Instrumentation
Aerosol Chemical Speciation Monitor and data analysis
An ACSM (Aerodyne Research, Inc., Billerica, MA, USA) was deployed to
measure PM1 components in Preila (Fig. 1, Sect. 2.1). A PM10
impactor-type inlet was utilized to remove coarse particles from the sample
stream. The sampling air (1.1 L min-1) passed through a vertical 2.5 m
long stainless steel tube with a 6 mm i.d. and a Nafion dryer (MD-110-12S-4,
PermaPure LLC, Toms River, NJ, USA) before reaching the device. Aerosol
particle diffusion losses in the sampling line were less than 4.0 % for
particles from 40 nm to 1 µm according to Gormley and Kennedy (Baron
and Willike, 2001) and the relative humidity lower than 50 % (by SATO
model SK-L200TH). Thus, the used sampling line and ambient relative humidity
did not affect aerosol mass concentration measured by ACSM. The transported
aerosol flow was split and directed to a scanning mobility particle sizer
(model 19.3.09 IFT/TT (TROPOS, Leipzig, Germany) and to the ACSM. In the
ACSM particles were directed onto a resistively heated surface at ∼600 ∘C
where NR-PM1 components are flash vaporized and the
resulting gases are subsequently ionized by 70 eV electron impact. ACSM was
operated with a time resolution of ∼28 min (for typical aerosol
loadings, i.e., several µg m-3) and a scan rate of 220 ms amu-1
from m/z 10 to 140 (approximately 31.9 s per scan and 1.126 s
pause), 56 scans and data interval 30 min. The data acquisition software
used was DAQ 1.4.4.4. The mass concentrations and mass spectra were
processed using ACSM standard data analysis software (v 1.5.3.0).
The instrument was calibrated using ammonium sulphate and ammonium nitrate.
The determined calibration parameters were response factor (RF)
RFNO3=2.75×10-11 and relative ionization efficiency (RIE)
RIENH4=6.16, RIESO4=0.92. The RIEOrg=1.4,
RIEChl=1.3 were set as default. However, the ACSM collection
efficiency varies depending on the acidity of aerosol particles, aerosol
composition, and particle phase water (Matthew et al., 2008). Many
atmospheric aerosol studies reported reasonable agreement and linear
correlations were obtained with other measurements by using a collection
efficiency of 0.5 (Aiken et al., 2009; Timonen et al., 2010). Middlebrook et
al. (2012) had proposed a collection efficiency calculation method. The
collection efficiency for each measurement and daily mean CE values were
calculated. The CE variation was small during the entire measurement
campaign (March, 2014), so the determined mean CE value was 0.52 with a
standard deviation of 0.08, which is very close to other studies (Aiken et
al., 2009; Timonen et al., 2010). This is not surprising because the sampled
aerosol was dried to RH < 50 %; moreover, the nitrate fraction was
quite low (15 % on average) and a high acidity of aerosols was not
expected at Preila station (EMEP, 2015). Thus, we used the CE = 0.52
in our investigation. The time series of organic aerosol mass spectra were
processed using PMF analysis.
PMF analysis
The ACSM measured data were averaged to 1 h time resolution. A graphical
user interface SoFi (Source Finder) (Canonaco et al., 2013), developed at
Paul Scherrer Institute was used to perform PMF for the source apportionment
of the non-refractory OA mass spectra collected during March 2014. Only
signals at m/z < 120 were used for PMF analysis (Paatero and Tapper,
1994; Paatero, 1997) due to the following reasons: (1) the signals above
m/z>120 account for a minor fraction of total signal, (2) the
m/z's > 120 have larger uncertainties because of poor ion
transmission and the large interferences of naphthalene signals on some
m/z's (e.g., m/z 127, 128, and 129) (Sun et al., 2012). A 2-factor solution
including a Primary Organic Aerosol factor (POA), and a Secondary Organic
Aerosol factor (SOA) was selected for this study. Twenty different PMF runs were
performed using a bootstrapping approach (Davison and Hinkley, 1997). The
bootstrap creates new input data matrices by randomly resampling measured
mass spectra from the original input matrices. Moreover, each PMF bootstrap
run is initiated from a different pseudorandom starting-point of the
algorithm (seed). The bootstrapping approach, together with the seed
approach allows a reasonable exploration of the PMF solution space (Paatero
et al., 2014). Higher order solutions (3 factors) were explored yielding
additional primary profiles, without a significant modification of the
secondary contributions. Moreover the retrieved additional profiles showed
very high time correlation (R2=0.98) with the POA factor,
suggesting a splitting of the same aerosol source. As the additional primary
factors could not be associated to specific primary emissions, those
solutions are not shown. Medium-long range transport of polluted air masses
resulted in a co-variability of the sources at the sampling site, hampering
a further separation of the primary organic aerosols.
