Observed and Modeled Mass Concentrations of Organic Aerosols and PM2.5 at Three Remote Sites around the East China Sea: Roles of Chemical Aging

Severe PM2.5 air pollution over the Asian continent is occasionally transported across the East China Sea by the westerly winds to Japan, continuing for long distances over the Pacific Ocean. Despite such polluted air masses causing health issues, conventional models tend to underestimate levels of organic aerosols (OA) and PM2.5. Here, PM2.5 and its major components recorded during three field campaigns carried out at Fukue Island (32.75°N, 128.68°E), Japan (spring 2009), Rudong (32.25°N, 121.37°E), China (spring 2010), and Jeju (33.35°N, 126.39°E), Korea (autumn 2012) around the East China Sea were used to test the performance of the Weather Research and Forecasting-Chem/ATRAS-MOSAIC model. Overall, model performance was improved by introducing chemical aging represented by a volatility basis-set scheme, whereby median values of the model/observation ratio for OA were raised to 0.34–1.28 from 0.30–0.35 in the case of conventional settings. In particular, the levels of OA at the Fukue site and daytime buildup of the OA levels at all three sites were reproduced by the model. OA levels were still sometimes underestimated. This suggests that either emission rates of organic precursors are being underestimated or other pathways of OA formation are also important. Our analysis also indicates that this region is characterized by high OH concentrations, promoting chemical aging. The predictions of PM2.5 levels in the model also improved, with median values of the model/observation ratio shifting from 0.67–0.91 to 0.68–0.95, when chemical aging of OA was taken into account.


INTRODUCTION
Heavy PM 2.5 air pollution in East Asia, which is widespread over several thousands of kilometers, is of concern as an important regional health risk factor.Continental outflow from East Asia, dominant in autumn/ winter/spring, carries pollutants for long distances across the East China Sea, while chemical aging on a time scale of days promotes secondary formation of sulfate, nitrate, and organics.Although fine particles in the outflow have been studied for a couple of decades (e.g., Huebert et al., 2003;Uematsu et al., 2010), it is only recently that quantification of PM 2.5 has been specifically targeted.This was motivated by the recent development of National Environmental Standards in Japan (2009( ), China (2012;;2016), andKorea (2015).Until 2009, the PM 2.5 mass concentration levels around the East China Sea were almost unknown.Kanaya et al. (2010) first reported that the Japanese Environmental Standard (35 µg m -3 as daily average) had been exceeded at Fukue Island (32.75°N, 128.68°E; 75 m asl), located in the eastern region of the East China Sea, based on their annual observations from 2009.There are no large local pollution sources on the island; hence, these levels were attributed to the influence of the East Asian continent, ca.700 km away.In subsequent years, a national monitoring network was developed in Japan (Ministry of the Environment of Japan, 2014).This monitoring revealed that above-threshold levels were widely observed over western Japan.However, comprehensive observations on the major chemical components of PM 2.5 including both organics and inorganics have not been often reported, relevant to such long-range transport; data were only available for specific research missions over western Japan (e.g., Hatakeyama et al., 2011;Takami et al., 2013).In the upstream region of Japan, data at Gosan (33.17°N, 126.09°E) on Jeju Island, Korea (Lim et al., 2012), Baengnyeong Island (37.97°N, 124.63°E) , Korea (Lee et al., 2015), and at Changdao Island (37.93°N and 120.72°E),China, between Bohai Bay and Yellow Sea (Feng et al., 2012) were reported.
It is critical to test and improve the capability of chemical transport models to describe the spatial and temporal variations of PM 2.5 , using observations as benchmarks.This will contribute toward a full understanding of processes and sources, leading to severe pollution events in specific areas.Using a Weather Research and Forecasting/Community Multi-scale Air Quality (WRF/CMAQ) model (Byun and Schere, 2006;Eder and Yu, 2006), Ikeda et al. (2014;2015) quantitatively assessed the contributions from continental and domestic sources to PM 2.5 at Fukue Island and for Japan.The results indicate that over 60% of PM 2.5 in western Japan is derived from sources in the Asian continent on annual average.These studies provided timely information to the Japanese government and improved public awareness of pollution issues.In their estimation, however, the capability of the model to accurately reproduce the mass concentration levels of major chemical constituents was not perfect.One major shortfall was that concentrations of organics were significantly underestimated, when a conventional chemical scheme (e.g., empirical two-product yield model) was used for their secondary formation.In Asia, this tendency to underestimate was originally identified during analysis of airborne observational data from the ACE-Asia campaign (Heald et al., 2005), and was commonly found for the simulation results from varieties of air quality models using the conventional scheme.Temporal variations or mass spectra of the observed organics implied that the mass concentrations of secondary organics were likely underestimated (e.g., Turpin et al., 1995;Zhang et al., 2005;Yu et al., 2007).
Meanwhile, Donahue et al. (2006) and Robinson et al. (2007) proposed a volatility basis-set (VBS) scheme that classifies organics using volatility, and takes into account enhanced gas-to-particle conversions in the course of chemical aging, i.e., the multiple steps of gas-phase oxidation by OH radicals that produce less volatile species.Threedimensional model simulations implementing this scheme show increased mass concentrations of organics, yielding better agreement with observations over the United States, Europe, and at a global scale (e.g., Fountoukis et al., 2011;Jathar et al., 2011;Ahmadov et al., 2012;Barbet et al., 2016).Recently, the VBS scheme has been introduced into regional model simulations for East Asia (Matsui et al., 2014;Morino et al., 2014Morino et al., , 2015;;Han et al., 2016;Lin et al., 2016).As some of these studies focused on urban conditions, only a subset has analyzed the regional-scale features in the context of continental outflows from East Asia.Specifically, data recorded at Fukue Island, Cape Hedo (26.87°N, 128.25°E; 60 m asl), Okinawa, and Gosan with quadrupole aerosol mass spectrometers (Q-AMS) over short periods were used for testing regional (Matsui et al., 2014) and global model simulations (Jathar et al., 2011;Jo et al., 2013).Han et al. (2016) included observations from the summit of Mt.Tai (36.27°N, 117.10°E; 1545 m above sea level (asl)), Shandong Province, China.Morino et al. (2015) used filter analysis data from six remote/rural sites in Japan (Tsushima, Oki, Kyotango, Tateyama, Sado, and Rishiri).Clearly more observational data from the East China Sea region are desired to obtain a better context, particularly for occasions of transport of continental air masses with high aerosol loadings.For this purpose, observational data from remote sites are useful, as they are representative of the grid sizes in regional-scale chemical transport models (typically 50-100 km).
In this paper, we present observational data consisting of mass concentrations of total PM 2.5 and its major chemical components (organics, inorganic ions, and black carbon; BC) at three remote sites around the East China Sea region.After examining mass closure, the observational data were compared with simulations of the WRF-Chem/ATRAS-MOSAIC model (Matsui et al., 2014a), including the VBS scheme to represent chemical aging.First, reproducibility of the OA mass concentrations was assessed, in terms of mean bias, and temporal variation including diurnal patterns.The concentrations of OH in the WRF-Chem model were also compared with those estimated using a photochemical box model, constrained by observations.Together with the evaluation of inorganic ions and BC, overall performance of the model in terms of its reproduction of observed total PM 2.5 concentration levels was evaluated.

