Introduction
The southeastern United States has experienced neutral to cooling shifts in
regional climate over the past century (Portmann et al., 2009; Saxena and Yu,
1998), in contrast to warming observed in the rest of the United States. This
has been partially attributed to increased formation of secondary organic
aerosol (SOA) with largely cooling effects due to efficient light scattering
and activity in cloud formation (Goldstein et al., 2009; Portmann et al.,
2009). Regionally, the main SOA source is oxidation of biogenic volatile
organic compounds (BVOCs), followed by condensation or reactive uptake onto
preexisting particles containing sulfate, nitrate, and ammonium (Anttila et
al., 2007; Boyd et al., 2015; Budisulistiorini et al., 2015a; Carlton et al.,
2010; Chameides et al., 1988; Hodas et al., 2014; Lee et al., 2010; Nguyen et
al., 2015; Weber et al., 2007; Xu et al., 2015a). Most studies of aerosol
climate impacts in the southeast have focused on the effects of SOA, as this
region has high concentrations of organic carbon, which, combined with
ammonium sulfate, contribute 60 %–90 % of fine particulate matter
(PM2.5) (Attwood et al., 2014; Boone et al., 2015; Cerully et al., 2015;
Nguyen et al., 2014). However, despite the importance of SOA, the mixing of
secondary species (SOA, sulfate, nitrate, etc.) with primary particles is not
fully known, particularly for forested locations impacted by regional
anthropogenic emissions. The distribution of chemical species in individual
particles across a population, i.e., mixing state, is critical for
climate-relevant properties including light scattering, water uptake,
particle acidity, and aerosol chemistry, such as metals dissolution (Artaxo
and Orsini, 1987; Ault et al., 2010; Cong et al., 2010; Craig and Ault, 2018;
Fang et al., 2017; Kunkel et al., 2012; Metternich et al., 1986; Violaki and
Mihalopoulos, 2010; J. Z. Xu et al., 2015). Therefore, it is important to
identify the sources of aerosol particles present in the southeastern United
States, as well as their size and mixing state, in order to accurately assess
their impact on aerosol direct and indirect effects (Li et al., 2016b; Moise
et al., 2015; Posfai and Buseck, 2010).
Mixing state is described in terms of external and internal mixtures: an
external mixture consists of particles that contain only one pure species
per particle, while an internal mixture describes particles that contain
equal amounts of all chemical species (Ault and Axson, 2017; Posfai and
Buseck, 2010; Riemer and West, 2013). The mixing states of ambient aerosol
populations are complex and can vary as a function of size, altitude, and
particle age (Ault et al., 2013a, 2009; Fu et al., 2012;
Healy et al., 2014a; Moffet et al., 2010b; Pratt and Prather, 2010). Aging –
or atmospheric processing such as coagulation, condensation of secondary
species, and heterogeneous reactions – leads to internal mixing, while freshly
emitted particles are more externally mixed (Schutgens and Stier, 2014;
Weingartner et al., 1997). Here, mixing state is used to describe the
distribution of chemical species in a population and is purely based on
composition (Ault and Axson, 2017). This does not including
particle morphology, phase, coatings, or other physical properties, which can
subsequently impact composition (Y. Zhang et al.,
2018). Although the representation of mixing state in models is still an
open research question (Riemer and West, 2013), an
appropriate description of mixing state is critical for modeling optical
properties (Chung and Seinfeld, 2005; Jacobson, 2001) and cloud
condensation nuclei (CCN) activity (Zaveri et al., 2010).
Riemer and West (2013) introduced the mixing state
index (χ) to quantify aerosol mixing state. This parameterization uses
single-particle mass fractions of individual components to calculate the
average particle-specific diversity and the bulk population diversity, from
which the mixing state index can then be determined. This methodology has
been applied to a handful of laboratory and field studies, to date. In the
laboratory, Dickau et al. (2016) used aerosol sizing and mass
instrumentation to quantify the volatile mixing state of soot.
Single-particle mass spectrometry data from field studies in London
(Giorio et al., 2015) and as part of the MEGAPOLI campaign in
Paris (Healy et al., 2014b) found mixing
state was dependent upon both time of day and air mass origin. Similarly,
mixing state parameters were applied to computer-controlled scanning
electron microscopy with energy-dispersive X-ray spectroscopy (CCSEM-EDX) and
scanning transmission X-ray microscopy–near-edge X-ray absorption fine
structure spectroscopy (STXM-NEXAFS) during the Carbonaceous Aerosol and
Radiative Effects Study (CARES) in the Central Valley of California (O'Brien et
al., 2015) and in the Amazon during the GoAmazon campaign (Fraund et
al., 2017). These studies found changes in mixing state were associated with
a buildup of organic matter and particle clusters were less diverse at
remote sites, respectively. However, additional studies are needed to
quantify the chemical mixing state of aerosols, particularly for rural
locations.
In this study, we analyzed individual atmospheric particles collected at a
rural location influenced by regional pollution in the southeastern United
States during the 2013 Southern Oxidant and Aerosol Study (SOAS) to identify
their size-resolved chemical composition and mixing state. CCSEM-EDX was
used to determine the size, elemental composition, and number fraction of
particles containing metal (nonvolatile) cations. STXM-NEXAFS was used to
characterize chemical bonding of carbonaceous components, specifically
distinguishing soot from organic carbon. Mass estimates of particle
elemental composition from CCSEM-EDX were calculated using a modified
version of the method from O'Brien et al. (2015) to quantify the mixing
state parameters for both submicron and supermicron particles during time
periods dominated by SOA/sulfate, dust, and sea spray aerosol (SSA).
Additionally, the variability in the mixing state index during
these three time periods of interest showed that submicron aerosol was
most internal when SOA particles dominated the aerosol population. However,
supermicron particles were most internally mixed when a single source (e.g.,
SSA or dust) dominated the population and most externally mixed when SOA was
present, along with SSA and dust. The variety of particle classes, varying
extent of secondary processing, and diverse chemical mixing states at this
rural, forested site may impact climate-relevant properties of aerosols in
the southeastern United States.
