Contribution of organic nitrates to organic aerosol over South Korea during KORUS-AQ

Abstract

The role of anthropogenic NO_x emissions in secondary organic aerosol (SOA) production is not fully understood but is important for understanding the contribution of emissions to air quality. Here, we examine the role of organic nitrates (RONO₂) in SOA formation over the Korean Peninsula during the KORUS-AQ field study in spring 2016 as a model for RONO₂ aerosol in cities worldwide. We use aircraft-based measurements of particle-phase and total (gas + particle) RONO₂ to explore RONO₂ phase partitioning. These measurements show that, on average, 1/4 of RONO₂ are in the condensed phase, and we estimate that ≈15% of the organic aerosol (OA) mass can be attributed to RONO₂. Further, we observe that the fraction of RONO₂ in the condensed phase increases with OA concentration, evidence that equilibrium absorptive partitioning controls the RONO₂ phase distribution. Lastly, we model RONO₂ chemistry and phase partitioning in the CMAQ chemical transport model. We find that known chemistry can account for 1/3 of the observed RONO₂, but there is a large missing source of semi-volatile, anthropogenically-derived RONO₂. We propose this missing source may result from oxidation of semi- and intermediate-volatility organic compounds and/or from anthropogenic molecules that undergo autoxidation or multiple generations of OH-initiated oxidation.

Graphical Abstract

1. Introduction

Organic aerosol (OA) constitutes a large, and often dominant, fraction of tropospheric aerosol mass^1-3. Much of this organic aerosol is secondary (secondary organic aerosol, SOA), produced from volatile organic compounds (VOCs) that are sufficiently oxidized in the atmosphere to be condensable and/or water-soluble^4-8. The chemical and physical processes that control SOA production, however, are complex and currently highly uncertain^1,9-16.

Particle phase organic nitrates (pRONO₂) have recently emerged as a significant component of SOA in areas dominated by biogenic emissions, including the Southeast US^17-22, in the Rocky Mountains²³, across Europe²⁴, in the boreal forest²⁵, in the California Central Valley^26,27, and in rural areas of both northern and southern China^28-30. A number of studies have also found significant contributions of pRONO₂ to SOA in regions of oil and gas production, including the Alberta Oil Sands³¹ and in the Uintah Basin³². Recent observations have shown that organic nitrates are a significant contributor to OA in Chinese cities^33,34. Specifically, Yu et al.³⁴ found that organic nitrates make up 9 – 25% of OA during spring - autumn in urban Shenzhen, and the dominant precursors to pRONO₂ included both biogenic (α-pinene, limonene, and camphene) and anthropogenic (styrene) VOCs.

Organic nitrates are produced from the oxidation of VOCs in the presence of NO_x( ≡ NO + NO₂), as shown in Figure 1. During the daytime when photochemistry is active, OH oxidation of VOCs generates RO₂ radicals (R1). The minor product (branching ratio α) of the reaction of NO with RO₂ radicals are gas-phase organic nitrates (gRONO₂, R2). In the nocturnal residual layer away from fresh NO emissions, NO₃ radicals can add to the double bonds of alkenes to generate gRONO₂ (R3) e.g., ³⁵.

$$\text{RH}+\text{OH}\stackrel{}{\to }{\text{RO}}_2+{\mathrm{H}}_2\mathrm{O}$$ (R1)

$${\text{RO}}_2+\text{NO}\stackrel{\alpha }{\to }{\text{RONO}}_2$$ (R2)

$${\text{NO}}_3+{\mathrm{R}}_1={\mathrm{R}}_2\stackrel{}{\to }{\mathrm{R}}_1({\text{ONO}}_2)-{\mathrm{R}}_2$$ (R3)

Figure 1:

Schematic of RONO₂ production and phase partitioning.

If the RONO₂ generated from either the OH-initiated or NO₃-initiated reaction pathways have sufficiently low volatility, they may partition into the aerosol phase as particle-phase organic nitrates, pRONO₂ (Figure 1). At 273 K, the addition of a nitrate functional group reduces the saturation concentration of a given molecule by 2.23 orders of magnitude³⁶, thereby generating a lower volatility compound that may condense to form SOA.

In this study, we examine the contribution of pRONO₂ to OA in Seoul, Korea. As a megacity, Seoul has a complex mixture of urban emissions, including from a number of chemical facilities and from transport of emissions from China, that contribute to the aerosol burden^37,38, though³⁷ determined that the dominant precursors for SOA production in Seoul were locally emitted VOCs. To better understand the sources of SOA in Seoul, here we aim to quantify the contribution of pRONO₂ to the total OA mass and determine the precursors and processes that control the production of pRONO₂ in Korea using observations from the 2016 Korea-United States Air Quality (KORUS-AQ) measurement campaign.

2. Methods

Here, we introduce the KORUS-AQ campaign (Section 2.1), the TD-LIF measurements of pRONO₂ and (Section 2.2), the CU-AMS measurements of pRONO₂ and OA (Section 2.3), and our CMAQ simulations of RONO₂ over Northeast Asia during the time period of the KORUS-AQ campaign (Section 2.4).