Observations
Data from three field campaigns were used in this study.1).The Fukue site (Takami et al., 2005;Kanaya et al., 2016a) is located on the northwestern part of Fukue Island.It is far from the populated area on the east of the island, and other large cities (e.g., it is ~100 km from Nagasaki, and ~200 km from Fukuoka).The site is ~200 km and ~700 km away from Korean and Chinese coastal lines, respectively.It is occasionally influenced by long-range transport events from the Asian continent.Typical transportation time from the continent during the springsummer season was 10-60 h (Kanaya et al., 2016b).The Rudong site is ~100 km north of Shanghai, and just 2 km west of the East China Sea coastline (Pan et al., 2012;Kudo et al., 2014).The major activity in the surrounding area is agriculture, and local sources of aerosols and precursor gases are limited.The Jeju site is on the western slope of Mt.Halla, located near the center of Jeju Island.Jeju city is ~20 km to the southwest.The nearby areas are farmland and forest.
Table 1 summarizes the instruments used during these three field campaigns.At the Fukue site, PM 2.5 and BC mass concentrations were measured using a SHARP5030 (Thermo Scientific, Waltham, MA, USA) and BCM3130 (Kanomax, Osaka, Japan).Organic carbon (OC) was measured with a semi-continuous ECOC analyzer (Sunset Laboratory, Inc., Tigard, OR, USA).A high-volume air sampler HV-500F (Sibata Scientific Technology, Ltd., Saitama, Japan), equipped with a slit-type impactor designed for PM 2.5 sampling, was used at a flow rate of 500 L min -1 for daily collection of particles.An ion chromatograph was used to quantify sulfate, ammonium, chloride, sodium, potassium, and calcium ions.Elemental metal mass concentrations (Fe, Al, Ti, Ca, and Mg) were analyzed using Inductively Coupled Plasma-Atomic Emission Spectroscopy (ICP-AES); these data were used to derive soil dust mass concentrations from the equation [soil dust] = 2.2 Al + 2.49 Si + 1.94 Ti + 1.63 Ca + 2.42 Fe (Hand et al., 2011).Si mass concentrations were always assumed to be 3.1 times that of Al, as specified for a standard sample of Gobi desert dust (CRM No. 30; National Institute for Environmental Studies).Mass concentrations of nitrate ion and organics were measured with the Q-AMS (Aerodyne Research Inc., Billerica, MA, USA).BC and CO data at the Fukue site are discussed by Kanaya et al. (2013Kanaya et al. ( , 2016b)).
At the Rudong site, online measurements of PM 2.5 , BC, OC, and particle filter sampling were carried out similarly to those at the Fukue site.All data on ion concentrations (sulfate, nitrate, ammonium, chloride, sodium, potassium, calcium) were obtained from ion chromatography, while metal concentrations (Fe, Al, Ti, Ca, and Mg) were analyzed using ICP-AES.Q-AMS observations were not conducted at this site.At the Jeju site, PM 2.5 and BC were measured with a BAM1020 (Met One Instruments Inc., Grants Pass, OR, USA) and an Aethalometer (AE22; Magee Scientific, Berkeley, CA, USA).A semi-continuous ECOC analyzer was used to provide OC mass concentrations.An online ambient ion monitor (URG-9000, URG Corporation, Chapel Hill, NC, USA), equipped with a PM 2.5 cyclone, was used to observe mass concentrations of sulfate, nitrate, ammonium, chloride, sodium, potassium and calcium ions.Filter sampling and chemical analysis conducted at another 1100 m-site on Jeju Island (33.35°N, 126.45°E; 1100 m asl) during the same period, suggest that these data were regionally representative.Metal analyses were only made with the PM 10 filters collected at this 1100m site; hence, the PM 2.5 metal concentrations were assumed to be 0.31 times those of PM 10 , based on median ratios of PM 2.5 /PM 10 mass concentrations for Mg 2+ and Ca 2+ .
The BC and ECOC analyzers were equipped with a 2.5 µm-diameter cutoff cyclone at all field sites.The temperature protocol for the ECOC analysis was similar to that used in the Interagency Monitoring of Protected Visual Environments (IMPROVE) at Fukue and Rudong sites, while the NIOSH (National Institute for Occupationa Safety and Health) protocol was used at the Jeju site.The systematic difference in measurements using these protocols was estimated to be 21% (Kondo et al., 2006).
The OH radical concentrations during the three field campaign periods were estimated, based on a photochemical box model, constrained by comprehensive observations of gases (ozone, CO, NO, NO 2 , volatile organic compounds (VOCs), and oxygenated volatile organic compounds (OVOCs)) and J values.The average O 3 , CO, benzene, and propane concentrations were 56.8, 189, 0.14, and 0.27 ppbv for Fukue (Kanaya et al., 2016a) , 56.1, 403, 1.0, and 1.2 ppbv for Rudong (Kudo et al., 2014;Pan et al., 2015), and 43.1, 172, 0.12, and 0.52 ppbv for Jeju.Briefly, using the Regional Atmospheric Chemistry Mechanism version 2 (RACM2, Goliff et al., 2013), OH concentrations were determined from the balance of production and loss rates in a time-dependent manner.Because of their observational constraint, the OH concentrations derived from the box model may be more reliable than the WRF-Chem model simulations.