SEARCH filter sample data for Centreville, AL, during SOAS
with purple boxes overlaid for time periods in which CCSEM was run and the
corresponding MOUDI stages that were analyzed. SOA-rich periods denoted with
green boxes (14–17 June and 7–11 July 2013) were studied by
Xiong et al. (2015), Pye et al. (2015),
Xu et al. (2015b), Hu et al. (2015), and
Rattanavaraha et al. (2016);
dust-rich periods marked with brown boxes (12–13 June and 26–28 June 2013)
were identified by Allen et al. (2015); and
SSA-rich periods marked with blue boxes (10–11 June and 3–6 July 2013) were
identified by Bondy et al. (2017b). Time periods
without samples analyzed are due to sample damage and identification of
mutually exclusive time periods. Based on periods identified in previous
studies, only 29 % of SOA-dominant periods, 40 % of dust periods, and
51 % of SSA periods were analyzed here.
Experimental
Aerosol sample collection
Samples of atmospheric particles were collected at the SOAS Centreville, AL,
site (32.9030∘ N, 87.2500∘ W; 242 m a.m.s.l.) between 5 June and 11 July 2013
(Bondy et al., 2017b; Carlton et al., 2018; Hidy et al., 2014). The
site was located in a rural, forested region near Talladega National Forest,
at a location that was part of the SouthEastern Aerosol Research and
Characterization Network (SEARCH). Meteorological and filter sample data
analyzed from the SEARCH network were used to aid selection of samples for
analysis (Figs. 1 and S1 in the Supplement). Particles were collected near ground level (1 m)
using a micro-orifice uniform deposit impactor (MOUDI, MSP Corp. Model 110)
sampling at 30 L min-1 with a PM10 cyclone (URG Model 786) to exclude
particles larger than 10 µm. The 50 % size cut points for the
MOUDI used in this analysis had aerodynamic diameters (Da) of 3.2, 1.8,
1.00, 0.56, 0.32 0.18, 0.10, and 0.056 µm (Marple et
al., 1991). Throughout SOAS, particles were impacted onto Cu 200 mesh transmission electron microscopy
(TEM) grids with Carbon Type B thin film (Ted Pella Inc.) for analysis with
SEM-EDX and STXM-NEXAFS. Substrates in the MOUDI were collected daily from
8:00 to 19:00 and 20:00 to 7:00 CST (with 1 h for substrate exchange),
except during intensive periods from 10 to 12 June, 14 to 16 June, 29 June to 1 July, and 7 to
9 July, when the sampling schedule was 8:00–11:00, 12:00–15:00,
16:00–19:00, and 20:00–7:00 CST (Table S1). Samples were collected more
frequently during intensive time periods, which were determined by
meteorological and gas phase concentrations
(Budisulistiorini et al.,
2015b). In Fig. 1, the MOUDI stages analyzed using CCSEM are noted for
each sample. After collection, all substrates were sealed in Teflon-wrapped
petri dishes within Ziploc bags and stored at -22 ∘C. Upon arrival
at the University of Michigan, Ziploc bags containing samples were removed
from the freezer and thawed to room temperature, and TEM grids were transferred
to grid storage boxes wrapped with Parafilm. Parafilm-wrapped grid boxes
were then stored in Ziploc bags in the freezer until imaging. Prior to
imaging the wrapped grid boxes were removed from the freezer and allowed to
reach room temperature before being opened. Although all samples were stored at
low temperatures after collection to minimize chemical reactions, the loss
of semivolatile species (e.g., water and organics) in the MOUDI and in the
SEM under vacuum may have led to the EDX results suggesting particles
contained slightly less organic carbon than in the ambient atmosphere, which
might make them appear slightly more externally mixed.
CCSEM-EDX analysis
Particles on MOUDI stages 3–10 (Da=0.056–1.8 µm, Fig. 1)
were analyzed using CCSEM (FEI Quanta environmental SEM) equipped with a
field emission gun operating at 20 kV and a high-angle annular dark-field
(HAADF) detector (Laskin et al., 2002, 2006, 2012). The SEM was equipped with an EDX spectrometer (EDAX, Inc.),
which was used to quantify X-rays of elements with atomic numbers
≥C(Z=6). A total of ∼34000 particles
were analyzed during time periods denoted in Table S2, which constitute a
representative cross section of the campaign. CCSEM analysis captured
particle physical parameters including projected area diameter, projected
area, and perimeter. Projected area diameter, which is equivalent to the
diameter of a circle with the same area as the particle silhouette, is
typically larger than aerodynamic diameters measured by other analytical
techniques (Bondy et al., 2017a; Hinds, 1999). For a
more accurate representation of particle size, projected area diameters were
converted to volume-equivalent diameter using a conversion factor of 0.49
for SOA/sulfate and biomass burning particles and 0.66 for SSA, determined
from atomic force microscopy (AFM) volume calculations of particles from
SOAS (Tables S3 and S4). EDX spectra from individual particles were analyzed
to determine the relative abundance of 14 elements: C, N, O, Na, Mg, Al, Si,
P, S, Cl, K, Ca, Ti, and Fe. Note that the Cu signal in the EDX spectra is
primarily due to the Cu grid from the substrate and was not included in
CCSEM-EDX analysis.
The CCSEM-EDX data sets were analyzed using k-means clustering of the
elemental composition following the method described in Ault et al. (2012) using codes written in MATLAB R2013b (MathWorks, Inc.).
