Oxidation flow reactors (OFRs) allow the concentration of a given atmospheric oxidant to be increased beyond ambient levels in order to study secondary organic aerosol (SOA) formation and aging over varying periods of equivalent aging by that oxidant. Previous studies have used these reactors to determine the bulk OA mass and chemical evolution. To our knowledge, no OFR study has focused on the interpretation of the evolving aerosol size distributions. In this study, we use size-distribution measurements of the OFR and an aerosol microphysics model to learn about size-dependent processes in the OFR. Specifically, we use OFR exposures between 0.09 and 0.9 equivalent days of OH aging from the 2011 BEACHON-RoMBAS and GoAmazon2014/5 field campaigns. We use simulations in the TOMAS (TwO-Moment Aerosol Sectional) microphysics box model to constrain the following parameters in the OFR: (1) the rate constant of gas-phase functionalization reactions of organic compounds with OH, (2) the rate constant of gas-phase fragmentation reactions of organic compounds with OH, (3) the reactive uptake coefficient for heterogeneous fragmentation reactions with OH, (4) the nucleation rate constants for three different nucleation schemes, and (5) an effective accommodation coefficient that accounts for possible particle diffusion limitations of particles larger than 60 nm in diameter.
We find the best model-to-measurement agreement when the accommodation
coefficient of the larger particles (
Aerosols impact the climate directly, through absorbing and scattering
incoming solar radiation (Charlson et al., 1992), and indirectly, through
modifying cloud properties (Rosenfeld et al., 2008; Clement et al., 2009).
Both of these effects are size-dependent, with larger particles dominating
both effects. Particles with diameters (
A large fraction of submicron aerosol mass is composed of organic aerosol
(OA) (Murphy et al., 2006; Zhang et al., 2007; Jimenez et al., 2009;
Shrivistava et al., 2017). OA is composed of thousands of often-unidentified
compounds (Goldstein and Galbally, 2007) and can be emitted directly in the
particle phase as primary OA (POA) or formed as secondary OA (SOA) through
gas-to-particle conversion. In SOA formation through the gas phase,
atmospheric oxidants (mainly OH,
Controlled studies of SOA formation have traditionally used large reaction
chambers with residence times of hours (often referred to as “smog
chambers”). Chambers are susceptible to loss of both gases and particles to
the walls of the chambers (e.g., Krechmer et al., 2016; Bian et al., 2017).
In order to enable the study of SOA formation from ambient air and limit wall
losses, oxidation flow reactors (OFRs, i.e., the potential aerosol mass (PAM)
reactor; Kang et al., 2007; Lambe et al., 2011a) were developed to produce
high and controllable oxidant concentrations and have short residence times
(usually
Nucleation, i.e., the formation of new
Condensation of vapors to newly formed aerosol particles as well as
pre-existing particles increases the total aerosol particle mass, but the net
condensation rate to differently sized particles is dependent upon the
volatility of the vapors. The lowest-volatility vapors condense essentially
irreversibly onto particles of all sizes (i.e., “kinetically limited” or
irreversible condensation; Riipinen et al., 2011; Zhang et al., 2012).
Semi-volatile vapors (with non-trivial partitioning fractions in both the
particle and gas phases at equilibrium) have a net condensation to particles
that is determined by reversible partitioning (i.e., quasi-equilibrium
condensation; Riipinen et al., 2011; Zhang et al., 2012). Kinetically limited
condensation is gas-phase-diffusion limited and only possible for compounds
with effective saturation concentrations (
The gas-phase oxidation rates of organic vapors as well as the competition
between gas-phase functionalization (the addition of polar, oxygen-containing
functional groups, generally lowering the volatility of the species) and
gas-phase fragmentation (the cleavage of C–C bonds, with each reaction
typically creating two higher-volatility products) influence the changes in
volatilities of organic species from atmospheric oxidation (e.g., Kroll et
al., 2009). Gas-phase oxidation rates have been well quantified for many
individual species in the lab (e.g., Atkinson and Arey, 2003a), but less is
known about gas-phase oxidation rates that may be appropriate for lumped
organic vapors in ambient air. Generally, a representative reaction rate
constant (
Particle-phase reactions also shape OA mass and the size distribution.
