Introduction
Organic matter (OM) in the cryosphere originates from different sources
(e.g. oxidation products of anthropogenic, biogenic, and biomass burning
volatile organic compounds – VOCs), is transported from short and long
distances, and is deposited via dry or wet deposition
(Antony et al., 2014). From the moment of the
emission, OM undergoes atmospheric chemistry processes, which profoundly
alter the chemical composition of OM, resulting in numerous chemical species
that are finally deposited on the snow/ice surface
(Legrand
et al., 2013; Müller-Tautges et al., 2016). Fingerprints of OM stored in
the snow and ice therefore potentially hold a rich historical record of
atmospheric chemistry processes and the transport pathways in the atmosphere
(Fu
et al., 2016; Giorio et al., 2018; Grannas et al., 2006; Pokhrel et al.,
2016).
The vast diversity of OM, which is found in snow and ice samples, is
impossible to characterise by one single method. The most used methods so
far in snow/ice OM research are based on gas chromatography (GC) and liquid
chromatography mass spectrometry (LC-MS)
(Giorio et al., 2018; Gröllert et
al., 1997). Novel high-resolution mass-spectrometry-based analytical
methods, such as Fourier-transform ion cyclotron resonance mass spectrometry
(FT-ICR-MS), Orbitrap mass spectrometry, and thermal desorption–proton
transfer reaction–mass spectrometry (TD-PTR-MS), have recently been
developed and can be used to characterise OM in the cryosphere with high
mass resolution
(Hawkes
et al., 2016; Kujawinski et al., 2002; Marsh et al., 2013; Materić et
al.,
2017). Therefore, numerous new proxies are now potentially available to interpret
the rich composition of OM in the cryosphere.
Reconstructing past atmospheric conditions from measurements of OM in the
cryosphere is analytically challenging because of (1) low concentrations of
target organics in the sample and (2) chemical changes that (might) happen
after OM deposition. A recently developed method using TD-PTR-MS has partly solved
the first issue, enabling the detection of low-molecular-weight OM ranging
from 28 to 500 amu (Materić et al.,
2017). However, chemical changes (e.g. photochemical, biological) and re-emission from the snow/ice surface
still remain challenging to quantify, especially in the context of the
diversity of OM species, both high- and low-molecular-weight OM.
The low-molecular-weight fraction represents an important part of OM in the
cryosphere and the group includes VOCs and semi-volatile organic
compounds (sVOCs) that deposit directly from the gas phase or as part of
secondary organic aerosols (SOAs). Low-molecular-weight OM has been
extensively studied in an atmospheric context by real-time or off-line PTR-MS
techniques
(Gkatzelis
et al., 2018; Holzinger et al., 2010b, 2013; Materić et al., 2015;
Timkovsky et al., 2015), however, not so in the context of deposited (e.g. dissolved) OM in the
cryosphere.
In this work, we applied a novel TD-PTR-MS method to measure concentrations
of OM present in alpine snow. The first application of this new technique is
investigation of snow–atmosphere interaction of OM during a dry weather
period.
Material and methods
Sampling site
The snow samples were taken at 3106 m in altitude at Hoher Sonnblick,
Austria, close to the research station Sonnblick Observatory. The sample
site was next to the southern precipitation-measuring platform, which is
about 50 m south-east of the observatory. The sampling location was carefully
chosen to be least affected by potential contamination coming from the
observatory. Average temperature of the site is about 1.1 ∘C in
summer and -12.2 ∘C in winter considering the meteorological data
being gathered since 1886.
The sample period spans the days from 20 March to 1 April 2017.
During this time period the Sonnblick Observatory experienced an
average day length of 12.5 h, an average temperature of
-4 ∘C, 78 % relative humidity, an average wind speed of
7.3 ms-1,
and a pressure of 696 hPa. There was no significant precipitation
observed but these days were mostly foggy in the morning with the exception
of 27 and 28 March 2017, which were nearly clear-sky days,
followed by less cloudy days till 1 April 2017. The measured air
temperatures (2 m above the surface) at the site were below zero for
the whole
time, with the exception of three brief instances when the temperature was
recorded at
0.1 ∘C for 10 min (Fig. 1a). However, hourly temperature
averages for these events were also <0 ∘C. If we use a
positive degree-day (PDD) model to assess the melting possibility of those
single 10 min periods, we calculated the depth of the meltwater to be 1.4–5.5 µm
(using the snow melt factor of 2–8 mm∘C-1day-1)
(Singh et al., 2000). Thus, we conclude that no
significant melting and runoff happened for the entire sampling period.
