A Box-Model Simulation of the Formation of Inorganic Ionic Particulate Species and Their Air Quality Implications in Republic of Korea

The Observation-Constrained Atmospheric BOX model (OCABOX) was used to analyze the formation of secondary inorganic PM species in the Seoul Metropolitan Area (SMA), South Korea. The measurement data of the ionic components of PM 2.5 and their gaseous precursors made at the Olympic Park ground site (37.53°N, 127.12°E) during the Korea-United States Air Quality field campaign were used to run OCABOX in observation-based mode and compare the simulation results. The use of the HNO 3 concentrations measured at a marine background site as the boundary conditions appeared to increase the accuracy of the model prediction of HNO 3 and particulate NO 3 - concentrations. For the primary precursors emitted considerably throughout the SMA, such as NO x and NH 3 , using the data measured inside the SMA as the boundary conditions could lead to more accurate predictions. OCABOX was shown to be a reliable tool to analyze the formation of secondary inorganic aerosol in the SMA if used with appropriate regional background concentrations and observation-based constraints


Nitrate (NO 3 -
) is an important secondary inorganic ionic particulate matter (PM) species produced by the oxidation of nitrogen oxides (NO x ).The chemical pathway from NO x via nitric acid (HNO 3 ) to particulate nitrate is complicated (Seinfeld and Pandis, 2016) and is affected by other atmospheric species, such as VOCs (Meng et al., 1997).In addition, the partitioning of total nitric acid (HNO 3 + NO 3 -) determined by the equilibrium between gaseous HNO 3 and particulate NO 3 -changes depending on temperature, relative humidity, and the concentrations of basic components, such as ammonia (NH 3 ), crustal species, and sea salt (Kim et al., 2004;Ro et al., 2001), that can neutralize HNO 3 , whereas the total sulfuric acid (H 2 SO 4 + SO 4 2-) exists primarily as particulate SO 4 2-.Therefore, the NO 3 -concentration varies considerably depending on the location and season.
Particulate NO 3 -can be a dominant ionic species in PM 2.5 where the ammonia A Box-Model Simulation of the Formation of Inorganic Ionic Particulate Species and Their Air Quality Implications in Republic of Korea concentration is high enough to neutralize both sulfuric acid and nitric acid (Wang et al., 2020), whereas it is usually present in the coarse mode where the ammonia concentration is insufficient (Ocskay et al., 2006;Wolff, 1984;Savoie and Prospero, 1982).NO x and NH 3 emissions are not being controlled very well in East Asia, including South Korea, compared to that of SO 2 (He et al., 2007;Richter et al., 2005), which may lead to an environment favorable to the formation of particulate ammonium nitrate (NH 4 NO 3 ), frequently making NO 3 the most abundant ionic particulate species (Kim et al., 2017;Lee et al., 2016;Park et al., 2013;Shon et al., 2012;Park and Kim, 2004;Choi et al., 2001).The formation of NH 4 NO 3 often coincides with the occurrence of high PM 2.5 concentration (Wang et al., 2020;Park and Kim, 2004).One study reported that the particulate NO 3 -concentration in Seoul, Korea increased in the 2010s despite the decreasing gaseous NO x concentration (Han and Kim, 2015), which warrants scientific understanding.
Box models are also helpful for investigating the causes of high air pollution events.Xue et al. (2014) examined a wintertime high-NO 3 -episode in Hong Kong using an observation-based box model and reported that both the gas-phase oxidation of NO 2 by OH radicals and heterogeneous N 2 O 5 hydrolysis contributed to HNO 3 production, which in turn was converted to particulate NO 3 under ammonia-rich conditions.However, Xue et al. (2014) used their box model only for diagnostic purposes; they reported only the reaction rates producing gaseous HNO 3 calculated from the measured data.Schroeder et al. (2020) used a box model to investigate the ozone production rate in the Seoul Metropolitan Area (SMA).
The strength of box models lies in their outstanding time efficiency stemming from the neglection or simplification of horizontal and vertical transports.The neglect of transport can be justified in the cases of extreme air stagnation.When the transport is considered using userdefined boundary conditions, the effects of boundary conditions on predicted pollutant concentrations may become considerable.For this reason, box models are generally used for "gaining conceptual understanding and testing hypotheses through targeted sensitivity simulations and comparison with observations" (Wolfe et al., 2016) rather than for obtaining accurate predictions.
