An evaluation of simulated particulate sulfate over East Asia through global model intercomparison

Abstract

Sulfate aerosols simulated by an aerosol module coupled to the Nonhydrostatic Icosahedral Atmospheric Model (NICAM) at a spatial resolution (220 km) widely used by global climate models were evaluated by a comparison with in situ observations and the same aerosol module coupled to the Model for Interdisciplinary Research on Climate (MIROC) over East Asia for January, April, July, and October 2006. The results indicated that a horizontal gradient of sulfate from the source over China to the outflow over Korea‐Japan was present in both the simulations and the observations. At the observation sites, the correlation coefficients of the sulfate concentrations between the simulations and the observations were high (NICAM: 0.49–0.89, MIROC: 0.61–0.77), whereas the simulated sulfate concentrations were lower than those obtained by the observation with the normalized mean bias of NICAM being −68 to −54% (all), −77 to −63% (source), and −67 to −30% (outflow) and that of MIROC being −61 to −28% (all), −77 to −63% (source), and −60 to +2% (outflow). Both NICAM and MIROC strongly underpredict surface SO₂ over China source regions and Korea‐Japan outflow regions, but the MIROC SO₂ is much higher than NICAM SO₂ over both regions. These differences between the models were mainly explained by differences in the sulfate formation within clouds and the dry deposition of SO₂. These results indicated that the uncertainty of the meteorological and cloud fields as well as the vertical transport patterns between the different host climate models has a substantial impact on the simulated sulfate distribution.

1 Introduction

Air pollution has been increasing in East Asia, and transboundary air pollution is an important issue for air quality and human health throughout the world [Janssen et al., 2012; Yu et al., 2012]. For example, aerosols and their precursors emitted from East Asian regions are often transported to western Pacific outflow regions and sometimes reach the United States, especially during spring [Jacob et al., 2003; Seinfeld et al., 2004; Nakajima et al., 2007; Singh et al., 2009; Yu et al., 2013]. The issue of transboundary air pollution should be addressed by monitoring and predicting pollutants over both source and outflow regions [Takemura et al., 2002b; Nakajima et al., 2007; Ramanathan et al., 2007]. Over East Asia, SO₂, a precursor of sulfate, is largely emitted as a result of fossil fuel combustion [Lu et al., 2010]; for this reason, sulfate aerosols have become the largest contributor to the submicron aerosol mass concentrations over East Asia [Takami et al., 2007; Sahu et al., 2009; X. Zhang et al., 2012].

Because sulfate has been more frequently measured than other aerosol components at various sites in East Asia, the sulfate aerosols simulated by aerosol transport models have been well evaluated using the observations from the EANET (Acid Deposition Monitoring Network in East Asia; http://www.eanet.asia/index.html), which is a network of more than 30 sites in Asia, including the outflow region in Japan [Carmichael et al., 2008; Lin et al., 2008b; Goto et al., 2011; Kuribayashi et al., 2012]. In addition, regional aerosol transport models have investigated source‐receptor relationships of sulfate throughout East Asia and have shown that sulfate aerosols originating from China have the largest contribution to background levels of sulfate throughout East Asia [Lin et al., 2008a; Wang et al., 2008; Kajino et al., 2012]. Aikawa et al. [2010] reported a significant horizontal gradient of sulfate from South Korea to Japan based on EANET measurements. However, these studies did not include observations over China due to the limited network of sulfate measurements throughout that country (EANET stations in China monitor wet deposition and concentrations of SO₂, NO₂, and PM₁₀ but do not monitor concentrations of aerosol chemical constituents). Recently, X. Zhang et al. [2012] and Y. Zhang et al. [2012], as part of the Chinese Meteorological Administration Atmosphere Watch Network (CAWNET), measured aerosol chemical compounds at 14 sites throughout China that EANET does not fully cover. The combination of EANET and CAWNET data enables the evaluation of simulated sulfate distributions in East Asia from China through Korea to Japan.

In the present study, we evaluated the sulfate distributions simulated by an aerosol transport model, the Spectral Radiation‐Transport Model for Aerosol Species (SPRINTARS) [Takemura et al., 2005; Suzuki et al., 2008] coupled to the Nonhydrostatic Icosahedral Atmospheric Model (NICAM), developed by Tomita and Satoh [2004] and Satoh et al. [2008, 2014], by focusing on the China source regions and Korea‐Japan outflow regions of East Asia in the four seasons of 2006. NICAM, which can be used at a range of resolutions up to the cloud system‐resolving scale [Miura et al., 2007; Satoh et al., 2008; Miyamoto et al., 2013], is used here at a typical resolution (approximately 220 km) of a global climate model [Niwa et al., 2011b; Dai et al., 2014a, 2014b; Goto, 2014]. NICAM uses the finite‐volume method for spatial discretization that completely conserves the mass of the tracers, whereas many general circulation models (GCM) use the finite‐difference method that requires some assumption to conserve them [e.g., Jöckel et al., 2001; Niwa et al., 2011a]. High‐resolution simulations of atmospheric aerosols using NICAM have also been conducted at a global scale [Suzuki et al., 2008] and a regional scale [Goto, 2014; Goto et al., 2015]. Therefore, NICAM can seamlessly treat various spatial scales as a multiscale single model; thus, in the present study, NICAM was used to evaluate simulated sulfate distributions with low resolutions.

The aerosol module, SPRINTARS, is originally coupled to a global climate model, MIROC (Model for Interdisciplinary Research on Climate), developed by the Atmosphere and Ocean Research Institute of University of Tokyo (AORI/UT), the National Institute for Environmental Studies (NIES), and the Japan Agency for Marine‐Earth and Technology (JAMSTEC) [K‐1 Model Developers, 2004; Watanabe et al., 2010]. The results of SPRINTARS coupled to MIROC have been compared with measurements at a global scale [Takemura et al., 2000, 2002a, 2002b; Goto et al., 2011], with measurements in springtime over East Asia [Takemura et al., 2003; Nakajima et al., 2007], and in international model intercomparisons such as AeroCom [Kinne et al., 2006; Textor et al., 2006; Myhre et al., 2013]. Therefore, a comparison of the results obtained by SPRINTARS coupled to NICAM with not only the measurements but also the results simulated by SPRINTARS coupled to MIROC becomes a “model intercomparison” that evaluates the impacts of the different host climate models on the modeled sulfate distribution, an approach employed by Liu et al. [2007] using aerosol models other than those investigated in our study. The use of model intercomparisons focusing on the host models for aerosol simulations is a limited but important method to estimate possible uncertainties related to basic questions of macrophysical processes [Stier et al., 2013].

This paper is organized as follows: the model formulations of the host climate models (NICAM and MIROC) and the aerosol module (SPRINTARS) and the measurements used in this study are described in section 2. In section 3, the NICAM‐simulated meteorological parameters and sulfur compounds are compared with the measurements and MIROC‐simulated results throughout East Asia for January, April, July, and October 2006. In section 4, we discuss the differences in the sulfur compounds between NICAM and MIROC in terms of their budgets and column amounts as well as the seasonality of sulfate. Finally, the conclusions are summarized in section 5. This paper highlights the uncertainty of the simulated sulfate distribution over East Asia arising from associated meteorological factors; therefore, the present study will also be useful for model intercomparison studies elsewhere in Asia, such as the United Nations Environment Programme (UNEP)/Atmospheric Brown Clouds (ABC)‐Asia [Ramanathan et al., 2008], the Model Inter‐Comparison Study (MICS)‐Asia [Carmichael et al., 2008] and AeroCom projects [Textor et al., 2006; Myhre et al., 2013].

