comparison of particle sizes determined with impactor, afm and sem

12
Comparison of particle sizes determined with impactor, AFM and SEM Patience Gwaze a,1 , Harold J. Annegarn b,1 , Joachim Huth c,2 , Günter Helas a, a Max Planck Institute for Chemistry, Biogeochemistry Department, P. O. Box 55020, Mainz, Germany b Department of Geography, Environmental Management and Energy Studies, University of Johannesburg, P. O. Box 524, 2006 Auckland Park, South Africa c Max Planck Institute for Chemistry, Particle Chemistry Department, P. O. Box 55020, Mainz, Germany Received 28 June 2006; received in revised form 27 February 2007; accepted 27 February 2007 Abstract Particles size comparisons were made between conventional aerodynamic and mobility sizing techniques and physical geometric sizes measured by high resolution microscopes. Atmospheric particles were collected during the wet and dry seasons in the Amazonian ecosystems. Individual particles deposited on four stages of the MOUDI (Micro-Orifice Uniform Deposition Impactor) were characterised for particle volumes, projected surface diameters and morphologies with an Atomic Force Microscope (AFM) and a Scanning Electron Microscope (SEM). AFM and SEM size distributions were verified against distributions derived from response functions of individual MOUDI stages as specified by Winklmayr et al. [Winklmayr, W., Wang, H.-C., John, W., 1990. Adaptation of the Twomey algorithm to the inversion of cascade impactor data. Aerosol Science and Technology 13, 322331.]. Particles indicated inherent discrepancies in sizing techniques. Particle volumes were systematically lower than expected by factors of up to 3.6. Differences were attributed to loss of mass, presumably water adsorbed on particles. Losses were high and could not be accounted for by measured humidity growth factors suggesting significant losses of other volatile compounds as well, particularly on particles that were collected during the wet season. Microscopy results showed that for hygroscopic particles, microscopy sizes depend on the relative humidity history of particles before and after sampling. Changes in relative humidity significantly altered particle morphologies. Depending on when changes occur, such losses will bias not only microscopy particle sizes but also impactor mass distributions and number concentrations derived from collected particles. © 2007 Elsevier B.V. All rights reserved. Keywords: Atmospheric aerosols; Single particle analysis; Particle size; Morphology 1. Introduction Aerosol particle size is a fundamental parameter in understanding contributions of particles to the radiation budget, and effects on human health and air quality (IPCC, 2001). Sizes are defined corresponding to specific physical properties measured on generally irregular particles, while most sizing techniques assume particles are spherical in shape. Direct physical particle Atmospheric Research 86 (2007) 93 104 www.elsevier.com/locate/atmos Corresponding author. Fax: +49 6131 305 579. E-mail addresses: [email protected] (P. Gwaze), [email protected] (H.J. Annegarn), [email protected] (J. Huth), [email protected] (G. Helas). 1 Department of Geography, Environmental Management and Energy Studies, University of Johannesburg, P. O. Box 524, 2006 Auckland Park, South Africa. Fax: +27 11 489 2430. 2 Fax: +49 6131 371 290. 0169-8095/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.atmosres.2007.02.009

Upload: wits

Post on 04-Dec-2023

0 views

Category:

Documents


0 download

TRANSCRIPT

86 (2007) 93–104www.elsevier.com/locate/atmos

Atmospheric Research

Comparison of particle sizes determinedwith impactor, AFM and SEM

Patience Gwaze a,1, Harold J. Annegarn b,1, Joachim Huth c,2, Günter Helas a,⁎

a Max Planck Institute for Chemistry, Biogeochemistry Department, P. O. Box 55020, Mainz, Germanyb Department of Geography, Environmental Management and Energy Studies, University of Johannesburg,

P. O. Box 524, 2006 Auckland Park, South Africac Max Planck Institute for Chemistry, Particle Chemistry Department, P. O. Box 55020, Mainz, Germany

Received 28 June 2006; received in revised form 27 February 2007; accepted 27 February 2007

Abstract

Particles size comparisons were made between conventional aerodynamic and mobility sizing techniques and physicalgeometric sizes measured by high resolution microscopes. Atmospheric particles were collected during the wet and dry seasons inthe Amazonian ecosystems. Individual particles deposited on four stages of the MOUDI (Micro-Orifice Uniform DepositionImpactor) were characterised for particle volumes, projected surface diameters and morphologies with an Atomic ForceMicroscope (AFM) and a Scanning Electron Microscope (SEM). AFM and SEM size distributions were verified againstdistributions derived from response functions of individual MOUDI stages as specified by Winklmayr et al. [Winklmayr, W.,Wang, H.-C., John, W., 1990. Adaptation of the Twomey algorithm to the inversion of cascade impactor data. Aerosol Science andTechnology 13, 322–331.]. Particles indicated inherent discrepancies in sizing techniques. Particle volumes were systematicallylower than expected by factors of up to 3.6. Differences were attributed to loss of mass, presumably water adsorbed on particles.Losses were high and could not be accounted for by measured humidity growth factors suggesting significant losses of othervolatile compounds as well, particularly on particles that were collected during the wet season. Microscopy results showed that forhygroscopic particles, microscopy sizes depend on the relative humidity history of particles before and after sampling. Changes inrelative humidity significantly altered particle morphologies. Depending on when changes occur, such losses will bias not onlymicroscopy particle sizes but also impactor mass distributions and number concentrations derived from collected particles.© 2007 Elsevier B.V. All rights reserved.

Keywords: Atmospheric aerosols; Single particle analysis; Particle size; Morphology

⁎ Corresponding author. Fax: +49 6131 305 579.E-mail addresses: [email protected] (P. Gwaze),

[email protected] (H.J. Annegarn), [email protected](J. Huth), [email protected] (G. Helas).1 Department of Geography, Environmental Management and

Energy Studies, University of Johannesburg, P. O. Box 524, 2006Auckland Park, South Africa. Fax: +27 11 489 2430.2 Fax: +49 6131 371 290.

0169-8095/$ - see front matter © 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.atmosres.2007.02.009

1. Introduction

Aerosol particle size is a fundamental parameter inunderstanding contributions of particles to the radiationbudget, and effects on human health and air quality(IPCC, 2001). Sizes are defined corresponding tospecific physical properties measured on generallyirregular particles, while most sizing techniques assumeparticles are spherical in shape. Direct physical particle

94 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

sizes can be determined with an Atomic ForceMicroscope (AFM) and for less volatile aerosol particleswith electron beam microscopy techniques (Scanning/Transmission Electron Microscope — S/TEM).

