizethese factors are relatively insignificant. As will be shown,particle aspect ratios do not appear to change with size, sothis is a reasonable assumption.[34] Particle area and volume distributions are presented

in Figure 8. Distributions are normalized by total area andvolume for particles with average diameter >3mm. Here wehave grouped size spectra for the two filters collected on 16July (Figures 8a and 8b), the two filters collected on 22 July(Figures 8c and 8d), and all remaining filter samples(Figures 8e and 8f). All particle distributions share somecharacteristics. Area distributions are dominated by a log-normal peak with area median diameter (AMD) on the orderof 5 to 7 mm and geometric standard deviations of 1.65–1.80. When higher resolution data are available, we consis-tently find that there is a shoulder that extends to �0.7 mmin diameter. Below this point dust particles do not haveappreciable surface area. Estimated volume distributions aremore lognormal, with volume median diameters (VMD)from 7 to 9 mm, and geometric standard deviations on theorder of 1.6 to 1.8.[35] As was reported in other PRIDE manuscripts [Mar-

ing, 2001; Reid et al., 2003a, 2003b] measured particle sizedistributions were relatively static during the campaign andmostly did not vary with altitude. This can be most clearlyseen in the higher moment distributions such as volume. Forthe 16 July case (at the tail of fairly strong dust event inwhich mid-visible optical depths were maximum of 0.5,0.25 during collection), single-particle analysis dust par-ticles collected in the MBL (altitudes <500 m) were onlyslightly larger than those at 3000 m in the SAL (VMD of6.8 versus 6.4 mm, respectively). Similarly, for the 22 Julycase two samples were collected. Particles collected in aregion at the bottom of the SAL and in the convectiveboundary layer (800–2700 m) were only slightly larger thanthose collected toward the top (2700–5000 m), with VMDsat 6 and 7 mm respectively. Hence, while some enhancementof larger particles can be seen at lower levels (presumablydue to settling), significant size separation is for the mostpart not evident. The one exception to this finding is thecase of 5 July. As described by Reid et al. [2003a], the 5July case was the one case where strong gradients were seenin particle size with altitude. While only one integrated dustsample was collected on the 5 July flight, we can see thatthe size distribution was like no other with wider distribu-tions and the largest VMD of the study at 10 mm.[36] Dust particle morphology statistics were also calcu-

lated from the electron backscatter images, and verified byhand calculations for some particles. Two principle quanti-ties, the aspect ratio, (Aspect = p(LProj)2(4A)�1), and shapefactor (Shape = P2(4pA)�1) are commonly reported in theliterature. The aspect ratio provides a basic approximationof particle roundness and is roughly equivalent to the ratioof major to minor axes of the ellipsoid best fit to theparticle. The shape factor, the inverse of particle circularity,is a dimensionless parameter used as an indicator of thecomplexity of the particle. As the ratio of the square of theperimeter to particle area, the shape factor describes how‘‘jagged’’ a particle is, by definition equal to 1 for a circle.

Figure 7. Detailed electron micrographs of individualparticles.


[37] Figure 9 presents cumulative probability plots ofparticle aspect ratio and shape factor as a function of particleaverage diameter. These plots were derived from the aircraftsampled particles measured during the study (median valuesof the aspect ratio and shape factors did not significantlyvary from sample to sample). As is expected for dustparticles, aspect ratios are relatively large and broad based.Median aspect ratios averaged 1.9 with a standard deviationof 0.9. Removal of NaCl-rich particles from the analysis did

not significantly change the distributions or the medianaspect ratios. The aspect ratio distributions for the dustappear to be independent of size, and the curves do notstatistically differ from one another for sizes less 10 mm.The largest particles (dp > 10 mm, representing roughly 3%of the analyzed particles) are slightly more elongated withmedian aspect ratio values of 2.2 ± 1.2.[38] In contrast to the aspect ratio, shape factors do

appear to have a strong dependence on size (Figure 9b).

Figure 8. Particle normalized cross-sectional area and volume distributions for particles collected onselected aircraft samples. Specified in the key are sample date at altitude region. MBL, marine boundarylayer; CBL, convective boundary layer; SAL, Saharan Air Layer.


Smaller particles have shape factors near unity while thelargest particles have a median shape factor of 3. However,for most particles in the 2 to 10 mm range (where most ofthe area and volume exists) values are a consistent 1.4.This trend is not unexpected as these are individualminerals. This trend is seen in Asian dust by Okada etal. [2001], and is cited as evidence of increased particlecomplexity. With increased particle size, there is a higherprobability that the particles are aggregates, and hence asubstantial increase in perimeter relative to area. This is

compounded by the possibility of aggregates flatting onimpact with the substrate.

4.2. Semiquantitative Single-ParticleElemental Analysis

[39] While the XRF analysis on DRUM sampling stripsprovides bulk elemental ratios from which dust chemicalstoichiometry can be performed, EDX analysis provides usinformation on the heterogeneity of individual particles.Particles larger than �1.5 mm were individually analyzedfor elemental ratios and for morphology. Three of thesefilters were also analyzed for particles in the range of 0.15 to3.0 mm diameter. From these results, elemental percentagesand molar percentages (relative to analyzed elements) weregenerated for each particle.[40] As an example of African dust, electron backscatter

images of particles collected in the SAL for the 21 Julyevent are presented in Figure 10 for two size ranges.Associated ternary plots for the two images are also shown.Given are Si-Na-Cl, to differentiate dust from sea salt, Si-Al-Fe, Si-Al-Ca and Si-Al-Mg, to differentiate various clayminerals from amorphous silicon, Si-Na-Mg to differentiateminerals such as feldspar and montmorillonite, and S-Na-Cato differentiate gypsum. Strong outliers from the analysisare individually labeled. Examination of Figure 10 revealsthe basic chemical nature of the dust.[41] Predominately, dust is made of alumino-silicates and

amorphous silicates with the remainder being trace amountsof gypsum, calcium carbonates and other species. Since thissample was taken in the SAL, sodium chloride from sea saltis almost nonexistent. A very strong alumino-silicate pop-ulation is the most striking feature of these plots with a massratio between Al and Si of �0.5. Iron, the principle lightabsorber for dust, was found in >90% of such particles atrelatively low concentrations (�0.08 mass ratio to silicon).Similarly, Mg, Ca and Na (not associated with Cl), otherkey tracer elements for clay minerals, were also stronglyand expectedly associated with aluminum-silicates species.A second population of more pure silicates is also visible.For these particles, while some Al is found, other traceelements such as Mg, Ca, and Na are not present insignificant quantities. Finally, a few outlier species arepresent. Since they are so few, they do not appear clearlyin the ternary plots but are labeled in Figure 10. Theseinclude; calcium enriched silicate particles (Ca-AlSi; prob-ably clay particles aggregated or layered with calciumcarbonate), calcium enriched particles (Ca; probably calci-um carbonate particles with some clay particle contamina-tion), gypsum (CaSO4), quartz (SiO2), iron-rich particles(Fe rich; probably clay aggregated with hematite), andtitanium enriched silicate particles (Ti-AlSi; probably TiO2

aggregated with clay particles).[42] While Figure 10 can give us a strong qualitative feel

for the chemical nature of the dust, more quantitativedescriptions require a thorough cluster analysis. Sevenaircraft samples were subjected to unsupervised clusteranalysis to group the particles into distinct types. Clusterswere identified by average compositions, consisting ofmolar percentages of the elements analyzed. Species suchas oxygen were not included in derivation of the molarpercentages. Because many of the particles are aggregatesand may contain conversion products, each cluster should

Figure 9. Cumulative probability plots of (a) aspect ratiosand (b) shape for variously sized particles collected onaircraft samples. Data from all analyzed aircraft samples areincluded in these plots.


be taken to represent a range of species and not a quantifiedcomposition.[43] Table 3 displays the dominant clusters and groups

derived from all filter data. Table 4 presents relative clusterscores for the samples analyzed and is discussed further insection 5. Normalized cross-sectional area distributions forthese clusters are presented in Figure 11. The clusters fall

roughly into 12 compositional groups. These groups werechosen not simply due to chemical composition ratios, butrather how the cluster scores tracked from event to event.Aluminum silicates account for 59% of the particles bynumber, and roughly 67% of the particle mass. Eightpercent of the particles and 17% of the mass belong inthe high Si group, which includes SiO2. Most of the

Figure 10. Production electron backscatter images taken from the 21 July 2000 Saharan Air Layer filter.Shown are (a) course-resolution and (b) fine-resolution images with their associated ternary plots of keyelements. More atypical particle types are labeled immediately to the right of the particle or by arrow.