In this study, mass concentrations of OA, rather than OC, were used.The OA/OC ratio was assumed to be 2.0 for all three field campaigns, with an estimated uncertainty of ± 25% (i.e., for the range 1.5-2.5).The OA mass is denoted here as "OC*2."The average mass fraction of m/z 44 of OA (f44) observed with the Q-AMS at Fukue Island during this period was 0.15, and the corresponding OA/OC ratio was estimated to be 2.0 from its empirical relationship (Aiken et al., 2008).Philip et al. (2004) also estimated the OA/OC ratio from satellite observations of NO 2 and obtained similar values.Using a mass balance method, Xing et al. (2013) estimated the OA/OC ratio over China to be 1.92 ± 0.39 year round, supporting the value assumed here.Except for the analysis of diurnal variation, the concentrations were averaged over a 22 h-period (typically from 0930 to 0730 LT) for the Fukue site, or over daytime (from 0930 to 1730 LT) and nighttime (from 1800 to 0800 LT) periods for the Rudong site to match the time resolution of the filter sample collection using the high-volume air sampler.Nine nighttime sampling periods during June 12-23, 2010 were excluded from our analysis at the Rudong site.These were the periods of open crop residues burning after the harvest in areas within 10 km, which had a strong influence on the aerosol data, as shown by Positive Matrix Factorization analysis (Pan et al., 2012).Such effects are sporadic in time and space, and thus, our model simulation with a 60 kmresolution had difficulty taking such effect into account.For the Jeju site, although all the data had an hourly resolution, daytime (from 0800 to 2000 LT) and nighttime (from 2000 to 0800 LT) averaging was carried out, similarly to the Rudong site, to study the reproducibility on a halfday time resolution.
The domain setting for the model simulation was the same as for the East Asia case of Matsui et al. (2014) (see their Fig.2(b)), having a nested inner domain with a horizontal grid spacing of 60 km.We considered anthropogenic and volcanic emissions (Streets et al., 2003), as well as biomass burning (GFED3, van der Werf et al., 2010) and online biogenic emissions from MEGAN2 (Guenther et al., 2012) in our model.
Another model run labeled as "conventional" was also carried out, in which SVOC and IVOC emissions were assumed to be zero, the saturation concentration of POA was assumed to be 10 -6 µg m -3 (non-volatile), and the oxidation rate coefficient of the surrogate species, originally 4 or 1 × 10 -11 cm 3 molecule -1 s -1 in the "aging" run, was assumed to be zero.Here, conventional runs without aging were intended to act as a baseline for comparison.Using the difference between modeled OA mass concentrations for aging and conventional runs, the impact of the aging scheme was assessed.
The results from the WRF-Chem model were comprehensively compared and evaluated with separate regional-scale model simulation results based on WRF/CMAQ (Ikeda et al., 2015), using an updated anthropogenic emission inventory (REAS ver. 2 for year 2008; Kurokawa et al. (2013)).WRF/CMAQ version 4.7.1 represents OA with a conventional two-product chemical scheme.For OA, inorganics and BC (Tables 2 and S1), except for nitrate at Fukue and sulfate/ammonium at Rudong, there were no large differences from WRF-Chem results.This suggested that the impact of using Streets et al. (2003) for year 2001 for the anthropogenic emission inventory was limited.More details are given in the subsequent sections.The simulated mass concentrations of major PM 2.5 components (OA, sulfate, nitrate, ammonium, and BC) were individually compared with observations at all field sites.Observed total PM 2.5 mass concentrations were also compared with observed or modeled sum concentrations.However, caution was exercised, as the observed values likely contained trace amounts of water as a mass fraction, although low relative humidity (35%) was targeted in the PM 2.5 observations.The E-AIM equilibrium model (http://www.aim.env.uea.ac.uk/aim/aim.php,Model III; Clegg et al., 1998) was used to estimate this water content at a relative humidity of 35% under meta-stable conditions, using observed and modeled dry compositions.This water mass was taken into account, before comparisons were made with the observed total PM 2.5 mass concentrations.

