Introduction
Black carbon is a primary aerosol tracer for high-temperature combustion
emissions and can be used to characterize the time evolution of its sources
(Lack et al., 2014). Investigations performed on ambient black carbon (BC) mass
equivalent concentrations have shown a stronger correlation
with adverse public health effects than the ones observed for the total mass
of particulate matter (Janssen et al., 2011, 2012; Grahame
et al., 2014; Olstrup et al., 2016). Black carbon absorbs sunlight very
efficiently and heats the atmosphere with the top-of-the-atmosphere forcing
exceeding 1 W m-2, and even though its lifetime is short compared to
CO2, it is still the second-most-important warming agent (Bond et al.,
2013). Models show that its short lifetime opens up the possibilities for
short-term mitigation of atmospheric warming (Shindell et al., 2012). While
health effects are of interest in the context of local air pollution,
climate change is investigated on regional and global scales, and co-benefits
from black carbon emission reductions are of great interest (Smith et al.,
2009). Measurements are conducted at locations suitable for all three
different scales, with absorption filter photometers being the most common
instrumentation employed to determine ambient black carbon mass
concentrations.
Filter photometers measure the change of light transmitted by or reflected
from the fiber filter on which particles are collected. These methods
feature high sensitivity and high time resolution. The measurements,
however, depend on the multiple light scattering within the filter and are
prone to several artifacts, including filter-loading effect and particle
scattering (Lack et al., 2014). Herein we focus on the filter-loading effect
(FLE), a saturation of the instrumental response due to the accumulation of
the sample in the filter matrix (Bond et al., 1999; Weingartner et al.,
2003; Moosmüller et al., 2009). This nonlinearity depends on the filter
material and the type of instrument (Virkkula et al., 2007; Collaud Coen et
al., 2010; Hyvärinen et al., 2013). It is important to compensate for
FLE to obtain accurate measurements of black carbon concentration and for
the determination of aerosol absorption spectra, which are used for source
apportionment (Kirchstetter et al., 2004; Sandradewi et al., 2008).
The filter-loading effect – that is, the reduction of the measurement
sensitivity with increasing filter loading – in Aethalometers has been
described by several models (Weingartner et al., 2003; Arnott et al., 2005;
Schmid et al., 2006; Virkkula et al., 2007; Park et al., 2010; Collaud Coen
et al., 2010; Drinovec et al., 2015). Offline methods for the quantification
of FLE are based on jumps of reported BC values just before and
immediately after the filter change (Weingartner et al., 2003; Virkkula et
al., 2007) or the BC vs. ATN method (Park et al., 2010; Segura
et al., 2014; Drinovec et al., 2015). These low-time-resolution estimations
of FLE showed big differences in the effect between seasons (Virkkula et
al., 2007). Uncoated soot induces a higher loading effect compared to coated
soot (Weingartner et al., 2003), and Virkkula et al. (2015) showed that the
loading parameter depends on the backscatter fraction and the single-scattering albedo (SSA).
Traditionally, Aethalometer data have been post-processed using fixed
parameters characterizing the loading effect. With the introduction of the
dual-spot Aethalometer it has become possible to measure FLE with high time
resolution (Drinovec et al., 2015). This allows for a much better
characterization of effects governing FLE. In this paper, we are testing the
hypothesis that the filter-loading effect depends on the optical properties
of particles present in the filter matrix, especially on black carbon
particle coating. Under laboratory conditions, coating can enhance optical
absorption of soot particles by up to factor of 2 (Weingartner et al.,
2003; Schnaiter et al., 2003). Ambient experiments show a smaller effect
from 6 to 60 % (Cappa and Wilson, 2012; Moffet and Prather, 2009; Liu et
al., 2015). Determination of the mixing state using FLE may also reduce
uncertainties in radiative forcing estimates (Jacobson, 2001; Shiraiwa et
al., 2008). There are two extreme mixing state models: an internal and an
external mixture as defined by Seinfeld and Pandis (2006), where the
external mixture represents two separate populations of two types of
particles and where an internal mixture represent one population of
particles composed of a homogenous mixture of materials from the two types.
Because black carbon cannot be homogenously mixed with non-absorbing
material, a third mixing state was introduced: a core-shell state (Jacobson,
2000). We use the term “mixing state” to refer to either of these states,
most notably the external and the core shell.
We conducted field campaigns in contrasting environments to determine the
influence of sources, particle age and coating on the magnitude of the FLE.
We show that the FLE during the 1-year-long campaign in Paris features
distinct winter and summer patterns. Winter dependence of the FLE is
discussed in the context of presented laboratory experiments with BC of
different particle size. Winter and summer intensive campaigns in Paris and
Payerne allowed us to determine a coating factor, which represents the
relative contribution of the material available for coating. The FLE
behavior during summer reveals an influence of particle coating on the FLE.
