We run CAM5 (Neale et al., 2010) with the three-mode Modal Aerosol Model (MAM3), which prognoses aerosol mass/number size distribution and mixing state in the Aitken, accumulation, and coarse modes (Liu et al., 2012). The simulated primary aerosol species include BC, primary organic matter (POM), sea salt, and dust, while the secondary aerosol species include sulfate and secondary organic aerosol (SOA). The aerosol species are assumed to be internally mixed within modes and externally mixed among modes. The physical, chemical, and optical properties of aerosols are simulated in a physically based manner. Aerosol processes include transport, gas- and aqueous-phase (in cloud water only) chemical reactions for sulfur species, microphysics (nucleation, condensational growth, and coagulation), dry deposition, wet scavenging, and water uptake. Efficient secondary formation of aerosol in Beijing, China, has been reported, characterized by frequent nucleation events preceding the pollution episodes followed by rapid condensational growth during the episodes (Qiu et al., 2013; Guo et al., 2014; Zhang et al., 2015). For treatment of these processes in CAM5, a binary H2SO4–H2O homogeneous nucleation scheme (Vehkamaki et al., 2002) is used, and a cluster activation scheme (Shito et al., 2006) is applied in the planetary boundary layer. Condensation of H2SO4 vapor and semi-volatile organics to the aerosol modes is treated dynamically using the mass transfer expressions (Seinfeld and Pandis, 1998) that are integrated over the size distribution of each mode (Binkowski and Shankar, 1995). Coagulation between Aitken and accumulation modes is considered. Water uptake is based on the equilibrium Köhler theory (Ghan and Zaveri, 2007). SOA formation is based on fixed mass yields, i.e., the percentage of semi-volatile organic compounds (VOCs) that could form SOA, with one additional step of complexity by explicitly simulating the emission and condensation/evaporation of the condensable organic vapors (i.e., the lumped semi-volatile organic gas species, SOAG) that are generated from VOCs. The aerosol optical properties are parameterized by Ghan and Zaveri (2007). The refractive indices for most aerosol components are taken from OPAC (Hess et al., 1998), but for BC the value (1.95, 0.79i) from Bond and Bergstrom (2006) is used. More details of the aerosol treatments can be found in Liu et al. (2012).

We conduct two CAM5 simulations with different anthropogenic emission inventories in China for the year 2009. The first simulation uses the emission inventory that follows the protocol of the IPCC AR5 experiments (the AR5 emission inventory hereinafter; see Lamarque et al., 2010). The second simulation is driven by an improved technology-based inventory (MEIC) developed at Tsinghua University (http://www.meicmodel.org/index.html). MEIC has the following advantages: (1) adoption of a detailed technology-based approach, (2) application of a dynamic methodology of rapid technology renewal, (3) re-examination of China's energy statistics, and (4) monthly emissions to represent species that have strong seasonal variations (Zhang et al., 2009). The MEIC emission inventory is verified to produce consistent aerosol precursor loadings with satellite observations (Li et al., 2010; Wang et al., 2010, 2012; Zhang et al., 2012; Liu et al., 2016). It has been widely used to study the trend of aerosol concentrations in China (Wang et al., 2013), the Asian air pollution outflow (Zhang et al., 2008; Chen et al., 2009), the relative contribution of emission and meteorology to the aerosol variability (Xing et al., 2011), and the sensitivity of air quality to precursor emissions (Liu et al., 2010).

The AR5 emission inventory is currently the default for CAM5. The method of mapping the MEIC emission inventory for CAM5 is described in the Supplement. In addition to the differences in the annual mean emissions in 2009 (Fig. S1 in the Supplement), there are large differences in the seasonal variations of two emission inventories (Fig. 1). The AR5 anthropogenic emissions of SO2, BC, and POM do not have seasonal variations. With the inclusion of emissions from biomass burning and shipping for SO2, BC, and POM, as well as volcanic source for SO2, the total BC, POM, and SO2 emissions have seasonal variations in the AR5 emission inventory. However, we should note that the seasonal variation in the AR5 emissions is rather weak since anthropogenic emission dominates in eastern China. This could be problematic since the severe winter haze events in northern China in recent years are often linked to the higher anthropogenic emission in winter. The MEIC emission is characterized by monthly variations for the emissions of SO2, BC, and POM that peak in winter. The emission of SOAG in both inventories shows a consistent seasonal variation that peaks in summer because the emissions of biogenic VOCs (isoprene and monoterpenes), which peak in summer, dominate the total SOAG emission.