Clusters were grouped into source-based classes by elemental composition
based on previous studies, including mineral dust (Axson et al., 2016b;
Coz et al., 2009; Creamean et al., 2016; Laskin et al., 2005; Sobanska et
al., 2003), SSA (Ault et al., 2013a; Bondy et al., 2017b; Hopkins et al.,
2008; Laskin et al., 2002; Prather et al., 2013), SOA/sulfate (Moffet et
al., 2013; O'Brien et al., 2015; Sobanska et al., 2003), biomass burning
aerosol (Li et al., 2003; Posfai et al., 2003), fly ash/metals (Ault
et al., 2012; Chen et al., 2012; Shen et al., 2016), biological particles
(Huffman et al., 2012), and fresh soot (Li et al., 2003). Soot forms
fractal aggregates of graphitic carbon (C) which contain tens to hundreds of
small spherical aggregates (Li et al., 2003). However, the intense carbon
signal from the carbon film substrate made chemical identification of soot
difficult, initially resulting in false positives from the substrate.
Because of their unique morphology, the size distribution of fresh soot
particles without a large secondary organic carbon coating altering the
fractal morphology was manually determined. Then, a scaling factor based on
the SEARCH network elemental carbon mass concentrations was applied to the
size distribution, and this factor was used in the subsequent analysis. More
information on this correction for soot can be found in the Supplement, specifically Table S5.
Mass calculations and mixing state parameters
Mole percent of elements analyzed using CCSEM-EDX was converted to mass
fractions using the method described by O'Brien et al. (2015) and detailed in
the Supplement. Briefly, particle volumes were calculated from volume-equivalent
diameters (calculated from projected area diameters using a conversion
factor determined from atomic force microscopy height images) assuming the
volume of a hemisphere. Particle masses were then calculated (μi= density × volume) assuming the following densities for each class:
1.3 g cm-3 for SOA/sulfate, biomass burning aerosol, and primary biological
particles (Li et al., 2016a; Manninen et al., 2014; Nakao et al., 2013);
2.0 g cm-3 for SSA particles (O'Brien et al., 2015); 2.6 g cm-3
for dust particles (Wagner et al., 2009); and 3.0 g cm-3 for fly ash particles (Buha et al., 2014). To
calculate the mass of each element, the elemental mole percent was converted
to a weight percent, which was multiplied by the estimated particle mass.
There are uncertainties in this conversion related to the material densities
and shapes assumed, as well as loss of semi-volatiles during analysis (water
and some organics), which lead to mass uncertainties on the order of
5 %–10 % on a per-particle basis.
Diversity parameters were calculated using two different methods in this
work: elemental diversity was calculated from CCSEM-EDX results similar to
O'Brien et al. (2015), and mixing state parameters due to aging were
calculated as described below (which use only two diversity species: the
mass fraction of elements associated with externally mixed particles and the
mass fraction of secondary species). To calculate elemental diversity
parameters, the mixing entropy of each particle (Hi) and average particle
mixing entropy (Hα) were calculated for each particle class as
described in detail by Riemer and West (2013):
Hi=∑a=1Apaiilnpia,Hα=∑i=1NpiHi,
where pi is the mass fraction of particle i in the population and
pia is the mass fraction of element a in particle i. The particle
diversity (Di) was then calculated by taking the exponent of the
particle-specific entropy Hi, and the average particle-specific
diversity (Dα) was calculated by taking the exponent of Hα. Dα was used as an indicator of elemental diversity for each
particle class: SOA/sulfate, biomass burning particles, fly ash, dust, SSA,
and biological particles.
In addition to elemental diversity, diversity parameters were calculated to
quantify the extent of particle aging. To calculate the mixing state aging
parameters for the three time periods of interest, two final mass values
were calculated: the mass of single particles in a class based on the sum of
elements characteristic to that class, and the mass of secondary species.
The elemental mass fractions as a function of size are depicted in Fig. S3. Due to the semi-quantitative nature of the lower Z elements
(Laskin et al., 2006) and substrate interferences, we
excluded C, N, and O from mixing state calculations, similar to O'Brien et al. (2015). The mass associated with SOA/sulfate was solely accounted for
by S (if present) and therefore was either ignored or severely
underestimated. Fresh biomass particles consisted of K and Cl; fly ash
particles contained Al and Si; unreacted dust particles consisted of Na, Mg,
Al, Si, K, Ca, Ti, and Fe; fresh SSA particles contained Na, Mg, Cl, K, and
Ca; and biological particles contained P, Cl, and K. As a metric for aging,
all sulfur was assumed to be secondary within particles, though trace
primary sulfur is present in SSA and may be in other classes. Thus, for the
purposes of this mixing state analysis each particle contained between one
and two components: a primary source-based composition and secondary aging
due to sulfur. Using the mass fractions of only these two components,
Hi, Hα, and the population bulk mixing entropy (Hγ) were
calculated for each particle class:
Hγ=∑a=1A-palnpa,
where pa is the mass fraction of element a in the population. The bulk
population diversity (Dγ) was then calculated by taking the exponent
of Hγ. The mixing state index (χ) is then defined as
χ=Dα-1Dγ-1×100,
where χ can range from 0 % for an external mixture to 100 % for
an internal mixture.
STXM-NEXAFS analysis
STXM-NEXAFS measurements of two MOUDI samples (stage 8, 50 % size
cut point of 0.18 µm, 100–200 particles analyzed per sample), 10 June
and 7 July 2013, were performed at the carbon K-absorption edge (280–320 eV)
to characterize chemical bonding of carbonaceous components, specifically
distinguishing soot from organic carbon. An example STXM image, a map, and
NEXAFS spectra for a biological particle are provided in the Supplement. STXM-NEXAFS
images and spectra of SOA will be the focus of a future publication. STXM
was conducted at the Advanced Light Source at Lawrence Berkeley National
Laboratory on beamline 5.3.2. The operation of the microscope has been
explained in detail by Kilcoyne et al. (2003). The
software programs MATLAB and Axis 2000 were both used for spectral analysis
of the STXM-NEXAFS data as described by Moffet et al. (2010a, 2016).