As particles get coated by secondary material during atmospheric ageing, we
show, using back-trajectory analysis, that the FLE depends on the particle
age. The ambient campaign in Ljubljana quantifies the increase of the
loading effect when the particle coating is removed.
Materials and methods
Instrumentation
BC and absorption coefficients are characterized with the dual-spot
Aethalometer (Magee Scientific Aethalometer model AE33) with real-time
loading compensation (Drinovec et al., 2015). The inlet air stream of the
AE33 is split, and the sample is collected on two filter spots concurrently
within the instrument. The flow through each of the two spots is different,
so the loading rates on the respective sample spots are different.
Different loading rates cause the accumulation of the sample to be different
between the two spots, resulting in a different magnitude of FLE between the
spots. Measurement of FLE enables the compensation of the data – using the
parametrization described in Drinovec et al. (2015), the compensation
parameter k can be derived. This parameter describes the linear
decrease of the instrumental sensitivity with loading of the spot with the
absorbing sample for each wavelength of light at which the measurement is
performed. The absorption coefficients are calculated at all wavelengths, and
BC is determined from the measurement at the wavelength 880 nm.
Absorption and BC are compensated using the following equation:
BCcompensatedλ=BCnon-compensatedλ1-kλ×ATNλ,
where ATN is optical attenuation of light passing through a sample
loaded filter. k1, k2, k3,
k4, k5, k6 and k7 are
compensation parameters determined at the following wavelengths: 370, 470,
520, 590, 660, 880 and 950 nm, respectively. The absorption Ångström exponent
(AAE) (Ångström, 1929) is calculated from absorption
coefficient b (Drinovec et al., 2015) measured at 470 and 950 nm: AAE =ln(b470 nm/b950 nm)/ln(950/470).
The instruments were run at 5 L min-1 total sample flow and 1 min time resolution.
The measurement of non-refractory submicron (NR-PM1) major chemical
species (i.e., organic matter, nitrate, sulfate, ammonium and chloride) was
performed using an Aerodyne® aerosol mass
spectrometer (AMS) and/or an aerosol chemical speciation monitor (ACSM),
fully described elsewhere (Canagaratna et al., 2007; Ng et al., 2011;
Fröhlich et al., 2013). Briefly, PM2.5 aerosols are sampled at
3 L min-1 and then sub-sampled at about 85 mL min-1 through an aerodynamic lens,
focusing submicron particles onto a 600 ∘C heated conical tungsten
vaporizer where non-refractory material is flash-vaporized and
quasi-instantaneously ionized by electron impact at 70 eV. The AMS mainly
differs from the ACSM system by allowing for size-segregated measurements of
the chemical composition by using a chopper in the AMS. Many AMSs also use
advanced mass spectrometers with higher resolution and better detection
limits. However, for the measurements of NR-PM1 species in ambient air,
a good agreement is commonly observed between data from these two
instruments (Crenn et al., 2015; Fröhlich et al., 2015).
Individual aerosol particles were analyzed using a field emission gun scanning
electron microscope (FEG SEM, JEOL JSM 7100F TTLS) coupled with an
energy-dispersive X-ray (EDX) system (X-Max Large Area Analytical Silicon
Drift Detector, Oxford Instrument). Aluminum foils used for sampling of
aerosol particles (sampling described in Sect. 2.2) were mounted directly
on the sample holder by double-sided conductive carbon tape. SEM-EDX
acquisitions were performed under high vacuum at 5 keV accelerating voltage.
Micrographs were acquired by a secondary electron detector at 4–6 mm
working distance (WD). The samples were first scanned under low
magnification to obtain the overall distribution of the particles and
afterwards closely examined at magnifications from 10 000 to 50 000×.
Representative areas were chosen for further EDX analysis, which was focused
on the identification of the coating material of soot agglomerates. The
microanalysis was performed at 10 mm WD with acquisition time of 50 to
150 s. Since the spatial resolution of the EDX analyses is around 1 µm,
the obtained results were considered as a qualitative assessment of particle
elemental composition.
Measurement campaigns
The Paris campaigns were conducted at the SIRTA observatory located about 20 km south of Paris,
France (48∘42′47′′ N, 2∘12′29′′ E). The observatory sits on a plateau, about 160 m a.s.l., in a
semi-urban environment being divided between agricultural fields, wooded
areas, housing and industrial developments. The prevailing winds are from
the southwest, blowing air of maritime origin over the site (e.g., Crippa
et al., 2013). Northeasterly winds also occur quite frequently under
high-pressure systems. The latter conditions are associated with the advection of
continental air masses as well as of polluted air from the Paris
metropolitan area. Beside continuous AE33 and ACSM measurements since 2013,
results from two intensive European Monitoring and Evaluation Programme (EMEP)
campaigns in summer 2012 (15 June–10 July 2012) and winter 2013 (24 January–16 February 2013) were available for
the present study.
The Payerne station is a rural background air quality monitoring station
located in southwestern Switzerland (46∘48′47′′ N,
6∘56′40′′ E, 489 m a.s.l.), between the Jura and the Alps.