2.1. KORUS-AQ

The KORUS-AQ campaign took place during May and June 2016 over the Korean peninsula and the Yellow Sea³⁹. Seoul, Korea is bordered to the west by the Yellow Sea and Gyeonggi Bay and bordered to the north, east, and south by forested and mountainous regions⁴⁰. During KORUS-AQ, winds were typically from the west or northwest, meaning that air over the Yellow Sea can be considered ‘background’ air for Seoul⁴¹. There are a number of large industrial facilities along the Northwest coast of South Korea, including the Daesan petrochemical complex which produces large amounts of VOC emissions⁴².

This analysis uses observations from the NASA DC-8 which flew 20 research flights out of Pyeongtaek, South Korea (≈ 60 km south of Seoul). Flights typically began around 08:00 LT (KST). During a typical flight, three missed approaches were performed over the Seoul Air Base (within 15 km of Seoul city center): one soon after takeoff around 08:00 LT, one around 12:00 LT, and one prior to landing around 15:00 LT. Each missed approach included 15-45 minutes of observations within the boundary layer in the Seoul Metropolitan Area. Flights also consisted of transects west of Seoul over the Yellow Sea, south of Seoul to Jeju, and/or southeast of Seoul to Busan at varying altitudes, as shown in Figure 2. We use 60-second measurement averages in this analysis.

Figure 2:

Maps of average (a) TD-LIF measured and (b) CMAQ modeled tRONO₂ on a log scale, gridded to 0.1°. Seoul, Jeju, Busan, and the Yellow Sea are labeled for reference. Note the modeling domain is larger than the domain of the map plotted here and covers much of Northeast Asia (17.4 - 47.2°N and 93.2-147.4°E).

During KORUS-AQ, the NASA DC-8 was equipped with the only two currently aircraft certified techniques to measure total pRONO₂: the UC Berkeley thermal dissociation laser induced fluorescence (TD-LIF) instrument and the University of Colorado-Boulder high-resolution time-of-flight aerosol mass spectrometer (CU-AMS). Though some other measurement techniques exist to measure certain specific RONO₂ species ^{e.g., 43,44}, the TD-LIF and CU-AMS measurement schemes allow for measurements of the sum of all pRONO₂ species.

2.2. TD-LIF measurements of tRONO₂ and pRONO₂

Measurements of tRONO₂ (gas + particle) were made using the UC TD-LIF instrument^45,46. Briefly, one channel of the instrument measures NO₂ by laser induced fluorescence (LIF). Two other channels first flow air through a heated quartz oven. One channel is set at 180°C, the temperature at which peroxy nitrates (RO₂NO₂) dissociate into RO₂ and NO₂. The second is set at 360°C, the temperature at which RONO₂ dissociate into RO + NO₂. The difference in NO₂ detected in adjacent channels gives the mixing ratio for each class of compounds: the RO₂NO₂ mixing ratio corresponds to the difference between the 180°C channel and the unheated channel, and the RONO₂ mixing ratio corresponds to the difference between the 360°C channel and the 180°C channel.

pRONO₂ concentrations were measured using a fourth channel configured as described in Rollins et al. ⁴⁷. Before entering the heated section of the instrument, air passes through a 10 cm long activated carbon honeycomb denuder with an outer diameter of 2 cm which removes gas-phase compounds (MAST Carbon International Ltd. carbon monolith with 89 cells cm⁻² where each open square is 0.63 by 0.63 mm with 0.43 mm thick walls in between). The particles that remain are then rapidly heated to vaporize the aerosols and dissociate the RONO₂ molecules present into RO and NO₂. NO₂ is then detected via LIF, giving a measurement of pRONO₂. We estimate a limit of detection of 20 ppt of pRONO₂, or 0.055 μg m⁻³ of NO₃. Though inorganic nitrate compounds will also be vaporized, volatile inorganic nitrate salts form HNO₃ when vaporized⁴⁸ and will therefore not interfere in this measurement. Empirical and theoretical studies confirm that NO₂, HNO₃, and gas-phase organic nitrates are all removed at nearly 100% efficiency in the charcoal denuder, while particles greater than 100 nm in diameter are transmitted with over 95% efficiency⁴⁷. Recent work has shown that HONO is removed with near-100% efficiency in dry air and with 85% efficiency at an RH of 46% with a similar denuder⁴⁹. Furthermore, during KORUS-AQ the denuder could be bypassed with a pair of 3-way valves, as shown in Figure S1. When bypassed, the NO₂ calibration mixture reached the pRONO₂ LIF cell. When not bypassed, the NO₂ calibration events served as checks for NO₂ breaking through the denuder. Compared to the 20 ppt limit stated above, no breakthroughs of the 12 or 24 ppb NO₂ calibration steps were detectable during the deployment as shown in Figure S6.

