2.1.1 CAS-ESM

CAS-ESM is the Earth system model developed by the Chinese Academy of Sciences. It is a coupled model incorporating the Atmospheric General Circulation Model of the IAP (IAP-AGCM) (Su et al., 2014), the Climate System Ocean Model (LICOM) (Liu et al., 2012), the Common Land Model (CoLM) (Dai et al., 2015), the sea ice model (CICE), the Dynamic Global Vegetation Model of the IAP (IAP-DGVM) (Zhu et al., 2018), the IAP-AACM, and the land and ocean biogeochemical models of the IAP (IAP-OBGCM) (Li and Xu, 2012). The IAP-AACM provides the mass concentration of trace gases and aerosols for the CAS-ESM and responds to the feedback of aerosols on meteorological fields. Currently, global climate and ecological environmental change is not only one of the core issues of international climate and environment diplomacy, but also an important factor governing the sustainable development of China. The Earth system model is a basic tool for understanding and solving these problems. The resolution of the CAS-ESM is currently $\mathrm{1}{}^{\circ }\times \mathrm{1}{}^{\circ }$ and will later be updated to $\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$. The CAS-ESM will be used for the climate numerical experiment with high resolution ($\mathrm{0.25}{}^{\circ }\times \mathrm{0.25}{}^{\circ }$) for 100 years (1950–2050) and will provide simulation results for phase 6 of the Coupled Model Intercomparison Project (CMIP6) and IPCC AR6.

2.1.2 IAP-AACM

The IAP-AACM was developed on the basis of the Nested Grid Air Quality Prediction Model System (NAQPMS; Wang et al., 2006b) and the Global Nested Grid Air Quality Prediction Model System (GNAQPMS; H. S. Chen et al., 2015). NAQPMS and GNAQPMS are widely used in the simulation of dust (Li et al., 2012), ozone (O₃) (Wang et al., 2006a; Li et al., 2007), deposition (Ge et al., 2014), air pollution policy control (Wu et al., 2011; Li et al., 2016; Wei et al., 2017), and the global transportation of mercury (H. S. Chen et al., 2015).

Like GNAQPMS, the IAP-AACM is a multi-scale nested model that describes atmospheric chemistry and aerosol process at both global and regional scales. In the IAP-AACM, sea salt and dust emissions are calculated online. The dust scheme originates from the wind erosion model developed by Wang et al. (2000) and improved by Luo and Wang (2006). The simulation of sea salt is based on the scheme from Athanasopoulou et al. (2008). Dry deposition processes are based on the resistance model approach of Zhang et al. (2003). The gas-phase chemistry scheme is Carbon Bond Mechanism Z (CBM-Z) (Zaveri and Peters, 1999). The cloud convection, aqueous chemistry, and in-cloud and below-cloud scavenging use the second generation of the Regional Acid Deposition Model (RADM2) (Stockwell et al., 1997). For aerosols, the thermodynamic equilibrium module ISORROPIA (Nenes et al., 1998, 1999) is used to calculate gas–particle partitioning of inorganic aerosols and aerosol water content. Furthermore, an aerosol microphysics dynamic module (APM) (Yu and Luo, 2009) was added to expand the simulation from mass concentration to size distribution (Chen et al., 2014, 2017). The secondary organic aerosol (SOA) module is based on the mechanism developed by Strader et al. (1999), considering two anthropogenic emission precursors (toluene and other aromatic hydrocarbons) and two bio-emission precursors (isoprene and monoterpene) (Li et al., 2011).