It is situated about 1 km southeast of the small rural town of Payerne
(Switzerland). The site is surrounded by agricultural land (grassland and
crops), forests and small villages. Intensive campaigns took place in
summer (15 June 2012–10 July 2012) and winter (24 January 2013–16 February 2013). In addition to the Aethalometer AE33, a high-resolution
ToF-AMS (Aerodyne Research) was operated throughout the campaigns.
The Ljubljana campaign was held between 6 August 2014 and 12 September 2014
in a suburban background location, 10 km southwest of Ljubljana, Slovenia
(46∘0′58′′ N, 14∘24′42′′ E, 293 m a.s.l.).
The site is at the edge of a village surrounded by agricultural fields and at
a distance of 500 m from a highway. It lies in the Ljubljana basin, which is
surrounded by hills rising from 500 to 1700 m above the valley floor. The
Ljubljana campaign used four AE33Aethalometers measuring four different
ambient air streams: non-treated, dried, thermodenuded, and thermodenuded
and dried samples. Ambient, dried and thermodenuded air streams had separate
inlets; the thermodenuded sample was separated in two streams, one connected
directly to the Aethalometer and the second guided through a silica gel drier.
The Dekati thermodenuder was set to a temperature of 300 ∘C with
10 L min-1 sample flow, which was split for the two Aethalometers. The
thermodenuder consists of a heated section and an air-cooled denuder where
semi-volatile compounds are adsorbed onto activated charcoal. The estimated
residence time for 10 L min-1 flow was 0.5 s. A diffusion membrane drier
(Magee Scientific, Aerosol Inlet Dryer) and a silica gel drier with dew
point reductions of 11.0 ± 1.0 and 11.5 ± 1.0 ∘C, respectively, were used. A scanning mobility particle sizer
(SMPS) with a differential mobility analyzer (DMA) (TSI, Model 3080)
connected to a water-based condensation particle counter (CPC) (TSI, Model 3785
UWCPC) was used to acquire the particle size distribution every 5 min.
A Dekati low-pressure impactor (DPLI) was used to sample particles into 13 size fractions
(from 30 nm to 10 µm) for SEM and EDX spectroscopy analysis. Impactor
samples were collected on aluminum foils (Dekati, CF-300) covered by a thin
layer of Apiezon L grease. Impactor stage 4 with an expected particle size
range between 170 and 260 nm was analyzed.
The K-puszta campaign took place at the K-puszta observatory (Hungary,
46∘58′12′′ N, 19∘33′ E, 125 m a.s.l.) during January
and February 2013. During winter ambient aerosols are dominated by wood
burning emissions from household heating. The laboratory biomass burning
profile was obtained during an experiment performed at the Paul Scherrer
Institute (Bruns et al., 2015). Laboratory diesel exhaust was characterized
at Aerosol d.o.o. in Ljubljana from a Euro 3 diesel engine. The exhaust was
transferred into the smog chamber, where it was diluted and analyzed.
Ljubljana roadside measurements were conducted during November 2012 in the
city center, close to a heavy-traffic road.
The Reno laboratory experiment was conducted at Desert Research Institute
(Reno, Nevada, USA). Pure mustard oil was burned in a simple ceramic bowl
with a cotton wick. The diluted emissions were collected and injected into a
distribution chamber (50 L) from which the sample was routed to multiple
instruments: the Aethalometer model AE33 run at 1 s time resolution, the
Photoacoustic Soot Spectrometer (model PASS-3, Droplet Measurements
Technologies) operated at 2 s time resolution, and the SMPS (TSI, DMA
model 2080 with CPC model 3775). Ponderosa pine needles (∼ 50 g) were
combusted in a biomass combustion chamber, a close replica of which has been
described by Tian et al. (2015). The combustion process could be divided into
the flaming and smoldering phases. Focusing on the flaming phase, we injected
the emissions from the first 35 s of the combustion into the distribution
chamber. The sample from the distribution chamber was analyzed for
approximately 15 min (time series of BC is shown in Supplement Sect. S.6).
PASS-3 absorption and scattering measurements were used to calculate the
SSA (Lewis et al., 2008).
Trajectory analysis
To illustrate air mass origin during specific pollution episodes during the
Paris and Payerne campaigns (please see Sect. 2.2), 72 h back trajectories
were calculated every 3 h using the PC-based version (v.4.9) of HYSPLIT
(Stein et al., 2015) with weekly Global Data Assimilation System (GDAS)
meteorological field data. Back trajectories were set to end at SIRTA and
Payerne at 500 m above ground level (a.g.l.).
The potential contribution source function (PSCF) was used to investigate the
potential advection of particulate material over large geographical scales
during the different measurement campaigns. In this methodology, the
probability that an air parcel may be responsible for high concentrations
measured at the receptor site is calculated based on back-trajectory
analyses, as described in Polissar et al. (1999). Briefly, at each ijth
grid cell, probabilities are calculated as
PSCFij=mijnij,
where nij represents the total number of back trajectories
passing through each ijth cell, and mij the number of
back trajectories passing through the same cell that are associated with
measured concentrations higher than a user-defined threshold. A
0.5∘ × 0.5∘ grid cell was used, covering western Europe.
