Seasonal variation of fine- and coarse-mode nitrates and related aerosols over East Asia: Synergetic observations and chemical transport model analysis

We analyzed long-term fineand coarse-mode synergetic observations of nitrate and related aerosols (SO4, NO3, NH4, Na, Ca) at Fukuoka (33.52°N, 130.47°E) from August 2014 to October 2015. A Goddard Earth Observing System chemical transport model (GEOS-Chem) including dust and sea-salt acid uptake processes was used to assess the observed seasonal variation and the impact of long-range transport (LRT) from the Asian continent. For fine aerosols (fSO4, fNO3, and fNH4), numerical results explained the seasonal changes, and a sensitivity analysis excluding Japanese domestic 20 emissions clarified the LRT fraction at Fukuoka (85% for fSO4, 47% for fNO3, 73% for fNH4). Observational data confirmed that coarse NO3 (cNO3) made up the largest proportion (i.e., 40–55%) of the total nitrate (defined as the sum of fNO3, cNO3 and HNO3) during the winter, while HNO3 gas constituted approximately 40% of the total nitrate in summer, and fNO3 peaked during the winter. Large-scale dust-nitrate (mainly cNO3) outflow from China to Fukuoka was confirmed during all dust events that occurred between January and June. The modeled cNO3 was in good agreement with observations 25 between July and November (mainly coming from sea salt-NO3). During the winter, however, the model underestimated cNO3 levels compared to the observed levels. The reason for this underestimation was examined statistically using multiple regression analysis (MRA). We used cNa, nss-cCa, and cNH4 as independent variables to describe the observed cNO3 levels; these variables were considered representative of sea salt-cNO3, dust-cNO3, and cNO3 accompanied by cNH4 (cNH4 term), respectively. The MRA results explained the observed seasonal changes in dust-cNO3 and indicated that the 30 dust-acid uptake scheme reproduced the observed dust-nitrate levels even in winter. The annual average contributions of each component were 43% (sea salt-cNO3), 19% (dust cNO3), and 38% (cNH4 term). The MRA dust-cNO3 component had a high value during the dust season, and the sea salt component made a large contribution throughout the year. During the


Introduction
Long-range transboundary transport of dust and pollutants in East Asia, and their complex interactions, is an important environmental issue due to the recent rapid economic developments and changes in these areas (e.g., Lawrence and Lelieveld, 2010;Li et al., 2017;Zhang et al., 2017).Because of the heavy air pollution occurring in China, a great deal of research focuses mainly on fine-mode aerosols (i.e., PM 2.5 , particle aerodynamic diameter < 2.5 µm).NO x emissions have been increasing rapidly over the past decade (e.g., Richter et al., 2005;Irie et al., 2016), and long-range nitrate transport is becoming increasingly important for regional nitrogen budget studies (e.g., Oita et al., 2016;Itahashi et al., 2016Itahashi et al., , 2017;;Uno et al., 2017a, b).A large proportion of nitrate exists in coarse mode (PM c , particle aerodynamic diameter > 2.5 µm) due to interactions with sea salt and mineral dust.The formation of nitrate on dust aerosols has been clearly observed using scanning electron microscopy both in laboratory experiments and using field measurements (Li and Shao, 2009).Acid uptake of pollutants over sea salt and dust surfaces is important when evaluating coarse NO − 3 (cNO − 3 ), as it modifies the chemical lifetime of nitric acid (including atmospheric loading and deposition).Continuous measurement of both fine-and coarse-mode aerosol compositions (including NO − 3 ) is critical for achieving a complete understanding of the fate of air pollutants and changes in the Asian atmospheric environment (Li et al., 2012;Pan et al., 2017;Wang et al., 2017a).
Ground-based and airborne aerosol observations have been studied to determine the physics and chemistry of highconcentration events (e.g., Huebert et al., 2003;Jacob et al., 2003;KORUS-AQ, 2016), but the duration of these observational campaigns is typically less than 1 month, which is insufficient for studying seasonal variation.Monitoring projects conducted by the Acid Deposition Monitoring Network in East Asia (EANET, 2014) and Asian Dust and Aerosol Lidar Observation Network (AD-Net; Sugimoto et al., 2008), among others, have been accumulating observational data for over 10 years, but detailed data on hourly aerosol compositions are rarely captured.Long-term aerosol observations (over at least 1 year), including of aerosol composition and with a high time resolution, are required because seasonal changes in Asian monsoons play an important role in the patterns and frequency of long-range pollutant transport; however, no such detailed observational studies have been undertaken to date.
We have made long-term synergetic observations of the behaviors of aerosols around the Chikushi Campus of Kyushu University, located in the suburbs of Fukuoka City (33.52 • N, 130.47 • E), since October 2013 (Pan et al., 2015(Pan et al., , 2016;;Itahahsi et al., 2017;Uno et al., 2017a, b;Osada et al., 2016).We used a state-of-the-art aerosol observation instrument to measure both fine-and coarse-mode aerosols.In this study, we report seasonal variation in both fine-and coarse-mode atmospheric aerosols based on long-term synergetic aerosol observations made at 1 h intervals in Fukuoka, Japan, from August 2014 to October 2015.During this period, several yellow sand and heavy pollutant transport episodes were observed.This paper reports on the major characteristics of anthropogenic aerosols and long-range dust transport, based on observations and chemical transport model (CTM) analysis.We focused on the seasonal variation in fine-and coarsemode long-range nitrate transport.
This report is structured as follows: Sect. 2 documents the observational dataset, Sect. 3 describes the CTM simulation in detail, Sect. 4 discusses temporal variations at observation sites and the model results, and Sect. 5 provides a summary and conclusions.