In the present work, a box model called the Observation-Constrained Atmospheric BOX model (OCA BOX), developed by modifying a three-dimensional (3D) chemical transport model EPA Models-3 Community Multiscale Air Quality Modeling System (CMAQ), was used to study the formation of secondary inorganic PM species in South Korea, where high PM concentrations are frequently associated with high concentrations of inorganic ionic species.The PM 2.5 concentration tends to be higher in winter and spring than in summer and fall because of higher emissions and poorer dispersion (in winter) or because of the long-range transport effects and intensive photochemical production of secondary species (in spring) (Kim et al., 2018;Kim et al., 2017;Kim et al., 2014).This study examined the formation pathways of NO 3 -and its contribution to the high pollution events using measured data and box modeling.(Crawford et al., 2021).The collected data have been used by many researchers trying to understand the factors influencing the air quality of Korea (e.g., Kim et al., 2022;Park et al., 2021;Jordan et al., 2020;Oak et al., 2019;Kim et al., 2018;Nault et al., 2018) because the KORUS-AQ campaign provided unprecedented comprehensive measurements of air pollutants, including both gas-and particulate-phase species.
In this study, a set of hourly data on major ionic components of PM 2.5 , including NH 4 , and Cl -, and their gaseous precursors, such as HCl, SO 2 , NO x , HNO 3 , and NH 3 , measured at the Olympic Park (OP) ground site (37.53°N,127.12°E) from 15 May to 17 June 2016 by Seoul Metropolitan Government Research Institute of Public Health and Environment (Choi et al., 2019) were used to run OCABOX in the observation-based mode and to compare the simulation results.The PM 2.5 mass concentra-tion was measured using a radiometric particulate mass monitor, FH62C14 (Thermo Scientific, USA), whereas a Monitor for Aerosol & Gasses in Ambient Air, MAR GA ADI 2080 (Metrohm Applikon., Netherlands), was used to determine the concentrations of the ionic particulate species and related gaseous species HCl, NH 3 , and HNO 3 .Gas species analyzers were used to measure the CO (Ecotech model EC9830, Australia), O 3 (Ecotech model EC9810, Australia), SO 2 (Ecotech model EC9850, Australia), and NO x (Ecotech model EC9841, Australia) concentrations.One can refer to the literature (Choi et al., 2019;Ghosh et al., 2012) for more detailed information on the measurement and data processing methods.Meteorological parameters, such as air temperature, relative humidity, and wind speed, were also measured at the OP site.
Fig. 1 shows the concentrations of several key pollutants measured during the period.There was a distinct event exceeding the Korean daily PM 2.5 standard (35 g/ m 3 ) during 25-31 May, which will be called the LRT period because it was caused by enhanced long-range transport, and several relatively short events during which local emissions dominated due to stagnant air (Park et al., 2021;Peterson et al., 2019;National Institute of Environmental Research and National Aeronautics and Space Administration, 2017).During the KORUS- AQ campaign, the air quality of Seoul was reportedly influenced strongly by the formation of secondary PM (Park et al., 2021;Nault et al., 2018;Kim et al., 2018).This is supported by the high correlations among the PM 2.5 and ionic species concentrations shown after 25 May.
As shown in Section 3, the supply of appropriate regional background concentrations appeared to be important for the reliable operation of OCABOX.The data measured at a marine background site in Baengnyeong Island (BI) (37.97°N, 124.63°E) collected during the same period were provided as the regional background concentrations in several sensitivity analyses.The Beta Attenuation Monitor Model 1020 (BAM1020) (Met One Ins., USA) was used to measure the PM 2.5 mass concentration, whereas the anion and cation particle and gas system, AIM URG-9000D (URG Corporation, USA), was used to measure the ionic species in PM 2.5 .The measured data of several gaseous species, CO (TELEDYNE API model T300, USA), O 3 (TELE-DYNE API model T400, USA), SO 2 (TELEDYNE API model 100A, USA), and NO 2 (TELEDYNE API model T200, USA), were also used.

