Modeling the Long-Range Transport of Particulate Matters for January in East Asia using NAQPMS and CMAQ

Two regional chemical transport models were applied to simulate high concentrations of particulate matters (PM) observed in East Asia in January 2015; the first model is the Nested Air Quality Prediction Modeling System (NAQPMS) and the second is the Community Multi-scale Air Quality Model (CMAQ). The variation of PM2.5 in both models showed well agreement with measurements over both eastern China and western Japan. Based on the model results and the aerosol compositions observed in Fukuoka in western Japan, three types of PM long-range transport (LRT) were identified: N-, S-, and D-type. The N episode showed higher fine-mode nitrate (fNO3) concentrations than fine-mode sulfate (fSO4), indicating the importance of NO3 LRT. The S episode showed the highest fSO4 concentrations (28.9 μg m), which were 3.4-fold higher than fNO3, due to high relative humidity. During the D episode, dust stagnated in Fukuoka for three days, due to the influence of lowand high-pressure systems; thus, dust LRT is also important in winter besides spring. Both models reasonable explained variations in aerosol components during both N and S episodes; however, both underestimated fSO4 especially during D episode, suggesting that they may miss certain emissions or chemical mechanisms. High coarse-mode NO3 (cNO3) concentrations (maximum: 6.3 μg m), and high cNO3/fNO3 ratios (maximum: 1.2) were observed during D episode. NAQPMS successfully captured this cNO3 peak after including heterogeneous reactions on dust. Our results emphasize the importance of such heterogeneous processes for understanding the LRT of dust and anthropogenic pollutants over East Asia.


INTRODUCTION
The long-range transport (LRT) of particulate matters (PM, e.g., mineral dust and anthropogenic aerosols) is an important environmental issue, especially in East Asia, where the emissions of both dust and anthropogenic pollutants are in large mount.Various intensive field studies were commissioned to resolve the transport and chemical evolution of dust and anthropogenic aerosols from the Asian continent; these included the Aerosol Characterization Experiments over Asia (ACE-Asia) (Huebert et al., 2003;Seinfeld et al., 2004) and the Transport and Chemical Evolution over the Pacific (TRACE-P) (Carmichael et al., 2003;Jacob et al., 2003).Both ACE-Asia and TRACE-P studies were conducted in the spring, during which time the prevalent winds were from the west, and dust occurred frequently over East Asia.In addition to these studies, numerical transport models have been used to illustrate the LRT of Asian dust during spring, such as the Chemical Weather Forecast System (CFORS) (Uno et al., 2008;Itahashi et al., 2010) and the Nested Air Quality Prediction Modeling System (NAQPMS) (Li et al., 2012).However, few studies have examined the LRT of dust in winter.
In recent years, severe anthropogenic aerosol pollution has become a critically important issue in China; specific instances of severe pollution usually occur in winter months, such as the pollution episode in January 2013 (Uno et al., 2014;Wang et al., 2014).Therefore the LRT of anthropogenic aerosols during the winter in East Asia is important, but not well understood.Sulfate (SO 4 2-) is an important component of anthropogenic aerosols that has drawn much attention.The first phase of Model Inter-Comparison Study in Asia (MICS-Asia Phase I) revealed the importance of SO 2 and SO 4 2-LRT over East Asia, involving eight LRT models (Carmichael et al., 2002).A recent modeling study determined that LRT drove the domination of SO 4 2-aerosols over west Japan during two air pollution episodes, even in summer when southern winds usually blow over Japan (Itahashi et al., 2012).
Compared with SO 4 2- , there is less research focus on the LRT of nitrate (NO 3 -), as the atmospheric lifetime of NO 3 is shorter than SO 4 2-.In addition, NO 3 -is produced via a reversible reaction between nitric acid (HNO 3 ) and ammonia (NH 3 ), and is thus more difficult to accurately measure than SO 4 2-.However, emission control measures in China instituted in 2006 have led to a dramatic decrease in SO 2 concentrations (Itahashi et al., 2014); since this time, NO 3 has become more prevalent, especially in winter when low temperatures favor the formation of NO 3 -.Recently, Itahashi et al. (2017) used the Community Multi-scale Air Quality Model (CMAQ) to demonstrate the importance of relative humidity (RH) towards SO 4 2-and NO 3 -pollution over East Asia in winter; further modeling studies are necessary to confirm such behavior, as the results of a single model may contain large uncertainties.
In this study, we applied two regional chemical transport models to simulate high concentrations of particulate matter less than 2.5 µm in diameter (PM 2.5 ) that occurred in East Asia in January 2015.The performance of the two models was evaluated by comparing model results to PM 2.5 and PM 10 observations over eastern China and western Japan, as well as fine and coarse mode secondary inorganic aerosols (SIA) measured at Fukuoka, Japan.Based on model results and observations, we found that three types of pollution caused the three high PM 2.5 episodes observed over western Japan.We also showed LRT patterns during the three pollution episodes using model-simulated horizontal aerosol distributions and backward trajectories.