Heterogeneous reactions between OH and organics at the surface of the
particle can yield fragmentation products with high-enough volatilities to
evaporate from the particle (e.g., Kroll et al., 2009), resulting in particle
mass loss. Heterogeneous reactions contribute to aerosol aging and influence
aerosol lifetime (George and Abbatt, 2010; George et al., 2015; Kroll et al.,
2015). Many laboratory studies have reported uptake coefficients of OH,
SOA uptake rates may be limited by the phase state of SOA through particle
diffusion limitations. Traditionally, SOA was viewed as a liquid mixture;
however, SOA have been observed in solid and amorphous phases in both
laboratory and field studies (Virtanen et al., 2010, 2011). Measurements
taken in 2013 and during the GoAmazon2014/5 campaign (Martin et al., 2016,
2017) found that SOA produced from oxidation products from the Amazonian
rainforest tended to be primarily liquid whereas SOA influenced by
anthropogenic emissions (both from the Manaus pollution plume and biomass
burning) tended to have higher fractions of semisolid and solid aerosol
(Bateman et al., 2015, 2017). Mixing in these solid or amorphous phases
could decrease (Cappa et al., 2011; Vaden et al., 2011), leading to
decreases in gas-particle partitioning rates (Shiraiwa and Seinfeld, 2012).
The impacts of the changes in phase state from liquid to solid/amorphous
matters less for SOA uptake at smaller particle sizes (
Each of the processes discussed above (nucleation, condensation of vapors, gas-phase functionalization and fragmentation reactions, heterogeneous reactions, accretion reactions, acid–base reactions, and particle diffusion limitations) could have very different timescales in the OFR as compared to the ambient atmosphere; for example, the chemistry timescale will typically be much shorter than the condensation and coagulation timescales in the OFR since the OFR OH concentrations can greatly exceed that of the ambient OH concentrations. Thus, models must be used to help interpret the OFR processes to determine how the observations relate to the ambient atmosphere. In this study, we use OFR measurements taken from two field locations. In the first, an OFR was deployed during the BEACHON-RoMBAS field campaign (Ortega et al., 2014) that took place in a montane ponderosa pine forest in Colorado, USA, during July–August 2011. The second is the GoAmazon2014/5 field campaign (Martin et al., 2016, 2017) that occurred from January 2014 to December 2015 in the state of Amazonia, Brazil, in the central Amazon Basin. OFR data from each of these two campaigns have been analyzed in previous work (Palm et al., 2016, 2017, 2018; Hunter et al., 2017) to understand the bulk OA mass and chemical evolution in the OFR. These analyses showed that the presence of unspeciated S/IVOCs contributes substantial OA mass production in the OFR at both locations. However, previous work has not analyzed the evolving aerosol size distribution in the OFR to gain insight into nucleation and growth processes. In this paper, we extend the analysis of these ambient datasets using the measured aerosol size distributions and a model of aerosol microphysics in the OFR.
The aerosol measurements investigated in this work were of ambient air before
and after oxidation in a PAM reactor, which is a type of OFR (Kang, 2007;
Lambe, 2011a). This OFR is a cylindrical aluminum tube with a volume of 13 L and a typical residence time of 2–4 min. OH
radicals were produced inside the OFR by photolysis of ambient
The BEACHON-RoMBAS field campaign (referred to as BEACHON hereafter) took
place in July–August 2011 at the Manitou Experimental Forest Observatory
near Woodland Park, Colorado (Ortega et al., 2014). The sampling site,
located in a ponderosa pine forest in a mountain valley, was influenced
mainly by 2-methyl-3-buten-2-ol (MBO) during the day and monoterpenes (MT)
at night. During BEACHON, an OFR was used to measure the amount and
properties of SOA formed from the oxidation of real ambient SOA precursor
gases and ambient aerosol. Ambient particles and SOA formation after OH
oxidation in the OFR (and also
The GoAmazon2014/5 field campaign (referred to as GoAmazon hereafter) took
place in the area surrounding Manaus, Brazil, in central Amazonia (Martin et
al., 2016, 2017), investigating the complex interactions between urban,
biomass burning, and biogenic emissions. OFR measurements of SOA formation
from OH oxidation of ambient air (and also
In this study, we use the TwO-Moment Aerosol Sectional (TOMAS) microphysics
zero-dimensional (box) model (Adams and Seinfeld, 2002; Pierce and Adams,
2009b; Pierce et al., 2011) combined with the Volatility Basis Set (VBS;
Donahue et al., 2006) as described in Bian et al. (2017). This version of
TOMAS-VBS simulates condensation, coagulation, and nucleation, and it has a
simple organic vapor aging scheme that moves an organic species down in