Particle sizing techniques of interest in this study aremicroscopy sizing and aerodynamic sizing with thecascade impactor. Some of the earliest studies onparticle sizing with the AFM were made by Friedbacheret al. (1995). The authors measured sizes of fineparticles collected with an 11-stage impactor andderived size distributions of particles with equivalentsurface diameters between 0.015 and 3.0 μm. Particlesize ranges were found to be comparable to expectedsize distributions of individual impactor stages. In otherstudies, volatile losses and hygroscopic growth ofammonium sulphate were quantified by measuringparticle sizes with the AFM (in ambient conditions)before and after shortly exposing particles to TEMvacuum (Pósfai et al., 1998). Köllensperger et al. (1999)found changes in morphology and surface properties ofaggregated soot-like test aerosols when particles wereexposed to different humidity states in the AFM.Recently, Barkay et al. (2005) showed that AFM andSEM are complementary techniques in determiningmorphology and mass of individual particles.

Several equivalent particle sizes can be acquiredfrom projections of particles in microscope images(Hinds, 1999). In this study, we will focus on the volumeequivalent diameter of the 3 dimensional (3D) volumeprojected in the AFM image — dve, and an averagedsurface diameter — da, averaged between the longestaxis of the particle's silhouette and its orthogonal axis.There are considerations to be made in AFM analyses.There are limitations in the AFM sizing techniquespertaining to the failure in recovering the actual volumeof an amorphous particle, and effects of the probe‘convolution’ with the sample as will be discussed later.However, microscopy descriptions of particle sizes areindependent of some of the physical properties thatcomplicate size measurements, e.g., shape factors,particle density or refractive index.

Scanning electron microscopy provides 2D informa-tion of particle surface (surface diameter and equivalentdiameter of the projected surface), complementaryelemental composition, and a detailed optical imagerywith a large field of view. The technique howeversuffers from mass loss from adsorbed water andvolatiles on particles in the vacuum. There is lack of3D information, and this might introduce bias onparticles that orient on substrates, e.g., fibres and flatparticles. Though it is possible to tilt the sample supportand measure particle size in different electron beam

projections, which can then be computed to particlevolume, the procedure is rather tedious and thereforehardly applied. For both AFM and SEM sizingtechniques, analysis is subject to personal bias — theoperator might overlook some particles and considerinteresting and more familiar features. Additionally, sizeparameters are derived from analysing a small numberof particles with comparatively inferior countingstatistics to the number deposited on substrates and thenumbers probed by most conventional spectrometers.

Size distributions of particles collected with cascadeimpactors are well characterised in the literature (Baronand Willeke, 2001). Recovery of size distributions havebeen explored using algorithms that assume no functionalform, non-linear smoothed Twomey algorithm (Donget al., 2004; Winklmayr et al., 1990) or an algorithmwhich assumes a multimodal lognormal distributionfunction (Dzubay and Hasan, 1990). These algorithmssuccessfully recovered mass distribution models to fitexperimental observations. Concerns with sizing particleswith impactors are due to particle losses and effects ofparticle shapes on the drag coefficient. Inter-stage lossesof small particles (dpb0.1 μm) due to Brownian motionare between 5% to 10% (Marple et al., 1991). These lossesmight significantly affect sizing measurements if notaccounted for in quantitative analysis (Dong et al., 2004).For non-spherical particles, a dynamic shape factor χ hasto be applied to the drag and settling velocity to accountfor particle shape on its motion (Hinds, 1999). Thoughvalues of χ can be found in the literature for variousparticle species, applying the shape factors to a hetero-geneous aerosol is complicated by the varying composi-tion, shapes and dynamic behaviour. For example, onfractal soot particles χ is a function of the particle massand fractal dimension (Baron and Willeke, 2001).

Ideally, sizes determined with different techniquesare consistent when among other parameters of thecarrying medium, shapes and density of sampledparticles are known. It has been shown recently formineral dust particles, that large dissimilarities canoccur when size distributions are compared over severalsizing techniques, particularly between microscopy,optical and aerodynamic sizing techniques (Reid et al.,2003). When different size distributions are implemen-ted into climate models to predict radiative forcings ofparticles these dissimilarities can be translated intosignificant forcing uncertainties.

The purpose of this study is to investigate and compareconsistency in particle sizes as characterised by micros-copy techniques and a cascade impactor. Physical sizes ofparticles on four stages of a cascade impactor aremeasured with the AFM and SEM and verified against

95P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

size distributions derived from known response functionsof impactor stages as specified by Winklmayr et al.(1990). Investigations will focus on stages 5 to 8, stageswith aerodynamic particle diameter 0.1bdpb2.5 μm, asthe large particles on stages 1 to 4 are beyond the limitingrange of 3μm in height for our AFM.The sub-micrometerrange studied here is significant since it accounts for mostof the aerosol number, mass, surface area and theseparticles are known to scatter and absorb light mostefficiently (Seinfeld and Pandis, 1998).

Given the numerous notations of equivalent dia-meters in literature, equivalent diameters used incharacterising particle sizes in this article are definedas follows:

1.1. Surface equivalent diameter (da )

The surface diameter is defined as the arithmeticaverage of the longest axis of the projected particlesilhouette and its orthogonal axis.

1.2. Volume equivalent diameter (dve )

The volume equivalent diameter is defined as thediameter of a spherical particle with the same volume asthe semi-ellipsoidal projection in the AFM images, withheight ha, longest axis and its orthogonal axis as thedimensions of the semi-ellipsoid.

1.3. Stokes diameter (dρ )

The Stokes diameter is defined as the diameter of aspherical particle with the same density and settlingvelocity as the particle in question. Stokes diameters aredefined for known particle densities ρ of 1.15, 1.4 and1.5 g cm−3. Definitions of the other equivalent particlediameters mentioned here and not explicitly comparedcan be found in the literature (Hinds, 1999).