Observation
Observations were made on the rooftop (4F) of the Fukuoka Institute of Health and Environmental Science and the Chikushi Campus of Kyushu University, Fukuoka.The horizontal distance between these two sites is about 5 km.Both sites are located in a suburban area of Fukuoka.The anthropogenic activity was very limited at both sites, and the air quality showed similar patterns.

Aerosol chemical speciation analyzer and NH x measurement
A continuous dichotomous aerosol chemical speciation analyzer (ACSA-12 monitor; Kimoto Electric, Osaka, Japan), was utilized to measure PM 10 (particulate matter < 10 µm in diameter) and PM 2.5 (particulate matter < 2.5 µm in diameter) with high temporal resolution (Kimoto et al., 2013).Particulate matter (PM) was collected on a tape filter made of Teflon (PTFE).Hourly observations were conducted to monitor SO 2− 4 , NO − 3 , optical black carbon (BC), and watersoluble organic compounds (WSOC) at Fukuoka.The mass concentrations of PM were determined using the beta-ray absorption method.The ACSA-12 measured NO − 3 and WSOC by an ultraviolet absorption-photometric method and SO 2− 4 by turbidimetry after the addition of BaCl 2 to form BaSO 4 and polyvinyl pyrrolidone as a stabilizer.Optical BC was measured using a near-infrared light scattering method, and the observed data showed a good correlation with the IM-PROVE protocol measurement (Hasegawa et al., 2004).The analytical period was within 2 h of sampling; therefore, the volatilization of particulate NH 4 NO 3 after collection was regarded as minimal compared with the traditional filter-pack observation method.ACSA has been tested previously (Osada et al., 2016) and used to identify aerosol chemical compositions at Fukuoka (Pan et al., 2016;Uno et al., 2017a, b).It should be noted that ACSA-12 measures both fine-and coarse-mode aerosols simultaneously and is an important for mass budget studies and evaluation of CTMs.

