Uncovering the dominant contribution of intermediate volatility compounds in secondary organic aerosol formation from biomass-burning emissions

ABSTRACT Organic vapors from biomass burning are a major source of secondary organic aerosols (SOAs). Previous smog chamber studies found that the SOA contributors in biomass-burning emissions are mainly volatile organic compounds (VOCs). While intermediate volatility organic compounds (IVOCs) are efficient SOA precursors and contribute a considerable fraction of biomass-burning emissions, their contribution to SOA formation has not been directly observed. Here, by deploying a newly-developed oxidation flow reactor to study SOA formation from wood burning, we find that IVOCs can contribute ∼70% of the formed SOA, i.e. >2 times more than VOCs. This previously missing SOA fraction is interpreted to be due to the high wall losses of semi-volatile oxidation products of IVOCs in smog chambers. The finding in this study reveals that SOA production from biomass burning is much higher than previously thought, and highlights the urgent need for more research on the IVOCs from biomass burning and potentially other emission sources.


Supplementary Text Text S1. Experimental details
As shown in Fig. S1, the biomass-burning emissions were introduced into the holding tank or smog chamber through heated lines (150 ℃) with the heated ejection diluter (180 ℃).The ejection diluter was operated with an input pressure of ~1.5 bar, resulting in a sheath flow of ~20 L min -1 and a central flow of ~2 L min -1 .The injecting time to the holding tank was ~10−20 min, while it was ~1−5 min to the smog chamber.During the ejection, CO, CO 2 , and THC analyzers were connected to the holding tank or smog chamber to monitor the gas concentrations.After 5−10 min of measurement, the concentrations stabilized and these gas analyzers were disconnected.
The volume of the laminar-flow oxidation reactor (LFOR) is 16 L and the total flow was set to 6 L min -1 , resulting in an average residence time of 160 s, which is the same as a previous study [4].The relative humidity (RH) in the LFOR was controlled at ~50%.The flow from the holding tank to the LFOR was controlled to ~0.1 L min -1 , realizing a diluting factor of 60.The ozone concentration in the LFOR was maintained at ~4 ppm by using a constant small flow (0.04 L min -1 ) through an ozone generator (UV pro).The sampling flow was ~2 L min -1 taking from the centre of the LFOR while the rest (~4 L min -1 ) was exhausted to minimize the wall effects.The OH radicals were produced by the photolysis of O 3 to O( 1 D) with Hg lamps (Jelight Inc.), followed by the reaction with water vapor: O( 1 D) + H 2 O → 2OH.The OH exposure (i.e., photochemical age) in the LFOR was adjusted by changing the number and input voltage of the Hg lamps [5,6].The OH exposure was determined by the decay of benzene and toluene measured with PTR-MS (see Text S2).For high-NO x experiments, percent-level N 2 O was introduced into the LFOR to generate NO: O( 1 D) + N 2 O → 2NO [5,7].The RO 2 fate was modelled with OFR Exposures Estimator v3.1 [8].By using different N 2 O concentrations (1%−10%) at various OH exposures, the RO 2 was estimated to react ~90% to NO, with the rest (~10%) mostly react with HO 2 , which are typical high-NO x conditions.
The smog chamber (SC) was located in a temperature-controlled house to maintain a constant temperature (20 ℃).A set of 40 × 100 W UV lights (Cleo Performance, Philips) was used for irradiation [9].The RH was initially set to ~50% when the lights were off.HONO was used as the OH source and was injected by passing a small flow of zero air through the mixture of H 2 SO 4 (0.1 mmol L -1 ) and NaNO 2 (0.2 mmol L -1 ) solution.The d9-butanol was introduced into the chamber for the OH exposure estimation [10].After the lights were turned on, the RH decreased to about ~40% due to the slight increase in temperature.The lights were on for about 3−5 h, resulting in a maximum photochemical age of 1.5−2.5 days with a global average OH concentration of 1.5 × 10 6 molecules cm -3 [11].The aerosol concentration was wall-loss corrected.Aerosol wall loss rate constant (k w,a ) in the chamber was determined using the decay of particle mass concentration at the beginning and end of each experiment when lights were off.The wall loss corrected aerosol mass concentration (Aerosol wlc ) was calculated as: where Aerosol meas is the measured aerosol mass concentration.Vapor wall losses were not corrected because of the following two reasons.First, the wall loss rates of unreacted NMOGs are very low according to our previous studies [2,12], while the correction of intermediate gas-phase products is highly challenging.Second, almost all the SOA yield data in previous studies (see Text S2 below) used to calculate the SOA production only consider particle wall loss.To get a reasonable SOA closure, vapor wall losses were not counted.

