Introduction
Organic aerosol (OA) accounts for 20–90 % of the submicron ambient
aerosol (Jimenez et al., 2009, and references therein), a great part of which
is secondary organic aerosol (SOA) formed via the condensation of oxidation
products of gas-phase precursors. Several direct (e.g. radiocarbon dating)
and indirect observations underline the key role of biogenic volatile organic
compounds (VOCs) for SOA formation (El Haddad et al., 2013, and references
therein). Current state-of-the-art models are unable to predict the burden of
biogenic SOA, especially in urban atmospheres (Hoyle et al., 2011),
highlighting a fundamental deficit in our knowledge of the chemical pathways
by which SOA accumulates and evolves in the atmosphere.
The ensemble of gaseous and particulate-phase species involved in SOA
formation is immensely complex. The compounds relevant for SOA formation are
often a minor fraction, resulting in yields (SOA formed to precursor reacted)
of only a few percent. Their chemical composition and volatility distribution
strongly depend on the oxidation conditions, most notably on the fate of
organic peroxy radicals (RO2), which react either with nitrogen oxides
(NOx) or other peroxy radicals (RO2 and HO2). The influence of
NOx on the oxidation mechanisms and SOA formation is commonly described
by the NOx / VOC ratio and has been under close scrutiny lately. For
most light precursors, such as isoprene and monoterpenes (including α-pinene), SOA yields appear to be strongly influenced by the
NOx / VOC ratio, with a general enhancement observed under low
NOx conditions (Ng et al., 2007; Presto et al., 2005).
Species involved in SOA formation are subject to ongoing chemical
degradation, which may lead to compounds of either lower (when
functionalization dominates) or higher volatility (when fragmentation
dominates). As a consequence, SOA yields and degrees of oxygenation
(described by the atomic oxygen to carbon ratio O : C) may depend on the
extent to which these species were exposed to oxidants (Kroll and Seinfeld,
2008; Donahue et al., 2012a).
SOA yields are generally described by the absorptive equilibrium partitioning
of condensable species to a well-mixed liquid phase (Odum et al., 1996),
which depends upon the chemical species' saturation vapour pressures
(classically two model products are considered) and their liquid-phase
activities (modified Raoult's law). Donahue and co-workers proposed the use
of a “volatility basis set” (VBS) for a better representation of the wide
range of OA in the atmosphere and the ongoing oxidation of semi-volatile
organics (Donahue et al., 2011, 2012b, and references therein). However,
several difficulties remain in estimating or measuring the saturation vapour
pressures (e.g. Huisman et al., 2013; Bilde et al., 2015), activity
coefficients and the mean molecular weight of the condensing species (e.g.
Clegg et al., 2008a, b).
Difficulties increase when considering the role of relative humidity (RH) and
of electrolyte particles in organic partitioning (Zuend and Seinfeld, 2013,
and references therein). From a thermodynamic point of view, water interacts
with SOA components by altering the water content of aerosol particles (also
at subsaturated conditions) and hence the equilibrium concentration of
water-soluble organic compounds. According to the equilibrium equation of
Pankow (1994), an increase in the condensed fraction of the organics can be
achieved by (1) increasing the absorptive particulate mass, (2) decreasing
the average molecular weight of condensed species or (3) decreasing the
activity coefficients of organic species. Therefore, it is expected that an
increase in the particulate water
content would in principle enhance SOA yields of water-miscible species.
Model calculations predict a pronounced effect, especially at low organic
mass loadings (Pankow, 2010). However, literature reports based on
experimental results seem contradictory: RH-dependent yields have been
reported for the ozonolysis of limonene, α-pinene and
Δ3-carene (Jonsson et al., 2006), while Prisle et al. (2010) found
a negligible impact of RH on SOA yields from α-pinene ozonolysis. A
substantial effect of RH on SOA yields was observed only for studies at
precursor concentrations < 1000 ppbv and with large variation in RH
(0.01 and 31 % RH (Bonn et al., 2002); < 2–58 % RH (Cocker
et al., 2001); < 2–85 % RH (Jonsson et al., 2006).
The impact of RH on the partitioning of organic species may be severely
suppressed by considerable deviations from ideal mixing between the
condensing organic species and the prevailing condensed phase. A growing
number of studies show that organic compounds are salted out in internally
mixed organic/inorganic/water aerosol particles and two stable liquid phases
may develop: an aqueous electrolyte solution and an organic solution
(Marcolli and Krieger, 2006; You et al., 2014). The miscibility of an organic
compound in aqueous droplets containing electrolytes depends on numerous
factors, including temperature, relative humidity, organic compound polarity,
the relative contribution of the
compound to the bulk particulate matter and the chemical nature of the
electrolytes (You et al., 2013; Zuend and Seinfeld, 2013). For example, phase
separation was observed to always occur for organic compounds with
O : C < 0.5 and at low relative humidity (You et al., 2013), and
is especially pronounced in the case of ammonium sulfate (compared to
ammonium hydrogen sulfate and nitrate). While neglecting phase separation of
organic compounds and the effect of RH thereon bears the potential for
invalid yield predictions (e.g. if the condensed phase is considered to
comprise a mixed electrolyte and organic solution), the way the complex
organic matrix interacts with water and the inorganic species remains
virtually unknown.
Another complication that might influence the interpretation of chamber
experiments conducted at different RHs is the enhancement of SOA yields by
the potential reactive uptake of organic products into the particle phase
(Kroll and Seinfeld, 2008; Kleindienst et al., 2006; Jang et al., 2002;
Iinuma et al., 2007). The mechanisms by which such reactions occur are not
fully identified, but may involve an ester or an aldol formation, which are
expected to be favourable in the absence of water and are possibly catalytic
under acidic conditions. Assessing the relative importance of particle-phase
processing at different particle water contents would require decoupling SOA
thermodynamics and additional reactivity.
In this study, we examine the impact of particle water content on the
chemical composition and yields of α-pinene SOA formed under low and
high NOx. This is performed by varying NOx/α-pinene
ratios, aerosol seed composition (hygroscopicity, acidity) and relative
humidity. Results are parameterized within a thermodynamic framework to
investigate whether changes in SOA non-ideal mixing properties with particle
water content may explain the variation in SOA yields with RH. SOA yields
reported here may aid the parameterization of the NOx and particulate
water dependence of α-pinene SOA production for further use in
atmospheric models.
Methods
Experimental set-up and instrumentation
Twenty experiments, listed in Table 1, were carried out
in the smog chamber (SC) of the Paul Scherrer Institute (PSI): a Teflon bag
of 27 m3 suspended in a temperature-controlled housing (Paulsen et al.,
2005). Photochemistry was initiated by four xenon arc lamps (4 kW rated
power, 1.55 × 105 lumens each, XBO 4000 W/HS, OSRAM), facing
parallel to the SC bag, and emitting a light spectrum similar to the solar
spectrum, and 80 black lights (Philips, Cleo performance 100 W) to
accelerate the aging process, located underneath the SC bag, with emission
between 300 and 400 nm wavelength (light
characterization in Platt et al.,
2013). A reflecting aluminium foil surrounds
the SC bag to maintain light intensity and light diffusion.
Overview of experimental conditions. Seed types (AHS: ammonium
hydrogen sulfate (NH4HSO4); SA: sulfuric acid (H2SO4);
AS: ammonium sulfate ((NH4)2SO4); CF: fluorinated carbon (see
Sect. 2.2)) and their assumed phase states: (L): liquid; (–): liquid and/or
solid. Initial seed mass concentrations; relative humidity (RH); measured
mean NOx concentrations during low NOx experiments (marked with an
asterisk) and measured initial NOx concentrations during high NOx
experiments; measured initial α-pinene concentrations (which reacted
before an OH exposure of
(2.0 ± 0.5) × 107 cm-3 h) and their estimated
fractions reacted with OH radicals in %; the rest reacted with O3).
NOx/α-pinene ratios; wall-loss-corrected organic mass
concentrations (COA) and corresponding yields Y averaged over
the OH exposure of (2.0 ± 0.5) × 107 cm-3 h.
Standard deviations (1 SD) given in brackets are measurement variability.
Horizontal lines separate experiments with different (i) seed composition,
(ii) RH, and (iii) NOx/α-pinene. Blank experiments (B1, B2 and B3)
are listed at the very bottom.
No.
Seed
RH
NOx
α-pin
α-pin
NOx /
COA
Yield, Y
decay
α-pin
(wlc)
(wlc)
type (phase)
initial
initial or
initial =
by OH
at OH exposure:
mean(*)
reacted
(2.0 ± 0.5) × 107 cm-3 h
µg m-3
%
ppbv
ppbv
%
µg m-3
1
AHS + SA (L)
8.0(0.5)
69(2)
44.4(0.8)
20.7
80.0
2.1
13.4(0.2)
0.115
2
AHS + SA (L)
12.3(0.5)
67(2)
70.4(1.3)
18.7
80.3
3.8
8.6(0.1)
0.081
3
AHS + SA (L)
4.9(0.3)
66(2)
19.6(0.7)
16.1
81.6
1.2
12.6(0.6)
0.138
4
AHS + SA (L)
4.7(0.2)
29(1)
23.6(0.6)
19.9
81.3
1.2
3.9(0.0)
0.035
5
AHS + SA (L)
8.0(0.3)
28(1)
52.1(0.6)
20.3
74.6
2.6
2.1(0.1)
0.018
6
AHS + SA (L)
5.2(0.3)
27(1)
1.3(0.4)*
18.3
87.8
0.071
12.0(0.5)
0.116
7
AS + AHS (L)
8.8(0.4)
67(1)
0.7(0.2)*
20.0
81.9
0.037
16.2(0.4)
0.143
8
AS + AHS (L)
4.3(0.6)
60(1)
1.0(0.2)*
18.7
86.5
0.052
12.3(0.4)
0.116
9
AS + AHS (L)
5.5(0.2)
56(2)
65.8(0.8)
30.9
65.0
2.1
11.1(0.1)
0.064
10
AS + AHS (L)
4.4(0.2)
50(1)
1.3(0.2)*
30.6
79.2
0.041
29.6(1.1)
0.171
11
AS + AHS (–)
4.1(0.2)
26(1)
0.7(0.3)*
18.9
81.3
0.039
5.5(0.2)
0.051
12
AS + AHS (–)
3.6(0.1)
26(1)
1.9(0.4)*
30.5
78.4
0.062
20.3(1.3)
0.118
13
AS + AHS (–)
8.2(0.3)
25(1)
1.1(0.3)*
19.6
87.4
0.055
5.3(0.1)
0.048
14
AS + AHS (–)
3.2(0.2)
23(1)
75.1(0.7)
27.8
69.4
2.7
2.4(0.1)
0.015
15
CF (L)
7.1(0.3)
58(1)
56.2(0.7)
31.7
67.9
1.8
9.8(0.1)
0.054
16
CF (L)
10.0(0.5)
58(2)
58.3(0.5)
31.3
73.9
1.9
10.7(0.1)
0.061
17
CF (L)
6.7(0.1)
26(1)
53.3(0.6)
30.5
69.7
1.7
4.7(0.1)
0.021
B1
CF (L)
0.3(0.1)
58(2)
53.0(0.6)
–
–
–
< 0.1
–
B2
AS + AHS (L)
4.5(0.7)
68(2)
0.9(0.5)*
–
–
–
2.8(1.0)
–
B3
AHS + SA (L)
3.8(0.2)
75(3)
1.4(0.3)*
–
–
–
0.5(0.1)
–
Various parameters were monitored in the SC. The temperature (T) and RH
measurement was optimized by
passing SC air through a radiation shielded sensor. One of two different
high-resolution time-of-flight aerosol mass spectrometers (HR-ToF-AMS,
Aerodyne Research, Inc., Billerica, MA, USA) was operated during three
different campaigns to measure the online size-resolved chemical composition
(organics, ammonium, nitrate, sulfate, chloride) of non-refractory particles
(DeCarlo et al., 2006). The HR-ToF-AMS were equipped with two different
PM2.5 lenses (Williams et al., 2013) to sample particles up to large
diameters above 1 µm. The sampled aerosol was dried
(∼ 10 % RH) before measurement. A supporting flow of
∼ 1.5 L min-1 was maintained parallel to the HR-ToF-AMS to
minimize diffusive losses in the sampling lines.