In addition, the IAP-AACM includes an updated DMS emission module from Lana et al. (2011). The DMS concentration in seawater is calculated using 47 313 observations from the “Global Surface Seawater Dimethylsulfide (DMS) Database” (http://saga.pmel.noaa.gov/dms/, last access: 14 June 2019) and an additional 63 observations in the South Pacific (Lee et al., 2010). The IAP-AACM also provides a simplified gas-phase chemistry mechanism specially designed for CAS-ESM to provide the major aerosol components (sulfate, OM, BC, dust, and sea salt). This mechanism retains aerosols with significant climatic radiative effects while cutting computational load; nitrate and its chemical reactions are excluded. This approach is common in global aerosol models such as the Integrated Massively Parallel Atmospheric Chemical Transport (IMPACT) model (Liu et al., 2005) and GLOMAP (Mann et al., 2010). The simplified scheme contains SO₂, DMS, sulfur acid gas (H₂SO₄), ammonia (NH₃), and hydrogen peroxide. Off-line monthly fields of the oxidants, such as the hydroxyl radical (OH), the nitrate ion radical (NO₃), and O₃, and the super oxidation of hydrogen (HO₂), generated from a simulation of the standard version of IAP-AACM, are read in and interpolated. Chemical processes in the simplified version are the same as those in the standard version except for the gas-phase scheme mentioned above. In this paper we focus on evaluating simulation results of the standard version model driven by a global version of Weather Research and Forecasting (WRF).

3.2.1 Hydroxyl radical (OH)

Oxidation is the basic characteristic of atmospheric chemistry. As the most important oxidant in atmosphere, OH is the crucial species in CTMs. OH in troposphere is mainly produced by the following reaction: ${\mathrm{O}}_{\mathrm{3}}+h\mathit{\nu }(\mathit{\lambda }\le \mathrm{320}\mathrm{nm})+{\mathrm{H}}_{\mathrm{2}}\mathrm{O}\to \mathrm{2}\mathrm{OH}+{\mathrm{O}}_{\mathrm{2}}$. The tropospheric mean concentration of OH in IAP-AACM is 13.0×10⁵ molec cm⁻³. This is a little higher than the mean OH concentration ($\mathrm{11.1}\pm \mathrm{1.6}\times {\mathrm{10}}^{\mathrm{5}}$ molec cm⁻³) given by the 16 ACCMIP models in Naik et al. (2013). The high concentration indicates a stronger atmospheric oxidation, which could explain the lower concentration of CO over ocean. The zonal mean OH concentrations for January, April, July, and October are shown in Fig. 3. As in other chemistry models, the OH concentration in the tropics is the highest throughout the year and decreases gradually from the tropics to the poles. This is due to the positive influence of solar radiation and the water vapour concentration. The seasonal north–south shift of the OH maximum is also ascribed to the seasonal variation of these two factors. The mean inter-hemispheric (N ∕ S) ratio of OH in the model is 1.26, which is in accordance with the multi-model mean ratio of 1.28±0.1 (Naik et al., 2013). Vertically, the highest concentration is in the 2–4 km layer over the tropics. In the Northern Hemisphere, the highest OH concentration appears in summer. The peak value is located at around 30^∘ N, in the atmosphere above 2 km. Generally, the distribution of the OH concentration is similar to that reported by other models (Huijnen et al., 2010; Badia et al., 2017).

Figure 3Zonal monthly mean concentration of OH in the troposphere for January, April, July, and October from the IAP-AACM (in 10⁵ molec cm⁻³).

Download

3.2.2 Trace gases

Global annual-averaged surface-layer trace gas distributions from the IAP-AACM are evaluated against site observations in Fig. 4a. Scatterplots of observations and simulations divided into 11 regions are displayed in Fig. 4b. Figure 5 shows the comparison of the annual surface concentrations of CO, O₃, and NO₂ between IAP-AACM and HTAP CTMs, including CAM-Chem (Lamarque et al., 2012), Oslo CTM3 (Søvde et al., 2012), and CHASER (Sudo et al., 2002).

Figure 4Annual mean concentration (ppb) of the surface layer in the IAP-AACM compared with observations. (a) The circles represent site observations. The first row is CO and O₃, and the bottom row is NO₂ and SO₂. (b) Scatterplots for Africa, Antarctica, the Arctic Ocean (ArcticO), Asia, the Atlantic Ocean (AtlanticO), Europe, the Indian Ocean (IndianO), North America (NAmerica), South America (SAmerica), Oceania, and the Pacific Ocean (PacificO). The abscissa shows the observations, and the ordinate shows the simulation. The colour of the points represents different regions.