Observation
Recordings of aerosol concentrations were conducted in the metropolitan area of Fukuoka, which is the fourth largest metropolitan area in Japan and the largest city on the island of Kyushu.Mass concentrations of PM, SO 4 2-, NO 3 -, and water-soluble organic compounds (WSOC) in both fine (particle diameter [Dp] < 2.5 µm) and coarse mode (2.5 µm < Dp < 10 µm) were measured using a continuous dichotomous aerosol chemical speciation analyzer (ACSA-12, KIMOTO Electric Co., Ltd) at 1-h time intervals (Kimoto et al., 2013) on the rooftop (4F) of the building housing the Fukuoka Institute of Health and Environmental Science (FIHES, Longitude: 130.48°E, Latitude: 33.51°N).The mass concentrations of PM were determined via beta-ray absorption.Mass SO 4 2-concentrations were determined using the BaSO 4 -based turbidimetric method, with the addition of BaCl 2 dissolved in a polyvinylpyrrolidone solution.The mass concentrations of NO 3 -and WSOC were determined using an ultraviolet absorption-photometric method.
In addition to these analytes, NH 4 + is also important, as it is a counterion for both SO 4 2-and NO 3 -.The concentration of NH 4 + was measured using a semi-continuous microflow analytical system (Kimoto Electric Co. Ltd., MF-NH 3 A, (Osada et al., 2011)).This system was located at the Chikushi Campus of Kyushu University (CCKU, Longitude: 130.47°E, Latitude: 33.52°N), ~5 km northwest of the ACSA-12 site (FIHES).Gaseous NH 3 was removed from the sample stream using a phosphoric acid-coated denuder.Then atmospheric NH 4 + was dissolved in ultrapure water using a continuous air-water droplet sampler and quantified by fluorescence.The cut-off diameter at the inlet impactor was ~2 µm, smaller than ACSA PM 2.5 cut-off.
The vertical distribution of aerosol extinction coefficients was continuously observed using a depolarized multiwavelength Mie-Raman lidar system, which was also located at CCKU (Hara et al., 2016).This location is one site within the Asian Dust and Aerosol Lidar observation network (AD-Net) (Sugimoto et al., 2008).The contribution of anthropogenic and dust aerosols to the extinction coefficient was estimated according to the depolarization ratio (DR), which assumes that anthropogenic (DR = 0.02) and dust (DR = 0.35) aerosols are externally mixed (Sugimoto et al., 2003).
The size and shape character of aerosol particles (0.5 µm < Dp < 10 µm) were measured using a newly developed polarization optical particle counter (POPC) (YGK Corp., Yamanashi, Japan, (Kobayashi et al., 2014)) located at CCKU (Pan et al., 2015).POPC uses a linearly polarized laser source at 780 nm and measures both forward and two compounds (perpendicular [S] and parallel [P] with respect to the scattering angle plane) of backward scattering intensity.The particle size is thus determined from the forward scattering intensity.The DR, defined as the fraction of the perpendicular polarization direction over total backward scattering, ([S/(S + P)]), is a good indicator of whether a particle is spherical.Particles that are both small in size (< 1 µm) and DR (< 0.2) are thus identified as anthropogenic aerosols; particles that are larger (> 3 µm) with a DR > 0.1 are recognized as dust aerosols (Pan et al., 2016).
In addition to observations in Fukuoka, hourly PM 2.5 concentrations were also measured on Goto Island and Tsushima Island, which are located to the west and northwest of Kyushu Island.These islands do not have significant anthropogenic emission sources and are thus regarded as remote sites.To evaluate aerosol concentrations simulated by our models, we also used hourly aerosol observations conducted by the China National Environmental Monitoring Center (http://106.37.208.233:20035)from three cities (Beijing, Qingdao, and Shanghai) in eastern China; the locations of these observation sites are shown in Fig. 1.