Meteorological data measured at Sonnblick station during the
sampling: (a) temperature, (b) global radiation, (c) wind direction, (d) relative
humidity. Vertical lines represent the sampling time. Note the wind
direction change in the period preceding the sampling on 29 March 2017.
More information on the meteorological conditions can be found in Figs. 1 and A1.
Sampling
Snow samples were taken every third day from the surface snow (<2 cm),
scooping the snow directly into clean 50 mL polypropylene vials. We also
took field blanks (ultrapure water) to assure that our blanks were exposed
to the same impurities as the snow samples. The samples were stored in a
freezer at -20 ∘C until the end of the sampling campaign and then
shipped on dry ice to the analysis lab, where they were kept frozen until
the analysis.
Analysis
Prior to the analysis, the samples (and blanks) were melted at room
temperature and filtered through a 0.2 µm PTFE filter. We loaded
1 mL
of each sample into clean 10 mL glass vials that had been prebaked at
250 ∘C overnight. The samples together with the field blanks were
dehydrated using a low-pressure evaporation–sublimation system and analysed
by TD-PTR-MS (PTR-TOF 8000, IONICON Analytik), following the method
described before (Materić et al.,
2017). The samples (triplicates) were run randomly
and the blanks (four replicates) were run in between covering the entire
period of the experiment. PTR conditions included a drift-tube pressure of 295 Pa,
drift-tube temperature of 120 ∘C, and drift-tube voltage of 603 V, yielding
E / N 122 Td. The thermal desorption procedure was optimised for snow-sample
analysis and has the following temperature sequence: (1) 1.5 min incubation
at 35 ∘C, (2) ramp to 250 ∘C at a rate of 40 ∘Cmin-1,
(3) 5 min at 250 ∘C, and (4) cooling down to <35 ∘C. The method is fast (<15 min per run), sensitive (e.g. limit of detection (LoD) <0.17 ngmL-1 for pinonic
acid, LoD <0.26 ngmL-1 for levoglucosan), requires a small
sample size (<2 mL of water), and provides reasonably high-mass-resolution data (>4500, full width at half maximum, FWHM).
For the data analysis, we used the custom-made software package PTRwid for
peak integration and identification and R scripts for statistical analyses
(linear regression, fitting, etc.)
(Holzinger, 2015). We used 3σ of the
field blanks for estimating the LoD, so only ions
that are above this value were taken into account for the scientific
interpretation (Armbruster and Pry, 2008). We
evaluated the impurities in the field blanks by comparing them with the
system blanks (clean vials) and discovered that the average impurity level
of a field blank was reasonably low (7.0 ngmL-1), which mostly (60 %)
originated from the ion m/z 81.035 (C5H4OH+). The
impurities here might originate from the polypropylene vials we used;
however, the levels are much lower compared to the methods used for
measuring total and dissolved OM (Giorio et al.,
2018). The impurities were taken into account by means of field blank
subtraction and LoD filtering (Materić et al.,
2017).
From the mass spectra, identified peaks were integrated over 8 min
starting when the temperature in the TD system reached 50 ∘C.
Extracted peaks were quantified by PTRwid and the concentration was
expressed in nanograms per millilitre of sample. We calculated the molar concentration
of C, H, O, and N for each sample, from which atomic ratios (O/C, H/C, N/C),
mean carbon number (nC), and mean carbon oxidation state (OSC) are
calculated as described earlier
(Holzinger
et al., 2013; Materić et al., 2017). For the elemental composition
calculation, we excluded ions m/z<100 as these are dominated by
thermal dissociation products of non-volatile high-molecular-weight
compounds. Taking into account these fragments of bigger molecules would
substantially alter elemental composition and atomic
ratios.
Total concentration of organic ions and cumulative metrics of
atomic ratio distribution. (a) Total concentration in nanograms per millilitre; the line
represents the fit from the simple deposition model explained in the text
(Eq. 1); (b) H/C ratio; (c) O/C ratio; (d) N/C ratio; (e) oxidative state of
carbon; (f) mean numbers of carbon. The error bars represent the standard
deviation of three replicates.
Results and discussion
Total ion concentration and simple mass balance model
During our sampling period, the total concentration of organics increases in
general over the time that the snow was exposed to the atmosphere (Fig. 2a).