2. Experimental techniques

Online particle measurements and sample collectionswere performed in Brazil during the LBA-EUSTACH(Large-Scale Biosphere-Atmosphere Experiment part ofthe EUropean Studies on Trace Gases and AtmosphericChemistry) and SMOCC Smoke Aerosols, Clouds,Rainfall and Climate campaigns (Guyon et al., 2003;Rissler et al., 2005). The campaigns were aimed atunderstanding atmospheric processes in the Amazonianecosystems, linkages between trace gas exchanges,aerosols and cloud formation mechanisms. Amazonianwet seasons represent clear atmospheric conditions with

average particle concentrations of 450 cm−3 typical of abackground aerosol (Zhou et al., 2002), while dryseasons are influenced by intense biomass burning withhigh particle concentrations exceeding 40000 cm−3

(Artaxo et al., 2002).Aerosol samples were collected onto aluminium

substrates using a 10 stage MOUDI (Model 110, MSPCooperation, Minneapolis, U.S.A). The first aerosolsample, hereafter identified as sample #1, was collectedduring the LBA-EUSTACH campaign from the 8–11May 1999 at Jaru Biological Reserve (10° 04′ 55″S, 61°55′ 48″W, 110 m above sea level). Relative humidityduring sampling was between 80% and 100%. Details onexperimental set-up are described elsewhere (Guyon etal., 2003). The sample was collected during the wetseason and considered ‘clean’ from trajectory analyses(Guyon et al., 2003). Sample #2 was collected duringSMOCC campaign on 15 September 2002 at a represen-tative pasture site, Fazenda Nossa Senhora Aparecida(7:40 to 17:45 local time, 10° 45′ 44″S, 62° 21′ 2″ W,315m above sea level). This sample was representative ofthe dry season and the averaged RH was 65% duringsampling.

The MOUDI aluminium substrates were conditionedand weighed in a temperature and relative humiditycontrolled environment prior to sampling (at RH of 50%and temperature of 20 °C for 24 h). After sampling,particulate mass collected on each stage was weighedwith a Mettler microbalance with a sensitivity of 1 μgand an accuracy of ±3 μg for the aluminium substrates(Guyon et al., 2003). Mass distributions of particles,dMinv /dlogdp, were recovered from the experimentalobserved mass distribution dMobs / dlogdp using aTwomey smooth non-linear iterative algorithm (Kandli-kar and Ramachandran, 1999; Winklmayr et al., 1990).Additional ‘artificial’ stages were introduced into theroutine before the inlet stage and after the last stage 10.Loadings on stages 1 and 10 were not measured butestimated from a linear extrapolation of dMobs /dlogdp.The artificial stages and estimated loadings wereintroduced to impose constraints in the inversion(Winklmayr et al., 1990). Number concentrations ofparticles dN /dlogdp were then determined from therecovered mass distribution. Size distributions withineach single stage were determined from the overall sizedistribution of the ambient samples as measured by theimpactor and weighed against the kernel functions of thespecific stage. The kernel functions represent fractionaldeposition of particles on each impactor stage as afunction of aerodynamic particle diameter (Winklmayret al., 1990). By assuming spherical particles withknown densities, the individual stage size distributions

96 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

were then expressed as distributions of Stokes dia-meters. For sample #1 a particle density ρp=1.5 g cm−3

was taken as the average of 1.47bρpb1.53 g cm−3 asobserved on the same samples (Guyon et al., 2003). Onsample #2, particle densities were 1.4 and 1.15 g cm−3

for dry and ambient particles, respectively (Rissler et al.,2005).

A DMPS (Differential Mobility Particle Sizer) and anAPS (Aerodynamic Particle Sizer) were used to measurenumber-size distributions of dry particles with diametersof 0.003 to 0.850 μm and 1.0 to 4.0 μm, respectively.Details of the experimental set-up are given elsewhere(Rissler et al., 2005). Size distributions were scanned atsynchronised 10 minute intervals in both instrumentsand combined to represent a size range of 0.003 to 4 μm.The scans were averaged over the time the MOUDI wasoperational — and the result is hereafter called theDMPS-APS size distribution. To recover the ambientaerosol distribution, the dry DMPS-APS size distribu-tion was simulated with computed growth factors atRH=65% using a procedure described elsewhere(Rissler et al., 2005; Zhou et al., 2002). Size distribu-tions of particles that could be deposited on stages 5 to8 were then computed from the DMPS-APS distribu-tions and Stokes diameter d1.15 — representing ambientparticles with particle density ρp =1.15 g cm− 3.Unfortunately, there were no DMPS-APS measurementsduring the collection of sample #1.

3. Microscopy analysis

The aerosol samples were analysed in ambientconditions with the AFM CP-Research Model (DigitalInstruments, SPM, Veeco Instruments, U.S.A). AFMlateral dimensions (xy) and heights (z) were calibratedbefore particle imaging using standard calibration gridsof lateral length of 9.9 μm and grids with heights of 19±1 nm, 104±1.5 nm and 540±2 nm. Commerciallyavailable gold coated conical tips of boron doped siliconwith a nominal half angle of 12° were used in bothcontact and non-contact modes of operation. Imagingforces were kept between 1 and 5 nN and scanning ratesbetween 0.01 and 2 Hz. Rates were chosen to avoid thetip damaging or dragging particles during scanning.Scan sizes were varied between 2 and 10 μm withresolutions of 256, 512 to 1024 pixels per length. Thesemicroscope settings were applied on all the imagesacquired. A reasonably large area was scanned first,with a scan size depending on the size of particlesexpected. Where particles were identified, higherresolution images were obtained by reducing both thescan size and scanning speed. Several spots were

imaged on each stage to compensate for depositioninhomogeneities.

After AFM analyses, the same samples wereinvestigated with a high resolution SEM (Leo 1530Electron Microscope Ltd., now Nano TechnologySystems, Zeiss). Images were acquired with an averagedsample current of 10 pA in a vacuum of between 10−5

and 10−4 Pa with magnifications between 30000 and80000, equivalent to 12.40 to 4.65 nm per pixel,respectively. Images contained 1024×768 pixels with256 grey levels. Some particles had haloes around themwhich had presumably been formed by the spreading ofa liquid component adsorbed on particles. Particlesinside haloes were characterised while excluding haloesin the measurements.