Chemical Transport Models
Two regional chemical transport models were used in this study: the NAQPMS (Wang et al., 2001;Li et al., 2012;Chen et al., 2017;Wang et al., 2017) and the CMAQ (Byun and Schere, 2006).All configurations selected for both NAQPMS and CMAQ simulations are summarized in Table 1.NAQPMS used a horizontal resolution of 45 km over East Asia (Fig. S1(a) in supplementary material), with 20 vertical layers spaced in a non-uniform manner using a sigma coordinate.The first CMAQ domain covered East Asia with an 81-km horizontal resolution; the nested domain covered eastern China and the whole of Japan at a 27-km horizontal resolution (Fig. S1(b)), with a 37-layer vertical grid for sigma-pressure coordinates.Both NAQPMS and CMAQ used the same aerosol thermodynamic module (ISORROPIA, (Nenes et al., 1998)) and aqueous chemistry module (RADM, (Chang et al., 1987)), but applied different gas-phase chemistry mechanisms.The gas-phase mechanism used for NAQPMS was the carbon-bond mechanism Z (CBM-Z, Zaveri and Peters, 1999), while the SAPRC99 mechanism was used within CMAQ (Carter, 2000).In the present study, the CMAQ model did not include mineral dust, while NAQPMS used a size-segregated dust module in which dust particles were separated into four bins according to size: 0.43-1 µm, 1-2.5 µm, 2.5-5 µm, and 5-10 µm, and heterogeneous reactions on dust particles were also included within the model (Li et al., 2012 and Table S1 in supplementary material).NAQPMS explicitly calculated the amounts of sulfate and nitrate produced from heterogeneous reactions.Sea salt emissions in NAQPMS were calculated online following Athanasopoulou et al. (2008), and were divided into 4 size bins same as dust.Sea salt emissions in CMAQ were calculated online based on the methodology in Kelly et al. (2010), and were treated on three size bins of the Aitken mode (< 0.1 µm), the accumulation mode (0.1-2.5 µm), and the coarse-mode (> 2.5 µm).NAQPMS considered aerosols from oxidation of DMS, while CMAQ did not consider DMS.Anthropogenic emissions within CMAQ were obtained from the Regional Emission Inventory in Asia (REAS) version 2.1 (Kurokawa et al., 2013), while NAQPMS used a mosaic Asian anthropogenic emission inventory for MICS-Asia and HTAP projects (MIX, version 1.1) (Li et al., 2017).Emissions in January 2008 and January 2010 were used in CMAQ and NAQPMS.
The largest difference between the two inventories is that NO x emissions in MIX were 13.8% higher than that within the REAS over China, but 13.4% lower over Japan.SO 2 emissions in MIX were 8.8%, 12.3% and 2.5% lower than REAS over China, Japan and South Korea, respectively.Differences in NH 3 emissions between the two inventories were less than 5% for all three East Asian countries.Meteorological variables used in both chemical transport models were provided by the Weather Research and Forecasting model (WRF, Skamarock et al., 2008), noting that different versions were used for different models (Table 1).The two WRF simulations used the same PBL scheme, land surface schemes, and cumulus parameterization, but used different cloud microphysics and radiation schemes.Boundary and initial conditions of WRF were based on final analysis data (FNL) from the United States National Center for Environmental Prediction (US-NCEP), which was also used in gridded nudging.Both WRF simulations showed good agreement with observed data, for example, the comparison of relative humidity (RH) was shown in Figs.S2 and 5. WRF slightly underestimate the observations at upwind sites (Beijing and Qingdao) during episode N, especially at the night of January 9 and 10.
The simulation period was from January 1st-23rd, 2015, and the first 6 days were discarded as model spin-up time.The 72-h backward trajectory starting from Fukuoka was calculated using HYSPLIT (Stein et al., 2015) to investigate the air mass origin for episodes of high PM 2.5 .

