Introduction
Soot particles formed during the combustion of organic fuels
are a major source of particulate matter (PM) from diesel engines,
open burning, and biofuel heating and cooking
. The black carbon (BC) in soot particles is
highly light absorbing, with a radiative forcing potentially
comparable to CO2 .
However, the magnitude of the radiative absorption remains highly
uncertain due to uncertainties related to the role of
atmospheric coatings of non-refractory PM (NR-PM) on
BC-containing particles. Such coatings may enhance light
absorption , cloud interactions
, and/or alter the
deposition rates and thus atmospheric distribution
of BC-containing particles. The
formation and composition of coatings on soot particles is
therefore a topic of major current interest
.
In addition to their role in climate, freshly formed soot
particles have been associated with negative impacts on vascular,
cardiopulmonary, and respiratory health
as well as lung cancer
. Multiple studies have shown
that ozonolysis enhances the apparent toxicity of soot or BC
particles
.
However, the mechanisms behind these effects remain
unclear. While the organic PM component (OM) may play a role
, BC toxicity generally persists after
washing in both polar and non-polar solvents
. This suggests that
the responsible species are either very strongly adsorbed or
chemically contiguous with the BC surface
; we refer to these species as “BC
surface groups”. The composition of such surface groups is
expected to vary widely between samples, fuel type, and
combustion conditions .
BC surface groups are formed during in-flame oxidation
or
reaction with O2 or other oxidants soon after
emission .
Offline analyses have measured oxygenated BC surface groups on BC
from diesel engines, aircraft turbines, wood combustion,
diffusion flames, and even particles formed after the
spark vaporization of graphite in argon
.
These surface groups have been identified as containing
C=O, C-O-C, and/or C-OH groups using
FTIR
and NEXAFS
.
The ubiquity of oxygenated surface groups on fresh BC surfaces
does not mean that fresh soot particles will appear hydrophilic
on a macroscopic scale. For example, fresh soot particles do not
generally activate as cloud-condensation nuclei (CCN) under
atmospherically relevant conditions
e.g.,. Instead, BC surface groups may
present sites for molecular adsorption, with consequences for the
heterogeneous chemistry of soot particles
. This heterogeneous chemistry includes
the health effects discussed above, as well as the reaction of
soot particles with trace gases in the atmosphere
. In particular, the heterogeneous
reaction of photoactivated BC surface species with NO2
and surface-adsorbed H2O may produce enough
HONO to contribute significantly to OH
concentrations in the urban atmosphere .
In addition to BC, aerosols emitted by combustion commonly
contain OM and inorganic ash .
Combustion-emitted organics are a major global source of primary
OM from both natural and anthropogenic sources
and may vary widely in composition
depending on the fuel and combustion process. For biomass
combustion in particular, the composition of OM varies
considerably between fuels and combustion conditions
.
For example, the same fuel burnt in the same stove may emit
organics of different composition and in different amounts,
depending on factors such as air flow and stove temperature
.
The organics emitted during biomass combustion are typically
semivolatile and partition
dynamically between the gas and particle phases of an aerosol.
Atmospheric oxidation of these organics may increase their degree
of oxygenation, altering the ability of aerosol particles to take
up water and/or act as CCN
. Oxidation may also lower the vapour
pressures of primary organic emissions, decrease their vapour
pressure, and lead to their condensation as OM within hours of
emission .
Such OM enhancement has been consistently observed for aged
wood-stove emissions
. For
open-biomass-burning emissions, OM enhancements are much more
variable and often negligible
.
Residential biomass combustion is a significant component of
global anthropogenic combustion emissions and
has been named the largest OM source in Europe
. This OM, together with co-emitted BC,
may dominate other wintertime pollution sources outside of cities
.
Residential biomass combustion is commonly performed using
logwood stoves , which may produce
particles with a toxicity comparable to or worse than that of
diesel-exhaust particles for normal or poor combustion,
respectively . An understanding of the
initial and aged composition of these pollutants is therefore
essential for the understanding and regulation of their emission
with regard to their health and climate effects.
This study evaluates the composition of BC and OM in soot formed
by combustion in a modern wood stove both before and after
simulated atmospheric aging. The stove was operated under optimal
conditions using beech wood as the fuel. Beech is one of the
major forest trees in Europe
, and its wood is
commonly purchased or gathered for combustion as a heating or
cooking fuel .
An online, dual-vaporizer soot-particle mass spectrometer was
used to characterize the fresh and aged soot. As described in
Sect. , the Soot-Particle Aerosol Mass Spectrometer (SP-AMS) vaporizer allowed
entire soot particle to be examined whereas the AMS vaporizer
detected only the NR-PM component (mostly OM). Thus the coatings
and composition of the soot could be studied. Data with the
SP-AMS laser on are referred to as “SP-AMS measurements” below;
with the laser off data are referred to as “AMS measurements”.
The data were analyzed using two approaches. First, selected ions
and representative mass spectra were investigated. Second, factor
analysis was used to obtain 3–4 representative factors to
allow the entire SP-AMS and AMS data sets to be succinctly
described and compared. The results are then discussed in terms
of BC surface oxidation, OM composition, potassium content, and
particulate water content.
Methods
Experimental
The experimental setup used here has been detailed in
. Beech-wood logs (Fagus sylvatica)
were burnt in a modern logwood stove (9kW, Rüegg Mars,
Switzerland) according to official Swiss type-approval protocols
at the officially certified testing facility of the University of
Northwestern Switzerland (Swiss register STS 396, European
register NB 2113). The stove was operated by facility personnel
resulting in burns that were, relative to household usage, extremely reproducible.
Wood logs were added to the stove in batches of ∼ 3 kg (see inset of Fig. ). The first
batch of each measurement day was ignited using a small amount of
tinder (a commercial product containing paraffin wax with the
appearance of wood wool) and kindling (small pieces of wood). The
tinder and kindling were placed on top of the first batch of wood
logs, which therefore burnt top-down. Subsequent batches were
allowed to self-ignite upon the embers of the previous batch,
burning bottom-up. In both cases, the stove operator increased
the airflow into the stove at the beginning of each burn to
facilitate the onset of flaming combustion. The consequential
increase in dilution was corrected for in
but not herein, for reasons discussed below.
Wood-combustion emissions were directed through an indoor chimney
where a heated sampling line (433K) extracted samples
for analysis. After eightfold dilution, samples were either
measured as is, oxidized (see below), or
filtered and oxidized
for a given burn.
The discussion below focusses on the first
case (fresh emissions) and oxidized case (aged
emissions).
Aging was performed using the Micro-Smog Chamber
MSC,. The MSC is described in detail
in and , so only a brief
description is given here. The reactor consists of three
UV-grade quartz tubes totalling only ∼ 225 cm3 in
volume. Losses of particles and low-volatility gases to the
UV-heated MSC walls are expected to be negligible
. Aerosols entering the MSC are exposed to
4W of UVC irradiation (254nm and
185nm emission lines) generated by five low-pressure
mercury lamps (Heraeus, type GPH212T5VH/2), forming
O3 from O2 and subsequently
OH⚫ from background water vapour. In the second tube,
30 W of UVA radiates from a high-pressure halogen lamp
(Panacol-Elosol, type UV-H 255) to drive further OH⚫
chemistry. Some NOx
(=NO + NO2) or SO2 chemistry
is also possible depending on the combustion emissions. The last
MSC tube allows the aerosol to cool slightly while reaction
continues. Some experiments were performed with the intermediate
UVA tube removed, resulting in no observed differences in this
study . For this study, OH exposures were
estimated at roughly 109 moleccm-3h-1
based on follow-up experiments .
This represents an extreme upper limit for atmospheric OH exposure.