The HR-ToF-AMS data were processed and analysed using the SQUIRREL
(SeQUential Igor data RetRiEvaL) v.1.52L analysis
software and PIKA (Peak
Integration by Key Analysis) v.1.11L for the IGOR Pro software package
(Wavemetrics, Inc., Portland, OR, USA). From the HR analysis of the mass
spectra, the O : C ratios of the bulk OA were determined based on the
parameterization proposed by Aiken et al. (2008). We note that while the
assessment of the uncertainties related to the O : C measurements by the
HR-ToF-AMS is not straightforward, a distinction should be made between
measurement precision and accuracy. We do not expect the accuracy of the
O : C ratios determined by the HR-ToF-AMS to be less than ∼ 20 %
(Aiken et al., 2008; Pieber et al., 2016; Canagaratna et al., 2015; Bozzetti
et al., 2017). For example, the use of a more recent parameterization
(Canagaratna et al., 2015) would yield higher O : C values (by 18 %)
and the O : C ratios reported here may be regarded as the lowest estimates.
By contrast, relative changes in the O : C ratios are expected to be
detected more precisely by the instrument (∼ 1–2 %). The influence
of potential biases and uncertainties in the determination of the O : C
ratios on our results will be discussed in the text.
Two scanning mobility particle sizers (SMPS) were additionally deployed for
the measurement of the aerosol size distributions. The first SMPS (a
custom-built differential mobility analyser, DMA: extended length
Leff= 0.93 cm, dmmax=1000 nm, recirculating
sheath flow, and a condensation particle counter, CPC 3022 (TSI)) was
connected to the HR-ToF-AMS sampling line to analyse the dried particles. A
second SMPS (SMPSwet, a TSI DMA classifier 3081 with recirculating
sheath flow and a TSI CPC 3022A) and a CPC (TSI: CPC 3025A) measured the wet
particle number size distribution and total number concentration
(d > 3 nm), respectively.
Gas-phase compounds with a higher proton affinity than water
(166.5 kcal mol-1) were measured with a quadrupole proton transfer
reaction mass spectrometer (PTR-MS, Ionicon). The PTR-MS was calibrated
before each experiment for α-pinene, detected at m/z 137 and
m/z 81; the accuracy of these measurements was estimated to be
∼ 5 %, based on the purity indicated on the calibration gas
cylinder.
A modified NOx instrument including a photolytic NO2-to-NO
converter (Thermo Environmental Instruments 42C trace level NOx analyser
equipped with a blue light converter) and two ozone monitors (Monitor Labs
8810 ozone analyser, Environics S300 ozone analyser) monitored NOx and
O3 in the chamber.
Chamber operation and aerosol seeding
SOA formation and growth from α-pinene were induced by the following
SC operation sequence: (1) humidification of the chamber, (2) addition of
seed aerosol, (3) introduction of VOCs, (4) addition of nitrous acid (HONO)
as an OH precursor, (5) addition of nitrogen oxides (equal amounts of
NO + NO2) if applicable, (6) an equilibration period (30–45 min),
(7) switching on of xenon and black lights to generate OH radicals, and (8) a
reaction time of 5 to 20 h (corresponding to
0–2 × 108 cm-3 h OH exposure; see Sect. 2.4).
Experimental conditions for each individual experiment are summarized in
Table 1. Prior to each experiment, cleaning of the SC was performed by the
injection of several ppmv of ozone (5 h) and the simultaneous irradiation
with black lights (10 h) at a temperature of 20 ∘C. This was
followed by a pure air flushing period at high relative humidity
(∼ 60 %) at a temperature of approximately 30 ∘C for at
least 20 h. Three blank experiments (seed aerosol, lights switched on, high
RH, but without adding α-pinene) were carried out to make sure that
the organic aerosol formed during the experiments is not significantly
influenced by background contamination in the SC. The organic mass
concentration formed was substantially lower (< 0.1 up
to 2.8 µg m-3) than during comparable experiments (similar
NOx and RH).
In the chamber, the temperature varied between 21 ∘C and
26 ∘C. Due to heat from the xenon and black lights, the temperature
increased, stabilizing only ∼ 1 h after experiment start. The increase
in temperature of 1–4 ∘C led to an absolute decrease in RH of
∼ 2–20 %; thus, the RH range in Table 1, 23–67 %, is given
for the stable temperature period. α-Pinene (98 %, Aldrich) and
an OH reactivity tracer (9-times deuterated butanol, 98 %, D9, Cambridge
Isotope Laboratories), hereafter referred to as butanol-d9 (1 µL
injected ≈ 10 ppbv in the SC), were sequentially injected into an
evaporation glass bulb heated to 80 ∘C. The two VOCs were
transferred into the bag by a dilution and flush flow (each
15 L min-1, maintained for 15 min) from an air purifier (737–250
series, AADCO Instruments, Inc., USA), further referred to as “pure air”.
Initial α-pinene concentrations were 16.1–31.7 ppbv.
HONO was used as a source of both NO and OH, produced by continuous mixing in
a reaction vessel of the reagents sodium nitrite (NaNO2,
1 mmol L-1 in milliQ-H2O) and three different concentrations of
sulfuric acid solutions (H2SO4, 1 mmol L-1
(experiments 1–6), 10 mmol L-1 (experiments 7, 8, 11, 13) and
100 mmol L-1 (experiments 9, 10, 12, 14) in milliQ-H2O) (Taira
and Kanda, 1990). HONO was carried by 2.5 ± 0.2 L min-1 pure air
flow into the SC; 2 ppbv (±10 %) of HONO were injected before lights
on to initiate photochemistry and the injection was continued throughout all
experiments. A chemiluminescence-based NOx instrument (Monitor Labs
9841A NOx analyser) was attached to the HONO source to monitor the
injected concentration throughout the experiment. In addition, equal
concentrations of NO (99.8 %; 1005 ppmv ± 2 %) and NO2
(purity: 98 %; 1005 ppmv ± 3 %), resulting in 19.6–75.1 ppbv
initial NOx, were added during experiments with NOx/α-pinene
> 1. Within the results section, the two terms “low NOx”
and “high NOx” refer to the following conditions.
Low NOx= NOx/α-pinene < 0.1, with
continuous HONO injection, indicated by an asterisk in Table 1 and figures.
High NOx= NOx/α-pinene > 1: initial
injection of NO + NO2 with continuous HONO injection.
In Table 1, we report for high NOx conditions the initial NOx
concentration (which decays with time), and for low NOx conditions the
mean NOx concentration. We note that the NO levels are the main driver
in determining whether RO2–NO or RO2–RO2 reactions would
prevail. Based on our calculations (not shown here; see Platt et al., 2014),
RO2 radicals would predominantly react with NO, when the concentration
of the latter is higher than only 1 ppbv. These conditions can be considered
high NOx. During high NOx experiments and throughout the period
when the majority of α-pinene was consumed, the NO concentration
remained higher than 5 ppbv, indicating that α-pinene oxidation
proceeded under high NOx. During low NOx experiments, the average
NOx concentration was around 1–2 ppbv, predominated by NO2, while
the NO concentrations were below detection limits (< 0.1 ppbv).
Under these conditions RO2–RO2 reactions may prevail.
During 14 experiments (nos. 1–14) and 3 blank experiments, an ammonium
hydrogen sulfate (NH4HSO4, Aldrich) solution in ultrapure Milli-Q
water (1 g L-1) was nebulized (0.6 L min-1) and introduced into
the SC with a pure air dilution flow of 10 L min-1 to act as seed
particles. To keep the seed aerosol in a liquid state, no drier was used
behind the nebulizer. We determined the acidity of the seed particles, here
described by the ratio NH4 / SO4, by the comparison between the
HR-ToF-AMS measurements of the seed particles and nebulized
(NH4)2SO4 and NH4HSO4 solutions. An
NH4 / SO4 ratio between 1 and 2 indicates a rather neutral seed
composition, representing a mixture of NH4HSO4,
(NH4)3(SO4)2 (letovicite) and (NH4)2SO4. For
simplicity, we replaced (NH4)3(SO4)2 with an equal mix of
NH4HSO4 and (NH4)2SO4 for the assumption of density
and growth factors. By contrast, NH4 / SO4≤1 indicates an
acidic seed, consisting of NH4HSO4 and H2SO4.
During experiments 7–14, the nebulized NH4HSO4 solution was partly
neutralized to (NH4)2SO4, presumably by background NH3
(Fig. S1 in the Supplement). During experiments 1–6 the seed was composed of
an aqueous mixture of NH4HSO4 and H2SO4, due to the in
situ production of gas-phase H2SO4 via the HONO injection
system, suppressing the seed neutralization by NH3 (even though
particulate H2SO4 was minimized by a Teflon filter applied between the HONO source and the SC).
Additionally, three α-pinene experiments (nos. 15–17) and one blank
experiment were conducted using an inert hydrophobic fluorinated hydrocarbon
seed (further referred to as the CF seed;
CF3CF2CF2O–[CF(CF3)CF2–O–]nCF2CF3;
Krytox® 1525). The CF seed was generated via
the evaporation of the pure compound at a temperature of 125–145 ∘C
and subsequent homogeneous nucleation in a pure air flow of
2.4 ± 0.1 L min-1. The CF-seed concentration was
6.7–10 µg m-3 when lights were switched on, and decayed very
rapidly to values below the detection limit of the HR-ToF-AMS
(0.1 µg m-3) during the course of the experiment. The CF-seed
mass spectrum in the HR-ToF-AMS is clearly distinct from that of α-pinene SOA, with main contributions at m/z 69 (CF3), m/z 169
(C3F7) and m/z 119 (C2F5) (Fig. S2 and Table S1 in
the Supplement).