Table 3. Cluster Analysis Results From an Ensemble of All Filters Collecteda

ParticleType Group

Abundance

Cluster Cluster Average CompositionPercent byNumber

Percentby Mass

Al-Si AlSi-1 3.9 14.4 C1 Si.46 Al.34 Mg.05 Na.04 Fe.04Al-Si AlSi-1 4.1 6 C2 Si.46 Al.27 Mg.08 Na.06 Fe.05Al-Si AlSi-1 5.8 11.4 C3 Si.49 Al.24 Mg.09 Na.05 Fe.04Al-Si AlSi-1 3.2 15.8 C4 Si.48 Al.30 Mg.06 Fe.04 Na.04Al-Si AlSi-1 2.6 2.7 C5 Si.41 Al.33 Na.07 Mg.06 Fe.04Al-Si AlSi-1 4.6 1.8 C7 Si.40 Al.26 Mg.09 Na.08 Fe.04Al-Si AlSi-1 4 1.9 C8 Si.41 Al.24 Na.12 Mg.09 Fe.04Al-Si AlSi-1 1.1 2.6 C16 Si.40 Al.23 Mg.17 Na.06 Fe.06Al-Si AlSi-1 2.1 0.8 C18 Si.48 Al.19 Na.08 Mg.09 Fe.04Al-Si AlSi-2 0.3 1.1 C23 Si.35 Al.23 Na.18 Cl.10 Mg.06 Fe.03Al-Si AlSi-2 2 0.7 C29 Si.32 Al.20 Na.19 Mg.10 Cl.05 Ca.04 Fe.03Al-Si AlSi-2 0.4 0.5 C34 Na.27 Si.26 Al.18 Cl.13 Mg.06Al-Si AlSi-2 0.2 0.6 C38 Na.34 Si.22 Cl.18 Al.14 Mg.05Al-Si AlSi-3(6 5.5 0.7 C13 Si.38 Al.20 Mg.12 Na.11 P.04 Fe.03 S.03Al-Si AlSi-3(6 7.1 0.4 C15 Si.31 Al.21 Na.13 Mg.12 P.05 S.04 Fe.03Al-Si AlSi-3(6 4.6 0.7 C17 Si.34 Al.28 Na.10 Mg.09 P.04 S.03Al-Si AlSi-3(6 2.7 0.5 C28 Si.22 Al.18 Na.17 Mg.15 P.06 S.06 Cl.04 Ca.04Al-Si AlSi-4(7 0.5 0.9 C31 Si.39 Al.23 Ca.12 Mg.08 Na.05 Fe.04Al-Si AlSi-5(8 0.2 0.5 C41 Si.38 Al.25 Fe.18 Mg.06 Na.05Al-Si AlSi-5(8 0.4 0.2 C42 Si.30 Al.24 Fe.23 Mg.07 Na.07Subtotal 55.3 65.4Si rich Si-1 1.6 3.5 C6 Si.79 Al.08 Na.04 Mg.03Si rich Si-1 1.8 1.7 C11 Si.68 Al.12 Na.05 Mg.05Si rich Si-1 0.8 4.1 C12 Si.82 Al.06Si rich Si-2 2.3 0.5 C24 Si.60 Al.11 Na.09 Mg.07 P.03Si rich Si-3 0.9 3.3 C25 Si.54 Al.23 Na.05 Mg.05 K.03Si rich Si-3 1.3 4.3 C9 Si.55 Al.22 K.07 Mg.05 Na.04Subtotal 8.7 17.4Ca rich Ca-1 0.3 0.3 C49 Ca.64 Na.08 Mg.07 Si.07 Al.05Ca rich Ca-2 1.1 0.8 C19 Ca.43 Si.18 Al.12 Mg.10 Na.09Ca rich Ca-2 0.7 0.9 C20 Ca.57 Si.11 Mg.08 Al.08 Na.07Ca rich Ca-3 2.1 0.1 C37 Ca.27 Na.18 Mg.14 Si.13 Al.11 P.04 S.04 Cl.03Ca rich Ca-3 2.4 0.1 C40 Ca.39 Na.20 Mg.15 Si.11 Al.07Subtotal 6.6 2.2Fine aggregates Agg-1 5.3 1.4 C14 Na.23 Mg.18 Al.14 Si.13 P.08 S.07 Cl.05 Ca.03 K.03Fine aggregates Agg-1 0.9 0.4 C45 Si.20 Al.15 Mg.12 Na.11 S.10 P.09 Cl.07 Ca.06 K.04 Sn.04 Ti.03Fine aggregates Agg-1 0.3 0 C56 Si.24 Al.12 S.10 Cl.10 Ca.10 K.09 Sn.07 Ti.06 Fe.04Fine aggregates Agg-1 0.3 0 C57 Si.28 Fe.25 Ca.10 K.10 Al.09 Ti.06 Cl.05 Sn.05 S.03Fine aggregates Agg-1 0.2 0 C58 Si.18 Ca.17 K.12 Ti.11 Fe.10 Cl.08 S.07 Al.05Fine aggregates Agg-1 0.2 0.1 C60 Si.39 Fe.12 K.10 Al.14 Ca.07 S.04 Cl.04 Ti.03 Sn.03Fine aggregates Agg-1 0.2 0 C65 Si.33 Al.12 K.08 Ca.08 S.08 Cl.07 Fe.06 P.06 Sn.05 Ti.05Fine aggregates Agg-2 1 0.4 C39 Na.25 Si.20 Al.15 Mg.12 Ca.07 Cl.07 S.06 P.03Fine aggregates Agg-2 0.7 0.3 C46 Na.31 S.16 Mg.11 Ca.10 Si.09 Al.08 Cl.08 P.04Subtotal 9.1 2.6NaCl rich NaCl 1.4 0.6 C30 Na.55 Cl.34 Mg.04NaCl rich NaCl 2 0.5 C32 Na.51 Cl.29 Si.06 Al.05 Mg.04NaCl rich NaCl 0.8 3 C35 Na.42 Cl.24 Si.13 Al.09 Mg.05NaCl rich NaCl 0.8 0.4 C43 Na.40 Cl.16 S.10 Si.07 Mg.07 Ca.07 Al.07NaCl rich NaCl 0.1 0.4 C53 Na.48 Cl.38 Si.03 Mg.03 Al.03Subtotal 5.1 4.9Ca, S rich CaS-1 4.4 0.1 C21 S.22 Ca.17 Na.16 Si.11 Mg.11 Al.10 P.05 Cl.03Ca, S rich CaS-2 0.7 0.9 C26 S.35 Ca.30 Si.08 Na.08 Al.06 Mg.05Ca, S rich CaS-2 0.2 0.3 C44 Si.24 S.21 Cl.18 Al.14 Na.08 Mg.07Subtotal 5.3 1.3Mg silicates MgSi-1 0.4 1.1 C33 Si.43 Mg.26 Al.14 Na.05 Fe.04Mg silicates MgSi-2 0.8 0.4 C36 Si.31 Mg.25 Al.20 Na.07 Fe.05Subtotal 1.2 1.5Ca silicates CaSi-1 1.3 0.6 C22 Si.29 Ca.24 Al.17 Mg.10 Na.09Ca silicates CaSi-1 0.2 0.3 C52 Si.34 Ca.26 Al.14 Mg.08 Na.06 Fe.04Subtotal 1.5 0.9Fe rich Fe-1 0.2 0.2 C48 Fe.41 Si.20 Al.17 Na.06 Mg.06Fe rich Fe-1 0.1 0.2 C66 Fe.58 Si.13 Al.08 Mg.04 Na.04Fe rich Fe-2 0.7 <0.1 C51 Fe.32 Na.15 Si.15 Al.13 Mg.09 S 06 P.03Subtotal 1.0 0.4Mg, Ca rich MgCa-1 0.1 0.2 C47 Mg.46 Ca.27 Si.09 Al.07 Na.05Mg, Ca rich MgCa-1 0.1 0.1 C50 Mg.36 Ca.20 Fe.18 Al.12 Na.07Subtotal 0.2 0.3Ti rich Ti-1 0.2 <0.1 C54 Ti.39 Si.19 Al.14 Na.09 Mg.07 Fe.04Ti rich Ti-2 0.1 0.1 C55 Si.31 Ti.22 Al.21 Mg.07 Na.06 Fe.05Subtotal 0.3 0.1Na rich Na-1 0.5 <0.1 C59 Na.43 Si.30 Al.15 Mg.05