Image analysis was performed manually to reduceerrors due to particles overlapping and contrast artefacts.The Scanning Probe Image Processor (SPIP) Version3.0.1.1 (Image Metrology A/S, Denmark) was used. Theprojected surface diameter of a particle's silhouette, da,was calculated as the arithmetic average of the longestaxis, the Feret diameter with no preferential orientationand its perpendicular axis (Hinds, 1999). The aspectratio AR, a measure of the roundness of a particleprojection, was defined as the ratio of the longest to itsorthogonal axis. Average heights in AFM images weretaken from the same axes profiles as the vertical distancebetween the base and top of the particle, ha. Anequivalent volume diameter (dve) of a spherical particlewas then calculated from the volume of the semi-ellipsoidal projection (of 2π) in the AFM images (withheight ha, longest axis and its orthogonal axis as thedimensions of the semi-ellipsoid). At least 100 particleswere characterised from each stage.

Output from scanning probe microscopes is a dilationof the probe (AFM tip or SEM electron beam) with thesample surface (Dongmo et al., 1996; Todd and Eppell,2001; Villarrubia, 1997; Williams et al., 1996). Thismeans that the probe adds to the projected length of aparticle an artefact that is a function of probe sharpnessand sample morphology. Probe dilation is substantialparticularly when particle sizes are comparable to theprobe dimension and these effects have to be removed inquantitative analyses. Dilation in images was investi-gated by measuring particles of known diameters in therange of expected particles sizes. Spherical polystyrenelatex (PSL) particles of diameters dp of 0.126±0.001 μm, 0.356±0.007 μm and 0.600±0.005 μmwere analysed with the AFM and SEM (40 particles ofeach PSL size).

Tip effects were characterised with a routine in theSPIP program similar to reported tip algorithms

Fig. 1. (a). SEM image of brochosomes collected on stage 6 of sample#1. Particles are hollow with projected surface diameters of 0.3 μm to1 μm. (b) AFM topographic image of a part of a brochosome. Coneopening is 200 nm in diameter.

97P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

(Villarrubia, 1997; Williams et al., 1996). First a blindreconstruction of the tip geometry was performed on theAFM images to characterise the geometry of a tip that ismost likely to have scanned the sample surface. Thereconstructed tip was then used to erode the images toobtain an upper bound surface on the actual samplesurface. PSL heights from the AFM and SEM projectedsurface diameters matched expected ranges with narrownormal distributions of standard deviation between0.008 and 0.015 μm. Probe dilations in the SEMmeasurements were insignificant because the width ofthe beam (of less than 5 nm) is diminutive compared tothe size of particles as expected. However, in the AFMmeasurements, tip dilation was observed to be substan-tial on all particle sizes measured. It should be notedtherefore that particle volumes calculated here representthe upper limits of the particle geometry, notably onsmall particles, while this effect should be of lessimportance on larger particles.

Analysis of aerosol particles on aluminium substratesusing an AFM can be complicated due to the roughnessof aluminium (Friedbacher et al., 1995). Three spotsvoid of particles were analysed for surface roughness onstage 8 of sample #1 (3.55 μm×3.55 μm and2.75 μm×2.75 μm) and on stage 7, sample #1(2.14 μm×2.14 μm). The average root mean squareroughness was found to be 12 nm, and about 10% of thesmallest surface diameter of particles that was measured.Therefore the roughness of the aluminium was con-cluded to be within the experimental error margin.Additionally, it is impossible to avoid dragging particleson substrates during scanning, especially those particlesthat deposit on top of other particles. These particles arenot as stable under the tip as those that are directlydeposited on the substrate. Images that showed suchevents were excluded from analyses.

4. Results

4.1. Particle morphology

A variety of particle morphologies were observed inthe AFM and SEM images. Morphologies of particlesbetween the two samples were clearly different,indicating different aerosol origins between the wetand dry seasons. On sample #1 (wet season sample),there was a high loading of biological particles on stages5 (nominal d50 = 1.0 μm) and stage 6 (nominald50=0.56 μm), e.g., brochosomes (Tulloch and Shapiro,1954; Wittmaack, 2005) shown in Fig. 1. Thesebrochosomes were hollow particles whose surfacediameters varied from 0.2 to 0.6 μm. There was also a

high concentration of fibrous particles on stages 5 and 6.If these fibrous particles were not hygroscopic and hadmaintained their morphology during sampling, theywere likely aligned perpendicular the settling velocity. Itis possible that they would bias surface diametersdetermined in microscope images.

Fig. 2 shows surface and topographic AFM imagesof aerosol particles from sample #1, stage 6. Apart fromthe sizes, this was the general morphology of mostparticles observed with the AFM on stages 7 (nominald50=0.32 μm) and stage 8 (nominal d50=0.18 μm). Inthe SEM vacuum, particles on stages 7 and 8 werehighly volatile on both samples, and no observationscould be made.

On sample #2 (dry season sample), stages 5 and 6,SEM images showed particles with haloes of volatilisedliquids. Haloes are formed from residues of what wereprobably salts and polar compounds dissolved in theaqueous component of particles. Particles within thehaloes consisted of soot inclusions, volatile euhedralcrystalline particles (volatile under SEM electron beam)and large stable spheroids of diameters up to 1 μm.Similar particle morphologies have been observed from

Fig. 2. AFM topographic image of aerosol particles on collected on stage 6 of sample #1. The image size is 3.89 μm×3.89 μm with a colour depthscale of 280 nm (0 black and 280 nm white). The 3D image is a shaded relief image of the surface topography, and is analogous to shining light onto a3D surface.

98 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

single particle analysis of biomass burning influencedaerosol (Li et al., 2003; Pósfai et al., 2003).

Particles were deposited along the longest axes, andtherefore averaged surface diameters, da, were alwayslarger than heights, ha, by factors of up to 25. Meanvalues of da/ha are listed in Table 1. The ratios decreasedwith increasing particle diameters, averaging 6.9 onstage 8 to 3.4 on stage 5 for sample #1. Sample #2showed a similar trend with larger ratios decreasingfrom 12.8 to 4.6. While both ratios are systematicallydecreasing with increasing particle diameters, theirdifferences point to different morphologies/chemicalcompositions of the AFM measured ‘dry’ particles andpossibly difference in source origins. Such high ratios ofda/ha have been observed elsewhere (Friedbacher et al.,1995). These authors pointed out that tip dilation alonecannot account for these high ratios. Jet velocities higherthan 100 m s−1 are encountered in the lower stages of aMOUDI impactor. Differences between heights anddiameters might be due to high jet velocities experi-enced upon impaction causing particles to deform asthey are pressed and spread on the substrate surface,particularly for hydrated particles.