Table 1 lists the expected physical state and seed composition of each
experiment dependent on RH (AHS: ammonium hydrogen sulfate
(NH4HSO4); AS: ammonium sulfate ((NH4)2SO4); SA:
sulfuric acid (H2SO4); CF: fluorinated carbon). Submicrometer AS
particles and letovicite particles ((NH4)3H(SO4)2) are
expected to effloresce at about 35 % RH, while more acidic particles
should remain liquid between 20 and 30 % RH (Martin, 2000; Ciobanu et
al., 2010).
Estimation of the hygroscopic growth factors and liquid water
content
The absolute liquid water content (LWC) of the aerosol particles was derived
for the bulk aerosol mass and the size-resolved mass distribution, based on
literature growth factors, the measured RH and chemical composition. The
growth factor GF (RH) of a particle is defined as the ratio of the wet
diameter at a given RH to the dry diameter:
GF(RH)=d(RH)/ddry.
Inorganic GFs were taken from the Aerosol Diameter Dependent Equilibrium
Model (ADDEM, Topping et al., 2005) for diameters of 360 nm. Organic GFs
were derived using the relationship between the hygroscopicity parameter
κ and the degree of oxygenation [κ = 0.29 × (O : C)] from Chang et al. (2010), well
representing the hygroscopicity of α-pinene SOA measured by Massoli
et al. (2010). The measured degree of oxygenation at an OH exposure of
(2.0 ± 0.5) × 107 cm-3 h was used to derive
κ, which in turn was converted to GF, assuming a negligible curvature
(Kelvin) effect (Kreidenweis et al., 2005):
GF(RH)=1+κRH100%1-RH100%13.
The mixed GFs for aerosol containing inorganic and organic species were
determined using as a first approximation the Zdanovskii–Stokes–Robinson
(ZSR) volume mixing rule (Stokes and Robinson, 1966):
GFmixedRH=∑iεi×GFiRH313,
where εi and GFi(RH) denote the volume fraction and GF
(RH) of species i, respectively. The H2O volume
(VH2O) was calculated using the definition of GF (RH) and the
dry volume (Vdry):
VH2O=Vdry×GFRH3-1.
VH2O multiplied by the density of water (1 g cm-3)
results in the LWC.
The dry (Sdry) and wet (Swet) surfaces from
HR-ToF-AMS size-resolved data were calculated with Eqs. (4) and (5),
respectively:
Sdry=6×(Vdry)/ddry,Swet=6×(Vdry+VH2O)/(dRH).
The LWC and surface distributions were calculated using size-resolved pToF
(particle time-of-flight) data of the HR-ToF-AMS. Due to the low pToF signal
of NH4, the NH4 surface distributions were estimated based on
SO4 pToF measurements.
Determination of OH exposure and extent of α-pinene
ozonolysis
The gas-phase composition, the OH concentration and the photochemical age of
a chemical reaction system may considerably differ between experiments of the
same duration. Furthermore, variation of the OH concentration within a single
experiment means that the photochemical age is not necessarily directly
proportional to the light exposure time. Consequently, we discuss reaction
time in terms of OH exposure (mol cm-3 h), defined as the OH
concentration integrated over time. OH exposures were derived based on the
decay of the OH tracer butanol-d9, detected by the PTR-MS as
M + H+–H2O at m/z 66, following the methodology
introduced by Barmet et al. (2012). The OH exposure was determined by the
integration of the following expression:
OHexposure=-∫t1=0t1kOH,butanol-d9×Δlnbutanol-d9Δt+fdilVdt,
where butanol-d9 and kOH,butanol-d9
(= 3.4 × 10-12 cm3 mol-1 s-1) are the
butanol-d9 concentration and its reaction rate constant against OH,
respectively, t is the time after lights on, V the chamber volume (we
assume as a first approximation a constant chamber volume of 27 m3),
and fdil the dilution flow due to HONO input (Sect. 2.2).
The percentage of α-pinene reacted with OH and O3 was derived
based on Eq. (7):
-dα-pindt=kO3,α-pinO3×α-pin+kOH,α-pinOH×α-pin+fdilV,
where [α-pin], [O3] and [OH] denote the concentrations of
α-pinene (measured by the PTR-MS at m/z 137 and 81), O3 and
OH, respectively, and kOH,α-pin
(= 5.3 × 10-11 cm3 mol-1 s-1) and
kO3,α-pin
(= 8.9 × 10-17 cm3 mol-1 s-1) the reaction
rate constants of α-pinene with OH and O3, respectively. The
percentage of α-pinene reacted with OH is presented in Table 1 and
discussed in Sect. 3.4.
Determination of suspended and wall-loss-corrected organic masses and
yields
Suspended OA mass. The suspended organic mass concentration
COAsus was derived by utilizing the chemical composition
measurements from the HR-ToF-AMS scaled to the total volume measured by the
SMPS (Figs. S3 and S4), using compound-specific densities
(ρorg= 1.4 g cm-3,
ρNH4HSO4= 1.79 g cm-3,
ρ(NH4)2SO4= 1.77 g cm-3,
ρH2SO4= 1.83 g cm-3).
For some experiments (9, 10, 12 and 14–17), the organic mass concentrations
determined by the HR-ToF-AMS were corrected for a sub-unity transmission
efficiency at the lower edge cut-off of one of the two PM2.5 lenses
employed (Figs. S3 and S4). Additionally, for organic mass calculation, we
assumed that the measured NO3 signals are entirely related to
organonitrates (RONO2), rather than NH4NO3. This assumption
mainly stems from (1) the observation of the NO3 signal in the same
particle size region as OA rather than SO42- (see the size-resolved
pToF data, Sect. 3.3), while inorganic nitrate would be expected to mix
within an electrolyte-rich aerosol and (2) the presence of NO3 under
acidic conditions, which are thermodynamically unfavourable for the
partitioning of nitric acid. This is also supported by the higher
NO+ / NO2+ ratios measured in the SC compared to ratios
recorded during NH4NO3 nebulization (on average 1–2.8 times
higher, Fig. S5), typically expected from organonitrates (Farmer et al.,
2010). We cannot exclude a part of the NO3 signal originating from
NH4NO3. However, even attributing all detected nitrate to
NH4NO3 would increase the calculated LWC by 1–13 % and
decrease the calculated OA mass by 2–7 % only, which would not influence
our conclusions. Finally, for yield calculations we assume the accuracy of
the aerosol-phase measurements to be 30 % (Canagaratna et al., 2007).
Wall-loss-corrected OA mass. To obtain the total COA
concentration corrected for losses of particles and vapours to the chamber
walls, we use Eq. (8), introduced by Hildebrandt et al. (2011), based on the
mass balances of the suspended organic aerosol mass,
COAsus, and the mass of the organic aerosols on the walls,
COAwalls (summed up to derive COA).
ddtCOAwallst=kOAwtCOAsust+ωt×kOAwtCOAsust+ddtCOAsust×COAwallstCOAsust
Here, kOAw represents the loss rate constant of organic
particles to the walls, derived by fitting the suspended organic mass
concentration 5–8 h after lights were switched on in the SC, when SOA
production is expected to be negligible (α-pinene concentration
< 1 ppbv, Fig. S6 and Table S2). We determine an average loss rate
of 0.13 µg m-3 h-1, corresponding to a particle
half-life of 5.3 h. The average kOAw was used in the
case of insufficient statistics to perform accurate fitting. In Eq. (8),
ω, ranging between 0 and 1, is a dimensionless proportionality
coefficient between the mass of organic vapours that partition onto the
wall-deposited particles and the mass of organic vapours that partition onto
the suspended particles. Here, we neglect the condensation of organic vapours
onto the wall-deposited particles, i.e. ω=0, consistent with
previous studies of α-pinene SOA production (Hildebrandt et al.,
2011, and references therein). This assumption gives a lowest estimate of SOA
yields, but does not influence the comparison between the experiments.
Considering the second limiting case ω=1, i.e. an equal
partitioning of organic vapours between the wall-deposited and suspended
particles, would increase the determined SOA yields (by up to 40 %, and
by 20 % on average).
Equation (8) does not take into consideration the loss of SOA-forming vapours
onto the clean Teflon walls, which may suppress SOA yields from laboratory
chambers under certain conditions. These processes may be related to the
vapours' reactive uptake onto walls (Loza et al., 2010; Nguyen et al., 2016)
or to their absorptive uptake (Zhang et al., 2014; Nah et al., 2016). The
reactive uptake of organic vapours onto chamber walls is only significant at
high RH. Loss rates for important reactive gases, including glyoxal, epoxides
and peroxides, have been documented at different RHs (Loza et al., 2010;
Nguyen et al., 2016). While these processes may also influence reactive
SOA-forming compounds under our conditions, they occur at timescales of hours
(Nguyen et al., 2016), much longer compared to the timescales of the
absorptive uptake, e.g. based on recent direct measurements of vapour losses
onto Teflon walls (∼ 10 min, Krechmer et al., 2016).
The absorption of organic compounds into the chamber walls obeys Henry's law
and depends on the compounds' accommodation coefficients and their activity
at the wall–gas interface (see for example Zhang et al., 2015). The
dependence of the compounds' absorption on RH (due to a change in
accommodation coefficients or in the activity of the wall-absorbed compounds)
has not yet been reported to the best of our knowledge and indeed merits
further investigations that are beyond the scope of the current study.
Nevertheless, we believe that vapour absorption onto the walls is unlikely to
be significantly affected by RH, due to the hydrophobic nature of Teflon and
its minor interaction with water under subsaturation conditions (RH
< 80 %). In our case, we have maintained chamber conditions
during our experiments such that vapour wall losses and their
inter-experimental differences can be minimized as much as possible. This is
done by (1) maintaining a relatively constant wall-to-seed surface ratio for
all experiments to avoid systematic biases between experiments and
(2) increasing SOA production rates, which rapidly provide a significant
particle condensational sink into which condensable gases can partition. We
also note that vapour wall losses were found to be minor for the α-pinene SOA system, where SOA formation is dominated by quasi-equilibrium
growth (Zhang et al., 2014; Nah et al., 2016).
SOA yields determination and parameterization. SOA mass yields,
Y (dimensionless quantity), are calculated from Eq. (9), as the
organic mass concentration formed, COA, per precursor mass
consumed, ΔCα-pinene:
Y=COAΔCα-pinene.