2 Operation of OCABOX
A box model OCABOX, developed by modifying the 3D chemical transport model CMAQ v.5.3.2 into 0-dimensional box model, was used in this study.Carbon Bond 6 (CB6) (Yarwood et al., 2010) was used as the gas-phase mechanism and the 6th generation aerosol module (AERO6) was used as the aerosol mechanism including aqueous-phase and heterogeneous reactions.The horizontal dimension of the box was set at 30 km and the planetary boundary layer height (PBLH) predicted by a mesoscale meteorological model Weather Research and Forecasting (WRF) v3.8 (Powers et al., 2017), which can be a proxy of the mixing layer height (MLH) (Banks et al., 2015), was used as the vertical dimension of the box.
The meteorological input file was generated using the  output of WRF and measured data.Fig. 2 shows the WRF domain (WO) with a horizontal grid size of 30 km on a Lambert Conformal map projection covering the whole Korean Peninsula.The terrain-following sigma vertical coordinate was employed with 29 unevenly spaced vertical levels ranging from the surface to 50 hPa.ERA-Interim reanalysis data were used for the initial and boundary conditions.WRF was run from 0000 UTC 20 April to 0000 UTC 30 June 2016 (Fig. 3).The first 10 days of WRF simulation were treated as spin-up time, and only the fields from 0000 UTC 30 April were supplied to OCABOX simulations.Table 1 provides detailed information of the WRF domain and physics options used for the WRF run.The measured data for air temperature, relative humidity, and wind speed were overwritten to the WRF output file to replace the model prediction.
The emission fields for the OCABOX run were gen-erated using Sparse Matrix Operator Kernel Emissions (SMOKE) v4.5.The Korean national emissions inventory called Clean Air Policy Support System (CAPSS) for 2016 (National Institute of Environmental Research, 2019) was used as the raw emissions data.The natural emissions were determined using the Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.1.Emissions speciation was performed using the chemical mechanism CB6.SMOKE was run from UTC 30 April to 0000 UTC 20 June 2016 (Fig. 3).
Table 2 summarizes the SMOKE run configuration used in this study.Fig. 2 shows the domain for the SMOKE run (SO).
OCABOX was run from 0000 UTC 10 May to UTC 17 June 2016, including a five-day spin-up (Fig. 3) for the five scenarios shown in Table 3: a base run (BASE) and four sensitivity analysis runs (SA1, SA2, SA3, and SA4).For the initial condition, the "profile" concentrations provided by the 3D model CMAQ were used.The profile concentrations were also used as the boundary conditions except for the species for which the concentrations measured at the BI marine background site (Na NH 3 , and HNO 3 ) were used, as shown in Table 3.For the build-up of the species whose concentrations are not measured at the OP site during the episode periods, the first five days of the OCABOX simulations were taken as the spin up time.Fig. 3 presents the simulation time set for the models used in this study.CMAQ v5.3 was needed to generate the species concentrations at the BI site to perform a sensitivity test for the marine background concentrations, which will be explained in Section 3.4.WRF v3.8 was used as the meteorological driver for this test.Fig. S1 shows the spatial domains used for WRF (WC1 and WC2) and CMAQ (C1 and C2) simulations for this test.Two-way nesting between Domain 1 (WC1 and C1) and Domain 2 (WC2 and C2) was used.A Lambert Conformal map projection was employed for all domains.Table 4 and Table 5 show detailed information on the run configurations for WRF and CMAQ, respectively, used for this sensitivity test.WRF was run from 0000 UTC 20 April to 0000 UTC June 2016, whereas CMAQ was run from 0000 UTC April to 0000 UTC 20 June, with the first 10 days for both WRF and CMAQ runs treated as spin-up time.The emission fields were prepared for the domain C2 in the same way described above.