Similarly to Waked et al. (2014), wet deposition was roughly estimated with
precipitation data along each trajectory, assuming that even low
precipitation would clean up the air parcel. For graphical purpose, results
were smoothed by a Gaussian filter and eventually normalized (see Fig. 9
with color scales from 0 to 1).
Typical wavelength dependence of the filter-loading compensation
parameter k obtained for ambient and laboratory campaigns. Campaign average
k and absorption Ångström exponent (AAE) are presented. The 590 nm
channel data for Payerne summer were not available (NA) due to instrument
malfunction.
K-puszta
Laboratory
Payerne
Laboratory
Ljubljana
Paris
Payerne
winter
biomass
winter
diesel engine
roadside
summer
summer
burning
exhaust
AAE
1.55
1.40
1.35
1.04
0.97
1.06
1.16
Filter-
370 nm
0.0047
0.0046
0.0044
0.0041
0.0045
0.0041
0.0043
loading
470 nm
0.0055
0.0052
0.0049
0.0045
0.0048
0.0036
0.0039
parameter
520 nm
0.0059
0.0056
0.0051
0.0046
0.0050
0.0035
0.0038
k at the
590 nm
0.0067
0.0060
0.0055
0.0048
0.0051
0.0034
NA
wavelength
660 nm
0.0072
0.0063
0.0059
0.0050
0.0052
0.0032
0.0034
880 nm
0.0081
0.0072
0.0063
0.0055
0.0052
0.0024
0.0019
950 nm
0.0084
0.0074
0.0064
0.0056
0.0053
0.0023
0.0012
Results
Variability of filter-loading effect
The magnitude of the FLE for aerosols collected on a filter is dynamically
determined by the dual-spot Aethalometer (Drinovec et al., 2015). The FLE
depends on the entirety of the sample accumulated in the filter matrix up to
the time of the measurement. The value of the FLE compensation parameter
k is calculated for each of the seven measurement wavelengths,
yielding a specific spectral fingerprint, which differs substantially
between different measurement campaigns (Fig. 1). The campaign-averaged
values of the loading parameter k for short wavelengths show
limited variability with a range of 0.004–0.005, while the variability in
the near infrared is much greater with a range of 0.001–0.008 (Table 1
and Fig. 1).
Table 1 also reports the average values of the absorption Ångström
exponent obtained for each campaign. Occurrence of higher k values
appears to be linked to higher values of the absorption Ångström
exponent, which is commonly used as a tracer for biomass burning aerosols
(Kirchstetter et al., 2004; Sandradewi et al., 2008). This is evident for
both laboratory and ambient data, with the most extreme case for field campaigns
being K-puszta winter data. For fresh diesel exhaust (laboratory and
Ljubljana roadside measurements), the obtained k values fall in the
middle of the parameter range, whereas lower k values are obtained
for the summer ambient (Paris and Payerne) campaigns. This variability will
be investigated in more detail in the following section.
Typical spectral fingerprints of the FLE compensation parameter k
obtained during ambient and laboratory campaigns. Campaign/experiment average
values are presented.
Black carbon mass concentration and AAE measured at the Paris urban background site (SIRTA
observatory) during 2013.
Filter-loading parameter k measured at the Paris urban background
site (SIRTA observatory) in 2013.
Seasonal variation of filter-loading effect
One-year-long black carbon mass concentrations at SIRTA show significant
day-to-day variability with yearly average and standard deviation of 1068 ± 1123 ng m-3 (Fig. 2).
Average concentrations are 913 ± 932 and 1263 ± 1299 ng m-3 for the
hot (April–September) and cold (October–March) periods, respectively.
The sources of black carbon impacting the measurement site vary during the
year; during summer the AAE is close to 1, indicating the dominance of traffic emissions, while during
winter its value is 1.4, suggesting the importance of biomass burning from
residential heating (Kirchstetter et al., 2004; Sandradewi et al., 2008),
which is consistent with previous BC source apportionment studies
performed in the Paris area during intensive campaigns (e.g., Favez et al.,
2009; Petit et al., 2014) as well as long-term investigations (Petit et al.,
2015).
Concomitantly with the absorption Ångström exponent, the parameter
k shows large seasonal variations (Figs. 3 and 4). Virkkula et
al. (2015) analyzed the k wavelength dependence by calculating the
slope of k vs. λ. It was shown that the slope is extremely
well correlated with the compensation parameter k6 (measured
at 880 nm), so either of the two can be used to describe the variation of
the parameter k. For the sake of simplicity, we focus our further
analysis on k6. High values of k6, between 0.006
and 0.012, are observed during winter, while in summer k6
remains low, featuring still large oscillations (0–0.005), with episodes
of low k6 that can last for several days, indicating stronger
influence of weather conditions than of daily cycles.