Temporal Variations in PM 2.5 for Eastern China and Western Japan
Fig. 2 shows the PM 2.5 concentrations in Beijing, Qingdao, and Shanghai in eastern China, and Fukuoka, Goto, and Tsushima in western Japan.Table 2 shows the statistical analysis used to determine model reproducibility.Fig. 2 and Table 2 indicate a high correlation between observations and the two models over all sites, noting a correlation coefficient (R) greater than 0.79 for all paired PM 2.5 datasets.Mean fractional bias (MFB) and mean fractional error (MFE) of CMAQ and NAQPMS satisfied model performance criteria (MFB ≤ ± 60%, and MFE ≤ + 75%) proposed by Boylan and Russell (2006).These results indicated that observed concentrations and variations in concentration for PM 2.5 in eastern China and in western Japan are well explained by both models, despite differences in model frameworks (Table 1).
Both models well reproduced the variability in PM 2.5 concentration, but underestimated the observed peak concentrations (446.9 µg m -3 ) in Beijing.The highest concentration simulated by CMAQ was 389.4 µg m -3 , while that by NAQPMS was 308.8 µg m -3 .In Qingdao and Shanghai, both models successfully captured the variability and peak PM 2.5 concentrations during each episode.The observed underestimation of PM 2.5 in Beijing may be due to the difficulty in capturing its complex topography within simulations, noting that the terrain in the Qingdao and Shanghai regions is relatively flat (Fig. 1).
Three high PM 2.5 concentration episodes were identified in western Japan (Fig. 2(d)).The first episode ('1st' in Fig. 2(d)), which lasted from 17:00 LT (local time, = UTC (Coordinated Universal Time) +09:00) on January 9 to 21:00 on January 11, had a maximum concentration of 86.4 µg m -3 at Fukuoka and 105.1 µg m -3 at Goto Island.The second episode ('2nd' in Fig. 2(d)) was shorter, lasting from 17:00 on January 16 to 6:00 on January 17, with a maximum concentration of 106.2 µg m -3 at Fukuoka and 104.8 µg m -3 at Goto Island.The third episode ('3rd' in Fig. 2(d)) lasted from 4:00 on January 19 to 3:00 on January 22 and had a maximum concentration of 57.1 µg m -3 at Fukuoka and 59.9 µg m -3 at Goto Island; this episode had the lowest peak concentration, compared with the first and second episodes, but was the longest in duration.
The first two episodes were well reproduced by both models.However, the CMAQ model underestimated PM 2.5 concentrations during the third episode due to the omission of dust.The average PM 2.5 concentration measured during this episode was 43.3 µg m -3 in Fukuoka; the CMAQ model estimated this as 24.7 µg m -3 , similar to the anthropogenic PM 2.5 estimated by NAQPMS (23.1 µg m -3 ).When dust is integrated into the model, the total PM 2.5 of NAQPMS was 40.5 µg m -3 , which is in good agreement with measured concentrations, which means about 43.0% of the PM 2.5 was contributed by dust aerosols.Therefore, due to the mixing of dust with anthropogenic pollutants, the daily-averaged mass concentration of PM 2.5 over three days (January 19-  '2nd' and '3rd' in (d).
21) exceeded the Japanese National Ambient Air Quality Standard (NAAQS; 35 µg m -3 ).This indicates that mineral dust may also be an important aerosol component in winter, as well as the spring.