Stacks of images taken at sequentially increasing photon energies were used
to obtain spatially resolved spectroscopic data at the carbon K-edge. For
organic identification, pixels were identified where the post-edge minus the
pre-edge (optical density (OD) at 320 eV minus OD 278 eV) was greater than
zero. For the inorganic component, particles with a ratio of the pre-edge to
the post-edge (OD 278 / OD 320) greater than 0.5 were identified. To
identify soot inclusions within particles, individual pixels of STXM images
were analyzed; if a pixel contained 35 % or greater C = C, a peak
which was identified as soot using graphitic carbon as a standard, then that
pixel was identified as a soot region. Additional details on identification
of SOA-containing soot inclusions are provided in the Supplement.
Results and discussion
Descriptions of particle classes during SOAS
Although SOAS took place in a rural, forested region frequently dominated by
SOA/sulfate, a variety of particle classes were observed. Based on the
chemical composition, seven main particle classes were identified:
SOA/sulfate, biomass burning particles, soot, and fly ash, which are
typically present in the submicron (< 1 µm) regime, and
mineral dust, SSA, and primary biological particles with characteristic
sizes > 1 µm. Figure 2 shows SEM images of representative
particles from each class and their corresponding EDX spectra. SOA/sulfate
particles were identified by the elemental composition of C and O, along
with S, N, or both S and N. As all SOA particles contained inorganic
species (e.g., ammonium sulfate) in addition to organic carbon (based on STXM
ODpre / ODpost=0.5, ∼20 % by weight), this
class is referred to as SOA/sulfate. EDX spectra of SOA/sulfate particles on
Si wafers (Fig. S4) confirmed C and O in the particles, as did STXM, as a
check due to the carbon film substrate on TEM grids that contributes to the
signal for C and O in the CCSEM-EDX analysis
(Kirpes et al., 2018). The presence of S and/or
N in addition to C and O is likely NO3- or SO42-, based
on Raman microspectroscopy (Ault et al., 2013b; Craig et al., 2017; Ebben
et al., 2013), or organonitrate or organosulfate compounds, which are
ubiquitous in the southeastern US (Ayres et al., 2015; Carlton et al.,
2009; Froyd et al., 2010; Hatch et al., 2011). SOA/sulfate particles were
typically circular (circularity ranging from 0.95 to 1, where 1 is perfectly
circular; equation in Supplement), though some SOA/sulfate exhibited liquid–liquid
phase separation (LLPS), such as core-shell or more complex morphologies,
which will be explained in a future publication. Biomass burning aerosol
particles were often circular as well (0.96 circularity), with high
concentrations of K and frequently S and Cl, in addition to C and O (organic
carbon) (Li et al., 2003; Posfai et al., 2003). Fly ash particles were
identified primarily by their spherical morphology (0.93 circularity) since
fly ash is generated through high-temperature processes (Ault et al.,
2012; Chen et al., 2012), in addition to high EDX signals from O along with
either Si or Al, likely in the form of SiO2 or Al2O3, respectively. A final class comprised primarily of submicron particles was
soot. Fresh soot particles were identified primarily by their morphology
consisting of agglomerated spheres, which had substantially smaller
diameters than fly ash (Li et al., 2003). However, fresh soot was not
very prevalent during SOAS and was typically present within other particles such
as SOA, which will be detailed below.
SEM images and corresponding EDX spectra for each of the main
particle classes identified during SOAS within the submicron
(a, SOA; b, biomass burning aerosol particles; c, fly ash; d, soot) and supermicron
(e, dust; f, SSA; g, primary biological) sizes. Note the elements with an
asterisk are not quantitative due to interference from the substrate or
detector.
In addition to submicron classes, three classes of particles predominantly
in the supermicron size range were observed during SOAS. Dust particles were
identified by strong signals from O, Al, and Si (aluminosilicates), often
along with other elements such as Na, Mg, K, Ca, Ti, and Fe (Coz et al.,
2009; Laskin et al., 2005; Sobanska et al., 2003). EDX spectra of SSA
particles contained a strong Na signal (Na+) and weaker Mg signal
(Mg2+) in a ∼10:1 ratio, as found in seawater
(Pilson, 1998); small contributions from K (K+) and Ca
(Ca2+); and were clearly distinguishable from freshwater aerosol
(Axson et al., 2016a; May et al., 2016, 2018b). Counter-ion
elements such as N, O, S, or Cl (NO3-, SO42-, or
Cl-) depended on whether the SSA was fresh or aged after transport
(Bondy et al., 2017b; May et al., 2018a). Finally, biological particles
typically contained primarily C due to organic macromolecules, along with
lesser amounts of N (likely in the form of amines/amino acids), O, P
(PO43-), and K (K+) (Huffman et al., 2012), as seen in
Fig. 2. Overall, numerous particle classes were detected during SOAS using
CCSEM-EDX based on unique chemical composition, morphology, and size.
STXM-NEXAFS was used to investigate carbonaceous particles since these
particles were prevalent during SOAS. The carbon K-edge was probed using this
technique, and high-spatial-resolution information was obtained regarding
sp2 C inclusions within SOA/sulfate, which were identified as soot at
285 eV (Moffet et al., 2010a). In two samples analyzed using
STXM-NEXAFS, 6.9 % and 9.9 % of particles by number contained sp2
C inclusions, suggesting that a small fraction of SOA/sulfate contained
soot. In comparison, organic carbon–elemental carbon (OC-EC) bulk analysis
by the SEARCH network detected ∼2 % elemental carbon by
mass, suggesting that, although little soot was present overall, a sizeable
fraction was present as small inclusions within SOA/sulfate. It is important
to consider the mixing state of aerosols when modeling radiative forcing in
the region, because internally mixed particles behave differently than pure
components. For example, soot coated with secondary organic material may
have an enhanced absorption compared to fresh soot or soot-less SOA, though
recent work has suggested that optical properties of coated soot are
challenging and nonlinear (Healy et al., 2015; Moffet et al., 2009;
Ramanathan and Carmichael, 2008; Zhang et al., 2008). These spectra
highlight that, although seven main particle classes were identified, many
particles, such as SOA and soot, were partially internally mixed.