Seven-wavelength aethalometer and Scanning Mobility Particle
Sizer
An aethalometer, Model AE31 Spectrum (Manufactured by Aerosol d.o.o.,
Ljubljana, Slovenia) provided continuous measurements of the black carbon
(BC) mass concentrations. The aethalometer was equipped with a PM2.5
impactor. The aethalometer data were recorded with a 5-min time
resolution. The optical transmission of light absorbing carbonaceous aerosol
particles was measured at seven wavelengths (370, 450, 520, 590, 660, 880,
and 950 nm). Measurements at 880 nm wavelength were used to determine BC
mass concentration (Lavanchy et al., 1999). The aethalometer converts light
attenuation measurements to BC mass using a specific attenuation absorption
cross-section (σ) of 16.6 m2 g-1 (at 880 nm)
(Aethalometer Operations manual). The default value for the near-infrared
wavelength of 880 nm was set by the manufacturer. An empirical algorithm for
loading effects compensation was used (Collaud Coen et al., 2010). The
Ångström exponent of the absorption coefficient computed by fitting
an exponential curve was evaluated.
Aerosol size distribution measurements were performed using a Scanning
Mobility Particle Sizer (SMPS) model 19.3.09 IFT/TT (TROPOS, Leipzig,
Germany), with automatic sheath flow, temperature and relative humidity (RH)
control (SMPS setup V2.6 TT 2006) as described in Wiedensohler et al. (2012)
applying a CPC UF-02M (Mordas et al., 2013). The SMPS measured particle size
(8.7 to 840.0 nm) with a time resolution of 5 min having 72 channels.
OC / EC, 14C, δ13C and δ15N analysis
Filter measurements were performed to determine OC, EC and total carbon (TC)
concentrations with a thermo-optical OC / EC analyser (Sunset Laboratory Inc,
USA) equipped with a non-dispersive infrared (NDIR) detector. A 1.5 cm2
filter punch was analyzed according to the EUSAAR2 protocol (Cavalli et al.,
2010). The blank filter was subtracted only from the measured OC and TC
concentrations, as for the EC the corresponding blank was below the
detection limit of the instrument.
14C in EC and TC was measured using the accelerator mass spectrometer
MICADAS, equipped with a gas-capable ion source (Szidat et al., 2014).
14C analysis of TC was determined after combustion of filter punches in
an elemental analyser, directly coupled to the MICADAS (Salazar et al.,
2015). The TC 14C raw data were corrected for a representative field
blank. For 14C analysis of EC, the filters were first water extracted
in order to minimize charring by removing the water-soluble OC (WSOC). Then
the Swiss_4S protocol (Zhang et al., 2012) was used to remove
the water-insoluble OC (WINSOC) and measure the EC 14C, by coupling of
the Sunset instrument to the MICADAS (Agrios et al., 2015). 14C in OC
was determined from the TC 14C and the EC 14C results with an
isotope mass balance calculation. All the data from the 14C analysis
were corrected for the decay of the 14C from 1950 until present. The
reported uncertainty for the non-fossil fraction of EC includes both
charring of OC (overestimation of EC) and EC loss (underestimation of EC)
during the WINSOC removal process (Zhang et al., 2012). Non-fossil fractions
of TC, EC and OC (i.e., TCnf, ECnf and OCnf) were determined
from the individual 14C analyses and 14C reference values. These
reference values represent emissions from purely non-fossil sources and
amount to 1.06 ± 0.03 for TC and OC and 1.10±0.03 for EC based
on the calculation of Mohn et al. (2008). The fossil fractions of TC, EC and
OC (i.e., TCf, ECf and OCf) were determined by subtraction of
the respective non-fossil fractions.
Bulk δ13C and δ15N values were derived by
measuring filter pieces (1.4 cm2) wrapped in tin capsules (8×5 mm,
Elemental Microanalysis) using an elemental analyser accompanying an isotope
ratio mass spectrometer (EA-IRMS, Flash EA1112 – Thermo V Advantage) via a
ConFlo III interface. The autosampler of the EA was continuously flushed
with He (180 mL min-1) to remove all atmospheric gases. Helium flow on
the oxidation column was 80 mL min-1. Flash combustion occurred in the
oxidation column with the presence of O2 (the O2 flow was 180 mL min-1 for 4 s).