We also carry out an additional CAM5 simulation using the decadal MEIC emission from 2002 to 2012 to examine the changes of ADREs due to emission. We choose these 11 years because China's economy recovered from a depression in 2002, and since then the SO2 emission started to grow dramatically and decreased after 2006 due to the application of flue-gas desulfurization devices. After 2012 the annual emission rates did not change as dramatically as in the previous years.

For the first two simulations, we run CAM5 for the year 2009 in a “constrained-meteorology” mode where the model winds are nudged towards ERA-Interim (Dee et al., 2011) on a 6 h relaxation timescale (Ma et al., 2013, 2014; Zhang et al., 2014). Climatological sea surface temperatures (SSTs) are prescribed in the two simulations. When simulating the decadal change from 2002 to 2012, we use the reanalysis data in 2009 cyclically to nudge the model meteorological fields. The constrained-meteorology technique facilitates the model–observation comparison of aerosols and gas species. Temperature and moisture are not nudged in this study. As evaluated in Zhang et al. (2014), nudging temperature and moisture creates a large perturbation to the model state, resulting in unrealistic behavior for cloud and convection parameterizations because these parameterizations are calibrated based on the free-running model climate. Because winds are constrained, the advection of heat and moisture are constrained to some degree when the difference in local temperature and moisture between two simulations is small, but local source and sink terms for atmospheric temperature and moisture are computed according to the model's fast processes (e.g., cloud processes) and land processes (due to prescribed SST). The changes in atmospheric temperature and moisture can in turn influence the gas- and aqueous-phase chemistry and aerosol loadings. The changes in aerosol loading will affect temperature through radiation. However, this local change in temperature is less than 1 K (see Sect. 8 in the Supplement).

We estimate the ADREs due to instantaneous impact of aerosol scattering and absorption on the Earth's energy budget. The ADREs are calculated by the difference between the “clear-sky” radiative flux in the standard model simulation and a diagnostic call to the model radiation code from the same simulation but neglecting the aerosol scattering and absorption.

The horizontal resolution is 0.9∘ × 1.25∘, and vertically there are 30 layers from the surface to 2.25 hPa, with the lowest four layers inside the boundary layer. We focus our analysis of model results over eastern China (22–44∘ N, 100–124∘ E; the red rectangle in Fig. 2), where the strongest anthropogenic emissions are located.

Satellite AOD retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Multi-angle Imaging Spectroradiometer (MISR) in 2009 are used to evaluate the model results. This study uses the monthly mean AOD from MODIS Terra Collection 6 (MOD08_M3 product, https://ladsweb.nascom.nasa.gov/). We use the combined AOD product from the Dark Target (Levy et al., 2010) and the Deep Blue (Hsu et al., 2004) algorithms. The MISR AOD retrievals are downloaded from the Atmospheric Science Data Center at NASA Langley Research Center (https://eosweb.larc.nasa.gov/project/misr/misr_table). We also compare our simulations with ground-based AERONET AOD and single-scattering albedo (SSA) retrievals at 12 sites in mainland China, Hong Kong, Taiwan, Japan, and Korea (shown in Fig. 2; https://aeronet.gsfc.nasa.gov/cgi-bin/webtool_aod_v3). Monthly-average AOD and SSA in 2009 are calculated from the daily averages, with the months that contain less than 3 daily values excluded.

Observation data of chemical compositions near the surface in China are collected from the literature (see Table S3 and the references in the Supplement). The chemical compositions of particulate matter with diameters smaller than 2.5 µm (PM2.5) are analyzed for sulfate, organic carbon (OC), BC, and SOA in these studies. The measured OC concentrations are multiplied by a factor of 1.4 for calculation of the total organic mass (i.e., POM) (Seinfeld and Pandis, 1998). Since the surface chemical composition data that cover a full year in 2009 are very limited, to compared at least one year's cycle of seasonal variation, we extend the range of time selection so that the observations are allowed to continue from 2009 to 2010. Many of the studies collected the samples continuously during April, July, October, and January to represent the concentrations in spring, summer, autumn, and winter. The observation in Xiamen was carried out for a full year of sample collection. Since we do not find the SOA measurements in 2009, we use the data in other years and are aware of the uncertainties due to the time difference. The geographical locations of these observations are shown in Fig. 2.