volatility upon reaction with an OH molecule (Bian et al., 2017). The
simulated aerosol species are sulfate, organics, and water within 40
logarithmically spaced size sections from 1.5 nm to 10
In this study, gas-phase functionalization is modeled by assuming that the
organic compounds within the VBS bins react with OH and products from this
reaction drop by one volatility bin (a factor of 100 drop in volatility). As
a base assumption of the rate constants of our vapors in the VBS bins
reacting with OH (
We account for gas-phase fragmentation reactions separately by allowing one
OH reaction with a molecule in the lowest volatility bin (
We further account for monoterpenes (MT) oxidation by OH for both campaigns
and isoprene oxidation by OH for GoAmazon in the model. Palm et al. (2016)
determined that on average during the BEACHON campaign, MT contributed
20 % of the measured SOA formation, with sesquiterpenes (SQT), isoprene,
and toluene contributing an additional 3 % of the measured SOA
formation. Since these other VOCs contributed a minor amount to the
measured SOA formation, they were not included in this analysis. S/IVOCs at
BEACHON contributed the remaining 77 % towards the measured SOA formation,
and were likely the main source for new particles in the OFR. It was
observed that for the GoAmazon campaign during the dry season, the
approximate average contribution to the measured SOA was 4 % from isoprene
and 4 % from MT, with an 8 % remaining contribution towards the measured
SOA coming from SQT, benzene, toluene, xylenes, and trimethylbenzene (TMB),
combined. Thus, less of the total SOA can be described by the VOCs included
in the model (isoprene and MT) for the GoAmazon simulations than can be
described for the BEACHON campaign. The remaining 83 % of measured SOA
formation was found to have come from unmeasured S/IVOCs, so again S/IVOCs
were likely the main source for new particles in the OFR. Including the
other VOCs would only increase the model-predicted SOA yield from the
initial VOCs by a few tenths of a
Product fractional mass yields for lumped monoterpenes and isoprene
(GoAmazon only) in each VBS bin in TOMAS. The monoterpene yields are based
on Henry et al. (2012), with the yield for the
The products of both MTs and isoprene oxidation enter the model's volatility
bins in the vapor phase. For MT SOA production, we use the product yields for
We simulate heterogeneous fragmentation reactions of particle-phase organics
in all VBS bins by OH. The resulting particle mass loss is modeled in TOMAS
through
All BEACHON-RoMBAS and GoAmazon2014/5 model inputs (assumed values for missing data points in bold). Each value
represents the ambient condition present at the beginning of each modeled
exposure. The OH concentration is calculated by assuming that 1 day of aging
is equal to a 24 h average atmospheric OH concentration of
In this work, we explore three different possible nucleation schemes. The
first two use a
We further explore the possibility of a sulfuric-acid only nucleation
scheme, as some nucleation schemes used in models only rely upon the
concentration of sulfuric acid (e.g., Spracklen et al., 2008, 2010;
Westervelt et al., 2014; Merikanto et al., 2016) by using an activation
nucleation scheme (Kulmala et al., 2006) for our third nucleation scheme,
referred to here as ACT, in which existing clusters are activated:
We include a simple approximation of potential vapor-uptake and/or particle
diffusion limitations by setting an adjustable accommodation coefficient
(
In this study, we do not simulate acid–base reactions and accretion reactions. No gas-phase bases (ammonia or amines) were measured during either campaign, making modeling acid–base reactions in TOMAS too unconstrained. Further, the model simulations point towards high concentrations of ELVOCs in the gas phase needed to facilitate nucleation (Sect. 3.1), indicating that gas-phase ELVOC production may be the dominant ELVOC-formation pathway over particle-phase ELVOC production (through accretion reactions and/or acid–base reactions). However, we cannot rule out ELVOC production in the particle phase through particle-phase reactions, as ELVOCs are in the particle phase at equilibrium.
We simulate loss of low-volatility vapors to the OFR walls using a
first-order rate constant,
For the BEACHON simulations, we use the residence time distribution (RTD) in the OFR of Palm et al. (2017) assuming non-Brownian motion (their Fig. S1). The RTD is less-well characterized for GoAmazon; we use the RTD for particles from Lambe et al. (2011a), but as discussed in Palm et al. (2018), the RTD from Lambe et al. (2011a) is likely more skewed than for the OFR used at GoAmazon, due to the larger inlet at GoAmazon. The SMPS data for both campaigns were corrected for diffusion losses to the walls of the sampling lines (Palm et al., 2016, 2018).
