The concentration , source and deposition flux of ammonium and nitrate in atmospheric particles during dust events at a coastal site in northern China

Asian dust has been reported to carry anthropogenic reactive nitrogen during transport from source areas to the oceans. In this study, we attempted to characterize NH+4 and NO − 3 in atmospheric particles collected at a coastal site in northern China during spring dust events from 2008 to 2011. Based on the mass concentrations of NH+4 and NO−3 in each total suspended particle (TSP) sample, the samples can be classified into increasing or decreasing types. In Category 1, the concentrations of NH+4 and NO − 3 were 20– 440 % higher in dust day samples relative to samples collected immediately before or after a dust event. These concentrations decreased by 10–75 % in the dust day samples in Categories 2 and 3. Back trajectory analysis suggested that multiple factors, such as the transport distance prior to the reception site, the mixing layer depth on the transport route and the residence time across highly polluted regions, might affect the concentrations of NH+4 and NO − 3 . NH + 4 in the dust day samples was likely either in the form of ammonium salts existing separately to dust aerosols or as the residual of incomplete reactions between ammonium salts and carbonate salts. NO−3 in the dust day samples was attributed to various formation processes during the long-range transport. The positive matrix factorization (PMF) receptor model results showed that the contribution of soil dust increased from 23 to 36 % on dust days, with decreasing contributions from local anthropogenic inputs and associated secondary aerosols. The estimated deposition flux of NNH4 +NO−3 varied greatly from event to event; e.g., the dry deposition flux of NNH4 +NO−3 increased by 9–285 % in Category 1 but decreased by 46– 73 % in Category 2. In Category 3, the average dry deposition fluxes of particulate nitrate and ammonium decreased by 46 % and increased by 10 %, respectively, leading to 11– 48 % decrease in the fluxes of NNH4 +NO−3 .


Introduction
Reactive nitrogen carried in dust particles can be transported over a long distance, and the atmospheric nitrogen deposition in oceans has been recognized as an important external source of the nitrogen supporting phytoplankton growth (Duce et al., 2008;Y. Zhang et al., 2010).This hypothesis has been evaluated through incubation experiments, in situ experiments and the use of satellite observational data (Banerjee and Kumar, 2015;Guo et al., 2012;Liu et al., 2013;Shi et al., 2012;Tan and Wang, 2014).However, the process is dynamic due to the worldwide changing emissions of NO x and NH 3 in the last few decades.For example, China and most of the developing countries in Asia experienced a large increase in emissions of NH 3 and NO x while a substantial decrease in emissions occurred in Europe over the last 3 decades (Grice et al., 2009;Liu et al., 2017;Ohara et al., 2007;Skjøth and Hertel, 2013).The change would affect the nitrogen carried by dust particles to some extent, and updated studies are thereby essential.sition flux of atmospheric particulate NH + 4 and NO − 3 during dust events.
2 Experimental methods

Sampling
Figure 1 shows the sampling site, which is situated at the top of a coastal hill (Baguanshan) in Qingdao in northern China (36 • 6 N, 120 • 19 E; 77 m above sea level) and is approximately 1.0 km from the Yellow Sea to the east.A highvolume air sampler (Model KC-1000, Qingdao Laoshan Electronic Instrument Complex Co., Ltd., China) was set up on the roof of a two-story office building to collect total suspended particle (TSP) samples on quartz microfiber filters (Whatman QM-A) at a flow rate of 1 m 3 min −1 .Prior to sampling, the filters were heated at 450 • C for 4.5 h to remove organic compounds.Our sample collection strategy involved collecting dust samples representing long-range-transported particles.We followed the definition of dust events adopted in the regulations of surface meteorological observations of China (CMA, 2004;Wang et al., 2008) and identified dust events based on the meteorological records (Weather Phenomenon) of Qingdao from the Meteorological Information Comprehensive Analysis and Process System (MICAPS) of the China Meteorological Administration.Due to no dust events lasting over 12 h (Lee et al., 2015;Su et al., 2017;Zhang et al., 2007), we collected one dust sample with a 4 h duration in a day.The sampling for dust particles started only when the measured PM 10 mass concentration in Qingdao (http://www.qepb.gov.cn/m2/) and the forecasted dust mass over Asia (http://www-cfors.nies.go.jp/~cfors/) had greatly increased.
On 20-21 March 2010, two dust events subsequently swept across Qingdao.The online data in high timeresolution can allow two dust events to be identified accurately from the start to the end.The data confirmed that the 4 h dust samples with IDs of 20100320 and 20100321 were well separated from each other for the two events, although they may not capture the entirety of the two events.The same was true for the dust samples with IDs of 20110501 and 20110502.Table 1 lists the sampling information.Based on the forecast, we also collected aerosol particle samples immediately before (or after, but only when no sample was collected prior to dust events), which were regarded as the reference samples.These reference samples were further classified into sunny day samples and cloudy day samples.For those events missing sampling prior to dust events, we collected post-dust samples under clear and sunny weather conditions as early as possible.
Asian dust events were mostly observed in the spring at the sampling site.Our intensive samplings were concentrated in the period of March to May in 2008-2011, when a smaller outbreak for Asian dust events was observed in northern China (Fig. S3 in the Supplement).Overall, a total of 14 sets of dust samples and 8 sets of reference samples were available for analysis in this study.
To facilitate the coastal sampling data analysis, sand samples were collected at the remote site of Zhurihe (42 • 22 N, 112 • 58 E) in the Hunshandake Desert, one of the main Chinese sand deserts, in April 2012.Sand samples were packed in clean plastic sample bags and were stored below −20 • C before the transfer.An ice box was used to store the samples during transport to the lab for chemical analysis.