The behaviors of NH 3 and NH +
4 are also important because they are the counterions for SO 2− 4 and NO − 3 .The concentrations of gaseous NH 3 and NH + 4 in fine particles were measured with a semi-continuous microflow analytical system (MF-NH 3 A, Kimoto Electric; Osada et al., 2011).Two inlet lines were used to differentiate the total amounts of NH x and particulate NH + 4 after gaseous NH 3 was removed using a phosphoric acid-coated denuder from the sample air stream.The cutoff diameter of the inlet impactor was about 2 µm (which is smaller than the ACSA PM 2.5 cutoff).Secondary inorganic aerosols (SO 2− 4 , NO − 3 , and NH + 4 ) were fully observed using our synergetic monitoring system.

Denuder-filter (D-F) pack method
During our observation period from August 2014 to October 2015, we conducted D-F pack measurements at the Chikushi Campus of Kyushu University, Fukuoka.An annular denuder-multi-stage-filter sampling system was used for HNO 3 and size-segregated aerosol sampling.The sampling interval was 6-8 h for intensive observation and 1-2 days for regular observation.At the inlet, coarse-mode aerosols were removed by nucleopore membrane filters (111114; Nomura Micro Science Co., Ltd., Atsugi, Japan; pore size = 8 µm), and then gas-phase HNO 3 was collected with the annular denuder (2000-30x242-3CSS; URG, Chapel Hill, NC, USA) coated with NaCl (Perrino et al., 1990).Fine-mode aerosols were collected with a PTFE filter (J100A047A; ADVANTEC, Tokyo, Japan; pore size = 1 µm), and a nylon filter (66509; Pall Co., Port Washington, NY, USA) captured volatilized nitrates from the PTFE filter (Appel et al., 1981;Vecchi et al., 2009).The sample air flow rate was 16.7 L min −1 (1 atm, 25 • C).Under these conditions, the aerodynamic diameter of the 50 % cutoff for the nucleopore filter was approximately 1.9 µm (John et al., 1983).The samples were analyzed by ion chromatography (IC).Finemode aerosols (< 1.9 µm in diameter) were underestimated by the D-F pack compared with the ACSA measurement, and coarse-mode aerosols (> 1.9 µm, with no upper limit) were overestimated by the D-F pack compared with the ACSA PM c measurements (2.5-10 µm) due to a difference in cutoff diameter between the methods (Osada et al., 2016).Details of the ACSA data comparison and validation were reported previously by Osada et al. (2016).