Chemical diversity observed within particle classes
Using SEM-EDX elemental mapping, morphology and the spatial distribution of
species within individual particles were examined. In Fig. 3, particles
(a–d) were identified as dust based on their morphology and elemental
composition. However, only (b–d) are aluminosilicate dust particles; (a) contains high concentrations of Ca and S instead. Based on its chemical
composition, this dust particle is most likely gypsum
(CaSO4 ⋅ 2H2O) (Hashemi et al., 2011). The elemental map highlights
that elements present within the dust class are not homogeneously
distributed among all particles. Rather, the dust class consists of
externally mixed particles with varying compositions. In addition to dust,
two other particle classes are represented in the elemental map in Fig. 3.
Particles (e–f) were identified as aged SSA due to the high concentration of
Na and Mg along with S and N (likely SO42- and NO3-)
(Bondy et al., 2017b), and particle (g) is a primary
biological particle, possibly coagulated with a small calcium oxide particle
based on the morphology and distinctly different elemental compositions of
the two components. In addition to differentiating particles among the seven
particle classes identified, SEM-EDX mapping allowed investigation into
whether coagulation or chemical aging of particles has occurred within
particles due to the presence of localized regions of elements or
surface-layer coatings (Conny and Norris, 2011). As seen in Fig. 3, very few of the particles have a homogeneous distribution of elements,
though vacuum analysis and drying can modify the internal distribution of
species within particles. Rather, Na, Ca, S, and Cl often appear in only a
few distinct regions within particles, which can likely be attributed to
heterogeneous reactions, limited diffusion, or other non-ideal behavior
(Y. Zhang et al., 2018). The aluminosilicate dust
particle (b) in particular has localized regions of Ca and S
(SO42-) on the edges of the particle, signifying that this
particle has undergone aging, resulting in a more diverse physicochemical
mixing state (Ault and Axson, 2017). Complex mixing states
like this have been observed previously for SSA and dust, showing that these
classes of particles can be externally mixed or have surface coatings and
inclusions leading to internal mixing, thereby altering their physical and
chemical properties (Deboudt et al., 2012; Fitzgerald et al., 2015; Gantt
and Meskhidze, 2013; Kandler et al., 2011; Kim and Park, 2012; Sobanska et
al., 2012; Sullivan et al., 2007, 2009).
SEM image (dark field) and EDX elemental maps of particles
indicated that these particle classes had various mixing states. Each of the
elemental map panels corresponds to two elements overlaid to show the
elemental distributions from the SEM image. The following particle classes
are shown: (a–d) dust, (e–f) aged SSA, and (g) primary biological.
Average histograms and
digital color histograms of different particle classes from SOAS:
(a) SOA, (b) biomass burning, (c) fly ash,
(d) dust, (e) SSA, and (f) primary biological.
Average spectra are shown on the left as the average mole percent of each
element analyzed by CCSEM-EDX (C, N, O, Na, Mg, Al, Si, P, S, Cl, K, Ca, and
Fe). On the right, the digital color histogram heights represent the number
fraction of particles containing a specific element, and the colors represent
the mole percent of that element. The average particle specific diversity
(Dα), representing the average number of elements in each
particle, is calculated for each submicron and supermicron class and has an
uncertainty of 5 %–10 %. Note the elements with an asterisk are not
quantitative due to interference from the substrate or detector and are not
included in Dα.
To probe the chemical diversity of each particle class, Fig. 4 shows the
average EDX elemental percentages for each particle class. The digital color
histogram height shows the number fraction of particles in a class containing
a specific element, while the color represents the mole percent of the
element. For example, 100 % of SOA/sulfate by number contain between
5 % and 100 % C
(mole %). To quantify elemental diversity of particles, Dα,
representing the average number of elements within particles in each class,
was calculated. Dα ranges from 1 (when a particle contains only
one element) to A number of elements. Note that, due to interference from
the substrate or detector, C, N, and O were not included in Dα
calculations. CCSEM-EDX results suggest that SOA/sulfate particles were
elementally the least diverse, as the primary quantifiable element was S,
leading to a Dα=1.0. However, other studies from SOAS that used
an aerosol mass spectrometer (AMS) (Guo et al., 2015; Xu et al., 2015b) or
ultra-performance liquid chromatography/electrospray ionization
high-resolution quadrupole time-of-flight mass spectrometry
(UPLC/ESI-HR-QTOFMS) (Budisulistiorini et al., 2015a; Riva et al., 2016)
discovered that a wealth of sources contribute to SOA, resulting in
hydrocarbon-like organic aerosol, isoprene-derived organic aerosol, and
more-oxidized and less-oxidized oxygenated organic aerosol (Zhang et al.,
2018). Their analyses also showed that SO42- is the most
abundant component aside from organic carbon, with significant concentrations
of NH4+ followed by NO3-, consistent with our
observations of 92 % of SOA by number containing S and 68 %
containing N (a challenging element for EDX) (Budisulistiorini et al., 2015a;
Guo et al., 2015; Riva et al., 2016; Xu et al., 2017, 2015b).
The composition of biomass burning particles was elementally more diverse
than SOA (Dα=1.9), with large contributions from
K+ (1 %–30 % by mole percent), as well as organic carbon (20 %–100 %
C and 2 %–50 % O by mole percent). However, in addition to these three
components, approximately 60 % of particles by number also contained
SO42- (1 %–15 % S by mole percent), 45 % contained
NO3-/NH4+ (1 %–10 % N by mole percent), and 15 % by
number contained 1 %–30 % Cl (mole percent). The presence of Cl suggests that
some of the biomass burning particles were fresh. However, because sulfate
and nitrate, which are indicative of aging (Li et al., 2003), were
present more frequently, biomass burning particles detected during SOAS
likely had sufficient time during transport for multiphase reactions to
occur (Washenfelder et al., 2015). The final submicron particle class,
fly ash (Dα=1.9), showed two distinct compositions in
addition to aging: fly ash from SOAS consisted of primarily SiO2,
although approximately 15 % by number contained aluminum oxides with
silicates. Low concentrations of SO42- and NO3-
(1 %–5 % S and N by mole percent) suggest acidic species, such as sulfuric and
nitric acid, reacted with 25 % of fly ash by number.