Analysis
The aerosol samples were weighted according to the standard protocol.The sample membranes were then cut into several portions for analysis.One portion of each aerosol sample was ultrasonically extracted with ultra-pure water in an ice water bath for determining inorganic water-soluble ions using ICS-3000 ion chromatography (Qi et al., 2011).The sand samples collected at the Zhurihe site were analyzed using the same procedure.
One portion of each aerosol filter was cut into 60 cm 2 pieces and digested with HNO 3 + HClO 4 + HF (5 : 2 : 2 by volume) at 160 • C using an electric heating plate.The concentrations of Cu, Zn, Cr, Sc and Pb were measured using inductively coupled plasma mass spectrometry (Thermo X Series 2), while the concentrations of Al, Ca, Fe, Na and Mg were measured using inductively coupled plasma atomic emission spectroscopy (IRIS Intrepid II XSP).Field blank membranes were also analyzed for correction.
One portion of aerosol sample was digested with an HNO 3 solution (10 % HNO 3 , 1.6 M) at 160 • C for 20 min in a microwave digestion system (CEM, U.S.).The Hg and As  2002) using cold vapor atomic fluorescence spectrometry (CVAFS).The detection limits, precisions and recoveries of water-soluble ions and metal elements are listed in Table 2.

Computational modeling
The enrichment factor of metal elements was given by where subscripts i and Re refer to the studied metal and the reference metal, respectively; (X i /X Re ) aerosols is the concentration ratio of metal i to metal Re in the aerosol samples; and (X i /X Re ) crust is the ratio of metal i to metal Re in the Earth's crust.For the calculation of the enrichment factor of the metal elements, scandium was used as the reference element (Han et al., 2012), and the abundance of elements in the Earth's crust given by Taylor (1964) was adopted.
The 72 h air mass back trajectories were calculated for each TSP sample using TrajStat software (Wang et al., 2009) and National Oceanic and Atmospheric Administration (NOAA) GDAS (Global Data Assimilation System) archive data (http://www.arl.noaa.gov/ready/hysplit4.html).The air mass back trajectories were calculated at an altitude of 1500 m to identify the dust origin.In addition, the distance over sea of the air mass for each sample was measured from the trajectory using TrajStat software (Wang et al., 2009).
The positive matrix factorization (PMF) is a commonly used receptor modeling method.This model can quantify the contribution of sources to samples based on the composition or fingerprints of the sources (Paatero and Tapper, 1993;Paatero, 1997).The measured composition data can be repre-sented by a matrix X of i by j dimensions, in which i number of samples and j chemical species were measured, with uncertainty u.X can be factorized as a source profile matrix (F) with the number of source factors (p) and a contribution matrix (G) of each source factor to each individual sample, as shown in Eq. (2).
where E ij is the residual for species j of the ith sample.
The aim of the model is to minimize the objective function Q, which was calculated from the residual and uncertainty of all samples (Eq.3), to obtain the most optimal factor contributions and profiles.
The EPA PMF 3.0 model was used to obtain the source apportionment of atmospheric particulates on dust and comparison days.Our modeled results satisfied the reasonable fit criteria; i.e., 90 % of the scaled residuals were located between the range −3 and +3 for each species.The correlation coefficient between the predicted and observed concentrations was 0.97.
Dry deposition velocities were obtained using Williams' model (Williams, 1982) by accounting for particle growth (Qi et al., 2005).Williams' model is a two-layer model used to calculate the dry velocity of size-segregated particles over the water.In an upper layer below a reference height (10 m), the deposition of aerosol particles is governed by turbulent transfer and gravitational settling.In the deposition layer, the gravitational settling of particles is affected by particle growth due to high relative humidity.To obtain the deposition velocity of different particle sizes, Williams' model needs many input parameters, such as the wind speed at 10 m height (U 10 ), air and water temperature and relative humidity.Relative humidity, air temperature and U 10 from the National Centers for Environmental Prediction (NCEP) were used in this study.Surface seawater temperature data were collected from the European Centre for Medium-Range Weather Forecasts (ECMWF).The meteorological and seawater temperature data had a 6 h resolution.According to a previously reported method (Qi et al., 2013), the dry deposition fluxes of the particles and the nitrogen species were calculated for dust and comparison days.