Numerical modeling
We used the 3-D Goddard Earth Observing System chemical transport model (GEOS-Chem) (v.09-02) (Bey et al., 2001;Park et al., 2004;Fairlie et al., 2007Fairlie et al., , 2010)).The model was run with the full GEOS-Chem NO x -O x -VOC-HO x -CO chemistry option to simulate the formation of aerosols, including mineral dust, sea salt, and secondary inorganic aerosols (SO 2− 4 , NO − 3 , NH + 4 ).We also modeled the emis-sion/transport of primary BC and organic carbon (OC).However, the detailed carbonaceous species and secondary organic aerosol (SOA) options in GEOS-Chem were not used in this study.Dust in GEOS-Chem is classified according to four size bins (radii of 0.1-1.0,1.0-1.8,1.8-3.0,and 3.0-6.0µm), based on Ginoux et al. (2004).The smallest size bin is further divided into four bins (radii 0.1-0.18,0.18-0.3,0.3-0.6,0.6-1.0µm) for optical properties and heterogeneous chemistry.This model uses the dust entrainment and deposition (DEAD) mobilization scheme from Zender et al. (2003), combined with the source function used in the Goddard Chemistry Aerosol Radiation and Transport (GO-CART) model (Ginoux et al., 2001), as described by Fairlie et al. (2007).Several modifications to capture seasonal changes in dust source function, the dust emission fraction within dust bins, and the change in wet scavenging efficiency that was inversely determined to fully capture Asian dust (Yumimoto et al., 2017) were used in this study.The general performance of our dust simulation was already reported by Yumimoto et al. (2017) and Uno et al. (2017a, b).Sea salt is distributed into two size bins (dry radii 0.01-0.5 and 0.5-8 µm).The sea salt emission scheme is based on the work of Jaeglé et al. (2011) and was well validated by those authors.It includes the effects of sea-surface temperature (SST), as more sea salt is emitted in the summer.
In this study, the reactive uptake of HNO 3 and SO 2 on dust (limited by dust alkalinity) and the uptake of gasphase H 2 SO 4 (limited by competition with other aerosol surfaces) (Fairlie et al., 2010) were used.Dust nitrate (mainly Ca(NO 3 ) 2 ) was simulated based on the heterogeneous reaction between dust and nitric acid, as follows: As described by Fairlie et al. (2010), the first-order reactive uptake rate constant (k) is calculated according to the uptake coefficient (γ ) and surface area density of dust particles (A) using the following equation, suggested by Jacob (2000): where C is the concentration of gas uptake (e.g., HNO 3 ), r is the aerosol particle radius, D g is the molecular diffusion coefficient, and c is the mean molecular speed.The uptake coefficient depends on the ambient relative humidity (RH), and we used the RH-dependent function in Fig. 1 of Fairlie et al. (2010).More details on dust-nitrate formation from a heterogeneous reaction can be found in Fairlie et al. (2010).
A similar heterogeneous reaction to that described by Eq. ( 1) for sea salt was also included in our calculation.
The model used the assimilated meteorological fields from the GEOS of the NASA Global Modeling and Assimilation Office (GMAO).The model has a horizontal resolution of 2  (Olivier and Berdowski, 2001) for the global domain and the Regional Emission Inventory in Asia (REAS, v. 2.1) for the Asian domain, as reported by Kurokawa et al. (2013).The REAS NH 3 emissions were modified to include seasonal variations in Asia based on Huang et al. (2012) and further changes in winter emissions recommended by Xu et al. (2015).Volcanic SO 2 emissions are based on the database of the Japan Meteorological Agency (http://www.data.jma.go.jp/svd/vois/data/ tokyo/volcano.html).The model simulation was conducted from the beginning of December 2013 to the end of October 2015, and results from the first 8 months were used to train the model.Other basic numerical settings were as reported in Uno et al. (2017a, b).
To investigate whether domestic or transboundary air pollution is dominant in Japan, we also performed a sensitivity simulation.Because the quantity of emissions from China was larger than that from Japan, to avoid large nonlinearities in the atmospheric concentration response to emissions variation (e.g., Itahashi et al., 2015) a sensitivity simulation was designed to include a 20 % reduction in all emissions from Japan (defined as JOFF20%).Based on differences between the baseline simulation (CNTL) and the JOFF20% simulation, the domestic contribution from Japan (JOFF) was estimated using a multiple of 5 for differences in CNTL and JOFF20% experiments.We also made a volcanic SO 2 sensitivity simulation without volcanic SO 2 emissions (VOFF) because the SO 2 to SO 2− 4 formation can be assumed to be linear.
Based on the analysis of PM concentrations, PM 2.5 and PM 10 were calculated by summing the individual aerosol (SO 2− 4 , NO − 3 , NH + 4 , BC, and OC), dust, and sea salt components of the model.Hereafter, we denote modeled ammonium nitrate (i.e., NH 4 NO 3 ) as A-NO − 3 , dust nitrate as D-NO − 3 , sea salt nitrate as SS-NO − 3 , and their sum as simply to the Na concentration based on the salinity and Na mass ratio of seawater (Keene et al., 1986).Japanese weather is controlled by changes in Asian monsoons.Under summer monsoon conditions (covered by S-SE wind from the hot and moist air mass of the North Pacific High), air temperature and RH are at their maximum, while during the winter monsoon (N-NW continental cold air outflow) low temperatures and less precipitation are observed in Fukuoka.Maximum wind speeds in excess of 10 m s −1 in summer-fall show the effects of typhoons (Fig. 2a).The precipitation difference between August 2014 and August 2015 is important for examining differences in the NH 3 concentration (August 2014 had more rainfall than August 2015), and we will discuss this in Sect.4.3.4.
The time variation of modeled coarse-mode Na + generally agreed well with observations except during typhoon events.There were five typhoons between July and September 2015, and the modeled cNa + was very high compared with observations.This indicated that the sea salt emission scheme for very strong wind conditions was overestimated.We also observed high precipitation during the typhoon events.Another important difference is that the typhoon in 2014 occurred in the fall (September-October), after the SST had cooled, while in 2015 it occurred in the summer (July-August) when the SST was warm and more sea salt was emitted.