KORUS-AQ is the first time pRONO₂ measurements have been made with TD-LIF on aircraft. Previous ground-based measurements of pRONO₂ by the TD-LIF were made in the Rocky Mountains during BEACHON-RoMBAS²³, in the Uintah Basin³², in the Southeast US during SOAS²¹, and in the California Central Valley during CalNex²⁷.

We apply a small correction for the loss of charged particles to TD-LIF measurements of tRONO₂ and pRONO₂. In the TD-LIF inlet configuration during KORUS-AQ, air for all channels goes through 10 - 20 cm of PFA Teflon before heating. We performed a series of laboratory experiments (detailed in Section S1) to determine the loss of charged particles in these lengths of PFA Teflon tubing. Taking into account the ambient distribution of charged particles⁵⁰ and the observed aerosol size distribution during KORUS-AQ, there is less than 20% loss for charged particles with diameters less than 280 nm in the TD-LIF inlet.

We also apply a correction for inertial losses of particles in the TD-LIF inlet. We model the inertial losses on the two bends (90°and 98°) in the inlet (see Section S1) for varying particle sizes. We apply the size-dependent modeled losses to the aerosol volume distribution measured by laser aerosol spectrometer (LAS, Langley LARGE group). On average, we estimate that the TD-LIF observes ≈ 60% of the particles observed by LAS. We apply both particle loss corrections (charged and inertial) to both the pRONO₂ and tRONO₂ TD-LIF measurements.

2.3. CU-AMS measurements of pRONO₂

A second measurement of pRONO₂ was made by the CU-AMS (Aerodyne Research, Inc.). The CU-AMS also measured organic aerosol (OA) concentrations. A description of the CU-AMS aircraft sampling can be found in DeCarlo et al.⁵¹, Canagaratna et al.⁵², and Nault et al.³⁷.

The CU-AMS uses NO_x ion ratios (NO₂⁺/NO⁺) to differentiate between inorganic nitrate (NH₄NO₃) and organic nitrate (pRONO₂)^23,53, described further in Section S2. Uncertainties in this method are greatest when pRONO₂ < 20% of the total measured aerosol nitrate; those CU-AMS measurements have been removed from this analysis. This ion ratio technique has been used previously in rural environments where VOCs are dominantly biogenic and RONO₂ concentrations are relatively small^17,23,54. However, the high NH₄NO₃ loadings in the urban environment measured during KORUS-AQ create uncertainty for the CU-AMS measurement of pRONO₂.

Though we applied a series of corrections for particle loss in the TD-LIF inlet (described in Section 2.2), we could not entirely reconcile the differences between the two measurements. Since the two measurements may be prone to larger uncertainties under different aerosol size and composition conditions, we conduct the following analyses using both the TD-LIF and CU-AMS pRONO₂ measurements separately and treat them as upper and lower bounds. A comparison of the TD-LIF and CU-AMS measurements, both before and after corrections, can be seen in Figure S9.

Because the TD-LIF and CU-AMS pRONO₂ measurements do not agree perfectly, we also use a CU-AMS-adjusted tRONO₂ to ensure a consistent comparison. In the following calculations that use CU-AMS pRONO₂, we subtract the TD-LIF pRONO₂ measurement from the TD-LIF tRONO₂ measurement to give an estimate of the gas-phase RONO₂ measured by the TD-LIF (gRONO₂). We then add the CU-AMS pRONO₂ to the estimated TD-LIF gRONO₂ to generate the CU-AMS-adjusted tRONO₂.

2.4. CMAQ modeling of RONO₂ chemistry and phase partitioning

We ran the Community Multiscale Air Quality Modeling System (CMAQ) model v5.2^55,56 with the RACM2_Berkeley2.1 chemical mechanism^22,57 over Northeast Asia with a 15-km horizontal grid (17.4 - 47.2°N and 93.2-147.4°E) and 27 vertical layers. Meteorological fields were generated by WRF v3.8.1 and processed for use in CMAQ by MCIP v4.5⁵⁸. The simulation period was April 17, 2016 - June 12, 2016, with the first 14 days as a spin-up period to minimize the impact of initial conditions.

We used the KORUSv5.0 anthropogenic emissions inventory developed at Konkuk University based on the CREATE emission inventory⁵⁹ which includes area, point, mobile, and ship emissions, MEGANv2.1 biogenic emissions⁶⁰, and FINNv1.5 fire emissions⁶¹, all processed through the Sparse Matrix Operator Kernel Emissions (SMOKE) system⁶². The KORUSv5.0 emissions inventory was prepared using the SAPRC07T AERO6 mechanism, which we then converted to RACM2_Berkeley2.1 (detailed in Table S1).

We made a few adjustments to the emissions inventory informed by a series of comparisons between CMAQ modeled VOC concentrations and aircraft VOC measurements made with whole air samples (WAS) analyzed with multi-column gas chromatography⁶³. We increased monoterpene emissions by a factor of three to improve the magnitude agreement between modeled and observed concentrations of monoterpenes (see Figure S12). Note we expect monoterpenes in Korea to have both biogenic as well as anthropogenic sources^64,65.