We simulate coagulation using the Brownian kernel in Seinfeld and Pandis (2006). However, we do not expect coagulation to be a dominant process in our OFR simulations. The condensation sink timescale for the measured size distributions were on the order of 0.5–5 min, which corresponds to coagulation sink timescales on the order of 1–10 min for 1 nm particles, 2.5–25 min for 2 nm particles, and 5–50 min for 3 nm particles (Dal Maso et al., 2002). Thus, in some cases the coagulation sink timescales for the freshly nucleated particles were similar to the residence time. However, in most cases, freshly nucleated particles grew to at least 20 nm within the OFR, so the nucleated particles spend only a small fraction (< 10 %) of the residence time at sizes smaller than 3 nm. Hence, the coagulation timescale of the growing particles is overall much longer than the residence time, and we expect on the order of 10 % or fewer of the nucleated particles to be lost by coagulation in these OFR experiments.
Inputs to TOMAS to initialize each OFR exposure simulated from the BEACHON
and GoAmazon field campaigns are given in Table 2; each input represents the
initial condition present at the start of the exposure. The initial ambient
size distribution from each campaign's SMPS is also used (Figs. 1 and S1 in
the Supplement, black lines). The initial S/IVOC concentration (as measured
by the TD-EIMS) is evenly divided between the
Data availability during BEACHON and GoAmazon caused data gaps that overlap some of the exposures modeled. For these cases with missing measurement data, we assume concentrations; assumed values are listed in bold in Table 2. Each assumed value is derived from either determining the trend from the nearest-available timepoints (for short data gaps) or by determining the concentration from different days with similar ambient conditions (for large data gaps).
All parameter value ranges for the suite of sensitivity simulations run in TOMAS.
In order to understand the evolution of the size distributions of the OFR exposures from the BEACHON and GoAmazon field campaigns, we use TOMAS to explore the parameter spaces of five uncertain parameters. These parameters are (1) the rate constant of gas-phase functionalization reactions with OH, (2) the rate constant of gas-phase ELVOC fragmentation reactions with OH, (3) the reactive uptake coefficient for heterogeneous fragmentation reactions with OH, (4) the nucleation rate constant for three different nucleation schemes, and (5) an effective accommodation coefficient that accounts for possible particle diffusion limitations of aerosol particles larger than 60 nm in diameter. Table 3 lists each uncertain parameter, the assumed base value, and the parameter space that we search through for each parameter (the “Multipliers” column).
As discussed in Sect. 2.3.1, we use as the base rate of
In the model, we treat fragmentation reactions separately from the
functionalization reactions. As discussed above, we select
As previously discussed, for the reactive uptake coefficient
For our primary nucleation scheme, NUC1, (Eq. 4), we use a base nucleation
rate constant value of
To account for possible particle-phase diffusion limitations, the effective accommodation coefficient is set to vary between 0.01 and 1 for particles larger than 60 nm in diameter (Table 3).
We simulate every combination of the uncertain parameters described above. In total, we run 10 125 sensitivity simulations for each BEACHON and GoAmazon OFR exposure for the first nucleation scheme (NUC1), going through each permutation for each of the five different uncertain parameters explored in this work. We further run 10 125 sensitivity simulations for both NUC2 and ACT for each experimental exposure. We acknowledge that there are further uncertainties in the measurements and modeling assumptions, including (1) potential but not modeled reactive uptake growth mechanisms, (2) uncertainties in the reported OFR OH concentration, (3) isoprene chemistry that may affect NPF, (4) whether some products from gas-phase functionalization reactions decrease more or less in volatility per reaction than the assumed factor of 100 drop in volatility, and likely other factors. However, exploring these uncertainties is outside of the scope of this paper (and some of these are not entirely orthogonal to the uncertain factors explored here) and will be left to a future study.
Figure 1 shows the measured initial and final SMPS volume size distributions for each exposure examined in this study from the BEACHON field campaign. We simulate these eight exposures between eq. ages 0.090 and 0.91 days in the TOMAS model for each combination of parameters (Table 3), initializing each run with the ambient conditions recorded at the time of each exposure (Table 2). Each modeled exposure was taken during the nighttime, when MTs were the dominant VOC. We limit this study to exposures less than 1 eq. day of aging in order to avoid the complications of modeling the different parameters in Sect. 2.3.3 across several orders of magnitude of OH, and since this is the range of exposures where NPF is most obvious experimentally.