remaining particles fall into the following 8 groups (anno-tated with percentage by count and percentage by massrespectively): calcium-rich particles excluding gypsum andanhydrite (6%, 2%), complex fine-mode aggregates (9%,3%), NaCl-rich particles (5%, 5%), sulfur- and calcium-richparticles like gypsum or anhydrite (5%, 1%), magnesiumsilicates (1%, 2%), calcium silicates (1%, 1%), iron-richparticles (0.2%, 0.4%), magnesium and calcium-rich par-ticles (0.5%, 0.4%), titanium-rich particles (0.3%, 0.1%),and sodium-rich particles excluding NaCl (0.7%, 0.1%). Wediscuss each group under the following headers.4.2.1. Alumino-Silicates Group[44] AlSi, contains the most common particle type seen in

the PRIDE dust samples, comprising roughly 55% and 65%of particle number and mass, respectively. There is someseparation with size, with 75% of the coarse-particle mass inthe SAL samples, and 60% of the fine particles. This groupincludes particles for which Al and Si are the predominantcomponents of the analyzed composition. Mineralogically,we expect them to be dominated by clay minerals such asillite, kaolinite, chlorite, and montmorillonite. Because theyare elementally similar to clay minerals, species such asfeldspars and micas will also fall into this group. As theseparticles make up the bulk of the dust, the area distributionfor this cluster (Figure 11a) is nearly identical to that givenin Figure 8 with an area modal diameter of 5 mm.[45] To simplify presentation of the AlSi data, they are

sorted into five groups that share common composition and/or occurrence. As most particles are aggregates, and evenindividual clay minerals vary considerably, they cannot beviewed as individual species. Rather, they should be viewedsimply as natural groupings of particles with similar ele-mental ratios. These groups are described below.[46] 1. AlSi-1 can be interpreted as clay mineral species

and accounts for 30% of all analyzed particles and roughly

60% of the total particle mass. Not surprisingly, The groupwas strongly associated with samples collected in the SALdust event (e.g., 21 and 24 July). Si and Al ranged from 40–49% and 19–33% of detectable molar percentages, respec-tively. Na and Mg molar percentages range from 4–9%,with Fe comprising 4–8% of the analyzed composition.Approximately 80% of Fe, the principal light absorber fordust, is associated with these clay particles. This clustershould be viewed as a continuum of clay minerals ratherthan an individual species. Stoichiometrically, the averageSi to Al ratio of 2:1 is consistent with illite, and excludessuch species as kaolinite, having a ratio of 1:1. However,based on individual samples and our ternary analysis,kaolinite is clearly present in the samples (�20% of theseparticles). However, because of the aggregate nature of mostparticles they do not appear in an unsupervised clusteranalysis but are rather grouped with other species. This isdemonstrated in Figure 12a, which shows the aluminum tosilicon ratio distributions for all silicon-rich clusters. Forcomparison, Figure 12b shows the aluminum to silicon ratiofor all analyzed aircraft samples.[47] 2. AlSi -2, representing 3% of the sample number

and mass, is enhanced in both Na and Cl. These likelyrepresent dust particles of similar composition as AlSi-1 thathave interacted with sea salt during transport. Not surpris-ingly, this cluster is more pronounced for samples taken inthe marine boundary layer. This group is primarily a coarse-mode feature, accounting for 10% of the 5 July coarseparticles, 13% and 25% of the 16 and 24 July MBL coarseparticles, respectively.[48] 3. AlSi -3, representing 20% of the sample number,

but only 2.3% of the mass, has lower Si concentrations, andsignificant levels of P, S, and Cl. This group accounts for30% of the fine particles on 5 July, and 35% of the fineparticles from the 16 July SAL sample. As there are

Table 3. (continued)

ParticleType Group

Abundance

Cluster Cluster Average CompositionPercent byNumber

Percentby Mass

Na rich Na-1 0.2 0.1 C62 Si.36 Na.31 Al.12 Cl.07 Mg.06Subtotal 0.7 0.1

aCluster average compositions are listed as molar percentages. Cluster species with <3% molar fractions are not listed in the composition. Clusters arelisted by qualitative particle type, followed by group and individual cluster.

Table 4. Summary of Particle Grouping

5 July Integrated 16 July SAL

16 July MBL>3 mm

20 July SAL>3 mm

21 July SAL 22 July SAL

24 July MBL>3 mm<3 mm >3 mm <3 mm >3 mm <3 mm >3 mm

Sample 1>3 mm,

Sample 2>3 mm,

Alumino-silicates 40.5 57.2 62.3 74.2 76.3 73.9 61.8 66.5 74.3 71.1 58.1Si rich 3.5 8.4 7.4 13.3 12.7 16.0 9.7 18.5 15.5 17.5 11.3Ca rich 3.3 5.2 2.4 3.4 1.6 3.7 16.7 4.9 3.9 3.4 0.4Fine aggregates 7.9 4.0 16.5 0.9 1.9 0.5 1.5 1.9 0.2 2.1 3.1NaCl rich 22.3 17.0 0.3 0.5 2.9 0.0 0.1 0.2 0.1 0.1 21.9Ca and S rich 18.0 3.8 3.1 1.2 1.4 1.0 0.1 1.5 0.9 1.3 1.5Mg Silicates 0.6 1.2 1.7 1.7 0.9 1.4 0.6 1.6 1.1 1.1 1.4Ca Silicates 0.1 0.4 1.2 1.1 0.5 1.5 4.2 1.9 1.6 1.5 0.1Fe rich 2.5 0.4 0.8 0.6 0.0 0.3 0.2 0.5 0.5 0.5 0.4Mg and Ca rich 0.1 0.3 0.3 0.5 0.3 0.7 0.1 0.9 0.7 0.6 0.1Ti rich 0.3 0.1 0.4 0.3 0.5 0.3 0.4 0.5 0.4 0.4 0.2Na rich 0 0.2 0 0 0.3 0 3.1 0 0 0 0.6


phosphate mines in southwestern Mauritania (a sourceregion), it is possible that these particles represent fine dustfrom that region. This group is not a significant presenceafter 16 July, indicating that the later samples may representdifferent transport or conversion regimes.[49] 4. AlSi -4 includes calcium-rich Al silicates, and

accounts for 0.5% of the analyzed particles. This group isabsent in the MBL and integrated samples, and poorlyrepresented in the 16 July fine and coarse SAL samples.However, it accounts for �1% of the particles in the later

SAL samples, suggesting that the absence of these particlesin the MBL and integrated samples may be due to conver-sion of Ca during transport to anhydrite.[50] 5. AlSi-5 contains iron-rich Al silicates with aver-

age composition of 18–23% Fe. This group is distributedthroughout the dusty layers, but only accounts for 0.5% ofthe samples. This group may simply represent iron-rich silicate minerals, or may indicate Fe contaminationof silicate minerals, as suggested by Falkovich et al.[2001].