2 Model and Observation Description

2.1 Nonhydrostatic Icosahedral Atmospheric Model

NICAM, which employs an icosahedral grid point method with a nonhydrostatic equation system [Tomita and Satoh, 2004; Satoh et al., 2008, 2014], is run with a maximum horizontal resolution of 0.87 km [Miyamoto et al., 2013] and can be coupled with transport models of aerosols and gases as a conventional atmospheric general circulation model [Suzuki et al., 2008; Niwa et al., 2011b; Dai et al., 2014a, 2014b]. In the present study, we adopted the globally uniform grid system with a horizontal resolution of 220 km. In the vertical, the model uses a Lorenz grid and z* (terrain‐following) coordinates with 40 layers including 10 layers below a height of approximately 2 km. The advection scheme for atmospheric tracers is based on the works of Miura [2007] and Niwa et al. [2011a], which extend the scheme of van Leer [1977], preserving both monotonicity and continuity. NICAM implements comprehensive physical processes for aerosol, radiation, turbulence, and cloud dynamics. The aerosol model used is known as SPRINTARS [Takemura et al., 2000, 2002a, 2005; Goto et al., 2011], as described in detail in section 2.3. The radiation transfer model is a two‐stream k distribution radiation scheme, which incorporates scattering, absorption, and emissivity by aerosol and cloud particles as well as absorption by gaseous compounds [Nakajima et al., 2000; Sekiguchi and Nakaima, 2008]. The vertical turbulent scheme comprises the level 2 turbulence closure scheme described by Mellor and Yamada [1974, 1982] and modified by Nakanishi and Niino [2004, 2009] and Noda et al. [2009]. The cloud dynamics apply both the prognostic Arakawa‐Schubert‐type cumulus convection scheme [Arakawa and Schubert, 1974; Emori et al., 2001] and the prognostic large‐scale clouds described by Le Treut and Li [1991]. The large‐scale cloud module is based on the one‐moment bulk method for cloud water mixing ratio. The liquid water content (LWC) and cloud fraction (CF) are empirically calculated by total amount of LWC and water vapor depending on the air temperature. The autoconversion rate is parameterized by Berry [1967] to consider the aerosol lifetime effect [Suzuki et al., 2004; Takemura et al., 2005]. For a comparison of vertical velocity with MIROC, omega (ω) in NICAM is calculated with the hydrostatic approximation. Meteorological fields (only wind) above approximately 2 km height were nudged every 6 h with the National Center for Environmental Prediction Final Analysis (NCEP‐FNL) data (http://rda.ucar.edu/datasets/ds083.2/). The monthly averaged sea surface temperature was also prescribed using the NCEP‐FNL data. In this study, the integrated period for spin‐up is at least 1 year.

2.2 Model for Interdisciplinary Research on Climate

The MIROC, as an atmospheric general circulation model (GCM) and atmosphere‐ocean‐coupled GCM, has been developed by AORI/UT, NIES, and JAMSTEC [K‐1 Model Developers, 2004; Watanabe et al., 2010] and has contributed to international projects such as the Intergovernmental Panel on Climate Change and Coupled Model Intercomparison Project (CMIP) for more than 10 years [Meehl et al., 2000, 2005; Intergovernmental Panel on Climate Change, 2001, 2007; Taylor et al., 2012]. The MIROC, as a spectral transform model with hydrostatic approximation, calculates advection using a fourth‐order flux form of the monotonic van Leer scheme [van Leer, 1977], except in the vicinity of the poles where the flux form semi‐Lagrangian scheme of Lin and Rood [1996] is employed. The aerosols, radiation, cloud dynamics, and nudged meteorological fields are the same as those used in NICAM, as noted in section 2.1. The vertical diffusion is calculated by level 2 of the turbulence closure scheme described by Mellor and Yamada [1974]. In the present study, MIROC version 3.2 was used with the horizontal resolutions of T42, that is, 2.8125° by 2.8125° in latitude and longitude. In the vertical, the model uses a sigma coordinate system with 56 layers of sigma levels, including 10 layers below a height of approximately 2 km of the surface in most areas. In MIROC, the integrated period for spin‐up is also at least 1 year.

2.3 Spectral Radiation‐Transport Model for Aerosol Species

The SPRINTARS aerosol module, developed by Takemura et al. [2000, 2002a, 2002b, 2005] and Goto et al. [2011], was used in this study. The SPRINTARS model calculates the mass mixing ratios of the main tropospheric aerosol components, that is, carbonaceous aerosol (black carbon and both primary and secondary organic carbon), sulfate, soil dust, sea salt, and the precursor gases of sulfate, namely, SO₂ and dimethylsulfide. The aerosol module adopts a one‐moment bulk method, which prescribes the particle size distribution and considers processes of emission, advection, diffusion, sulfur chemistry, wet deposition, and dry deposition, including gravitational settling. All chemical components, except for a part of carbonaceous aerosols, are externally mixed with each other, whereas primary organic carbon and black carbon from burning sources are internally mixed. The particle size distributions of sulfate, internally mixed organic carbon with black carbon, pure organic carbon, and pure black carbon particles were assumed to be a logarithmic normal distribution with dry‐mode radii of 69.5, 100, 100, and 11.8 nm, respectively [Takemura et al., 2002a; Goto et al., 2011]. As for the sea salt, the particle size distribution is calculated by a bin‐type method with four bins ranging from 0.174 to 5.64 µm. The aerosol densities and refractive indices of all aerosols were set to the same values used by Takemura et al. [2002a] and Dai et al. [2014a]. Combinations of the precalculated cross sections of the extinction and simulated mixing ratios for each aerosol species provide the simulated aerosol extinction coefficient for each time step of the model [Dai et al., 2014a]. Deposition processes of the aerosols include both wet (through both rainout and washout) and dry (through both turbulence and gravity), whereas those of the precursor gases only include rainout and dry deposition by turbulence. Rainout represents the process through which aerosol or gas compounds in a cloud droplet are scavenged by precipitation. The interstitial fractions of sulfate were fixed at 0.5, whereas those of SO₂ were determined by Henry's law [Takemura et al., 2002a]. For the other aerosol components, the interstitial fractions of the hydrophilic components, that is, internally mixed organic carbon with black carbon, pure organic carbon, and sea salt, were fixed at 0.7, whereas those of the hydrophobic components, that is, pure black carbon and soil dust, were fixed at 0.9. The washout process, that is, the process through which aerosols in the atmosphere (not in clouds) are scavenged through collision with raindrops, is implemented as in Takemura et al. [2002a].

We used the emission inventories of anthropogenic SO₂ and black carbon provided by Zhang et al. [2009] during the Intercontinental Chemical Transport Experiment Phase B project in spring 2006 over Asia (the emission fluxes of the anthropogenic SO₂ are shown in Figure 1 with sites used for the model validation). As for sea‐salt particles, we calculated the emission fluxes from the oceans using a function of wind speed at the 10 m height [Monahan et al., 1986]. For the other areas, sources, and chemical compounds, we used the same emission inventories as those described by Goto et al. [2011]. These emissions were put into the lowest model level, and seasonality of all anthropogenic sources was not considered in this study. The sulfate is formed from both gas phase and aqueous phase reactions. In the gas phase, SO₂ is oxidized by OH, whereas SO₂ in the aqueous phase is oxidized by H₂O₂ and O₃. The monthly averaged oxidant distributions were prescribed from the global chemical transport model CHASER [Sudo et al., 2002]. The O₃ and H₂O₂ concentrations in the aqueous phase were restored to the CHASER values at the beginning of the next time step, although this method may cause an overestimation of the sulfate formation from aqueous phase reactions especially in winter [e.g., Koch et al., 1999].

Figure 1

Spatial distributions of the SO₂ emission fluxes in annual mean of 2006. The observation sites indicate their characteristics (urban or nonurban). The squares and circles represent urban and nonurban sites under both CAWNET and EANET measurements. The two crosses in black represent the stations of Cape Hedo and Fukue. The large rectangle in blue indicates the target region (110°–140°E, 25°–40°N) in this study.