In order to further test the validity of our results, we apply the TOMAS
model version developed to simulate OFR exposures from the BEACHON field
campaign to OFR exposures taken between 31 August and 4 September 2014 during
the dry season of the GoAmazon field campaign. Figure S1 shows the initial
and final SMPS volume size distributions for each exposure examined in this
study from the GoAmazon field campaign. We simulate each of these exposures
for the same combination of parameters as used for the BEACHON simulations,
initializing each run with the ambient conditions at the corresponding times
(Table 2). However, unlike the BEACHON simulations, we include isoprene as a
source of SOA in the model, with VBS yields given in Table 1. Again, like
BEACHON, each modeled exposure was taken during the nighttime and is limited
to exposures less than 1 eq. day of aging. During IOP2, it was observed that
isoprene would peak during the day around 15:00–16:00 local time and MT
would peak later, around 18:00 local time (Liu et al., 2016; Martin et al.,
2016). Isoprene was primarily depleted through oxidation reactions by
nighttime, but MT had a background level that remained approximately constant
between midnight and noon (local times) when the concentrations would begin
to rise again (Fig. S2). We model fewer exposures for GoAmazon than BEACHON
(four vs. eight) as few of the GoAmazon OFR exposures during this time period
showed significant SOA growth on top of the already-high ambient SOA
concentrations as compared to BEACHON. Also, many of the OFR exposures were
either between 0.4 and 0.5 eq. days or
BEACHON-RoMBAS initial (i.e., ambient air, black line) and final (i.e., after OFR processing, blue line) SMPS-derived volume distributions for each individual exposure modeled in this study. The differences in SOA production between exposures of similar ages are due to the fact that the exposures were taken from different times during the campaign and thus different precursor concentrations were present (Table 2).
Bulk S/IVOCs were not measured during the GoAmazon campaign and instead we use the model to estimate the S/IVOC concentrations required to explain the aerosol particle growth. We use as base values of S/IVOC concentrations the average S/IVOC : MT ratio from the BEACHON campaign, 1.4, as MT data are available during GoAmazon, and use the model to determine which S/IVOC concentrations are needed to help explain observed growth. This analysis is described in Sect. 3.2.
Example model
In order to determine the goodness-of-fit of each model simulation to the
observed size distribution from the SMPS, we compute the normalized mean
error (NME) statistic of the first four moments of the size distribution for
each model simulation:
Example case of a 0.23 eq. day aging exposure from the
BEACHON-RoMBAS campaign. The panels represent the moments used to calculate
the normalized mean error (NME), with
To determine the contribution to aerosol formation and growth for the OFR exposures studied here from the input VOCs vs. the input S/IVOCs, we compare the predicted change in the OFR in total aerosol particle number and volume between simulations with S/IVOCs to simulations with no S/IVOCs. We do this comparison for the six best-fitting simulations with S/IVOCs for each exposure and calculate the mean volume changes for these six simulations with and without S/IVOCs. With these number and volume changes, we calculate the fractional contribution of S/IVOCs to aerosol particle volume production in the OFR. We use the same technique to determine the contribution of isoprene to aerosol formation and growth for the GoAmazon OFR exposures studied here using the same methods.
Representation of the parameter space for the average across the
0.09–0.9 day eq. aging exposures from BEACHON-RoMBAS examined in this study
for the NUC1 nucleation scheme and base value of the reactive uptake
coefficient of 0.6. The effective accommodation coefficient increases across
each row of panels; the rate constant of gas-phase fragmentation increases up
each column of panels. Within each panel, the rate constant of gas-phase
reactions with OH increases along the
Figure 4 represents the averaged NME summed across the eight
0.09–0.9 eq. day aging exposures modeled from the BEACHON field campaign,
for the NUC1
For the parameter combinations of
Figures 2b and 3 show an example of the final volatility distribution and
size distributions for the best-fit case for an exposure of 0.23 eq. days,
corresponding to the model parameters of
Figures S3, S5, S7, S9, S11, S13, S15, and S17 show the same analysis as
presented in Fig. 4 for each individual exposure modeled for the base value
of
Figure S19 shows the same analysis as Fig. 4, but for the NUC2 nucleation
scheme. It is qualitatively quite similar to NUC1 but with the wells of
averaged best-fit regions shifted and expanded slightly for some cases. Since
