Spatial and temporal variations of new particle formation in East Asia using an NPF-explicit WRF-chem model: North-south contrast in new particle formation frequency

Authors


Abstract

[1] A recently developed new particle formation (NPF)-explicit version of the Weather Research and Forecasting Chemistry (WRF-chem) model can explicitly calculate the growth and sink of nucleated clusters with 20 aerosol size bins from 1 nm to 10 µm in diameter. In this study, the model was used to investigate spatial and temporal variations in NPF event frequency and the concentrations of aerosols (condensation nuclei, CN) and cloud condensation nuclei (CCN) within the boundary layer in East Asia in spring 2009. We found a distinct north-south contrast in the NPF frequency and mechanism in East Asia. NPF occurred mainly during limited periods over certain regions between 30° and 45°N (northeast China, Korea, and Japan, including regions around the active volcanoes Miyakejima and Sakurajima). In these latitudes, NPF was suppressed by high concentrations of preexisting particles under stagnant air conditions associated with high-pressure systems, although nucleation occurred more extensively during most of the simulation period. In contrast, south of 30°N, nucleation and NPF were both infrequent because of low SO2 emissions and H2SO4 concentrations. The period-averaged NPF frequency at 30°–45°N was three times that at 20°–30°N. This north-south contrast in NPF frequency was validated by surface measurements in outflow regions of East Asia. The simulated period- and domain-averaged contribution of secondary particles was estimated to be 44% for CN (>10 nm) and 26% for CCN at a supersaturation of 1.0%, though the contribution was highly sensitive to the amount and size distribution of primary aerosol emissions and the rate coefficient of the nucleation parameterization.

1 Introduction

[2] Aerosol particles are produced by two sources in the atmosphere and play an important role in the Earth's climate through direct and indirect effects. Primary particles are emitted directly to the atmosphere, and secondary particles are formed through nucleation and new particle formation (NPF). The formation of secondary particles has a large impact on the concentrations of condensation nuclei (CN) [e.g., Spracklen et al., 2006, 2010; Yu et al., 2010; Matsui et al., 2011a, hereafter M2011], cloud condensation nuclei (CCN) [e.g., Spracklen et al., 2008; Kuang et al., 2009; Merikanto et al., 2009; Yu and Luo, 2009; Luo and Yu, 2011; M2011], cloud droplets [e.g., Pierce and Adams, 2009; Makkonen et al., 2009, 2012], and, potentially, aerosol indirect effects [e.g., Kerminen et al., 2005; Wang and Penner, 2009; Merikanto et al., 2010; Makkonen et al., 2012]. It also adversely affects human health through the formation of nanoparticles.

[3] Previous studies have suggested that the occurrence of NPF is determined by two processes: (1) formation of molecular clusters by nucleation, and (2) subsequent growth of clusters to aerosols [e.g., Kulmala et al., 2000, 2004a, 2007; Zhang et al., 2012]. The growth of clusters is represented by the balance between the condensational growth rate (GR) of clusters and the condensation and coagulation sink (CS) of clusters [e.g., Kerminen and Kulmala, 2002; Kulmala et al., 2004a; McMurry et al., 2005; Kuang et al., 2010, M2011]. Sulfuric acid (H2SO4) and ammonia (and possibly amines) are considered to be the dominant species driving cluster formation [e.g., Kulmala et al., 2000; Berndt et al., 2005, 2006; Sipilä et al., 2010], while organic vapors, in addition to H2SO4 and ammonia, may contribute to the growth of nanoparticles [e.g., O'Dowd et al., 2002; Kulmala et al., 2004b, 2004c, 2006; Kerminen et al., 2004; Paasonen et al., 2010; Ristovski et al., 2010; Wang et al., 2010; Zhang, 2010].

[4] Several mechanisms have been proposed to describe the concentrations of clusters and nanoparticles: binary homogeneous nucleation (BHN) [e.g., Wexler et al., 1994; Pandis et al., 1994; Vehkamäki et al., 2002; Yu, 2008], ternary homogeneous nucleation [e.g., Merikanto et al., 2007], ion-mediated nucleation [Yu, 2006; Yu et al., 2008], activation-type nucleation (AN) [Kulmala et al., 2006; Sihto et al., 2006], and kinetic nucleation (KN) [e.g., McMurry, 1980, 1983; Kuang et al., 2008]. Many three-dimensional modeling studies using these parameterizations have been conducted on global [e.g., Spracklen et al., 2006, 2008, 2010; Merikanto et al., 2009, 2010; Makkonen et al., 2009, 2012; Pierce and Adams, 2009; Wang and Penner, 2009; Yu and Luo, 2009; Yu et al., 2010; Luo and Yu, 2011] and regional scales [e.g., Elleman and Covert, 2009; Zhang et al., 2010; Sotiropoulou et al., 2006; Jung et al., 2010].

[5] East Asia is one of the largest sources of aerosols and their precursor species in the world [Streets et al., 2003; Bond et al., 2004; Dentener et al., 2006; Zhang et al., 2009a]. Because pollution from East Asia can be transported to the Pacific Ocean, North America, and the Arctic [e.g., Koike et al., 2003; Liu et al., 2003, 2008a; Jaffe et al., 2003; Adhikary et al., 2010; Matsui et al., 2011b, 2011c], its potential for affecting climate and air quality is large on both regional and hemispheric scales. Therefore, understanding the behavior of CN and CCN over East Asia and its outflow region, including spatial and temporal variations of CN and CCN concentrations, NPF frequency, and the contributions of primary and secondary particles, is of great importance. Various measurements related to NPF have been carried out previously in East Asia and over the western Pacific, for example, surface measurements in China [Wehner et al., 2004; Wu et al., 2007; Liu et al., 2008b; Yue et al., 2009, 2010, 2011; Wiedensohler et al., 2009; Gong et al., 2010; Shen et al., 2011; Zhang et al., 2011] and Korea [Lee et al., 2008; Song et al., 2010; Yum et al., 2007; Kim et al., 2013] and aircraft measurements during the TRACE-P and ACE-Asia campaigns [Weber et al., 2003; McNaughton et al., 2004; Buzorius et al., 2004]. Few three-dimensional modeling studies, however, have focused on NPF and its impact on CN and CCN concentrations in East Asia.