Overview of Observations: Total PM 2.5 and its Various Components
The actual campaign-average mass concentrations of major chemical components and the total PM 2.5 mass concentrations are shown in Fig. 2 (leftmost bar of the three bars shown for each campaign).At the Fukue site, the average sulfate, nitrate, ammonium, BC, and OA (OC*2) mass concentrations were 7.4, 0.4, 2.5, 0.5, and 3.9 µg m -3 , respectively.Observed temporal variations in PM 2.5 (Fig. 7(f), blue line) were dominated by pollution transported from the Asian continent.For example, PM 2.5 levels exceeding 30 µg m -3 were recorded on April 29-30, May 7-9, May 11-12, May 24-25 and June 2-3, 2009.Backward trajectories calculated using the hybrid singleparticle Lagrangian-integrated trajectory model (HYSPLIT; Draxler and Rolph, 2013) for these days showed that air masses arriving at the Fukue site had traveled over China or Korea.The concentrations of other components (soil dust and minor ions) and the estimated water content were 1.2 and 3.2 µg m -3 on average.The sum of individually measured components (19.1 µg m -3 ) comprised 90% of the total PM 2.5 mass concentrations (21.2 µg m -3 on average) measured with the SHARP5030.When we used the OA values from the Q-AMS, which were higher than OC*2, a better mass closure (99%) was achieved.
At the Rudong site, the average sulfate, nitrate, ammonium, BC, and OA mass concentrations were 13.0, 4.7, 5.8, 1.2, and 12.3 µg m -3 (Fig. 2).A heavy pollution episode occurred around May 20, 2010 (see Fig. 8(a)), influenced by emissions from the Shanghai-Nantong area, located 20-100 km south.In contrast, easterly winds carried clean air masses to this site from May 28 to June 9.After June 10, air pollution was again recorded with the influence from postharvest wheat residue burning (data having a large influence from this biomass burning were removed, as discussed in the previous section).Other components (soil dust and minor ions) and the estimated water content had average mass concentrations of 3.8 and 6.9 µg m -3 .The sum of individual components (47.8 µg m -3 ) exceeded the observed total PM 2.5 mass concentration (41.5 µg m -3 ) measured with the SHARP5030.This implies that the actual OA/OC ratio was lower than 2.0.However, the prescribed OA/OC ratio of 2.0 was retained for our discussion, because the uncertainty in the total PM 2.5 was large.
At the Jeju site, the average sulfate, nitrate, ammonium, BC, and OA mass concentrations were 2.5, 0.4, 1.6, 0.6, and 3.3 µg m -3 (Fig. 2).High concentrations were recorded on October 10, 15, 20, 28 and 29, when air mass traveled over northern parts of central East China (Fig. 8(b)).The average mass concentrations of other components (soil dust and minor ions) and the estimated water content were 1.6 and 2.6 µg m -3 , yielding a sum for individual components of 12.5 µg m -3 .This corresponded to 87% of the total PM 2.5 mass concentrations measured with the BAM1020.
The relative mass fractions of OA compared with the observed total PM 2.5 were 18%, 30%, and 23% for the Fukue, Rudong, and Jeju sites, respectively, suggesting that OA fractions were large.This indicates that the reproducibility of OA levels in our model is key to satisfactory representation of PM 2.5 at these sites.At the Fukue site (Fig. 3(a)), the observed OA concentrations averaged 3.9 µg m -3 for OC*2, but 5.7 µg m -3 from the Q-AMS.Meanwhile, conventional and aging runs had average concentrations of 1.5 and 5.6 µg m -3 , respectively.The increase in the organic fraction related to introducing chemical aging into our model was large, increasing by as much as 3.7 times.This implies there is a significant influence of chemical aging (OH oxidation) of precursor species in this region.For selected days with continental outflow (hatched in Fig. 3(a)), the observed average OA mass concentration increased to 4.2 µg m -3 (or 6.0 µg m -3 from the Q-AMS).Similar increases occurred in the aging run (6.8 µg m -3 ), while levels in the conventional run were almost unchanged (1.7 µg m -3 ), clearly suggesting a larger impact of aging on these days.The factor of increase was similar to values (factors of 2-5) reported by Jathar et al. (2011) for the OA mass concentrations over East Asia.A direct comparison with modeling results from Matsui et al. (2014a), for which a domain average increase of total OA by a factor of 5.4 (24 March-26 April 2009, East Asia) was reported, is not possible, because their "aging-off" simulation was quite different from our reference one (i.e., our conventional run).In particular, our assumptions regarding the emission rates for POAs/IVOCs/SVOCs were modified.Table 2 summarizes the performance of our model using various statistical indices.Our model showed remarkable agreement with Q-AMS data (shown in parentheses in Table 2); in this case, the model/observation median ratio for OA was 0.99 with the aging run, which improved markedly from 0.25 for the conventional run.In fact, the mean bias of -4.1 µg m -3 calculated for the conventional run disappeared in the aging run.