Table 1Ratios of surface diameters to heights da /ha, and aspect ratio AR ofprojected surfaces

Stageno.

da/ha AR

Sample #1 Sample #2 Sample #1 Sample #2

5, AFM 3.4±2.1 4.6±2.2 1.51±0.87 1.18±0.185, SEM – – 1.33±0.65 –6, AFM 3.1±1.6 7.7±3.5 1.25±0.74 1.32±0.476, SEM – – 1.21±0.90 –7, AFM 4.3±1.9 11.0±5.6 1.21±0.15 1.20±0.198, AFM 6.9±7.1 12.8±4.6 1.31±0.35 1.23±0.20

Table 1 also lists particle aspect ratios, AR. Silhouettesof small particle were nearly round, with the amount offibrous particles increasing towards large particle stageson sample #1, as is indicated by the AR which increasesfrom 1.31 on stage 8 to 1.51 on stage 5.

4.2. Particle size distributions

AFMequivalent volume distributionswere normalisedto maximum frequencies on each stage and compared toexpected Stokes diameter distributions of each impactorstage. Absolute concentrations were not determined fromthe AFM and SEM distributions because the size of thedeposition spot cannot bemeasuredwith certainty in orderto derive total count of particles deposited at each spot.These counts are also likely biased by inhomogeneity ofthe deposition spot since particles tend to accumulateclose to the centre of the spot.

4.3. Sample #1

Fig. 3(a to d) shows number distributions ofindividual stages as functions of AFM equivalentvolume (dve) and Stokes diameters (d1.5) on sample#1. Parameters of lognormal functions fitted to thesedistributions are listed in Table 2. Distributions ofmeasured volume diameters, dve, span an order ofmagnitude on each stage, and are all lower than theequivalent Stokes diameters expected on individualMOUDI stages. These differences suggest mass loss onindividual particles. Sample #1 was collected under highhumidity conditions and particles were likely collectedas solution droplets. Losses of water and/or volatiles canbe expected, and only the residuals remain formicroscopy measurements.

Fig. 3. (a–d). Number distributions on individual stages as functions of AFM equivalent volume (dve) and Stokes diameters (d1.5) on sample #1,stages 8 to 5, respectively. The distributions were fitted with lognormal functions of parameters given in Table 2.

99P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

On sample #1, about 50% of collected particles onstages 5 and 6 were brochosomes. The pore spaces ofthese particles were calculated to be 30–40% of the total

Table 2Geometric mean diameters derived from fitting lognormal functions to size

Sample #1

Stageno.

Nominal 50% cut-off MOUDI

d50 (μm) d1.5 (μm) σ1.5

5 1.0 1.25±0.01 1.61±0.035 a

6 0.56 0.64±0.002 1.22±0.037 0.32 0.38±0.01 1.15±0.028 0.18 0.24±0.001 1.11±0.02

Sample #2

Stageno.

Nominal 50% cut-off MOUDI DMPS-APS

d50(μm) d1.4(μm) σ1.4 d1.15(μm) σ

5 1.0 1.10±0.01 1.36±0.03 1.27±0.013 16 0.56 0.66±0.002 1.21±0.03 0.60±0.002 16 a 0.60±0.002 17 0.32 0.42±0.001 1.20±0.02 0.38±0.002 18 0.18 0.25±0.001 1.11±0.01 0.26±0.001 18

dve=Equivalent volume diameter projected in AFM images; da=averaged su1.4 and 1.5 g cm−3; d50=nominal single plate cut-off diameter.a Additional peak.b After tip deconvolution (see text).

particle volume, and increasing with the size of particles.Brochosomes may take up water by capillary condensa-tion in the pore spaces without necessarily growing, until

distributions in Figs. 3–5 (R2N0.78 in all the fits)

Volume equivalentdiameter

AFM surfacediameter

SEM surfacediameter

dve (μm) σve da (μm) da (μm)

1.00±0.02 1.39±0.10 1.41±0.03 1.09±0.110.35±0.02 1.36±0.14 0.46±0.11 –0.50±0.01 1.66±0.02 0.70±0.01 0.54±0.120.16±0.01 1.23±0.10 0.27±0.01 –0.12±0.01 1.44±0.12 0.18±0.01 –

Volume equivalentdiameter

AFM surfacediameter

1.15 dve(μm) σve da(μm)

.73±0.08 0.67±0.06 1.96±0.06 0.93±0.03

.13±0.04 0.60±0. 01 1.10±0.01 0.79±0.03

.13±0.04 0.36±0.01 1.24±0.08 –

.15±0.02 0.40±0.02 1.28±0.12 0.81±0.01

.11±0.02 0.31±0.01 1.33±0.03 0.67±0.020.28±0.01 b 1.37±0.17 b 0.60±0.02

rface diameter; dρ=Stokes diameter for particles densities of ρp=1.15,

100 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

the pores are filled. This might explain the narrow shiftbetween the expected and measured volumes on stage 6and on the first peak close to dve=1.0 μm on stage 5,stages where high concentrations of brochosomes werefound. The second peak on stage 5 at dve=0.4 μm is likelyto be a contribution of particles with a large solublefraction.

To compare average surface diameters between thetwo microscopes, particles on sample #1 stages 5 and 6were imaged first with the AFM and then the SEM.These distributions are plotted in Fig. 4. Note that here astraightforward averaged projected diameter from the2D projections da is used, which is directly availablefrom both images. On stage 6, AFM particle diametersshowed a narrower distribution with a peak atda=0.63 μm, and SEM diameter peaks are aroundda=0.25 μm and 0.54 μm. Lower diameters from theSEM are expected from a further mass loss of adsorbedwater and volatile compounds due to the high vacuumand possible thermal degradation by the electron beam.Assuming a uniform loss of mass around particles onstage 6, changes between AFM and SEM are equivalentto a reduction of 54% in individual particles' volume.Similar observations have been made elsewhere (Pósfaiet al., 1998). By measuring AFM sizes before and aftersubjecting ammonium sulphate particles to TEMvacuum for 5 min (with the electron beam switchedoff), the authors observed volume decreases of between30% and 75% on particles with aerodynamic diametersof 0.2 to 0.7 μm (their tip effects were within 5% ofmeasured surface diameters). Their particles weresampled at RH=83%, and under these conditions,particles were likely deliquesced and were sampled assolution droplets (Tang, 1996). It will be shown that ourparticles also contained significant amount of waterduring sampling. Distributions of da cannot be com-

Fig. 4. Distributions of surface diameters (da) measured with AFM and SEM oby losses of water and volatiles in the vacuum.

pared to those of d1.5 because 2D surface diameters donot have a direct physical meaning, particularly sincethe particles were deformed upon impact as discussedbefore.