RESULTS AND DISCUSSION
The concentrations of all the species except NH 3 , NH 4 + , NO, NO 2 , HNO 3 , and NO 3 -measured at the OP site were used as the constraints for the basis of the OCABOX run.For the missing data, the average values of the previous and next hours (in the cases of isolated missing data) or the average values for the whole simulation period (in the cases of consecutive missing data) were used in principle.On the other hand, this method could not be applied to Na + , K + , Ca 2+ , and Mg 2+ because there were too many missing values.In particular, there were only six samples that provided valid Na + concentrations.However, the average Na + concentration of those six samples was 0.46 μg/m 3 , which is similar to the values reported previously for Seoul (Choi and Kim, 2010;Kang et al., 2006Kang et al., , 2004;;Lee et al., 2005).For these cations, the average values of the measured concentrations were used for the whole simulation period.
Table 6 lists the average values measured for those cations.An anion Cl -and gas species HCl are shown together.These six species are measured only occasionally (e.g. during intensive campaigns like the present study) in Korea and, therefore, box models often have to be run without their measured concentrations.Although the concentrations of these species are not very high, the possible errors due to the lack of information on their concentrations need to be estimated for wide applications of OCABOX.For the species whose concentrations are not constrained, the OCABOX calculates their concentrations by solving the governing mass balance equations with the initial and boundary conditions pro-  A sensitivity test was conducted on the concentration input for the above-shown six species.The differences between the concentrations of HNO 3 , NH 3 , NO 3 -, and NH 4 + calculated by OCABOX with and without the constraints for the six species listed in Table 6 were all less than 6%.The simulation without the constraints for the six species resulted in the increased concentrations of HNO 3 (by 5.1%) and NH 4 + (by 5.9%) and decreased concentrations of NO 3 -(by 1.5%) and NH 3 (by 4.8%) because the profile values for the cations were lower than the measured ones while the profile value for Cl -was higher than the measured one.These differences, however, are relatively small compared to the overall uncertainty of the OCABOX simulations.Hence, using the CMAQ profile as the initial and boundary conditions for these species can be justified when the measured data are not available.A previous study also showed that the effects of the crustal cations and Na + on the aerosol thermodynamics were very small (Kim et al., 2022).
In the following subsections, the predictions of OCA BOX are compared with observations.The two secondary inorganic ionic aerosol species NO 3 -and NH 4 + that were not constrained by observations and their gaseous precursors NO, NO 2 , HNO 3 , and NH 3 are the target species in the comparisons.

1 Base Run (BASE)
As shown in Table 3, the CO, O 3 , SO 2 , SO 4 2-, Na + , K + , Ca 2+ , Mg 2+ , Cl -, and HCl concentrations measured at the OP site were used as constraints and the CO, O 3 , SO 2 , and NO 2 concentrations measured at the BI site were used as regional background conditions for the BASE simulation.
Fig. 4 compares the OCABOX-predicted concentrations with observations.The results of the integrated process rate (IPR) analysis, which is a part of the process analysis (PA), indicating the effects of each process, such as gas-phase chemistry (CHEM), horizontal advection (HADV), vertical diffusion (VDIF), emission (EMIS), dry deposition (DDEP), and aerosol processes (AERO), on the species concentrations are shown together.NO was supplied mostly by emission and converted to NO via chemistry (Fig. 4a), whereas NO 2 was produced from the chemical reaction of NO and removed by chemistry and horizontal advection (Fig. 4b).Compared to the observations, both NO and NO 2 were underestimated by OCABOX considerably.The reason for underpredicting NO and NO 2 may be either the underestimation of NO x emissions or too low background concentrations provided by the CMAQ profile, as was indicated by the huge effect of HADV on NO 2 .This will be discussed in more detail in the following subsections presenting the sensitivity analysis results.
HNO 3 (Fig. 4c) and NH 3 (Fig. 4e) were also underpredicted, but the aspects were quite different; the underprediction of HNO 3 was remarkable mostly at night, whereas NH 3 was underpredicted during the day.The characteristic evolutions of NH 3 concentration and the emission rate (Fig. 4e) stem from the diurnal variation of mixing height (not shown); the low mixing height at night leads to a strong influence of emission, and the opposite is the case during the day.Regarding HNO 3 , the story becomes completely different because HNO is not a primary pollutant; the underestimation of HNO 3 is prominent at night.
During the day, HNO 3 is produced by whereas during the night, its production takes place via the following reactions: When Reactions ( 2)-( 5) are not important, HNO 3 reaches its maximum concentration during the day and minimum during the night in the presence of basic components (Mehlmann and Warneck, 1995), which can be attributed to the photochemical production during the day by Reaction (1) followed by its conversion to partic-ulate NO 3 -during the night at high humidity and low temperature.A similar pattern is shown in the time series of the model-predicted HNO 3 concentration but the measured HNO 3 concentration did not show the nighttime minimum (Fig. 4c), particularly during the LRT period.
Fig. 5 shows the result of integrated reaction rate (IRR) analysis, which is the other part of PA, for HNO 3 .The two pathways shown above were clearly distinguished in terms of the time of day and HNO 3 produc- Several sensitivity analyses were performed to determine how the improved model prediction can be achieved.

