Introduction
Over the past decade, studies of aerosol optical properties have become an
important topic of research motivated by large uncertainties that limited
our predictive capabilities of climate (IPCC, 2007). While our climate
predictions have become more robust in recent years, radiative forcing (RF)
by aerosols “continue to contribute the largest uncertainty to the total RF
estimate” (IPCC, 2014). As work has progressed in recent years via
modeling, laboratory studies, ambient measurements, and satellite
retrievals, complicated relationships have been revealed between aerosol
primary sources and secondary production and processing mechanisms. Aerosol
composition, particle size and shape, the mixing state of an aerosol
population, and particle number concentrations all play a role in the
optical properties of aerosols in the ambient atmosphere. Hence, many
efforts are underway to investigate these features in ambient aerosols to
better characterize their optical properties, in the hopes of reducing
uncertainties in calculating the Earth's radiation budget. Conversely, it
has also been suggested that optical properties retrievals hold promise for
remotely sensing aerosol size and composition throughout the global
troposphere as well as to forecast regional- and local-scale air quality.
Aerosols interact with light both through scattering and absorption; the sum
of these is known as extinction. This interaction is wavelength (λ)-dependent, often characterized by a power law, where the Ångström
exponent (α) is the power that defines the slope of that relationship
to wavelength (Ångström, 1929; Moosmüller and Chakrabarty, 2011):
p(λ)=λ-α,
where p(λ) represents any parameter that can be represented with a
power law. Ambient aerosol optical properties (extinction, scattering, and
absorption) are each represented in this way. Many currently available
commercial aerosol instruments (e.g., laser-based methods) obtain these
parameters at relatively few (three or fewer) visible wavelengths. For these
instruments, α is calculated for two wavelengths (λ1,
λ2) and it is assumed that a power law represents the full
wavelength range between those two endpoints (Moosmüller and Chakrabarty,
2011), e.g.,
pλ1pλ2=λ1λ2-α,αλ1,λ2=-lnpλ1/pλ2lnλ1/λ2.
However, there is evidence that other mathematical relationships (e.g.,
nonlinear expressions to account for curvature in log space) may offer a
better fit for certain optical parameters of specific types of aerosol over
particular wavelength ranges (e.g., Eck et al., 1999; Schuster et al., 2006).
This suggests that extrapolating α to wavelengths outside of the
measured range may be problematic, especially when extrapolating into the UV
from measurements made at visible wavelengths. Hence, observations at several
wavelengths (or over a broad range of wavelengths) particularly into the UV,
would be useful in more fully characterizing wavelength-dependent optical
properties of ambient aerosols (Laskin et al., 2015).
Wavelength-dependent optical properties are used to distinguish four aerosol
types from each other: black carbon (BC), brown carbon (BrC), dust, and
nonabsorbing aerosols. The wavelength dependence of aerosol light scattering
depends on the size distribution and particle shape, and less on the chemical
composition of the aerosol. Aerosol light absorption, however, depends
heavily on the chemical or molecular structure within the particles. BC is
the strongest absorber of visible light, with its absorption characterized by
αabs∼1 (e.g., Yang et al., 2009; Desyaterik et al., 2013;
Bond et al., 2013). This wavelength dependence arises from the molecular
structure of BC, which is emitted with a graphite-like sheet structure (Bond
and Bergstrom, 2006; Yang et al., 2009). Such structures feature many π-bonded carbon atoms with delocalized electrons that readily absorb light
across the ultraviolet (UV) through infrared (IR) wavelength range
(Desyaterik et al., 2013), giving rise to the black color of soot. The
graphitic sheets fold into spherules that then rapidly coagulate into loose
aggregates, small enough with respect to visible and near-visible wavelengths
that their interaction with light falls within the Rayleigh regime, i.e.,
αabs∼1 (Yang et al., 2009). In this regime, refractive
indices are assumed to be wavelength-independent (Moosmüller and
Chakrabarty, 2011, and references therein).
In contrast, BrC (light-absorbing organic carbon) has far fewer conjugated
π electrons to absorb light. The wavelength range of absorption depends
on the number of conjugated bonds in a molecule, along with the presence of
heteroatoms, e.g., O or N, such that more bonds or heteroatoms shift the
absorption to longer wavelengths (e.g., Jacobson, 1999; Apicella et al.,
2004; Chen and Bond, 2010). Where BC has many such bonds, absorption occurs
into the IR wavelength range, but for BrC absorption is limited to UV and
visible wavelengths. Aromatic rings offer a stable structure for conjugated
bonds in atmospheric molecules; however a single ring has too few bonds to
absorb light in the near-UV or visible range. Substituted O or N molecules on
a single ring, or multiple rings (i.e., polycyclic aromatic hydrocarbons,
PAHs) shift the absorption into the UV/visible range, leading to a characteristic
wavelength dependence for these BrC aerosols that features strong UV
absorption such that αabs>1 (Andreae and Gelencsér, 2006;
Moosmüller et al., 2009; Chen and Bond, 2010; Desyaterik et al., 2013).
This results in BrC exhibiting a yellow-to-brown color, hence the term
“brown carbon”. BrC may be expected to optically evolve in the atmosphere
since terminal double bonds (and heteroatoms) are more susceptible to
atmospheric oxidant attack, thereby reducing the number of conjugated π
electrons in the molecules that comprise BrC. Conversely, secondary organic aerosol (SOA) formation from
oxidation products with suitable structure may lead to BrC formation in the
atmosphere.
Dust particles are more weakly absorbing per mass in the mid-visible than
either BC or BrC, but are important absorbers in the atmosphere due to the
mass of material emitted (Bond et al., 2013). The optical properties of dust
depend on its mineral composition, particle shape, aerosol size distribution,
and modifications during atmospheric transport (Yang et al., 2009). Dust that
contains iron oxides and clay is known to significantly absorb light in the
UV/visible range (Sokolik and Toon, 1999), with reports of αabs up
to ∼3 over 325–660 nm (Alfaro et al., 2004; Yang et al., 2009). Many
other organic and inorganic compounds in aerosols do not absorb light and are
classified optically as nonabsorbing (or pure scattering) aerosols, e.g.,
polystyrene latex spheres (PSLs), ammonium sulfate (AS), and sea salt
(Washenfelder et al., 2013).
The characteristic wavelength dependence of absorption distinguishes these
groups from each other. Extinction, however, includes the effects of both
scattering and absorption, where scattering dominates the two contributions.