Within the supermicron particle classes, a range of elemental compositions
were observed for dust and SSA (Fig. 4b). Dust was primarily composed of
aluminosilicates (10 %–100 % O, 1 %–50 % Si, and
1 %–50 % Al by mole percent), with minor contributions from other
chemical species, including CO32- or organic coatings
(5 %–100 % C by mole percent), NO3- (1 %–10 %
N by mole percent), SO42- (1 %–10 % S by mole percent),
Mg2+ (1 %–10 % by mole percent), K+
(1 %–5 % by mole percent), Na+ (1 %–10 % by mole
percent), Ca2+ (1 %–10 % by mole percent), and
Fe2+/Fe3+ (1 %–30 % by mole percent). The
frequency of these minor elements in dust particles varied widely, resulting
in a high average particle species diversity (Dα=4.4), with
nitrate present in approximately 75 % of the dust population by number
and titanium oxides present in less than 5 %. The diversity of dust
indicates various sources and processing throughout SOAS, which likely
contributed to time periods with distinct dust compositions due to wind
speed, wind direction, and pollution levels. Allen et al. (2015) detected two
high-nitrate events in the coarse-mode during SOAS, the first
corresponding to high levels of SSA and dust, and the second primarily dust.
The first event exhibited a higher percent of Na+, not all of which
was attributed to SSA due to the low Mg2+ / Na+ molar
ratio, while the second event had a higher percent composition of
Ca2+. Back trajectory analysis of the air mass origin during the
two coarse-particle events indicates that, although the overall pattern in
wind trajectories was similar, slight differences in wind patterns at the
beginning of each event may have contributed to the observed differences in
composition of the aerosol, suggesting a relatively local origin of the dust
during the second event. The elements of SSA were more homogeneously
distributed throughout the population than dust (Dα=2.9), with
100 % of particles by number containing C, O, and Na; 75 % by number
containing S; and ∼70 % by number containing > 1 %
(mole %) N and Mg. SSA particles also showed various degrees of aging
with respect to the anions, since chloride can be liberated through
multi-phase reactions with acidic species such as HNO3,
H2SO4, and organic acids (Bondy et al., 2017b). Partially aged
SSA comprised approximately 20 % of particles by number, indicated by
Cl- (1–10 % Cl by mole percent) in addition to nitrate and
sulfate. Complete chloride depletion and aging by nitrate (1 %–30 %
N by mole percent) and sulfate (1 %–30 % S by mole percent) was more
ubiquitous though, with each secondary species present in ∼90 % of
SSA by number. A thorough discussion of the degree of reactive processing of
SSA transported inland to Centreville can be found in Bondy et al. (2017b).
Size distributions for specific particle-rich time periods:
(a) SOA-rich periods (14–17 June and 7–11 July 2013), (b) dust-rich periods
(12–13 June and 26–28 June 2013), and (c) SSA-rich periods (10–11 June and
3–6 July 2013). SOA periods were dominant throughout the times when samples
were analyzed during SOAS (61 %), high-dust periods (24 %) (Allen et
al., 2015), and SSA periods (19 %) of the time. SSA periods were defined in
Bondy et al. (2017). * Literature-identified SSA periods and dust periods
overlap from 11 to 13 June 2013; thus the percentage of SOA, dust, and SSA
periods is greater than 100 % due to double counting of that time period.
Only particles with volume-equivalent diameters between 0.2 and 5 µm
are shown due to too few particles present at larger sizes for statistical
analysis.
Primary biological particles contained primarily organic carbon (50 %–100 %
C and 5 %–20 % O by mole percent) with minor contributions from
PO43-, SO42-, and K+, in addition to other minor
elements (Dα=6.2). The minor constituents (P, K, S)
were not detected in all particle (20 % by number). The absence of these
minor constituents from EDX spectra is likely the result of low
concentrations compared to carbon, and signal below the 1 % detection
threshold. Although sulfate is typically an indication of aging by
H2SO4 in aerosol particles, it is also naturally present in
biological particles. Furthermore, because the sulfur signal intensity is on
the same scale as the other minor constituents, it is not necessarily from
secondary processes. Overall, throughout both the submicron and supermicron
particle regimes, particle diversity varied, indicating sources of long- and
short-range transport, and various degrees of aging of particles within each
class.
Variations in particle classes observed during key SOAS events
Three main time periods (SOA, dust, and SSA) were identified during SOAS
that had distinctly different sources and processing (Fig. 5). Figure 5a
depicts the size-resolved chemical composition averaged over two
SOA-dominated time periods (14–17 June and 7–11 July 2013), two dust
events (11–13 and 26–28 June 2013), and two SSA events (10–11 June 2013 and 3–6 July 2013), though only select MOUDI stages were
analyzed for each sampling period. During each time period depicted,
SOA/sulfate averaged > 60 % of accumulation mode (0.2–1.0 µm) and 2 % of the supermicron (1.0–5.0 µm) particles by
number fraction. However, the number fraction of SOA/sulfate was highly
variable between the SOA, dust, and SSA periods. During the two periods
dominated by SOA/sulfate depicted in Fig. 5a, the number fraction of
SOA/sulfate reached up to 95 % in the accumulation mode and up to 70 %
of supermicron particles. Because Centreville, AL, is a forested site and
BVOC emissions, such as isoprene, are high in this region, it is not
surprising that SOA/sulfate dominated throughout the majority of the
campaign, particularly at small particle sizes. However, the fraction of
SOA/sulfate > 1 µm is noteworthy, as SOA/sulfate particles
are typically considered submicron in size.