2.4 Dataset for Model Evaluation

The mean sulfate mass concentrations over China were provided as part of CAWNET by X. Zhang et al. [2012] and Y. Zhang et al. [2012], whereas the SO₂ concentrations were not measured. Particles less than 10 µm were collected using a MiniVol™ air sampler (Airmetrics, OR) for 24 h at 13 sites (seven urban and six nonurban sites). According to X. Zhang et al. [2012] and Y. Zhang et al. [2012], the sampling points in the urban sites were typically 50–100 m above the city average elevation to capture well‐mixed aerosols in the planetary boundary layer. At the nonurban sites, the sampling points were selected as a representative area to escape the influence of local sources. The sulfate components were analyzed by ion chromatography. In other countries of East Asia, mean sulfate mass and mean SO₂ concentrations near the surface at more than 30 sites were observed by EANET (http://www.eanet.asia/index.html). Particles were collected on a Teflon filter using the filter pack method (without size cut, but it has been evaluated that sulfate particles up to diameters of 10 µm were collected efficiently, because the sizes of most sulfate particles are smaller than 2.5 µm) every 1 or 2 weeks and analyzed by ion chromatography under the quality control guidelines of EANET [Acid Deposition Monitoring Network in East Asia, EANET, 2003]. However, it should be noted that at three sites in Korea, the observed sulfate concentrations might not be representative accurate values for a month data, because the measurements are sparsely taken (once in 6 days). The EANET data sets have often been used for analysis and model evaluations [Carmichael et al., 2008; Aikawa et al., 2010; Kajino et al., 2012; Kuribayashi et al., 2012]. At Cape Hedo (128.25°E, 26.87°N) and Fukue (128.68°E, 32.75°N) over the East China Sea, sulfate aerosols (less than 1 µm) were measured every 10 min with an Aerodyne Aerosol Mass Spectrometer (AMS; Aerodyne Research, Inc., MA) under the operation of the NIES group [Takami et al., 2005, 2007]. A description of the AMS and its accuracy can be found elsewhere [Jayne et al., 2000; Allan et al., 2003a, 2003b]. These sites are shown in Figure 1.

At both Cape Hedo and Fukue, lidar measurements operated by the NIES group were also available for the model evaluation in terms of vertical profiles [Sugimoto et al., 2003; Shimizu et al., 2004]. The lidar measured the vertical profiles of the backscattering intensity at 532 and 1064 nm and the depolarization ratio at 532 nm. The backscattering intensity was converted to extinction, and the depolarization ratio distinguished extinction between spherical and nonspherical particles. In this study, we only used the vertical profile of extinction for the spherical particles. The detailed algorithm was provided by Sugimoto et al. [2003] and Shimizu et al. [2004].

For the meteorological fields, the simulated relative humidity (RH) and precipitation were evaluated by estimated RH from NCEP‐FNL data and estimated precipitation from the Global Precipitation Climatology Project (GPCP) [Adler et al., 2003], respectively.

3 Results

3.1 Meteorological Fields

The meteorological parameters that can strongly influence the distribution of aerosols and their precursors are cloud and precipitation, which are mainly determined by basic parameters, that is, RH, vertical velocity, and so on. In this study, the horizontal wind above 2 km height was nudged to minimize differences between NICAM and MIROC, whereas the NICAM‐simulated RH, cloud, precipitation, and vertical air motion would be different from the MIROC‐simulated fields.

The liquid water content (LWC), the cloud fraction (CF), and the cloud water removal rate, defined as a conversion rate from cloud to precipitation (CRCTP), below 2 km height strongly affect the sulfur concentrations. In Figure 2, RH, LWC, CF, and CRCTP below 2 km height and the precipitation simulated by NICAM and MIROC in latitudinal average (25°–30°N, 30°–35°N, 35°–40°N) are illustrated in 5° longitude intervals between 110°E and 140°E. The NICAM‐simulated and MIROC‐simulated RH below 2 km height were overestimated compared to the NCEP‐FNL‐estimated results in most seasons and regions by less than 20%. The differences in the RH between NICAM as well as MIROC and NCEP‐FNL in the region 110°–120°E were generally larger than those in the other areas. Since below 2 km height most clouds are formed by large‐scale condensation, we excluded cumulus clouds in this analysis. The NICAM‐simulated LWCs were generally greater than the MIROC‐simulated results, especially in the region 110°–125°E in January, April, and July by the mean range from 0.03 to 0.17 g/m³. In contrast, the NICAM‐simulated LWCs were smaller than the MIROC‐simulated results in the region 120°–140°E and 25°–30°N in January, the region 125°–140°E and 25°–35°N in July and the region 110°–140°E and 30°–40°N in October by less than 0.06 g/m³. In the region 110°–120°E and 30°–40°N in October, the magnitudes of the NICAM‐simulated LWC were comparable to those of the MIROC‐simulated results. In the spatial and four‐seasonal mean, NICAM‐simulated LWCs were greater than MIROC‐simulated ones by 0.08 g/m³ (or 59% relative difference, in the source region of 110°–120°E and 25°–40°N) and 0.05 g/m³ (or 56% relative difference, in the outflow region of 130°–140°E and 25°–40°N). The magnitudes of the NICAM‐simulated CF were generally greater than the MIROC‐simulated results by 0.16 (or 62% relative difference, in the source region of 110°–120°E and 25°–40°N) and 0.06 (or 30% relative difference, in the outflow region of 130°–140°E and 25°–40°N) in the spatial and four‐seasonal mean.

Figure 2

Relative humidity (RH) and large‐scale cloud parameters simulated by NICAM and MIROC: liquid water content (LWC), cloud fraction (CF), conversion rate of cloud water to precipitation (CRCTP), and surface precipitation flux. Results are regional means over 5° latitude and longitude intervals between 25°N–40°N and 110°E–140°E. The RH, LWC, CF, and CRCTP are further vertical means between the surface and 2 km height. The different colors and symbols are represented below the panel. The ‘REF’ means reanalysis results of GPCP regarding precipitation and NCEP/NCAR regarding RH. The line type and color indicate the two different models and the latitude bands.

The CRCTP and precipitation determine the removal rates of sulfate and SO₂ through wet deposition. Over the source region (110°–120°E), the NICAM‐simulated CRCTPs tend to be generally larger than the MIROC‐simulated ones, whereas over the outflow region (130°–140°E) the NICAM‐simulated ones tend to be smaller than the MIROC‐simulated ones. The differences in the regional mean values of CRCTP between NICAM and MIROC were estimated to be less than 20%. As for the precipitation, the NICAM‐simulated and MIROC‐simulated fluxes are compared with the results from GPCP in Figure 2. In addition, Figure 3 illustrates the seasonal and spatial pattern of the NICAM‐simulated, MIROC‐simulated, and GPCP‐estimated precipitation in the region (80°–160°E, 10°–60°N). Figure 3 also shows values of relative bias (B_r) and absolute bias (B_a) in the target region (110°–140°E, 25°–40°N) between the models and GPCP. In the target region in January, the NICAM‐simulated precipitation was overestimated compared to the GPCP result, with a B_r value of 84% and a B_a value of 25 mm/month, whereas the MIROC‐simulated precipitation was generally comparable to the GPCP result, with a B_r value of 21% and a B_a value of −2 mm/month. In April, the NICAM‐simulated precipitation was underestimated compared to the GPCP result, with a B_r value of −12% and a B_a value of −49 mm/month, whereas the MIROC‐simulated precipitation was overestimated compared to the GPCP result, with a B_r value of +41% and a B_a value of 20 mm/month. Especially, in the specific region (110°–120°E and 25°–35°N), the MIROC‐simulated precipitation was overestimated (Figure 2). In contrast, in July and October, the NICAM‐simulated and MIROC‐simulated precipitation were generally comparable to the GPCP results, with B_r values of less than ±5% and B_a values of approximately −30 mm/month (July) and less than −20 mm/month (October). Therefore, the differences in the CRCTP and precipitation fluxes between NICAM and MIROC depended on seasons and regions, whereas the NICAM‐simulated LWC and CF below 2 km height were generally larger than those simulated by MIROC in all seasons.

Figure 3

Spatial distributions of the precipitation fluxes simulated by NICAM and MIROC and estimated from GPCP. The square and numbers in white indicate the target region (110°–140°E and 25°–40°N) and values of relative bias (B_r) and absolute bias (B_a) in the target region. The B_r and B_a are defined as (C_m − C_o)/C_o in percentage and C_m − C_o in millimeter per month, respectively. The C_m and C_o represent the precipitation flux from the simulations and GPCP, respectively.