For example, even though dust may have αabs∼3, the large
size (typically supermicron diameters) of dust particles results in
αscat near 0, leading to αext∼0. For the
other three groups, with smaller (submicron diameter) particle sizes that may
be comparable to each other in the ambient atmosphere, the scattering
contribution may be expected to be similar, while the absorption contribution
will clearly differ. In visible wavelengths, the relative difference in
absorption may be difficult to discern from the much larger scattering term
when measuring extinction. However, in the UV, particularly at wavelengths as
short as 300 nm, the absorption term may influence the shape of the
extinction spectrum revealing differences among these groups.
Of these four groups of optically defined aerosols, BrC may be the most
complex to quantify with myriad potentially important sources identified in
the literature (see the recent review by Laskin et al., 2015, and references
therein). Sources of BrC include combustion (biogenic and fossil fuels),
humic-like substances, and SOA compounds that are derived from anthropogenic
and biogenic precursors (e.g., Andreae and Gelencsér, 2006; Hecobian et
al., 2010). Identifying the specific compounds that contribute to BrC is
challenging. Several studies have examined the absorption properties of
solution-phase extracts (e.g., in water, methanol, or hexane solvents; e.g.,
Chen and Bond, 2010; Hecobian et al., 2010; Zhang et al., 2011, 2013);
however, the solubility of the compounds in each solvent determines which
compounds are extracted, and hence, the wavelength dependence will reflect
the compounds in solution. Other factors will also affect solution-phase
absorption, beyond the choice of solvent, which include the degree of
dilution (Zhang et al., 2013), pH (Jacobson, 1999), and degree of dissolution
(i.e., ionic versus molecular forms) (Jacobson, 1999).
Given that different sources have been identified for BrC, where mixtures
from various sources occur (whether primary organic aerosol (POA), SOA, or both), photooxidative
competition among the compounds present may be expected to lead to differing
evolution of BrC in the atmosphere. In some cases, photooxidation may reduce
BrC over time; in others it may produce BrC. Since not all SOA-forming
products absorb light and those that do absorb at varying peak wavelengths
and at varying strengths (e.g., Jaoui et al., 2008; Zhang et al., 2011), it
is expected that BrC formation, evolution, and destruction in the atmosphere
may be highly variable. It must also be noted that chromophores may comprise
a minor fraction of ambient aerosol mass, such that large changes in optical
properties with aerosol age are not necessarily accompanied by large changes
in mass spectra, or Fourier transform infrared (FTIR) or nuclear magnetic resonance (NMR) spectra (Bones et al., 2010). Hence, tools are
needed to assess the optical evolution of ambient aerosols, along with tools
used to investigate their chemical and physical qualities.
In addition to contributing to studies of ambient aerosol optical, physical,
and chemical evolution, directly measured aerosol extinction is expected to
be useful for studies of other important optical information as well. For
example, single-scatter albedo (ω(λ)) is a measure of the
fraction of light scattered from the total extinction defined by the ratio of
the scattering coefficient (σscat(λ)) to the extinction
coefficient (σext(λ)),
ω(λ)=σscat(λ)σext(λ)=σscat(λ)σscat(λ)+σabs(λ),
where σext(λ) may also be expressed as the sum of
σscat(λ) and the absorption coefficient
(σabs(λ)). This is an intensive property of a particle,
a function of its composition and independent of total aerosol concentration
(Yang et al., 2009). For nonabsorbing particles, ω=1 at all
wavelengths. This parameter is used in models to determine the sign and
magnitude of RF (Moosmüller and Chakrabarty, 2011). Often for in situ
measurements, scattering is measured (e.g., nephelometry) and absorption is
measured (e.g., aethalometry), then summed to obtain extinction for the
calculation of ω(λ). Commercially available instruments are
also available for the direct measurement of extinction. However, these
measurements are restricted to visible wavelengths and are often at a few
(one to three) wavelengths. See Washenfelder et al. (2013) for a recent
synopsis of various measurement techniques for aerosol scattering,
absorption, and extinction. Direct extinction measurements of
σext(λ) over both the UV and visible range would
provide a useful addition to current measurement techniques for the
calculation of ω(λ).
With this motivation, to investigate linkages between ambient aerosol
chemistry, physical attributes, and optical properties, the instrument to
study spectral aerosol extinction across the UV/visible range (abbreviated
SpEx) has been developed. In this work we introduce the new instrument and
show laboratory test results performed at NASA Langley Research Center in
March 2014 that characterize both the instrument performance as well as
spectral properties of various laboratory-generated aerosols. The
rack-mounted prototype configuration was deployed aboard the ground-based
NASA Langley Mobile Aerosol Characterization (MACH-2) laboratory during
DISCOVER-AQ (Deriving Information on Surface conditions from Column and
Vertically Resolved Observations Relevant to Air Quality) and obtained
ambient spectra while parked at several ground sites. A subsequent paper
will present data obtained during the DISCOVER-AQ Colorado field campaign in
July/August 2014.
Schematic of SpEx, along with other instrumentation used in the
laboratory experiments. The laboratory aerosol instrumentation downstream of
the optical cell sampled from a common manifold.
Experimental
Tests were performed with a calibrated NO2 standard as well as
laboratory-generated aerosols to characterize the new instrument. A detailed
description of SpEx follows, along with brief descriptions of the other
instruments used to validate the new technique.
SpEx description
SpEx is a modified version of the aerosol extinction differential optical
absorption spectrometer (AE-DOAS), a custom-built instrument that has been
used in laboratory studies (Chartier, 2010; Chartier and Greenslade, 2012).
Development of SpEx adopted the measurement approach of AE-DOAS with
improvements to allow for faster data acquisition and enhanced sensitivity.
These instruments use a White-type optical cell (White, 1942) in which the
primary and two secondary concave mirrors have the same radius of curvature
separated by a distance of twice the focal length. Light intensity in the
cell (with an optical path length, L) is measured for a sample (air with
aerosols, I) and for a reference (filtered air, I0) allowing for the
calculation of aerosol extinction via Beer's Law:
σext=-lnI/I0L.
Extinction is expressed in units of inverse length. For convenience, in the
ambient atmosphere, the units typically used are inverse megameters
(Mm-1).