Formed gases were taken to the reduction column in
which molecular nitrogen was obtained from any nitrogen oxides followed by a
water trap (magnesium perchlorate). The nitrogen and the carbon dioxide were
separated on a packed gas chromatographic (GC) column (PoraPlot, 3 m × 2 cm,
35 ∘C) and delivered to the isotope ratio mass spectrometer (via the
ConFlo interface) where the measurement of carbon and nitrogen isotope ratio
was made. The amount of nitrogen and carbon in the sample was determined by
a thermal conductivity detector which is a part of the elemental analyser.
These measurements were used in the isotope mass balance calculations (Eq. 1).
The total carbon and total nitrogen fractions of the aerosol particles were
used for the isotopic ratio measurements. Stable carbon and nitrogen
isotopic ratio measurements were expressed relative to the Vienna Pee Dee
Belemnite (VPDB) standard using the formula:
δ13C=Rsample/Rsample-1×1000(‰),
where Rsample and Rstandard are the ratios of 13C to 12C
(or 15N to 14N) in the sample and the standard (referred to as
VPDB), respectively.
Repeated analysis of certified reference material (caffeine IAEA-600) and
oil (NBS 22) gave an average δ13C value: mean ±σ=-27.77±0.08 ‰ (certified value: mean
±σ=-27.771±0.043 ‰VPDB)
and -30.03±0.09 ‰ (certified value: mean ± σ=-30.031±0.043 ‰VPDB),
respectively. These values were used for δ13C measurements in
order to evaluate an analytical precision and calibration of a reference gas
(CO2) to VPDB. Meanwhile, the IAEA-600 standard gave an average δ15N value: mean ±σ=1.0±0.2‰ which was used for calibration of a reference gas
(N2) to air (for δ15N measurements).
Stable carbon and nitrogen isotope ratios were measured in the samples with
the signal intensity reaching 1000 mV or more, due to analytical
restrictions (the isotope values measurements below 1000 mV did not fulfil
linearity requirements of 0.07 ‰/V for the internal
standard).
The mass balance equation was used to calculate the real δ values of
carbon or nitrogen of the aerosol samples (blank correction):
mmeasured×δXmeasured=mreal×δXreal+mblank×δXblank,
where mmeasured was the mass of measured material (carbon or
nitrogen) in the measured sample, δXmeasured was the measured
(aerosol + filter) δ value (carbon or nitrogen), mreal was
the mass of real aerosol material (carbon or nitrogen), δXreal
was the isotope ratio of the real aerosol material (carbon or nitrogen);
mblank and δXblank were the mass and isotope ratio (of
carbon or nitrogen) of the blank filter, respectively.
Radiocarbon-based source apportionment of carbonaceous
aerosols
An estimate of fossil and non-fossil primary and secondary organic carbon
(POCf, POCnf, SOCf, SOCnf) was achieved by coupling
ACSM-PMF results, 14C data, and organic marker measurements using a
chemical mass balance-like approach. The sensitivity of POCf,
POCnf, SOCf, and SOCnf contributions to the assumed
parameters and measurement errors are described in details in this section.
The approach is based on the POCnf estimate, for a subsequent
determination of SOCnf, SOCf, and POCf as follows:
SOCnf=OCnf-POCnf,SOCf=SOC-SOCnf,POCf=OCf-SOCf.
14C measurements and ACSM-PMF results were coupled as follows. Daily
OCnf measurements from radiocarbon analysis as well as average daily
POA from ACSM-PMF results provided two upper boundaries for the daily
POCnf contribution. In this manner we identified a possible daily range
of POCnf contributions. In order to determine more precisely the
POCnf daily contributions within the aforementioned possible daily
ranges, we performed a sensitivity analysis. Briefly, in the sensitivity
analysis we considered a uniform distribution of possible POCnf
contributions within the identified possible daily ranges, meaning that each
POCnf value in the selected ranges was considered as equally probable
(however, as discussed in the next section, in order to explore the
influence of this assumption we also performed the same sensitivity analysis
assuming a non-uniform distribution). Assuming no POCnf contribution
from other sources than biomass burning organic carbon (BBOC), each
POCnf contribution in the acceptable daily ranges could be written
either as [BBOC] = [levoglucosan] /α or as [BBOC] =
[ECnf] /β, where α represents the levoglucosan / BBOC ratio
and β represents the ECnf/ BBOC ratio. In two separated
sensitivity analyses we scanned broad α and β ranges covering
the possible POCnf daily ranges and we retained only POCnf,
[levoglucosan] /α, and [ECnf] /β combinations associated
to selected acceptance criteria described in the following. From the
acceptable solutions we then derived the daily probability distribution
function of POCf, SOCnf, SOCf, POCf, α, and
β.