[6] We previously developed an NPF-explicit version of the WRF-chem model that represents the aerosol size distribution between 1 nm and 10 µm with 20 bins and includes the AN and KN nucleation parameterizations [M2011]. This model was applied to the Beijing region in summer 2006 and successfully reproduced the timing of observed NPF events through theoretical calculations of the GR and CS of nucleated clusters [M2011]. In this study, we apply this NPF-explicit WRF-chem model to East Asia for spring 2009 (section 2), when both surface and aircraft CN measurements were conducted in outflow regions in East Asia (section 3). These measurements are used to evaluate the CN concentrations and the frequency of NPF events in these regions predicted by the model simulations (section 4). The evaluation of NPF events in Asian continental outflow regions is a challenging issue, compared with their evaluation over source regions, because in outflow regions NPF may be controlled not only by the concentrations of precursors (e.g., sulfuric acid) and preexisting particles but also by synoptic-scale meteorological and chemical conditions during the transport from source areas to the measurement sites. Then, we examine the spatial and temporal variability of NPF and CN concentrations within the boundary layer over East Asia in relation to synoptic-scale meteorological variations (section 5.1). Finally, we estimate the frequency of NPF and the contribution of secondary particles in the boundary layer over East Asia (section 5.2). This study is the first attempt to apply and evaluate a three-dimensional model that can explicitly calculate NPF processes in East Asia.

2 Model Simulations

2.1 NPF-Explicit WRF-Chem Model

[7] The NPF-explicit version of the WRF-chem model [M2011] used in this study is based on WRF-chem model version 3.1.1 [Grell et al., 2005], which calculates number and mass concentrations of the following aerosols species: sulfate (SO42−), nitrate, ammonium, black carbon (BC), primary organic aerosol (POA), nonreactive dust, sea salt, and aerosol water [Fast et al., 2006]. The aerosol size distribution is represented by using 20 bins from 1 nm to 10 µm (dry diameter), and nucleated clusters and their condensation and coagulation growth are simulated explicitly. To provide the cluster formation rate at 1 nm, we used the AN parameterization [Kulmala et al., 2006] in the boundary layer with the constant rate coefficient of A = 2 × 10−7 s−1, which value was adopted in M2011. The results with this coefficient are mainly discussed in this study. The uncertainties in the coefficient A chosen in this study are discussed by M2011 (e.g., the sulfuric acid concentration, possible contributions from organics). Because these uncertainties are large [e.g., Spracklen et al., 2006; M2011], sensitivity simulations were conducted with other coefficients (A = 1 × 10−6–1 × 10−8 s−1). In the free troposphere, we applied a BHN parameterization [Wexler et al., 1994]. CCN concentrations at supersaturations of 1.0% and 0.1% (CCN1.0 and CCN0.1) were calculated based on Köhler theory [M2011]. The other chemical schemes used in this study are similar to those described by Matsui et al. [2009, 2010] and M2011 (Table 1). Gas- and aqueous-phase chemistry and dry and wet removal processes are considered in the simulations. The meteorological schemes are similar to those used by Matsui et al. [2009]. More details of the NPF-explicit WRF-chem model are given by M2011.

Table 1. Chemical Options of the WRF-Chem Model Used in This Study
Chemical ProcessModel OptionReference
Gas-phase chemistryCarbon Bond Mechanism (CBM-Z)Zaveri and Peters [1999]
Aerosol nucleationActivation-type nucleation (boundary layer), binary homogeneous nucleation (free troposphere)Kulmala et al. [2006]; Wexler et al. [1994]
Aerosol condensationMOSAICZaveri et al. [2005a, 2005b, 2008]
Aerosol coagulationSemi-implicit coagulation solver (COAGSOLV)Jacobson et al. [1994]
Moving binsTwo-moment advectionSimmel and Wurzler [2006]
Aqueous-phase chemistryBulk approachFahey and Pandis [2001]
PhotolysisFast-JFast et al. [2006]

[8] Recent laboratory studies and field measurements suggest that organic compounds may influence both nucleation and the subsequent growth of new particles under various atmospheric conditions [e.g., O'Dowd et al., 2002; Ristovski et al., 2010; Kerminen et al., 2010; Zhang, 2010; Metzger et al., 2010; Bzdek et al., 2010; Smith et al., 2010; Wang et al., 2010; Paasonen et al., 2010; Zhang et al., 2009b, 2012; Zollner et al., 2012; Yu et al., 2012]. However, as shown in M2011, condensable organic gases and secondary organic aerosols are not considered in the NPF-explicit WRF-chem model, because the mechanisms for organics have yet to be fully established, and parameterizations accounting for amines and other organic vapors are still very preliminary. Therefore, the addition of organic vapors and secondary organic aerosols to our NPF scheme is left for the future, and their treatment is beyond the scope of this paper. Simulations that do not take into account organic growth may underestimate CCN concentrations [e.g., Riipinen et al., 2011]. The importance of this effect will be examined in the future, after the model is modified to implement organic aerosol formation.

2.2 Simulation Setups

[9] Figure 1a shows the model domains used in this study. The outer (orange) domain covers most of East and Southeast Asia, and the inner (red) domain covers eastern China, Korea, and Japan. The horizontal grid spacing of the outer domain is 180 km (60 × 40 grids), and that of the inner domain is 60 km (75 × 45 grids). The spatial resolution employed by our simulations was not so fine because of the detailed (heavy) simulations of the NPF-explicit WRF-chem model. Subgrid scale formation of H2SO4, sulfate, and new particles is not considered in our model. Some studies report that H2SO4 and sulfate can form very close to the source, for example, in the plumes of volcanoes and power plants, suggesting that their subgrid scale formation could affect NPF rate [e.g., Stevens et al., 2012]. The model uses 26 vertical layers from the surface to 100 hPa (11 layers from the surface to 800 hPa). The vertical resolution used in this study is similar to that used by previous aerosol simulation studies over East Asia [e.g., Adhikary et al., 2010; Uno et al., 2003; Kondo et al., 2011a; Matsui et al., 2013b; N. Oshima et al., Vertical transport mechanisms of black carbon over East Asia in spring during the A-FORCE aircraft campaign, submitted to Journal of Geophysical Research, 2013]. The simulation period was from 21 March to 26 April 2009 (37 days). The first 3 days were used for model spin-up, and statistics were calculated for the period between 24 March and 26 April. In this paper, we describe the results for the inner domain. The temporal resolution of the model output was 1 h.

Figure 1.

(a) Simulation domains used in this study. The outer domain (orange) covers most of East and Southeast Asia, with a horizontal resolution of 180 km (60 × 40 grids). The inner domain (red) covers China, Korea, and Japan, with a horizontal resolution of 60 km (75 × 45 grids). Light blue squares show the locations of Miyakejima and Sakurajima (active volcanoes). (b) Locations of surface measurements (circles) and flight tracks of the A-FORCE aircraft campaign (blue lines) used in this study.