Monthly spectral fingerprints of the filter-loading effect
compensation parameter k, obtained during the Paris campaign (SIRTA
observatory) in 2013.
Figure 4 shows monthly spectral fingerprints of the parameter k,
which are in agreement with the offline analysis of the Aethalometer data
(Virkkula et al., 2007, 2015): we measured the lowest
values of the parameter k during summer (red) and highest during
winter (blue) months, with spring and autumn somewhere in between. Again,
the variation of the parameter k6 roughly correlates with the
AAE, except for the months of April, May and October, when the
k6 monthly mean varies between 0.0028 and 0.0041, while the AAE mean changes very little between these months (Table 2 and
Fig. S5 in the Supplement). This seems to indicate that two distinct factors – meteorology
and sources – may significantly influence the value of the parameter
k.
To analyze the influence of black carbon sources, we investigated in more
detail the dependence of k6 on the AAE during the
heating season. The November–December period was analyzed because of
stable weather conditions. The results confirm that for AAE values above 1.2
the loading parameter k6 increases with AAE (Fig. 5), indicating higher values for biomass burning emissions
than traffic emissions. This is in agreement with the laboratory biomass burning
experiment, where high k6 values were also obtained (Fig. 1). Figure 5 shows a value for k6 close to 0.0055 at an
absorption Ångström exponent of 1, as expected for diesel exhaust
from source and ambient measurements (Table 1). We hypothesize that higher
values of k6 for biomass burning compared to diesel exhaust
are related to the difference in black carbon agglomerate size, which is
investigated in more detail in the following section. Indeed, the diesel
engine exhaust soot mobility diameter was reported in the 30–100 nm range
(Harris and Maricq, 2001; Keskinen and Rönkkö, 2010; Ning et al.,
2013), in contrast to the 130–160 nm range for wood-stove emissions
(Laborde et al., 2013) and 260–590 nm for wildfires (China et al., 2013).
In special cases of large-scale sooty and turbulent fires with long flame
residence time it is possible to obtain super-aggregates with geometric
diameter larger than 1 µm (Chakrabarty et al., 2014).
Monthly averages and standard deviation of black carbon (BC) mass
concentration, AAE and
compensation parameter k6 for the Paris campaign (SIRTA observatory) in
2013.
Month
BC (ng m-3)
AAE
k6
February
1390 ± 1320
1.28 ± 0.47
0.0059 ± 0.0023
March
1370 ± 1070
1.29 ± 0.23
0.0054 ± 0.0024
April
570 ± 640
1.17 ± 0.43
0.0041 ± 0.0016
May
640 ± 660
1.16 ± 0.46
0.0028 ± 0.0011
June
750 ± 670
1.10 ± 0.37
0.0023 ± 0.0010
July
1250 ± 1000
1.04 ± 0.31
0.0023 ± 0.0010
August
870 ± 690
1.09 ± 0.32
0.0018 ± 0.0012
September
1310 ± 1380
1.10 ± 0.29
0.0027 ± 0.0010
October
960 ± 910
1.18 ± 0.33
0.0035 ± 0.0015
November
1180 ± 950
1.25 ± 0.25
0.0052 ± 0.0009
December
1580 ± 1930
1.41 ± 0.26
0.0063 ± 0.0010
Correlation between the FLE compensation parameter k6 and the
AAE for November and December 2013 in
Paris (SIRTA observatory).
Filter-loading parameter k for laboratory measurements of
emissions from mustard oil lamp and flaming Ponderosa pine
needles (a). Normalized aerosol number size distributions for both
sources (b).
Filter-loading parameter k6, aerosol mobility diameter (mode of
number size distribution), single-scattering albedo (SSA) at 532 nm and
AAE of soot generated by mustard oil lamp and flaming Ponderosa pine needles.