Soot-Particle Aerosol Mass Spectrometer
The SP-AMS used in this study is an HR-ToF-AMS (hereafter
shortened to AMS) equipped with a switchable 1064nm
continuous-wave soot-particle vaporization module
. The SP module was switched on and off
periodically, whereas the AMS vaporizer (described below)
remained on continuously. Thus the data reported here represent
either standard AMS measurements or dual-vaporizer
SP-AMS measurements. Here, we use these terms to indicate
the laser state where relevant, also referring to the
instrument itself as the SP-AMS.
The design of the SP-AMS has been described in detail elsewhere
. In brief, the instrument samples
aerosol through an aerodynamic lens, which efficiently focusses
particles with aerodynamic diameters between
approximately 60 and 600 nm into a narrow beam. The lens opens
into a vacuum-pumped chamber, where gas expansion accelerates
particles to their size-dependent terminal velocity. Particles
then transit a sizing chamber to be vaporized within an
ionization chamber. Vaporization is achieved by a switchable
1064nm continuous-wave laser (SP-AMS mode) or
by impaction upon a porous-tungsten thermal vaporizer held at
873K (600 ∘C; AMS mode).
In SP-AMS mode, 1064nm light-absorbing refractory PM
(LR-PM) is
heated to vaporization by the continuous-wave laser.
Due to this heating, any internally mixed material which is refractory below
the LR-PM vaporization temperature may also be vaporized. In this study, the
LR-PM was rBC, which vaporizes at ∼ 4000 K
, so that NR-PM (here mostly
OM; ), ash, or BC surface functionalities
are vaporized when internally mixed with rBC. NR-PM
that is not internally mixed with rBC will pass through the SP-AMS laser and
be vaporized on the AMS thermal vaporizer.
In AMS mode, only NR-PM is vaporized.
In both cases, the vapour is ionized by electron
impact. The resulting ions are mass analyzed by pulsed
extraction into an ion-time-of-flight chamber.
The SP-AMS can provide free-molecular-regime aerodynamic-size
measurements by modulating the sampled particle beam with a
mechanical “chopper” and monitoring the time taken for
signals to be
observed (particle time-of-flight, or PToF,
mode). In this study, PToF data were accumulated over
15–20 s min-1 and were further averaged before
analysis. For the remainder of each minute, mass spectrum
(MS) mode data were acquired by removing the chopper entirely
(“open” position) for 5 s of particle-beam
measurements, then replacing it entirely (“closed” position)
for 5 s of background measurements. The closed signals
were subtracted from the corresponding open signals after data
analysis (described below).
Using a camera, the laser was observed to stabilize after roughly
5 s when switching on (going from AMS to SP-AMS mode) and almost
instantly when switching off (vice versa). So, since each chopper cycle began
with a closed measurement (no rBC), no additional wait time was specified
when switching the laser on or off.
The ion-time-of-flight chamber was
operated in single-reflectron “V” mode with a resolution of
4200 at m/z 91. This mode was sufficient for the unambiguous
distinction and determination of elemental composition for
virtually all ions, with exceptions discussed below. In SP-AMS
mode, the instrument was operated in two separate mass-spectral
configurations, one of which extended to 1000 m/z to monitor
for the fullerene-ion signals that have been previously
observed
.
No fullerenic signals were observed, so these data are not
presented.
Therefore, in the data presented below, one SP-AMS measurement is available
for every two AMS measurements.
SP-AMS and AMS data analysis
Ion signals in the SP-AMS/AMS were identified and integrated
using a modified version of the open-source PIKA software
version 1.10H; as well as custom code
written in Igor Pro (version 6.2 and 6.3; WaveMetrics, OR,
USA). Two major modifications were made to PIKA. First, the
robustness of the peak-shape and peak-width-calibration
determinations were improved as described in the Supplement.
Second, peak-integration uncertainties were estimated as
discussed in and described briefly in
Sect. .
CO+ signals
Special attention was given to the ions
HxO+ and CO+, which are
sensitive to interferences from background H2O and
N2, respectively, and are normally estimated from the
signal at CO2+ .
CO+ signals were directly quantified in PIKA as
described in detail in .
That is, the height of an empirically defined pseudo-Gaussian peak was
fitted simultaneously to each of CO+ and N2+. The
fitted pseudo-Gaussians were constrained in position by the m/z calibration
and in width by a peak-width calibration.
While these signals
normally suffer from poor resolution from N2+,
the current mass spectrometer was able to resolve the two peaks when
CO+ signals were very high above 1000 counts per second;
. As shown in Figs. and and in more detail in , the
CO+ fits resulted in reproducible
CO+:CO+ ratios for each experiment. In the aged
experiment, this ratio was 1.22± 0.01 in the AMS; in the
filtered-and-aged experiment not discussed herein see
, this ratio was 0.86± 0.02 in the AMS. These values
are consistent with literature-observed ratios of 0.9–1.25
and are discussed further in .
The
possibility that other unidentified ions remained
poorly resolved from N2+ is very unlikely, as
also described in that publication.
Gas-phase interferences at CO+ and at CO2+
Gas-phase interferences at CO2+ were corrected for using
filtered-aerosol measurements taken periodically throughout each experiment.
This method has been detailed by . The CO2+
correction was smaller than that needed for atmospheric measurements due to
the factor-of-200 dilution by synthetic air (80 % N2,
20 % O2, 99.999 % purity), and the correction factor was
dynamically scaled by continuous gas-phase carbon-dioxide measurements. The
resulting correction factor was 7–15 % of the measured
CO2+ signals (interquartile range, both experiments) and
depended on the stage of combustion.
Gas-phase interferences at CO+ were not corrected for. The high
modified combustion efficiency (MCE) during these experiments meant that CO(g) emissions were much
smaller than CO2(g) emissions, especially during the periods when
CO+ could be quantified. In particular, for the periods when
CO+ signals were high enough to be quantified (due to the
peak-fitting limitations described above) the CO2+ correction
factor was 0.6–2 % (interquartile range, both experiments). These
periods corresponded to MCE values greater than 0.97, so the CO+ correction
factor would have been less than < 3 % of the CO2+
correction factor (note that CO+ signals were generally higher
than CO2+ when quantifiable). This correction would be
negligible relative to the uncertainty in the CO+ due to the fit
described above.
Time series for the fresh beech-wood-combustion experiment. Selected
SP-AMS (solid lines with circles) and AMS (dashed lines) data are
shown in nitrate-equivalent mass (proportional to ion counts). Error
bars are smaller than the data symbols. The uppermost panel shows
gas-phase CO2(g) concentrations in the chimney (proportional
to combustion rate), shaded by the modified combustion efficiency (MCE). The second-from-top
panel shows the CO+:CO2+
ratio (green: AMS; red: SP-AMS), which was constant for the AMS (dashed line shows the result of a linear regression of CO+ against CO2+).
Shaded regions exemplify “starting” (green shading) and “flaming”
(red shading) phases, defined using the CO+:CO2+ ratio.
Time series for the aged beech-wood-combustion experiment. All
traces
are analogous to Fig. but are scaled differently.
Note especially the change of scale for C2H4O2+.
In this experiment, the second batch of wood failed to start (grey
shading), causing OM concentrations tenfold greater than normal.
This
especially increased the signal at C2H4O2+ (a
tracer ion for holocellulose pyrolysis), which has been plotted off-scale
during that period to allow features at other times to be visible.
The MCE during this period was also off-scale, with mean
0.82.
Water signals
The ions HxO+ (=O+,
HO+, and H2O+) were not quantifiable
in MS mode due to interferences from background water. That
these three ions originated from water, and not from the
fragmentation of other molecules, was confirmed with the ratios
of particulate (PToF) signals at
O+:HO+:H2O+ ,
which was 4:25:100 in both AMS and SP-AMS. The proportion of
this water signal originating from the thermal decomposition of
OM was estimated in proportion to CO2+ by
assuming a 1:1 H2O+:CO2+ ratio,
based on laboratory measurements of fulvic acid
and smog-chamber secondary OM
. However, the
molecular composition of fresh wood-smoke aerosols is
considerably different from such OM, containing an abundance of
aldehydic, phenolic, and alcoholic functional groups from
lignin, cellulose, and hemicellulose
.