Download

Figure 5Annual mean surface distributions (ppb) from IAP-AACM compared with HTAP models. Rows from top to bottom represent IAP-AACM, CAM-Chem, Oslo CTM3 and CHASER respectively. The left column displays CO, the middle column displays O₃ and the right column is NO₂.

Download

Overall, the global surface CO simulation of the IAP-AACM is lower than the observations, especially in natural source regions. The difference over ocean reaches ∼100 ppb, with NMB values ranging from −0.59 to −0.45 (shown in Table S3). In anthropogenic source regions, the model is close to site records in North America (NAmerica), EA, and Europe with NMB values of −0.23, −0.34, and −0.39 respectively. The scatterplot clearly shows a negative bias between the model and observations. The lower model results should be related to underestimated emissions and overestimated OH. As discussed in Sect. 3.2.1, the OH concentration in the troposphere is slightly higher than the other models. As shown in Fig. 3, in the IAP-AACM, the peak concentration of OH (30–35 molec cm⁻³) is higher than the other models (under 30 molec cm⁻³). Due to the sink reaction of CO ($\mathrm{CO}+\mathrm{OH}\to {\mathrm{CO}}_{\mathrm{2}}+\mathrm{H}$), the CO loss is larger in the IAP-AACM. Moreover, the anthropogenic emission of CO is 546.4 Tg yr⁻¹ in the IAP-AACM (shown in Table S2). This is lower than other emission inventories (e.g. MOZART-4 with 642 Tg yr⁻¹, Emmons et al., 2010; ACCMIP with 610.5 Tg yr⁻¹, Badia et al., 2017). Janssens-Maenhout et al. (2015) pointed out that CO emission from HTAP_v2 has an uncertainty of 15 %–100 % and 35 %–150 % in data from well-maintained countries and poorly maintained countries respectively. Furthermore, Shindell et al. (2006) evaluated 26 global models and showed that all of the model results were lower than observations in the Northern Hemisphere (NH) except in the tropics. This is related to a lower CO emission source. The spatial distribution of CO concentrations in IAP-AACM is similar to that in other models from HTAP in 2010. High values are found in industrial areas such as North America (NAmerica), Europe, and EA, and biomass burning areas (BBAs) such as South Africa (SAfrica) and South America (SAmerica). Overall, IAP-AACM shows lower concentration in BBAs. This difference probably relates to the different fire emissions used in the IAP-AACM and HTAP models (GFED4 vs. GFED3) (Janssens-Maenhout et al., 2015). Due to the impact of a reduction in combustion area and decreasing fuel consumption, there is a reduction of about 20 %–30 % in BBAs for CO emissions in GFED4 compared with GFED3 (van der Werf et al., 2017).

The surface distribution of O₃ simulated by the IAP-AACM is in a good agreement with observations. The O₃ simulations at most sites are within a factor of 2 of observations, and all the regions have NMB values within ±0.2 (Table S3) except for Antarctica and Asia. There are three sites in Southeast Asia that exhibit concentrations more than twice the observed values. As these sites are located at the coast, simulations in the coarse grids cannot capture the steep variation. Additionally, South Asia is a high-emission area for biogenic volatile organic compounds (VOCs). Uncertainties in the biogenic emission inventory also cause large errors in the O₃ simulation due to photochemical processes. As shown in Fig. 5, the IAP-AACM exhibits a generally similar surface distribution of O₃ to the other models. However, the model shows lower concentrations in high-latitude areas (especially over ocean) than the other models. That could be partly related to the dry deposition over sea ice. The dry deposition velocity to ice is under 0.02 cm s⁻¹ across the 15 HTAP models (Hardacre et al., 2015). In the IAP-AACM, it is higher (0.035–0.048 cm s⁻¹) than in the abovementioned models, as shown in Fig. S2. Additionally, the dry deposition velocity over ocean is 0.042–0.05 cm s⁻¹ in IAP-AACM compared with the HTAP models (around 0.05 cm s⁻¹); according to Ganzeveld et al. (2009) there should be a difference of less than 12 %.