Daily variation and validation of modeling reproducibility
The monthly variation depends on the how well the model can reproduce the daily variation in the long-range transport (LRT) from continental Asia to Japan, as well as sporadic dust transport.In this section, we analyzed this short-term daily variation and sporadic dust phenomena and validated the model performance statistically.The monthly (seasonal) variation of aerosols is discussed in Sect.4.3.The observed PM 2.5 clearly exhibited frequent spike-like peaks from winter to early spring, but the frequency decreased after April according to the frequency of "polluted" cold air outbreaks from the Asian continent.The modeled PM 2.5 reproduced most of the observed variation.The modeled PM 2.5 had a high correlation (PM 2.5_model = 0.58 (PM 2.5_obs ) + 1.16 µg m −3 , R = 0.72) but was underestimated (normalized mean balance (NMB) = −35.7 %) because the observation data included aerosol compositions that were not incorporated into the model (e.g., SOAs, many trace metals).
The observed PM 10 included coarse aerosols (e.g., dust and sea salt), and peaks corresponded to major dust events.Symbols A-G in Fig. 3b are the major dust events that occurred during our observation period (Uno et al., 2017b).The time variation of modeled PM 10 was in good agreement with observations but was consistently underestimated (NMB = −31.3%) for the same reasons as for PM 2.5 (i.e., more coarse aerosols were included in the observed data than in the model) and showed some uncertainty for large dust concentrations (Uno et al., 2017b).
The modeled BC was systematically underestimated from July to December but was at reasonable levels during the winter.The modeled BC also shows a similar underestimation (NMB = −32 %), although R had a smaller value (0.34) than for PM 2.5 .One possible explanation is the underestimation of Japanese domestic BC emissions (e.g., Itahashi et al., 2017).
We can see very good agreement for fine SO 2− 4 , with NMB = −1.5 % and R = 0.68.This indicated that the modeled fSO 2− 4 reproduced most of the observed variation quite well.Figure 3d shows intermittently high SO 2− 4 concentrations.The high SO 2− 4 levels in the winter depended on the frequency of LRT from the Asian continent based on the synoptic weather change that occurred once or twice a week, while high SO 2− 4 levels are usually observed in summer (see also Fig. 4b).This is related to the meteorological conditions, i.e., the high RH in summer.Figure 3d shows high SO 2− 4 levels from the end of July until early August 2015 (designated SVolc in the figure); this was used to demonstrate that the period of volcano impact is important for SO 2− 4 levels.These high SO 2− 4 concentrations were due to Japanese domestic emissions and volcanic SO 2 emissions.Because volcanic emissions are a natural phenomenon, and the day-byday changes are difficult to predict, the modeled SO 2− 4 level during SVolc misses the observed peak SO 2− 4 (see the Appendix for a more detailed analysis).
For fNO − 3 , the daily maximum reached 6-9 µg m −3 and was sometimes higher than the fSO 2− 4 , with the same timing regarding the peaks as for SO 2− 4 .This indicated that fNO − 3 was also controlled by LRT from the Asian continent (as discussed by Itahashi et al., 2017, andUno et al., 2017b): NMB = −8.5 % and R = 0.68.The correlation is the same as for fSO 2− 4 but has a larger NMB than fSO 2− 4 .Coarse NO − 3 had a distinct peak value during dust events A-G (D-F cNO − 3 levels are higher than ACSA cNO − 3 levels due to different upper cutoff limits).The modeled cNO − 3 levels were higher (see Figs. 3c and 4g) during the fall (October to November) because of sea salt NO − 3 and during the dust season (typically from February to June).Coarse NO − 3 takes a larger NMB compared with fine NO − 3 and is underestimated.This can be understood because cNO − 3 depends on the prediction accuracy of dust and sea salt.
As shown in Fig. 3, a regression slope that is less than 1 could be a natural phenomenon owing to representation issues regarding the model and observations.The GEOS-Chem model result is an average value over a 0.5 • × 0.667 • grid, while the observation is conducted over a station.This is another reason why the simulated model results tend to be underestimated at the high end and overestimated at the low end, which results in a slope less than 1.
For all of the components shown in Table 1, the mean fractional bias (MFB) ranges between −51.0 and −8.96 %, and the mean fractional error (MFE) ranges between 47.99 and 85.75 %.Except for fNO − 3 , these results satisfied the model performance criteria (MFB < ±60 % and MFE < ±75 %) proposed by Boylan and Russell (2006).Wang et al. (2017b) also reported a similar score for fNO − 3 for different chemical transport models (CMAQ and NAQPMS) and this may be a limitation of current CTMs.filter to catch the volatilized HNO 3 from the Teflon filter surface.Thus, both ACSA fNO − 3 and D-F fNO − 3 measurements showed consistent concentrations during the summer (average = 0.43-0.68µg m −3 ; see Fig. 3e).From mid-May to October, the Japanese contribution becomes dominant (annual mean Japanese domestic contribution = 53 %).