Text S2. SOA and NMOG measurements and calculations
Organic aerosol (OA) mass concentration was calculated by multiplying particle volume concentration (measured with the SMPS) by the particle density (1.6−1.7 g cm -3 ), which was calculated from the vacuum aerodynamic diameter (obtained from the AMS) and the electric mobility diameter (obtained from the SMPS).The primary OA (POA) mass concentrations in the LFOR and SC were 15−45 μg m -3 and 10−35 μg m -3 , respectively.By performing control experiments that remove the NMOGs with denuder, we found that the POA mass concentration change little during oxidation (< 5%).Therefore, the SOA mass concentration was calculated as the difference between total OA mass concentration and the POA mass concentration.
A novel instrument for real-time measurement of organic aerosol composition, extractive electrospray ionization time-offlight mass spectrometer (EESI-TOF-MS) was deployed for the biomass-burning SOA in this study.EESI can measure aerosol composition without thermal decomposition or ionization-induced fragmentation.The details of the instrument is discussed elsewhere [13] and only briefly overviewed here.Air from LFOR or smog chamber was sampled at 1 L min -1 , either directly or through a particle filter to get a background measurement.The flow passed through a multichannel denuder, removing gas-phase species, and then intersected a spray of charged droplets generated by a conventional electrospray capillary.Particles collide with the electrospray droplets and the soluble components are extracted, ionized by Coulomb explosion of the charged droplets, and detected by TOF-MS (mass resolution ~10000).The electrospray working fluid is 100 ppm sodium iodide (NaI) water solution.Spectra are recorded in positive ion mode, in nearly all cases as adducts with Na + .
SOA production from precursors was mainly estimated using the NMOGs measured with PTR-MS.Although PTR-MS cannot distinguish different isomers and can only provide molecular formulas without identifying the molecular structure, previous studies have identified lots of NMOG structures [12].As shown in Table S2, we identified 18 NMOGs with reported SOA yields.For other NMOGs with carbon number (n C ) ≥ 6 (with or without an identified structure), the average of applied yields for n C ≥ 6 compounds (0.32) was used as their SOA yield.The SOA mass concentration produced from each NMOG was calculated by multiplying the reacted concentration and the SOA yield.The reacted NMOG concentration was calculated by their reaction rate constants with OH (k OH ) and the estimated OH exposure.For NMOGs without literature k OH or a molecular structure, the k OH was calculated from molecular formulas using a parameterization method [14].The volatilities of individual NMOGs with reported structures were taken from literature [15], while those with unknown structure were estimated based upon their molecular formulas using a parametrization described previously [16].
To quantify the NMOGs measured with PTR-MS and Vocus-PTR, the reaction rate constant of each species with the reagent ion (H 3 O + ) and a transmission function are needed to convert signal intensity to concentration.When available, individual reaction rate constants from the literature were applied [17] and a default reaction rate constant of 2×10 -9 cm 3 s - 1 was applied to all other ions.The transmission function was determined using 12 NMOGs in a gas standard (acetaldehyde, methanol, acetonitrile, acetone, acrylonitrile, isoprene, methyl ethyl ketone, benzene, toluene, m-xylene, α-pinene, 1,2,4trimethylbenzene).

Text S3. Correction between LFOR and smog chamber SOA yields
As the SOA yields from LFOR experiments are largely limited and there are more available SOA yields data from smog chambers, we had to use the reported smog chamber SOA yields for the SOA production calculation in this study.However, it was noticed previously that the SOA yield changed at different photochemical ages in the LFOR [6,18], while the chamber SOA yields are mostly at <2 photochemical days.Therefore, a correction is needed to apply this yield to a wider range of photochemical ages.
The idea is to perform LFOR experiments with several single NMOGs (the representatives in the biomass-burning emissions), and get the SOA production pattern under different photochemical ages.The pattern (normalized SOA yield curve) is then used to correct the difference observed under low (~1 day) and high (up to ~14 days) photochemical ages.We chose 5 NMOGs to perform the LFOR experiments, including 3 VOCs (benzene, toluene, and α-pinene) and 2 IVOCs (o-cresol and naphthalene), which are among the most important contributors to SOA production from the wood burning according to PTR-MS measurements in this study.These NMOGs were introduced into the LFOR by using a small flow passing on top of the glass vial with liquid/solid chemicals, reaching a concentration of 10-50 ppb (see Table S3 and Fig. S4a).The SOA yields normalized by the maximum were put together and were fitted to get a curve (Fig. S4b).As smog chamber yields were mostly obtained at the photochemical age of ~1 day, this yield curve was divided by the value at 1 day (0.52) to get a correction factor curve (Fig. S4c).This correction factor curve was used to correct all the SOA production from NMOGs in the LFOR.