Figure 11. Particle normalized size distribution plots for cluster analysis groupings; (a) Al- and Si-richand Mg- and Si-rich clusters, (b) Si-rich and Mg- and Ca-rich clusters, (c) Ca- and Si-rich and Ca-richclusters, (d) aggregate and NaCl-rich clusters, (e) Ca- and S-rich and Fe-rich clusters, (f) Ti-rich and Na-rich clusters. Data from all analyzed samples are included in these plots.


4.2.2. Si-Rich Group[51] The Si-rich group is the second largest cluster group,

accounts for roughly 9% of the particles analyzed, and ischaracterized by molar fractions of Si greater than 50%. Forthe individual samples, 13–18% of the coarse SAL par-ticles, and 3–10% of the fine particles belonged to thisgroup. The derived clusters can be grouped into threedistinct subgroups; probable SiO2, fine alumino-silicate,and coarse alumino-silicate particles. Several particles thatappear to be diatom skeletons were seen, especially on the 5July sample, but the SiO2 particles were primarily crystal-line fragments, either alone, or as part of larger aggregates.Like other dust particle types, these too have an area mediandiameter of �5 mm, although with a noticeably smallerstandard deviation.[52] 1. Si-1 has average composition of >68% Si with

less than 10% Al, Mg and Na, and is probably SiO2

particles. With the exception of the 5 July sample, thisgroup accounts for 8–12% of the coarse particles from eachsample, and 3–5% of the fine particles. The 5 July samplecontains only 6% coarse and 1% fine Si-rich particles.[53] 2. Si-2 is primarily fine-mode particles, representing

3% of the fine particles from 5 July, and 4% of the fineparticles from 16 July, but only 1% of the fine SAL particlesfrom 21 July. It has average composition of 60% Si, 11%Al, 9% Na and 7% Mg with some trace elements.[54] 3. Si-3 has an average composition of 55% Si,

23%Al, and 3–7% Na, Mg, and K, indicating that theseare probably silicate minerals such as micas or clays. Thisgroup is nearly absent in the 5 July samples and the fine 16July SAL sample. However, it is relevant in the latersamples, making up 4–8% of the coarse SAL samples.4.2.3. Calcium-Rich Group[55] The calcium-rich group, Ca, excluding probable

gypsum and anhydrite particles, represents about 7% ofthe samples. This group is less prevalent in the MBLsamples from 16 July and 24 July but accounts for 3–5%of the coarse particles in the SAL samples, and 2–3% of thefine particles for 5 July and 16 July. In contrast, this groupaccounts for almost 17% of the 21 July fine SAL sample.The calcium-rich group is the first prevalent cluster to have

a size distribution that deviates from the bulk analysis, withan enrichment of submicron particles. Three subgroupssimplify the analysis of this cluster.[56] 1. Ca-1 is primarily Ca (64%), and accounts for 0.3%

of the analyzed particles. These are likely calcium carbonateparticles.[57] 2. Ca-2 accounts for 1.8% of the particles analyzed,

and ranges in average composition from roughly 40 to 60%Ca, with 10–20% Si, 8–12%Al, 8–10%Mg, and 7–9% Na.This group is primarily comprised of coarse-mode particles.Enrichment in elements like Si and Al are likely due toaggregation with clay particles.[58] 3. Ca-3 is found almost entirely in the fine mode, and

accounts for nearly 5% of the analyzed particles. This groupis Ca-rich, but has Na, Mg, and Si as major elementalcomponents. This group accounts for 12% of the fine SALparticles in the 21 July dust layer sample, 3% of the fineintegrated particles from 5 July, and 1.5% of the fine SALparticles from 16 July.4.2.4. Small-Mode Aggregates Group[59] The small mode aggregates group comprises 9% of

the analyzed particles, but is primarily fine particles. Theseparticles are complex and varied combinations of Na, Mg,Al, Si, P, S, Cl, and K, with some Ti and Sn. Sizedistributions of these particles (Figure 11d) show they arequite clearly unlike other alumino-silcate aggregate par-ticles, having an area median diameter on the order of 2 mm.As these particles appear in the coarse and fine-modesample populations, we do not think it is an artifact ofthe analysis. Two subgroups expedite presentation of thisgroup and were found based on sampling location.[60] 1. Agg-1 is primarily associated with the 16 July dust

layer sample, accounting for 18.7% of the fine 16 July SALparticles, though it has a component (�2%) in the 5 Julyfine- and coarse-mode samples, and in the coarse 22 JulySAL sample aloft.[61] 2. Agg-2 is primarily in the integrated and MBL

samples. In the 5 July sample, it accounts for 5.8% of thefine mode and 2.2% of the coarse-mode particles. It alsorepresents 2.7% of the coarse-mode surface sample from 24July, and 1.6% of the MBL sample from 16 July. This group

Figure 12. Cumulative probability plots of the molar ratio of aluminum to silicon for all analyzedaircraft sample particles. Specified in the key are sample date and altitude region. MBL, marine boundarylayer; CBL, convective boundary layer; SAL, Saharan Air Layer.


is a minor component, accounting for 1.7% of the analyzedparticles.4.2.5. NaCl-Rich Group[62] The sodium and chlorine-rich group, NaCl, repre-

sents 5% of the analyzed particles, and involves integratedand surface sample particles. It includes 22% of the fine and17% of the coarse particles from 5 July, and 22% of thecoarse MBL particles from 24 July. This group makes uponly 3% of the 16 July coarse MBL sample when dust wasthe dominant species near the surface. The size distributionfor these particles is surprisingly similar to the aggregatecluster. This group ranges in composition from 40–55% Na,16–40% Cl, and 3–7% Mg, with some clusters alsocontaining Si, and Al.4.2.6. Ca- and S-Rich Group[63] The CaS group includes roughly 5% of the analyzed

particles, but is primarily associated with the 5 July fineintegrated sample. As discussed in the sections on theDRUM sampler analysis, a Ca and S spike was alsoobserved in the bulk analysis. Particles in this group showthe strongest deviation in size from the bulk dust, with aprominent fine-mode component. Two subgroups best char-acterize these particles.[64] 1. CaS-1 represents fine particles, many associated

with larger branching conglomerate structures, and somenearly circular. This group includes 4.4% of the analyzedparticles. It accounts for 18% of the fine-mode particlesfrom 5 July, and 3% of the fine-mode SAL particles from 16July. This group has an average molar composition of 22%S, 17% Ca, 16% Na, 11% Si, 11% Mg, 10% Al, 5% P, and3% Cl, and may represent conversion products of Ca-richparticles.[65] 2. CaS-2 represents 4% of the coarse and 2% of the

fine 5 July integrated samples, but also represents about1.5% of the coarse particles from each of the other samples.With its average composition of 35% S, 30% Ca, 8% Si, 8%Na, 6% Al and 5% Mg, this group probably representsnative gypsum or anhydrite from crustal sources. Somecharacteristic lenticular and spar crystal shapes are seen,but most particles are relatively rounded. The median shapeand aspect values reflect the fact that many of these particlesare part of loose aggregates.4.2.7. Mg-Silicate Group[66] The MgSi group comprises 1% of the analyzed

particles. It has a size distribution almost identical to theAl-Si group, suggesting that chemically and morphologi-cally they behave the same. Two subgroups are useful todescribe these particles.[67] 1. MgSi-1 is nearly absent from the 5 and 16 July

fine SAL samples. However, it is associated with the SALand MBL coarse particles from 16, 20, 21, 22, and 24 Julysamples. This subgroup has an average composition of 43%Si, 26% Mg, 14% Al, 5%Na, 4% Fe.[68] 2. MgSi-2 is associated with the 5 and 16 July

sample periods, but is less prevalent in the later sampleperiods, perhaps indicating a different source region. Thissubgroup comprises 0.8% of the analyzed particles and hasan average composition of 31% Si, 25% Mg, 20% Al, 7%Na, and 5% Fe.4.2.8. Ca-Silicate Group[69] The CaSi group comprises about 1.5% of the ana-

lyzed particles, and represents 4.2% of the fine particles

from the 21 July SAL sample. The average compositionranges from roughly 29 to 34% Si and 24–26% Ca, withsignificant Al, Mg and Na. The near absence of CaSiparticles from the MBL probably reflects conversion toaggregates or Ca- and S-rich particles, as these groups areenhanced in MBL samples.4.2.9. Fe-Rich Group[70] The Fe group, comprises about 1% of the analyzed

particles, and accounts for 2.4% of the fine-mode particlesfrom 5 July. The iron group size distribution tracks that ofthe Ca-S group, another species that was enhanced on 5July. It is best described by two subgroups.[71] 1. Fe-1 is found in all samples, but is more prevalent

in integrated and SAL coarse-mode samples. This subgroupranges in average composition from 41–58% Fe, 20–12%Si, 17–8% Al, 6–4% Na and Mg.[72] 2. Fe-2 consists of fine-mode particles from 5 and 16