Dust was the dominant particle source during two coarse-mode nitrate events
(Fig. 5b) detailed previously by Allen et al. (2015) and defined more
narrowly herein as 11–13
and 26–28 June 2013 to differentiate from SSA transport time periods
and account for available CCSEM data. During the dust-dominated time periods
analyzed, dust constituted > 55 % of supermicron particles
(1.0–5.0 µm) by number but also contributed, on average, 26 % of
accumulation mode particles (0.2–1.0 µm) by number.
Similar to dust, SSA contributed significantly to the overall particle
population multiple times throughout the study, comprising approximately 35 % of particles, by number, analyzed during an event in the middle of June
(10–11 June 2013) and at the beginning of July (3–6 July 2013). Both of
these SSA events were also characterized by high number fractions of dust,
as observed in Fig. 5c. During these SSA-rich periods, SSA particles were
predominately larger than 1 µm (38 % by number), although notable
contributions to accumulation mode number fractions of SSA were also
observed (22 % by number from 0.2 to 1.0 µm). During these two events
the degree of atmospheric processing varied, with a considerable number
fraction of partially aged SSA present during the second event compared to
the first event, which was primarily fully aged SSA
(Bondy et al., 2017b).
(Left) Size-resolved compositions indicate the number
fraction of particles containing nonvolatile cations Na, Mg, K, Ca, and Fe
during the (a) SOA period, (b) dust period, and (c) SSA period. (Right) The
number fraction of submicron and supermicron particles during each period
containing each nonvolatile cation.
Nonvolatile cations during SOAS
Recently, the potential for soluble nonvolatile cations such as Na+,
Mg2+, K+, and Ca2+ to improve thermodynamic modeling results
of aerosol acidity when included as inputs has been suggested, assuming all
species are internally mixed (Guo et al., 2017). As
CCSEM-EDX can readily detect metals within individual particles, the number
fraction of particles containing Na, Mg, K, Ca, and Fe at sub- and
supermicron sizes during the SOA, dust, and SSA events is shown in Fig. 6.
In addition to these metals, Mn was detected within < 3 %
particles by number during SOAS at > 2 % in a particle, and,
given its low fraction, Mn was not included in further analysis. During all
events, the number fraction of particles containing nonvolatile cations
increased as a function of particle size, with a higher number fraction of
metal-containing particles at supermicron sizes (19 %–94 %) compared to
submicron sizes (1 %–50 %). During all the time periods depicted, Na was
present most frequently, closely followed by Mg, indicative of SSA
particles. Fewer particles contained K and Ca by comparison, and Fe was
present within the lowest number fraction of particles, except for during
the dust period when Fe was more frequent. The number fraction of
metal-containing particles was not consistent throughout SOAS but varied
dramatically between the SOA, dust, and SSA periods. In general, particles
during the dust and SSA events contained higher number fractions of all
nonvolatile cations, particularly Na and Mg. However, the variation between
specific metals was largely dependent on the dominant particle class during
each period.
Figure 7 focuses on the SOA time period and shows the number fraction of
particles within each particle class containing Na, Mg, K, Ca, or Fe (dust
and SSA periods are shown in Fig. S5). Within both submicron and
supermicron particles, the nonvolatile cations within each class are
consistent, though a marginally larger number fraction of supermicron
particles contained nonvolatile cations, likely due to detection limits for
smaller particles. Less than 5 % of SOA/sulfate particles by number
contained any Na, Mg, K, Ca, or Fe. Conversely, all other particle classes
contained metals within a substantial number of particles. Specifically, all
biomass burning particles contained K. Fly ash, on the other hand, most
frequently contained Na, though most fly ash contained Al or Si instead
(Fig. 4). Additionally, a considerable fraction of dust particles
contained Na, Mg, K, Ca, or Fe; all SSA contained Na and many contained Mg;
and most primary biological particles contained Na, Mg, and K. When the
nonvolatile cations are compared within each class during the SOA, dust, and SSA periods
(Figs. 7 and S4), the number fractions of metal-containing particles are
consistent for each particle class, suggesting that an internal mixing
assumption for nonvolatile cations and their presence in SOA/sulfate
particles does not reflect overall particle composition.
Size-resolved particle class compositions indicate the
number fraction of particles in each class containing nonvolatile cations
Na, Mg, and Fe during the SOA period in the (a) submicron and
(b) supermicron size range.
Particle aging
In contrast to nonvolatile cations, the contribution of secondary components
within each class varied drastically throughout SOAS. Due to atmospheric
reactions and transport of gases from nearby cities, many of the particles
analyzed from SOAS were likely not “fresh” from their source but had
undergone secondary processing by species such as HNO3,
SO2/H2SO4, or organic acids. Secondary processing of
particles is important because changing their chemical composition can
impact light scattering and CCN properties (Chang et al., 2010; Chi et
al., 2015; Ghorai et al., 2014; Giordano et al., 2015; Hiranuma et al.,
2013; Lu et al., 2011; Moise et al., 2015; Robinson et al., 2013; Sedlacek
et al., 2012; Tang et al., 2016). As the chemical composition of particles
varied over time, the mixing state index was used to quantify the degree of
aging. The degree of secondary processing for each particle class was
calculated as the average mass fraction of sulfur per particle; see Fig. S6 and details in the Supplement. Only sulfur was used as an indicator of aging in
this study since carbon had interference from the background and nitrogen is
only semi-quantitative with CCSEM-EDX (Laskin et al.,
2006).