For comparison purposes, Y values are reported in Fig. 2 and Table 1 and
discussed in Sect. 3 at an OH exposure of
(2.0 ± 0.5) × 107 cm-3 h, reached during all
experiments. The parameterization of smog chamber SOA yields measured is
based on the absorptive partitioning theory of Pankow (Eq. 10), which
expresses the production of a set of semi-volatile surrogate products (total
number N) as a function of the mass yield of these products,
αi, and their partitioning coefficients, ξi (a dimensionless quantity reflecting the condensed-phase mass fraction
of these products). The critical parameters driving the partitioning of these
products are their effective saturation concentration, Ci∗, and
the total concentration of the absorptive organic phase, COA. As
discussed below, we consider the absorptive organic mass as the sum of the
total OA concentration and the liquid water in this phase (Sect. 3.3).
Y=∑iNαiξi=∑iNαi1+Ci∗COA-1
In Eq. (10), Ci∗ (in µg m-3) is a semi-empirical
property (inverse of the Pankow-type partitioning coefficient,
KP,i) reflecting the saturation vapour pressure of the pure
constituents pL,io and the way they
interact with the organic mixture (effectively including liquid-phase
activity coefficients, γi), as expressed in Eq. (11):
Ci∗=106MiγipL,i0760RT.
Here, Mi denotes the compound molecular weight, R the ideal gas
constant and T the temperature. Smog chamber yields from single
experiments are fitted as a function of COA using the volatility
basis set (VBS) (Donahue et al., 2006), which separates semi-volatile
organics into logarithmically spaced bins of effective saturation
concentrations Ci∗. Figure 5 shows the resulting
parameterizations (lines) in comparison to the measured data (symbols) for
each experiment.
To determine the yields per volatility bin (αi),
wall-loss-corrected SOA yields, Y as a function of wall-loss-corrected
absorptive mass concentration COA (data presented in Fig. 5),
were used. We assumed a total number of five bins: N = 5 with
Ci∗= 0.01, 0.1, 1, 10 and 100 µg m-3. To
solve Eq. (10) for the parameters αi, we introduced a novel
approach using a Monte Carlo simulation. This approach provides best
estimates of αi values (data shown in Figs. 6 and S7), together
with a measure for the uncertainties related to the determination of the
volatility distributions from SC experiments. The calculation proceeded as
follows.
From the calculated yields (Y), lower (Y1) and
upper (Y2) yield curves were determined based on the estimated
measurement accuracy of COA and α-pinene mass. A possible
yield domain was inscribed against COA by plotting Y1 and
Y2 versus their corresponding lower and upper COA,
respectively. The parameterized yield curves are shown in Fig. 5.
A range of possible inputs was defined for each of the parameters
αi. This range is restricted within the following interval: [0;
2 × (Yi-Yi-1)]. This step was only necessary for
computational reasons.
αi parameters were randomly generated over the
defined intervals.
Deterministic computations of Y versus COA were performed using
Eq. (10) and the generated αi inputs.
Volatility distributions that fell within the domain defined in step (1) were retained, aggregated and presented as probability distribution functions
for each of the five effective saturation concentrations Ci∗
(probability density function, PDF, Figs. 6 and S7).
Thermodynamic modelling
General principles. Using thermodynamic modelling, we seek to
understand whether observed changes in SOA bulk properties (yields and
O : C ratios) with RH can be explained by a change in the particles'
thermodynamic properties. Gas–liquid and liquid–liquid phase partitioning
calculations are performed following the methods developed in Zuend et
al. (2008, 2010, 2012) and Zuend and Seinfeld (2013), using the thermodynamic
group-contribution model, AIOMFAC (Aerosol Inorganic-Organic Mixtures
Functional groups Activity Coefficients), to calculate activity coefficients.
This approach enables prediction of the phase partitioning of known organic
compounds knowing their abundances in a given known mixture of organic
species and electrolytes at a given RH and temperature.
The modelling requires the use of explicit surrogate compounds. Based on
Eq. (10), the partitioning of these compounds is driven by their volatility
distributions, which depends on (1) the compounds' effective saturation
concentrations and (2) their relative abundances. Compounds' effective
saturation concentrations, used as model inputs, were calculated based on
Eq. (11) initially assuming ideal mixing (γ=1) and utilizing vapour
pressures estimated using EVAPORATION (Compernolle et al., 2011). The
relative abundances of these compounds in the model mixtures are based on the
volatility distributions derived from experimental data (Sect. 2.5).
We assumed instantaneous reversible absorptive equilibrium of semi-volatile
organic species into ideal and non-ideal liquid-phase aerosols. In the case
of non-ideal solutions, positive and negative deviations of mole
fraction-based activity coefficients from unity indicate the degree of
non-ideality in a mixture. The activity coefficients take into account the
compounds' affinity towards the solution (interactions with other organic
species, electrolytes and water) and hence depend on the solution's chemical
composition. Therefore, the activity coefficients for the different organic
species cannot be set a priori, but are calculated iteratively in the
model,
until convergence of the compounds' abundances in the different phases; see
(Zuend et al., 2008, 2011).
We considered cases with and without interactions between the electrolyte and
organic phases, which enables one to assess the solubility of the organic
compounds in the inorganic seed aerosols. Seed concentrations in
µg m-3 were transformed into moles of seed per volume
(mol m-3) assuming equal shares of AHS and SA for acidic seeds and
equal shares of AHS and AS for neutral seeds. Because interaction parameters
of some organic functional groups with HSO4- are missing in AIOMFAC,
we assumed the interactions of organic compounds with HSO4- and with
SO42- to be similar. For all computations, metastable supersaturated
salt solutions were allowed.
Model compounds are only surrogates. The choice of these surrogates should
reflect the wide range of volatility and hydrophilicity of SOA species. We
have selected as surrogates α-pinene photo-oxidation products
reported in the literature, identified under different NOx and aerosol
seed conditions, covering the wide range of volatility relevant to SOA.
Compound hydrophilicity instead depends heavily on the number of functional
groups present in the molecule, which can be largely simulated by the
compound O : C ratio (Zuend and Seinfeld, 2012). As increasing the compound
O : C ratio also decreases its volatility, we have considered cases with
and without fragmentation products with a shorter carbon backbone chain but
high O : C ratios, expected to be representative of later-generation
photolysis and photo-oxidation products (Mutzel et al., 2015; Krapf et al.,
2016). This approach would effectively decouple compounds' hydrophilicity and
volatility and allow scanning of both properties independently. While we
recognize that oxidation conditions, e.g. NOx concentrations, may
significantly alter the product distribution, we did not select different
sets of products for the different conditions. This is because
such a separation would implicitly suggest that the chemical composition
of the few compounds reported at different conditions can be extrapolated to
the bulk OA under our conditions;
such a separation would significantly limit the number of surrogates at
each condition, increasing the sensitivity of the model to the compounds'
selection; and
the model is less sensitive to the compounds' chemical structure than to
their elemental composition (see below and Li et al., 2016), e.g. number of
oxygen and carbon, which has been taken into account by including
fragmentation products.
The relative abundance of the selected compounds is optimized in the model
for each experiment at the prevailing RH such that the modelled and measured
SOA yield and O : C ratio match. Then, the RH is modified in the model, and
changes in the SOA yields and O : C ratios are compared to the
observations. Model calculations were performed at an OH exposure of
(2.0 ± 0.5) × 107 cm-3 h. In the following we
thoroughly describe the different steps involved in the model setting.
Simulated cases. The following simulations were performed.
Case org: non-ideal partitioning, including liquid–liquid phase separation (LLPS; activity coefficients calculated with AIOMFAC), of the organic
compounds between the gas phase and a purely organic aerosol phase neglecting
the presence of the seed aerosol and using reported model compounds only;
Case id: ideal partitioning (activity coefficients all set to unity; LLPS
cannot occur) between gas phase and organic aerosol phase using reported
model compounds only;
Case sd: non-ideal partitioning including the seed aerosol to an
internally mixed organic/AS aerosol using reported model compounds only;
Case sdfr: non-ideal partitioning including the seed aerosol to an
internally mixed organic/AS aerosol including the formation of fragmented
oxidation products (see below);
Case orgfr: non-ideal partitioning including LLPS to a purely organic
aerosol phase neglecting the presence of the seed, including the formation of
fragmented oxidation products.
αi parameters determined through the Monte Carlo simulations assume
that the absorptive mass consists of the organic phase (in accordance with
assumptions in chemical transport models). This assumption is violated for
cases sd and sdfr as compounds may partition into the inorganic phase.
Nevertheless, the results obtained for these cases may still be examined in
relative terms to inspect the effect of RH on SOA yields in the presence of
an inorganic seed and the partitioning of SOA compounds between the inorganic
and organic phases.
The 20 min averaged wall-loss-corrected (symbols and lines) and
suspended (lines) organic mass concentrations as a function of OH exposure.
Data are separated according to similar initial α-pinene
concentrations (20 ppbv – top panel; 30 ppbv – bottom panel). Experiment
numbers are given in the legend and classified by seed composition; asterisks
indicate low NOx experiments.
Model compounds. To simulate SOA partitioning in AIOMFAC, α-pinene photo-oxidation products reported in the literature (Eddingsaas et
al., 2012; Jaoui and Kamens, 2001; Kleindienst et al., 2007; Valorso et al.,
2011) were chosen as model compounds for cases org, id and sd, namely
ValT4N10 (10th compound in Table 4 from Valorso et al., 2011),
3-hydroxyglutaric acid, pinic acid, hopinonic acid, norpinic acid,
2-hydroxyterpenylic acid, 10-oxopinonic acid, and 4-oxopinonic acid. These
compounds are listed in Table S4 together with their relevant physicochemical
properties (MW, O : C ratio, vapour pressures and chemical structures). For
cases sdfr and orgfr, additional oxidized fragmented products (3-oxoadipic
acid, glutaric acid, 5-COOH-3-OH-pentanal, and succinic acid) were included
(Table S4). While these compounds were not reported to derive from α-pinene oxidation, their structure, including carbon and oxygen numbers, is
very similar to the most abundant compounds detected by Chhabra et al. (2015)
and Mutzel et al. (2015) using chemical ionization mass spectrometry.
Volatility distributions could reliably be determined for volatility bins
C∗=0.1–100 µg m-3. Nevertheless, lower
volatility products with C∗ = 0.01 µg m-3 are
also formed. Therefore, this bin was loaded for all phase partitioning
calculations with equal fractions of the model compounds diaterpenylic acid
acetate, 3-MBTCA and ValT4N9 with mass yields of 10-5 each. This low
mass fraction does not influence the organic yield, but proved to aid the
convergence of the phase partitioning calculation. To achieve mass closure in
the model, pinonaldehyde (MW: 168 g mol-1) was assumed to represent
the more volatile products, which do not partition to the condensed phase.
Because pinonaldehyde resides almost totally in the gas phase, it was not
explicitly modelled.