The following meteorological parameters can directly determine the vertical transport and affect the vertical distribution of the aerosols and their precursors in the lower atmosphere: vertical velocity of the airmass, vertical turbulent mixing, and convective clouds (even though in the target region convective clouds are limited and less important for the vertical transport). The vertical velocities of airmass are determined by omega (ω), defined as a vertical pressure velocity, and its standard deviation (σ_ω), which is governed by transient synoptic or mesoscale systems. The turbulent mixing processes are mainly determined by the diffusion coefficients (K) and the height of the planetary boundary layer (PBL). The meteorological parameters related to the vertical transport also include the bulk coefficient, which is used in the kinematic vertical turbulent sensible heat flux in the surface layer and strongly affects the dry deposition processes for SO₂. K and the dry deposition velocity for SO₂ are mainly governed by the subgrid‐scale turbulence closure scheme. In Figure 4, ω, σ_ω, and K below 2 km height, the PBL heights, and the dry deposition velocity for simulated SO₂ by NICAM and MIROC in latitudinal average (25°–30°N, 30°–35°N, 35°–40°N) are illustrated in 5° longitude intervals between 110°E and 140°E, as in Figure 2. To simplify the analysis in this section, we mainly focused on their four‐seasonal mean values in the two regions: source (110°–120°E and 25°–40°N) and outflow (130°–140°E and 25°–40°N).

Figure 4

Omega mean and its standard deviation, diffusion coefficient (K), PBL height, and dry deposition velocity for SO₂ (V_dep) simulated by NICAM and MIROC. Results are regional means over 5° latitude and longitude intervals between 25°N–40°N and 110°E–140°E. All parameters are further vertical means between the surface and 2 km height. The different colors and symbols are represented below the panel.

The NICAM‐simulated and MIROC‐simulated ω values in monthly averages were generally within ±0.05 Pa/s, and their differences between NICAM and MIROC (MIROC minus NICAM) were estimated to be −0.02 Pa/s in the source region and −0.01 Pa/s in the outflow region. The differences in σ_ω between NICAM and MIROC were also generally within ±0.05 Pa/s. The larger σ_ω values with the mean ω value of almost zero indicate stronger vertical transport in the lower atmosphere, due to transient synoptic or mesoscale systems. Although the NICAM‐simulated ω values tend to be positive especially in the region 110°–120°E and 35°–40°N in January, April, and October, the model results of ω and σ_ω indicate that both NICAM and MIROC have similar frequencies of upward and downward velocities. This may suggest that differences in the vertical transport caused by transient synoptic or mesoscale systems between NICAM and MIROC were small. The spatial and four‐seasonal mean values of the difference in σ_ω between NICAM and MIROC (MIROC minus NICAM) were calculated to be approximately −0.01 Pa/s in both the source and outflow regions.

Strong spatial and seasonal variations of K were found in both NICAM and MIROC, whereas the NICAM‐simulated K values were generally lower than the MIROC‐simulated ones, especially in the region 125°–140°E and 25°–40°E in January and the region 110°–120°E and 25°–40°E in October. In the source region, the regional mean values of NICAM‐simulated K were calculated to be less than 60 m²/s, whereas those of MIROC‐simulated K were calculated to be at most 360 m²/s in October and at most 190 m²/s in the other months. The spatial and four‐seasonal mean values of the difference in K between NICAM and MIROC (MIROC minus NICAM) were calculated to be +117 m²/s (or +530% relative difference) in the source region and +76 m²/s (or +97% relative difference) in the outflow region. In the source region, the PBL heights were approximately from 0.5 to 1.5 km without clear seasonality, whereas in the outflow region they were seasonally varied and ranged from 0.5 km (April and July) to 1.5 km (January and October). Generally, the differences in the PBL height between NICAM and MIROC were small. The spatial and four‐seasonal mean values of the difference in PBL height between NICAM and MIROC (MIROC minus NICAM) were calculated to be +0.2 km in the source region and −0.1 km in the outflow region.

The SO₂ dry deposition velocity also strongly affects the SO₂ concentrations, because the dry deposition contributes significantly to the SO₂ budget [e.g., Goto et al., 2011]. In both NICAM and MIROC, the magnitudes of SO₂ deposition velocity in the source region were larger than those in the outflow region by approximately 0.2 cm/s, because of differences in the surface resistance between land and ocean. The differences in the SO₂ dry deposition velocity between NICAM and MIROC were small. The spatial and four‐seasonal mean values of the difference in SO₂ deposition velocity between NICAM and MIROC (MIROC minus NICAM) were calculated to be −0.10 cm/s in the source region and +0.05 cm/s in the outflow region.

3.2 Sulfate

Figure 5 compares the sulfate mass concentrations between the simulations and observations over East Asia. Table 1 shows the statistical parameters (normalized mean bias, NMB, and correlation coefficient, R) calculated from the data in Figure 5 for all sites, the source region sites in China, and the outflow sites in Korea and Japan. The NMB is defined as NMB = (C_sim − C_obs)/C_obs where C_sim and C_obs are average concentrations from the simulations and observations. Figure 6 shows the spatial distributions of the surface sulfate mass concentrations simulated by NICAM and MIROC with the observed results by the CAWNET and EANET. In Figure 5 and Table 1, the NICAM‐simulated sulfate concentrations at all sites were underestimated compared to the observations, with NMB values ranging from −68% (January) to −54% (July), and moderate to high R values ranging from 0.49 (October) to 0.89 (July). At the source region sites in China, the underestimation of the NICAM‐simulated sulfate concentrations was remarkable, with NMB values ranging from −77% (January) to −63% (July) but moderate to high R values ranging from 0.48 (January) to 0.89 (July). For example, at the urban site of Xian (108.97°E, 34.43°N), the observed sulfate mass concentrations, ranging from 29.4 µg/m³ (April) to 91.5 µg/m³ (January), were much larger than the NICAM‐simulated ones, ranging from 3.8 µg/m³ (April) to 15.7 µg/m³ (July). The underestimations of the models were mainly because the horizontal resolution in this study could not fully resolve the heterogeneity of the air pollution around the urban sites, especially in China, and because at Dunhuang (94.68°E, 40.15°N), Lhasa (91.13°E, 29.67°N), and Mongolian and near Mongolian sites, the emission inventory used in this study was underestimated. In contrast, at the outflow sites (Korea and Japan), NMB values simulated by NICAM ranged from −67% (October) to +3% (July) and high R values ranged from 0.62 (July) to 0.92 (April). At all sites, the results of the MIROC‐simulated sulfate concentrations show the same tendencies as those obtained by NICAM, except for at the source region sites in October and at the outflow sites in January. At the source region sites in October, the MIROC‐simulated sulfate was closer to the observation than the NICAM‐simulated sulfate, with an NMB value of −38% and a high R value of 0.93. This difference in the sulfate between NICAM and MIROC is further discussed in section 4.2. At the outflow sites in January, the MIROC‐simulated sulfate was much underestimated compared to that simulated by NICAM, with an NMB value of −60% and an R value of 0.74 (which is comparable to the one obtained by NICAM). Generally, both the NICAM‐simulated and MIROC‐simulated sulfate mass concentrations were underestimated compared to the observations, but they were not fully explained by the coarse resolution (200–300 km grids) in this study, because results of regional aerosol transport models [e.g., Kajino et al., 2012; Matsui et al., 2011] using finer resolutions (50–60 km grids) would still be underestimated compared to the measurements. Therefore, the underestimation of the simulated sulfate mass concentrations may be a common problem among the aerosol modeling community.

Figure 5

Scatterplots of the sulfate concentrations simulated by NICAM and MIROC at the sites of CAWNET and EANET shown below the panel using different colors depending on locations. The squares in yellow, circles in red, triangles in green, diamonds in blue represent sites in the region of 25°N–42°N China, Korea‐Japan, the regions south of 25°N in China, and north of 42°N in China‐Mongolia‐Russia. The asterisks in purple represent remote sites in the Pacific Ocean. The open and closed symbols represent urban and nonurban sites.