Evaluation of Model Results and Implications for Organic Aerosols
In the case of the Rudong site (Fig. 3(b)), average OA concentrations increased from 4.3 to 7.2 µg m -3 , when chemical aging was taken into account, yielding better comparison with observations (12.3 µg m -3 ).The increase in OA at the Rudong site was smaller than that at the Fukue site, having a factor of 1.7 only.In the time series plot, observed temporal variations were sometimes reproduced using the aging run, particularly during the period June 9-24, 2010.This resulted in a better correlation coefficient (0.81, Table 2).The mean bias of -4.2 µg m -3 for the aging run was still large, but it could be relaxed, when assuming a lower limit for the OA/OC ratio of 1.5 (e.g., Matsui et al., 2009).This is justified because the Rudong site was nearer to pollution sources and therefore less chemical aging was expected than for the other cases (Fukue and Jeju).In addition, the consistency between the summed individual concentrations and the total PM 2.5 would be improved.
At the Jeju site in autumn 2012, the OA levels only slightly increased from 1.4 to 1.5 µg m -3 , when aging was included in the model.These levels were both lower than the observation (3.3 µg m -3 ).At times, for the Rudong and Jeju sites, the aging run yielded lower OA concentrations than the conventional run.Such a magnitude relationship was observed when fresh pollution was relatively important; POA with low saturation concentration (10 -6 µg m -3 ) persisted in conventional runs, whereas in aging runs, a sizable fraction of the aerosols were assumed to be more volatile (included as SVOCs and IVOCs) and were lost to the gas phase as dilution proceeded.Only when oxidation overrode this loss, did OA content increase in aging runs.
For selected days at the Jeju site having continental outflow 14,17,(19)(20)(22)(23)(24)(25)(27)(28)(29), October 31-November 2, and November 4-7 in 2012), identified by the backward trajectory analysis, the average OA concentration levels in the aging run increased to 1.6 µg m -3 , although there was still a large difference to observation levels of 3.3 µg m -3 for those days.Similar to the Rudong site, this gap was diminished, when an OA/OC ratio of lower than 2.0 was assumed for the observations.However, a lower value is not supported by the fact that the gap between the summed concentrations of individual components and the total PM 2.5 levels would become larger.A low sensitivity to an aging effect was also recognized at the Fukue site during the same period (autumn 2012); at this time, the OA levels were quite similar, 1.3 and 1.4 µg m -3 for both aging and conventional runs, with both being lower than observed OA levels (4.7 µg m -3 ).At the Jeju site, the fraction of K + in the total PM 2.5 (1.6%) was larger than for the spring campaigns at the Fukue and Rudong sites (0.5%, 1.3%), indicating the possibility that primary organic species emitted from biomass burning were not adequately taken into account in the model.
Fig. 4 shows average diurnal variations in observed and modeled OA mass concentrations.Generally, increases in the afternoon related to photo-oxidation became evident, when aging was considered.This was most apparent at the Rudong site, which is nearest to the source region; the difference between the daytime maximum and the nighttime minimum for its diurnal variation was 10.1 µg m -3 on average, while values from the aging and conventional runs were 6.4 and 4.5 µg m -3 .The timing of this increase (during 0600-1200 LT) was reproduced in the aging run.At the Fukue site, the observed maximum and minimum difference was 1.1 µg m -3 (1.4 µg m -3 based on Q-AMS data), while values for the aging and conventional runs were 1.0 and 0.1 µg m -3 , respectively.At the Jeju site, the observed increase toward late afternoon (1800 LT) was also captured by the aging run.The observed maximum and minimum difference of 2.1 µg m -3 was better reproduced by the aging run (1.4 µg m -3 ) than the conventional run (0.6 µg m -3 ), suggesting that aging was important also for the Jeju site.
OH oxidation played critical roles in our aging simulation.To date, OH concentrations in model simulations using the VBS scheme have only been evaluated in urban environments (Tsimpidi et al., 2010;Matsui et al., 2014;Morino et al., 2014).However, the ability of a model to represent reasonable OH concentration levels is also important in non-urban environments.Given that OH concentrations in urban areas could vary significantly over time and space, evaluation under rural conditions may be even more meaningful.Here, the OH concentrations in the aging run were compared with those independently estimated using the photochemical box model, constrained by observed levels of ozone, CO, NO, NO 2 , VOCs, OVOCs, and J values, as outlined in the previous section.The OH concentrations from the box model (labeled as RACM2) showed daytime peaks of 7.3 and 3.9 × 10 6 radicals cm -3 , while the WRF-Chem model with aging reached even higher values (1.5 × 10 7 radicals cm -3 ) at the Fukue and Rudong sites (Figs.5(a) and 5(b)).The 24-h averages of 2.2 and 1.3 × 10 6 radicals cm -3 from the box model were lower than values of 4.3 and 4.6 × 10 6 radicals cm -3 from the WRF-Chem model.Given the uncertainty in the box-model-based OH level, which was typically estimated to be ± < 50% (1σ) (e.g., Kanaya et al., 2007Kanaya et al., , 2012)), the WRF-Chem model likely overestimated OH for the Fukue and Rudong sites by at least factors of 1.3 and 2.3, respectively.The WRF-Chem model underestimated concentration levels of CO (median of the model/observation ratio was 0.70 and 0.74 for Fukue and Rudong, respectively), an indicator of OH loss term.This might be a partial cause for the high OH levels.Full characterization of the OH budget will be a subject of future studies.On the other hand, the reproducibility of OH levels at the Jeju site in autumn 2012 (Fig. 5(c)) was good, with 24-h averages of 2.0 and 1.7 × 10 6 radicals cm -3 for the box model and the WRF-Chem model, respectively.
Considering this fact, together with the relatively high k value (up to 4 × 10 -11 cm 3 molecule -1 s -1 ) adopted in our model (i.e., near the collision frequency of neutral molecules in the gas phase), we surmise that the oxidation frequency k[OH] is around its upper limit in the aging run.Therefore, the underestimation of the OA levels at the Jeju and Rudong sites likely indicates underestimation of the primary emission rates for SVOCs/IVOCs.Such underestimation may be related to the rapid economic growth in East Asia till the years of observation since the base year of the adopted emission inventory used in this study.Another possibility is that other oxidation pathways producing organic aerosols, such as heterogeneous oxidation and aqueous phase reactions not represented in the aging run, also play important roles at the Rudong and Jeju sites.The 24-h averages of OH concentrations from the box model were a factor of ~2 higher than the zonal mean OH concentrations (Spivakovsky et al., 2000) of 0.9-1.5 × 10 6 radicals cm -3 in spring and 6.3 × 10 5 radicals cm -3 in autumn for the studied latitudinal band.Considering the 24-h average OH concentrations of 2.2 × 10 6 radicals cm -3 , k of 4 × 10 -11 cm 3 molecule -1 s -1 , and a typical transport time from the continent to the Fukue site of 40 h (Kanaya et al., 2016b), > 10 generations of oxidation could occur for surrogate SVOC/IVOC species.Thus, we conclude that the continental outflow air masses were highly oxidative, irrespective of the fact that the WRF-Chem model tended to overestimate OH concentrations.This is consistent with the high f44 values of ~0.15.Irei et al. (2016) also suggests that high oxidizing capacity in the Fukue Island region has led to efficient formation of secondary organic/inorganic aerosol species.Future studies are necessary to verify real OH levels in this region and to determine a reasonable rate coefficient k.
Left panels of Fig. 6 show the breakdown of the modeled OA into volatility bins for the aging run for all field sites.The basic feature was that the species with saturation concentrations similar to the total OA level were dominant.This is related to partitioning between gas and aerosol phases.However, occasionally highly oxidized species (C* ≤ 10 -1 µg m -3 ) become abundant during chemical aging.Such features could be used for testing the mechanisms in the future, when more advanced OA measurements with volatility segregation are conducted.Right panels of Fig. 6 indicate the average fractions of POA, OPOA, secondary organic aerosols from anthropogenic gases (ASOA), and secondary organic aerosols from biogenic gases (BSOA) for all three campaigns.In all cases, dominance of OPOA and ASOA was suggested, as opposed to the dominance of POA in the conventional run.This is consistent with observations showing that oxygenated organics were dominant in continental outflows (Zhang et al., 2007).A bivariate regression analysis, subdividing OA into hydrocarbon-like OA (HOA) and oxygenated OA (OOA) using the m/z 44 and 57 signals from the Q-AMS, was carried out at the Fukue site.This indicated the dominance of OOA (5.1 µg m -3 or 91%) over HOA (0.58 µg m -3 , 9%), consistent with our aging run results.However, the HOA level was larger than the modeled POA in the aging run (0.11 µg m -3 ), supporting the possibility that the primary emission rate of SVOC/IVOC might be larger.