This chemical composition and the high
CO+:CO2+ ratio of the present sample
(discussed below) suggest that this 1:1 ratio is likely to
be biased low.
Potassium signals
SP-AMS and AMS signals are generally proportional to vaporized
particulate mass due to the two-step vaporization/ionization
process in the instrument. However, species with especially
lower ionization potentials such as potassium may undergo
one-step thermal ionization during the vaporization process
e.g.,. Due to this unique
ionization mechanism, K+ ions are normally observed
with a unique distribution of kinetic energies, i.e., a
uniquely broad peak shape in the mass spectrum. This means
that the peak-integration routines programmed into PIKA are not
applicable to K+. Potassium-ion signals were
therefore estimated as the total, baseline-subtracted signal at
m/z 39±0.2 minus the PIKA-fitted
(Sect. ) signal for the two ions
which fell within that range, C3H3+ and
13CC2H2+. The closed background
(Sect. ) obtained in this way was
subtracted from the analogous open signal. The unique
behaviour of potassium ions in this study provides validation
that all other ions, including those from BC, underwent
two-step vaporization/ionization in the SP-AMS.
Particle collection efficiencies (CEs)
For SP-AMS and AMS signals to be quantified, the
efficiency by which particles of different size, composition, and morphology
are vaporized must be known. This efficiency relies on (i) the transmission
of particles from sample to vaporizer, and (ii) the successful vaporization
of transmitted particles.
The geometry of the AMS and its thermal vaporizer has been designed such that
point (i) above is negligible for particles of aerodynamic diameter between
60 and 600 nm . For such particles, the CE of the AMS is governed by the
probability of particle bounce at the thermal-vaporizer surface
e.g.,.
As discussed extensively in a separate publication , we
evaluated the available literature on wood-combustion OM to arrive at a CE of
1.0 for both fresh and aged emissions.
Point (ii) above is negligible for BC-containing particles as the laser power
is normally operated in a regime of excess power .
However, point (i) above is not negligible , because
the SP-AMS laser vaporizer is physically smaller than the AMS thermal
vaporizer (with an estimated 40 % less effective area;
).
Thus, point (i) may lead to a negative SP-AMS bias if particles are too
small to be adequately focussed into the SP-AMS laser beam. In this context,
small refers to the aerodynamic size of the particles, which is a function of
particle volume, morphology, and density. demonstrated
that particle focussing is an issue for the SP-AMS for nebulized carbon-black
particles of mobility diameter 200 nm. Although the aerodynamic diameter of
those particles was not reported, this is a relatively large size when
considering freshly formed soot , and so the effects
observed by also apply to this study. Moreover,
nebulized carbon black is likely to have a significantly smaller dynamic
shape factor than the freshly formed fractal-like soot
emitted by flames and combustion engines and likely emitted during the
present experiments.
We thus expect that particle focussing lends a significant bias to the
SP-AMS signals in this study, corresponding to a CE less than unity for the
SP-AMS. However, the present particles were generally small relative to the
lower limit of the SP-AMS aerodynamic lens as discussed further for
this data set in, so that coating of these particles by
secondary OM would likely change their ability to be focussed into the SP-AMS
laser. In addition, particle morphology is expected to change significantly
within burns . A single collection efficiency based
on an external rBC mass reference, as has been applied in previous SP-AMS
studies
,
would therefore not be possible. A time-resolved collection efficiency would
require the assumption of internal mixing, and a reference instrument was in
any case not available. Below, we focus instead on the changing SP-AMS
signals and their relationship to AMS signals.
Relative ionization efficiencies (RIEs)
RIEs were applied to the
SP-AMS and AMS data to represent variations in the instrumental
sensitivity to different species . The RIE
is the efficiency with which a given gas molecule is ionized
relative to the AMS calibration standard, NO3-
in ammonium nitrate.
The RIE(Cx+) was set to 0.2 based on . This is
likely an underestimate as it was derived without accounting for the SP-AMS
CE and because it did not account for refractory
COx+ species (rCOx+; ). However, in
the absence of additional information on the SP-AMS CE (see above), we have
used this RIE to provide an estimate of the relative intensity of Cx+
signals in the mass spectra shown below.
The RIE(OM) was set to 1.4 based on . This
value was extrapolated to RIE(rCOx+) because
both OM and rBC yield molecular CO and
CO2 upon thermal decomposition
. Although the SP-AMS sensitivity to BC
surface groups has not been established experimentally, it
appears to be higher than that of Cx+ based on
the fact that the raw signals of rCOx+ were much
higher than those of Cx+ in this study. These
high signals indicate that RIE(rCOx+) > RIE(Cx+), since rBC is, by definition,
composed primarily of graphitic carbon
and since rBC
particles are completely vaporized in the SP-AMS
. In addition, the assumed
RIE(rCOx+) of 1.4 implies an elemental
composition of about 10 % oxygen in the BC,
consistent with elemental analyses from the literature
. However, this
value should be carefully validated before being used for mass
quantification.
The SP-AMS and AMS RIE(K+) are likely different
given that potassium was thermally ionized and that the
temperatures experienced by potassium in the SP-AMS and AMS
vaporizers are different. For the AMS, an RIE(K+)
of 2.9 has been reported by ,
while no data are available for the SP-AMS. To reflect a lack
of information, RIE(K+) was simply set to unity.
All other ions discussed below are considered organic and
therefore assigned an RIE of 1.4 in PIKA. SP-AMS signals were
not calibrated to mass in this study because the fraction of
rBC focussed into the laser is dependent on soot-particle shape
and coating
.
SP-AMS and AMS uncertainty
analysis
Application of new positive matrix factorization (PMF) error model
Independent of peak-overlap uncertainties, peak integration in PIKA leads to
a constant relative imprecision (fractional imprecision) which is expected to
be approximately constant for a given peak in a given data set
. Peak-integration imprecisions arise from (i)
peak-width-calibration imprecision and (ii) peak-fitting imprecisions that
arise due to m/z-calibration biases (i.e., the finite mass accuracy
of the instrument) and m/z-calibration imprecisions
.
Although peak-fitting imprecisions may become much larger for overlapping
peaks relative to isolated peaks , peak
overlap was not addressed quantitatively here (the final details of the
method described by , were not yet available). We have applied a simplified
approach by assuming an approximately constant value of ∼ 4 % for
(ii) above, which was estimated from the Monte-Carlo-estimated fitting
imprecision for isolated peaks in the present data set .
For (i) above, we used the measured imprecision of in our peak-width
calibration of 2.5 % . The quadratic sum of these
fractional uncertainties,
4.7 %, gave the peak-integration uncertainty, as described in .
This percentage peak-integration uncertainties were combined in
quadrature with a Poisson uncertainty term to yield an
uncertainty model which is dominated by ion-counting
uncertainty for low signals and by peak-integration
uncertainties for high signals
for this data set, those greater than ∼ 1000 counts per second
for a given peak; .
Additional uncertainties
In addition to the error model discussed above, additional
uncertainties were assigned to CO+,
CHO+, CHO2+, and
C3H3+, for various physical reasons, as shown
in Table . These additional uncertainties
were only calculated for ions which were of significant signal
in the mass spectrum. Table also shows the
default uncertainties for above-detection-limit
and below-detection-limit
signals.