Table 1. Statistical Parameters, That Is, Mean Concentration, Normalized Mean Bias (NMB), and Correlation Coefficient (R), Calculated Between the Simulated and Observed Sulfate and SO₂ Concentrations at the Sites of CAWNET and EANET, the Source Region Sites in China, and Outflow Sites in Korea and Japana^a The NMB is defined as NMB = (C_m − C_o)/C_o where C_m and C_o represent concentrations from the simulations and observations.

Characteristics	Model/Observation	January	April	July	October
Sulfate
Mean (µg/m^3)
All sites	Observation	14.02	10.80	9.86	12.46
Source region sites in China	Observation	33.69	23.38	24.73	27.96
Outflow sites in Korea and Japan	Observation	4.80	5.28	3.41	6.06
NMB (%)
All sites	NICAM	−68	−61	−54	−58
	MIROC	−61	−60	−54	−28
Source region sites in China	NICAM	−77	−74	−63	−76
	MIROC	−71	−68	−59	−38
Outflow sites in Korea and Japan	NICAM	−29	−17	+3	−67
	MIROC	−60	−22	+2	−52
R
All sites	NICAM	0.64	0.60	0.89	0.49
	MIROC	0.61	0.70	0.75	0.77
Source region sites in China	NICAM	0.48	0.51	0.89	0.55
	MIROC	0.44	0.56	0.63	0.93
Outflow sites in Korea and Japan	NICAM	0.73	0.92	0.62	0.76
	MIROC	0.74	0.92	0.22	0.50
SO2
Mean (ppbv)
All sites	Observation	5.38	2.84	1.80	2.37
Source region sites in China	Observation	24.08	11.05	7.35	10.10
Outflow sites in Korea and Japan	Observation	1.78	1.06	0.73	0.81
NMB (%)
All sites	NICAM	−91	−81	−63	−85
	MIROC	−65	−40	−21	−40
Source region sites in China	NICAM	−94	−87	−79	−86
	MIROC	−67	−40	−23	−39
Outflow sites in Korea and Japan	NICAM	−86	−62	−27	−80
	MIROC	−67	−28	+22	−48
R
All sites	NICAM	0.65	0.71	0.66	0.74
	MIROC	0.90	0.79	0.86	0.77
Source region sites in China	NICAM	0.18	0.74	0.54	0.73
	MIROC	0.91	0.83	0.94	0.63
Outflow sites in Korea and Japan	NICAM	0.53	0.28	0.34	−0.07
	MIROC	0.51	−0.24	−0.39	0.36

• ^a The NMB is defined as NMB = (C_m − C_o)/C_o where C_m and C_o represent concentrations from the simulations and observations.

Figure 6

Spatial distributions of the surface sulfate mass concentrations simulated by NICAM and MIROC with those observed by CAWNET and EANET. The square and circle symbols represent nonurban and urban sites, respectively.

To evaluate how simulated sulfate concentrations compared to observed values in terms of their temporal variation, we compared the model results with the observations using the AMS in the outflow regions at Cape Hedo and Fukue (Figure 7). Figure 7 represents the normalized frequency of the sulfate concentrations obtained by NICAM, MIROC, and the observations. At Cape Hedo, the normalized frequency distributions of the NICAM‐simulated and MIROC‐simulated sulfate concentrations in January, April, and July were generally comparable to that obtained by the AMS observations. Especially, the contrast between the normalized frequency distributions of July and those of other seasons (January and April) was prominent in both the simulations and observations, because the sulfate mass concentrations at Cape Hedo were lower in July, mainly because in summer the southern wind from the oceans was stronger due to the North Pacific High (or Ogasawara High). In January and April, approximately 30% of the NICAM‐simulated sulfate concentrations and 22–30% of the MIROC‐simulated sulfate concentrations exceeded 5 µg/m³, whereas approximately 50% of the observed sulfate concentrations exceeded this values. In October at Cape Hedo, however, the NICAM‐simulated and MIROC‐simulated sulfate concentrations were underestimated compared to that obtained by the AMS observations, as shown in Figure 6 where the NICAM‐simulated and MIROC‐simulated sulfate concentrations at the western part of Japan were underestimated compared to the observations. The underestimation is possibly caused by the underestimation of transboundary air pollution from the continents. In contrast, at Fukue in April, the normalized frequency distributions of the NICAM‐simulated sulfate concentration above 5 µg/m³ were overestimated compared to the observations. Approximately 50% of the NICAM‐simulated and MIROC‐simulated sulfate concentrations exceeded 5 µg/m³, whereas only 12% of the observed sulfate concentrations exceeded that value. In Figure 6, however, the NICAM‐simulated and MIROC‐simulated sulfate concentrations at the western part of Japan in April were comparable to the observation. It should be noted that such differences in the average values between the observations by the filter pack method and by the AMS were sometimes seen, because the AMS can measure the temporal variations of the aerosol mass concentrations, but it is less accurate than the filter measurements.

Figure 7

Normalized frequency distributions of the surface sulfate mass concentrations at Cape Hedo and Fukue using the simulations by NICAM and MIROC and the AMS measurements.

3.3 SO₂ and Sulfate‐to‐Sulfur Ratio

To validate the model‐derived sulfur compounds (sulfate and SO₂), we evaluated the near‐surface SO₂ and sulfate‐to‐sulfur ratio, the latter defined as molar mixing ratios of sulfate to the sum of sulfate and SO₂. Figures 8 and 9 illustrate the SO₂ and the sulfate‐to‐sulfur ratio obtained from NICAM, MIROC, and the EANET observations. Because SO₂ is oxidized to sulfate in the atmosphere during long‐range transport, the sulfate‐to‐sulfur ratio is lower in urban areas and higher in downwind regions. Therefore, the agreement in the sulfate‐to‐sulfur ratio between the model and observation suggests a consistent ratio between removal and oxidation in the models. At the sites over the eastern part of Japan and two remote islands, that is, Hedo (number 26 in Figure 5) and Ogasawara (number 27 in Figure 5), the NICAM‐simulated SO₂ values were closer to the observations compared to the MIROC‐simulated results (Figure 8). At China and Korea sites as well as the sites over the western part of Japan, however, the NICAM‐simulated SO₂ concentrations were underestimated compared to the observed and MIROC‐simulated results. Especially, at the source region sites in China, the NICAM‐simulated SO₂ concentrations reached less than 5 ppbv, whereas the observed and MIROC‐simulated SO₂ concentrations reached more than 10 ppbv. NMB values between NICAM and MIROC were calculated to be 26% (January), 47% (April), 56% (July), and 47% (October). Table 1 (sulfate) shows NMB and R of SO₂ calculated from the data of Figure 8. At all sites, the underestimation of SO₂ concentration obtained by NICAM was remarkable with NMB values ranging from −91% (January) to −63% (July) and moderate to high R values ranging from 0.65 (January) to 0.74 (October). MIROC‐simulated SO₂ concentrations were also underestimated compared to the observations but larger than the NICAM ones with NMB values ranging from −65% (January) to −21% (July) and high R values ranging from 0.77 (October) to 0.90 (January). The high R values obtained by both NICAM and MIROC mean that the horizontal distributions of SO₂ simulated by NICAM and MIROC were generally close to those obtained by the observations. In contrast, at the outflow sites, R values tend to be lower in both NICAM and MIROC, except for in January, with an R value of less than 0.4. Generally, discrepancies in SO₂ between the simulations and observations are larger than those in sulfate in both global models [e.g., Chin et al., 2000; Easter et al., 2004] and regional models [e.g., de Meij et al., 2006; X. Zhang et al., 2012; Y. Zhang et al., 2012]. Because the atmospheric lifetime of SO₂ is approximately within 1–2 days [e.g., Roelofs et al., 2001], SO₂ tends to be localized and the SO₂ concentrations in monitoring sites can be strongly affected by the nearest SO₂ emission sources in the simulation and probably in the real atmosphere. In addition, the observed SO₂ had a clear seasonality with the peak in January, whereas the both NICAM‐simulated and MIROC‐simulated SO₂ do not have such strong seasonality, primarily because the seasonality of SO₂ emissions in China (higher in winter) was not taken into account for the emission inventories used in this study and possibly because the seasonality of some removal processes may be missed.