we do not have measurements to further constrain the system, we acknowledge
that we cannot definitively select NUC1 or NUC2 as being the better
nucleation parameterization and instead note that both nucleation schemes
appear to provide physically meaningful results and require further study. In
contrast, Fig. 5 shows the same analyses of Fig. 4 but for the ACT nucleation
scheme (Eq. 5). Figure 5 shows that there are regions of moderate NME values
between 0.45 and 0.5 for
Further, as the best fits in the model come from the
Representation of the parameter space for the average across the
0.09–0.9 day eq. aging exposures from BEACHON-RoMBAS examined in this study
for the ACT nucleation scheme and base value of the reactive uptake
coefficient of 0.6. The effective accommodation coefficient increases across
each row of panels; the rate constant of gas-phase fragmentation increases up
each column of panels. Within each panel, the rate constant of gas-phase
reactions with OH increases along the
It is of note that in general, the simulations using
The BEACHON simulations show very little sensitivity towards the reactive
uptake coefficient (
As discussed in Sect. 2.3.1, the first term of Eq. (1) relies on log10(
Palm et al. (2016) compared the total SOA formed in the OFR during the
BEACHON campaign to the predicted yield from the measured VOCs for OH
oxidation in the OFR. For the analysis, they included the measured MT,
sesquiterpene (SQT), isoprene, and toluene concentrations and used
low-
To determine the contribution towards the change in total number and volume, we compare the changes in total volume between the averaged change in total volume for the six cases with the lowest (best) NME values of the original model runs for the NUC1 nucleation scheme to the same six cases (matching parameters) but with the initial S/IVOC concentration set to zero (See Sect. 2.5 for calculation details). Table 4 summarizes the fractional contribution of the measured initial S/IVOCs (Table 2) towards the total change in number and volume. The model predicts that the S/IVOCs contribute on average 85 % towards the total new number formed in the OFR, indicating a strong dependence on S/IVOCs for new particle formation in the OFR at BEACHON. The contribution of S/IVOCs towards the total change in volume is lowest for the lowest exposures, and increases with increasing eq. age of each exposure. This is primarily due to the increasing equivalent timescales of the increasing OH exposures: within our model it takes more reactions with OH for S/IVOC species to reach the lowest volatility bins than the MT and isoprene species. Thus with increasing timescales (or eq. ages), the contribution of S/IVOCs towards SOA formation and growth will increase as a higher fraction of these species reach the lowest volatility bins; the results in Table 4 corroborate this. However, given that the chemical evolution of S/IVOC is probably more complex than is represented here, we do not know if this result of S/IVOCs contributing a lower fraction of volume for low exposures is a robust conclusion. Overall, we predict that the average fractional contribution of the initial ambient S/IVOCs towards the change in total volume is 39 % for the BEACHON exposures, and that the initial ambient MT contributes the remaining 61 % towards the change in total volume. Palm et al. (2016) and Hunter et al. (2017) estimated from two independent analyses that S/IVOCs contributed on average 77 %–78 % towards the total mass SOA formation during BEACHON. It is likely that part of the difference between our model findings and Palm et al. (2016)'s findings is due to the difference in number of samples examined between the two studies as well as differences in the length of exposures analyzed, since Palm et al. (2016) included multi-day exposures in their analysis. It is important to note that running the model with the initial S/IVOCs set to zero (“S/IVOCs off”) does not perfectly inform us of the theoretical SOA yield of the MT concentration because the overall particle-phase yield of MTs products decreases with S/IVOCs off due to less mass to partition to.
Modeled fractional contribution of initial S/IVOCs towards the total change in number and volume between the initial and final number volume size distributions of each exposure modeled in this study. We use the measured S/IVOCs for the BEACHON-RoMBAS calculations and the best-fit initial S/IVOC concentration found for the GoAmazon calculations. The remaining fractional contribution towards the total change in number and volume is attributable to the measured initial monoterpenes (both campaigns) and measured initial isoprene (GoAmazon). Each exposure's fractional contribution is calculated using the averaged contributions of the six model cases with the lowest (best) NME values from the full model parameter space.