[10] Boundary conditions are assumed for the outer domain in our simulations: a CN concentration of 666 cm−3 and a SO2 concentration of 0.1 ppbv in the upper troposphere (~10 km). The particles from the boundaries were not dominant, accounting for less than 20% of the simulated CN concentrations (>10 nm) in the upper troposphere (>7 km) during our aircraft campaign (section 3). Therefore, the influence of the boundary conditions on the CN and CCN concentrations in the boundary layer in our simulations is probably limited.

[11] In this study, we used the anthropogenic, biomass burning, and volcanic emission inventories for East Asia [Streets et al., 2003], springtime biogenic emissions from Global Emissions Inventories Activity [Guenther et al., 1995], and online sea salt [Gong et al., 2002] and dust emissions [Shaw et al., 2008]. The anthropogenic emission inventories of Streets et al. [2003] were used in this study to be consistent with our previous study using the WRF-chem model [Matsui et al., 2013a]. Since the volcanic emissions are for the period when the Miyakejima volcano was erupting, volcanic SO2 emissions from it were modified to 1500 ton/day based on measurements made by the Japan Meteorological Agency in 2009 (1000–2000 ton/day) (http://www.seisvol.kishou.go.jp/tokyo/320_Miyakejima/320_So2emission.htm, in Japanese). A lognormal size distribution, with a count median diameter of 50 nm and a standard deviation (σ) of 1.82, was assumed for primary aerosol emissions (BC and POA), considering the results of measurements in Tokyo and Beijing [Kondo et al., 2011b; M2011] and the treatments of emissions by global modeling studies [e.g., Dentener et al., 2006; Stier et al., 2005; Reddington et al., 2011], in which median diameters from fossil fuel sources were assumed to be 30–60 nm. Our value is within this range, though it may be small for biomass burning and biofuel emissions. The uncertainty in the size of emitted primary particles is a major source of the uncertainties in CN and CCN concentrations [e.g., Reddington et al., 2011], but we do not focus on this uncertainty in this paper. We assumed that off-line aerosol and precursor emissions (anthropogenic, biomass burning, and volcanic emissions) did not vary temporally during the simulation period, following our previous simulations over East Asia [Matsui et al., 2013a; N. Oshima et al., submitted manuscript, 2013].

[12] To estimate the contribution of NPF to CN and CCN concentrations, two simulations were conducted in this study: one with and one without nucleation parameterization (nucleation-on and nucleation-off runs). Although nucleation and NPF are highly nonlinear processes, the difference in CN and CCN concentrations between the nucleation-on and nucleation-off runs (ΔCN and ΔCCN) can be used as an approximate measure of secondary particle formation (contribution from NPF). Because we assumed no primary aerosol emissions of particles smaller than 10 nm (bins 1–5), the CN concentration in bin 5 (secondary particles grown to 6.3–10 nm, CNBIN5) is also used as an indicator of an NPF event.

3 Measurements

[13] We used surface and aircraft aerosol measurements to evaluate our model simulations. Figure 1b shows the locations of surface measurements during the simulation periods: Fukue (32.75°N, 128.68°E) and Hedo (26.87°N, 128.25°E) stations in Japan and Anmyeon (36.53°N, 126.32°E) and Gosan (33.28°N, 126.17°E) stations in Korea. At the Fukue and Hedo stations, CN concentrations larger than 300 nm in diameter were measured with an optical particle counter (RION KC-01D) under dry conditions (relative humidity < 50%) [e.g., Hodkinson and Greenfield, 1965; Quenzel, 1969; Pinnick et al., 1973]. BC mass concentrations were measured with a filter-based absorption photometer, the continuous soot monitoring system (COSMOS) [Miyazaki et al., 2008; Kondo et al., 2011b]. SO42− mass concentrations were measured with an aerosol mass spectrometer (AMS) with an assumed collection efficiency of 0.5 at Fukue and 1.0 at Hedo [Takami et al., 2005, 2007]. At the Anmyeon and Gosan stations, aerosol size distributions were measured by a scanning mobility particle sizer (TSI SMPS 3034) [Lee et al., 2008; Flowers et al., 2010; Kim et al., 2011]. Aerosol number concentrations were collected for 54 size bins from 10 to 470 nm in diameter.

[14] We also used the results of aircraft measurements obtained during the Aerosol Radiative Forcing in East Asia (A-FORCE) campaign [Oshima et al., 2012]. This campaign was conducted over the Yellow Sea and the East China Sea during March and April 2009 (Figure 1b), and a number of vertical profiles of aerosols were obtained from near the surface to 9 km in altitude. Number concentrations of total (dry diameters of 10–1000 nm, CN10) and Aitken-mode particles (dry diameters of 10–130 nm, CNAitken) were measured by two condensation particle counters (TSI CPC 3772) [Takegawa and Sakurai, 2011; Takegawa et al., 2013]. Because the focus of this study is NPF in the boundary layer, we use only the data from the lower troposphere (<4 km in altitude). NPF events in the free troposphere will be discussed in a future paper. More details of the A-FORCE campaign are described by Oshima et al. [2012].

4 Comparison With Measurements

4.1 Accumulation-Mode Aerosols at Fukue and Hedo

[15] Figures 2a–2d show time series plots of BC and SO42− mass concentrations at Fukue and Hedo. Simulated aerosol concentrations were chosen from a horizontal and vertical grid closest to each site (for surface measurements, sections 4.1 and 4.2) or flight track (for aircraft measurements, section 4.3). Temporal variations of observed BC and SO42− mass concentrations generally corresponded to changes in meteorological conditions: when the sites were covered by high-pressure systems, high concentrations were measured, and concentrations decreased rapidly after the passages of a cold front. Both the absolute BC mass concentrations and their temporal variations were reproduced well by the simulations. Simulated BC concentrations underestimated observed concentrations by 17% at Fukue (R = 0.69) and by 14% at Hedo (R = 0.54) on average over the simulation period. SO42− mass concentrations were underestimated by 64% at Fukue (R = 0.65) and 18% at Hedo (R = 0.29; this low R value is likely due to the small variability of the simulated SO42− at Hedo). Possible reasons for the underestimation of BC and SO42− are the treatment of the BC mixing state (all BC particles were assumed to be internally-mixed, leading to the overestimation of wet removal processes) [Matsui et al., 2013a], the underestimation of SO2 to SO42− conversion [e.g., Matsui et al., 2009], and the uncertainties in the measurements and emission inventories.

Figure 2.

Time series of (a–b) black carbon, (c–d) sulfate mass concentrations, and (e–f) aerosol number concentrations (>300 nm) at Fukue and Hedo stations (24 March to 26 April 2009). Double-headed red arrows and vertical blue arrows in Figures 2a and 2b denote periods when the observation sites were covered by high-pressure systems and the passage of a cold front, respectively, at Fukue and Hedo. Measurements were available for 27 March to 26 April 2009 at Fukue and for 24 March to 26 April 2009 at Hedo.