Comparison between modeled and observed BTEX (benzene, toluene, ethyl benzene, and xylenes) indicated that these species were also underestimated in the emissions inventory (see Figure S12). We updated BTEX emissions over the Daesan petrochemical complex to match emission fluxes calculated from observations using a mass balance approach by Fried et al.⁴². Elsewhere, we note that the spatial pattern of modeled TOL (defined as toluene and less reactive aromatics, for measurement comparison purposes we approximate as the sum of toluene and ethyl benzene) corresponds well to the spatial pattern of the sum of measured toluene and ethyl benzene (see Figure S10). However, without any emissions corrections, the model underestimates boundary layer TOL by a factor of 1.4. We also note that measurements of other reactive aromatics (xylenes and 1,2,4-trimethyl benzene) correlate well with the sum of measured toluene and ethyl benzene (see Figure S11). As such, we scale TOL emissions by 1.4 and define the emissions of the other reactive aromatics based on their measured ratios to the sum of toluene and ethyl benzene. We use measured o-xylene as a proxy for model species XYO, the sum of measured m-xylene and 1,2,4-trimethyl benzene as a proxy for model species XYM, and measured p-xylene as a proxy for model species XYP. This method results in setting XYO emissions as 0.05× TOL, XYM emissions as 0.08× TOL, and XYP emissions as 0.07× TOL, such that XYO, XYM, and XYP emissions all follow the same spatial pattern as TOL.

We use the default initial conditions and boundary conditions from the initial condition (ICON) and boundary condition (BCON) processors in CMAQ v5.2. However, measurements of isoprene-derived nitrates by Caltech’s Chemical Ionization Mass Spectrometer (CIT-CIMS)⁶⁶ indicated that longer-lived propanone nitrate and ethanal nitrate were underestimated in CMAQ. Consequently, we increased the boundary and initial condition concentrations of propanone nitrate and ethanal nitrate to match the CIT-CIMS observations of both nitrates over the Yellow Sea (propanone nitrate = 21.5 ppt; ethanal nitrate = 4.1 ppt).

The original RACM2 (Regional Atmospheric Chemistry Mechanism) mechanism⁶⁷ is available in CMAQ v5.0.2 and later versions⁶⁸. Browne et al.⁶⁹ modified the mechanism to RACM2_Berkeley to expand the organic nitrate chemistry. New species, along with their corresponding oxidation rates and branching ratios, were added to further classify anthropogenic nitrates^70-72 and to represent monoterpene nitrates^73-76. The parameterization of OH-initiated isoprene oxidation was also updated^77-79. RACM2_Berkeley was evaluated using aircraft observations over the Canadian boreal forest⁶⁹.

RACM2_Berkeley was updated to RACM2_Berkeley2 in Zare et al.⁵⁷ to reflect recent advances in the representation of OH- and NO₃- initiated BVOC oxidation under both low- and high-NO_x conditions, with a focus on a detailed representation of nitrates derived from NO₃-initiated oxidation of isoprene and on the fate of the most important individual biogenically-derived organic nitrates. Deposition rates were also updated.

Zare et al.²² revised RACM2_Berkeley2 to RACM2_Berkeley2.1 to include an explicit representation of multi-phase organic nitrate formation and loss, including vapor-pressure driven partitioning into organic aerosol, aqueous-phase uptake, and condensed-phase reactions. Further updates were also done to explicitly represent isoprene nitrates from NO₃ oxidation that are subject to reactive uptake to the aerosol phase. As such, the RACM2_Berkeley2.1 mechanism represents our current understanding of RONO₂ chemistry and phase partitioning. Zare et al.²² evaluated this mechanism (implemented in CMAQ) using observations from the Southern Oxidant and Aerosol Study (SOAS) campaign in the Southeast US during summer 2013. Inclusion of the particle-phase pathways for RONO₂ improved the model-measurement agreement for tRONO₂, and the modeled fraction of tRONO₂ in the particle phase (F_p) was within the range of observed F_p.

To compare modeled and measured concentrations, we sample CMAQ coincidentally in time (hourly resolution) and horizontal space with each observation. All comparisons in the following analysis use boundary layer measurements (< 1, 000 m) and the average of the bottom three model layers.

3. Results

Maps of average TD-LIF measured and CMAQ modeled tRONO₂ used in the following analysis are shown in Figure 2. Both the measurements and model indicate that tRONO₂ concentrations are highest in and around Seoul. However, the model consistently underpredicts tRONO₂ concentrations throughout the region. For reference, CMAQ predicts that > 95% of pRONO₂ are derived from vapor-pressure dependent partitioning into organic aerosol, whereas < 5% of pRONO₂ enter the particle phase through aqueous pathways, similar to what Zare et al.²² found for the Southeast US.