Text S4. Estimation of uncertainty in the contribution of measured NMOGs to SOA
The overall uncertainty associated with the PTR-MS-NMOG contribution to SOA contains the uncertainties from the following three parts.
(1) SOA mass concentration.SOA mass concentration was calculated by volume concentration and particle density, which has a small uncertainty of ±10%.
(2) NMOG concentration.The uncertainties in PTR-MS measurements are mainly from: a. uncertainty in reaction rate constant with H 3 O + ; b. uncertainty in transmission efficiency.According to previous studies [4,19], the overall uncertainty in PTR-MS measurement is ±30%.
a.The uncertainty in original chamber yield of individual precursor is estimated as ±50%.
b.The uncertainty in the scaling factor between SC and LFOR.When using the 5 NMOGs to get a fitting curve, an uncertainty of 0.15 in the normalized yields is estimated (Fig. S4b).This leads to 30−40% uncertainty in the applied factor (depends on photochemical age, see Fig. S4c).We use ±40% for the estimation below.
The uncertainty in applied SOA yield from individual NMOGs (i.e., yield (±50%) × factor (±40%)) is ±64%.When summing them up, the error of each NMOG is uncorrelated [19], so the summed uncertainty for these errors is estimated as the quadrature addition of each absolute uncertainty.This leads to an uncertainty in overall SOA yield to be 15−20%.
Overall, the uncertainty in NMOG concentration is ±30%, which is mainly from the calibration and can be a systematic error.Therefore, the SOA production (NMOG concentration × SOA yield) uncertainty is estimated by directly adding the two errors together, which is 45−50%.After considering errors in SOA mass concentration, the overall uncertainty changes very little, so we estimate the overall uncertainty in the contribution to be ±50%.
When photochemical age is shorter than 2 days, the uncertainty in reaction rate constant with OH (k OH ) can also influence the overall uncertainty.The uncertainty of 40% k OH is assumed and accounted in the results shown in Fig. 2b.

S5.1 LVOC and SVOC fates without considering reactions
The fates of lower-volatility organic vapors (i.e., LVOC and SVOC) in the smog chamber and LFOR is modeled with the KinSim (v.4.1.6)chemical kinetic integrator [20] to provide a simplified understanding of how these vapors evolve in different reactors.In this part, we follow the framework of Krechmer et al. [21] but remove the chemical transformation, i.e., only the condensation on aerosol particles and wall and the evaporation are considered.The parameters applied in the model are listed in Table S5.The initial organic aerosol mass concentration (C OA ) is set to 30 μg m -3 , which is the typical POA concentration used in our smog chamber and LFOR experiments.The initial concentrations of SVOC and LVOC are both set to 1 ppb, and no formation sources are presented during the simulation (i.e., precursor concentration is set to 0).The saturation mass concentrations (C*) of SVOC and LVOC are assumed to be 10 μg m -3 and 0.001 μg m -3 , respectively.
Both SVOC and LVOC condense on the preexisting organic aerosol particles and wall.The condensation rate onto the particles (k cond,p ) is calculated based on the suspended aerosol condensation sink (CS) [6,22]: where D is the diffusion coefficient, which is estimated to be 6 × 10 -6 m 2 s -1 [21,23].The model calculates a CS at each time step based on the total amount of aerosol present at each point in the simulation using the following equation: where r is the particle radius, N(r) is the particle number size distribution, and F FS is the Fuchs−Sutugin correction for gasphase diffusion [21,22,24]: where  is the mass accommodation coefficient (also known as the sticking coefficient) of condensing vapor, which is assumed to be 1 [25].Kn is the Knudsen number: where M is the molecular weight of the condensing vapor, which is assumed to be 200 g mol -1 ; R is the gas constant; T is the temperature.
We assume that SVOC re-evaporates from particles and wall but LVOC does not, since the C* of LVOC is very low.The evaporation rate of SVOC from particles (k evap,p ) is calculated from k cond,p and equilibrium constant as:

𝐶 * 𝐶
In the smog chamber, the condensation rate of organic vapor onto the wall (k cond,w ) is estimated to be 1.7 × 10 -3 s -1 , corresponding to a lifetime of ~10 min, which is the typical timescale for SVOC and LVOC [26,27].The evaporation rate of SVOC from the wall (k evap,w ) is calculated from k cond,w and equilibrium constant as: where C w is the equivalent wall mass concentration, expressed as the equivalent amount of organic aerosol particles that would result in the same gas-wall equilibrium condition [28].Follow Krechmer et al. [27], we calculate the C w of SVOC to be 63.7 μg m -3 by:  16  * .
In the LFOR, the wall loss rate is very low.For example, we can get a ~100% yield of semi-volatile sulfuric acid aerosol in the LFOR [6].Therefore, we assume that the k cond,w in the LFOR is half of that in the smog chamber, which is 8.5 × 10 - 4 s -1 .The k evap,w in the LFOR is also calculated with Eq. (S8).
The other parameters used in the model are as follows: pressure 1 bar, temperature 298 K, RH 50%, duration time 3 h for the smog chamber and 160 s for the LFOR.When the model is running, the aerosol concentration and CS are calculated in real-time.For simplification, particle wall loss is not considered in the model.With this KinSim kinetic integrator, we can model the concentrations of LVOC and SVOC that get lost to the wall, condense on the aerosols, and stay in the gas phase during the residence time in the smog chamber and in the LFOR, which are shown in Fig. S9.

S5.2 LVOC, SVOC, and IVOC fates considering OH reactions
The above model only considers the condensation and evaporation of LVOC and SVOC, which shows that the wall loss is a large trap for SVOCs in smog chambers.However, the reactions with OH are also an important fate of organic vapors.Therefore, we further expand the model by: (1) considering OH reactions; (2) including IVOC; (3) adding a scenario of polluted ambient air.
(1) The OH exposure is set to 2.59 × 10 11 molecules cm -3 s for all scenarios, which corresponds to an equivalent photochemical age of ~2 days (same age as the SOA composition data shown in Fig. 3a).The reaction rate constants with OH (k OH ) of IVOC is estimated as 2.65 × 10 -11 cm 3 molecule -1 s -1 (see Text S6).We assume that IVOC is converted to SVOC after two steps of OH reactions, while SVOC is converted to LVOC after two steps of OH reactions based on the volatility change when adding functional groups [29].The average carbon number and average oxygen number of IVOCs measured in this study are 6.4 and 1.9.Based on this, the number of oxygen atoms is assumed to be 3.9 for SVOC and 5.9 for LVOC, which are 2 and 4 oxygen atoms more than LVOC.We assume that the carbon number of SVOC and LVOC is the same as IVOC.We then calculated the k OH of SVOC and LVOC based on the carbon and oxygen numbers using the equation from Donahue et al. [14]: where n C and n O are carbon number and oxygen number, respectively.The k OH of SVOC and LVOC are calculated as 4.4 × 10 -11 cm 3 molecule -1 s -1 and 6.1 × 10 -11 cm 3 molecule -1 s -1 , respectively, which are similar to the results estimated previously [14].
(2) For IVOC, the initial concentration is also set as 1 ppb.The k cond,p and k evap,p are calculated using the same method as SVOC shown in Text S5.1 It is found that the condensation rate of IVOC is much lower than SVOC from our measurement and according to previous studies [27], thus the k cond,w of IVOC in smog chamber is assumed to be 10 times lower than that of SVOC (i.e., 1.7 × 10 -4 s -1 ).The k evap,w of IVOC in smog chamber is calculated using Eq.(S8) and Eq.(S9).For LFOR, the k cond,w and k evap,w are assumed to be half of those in smog chamber, which is the same assumption as SVOC.
(3) For the polluted ambient air, the k cond,w and k evap,w are set to zero for all organic vapors since there are no wall losses.The duration time is 2 days while the OH concentration is set as the global average value of 1.5 × 10 6 molecules cm -3 .All parameters are summarized in Table S5.
In addition to the two condensational fates (i.e., particle and wall), the third fate of OH reaction is added.For IVOC and SVOC, it is assumed that two steps of OH reactions would change the product category to lower volatility compounds.Therefore, the lifetime against OH reaction is calculated as 2 / (k OH × [OH]).
For LVOC, a few steps of oxidation may lead to the formation of ELVOC but the main fate of ELVOC is highly similar to that of LVOC, i.e., condensation into particles.More steps of OH oxidation may lead to fragmentation that forms higher volatility compounds that can stay in the gas phase.Therefore, we assume that 5 steps of oxidation are required for the fragmentation, same as a previous study [22].This leads to the lifetime against OH reaction for LVOC to be 5 / (k OH × [OH]).The model results considering OH reactions are shown in Fig. S10.