Evaluation of Model Results and Implications for Other Major Species and PM 2.5
Fig. 7 shows the capacity of our model to reproduce concentrations of other major species (sulfate, nitrate, ammonium, and BC), and their sum for the Fukue site during spring 2009.Fig. 8 shows PM 2.5 comparisons for the Rudong and Jeju sites.Their major chemical species are shown in Figs.S1 and S2.As average concentrations (left panels) show, the aging run did not change the levels of inorganic species or BC.Our model was better at reproducing average levels and temporal variations in these species than for OA.In fact, median model/observation ratios were 0.59-1.4for sulfate, 1.8-3.4 for nitrate, 0.86-1.2for ammonium, Fig. 6.Breakdown of the modeled organic aerosol (OA) into volatility bins for the aging simulation (left panel).Breakdown of the modeled campaign-average OA to primary organic aerosols (POA), oxygenated POA (OPOA), anthropogenic secondary organic aerosols (ASOA), and biogenic secondary organic aerosols (BSOA) (right panel).and 0.72-0.94for BC in all three field campaigns (Table S1).Nitrate concentrations were commonly overestimated, possibly because the uptake of nitrate by larger particles (Itahashi et al., 2016) was not well represented in our model, as emissions of dust and sea-salt particles were not taken into account, or else the related gaseous deposition rate was too small (e.g., Shimadera et al., 2014;Morino et al., 2015).Sulfate at the Fukue and Rudong sites was slightly underestimated by our model, possibly reflecting the underestimated emission rates.
The sum concentrations modeled for these four major inorganic species and OA were added to "others" and water mass concentrations, and compared with total PM 2.5 mass concentrations observed by SHARP or BAM instruments.The "others" category in our model, having values of 1.3, 4.5, and 1.2 µg m -3 for Fukue, Rudong, and Jeju sites, is derived from PM 2.5 that was not categorized into other classes.The average water mass concentrations were estimated to be 2.4, 3.6, and 1.5 µg m -3 for Fukue, Rudong, and Jeju sites; therefore, their contributions to the PM 2.5 mass concentrations were non-negligible.The sums of modeled average PM 2.5 mass concentrations from aging runs had values of 18.6, 39.3, and 11.0 µg m -3 for Fukue, Rudong, and Jeju sites, consistent with observed averages of 21.2, 41.5, and 14.4 µg m -3 (Figs. 7 and 8).This agreement was better than for the PM 2.5 concentrations from conventional runs, yielding 14.6, 36.3, and 10.9 µg m -3 for these sites.Clearly, the median model/observation ratio for PM 2.5 was improved by introducing the chemical aging scheme from 0.67 to 0.83, from 0.91 to 0.95, at Fukue and Rudong sites, respectively (Table 2).The mean biases were also reduced from -6.5 to -2.6 µg m -3 and from -5.3 to -2.2 µg m -3 at these two sites.The correlation coefficients for PM 2.5 in the aging runs were in the range of 0.63 to 0.69, similar to the range for sulfate (0.59-0.70).
In the scatterplots showing observed and modeled PM 2.5 and OA, in which data from all three field campaigns are plotted together (Fig. 9), the slopes of the regression lines increased from 0.69 to 0.76 for PM 2.5 , and from 0.34 to 0.56 for OA through introduction of aging in the model, while high correlation coefficients were maintained (R = 0.77 and 0.76).The slope value for PM 2.5 (0.76) was still lower than unity; large uncertainty in values for studied species (e.g., OA) may be important, although fine-mode fractions of dust and sea-salt particles, not considered in the model simulation, may also be responsible.Thus, we conclude that the capability of the WRF-Chem model to reproduce PM 2.5 and OA levels and their diurnal variations in East Asia was improved by taking chemical aging into account, although details (e.g., OH concentrations, and precursor emission rates) need to be tested in the future studies towards better characterization and representation in the day-to-day variations.