Characteristics of the Three Air Pollution Episodes in Fukuoka
We used observations of lidar, POPC, and ACSA in Fukuoka to reveal characteristics of the air pollution episodes observed in western part of Japan.Fig. 3 shows the vertical distribution of mineral dust and anthropogenic particle extinction coefficients measured using lidar.Fig. 4 shows aerosol volume distribution as a function of DR and size (observed using the POPC), as well as the percentage of each species within the fine-mode secondary inorganic aerosols (fSIAs) during all three episodes.During the first episode, the anthropogenic aerosol extinction coefficient (< 0.15 km -1 ) dominated the dust extinction coefficient (< 0.06 km -1 ), as observed by lidar ('1st' in Fig. 3).At January 11 13:00 during the first episode (Fig. 4(a)), when PM 2.5 concentration is highest (86.4 µg m -3 ) in Fukuoka, POPC also showed the anthropogenic aerosol (small size and DR) was dominant and mixed with a small amount of dust aerosols (large size and DR).Aerosol component observations also suggest that anthropogenic SIA is important, as it accounted for 48.5% of the PM 2.5 .Among three SIA species, NO 3 -was the most prevalent (42%), while fractions of SO 4 2-(34%) were 8% lower.This underscores the importance of NO 3 -LRT during the winter which has not been the focus of previous work (Itahashi et al., 2012;Uno et al., 2014).We therefore refer to the first episode as 'type N'.
The second episode was distinctive from this first episode.Lidar results ('2nd' in Fig. 3) showed that the dust extinction coefficient was very small (about 0.01 km -1 ); the anthropogenic aerosol extinction coefficient reached 0.2 km -1 , suggesting that this episode was driven by anthropogenic pollution.Lidar also showed high sphere extinction coefficient before the second episode due to aerosol hygroscopic growth under high RH condition (Fig. 5(a)).POPC also resulted in a high volume of small anthropogenic particles with low DR, with almost no dust aerosols (Fig. 4(b)).Sulfate accounted for 65% among the three SIA species, and was ~3.4-fold greater than nitrate.This episode also resulted in the highest SO 4 2-concentration (28.9 µg m -3 ) over the entire period, and is thus referred to as 'type S' within this study.
The third episode showed high dust extinction coefficients that lasted for about three days, with a maximum of ~0.1 km -1 according to lidar results ('3rd' in Fig. 3).During this period, the anthropogenic extinction coefficient was low (< 0.1 km -1 ) on January 19, and increased to ~0.15 km -1 on January 21.The POPC (Fig. 4(c)) clearly showed a large volume of dust aerosols mixing with anthropogenic aerosols.The ratio of SIA to PM 2.5 was 30.1% at 12:00 on January 19, which was lower than during episodes N and S due to  high concentrations of dust within the PM 2.5 .This confirms that dust was an important component in PM 2.5 during this episode, as simulated by NAQPMS.We refer to this third episode as 'type D' within this study.).Both models reasonably reproduced variations in fSIA, as well as its discrete components, especially during the three episodes of high fSIA.The correlation coefficient (R) between fSIA observations and both models was greater than 0.8 (Table 2).