July, accounting for 2.4% of the integrated 5 July fine-modeparticles, and 0.5% of the 16 July fine-mode particles. Thissubgroup has an average composition of 32% Fe, 15% Na,15% Si, 13% Al, 9% Mg, 6% S, and 3% P. The phosphorouscould be from ocean aerosols, but may reflect a source regionin southern Mauritania where phosphate mining occurs.4.2.10. Mg- and Ca-Rich Group[73] MgCa, comprises roughly 0.2% of the analyzed

particles and occurs in the coarse-mode dust layer samplesfrom 16 July, 20 July, 21 July, and 22 July. These particlesrange from 46–36% Mg and 27–20% Ca, and include someSi, Al and Na. These are probably dolomitic minerals withsilicate contamination. This cluster is not seen in the 5 Julysamples and, as the Ca- and S-rich particle groupings havelow Mg levels (on average,<10%), the absence cannot beexplained by conversion to anhydrite. Size is similar to theSi-rich cluster, with a more pronounced peak in area at 5 mmthan other species.4.2.11. Ti-Rich Group[74] Ti, comprises 0.3% of the analyzed particles, and

contains particles having >20% Ti. These particles are rare,but occur in both the fine and coarse mode, often in smallaggregates. Particle size shows more a bi-modal behaviorthan other species and may suggest two separate mineral-ogical types. But, as this is a trace species and the modescannot be cross-verified between the fine- and coarse-modeanalysis, we cannot exclude the possibility of artifact.4.2.12. Na-Rich Group[75] Na, comprises 0.7% of the analyzed particles and has

average composition ranging from 30–45%Na, 30–36%Si,12–15%Al, with some Mg and Cl. Note that this group isnot expected to include many NaSO4 particles; the mini-mum size range for analysis (0.15 microns) was chosen toexclude the majority of these particles in order to get betterstatistics on dust particles. Like the Ti-rich group, theseparticles exhibit bi-modal behavior.

5. Discussion: Bulk Approximations andEstimates of Chemistry

5.1. Derivation of Mineralogical Information

[76] A goal of the PRIDE study was to use bulk andsingle-particle analysis to help derive mineralogical infor-mation to be used in further dust modeling studies. It is firstprudent to ensure that the DRUM and single-particle


analyses are consistent. For comparison with the bulkDRUM sampler data, the individual particle data masseswere approximated by assuming an elliptical volume (withaxes of Long Proj, Minor Proj, 1/2 Minor Proj). The particlemass approximations, weighted by the elemental ratios,were used to generate bulk elemental mass relative to Sifor comparison (Table 5). The most important ratio, Al to Si,is within 10% with the bulk drum analysis. K and Ti are alsowell matched. Fe is enriched in the single-particle analysisby 10–30%. Mg is the only element significantly enrichedin the single-particle analysis (on the order of 2X). This isnot surprising since Mg is a trace species, and in XRFanalysis it can be swamped by noise. Conversely S wasenriched in the DRUM sample by a similar amount,potentially due to fine sulfates that, while analyzed on theDRUM strips, were smaller than the single-particle analysissize threshold, and so were not analyzed on the aircraftsamples. While these differences may appear large, giventhe relative uncertainties in deriving a mass ratio, weconsider the agreement to be quite good.[77] Given this reasonable comparison, the bulk analysis

from the DRUM sampler, and the cluster analysis from thesingle-particle analysis, we can derive a mineralogicalmodel for the PRIDE field study. With the exception ofthe 5 July event (to be discussed below), dust at Puerto Ricowas remarkably stable. Elemental ratios from the DRUMand single-particle analyses for species with good signal tonoise showed high correlation coefficients and little vari-ance. The ratios between magnesium, aluminum, silicon,potassium and iron were nearly static for the entire studyperiod.[78] As much of the Saharan region overlies ancient

mudstone and shale layers, illite, a main constituent ofweathered shale and mudstone, is expected in the claysignature. Soils of the more humid Sahel region weatherthrough leaching of soluble cations (Na, Mg, Si, K, Ca, Mn,etc.), producing a material that is relatively enriched in ironoxides, aluminum oxides, and 1:1 Al:Si crystalline clays,such as kaolinite [Food and Agricultural Organization(FAO), 2001]. Feldspar and plagioclase minerals from aridregions, exposed to heat but little water, tend to weather toform chlorites, or smectites such as montmorillonite. Final-ly, carbonates, gypsum and anhydrite, and various saltsfound in depressions and seasonal lakes in the Sahara areexpected to be mobilized during strong dust events. Whilemany of these minerals cannot be definitively identified

without X-ray diffraction analysis, the elemental composi-tion and morphology of the individual particles analyzedfrom the aircraft samples are not inconsistent with thismineralogy.[79] From previous X-Ray diffraction (XRD) and optical

analysis of Saharan dust [e.g., Falkovich et al., 2001;Anderson et al., 1996; Glaccum and Prospero, 1980;Prospero et al., 1981; Sokolik and Toon, 1999; Caquineauet al., 2002], Saharan dust minerals tend to be comprisedprimarily of illite, palygorskite and kaolinite, with nonclayminerals including calcite, quartz, dolomite, feldspar (mi-crocline and plagioclase), halite, and gypsum. This isconsistent with the elemental data and cluster analysis fromPRIDE, except that higher magnesium levels and reducedpotassium levels were seen in the particles analyzed in thisstudy. This would suggest that the collected silicates mayhave included more mafic feldspars, micas, and chlorites(until XRD analysis is performed on these samples, this willremain unresolved). Regardless, �60% of dust particlenumber and 70% of mass can be attributed to alumino-silicates, probably clay and feldspar minerals. A further�10% of number and �20% of mass can be grouped into Sienriched particles, like quartz or amorphous silicates, thusmaking a total of 70% and 90% of dust particle mass andvolume (respectively), which can be categorized into thesetwo broad groups.[80] Many of the collected particles visually appear to be

mineral aggregates. In fact, the most typical environment forclay minerals is as mixed layers or aggregates of varioustypes of clay particles. Clay particles often coat mineralgrains, form aggregates, bridge spaces between mineralgrains, or form layered structures. Table 6 lists formulae,density, and optical properties at a standard 589 nm wave-length [Gribble and Hall, 1985] of some of the commonminerals expected in Saharan or Sahel soils and dusts. Amorein-depth discussion of optical properties, including wave-length dependence of index of refraction for some commonSaharan minerals, is given by Sokolik and Toon [1999].[81] In these dust particles, iron is an important component

since, in its oxidized form, it is considered the most dominantlight absorbing species [Sokolik et al., 1999]. Consequently,the chemistry and morphology of iron in dust is important forradiative transfer modelers. On the basis of the DRUM andsingle-particle analyses, elemental iron composes �2.5–3%of total dust mass (this based on the assumption that alumi-num is 8% of total mass). Iron, with a molar ratio to silicon of

Table 5. Reconstructed Bulk Mass Relative to Si for All Analyzed Particlesa

DRUM 5 July Int 16 July MBL 16 July SAL 20 July SAL 21 July SAL 22 July SAL 24 July MBL 24 July SAL

Na – 0.48 0.25 0.12 0.07 0.11 0.07 0.54 0.09Al 0.47 0.55 0.51 0.52 0.50 0.49 0.50 0.50 0.51Mg 0.06 0.20 0.14 0.14 0.12 0.13 0.12 0.15 0.13Si 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0P – 0.05 0.05 0.04 0.03 0.04 0.03 0.04 0.05S 0.1 0.11 0.06 0.05 0.03 0.05 0.03 0.07 0.05Cl – 0.42 0.18 0.05 0.02 0.05 0.02 0.48 0.03K 0.08 0.10 0.09 0.08 0.07 0.08 0.07 0.09 0.07Ca 0.18 0.17 0.07 0.10 0.10 0.13 0.11 0.08 0.10Ti 0.02 0.03 0.03 0.02 0.02 0.03 0.02 0.02 0.03Fe 0.14 0.21 0.15 0.17 0.16 0.17 0.16 0.15 0.15Sn – 0.07 0.07 0.05 0.03 0.05 0.04 0.04 0.06

aParticles are labeled by date and altitude range of samples. Also shown is the average DRUM sampler ratio as given in Table 2. Int,integrated; MBL, marine boundary layer; SAL, Saharan Air Layer.