The concentration of organics in the snow surface reflects a dynamic balance
between two opposing processes that work independently: deposition as source
and loss. If we consider just dry deposition (it was a period without
precipitation), the retained (actual) concentration of the organics in the
snow can be described as
dmdt=D-L,
where m is the concentration of organics remaining in the snow, D is the total
dry deposition rate, and L is the overall loss rate due to re-volatilisation,
photochemical reactions, biological processes, etc. As our samples generally
show an increase in the ion concentrations (Figs. 2 and 3), the loss rate by
re-volatilisation, photochemical reaction, and biological decay is lower than
the total deposition rate (D>L). A negative mass balance, i.e. D<L, can happen,
for example, in periods of extensive photochemical reactions together with
snow exposure to an air mass with a low concentration of OM.
Box plots of concentration for ions representing four distinctive
groups (tick line of a box represents the median and upper and lower line
maximum and minimum values): (a) ion m/z 115.070 – pinonic acid, (b) ion
m/z 85.029 – levoglucosan, (c) ion m/z 99.008, and (d) ion m/z 159.065. The
lines illustrate the change in the concentration over the time that is
typical for each group. The first sample is taken just after the
precipitation (snow symbol), followed by a non-precipitation period for the
rest of the experiment (other weather symbols).
Our total concentration data (as well as many individual ion groups; see
below) indicate a relaxation towards a source–sink equilibrium.
Mathematically, the simplest model that has these characteristics is a
system with quasi-constant deposition rate D (i.e. changes in deposition are
much slower than changes in the loss rate) and a first-order loss rate
(L=-km in Eq. 1), which can be integrated to yield
m=m0e-kt+Dk(1-e-kt),
where m0 is the initial concentration of m, k is the first-order loss rate
coefficient, and t is time. In our experiment, we measured m with a time step
t of 3 days and consider m0 to be our measurement of the fresh snow in
the beginning of the analysis period. Equation (2) can then be fit to the data and
the best fit for the total concentration of semi-volatile organic traces
(R2>0.9899 and rRMSE<3.5 %) was found for k=0.31 day-1 and D=206 ngmL-1day-1. When the fit is
applied to the mass of carbon in the detected organics, the best-fit values
for the two parameters are k=0.30 day-1 and D=114 ngmL-1Cday-1,
respectively. Considering reported average organic aerosol (OA)
concentration we assume the winter air concentration (C) to be at most 2 µgCm-3
(Guillaume
et al., 2008; Holzinger et al., 2010a; Strader et al., 1999). Further taking
an average sampling depth of 2 cm and a snow density of 250 mgmL-1 we
calculated a deposition velocity of 0.33 cms-1 according to Eq. (3):
v=DC×A,
where D is the measured deposition rate, C is concentration (2 µgCm-3),
and A is the area that was typically sampled (combining sampling
depth and snow density relates 1 mL of the sample to an area of 2 cm2).
Assuming slightly higher (3 µgCm-3) or lower (1 µgCm-3)
OA concentration in air and sampling variation
between 1.5 and 2 cm in depth we calculated a positive error of a factor of 2
and a
negative error of a factor of 2-1. Thus, a deposition velocity for OAs of 0.17–0.66 cms-1 would be required to be consistent with
the observations. However, the deposition velocities for OA
were previously estimated to be 0.034±0.014 and 0.021±0.005 cms-1
for particles in the 0.15–0.3 and 0.5–1.0 µm size ranges
(Duan
et al., 1988; Gallagher et al., 2002). The required deposition velocities
are approximately an order of magnitude higher than the previously reported
estimates even if we use the upper limit of expected OA
concentration (2 µgCm-3). Therefore, we conclude that the
dominating contribution to OM in the snow is from gas-phase sVOCs. As direct
measurements of bulk sVOCs do not exist, we estimated the required average
loads of sVOCs in the air passing the sampling location to explain the
observations. Using the deposition rate calculated from our measurements
(Eq. 2), the concentration-weighted average molecular mass of measured
compounds, and deposition velocities of 1 cms-1 (assuming that sVOC
deposition velocities are similar to that of formic acid)
(Nguyen et al.,
2015), we calculated an average gas-phase sVOC burden of 883 ngm-3 of
air which is equivalent to 247 ppt. Assuming slightly higher or lower
deposition velocities (±0.2 cms-1) yields errors of +221 and
-148 ngm-3, or +62 and -41 ppt. Our calculated value of average sVOC
concentration agrees with previous estimates of 600 ngm-3
(Zhao et al., 2014). Thus, our
data suggest that dynamic processes of dissolved organic matter (DOM) on the surface snow are dominated
by deposition and re-volatilisation of gas-phase sVOCs. This has important
implications for our understanding of the snow surface processes. Our
analysis suggests that air masses with different sVOC composition can leave
different OM fingerprints in the snow (discussed in the sections below).