Long-Range Transport of Anthropogenic Pollutants during Type N and S Episodes
Both the MFB and MFE of CMAQ and NAQPMS satisfied model performance criteria previously proposed (Boylan and Russell, 2006).
Although both NAQPMS and CMAQ well reproduced variations in fSIA, both models underestimated fSO 4 2concentrations (Table 2), which is a common problem of air quality models especially in winter time, e.g., WRF-Chem (Huang et al., 2014) and GEOS-CHEM (Uno et al., 2017a).There would be several reasons for sulfate underestimation, e.g., inaccurate emission or meteorology field, missing mechanisms.During episode N, WRF slightly underestimate RH at upwind sites (Beijing and Qingdao) at the night of January 9 and 10 (Fig. S2), which may be one reason for underestimation of sulfate.However during episode D, the simulated fSO 4 2-by the two models were both significantly lower than the observations, while the models did not underestimate RH (Figs.S2 and 5).The observed average concentration of fSO 4 2-was 8.9 µg m -3 in episode D, while NAQPMS and CMAQ simulations of anthropogenic fSO 4 2indicated 3.2 µg m -3 and 2.9 µg m -3 , respectively.After considering soil-derived fSO 4 2-(4% of fine mode dust (Wu et al., 2012)) and heterogeneous reactions on dust particles, NAQPMS simulated a total fSO 4 2-concentration of 4.8 µg m -3 , 42.7% lower than the observation.This indicates that current models may miss certain emissions of SO 2 or chemical mechanisms that promote the conversion of SO 2 to SO 4 2-when dust is present.For example, smog chamber results show that NO 2 and SO 2 have a synergistic effect when they react at the surface of mineral dust, leading to the rapid conversion of SO 2 to SO 4 2- (He et al., 2014).From Fig. 5, it is clear that both observations and model results showed higher fNO 3 -than fSO 4 2-during the type N episode, with higher fSO 4 2-than fNO 3 -during the type S episode, as discussed previously.To reveal the importance of LRT on high concentrations of fSIA during type N and S episodes, the spatial distributions of fSIA and the fSO 4 2-/fSIA ratio simulated by NAQPMS and backward trajectories from Fukuoka were shown in Fig. 6.The spatial distribution in fSIA showed similar patterns during the two episodes, in which high-concentration regions of fSIA stretched from the eastern coastline of China to the East China Sea and western Japan.These spatial distribution patterns showed the outflow of fSIA from continental Asia to western Japan, which is consistent with corresponding PM 2.5 peaks at both Qingdao and Japan (Fig. 2).
The fSO 4 2-/fSIA ratio increased from eastern China to western Japan, which is due to the irreversible formation of (NH 4 ) 2 SO 4 and decomposition of NH 4 NO 3 when concentrations of NH 3 and HNO 3 decrease during transport.Backward trajectories during both type N and S episodes showed similar patterns, in which an air mass left the south Shandong province in China and reached Fukuoka within 24 h.As the distance from the Chinese coastline to Fukuoka is ~1000 km, the speed of the air mass must have exceeded 40 km h -1 .
Fig. 6(a) shows the spatial distribution for the type N episode, during which time the air mass was located over China.A high concentration of fSIA (> 100 µg m -3 ) occurred over the east coast of China before the air mass moved to Fukuoka, whereas the fSO 4 2-/SIA ratio was less than 0.25 over the eastern part of China and the East China Sea.When the air mass arrived in Fukuoka, the fSO 4 2-/fSIA ratio was ~0.25 over western Japan (Fig. 6(b)), which meant that (NH 4 ) 2 SO 4 (1.4 times that of SO 4 2-) comprised ~35% of the fSIA, while NH 4 NO 3 was ~ 65% of the fSIA (Fig. 4(a)).
During the type S episode, model results showed that the atmosphere above Fukuoka had an fSO 4 2-/fSIA ratio of more than 0.5 (Fig. 6(d)); thus, (NH 4 ) 2 SO 4 was more than 70% that of the fSIA while NH 4 NO 3 was less than 30% (Fig. 4(b)).Fig. 6(c) shows the spatial distribution of the air mass when it was located over China, during which time high concentrations of fSIA (> 100 µg m -3 ) occurred, similar to the type N episode.The fSO 4 2-/fSIA ratio increased when the air mass arrived at Fukuoka compared with that in China, suggesting that the production of fSO 4 2-occurs quickly during the transport process.
Values of RH were significantly higher (~100%) during the beginning of the S-type episode compared with the Ntype episode (< 70%; Fig. 5(a)).This signifies that SO 4 2formed quickly through aqueous-phase reactions under high RH conditions.More details regarding the differences between these two episodes, outlined using the CMAQ model, can be found in Itahashi et al. (2017).

Mechanism of Dust LRT in Fukuoka during the D-Type Episode
Fig. 7 shows the simulated spatial distribution of mineral dust and backward trajectories from Fukuoka during the D-type episode.Northeast China was controlled by a lowpressure system at 12:00 on January 18 (L1 in Fig. 7 during which time southern China was controlled by a large high-pressure system (H1).Northern China (including the provinces of Inner Mongolia, Shaanxi, Shanxi, and Hebei) were between the two weather systems, and covered by strong northwestern winds and mineral dust at maximum concentrations of more than 150 µg m -3 .Backward trajectories at Fukuoka from 12:00 on January 19 (line with squares) and January 20 (line with circles), are shown in Fig. 7(a); the air mass was in Northern China at 12:00 on January 18, which then moved quickly and directly to Fukuoka within 24 h.This distance is ~1500 km, meaning that the speed of the air mass was greater than 60 km h -1 .
After this, the low-pressure system (L1) moved to northeastern Japan, and the high-pressure system (H1) controlled over eastern China and the East China Sea (12:00 on January 19; Fig. 7(b)).South Korea and western Japan were between these two systems and were thus covered by a plume of dust transported from Asia through strong northwestern winds.The backward trajectory from Fukuoka at 12:00 on January 20 (line with circles) also indicated that the air mass was on South Korea's western edge at 12:00 on January 19, at which time the dust concentration was over 40 µg m -3 .This air mass was then quickly transported to northern Kyushu Island within 12 h, after which time it moved slowly to Fukuoka over another 12 h.
At 12:00 on January 20, the high-pressure system (H1) moved to the northeast to cover a large part of eastern Asia, including northeast China, Korea, and western Japan (Fig. 7(c)).Winds within the Kyushu area were low while under the influence of this high-pressure system, which slowed the movement of the air mass after 00:00 on January 20.Thus, high atmospheric concentrations of dust were stagnant over the Kyushu area.
At 12:00 on January 21, the high-pressure system (H1) moved to the northeast of Japan and an inverted trough (L2) formed at the southern end of Kyushu Island (Fig. 7(d)).The dominant wind direction over Kyushu Island was southeast due to the influences of the two weather systems; thus, the dust plume moved slowly to the northwest.The backward trajectory from Fukuoka at 12:00 of January 21 (line with triangles) indicated that the air mass had already reached the south Kyushu Island at 00:00 of January 20, and stayed on the island for 36 h to reach Fukuoka.The daily variation of dust spatial distribution implied the direct transport of mineral dust from continental Asia to the downwind regions, and the dust plume was stagnant over western Japan for three days under the control of high pressure and low pressure systems.