�1:11 from the DRUM sampler, is also close to the empiricalillite model [illite = K0.6(H3O)0.4Al1.3Mg0.3Fe0.1

2+

Si3.5O10(OH)2 � (H2O)]. Based on the size partitioning inthe DRUM sampler, the Si:Fe mass ratio was fairly consistentas a function of size. A ternary plot of the individual particleanalysis data from the aircraft samples, Figure 13, shows therelationship between aluminum, silicon, and iron in all theaircraft samples. The vast majority of the particles have lessthan 10% Fe (molar), which is normal mineralogy for illiteand other silicates. Particles with higher molar percentages ofFe tended toward balanced Al and Si molar percentages,suggesting Sahel origins. Iron-rich particles ranged from10% Fe (molar) to a high of 90% Fe, though these particleswere extremely rare. The cluster analysis gives good averagecompositions for the Fe-rich particles. The highest relativeproportion of iron-rich particles is seen in the 5 July integrat-

Table 6. Density and Real Index of Refraction of Minerals Found in Saharan Dusta

Type Common Name Formula Mean Density Real Index of Refraction

Clay illite K0.6(H3O)0.4Al1.3Mg0.3Fe0.1Si3.5O10(OH)2�(H2O)

2.75 a = 1.535 � 1.57b = 1.555 � 1.60g = 1.565 � 1.605

Clay kaolinite Al2Si2O5(OH)4 2.6 a = 1.553 � 1.563b = 1.559 � 1.569g = 1.56 � 1.57

Clay montmorillonite (Na,Ca)0.5(Al,Mg,Fe)4(Si,Al)8O20(OH)4�n(H2O)

2.35 a = 1.485 � 1.535b = 1.504 � 1.55g = 1.505 � 1.55

Clay smectite (Na,Ca)Al4(Si,Al)8O20(OH)4�2(H2O) 2.34 a = 1.519b = 1.55g = 1.559

Clay chlorite Na0.5(Al,Mg)6(Si,Al)8O18(OH)12�5(H2O)

2.42 a = 1.542 � 1.564b = 1.545 � 1.581g = 1.545 � 1581

Ca rich calcite CaCO3 2.71 e = 1.486w = 1.64 � 1.66

Ca rich dolomite CaMg(CO3)2 2.84 e = 1.5w = 1.679 � 1.681

Ca rich gypsum CaSO4�2(H2O) 2.3 a = 1.519 � 1.521b = 1.522 � 1.523g = 1.529 � 1.53

Ca rich anhydrite CaSO4 2.97 a = 1.569 � 1.573b = 1.574 � 1.579g = 1.609 � 1.618

SiO2 quartz SiO2 2.62 w = 1.543 � 1.545e = 1.552 � 1.554

Feldspars microcline KAlSi3O8 2.56 a = 1.518b = 1.522g = 1.525

Plagioclase feldspar Var oligoclase (Na,Ca)(Si,Al)4O8 2.65 a = 1.533 � 1.543b = 1.537 � 1.548g = 1.542 � 1.552

Plagioclase feldspar Var albite NaAlSi3O8 2.62 a = 1.528 � 1.533b = 1.532 � 1.537g = 1.538 � 1.542

Plagioclase feldspar Var anorthite CaAl2Si2O8 2.73 a = 1.572 � 1.576b = 1.578 � 1.583g = 1.583 � 1.588

Oxides hematite Fe2O3 5.27 –Oxides goethite FeO(OH) 3+ a = 2.26

b = 2.393g = 2.398

Oxides gibbsite Al(OH)3 2.34 a = 1.568 � 1.57b = 1.568 � 1.57g = 1.586 � 1.587

Oxides rutile TiO2 4.25 w = 2.621e = 2.908

Salt halite NaCl 2.17 n = 1.544aGiven is type, common name, formula, mean density (g/cm3), and index of refraction for biaxial minerals (a, b, g), and uniaxial minerals (w,e).

Table 7. Molar Ratios of Na and Cl for All Samples of R2 Values

for the Na to Cl Regressiona

NaCl R2 Yint

5 July 1 to 0.62 0.93 �0.0416 July 1 to 0.51 0.87 �0.0224 July 1 to 0.64 0.93 �0.0516 July 1 to 0.37 0.58 020 July 1 to 0.14 0.15 021 July 1 to 0.20 0.30 022 July 1 to 0.13 0.21 022 July 1 to 0.09 0.23 0

aNote the SAL samples show poor R2 values; little of the Na appears tobe from NaCl.


ed sample, in the fine mode. The fine mode, iron-richparticles are characterized by nearly balanced Al and Simolar percentages, supporting a kaolinite mineralogy. Thelarger, iron-rich particles show more widely varying Al to Siratios. This may reflect iron contamination of nonkaoliniticminerals, or may simply indicate aggregation of fine, iron-rich kaolinitic particles with other silicates. Other than the5 July sample, the samples show relatively consistent percen-tages of iron-rich particles, though relatively fewer fine-modeiron-rich particles are seen.[82] Interestingly, the MgSi-2 group of the Mg enriched

silicates also tended to have near 1:1 Al Si molar ratios.These particles were more prevalent in the earlier samplingdates, 5 and 16 July, and were relatively absent from the 21,22, and 24 July samples, replaced by increased prevalenceof the MgSi-1 group particles, having Si:Al ratios closer to2:1. Without XRD analysis, determining whether this rep-resents a single species, such as amphibole or garnet, or astable aggregate is not possible, though the apparent shift inmineral type does support a difference in source region.[83] Bulk mass ratios of Na:Cl, and Ca:S from the aircraft

samples suggest that SAL samples are enriched in bothsodium and calcium-rich particles. Stoichiometry does not,in general, favor the formation of NaCl or CaSO4 (gypsum).Calcium and sodium are more likely to be in the form offeldspars (e.g., cluster group AlSi-4, Ca-rich, and Na-rich)accounting for �8% of particles and 3% of mass.[84] Although not a dominant species, calcium sulfate, as

gypsum or anhydrite, is present in the analyzed samples.The continuous range of the calcium sulfate concentrationrelative to Si suggests that some of these particles may beconversion products of calcium-rich minerals with SO2 inthe atmosphere. The coarse-mode particles tend to havemore nearly balanced calcium and sulfur molar fractionsthat probably represent crustal gypsum or anhydrite, and arange of calcium-rich but sulfur poor particles. The lack ofcalcium-rich particles in the fine mode supports some of themeasured calcium sulfate being due to conversion.[85] In the MBL, sodium was overwhelmingly in the

form of sea salt (see Table 7 and Figure 14). Three clear

particle groupings were seen that involve sodium: (1)Particles with clear NaCl X-ray lines (either pure NaCl orNaCl combined with silicate particles, (2) a small sodium-rich silicate group, and (3) an indeterminate population thatis either very well aged NaCl-rich particles or high sodiumsilicates. Particles from group 2 may be Na-feldspars, whileparticles of group 3 were rare. NaCl was not often found asindependent particles; rather it was typically found incombination with silicate particles.[86] For the samples re-analyzed at higher magnification,