The D/k ratio quantifies the equilibrium point (asymptote) for the model
described in Eq. (2). This represents a point at which the equilibrium is
established between deposition and losses. The derived time constant for
loss of about 3 days implies that 90 % equilibrium is established for the
total ion concentration in only 6 days. This value, however, represents an
average equilibrium time for total measured DOM, and it is reasonable to
assume that this equilibration timescale differs among different
compounds. In particular, it is estimated to be established much faster for
the gas-phase sVOCs compared to SOA.
Similar mass balance calculations will be carried out in the following
section for individual ion groups.
Grouping of ions with similar time evolution
In the data analysis of TD-PTR-MS spectra, we found 270 organic ions above
the detection limit present in the samples. Compounds that have the same
origin (similar sources or atmospheric chemistry processes) should feature
similar time evolution, if the lifetime is not so short that such a common
time evolution is lost. Based on the pattern of concentration change over
time (using a linear regression model) we identified four groups of ions with
a
similar time evolution (Fig. 3, Table A1). In groups 1, 2, 3, and 4 we
assigned 25, 33, 9, and 21 ions, respectively (88 ions in total 33 %), and
175 ions did not fall into any of these groups. Ions which we did not assign
to any group either showed different time evolution or had concentrations
close to the detection limit causing poor correlation. On average, the total
concentration levels of the ions within the four groups were 30, 56, 16, and 57 ngmL-1 and 315 ngmL-1 for ions which did not fall
into any of the described groups. Specific information can be found in Materić (2019). These levels of OM retrieved by PTR-MS agree with
previous measurements at the site, although different methods have been used
(Gröllert et al., 1997).
In the first two groups (Fig. 3a and b), among the numerous ions we
identified masses that we tentatively attribute to pinonic acid (m/z 115.07
fragment) and levoglucosan (e.g. m/z 85.03 and 97.03 fragments)
(Salvador et
al., 2016). Pinonic acid is an oxidation product of monoterpenes and the
main source is expected to be emissions from surrounding alpine conifer
forests; thus group 1 ions indicate air masses that were originally rich in
biogenic VOCs, which have been processed during transport. Levoglucosan is a
clear indicator of biomass burning and the most likely source during this
period is domestic wood combustion. Therefore, we associate group 2 ions with
the anthropogenic wood combustion sources and their products in complex
atmospheric processing.
The compounds that fall in group 3 show, after an initial increase in the
concentration on 23 March 2017, a decreasing trend (Fig. 3c; see also Table A1).
The change in the concentration of the compounds constituting this group may
point to a one-time significant pollution event which happened between
20 and 23 March. The total concentration of ions in this
group was measured to be 34 ngmL-1 (8.2 % of the total organics) on
23 March 2017. This deposition event could have come from a single source;
however higher-time-resolution measurements are needed to further
characterise the potential source. As total concentration of ions in this
group drops in 6 days below 20 ngmL-1 (3.1 % of the total organics),
this group is also an example of how contaminated snow equilibrates with the
cleaner atmosphere on timescales similar to those we derived from the simple box
model.
As for total concentration, most of the ions and ion groups show an increase
in the ion concentrations throughout the sampling period. Group 4 (Fig. 3d,
Table A1) represents the compound group for which the concentration seems
steadily increasing towards an equilibrium. This indicates that the simple
mass balance model may be applicable, i.e. the assumption of a (close to)
constant deposition and first-order loss rate. Therefore we also applied the
simple mass balance model (see Sect. 3.1) to the individual ions in group 4
to investigate whether individual organic compounds have different k values.
This is expected due to the different chemical and physical properties (such
as volatility, susceptibility to photolysis, etc.) as well as different
nutrition adequacy for potential biodegradation. For the sum of organic ions
in group 4 (Fig. 3d, Table A1), k=0.20 day-1. Generally, the
lower k value of this group compared to the total sVOC could be related to
the fact that most of the ions here are heavier (thus less volatile).
However, within this group k values of individual ions were found to be
independent of the molecular weight and also independent of the
composition, i.e. O/C, H/C, OSC, and nC (R2<0.12). As the
volatility of sVOC is expected to depend on molecular weight and functional
groups (longer sVOCs are in general less volatile, unless additional
functional groups are involved), this suggests that volatility might not
play the only role in the loss processes of this group.