4.4. Sample #2

Fig. 5(a to d) shows number distributions of individualstages as functions of AFM equivalent volume (dve) andStokes diameters (d1.4 for the MOUDI distributions andd1.15 for the DMPS-APS distributions) on sample #2.Parameters of the lognormal fits are listed in Table 2. Onstages 7 and 8, physical volume of particles compared verywell with expected sizes. The fact that expected sizedistributions of ambient particles could be reproduced inAFM measurements indicates that small particles did notchange their morphology after collection as they exhibitrelatively less water absorption (Winkler, 1988). On stages6, distribution of dve shows two peaks, one of particleswithin the expected size range where dve=0.60 μm andmuch smaller particles with dve=0.36 μm. These peaksmight correspond to two particle species with differenthygroscopic properties.

Projected surfaces of particles on stage 8, sample #2were corrected for tip dilation with the algorithmdescribed before. After tip correction, particle heightsremained the same (mode of 0.074±0.006 μm beforeand after tip correction, and not altered as expected),while the geometric mean dve was reduced by 10% from0.31 μm to 0.28 μm (and from 0.67 μm to 0.60 μm forda). Confidence in the convolved output (measured asthe probability of a surface being probed by just the tipapex) was low on projected edges of particles unless oneopted for an unrealistically sharp tip. Since these are theareas of interest in our measurements, tip deconvolutionwas not applied to the rest of the samples. Though tip

n stages 5 and 6 of sample #1. Sizes of particles are reduced in the SEM

Fig. 5. (a–d). Number distributions on individual stages as functions of AFM equivalent volume (dve) and Stokes diameters (d1.4 for the MOUDIdistributions and d1.15 for the DMPS-APS distributions) on sample #2, stages 8 to 5, respectively. Distributions were fitted with lognormal functionsof parameters given in Table 2.

101P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

effects can be removed from images, there still remainsuncertainty around particle edges and further investiga-tions on tip dilation still need to be carried out.

5. Discussion and conclusions

There was no concurrence in size distributionsbetween physical sizes defined by microscopy techniquesand aerodynamic size of particles. Except on sample #2,stages 7 and 8, the AFM volume distributions did notadequately reproduce expected distributions. Equivalentvolume diameters were smaller than Stokes diameters byfactors up to 3.6. These observations indicate significantloss of particle masses, otherwise unrealistically highspecific weights would be needed to explain theoccurrences of such particles on these stages. In thesubsequent paragraphs, we will discuss the samplingconditions and observed mass losses, and explain howthey influenced size measurements.

Firstly, the relative humidity during sampling wasbetween 80% and 100% on sample #1 and averaged65% on sample #2. In these high humidity conditions,hygroscopic particles absorb water, with the uptake ofwater depending on the soluble fraction and size ofparticles. In the wet season, the Amazonian fine modeaerosol mass is primarily composed of 40% solublebiogenic emissions and 15% soluble inorganic salts,

predominantly ammonium bisulphate (Roberts et al.,2002). A soluble fraction of 5–10% sulphate wasobserved to be sufficient in explaining growth ofactivated CCN (Cloud Condensation Nuclei) in thesize range of dry diameter of 0.05bdpb0.5μm (Rissleret al., 2005; Roberts et al., 2002). Our particles werelikely sampled with high water content and weretherefore spherical shaped. This implies that particlebounce between stages and effects of particle shape onthe drag coefficient were negligible.

The particles changed their morphology during orafter deposition on aluminium substrates. Distortionfrom spherical particles by flattening can easily be seenfrom observed differences between surface diametersand heights of particles. The liquid component ofparticles might have evaporated after deposition andwhat we are now measuring are the dehydratedcomponents of aerosol particles, at less than 10%relative humidity in the SEM and ambient laboratoryrelative humidity (about 50%) in the AFM. However, itis not possible to confirm whether the particles hadrecrystallised, or still contained some adsorbed water inequilibrium with the ambient air during AFM analyses.

SEM images showed large haloes around most ofparticles suggesting the presence of a liquid, whichaffected the aluminium substrates. Fig. 6 shows such atypical particle, from sample #2 on stage 6. In this

102 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

image, there are smaller particles in the halo, which hadeither recrystallised and separated from the main particleduring evaporation of water or disintegrated onimpaction. The averaged surface diameter of the halois da=5.70 μm; for the large particle (treated as anagglomerate) da=1.70 μm; and for the small particles dais about 0.37 μm. The haloes can be identified in theSEM images because of the large field of view andcontrasting colour depths depending on the topographyand chemical composition. However, in the AFMtopography mode used in this study, haloes do notprovide enough height contrast to be identified. It ispossible that smaller particles in haloes were treated asdistinct individual particles, further biasing size dis-tributions to smaller sizes. Losses of water are high asindicated by large haloes in the SEM images.

The deviations between wet DMPS-APS size distribu-tions and MOUDI size distributions are small, as wasexpected because of the low growth factors. Humiditygrowth factors were 1.02 and 1.06 for dry diameters of0.020μmand 3.0μm, respectively at RH of 65% and 1.07to 1.25 at RH of 90% (Rissler et al., 2005). There were nohygroscopic growth measurements during the collectionof sample #1, but representative measurements of the wetseason aerosol from March to April 1998 at Balbina,Manaus (1°56′S, 59°25′W) showed averaged growthfactors for dry diameters of 0.035 to 0.265 μm to be 1.16to 1.32 at RH of 90% (Zhou et al., 2002). When weconsider ratios dρ /dve in this study, the changes in particlevolumes reflect higher water content on particles (theseratios represent expected diameters/AFM measureddiameters, though they are not a direct measure of growthwith humidity). Losses of water higher than expected canbe attributed to increased hygroscopicity by the formation

Fig. 6. SEM image of particles with a large halo of residue formed bywhat were probably inorganic salts dissolved in the aqueouscomponent of the particles. The averaged surface diameter da of thehalo is 5.7 μm and the larger agglomerated particle da is about 1.7 μm.

of organic coatings on aqueous aerosol particles (Buseckand Pósfai, 1999). The ratios on wet season sample #1 areall higher than on dry season sample #2, consistent withobserved higher hygroscopic growth factors in the wetseason. The ratios dρ /dve, are also higher for largerparticles, consistent with the dependence of humiditygrowth on the size of dry particles.