The increased CO+ detection limit accounts for the
failure of the PIKA peak-integration algorithm to integrate
peaks below a ∼1kHz “detection limit” caused
by the neighbouring N2+ signal. This behaviour
is expected for peaks with a very small intensity relative to
an overlapping peak . The validity of
CO+ above this DL was confirmed by their linear
variation with CO2+ in the AMS, with slopes
fully consistent with the literature as detailed elsewhere
. To address this detection limit,
backgrounds were subtracted prior to fitting (“Diff” mode in
PIKA) for CO+, reversing the normal procedure
(i.e., “OminusC” mode in PIKA). An uncertainty of
3.5×DL was assigned to below-DL values according
to common practice in the PMF community
, although the
results were not sensitive to the chosen value. Below-DL values
were not replaced since this was not a detection limit in the
conventional sense. The additional uncertainty in
CO+ signals due to their overlapping with
N2+ was
not modelled.
AMS and SP-AMS uncertainties σ used in this study and their detection
limits (DLs). IX+ is the rate of counts of
ion X+; I^X+ is IX+
normalized to its maximum value; RX+ is the
abundance of an ion X+ relative to its most abundant isotopologue;
and k is a constant determined as 1µgm-3 as discussed
in the text.
Species
Signal
Uncertainty
Rationale
Default
I
σ=σcounting2+σA2 if >DL
Combined poisson and peak-integration
σ=σcounting(1) if <DL of 1 ion.
uncertainties .
CO+
I
σCO+ if >DL,
Higher-than-normal DL due to poor
3.5×DL if <DL of 1kHz.
resolution from N2+.
CHO+
I
σCHO+2+R15NN+2
Fit to the overlapping 15NN+ peak
constrained by N2+.
CHO2+
I
σCHO2+2+R13CO2+2
Fit to the overlapping 13CO2+ peak
constrained by CO2+.
C3H3+
IC3H3+-
σC3H3+2+3k×I^K+2
Interference by tailing of extremely
k×I^(K+)
high SP-AMS K+ signals from rBC.
The uncertainty in CHO+ (m/z 29) was increased to
reflect the fact that the signal intensity of
15NN+ (m/z 29) was constrained by the signal
intensity at N2+ (m/z 28) in PIKA using a
relative abundance of 15N predicted according to
. In contrast to the
majority of isotopically constrained fits, the actual
uncertainty of the signal assigned to 15NN+ was
therefore very large relative to the initially estimated
CHO+ uncertainty. To account for this,
σ15NN+ was added in quadrature
to the original uncertainty in CHO+ (Table
). This procedure provides only a rough
estimate of the true uncertainty, as it assumes that the
uncertainty σCHO+ is independent
of the constrained-fit procedure, which is generally not true
.
The uncertainty in CHO2+ suffered from a
constrained 13CO2+ fit, similarly to the case
of CHO+ and 15NN+, and was treated
similarly.
The final ion in Table ,
C3H3+, was overestimated in the presence of
high SP-AMS K+ signals. This overestimation was
made apparent by examining background mass
spectra (chopper blocking the particle beam), in which
K+ signals were negligible
but C3H3+ signals remained.
These background
spectra were used to estimate a C3H3+
overestimation of ∼2µgm-3 when
K+ was highest. This simple estimate was
represented numerically as 1±3µgm-3,
scaled by the time series of SP-AMS K+, and
subtracted from the SP-AMS C3H3+ signals.
The corresponding uncertainty was estimated according to this
subtraction, which is considered a reasonable estimate since
the influence of the K+ background on the peak
shape appeared to be negligible, by inspection. No
41K+ interference was observed so no changes
were made at m/z 41. Similarly, AMS potassium signals (i.e.,
when the SP laser was off) were too low for interference from
K+ to be an issue.
Positive matrix factorization
PMF assumes that a matrix of data can be explained by a linear
combination of factors with characteristic profiles and varying
temporal contributions . The
PMF model has been widely and successfully applied to AMS data
. Applying PMF to
AMS or SP-AMS data entails the assumption that the overall mass
spectrum can be described by a small number of characteristic
mass spectra ,
which is evaluated during PMF analysis by inspection of the
model residuals
.
PMF analysis was conducted using the PMF Evaluation Tool
. Before analysis, signals
were integrated and exported using the peak-integration
approach and new uncertainty model described in the Supplement
and in Sect. . Ash signals, defined as
ions containing Cl, Si, K, or other metals, were excluded from
the model. The number of PMF factors used, P, was chosen
based on the degree to which the model improved with increased
P, as evaluated by the structure of model residuals in
temporal and m/z space. In general, increasing P beyond the
values presented herein only served to better explain spikes in
OM concentration at the start of each burn. Therefore, the
conclusions discussed below are insensitive to the chosen P.
The large number of zeroes in this data set meant that
ambiguity due to possible linear transformations of the PMF
solution (“rotational ambiguity”) was negligible in this data
set . Similarly, there was no
evidence for local minima (see Supplement).
The large range of signals observed in the present study led
PMF to report “unique factors” containing
a single ion such as CO+ when using the old
uncertainty model. This was corrected by the
uncertainty model
(Sect. ), which provides a more even
weighting for high and low signals. We note that the linear
uncertainty term in the model is similar
to the ad hoc “C3” parameter of PMF labelled “model
error” in PET. The difference is that
the value of the linear uncertainty term was estimated from the
data, and that it was added in quadrature rather than linearly
to the pre-existing Poisson uncertainties, since its physical
origin is independent of counting uncertainties
.
The new uncertainty model also generally reduced
signal-to-noise ratios (SNRs) such that previous outliers (in
weighted-residual space) from major ions such as
CO+ and C3+ became more comparable
to other species. show that such
outliers may arise purely from the omission of a linear
uncertainty term when such an uncertainty exists. The outliers
therefore do not indicate measurement errors, peak-integration
errors, or transient signals. The new uncertainty model also
increased the number of “weak” variables, defined as having
SNR below 2 , which were
downweighted by a factor of 2. No variables were “bad” in
the sense of having SNR < 0.2
. Additional details on the PMF
analysis, including residual plots, are available in the
Supplement.
Results
Temporal evolution of burns
Figures and illustrate
the temporal evolution of the fresh and aged emissions using
selected SP-AMS and AMS marker ions. Also shown in the figure
is the pre-dilution CO2(g) concentration, which is
proportional to the combustion rate. The CO2(g)
also indicates the airflow into the chimney, which implicitly
dilutes the emissions to a varying degree. Following the
addition of a batch of wood (black arrows along the abscissa),
CO2(g) increases rapidly (“starting phase”)
before reaching a relatively stable level (“flaming phase”)
and finally dying off (“smouldering phase” with negligible
emissions).
During the starting phase, the wood was only partially aflame,
typically in only one region. Both fresh
(Fig. ) and aged (Fig. )
OM emissions were highest during this phase. The degree of
oxidation and emission factors of this OM has been reported in
detail elsewhere .
The starting-phase OM emissions are illustrated by the
C2H4O2+ signal in
Fig. . This ion is correlated with
cellulose/hemicellulose pyrolysis products such as levoglucosan
and other anhydrosugars and its signal spiked
when each batch of wood was added (black arrows on the
abscissa) and died away thereafter. Similar trends were seen
for the oxidized emissions (Fig. ), although
the C2H4O2+ signal was much lower in that
case. The SP-AMS and C2H4O2+ signals were
not always well correlated, as discussed in Sect. .
The starting phase ended once flames had completely engulfed
the wood. In this “flaming phase” most volatilized organics
were destroyed in the flames such that emissions were comprised
of little OM but significant amounts of refractory black carbon. This is illustrated in Figs. and by the C3+ ion, which in the
SP-AMS represents rBC .
Black carbon was not the only SP-AMS species observed during
the flaming phase. Signals from two other species,
K+ and COx+ (=CO++CO2+), remained extremely high
during this phase (Figs. and ). These signals dropped to negligible
levels in the AMS (dashed lines in Fig. ),
confirming that they were generated from LR-PM particles.