k6
Particle size
SSA532 nm
AAE
mode (nm)
Mustard oil lamp
0.0074 ± 0.0003
210
0.222 ± 0.003
1.02 ± 0.02
Flaming needles
0.0064 ± 0.0014
50
0.43 ± 0.05
1.25 ± 0.03
Particle size and the filter-loading effect
To investigate the influence of black carbon particle size on k, we
characterized emissions from a mustard oil lamp and from flaming ponderosa
pine (Pinus ponderosa) needle combustion under laboratory
conditions. Mustard oil burning produces particles with a mode of mobility
number size distribution of 210 nm (Fig. 6b), which compares fairly well
with the median aggregate mobility diameter of 297 ± 74 nm reported by
Chakrabarty et al. (2013) for mustard oil lamp emissions. These particles
are dominated by black carbon as indicated by an absorption Ångström
exponent of 1 and low single-scattering albedo of 0.222 (Table 3). Our
Ångström exponent of 1 (470, 950 nm) is somewhat lower than that
determined by Chakrabarty et al. (2013, Fig. 2a) as 1.331 ± 0.004 (405, 781 nm)
with a three-wavelength photoacoustic instrument. Our single-scattering albedo is in agreement with the lower values of 532 nm
SSA measured by Chakrabarty et al. (2013, Fig. 4a) with an
integrated photoacoustic nephelometer. Chakrabarty et al. (2013) argue that
higher values (up to ∼ 0.8) of single-scattering albedo shown
in their Fig. 4 were caused by flickering of the flame increasing organic
matter emissions, which may also have caused their increased absorption
Ångström exponents. Flaming pine needle combustion produces
considerably smaller particles with a mean diameter of 50 nm. We sampled
only the flaming phase of Ponderosa pine needle combustion, where the
emissions feature low values of the AAE (i.e., 1.25) and
SSA (i.e., 0.43) as shown in Table 3. This is in agreement with the
low SSA of 0.41 ± 0.14 at 532 nm that can be derived from the
ratio of fuel-based emission factors of absorption and extinction cross
sections measured by Chen et al. (2007, Table 1) with a photoacoustic
absorption instrument and a cavity ring-donw (CRD)/cavity-enhanced detection (CED) extinction instrument, respectively, for
flaming combustion of Ponderosa pine needles. We observed a larger
filter-loading effect for mustard oil lamp emissions (Fig. 6a), which we
believe is related to the larger particle size. This might explain the
correlation between k and AAE in Fig. 5. The values of
the loading parameter for both sources and their wavelength dependence are
similar to the ambient observations in winter (Figs. 1 and 4). The loading
parameter values for fresh combustion products seem to be influenced by the
size of the combustion products, with an approximately 15 % increase in the
infrared (IR) wavelength region for an aerosol mobility diameter increase of about 4 times. The increase of the loading parameter in the IR with size is
consistent with the reported soot particle diameters for different ambient
sources (Ning et al., 2013; China et al., 2013). The difference in size is
accompanied by the difference in the absorption Ångström exponent
and single-scattering albedo. In addition to the size dependence of
k6, the dependence of the loading parameter on SSA we
report here agrees with the one reported for Nanjing (Virkkula et al., 2015)
– the decrease in SSA increases the loading parameter.
Influence of the mixing state on the filter-loading effect: material
available for coating
The second factor influencing k was assumed to be related to the
meteorological conditions. To test this assumption, high-time-resolution
variation of k was investigated during EMEP campaigns taking place during
summer and winter in Paris and Payerne (Fig. S1). During summer, we observed
periods with k6 close to 0, which lasted for several days with short
intervals with k6 close to 0.005. During the winter campaigns, we
observed higher values of k6 (up to 0.01) at both locations, whereas in
Paris there were also two periods with low k6 values. To investigate the
dependence of the loading parameter k on the available coating material, we
combined the Aethalometer data with the Payerne and Paris AMS/ACSM data. Here
we define the coating factor (CF) as a mass of the potential material
available for coating (a sum of sulfate, ammonium, nitrate and organic mass)
normalized to the black carbon mass:
CF=SO42-+NO3-+NH4++ORGBC.
The time series of CF and k6 from the Paris summer
campaign shows a similar time variation pattern (Fig. 7). High CF
coincides with low value of parameter k6, and vice versa.
However, there is a time lag between both measurements which is related to
the way the loading parameter k is determined (Drinovec et al.,
2015): the parameter k is a cumulative property of all particles on
the filter; hence all particles on the filter, collected from the last time
the tape has been advanced, contribute to the loading effect. This means
that the time lag varies with and depends on the ambient black carbon
concentration. The second reason is the geometric smoothing employed in the
algorithm, which takes the last k value determined from the
previous filter spot into account. The time lag impairs the k6
vs. CF correlation resulting in a higher standard deviation of
k6 averaged in each CF bin (Fig. 8).
Coating factor (CF) and filter-loading parameter k6 time series
during the Paris summer campaign. Note the inverted scale for k6.
Correlation between the filter-loading parameter k6 and the
CF during winter and summer campaigns in Paris and Payerne.
Hourly data are averaged in CF bins 2 units wide. The error bars represent
standard deviation for data points inside each bin.
We have observed a change of k6 with increasing coating factor
for all four campaigns. The decrease is of different magnitude and shape for
summer and winter. The decrease is larger in summer than in winter. In
summer in Paris and in Payerne in both seasons, the relationship is linear,
while the shape is more complex in Paris in winter.
The results are most pronounced for the Paris summer campaign with the
largest absolute slope of the linear regression line. In Payerne in summer,
the slope is slightly lower. In summer, k6 reaches 0 at
coating factor values of approximately 30 for both locations (Table 4).
These results show that during summer the coating material causes a
reduction of the filter-loading effect. In Paris, CF is much larger
than it is in Payerne. Because most of the non-refractory material in Paris
during the summer is of secondary origin (Freutel et al., 2013), it is
possible to form internally mixed aerosols in the atmosphere. The
correlation of particle size with coating factor in Paris supports this
assertion (Fig. S2).