Average wall-loss-corrected yields Y at
(2.0 ± 0.5) × 107 cm-3 h OH exposure as a function
of RH. Symbol sizes represent α-pinene reacted, symbol colours
represent NOx/α-pinene and symbol shapes represent seed
composition according to Table 1. Experiments with similar NOx/α-pinene, seed composition and α-pinene reacted are connected with
dashed lines showing the increase in yield for increased RH. Experiment 6 has
no counterpart experiment at high RH. Experiments with similar RH, α-pinene reacted and seed composition are connected with solid lines showing
the increased yield with a decreasing NOx/α-pinene ratio.
Model compound Cj∗ calculation. Model
compounds j were assigned to the volatility bins based on the calculated
Cj∗ values using Eq. (11) and assuming ideal mixing
(γj = 1). Using the actual activity coefficients calculated
with AIOMFAC is not possible because they are a result of the phase
partitioning calculation. However, because activity coefficients proved to be
generally in the range of 0.1 to 10, they did not alter the initial product
assignments to the volatility bins, which cover 1 order of magnitude in
effective saturation concentration. Therefore, the initial allocation of the
compounds to the volatility bins in the VBS remains valid after taking the
non-ideality into account. pL,jo used in the
calculations were vapour pressures of pure compounds in liquid or subcooled
liquid state, at 298 K estimated using EVAPORATION (estimation of vapour
pressure of organics, accounting for temperature, intramolecular, and
non-additivity effects, Compernolle et al., 2011), without using the
empirical correction term for functionalized diacids.
Setting model compounds' relative abundances in the model. From
the αi parameters generated through the Monte Carlo simulations in
Sect. 2.5, 10 sets per experiment were randomly selected for the phase
partitioning calculations (provided that these parameters fall within the
10th and 90th percentiles, to avoid outliers). The 11 chosen experiments
exclude experiments 15–17 with hydrophobic seeds, experiment 6 which has no
counterpart experiment at high RH, and experiments 5 and 14 where the COA
concentration range was very limited to accurately derive αi
parameters (Sect. 2.5). The chosen parameters are listed in Table S5. As
several model compounds are assigned to a volatility bin i, the yield of
a compound j is expressed as its relative contribution within the
volatility bin i, χj,i, times the relative abundance of the bin
αi. For each experiment and each simulated case the fitted
χj,i values are listed in Tables S6–S8. For case org the χj,i values were optimized to match the experimental organic yields and
the measured O : C at the actual RH of the experiment. RH was then changed
in the model and the effects of RH on SOA yields and degree of oxygenation
were evaluated. For cases id and sd the same χj,i were used as for
case org. For cases sdfr and orgfr, more fragmented compounds were added and
their χj,i values were optimized to achieve agreement between
measured and modelled organic mass yields and O : C ratios. Likewise, the
effects of RH on SOA yields and on O : C ratios were then evaluated by
changing the RH in the model.
Van Krevelen diagrams: mean (and 1 SD measurement variability)
H : C versus O : C at OH exposure
(2.0 ± 0.5) × 107 cm-3 h (symbols). Symbol colours
indicate RH, symbol shapes the seed composition and the asterisk low NOx
experiments. (a) The dashed line represents the (least orthogonal
distance) fitted slope of experiments (1–14) with inorganic seed: -0.84.
The triangular-shaped solid lines represent the range of ambient SOA (Ng et
al., 2011a); 1 h averages of H : C versus O : C (for OH exposure
> 2 × 106 cm-3 h) are given by the coloured
lines. Experiments 15–17 show the influence of organic seed compounds (data
shown from suspended organic mass > 0.3 µg m-3).
(b) The data were split into three groups according to their
wall-loss-corrected organic mass concentrations to exclude concentration
effects (left panel: 2–6 µg m-3; middle panel:
8–14 µg m-3; right panel: 16–30 µg m-3).
The grey shaded areas include all low NOx experiments.
The following pairs of corresponding experiments were simulated:
experiments 3 and 4 (66 % RH, 29 % RH), experiments 7 and 11
(67 % RH, 26 % RH), experiments 8 and 13 (60 % RH, 25 % RH),
and experiments 10 and 12 (50 % RH; 26 % RH). Equilibrium
calculations were performed starting from both, low and high RH experiments.
For experiments 5 and 14 performed at low RH values, the volatility
distribution parameters could not be determined for the most volatile bin
(α5). Therefore, equilibrium partitioning calculations were only
performed starting from their corresponding experiments, namely experiment 1
(69 % RH), experiment 2 (67 % RH), and experiment 9 (56 % RH).
Experimental results
Figure 1 shows the suspended and wall-loss-corrected organic mass
concentrations for all experiments listed in Table 1 as a function of OH
exposure. SOA mass is rapidly formed and the wall-loss-corrected mass reaches
a plateau at an approximate OH exposure of
2 × 107 cm-3 h. In the following, comparisons between
the different experiments are carried out at an OH exposure of
2 × 107 cm-3 h. For illustrative purposes, the
wall-loss-corrected aerosol yield is shown in Fig. S8, as a function of
α-pinene reacted to demonstrate its variability between different
experimental conditions.
SOA yield dependence on RH, NOx/α-pinene and aerosol seed
composition
In Fig. 2, we examine the relationship between the determined yields and the
prevailing experimental conditions: RH, NOx/α-pinene and seed
composition. Lines connect experiments conducted under comparable conditions,
but at different RH (dashed lines) and at different NOx/α-pinene
(solid lines). The visible effects of RH and NOx conditions on the yield
in Fig. 2 were statistically examined using a multilinear analysis (Fig. S9
and Table S3). Three yield parameterizations were inter-compared and only the
simplest model which represented significantly better the observations was
considered for discussion.
The following features can be deduced from the analysis.
SOA yields increase with α-pinene concentrations, consistent
with the semi-volatile nature of SOA compounds formed. We estimate that
yields increase by approximately 2 percentage points when
α-pinene concentrations increase from 20 to 30 ppbv.
A significant effect of the seed initial concentrations (or surface) on
SOA yields was not observed. This may suggest (1) that SOA most likely forms
its own phase and does not significantly partition into the seed aerosol (see
below) and (2) that SOA condensation is not significantly affected by the
vapour losses to the SC walls, which can be diminished by increasing the
aerosol surface. This is consistent with recent smog chamber results
suggesting that for the α-pinene system SOA formation is dominated by
quasi-equilibrium growth and vapour losses to the walls do not depend on the
seed concentrations, but rather on SOA (precursor) formation (oxidation)
rates (Nah et al., 2016).
SOA yields are significantly reduced under high NOx conditions (-3.3 % ± 0.6 %, p<0.001), in agreement with literature
data (Ng et al., 2007). Such a decrease indicates that SOA compounds formed
under low NOx conditions are less volatile than those formed under high
NOx conditions.
We observed a clear influence of the RH on the yields, which increase on
average by 1.5 ± 0.6 % per 10 % RH, for the range explored.
This indicates that the particulate water content plays a central role in the
partitioning of SOA compounds, either by altering the thermodynamic
properties of the bulk phase (e.g. increasing the absorptive mass or
decreasing the compound activity coefficients, non-reactive uptake) or by
providing a reactive sink for semi-volatile species (e.g. formation of lower
volatility compounds/oligomers in the bulk phase, reactive uptake).
Additionally, the multilinear analysis suggests that the magnitude of the RH
influence on SOA depends on the seed chemical nature (yields correlate with
the interaction term between RH and seed composition), with a greater
influence for the acidic seed (0.21 ± 0.03 % per 1 % RH, p<0.001), the most hygroscopic aerosol, compared to the non-acidic
seed (0.15 ± 0.03% per 1 % RH, p<0.001) and the
hydrophobic seed (0.09 ± 0.04 % per 1 % RH, p = 0.05).
Overall, these results highlight the sensitivity of SOA yields to the
prevailing oxidation conditions and the particle bulk-phase composition and
water content and therefore the need to consider such conditions to obtain an
accurate prediction of the SOA burden in the atmosphere. Nonetheless, results
from this analysis should be regarded with some level of caution, owing to
the limited size of the dataset. Despite the methodical assessment of the
significance of the multilinear analysis results, we cannot unambiguously
propose a mechanism by which water and seed hygroscopicity/acidity enhance
SOA yields or determine whether there is interplay between these two
parameters. Nevertheless, we provide evidence that both of these parameters
play a significant role in the formation or condensation of SOA species. In
Sect. 4, using phase partitioning computation and AIOMFAC, we shall assess to
what extent and under which conditions particulate water may alter the
organic species' activity coefficients and as a consequence their absorptive
partitioning.
SOA elemental composition
The effect of the experimental conditions on SOA chemical composition was
investigated using the Van Krevelen space (Fig. 3a). The overall region of
the experimental data is very comparable for all experimental conditions and
comparable to ambient data (Ng et al., 2011a). Data for experiments 1–14,
with inorganic seeds, follow a similar slope with aging (-0.84): an
increase in O : C during aerosol aging takes place under all conditions.
The O : C and H : C ratios of SOA produced with hydrophobic seed aerosol
(experiments 15–17) show lower values than with inorganic seeds.
Evolution of size distributions from the HR-ToF-AMS. Measured
organic, SO4, NH4, and NO3 mass distributions for OH exposures
of 0×, (0.5 ± 0.2)×, (1.0 ± 0.3)× and
(2.0 ± 0.5) × 107 cm-3 h. Black lines represent
estimated liquid water content (method: Sect. 2.3; RH and individual GFs
given in the legend, percentage in brackets: fractions of SO4). The
calculated dry and wet surface distributions are shown as dashed lines on the
right axes. (a) High NOx with more acidic seed [H2SO4
and NH4HSO4]: experiments 3 and 4 with NOx/α-pinene = 1.2 and experiment 2 with NOx/α-pinene = 3.8.
(b) Low NOx with less acidic seed [(NH4)2SO4 and
NH4HSO4]: experiments 8 and 13 with NOx/α-pinene ≈ 0.05. Additional figures in the Supplement (Fig. S12).
In Fig. 3b, data are separated according to the wall-loss-corrected organic
mass concentrations COA (left panel: 2–6 µg m-3;
middle panel: 8–14 µg m-3; right panel:
16–30 µg m-3), to isolate the NOx and RH effects on the
chemical composition from the possible influence of enhanced partitioning of
semi-volatile organic species to the aerosol phase due to a higher SOA
loading (Pfaffenberger et al., 2013). We observe that NOx levels have
the highest influence on SOA elemental composition; namely, SOA formed at low
NOx (marked with asterisks) is characterized by higher H : C than that
formed at high NOx. This is even more pronounced at an early reaction
stage (OH exposure < 2 × 106 cm-3 h, not shown
in Fig. 3), where fresh SOA products formed during all low NOx
experiments have higher H : C and lower O : C. Such an influence of
NOx levels on SOA elemental composition is consistent with our general
understanding of gas-phase chemistry: under low NOx, substantial amounts
of hydroperoxides and alcohols would result in a higher H : C ratio than
e.g. carbonyls formed under high NOx conditions (Atkinson, 2000). Also,
as hydroperoxides and alcohols are substantially less volatile than carbonyls
and less fragmentation is expected at lower NOx conditions, less
oxygenated species with higher carbon numbers may partition to the particle
phase at low NOx, which would lead to lower O : C ratios.