These instruments (AE-DOAS and SpEx) are comprised of three key components
coupled with fiber optics: a broadband xenon lamp, a closed White-type
multi-pass cell, and a spectrometer. The main improvements to AE-DOAS are
detailed here and include a longer optical path length and a 16-bit
spectrometer to improve sensitivity; larger orifices to allow for faster flow
rates, more rapid air exchange in the cell, and hence faster data
acquisition; automated operation for 24 h sampling in the field that includes
flow control, switching between filtered and unfiltered lines, and data
acquisition. SpEx offers a rack-mountable design for field studies of ambient
aerosol monitoring.
SpEx (Fig. 1) was custom-designed based on the UV 5000 gas analyzer system
(Cerex Monitoring Solutions, LLC, Atlanta, GA) with a 150 W xenon lamp
source (Cerex P/N CRX-X150W), integrated with an Ocean Optics, Inc. (Dunedin,
FL) QE65Pro 16-bit spectrometer. These components are coupled with
600 µm fiber optic cables to a custom-designed White-type optical
cell fabricated by Cerex Monitoring Solutions, LLC., out of aluminum with
rectangular dimensions of 6′′×4.25′′×56.18′′ (15.24×10.8×142.7 cm) and an internal volume of approximately 17 L. The
QE65Pro includes their standard 1024 pixel detector and a composite grating
providing 1.36 pixels per nm over a range of 750 nm starting at 200 nm,
resulting in a spectral resolution of approximately 0.7 nm. The built-in
cooling system of the spectrometer allows the detector to be cooled down to
-20 ∘C to minimize noise. The fiber optic cables are 2 m long
with heavy-duty metal jacketing.
Constructed from a rectangular tube, two ports were cut in the top of the
cell above the optics to allow for adjusting the optical alignment. Four side
ports, two at each end, allow for flexible configuration of air inflow and
outflow ports for rack mounting. For laboratory and ambient sampling
(Fig. 1), ports on opposite sides were used to maximize mixing in the cell
and to ensure that air flows across the optical path. Half-inch (1.27 cm)
conductive silicone inlet tubing allowed for flow rates up to
80 L min-1 for rapid air exchange
within the volume, and were actively controlled using a mass-flow controller
(MFC). To ensure complete air exchange between the sample and reference air
within the optical cell, a flush volume of at least 3 times the internal
volume of the cell (17 L ×3=51 L) must be used. At
80 L min-1 this takes approximately 40 s. In the laboratory tests, a
flush time of 90 s was used, with the particle counts in the outflow of the
optical cell monitored to ensure the air was fully exchanged between the
sample (unfiltered) and reference (filtered) lines, which were autonomously
cycled using a pair of electrically actuated, straight-path ball valves. The
change in pressure, measured downstream of the sample cell, when switching
from sample and reference lines, was minimized (to less than 1 torr) by
slightly restricting flow in the sample line and manually setting the
filter-line pressure drop (at a constant flow rate) using a needle valve.
This allowed accurate measurements and negligible particle transmission loss.
The White-type cell (Fig. 1) is designed to allow for a range of optical path
lengths. As is evident from Eq. (4), for any given difference between I and
I0, a longer path length (L) offers greater sensitivity in extinction,
while for any given path length, maximizing I0 improves sensitivity of
σext to I. With a 48-inch (1.23 m) basepath between
mirrors, an alignment with six spots across the top of the primary mirror
(28 passes total) provides an optical path length of 34.4 m, while the
maximum of eight spots (36 passes total) extends the path to 44.3 m. Since a
little light is lost at each reflection of the light beam, intensity
decreases as the path length (i.e., the number of spots) increases. Hence,
determining the optimum path length requires a balance in maximizing the
intensity versus maximizing the number of spots. In tests at Cerex Monitoring
Solutions, seven spots (32 passes total) were found to offer the optimum path
length of 39.4 m, given the light source and the efficiency of the mirror
coatings over the UV/visible range. Note that the optical path results in two
rows of light spots on the primary mirror. Notches cut out of the top half of
the primary mirror at either end allow incoming and outgoing light to enter
and exit the cell via the fiber optic ports. Collimating lenses focus the
light as it enters and exits the cell.
As this is a broadband light source, the spot size is large (on the order of
1 cm). Longer wavelengths of the light wave are toward the center of the
spot, while the shorter wavelengths are toward the outside of the spot. Since
the spot is larger than the fiber diameter (600 µm), as the mirror
alignment is adjusted slightly, the sampled spectrum from the light source
may be optimized. Tuning results in large changes in the measured intensity
spectrum for the same optical path length. Here, the QE65Pro records the lamp
intensity from 200 to 995 nm (Fig. S1 in the Supplement). The spectral
characteristics of the lamp and the mirror coatings are such that wavelengths
between 200–250 and 750–800 nm exhibit low intensity, while those
> 800 nm tend to be most intense. In order to optimize for
sampling over the 300–700 nm wavelength range, the spectral sampling is
adjusted to maximize the intensity over that range, while the integration
time is set such that the infrared channels saturate (Fig. S1). When properly
aligned, an integration time of 20–50 ms maximizes the intensity counts
without saturation over the 300–700 nm range.
Allan variance plot for select wavelengths. Test conducted with
filtered air at a flow rate of 90 L min-1 using 50 ms integration time per
spectrum.
SpEx operation, variance tests, and limit of detection
For typical ambient atmospheric particle concentrations, extinction
represents a small difference between two large numbers, I and I0. As
a result, care must be taken to minimize sampling bias and noise. Prior to
any sample collection, the instrument is turned on and allowed to warm up for
∼1 h to allow the lamp to thermally stabilize. Measured intensity may
drift slightly between spectra (typically < 1 %). This may
arise from thermal variability of the lamp (which was not
temperature-controlled for these tests), temperature variations of the gas
cell, inducing minor fluctuations of the optics, or variability in the power
source for the lamp. To limit any variability in the extinction calculation
that may arise from a drift in intensity, the extinction is automatically
calculated by averaging the reference spectra collected immediately prior to
and following a given sample spectra.
Noise arises in optical instrument systems both from high- and low-frequency
sources that can be characterized by an Allan variance calculation. Such a
calculation is used to identify the optimum sampling interval, i.e., sampling
over too few spectra results in noise from high-frequency sources, while
sampling over too many results in noise from low-frequency sources. A plot of
Allan variance shows the optimum sample size to minimize noise in the signal.