The CMAQ model (v5.0.2) was applied over the East Asia area to simulate the concentrations of PM 10 , NO x and NH 3 for 14 dust samples.The simulated domain contains 164 × 97 grid cells with a 36 km spatial resolution, and the centered point was 110 • E, 34 • N. The vertical resolution includes 14 layers from the surface to the tropopause, with the first model layer at a height of 36 m above ground level.The meteorological fields were generated by the Weather Research and Forecasting (WRF) Model (v3.7).Considering that the simulated area is connected to the Yellow Sea, the CB05Cl chemical mechanism was chosen to simulate the gas-phase chemistry.Zhang et al. (2009) generated the emissions of air pollutants in 2006 including NO x and NH 3 over East Asia and they updated the emission inventory in 2008, which we have used in this study.Initial conditions (ICONs) and boundary conditions were generated from a global chemistry model of GEOS-CHEM.All the dust events simulations are performed separately, each with a 1-week spin-up period to minimize the influence of the ICONs.The validation of the application of the CMAQ model in China has been reported by Liu et al. (2010a, b).

Other data sources and statistical analysis
Meteorological data were obtained from the Qingdao Meteorological Administration (http://qdqx.qingdao.gov.cn/zdz/ystj.aspx) and the MICAPS of the Meteorological Administration of China.Different weather characteristics, such as sunny days, cloudy days and dust days, were defined according to information from the MICAPS and Qingdao Meteorological Administration.According to the altitude, longitude and latitude of the 72 h air mass back trajectory of each dust sample, the pressure level, temperature and relative humidity (RH) data along the path of the air mass were derived from the NCEP/NCAR reanalysis system (http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis.html)for each sample.The mixed layer depth during the air mass transport of dust samples was obtained from the HYSPLIT Trajectory Model (http://ready.arl.noaa.gov/hypub-bin/trajasrc.pl) using the same method.Then the average mixing layer, transport altitude, air temperature and RH were calculated as an average of all points on the air mass back trajectory of each sample.Spearman correlation analysis was applied to examine the relationships of nitrate and ammonium with transport parameters, and P values of < 0.05 were considered to be statistically significant.

Characterization of aerosol samples collected during dust events
We first examined the mass concentrations of TSP samples and the concentrations of crustal and anthropogenic metals therein through a comparison with the samples collected on dust days and reference samples immediately before or after dust days, providing the background information for our target species analyzed later.The comparative results are highlighted below.For these reference samples, the TSP mass concentrations ranged from 94 to 275 µg m −3 , with an average of 201 µg m −3 (Fig. 2, Table S1).The TSP mass concentration increased substantially to 410-3857 µg m −3 in dust day samples, with an average of 1140 µg m −3 .In each in- dividual pair of a dust day sample and a reference sample, a net increase in the mass concentration of TSPs was observed.The percentages varied from 82 to 1303 % for dust samples compared to references, with a mean value of 403 % (Table S1).A similar increase was present in the crustal elements in each pair of samples.For example, the mean concentrations of Sc, Al, Fe, Mg and nss-Ca (usually used as a typical dust index) increased by more than a factor of 2.
On the other hand, the enrichment factors (EF) of Al, Fe, Ca and Mg were less than three in dust day samples with values less than 14 in the reference samples (Table 3).Lower values are indicative of elements from a primarily crustal origin.
The average mass concentrations of anthropogenic elements, such as Cu, Pb, Zn, Cr, Hg and As, in dust day samples increased by 107 to 722 % compared to those in the reference sample; however, the EF of the anthropogenic metal elements decreased in the former.This indicates that dust particles likely carried more anthropogenic elements, although their relative contribution to the total mass was lower than that in the reference sample.Note that Sample 20110415 was excluded for further analysis.It was judged as a local blowing dust event because no corresponding dust event existed upwind.