2 Sensitivity Test Run 1 (SA1)
As shown in Fig. 4, the base run resulted in a general underestimation of NH 4 + and NO 3 -, due to the underestimation of primary precursors NH 3 , NO, and NO 2 .This can be attributed to two possible reasons: underes- A simple order-of-magnitude calculation based on the used emissions data showed the effects of the emissions to be ~0.1 ppb/h for both NH 3 and NO x , which are too low to achieve the observed concentration levels.A previous study reported that even 50%-increased NO x emissions led to an underestimation of the NO x concentration in Seoul during the KORUS-AQ period (Oak et al., 2019).On the other hand, the underestimated vehicleemitted ammonia may lead to an underestimation of NH 4 NO 3 (Link et al., 2017).Therefore, the first sensitivity test conducted in this study was to increase NO, NO 2 , and NH 3 emissions by a factor of 2 to see if this could reduce the gap between the model prediction and observation satisfactorily.
Fig. S2 shows the OCABOX predictions and PA results obtained with the doubled emissions of NO, NO 2 , and NH 3 .Simply doubling the emissions of the primary species did not improve the model performance very much.Although the NH 3 and HNO 3 concentrations were increased by emissions doubling, the underprediction of HNO 3 during the night and the underprediction of NH 3 during the day were not mitigated, resulting in a minor increase in the NH 4 + and NO 3 -concentrations.
A possible reason why the increased emissions could not improve the model performance may be found by examining the free ammonia ratio (FAR) (Blanchard and Hidy, 2003): ] are the total ammonia and nitric acid molar concentrations, respectively.In the right-hand side of Eq. ( 6), the numerator means the amount of ammonia remaining after neutralizing SO 4 2-.Therefore, FAR is the ratio of free ammonia over total nitric acid.If FAR>1, the atmosphere is in an ammonia-rich condition.Fig. 6 shows the FAR values predicted by the BASE and SA1 runs with observed values.The observation gave the FAR values always larger than 1 (in the range of 1.27-138), often exceeding 10, except for the missing data, indicating an ammonia-rich condition that prevailed in SMA during the studied period.This is in agreement with the previous reports that total ammonia is in excess in the atmosphere of Seoul due to the local emissions from vehicles, agricultural activities, and industry (Kim et al., 2018(Kim et al., , 2017;;Han and Kim, 2015;Kim, 2006).On the other hand, the model-predicted FAR values often decreased to below 1 for both BASE and SA1 runs, indicating that the model did not provide sufficient ammonia.The possibility of underestimating the emissions by more than a factor of 2 does not appear feasible.Therefore, additional sensitivity analyses needed to be conducted.