Although high K+ signals were only observed in the SP-AMS,
relatively high COx+ signals were observed in both AMS and SP-AMS.
These signals therefore appear to have originated from both OM and rBC.
The CO+:CO2+ ratio for these signals provided
a useful metric of comparison between SP-AMS and AMS across
burn periods. In the AMS, this ratio was 3.92±0.01 for
fresh emissions but 1.22±0.01 for aged emissions
. This latter ratio is in good agreement
with observations of atmospherically oxidized OM
0.9–1.25,, indicating a relative increase
of CO2+ due to the thermal degradation of
oxidation-formed carboxylic acids or peroxides on the AMS
vaporizer . In contrast, the phenols, alcohols,
and aldehydes that are abundant in fresh wood
are more
likely to yield CO+ than CO2+ when
vaporized, giving the high CO+:CO2+ ratio of
3.92.
The SP-AMS CO+:CO2+ ratio was not different
from the AMS ratio during the starting phases
(Figs. and ) nor during
the filtered and oxidized experiment (not shown). This suggests
that OM fragmentation in the SP-AMS and AMS was comparable, at
least in terms of COx+ fragments.
In contrast to the OM-dominated starting phase, the SP-AMS
CO+:CO2+ ratio increased significantly during
the flaming phase, increasing from ∼4 to ∼6 for the
fresh case and from ∼1 to ∼2 for the oxidized case.
Fresh-emissions SP-AMS and AMS mass spectra for the starting-phase
(highest OM emissions) and flaming-phase (lowest OM emissions, high
BC emissions, Fig. ) beech-combustion experiments.
All signals have been scaled by representative RIEs.
The estimated signals at O+, OH+, and H2O+
have been estimated from CO2+ (Sect. ) and therefore plotted with thinner bars.
The insets show photographs of a typical burn.
The mean and standard deviation of the MCE over the averaged period of time
is included. Note the changes of
scale in panels (c) and (d).
In panel (d), no CO+ signal is shown as that species was
below its detection limit (2.4 units).
SP-AMS and AMS mass spectra
Fresh emissions
Figure shows SP-AMS and AMS carbonaceous-ion
mass spectra for the starting- and flaming-phase periods of the
fresh-emissions experiment that are highlighted in
Fig. . Ions are coloured according to
their oxygen content, and signals at integer m/z are stacked
for clarity. Ash species such as K+ (the most
intense ion), Cl+, HCl+,
Zn+, SO+, and SiO2+ (all
low intensity) were excluded since these species are not
susceptible to oxidation in the MSC.
Aged-emissions SP-AMS and AMS mass spectra for the starting-phase
(highest OM emissions) and flaming-phase (lowest OM emissions, high
BC emissions) beech-combustion experiments, in analogy to Fig. .
Note the change of scale in panel (d).
The estimated signals at O+, OH+, and H2O+
have been estimated from CO2+ (Sect. ) and therefore plotted with thinner
bars. In panel (d), no CO+ signal is shown as that species
was below its detection limit (2.4 units).
PMF factors for fresh-emissions data. The left column shows SP-AMS
factor time series (black) together with AMS time series (blue) and
selected SP-AMS tracer ions (symbols). The right column and insets shows
the corresponding SP-AMS factor mass spectra using the same colour
scheme as Fig. : black ions are carbon clusters,
green ions are hydrocarbon fragments, and pink/purple are more- or
less-oxygenated carbon-containing ions. The right-column blue symbols
show the AMS mass spectra for data >1 % (or >0.1 %)
of the spectrum in the main panel (or inset). Signals for (c) POM–Start
are drawn off-scale to show key structural features.
In the mass spectra of (b), no CO+ signal is shown as that
species was below its detection limit (see text).
The SP-AMS and AMS mass spectra in Fig. a and
b are similar during the starting phase when
OM was highest. This explains why the SP-AMS
CO+:CO2+ ratios in
Figs. and are similar
to the AMS ratios.
In contrast to the starting phase, Fig. c and d show that the flaming-phase PM consisted
mostly of refractory PM. The AMS signals during this phase
were negligible. (Note that the ordinate maximum of
Fig. d is 2 orders of magnitude smaller than
Fig. c.) Thus the SP-AMS mass spectrum shows
that the rBC particles yielded Cx+ and
COx+ (and K+) from refractory
species.
The other major signals in the flaming-phase SP-AMS mass
spectrum are the carbon-cluster ions Cx+ with
1≤×≤3. Another study by our group
found that the ratios between these ions
was indicative of the underlying structure of the rBC, with rBC
particles that generated fullerenic ions having
C1+:C3+ ratios close to and other
samples having C1+:C3+ ratios below
0.8. The C1+:C3+ ratio observed in
Fig. c, where AMS signals are negligible, is
close to unity. This is an exception to the trends observed by
and others
. In contrast,
the more robust but lower-sensitivity ratio
C4+:C3+ was well within the
0.01–0.07 range reported by , for
both fresh and aged emissions. The
C4+:C3+ ratio and not the
C1+:C3+ should therefore be used in
source apportionment studies.
Aged emissions
Figure shows mass spectra from the third
burn of the aged-emissions experiment. The overall trends
between SP-AMS and AMS are similar to those discussed in the
previous subsection. However, both SP-AMS and AMS
starting-phase mass spectra (Fig. a and b, respectively) show relatively lower
signals from hydrocarbon fragments (green bars) and
relatively higher signals from COx+ ions due to
oxidative functionalization of the OM.
Similarly to the fresh-emissions case, flaming-phase
COx+ signals were of comparable magnitude to
Cx+ signals in the SP-AMS
(Fig. c). However, CO2+
increased relative to both C3+ and
CO+, indicating oxidation of the species which
generated COx+. We note that surface oxidation
is not expected to significantly influence SP-AMS
Cx+ signals, since these signals represent the
bulk composition of the solid rBC. The chemical species
generating Cx+ and COx+ signals
remained largely refractory after oxidation as indicated by
the AMS data, which are plotted in Fig. d with
an order-of-magnitude-smaller ordinate maximum.
Positive matrix factorization
The discussion in Sect. highlighted
selected ions and mass spectra from selected time periods in
the burn cycle. To generalize this discussion to the entire
mass spectrum and the entire burn cycle, PMF was performed on
the fresh and aged data. The utility of PMF was to reduce 399
measured ions in over 100 mass spectra to 3–4 “factors”
with mass spectra and time series describing > 92 %
of the variance in the data. (Over 97 % of the
variance would have been explained had the failed-start burn
been excluded.) These PMF factors are discussed below.
Fresh emissions
With the exception of BC factors, each fresh SP-AMS PMF factor
had an analogous factor in the AMS data. This is shown in
Fig. a–c. Each panel in the figure shows
the time series for each PMF factor in black (SP-AMS) or blue
(AMS). The smoothness of each factor time series should not be
compared because only half as many SP-AMS data were available
as for the AMS (Sect. ). The time series
in Fig. also include arbitrarily scaled raw
signals of selected ions.
Figure a shows the first PMF factor,
“Fresh–BC”, with a mass spectrum similar to the SP-AMS
flaming-phase mass spectrum (Fig. ). The mass
spectrum is plotted following the scheme introduced in
Fig. . The time series of this factor closely
followed that of C3+ except during spikes in
concentration. These exceptions may reflect a change in
instrument response, for example due to detector
saturation. Alternatively, they may reflect a change in PM
composition, for example due to the wood logs shifting position
during combustion (cf. Fig. a, inset) and
causing a transient change in rBC composition.
The observation of both COx+ and
Cx+ signals in the Fresh–BC factor suggests that the
two species originated from the same physical source
(see also Fig. ).