Coarse-mode NO −
3 Figure 4g shows the clear seasonal cycle of modeled cNO − 3 .The modeled results showed that dust-nitrate concentrations increased during dust episodes.Modeled sea salt nitrate was consistently in the order of 0.5-1.0 µg m −3 as the baseline cNO − 3 , and it was dominant except during the dust season (showing good agreement with observed values).
Table 2 summarizes the comparison of cNO − 3 levels.Except for January to June, sea salt NO − 3 was dominant within cNO − 3 , and the model results were in good agreement with actual observations.From January to June, the ratio of D-cNO − 3 to SS-cNO − 3 ≈ 1 : 1 and the modeled total cNO − 3 corresponded to two-thirds of the observed total cNO − 3 (annual average ratio of D-cNO − 3 to SS-cNO − 3 ≈ 1 : 2.4).From January to June 2015, we observed several dust events (designated A-G).Uno et al. (2017b) described the typical onset of dust events B and C and pointed out underestimation of the modeled cNO − 3 during these cold dust cases.In this paper, we examine dust event G as a case study of warm weather dust-acid uptake validation.
Figure 5 shows the daily changes in dust (colored region) and total D-cNO − 3 (contoured area) horizontal distributions from (a) 12 June to (b) 13 June and the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model back trajectory starting from Fukuoka.It also shows a comparison of modeled and observed (c) fine-and (d) coarse-mode NO − 3 .The dust transport path was very similar to those of dust events B and C, as described by Uno et al. (2017b).For dust events B and C (see Fig. 3f), modeled cNO − 3 was underestimated, but for dust event G modeled cNO − 3 levels were in good agreement with observed levels.Modeled results show that D-cNO − 3 mainly formed over the Yellow Sea and the East China Sea.This indicates that the underestimation of dust NO − 3 over the winter was independent of the dust-acid uptake scheme.These results indicate the importance of further studies on the mechanism of cNO − 3 formation.40 % of total NO − 3 ); this was due to the change in thermal equilibrium between gases and particles.Notably, cNO − 3 was always higher than fNO − 3 (cNO − 3 made up 27-55 % of the total NO − 3 , and it exceeded 45 % in winter).We observed 0.5-1.0 µg m −3 (0.2-0.4 ppb) of HNO 3 even in winter.
The observations showed that fine NH + 4 is higher throughout the year, and the NH 3 gas level exceeded 1 µg m −3 (0.4 ppb), even in winter.This high NH 3 concentration may be influenced by local agriculture and poultry farming 5-10 km south of the observation site.The high NH 3 concentration (see Fig. 4d) in August 2015 (3-4 times higher than in August 2014) might be due to differences in the high temperature (the monthly mean was 28.4 • C, which was 1.7 • C higher than in 2014) and less precipitation (month total was 186 mm, which was 228 mm smaller than in 2014).Roelle et al. (2002) indicated that the NH 3 emission from soil increased exponentially as soil temperature increased, and due to precipitation more soil water fills the pores in the soil matrix and hinders the diffusion of NH 3 from the soil to the air.Their results suggested that our observed meteorological conditions in each year can explain the NH 3 concentration