Using an integration time of 50 ms per spectrum (the upper end of the range
used for sampling prior to realigning the optics) and a flow rate of
90 L min-1, filtered air was pulled through the system and spectra
were acquired in a 13 h test. The Allan deviation of the average versus the
number of scans shows that for SpEx ∼ 1000, spectra provide the best
sensitivity over the 300–700 nm wavelength range (Fig. 2), resulting in
Allan deviations of ±3.5–5.7 Mm-1 (Table S1 in the Supplement).
At 50 ms, it takes 50 s to collect 1000 spectra. From Fig. 2 however, it is
evident that fewer spectra can offer sensitivity nearly as well, while
allowing for faster data acquisition. On the basis of laboratory tests (data
not shown) and the Allan variance calculations, it was found that a 30 s
sampling time provided good sensitivity with low noise. For 600 spectra
(30 s sampling time for 50 ms integration) the Allan variance was
± 4.1–7.6 Mm-1 over 300–700 nm. Intensity spectra were acquired
at a temporal resolution of 125 s (90 s flush, 30 s sample, plus a 5 s
delay added to allow time for valves to open and close to switch between the
filtered and unfiltered lines). As a result, aerosol extinctions were
obtained with a 250 s time resolution.
These data suggest a 3σ limit of detection of
∼ 12–23 Mm-1 over 300–700 nm wavelengths for the individual
laboratory spectra presented here. This detection limit is adequate for
ambient aerosol measurements in most ambient boundary layer environments
(typical mid-visible scattering is 57 Mm-1 for average background
environments and 23 Mm-1 for clean continental background environments,
Seinfeld and Pandis, 1998). For ground-based measurements in clean
continental environments, by making use of the standard error of the mean,
individual extinction spectra may be averaged, thereby reducing the error by
the square root of the number of samples in the mean. Hourly averages
(16 spectra) have the effect of reducing the error by a factor of 4, thereby
reducing the detection limit to 3–6 Mm-1 over the 300–700 nm range.
Thus, this data reduction scheme is able to sufficiently characterize both
concentrated plumes and background conditions on reasonable timescales at
ground sites. Future modifications to SpEx are anticipated to reduce the
limit of detection such that airborne deployments for measurements in the
free troposphere will be feasible.
Validation instruments
The Cavity Attenuated Phase Shift extinction (CAPS PMex) monitor,
manufactured by Aerodyne Research, is an optical extinction spectrometer
which has a light-emitting diode (LED) as a light source, a 26 cm long
sample cell with two high reflectivity mirrors and a vacuum photodiode
detector. Due to its configuration, the cell cavity has an effective path
length of approximately 2 km. Similar in principle to a cavity ring-down
instrument, the CAPS PMex monitor relies on measuring the average time spent
by the light within the sample cell. Particles in the cell decrease this time
due to scattering and absorption. The particle light extinction is determined
from changes in the phase shift of the distorted waveform of the square-wave-modulated LED light that is transmitted through the optical cell. Three CAPS
PMex monitors were used with differing wavelengths (450, 530, and 630 nm).
The CAPS PMex particle optical extinction monitor has a range of
0–4000 Mm-1, resolution of 0.1 Mm-1 and a sensitivity of
2.5 Mm-1 in a 1 s sampling period.
The TSI Inc. Integrating Nephelometer (model 3563) measures aerosol
scattering at three visible wavelengths (450, 550, and 700 nm) for the total
angular range between 7 and 170∘. The measurement can be corrected
for truncation (i.e., the missing scattering not measured for angles 0–7 and
170–180∘) based on Anderson and Ogren (1998). The instrument was
regularly zeroed with particle-free air and calibrated with CO2.
Scattering is measured at 1 Hz and a sensitivity of 0.1 Mm-1 for a 60 s
averaging period.
Aerosol generation
Dust aerosol samples were generated by a Wrist Action Shaker (Burrell model
75) with the speed varied to change the aerosol concentration. The motion of
the shaker suspends the dust into the sample air stream via saltation (Kumar
et al., 2011). The dust was either sampled as produced or through a stainless
steel cyclone (URG) with a 1 micron cut at 10 L min-1. The flow
through the cyclone could be changed to result in a differing aerosol size
distribution. Changes between the PM1 and PM2.5 tests described in
Sect. 3 were verified with an Aerodynamic Particle Sizer (APS, TSI, Inc.
model 3321). Soot samples were generated by a quenched propane burner
(miniCAST Series 4202, Jing Ltd.) with variable nitrogen-to-fuel ratios used
to vary soot size and soot composition (Moore et al., 2014). Liquid aerosols
were generated from aqueous solutions by a TSI Inc. aerosol nebulizer (model
3076). In all cases the generators did not produce enough flow
(1–10 L min-1) to accommodate all the instrumentation, so
particle-free air was added to the output from the aerosol generators. For
some tests, aerosols were size-selected using a differential mobility
analyzer (DMA, TSI Inc. model 3081). For these size-selected tests, we did
not correct the size distributions for multiply charged particles, as all
compounds selected at the same size using the same flow rates will have
similar percentages of multiply charged species, resulting in a similar size
distribution for all cases. Particle number concentrations were measured (by
a condensation particle counter, CPC, TSI 3775) to verify sufficient flush
timing and to verify constant test conditions. Measurements of size
distributions of submicron (by a scanning mobility particle sizer, SMPS, TSI)
and supermicron (by a laser aerosol spectrometer, LAS, TSI Inc.) aerosols
during laboratory testing were used to confirm the generated aerosol
characteristics.
Results
Accuracy of measured spectra
O2, NO2, and PSL tests
Laboratory tests were designed to document the accuracy of SpEx extinction
spectra by assessing (1) whether the filter used to remove the aerosols also
removed NO2 such that measured spectral characteristics include
gas-phase differences unrelated to aerosol extinction, (2) whether the
intense UV lamp light photolyzed NO2 to O3 in the optical cell, and
(3) whether the measured spectral shapes were consistent with expectations
from theoretical calculations. NO2 efficiently absorbs light in the
UV/visible range, motivating test no. 1. It is also readily photolyzed by UV
light, making it useful for testing photolysis in the optical cell by the
lamp (test no. 2).
100 ppb NO2 (diluted from a calibrated cylinder of 5 ppm NO2) was
measured (using a Los Gatos Research instrument model 907) across the
sample-line filter showing a negligible difference in mixing ratio between
downstream and upstream locations, suggesting the filter is not modifying the
NO2 concentration between the filtered and unfiltered sampling lines.