3 Sensitivity Test Run 2 (SA2)
In the second sensitivity analysis, NO, NO 2 , NH 3 , and NH 4 + were also constrained by the measurements (Table 3) to mitigate the effects of inaccurate emissions of NO x and total ammonia.Fig. 7 and Fig. S3 (for NO and NO 2 ) present the results of the SA2 run.Although this led to naturally improved NH 3 and NH 4 + , NH 3 was overpredicted, while NH 4 + was underpredicted, accompanied by an underprediction of HNO 3 and NO 3 -during the LRT period.This result can be interpreted as the excessive evaporation of NH 4 + to NH 3 owing to the insufficient supply of HNO 3 by the model, again raising the possibility of the omission of HNO 3 transported from the Yellow Sea.On the other hand, the predictions of both NO 3 -and NH 4 + were improved considerably during relatively short high PM event periods reportedly influenced by stagnant air (e.g., 7-8 June).Based on these results, it is important to test the hypothesis that the reason for the underprediction of nighttime total HNO 3 during the LRT period is due to the omission of the inflow of HNO 3 produced over the Yellow Sea.

4 Sensitivity Test Run 3 (SA3)
Until now, the data measured at the BI site (37.97°N,124.63°E) for CO, O 3 , SO 2 , and NO 2 and the 'CMAQ profile' values for the other species were used as the boundary conditions for horizontal advection.Because the measured concentrations at the BI site and the CMAQ profile values are much lower than the concentrations in SMA except for O 3 , the use of these boundary conditions may lead to strong ventilation for the box modeling, resulting in considerable underestimations.Therefore, two sensitivity tests were performed to account for the effects of polluted background.
The first test for the 'polluted background' hypothesis was conducted with respect to the possibility of the inflow of HNO 3 produced over the nearby Yellow Sea during the night.As was mentioned above, Reactions (2)-( 5) are generally not crucial inside the SMA because the nighttime O 3 concentration tends to be low due to the active ozone titration by high NO emission.Over the Yellow Sea, which is located to the west of SMA, Reactions (2)-( 5) can be significant because the maritime ozone tends to survive during the night due to negligible NO emission.If HNO 3 produced in this way flows into the SMA, where the ammonia-rich condition prevails, it may react efficiently with NH 3 to produce NH 4 NO 3 .
In SA3, all the measured data on relevant species (Na + , K + , Ca 2+ , Mg 2+ , Cl -, HCl, SO 4 2-, NO 3 -, and NH 4 + ) made at the BI site were added to the background field.
The problem was that NH 3 and HNO 3 were not measured at the BI site.Therefore, their concentrations were estimated from the measured concentrations of NH 4 + and NO 3 -as follows.First, the 3D model CMAQ was run, as described in Subsection 2.  Fig. 8 compares the NH 3 and HNO 3 concentrations estimated in this way for the BI site with those measured at the OP site.The estimated BI NH 3 concentration was much lower than that measured at the OP site, as expected, because of the low NH 3 emission over the sea.On the other hand, the estimated BI HNO 3 concentration was as high as or even higher than the measured OP concentration during the LRT period.By contrast, during the non-LRT period, it was generally lower than the OP concentration.These results are very similar to the report of another research group who measured HNO 3 in both the BI and OP sites during the KORUS-AQ campaign (Korean Society for Atmospheric Environment, 2016).
Fig. 9 and Fig. S4 (for NO, NO 2 , and NH 3 ) show the results of SA3.Despite the unchanged underestimation of NO and NO 2 compared to the BASE run, the HNO 3 and NO 3 -concentrations were increased significantly.This, in turn, led to considerable improvement in the prediction of high NH 4 + concentrations during the LRT period.The degree of agreement with observation in terms of NH 4 + concentration predicted for the LRT period was even better than SA2, in which NH 4 + was constrained directly by observation.This result strongly suggests that the model-predicted NH 4 NO 3 formation in that period was hindered by the shortage in supply of HNO 3 from the sea.On the other hand, the NH 4 + prediction of SA3 during the non-LRT period was no better than that of SA2.