Concentrations of NH + 4 and NO − 3 in dust day samples
When the mass concentrations of NH + 4 and NO − 3 in each pair of TSP samples were compared, the concentrations of NH + 4 increased by 8-473 % in some dust day sam-  ples (20080301, 20080315, 20090316, 20100315, 20100320, 20100321, 20110418 and 20110502), but decreased by 28-84 % in other dust day samples (Fig. 3, Column NH + 4 and NO − 3 in Table S1).The same was generally true for the measured concentrations of NO − 3 .Considering the relative values of NH + 4 and NO − 3 in dust day samples relative to the reference samples, we classified the dust day samples into three categories (Table 4).In Category 1, the mass concentrations of NH + 4 and NO − 3 were larger in dust day samples compared to the reference samples.In Category 2, the reverse was true.In Category 3, the mass concentrations of NO − 3 were lower in the dust samples than in the reference samples, whereas the concentrations of NH + 4 were close to the reference.As reported, the Yellow Sea encountered dust storms mainly derived from the Hunshandake Desert (Zhang and Gao, 2007).We thereby compared our observations with the sand particles collected from this desert (Table 5).The ratios of mass concentrations of nitrate and ammonium to the total mass of sand particles were very low, i.e., less than 81 µg g −1 , which are approximately 3 orders of magnitude less than the corresponding values in our dust samples.The values obtained from atmospheric aerosols at the urban sites of Duolun (Cui, 2009) and Alxa Right Banner (Niu and Zhang, 2000), which are closer to the desert, increased on dust days, but were still over 1 order of magnitude lower than the corresponding values in this study (Table 5).The mixing and chemical interaction between anthropogenic air pollutants and dust particles during transport from the source zone to the reception site likely played an important role in increasing the ratios, leading to extremely larger ratio values at this site relative to those in source dust and in upwind atmospheric particles (Cui et al., 2009;Wang et al., 2011;Wu et al., 2017).Since air pollutant emissions, meteorological conditions, chemical reactions and others can affect the concentrations of NH + 4 and NO − 3 in atmospheric particles collected on dust days, the observed increase or decrease in the mass concentration of nitrate and ammonium in different dust samples compared to the reference implied the combined effect of those factors.