CONCLUSIONS
We conducted three intensive field campaigns around the East China Sea, measuring mass concentrations of total and major chemical components of PM 2.5 at Fukue (Japan), Rudong (China), and Jeju (Korea), under the influence of Fig. 7. Campaign averages and time series of observed and modeled major components of PM 2.5 (a-e), and total PM 2.5 (f) for the Fukue site.continental outflows containing elevated concentrations of aerosols and precursors.Mass closure of PM 2.5 was almost satisfied with observed quantities at all three sites.These results were used as benchmarks to test the regional-scale WRF-Chem/ATRAS-MOSAIC model simulations representing chemical aging, using the VBS scheme.We demonstrated that OA levels were markedly increased, up to a factor of 3.7, through introducing the chemical aging scheme into our model, resulting in better agreement with observations.Although the WRF-Chem model tended to overestimate OH concentrations, compared with levels estimated using a constrained photochemical box model,  we conclude that a highly oxidative atmosphere occurs over the East China Sea.This results in marked increases in OA levels, when chemical aging was taken into account.For data for the Jeju site in autumn, the OA increases were not significant, and the impact of aging was weaker than in spring, potentially reflecting lower OH concentrations.The underestimation of OA levels by our model, even with the chemical aging, suggests the possibilities of unaccounted primary emissions or underrepresented pathways forming OA, other than oxidation in the gas phase.Our model successfully reproduced the sums of major PM 2.5 chemical components other than OA (i.e., sulfate, nitrate, ammonium, and BC).After considering the associated water mass concentrations, the estimated total PM 2.5 mass concentrations in the aging run were compared well with observations, yielding median model/observation ratios of 0.68-0.95for all field campaigns.To our knowledge, implications of increased OA levels, related to the VBS scheme on total PM 2.5 concentrations, have rarely been discussed.We suggest that the VBS scheme could better represent secondary formation of organic aerosols and PM 2.5 .Specifically, identifying oxidation rates of organics by OH and primary emissions of all precursors are the key factors to consider in future studies.