In order to model GoAmazon size distributions with TOMAS, we assumed an
initial S/IVOC concentration, as no instrumentation was present during the
campaign to measure total S/IVOC mass. For a starting total S/IVOC
concentration, we used the same measured ratio of S/IVOCs to MTs from BEACHON
of 1.4 (Table 2). This initial S/IVOC concentration was not sufficient to
explain the observed change in aerosol volume, nucleation, and new-particle
growth in the OFR for GoAmazon (see Figs. S21–S22 for an example). We found
that the initial S/IVOC concentration needed to be increased by between 20
and
Representation of the parameter space for the average across the
0.3–0.6 day eq. aging exposures from GoAmazon examined in this study for the
NUC1 nucleation scheme, base value of the reactive uptake coefficient of 0.6,
and assumed S/IVOC : MT ratio of
Figure 6 represents the averaged NME across the four 0.3–0.6 eq. day aging
exposures modeled from the GoAmazon field campaign for the NUC1
Previous ambient observations of the Amazon rainforests have not observed
nucleation at the surface (e.g., Spracklen et al., 2006; Martin et al., 2010;
Kanawade et al., 2011). Reasons could include low sulfuric acid (Kanawade et
al., 2011), high condensation sinks resulting from a strong source of primary
biogenic aerosols during the dry season (Lee et al., 2016), and possible yet
currently unexplained suppression mechanisms from isoprene and its oxidation
products (Lee et al., 2016), the dominant biogenic VOC of the region
(Guenther et al., 2012). Wang et al. (2016) found high concentrations of
small particles in the lower free troposphere during the wet season of
GoAmazon; however, they found that these particles appeared to be from NPF
and subsequent condensational and coagulational growth from the outflow
regions of deep convective systems, such as those common to the Amazonian
rainforest during the wet season. These particles could then be transported
to the boundary layer through vertical transport. By contrast, in some of the
OFR-oxidized air during the GoAmazon campaign the size distributions show
clear evidence of NPF and growth (e.g., Fig. S1) and the TOMAS model
simulations corroborate the observed NPF (Figs. S28, S30, S32, and S34), even
at the initial S/IVOC inputs (Fig. S24). The OFR shifts the relative
timescales of chemistry vs. condensation, which may create higher
concentrations of low-volatility vapors capable of participating in
nucleation and early growth relative to the ambient atmosphere during
GoAmazon. The lowest NME values (best fits) from the averaged BEACHON modeled
exposures (Fig. 4) for the highest two
Tests of NUC2 and ACT show similar changes from NUC1 for GoAmazon to BEACHON
(Figs. S25 and S26). NUC2 results were qualitatively similar to NUC1, and we
cannot determine which scheme performed better. The regions of lowest NME
values (best fits) shifted for the ACT scheme relative to the NUC1 and NUC2
schemes, and generally the NMEs are not quite as low as for NUC1 and NUC2,
although better fits are found for the ACT nucleation scheme for GoAmazon
than BEACHON. Thus it would seem that either a
Similar to BEACHON, more good fits for each nucleation scheme occur at lower
values of
Unlike the BEACHON campaign, bulk S/IVOCs were not measured directly during
the GoAmazon campaign. However, Palm et al. (2018) applied a similar analysis
to that of Palm et al. (2016) to determine the measured vs. predicted SOA
yield. They found that on average, OH oxidation of ambient air during the
GoAmazon campaign (dry season) produced
Modeled fractional contribution of the measured initial isoprene concentrations towards the total change in number and volume between the initial and final number and volume size distributions modeled from the GoAmazon2014/5 campaign, using the best-fit S/IVOC estimate. (Isoprene was not included in the model for the BEACHON-RoMBAS distributions.) The remaining fractional contribution is attributable to the MT and S/IVOC concentrations in the model. Each exposure's fractional contribution is calculated using the averaged contributions of the six model cases with the lowest (best) NME values from the full model parameter space.
To determine the contribution of MT and isoprene towards the change in total
number and volume for the GoAmazon exposures, we repeat the analysis done for
the BEACHON exposures (Sect. 3.1.2) and the results are summarized in
Table 4, using the S/IVOC concentrations of
The TOMAS box model does not include isoprene-specific oxidation pathways and
instead allows it to oxidize in the VBS scheme along with the other lumped
oxidized species. We determine the fractional contribution of the initial
isoprene concentration towards the change in total number volume for each
exposure modeled (Table 5); the remaining fraction is the total volume change
attributable from initial MT and the optimized initial S/IVOC concentrations
(
In this study, aerosol size distributions between 0.09 and 0.9 days of eq.
aging formed under OH oxidation in an OFR during the BEACHON-RoMBAS (BEACHON)
and GoAmazon2014/5 (GoAmazon) field campaigns were modeled in the TOMAS box
model in order to better understand the microphysical processes that shape
the size distribution under oxidative aging. We explored the following
parameter spaces to find regions of best-fit model-to-measurement agreements:
(1) nucleation rate constants for two
In order to limit the scope of this study, several uncertain processes and
values were not included in this analysis. We did not include the formation
of low-volatility organics through particle-phase acid–base reactions or
accretion reactions, as (1) no measurements of gas-phase bases were made at
either campaign and (2) the model results indicate the importance of the
gas-phase ELVOC creation pathway must be fast in order to drive nucleation,
which may limit the importance of particle-phase pathways. We did not
consider the model sensitivity towards the input OH concentration, although
there is uncertainty associated with the estimated OH exposure (Palm et al.,
2016, 2018). We further did not explore the model sensitivity towards the
assumed decrease of a factor of 100 in volatility for each product from OH
functionalization reactions, nor did we explore the sensitivity of including
fragmentation reactions for volatility bins higher than the ELVOC bin. These
two uncertainties are not entirely orthogonal to the uncertainties in
We found that we could not explain the observed size-distribution shift without slowing the uptake of SOA to the accumulation-mode particles. With an accommodation coefficient of 1 assumed for the full size distribution, these larger particles underwent too much condensational growth relative to the nucleation mode for all test cases. We speculate that this slowed uptake of larger particles may be indicative of particle-phase diffusion limitations. We approximate vapor-uptake limitations by allowing the accommodation coefficient of particles larger than 60 nm in diameter to vary between 0 and 1. We found that we can achieve the best fits of the size distribution when the accommodation coefficient of these larger particles was 0.1 or lower (if we similarly lowered the accommodation coefficient of smaller particles, we would not have gotten good fits as the new particles did not grow enough). Whether this is representative of ambient aerosol processes or just representative of conditions within the OFR is the subject of a future study.