Fitting parameters of the filter-loading parameter k6 versus
the coating factor and AAE during winter and summer campaigns in Paris and
Payerne.
slope
intercept
AAE
(10-4)
at k6=0
Paris summer
-1.2 ± 0.1
27 ± 3
1.06
Payerne summer
-0.8 ± 0.5
35 ± 20
1.16
Paris winter
-0.8 ± 0.3
57 ± 22
1.34
Payerne winter
-0.2 ± 0.1
340 ± 180
1.35
The fitted slope obtained for the Paris winter campaign is lower compared to
summer, and k6 reaches 0 at CF = 57, with the dependence being
more complicated than the linear one observed in summer. There is almost no
decrease of k6 with increasing CF during winter in Payerne. This might
be a consequence of biomass burning emissions which are dominated by organics
usually externally mixed with black carbon – as confirmed for the Paris
region during the MEGAPOLI campaign (Healy et al., 2013). The amount of black
carbon and organics from biomass burning depends strongly on the combustion
conditions, with black carbon being created at high-temperature flaming
combustion and organics at low-temperature smoldering combustion (Chakrabarty
et al., 2013; Pagels et al., 2013). Especially during summer, elevated
amounts of secondary non-refractory material are available for coating, and
we observe a good correlation of k6 with CF, which indicates that
coating is responsible for reduction of the filter-loading effect. The
decrease of k6 with increasing CF is at first counterintuitive when it
is compared with the influence of the measured aerosol mobility diameter on
k6. It was observed that the increase in the mobility diameter may
induce a k6 increase (Fig. 6), and one could speculate that the
availability of material for coating black carbon cores, and hence a high
value of CF, would increase the aerosol size and thereby increase the
k6 value. In fact, the contrary happens; increasing CF decreases
k6. However, increase in CF should increase SSA as well, and it has
been shown for ambient aerosols in Nanjing (Virkkula et al., 2015) that the
increase in SSA decreases k6. To resolve this question in the
following sections, we will investigate first the influence of aerosol age
and continue with removing or reducing aerosol coating.
Aerosol age and origin influence the filter-loading effect
The influence of the aerosol age was indirectly investigated by using
the PSCF. We divided the dataset into
two subsets distinguished by k6 being bigger or smaller than
the campaign average of 0.002. The analysis was carried out for the summer
period at Paris and Payerne (Fig. 9). For both sites, k6 > 0.002 and k6 < 0.002 occurred around
60 and 40 % of the time, respectively. There are sufficient data in
both subsets to provide comparable results. In Paris, the compensation
parameter k is very well correlated with aerosol age: high values
indicate local sources, while values close to 0 indicate long-range
transport (Fig. 9). High k6 values are associated with a
local pattern of air pollution, since the highest PSCF values are located near
the station. The map resembles results previously obtained in Paris for
local sources like traffic and wood burning (Bressi et al., 2014).
Measurements during the MEGAPOLI campaign showed that fresh traffic BC
particles are non-coated, whereas BC particles from long-range transport
exhibit substantial coatings of non-refractory material (Laborde et al.,
2013). For low k6 values (k6 < 0.002)
the sources are distributed over France, part of Great Britain and the
Atlantic Ocean. PSCF results for CF follow the same pattern,
indicating a potential coating material that is brought to Paris by mid- and
long-range transport during which the emissions are aged and aerosols get
coated, and also mixed with more local sources of BC (Bressi et al., 2013;
Petit et al., 2015).
Potential source contribution function analysis in Paris and Payerne
for measurement with low/high values of the filter-loading compensation
parameter k6, for the CF and BC being higher than the 75th percentile.
Theoretical studies show that coating consisting of secondary organic and
inorganic species is mostly transparent in the visible part of the spectrum
and can affect the absorption of light by the particles (Fuller et al.,
1999; Bond et al., 2006). The fact that coating can increase absorption has
been confirmed in ambient studies (Wei et al., 2013b; Liu et al., 2015), and the
coating may also be responsible for the decrease of parameter k;
the possible mechanism would be the increase in SSA and decrease of
the backscatter fraction caused by the coating (Virkkula et al., 2015). In
Payerne, the differences between high and low k6 are not
clearly appreciable, which can be explained by a different meteorology and
orography. Indeed, the station is located between the Jura Mountains and the
Alps, where the air is in general more stagnant compared to Paris and is
channeled from southwest to northeast due to the mountain ranges. This
limits the interpretation of our PSCF results and should be further
investigated on a longer temporal scale (e.g., several months), which would
provide more representative results for this station.
Influence of the mixing state on the filter-loading effect: removing the
coating of ambient aerosols
Our coating hypothesis was put to a definite test during the Ljubljana
campaign; we investigated the effect of the ambient aerosol coating
thickness reduction on the loading effect. The campaign was carried out in
summer, when the atmospheric mixing is stronger, so we expected aged and
internally mixed aerosols to be ubiquitous. We have used two approaches to
reduce the coating thickness: drying and thermo-denuding the inlet stream.