Additional tests using both O3 (Thermo Environmental Instruments model
49C ozone analyzer) and NO2 showed that O3 is not being produced
(conversely NO2 is not lost) in the optical cell by the light sources
with 70–80 L min-1 flow. Minimal surface losses across the optical
cell were found in both the O3 (9 ppb in the cell vs. 15 ppb in room
air, on average) and NO2 (92 vs. 100 ppb calibrated inflow) tests.
However, the advantage of the difference method employed in the SpEx
measurement is that gas-phase losses in the optical cell affect both the
sample and reference lines in the same way, canceling each other out as long
as ambient mixing ratios are constant throughout the full sample/background
cycle.
Two sets of tests were performed, measuring absorption spectra of NO2 and
extinction spectra of PSLs to verify measurements with theory. For NO2,
a series of four tests were done starting with an undiluted 5 ppm test from
the calibrated cylinder, followed by a series of three dilutions (1, 250, and
100 ppb). NO2 tests were performed by slowly filling the cell, capping
it, then collecting spectra, in order to minimize the volume of calibration gas
that would have been consumed with the 17 L volume and typical
80 L min-1 flow rate. Stopping the flow led to increased surface
losses. Correcting the SpEx absorption spectra for these losses, it is clear
that SpEx obtained a curve that captured the expected features calculated
from GOME/SCIAMACHY NO2 cross-section data binned to 1 nm resolution
(Fig. 3). Similar results were obtained in all four tests; only the 5 ppm
results are shown here. A minor offset of ∼1 nm in the SpEx absorption
spectrum was found compared to the theoretical curve, requiring a correction
of +1 nm to the nominal SpEx wavelength. The comparison in Fig. 3 shows
the ability of SpEx to capture the broad absorption peak centered around
400 nm over the 300–500 nm range, along with high-spectral-resolution
features down to a few nanometers' width. In laboratory studies, similar
high-resolution features in spectra of monodisperse aerosols have been used
to retrieve refractive indices (e.g, Chartier and Greenslade, 2012).
Theoretical NO2 absorption calculated from GOME-SCIAMACHY
cross sections averaged to 1 nm (dashed black line) compared with measured
wavelength-corrected SpEx absorption binned to 1 nm (solid blue line), and
CAPS PMex extinctions (red diamonds) from 300 to 600 nm (top panel). Detail
of spectra from 440 to 540 nm are shown in the bottom panel.
The wavelength-dependent refractive index for polystyrene latex
spheres used in MiePlot (top panel). Extinction cross sections
(Cext, m2 particle-1, right axes) calculated with MiePlot for
two nominal sphere diameters: 600 nm (middle) and 900 nm (bottom) with curves
for monodisperse size distributions (black dashed lines) and best fit curves
(solid black lines), along with measured wavelength-corrected SpEx extinction
spectra (Mm-1, dotted colored lines, left axes). Several spectra were
measured in each SpEx aerosol test, here seven spectra for PSL 600 nm and
six spectra for PSL 903 nm. Variability in the measured spectra arises from
variability in the concentrations of aerosols generated over the course of
each test.
The second set of tests measured the extinction spectra from 600 and 903 nm
PSL spheres (Fig. 4). Mie theory calculations were performed with MiePlot
v4.3 (http://philiplaven.com/mieplot.htm) using the
wavelength-dependent refractive index for PSLs provided with that code. The
wavelength dependence of the refractive index (Fig. 4, top panel) is
consistent with results reported by Washenfelder et al. (2013) and references
therein. The measured extinction spectra for 600 and 903 nm PSLs capture the
structure of spectra (Fig. 4) calculated from the extinction cross sections
(Cext, m2 particle-1). The 600 nm PSL spectra also
agree well with that shown in Chartier (2010). Sensitivity tests of the Mie
cross-section calculations to the particle radius and the dispersion of the
aerosol size distribution show peaks in the extinction spectrum shift to
longer wavelengths with increasing particle size, while the amplitude of the
peaks flatten as the size distribution broadens (Fig. S2). The high-spectral-resolution features in the spectral curves for these PSL tests arise
from the monodisperse size distribution of these manufactured particles. For
both the 600 and 903 nm particles, the best fit to the measured spectra was
found using slightly larger particle diameters (604 and 904 nm,
respectively) and dispersions with 1 and 0.5 % standard deviations,
respectively (Fig. 4). These results are within the manufacturer calibration
specifications for the two lots used here with 600±9 nm and 903±9 nm for the mean diameters with size distributions of 10.0 nm standard
deviation; 1.7 % CV (coefficient of variance) and 9.3 nm standard deviation; and 1.0 % CV,
respectively.
Comparison of SpEx with CAPS PMex extinction and nephelometer
scattering
Wavelength-dependent refractive indices are not known for many aerosols
generated in the laboratory. In such cases, comparisons to other measurements
are useful for testing the performance of the SpEx instrument. A series of
tests were performed using a variety of laboratory-generated aerosols ranging
from pure scatterers to various materials that absorb light (Table S2).
Fullerene soot and Aquadag are commercially available and have been suggested
to be used as surrogates for combustion-produced soot (Gysel et al., 2011;
Baumgardner et al., 2012; Beyersdorf et al., 2014). All of the measured SpEx
extinction spectra (30 s sampling period) were compared with extinction and
scattering measured at three visible wavelengths by CAPS PMex (450, 530, and
630 nm) and nephelometer (450, 550, and 700 nm) instruments, respectively.
CAPS PMex and nephelometer record 1 s data, which were averaged over the
30 s intervals coinciding with the SpEx sampling periods. The shape of the
CAPS PMex and nephelometer curves in these tests (e.g., Fig. 5a and c)
reflects the switching between filtered (zero extinction and scattering) and
unfiltered (nonzero extinction and scattering), with each period about 2 min
in length (90 s flush followed by 30 s sample collection by SpEx). CAPS
PMex and nephelometer periodically zero, resulting in channels that
occasionally switch between zero and sample values independently of the air
flow through SpEx. Note, part of the offset between the nephelometer and CAPS
PMex evident in Fig. 5a and c arises from the different wavelengths measured.
However, in Fig. 5b good agreement (typically within 15 %) is found with
the measured spectra from SpEx at each instrument's specific wavelengths.