[16] Figures 2e and 2f show time series plots of CN concentrations larger than 300 nm in diameter. Temporal variations of the number concentrations are similar to those of the BC and SO42− mass concentrations at both Fukue and Hedo: high concentrations were observed when the station was influenced by a high-pressure system, and concentrations decreased rapidly after the passage of a cold front. This is reasonable because large-diameter aerosols (>300 nm) account for a substantial proportion of the total aerosol mass. The simulation generally reproduced the temporal variations of the CN (>300 nm) number concentrations at both sites (R = 0.72 at Fukue and R = 0.71 at Hedo), but the simulation underestimated the measured CN concentrations by 31% at Fukue and 35% at Hedo on average over the simulation period.

4.2 CN10 and NPF at Anmyeon, Gosan, and Hedo (North-South Contrast of NPF Event Frequency)

[17] Figure 3 shows time series plots of CN10 concentrations at Anmyeon and Gosan. The meteorological conditions at these sites were generally similar to those at Fukue. At Anmyeon, period-averaged CN10 concentrations were reproduced by the simulation within a factor of 2 (overestimated by 60%), whereas temporal variations were not reproduced well (R = 0.15). CN10 concentrations were higher during the first and last 10 days of the simulation period (24 March–1 April and 14–26 April), with larger variability (3700 ± 1600 cm−3 on average), while they were lower during the middle of the simulation period (2–13 April), with smaller variability (3400 ± 1000 cm−3 on average). Model simulations reproduced these temporal variations of observed CN10. The high concentrations of CN10 during the first and last 10 days of the simulation period were generally due to the occurrence of NPF events, as described below (Figure 4). At Gosan, measurements were available for only 7 days of the simulation period, but the period-averaged CN10 concentrations were reproduced within a factor of 2 (overestimated by 70%, R = 0.30). Simulations using rate coefficients of A = 1 × 10−6–1 × 10−8 s−1 (red shading in Figure 3) showed that CN10 concentrations were not very sensitive to the A coefficient values at either station. Even when CN10 concentrations were simulated in a nucleation-off run (primary aerosols only), they were overestimated by 20% (orange lines in Figure 3). This result suggests that primary particles around Anmyeon were overestimated in our simulation.

Figure 3.

Time series of CN10 concentrations at (a) Anmyeon and (b) Gosan stations (24 March to 26 April 2009). Double-headed red arrows and vertical blue arrows in Figure 3a denote the periods covered by high-pressure systems and the passage of a cold front, respectively, at Anmyeon station. Red shading shows the range of CN10 concentrations simulated with rate coefficients of the nucleation parameterization of A = 1 × 10−6–1 × 10−8 s−1. The orange lines show simulated CN10 concentrations without nucleation parameterization. Measurements were available for 24 March to 26 April 2009 at Anmyeon and for 11–13 and 16–20 April 2009 at Gosan.

Figure 4.

Time series of aerosol number size distribution (3–1000 nm) for (a) measurements and (b) model simulations at Anmyeon station (24 March to 26 April 2009). Double-headed arrows and vertical blue arrows in Figure 4a denote the periods covered by high-pressure systems and the passage of a cold front, respectively. Closed circles show NPF days defined as periods when banana-shaped growth curves (particles around 10–20 nm and their subsequent growth) were observed. All circles (both closed and open) show NPF days defined as number concentrations (dN/dlogD) greater than 3000 cm−3 at 20 nm for more than three consecutive hours in a day (section 4.2).

[18] Figures 4a and 4b show time series plots of the number size distribution at Anmyeon. The model did not always reproduce the exact timing of observed individual NPF events. Because M2011 reported that the timing of NPF events was reproduced well at Beijing (which is near emission sources), the poor reproduction of the timing in this study suggests that NPF events at Anmyeon are not controlled locally and clusters can form upwind of the measurement site (nonlocal events) (see section A1). Currently, it is difficult to reproduce the exact timing of individual NPF events in the outflow regions, and future studies should investigate which parameters are key for the reproduction of individual NPF events in outflow regions.

[19] Although it was difficult for the model to reproduce individual NPF events, the model simulations captured broad features of the observed NPF events (e.g., contrasts due to synoptic-scale meteorology). NPF events occurred frequently during the first and last 10 days, while few NPF events occurred during the middle of the period. When we defined NPF days in both the measurements and the model simulations as number concentrations (dN/dlogD) of 20-nm particles greater than 3000 cm−3 for more than three consecutive hours in a day, the frequency of NPF days was estimated to be 50% (11 days) for measurements and 55% (12 days) for model simulations during the first and last 10 days of the simulation period (Table 2 and Figure 5). During the middle of the simulation period, the frequency was estimated to be 0% (0 days) for measurements and 17% (2 days) for model simulations (Table 2 and Figure 5). In this respect, the frequency of NPF events (the contrast under different synoptic-scale meteorological conditions) at Anmyeon was generally reproduced by the model during the simulation period.

Table 2. Frequency of Observed and Simulated NPF Events (%) at Anmyeon, Gosan, and Hedo Stationsa
SitePeriodMeasurements (1)Simulations (1)Measurements (2)Simulation (2)
  1. a

    NPF days and frequencies were estimated by two definitions (section 4.2): (1) NPF day was defined as number concentrations (dN/dlogD) greater than 3000 cm−3 at 20 nm for more than three consecutive hours in a day. (2) NPF day was defined as a day when a banana-shaped growth curve (particles around 10–20 nm and their subsequent growth) was observed.

Anmyeon24 March–26 April 200932 (11/34)41 (14/34)18 (6/34)21 (7/34)
 24 March–1 April, 14–26 April 200950 (11/22)55 (12/34)27 (6/22)32 (7/22)
 2–13 April 20090 (0/12)17 (2/12)0 (0/12)0 (0/12)
Gosan18 February–12 April 200829 (16/55)35 (19/55)25 (14/55)16 (9/55)
Hedo18 February–12 April 20089 (5/55)16 (9/55)5 (3/55)5 (3/55)
Figure 5.

Frequency of NPF events (%) at Anmyeon, Gosan, and Hedo stations using the definition of NPF days given in section 4.2 during 24 March to 26 April 2009 at Anmyeon and during 18 February to 15 April 2008 at Gosan and Hedo.

[20] Simulated CS showed distinct peaks during the middle of the simulation period (under high-pressure conditions) (section A1). Many (few) NPF events were simulated during the periods when CS values were low (high), suggesting that high CS conditions effectively suppress NPF events. These results are consistent with measurements in Korea and over the western Pacific [Lee et al., 2008; McNaughton et al., 2004] in which NPF were observed frequently after the passage of a cold front, when the concentrations of preexisting particles were sufficiently low.