Figure 8

Spatial distributions of the surface SO₂ concentrations simulated by NICAM and MIROC compared with those observed by EANET. The square and circle symbols represent nonurban and urban sites, respectively.

Figure 9

Spatial distributions of the sulfate‐to‐sulfur ratios simulated by NICAM and MIROC compared with those observed by EANET. The square and circle symbols represent nonurban and urban sites, respectively.

As shown in the previous section, the differences in the simulated sulfate were smaller than those in the simulated SO₂. Therefore, the NICAM‐simulated sulfate‐to‐sulfur ratios were higher than the MIROC‐simulated results. Figure 9 indicates that in most areas in all seasons, the NICAM ratios were approximately 0.2 greater than the MIROC ratios. Over the outflow region in Japan, the values of the NICAM‐simulated fraction were closer to those obtained by the observations, compared to those obtained by MIROC. In addition, over the same region, the values of the NICAM‐simulated and MIROC‐simulated fractions as well as the observations were highest in October and lowest in January. Over the same region, the sulfate‐to‐sulfur ratios in October were estimated to be 0.5–0.9 (NICAM), 0.3–0.6 (MIROC), and 0.6–0.9 (observation), whereas those in January were estimated to be 0.3–0.8 (NICAM), 0.1–0.5 (MIROC), and 0.3–0.8 (observation). These results indicate that NICAM more accurately reproduced the observed sulfate and SO₂ values over the outflow regions around Japan, although the NICAM‐simulated sulfate and SO₂ concentrations tend to be underestimated compared to the observations.

3.4 Vertical Distribution

The vertical profiles of NICAM‐simulated and MIROC‐simulated extinction were also evaluated in the outflow region (Cape Hedo and Fukue) using lidar‐derived extinction for the spherical particles, as shown in Figure 10. The extinction values depend on the sizes and total mass concentrations of spherical particles, which includes water that in turn depends on RH and particle hygroscopicity. The total mass concentrations of the spherical particles over these two sites are primarily sulfate, carbonaceous components, sea salt, and water. The impact of dust on the extinction values even in April is small. Although aged and wetted dust particles may be spherical, these mostly occur over these sites on days when dust concentrations and contributions to extinction are small. In contrast, during the dusty days, most of the dust is observed as nonspherical particles. Therefore, Figure 10 shows the vertical profiles of the simulated extinction of the sum of sulfate, carbonaceous, and sea‐salt aerosols under ambient RH conditions as well as the simulated extinction calculated using sulfate particles in dry conditions and under the ambient RH conditions. Above 1–2 km, the vertical profiles of the NICAM‐simulated extinction for total spherical particles were closer to the observations than the MIROC extinction in January, April, July, and October. Below 1–2 km, both the NICAM and MIROC extinction were lower than the observations. Above approximately 1 km height, the magnitudes of the NICAM‐simulated extinction for the total spherical particles in the ambient conditions as well as for the pure sulfate at the ambient and dry conditions were larger than those obtained from MIROC at both sites in all seasons. Therefore, the differences in the pure sulfate between NICAM and MIROC directly influence the differences in the total spherical particles between MIROC and NICAM. In contrast, below approximately 1 km height, the differences in the extinction due to pure sulfate in dry conditions between NICAM and MIROC are larger than those in the ambient conditions between NICAM and MIROC. That means that below 1 km height the differences in the RH between NICAM and MIROC affect the simulated extinction in the ambient conditions.

Figure 10

Vertical profiles of the extinction for total spherical particles (sulfate, carbonaceous, and sea‐salt aerosols) under the ambient RH using NICAM and MIROC and the NIES lidar observations as well as the extinction for only sulfate under dry and ambient RH conditions obtained by NICAM and MIROC in units of 1/Mm. The bars represent the 25th and 75th percentiles of the lidar observations. The solid and dashed lines in different colors are represented below the panel.

To clarify the differences in the sulfur compounds between above and below 2 km height over the main target region (110°–140°E and 25°–40°N) including the two sites (Cape Hedo and Fukue), we showed mixing ratios of sulfate and SO₂ simulated by NICAM and MIROC in 0–2 and 2–4 km height averages in Figure 11. In Table 2, we also defined the four‐seasonal mean values of the mixing ratios in the two regions: source (110°–120°E and 25°–40°N) and outflow (130°–140°E and 25°–40°N). Table 2 indicates that below 2 km height, in both the source and outflow regions, the NICAM‐simulated sulfate mixing ratios were higher than the MIROC‐simulated ones, whereas the NICAM‐simulated SO₂ mixing ratios were lower than MIROC‐simulated ones. Figure 11 also indicates these general differences in sulfate and SO₂ between NICAM and MIROC. An exception is in the region 110°–120°E and 30°–40°N in October, when the NICAM‐simulated sulfate mixing ratios were lower than the MIROC‐simulated ones, and this is discussed in section 4.2. In addition, the mixing ratios of the NICAM‐simulated sulfate were higher than those of the NICAM‐simulated SO₂, whereas those of the MIROC‐simulated sulfate were lower than those of the MIROC‐simulated SO₂ in the source region and comparable to those of the MIROC‐simulated SO₂ in the outflow region. These results are also found in the results obtained by the comparison at the surface shown in Figures 6, 8, and 9. Furthermore, in the source region, the mixing ratios of the sum of sulfate and SO₂ obtained by NICAM (4.05 nmol/mol) were lower than those obtained by MIROC (6.70 nmol/mol), whereas in the outflow region, those obtained by NICAM (0.82 nmol/mol) were comparable to those obtained by MIROC (0.87 nmol/mol). As a result, in the source region, the sulfate‐to‐sulfur ratios obtained by NICAM (71%) were much higher than those obtained by MIROC (35%).

Figure 11

The mixing ratios of sulfate and SO₂ in 0–2 and 2–4 km height averages simulated by NICAM and MIROC in latitudinal average (25°–30°N, 30°–35°N, 35°–40°N) over 5° longitude intervals between 110°E and 140°E. The solid and dashed lines in different colors are represented below the panel.

Table 2. Spatial and Four‐Seasonal Mean Values of the Mixing Ratios of Sulfate, SO₂, and the Sum of Them, and Sulfate‐to‐Sulfur Ratios in the Source and Outflow Regions

	Source Region (110°–120°E and 25°–40°N)		Outflow Region (130°–140°E and 25°–40°N)
	NICAM	MIROC	NICAM	MIROC
0–2 km Height Average
Sulfate (nmol/mol)	2.89	2.28	0.64	0.44
SO2 (nmol/mol)	1.17	4.42	0.18	0.44
Sulfate + SO2 (nmol/mol)	4.05	6.70	0.82	0.87
Sulfate‐to‐sulfur ratio (%)	71	35	79	53
2–4 km Height Average
Sulfate (nmol/mol)	1.60	0.34	0.50	0.11
SO2 (nmol/mol)	0.51	0.30	0.12	0.09
Sulfate + SO2 (nmol/mol)	2.10	0.64	0.62	0.20
Sulfate‐to‐sulfur ratio (%)	73	57	80	60

At 2–4 km height in all seasons, the important feature is that the mixing ratios of both sulfate and SO₂ simulated by NICAM were generally higher than those simulated by MIROC (Figure 11 and Table 2). In addition, Table 2 indicates that the mixing ratios of the sum of sulfate and SO₂ obtained by NICAM in both the source and outflow regions were higher than those obtained by MIROC. In terms of the sulfate‐to‐sulfur ratios in the source region, Table 2 also shows that NICAM‐calculated values at 2–4 km height (approximately 70%) were comparable to those at 0–2 km height, whereas MIROC‐calculated values at 2–4 km height (57%) were higher than those at 0–2 km height (35%). In the outflow region, NICAM‐calculated and MIROC‐calculated values of the sulfate‐to‐sulfur ratios at both 0–2 and 2–4 km heights were approximately 60% (Table 2).