To test
whether the COx+ might have been attributed to a
separate factor from Cx+ with more PMF factors,
we increased the number of PMF factors as high as 10. The
result was two separate Cx+-containing factors,
both of which remained associated with COx+ in a
similar manner to Fresh–BC. This suggests that these two
species were physically related, possibly originating from a
single process in the combustion.
PMF factors for aged-emissions data. Panels are analogous to
Fig. .
Signals for (c) OOM–Start are drawn off-scale to show key structural
features. Time series (d) is plotted on a log-scale due to its low
intensity, with an axis minimum of the OM limit of detection (3σ).
In the mass spectra of (d), no CO+ signal is shown as that
species was below its detection limit (see text).
The next factor in the figure is “POM–Flame”
(Fig. b). This primary OM factor is
named for its sustained signal during the flaming phase
discussed above, though it is important to note that the other
OM factor (“POM–Start”) was simultaneously present at
comparable intensity.
POM–Flame was better correlated with the alkyl ion
C4H7+ (Fig. b) than with
the pyrolysis tracer C2H4O4+. The SP-AMS
mass spectrum for this factor is shown in
Fig. b. An analogous AMS factor, with
similar temporal trend and mass spectrum, was also identified
and is included in the figure. The AMS mass spectrum is
included as the summed signal at integer m/z since one ion
typically dominated this sum, as shown by the coloured SP-AMS
signals. For simplicity, AMS data are only shown for signals
contributing > 1 % (or > 0.1 %) of the total in the
main panel (or inset) mass spectrum.
The AMS and SP-AMS POM–Flame mass spectra were virtually
identical. Both showed considerable amounts of large
hydrocarbon fragments (Fig. b,
inset). Neither POM–Flame spectra included CO+
because the low concentrations of POM–Flame meant that this
ion was below its detection limit.
The final factor in Fig. , POM–Start,
contributed more to overall OM signals than POM–Flame and made
most of its contribution during the starting phase of each burn
(Fig. c).
POM–Start was absent for the first burn, along with the
anhydrosugar tracer C2H4O4+. This is most
likely because in the first burn combustion was initiated by
tinder placed atop the wood, whereas in subsequent burns
combustion was initiated from below by the hot embers of the
previous burn (Sect. ). Using
C2H4O4+ as a wood-combustion tracer in an
atmospheric context may therefore underestimate wood-combustion
emissions in some cases.
Aged emissions
The aged-emissions PMF factors could be viewed as analogous to
the fresh-emissions results. The lesser signals at higher m/z
were reduced in each analogous aged mass spectrum, as expected
since highly oxidized species fragment during
vaporization/ionization to a much greater extent than reduced
ones .
An Aged–BC factor was well correlated with C3+
(Fig. a). The Aged–BC mass spectrum was
dominated by the same Cx+ and
COx+ ions as the Fresh–BC but in different
relative intensities. The ratio C3+:CO2+
decreased from 3.5 to 2.4 after aging, while the ratio
CO+:CO2+ decreased from 5.4 to 3.5. The
source of COx+ therefore both increased and
became more oxidized following aging.
The absence of an analogous AMS factor suggested that the source of this
BC-associated COx+ remained refractory after aging.
An OOM–Flame (oxidized OM, flaming phase) factor analogous to
POM–Flame was also observed. Figure b shows that
this factor was well correlated with
C2H4O2+. Here, the
C2H4O2+ tracer is much lower in intensity
than for the fresh-emissions case
(cf. Figs. and ) due to
its susceptibility to oxidative aging
. However,
this reduction in C2H4O2+ intensity does
not invalidate its use as a pyrolysis tracer. The more-common
oxidized-OM tracer, CO2+
,
was inappropriate in this case due to its confounding
refractory source in Aged–BC. The resulting ambiguity in the
interpretation of CO2+ in BC-rich aerosols is
likely to be significant in atmospheric SP-AMS studies.
The OOM–Start factor corresponded to higher OM loadings than,
and followed similar trends to, POM–Start. This factor was
present at extremely high loadings during the failed-start case
(off-scale in Fig. c; re-plotted in
Fig. S2 in the Supplement). Its signals at higher m/z
(Fig. c, inset) were relatively higher than for
OOM–Flame, possibly because oxidation in the MSC was less
extensive when organic vapour concentrations were higher.
The fourth aged-emissions factor was the tinder factor,
corresponding to the paraffin-soaked wood shavings used to
start the fire (Sect. ) and to re-ignite
the failed-start burn (Fig. d, ∼01:10).
(Prior to this re-ignition, the air flow into the stove was
changed and the door was afterwards opened briefly.) The tinder
mass spectrum was dominated by hydrocarbon fragments
CxHy+. It also contained some
CO2+ (and probably below-detection-limit
CO+) due to oxidation in the MSC.
Note that tinder was also used at the beginning of the
fresh-emissions experiment, and an analogous PMF factor did initially
result for that experiment.
However, only the first two measurements of the fresh-emissions experiment
were strongly impacted by this contaminant.
To simplify the PMF model needed to describe the data, the first two
measurements of that experiment were downweighted threefold (see Supplement) for the fresh emissions.
Discussion
The measurements presented above provide evidence for the
presence and oxidative enhancement of refractory rBC surface
groups and for a pyrolytic origin of the OM in beech-wood
soot. These and other compositional features of the soot are
discussed below. A detailed discussion of the oxidized OM can
be found elsewhere .
rCOx+ from BC surface groups
The combination of non-refractory AMS and refractory SP-AMS
measurements of CO+ and CO2+
(Figs. c–d and c–d) clearly
showed that the majority of these signals originated from
refractory species (rCOx+). These
refractory CO2+ signals became more intense after
oxidative aging, indicating that they originated from
incompletely oxidized species. The change in the ratio of
the rCOx+ signals after aging
indicated a change in the chemical species producing these signals.
These two observations reduce the likelihood of confounding
sources of rCOx+ generating the majority of the
signal. Carbonates such as CaCO3
cannot be oxidized as observed. Refractory
organics such as large, oxidized polyaromatic hydrocarbons may
generate COx+ but would also produce other,
hydrogen-containing,
ions in the mass spectrum. Additionally, if the refractory organics had not
contained hydrogen, the observed increase in
COx+ after aging would be impossible.
This increase in COx+ is consistent with soot formation mechanisms:
soot only forms in oxygen-deprived environments
and initially condenses as PAH-like
nanoparticles before graphitizing via loss of hydrogen
. Oxygen
is normally gained during later oxidation, before exiting
the flame, for example by flame-produced radicals like
OH⚫ .
Since well over 90 % of soot produced in a flame is destroyed
the abovementioned in-flame oxidation, all soot is expected to
contain oxygenated surface groups
.
The BC surface functional groups are observed as
COx+ and not, say, C3CO+ because
BC surface groups thermally decompose at much lower
temperatures than the vaporization temperature of BC
.
We cannot completely rule out the role of adsorbed CO(g)
or CO2(g) in forming rCOx+. This is
an inherent weakness of the destructive SP-AMS technique
relative to FTIR and NEXAFS. However, the significant change in
COx+ upon oxidation indicates that such a role
was not dominant. Additionally, if the thermal stability of
covalently bonded or adsorbed species at the BC is similar,
reaction mechanisms may
be also be similar in both cases.
Atmospheric implications of BC surface groups
Two atmospherically relevant observations can be made regarding
BC surface groups. First, the PMF results show that the
relative amount of these surface groups can be regarded as
constant throughout the burn. Regardless of burn stage, stove
temperature (cf. first burn in Fig. ), or the
concentration of emitted BC, a single PMF factor adequately
represented these signals. It may therefore be possible to
model the surface of beech-wood BC as a single chemical
species. This single chemical species is likely to undergo
complex chemistry given that laboratory-generated alkane soot samples
display different regimes of reactivity and are sensitive to
photochemistry . Soot
surrogates such as the n-hexane soot recommended by the
International Steering Committee on Black Carbon Reference
Materials may therefore not
provide an accurate representation of wood-combustion soot.