5 Sensitivity Test Run 4 (SA4)
SA4 was the last sensitivity analysis and the second test for polluted background.In this test, the effects of horizontal advection were removed.The SMA surrounding the box domain of this study is much larger than the box itself and usually maintains a high pollution level.Therefore, the air flowing into the box may have similar levels of pollutant concentrations, having little + , which may be influenced strongly by the emission sources in the vicinity, were replaced by the measured data at the OP site.This method was selected to incapacitate the ventilating effect of horizontal advection instead of simply skipping the horizontal advection process in the model because the latter may lead to the steady build-up of emitted primary pollutants in the box.
Fig. 10 presents the results of the SA4 run.The use of observations made at OP site as the regional background concentrations, without constraining them directly, led to accurate predictions of NO and NO 2 , showing that the inflow from the polluted background is sufficient to maintain the high concentrations inside the box even without a large emission.This result suggests that the main reason for the underprediction of NO x in the BASE run might not be the low emissions but the venti- and NO 3 -all were improved for both the LRT and non-LRT periods because the supply of both the primary and secondary precursors from the regional background was accounted for.
The results of SA4 show that the assignment of appropriate regional background concentrations may be nec-essary for reliable box model predictions.For the secondary precursor HNO 3 , for which nighttime production over the sea may be significant, the measurement made at the BI site appears adequate as the marine background condition, whereas the observations made inside the SMA may be better for the primary precursors, such as NO, NO 2 , and NH 3 .Overall, the box model OCA BOX can be a reliable tool for analyzing the formation of

CONCLUSIONS
A box model OCABOX was used to study the formation of secondary inorganic aerosol in South Korea.A set   tion results suggested that although the photochemical inland production during the day was the main pathway of the production of HNO 3 and NO 3 -, the maritime production during the night made nonnnegligible contribution during the LRT period.Appropriate regional background concentrations appeared to be essential for reliable box model simulation.The use of the observations made at a marine background site as the background concentrations could remedy the underestimation of HNO 3 by OCABOX.In the case of the primary precursors, the use of measurements made inside the SMA as background concentrations, without constraining them directly, led to better model performance.

SUPPLEMENTARY MATERIALS
tions and meteorological parameters are required to run and validate OCABOX.The Korea-United States Air Quality (KORUS-AQ) field study conducted jointly by the National Institute of Environmental Research of South Korea and the National Aeronautics and Space Administration of the United States in May-June 2016 offered extensive data sets measured from ground stations, aircrafts, and ships

Fig. 1 .
Fig. 1.Overview of the time evolutions of the concentrations of several key pollutants during the studied period.

Fig. 2 .
Fig. 2. Model domains for preparing meteorology and emissions inputs: the WRF domain (WO) and the SMOKE domain (SO).

Fig. 5 .
Fig. 5. HNO 3 production rates from the reactions with OH during the day and from the reactions with N 2 O 5 during the night.

-
each hour during the whole period were calculated, and the average [NH 3 ]/[NH 4 + ] and [HNO 3 ]/[NO 3 -] ratios were determined at each hour.Finally, the NH 3 and HNO 3 concentrations were estimated by multiplying these ratios by the measured concentrations of NH 4 + and NO 3

Fig. S1 .
Fig. S1.The spatial domains used for the sensitivity test on the background concentrations.

Table 1 .
Information on the domain and physics options used for WRF running.

Table 2 .
SMOKE configuration used in this study.

Table 4 .
WRF configurations used for the sensitivity test for background concentrations.

Table 5 .
CMAQ configurations used for the sensitivity test for background concentrations.Inherited from the 3D model CMAQ, the profile values represent the minimal background concentrations.Table6also lists the profile values for the six species.The differences between the profile values and measured ones are not very large; all the species showed agreement within a factor of two except for Ca 2+ and K + because the concentrations of the crustal and sea-salt species in fine particles (PM 2.5 ) are not very high in the SMA except in the Asian dust transport periods.

Table 6 .
Measured and profile values for the crustal and sea-salt species and related gas species concentrations.