The ADREs have been estimated based on ground-based and satellite observations at different locations in China (Z. Li et al., 2016). Table S4 summarizes the observations used in this study. Most of the data are from the Chinese Sun Hazemeter Network (CSHNET) (Xin et al., 2007; Li et al., 2010). The ADREs are consistently defined as the difference of the irradiance at the top of the atmosphere (TOA), at the surface, and in the atmosphere with and without the presence of aerosols. The ADREs are either calculated by radiative transfer models using the measured or retrieved aerosol properties (AOD, SSA, phase function, Ångström exponent, and size distribution) and surface reflectance (Xia et al., 2007a; Li et al., 2010; Liu et al., 2011; Zhuang et al., 2014), or else derived from the fitting equation of irradiance measurements as a function of AOD (Xia et al., 2007b, c). Since the MEIC emission inventory is for anthropogenic aerosols, we only compare with observations at locations away from deserts that are less impacted by dust aerosol. For the same reason, the shortwave radiation is discussed since anthropogenic aerosols are mostly fine particles, the impact of which on the longwave radiation can be ignored. All data analyses are performed after cloud screening to ensure clear-sky conditions. Since the solar irradiance depends on solar zenith angle (i.e., the time of the day), we compare with the measurements that are averaged over 24 h. If both the TOA and the surface ADREs are provided, we calculated the ADRE in the atmosphere by subtracting the ADRE at TOA by the ADRE at the surface.

Observed surface concentrations at 10 locations in China show that the primary and secondary aerosols have distinct seasonal variations (Fig. 8). The observed surface concentrations of primary aerosols (BC and POM) at all locations show maximums in winter, suggesting that their seasonal variations are mainly controlled by the emission. The MEIC emission significantly improves the modeled seasonal variations compared with the AR5 emission that has no seasonal variations of POM and BC emissions. In contrast, the observed concentrations of sulfate in northern China (Chengde, Shangdianzi, Beijing, Tianjin, Shijiazhuang, Zhengzhou) are characterized by summer maximums. This is due to a higher photochemical production rate in summer (Wen et al., 2015). The modeled concentrations of sulfate also show their maximum in summer. This feature is commonly seen for many climate models. The concentrations of sulfate in the southern cities (Xiamen and Guangzhou) do not have summer maximums due to the Asian summer monsoon with strong winds and precipitation.

We examine the processes that determine the concentrations of sulfate in the model, including gas-phase and aqueous-phase production, dry and wet scavenging, and the controlling meteorological variables (Fig. 9). The MEIC emission of SO2 peaks in winter in northern China due to heating in the domestic section, whereas the AR5 emission does not have seasonal variations (Fig. S7). Obviously, the surface concentrations of sulfate aerosol cannot be explained by emission alone, and the atmospheric processes are more likely responsible for the seasonality. We find that the simulated seasonal variations of surface concentrations of sulfate aerosol are controlled by the gas-phase and aqueous-phase production processes and to a lesser extent by the emission of SO2. The gas-phase chemistry is most active in summer due to the temperature dependence of the oxidation rate of SO2 by OH. Also the oxidation rate depends on the concentration of the OH radical, which is highest due to efficient photochemical reactions in summer. The aqueous-phase formation of sulfate aerosol also peaks in summer due to higher relative humidity and thus more cloud water. Although the MEIC SO2 emissions peak in winter, both the gas-phase and aqueous-phase oxidations are less efficient in winter, which results in lower concentrations of sulfate aerosol than in summer.