There are limitations we have not considered incomputing dve. Equivalent volume diameters werecalculated by assuming ellipsoidal hemispheres asprojected volumes in AFM images. This assumption isvalid for particles with smooth and rounded surfaces onboth projected surface silhouettes and height profiles.The equivalent volume for non-spherical particles likecylinders (e.g. fibres) and prisms (minerals and salts) areunderestimated when calculated from ellipsoidalvolumes of the same dimensions. Such effects areambiguous on amorphous particles. The second as-sumption made was that particles were compact, andthey had deposited firmly on substrates without voids.Particle volumes are therefore overestimated if therewere gaps between particles and substrates. Largerparticles were probably systematically undercountedbecause they were likely concentrated in the piled upcentre, which was avoided in order to image isolatedparticles. Additionally, samples #1 and 2 were collectedfor 70 and 10 h, respectively, and over such long timeintervals variations in aerosol compositions, e.g., due tochanges in meteorological conditions, are likely tooccur. In the Amazonia, water-soluble inorganic aerosolspecies were observed to exhibit strong diurnal varia-tions due to changes in micrometeorological conditions(Trebs et al., 2004) and also to a less extent the elementalcomposition of the fine mode aerosol (Guyon et al.,2003). Due to long sampling times, our measurementsmight be generalising particles with different origins andchemical compositions.

Despite these possible uncertainties, our observationsshow systematic shortcomings inherent to size or massmeasurements of collected hygroscopic particles. Firstly,particle sizes are dependent on: definition of ‘size’projected in images, i.e., projected volume in 3D orsurface area in 2D; and sampling conditions under whichparticles were collected. Microscopy sizes depend on therelative humidity history of particles before and aftersampling. Sizing discrepancies can occur when micros-copy sizes are compared with online sizing techniques.Although microscopy techniques have become effectiveand powerful tools in atmospheric research (Fletcheret al., 2001), this study shows there is clear misrepresen-tation of microscopy particle sizes when there are masslosses, even in the AFM ambient state. We cannot

103P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

determine how much observed mass distributions of theMOUDI were also underestimated by these losses, sincewe do not know when losses occurred. It should be notedhere, due to an erroneous size assessment, a mismatch ofthe aerosol size/composition distributions can occur andthus attribution of elements or chemical compounds tospecific size modes may be distorted. There is therefore aneed to conduct closure studies to assess and quantifyuncertainties on impactor distributions since particle massdistributions and number-size distributions are widelyderived from collected and stored samples. One suchprocedure could be the measurement of aerosol propertieswith particle number counters in parallel with an impactoroperating at varying humidity conditions. As seen here,reconciling mass losses with humidity growth factorsmight not be enough to account for water losses. Thisreconciliation is further complicated by the multimodalsize distributions seen on individual stages, indicative ofdifferent hygroscopic growth factors of particles collectedeven on a single stage.

Adsorbed water on particles also alters morphologiesof particles deposited on impactor substrates. While thiseffect might not be true for other sampling devices,particles in impactors experience high jet velocitieswhich flatten and deform particles on deposition.Results are evident on 3D projection of particles in theAFM, rather than 2D SEM surface projections. In thisstudy, average ratios of lateral diameters to particleheights were between 3 and 13, ratios that indicatesignificant deformities for particles that had beencollected as droplets. Because of these deformations,diameters derived from 2D projected surfaces tend tooverestimate particles sizes. This is also true forparticles that tend to orient on substrates like fibresand flat dust particles (Reid et al., 2003). Certainly,morphological parameters like sizes, aspect ratios andcircularity determined from 2D projections of suchparticles do not represent geometrical properties of theoriginal particles.

Acknowledgements

P. Gwaze would like to acknowledge the scholarshipprovided by the Max Planck Society (MPG). Part of thiswork was carried out within the framework of theSmoke, Aerosol, Clouds, Rainfall, and Climate(SMOCC) project, a European contribution to theLarge-Scale Biosphere-Atmosphere Experiment inAmazonia (LBA). It was financially supported by theEnvironmental and Climate Programme of the EuropeanCommission (contract No. EVK2-CT-2001-00110SMOCC), the Max Planck Society, the Fundação de

Amparo à Pesquisa do Estado de São Paulo, and theConselho Nacional de Desenvolvimento Científico(Instituto do Milênio LBA). We thank all members ofthe LBA-SMOCC and LBA-RACCI Science Teams fortheir support during the field campaign especially A. C.Ribeiro, M. A. L. Moura and J. von Jouanne.

References

Artaxo, P., Martins, J.V., Yamasoe, M.A., Procópio, A.S., Pauliquevis,T.M., Andreae, M.O., Guyon, P., Gatti, L.V., Leal, A.M.C., 2002.Physical and chemical properties of aerosols in the wet and dryseasons in Rondônia, Amazonia. Journal of Geophysical Research107, 8081. doi:10.1029/2001JD000666.

Barkay, Z., Teller, M.A., Ganor, E., Levin, Z., Shapira, Y., 2005.Atomic force and scanning electron microscopy of atmosphericparticles. Microscopy Research and Technique 68, 107–114.doi:10.1002/jemt.20241.

Baron, P.A., Willeke, K. (Eds.), 2001. Aerosol Measurement:Principles, Techniques, and Applications. Wiley-Interscience,New York, p. 1160.

Buseck, P.R., Pósfai, M., 1999. Airborne minerals and related aerosolparticles: effects on climate and the environment. Proceeding of theNational Academy of Sciences of the USA, vol. 96, pp. 3372–3379.

Dong, Y., Hays, M.D., Smith, N.D., Kinsey, J.S., 2004. Invertingcascade impactor data for size-resolved characterization of fineparticulate source emissions. Journal of Aerosol Science 35,1497–1512.

Dongmo, S., Troyon, M., Vautrot, P., Delain, E., Bonnet, N., 1996.Blind restoration method of scanning tunneling and atomic forcemicroscopy images. Journal of Vacuum Science and Technology B14, 1552–1556.