3.1. RONO₂ partition into the aerosol phase and can be a significant contribution to SOA

We explore the average phase partitioning behavior of RONO₂ during KORUS-AQ in Figure 3. Our observations from both the TD-LIF and CU-AMS indicate that, on average, 1/4 of tRONO₂ is in the condensed phase and therefore contributes to the OA burden. We also consider a line, drawn above most measurement means, that represents a reasonable upper limit of 35% for the fraction of tRONO₂ in the particle phase.

Figure 3:

(a) Plot of pRONO₂ versus tRONO₂ mixing ratios as measured by TD-LIF and CU-AMS. Data are binned by tRONO₂ mixing ratio, and the average pRONO₂ in each bin is plotted. The York fit shown corresponds to the average fraction of RONO₂ in the particle phase (F_p). We draw an estimated upper limit (≈35%) for the fraction of RONO₂ in the particle phase, as shown in the blue dashed line, drawn above the mean of most measurements. (b) Plot of pRONO₂ mass concentration (using an estimated average molecular weight of 300 g mol⁻¹) versus OA mass concentration. Data are binned by OA concentration, and the average pRONO₂ in each bin is plotted. The York fit shown corresponds to the average fraction OA mass that can be attributed to pRONO₂. Again, we draw an estimated upper limit (≈40%) for the fraction of OA mass attributable to pRONO₂, as shown in the blue dashed line, drawn above most measurement means. We do not understand why AMS data above 15 μg m⁻³ deviates so strongly from the trend measured at lower OA concentrations. In both plots, the larger, dark colored error bars correspond to the standard deviation of measurements within each bin to represent observed variability whereas the smaller, light colored error bars correspond to the standard error of measurements within each bin to represent measurement uncertainty. We apply a threshold requirement of 20 observations per bin to include in plot.

To quantify the contribution of pRONO₂ to total OA concentrations, we assume an average molecular weight for pRONO₂ of 300 g mol⁻¹²⁷. We expect condensable RONO₂ to be highly oxidized, contain at least one nitrate group (molecular weight = 62 g mol⁻¹) and to therefore have relatively high masses. With this assumption, we estimate that ≈ 15% of the OA mass can be attributed to pRONO₂, as shown in Figure 3. Note that this estimate does not include CU-AMS measurements when pRONO₂ ¡ 20% of total measured aerosol nitrate. We again consider a reasonable upper limit, drawn above most measurement means, to estimate that a maximum of 40% of OA can be attributed to pRONO₂. This is within the range of pRONO₂ contributions to OA mass measured across Europe (42%)²⁴, in a suite of studies across the eastern US, western US and Europe (5-73%)⁸⁰, and in recent studies in urban and rural China (9-28%)^34,81.

3.2. Observations indicate RONO₂ phase partitioning is controlled by absorptive partitioning into OA

Previous studies have shown that vapor pressure controls the phase of organic nitrates^22,27. This equilibrium absorptive partitioning follows Raoult’s Law: the fraction of RONO₂ in the particle phase increases with increasing mass of the absorbing or solvating aerosol, namely total organic aerosol^82,83. Accordingly, the equilibrium fraction of an individual RONO₂ species i in the particle phase (F_p,i) is given by

$$F_{p,i}=\frac{C_{p,i}}{C_i}=\frac{C_{OA}∕C_i^{\ast }(T)}{1+C_{OA}∕C_i^{\ast }(T)}={\left(1+\frac{C_i^{\ast }(T)}{C_{OA}}\right)}^{-1}$$ (1)

Here, C_p,i and C_i are the particle phase and total concentrations of species i, respectively. $C_i^{\ast }(T)$ is the temperature-dependent saturation concentration (μg m⁻³) of species i, and C_OA is the concentration of total OA.

For both the TD-LIF and CU-AMS measurements of pRONO₂, the fraction of RONO₂ in the particle phase (F_p) increases with increasing OA concentration and increases with decreasing temperature, as shown in Figure 4. Assuming the speciation of RONO₂ is invariant with temperature, these relationships between F_p, OA, and temperature indicate that the phase partitioning of RONO₂ during KORUS-AQ is indeed controlled by equilibrium absorptive partitioning.

Figure 4:

[Top] Plots of the fraction of RONO₂ in the particle phase (F_p) versus OA concentration. Data were separated into three temperature bins (centered at 286, 293, and 300 K) and binned by OA concentration. The average F_p in each OA bin is plotted, and error bars represent the standard deviation of F_p in each bin. As suggested by absorptive partitioning theory, measured F_p increases with increasing available solvating aerosol (in this case, OA). [Bottom] Temperature-dependent fractional distribution (f_j) of saturation concentrations ($C_j^{\ast }$) fit to a volatility basis set. Each set of plots is shown for the TD-LIF measurements (a,e), the CU-AMS measurements (b,f), unmodified CMAQ output (c,g), and CMAQ output with an unknown source of RONO₂ added (d,h).