Importance of Heterogeneous Reactions on Mixing of Dust and Anthropogenic Pollutants during LRT
Fig. 8 shows the variations of coarse mode PM (PMc) in Beijing and Fukuoka, as well as coarse mode NO 3 -(cNO 3 -) and SO 4 2-(cSO 4 2-) in Fukuoka.Mineral dust is an important aerosol of PMc; however, the CMAQ model in this study did not include it.Thus, we only show the simulation results of NAQPMS.NAQPMS well reproduced the temporal variations of coarse-mode aerosols in Beijing, but underestimated PMc in Fukuoka.After considering the dust component, NAQPMS simulated PMc increased by 5.5 µg m -3 (46.2%) to 11.9 µg m -3 , and was 3.6 µg m -3 (21.2%) lower than observations (Table 2).MFB and MFE were -49.4% and 74.0%respectively, and satisfied the model performance criteria proposed by Boylan and Russell (2006).From Fig. 8, it can be seen that during episode D, the high concentration of dust lasted about 1.5 days in Beijing and about 3 days in Fukuoka; the transport mechanism was discussed earlier.The model showed good agreement with observed cSO 4 2-measurements, but slightly underestimated cNO 3 -; this may be due to underestimation of the uptake of HNO 3 on sea salt particles, and without considering anthropogenic NO 3 -in coarse mode.The observed concentration of cSO 4 2-with a maximum of 2.4 µg m -3 was significantly lower than that of fSO 4 2-, while the maximum ratio of cSO 4 2-to fSO 4 2-was 0.3, indicating that fSO 4 2-was much more important for SO 4 2-.However, NO 3 -was quite different.The observed cNO 3 showed a high concentration, with a maximum of 6.3 µg m -3 , and the maximum ratio of cNO 3 -to fNO 3 -was 1.2 during episode D, indicating that cNO 3 -could be more important than fNO 3 -during the dust period.NAQPMS reasonably reproduced the observed concentration variation and peak of cNO 3 -, after considering heterogeneous reactions on dust particles.As a result of heterogeneous reactions, NO 3 formed on the surface of dust particles and they were mixed internally.The maximum concentration of dust NO 3 simulated by NAQPMS was 5.7 µg m -3 , accounting for 14.8% of the dust concentration; this may be important for the size, shape, and hydrophilic property of dust aerosol and its direct and indirect effects on climate.