NOx levels also affect the amount of organonitrate formed. Assuming that
the entire nitrate signal arises from organonitrates (i.e. a maximum
organonitrate contribution), we estimate molar ratios (Fig. S10) of NO3
to carbon of ∼ 1:30 for high NOx and ∼ 1:100 for low
NOx. Assuming oxidation product molecules with 10 carbon atoms, these
ratios imply that every 3rd and 10th molecule contains one NO3
functional group, for high and low NOx conditions, respectively.
Conversely, we could not observe a significant effect of RH and seed
composition on SOA elemental composition and degree of oxygenation, despite
their significant influence on SOA yields. This may be due to the limited
range of the O : C ratio spanned by the different experiments in our case
(on average O:C∈0.56-0.75 at an OH exposure of
2 × 107 cm-3 h). Furthermore, we did not observe any
significant dependence between these parameters and the ratio of organic
fragments larger than m/z 150 to the total organic mass, a proxy for
oligomers measured with the HR-ToF-AMS, which would explain the observed
increase in yields with the aerosol liquid water content (Fig. S11).
Measured (symbols) and parameterized (lines) wall-loss-corrected SOA
yield, Y, as a function of wall-loss-corrected absorptive mass
concentration (COA: organics + NO3+ H2O) for low
NOx experiments (upper panel) and high NOx experiments (lower
panel). Data were limited to an OH exposure of
2 × 107 cm-3 h and averaged over 20 min. Symbol colours
indicate the RH, symbol shapes the seed composition and the asterisks the
NOx / VOC ratio.
Probability density functions (PDFs) of αis for
volatility bins (Ci∗= 0.01, 0.1, 1, 10 and
100 µg m-3) for one high RH (exp. 3, upper panel)
experiment and one low RH (exp. 4, lower panel).
Calculated average organic yields for experiments at the measured RH
(columns in full colours) and at the RH of the corresponding experiments
(light colours; see Table 2 for reference) based on 10 randomly chosen
volatility distributions for the cases org, id, sd, sdfr, and orgfr
(identified by different colours). The vertical black lines in the columns
indicate the range of values obtained for the calculations with the
individual volatility distributions. The horizontal dotted line marks the
measured value for the experiment performed at low RH, the dashed line the
value for the experiment performed at high RH.
Analysis of aerosol size-resolved chemical composition
As the seed composition and particulate water content appear to greatly
influence SOA yields, we examine in this section the interaction between
these parameters and SOA, through the investigation of the aerosol
size-resolved chemical composition. This information is used later to infer
the behaviour of the absorptive organic phase and its mixing with the
inorganic seed, while modelling the aerosol dynamics in the SC is beyond the
scope of this study. Figure 4 shows the aerosol size-resolved chemical
composition at 0 × 107, (0.5 ± 0.2) × 107,
(1.0 ± 0.3) × 107 and
(2.0 ± 0.5) × 107 cm-3 h OH exposure for five
experimental conditions (data for additional experiments are available in
Fig. S12). Figure S13 shows a 3-D representation of the time-dependent number
and volume size distributions, measured by the SMPS.
For all experiments, the aerosol size distributions show two externally mixed
aerosol populations, with a mode at lower diameters (∼ 200 nm, mode 1)
mostly containing SOA and another mode at higher diameters (∼ 400 nm,
mode 2) mostly consisting of the seed. We note that the particle size
distribution evolved consistently under different conditions, with the
smallest seed particles growing with SOA condensation or coagulation
(Figs. S12 and S13). While we have detected an increase in particle number
(Fig. S14), we note that intense nucleation events did not occur. For higher
yields, the main SOA mass occurs in mode 1 and despite the sizeable increase
in the yield with particulate water content (e.g. under acidic conditions and
high RH), we did not observe a significant enhancement of SOA in mode 2. In
addition, we did not note any statistically significant correlation between
the initial seed volume and SOA yields; instead SOA growth seems to be driven
by the favourable partitioning of semi-volatile species to smaller particles
at an early stage of the experiment (Fig. 4). Such behaviour would imply that
semi-volatile compounds do not additionally partition or react in the
electrolyte-rich phase on the timescale of this experiment, but rather the
reactive or non-reactive uptake of these products onto the particles is
enhanced with the increase in the initial particulate water content.
Calculated average O : C ratios for experiments at the measured RH
(columns in full colours) and
at the RH of the corresponding experiments (light colours; see Table 2 for
reference) based on 10 randomly chosen volatility distributions for the cases
org, id, sd, sdfr, and orgfr (identified by different colours). The vertical
black lines in the columns indicate the range of values obtained for the
calculations with the individual volatility distributions. The horizontal
dotted line marks the measured value for the experiment performed at low RH,
the dashed line the value for the experiment performed at high RH.
Prevalent oxidation reagent and its influence on SOA yields and chemical
composition
Based on the mixing ratios of OH and ozone, we estimate that a greater part
of α-pinene has reacted with OH (on average 0.78 ± 0.07).
Moreover, it is worthwhile mentioning that the further processing of the
first-generation products – which do not contain C=C bonds – would almost
exclusively proceed through OH oxidation. Accordingly, we conclude that SOA
compounds detected are mainly from OH chemistry, independent of the NOx
level and relative humidity.
We note that the fractions of α-pinene that reacted with OH under low
RH (0.79 ± 0.07) and high RH (0.77 ± 0.07) are not statistically
different (t test, p = 0.71), within our experimental variability.
Therefore, we do not expect that differences in SOA yields observed between
experiments at low and high RH will be due to a change in the prevalent
oxidant. We recognize that water vapour may change the oxidation product
distributions, via its reaction with the stabilized Criegee intermediates,
produced upon the ozonolysis of α-pinene. However, we note that not
only the fraction of α-pinene that reacts with O3 is not
substantial (in comparison with that reacted with OH), but a major fraction
of α-pinene Criegee intermediates also undergoes unimolecular
decomposition to form OH and does not react with water (Atkinson and Arey,
2003). Therefore, it is unlikely that a change in RH would sizably modify the
distribution of the products formed via gas-phase chemistry.
By contrast, the fraction of α-pinene that reacted with OH is found
to be sensitive to the NOx concentration. The production of O3 was
faster under high NOx (average [O3] = 35 ppbv) compared to low
NOx (average [O3] = 22 ppbv), due to an efficient VOC–NOx
catalytic cycle. Consequently, the fraction of α-pinene that reacted
with OH under high NOx (0.75 ± 0.06) is lower than that under low
NOx (0.83 ± 0.04). These small but statistically significant
differences (8 percentage points; t test, p = 0.004) in the
contribution of ozone/OH to α-pinene oxidation are expected to
explain a small part of the differences in SOA yields and chemical
composition observed at low and high NOx, with a higher fraction of
ozonolysis products under high NOx conditions. Despite this, we expect
the influence of NOx on the fate of RO2 to be the main driver of
the observed differences in SOA yields and chemical composition between low
and high NOx conditions (because ozonolysis products are only a minor
fraction and differences between the two conditions are rather small).
Phase partitioning calculation results
The organic yields and O : C ratios of the phase partitioning calculations
are presented in Figs. 7 and 8, respectively, and listed in Table S9. Each
panel in Fig. 7 compares organic yields for the cases org, id, sd, sdfr, and
orgfr at the actual RH of the experiment (full colour) and at the
lower/higher RH of the corresponding experiment (light colour). Table 2 gives
the increase in organic yields from low to high RH of the corresponding
experiments as ratios (org yield (high RH)/org yield (low RH)).
Increase in organic yield from low to high RH for the different
experiments as org yield (high RH)/org yield (low RH). The second column
lists the ratio of measured organic yield enhancements with RH. In the five
last columns the ratios of organic yields calculated at high and low RH for
the different experiments are given (average value of the 10 volatility
distributions with lowest and highest values in brackets in the second row).
Experiments
Calculations
Exp.
Yield
Exp.
Case org
Case id
Case sd
Case sdfr
Case orgfr
no.
ratio
1/5
6.38
1
1.62
2.02
1.54
4.23
4.34
(1.48–1.86)
(1.67–2.57)
(1.37–1.77)
(2.50–6.70)
(2.89–6.62)
2/5
4.10
2
1.42
1.61
1.31
2.74
2.24
(1.23–1.47)
(1.31–1.73)
(1.18–1.35)
(1.33–6.24)
(1.35–2.92)
3/4
3.23
3
1.64
2.07
1.57
3.67
3.13
(1.44–1.86)
(1.88–2.38)
(1.48–1.73)
(2.74–5.29)
(2.53–4.14)
3/4
3.23
4
1.76
2.09
1.58
2.93
3.12
(1.58–1.90)
(1.78–2.41)
(1.44–1.71)
(1.91–4.88)
(2.13–4.09)
7/11
2.95
7
1.35
1.52
1.28
2.16
2.38
(1.25–1.47)
(1.35–1.80)
(1.20–1.39)
(1.48–3.98)
(1.64–3.69)
8/13
2.32
8
1.43
1.61
1.36
2.66
2.44
(1.30-1.58)
(1.43-1.87)
(1.27-1.50)
(1.60-4.79)
(1.73-3.57)
9/14
4.63
9
1.43
1.56
1.33
2.45
2.69
(1.32–1.59)
(1.38–1.82)
(1.23–1.48)
(1.61–4.70)
(1.75–4.24)
10/12
1.46
10
1.29
1.38
1.28
1.44
1.55
(1.13–1.52)
(1.22–1.67)
(1.17–1.49)
(1.18–1.66)
(1.26–2.17)
7/11
2.95
11
1.80
2.22
1.67
2.74
3.38
(1.52–2.21)
(1.69–3.09)
(1.40–2.05)
(1.75–4.03)
(1.87–5.61)
10/12
1.46
12
1.34
1.48
1.33
1.34
1.70
(1.21–1.48)
(1.27–1.71)
(1.19–1.49)
(1.19–1.65)
(1.33–2.06)
8/13
2.32
13
1.54
1.74
1.41
2.73
2.54
(1.39–1.68)
(1.52–1.91)
(1.30–1.50)
(1.29–6.06)
(1.39–4.90)
The main objectives of the phase partitioning calculations are (1) estimating
the impact of liquid water on SOA mixing properties (activity coefficients
and liquid–liquid phase separation (LLPS)) and (2) determining the
conditions or the potential model mixtures that can explain the observed
yields and O : C ratios and their variation with RH. We note that model
results are highly sensitive to the surrogates assumed, and the determination
of SOA composition on a molecular level would considerably help in confirming
our results. Nevertheless, fitting both organic yields and O : C ratios
significantly aids in constraining the type of compounds that participate in
partitioning (i.e. from a compound O : C ratio and vapour pressure, its
carbon number can be inferred). For example, highly oxygenated compounds
cannot be very volatile without significant fragmentation, whereas
oligomerization leads to a significant decrease in the compounds' vapour
pressure without necessarily increasing their O : C ratios.