To determine the volatility distribution of RONO₂ observed during KORUS-AQ, we define a saturation concentration basis set of $\{C_j^{\ast }\}=\{3,30,300\}$, μg m⁻³, mfollowing the convention of Donahue et al.⁸². Though we expect some RONO₂ species to have volatilities outside of this range, because the OA concentrations we observe during KORUS-AQ do not exceed 40 μg m⁻³ we cannot reasonably constrain volatilities outside of this defined basis set. Given this basis set, the total fraction of organic nitrates in the particle phase (F_p,tot) can be represented as

$$F_{p,tot}=\frac{{\Sigma }_i{C}_i{F}_{p,i}}{{\Sigma }_i{C}_i}=\sum _{j=1}^n{f}_j{\left(1+\frac{C_j^{\ast }}{C_{OA}}\right)}^{-1}$$ (2)

Here, f_j is the fraction of organic nitrates that can be classified as having saturation concentration $C_j^{\ast }$, and n = 3 for the basis set defined earlier.

We solve for each f_j in Equation 2 using our observations of F_p,tot (= pRONO₂ / RONO₂) and organic aerosol concentrations (C_OA). Moreover, because saturation concentration is dependent on temperature, we separate the observations into a series of temperature bins and solve for fitting parameters f_j in each temperature bin, as shown in Figure 4 for both TD-LIF and CU-AMS observations. As expected, organic nitrates become less volatile at lower temperatures. At all temperatures, 10-39% of organic nitrates can be represented with C* ≤ 3 μg m⁻³, meaning they will dominantly be condensed at the average observed organic aerosol concentrations of ≈ 9.8 μg m⁻³. At high temperatures (≈ 300 K), 73-76% of organic nitrates can be represented with C* ≥ 300 μg m⁻³, meaning that they will dominantly remain in the gas phase at observed OA concentrations. At low temperatures (≈ 286 K), the TD-LIF measurements suggest that 67% of organic nitrates can be represented with C* ≥ 300 μg m⁻³ and the CU-AMS measurements suggest 61% of organic nitrates can be represented with C* ≥ 30 μg m⁻³.

We also fit the data to Equation 2 using an empirical relationship between C* and ΔH_vap from Epstein et al.⁸⁴ to examine the variation of RONO₂ volatilities observed at different temperatures but referenced to 300 K. Figure S15 shows the distribution of C*(300 K) for RONO₂ during KORUS-AQ.

3.3. CMAQ modeling misses a large source of semivolatile, anthropogenically-derived RONO₂

To test the efficacy of our simulations, we compare modeled and measured NO_x, O₃, and O_x( ≡ O₃ + NO₂) mixing ratios at the Olympic Park ground site in Seoul, as shown in Figure S13. Our CMAQ simulation is able to capture the diurnal patterns in NO_x, O₃, and O_x and does not show a systematic under- or over-estimation. Additionally, we compare measured and modeled OA concentrations at KIST (Korea Institute of Science and Technology, Seoul) during the campaign (see Figure S13). The CMAQ simulation is able to accurately capture the regional OA background concentration and many of the episodic events with elevated OA. Moreover, a multi-model inter-comparison study of air quality simulations for the KORUS-AQ campaign used comparisons of PM₁ to show that models, including CMAQ, were generally able to capture the synoptic meteorological patterns during the campaign⁸⁵. We also include a comparison of measured and modeled temperature on the DC-8 in Figure S13 which shows that our CMAQ simulations are able to capture the observed variability in temperature.

However, our CMAQ simulation underpredicts measured tRONO₂ concentrations by a factor of ≈ 3, as shown by the slopes reported in Table 1 and plotted in Figure S14. Moreover, our CMAQ simulation underpredicts measured pRONO₂ concentrations by a factor of ≈ 10, indicating that the RONO₂ in CMAQ are too volatile. These underpredictions for both tRONO₂ and pRONO₂ indicate that our simulation is missing a large source of condensable RONO₂. We note that the RACM2_Berkeley2.1 mechanism does not include Cl-initiated oxidation of VOCs; however, based on observations, we estimate that the production rate of RO₂ radicals from Cl oxidation is an order of magnitude slower than that from OH-initiated oxidation (see Section S6) and is therefore insufficient to explain the missing source of RONO₂.

Table 1:

Comparison of the York fit slopes between measured (TD-LIF and CU-AMS) and CMAQ modeled concentrations of tRONO₂, pRONO₂, and F_p. Comparison is shown for both the unmodified CMAQ output and CMAQ output with an unknown source of condensable RONO₂ added. Scatter plots of these comparisons can be seen in Figure S14.

	tRONO2		pRONO2		Fp
	TD-LIF	CU-AMS	TD-LIF	CU-AMS	TD-LIF	CU-AMS
CMAQ	0.30	0.35	0.12	0.09	0.56	0.35
CMAQ add unknown	0.88	0.98	0.92	0.79	0.61	0.44

To help determine the origin of the missing source of RONO₂, we examine the correlation between the model-measurement RONO₂ difference (RONO_2,diff) and measurements of various VOC classes. We find R² < 0.05 for the correlation between RONO_2,diff and both isoprene and α-pinene, whereas there are relatively stronger correlations between RONO_2,diff and anthropogenic alkanes (R² = 0.15), alkenes (R² = 0.12), aromatics (R² = 0.23), and aldehydes (R² = 0.67). The weak correlations between the RONO_2,diff and VOCs of biogenic origin and the relatively stronger correlations between RONO_2,diff and VOCs of anthropogenic origin suggest that the missing source of condensable RONO₂ is derived from anthropogenic VOCs.