Figure 6Mean seasonal variation of O₃ (in ppb) over the NAmerica, Europe, Asia, AtlanticO, PacificO, Antarctica, SAmerica, Africa, and Oceania sites. Black and red lines represent the average of observations and simulations respectively. Grey shaded areas and red vertical bars show ±1 standard deviation for observations and for model results respectively.

Download

As shown in Fig. 6, the model shows good skill in capturing the seasonal variation of surface O₃ in the Southern Hemisphere (SH) and NH oceans. Over ocean, the O₃ concentration is higher in spring and lower in summer. In contrast, it is higher in autumn or summer over land in the SH. The model shows poor performance regarding the seasonal cycle of surface ozone in the NH, with an underestimation in the NH over land in cold seasons and an overestimation in Europe and EA in summer. In the IAP-AACM, the stratospheric–tropospheric exchange is not considered. This leads to a large negative bias in the simulation. The stratosphere-to-troposphere ozone transport event occurs widely across mid-latitudes in the NH (Monks, 2000; Akritidis et al., 2018). Research (Munzert et al., 1985; Austin and Follows, 1991) has shown that the maximum stratosphere to troposphere flux occurs in late winter/spring. The surface O₃ concentrations over EA (sites mainly located in Japan) are overestimated in summer and early autumn. The same pattern is also found in the multi-model inter-comparison of 21 HTAP models (Fiore et al., 2009). The simulations in island countries located in EA are sensitive to the timing and extent of the Asian summer monsoon (Han et al., 2008). The positive model bias in this season may stem from an inadequate representation of the southwesterly inflow of clean marine air. Furthermore, the underestimation of cloud cover in summer may also be responsible for the overestimation of O₃ due to stronger photochemistry. Additionally, it is difficult for a global model with a coarse resolution to resolve local orographically driven flows or sharp gradients in mixing depths under complex underlying surface conditions over land. As the seasonal variation of surface O₃ should be different in different environments, a seasonal cycle comparison between these NH sites which are separated into land, mountain, and marine is also shown in Fig. S3. The underestimation in cold seasons and overestimation in summer are found in different environments to varying degrees. For inland areas, the model tends to overestimate O₃ concentrations in summertime. Uncertainties in VOC–NO_x–O₃ chemistry may also contribute to this overestimation. The natural source of isoprene from vegetation is important regarding O₃ formation due to its high proportion of VOC emissions in summer (estimated to be 40.9 Tg yr⁻¹ in China by Fu et al., 2012). In IAP-AACM, the O₃ concentration is about 5–15 ppb lower than site observations in Antarctica. This may be caused by the lack of halogen chemistry in the model. Remarkable ozone depletion events driven by halogen chemistry (mostly notably as bromine) are observed in the polar boundary layer (Simpson et al., 2007). A model study by Falk and Sinnhuber (2018) indicated that there are missing sources of bromine release from ice and snow in EMAC v2.52. The over-prediction of dry deposition velocity to sea ice also plays a role here, as previously mentioned.