In summary, in both layers (0–2 and 2–4 km) and regions (source/outflow), the mixing ratios of the NICAM‐simulated sulfate were generally higher than those of the MIROC‐simulated sulfate and the NICAM‐simulated SO₂. Below 2 km height, the mixing ratios of the NICAM‐simulated SO₂ were lower than those of the MIROC‐simulated SO₂, whereas above 2 km height, those of the NICAM‐simulated SO₂ were higher than those of the MIROC‐simulated SO₂. These differences suggest that NICAM‐simulated SO₂ was more frequently converted into sulfate, and SO₂ as well as sulfate was more distributed above 2 km height, compared to the results obtained by MIROC.

4 Discussion

4.1 Sulfur Budget

We found some differences in the sulfate and SO₂ concentrations between NICAM and MIROC over the source/outflow regions above/below 2 km heights. To further investigate these differences between NICAM and MIROC, we plotted the column burden tendency of SO₂ oxidation to sulfate in the aqueous and gas phases and the budget fluxes of the sulfur compounds (sulfate and SO₂) in latitudinal average (25°–30°N, 30°–35°N, 35°–40°N) in 5° longitude intervals between 110°E and 140°E (Figure 12). We also calculated the four‐seasonal mean values of these parameters and additional ones (the emission flux, the inferred export flux, and the column burden of sulfate and SO₂) for the overall source region (110°–120°E and 25°–40°N), which are shown in Table 3. The sulfur budget includes the oxidation of SO₂ to sulfate in the aqueous phase (within clouds) and in the gas phase (outside of clouds) and the removal through dry and wet deposition of the sulfur compounds. The total sulfur column burden (sulfate plus SO₂) for the source region obtained by NICAM was about 30% higher than that obtained by MIROC. This is because the combined sink processes in MIROC were somewhat more rapid, as indicated by the shorter lifetime for total sulfur in MIROC (1.1 days) compared to NICAM (1.6 days). Much larger differences are seen in the individual SO₂ and sulfate burdens and sinks. In the source region, the greatest contributor to the SO₂ budget in NICAM was SO₂ oxidation within clouds to form sulfate, with a spatial and four‐seasonal mean value of 200 mg S/m²/month (64% of the total SO₂ loss processes), whereas the greatest contributor in MIROC was SO₂ dry deposition, with a spatial and four‐seasonal mean value of 148 mg S/m²/month (41% of the total SO₂ loss processes). In contrast, the spatial and four‐seasonal mean value of the SO₂ dry deposition in NICAM was calculated to be 33 mg S/m²/month (10% of the total SO₂ loss processes), whereas that of the SO₂ oxidation within clouds in MIROC was calculated to be 70 mg S/m²/month (20% of the total SO₂ loss processes). In NICAM, the amount of wet deposition for SO₂ (39 mg S/m²/month or 12% of the total SO₂ loss processes) was larger than that of dry deposition (33 mg S/m²/month). In terms of the inferred export flux of SO₂ from the source region, the mean value obtained by NICAM (22 mg S/m²/month or 7% of the total SO₂ loss processes) was lower than that obtained by MIROC (76 mg S/m²/month or 21% of the total SO₂ loss processes). Note that these are net exports, and the actual exports are likely somewhat larger because these is some SO₂ inflow along the western boundary of the source region. This finding suggests that the difference in the greatest contributors to the sulfur budget in the source region between NICAM and MIROC can be explained by the difference in the cloud distribution, the removal process, and the vertical transport of sulfate and SO₂, especially below 2 km height. Over China in the source region (110°–120°E and 25°−40°N), NICAM simulated more LWC below 2 km height than MIROC, as shown in Figure 2, which can cause larger budgets of SO₂ oxidation within clouds in NICAM. The difference in LWC between NICAM and MIROC was probably caused by the difference in RH between NICAM and MIROC (Figure 2). In addition, Figure 11 and a previous study by Goto [2014] indicate that the aerosol and its related tracers in NICAM tended to be lifted to higher levels (up to approximately 2 km height) compared to those in MIROC. As a result, more SO₂ in NICAM is easily converted to sulfate within plenty of clouds, whereas more SO₂ in MIROC is either removed (by dry deposition primarily) or transported to the outflow regions. Therefore, the NICAM‐simulated SO₂ concentrations near the surface were generally lower than the MIROC‐simulated SO₂ ones. In conclusion, the difference in the cloud distribution as well as the vertical transport patterns between the models causes the difference in the modeled sulfate and SO₂ distribution, as a result of the difference in the budget of SO₂ oxidation in clouds, and dry and wet deposition of SO₂.

Figure 12

The budget fluxes of the sulfur compounds (sum of SO₂ and sulfate and sulfate) in latitudinal averages (25°–30°N, 30°–35°N, 35°–40°N) over 5° longitude intervals between 110°E and 140°E. The solid and dashed lines in different colors are represented below the panel. The terms “CFSO2A,” “CFSO2G,” “SO2DRYF,” “SO2WETF,” “SUWETF,” and “TDEP” represent the column burdens tendency of SO₂ oxidation to sulfate in the aqueous phase, column burdens tendency of SO₂ oxidation to sulfate in the gas phase, dry deposition flux of SO₂, wet deposition flux of SO₂, wet deposition flux of sulfate, and total (dry and wet) deposition flux of sulfur (sum of SO₂ and sulfate), respectively.

Table 3. Spatial and Four‐Seasonal Mean Values of the Budget Fluxes of the Sulfur Compounds (Sulfate and SO₂) in the Source Region

Regional Budgeta^a Units are mg S/m^2/month (flux), mg S/m^2 (burden), and days (lifetime). The percentage in bracket represents a ratio of each flux to the total loss processes.	Source Region (110°–120°E, 25°–40°N)
	NICAM	MIROC
SO2
Emissiona^a Units are mg S/m^2/month (flux), mg S/m^2 (burden), and days (lifetime). The percentage in bracket represents a ratio of each flux to the total loss processes.	314d^d Although the same emission inventory was used in NICAM and MIROC, relative difference in the emission flux between NICAM and MIROC is approximately 10%, because of a difference in the grid system between NICAM and MIROC under the coarse resolution.	356d^d Although the same emission inventory was used in NICAM and MIROC, relative difference in the emission flux between NICAM and MIROC is approximately 10%, because of a difference in the grid system between NICAM and MIROC under the coarse resolution.
SO2 oxidation in the gas phase	21 (7%)	35 (10%)
SO2 oxidation in the aqueous phase	200 (64%)	70 (20%)
Dry deposition	33 (10%)	148 (41%)
Wet deposition	39 (12%)	27 (8%)
Inferred exportb^b Inferred export is estimated from the difference in a flux between the emission and the total loss processes.	22 (7%)	76 (21%)
Column burden	5.1	9.2
Sulfate
Total production	221	105
Dry deposition	19 (8%)	13 (13%)
Wet deposition	134 (61%)	39 (37%)
Inferred exportb^b Inferred export is estimated from the difference in a flux between the emission and the total loss processes.	68 (31%)	53 (50%)
Column burden	11.9	3.9
Sulfate + SO2
Column burden	17.0	13.1
Lifetimec^c Lifetime is defined as a ratio of column burden to total sinks.	1.6	1.1

• ^a Units are mg S/m²/month (flux), mg S/m² (burden), and days (lifetime). The percentage in bracket represents a ratio of each flux to the total loss processes.
• ^b Inferred export is estimated from the difference in a flux between the emission and the total loss processes.
• ^c Lifetime is defined as a ratio of column burden to total sinks.
• ^d Although the same emission inventory was used in NICAM and MIROC, relative difference in the emission flux between NICAM and MIROC is approximately 10%, because of a difference in the grid system between NICAM and MIROC under the coarse resolution.

4.2 Seasonal Analysis of Sulfate

In this section, we discuss the differences in the seasonality of the sulfate between NICAM and MIROC. Table 1 indicates that both NICAM‐simulated and MIROC‐simulated sulfate mass concentrations generally had similar seasonal variations to the observed ones, with maximum in January and October and minimum in April and July. However, as mentioned in sections 3.2 and 3.4, relatively large differences in the simulated sulfate between NICAM and MIROC over the source region in October were found.