We found that gas-phase fragmentation reactions also had a significant impact
upon the modeled size distributions. Our best-fit gas-phase fragmentation
rate constants were higher than that of a previous mass-based study of OFR
exposures from BEACHON (Palm et al., 2016) required to model the
distributions. However, these higher rates may be because our model only
simulated fragmentation reactions of the lowest-volatility compounds, that of
In general, the
This study has shown the potential for using OFRs to study factors that
control NPF and size-distribution evolution using ambient-air mixtures. The
fact that coagulation plays a small role in the measured number concentration
indicates that this type of reactor is useful to evaluate model
parameterizations of the number of nucleated particles and their growth, as a
function of ambient and OFR conditions. Using an OFR greatly expands the
parameter space over which comparisons can be made as well as the number of
cases that can be studied, compared to using only ambient data where
parameter variations are more narrow, and where NPF is not observed under
many conditions. Future studies could use OFRs in nucleation studies to both
better understand the dependencies of nucleation on input species (e.g.,
The data used from the BEACHON-RoMBAS campaign in the
publication are available at:
The supplement related to this article is available online at:
JLH, JRP, BBP, and ALH defined the scientific questions and scope of this work. ALH performed all model simulations with help from QB and JRP. BBP, PCJ, ESC, DAD, AH, JFH, WJ, TK, JHK, and ZP carried out the primary measurements and data processing for the BEACHON-RoMBAS field campaign. SSdS, BBP, PCJ, DAD, MLA, ABG, JHP, RS, JNS, and SK carried out the primary measurements and data processing for the GoAmazon2014/5 campaign as well as campaign supervision and design. ALH prepared the paper with substantial contributions from JRP, BBP and JLJ, with additional contributions from all other coauthors.
The authors declare that they have no conflict of interest.
This research was supported by the US Department of Energy's Atmospheric
System Research, an Office of Science, Office of Biological and Environmental
Research program, under grant no. DE-SC0011780, by the U.S National Oceanic
and Atmospheric Administration, an Office of Science, Office of Atmospheric
Chemistry, Carbon Cycle, and Climate Program, under cooperative agreement
award no. NA17OAR430001, and by no. NA17OAR4310002 and the U.S. National
Science Foundation, Atmospheric Chemistry program, under grant no.
AGS-1559607 and AGS-1558966. The CU-Boulder group was supported by US DOE
(BER/ASR) DE-SC0016559, and US EPA STAR 83587701-0. This manuscript has not
been formally reviewed by EPA. The views expressed in this document are
solely those of the authors and do not necessarily reflect those of the
Agency. EPA does not endorse any products or commercial services mentioned in
this publication. Institutional support was provided by the Central Office of
the Large Scale Biosphere Atmosphere Experiment in Amazonia (LBA), the
National Institute of Amazonian Research (INPA), and Amazonas State
University (UEA). We acknowledge support from the Atmospheric Radiation
Measurement (ARM) Climate Research Facility, a user facility of the United
States Department of Energy (DOE), Office of Science, sponsored by the Office
of Biological and Environmental Research, and support from the Atmospheric
System Research (ASR) program of that office. Additional funding was provided
by the Amazonas State Research Foundation (FAPEAM), the São Paulo State
Research Foundation (FAPESP), the USA National Science Foundation (NSF), and
the Brazilian Scientific Mobility Program (CsF/CAPES). The TD-EIMS
measurements were supported by NOAA grant NA10OAR4310106 (MIT). The research
was conducted under scientific license 001030/2012-4 of the Brazilian
National Council for Scientific and Technological Development (CNPq).
PTR-TOF-MS measurements were supported by the Austrian Science Fund (FWF)
project no. L518-N20. We are grateful to Andrew Turnipseed for