From STXM-NEXAFS, we know that most SOA particles are mixtures of organic
and inorganic (mostly ammonium sulfate) components, and there are almost no
externally mixed organic or ammonium sulfate particles present. As such,
based on elemental composition shown in Fig. 4 and the fact that C, O, and
N could not be quantified in this study, particles containing only S in our
mixing state analysis are presumed to be SOA/sulfate. A large fraction of
sulfur in SOA was likely in the form of sulfate, since sulfate was
identified as the major inorganic component within SOA (24 % wt in fine
aerosol) (Budisulistiorini et al.,
2015b). Additionally, IEPOX-derived organosulfates and other organosulfates,
known to contribute to the organic aerosol fraction in Centreville (Bondy
et al., 2018; Boone et al., 2015; Budisulistiorini et al., 2015b; Riva et
al., 2016; Xu et al., 2015a), also contributed to the EDX sulfur content of
SOA. The other five particle classes contained substantially less sulfur
than SOA: SSA (20 % S–30 wt % S), biomass burning particles (15 % S–25 wt % S), dust (5 % S–15 wt % S), fly ash
(2 % S–10 wt % S), and biological
particles (15 % S–25 wt % S). SSA and biomass burning particles are both
readily aged by sulfuric acid forming Na2SO4 and K2SO4,
respectively (Chen et al., 2017; Hopkins et al., 2008; Li et al., 2003),
although up to 8 % of sulfate in SSA may have marine origins
(Pilson, 1998). Aluminosilicate dust, the most common type of
mineral dust detected during SOAS, is also aged by sulfuric acid (Fitzgerald
et al., 2015; Perlwitz et al., 2015; Song et al., 2007; Sullivan et al.,
2007). Fly ash detected during SOAS did not contain much sulfur, indicating that
it was relatively fresh (Shen et al., 2016), or was aged
more by other species such as organics, relative to sulfuric acid
(Li et al., 2017). Primary biological particles also contained low
mass fractions of sulfur. However, as heterogeneous chemistry of this class
of particles has not been explored as extensively as the other classes and
the sulfur mass fractions did not follow the same trends for the three time
periods (Fig. S6), the sulfur content in biological particles may have
been, but was not necessarily, the result of aging (Estillore et
al., 2016).
In addition to differences in aging by sulfur for each particle class, the
average mass fraction of sulfur within each class varied during the
SOA-rich, dust-rich, and SSA-rich time periods (Fig. S6). Specifically,
the average mass fraction of sulfur was significantly higher during the
SOA-dominated time period compared to the dust and SSA periods at the 95 % confidence interval for all particle classes aside from biological
(Tables S6–S7). However, the mass fraction of sulfur was not statistically
different between the dust and SSA periods for any particle classes (Table S8). Stagnant air masses, indicated by slower average wind speed at
Centreville during the SOA/sulfate period (1.6±0.7) compared to the
dust period (2.3±0.9) and SSA period (2.1±0.9), may have led
to more aging during the SOA events.
Quantification of mixing state using aging diversity measures
To quantify the differences in aging during three described events, the
mixing state index was calculated for the SOA, dust, and SSA time periods
(Fig. 8). Because the particle classes present at sub- and supermicron
sizes vary dramatically, mixing state indices were calculated separately for
the two size ranges. From calculations of the average particle diversity and
the bulk diversity (Fig. 8a), the mixing state index, a ratio measuring
how close the population is to an external or internal mixture, could be
determined for each time period (Fig. 8b). The mixing state indices for
supermicron particles were generally the highest (χ=19 %, 15 %, and 11 % during the SSA, dust, and SOA periods, respectively),
signifying that supermicron particles were less diverse than submicron
particles. The supermicron SOA period is more externally mixed than the SSA
or dust periods for two reasons: (1) it contains the most individual
particle classes (largest bulk diversity, ∼5), and (2) the
particle-specific diversity is highest as well, indicating that this period
has a lot of aging by sulphur (and likely organic carbon), contributing to
the relatively high mixing state index. The mixing state indices for
accumulation mode particles during the SOA and SSA periods were comparable
(χ=10 % and 9 %, respectively), while during the dust time
period many more elements were present in separate particles, leading to the
most external mixture observed. The SSA time period had a more external
accumulation mixing state index than the supermicron mode, which is logical
as SSA dominated the supermicron during SSA periods, but the accumulation
mode still had distinct classes (e.g., SOA/sulfate). For the SOA time
periods, the accumulation mode had the lowest bulk diversity and
particle-specific diversity, while the supermicron mode had both high
particle diversity and greater particle-specific diversity, which led to
them having similar mixing state indices. Overall, Fig. 8a demonstrates
that time periods with low bulk diversity, which contain fewer particle
classes, have mixing state indices closer to 100 % (more internally
mixed).
(a) Mixing state diagram showing the bulk diversity and average
particle-specific diversity and (b) mixing state indices for sub- and
supermicron particles during the SOA, dust, and SSA periods. For submicron
particles, contributions by different sources impact mixing state.
Mixing state indices of this work are lower than previous reports by
Fraund et al. (2017) and O'Brien et al. (2015) (χ > 80 % and χ > 40 %, respectively), who
used CCSEM-EDX and STXM-NEXAFS to analyze particles collected in the Amazon
and Central Valley, CA. In both studies, calculations using STXM-NEXAFS
resulted in low diversity and high mixing state indices, likely due to the
inclusion of organic carbon, which increased particle homogeneity. O'Brien
et al. also calculated mixing state indices using solely CCSEM-EDX, similar
to our study. From this method, O'Brien found lower mixing state indices
using CCSEM-EDX (χ=41 %–90 %) compared to mixing state index
calculations from STXM-NEXAFS results (χ > 60 %).
However, the mixing state index in this previous work increased during
high-SSA periods and periods characterized by increased mass fractions of K, Ca,
Zn, and Al, suggesting that periods with higher average particle-specific
diversity were more homogeneous, simply because they contained more elements
than periods dominated by carbonaceous particles. To address this inherent
challenge associated with quantifying mixing state using CCSEM-EDX, in the
current study we calculate mixing state based on the number of particle
classes and secondary species (in this study, sulfur) rather than the number
of elements within particles. Quantifying mixing state parameters using this
approach is consistent with our concept of atmospheric aging since the
mixing state index increases as bulk diversity decreases and the mass
fraction of secondary species increases, signifying aging increases the
degree of internal mixing in a population. This method of quantifying
aerosol mixing state using single-particle methods can be used to show the
varying impact of sources and aging between different air masses at the same
location and will be expanded upon in future work.