Examples of laboratory tests: PSL 903 nm (a and b) and
Aquadag (c and d). One second CAPS PMex extinction (solid lines) and
nephelometer scattering (dashed lines) at 450 nm for both (blue curves, b),
530 and 550 nm, respectively (green curves, g), and 630 and 700 nm,
respectively (red curves, r) are shown in (a) and (c). SpEx extinction
spectra are shown (b and d) with 30 s averages (start and end times
indicated by vertical dashed and solid black lines, respectively, a
and c) of CAPS PMex and nephelometer data. Each numbered spectrum (e.g.,
SpEx_1) and 30 s average (e.g., AveCAPS_1 and
AveNeph_1) in (b) and (d) correspond in order to each
nonzero peak in (a) and (c).
Two examples from this series of aerosol tests are shown in Fig. 5: PSL
903 nm and Aquadag. Here, the SpEx extinction spectra are shown without any
smoothing and are plotted using the nominal (rather than corrected)
wavelengths. Most of the test data do not exhibit the high-spectral-resolution features evident in the PSL 600 and 903 nm tests. This
is due to the relatively broad size distributions of the particles generated,
compared to the monodisperse PSL particles. Hence, the 1 nm wavelength
correction is likely insignificant for the majority of laboratory and ambient
aerosols. For clarity, error bars of ± 6 Mm-1 (1σ) for the
extinction spectra are not shown. Similarly, error bars are omitted for CAPS
PMex and nephelometer average data, however, the markers are scaled to
approximate their ±1σ range. The two examples presented in Fig. 5
were chosen to illustrate the typically good agreement found between CAPS
PMex and SpEx extinction. They also illustrate the small difference between
extinction and scattering for aerosols that do not absorb light (e.g., PSL
903 nm), versus the large difference found for aerosols that do absorb light
(e.g., Aquadag).
Extinctions measured at 450 (top panel), 530 (middle panel), and 630
(bottom panel) nm by SpEx versus by CAPS PMex (as shown in Fig. 5) for all
aerosol tests listed in Table S2. Linear fits show the slope (b) and the
intercept (a) in each case, along with r2 values.
Twenty-two aerosol tests were performed (Table S2) recording two to seven spectra in
each test (typically three to four), resulting in a total of 87 spectra. SpEx data
were averaged over 5 nm corresponding to each CAPS PMex wavelength in order
to plot SpEx versus CAPS PMex extinctions (Fig. 6). Good agreement was found
between these measurements with slopes near unity at all three wavelengths
and r2 values of 0.94–0.96. These results, along with those of the
NO2 and PSL tests, indicate that SpEx provides accurate spectral
information within the precision of the instrument. A similar comparison
between SpEx extinction and scattering measured by the TSI nephelometer for
the purely scattering aerosol tests shows similarly good agreement (Fig. S3)
with slopes near unity and r2 values of 0.94–0.95. Single scattering
albedo values calculated from the nephelometer and SpEx data shown in Fig. S3
are given in Table S3.
Comparison of normalized aerosol extinction spectra
The magnitude of the extinctions measured for these aerosol tests ranged from
∼ 50 to 500 Mm-1 (Fig. 6), a reasonable range for polluted ambient
conditions. In order to compare the shape of each spectrum with others from
aerosols of similar size and different composition or of the same composition
with different sizes, each spectrum was normalized such that the maximum
extinction observed was scaled to a value of 1 (Fig. 7). A comparison of the
PSL spectral shapes illustrates the shift of peak extinction from longer to
shorter wavelengths with decreasing particle diameter from 903 to 200 nm
(Fig. 7a). This is consistent with the expectation that particles interact
most efficiently with wavelengths of light that are similar to their
diameter. This is also evident in the AS and citric acid tests where data
were obtained for 200 and 600 nm particle sizes (Fig. 7b).
Normalized mean spectra (wavelength-corrected) from aerosol tests:
PSL (a), AS and citric acid (b), montmorillonite and Luberon Natural Red
dusts (d) as a function of particle size; 600 nm (c) and PM1 (e) as a
function of aerosol composition.
Variability in the wavelength dependence of the spectra is also found among
600 nm particles of three different compositions (Fig. 7c). This variability
in shape arises from differences in the refractive indices of these
materials, along with some variability that may be due to minor differences
in the size distributions (e.g., the high-spectral-resolution structure in
the PSL spectrum arises from a nearly monodisperse size distribution, whereas
the DMA size-selected AS, and citric acid particles were more broadly
dispersed). Spectral comparisons of particles with a constant size illustrate
the utility of SpEx to measure spectral differences that arise from
variability in chemical composition.
Dust particles are known to more efficiently extinguish light across all
wavelengths and the SpEx spectra are consistent with this expectation
(Fig. 7). Five powders (Table S2, Fig. 8) were used for PM1 size
distribution tests, while only two (montmorillonite and Luberon Natural Red
powders) were used for PM2.5 tests. Note, four of the five dust samples
tested are sold as pigments; hence, their commercial names are used here
(Natural Pigments, Willits, California). This particular set of dust samples
was selected to represent a broad range of pigments (Fig. 8).
Pigmented dusts used in aerosol tests (from left to right:
montmorillonite, Blue Ridge Violet, Luberon Natural Red, Italian Yellow
Earth, goethite).
Normalized mean extinction spectra (wavelength-corrected) scaled to
their values at λ=700 nm for ∼ 200 nm particles
(left panel) and brown carbon (BrC) and black carbon (BC) surrogate compounds
(right panel).
The extinction spectra for montmorillonite are consistent with the previous
tests in that the maximum extinction is found at longer wavelengths for the
larger particles (Fig. 7d). However, for Luberon Natural Red, there is
surprisingly little difference between the PM1 and PM2.5 spectra.
This suggests the pigment of Luberon Natural Red affected the shape of the
spectrum in both tests. Consider the extinction spectra for all five PM1
dust samples (Fig. 7e). Both Blue Ridge Violet and Luberon Natural Red
exhibit peak extinctions between 600 and 700 nm, consistent with the visibly
reddish hue of these samples (Fig. 8), while both Italian Yellow Earth and
goethite peak between 500 and 600 nm. Only the bright white montmorillonite
peaks at the short UV wavelengths. The brown colored goethite exhibits the
smallest wavelength dependence in its spectrum, while montmorillonite
exhibits the largest dependence. These results indicate that the mineral
composition influences the spectral shape in addition to the dominant role of
size distribution on extinction.