variation.The annual average cNH + 4 / total NH x ratio was 10 %, but it increased to 15 % (JFM average).This indicates that the cNH + 4 counterpart in winter is important for understanding cNO − 3 and cSO 2− 4 .The model results underestimated NH 3 (Fig. 4d) and overestimated HNO 3 (Fig. 4e); however, the modeled total NO − 3 , fNO − 3 and fNH + 4 levels were in good agreement with the observed values (compared with Fig. 4h, g, and c, respectively).The NH 3 emissions inventory is at 25 km resolution and cannot reflect the impact of local agriculture and poultry farming, which has large uncertainty; this results in the underestimation of NH 3 emissions in our study.However, most of the cNO − 3 formation occurred before arrival in Japan (it occurred mainly over the ocean) and was also not very sensitive to local NH 3 emissions.
The present model does not include the formation scheme of cNH + 4 .The counterparts of cNH + 4 can be cNO − 3 and cSO 2− 4 , and this is one of the reasons for cNO − 3 underestimation by the model.It is important to describe the possible reasons for this underestimation to conduct a detailed N budget study, because the cNO − 3 fraction is very large.Another interesting question is whether the modeled acid uptake scheme can explain the observed cNO − 3 levels in dust and sea salt.We used statistical analysis to investigate this.cNa + , non-sea-salt cCa 2+ (nss-cCa 2+ ), and cNH + 4 as independent variables to describe the observed cNO − 3 (cSO 2− 4 was excluded from the MRA due to its strong colinearity with cNH + 4 ).One important point is that the observation site is surrounded by school grounds and a large city park, providing background local dust.We found that the median value of nss-cCa 2+ during the non-dust season was 0.25 µg m −3 , and the residence time of local dust coming to the observation site for producing cNO − 3 was short (< 1 h).We excluded observations less than nss-cCa 2+ < 0.25 µg m −3 from the MRA.
Figure 8 shows the monthly averaged MRA results and clearly explains the observed seasonal variation.Table 2 summarizes each term and shows the comparison with CTM.The average annual contribution of each term was 40 % for SS-cNO − 3 , 20 % for D-cNO − 3 , and 40 % for cNH + 4 term.The D-cNO − 3 value was high during the dust season, and the sea salt component made the largest contribution throughout the year.The cNH + 4 term made a large contribution during the winter, and one possible reason for this is the condensation/coagulation of small NH 4 NO 3 and (NH 4 ) 2 SO 4 particles onto large particles (e.g., sea salt and dust).The annual average D-cNO − 3 to SS-cNO − 3 ratio in MRA was 1 : 2, which is close to the modeled ratio (1 : 2.4), as shown in Table 2.
Our results suggest that inclusion of aerosol microphysical processes (such as condensation and coagulation of the fine anthropogenic aerosols NO − 3 and SO 2− 4 onto the coarse particles) is important for exploring the observed cNO − 3 concentrations.A CTM incorporating advanced particle microphysics is one potential option (Yu and Luo, 2014).Such a modeling approach, incorporating interactions with mineral dust and sea salt, has not yet been fully explored in East Asia and is a future research direction.