Note that the missing source of RONO₂ over the Korean peninsula is likely derived from both transport of RONO₂ produced in China as well as from locally-produced RONO₂. As shown in Figure 2, CMAQ underpredicts RONO₂ over the Yellow Sea, a region influenced by transport of air from China during parts of the KORUS-AQ campaign, as well as over the urban centers of the Korean peninsula where local chemistry contributes to the observed RONO₂.

Furthermore, the RACM2_Berkeley2.1 mechanism was initially tested and validated on a regional scale over the Southeast US, an area dominated by biogenic emissions^22,57. Additionally, as described in Section 2.4, we adjusted the emissions of monoterpenes to improve the model-measurement agreement for biogenic VOCs. Though we expect some change in the oxidation product distribution between low-NO_x environments (e.g., Southeast US) and high-NO_x environments (e.g., Seoul), we are reasonably confident that our CMAQ simulation is accurately capturing the production and fate of RONO₂ derived from biogenic VOCs. We therefore attribute the missing source of RONO₂ in our simulations to RONO₂ of anthropogenic origin. This previous work evaluating RACM2_Berkeley2.1 in the Southeast US^22,57 did not look at urban RONO₂ in the US, but we have no reason to suspect that this missing source of condensable RONO₂ is not a general phenomenon.

The relationship between CMAQ-modeled F_p(RONO₂), OA, and temperature is shown in Figure 4c. In contrast to the observations (Figure 4a,b) and in contrast with absorptive partitioning theory, the modeled F_p increases with increasing temperature and decreases with increasing OA. Exploration of the speciated distribution of modeled RONO₂ (shown in Figure S16) indicates that the increase in modeled F_p with temperature is driven largely by a temperature-dependent change in the RONO₂ speciation. The phase partitioning of each RONO₂ species is controlled by absorptive partitioning, meaning the fraction of an individual RONO₂ species in the particle phase increases with decreasing temperature. However, the modeled increase in the total concentration of low-volatility monoterpene nitrates (HONIT) with temperature is larger than the modeled change in concentration of other higher-volatility nitrates with temperature. As a result, the concentration of pRONO₂ increases with increasing temperature faster than the concentration of gRONO₂ increases with temperature, causing the total F_p to increase with increasing temperature. This modeled relationship between F_p and temperature stands in stark disagreement with the observations and therefore indicates that the species distribution of RONO₂ over Korea is incorrectly captured in our CMAQ simulation.

To test and quantify our hypothesis that our CMAQ simulation is missing a large source of condensable, anthropogenic RONO₂, we test the effect of adding an additional source of RONO₂. Because our CMAQ simulation underpredicts measured tRONO₂ concentrations by a factor of ≈ 3 (Table 1), we assign this additional source to have double the concentration of the existing simulated RONO₂. To determine the average volatility of this missing source of RONO₂, we iteratively vary its assigned C* by order of magnitude (e.g., C* = 30, 300, 3000 μg m⁻³) and use an empirical relationship between C* and ΔH_vap from Epstein et al.⁸⁴. We find the best agreement between modeled and measured pRONO₂ and F_p with C* = 300 μg m⁻³ as shown in Table S2. Though comparison between modeled and measured RONO₂ remains relatively scattered (see Figure S14) and the missing source likely includes a variety of molecules with a range of volatilities, adding this missing semivolatile RONO₂ source improves the magnitude of the model-measurement agreement for tRONO₂, pRONO₂, and F_p, as shown in Table 1. Moreover, as shown in Figure 4, addition of this unknown source of relatively condensable RONO₂ results in an increase in F_p with decreasing temperature and increasing OA concentration. This relationship between F_p, temperature, and OA is in agreement with the observations and with equilibrium absorptive partitioning theory.

4. Discussion

The RACM2_Berkeley2.1 mechanism represents our state-of-the-science understanding of RONO₂ chemistry, where the only sources of semi-volatile RONO₂ are biogenic. However, this mechanism only captures one third of the RONO₂ production over the Korean peninsula. Moreover, the unknown source of organic nitrates consists of RONO₂ that are lower volatility than most of the existing RONO₂ in the model. Consequently, our current understanding of RONO₂ chemistry is missing pathways for semivolatile RONO₂ production as a result of either missing oxidation pathways (first- or multi-generation, bimolecular or unimolecular) or an underestimation of RONO₂ yields.