CONCLUSIONS
Two regional chemical transport models (NAQPMS and CMAQ) were used to simulate several episodes of high PM 2.5 concentration observed in January 2015 over eastern China and western Japan.Simulation results from both models reasonably explained observed PM 2.5 levels, as well as the variation observed within three sites in both eastern China and western Japan.Some bias existed between these models due to differences in frameworks, including model domains, horizontal resolution, vertical layers, and emissions.
Based on the model results and the synergetic aerosol observations in Fukuoka, Japan, three types of LRT of air pollutants were observed.The first episode showed increased fNO 3 -concentrations relative to fSO 4 2-(type N), indicating the importance of NO 3 -LRT in winter.The second episode showed fSO 4 2-concentrations, which were ~3.4-fold greater than fNO 3 -(type S).The third episode showed high dust concentrations mixed with anthropogenic pollutants (type D), indicating that the LRT of dust was also important in winter, as well as spring.
Both models reasonably explained variations in aerosol components during episodes N and S. Simulated spatial distribution variations indicated the outflow of fSIA from continental Asia to western Japan, consistent with the corresponding PM 2.5 peak at Qingdao and over Japan.During episode S, RH was significantly higher than episode N, therefore, SO 4 2-formed quickly due to aqueous-phase reactions under high RH conditions.
During episode D, mineral dust transported from continental Asia was quickly transported to downwind regions, stagnating over the south of Japan for three days.Measurements showed high cNO 3 -concentrations and high cNO 3 -/fNO 3 -ratio during episode D. These findings were well reproduced by the NAQPMS model after considering heterogeneous reactions on dust particles, which indicates the importance of heterogeneous processes for the LRT of dust and anthropogenic pollutants over East Asia (Pan et al., 2017;Uno et al., 2017b, c).During this period, both models underestimated fSO 4 2-levels, indicating that current models may miss certain emissions of SO 2 and mechanisms promoting the conversion of SO 2 to SO 4 2-.

Fig. 2 .
Fig. 2. Time series of the PM 2.5 concentration in Beijing, Qingdao, and Shanghai, China and at Fukuoka, the Goto Islands, and Tsushima Island, Japan during January 7-23, 2015.Black circles indicate observations.Magenta and yellow shadings indicate simulated anthropogenic and mineral dust PM 2.5 by NAQPMS.Green lines indicate simulated PM 2.5 by CMAQ.Three high PM 2.5 concentration episodes in Fukuoka are shown as '1st','2nd' and '3rd' in (d).

Fig. 3 .
Fig. 3. Temporal variation of the vertical distribution of (a) mineral dust and (b) anthropogenic aerosol extinction coefficients by lidar at Fukuoka.

Fig. 4 .
Fig. 4. Aerosol volume distribution as a function of depolarization ratio (DR) and size, as well as percentages of each species in fine-mode secondary inorganic aerosols (fSIAs) during the (a) first (type N), (b) second (type S), and (c) third (type D) episodes.

Fig. 5
shows temporal variations in RH and fSIAs in Fukuoka; the fSIAs were composed of fine mode sulfate (

Fig. 5 .
Fig. 5. Time series of (a) relative humidity (RH), (b) total concentration of fSIAs, and (c-e) concentration of SO 4 2-, NO 3 -, and NH 4 + , respectively, in Fukuoka, Japan.Black circles indicate observations.RH values used in NAQPMS and CMAQ are shown as magenta and blue lines, respectively, in (a).Magenta and yellow shadings indicate simulated anthropogenic and dust heterogeneous reactions-produced fSIA by NAQPMS, while green lines indicate simulated fSIA by CMAQ in (b-e).

Fig. 6 .
Fig. 6.Simulated daily variation of fSIA spatial distribution during type N (a, b) and type S (c, d) period when the air mass left China (a, c), and when the air mass reached Fukuoka (b, d).Contours shown by thin and thick blue lines for SO 4 2-/SIA ratio represent 0.25 and 0.5, respectively.The 72-h HYSPLIT backward trajectory from Fukuoka is overlaid by green lines with triangles at 6-h intervals.The light green part indicates the trajectory before the time of the figure, while the dark green part indicates the trajectory after.
Fig. 7. Simulated daily variation of dust spatial distribution during type D period.Contours represent isobaric lines at sea level.The 72-h HYSPLIT backward trajectory from Fukuoka is overlaid by magenta lines with markers at 6-h intervals.Backward trajectories with squares, circles, and triangles represent air masses starting from Fukuoka at 12:00 of January 19, 20 and 21 (LT), respectively.The light magenta part indicates the trajectory before the time of the figure, while the dark magenta part indicates the trajectory after.

Fig. 8 .
Fig. 8. Time series of coarse-mode PM (PMc) concentration in Beijing (a) and Fukuoka (b), as well as coarse-mode SO 4 2-(c) and NO 3 -(d) concentration in Fukuoka during January 7-23, 2015.Black circles indicate observations.Magenta, blue, and yellow shadings indicate anthropogenic, sea salt, and dust heterogeneous reactions produced PM simulated by NAQPMS.

Table 1 .
Configurations and emissions of NAQPMS and CMAQ.

Table 2 .
Statistical summary of comparisons of the model results with observations.