We notice that some other observations show different seasonality of sulfate aerosol from the model results. For example, observations from the China Meteorological Administration Atmosphere Watch Network (CAWNET; Zhang et al., 2012) show that concentrations of sulfate aerosol in the northern Chinese cities (e.g., Gucheng and Zhengzhou in Fig. S8) peak in winter as opposed to summer in spite of a minor maximum in summer. The observed seasonal variations at two pairs of nearby sites from CAWNET and our study (Gucheng 2006–2007 versus Beijing 2009–2010, Zhengzhou 2006–2007 versus 2009–2010) are different from each other. This may reflect that the relative contributions of the emissions and the atmospheric processes in determining the concentration of sulfate change with years and locations. It is also possible that some mechanisms of sulfate aerosol formation for these CAWNET sites, which are especially important in winter, are not properly modeled or are missing in the model. For example, the aqueous-phase oxidation of SO2 in preexisting aerosols is not modeled, which is important in explaining the winter haze in China (Wang et al., 2016; Cheng et al., 2016).

Having the same constrained meteorology for the two simulations with different emission inventories provides us with an opportunity to examine the impact of emission versus atmospheric processes on the seasonality of aerosols. The longitudinal-average BC burden in the simulation with the MEIC emission shows a strong seasonal variation between 25 and 40∘ N with a higher burden in winter (Fig. 10a), which corresponds with the seasonal variation of BC emission (Fig. 10b). Since there is no seasonal variation for BC aerosol in the AR5 emission (Fig. 10d), the seasonal variation of BC concentrations can only be due to the impact of atmospheric processes in the AR5 emission run (Fig. 10c). The winter peak is also seen for the AR5 run most likely due to stagnant wind fields for dispersion in winter. The summer minimums are due to wet scavenging by the monsoon precipitation. Figure 10 indicates that seasonal variations of both the emission and atmospheric processes play important roles in determining the seasonal variation of BC concentrations.

The distinct impacts of emissions and atmospheric processes that are associated with meteorological factors on the seasonal variations of primary (e.g., BC) and secondary aerosols (e.g., sulfate) are further demonstrated in Fig. 11. The seasonal variation of differences in the longitudinal-average burden of BC between the two emission runs closely resembles the pattern of differences in the emission of BC (Fig. 11a, b). However, the dependence of seasonal variation of the burden of sulfate on the emission of SO2 is less evident (Fig. 11c, d). The difference of SO2 emission between the two inventories at 30–45∘ N is amplified by the production and condensation of H2SO4 gas, which are favored at higher temperatures in summer. This larger difference of the sulfate burden between the two emission runs is obviously aligned with higher temperatures between 30 and 45∘ N in summer (May to July) (Fig. 11e). In contrast, although there is a comparable difference in the SO2 emission between 35 and 40∘ N in cold seasons (November to March), the difference of the sulfate burden is not as evident due to the fact that low temperatures inhibit the production of sulfate. Wet scavenging by clouds and precipitation helps to reduce the concentrations and their absolute differences in southern China during spring and summer (Fig. 11f, g). Higher wind speeds north of 35∘ N in winter (Fig. 11h) for aerosol dispersion help to explain the small difference of the sulfate burden in spite of the evident difference of SO2 emission there. The stagnant wind field that propagates from 22 to 35∘ N in spring and from 35 to 22∘ N in autumn makes the impact of the difference in emissions on the large differences of BC and sulfate burdens between the two simulations prominent in the corresponding seasons and regions. Due to the complex atmospheric processes, the spatiotemporal patterns of secondary aerosol burdens less closely follow their precursor gas emissions than primary aerosols.

In this study, changes in the aerosol radiative forcing will alter atmospheric temperature and moisture in the model and can, in turn, influence gas- and aqueous-phase chemistry and aerosols. However, differences in temperature (< 1 K) and moisture (< 3 %) are small enough compared to seasonal variations and therefore do not affect our finding on the impacts of emissions and atmospheric processes on the aerosol burden. More discussion on the effect on aerosol–meteorological interactions is provided in Sect. 8 of the Supplement.

Figure 12 shows the spatial distribution of annual-average shortwave ADREs in China simulated using the MEIC and the AR5 emissions due to all aerosol species at TOA, at the surface, and in the atmosphere. The TOA radiative cooling effect is evident in eastern China due to anthropogenic aerosols. In some parts of the southwestern China the ADRE at TOA is positive due to strong BC absorption in the atmosphere. The most pronounced surface cooling and atmospheric warming are located in northern China and the Sichuan Basin, which is consistent with the spatial patterns of the emissions. At these locations the surface and atmospheric differences of the ADREs between the two simulations are also significant.