Drying is expected to influence particles coated with hygroscopic material;
thermo-denuding was used to remove certain organic compounds (Huffman et
al., 2009; Cappa and Wilson, 2012; Liu et al., 2015), but it also removes
nitrates and sulfates which decompose above 280 ∘C (CRC, 1920).
The thermo-denuder in our experiment was set to an operation temperature of
300 ∘C. At this temperature, part of the low-volatility oxygenated
organic aerosols (LV-OOAs) is not removed from the sample (Poulain et al.,
2014).
The de-coating efficiency of the thermo-denuder was tested using SEM
analysis of the impactor samples taken during midday, the time when a
well-mixed atmosphere is expected. Coating residues were most pronounced for
above-average size particles (Fig. S4); thus the impactor stage with a 170–260 nm particle size range was selected for the analysis (Fig. 10). We
can observe soot agglomerates (white patches marked by red arrows), organic
residues (dark patches around agglomerates) and brain-shaped dendritic
residues (marked by yellow arrows). Dark patches around soot agglomerates
can be also found on the thermo-denuded sample, indicating low volatility of
these compounds. The main difference between the non-perturbed and
thermo-denuded sample is the removal of the dendrite-shaped residue; a
similarly shaped residue has been reported in different studies (Buseck and
Posfai, 1999; Adachi et al., 2014) as secondary ammonium sulfate particles.
To confirm the presence of secondary inorganic particles and analyze their
chemical composition, we utilized EDX spectroscopy. In the background area (Fig. 10c, spectrum 1), we identified
spectral peaks specific for carbon, oxygen and aluminum, as expected for
impactor aluminum foil covered with a thin layer of grease. In the residue
area (spectrum 2) we observed carbon and oxygen peaks with higher amplitude
than in the background area, and additional sulfur and nitrogen peaks. This
indicates the presence of oxidized organic compounds, ammonium sulfate and
possibly nitrates, further confirming the presence of coatings and in line
with findings of previous studies (Adachi et al., 2014).
The de-coating experiment consisted of four instruments sampling ambient,
dried, thermo-denuded, and thermo-denuded and dried air streams
simultaneously. The diurnal profiles of k6 show qualitative
and quantitative differences between the four treatments (Fig. 11): the
loading parameter k6 for ambient aerosols reaches its highest
value during the night and decreases during the afternoon, when the
atmosphere is completely mixed. Drying increases the average value of
parameter k6 and amplifies the night–day variation. For the
thermo-denuded samples, k6 is almost constant during the day,
with higher average values for the thermo-denuded and dried sample stream.
The average campaign value of parameter k6 shows a
statistically significant difference between all treatments (p=0.05). The k6 for thermo-denuded samples is similar to that of
fresh diesel exhaust (Table 1) as expected for ambient aerosol after removal
of inorganic coating by the thermo-denuder. These results are consistent
with the findings of the coating factor analysis above, where coating causes
reduction of k6.
SEM images of ambient (a, c) and
thermo-denuded (b) impactor samples (size range 170–260 nm) taken
during the Ljubljana campaign. The energy-dispersive X-ray
spectra (d) were obtained at the selected areas on
image (c). Red and yellow arrows mark soot agglomerates and
secondary ammonium sulfate residues, respectively.
Influence of dried and/or thermo-denuded ambient aerosols on the
filter-loading parameter k6 – diurnal plot. The average values of
k6 for the different treatments are significantly different with p=0.05.
Freshly emitted BC particles are hydrophobic (Weingartner et al., 1997;
Zuberi et al., 2005) and show low relative-humidity-induced particle growth
(Laborde et al., 2013). Immediately after sunrise, soot particles in the
atmosphere begin to age by developing a coating of secondary species
including ammonium nitrate, sulfate and organics (Moffet and Prather,
2009). This coating makes particles hydrophilic and sensitive to
relative-humidity-induced growth (Hu et al., 2010; Zhang et al., 2015), which
influences scattering and absorption (Wei et al., 2013a, b).
Drying the inlet stream during the Ljubljana campaign reduced the thickness
of the hydrophilic coating and caused a large increase of k6.
An even greater effect was observed for the thermo-denuded sample, where
most of the inorganic coatings were removed. Interestingly, we observed the
effect of drying also for thermo-denuded particles. This is attributed to
the presence of oxygenated organic aerosols that are less efficiently
removed because of their low volatility (Huffman et al., 2009; Cappa and
Jimenez, 2010; Poulain et al., 2014). For the urban environment, it was
shown that most of the secondary (oxygenated) organic aerosols are
water-soluble (Kondo et al., 2007) and could be prone to relative-humidity-induced
growth. This experiment further supports the hypothesis that coating
is responsible for the reduction of the filter-loading effect in the
Aethalometer.