Theoretical analysis of the three categories
Ammonium salts are common in atmospheric particles with diameters of less than 2 µm (Yao et al., 2003;Yao and Zhang, 2012).Many modeling studies have shown that the gas-aerosol thermodynamic equilibrium is assumed to be fully attained for inorganic ions, including ammonium salts in PM 2.5 (Dentener et al., 1996;Underwood et al., 2001;Z. Wang et al., 2017;Zhang et al., 1994;Zhang and Carmichael, 1999).Reasonably good agreements between ammonium salt modeling results and observations reported in the literature support the validity of this assumption (Chen et al., 2016;Penrod et al., 2014;Walker  Golmud, Qinghai (Sheng et al., 2016) 892.9 -c Hohhot, Inner Mongolia (Yang et al., 1995) 588.1 No data a Relative concentration of NH + 4 and NO − 3 per aerosol particle mass.b Samples collected on a floating dust day (horizontal visibility less than 10 000 m and very low wind speed).c The ammonium concentration was lower than the detection limit of the analytical instrument.et al., 2012).Supposing that a thermodynamic equilibrium had been attained by the ammonium salts in Category 1, the reactions between carbonate salts and ammonium salts, such as (1) (NH 4 ) 2 SO 4 + CaCO 3 ⇒ CaSO 4 + NH 3 (gas) + CO 2 (gas) + H 2 O and (2) 2NH 4 NO 3 + CaCO 3 ⇒ Ca(NO 3 ) 2 + 2NH 3 (gas) + CO 2 (gas) + H 2 O, will release NH 3 (gas) until CaCO 3 has been completely used up.During dust events, very high concentrations of Ca 2+ were observed, and high CaCO 3 concentrations were therefore expected.For example, the single-particle characterization showed that Asia dust from the Gobi and Inner Mongolian deserts had rich CaCO 3 , with a ratio of 4.3-6.7 % for reacted CaCO 3 and 3.0-4.6% for unreacted CaCO 3 (Hwang et al., 2008).Heterogeneous chemical reactions of mineral dust mostly occurred on CaCO 3 mineral dust (Hwang and Ro, 2006).However, when Category 1 was considered alone except for Sample 20100321, a good correlation was obtained for [NH + 4 ] equivalent concentration = 0.98 × [NO − 3 + SO 2− 4 ] equivalent concentration (R 2 = 0.83, P < 0.05).The good correlation, together with the slope of 1, strongly indicated that the NO − 3 and SO 2− 4 were almost completely associated with NH + 4 in these dust day samples.Anthropogenic ammonium nitrate and ammonium sulfate were thought to be produced by gas, aqueous-phase reactions and thermodynamic equilibrium processes, and they were usually internally mixed (Seinfeld and Pandis, 1998).Conversely, the poor correlation of Ca 2+ to NO − 3 and SO 2− 4 showed that the formation of CaSO 4 and/or Ca(NO 3 ) 2 was probably negligible.Thus, ammonium salt aerosols very likely existed separately to dust aerosols in these dust day samples.Z. Wang et al. (2017) also found that coarse-mode ammonium was quite low and fine-mode dust particles existed separately to anthropogenic ammonium nitrate and ammonium sulfate.The observed NO − 3 and NH + 4 in Asia dust samples were argued to be due to the physical mixing of two types of particles rather than the heterogeneous formation of nitrate and ammonium (Huang et al., 2010).The hypothesis appeared to be valid in Category 1, where NH + 4 was negatively correlated with Ca 2+ (Fig. S4).In Sample 20100321 collected on 21 March 2010, [NH + 4 ] only accounted for ∼ 70 % of the observed [NO − 3 + SO 2− 4 ] in an equivalent concentration.This result suggested that ∼ 30 % of (NO − 3 + SO 2− 4 ) may be associated with dust aerosols via the formation of metal salts of the two species.This hypothesis was supported by the correlation result; i.e., NO − 3 was positively correlated with NH + 4 and Cu, and SO 2− 4 was correlated with K + , Na + and Mg 2+ (Fig. S4).Scheinhardt et al. (2013) found that Cu 2+ showed mixed organic and nitrate complexation in aerosol particles, using a thermodynamic model (E-AIM III).Cu was also detected to be partly in the form of nitrate in aerosol particles by single-particle mass spectrometry (H.Wang et al., 2016;Zhang et al., 2015).Cu was once used as an effective marker of diesel and biodiesel-blend exhaust (Gangwar et al., 2012), while it can also be derived from copper pyrites (CuFeS 2 ) in Inner Mongolia mines (Huang et al., 2010).The increase of Cu in the mass concentration in dust samples implied dust particles mixed with anthropogenic particles, particularly from industrial emissions, during transport.In addition, many studies showed that SO 4 ] equivalent concentration existed.When Category 2 was considered alone except for one Sample 20110501, the equivalent ratios of NH + 4 to NO − 3 + SO 2− 4 were generally much smaller than 1, suggesting that a larger fraction of NO − 3 + SO 2− 4 may exist as metal salts due to reactions of their precursors with dust aerosols.NO − 3 and SO 2− 4 showed no correlation with NH + 4 but did show significant correlation with Pb (Fig. S4).The average concentration of Ca 2+ in Category 2 (0.43 ± 0.40 µmol m −3 ) was evidently higher than that in Category 1 (Ca 2+ : 0.17 ± 0.04 µmol m −3 ), implying the probable formation of CaSO 4 and/or Ca(NO 3 ) 2 and the release of NH 3 (gas).Moreover, except for 20080502, the remaining dust samples in Category 2 were transported from the desert relatively enriched with CaCO 3 (1-25 % in wt %) (Formenti et al., 2011).A positive correlation between NO − 3 and SO 2− 4 in Category 2 compared to a negative correlation in Category 1 also implied that the dust particles enriched with CaCO 3 in Category 2 might play an important role to form SO 2− 4 and NO − 3 .Ca-rich dust particles coated with highly soluble nitrate were observed at Kanazawa in Japan during Asian dust storm periods using SEM/EDX (scanning electron microscopy equipped with an energy dispersive X-ray spectrometer) (Tobo et al., 2010).The single-particle observation conducted by Hwang and Ro (2006) showed that CaCO 3 in dust particles was almost completely consumed to produce mainly Ca(NO 3 ) 2 species.] in an equivalent concentration.As discussed above, ∼ 20 % of (NO − 3 + SO 2− 4 ) may be associated with dust aerosols via the formation of metal salts of the two species.The equivalent ratio of NH + 4 to NO − 3 + SO 2− 4 was only 0.14 for Sample 20100320, and Ca 2+ for this sample (0.47 µmol m −3 ) was evidently higher than that for Sample 20100315 (Ca 2+ : 0.12 µmol m −3 ) and 20110418 (Ca 2+ : 0.12 µmol m −3 ), suggesting that a larger fraction of NO − 3 + SO 2− 4 may exist as metal salts.However, the unique changes in NH + 4 and NO − 3 , different from Category 1 and 2, need further investigation.