Figure 11 also shows that below 2 km height in the source region (110°–120°E and 25°–40°N) the MIROC‐simulated sulfate concentrations in October were higher than the MIROC‐simulated ones in the other months and the NICAM‐simulated ones in the majority of the source region (110°–120°E and 30°–40°N). The differences in MIROC‐simulated sulfate concentrations between October and the other months are attributed to the larger values of the chemical production of sulfate in the cloud (CFSO2A in Figure 12) compared to ones in the other months. In October, at the majority in the source region (110°–120°E and 30°–40°N), the MIROC‐simulated LWC was higher than the NICAM‐simulated one, whereas in the other months and other regions in the target area, the MIROC‐simulated LWC was lower than the NICAM‐simulated one. The difference in the LWC between NICAM and MIROC is possibly caused by the difference in the RH (Figure 2). In addition, in October at the source regions, the differences in the diffusion coefficient, K, shown in Figure 4 were greater than those in the other months, whereas the other physical parameters shown in Figures 2 and 4 were not so different from those in the other months. Therefore, in MIROC, the larger values of K below 2 km height over the source regions in October probably also cause the increases in the sulfate production in the clouds through the acceleration of vertical transport of SO₂ to the heights where lower level clouds exist. Conversely, NICAM reproduced relatively small seasonal variations of simulated sulfate concentrations in China, probably because of the small seasonal variations of NICAM‐simulated K and clouds (LWC and CF). Above 2 km height at all regions, the NICAM‐simulated sulfate and SO₂ concentrations were higher than the MIROC‐simulated ones even in October. It might be caused by the difference in the model framework including the advection scheme, the coordinate system, the grid system, and/or the hydrostatic approximation between NICAM and MIROC. The uncertainties of the simulated LWC among global models participating in CMIP as well as retrieved LWC from various satellites are very large [e.g., Li et al., 2008]. This implies that the differences in the basic meteorological and cloud fields including LWC between NICAM and MIROC are within the model uncertainty. The present study indicates that the possible differences in the clouds and vertical transport have impacts on the sulfate distribution.

5 Summary

Over East Asia, the spatial gradient of sulfate aerosols from source regions such as China to outflow regions such as Japan has not been fully evaluated using a combination of measurements and simulations, primarily due to a lack of available measurements in East Asia, especially in China. Recently, new measurements in China under the CAWNET project have become available for use in model validation. Using CAWNET, EANET, highly time‐resolved measurements by AMS and vertical profiles of aerosol extinction for spherical particles from lidar, we executed a global aerosol transport model, SPRINTARS, coupled to NICAM at a spatial resolution (220 km) that is widely used in climate models for January, April, July, and October 2006. Because SPRINTARS coupled to a different host climate model, MIROC, has been compared with various measurements and other aerosol transport models as part of international model intercomparisons such as AeroCom, SPRINTARS coupled to MIROC was also used for the model evaluation of SPRINTARS coupled to NICAM.

In the present study, the meteorological and cloud parameters simulated by NICAM and MIROC were compared over East Asia. Especially, in the main target region (110°–140°E and 25°–40°N) including the aerosol sources over China (110°–120°E and 25°–40°N) to the outflow over Japan (130°–140°E and 25°–40°N), the NICAM‐simulated large‐scale cloud liquid water content below 2 km height was greater than MIROC‐simulated one by 0.08 g/m³ or 59% (source region) and 0.05 g/m³ or 56% (outflow region) in the spatial and four‐seasonal mean. The magnitudes of the NICAM‐simulated cloud fraction below 2 km height were generally greater than the MIROC‐simulated results with differences of 0.16 (62%) over the source region and 0.06 (30%) over the outflow region in the spatial and four‐seasonal mean. Among the meteorological parameters related to the vertical transport, the diffusion coefficient (K) has a remarkable difference between NICAM and MIROC. The MIROC‐simulated K values were greater than those simulated by NICAM by +117 m²/s (or +530% relative difference) in the source region and +76 m²/s (or +97% relative difference) in the outflow region.

The horizontal gradient of sulfate distributions from China through Korea to Japan with longitude was clearly present in both the simulations and the observations. The correlation coefficients of the sulfate concentrations between the simulations and the observations were high (NICAM: 0.49–0.89, MIROC: 0.61–0.77), whereas the simulated sulfate concentrations were lower than those observed with the normalized mean bias (NMB) of the NICAM‐simulated sulfate concentrations to observed ones being −68 to −54% (all sites), −77 to −63% (source region sites), and −67 to +3% (outflow sites) and the NMB of the MIROC to the observation being −61 to −28% (all sites), −71 to −38% (source region sites), and −60 to +2% (outflow sites). The NICAM‐simulated SO₂ and the sulfate‐to‐sulfur ratios were also generally comparable to the observations in the western part of Japan, whereas the NICAM‐simulated SO₂ was less than the MIROC‐simulated SO₂, especially at Chinese sites, by more than 5 ppbv (or NMB values between NICAM and MIROC range from 26% in January to 56% in July).

To explore the differences in the sulfur compounds between NICAM and MIROC, we compared the vertical profiles and their budgets in latitudinal averages (25°–30°N, 30°–35°N, 35°–40°N) over 5° longitude intervals between 110°E and 140°E. We found that the NICAM‐simulated sulfate concentrations, especially above 2 km height, were higher than the MIROC‐simulated ones in all seasons at least in the main target region (110°–140°E and 25°−40°N) including two focusing sites (Cape Hedo and Fukue). We also found that in the source region (110°–120°E and 25°–40°N), the production rate of sulfate from SO₂ within clouds in NICAM was greater than that obtained by MIROC, whereas the dry deposition rate of SO₂ in NICAM was less than that obtained by MIROC. As for seasonal variation of the sulfur components, relatively large differences in the simulated sulfate between NICAM and MIROC over the source regions were found in October. The analysis of the difference indicated that the distribution of the sulfur compounds was mainly affected by the cloud fields and partly affected by the vertical transport through modulating the sulfate formation within clouds.

Therefore, the differences in the meteorological and large‐scale cloud parameters between NICAM and MIROC cause the differences in the budget fluxes of the sulfur compounds and thus the sulfate and SO₂ distributions. Our findings clearly indicate that the uncertainty of the meteorological and the cloud fields as well as the vertical transport patterns, especially below 2 km height, between the different host climate models has a substantial impact on the simulated sulfate distribution.

Acknowledgments

Data used to support this article can be obtained by request to the corresponding author ([e-mail]). We thank the model developers of SPRINTARS (http://sprintars.riam.kyushu‐u.ac.jp/indexe.html), NICAM (http://nicam.jp/), and MIROC (http://www.jamstec.go.jp/sousei/eng/index.html). We also acknowledge the measurements of CAWNET (shown in the literature of http://www.atmos‐chem‐phys.net/12/779/2012/acp‐12‐779‐2012.html), EANET (available at http://www.eanet.asia/index.html), the AMS data at Cape Hedo and Fukue (acquired by contacting [e-mail]), NIES‐lidar at Cape Hedo and Fukue (acquired by contacting [e-mail] and [e-mail]), NCEP (downloaded from http://rda.ucar.edu/datasets/ds083.2/), and GPCP (downloaded from http://www.gewex.org/gpcp.html). Some of the authors were supported by projects from the Grant‐in‐Aid for Scientific Research on Innovative Areas (grant 24110002), the Grant‐in‐Aid for Young Scientist B (grant 26740010), the UNEP/ABC‐Asia project, the SALSA project of the Research Program on Climate Change Adaptation (RECCA) in the Ministry of Education and Sports in Japan (MEXT) (grant 10101026), the Global Environment Research Fund S‐12 (grant 14426634) of the Ministry of Environment in Japan, MOE/GOSAT, JST/CREST/EMS/TEEDDA (grant 12101625), JAXA/EarthCARE, GCOM‐C, MEXT/VL for climate diagnostics, and the MEXT/KAKENHI/Innovative Areas 2409. The model simulations were performed using the supercomputer systems of the HITACHI SR16000 System (yayoi) in the Information Technology Center at the University of Tokyo, Japan, and the NEC SX‐9/A(ECO) at the National Institute for Environmental Studies, Japan. We also acknowledge H. Yashiro, Y. Niwa, N. Oshima, and K. Aoki for providing helpful comments on this paper.