Simulations with reported α-pinene photo-oxidation products (cases
org, id, sd)
For case org, the contribution of model compounds (χj,i) to the
volatility bins at the actual RH could be optimized such that agreement was
achieved between measured and calculated yields with deviations of less than
10 %, constituting a proof of concept of the applied approach. For most
simulations, the χj,i values optimized for case org were also valid
for cases id and sd.
In general, simulations assuming cases org, id and sd failed in predicting
the change in SOA yields with RH (Fig. 7) and led to a significant
underestimation of the SOA O : C ratios (Fig. 8). The only exception is the
pair of experiments 10 and 12 performed at low NOx (1.9 ppbv) and high
α-pinene levels (30.4 ppbv) for which the model could account for
87–99 % of the increase in SOA yield from low to high RH. This agreement
was achieved although the model underestimated the O : C ratios by 24 %
and predicted a smaller increase in the O : C values with RH than observed.
The modelled O : C remained almost constant at 0.45–0.48 at low and high
RH (difference of 6 %), while the measured O : C increased from 0.56 at
low RH to 0.64 at high RH (change of 14 %). We also note that the use of
the more recent parameterization by Canagaratna et al. (2015) would yield
even higher O : C values, widening the gap between measured and modelled
O : C ratios.
For the other experiments performed under low NOx conditions, model
simulations accounted for only 43–75 % of the observed yield increase
with RH. Simulated O : C values ranged from 0.48 to 0.55 and failed to
reproduce the observed increase from 0.57 to 0.62 for experiments 8 and 13
and from 0.6 to 0.64 for experiments 7 and 11. Likewise, for high NOx
conditions, model simulations for experiments 1 and 5, 2 and 5 and 9 and 14
could only account for 24–39 % of the observed yield increase from low
to high RH and for 49–65 % for experiments 3 and 4. Simulated O : C
values remained almost constant at 0.49–0.54 for all high NOx
experiments and were considerably lower than the observed O : C ratios.
For case id the increased SOA yield at high RH, attributed to the additional
partitioning of semi-volatile compounds (norpinic acid, 2-hydroxyterpenylic
acid, 10-oxopinonic acid and 4-oxopinonic acid), is a direct consequence of
the increased absorptive mass due to the higher water content. For cases org
and sd, partitioning to the condensed phase is reduced compared to case id as
AIOMFAC predicts activity coefficients greater than 1 for higher-volatility
compounds, e.g. 10-oxopinonic acid and 4-oxopinonic acid. This effect is even
enhanced for case sd, when partitioning to the total condensed phase
including the seed aerosol is simulated. Although LLPS is predicted for all
simulations, the salting-out effect of AS which also partitions to some
degree to the organic phase leads to a further decrease in the organic yield
at high RH compared to case id.
In summary our results suggest that with the reported model compounds for
α-pinene photo-oxidation, the measured O : C and the increase in
organic yields with RH cannot be simulated. Therefore, we explored whether
the formation of fragmented and more oxidized products may explain the high
O : C ratio observed and the high sensitivity of the yields to RH.
Equilibrium phase partitioning (mol m-3) for case sdfr between
gas phase, electrolyte phase, and organic phase at low RH (25 %) and at
high RH (60 %) for experiment 8, volatility distribution 11 (upper
panels) and experiment 13, volatility distribution 45 (lower panels).
Compound numbers are c1: diaterpenylic acid acetate, c2: 3-MBTCA, c3:
ValT4N9, c4: ValT4N10, c5: 3-hydroxyglutaric acid, c6: ValT4N3, c7:
3-oxoadipic acid, c8: pinic acid, c9: hopinonic acid, c10: glutaric acid,
c11: norpinic acid, c12: 2-hydroxyterpenylic acid, c13: 5-COOH-3-OH-pentanal,
c14: succinic acid, c15: 10-oxopinonic acid, c16: 4-oxopinonic acid, ammonium
sulfate (AS), water (w), and the sum of the compounds c1–c16.
Equilibrium phase partitioning (mol m-3) for case sdfr between
gas phase, particle phase, electrolyte phase, and organic-rich phase at low
(29 %) and high (66 %) RH for experiment 4, volatility distribution 2
(upper panels), and volatility distribution 93 (lower panels). Compound
numbers are c1: diaterpenylic acid acetate, c2: 3-MBTCA, c3: ValT4N9, c4:
ValT4N10, c5: 3-hydroxyglutaric acid, c6: ValT4N3, c7: 3-oxoadipic acid, c8:
pinic acid, c9: hopinonic acid, c10: glutaric acid, c11: norpinic acid, c12:
2-hydroxyterpenylic acid, c13: 5-COOH-3-OH-pentanal, c14: succinic acid, c15:
10-oxopinonic acid, c16: 4-oxopinonic acid, ammonium sulfate (AS), water (w),
and the sum of the compounds c1–c16.
Simulations including fragmented products (cases sdfr and orgfr)
Non-fragmented products (e.g. highly oxygenated C10 or dimers) would be
low-volatility (LVOC) or extremely low-volatility organic compounds (ELVOC)
(Zuend and Seinfeld, 2012; Donahue et al., 2006, 2012b; Tröstl et al., 2016) with effective
saturation concentrations C∗≤ 0.1 µg m-3, when
their O : C ratio is as high as the observed O : C ratio. Therefore, they
are expected to be in the particle phase independent of the prevalent RH and
the inclusion in the model of additional amounts of these products would not
help explain the observed difference in the yields at different RHs, leading
to an overestimation of OA mass at low RH. This implies that to better
capture the measured O : C range without overestimating SOA yields,
fragmented products with shorter backbones need to be introduced.
Low molecular weight compounds resulting from fragmentation were therefore
added to the volatility bins with
C∗ = 1–100 µg m-3, namely, 3-oxoadipic acid to
the volatility bin with C∗ = 1 µg m-3, glutaric
acid to the volatility bin with C∗ = 10 µg m-3
and 5-COOH-3-OH-pentanal and succinic acid to
C∗ = 100 µg m-3. For case sdfr, αi
parameters were optimized assuming equilibrium partitioning of SOA to the
whole condensed phase, including the seed aerosol. For case orgfr absorption
to the organic phase only was assumed. While the organic and inorganic phases
seem to form externally mixed particles based on the chemically resolved size
distribution, examining both cases provides valuable insights into the impact
of the presence of an inorganic aerosol seed in the system. Tables S7 and S8
of the Supplement provide the contribution of model compounds (χj,i
values) to volatility bins Ci∗ for these cases. The organic
yields and O : C ratios for cases sdfr and orgfr are shown in Figs. 7 and
8, respectively. In Table 2 the relative increase in SOA yields from low to
high RH is given. Cases sdfr and orgfr are similarly successful in
reproducing SOA yields and O : C ratios. However, the variability in the
modelled yields when using different volatility distributions for cases sdfr
and orgfr is larger than for cases org, id, and sd (Fig. 7).
For experiments carried out at low NOx conditions (experiments 7 and 11,
8 and 13 with NOx < 2 ppbv), both cases explained the
observed increase in the organic yields with RH and improved the agreement
between modelled and observed O : C ratios. For experiments 8 and 13, both
cases accurately captured the increase in SOA yields from low to high RH with
deviations < 2 %. Modelled O : C agreed very well with
measurements for case sdfr with deviations ≤ 1 % and slightly worse
for case orgfr with deviations of ≤ 5 %. For experiments 7 and 11,
cases sdfr and orgfr reproduced satisfactorily (deviations ≤ 11 %)
the measured yields for calculations with the volatility distributions
determined for experiment 11, but overestimated the observed increase in
O : C from low (O : C = 0.60) to high RH (O : C = 0.64). Using
the volatility distributions determined for experiment 7, the low yields at
low RH could not be achieved. For experiments 10 and 12, cases org, id, and
sd were able to simulate the increase in organic yield with the oxidation
products reported in the literature, but failed to reproduce the observed
O : C ratios. Cases sdfr and orgfr with additional oxidized and fragmented
products showed improved agreement of observed yields and reproduced
successfully the observed O : C ratios.
Under high NOx conditions, cases sdfr and orgfr could only simulate
satisfactorily SOA yields and O : C ratios observed during experiments 3
and 4. For experiments 1–2 and 5, and 9 and 14, model and measurement
agreement was less satisfactory. The model could not reproduce the low yields
at low RH observed during experiments 5 and 14, and hence the strong
reduction of the yields observed with the decrease in RH is underestimated
(the simulations predict only 52–66 % of the observed change).
Additionally, for these experiments the model underestimates the O : C
ratios (average measured O : C = 0.66 and 0.68, at low and high RH,
respectively). In the following, we examine and discuss the potential reasons
that might explain model–measurement disagreements.
Simulations including organonitrates
We examined whether the discrepancy between modelled and measured yields may
be ascribable to the selection of the model compounds at high NOx, by
introducing additional surrogates – specifically organonitrates – expected
to be representative of the compounds formed under these conditions. Assuming
that the NO3 signal in the HR-ToF-AMS originates from organonitrates,
every third molecule should contain one ONO2 functional group at high
NOx conditions. We tested whether adding organonitrates described by
Valorso et al. (2011) would improve the agreement between measurements and
observations for the high NOx experiments. However, this was not the
case (not shown). Such sensitivity tests could not be performed for cases
including partitioning to the seed aerosol because interaction parameters
between organonitrates and sulfate are not available in AIOMFAC.
Simulations including higher volatility oxidation products
The Monte Carlo simulations enabled the determination of αi
parameters for volatility bins
C∗ = 0.1–100 µg m-3. αi parameters in
the volatility bin with C∗ = 1000 µg m-3 could
not be reliably extracted. Nevertheless, SOA products belonging to this bin
are present in the chamber and could partition to some extent to the
condensed phase. We investigated whether the presence of substances with an
effective saturation concentration of 1000 µg m-3 could
reproduce the low SOA yields at low RH and its strong increase at high RH
observed for the high NOx experiments 1 and 5, 2 and 5, and 9 and 14.