NO₂ is mainly distributed in the anthropogenic source areas, which is well captured by IAP-AACM (see Fig. 4a). In the NH, the maxima are located in urban areas due to fossil fuel combustion from traffic and industry. The surface concentration of NO₂ is much higher in eastern China (>20 ppb) than in eastern NAmerica and Europe (3–10 ppb). In the SH, the maxima are located in SAmerica and SAfrica due to biomass burning, where NO₂ ranges from 1 to 10 ppb. The simulations are in good agreement with observations in NAmerica, Europe, and most parts of EA. As shown in Fig. 4b, simulations are within a factor of 2 of observations at most sites, with NMB values of −0.14, 0.16, and −0.14 for Asia, Europe, and NAmerica respectively. As we use the same anthropogenic emission inventory, the distribution of NO₂ in the NH between IAP-AACM and other models is similar (shown in Fig. 6). The concentrations in SAmerica and SAfrica are slightly lower (∼3 ppb) than in the other models, due to the aforementioned different version of GFED data used in the IAP-AACM and HTAP models. Compared with the other models, the surface NO₂ over ocean is higher in IAP-AACM. In IAP-AACM, the emissions of energy and industry are emitted in the first five layers considering the stack height. The higher injection height of emissions leads to a larger transportation distance and lower NO_x at the surface in source areas (Badia et al., 2017). Consequently, it also leads to a higher concentration of surface ozone in NH source areas due to weak NO_x titration.

Similar to NO₂, SO₂ shows high values in the NH over land, mainly due to human activities associated with fossil fuel combustion. Maximum concentrations are mainly found in NAmerica, Europe, India, and EA, ranging from 5 to 20 ppb; concentrations can even reach over 20 ppb in eastern China. SO₂ over ocean is mainly due to DMS oxidation from marine organisms, ranging from 0.1 to 1 ppb. The model shows a distribution of 0.1–20 ppb in EA and 0.1–5 ppb in western NAmerica, which is consistent with observations. The simulated values of SO₂ in eastern NAmerica and Europe are about 1–10 ppb and are overestimated, with NMB values of 3.51 and 3.79 respectively (as shown in Table S3).

3.2.3 Aerosol composition

Figure 7 shows the annual surface concentrations of aerosols in IAP-AACM in comparison with site observations. The formation of sulfate, nitrate, and ammonium (SNA) are due to gas-to-particle conversion in the atmosphere. The distribution of SNA is consistent with the precursor gases (SO₂, NO_x, and NH₃), mainly in the NH over land. As shown in Fig. 7a, the surface distribution of SNA in IAP-AACM is generally close to the site records. As shown in Fig. 7b, sulfate is more or less overestimated. Specifically in Asia, the simulations at most sites are within a factor of 2 of observations, with a NMB of 0.36. However, in NAmerica and Europe, it is overestimated with NMB values of 1.94 and 1.1 respectively. The site-averaged simulation value is 1.67 and 1.56 µg m⁻³ higher than observations in NAmerica and Europe respectively. This overestimation is consistent with the high level of SO₂ described previously. There are more uncertainties in the simulation of nitrate, as the volatility of HNO₃ makes nitrate formation more sensitive to environmental factors such as temperature and relative humidity in the atmosphere. The complex photochemical reactions of NO_x also contribute to the uncertainties. The concentration of nitrate is higher in eastern America and lower in western America. IAP-AACM reproduces the distribution of nitrate in western America well but overestimates it in eastern America. The model does not fully capture the spatial variation over Europe, with an overestimation at most of the sites. For EA, there is an underestimation of ∼5 µg m⁻³ in Southeast Asia and Japan. Overall, the NMB values are within ±0.8 (0.5, 0.74, and −0.61 for NAmerica, Europe, and EA respectively). The performance of ammonium varies in different regions, as there are more uncertainties in the emission of NH₃ (the precursor of ammonium) from croplands (Xu et al., 2019). There is slight negative bias in America and a positive bias in Asia, with NMB values less than ±1 (−0.46 and 0.85 respectively). In Europe, there is a significant positive bias with a NMB of 1.49.

Figure 7(a) The same as Fig. 4, except for sulfate, nitrate, ammonium, BC, OC, and PM_2.5 (in µg m⁻³). The top row is sulfate, nitrate, and ammonium, and the bottom row is BC, OC, and PM_2.5. (b) Scatterplot of annual mean concentration. The order of the plots in (b) are in accordance with panel (a). Solid circles in blue, yellow, and magenta represent Asia, Europe, and NAmerica respectively.