As shown in Table 2 the annual-average cooling effect at TOA is reduced (more negatively) by -0.91 W m-2 (22.3 %) by all aerosols using the MEIC emission (-5.02 W m-2) compared with that using the AR5 emission (-4.11 W m-2). At the surface there is a strong cooling effect of -18.47 W m-2 using the MEIC emission, which is reduced (more negative) by -3.48 W m-2 (23.3 %) compared with that using the AR5 emission (-14.99 W m-2). The atmospheric warming effect of all aerosols using the MEIC emission is estimated to be 13.45 W m-2, which is 2.57 W m-2 (23.6 %) stronger than the estimation made by the AR5 emission (10.88 W m-2) over eastern China.

Table 2 also shows the annual-average ADREs over eastern China by individual aerosol species. The ADREs of SOA are not shown due to its large emission uncertainty. Due to larger AODs simulated with the MEIC emission, the ADREs of each aerosol species are larger than the ADREs using the AR5 emission by 33.6 to 47.2 % at TOA. Tables 1 and 2 show that over eastern China a 12.0–46.9 % difference of the anthropogenic emission rates of various aerosol species results in a 30.4 % difference of the total AOD of all species (including anthropogenic and natural aerosols) and 22.3, 23.3, and 23.6 % differences of the ADREs at TOA, the surface, and in the atmosphere, respectively. The impacts of the emission on AOD and ADREs are significant.

The normalized radiative effect (NRE) represents the radiative effect efficiency per unit aerosol optical depth (Schulz et al., 2006). The light-scattering aerosols (sulfate and POM) have very similar negative NREs (-31.77 and -33.84 W m-2τaer-1 with the MEIC emission, respectively). The light absorbing BC aerosol shows a much higher positive NRE (100.52 W m-2τaer-1), which is comparable to the mean NREs of the AeroCom models (153 W m-2τaer-1) considering the wide range of the estimates among the models (28 to 270 W m-2τaer-1) (Schulz et al., 2006). The NREs of BC are much higher than the other aerosol species, especially the warming in the atmosphere (305.50 W m-2τaer-1). This indicates that the ADREs are much more sensitive to the BC aerosol burden than the other aerosol species and highlights the importance of the BC emission and concentration to correctly represent the ADREs in the model. BC also makes the largest contribution to the ADRE in the atmosphere and at the surface. We note that the ADREs of light-scattering aerosols (sulfate and POM) in the atmosphere are also warming effects. The explanation is that coating of these scattering aerosols on BC increases the absorption capability of the internally mixed aerosol particles (i.e., particles in the same aerosol mode with BC) (Chung et al., 2012).

The modeled spatial distributions of ADREs of BC in summer and winter at the surface and in the atmosphere are shown in Figs. 13 and 14, respectively. With the AR5 emission, the average ADREs over eastern China in winter (-4.40 W m-2 at the surface and 6.11 W m-2 in the atmosphere) are close to the ADREs in summer (-4.40 W m-2 at the surface and 6.28 W m-2 in the atmosphere). Due to the higher MEIC BC emission in winter, the cooling effect of BC at the surface is much more significant when using the MEIC emission (-7.35 W m-2) than the AR5 emission averaged over eastern China (Fig. 13). Likewise the warming effect of BC in the atmosphere with the MEIC emission (10.50 W m-2) is nearly twice as much as that using the AR5 emission (Fig. 14). Driven by the same constrained meteorology, the MEIC emission results in much stronger seasonal variation of ADREs of BC than the AR5 emission.

Figure 15 shows the comparison between the measured and modeled ADREs at TOA, at the surface, and in the atmosphere over China. Observations from 25 nationwide stations shows that clear-sky ADREs are characterized by a strong radiative heating in the atmosphere, which implies a substantial warming in the atmosphere and cooling at the surface (Li et al., 2007, 2010). Model simulations show small ADREs (∼ -10 to -2 W m-2) at TOA with both the MEIC and AR5 emission inventories, while the measurements give a larger range of ADREs at TOA (∼ -14 to 2 W m-2). At the surface and in the atmosphere, the modeled ADREs using the MEIC emission inventory at most locations are within a factor of 2 of the observations. The MEIC emission inventory produces better agreement with the observations than the AR5 emission inventory.