Source apportionment of aerosols during dust and non-dust events
The sources of atmospheric aerosols in dust and reference samples were determined by PMF modeling (Paatero and Tapper, 1993;Paatero, 1997).Figure 4 shows that atmospheric aerosols in the reference samples mainly included six sources, i.e., industry, soil dust, secondary aerosols, sea salt, biomass burning and coal combustion/other sources.In these dust samples, including Categories 1-3, oil combustion, industry, soil dust, secondary aerosols and coal combustion/other sources were identified as five major sources (Table 6).The contribution of soil dust evidently increased from 23 to 36 % in the dust samples relative to the reference, consistent with the high concentrations of TSPs and crustal metals observed on dust days.The calculated contribution of nitrate plus ammonium from the soil dust source to the total mass of nitrate plus ammonium in the dust samples greatly increased.The source profile for coal combustion in the dust day samples showed a high percentage of K + , Cl − , Ca, Mg, Co, Ni, As, Al and Fe, indicating that coal combustion particles may exist contemporaneously with other anthropogenic pollutants emitted along the transport path.Liu et al. (2014) also found a larger net increase in the contribution of dust aerosols to the mass of PM 10 , i.e., 31-40 %, on dust days compared to non-dust days in Beijing, which is approximately 600 km upwind of Qingdao.Accordingly, they reported that the contributions of local anthropogenic sources decreased on dust days, especially those from secondary aerosols, consistent with the EF of anthropogenic metals observed on dust days.The calculated air mass trajectories for 13 out of 14 samples showed that the air mass originated from Inner Mongolia, northern China (Fig. 5), generally consistent with the results of Zhang and Gao (2007).The remaining one, with an ID of 20110418 originated from Northeast China.The calculated trajectories showed that the entire dust air mass passed over these highly polluted regions with strong modeled emissions of NO x and NH 3 , shown in Fig. 6, and experienced different residence times in these regions.Figure 5 shows that all air mass trajectories in Category 1 were transported from either the north or northwest over the continent, except for Sample 20110502.In Category 2, the air masses always traveled 94-255 km over the sea prior to arriving at the reception site.NH 3 -poor conditions in the marine atmosphere disfavored the formation and existence of ammonium nitrate.On the other hand, the humid marine conditions (the calculated average RH ranged from 50 to 75 % over the Bohai and Yellow seas in 2006-2012 using NCEP/NCAR reanalysis data) might have enhanced hetero-coagulation between dust and smaller anthropogenic particles, leading to the release of NH 3 via reactions between preexisting ammonium salts and carbonate salts.The average mixing layer was less than 900 m along the air mass transport routes for most sampling days in Category 1 (Table 7), favoring the trapping of locally emitted anthropogenic air pollutants in the mixing layer.The air masses in Category 1 took over 11-39 h to cross over the highly polluted area with appreciable modeled concentrations of NO x (5.7 ± 1.4 ppb) and NH 3 (7.6 ± 3.3 ppb).Except for two samples (IDs of 20080529 and 20110319), air masses in Category 2 took less than 10 h to cross over the polluted areas, with lower concentrations of NO x (modeled value:  3.6 ± 3.4 ppb) and NH 3 (modeled value: 4.7 ± 4.7 ppb), and the mixing layer height along the route was 916-1194 m (on average) for each dust event.Moreover, the averaged wind speed at sampling site was 2.8 m s −1 in Category 1, but 6.2 m s −1 in Category 2. The lower wind speed in Category 1 was unexpected, implying dust particles very likely traveled aloft at a high speed and then mixed down to the ground through subsidence.This further led to the external mixing of anthropogenic particulate matter and dust.The correlation analysis results in Table S2 indirectly support these conclusions.
The concentrations of PM 10 and its major components NO − 3 and NH + 4 over East Asia on dust days and comparison days were modeled using the WRF-CMAQ model (Fig. S5-S6).Spatial distributions of simulated PM 10 during each dust events were consistent with the records in the Sanddust Weather Almanac (CMA, 2009(CMA, , 2010(CMA, , 2012(CMA, , 2013)).The dust particles were transported eastward by passing over the sampling site, the East China Sea, and arriving at the far remote ocean region, except for the local blowing dust sample with an ID of 20110415, as mentioned previously.NMB (normalized mean bias) values of simulated NO − 3 were −4 and −12 % in dust and non-dust reference samples, respectively, indicating that CMAQ results reasonably reproduce the mass concentrations of NO − 3 (Fig. S6).Simulated NH + 4 concentrations in dust samples were severely underpredicted, with NMB values of −71 %.For reference samples, simulated NH + 4 concentrations sometimes can well reproduce the observational values, but the simulation was sometimes severely deviated from the observation.The deviation could be related to many factors which were outside of the scope of this study.The separate mixing mechanism proposed in this study urgently needs to be included in the model for accurately predicting concentrations during dust events.