[21] We also compared NPF events between measurements and model simulations for Gosan and Hedo stations in spring 2008 (section A2). Hedo is located at a lower latitude than both Anmyeon and Gosan stations (Figure 1b), but the longitudes of the three stations are similar. The frequency of NPF events at Gosan and Hedo in spring 2008 was reproduced by model simulations (Figure 5 and Table 2). At Gosan, the observed frequency of NPF days was 29% and the simulated frequency was 35% (Figure 5 and Table 2). These frequencies are equivalent to those at Anmyeon in spring 2009. At Hedo, however, the frequencies of NPF days were systematically smaller than the other stations: 9% in measurements and 16% in the model simulations. Therefore, there is a clear north-south contrast in the frequency of NPF in both measurements and model simulations. This interesting result is discussed further in section 5.

[22] Our definition of NPF events is objective, but it includes both clear and unclear events. We therefore estimated the number of clear NPF events, defined subjectively as periods when banana-shaped growth curves (particles around 10–20 nm and their subsequent growth) were observed. DalMaso et al. [2005] suggested a similar guideline to define NPF events from observations. About half of the NPF events in both measurements and model simulations of this study were clear NPF events by this definition (Table 2). Though the number of NPF days according to our definition is about twice the number of clear NPF days, NPF frequencies agree well between measurements and model simulations and they show a north-south contrast (Table 2) by both definitions.

4.3 CN During the A-FORCE Aircraft Campaign

[23] The blue line in Figure 6a shows the mean vertical profiles (0–4 km) of observed CNAitken (dry diameters of 10–130 nm, at standard temperature and pressure (STP) along the flight tracks during the A-FORCE campaign. Maximum CNAitken concentrations were observed in the boundary layer (0–2 km, 3266 cm−3 on average), and they decreased with altitude (890 cm−3 on average between 3 and 4 km). This vertical profile was reproduced well by simulations with nucleation parameterizations (nucleation-on, Figure 6a, red line), although the mean concentrations were overestimated (6534 cm−3 within 0–2 km). In contrast, without nucleation parameterization (nucleation-off, Figure 6a, orange line), the simulation could not reproduce the profile. The CNAitken concentrations modeled without nucleation parameterization changed little with altitude and were underestimated (962 cm−3 within 0–2 km). These results suggest that model simulations must consider NPF in the boundary layer to capture realistic CN vertical profiles.

Figure 6.

(a) Vertical profiles of observed (blue) and simulated CNAitken (10–130 nm) mean concentrations with (red) and without (orange) nucleation parameterizations during the A-FORCE campaign (24 March to 25 April 2009, Oshima et al. [2012]). Averages (squares) and standard deviations (horizontal bars) are shown, except for the 1 min data (gray). Red shading shows the range of CNAitken concentrations simulated using the rate coefficients of A = 1 × 10−6 to 1 × 10−8 s−1 in the nucleation parameterization. (b) Scatterplot of observed and simulated CNAitken concentrations within the planetary boundary layer (0–2 km) for both nucleation-on (red) and nucleation-off (black) model simulations. (c) Vertical profiles of mean observed (blue) and simulated CN concentrations in the accumulation mode (130–1000 nm) with (red) and without (orange) nucleation parameterization during the A-FORCE campaign.

[24] The red shading in Figure 6a shows the range of CNAitken concentrations simulated with the coefficients of A = 1 × 10−6–1 × 10−8 s−1. The simulated CNAitken concentrations were overestimated even with the smallest coefficient (A = 1 × 10−8 s−1). This result implies that the overestimation of CNAitken compared with the A-FORCE observation cannot be explained by the uncertainty in the rate coefficient A (nucleation parameterization). Uncertainties in other factors (e.g., emissions and their size distributions, processes in model) likely contribute more to the overestimation in this case.

[25] Figure 6b shows a scatterplot between measured and simulated CNAitken concentrations within the boundary layer (0–2 km). CNAitken concentrations simulated with nucleation parameterization correlated positively with measured values (red circles, R = 0.58). This result suggests that, to some extent at least, model simulations can capture the timing of high observed CN concentration within the boundary layer. In contrast, without nucleation parameterization, the correlation between simulated and observed concentrations was much less (black squares, R = 0.30).

[26] Figure 6c shows mean vertical profiles (0–4 km) of CN concentrations in the accumulation mode (CN10–CNAitken, dry diameters of 130–1000 nm, at STP) during the A-FORCE campaign. The simulated CN concentrations in the accumulation mode were very similar between the nucleation-on and nucleation-off simulations. In the boundary layer, mean measured concentrations were 1975 cm−3, and concentrations simulated with nucleation parameterization were 1766 cm−3. Both the absolute concentrations and the vertical profile of CN concentrations in the accumulation mode were reproduced well by the model simulations.

[27] The number and mass (volume) concentrations of BC and BC-free particles were evaluated by Matsui et al. [2013a]. Matsui et al. [2013a] used a similar simulation set up (e.g., meteorological and emission data, chemistry schemes) and reproduced the vertical profiles of both BC and BC-free particles reasonably well.

5 Frequency and Contribution of NPF in East Asia

5.1 North-South Contrast in NPF Events and Their Frequency

[28] Because the model simulations generally reproduced the observed temporal variations of CN10 and the frequency of NPF events in the boundary layer, we examined the spatial distribution of the frequency and the contribution of NPF in the boundary layer over the entire East Asian region.

[29] Figure 7 shows latitudinal and temporal variations of the concentrations of H2SO4, CNBIN5, CN10, and PM2.5 at a sigma level of 0.895 (~1 km) along the 125°E meridian. H2SO4 and CNBIN5 can be used as measures of nucleation rates and NPF events (section 2), respectively. Daytime H2SO4 concentrations were high north of 30°N on most days during the simulation period (Figure 7a). These high concentrations reflect the regional distribution of SO2 emissions over East Asia. Because a large proportion of SO2 is emitted north of 30°N in China (Figure 8a), nucleation occurred on most days during the simulation period over the regions from 30° to 45°N (Figure 7a). In contrast, enhancements in CNBIN5 and CN10 were found in limited regions and periods (Figures 7b and 7c), corresponding to regions and periods with low PM2.5 concentrations (Figure 7d). NPF was suppressed when PM2.5 concentrations were high (high CS values) because of stagnant atmospheric conditions in areas covered by high-pressure systems (e.g., 25°–40°N during the middle of the simulation period, 30°–35°N during 24–30 March, and 25°–35°N during 15–25 April). Neither nucleation nor NPF was frequent south of 30°N, where concentrations of SO2 and H2SO4 were low.

Figure 7.