As is evident in Fig. 7, small particles exhibit extinction peaks at the
shortest wavelengths measured by SpEx. Normalizing the mean curves to their
maximum value (at 300 nm in the remaining tests performed) has the effect of
visually suppressing the variability of such spectra in the UV region, while
highlighting differences in the longer wavelength visible range. Since we are
trying to investigate what aerosol information may be found in extinction
spectra, particularly in the UV, we have scaled the rest of the normalized
spectra to their values at 700 nm. This scaling produces the effect of
visually highlighting the differences in the spectral shapes in the UV
(Fig. 9).
Not all of the materials used generated enough aerosol mass in the liquid
nebulizer to allow for size selection. The slow flow rate of the DMA
(1–2 L min-1) used for the size selection from the nebulized
polydisperse aerosol leads to a highly diluted contribution in the
80 L min-1 air flow into the 17 L optical cell. Hence, 200 nm
diameter particle spectra were only obtained for PSL, AS, citric acid, and
2-carboxybenzaldehyde (2-CB), i.e., those materials that produced high
concentrations of particles from which to select the 200 nm size range. The
remaining spectra were obtained from the polydisperse size distributions
generated from Fullerene soot, Aquadag, and cinnamaldehyde.
The 200 nm laboratory test aerosols from four different solutions all exhibit
larger extinctions at 300 nm than 700 nm, ranging from a factor of 1.6 to 2.4
larger (Fig. 9, left panel). In addition, 190 nm soot particles generated
from a mini-CAST instrument burning propane have an extinction ∼3.25
times larger at 300 than 700 nm. The extinctions obtained from polydisperse
Fullerene soot, Aquadag, and 2-CB are ∼1.5 times larger at 300 than
700 nm. Unlike the other compounds tested, polydisperse cinnamaldehyde has a
strongly curved spectrum where extinction at 300 nm is 7.5 times larger than
at 700 nm (Fig. 9, right panel). The shape of this spectrum is suggestive,
especially since cinnamaldehyde has a molecular structure consistent with
expectations for a BrC compound. The characteristic trait of BrC is strongly
enhanced absorption in the UV spectral range. Given the variable sources of
BrC and the differing photochemical fates of the diverse chemical compounds
that are likely to contribute to BrC (e.g., Lee et al., 2014; Laskin et al.,
2015), there is no widely adopted standard BrC surrogate. Both 2-CB and
cinnamaldehyde were tested here specifically to try to identify a possible
BrC surrogate (based on Scheme 1 in Lee et al., 2014). Of the two,
cinnamaldehyde exhibits the curvature that might arise from enhanced UV
absorption (assuming that absorption contributes a significant fraction to
the extinction at the lower UV wavelengths). For the rest of this discussion,
we will treat cinnamaldehyde as a BrC surrogate compound.
Extinction Ångström exponents calculated for the compounds
in Fig. 10 based on visible wavelengths that can be measured by current
commercially available instrumentation and UV/visible wavelengths which can
be measured by SpEx.
CAPS PMex wavelengths
Difference
CAPS PMex
SpEx
Difference
Fit to all
(absolute)
wavelengths
wavelengths
(absolute)
SpEx data
Wavelengths
450 and 530
530 and 630
–
450 and 530
300 and 530
–
300–700
Fullerene
0.18
0.51
0.33
0.18
0.18
0.00
0.19
Aquadag
0.32
0.38
0.06
0.32
0.50
0.18
0.56
Low O : C soot
0.65
0.56
0.09
0.65
0.76
0.11
0.94
High O : C soot
1.07
0.86
0.21
1.07
1.17
0.10
1.26
Cinnamaldehyde
1.71
1.84
0.14
1.71
2.50
0.79
2.20
To test the spectral shapes of soot generated from propane in the mini-CAST,
two settings were used that generated either more or less oxygen in the soot
based on the work of Moore et al. (2014). Neither is strictly brown or black
carbon, but one has a higher O : C ratio than the other. Small particles,
∼ 40 nm in diameter, were generated in both tests. As expected the
“high” O : C soot exhibits enhanced UV extinction (via absorption)
compared to the “low” O : C soot (Fig. 10). Here, both soot tests show
greater enhancement at 300 nm than observed for either BC surrogate
(Fullerene soot or Aquadag), which is not surprising given the smaller
particle size generated by the propane flame. Neither soot test is as
strongly enhanced as cinnamaldehyde, even given its larger particle size.
While the comparison between the two soot tests should reflect differences in
composition, some of the difference among the three polydisperse curves
undoubtedly arises from size distribution differences. Hence the comparison
here is mainly to illustrate the different spectral shapes expected for BC
and BrC when absorption makes a significant contribution to the extinction
signal and to use the surrogates simply to provide context for the 40 nm
soot tests.
Normalized mean extinction spectra (wavelength-corrected) scaled to
their values at λ=700 nm for two smoke tests with relatively
higher or lower O : C ratios in the soot particles, with the former expected to
contain more BrC than the latter. Cinnamaldehyde is used as a BrC reference.
As discussed in the introduction, historically the extinction spectrum has
been described with an inverse power law relationship with wavelength as in
Eq. (1) (Ångström, 1929; Moosmüller and Chakrabarty, 2011). An
Ångström exponent can be calculated based on any pair of wavelengths
via Eq. (2) and should be constant if extinction is truly power-dependent.
Extinction Ångström exponents (αext) calculated from
two pairs of SpEx wavelengths (chosen to match those measured by CAPS PMex:
450 and 530 nm, and 530 and 630 nm) for the compounds shown in Fig. 10,
clearly exhibit some variability in αext depending on the
wavelength pair (Table 1). The differences between the two visible wavelength
pairs range from 0.06 to 0.33 (Table 1). Taking advantage of the capability
of SpEx to measure into the UV down to 300 nm, these visible wavelength
pairs may also be compared to a UV-visible wavelength pair, e.g., 300 and
530 nm (Table 1). Absolute αext differences between the 450,
530 nm pair and a SpEx-enabled 300, 530 nm pair were small for the BC
surrogates (0.00–0.18) but significant for the BrC surrogate
(cinnamaldehyde, 0.79). This result supports prior studies (e.g., Eck et al.,
1999; Schuster et al., 2006) indicating that for some aerosols, mathematical
descriptions other than power laws may better describe the spectral shape of
αext.