This was achieved by replacing a part of the substances in the volatility bin
C∗ = 100 µg m-3 with pinalic acid, terpenylic
acid and 3-2-oxopropanylolxypropanoic acid (Table S4). While the addition of
these compounds in the model could reproduce the low yields at low RH, the
yields at high RH were strongly underestimated (not shown). Accordingly, the
introduction of organonitrates or higher-volatility compounds in the model
did not improve the agreement between modelled and observed yields and
O : C ratios (not shown here), suggesting that equilibrium partitioning
alone cannot explain the strong SOA yield increase under high NOx
conditions.
Partitioning of individual components into the gas and condensed
phases
The phase partitioning calculations do not only allow the simulation of the
total organic yield and average O : C ratio of the condensed products, but
also provide insights into the partitioning of individual compounds to the
gas and condensed phases. Phase partitioning of individual model compounds
into the gas and condensed phases for case sdfr is examined in Fig. 9 for
experiments 8 and 13 and in Fig. 10 for experiments 3 and 4. Detailed results
of the phase partitioning calculations are listed in Tables S10 and S11. For
experiments 8 and 13, the model predicts an LLPS into an organic-rich phase
(op) and a predominantly electrolyte-like phase (ep). Overall, organic
compounds are predominantly in the organic phase at both RH, with ep/op
< 1 × 10-5 at 25 % RH, and ep/op ≈ 0.003–0.008 at 60 % RH. Compounds in volatility bins
C∗ = 0.01 and 0.1 µg m-3 are mainly present in
the condensed phases, while compounds in volatility bin
C∗ = 100 µg m-3 show preferred partitioning to
the gas phase. The strongest increase in the condensed phase when RH is
increased from 25 to 60 % is observed for the model compounds assigned to
C∗ = 100 µg m-3. Moderately oxygenated species
(4-oxopinonic acid and 10-oxopinonic acid) in this bin show a moderate
increase (about a factor of 2) driven by the increase in the absorptive mass.
This increased partitioning is limited by an increase in the activity
coefficients of these compounds (for experiment 8 from 1.69 and 1.63 at low
RH to 2.70 and 2.93 at high RH and for experiment 13 from 1.49 and 1.54 at
low RH to 2.88 and 3.24 at high RH). Conversely, the 5-fold enhanced
partitioning of the fragmented and more functionalized compounds
(5-COOH-3-OH-pentanal and succinic acid) into the condensed phase at high RH
is driven by the increase in the absorptive mass and the slight decrease in
the compounds' activity coefficients (for experiment 8 from 0.84 and 0.51 at
low RH to 0.68 and 0.43 at high RH and for experiment 13 from 0.90 and 0.56
at low RH to 0.69 and 0.44 at high RH).
For experiments 3 and 4, model results obtained from the parameterization of
the yields in experiment 4 are highly sensitive to the assumed volatility
distribution. As shown in Fig. 10, when introducing volatility distributions
characterized by high contributions of semi-volatile oxygenated compounds
pertaining to the volatility bin C∗ = 100 µg m-3,
the model predicts a liquid-phase mixing. Note that when mixing is predicted,
the model tends to overestimate the O : C ratio (O : C = 0.81 at high
RH for volatility distribution 2 shown in the upper panels of Fig. 10) and
the yields at both high and low RH. Conversely, when volatility distributions
with less volatile and less oxygenated compounds are used (e.g. volatility
distribution 93), lower yields and lower O : C ratios (0.69 for volatility
distribution 93 shown in the lower panels of Fig. 10) are obtained and LLPS
is predicted. For such O : C ratios LLPS has also been observed for
experiments performed with model mixtures (e.g. Song et al., 2012). For
experiment 4, the volume of the electrolyte phase is larger than that of the
organic-rich phase and there is considerable partitioning of organic
compounds to the electrolyte phase (ep/op = 2.26). These observations
illustrate that for compounds with high O : C ratios, small differences in
the volatility distribution parameters can lead to totally different phase
partitioning. Irrespective of these differences, none of the volatility
distributions used could reproduce the measured yields at both high and low
RH, likely due to the deficient representation in the model of the
interactions between the acidic aerosol and the organic compounds – and of
the chemical processes occurring under acidic/high NOx conditions.
Modelling considerations and limitations
Model simulations were carried out by considering two α-pinene
C8–C10 photo-oxidation products per volatility bin for cases org, id and sd.
This number was sufficient to represent the species within one volatility
bin, because the hydrophilicity of these compounds with a consistently high
carbon number correlates with their volatility. Therefore the sensitivity of
model predictions to the contribution of these compounds to the volatility
bins (described by χj,i values) is relatively low. This model
setting, corresponding to a pseudo-one-dimensional VBS where the compounds'
vapour pressure and degree of oxidation correlate, was proven to be
insufficient for an accurate description of both SOA mass and degree of
oxygenation. By contrast, for cases sdfr and orgfr, which include more
oxidized short chain products, hydrophilicity and volatility may be varied
more independently, which introduces an additional degree of freedom and
would correspond to a pseudo-two-dimensional VBS (2D-VBS), as described by
Donahue et al. (2011). While we show that model predictions based on this
setting are highly dependent on the volatility distribution parameters and
model compounds assumed, in general the introduction of fragmented more
oxidized compounds reproduced well the high observed O : C ratios and the
increase in O : C and SOA yields with RH, specifically at low NOx. The
estimated volatility distributions and average carbon (C∼ 6) and oxygen
(O∼ 4) numbers when considering fragmented products are in agreement
with chemical speciation analysis previously reported for the same system
(Chhabra et al., 2015). The analysis shows that for such semi-volatile
products with O : C ratios of the order of ∼ 0.6, an increase in RH
from 23–29 to 60–69 % induces a mass increase by up to a factor of 3,
driven by the higher particle water content and the lower activity
coefficients of the more fragmented products at high RH.
The measured O : C ratio is a key parameter for constraining the model.
Here, we have used the high-resolution parameterization proposed by Aiken et
al. (2007), while the use of the more recent parameterization by Canagaratna
et al. (2015) would result in even higher O : C ratios (by 18 %).
Higher O : C ratios would require increasing even further the contribution
or the degree of oxidation of the fragmented compounds and would imply that
the model predicts an even higher sensitivity of the yields to RH. Therefore,
the O : C values used here yield more conservative estimates of the
contributions (or the degree of oxidation) of fragmented products in the
model and of the sensitivity of the yields and O : C ratios to the RH.
Under high NOx (and low pH), the model could not reproduce the factor of
6 increase in yields at high RH, using a variety of chemically dissimilar
surrogate compounds. This indicates that the increased absorptive uptake of
these compounds due to the increase in SOA mass and the decrease in the
compounds' activity coefficients cannot explain the observed enhancements
alone. Under these conditions, additional processes may likely play a role in
the enhancement of OA with RH. NOx concentrations have a dramatic
influence on SOA chemical composition and volatility. Therefore, the
discrepancy between model and measured yields at high NOx conditions may
be explained by either an inadequate representation of SOA surrogates in the
model and their interaction parameters with the seed or by an enhanced
reactive uptake of SOA species formed under high NOx conditions, such as
carbonyls that are reported to instigate the formation of lower-volatility
compounds in the particle phase (Shiraiwa et al., 2013, and references
therein). Given the dearth of additional chemically resolved measurements of
the particle-phase species formed under different conditions, the mechanism
by which RH enhances the uptake of SOA species under high NOx (low pH)
remains currently undetermined.
Cases org, id, and orgfr assumed gas-particle partitioning into the organic
aerosol only. In this case, the role of the seed aerosol is restricted to
providing a substrate for nucleation of organic vapours. This is the case for
effloresced seed particles. For liquid seed aerosols this is equivalent to a
complete organic/electrolyte phase separation with no partitioning of
inorganic ions to the organic phase. For cases sd and sdfr equilibrium
partitioning between the gas phase and the entire condensed phase including
the seed aerosol is assumed, leading in most cases to LLPS. The current model
does not yet contain interaction parameters of bisulfate with all involved
organic functional groups. Therefore, ammonium bisulfate was treated as
ammonium sulfate in the model, and we are not capable of distinguishing
whether the enhanced partitioning of semi-volatile vapours in the acidic
medium is attributable to additional reactions in the bulk phase (catalysed
at lower pH) or to an enhanced solubility of SOA species.
Considering the size-resolved particle chemical composition discussed in
Sect. 3.3 (Fig. 4), LLPS is likely not realized within single particles, but
the aerosol population splits up into a predominantly organic mode at
∼ 200 nm and a predominantly inorganic mode at ∼ 400 nm. The
formation of these two populations may occur by the homogeneous or
heterogeneous nucleation of highly oxidized non-volatile products.
Homogeneous nucleation implies new particle formation (which would only be
moderate – see Figs. S13 and S14 – due to the high condensation sink in the
chamber), while heterogeneous nucleation proceeds via condensational growth
(which would occur on smaller particles with a higher surface). Both
processes are expected to create small organic-rich particles, providing an
organic absorptive phase into which additional semi-volatile compounds may
preferentially partition. When the organic and electrolyte phases are present
in different particles, the two phases communicate via gas-phase diffusion,
and equilibration timescales depend on the components' volatility. For
compounds with C∗ = 0.1–100 µg m-3,
equilibration occurs within timescales of minutes to tens of minutes,
assuming no bulk-phase diffusion limitations (Marcolli et al., 2004). In the
larger particle electrolyte-rich mode, the inorganic ions would exert a
salting-out effect driving the organic compounds to partition to the gas
phase or into the smaller organic-rich particles. This would prevent the
organic compounds from partitioning in significant amounts into the seed
aerosol from the beginning and depleting even further these larger particles
from the organic material. Under such a scenario an externally mixed
phase-separated aerosol may evolve in the smog chamber. Overall, the
size-resolved chemical composition information confirms the modelling
results, providing compelling evidence for organic/electrolyte LLPS.
The scenario outlined above is based on equilibrium partitioning and does not
invoke diffusion limitations within the condensed phase. Recent evidence may
challenge this assumption, suggesting that SOA may adopt a highly viscous
state (e.g. Virtanen et al., 2010; Koop et al., 2011), where bulk diffusion
and evaporation are kinetically limited. However, while such behaviour occurs
under certain conditions, e.g. low temperature and low relative humidity, we
expect that this is not the case for the aerosol investigated in this study.
Saleh et al. (2013) showed that SOA from α-pinene ozonolysis reaches
equilibrium with the gas phase within tens of minutes at low mass loadings
(2–12 µg m-3) upon a step change in temperature. Robinson et
al. (2013) determined by aerosol mixing experiments that the diffusion
coefficient in α-pinene-derived SOA is high enough for mixing on a
timescale of minutes. Fast mixing is further supported by measurements of
ambient OA (Yatavelli et al., 2014) showing that biogenic SOA reaches
equilibrium within atmospheric timescales, under similar conditions to those
in our chamber. Therefore, we expect that, if thermodynamically favourable,
liquid–liquid mixing would have occurred under the timescales of our
experiments and a unimodal particle population would have emerged. However,
consistent with model predictions, this is not the case.