Latitudinal and temporal variations of (a) H2SO4, (b) CNBIN5, (c) CN10, and (d) PM2.5 concentrations at a sigma level of 0.895 (~1 km) along the 125°E meridian.

Figure 8.

(a) SO2 emissions used in this study. Period-averaged concentrations of (b) SO2, (c) H2SO4, (d) CNBIN5, and (e) PM2.5 at a sigma level of 0.895 (~1 km). (f) Frequency of NPF (ratio of NPF days to total simulation days) over East Asia during the simulation period (24 March to 26 April 2009). NPF days were defined for each grid cell and day as a daily maximum concentrations of CNBIN5 greater than 2000 cm−3.

[30] Figures 8b–8e show period-averaged simulated concentrations of SO2, H2SO4, CNBIN5, and PM2.5 at a sigma level of 0.895 (~1 km). Simulated SO2 concentrations were high over central and northern China, over Korea, around the Sichuan Basin, and around Miyakejima and Sakurajima volcanoes (Figure 8b). This distribution reflects the spatial distribution of SO2 emissions (Figure 8a). The spatial distribution of H2SO4 was similar to that of SO2, with high concentrations widespread over East Asia (central and northern parts of China, Korea, the Yellow Sea, the East China Sea, and Japan), indicating that nucleation occurred extensively in East Asia between 30° and 45°N during the simulation period (Figure 8c). In contrast, the distributions of high CNBIN5 concentrations were limited to portions of northeastern China and Japan (Figure 8d), probably because NPF was suppressed by large amounts of preexisting particles. Over China in particular, high CNBIN5 concentrations were simulated along the western and northern borders of the area with high H2SO4 (nucleation area). These areas were upwind of the areas with high anthropogenic emissions, and they were characterized by relatively high SO2 concentrations and low amounts of preexisting particles. For example, very high CNBIN5 concentrations west of Beijing (but not in central Beijing) were likely due to the lower amounts of preexisting particles, as pointed out by M2011. As the suppression of NPF was determined by the spatial and temporal variations of preexisting particles, which were controlled by synoptic-scale meteorological variations, regions with high period-averaged PM2.5 concentrations did not necessarily correspond to regions of NPF suppression (Figure 8e). The period-averaged PM2.5 distribution reflects both the distribution of anthropogenic aerosols over northern China (with peak concentrations around 110°–120°E, 35°–40°N) and dust around Mongolia.

[31] Figure 8f shows the frequency of NPF over East Asia, defined as the ratio of NPF days to total simulation days. In this section, NPF days were defined for each grid cell and day as a daily maximum concentrations of CNBIN5 greater than 2000 cm−3 (dN/dlogD > 10000 cm−3), the same definition used by M2011. (Note that in section 4.2, where we compared the frequency of NPF days between measurements and simulations, we used a different definition of NPF days.) NPF days defined here include not only NPF events occurring locally but also the transport of nanoparticles (<10 nm) from upwind regions (nonlocal NPF events). NPF frequency was estimated to be 40–80% over northern China around Beijing, 30–70% over northeastern China and Korea, and 30–80% around Miyakejima and Sakurajima volcanoes during the simulation period. These NPF event frequencies are generally consistent with reported NPF frequencies observed in China and Korea in the springtime, though the definition of NPF events is not exactly the same among these studies: 40–70% in Beijing [Shen et al., 2011; Wehner et al., 2004; Wu et al., 2007] and 20–40% at Gosan and Anmyeon [Lee et al., 2008; Song et al., 2010; Yum et al., 2007; Kim et al., 2013]. NPF events were not frequent within the boundary layer over central or southern China or over the southwestern Pacific (NPF frequency generally less than 20%) during the simulation period because over these regions low H2SO4 concentrations caused nucleation rates and GR to be low.

[32] In summary, nucleation occurred extensively on most simulation days between 30° and 45°N in the boundary layer (northeast China, Korea, and Japan, including regions around active volcanoes). NPF events at these latitudes were considerably suppressed by high concentrations of preexisting particles under stagnant air conditions associated with high-pressure systems. In fact, when high (low) PM2.5 periods were defined for each grid cell at a sigma level of 0.895 (~1 km) as the 5 days with the highest (lowest) daily mean PM2.5 concentrations during the simulation period, mean NPF frequency was estimated to be 16% and 31% during the high- and low-PM2.5 period, respectively, at latitudes 30°–45°N (Figure 9). Conversely, neither nucleation nor NPF were frequent south of 30°N because SO2 emissions and H2SO4 concentrations were relatively low there (Figure 9). The suppression of NPF by high PM2.5 concentrations was not seen at latitudes 20°–30°N (Figure 9). The period-averaged NPF frequency was estimated to be 24% at latitudes 30°–45°N and 8% at latitudes 20°–30°N (Figure 9). These results show a clear north-south contrast in NPF frequency and mechanism during springtime in East Asia.

Figure 9.

Period-averaged SO2, H2SO4, PM2.5, and CNBIN5 concentrations and the frequency of NPF at latitudes of 30°–45°N (blue) and 20°–30°N (red). High- and low-PM2.5 periods were defined for each grid cell at a sigma level of 0.895 (~1 km) as the 5 days having the highest or lowest daily mean PM2.5 concentrations during the simulation periods.

5.2 Contribution of NPF to CN and CCN

[33] In this section, we describe the contribution of NPF (secondary particles) to CN and CCN concentrations in East Asia. As shown in section 2, we mainly show the simulation results obtained using a constant rate coefficient of A = 2 × 10−7 s−1 in the AN parameterization. The results of sensitivity simulations with other coefficients (A = 1 × 10−6–1 × 10−8 s−1) are also shown in order to understand the sensitivity of NPF to the coefficients in the nucleation parameterizations.