Because the known chemistry can only account for one third of the observed RONO₂, the missing source is approximately double in magnitude to the known sources. During KORUS-AQ, the average reactivity of all measured VOCs with OH was 2.4 s⁻¹, and the effective average RONO₂ yield (α), weighted by reactivity, was 1.3%. If the unknown source of RONO₂ has a low α of 1%, the missing reactivity must be ≈ 3 s⁻¹. On the other hand, if the unknown source of RONO₂ has a higher α of 20%, the missing reactivity must be ≈ 0.15 s⁻¹. For reference, during KORUS-AQ the average isoprene reactivity was 0.051 s⁻¹ and the average toluene reactivity was 0.054 s⁻¹.

We hypothesize three potential missing sources of semivolatile RONO₂: (1) missing source(s) of semi- and intermediate-volatility organic compounds (S/IVOCs) that are oxidized to RONO₂; (2) unrepresented autoxidation mechanisms that produce highly oxygenated organic peroxy radicals (RO₂) which could react with NO to form RONO₂; or (3) more generations of bimolecular oxidation than are currently represented.

S/IVOCs are considered major SOA precursors ^{e.g., 12,14,37,86-93}, but their concentrations are challenging to measure in the atmosphere due to condensation within instruments ^{e.g., 94}, and their chemistry is difficult to measure in chamber experiments due to wall loss ^{e.g., 95}. Nault et al.³⁷ concluded that, during KORUS-AQ, S/IVOCs and reactive aromatics contributed to 70% of the total SOA over Seoul. Because they are emitted with relatively low volatility, oxidation of S/IVOCs to form RONO₂ could contribute to the missing source of semivolatile RONO₂. Because the addition of a nitrate group decreases a molecule’s volatility by ≈ 2.2 orders of magnitude³⁶, a missing RONO₂ source with saturation concentration 300 μg m⁻³ implies a precursor with C* = 10⁵ μg m⁻³, namely an IVOC. The contribution of S/IVOCs to pRONO₂ is not unprecedented; Lee et al.³¹ determined that much of the pRONO₂ formation in the Alberta oil sands occurred via photo-oxidation of IVOCs under high-NO_x conditions.

Autoxidation, a mechanism involving an intramolecular hydrogen-shift followed by addition of molecular oxygen in RO₂ radicals, can quickly (in seconds) generate highly oxygenated molecules, or HOMs^96,97 and references therein. Because of their high oxygen content, HOMs have significantly reduced volatility compared to their parent VOCs ^{e.g., 98-100}. Although autoxidation becomes relatively more competitive with bimolecular oxidation pathways as NO_x decreases, absolute rates of autoxidation increase with increasing NO_x due to increased oxidant availability¹⁰¹. In Korea’s high-NO_x environment, autoxidation may generate highly oxidized RO₂ which could produce RONO₂ via reaction with NO (R2). While most previous studies of HOMs have focused on autoxidation of RO₂ derived from biogenic VOCs, theoretical calculations by Wang et al.¹⁰² indicate that substituted benzenes, which were measured in high abundance during KORUS-AQ^42,63, may also produce HOMs through autoxidation of bicyclic peroxy radicals. Beginning with a substituted benzene molecule with C* ≈ 10⁷ μg m⁻³, one hydrogen-shift reaction resulting in the addition of a hydroperoxide group would reduce the volatility by ≈ 2.5 orders of magnitude, and further addition of a nitrate group would reduce the volatility by ≈ 2.2 orders of magnitude³⁶, resulting in a C* ≈ 10² μg m⁻³ compound. Consequently, only one hydrogen-shift reaction is necessary to convert a substituted aromatic compound to a nitrate of the missing volatility.

Additionally, multiple recent studies have suggested that multi-generation OH oxidation of aromatics can lead to highly oxygenated oxidation products, many of which, particularly under high-NO_x conditions, contain nitrogen ^{e.g., 103-105}. Some of these nitrogen-containing products are likely organic nitrates, but the nitrogen-containing product distribution also includes peroxy nitrates and nitro aromatics. Because aromatics are a large contributor to total VOCs over Korea^42,63, there could be significant production of semivolatile, multi-functional, oxygenated organic nitrates from multi-generation oxidation of aromatic VOCs.

In summary, exploration of the phase partitioning of RONO₂ over the Korean peninsula using our aircraft-based measurements of pRONO₂ and tRONO₂ during KORUS-AQ, as an example of urban chemistry, indicate that organic nitrates contribute ≈ 15% of the total OA. This significant contribution of organic nitrates to the OA burden, as has been observed elsewhere, reinforces the notion that a better understanding of the processes that control the production, loss, and phase partitioning of RONO₂ are crucial for understanding the processes that control SOA production and loss. Our current understanding of RONO₂ chemistry can only explain one third of the observed RONO₂ in Korea and is therefore missing a source of semi-volatile, anthropogenically-derived RONO₂ in and around Seoul. We recommend further laboratory and field research to determine the source VOCs and mechanisms that drive the production of this missing source of organic nitrates.

Synopsis:

We use aircraft-based measurements and modeling to explore the organic nitrate aerosol in Korea as a model for cities worldwide.