The assumption that each input POCnf contribution in the selected
possible range is equally probable (hereafter referred to as “uniform
distribution approach”) has advantages and drawbacks: while this assumption
does not consider any a priori information about levoglucosan / POCnf and
ECnf / POCnf, it considers those ratios as equally possible. To
explore the influence of this assumption on our results we performed the
same sensitivity analysis assuming an input levoglucosan / POCnf
distribution derived from 33 profiles for combustion of hard or softwoods in
domestic fireplaces or woodstoves (Fine et al., 2001, 2002, 2004a, b;
Schmidl et al., 2008, the approach is hereafter referred to as “non-uniform
distribution approach”). We eventually derived the probability distribution
functions of the levoglucosan / POCnf and ECnf / POCnf ratios
relative to the acceptable solutions. The two approaches provided similar
results. From the uniform distribution approach, a median
levoglucosan / POCnf ratio of 0.18 (1st quartile = 0.14; 3rd
quartile = 0.23) and a median ECnf / POCnf ratio of 0.32 (1st
quartile = 0.28; 3rd quartile = 0.36) were retrieved, whilst from
the non-uniform distribution approach a median levoglucosan / POCnf ratio
of 0.15 (1st quartile = 0.13; 3rd quartile = 0.18) and a
median ECnf / POCnf ratio of 0.33 (1st quartile = 0.28;
3rd quartile = 0.36) were obtained.
In the following section a technical description of the sensitivity analysis
implementation is reported. For each filter sample i, 10000 random
combinations (r) of input data, [TC]i,r, [EC]i,r,
[ECf]i,r, [OCf]i,r, and
[Levoglucosan]i,r, were generated. In this process, we assume a
normal distribution of the errors around the average [X]i value
(X being one of the input values mentioned above), and a distribution width
equal to the standard deviation σ[X]i.
(a) Combined MODIS images observed from the Aqua satellite
on 10 March 2014, showing numerous fires due to seasonal grass burning and
72 h air mass backward trajectories from the fire regions arriving at Preila
at 100 (red), 200 (blue) and 500 (green) m above ground level (a.g.l.).
(b) NAAPS model results showing surface smoke concentrations for the
strongest stage (10 March 2014) (the color scale (from blue to purple)
corresponds to the seven levels of the contours that indicate the smoke mass
mixing ratio (µg m-3) at the surface). Smoke optical depth at
a wavelength of 0.55 µm. The contouring begins at
1 µg m-3 and doubles in magnitude for each successive
contour. (c) Pressure level in Pa at the surface for 2.5∘
latitude × 2.5∘ longitude global grids (NCEP/NCAR Reanalysis
1, 10 March 2014). (d) PM2.5 concentration
(µg m-3) forecast utilized by the SILAM chemical transport
model during the event of grass fires. (e, f) ACSM organics
concentration (µg m-3) (measured in Preila) weighted air mass
back trajectories of 48 h (for an arrival on 8 (e) and 10
(f) March 2014) with an altitude endpoint of 500 m a.g.l.
For each random combination of input data, the corresponding
[OC]i,r, [ECnf]i,r, and [OCnf]i,r
values were determined as
OCi,r=TCi,r-ECi,r,ECnfi,r=ECi,r-ECfi,r,OCnfi,r=OCi,r-OCfi,r.
10000 random [SOC]s values were generated by randomly selecting a daily
average [SOA]s value from one of the 20 ACSM-PMF runs (s). The
corresponding [SOC]s values were derived as
SOCs=SOAs/(OM/OC)SOAs.
(OM / OC)SOAs and σ(OM / OC)SOAs were calculated according
to Aiken et al. (2009) as function of the fractional contribution of the
m/z 44 (f44) to the SOAs mass spectra. Fröhlich et al. (2015) showed a
systematic difference between f44 measured from ACSM and AMS; therefore an
empirical correction factor was accordingly applied to rescale f44 from ACSM
(f44ACSM) data to the corresponding AMS f44 value (f44AMS). The uncertainty
relative to the f44 correction factor was propagated into σ(OM / OC)SOAs which includes the O / Cs uncertainty as well. Each
[SOC]i,r value was obtained by randomly varying [SOC]s
assuming a normal distribution of errors around the average value
[SOC]s and a distribution width equal to σ(OM / OC)SOAs.