[34] Figure 10 shows period-averaged CN10 and CCN1.0 concentrations at a sigma level of 0.895 (~1 km) for both nucleation-on and nucleation-off simulations. In the nucleation-on simulation, maximum CN10 concentrations were simulated over northern China, Korea, and Japan, corresponding to the regions where NPF was frequent (Figure 8f). The contribution of secondary particle formation (ΔCN10/CN10) was estimated to be more than 40% over these regions (Figure 10e): 40–70% over China and Korea and 60–80% around Miyakejima and Sakurajima volcanoes. Large amounts of secondary particles were produced by SO2 emissions from Miyakejima and Sakurajima volcanoes in Japan, because they created conditions of high GR and low CS values during the daytime. The secondary particles produced by the volcanoes could be a large source of CN and CCN over the Pacific, even though their mass contributions were small. Over the Yellow Sea and the East China Sea, the contribution of secondary particle formation was estimated to be 40–60%. In contrast, CN10 simulations without nucleation parameterization (primary particles) produced the highest concentrations over central China and around the Sichuan Basin (Figure 10c). Most of CN10 was primary particles between 20° and 30°N in China, where nucleation did not occur efficiently because of low SO2 emissions and H2SO4 concentrations (Figure 8). The contribution of secondary particle formation was generally less than 20%. The period- and domain-averaged CN10 concentrations were 4711 and 2642 cm−3 for the nucleation-on and nucleation-off simulations, respectively. The contribution of secondary particle formation to CN10 concentrations was estimated to be 44% on period and domain average (Table 3). The period- and domain-averaged CN10 concentrations were 5803 cm−3 and 3901 cm−3 in simulations with rate coefficient values of A = 1 × 10−6 s−1 and A = 1 × 10−8 s−1, respectively. The contribution of NPF to CN10 concentrations was 54% and 32%, respectively, in these sensitivity simulations (Table 3), suggesting that the CN10 concentration in the simulation domain was sensitive to the rate coefficient value in the nucleation parameterization.

Figure 10.

Period-averaged concentrations of CN10 ((a) nucleation-on run, (c) nucleation-off run) and CCN1.0 ((b) nucleation-on run, (d) nucleation-off run) at a sigma level of 0.895 (~1 km). Period-averaged contributions of NPF to (e) CN10 (ΔCN10/CN10) and (f) CCN1.0 (ΔCCN1.0/CCN1.0). ΔCN10 (ΔCCN1.0) is the difference in the CN10 (CCN1.0) concentration between the nucleation-on and nucleation-off runs.

Table 3. Period- and Domain-Averaged CN and CCN Concentrations in East Asia a
ComponentNucleation-OffNucleation-OnNucleation-OnNucleation-On
(A = 2e − 7)(A = 1e − 6)(A = 1e − 8)
  1. a

    The values were calculated at a sigma level of 0.895 (~1 km).

  2. b

    The values within parentheses indicate the contributions of secondary particles (%).

CN10 (cm−3)26424711 (44%b)5803 (54%)3901 (32%)
CCN1.0 (cm−3)19312598 (26%)2793 (31%)2355 (18%)
CCN0.1 (cm−3)306306307302

[35] Primary particles were more important for the case of CCN1.0 concentrations. High CCN1.0 concentrations were simulated over northern and central China and around the Sichuan Basin in both the nucleation-on and nucleation-off simulations (Figures 10b and 10d). The contribution of secondary particle formation to CCN1.0 was estimated to be less than 30% over most of China (Figure 10f). The contribution of secondary particles was larger over Korea, Japan, and the western Pacific: 30–40% over the Yellow Sea, 30–60% over Korea, and 40–70% around Miyakejima and Sakurajima volcanoes (Figure 10f). The volcanoes could be an important source of secondary particles over East Asia on period average, in terms of both NPF frequency (Figure 8f) and secondary particle contribution to CN and CCN concentrations (Figures 10e and 10f). The period- and domain-averaged CCN1.0 concentrations were 2598 and 1931 cm−3 in the nucleation-on and nucleation-off simulations, respectively. The contribution of NPF to CCN1.0 concentration was estimated to be 26% for the period and domain average (Table 3). The period- and domain-averaged CCN1.0 concentrations were 2793 and 2355 cm−3 in the nucleation-on simulations with the rate coefficient A = 1 × 10−6 s−1 and A = 1 × 10−8 s−1, respectively. The contribution of NPF in these sensitivity simulations was 31% and 18%, respectively (Table 3), suggesting that the CCN1.0 concentration in the simulation domain was moderately sensitive to the rate coefficient value used in the nucleation parameterization.

[36] The contribution of NPF to the CCN1.0 concentration estimated in this study is smaller than the estimate of Merikanto et al. [2009] for Asia (51–71% from both boundary layer and tropospheric nucleation mechanisms), though the regions and periods evaluated differ between their study and ours. We note that the contribution of NPF would be highly sensitive to the treatment of emissions (both the amount and particle size distribution), nucleation parameterization, and the values used for parameters related to NPF, such as the concentrations of sulfuric acid and preexisting particles, in the simulations. For example, if we assumed larger emitted particles in our simulations, the contribution of NPF would increase considerably (e.g., a change in the median diameter for emissions from 50 nm to 70–80 nm would decrease the emitted number concentrations to about one-third). More detailed studies are necessary for accurate estimation of primary and secondary particle contributions to CN and CCN over East Asia.

6 Summary

[37] CN and CCN concentrations were simulated over East Asia and its outflow regions with an NPF-explicit version of the WRF-chem model, which can calculate the growth and sink of nucleated clusters explicitly with 20 aerosol size bins from 1 nm to 10 µm. The model was used to understand the spatial and temporal variations of CN and CCN number concentrations, the frequency of NPF events, and the contribution of primary and secondary particles within the boundary layer over East Asia and its outflow region during spring 2009.

[38] Model simulations did not necessarily reproduce the exact timing of individual NPF events at surface measurement sites (Anmyeon, Gosan, and Hedo) in the outflow region in East Asia. This is likely because NPF events in those regions are not controlled only by local NPF (nonlocal factors during transport of air masses from the source area are also important), and also because it is difficult for the model to reproduce both local and nonlocal parameters related to NPF (e.g., concentrations of sulfuric acid and preexisting particles) precisely. However, the model reproduced the temporal variations in aerosol mass and number concentrations and the frequency of NPF events at surface measurement sites (Fukue, Hedo, Anmyeon, and Gosan) reasonably well. The simulation also reproduced the timing of high CN concentrations within the boundary layer and the vertical profile of CN concentrations observed during the A-FORCE aircraft campaign, conducted over the Yellow Sea and East China Sea in March and April 2009. These features were not reproduced well in simulations without nucleation parameterization, indicating that nucleation and NPF processes are very important for realistically capturing the temporal variability and vertical profiles of CN concentrations in model simulations.

[39] Nucleation occurred extensively at latitudes 30°–45°N in East Asia (central and northern China, Korea, the Yellow Sea, the East China Sea, and Japan) on most days during the simulation period. This is because SO2 emissions and H2SO4 concentrations were sufficiently high at these latitudes. However, NPF events occurred on fewer days and were confined in northeastern China, Korea, and around Miyakejima and Sakurajima volcanoes. This is because NPF was considerably suppressed by the presence of large amounts of preexisting particles under stagnant air conditions associated with high-pressure systems. The frequency of NPF events was significantly higher when concentrations of preexisting particles were very low (such as just after the passage of a cold front) at latitudes 30°–45°N. In contrast, neither nucleation nor NPF occurred frequently south of 30°N because SO2 emissions and H2SO4 concentrations were low there. The suppression of NPF by high PM2.5 concentrations was not remarkable at latitudes 20°–30°N. The period-averaged NPF frequency at latitudes 30°–45°N was three times the frequency at latitudes 20°–30°N. Thus, there was a distinct north-south contrast in the frequency and the mechanism of NPF during springtime in East Asia.