A deviation in the general concentration trend in individual ions (from the
expected growth, Sect. 3.1) was observed on 29 March 2017, particularly in groups 1 and 2 represented by pinonic acid and levoglucosan (Fig. 3a and
b). Elevated levoglucosan and lower pinonic acid levels observed on
29 March are temporally related to a change in wind direction. On
29 March, the air masses originate from the north-east direction, rather
than the north-west direction seen for other samples (Fig. 1c), so this event is
attributed to the meteorological situation and possibly a more pronounced
source of biomass burning following the transport regime at the time.
Presence of such distinctive patterns of concentration change over time, ion
grouping, and their relation with the meteorological data indicate that
meteorology and deposition of sVOCs after fresh precipitation strongly
affect the organic composition in snow, which questions the most
straightforward approach of interpreting OM signals in terms of OA in the air.
Elemental composition
We further investigate the processing of OM in snow during the study period
by calculating cumulative metrics of the OM composition from the PTR-MS
data, namely the elemental ratios O/C and H/C, nC, and OSC of the organic carbon, in
order to further characterise the processes behind the observed changes. The
fresh snow sample (20 April 2017) has the lowest total concentration of all
measured organics, low OSC, the lowest O/C and N/C values, and high H/C and
nC values (Fig. 2), which all indicate “fresh” OM in the air
(Kroll et al., 2011), which was captured in
the snow.
An interesting signature in the metrics is observed on 29 March 2017 when the
prevailing air flow regime was interrupted (wind direction change, Fig. 1c).
This sample showed the highest value of nC, the lowest OSC, and
elevated H/C and low O/C ratios (Fig. 2). This all indicates photochemically
younger (fresher) emissions of VOCs and semi-volatiles, originating from
air masses rich in biomass burning aerosols (Fig. 3b), which is in agreement
with previous results linking low OSC and high nC to biomass burning
aerosols (Kroll et al., 2011). However, on
29 March we also observed lower average total OM concentration in the
sample compared to the previous period, which clearly indicates a net loss of
OM. Potential processes that could explain such a loss of OM involve
photolysis-induced re-volatilisation, OM runoff (e.g. snow melting), or
oxidation. The photolysis-induced volatilisation should be higher for this
sample as the previous days (27 and 28 March) had the highest global
radiation values (33 % higher than the average for the sampling period) and the
longest sunshine duration (>12 h) (Figs. 1b and A1). Conversely, no significant temperature increase has been measured to
support increased melting and OM runoff. Loss by oxidation (referring to
“dark” oxidation that is uncoupled from photooxidation) is also unlikely
as a main process since the O/C ratio did not increase for 29 March
(Fig. 3c). Thus, the most likely cause of the lower total OM concentration
observed on 29 March is re-volatilisation, possibly enhanced by photolysis,
which would indicate that the air contained a lower burden of SVOCs. In
addition, new OM material with different characteristics was deposited
before that sample was collected. Combining all metrics (Fig. 2) and
meteorological data available (Fig. 1), we can conclude that the air passing
the site prior to 29 March 2017 was cleaner and photochemically younger and
contained higher molecular weight compounds that might have originated from anthropogenic
emissions such as biomass burning (high levels of levoglucosan, Fig. 3).
Conclusion
In this work, we analysed the concentrations of low-molecular-weight organic
matter (20–500 amu) in alpine snow samples during a 12-day
no-precipitation period, 20 March–1 April 2017. We noticed four distinctive
groups of ions with a similar concentration trend over that time (R2>0.9),
suggesting common sources, chemistry processes, or
transport pathways. The largest two groups of ions came from (a) surrounding
forests (e.g. pinonic acid – associated with monoterpene oxidation) and (b) residential
fires (levoglucosan – common biomass burning marker). The snow
sample taken on 29 March showed a change in the general
concentration trend, consistent with a shift in wind direction, indicating
different air mass origin. This is also in agreement with a change in atomic
ratio metrics (O/C, H/C, OSC, and nC), which also indicated that
re-volatilisation is the most important pathway of OM loss here, suggesting
that the advected air was cleaner during this period. Dry deposition can be
approximated by a mass balance model with a roughly constant deposition rate
of D=206 ngmL-1day-1 and a first-order loss rate constant
k=0.31 day-1. Calculated deposition velocities were inconsistent
with the idea that OAs contribute the bulk of deposited OM;
instead we suggest a dominant contribution of gas-phase sVOCs over the OA in
the total bulk organic matter. This all indicates that, at least for this
site and location, snow–atmosphere DOM exchange processes are mostly driven
by gas-phase sVOCs, for which equilibration with air is fast. This has
implications for the reconstruction of recent atmospheric conditions by
analysis of organics in the snow.