Second, the raw data, supported by PMF, show that the
BC surface functionality changed considerably after aging. This
is demonstrated by the change of
CO+:CO2+ ratio after aging. Based on
the PMF factors, this ratio decreased from 5.4 to 3.5 (a 54 %
decrease).
As mentioned above, these rCOx+
signals originate from the thermally driven desorption of
different functional groups on the BC surface as CO(g)
or CO2(g) . Whether
CO(g) or CO2(g) desorbs is governed by the
nature of the functional groups themselves. The mass of
desorbed gases can be directly related to the mass of the
initial functional groups
. Although the exact path
of decomposition is sensitive to the heating rate
, the decrease of the
CO+:CO2+ ratio after aging indicates
an increased average oxidation of the BC surface groups. For
example, the decreased ratio may have corresponded to the
oxidation of phenolic or carbonyl groups at the BC surface
. In addition to the increased average
oxidation of the BC surface upon aging, the absolute quantity
of BC surface groups changed. This can be considered in terms
of the C3+:CO2+ ratio, which
decreased from 3.5 to 2.4 (a 46 % decrease). This corresponds
to an increase in the amount of functionalized carbon and
indicates that the BC surface became more oxidized upon
exposure to OH⚫ and O3. However, the
data cannot be used to estimate the absolute oxygen content of
the BC until the SP-AMS sensitivity to rCOx+ is
established by future studies (Sect. ).
Our data represent the first time-resolved measurements of
in situ BC surface aging. The SP-AMS may therefore be useful in
the online measurement of BC surface groups in atmospheric
studies. This would allow the competition between BC oxidation
and organic oxidation to be investigated, although oxidant
concentrations in this study were higher than would be expected
in the atmosphere (Sect. ).
The majority of BC emissions always occurred in the flaming
phase, during which organic concentrations were at their lowest
in both the particle phase (Figs. and ) and the gas phase . This
is because on the one hand large amounts of organic vapours are
yielded by pyrolysis (Sect. ) at
temperatures lower than the ignition point of the fire, while
on the other hand ignition triggers the flaming combustion
which simultaneously generates soot and converts the emitted
organics to CO(g) and CO2(g)
. The organic vapours which are
emitted by flaming-phase combustion appear may have followed
trajectories that avoid the flames, given that their mass
spectra resemble the aliphatic products of lignin pyrolysis
().
Given that most BC is emitted when little organics are emitted,
and that BC was overall the major species emitted by this fire
, the BC surface may represent
a significant oxidant sink during the initial aging of a similar combustion plume. The actual significance of
the BC in this case would depend on the degree of mixing
between starting and flaming phases after emission and on the
fate of co-emitted nitrogen and sulfur oxides.
It might be hypothesized that a functionalized BC particle
would become more hygroscopic and therefore more likely to act
as a CCN or ice nucleus. However, while our measurements are the first to directly
observe the BC functionalizations in a wood-combustion aerosol,
they do not directly affect the observations of previous
studies on beech-wood soot which suggest a minor role in this
regard . In the case of CCN, these
and other studies have indicated
that mixing of beech-combustion soot with other aerosols via
condensation or coagulation is the most likely pathway for
their becoming CCN active.
In general, our conclusions apply only to a well-operated stove
and, moreover, only to wood stoves. Under different operating
conditions, stove emissions may change considerably
. Less-efficient combustion
systems such as open burning may produce more primary as well
as secondary OM . These higher
organic emissions lead to a greater role of secondary OM in the
evolution of particle hygroscopicity, which is also strongly
dependent on the fuel .
Pyrolysis-formed OM
In general, the SP-AMS and AMS mass spectra were highly
similar. Excluding Cx+, the uncentred
correlation coefficients (rUC) between the two
POM–Start and POM–Flame mass spectra were 0.99 and 0.93,
respectively. Excluding all ions below m/z 44 (to account for
refractory COx+ signals and to reduce the
influence of the highest signals) increased this
rUC to 0.993 and 0.997 for POM–Start and
POM–Flame, respectively.
The mass spectrum of POM–Start was similar to that of pure
levoglucosan
, as shown
in Table . This is consistent with the fact
that C2H4O2+ signals (m/z 60), commonly
used as a tracer for biomass-burning pyrolysis products like
levoglucosan
,
were almost entirely explained by POM–Start in the PMF
model. POM–Start is thus interpreted as reflecting wood
pyrolysis, in particular the pyrolysis of cellulose,
hemicellulose, and other carbohydrates (“holocellulose”).
Holocellulose pyrolysis produces levoglucosan and other
anhydrosugars and occurs at
appreciably lower temperatures than pyrolysis of the other
major polymer in wood, lignin
.
This thermal instability, together with the fact that
holocellulose comprises about 70 % of beech wood
, explains why
POM–Start was the more-abundant POM factor. The high
CO+ signals associated with POM–Start
(Fig. c) may therefore be explained as
originating from the polyalcoholic sugars such as glucose and
xylose which comprise holocellulose. The relatively low signal
of CO2+ from POM–Start may be related to
decarboxylation reactions within the stove during pyrolysis
, similarly to the decarboxylation which
produces CO2+ within the AMS and SP-AMS during
vaporization .
An interesting feature of the POM–Start mass spectrum was the
presence of aromatic ions such as C6H5+
(phenyl) and C7H7+ (benzyl). The fact that
these ions originated from aromatic molecules was confirmed by
their persistence in the OOM–Start mass spectrum due to the
unusual stability of oxygenated aromatic molecules against
fragmentation upon electron impact
. These aromatics may have
formed during pyrolysis
or from the flame itself
. Higher
starting-phase signals of aromatics, including PAHs, have been
observed by previous studies .
Since POM–Start was ascribed to holocellulose pyrolysis, it
was hypothesized that POM–Flame may have been more
closely associated
with lignin pyrolysis, which generally requires higher
temperatures . The alkyl fragments
CnH2n-1+ and CnH2n+1+ seen
in the POM–Flame mass spectrum (Fig. b) may
be related to the cyclic aliphatic molecules, phytosterols,
emitted together with substituted phenols during lignin
pyrolysis
.
The uncentred correlation coefficient between the mass spectra
of pure burnt lignin
and POM–Flame for the SP-AMS and AMS was relatively high
(Table ), suggesting that this association was
reasonable. However, the POM–Flame mass spectra were also
well correlated with levoglucosan (Table ),
suggesting that either POM–Flame was not clearly separated
from POM–Start during factor analysis or that POM–Start
contained contributions from both lignin and
holocellulose. Both are likely to be true to some degree (the
factor separation issue is further discussed in the
Supplement). In particular, although holocellulose pyrolyzes
at lower temperatures, both holocellulose and lignin pyrolyze
across a range of overlapping temperatures
,
which would have led to a range of mass spectra being observed.
Moreover, ash species such as potassium catalyze the pyrolysis
process , so the
pure-lignin mass spectrum used here is not an ideal reference.
A second, distinct hypothesis for the origin of POM–Flame is
the in-flame synthesis of aliphatic functionalities, as has
recently been observed by
. This hypothesis
is considered unlikely given the degree of oxygenation of
POM–Flame.
Uncentred correlations, excluding HxO+,
of the OM mass spectra from PMF with literature AMS spectra .
Correlations over 0.75 are highlighted.
SP-AMS
AMS
POM–Start
POM–Flame
POM–Start
POM–Flame
Levoglucosan
0.78
0.41
0.72
0.59
m/z>44
0.84
0.58
0.84
0.55
Lignin-c.∗
0.70
0.56
0.63
0.60
m/z>44
0.71
0.80
0.74
0.79
∗ Lignin-c.: OM from the
combustion of pure lignin; lignin in wood may pyrolyze differently
.