Conclusions
Long-term synergetic fine-and coarse-mode aerosol observations were analyzed at 1 h intervals in Fukuoka, Japan, from August 2014 to October 2015.A GEOS-Chem chemical transport model including dust and sea salt acid uptake processes was used for detailed analysis of observation data to understand the effects of LRT from the Asian continent.The findings from this study can be summarized as follows:

4Figure 1 .
Figure 1 shows the location of the observation sites, Fukuoka and the Mt.Sakurajima volcano, and the anthropogenic SO 2 emission distribution used in the model calculation.Figure 2 shows (a) the daily mean and maximum wind speed, (b) the daily mean temperature, RH, and precipitation in Fukuoka, as observed by the Japan Meteorological Agency, and (c) the observed coarse-mode Na + by D-F, and GEOS-Chem-simulated coarse-mode sea salt.The GEOS-Chemsimulated coarse-mode sea salt concentration was converted

Figure 3
Figure 3 shows the temporal variation in (a) PM 2.5 , (b) PM 10 , (c) optical BC, (d) fine SO 2− 4 , (e) fine NO − 3 , and (f) cNO − 3 observations at Fukuoka.The ACSA observations are indicated in blue; D-F pack observations are indicated by the red line (spike-like high peaks during the dust event are due to large particles > 10 µm, which is the cutoff in ACSA sampling).The figures also include the correspond-

Figure 3 .
Figure 3. Daily average (a) particulate matter < 2.5 µm diameter (PM 2.5 ), (b) particulate matter < 10 µm diameter (PM 10 ), (c) black carbon (BC), (d) fine SO 2− 4 , (e) fine NO − 3 , and (f) coarse NO − 3 .The blue line and dots show aerosol chemical speciation analyzer (ACSA) observations, the red line indicates measurements from the denuder-filter pack (D-F) method, and gray shading indicates the model simulation.D-F data are sampling period averages.Observed data are on the left axis and numerical results are on the right axis, scaled by the regression results (except for panel c).Symbols A-G in panel (b) indicate major dust events in Fukuoka.

Figure 4 Figure 5 .
Figure 4 shows the monthly average (a) PM 2.5 , (b) fine-and coarse-mode SO 2− 4 , (c) fine and coarse NH + 4 , (d) NH 3 gas, (e) HNO 3 , (f) total NO − 3 , (g) coarse NO − 3 , and (h) fine NO − 3 levels.The box-and-whisker plots show the observed fine aerosol levels (10th, 25th, 50th, 75th, and 90th percentile values are marked).The observed average monthly coarse-mode aerosols are shown by the red dashed line.The modeled average monthly fine aerosol concentration for the CNTL is depicted by the straight black line and for the JOFF by the black dashed line.

Figure 6 .
Figure 6.Monthly average values for (a) NH x , (b) NO 3 , and (c) the mass fraction (%) of NO 3 , observed by D-F.

4. 4 Figure 7 .
Figure 7. (a) D-cNO − 3 estimated by multiple regression analysis and modeled D-cNO − 3 ; (b) the same parameters for sea salt NO − 3 .Modeled results averaged over the same time period as for the D-F measurements.

Figure A2 .
Figure A2.Daily average variation in observed (gray shading) and modeled fine SO 2− 4 (blue line is CNTL and red line is VOFF).
S-55 • N, 70-150 • E), both having 47 vertical levels from the surface to 0.01 hPa.The lowest model layer thickness was approximately 130 m.We used anthropogenic emissions data from the Emission Database for Global Atmospheric Research(EDGAR, v. 3) • × 2.5 • for global runs and 0.5 • × 0.667 • for Asian one-way nesting runs (11 •

Table 1 .
Statistical summary of comparisons of the model results with observations between August 2014 and October 2015.

Table 2 .
Comparison of coarse-mode NO − 3 levels (µg m −3 ) ACSA observation.b D-F observation.c Average level between September 2014 and July 2015 (excluding typhoon period). a