[40] The contribution of NPF was also high at latitudes 30°–45°N (30–80% for CN10 and 20–60% for CCN1.0), whereas it was less than 20% at latitudes 20°–30°N. Secondary particles from volcanoes in Japan could be a large source of CN and CCN over the Pacific. The contribution of secondary particles was estimated to be 44% for CN10 and 26% for CCN1.0 on period and domain average, suggesting that secondary particles could substantially influence the spatial and temporal variability of CN10 and CCN1.0 in East Asia, and possibly the variability of cloud droplet number concentrations and aerosol indirect effects as well. Because NPF frequency and CN10 and CCN1.0 concentrations are highly sensitive to the amount and size distribution of primary aerosol emissions and to the rate coefficient used in the nucleation parameterization, the uncertainties in these parameters must be reduced to increase the accuracy of the estimation of the impact of nucleation on NPF, CN and CCN concentrations, and aerosol indirect effects.

Appendix A

A1 Nonlocal NPF Events at Anmyeon

[41] Figure A1a shows observed and model-simulated CS (s−1) at the Anmyeon site. Figure A1b shows GR (nm h−1) for nucleated clusters at the Anmyeon site. These values were estimated using equations derived by Kerminen and Kulmala [2002] (equations (6) and (7) of M2011) by assuming the mass accommodation coefficient of condensing vapor (α) of unity. GR and CS are proportional to the concentrations of condensable gases and preexisting particles, respectively. In this study, GR was calculated for sulfuric acid only (GRSULF). The observed values of CS and their temporal variations were generally reproduced by the simulations (Figure A1a). Simulated CS showed distinct peaks during the middle of the simulation period (high-pressure conditions). Many (few) simulated NPF events were seen during the periods of lower (higher) CS values, suggesting that high CS conditions effectively suppressed NPF events.

Figure A1.

Time series of (a) observed and model-simulated condensation and coagulation sink (CS) and (b) the model-simulated growth rate (GR) of nucleated clusters at Anmyeon station (24 March to 26 April 2009). CS and GR values were calculated using equations (6) and (7) of M2011. GR was calculated for sulfuric acid only (GRSULF).

[42] In contrast, GRSULF showed no clear difference between periods with and those without NPF. NPF events did not necessarily coincide with higher GRSULF values in the model simulations; they occurred even when the GRSULF value was quite low (e.g., on 27 March and 22 April), and they did not necessarily occur when the GRSULF value was high (e.g., on 11 and 13 April). These results suggest that the local H2SO4 concentration was not critical for the occurrence of NPF at Anmyeon during the simulation period, and that the amount of preexisting particles (CS) was key to the occurrence or suppression of NPF events, as described above. In most NPF events simulated by the model, nucleation did not necessarily occur at the measurement site. Clusters could also be formed in upwind regions and then transported to the measurement site with their growth. This interpretation is reasonable because peak diameters were already several nanometers at the beginning of most of the NPF events at the Anmyeon site (Figure 4b). Therefore, the model should reproduce both local and nonlocal parameters related to NPF, including the histories of air masses from sources to the outflow regions. This may be a major reason why NPF events in outflow regions are difficult to reproduce and why our model could not capture individual NPF events at the Anmyeon site. In this regard, NPF events at Anmyeon differ from those at an urban site in Beijing, where most NPF events started from nucleation clusters with diameters of 1 nm and were controlled by local parameters [M2011]. M2011 reported that the timing of NPF events was reproduced well at Beijing (near the emission sources). More studies are needed to understand the key parameters to reproduce individual NPF events in outflow regions.

A2 NPF Events at Gosan and Hedo in Spring 2008

[43] We also compared NPF events between measurements and model simulations at Hedo and Gosan stations in spring 2008. At Hedo, aerosol size distributions between 10 and 500 nm in diameter were measured by a wide-range particle spectrometer (WPS) between 18 February and 12 April 2008. The instrument included a scanning mobility spectrometer (SMS) comprising a differential mobility analyzer (DMA) and a condensation particle counter (CPC). At Gosan, aerosol size distributions between 10 and 470 nm in diameter were measured with a scanning mobility particle sizer (TSI SMPS 3034) [Flowers et al., 2010; Kim et al., 2011] during the same periods. Model simulations were conducted for these periods (15 February to 12 April) with the same settings of simulation domains, emissions, and model options as the simulations of spring 2009 (main text). NPF days defined as in section 4.2 (number concentrations (dN/dlogD) greater than 3000 cm−3 at 20 nm for more than three consecutive hours in a day) were used with both measurement data and model simulations.

[44] At Gosan, there were 16 NPF days in measurements and 19 NPF days in model simulations during the simulation periods (55 days). The proportions of NPF days were 29% and 35% for measurements and model simulations, respectively (Figure 5 and Table 2). These proportions are similar to those at Anmyeon in spring 2009 (Figure 5). At Hedo, the frequency of NPF events was much smaller: the proportion (number) of NPF days was 9% (5 days) and 16% (9 days) for measurements and model simulations, respectively.

[45] Similar to the results at Anmyeon station in spring 2009 (section 4.2), model simulations could not necessarily reproduce the exact timing of individual NPF events in spring 2008 at Gosan and Hedo stations in the Asian outflow region (not shown). However, the frequency of NPF events, including the north-south contrast, was reproduced by the model simulations reasonably well (Figure 5 and Table 2).

Acknowledgments

[46] This work was supported by the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), the strategic international cooperative program of the Japan Science and Technology Agency (JST), the global environment research fund of the Japanese Ministry of the Environment (A-0803, A-1101, and B-1006), and by the Alliance for Global Sustainability (AGS) project, University of Tokyo. This study was conducted as a part of the Mega-Cities: Asia Task under the framework of the International Global Atmospheric Chemistry (IGAC) project. Support for MOSAIC and WRF-Chem was provided by the U.S. Department of Energy's (DoE) Atmospheric System Research program under Contract DE-AC06-76RLO 1830 at PNNL. PNNL is operated for the U.S. DoE by the Battelle Memorial Institute. For a part of the simulations, we used the HA8000 computer system operated by the Supercomputing Division, Information Technology Center, University of Tokyo. S.-C. Yoon and S.-W. Kim were supported by the BK21 program of the School of Earth and Environmental Sciences, Seoul National University, and by the Korea Meteorological Administration Research and Development Program under Grant CATER 2012-3020.