the relationship between the coarse and fine-mode particlesis of interest. Three samples were reanalyzed; the 5 Julyintegrated MBL sample, the 16 July SAL sample, and the21 July SAL sample. For the 5, 16, and 21 July samples,normalized Al/Si ratio histograms were calculated for thecoarse and fine-mode particles. The difference plots for thecoarse - fine mode Al/Si ratio histograms are shown inFigure 15. The 5 and 16 July samples show similardifference plots with the coarse data slightly enhanced inparticles having Al/Si ratio less than 0.8, and the fine modeenhanced for particles with Al/Si ratios greater than 0.8.Two coarse-mode peaks are seen near 0.1 and 0.65, and afine-mode peak near Al/Si ratio of 1 (possibly kaoliniticclay particles). The 21 July difference plot, in contrast,shows three significant peaks; the coarse-mode data areenhanced for Al/Si ratios of 0.1 and 0.75, and the fine-modedata for particles with Al/Si ratios near 0.55. Little differ-ence is seen between the fine and coarse mode for Al/Siratios near 1, typical for Kaolinite. The 21 July sample alsoshows enhanced Al, S, Ca, Mg and Fe levels relative to Si,lending support to the later MBL and SAL samples having adifferent source or transport regime.

5.2. The 5 and 16 July Chemistry Anomaly

[87] From the Naval Aerosol Analysis and PredictionSystem (NAAPS www.nrlmry.navy.mil/aerosol), the prima-ry dust source regions in the Sahara and Sahel shifted fromcentral Saharan Africa toward western Africa during thestudy period of 25 June to 25 July (Figure 16). The sourceregions between 25 June and 3 July were located at the edgeof the Sahel in the area of the Mali, Niger, and Algeriaborder, extending northward into Algeria to include a regionof seasonal saline lakes to the east of Erg Chech. Soils-richin salt, gypsum, and calcium are common in the saline lakeregion of central Algeria near Erg Chech [FAO, 2001]. Thesource area at the edge of the Sahel centered on 20�Nlatitude (Figure 16a) is expected to be high in Kaolinite andrelatively low in Illite [Caquineau et al., 2002]. We did notsee a significant difference in the Al:Si ratio in the DRUMdata, though we did see a slight enhancement in particleswith Al:Si ratio near 1:1 in the aircraft sample data. The lackof a clear Kaolinite signature, coupled with the observedenhancement in Ca- and S-rich particles from this timeperiod, probably reflects the influence of the more northerlysource region, between 25� and 30� latitude near Erg Chech,which is expected to have a higher Illite/Kaolinite ratio[Caquineau et al., 2002]. Beginning 4 July, the sourceregions shifted west into Mauritania and Western Saharato include the sand dunes of the southern region of WesternSahara, and the sandy desert areas of Mauritania andnortheastern Mali. These soils are exposed to little water,so remain rich in potassium, magnesium, calcium, and other

Figure 13. Ternary diagram of molar fractions of iron,aluminum, and silicon for all analyzed aircraft sampleparticles.


bases. A few later events included minor inputs from themore eastern source regions, occurring 3–4 July, 8–9 July,and 13–14 July, but the main source areas after 3 July werein western regions where Illite is the predominant clayspecies [Caquineau et al., 2002]. Indeed, the later samples,especially those in the Saharan Air Layer from 20 21, and22 July, show similar chemistry and morphologies that aredistinct from the 5 July sample.[88] With an expected �7 day transport time from Africa

to Puerto Rico, the eastern sources would be responsible forthe 5 July anomalous chemistry. Saline lakebeds, calcium-rich, and gypsum-rich soil areas near Erg Chech potentiallyexplain both the high salt and gypsum levels, and the largenumber of diatom fragments collected. Calcium and Sulfurpeaks seen in the DRUM data on 13 July, and in the DRUMand aircraft samples on 22 July may be related to latersource events in this area (events seen on 8 July and 16July). However, concurrent emissions from western regionsof the Sahara in Mauritania, which include saline lakes andgypsum-rich playas, cannot be ruled out as significantsources for these later peaks.

6. Summary and Conclusions

[89] In this manuscript we present chemistry and mor-phology data from Saharan dust particles collected duringPRIDE from 3–24 July 2000. Bulk elemental analysis ofDRUM impactor strips from a surface site and single-particle

Figure 15. Normalized difference plot of coarse-finemode particles versus aluminum/silicon ratio. Specified inthe key is sample date.

Figure 14. Ternary diagrams of molar fractions of sodium, chlorine, and silicon for four analyzedaircraft sample dates. The 5 July and 21 July plots also display 5000X magnification particle data(average diameter <2 mm). Note the lack of NaCl in the SAL samples. (a) 5 July 2000 Integrated samples,(b) 21 July 2000 SAL samples, (c) 22 July 2000 SAL samples, and (d) 24 July 2000 MBL surfacesample.


analysis of samples collected from an aircraft platformshowed significant correlations for those species with goodsignal-to-noise ratios. The derived molar ratios were consis-tent with a predominantly illite dust source, though evidencefor kaolinite, feldspars and other species is present in thesingle-particle analysis data. An observed continuum in theAl:SI ratio probably reflects particle agglomeration, minerallayering, analysis uncertainty, and the fact that many mineralspecies range widely in composition.[90] During the sampling period, the DRUM impactor

recorded significant increases in dust surface concentrationson 5, 10, 15–16, and 21 July 2000 that were coincidentwith high aerosol optical thicknesses (AOTs). Additionalsurges of dust at the surface that were not accompanied byhigh column AOTs occurred on 13 and 24 July. Hence dustsurface concentrations did not strongly correlate with dustAOT.[91] Key elemental ratios from the bulk analysis for the

dust were for the most part in agreement with previousfindings. Magnesium, aluminum, silicon and iron ratioswere consistent with a dust dominated by the clay mineralillite {K0.6(H3O)0.4Al1.3Mg0.3Fe0.1

2+Si3.5O10(OH)2 �(H2O)},although potassium was underrepresented. The DRUM dataalso showed that while elemental ratios were nearly staticfor most of the field study, the dust event that occurred on 5and 16 July were significantly enriched in calcium andsulfur.[92] Seven of the most heavily loaded filters collected by

the research aircraft were analyzed by single-particle anal-ysis. Cross-sectional area distributions showed an areamedian diameter of �6 mm. As particles tend to lie flat onthe filter substrate, this value is larger than the true areadistribution of the ambient particles. By estimating particle

depth, we estimate the volume median diameter to be on theorder of 7 mm (again, this is likely to be an over estimate).Average particle aspect ratios, related closely to the averageratio of major to minor axes, were found to have a medianvalue of 1.9. Particle shape factors suggest a higher prob-ability of aggregates for larger dust particles.[93] Elemental speciation of individual particles was

performed on the 60,500 particles for which particle sizeswere measured. Ternary and cluster analyses support thestoichiometry of the DRUM impactor samples. More than70% of dust particle mass can be attributed to alumino-silicate clay minerals such as illite, Kaolinite, and montmo-rillonite. Silicates with lesser amounts of aluminum (such asamorphous silicon and quartz) make up the next largestgroup, comprising another 10–15%. Samples collected inthe marine boundary layer showed sea-salt particles weretypically combined with dust particles. The remainingparticles appear to be carbonates, sulfates, salts, and othertrace minerals. As seen in the DRUM sampler, the aircraftsample for 5 July also showed enhancement of calcium andsulfur in the form of gypsum or anhydrite (CaSO4).[94] The ratio of Si to Al remained fairly constant

throughout the study, both from the DRUM analysis andthe single-particle analysis of the aircraft data. As our dustsource regions appeared to shift from the southern centralSahara to the western Sahara during the course of the study,we expected to see some change in the silicates. For the 5July aircraft sample, we did observe a relative enhancementin particles with 1:1:1 Al to Si to Mg ratios, as well as asubstantial enhancement in the number of Ca- and S-richparticles, and of diatom fragments. However, the mean Al toSi ratio remained stable, and relative percentage of kaolin-itic particles was only slightly enhanced. Examination of

Figure 16. NAAPS weekly average surface dust emissions for the Sahara and Sahel regions of Africa inyear 2000: (a) 26 June to 3 July, (b) 3–10 July, (c) 10–17 July, and (d) 17–24 July.


dust source regions in West Africa suggests that the anom-alous chemistry associated with the 5 July event is related toa more eastern source region than later events.