and metals
Dust events are known to increase the deposition fluxes of aerosol particles along the transport path because of high particle loadings.For example, Fu et al. (2014) found that the long-range transported dust particles increased the dry deposition of PM 10 in the Yangtze River Delta region by a factor of approximately 20.In terms of atmospheric deposition in the oceans, a few studies reported enhancements in oceanic chlorophyll a following dust storm events (Banerjee and Kumar, 2015; Tan and Wang, 2014).In addition to that in high-nutrient and low-chlorophyll (HNLC) regions, the input of nitrogen and other nutrients associated with dust deposition is expected to promote the growth of phytoplankton in oceans with varying nutrient limitation conditions.Thus, we calculated the dry deposition fluxes of aerosols particles, N NH + 4 +NO − 3 and metal elements during dust and reference periods using the measured component concentrations and modeled dry deposition velocities (Table 8).We also compared the calculated dry deposition flux of TSP and N NH + 4 +NO − 3 with previous observations in the literature.The calculated dry deposition fluxes of atmospheric particulates increased on dust days compared to the reference to some extent.For example, the particle deposition fluxes varied over a wide range from 5200 to 65 000 mg m −2 month −1 on different dust sampling days, with an average of 18 453 mg m −2 month −1 , in comparison with the dry deposition flux of TSP of 2800 ± 700 mg m −2 month −1 from the reference periods in the coastal region of the Yellow Sea.S3).In Categories 2 and 3, the dry deposition fluxes of TSP increased by 126 to 2226 % compared to the reference.The dry deposition fluxes of particulate N NH + 4 +NO − 3 decreased by 50 %, on average, in Categories 2 and 3, although the fluxes of ammonium of two samples in Category 3 increased.A larger decrease compared to the reference in the flux of nitrate was present in Categories 2 and 3, i.e., decreases of 73 and 46 %, respectively.The ammonium deposition flux also decreased by 47 % in Category 2 but increased by 10 % in Category 3.
Except for Pb and Zn in Category 2, the calculated dry deposition fluxes of Cu, Pb and Zn increased with those of nitrogen on dust days.Trace metals were found to have a toxic effect on marine phytoplankton and inhibit their growth (Bielmyer et al., 2006;Echeveste et al., 2012).Liu et al. (2013) found that inhibition coexisted with the promotion of phytoplankton species in incubation experiments in the southern Yellow Sea in the spring of 2011 by adding Asian dust samples to collected seawater.However, the calculated dry atmospheric deposition fluxes of Fe increased by a factor of 124-2370 % in dust day samples.F. J. Wang et al. (2017) recently reported that Fe can alleviate the toxicity of heavy metals.Moreover, atmospheric inputs of iron to the ocean have been widely proposed to enhance primary production in HNLC areas (Jickells et al., 2005).
Due to anthropogenic activity and economic development, NO x and NH 3 emissions were reported to increase in China from 1980 to 2010 (Fig. S3; Liu et al., 2017).The dry deposition flux of N NH + 4 +NO − 3 should have theoretically increased with the increase in the emission of inorganic nitrogen.Considering the different dry deposition velocities to be used in various studies, we recalculated the dry deposition flux of N NH + 4 +NO − 3 in the literature using the dry deposition velocities of 1 cm s −1 for nitrate and 0.1 m s −1 for ammonium, as reported by Duce et al. (1991).We thereby found that dry deposition fluxes of N NH + 4 +NO − 3 over the Yellow Sea during the dust days increased greatly from 1999 to 2007, but the values in Qingdao varied narrowly within a range of 94.75-99.65 mg N m −2 month −1 during the dust days from 1997 to 2011 (Table 9).The complicated results implied that even more updated works are needed in the future.