Curve fits (blue solid lines) applied to SpEx mean extinctions
normalized to 700 nm (black solid lines). Power law fits (left panels) and
polynomial fits (right panels) for AS 200 nm (top panels), polydisperse
Aquadag (middle panels), and polydisperse cinnamaldehyde (bottom panels). Fit
residuals (blue dots) are plotted above each curve fit. The residual is
calculated by subtracting the fit function value from the raw data value; hence, the units are the same as the normalized extinctions.
In addition to calculating αext for particular wavelength
pairs, power law line fits were performed for the mean spectrum obtained from
each aerosol test. Here, we modify Eq. (1) for the fitting as σext(λ)=Aλα where typically A∼1 and
α is constrained to always be negative. Power laws are typically used
when decades of data are spanned in log-log space over the relevant intervals
in x and y. However, over 300–700 nm in wavelength, none of the test
data spanned more than a decade. Most of the tests in log-log space exhibit
only a narrow range of extinction and nearly all of the test spectra exhibit
curvature. This curvature is why αext is sensitive to the
choice of wavelength pairs used in its calculation. In all cases polynomials
provided a better fit to the data than power law functions. If a data set is
well fit by a mathematical function, then the residuals (values resulting
from subtracting the fit values from the raw data values) should reflect
random noise distributed around zero. If the residuals exhibit a trend, then
the function used is not a good fit to the data. In all cases, the power law
fit residuals exhibited trends, whereas polynomial fits of order 7–10
provided good fits with residuals exhibiting random noise (e.g., Fig. 11).
These results suggest that the spectra obtained carry more information than
can be conveyed by a power law fit and that examination of features in
certain spectral ranges (e.g., in the UV for identification of BrC) may offer
greater insight into the characteristics of the aerosols when measured with
SpEx.
Conclusions
A new instrument, the spectral aerosol extinction (SpEx) instrument, for the
measurement of ambient aerosol extinction spectra over the 300–700 nm
wavelength range has been described and characterized in the laboratory with
NO2 and a variety of aerosols, generated from nonabsorbing materials,
dusts, smoke, and BC and BrC analogs. SpEx is still under development with
ongoing work to further reduce noise, improve precision, and reduce the
detection limit. It is anticipated that prior to the next field deployment,
the detection limits of individual spectra will be significantly reduced.
Nonetheless, as reported herein SpEx performed well in all laboratory tests.
Good agreement with theoretical calculations for NO2 and PSL spheres was
found. Correlations between SpEx and CAPS PMex extinction measurements at
450, 530, and 630 nm showed good agreement for all of the aerosol tests with
r2 values of 0.94–0.96. The aerosol test results reveal distinct
differences in the spectral shapes obtained from the aerosols both as a
function of size and composition. These results suggest that measurements of
extinction spectra of ambient aerosols may also reveal unique characteristics
related to size and composition over this wavelength range.
Polynomials were found to fit the measured spectra better than power laws due
to the curvature of the spectra. This result indicates that detailed spectral
information as provided by SpEx may offer greater insight into aerosol
characteristics than can be obtained from commercially available instruments
that make measurements at only a few wavelengths. The observed spectral
curvature resulted in differences in αext depending on the
wavelength pair used in calculations based on Eq. (2), as well as that
obtained from a power law fit to the spectrum. In some cases the differences
were small, suggesting extrapolation of αext into the UV from
paired visible wavelengths may be acceptable, but large differences found for
the BrC surrogate indicate extrapolation cannot be applied to such materials.
With its lower wavelength limit of 300 nm, SpEx will facilitate exploration
of spectral UV optical characteristics.
SpEx is similar in some respects to the broadband cavity-enhanced
spectroscopy (BBCES) instrument (Washenfelder et al., 2013, 2015). BBCES has
been used in the laboratory to retrieve wavelength-dependent complex
refractive indices (m(λ)) for specific aerosols (Washenfelder et
al., 2013) and it has been used in the field to perform an optical closure
study to assess the contribution of scattering to extinction, as well as the
contribution of BC and BrC to aerosol absorption at 365 nm (Washenfelder et
al., 2015). Further in the latter study they found in rural Alabama in summer
that BrC mass was principally associated with biomass burning; biogenic SOA
contributed only a minor fraction. Note, their source apportionment relied
both on BBCES measurements and absorption measurements obtained with the
PILS–LWCC (Particle Into Liquid Sampler coupled to a Liquid Waveguide
Capillary Cell). BBCES uses two cavities with light emitting diode sources
that offer a combined spectral range of 355–420 nm (Washenfelder et al.,
2015). The smaller cavities (each with an approximately 1.5 L volume),
slower flow rate, and lower detection limit makes BBCES more suitable than
SpEx for retrievals of m(λ) due to the need to size select aerosols
to perform the retrieval. Although we anticipate using SpEx in closure
studies for specific wavelengths similar to that performed by Washenfelder et
al. (2015), its real power lies in its ability to obtain detailed aerosol
spectra over a wide spectral range.
When applied to ambient aerosol populations, it is anticipated that this
spectral measurement capability will provide a novel tool for characterizing
aerosols, enabling investigations of their optical evolution in the ambient
environment. SpEx is a rack mountable instrument that may be deployed in
mobile laboratories allowing for studies of the optical evolution of
extinction as fresh emissions are transported downwind from sources. Such
measurements offer a complement to related extant in situ measurements of
aerosol chemical, physical, and optical properties. For example, SpEx
extinction spectra coupled with PILS–LWCC absorption spectra will allow for
the calculation of scattering spectra, single-scattering albedo spectra, and
other spectral optical properties. This instrument for measuring in situ
aerosols also offers a new measurement capability for calibration and
validation studies of aerosol retrievals from remote sensors.
In a recent review of BrC (Laskin et al., 2015), results from studies
examining the role of absorbing aerosols on photolysis rates of atmospheric
gases and biogenic gas emissions were reported that offer additional
intriguing possibilities for the utility of the spectral information that can
be obtained with SpEx. By acquiring data on aerosol light extinction down to
300 nm in the ambient environment, SpEx may help inform our understanding of
how UV-absorbing aerosols influence the photochemical environment for ambient
gases (i.e., via their potential to reduce photolysis rates) and affect
biogenic emissions of gases such as isoprene by altering the light
environment. In their discussion of aerosol absorption, Laskin et al. (2015)
note that wavelength-resolved measurements are rare and are preferable to
assumed power law wavelength dependencies. SpEx will help address the need
for more wavelength-resolved measurements to examine the complex interactions
in our ambient atmospheric–biogeochemical environment.