[BBOC]i,r contributions for each sample i were derived as follows:
BBOCi,r=levoglucosani,r/α,BBOCi,r=ECnfi,r/β,
where α represents the levoglucosan / BBOC ratio. This ratio was
systematically varied between 0.01 and 0.31 according to Huang et al. (2014)
and references therein (scan step equals 0.01). β corresponds to the
EC / BBOC ratio. Values of β were systematically varied between 0.1 and
0.4 according to Zhang et al. (2015) and references therein (scan step equal
to 0.01). 10 000 [BBOC]i,r,α and 10 000
[BBOC]i,r,β were determined as in Eqs. (8) and (9). Only
acceptable [BBOC]i,r,α/β
(= [POCnf]i,r,α/β) values were considered for
the sensitivity analysis. The criteria to consider a
[BBOC]i,r,α/β value as acceptable were
(a)BBOCi,r,α/β≤[POC]i,rand(b)BBOCi,r,α/β≤[OCnf]i,r
[POC]i,r was determined as follows:
POCi,r=OCi,r-SOCi,r
Only acceptable [POC]i,r values were considered. The criterion to
consider a [POC]i,r value as acceptable was
(c) [POA]s / [POC]i,r≥1.3 according to Mohr et al. (2009),
Aiken et al. (2009).
[SOCnf]i,r values were then derived as
SOCnfi,r=OCnfi,r-POCnfi,r
Only acceptable [SOCnf]i,r values were considered, where
(d)SOCnfi,r≤SOCi,r
Only solutions where all 4 criteria (a), (b), (c), and (d) held were considered
acceptable and retained.
Finally, [SOCf]i,r and [POCf]i,r were
calculated as:
SOCfi,r=SOCi,r-SOCnfi,rPOCfi,r=OCfi,r-SOCfi,r
Organic markers and satellite products
Determination of organic marker concentrations was performed using a
developed in situ derivatization thermal desorption gas chromatography time
of flight mass spectrometry (IDTD-GC-MS) method (Orasche et al., 2011).
Biomass burning episodes were explored using a variety of remote-sensing
data sets and their derived properties. Satellite data and ground-based
observations of aerosol properties from the MODIS, HYSPLIT and SILAM (Sofiev
et al., 2006) were coupled to analyze the variability of carbonaceous
aerosols in Lithuania (Fig. 2).
The MODIS sensors on-board NASA's Terra and Aqua satellites provide multiple
thermal observations of the Earth on 9–10 March 2014 at a spatial
resolution of 1 km using the latest version of the MODIS Active Fire Product
(MOD14/MYD14) algorithm (MODIS, 2011). To identify the influence of air
masses from different transport pathways on the large biomass burning (BB)
event occurring at Preila, 72 h back trajectories at an arrival height of
100, 200 and 500 m were calculated by the Hybrid Single Particle Lagrangian
Integrated Trajectory (HYSPLIT) Model Version 4.8 (Stein et al., 2015). All
air mass back trajectories were generated using Gridded Meteorological Data
archives of the Air Resource Laboratory (ARL), National Ocean and
Atmospheric Administration (NOAA) (Fig. 2a).
The Navy Aerosol Analysis and Prediction System (NAAPS) model results were
used to define the distribution of BB aerosols from wildfire areas (model
description and results are available from the web pages of the Naval
Research Laboratory, Monterey, CA, USA; http://www.nrlmry.navy.mil/aerosol/) (Fig. 2b). The NAAPS model has been
adapted to combine real-time observations of biomass burning based on the
joint Navy/NASA/NOAA Fire Locating and Modelling of Burning Emissions system
(FLAMBE, http://www.nrlmry.navy.mil/aerosol/) (Reid et al.,
2004). The method has proven helpful in previous studies of long-range and
regional transport of smoke (Honrath et al., 2004). The resolution of
2.5∘ longitude ×2.5∘ latitude National Centers
for Environmental Prediction (NCEP) reanalysis data (Kanamitsu et al., 2002)
during the grass burning episode were analysed to illustrate the sub
synoptic-scale weather feature among the biomass burning events over
Lithuania issued every 6 h for March 2014 (Fig. 2c). SILAM is an air quality
and emergency open code system (http://silam.fmi.fi/) providing
PM2.5 emission maps by Eulerian dynamics and a combination of basic
acid and ozone chemistry with inert particles for fire and anthropogenic
primary PM emission to account for the fire-induced aerosol contribution
(Fig. 2d).