Dzubay, T.G., Hasan, H., 1990. Fitting multimodal lognormal sizedistributions to cascade impactor data. Aerosol Science andTechnology 13, 144–150.

Fletcher, R.A., Small, J.A., Scott, J.H.J., 2001. In: Baron, P.A.,Willeke, K. (Eds.), Analysis of Individual Collected Particles.Wiley-Interscience, New York, pp. 295–363.

Friedbacher, G., Grasserbauer, M., Mesimani, Y., Klaus, N.,Higatsberger, M.J., 1995. Investigation of environmental aerosolby atomic force microscope. Analytical Chemistry 67, 1749–1754.

Guyon, P., Graham, B., Roberts, G.C., Mayol-Bracero, O.L.,Maenhaut, W., Artaxo, P., Andreae, M.O., 2003. In-canopygradients, composition, sources, and optical properties of aerosolover the Amazon forest. Journal of Geophysical Research 108,4591. doi:10.1029/2003JD003465.

Hinds, W.C., 1999. Aerosol Technology: Properties, Behavior, andMeasurements of Airborne Particles. Wiley-Interscience, NewYork, p. 483.

IPCC, 2001. In: Houghton, J.T., Ding, Y., Griggs, D.J., Noguer, M.,van der Linden, J., Dai, X., Maskell, K., Johnson, C.A. (Eds.),Climate Change 2001: The Scientific Basis. Contribution ofWorking Group I to the Third Assessment Report of theIntergovernmental Panel on Climate Change. Cambridge Univer-sity Press, Cambridge, United Kingdom.

Kandlikar, M., Ramachandran, G., 1999. Inversion methods foranalysing aerosol spectrometer measurements: a critical review.Journal of Aerosol Science 30, 413–437.

Köllensperger, G., Friedbacher,G., Kotzick, R., Niessner, R.,Grasserbauer,M., 1999. In-situ atomic force microscopy investigation of aerosols

104 P. Gwaze et al. / Atmospheric Research 86 (2007) 93–104

exposed to different humidities. Fresenius' Journal of AnalyticalChemistry 364, 296–304.

Li, J., Pósfai, M., Hobbs, P.V., Buseck, P.R., 2003. Individual aerosolparticles from biomass burning in southern Africa: 2. Compositionand aging of inorganic particles. Journal of Geophysical Research108, 8484. doi:10.1029/2002JD002310.

Marple, V.A., Rubow, K.L., Behm, S.M., 1991. A MicroorificeUniform Deposit Impactor (MOUDI): description, calibration, anduse. Aerosol Science and Technology 14, 434–446.

Pósfai, M., Xu, H., Anderson, J.R., Buseck, P.R., 1998. Wet and drysizes of atmospheric aerosol particles: an AFM–TEM study.Geophysical Research Letters 25, 1907–1910.

Pósfai, M., Simonics, R., Li, J., Hobbs, P.V., Buseck, P.R., 2003.Individual aerosol particles from biomass burning in southernAfrica: 1. Compositions and size distributions of carbonaceousparticles. Journal of Geophysical Research 108, 8483. doi:10.1029/2002JD002291.

Reid, J.S., Jonsson, H.H., Maring, H.B., Smirnov, A., Savoie, D.L.,Cliff, S.S., Reid, E.A., Livingston, J.M.,Meier,M.M., Dubovik, O.,Tsay, S.-C., 2003. Comparison of size and morphologicalmeasurements of coarse mode dust particles from Africa. Journalof Geophysical Research 108, 8593. doi:10.1029/2002JD002485.

Rissler, J., Vestin, A., Swietlicki, E., Fisch, G., Zhou, J., Artaxo, P.,Andreae, M.O., 2005. Size distribution and hygroscopic propertiesof aerosol particles from dry-season biomass burning in Amazonia.Atmospheric Chemistry and Physics Discussions 8149–8207.

Roberts, G.C., Artaxo, P., Zhou, J., Swietlicki, E., Andreae, M.O.,2002. Sensitivity of CCN spectra on chemical and physicalproperties of aerosol: a case study from the Amazon Basin. Journalof Geophysical Research 107, 8070. doi:10.1029/2001JD000583.

Seinfeld, J.H., Pandis, S.N., 1998. Atmospheric Chemistry andPhysics: From Air Pollution to Climate Change. Wiley-Inter-science, New York, p. 1326.

Tang, I.N., 1996. Chemical and size effects of hygroscopic aerosols onlight scattering coefficients. Journal of Geophysical Research 101,19245–19250.

Todd, B.A., Eppell, S.J., 2001. A method to improve the quantitativeanalysis of SFM images at the nanoscale. Surface Science 491,473–483.

Trebs, I.,Meixner, F.X., Slanina, J.,Otjes, R., Jongejan, P.,Andreae,M.O.,2004. Real-time measurements of ammonia, acidic trace gases andwater-soluble inorganic aerosol species at a rural site in the AmazonBasin. Atmospheric Chemistry and Physics 4, 967–987.

Tulloch, G.S., Shapiro, J.E., 1954. Brochosomes and leafhoppers.Science 120, 232.

Villarrubia, J.S., 1997. Algorithms for scanned probe microscopeimage simulation, surface reconstruction and tip estimation.Journal of Research of the National Institute of Standards andTechnology 102, 425–454.

Williams, P.M., Shakesheff, K.M., Davies, M.C., Jackson, D.E.,Roberts, C.J., Tendler, S.J.B., 1996. Blind reconstruction ofscanning probe image data. Journal of Vacuum Science andTechnology B 14, 1557–1562.

Winkler, P., 1988. The growth of atmospheric aerosol particles withrelative humidity. Physica Scripta 37, 223–230.

Winklmayr, W., Wang, H.-C., John, W., 1990. Adaptation of theTwomey algorithm to the inversion of cascade impactor data.Aerosol Science and Technology 13, 322–331.

Wittmaack, K., 2005. Brochosomes produced by leafhoppers—awidely unknown, yet highly abundant species of bioaerosols inambient air. Atmospheric Environment 39, 1173–1180.

Zhou, J., Swietlicki, E., Hanson, H.C., Artaxo, P., 2002. Submicrom-eter aerosol particle size distribution and hygroscopic growthmeasured in the Amazon rain forest during the wet season. Journalof Geophysical Research 107, 8055. doi:10.1029/2000JD000203.