Conclusion
The concentrations of nitrate and ammonium in TSP samples varied greatly from event to event on dust days.Relative to the reference samples, the concentrations were both higher Atmos.Chem.Phys., 18, 571-586, 2018 www.atmos-chem-phys.net/18/571/2018/ in some cases and lower in others.The observed ammonium in dust day samples was explained by NH + 4 was likely either in the form of ammonium salts existing separately to dust aerosols or as the residual of incomplete reactions between ammonium salts and carbonate salts.NO − 3 in the dust day samples can be due to either mixing or reactions between anthropogenic air pollutants and dust particles or a combination of both during the transport from the source zone to the reception site.However, this process was generally much less effective and led to a sharp decrease in nitrate in TSP samples of Category 2. The existence of ammonium salt aerosols separate to dust aerosols and the extent of the reactions between ammonium salts and carbonate salts were evidently associated with the transport pathway, metrological conditions and precursor emissions and other factors.Due to a sharp increase in dust loads on dust days, the contribution of dust to the total aerosol mass increased compared to the samples collected on other days.The contributions from local anthropogenic sources were accordingly lower on dust days.
Overall, this study strongly suggested that atmospheric deposition of N NH + 4 +NO − 3 on dust days varied greatly.A simple assumption of a linear increase in N NH + 4 +NO − 3 with increasing dust load, like that in the literature, could lead to a considerable overestimation of the dry deposition flux of nutrients into the ocean and the consequent primary production associated with dust events.

Figure 2 .
Figure 2. Mass concentrations of TSP, Al, Fe and nss-Ca in aerosol samples collected at the Baguanshan site on dust and reference days during March-May from 2008 to 2011.

Figure 3 .
Figure 3. Mass concentrations of NH + 4 and NO − 3 in aerosol samples collected at the Baguanshan site on dust and reference days during March-May from 2008 to 2011.

Figure 4 .
Figure 4. Source profiles of atmospheric aerosol samples collected on reference (a) and dust (b) days using the PMF model.

Figure 5 .
Figure 5.The 72 h backward trajectories for dust samples from 2008 to 2011 (the yellow domains in the map represent the dust source regions in China).

4. 3
Influence of transport pathways on NH + 4 and NO − 3 in dust samples

Figure 6 .
Figure 6.Seasonal mean emissions of NO x (a) and NH 3 (b) over East Asia during March-May 2008.

Table 1 .
Sampling information for the aerosol samples collected at the Baguanshan site in the coastal region of the Yellow Sea.
Figure 1.Location of the aerosol and dust sampling sites.

Table 2 .
Detection limits, precisions and recoveries of water-soluble ions and metal elements.
in sample extracts were analyzed following the U.S. Environmental Protection Agency method 1631E (U.S. EPA,

Table 3 .
The average concentrations and EFs of metal elements on dust and non-dust days.
* EF values less than 10 indicate that the studied element is mainly derived from crustal sources, whereas EF values much higher than 10 indicate an anthropogenic source.

Table 5 .
Comparison of the NH + 4 and NO − 3 content in sand and aerosol particles on dust days or close to the dust source region (unit: µg g −1 ).

Table 6 .
Sources and source contributions (expressed in %) calculated for aerosol samples collected during dust and non-dust events.

Table 7 .
Concentrations of TSP, NO − 3 and NH + 4 ; transport speed; transport distance over the sea; transport distance; air temperature; RH; average mixed layer during transport and transport time in polluted regions for atmospheric aerosol samples on dust days.Residence time of the air mass passing over parts of highly polluted regions according to the trajectories of samples.
a b Average air temperature; the definition is given in Sect.2.4.c Average relative humidity; the definition is given in Sect.2.4.d Reference samples.

Table 9 .
Comparison of dry deposition flux and normalized flux of TSP (mg m −2 month −1 ) and N NH + −2 month −1 ) with observations from other studies (mg N m −2 month −1 ).The calculation method of the normalized flux of N