Fig. 1 .
Fig. 1.Locations of the three field sites (Fukue, Rudong, and Jeju) around the East China Sea.The background color map shows average aerosol optical depth measured by a MODIS sensor at 550 nm from Jan 2009 to Dec 2012 (http://giovanni.sci.gsfc.nasa.gov/giovanni/).

Fig. 3 Fig. 2 .
Fig. 2. Stacked bar charts of the observed (obs) and modeled average mass concentrations of major chemical components, others, and water of PM 2.5 for individual field campaigns.Summed mass concentrations are compared with the total PM 2.5 monitoring data using a horizontal dashed black line overlain on the observational bars.Modeled results are for conventional (conv) and aging runs.

Fig. 3 .
Fig. 3. Campaign averages (left) and time series (right) of observed and modeled organic aerosol (OA) mass concentrations for (a) Fukue, (b) Rudong, and (c) Jeju sites.The days having a hatched pattern for the Fukue and Jeju sites are those when air masses were transported from the continent.

Fig. 4 .
Fig. 4. Diurnal variations in the observed and modeled organic aerosol (OA) mass concentrations.The light blue bands represent the range of observed OA, when the OA/OC ratio varied between 1.5 and 2.5.OC = organic carbon.

Fig. 5 .
Fig. 5. Average diurnal variations in OH concentrations at the three field sites.Red circles are the WRF-Chem simulations with aging.They are compared with those estimated using a constrained photochemical box model (blue circles).The zonal mean levels estimated by Spivakovsky et al. (2000) for April to July (gray area) are also shown.

Fig. 8 .
Fig. 8. Campaign averages and time series of observed and modeled total PM 2.5 for (a) Rudong, and (b) Jeju sites.

Fig. 9 .
Fig. 9. Scatterplots showing observed and modeled (a) total PM 2.5 , and (b) organic aerosol (OA) using data from the three field campaigns.Circles and solid lines (least-square regression lines) show data from aging runs, while crosses and dashed lines show conventional runs.

Table 1 .
List of instruments used during the three field campaigns.

Table 2 .
Statistical parameters of correlations between observed and modeled organic aerosols (OA) and PM 2.5 for individual model simulations.Values in round parentheses for Fukue site are for OA measurements with quadrupole aerosol mass spectrometers.Modeled OA concentrations in square brackets are from WRF/CMAQ.MB and NMB are mean bias and normalized MB.Average concentrations are for data pairs (i.e., observation and model simulation) and could be different from campaign averages mentioned in text.