Comparison of AMS and SP-AMS OM signals
The interpretation of the POM mass spectra as dominated by
pyrolysis products provides insight into the relationship
between the SP-AMS and AMS data, in particular the excellent
correlation between the two sets of mass spectra
in spite of the SP-AMS signals being frequently higher
(Figs. b–c and b–d).
The
excellent mass-spectral correlation is much better than
expected given the possibility of different vaporization
temperatures in SP-AMS and AMS
(Sect. ). Differences in vaporization
temperature have previously been invoked to explain
fragmentation differences in the SP-AMS relative to the AMS
for OM coatings of diesel-exhaust soot and of a branched-chain
laboratory diester . Since fresh
diesel-exhaust-soot coatings consist mostly of lubricating oil
,
both of these samples are chemically distinct from
pyrolysis-generated OM. The similarity in AMS and SP-AMS
pyrolysis-OM mass spectra may be due to fact that pyrolysis
products have already undergone thermal bond rearrangement and
dehydration reactions and are thus
less likely to do so when heated in the AMS or SP-AMS.
The higher SP-AMS signals in
the OM time series in
Fig. b–c
might be hypothesized to reflect a difference in the SP-AMS sensitivity to
OM when it is internally mixed with BC due to a change in either
vaporization temperature or physical position of the vaporized particle
(Sect. ; ). However, this difference was
observed even in the absence of rBC, as shown
by the second fresh-emissions burn.
In addition, the difference between SP-AMS and AMS was smaller for the
aged-emissions experiment (Fig. b–c) than the
fresh-emissions experiment (Fig. b–c). The apparent
influence of aging is unlikely to be related to a change in mixing state:
both aerodynamic- and mobility-size distributions were unimodal. A
difference in particle focussing efficiency (Sect. ) also does not explain the
difference, as aging would have increased the size of the particles and
therefore focussed a larger fraction of them into the SP-AMS laser. The
difference between fresh and aged samples may therefore reflect an influence
of the chemical composition of the OM.
It is therefore hypothesized that the observed discrepancy was caused by
1064nm light-absorbing carbonaceous species other
than rBC brown carbon;. Brown-carbon absorption
would explain why the SP-AMS/AMS discrepancy was reduced after
aging (Figs. b–c and b–c),
since oxidation may reduce the conjugated or aromatic bonds
required for light absorption. Lignin
and its pyrolysis products
contain the majority of the aromatic
species from wood and is itself brown
. The water-soluble
component of aerosol from inefficient beech-wood combustion is
also brown . Brown carbon would explain why
the SP-AMS/AMS discrepancy in the aged experiment was highest
for the anomalously high emissions of the second, failed-start
burn (Fig. b and Fig. S2).
Refractory sources of potassium
Potassium ions, K+, were observed as a dominant
species in the SP-AMS mass spectrum but were negligible in the
AMS. The SP-AMS therefore provides the possibility to
specifically measure rBC-bound K+ with high
sensitivity, allowing its use as an atmospheric tracer for
biomass-combustion aerosols. The requirement of internal mixing
with rBC would avoid interference from other atmospheric
sources of potassium like dust, vegetative debris, and sea salt
, although a minor
contribution to K+ from vehicular emissions may be
expected .
Refractory sources of water
As discussed in Sect. ,
H2O+ quantification in the SP-AMS is routinely
confounded by background signals from water vapour and
particulate water. The size-resolving PToF mode of the SP-AMS
was used to separate these background signals from particulate
signals, which was only possible when mass loadings were
sufficiently high
(Fig. S8).
The particulate H2O+ signals
were virtually negligible in the AMS but extremely high in the
SP-AMS: a factor of 40 higher than C3+ (even
after including the C3+ RIE;
Sect. )
during the flaming phase.
Unlike COx+,
the H2O+ intensity did not change after aging
and was constant (when normalized to C3+) to
within 10 % between fresh and aged experiments.
The fact that the relative amount of H2O+ did
not change after aging suggests that the dominant source was
not surface functional groups
. Rather, the large amount of
H2O+ observed and its thermal stability in the
AMS point towards an origin of liquid water adsorbed in pores
on the BC itself .
Such water may have been produced during combustion or evaporated from the
logs, which had a moisture content of 12–20 % according to Swiss testing
standards. Such adsorbed
water is known to be
thermally stable and does
not evaporate easily due to the inverse Kelvin effect
.
has proposed that the nucleation of
ice on soot particles which are not immersed in water droplets
(deposition-mode nucleation) is governed by such in-pore water.
In a study of deposition-mode ice nucleation on beech-wood soot
produced by a stove similar to ours, as well as diesel soot,
found that oxidative aging did not influence
the ice-nucleating activity of the soot particles, although the
activity of the diesel and wood soot particles was different.
Our observation that the H2O+ signal was
unchanged upon aging is consistent with those results and
suggests that the SP-AMS H2O+ signals may relate
to the ice-nucleating potential of a given soot sample. More
work is needed to explore this hypothesis.
Summary and conclusions
Dual-vaporizer aerosol-particle mass spectrometry was used to
investigate the composition of beech-wood soot as a function of
combustion time and simulated atmospheric aging. Vaporization
via contact with an 873 K vaporizer (AMS) or via
radiative heating by a 1064 nm continuous-wave laser
(SP-AMS) allowed either the OM component of the particles or
the entire soot particles to be probed, respectively.
The repeated addition of new logs to the wood stove led to a
natural definition of a “starting” and “flaming” phase of
combustion. The starting phase of combustion generated large
amounts of OM with a similar mass spectrum in both AMS and
SP-AMS. Analysis of the starting-phase mass spectrum showed
that it was very similar to levoglucosan, indicating that the
OM was dominated by the products of holocellulose pyrolysis.
The corresponding flaming-phase OM appeared to consist of a
mixture of holocellulose and lignin pyrolysis. These
flaming-phase OM signals were virtually negligible relative to
signals from refractory black carbon.
The near-absence of AMS signals during flaming combustion
allowed refractory sources of signals in the SP-AMS to be
distinguished. These signals were dominated by
C1-3+, CO1-2+, and
K+. The ratio C4+:C3+
was within the range reported by previous studies for
mass-spectrally similar rBC samples.
Factor analysis showed that the refractory, oxygenated
carbonaceous ions, rCOx+, were strongly
associated with the rBC ions (Cx+). Moreover,
rCOx+ signals increased upon oxidation relative
to Cx+, and the CO2+ increased
relative to CO+. It was thus inferred that BC
surface groups were the source of the rCOx+
signals. The degree of BC surface oxidation did not vary for
different stages of the burn in either the fresh or aged
aerosols and did not appear to be dependent on the
concentration of co-emitted organics. For slightly or
moderately aged aerosols, these conclusions may not hold and
should be explored in future work. To our knowledge, these are
the first measurements to address the surface oxidation of BC
in the presence of co-emitted gases.
Significant signals from K and from H2O
were generated by soot particles in the SP-AMS only. The
potassium signals would allow the SP-AMS to differentiate
biomass-combustion soot from other sources in the
atmosphere. The water signals did not change with oxidation, as
would be expected had they originated from
thermal-decomposition reactions and were much higher in signal
than COx+. They were therefore inferred to have
originated from pores in the BC itself
, which have been implicated
in the heterogeneous ice-nucleating activity of soot
.
These results indicate that wood-combustion soot is
significantly oxygenated and contains OM impurities of similar
chemical composition to the wood itself. The hygroscopicity,
CCN activity, heterogeneous chemistry, and ice-nucleating behaviour
of this soot (when either fresh or aged) are therefore likely
to be different than that of laboratory surrogates. However, as
only one fuel has been studied under controlled conditions in
this work, more data are needed to constrain the compositional
properties of biomass-combustion soot.