Text S6. IVOC SOA yields estimation
Using the concentration, chemical information, and estimated SOA production of the IVOCs (~70% of total SOA in LFOR experiments), we can estimate their average SOA yield.From the molecular formula and their relative intensities, the average molecular weight of 120 ± 24 Da is determined for these IVOCs.The reaction rate constants with OH (k OH ) are taken from literature when available (e.g., cresol and naphthalene) and are calculated based on molecular formula [14] for other NMOGs.The average k OH is estimated to be 2.65 ± 1.06 × 10 -11 cm 3 molecule -1 s -1 .The average SOA yield of IVOCs was then calculated as dividing the 70% SOA mass concentration by the reacted IVOC mass concentration.

Text S7. Potential OFR limitations
The high radical (including OH, HO 2 , etc.) concentrations and the UV light source for OFRs have the potential to lead to chemistry in the OFR which may be less relevant to the troposphere.A series of modeling studies have addressed these issues recently [30], indicating that reactions in OFRs can often be similar to ambient conditions when atmospheric oxidation is dominated by OH.The possible tropospherically-irrelevant/undesired reactions under some conditions are mainly considered to be from OH suppression caused by operating experiments with extremely high precursor concentrations (i.e., external OH reactivity, OHR ext ), an overly high UV photon flux and/or minimal humidity [30].To prevent these issues in this study, we performed the LFOR experiments following the recommendations of previous studies [30][31][32]: high humidity (RH = ~50%), low ozone concentration (4 ppm), relatively low precursor concentration (OHR ext < 30 s -1 ), and relatively long residence time (~160 s).
A potential issue associated with OFRs is that the OH/HO 2 ratio at higher photochemical ages is likely to be higher than that of the typical ambient atmosphere, leading to a higher contribution from the RO 2 +OH reaction to the fate of RO 2 [30,32].Another potential issue is that the RO 2 lifetime in the OFR might be shorter than in ambient air (due to higher radical concentrations), leading to insufficient time for RO 2 isomerization [30,32].These are common issues for all OFRs at higher photochemical ages and may lead to less relevant OFR chemistry/conditions, which represents the most significant limitation of OFRs.

Text S8. Sensitivity to SOA yield of nC ≥ 6 compounds
As shown in Text S2, we used a SOA yield of 0.32 for the n C ≥ 6 compounds with no reported yield data.However, NO x condition and OA mass concentration influence the SOA yield.In a previous parameterization study [33], the SOA yield of n C ≥ 6 compounds in wood-burning emissions was determined in the range of 0.24−0.4for OA mass concentration of 20−200 μg m -3 (the OA mass concentration range in this study).Therefore, we tested the influence of using lower and higher limits of the SOA yield (i.e., 0.24 and 0.40) on the SOA closure.As shown in Fig. S12a and 12b, the lower and higher limits of the SOA yield lead to the explained SOA fractions of 24% and 30% for LFOR experiments, which are close to the fraction (27%) with a yield of 0.32.For SC experiments, the variations in SOA yield also influence little to the SOA closure results.This indicates that the SOA closure results are not highly sensitive to the SOA yield of n C ≥ 6 compounds.

Figure S2 .b
Figure S2.SOA production from open and stove burning.(a) Low-NO x condition.(b) High-NO x condition.
Text S5.1 for the calculation details. b.Only included in the model of Text S5.2.

Table S1 .
Summary of the LFOR experiments for biomass-burning emissions.

Table S2 .
Smog chamber SOA yield data applied to the calculation.

Table S3 .
Summary of the LFOR experiments for single NMOGs.

Table S4 .
Summary of the smog chamber experiments.

Table S5 .
Parameters for IVOC, SVOC, and LVOC applied in the kinetic model.