[95] Acknowledgments. We are grateful to the personnel at NavalStation Roosevelt Roads, Daniel Eleuterio, Roger Hahn and the entire staffat North Atlantic Meteorology and Oceanography Detachment, RooseveltRoads. We also would like to thank the whole Gibbs Flite Center crew,including William ‘‘Buzz’’ Gibbs, Michael Kane, Michael Hubble, and LyleRichards. Additional thanks for Navajo support are due to Duane Allen,NASA Ames Research Center. We appreciate comments from RichardPaulus, SSC San Diego. PRIDE funding was provided by the Office ofNaval Research Code 322, N0001401WX20194, and the NASA Mission toPlanet Earth program office.

ReferencesAnderson, J. R., P. B. Buseck, T. L. Patterson, and R. Arimoto, Character-ization of the Bermuda tropospheric aerosol by combined individual-particle and bulk aerosol analysis, Atmos. Environ., 30, 319–338, 1996.

Cahill, T. A., C. Goodart, J. W. Nelson, R. A. Eldred, J. S. Nasstrom, andP. J. Feeny, Design and evaluation of the DRUM impactor, in Proceed-ings of the International Symposium on Particulate and Multi-PhaseProcesses, edited by T. Ariman and T. Nejat, vol. 2, pp. 319–325, Taylorand Francis, Philadelphia, Pa., 1985.

Caquineau, S., A. Gaudichet, L. Gomes, and M. Legrand, Mineralogy ofSaharan dust transported over northwestern tropical Atlantic Ocean in re-lation to source regions, J. Geophys. Res., 107(D15), 4251, doi:10.1029/2000JD000247, 2002.

Coude-Gaussen, G., P. Rognon, G. Bergametti, L. Gomes, B. Strauss, J. M.Gros, and G. M. Le-Coustumer, Saharan dust on Fuerteventura Island(Canaries): Chemical and mineralogical characteristics, air mass trajec-tories, and probable sources, J. Geophys. Res., 92, 9753–9771, 1987.

d’Almeida, G. A., and L. Schutz, Number, mass and volume distributionsof mineral aerosols and soils of the Sahara, J. Clim. Appl. Meteorol., 22,233–243, 1983.

Falkovich, A. H., E. Ganor, Z. Levin, P. Formenti, and Y. Rudich, Chemicaland mineralogical analysis of individual mineral dust particles, J. Geo-phys. Res., 106, 18,029–18,036, 2001.

Food and Agriculture Organization, Lecture Notes on the Major Soils of theWorld, edited by P. Driessen et al., United Nations, Rome, 2001.

Gao, Y., and J. R. Anderson, Characteristics of Chinese aerosols determinedby individual-particle analysis, J. Geophys. Res., 106, 18,037–18,045,2001.

Glaccum, R. A., and J. M. Prospero, Saharan Aerosols over the tropicalNorth Atlantic: Mineralogy, Mar. Geol., 37, 295–321, 1980.

Gribble, C. D., and A. J. Hall, A Practical Introduction to Optical Miner-alogy, Allen and Unwin, Concord, Mass., 1985.

Intergovernmental Panel on Climate Change, Climate Change 2001: TheScientific Basis, edited by J. T. Houghton et al., Cambridge Univ. Press,New York, 2001.

Keene, W. C., R. Sander, A. A. P. Pszenny, R. Vogt, P. J. Crutzen, and J. N.Galloway, Aerosol pH in the marine boundary layer: A review and modelevaluation, J. Aerosol Sci., 29, 339–356, 1998.

Keene, W. C., et al., Composite global emissions of reactive chlorine fromanthropogenic and natural sources: Reactive chlorine emissions inven-tory, J. Geophys. Res., 104, 8429–8440, 1999.

Koren, I., E. Ganor, and J. H. Joseph, On the relation between size andshape of desert dust aerosol, J. Geophys. Res., 106, 18,047–18,054,2001.

Maring, H. D., Particle size distribution measurements at Puerto Rico dur-ing PRIDE, paper presented at IAMAS 2001 Meeting, Int. Assoc. ofMeteorol. and Atmos. Sci., Innsbruck, Austria, 12 –16 July 2001.

Maring, H., D. L. Savoie, M. A. Izaguirre, C. McCormick, R. Arimoto,J. M. Prospero, and C. Pilinis, Aerosol physical and optical properties andtheir relationship to aerosol composition in the free troposphere at Izana,Tenerife, Canary Islands, during July 1995, J. Geophys. Res, 105,14,677–14,700, 2000.

Okada, K., J. Heintzenberg, K. Kai, and Y. Qin, Shape of atmosphericmineral particles collected in three Chinese arid-regions, Geophys. Res.Lett., 28, 3123–3126, 2001.

Prospero, J. M., R. A. Glaccum, and R. T. Nees, Atmospheric transport ofsoil dust from Africa to South America, Nature, 289, 570–572, 1981.

Reid, J. S., T. A. Cahill, and M. R. Dunlap, Geometric/aerodynamic sizeratios of ash aggregates from burning Kuwaiti oil fields, Atmos. Environ.,28, 2227–2234, 1994.

Reid, J. S., D. L. Westphal, J. M. Livingston, D. L. Savoie, H. B. Maring,H. H. Jonsson, D. P. Eleuterio, J. E. Kinney, and E. A. Reid, Dust verticaldistribution in the Caribbean during the Puerto Rico Dust Experiment,Geophys. Res. Lett., 29(7), 1151, doi:10.1029/2001GL014092, 2002.

Reid, J. S., et al., Comparison of size and morphological measurements ofcoarsemode dust particles fromAfrica, J. Geophys. Res., 108, doi:10.1029/2002JD002485, in press, 2003a.

Reid, J. S., et al., Analysis of measurements of Saharan dust by airborne andground-based remote sensing methods during the Puerto Rico Dust Ex-periment (PRIDE), J. Geophys. Res., 108, doi:10.1029/2002JD002493, inpress, 2003b.

Sokolik, I. N., and O. B. Toon, Incorporation of mineralogical compositioninto models of the radiative properties of mineral aerosols from UV to IRwavelengths, J. Geophys. Res., 104, 9423–9444, 1999.

Sokolik, I. N., D. M. Winker, G. Bergametti, D. A. Gillette, G. Carmichael,Y. J. Kaufman, L. Gomes, L. Schuetz, and J. E. Penner, Introduction tospecial section: Outstanding problems in quantifying the radiative im-pacts of mineral dust, J. Geophys. Res., 106, 18,015–18,028, 2001.

Taylor, S. R., and S. M. McLennan, The geochemical evolution of thecontinental crust, Rev. Geophys., 33, 241–265, 1995.

��E. A. Reid and J. S. Reid, Aerosol and Radiation Modeling Section,

Marine Meteorology Division, Naval Research Laboratory, 7 Grace HopperStreet, Stop 2, Monterey, CA 93943-5502, USA. ([email protected])A. Broumas, S. S. Cliff, M. R. Dunlap, and M. M. Meier, Material

Science and Chemical Engineering Department, University of California,Davis, CA 95616, USA.K. Perry, Department of Atmospheric Sciences, University of Utah, Salt

Lake City, UT 84112, USA.H. Maring, Rosenstiel School of Marine and Atmospheric Science,

University of Miami, Miami, FL 33149, USA.


Documents

Characterization of African dust transported to Puerto ...dust.ess.uci.edu/ppr/ppr_RRM03.pdf · microphysical properties of dust particles collected on surface and airborne filter