Download

Due to the large contribution of biomass burning and fossil fuel combustion associated with old technology, carbonaceous aerosols are higher in developing countries than in developed regions such as NAmerica and Europe. The concentration of BC and OC ranges from 2 to 10 and 5 to 10 µg m⁻³ in China and India respectively, whereas it ranges from ∼1 and 3 to 10 µg m⁻³ respectively in SAfrica and SAmerica. In general, the model results are consistent with observations in the three regions shown in Fig. 7, with NMB values within ±0.65 and ±0.7 for BC and OC respectively. The accuracy of the simulation mainly depends on emissions and deposition processes, as BC is quite inert to chemical reactions.

The simulation of BC in China is accurate with 70 % of the stations within a factor of 2 of observation, whereas OC is underestimated by about 5–10 µg m⁻³. The meteorological conditions and emission inventories in the model are inconsistent with the observation year (2006) of carbonaceous aerosols in China. This may be partially responsible for the bias of OC. According to recent studies, there was a slight increase (less than 0.1 Tg) in both BC and OC emissions from 2006 to 2010 in China (Lu et al., 2011; Fu et al., 2012). As shown in Fig. 7, the simulated BC concentrations at most sites are close to observations, whereas the simulated OC is significantly underestimated. The study by Fu et al. (2012) showed a significant underestimation of OC emissions over China. Furthermore, Zhao et al. (2016) found that the pathway of intermediate volatile organic compounds (IVOC) to SOA is very important regarding the formation of SOA. Their model experiments suggested that IVOCs constitute over 40 % of the OM concentrations in eastern China. Yang et al. (2018) also showed a significant increase in the SOA concentration in an observation-based box model which included IVOC reactions. IVOC reactions are not included in our SOA module. The SOA module in IAP-AACM is a two-product scheme. Model studies using a two-product scheme estimated an underestimation of OM by 40 %–78 % in China (Lin et al., 2016; Han et al., 2016). Thus, the close simulation of BC but large underestimation of OC calls for an improvement in the SOA formation mechanism in IAP-AACM.

Generally, the model shows good skill in simulating PM_2.5. Model results at most sites are close to observations, as shown in Fig. 7b, especially in Europe and Asia with NMB values of −0.35 and −0.36 respectively. The underestimation in western China could be related to uncertainties in emissions and unrepresentative simulation with coarse grids.

3.2.4 Comparison with satellite data

The VTC of O₃ is compared against satellite observations derived from OMI (shown in Fig. 8). In general, the pattern of the seasonal cycle was covered by the model. On the Northern Hemisphere mainland, higher O₃ VTC appears during June–July–August (JJA), whereas in the SH it appears during September–October–November (SON) with a range of 40–60 DU. The model maintains a high value (40–50 DU) in the tropics during December–January–February (DJF), which is possibly due to the high concentration of CO emitted from biomass burning. The underestimation of cloud cover in the intertropical convergence zone may also contribute. The O₃ VTC is significantly underestimated over ocean in middle–high latitudes. As the stratospheric chemistry is not considered in IAP-AACM, the lack of stratospheric–tropospheric exchanges could partly be responsible for the underestimation of the column burden.

Figure 8Seasonal mean column concentration of O₃ from IAP-AACM (left column) and OMI (right column). Seasons are defined as March–April–May (MAM), June–July–August (JJA), September–October–November (SON), and December–January–February (DJF).

Download

The VTC of NO₂ is also compared against satellite observations derived from GOME-2A (shown in Fig. 9). The NO₂ VTC has a range of 20–150×10¹⁴ molec cm⁻² in most source areas. In general, IAP-AACM reproduced the magnitude in different regions. In addition, the model captured seasonal variations in the NO₂ concentration in the vertical troposphere well. In anthropogenic source areas in the NH (e.g. NAmerica, Europe, and EA), the NO₂ VTC is higher in SON and DJF and lower in JJA. The column concentration is higher during JJA in SAmerica and SAfrica, whereas it is higher during DJF in central Africa due to vegetation burning in dry season. Compared with GOME-2A, IAP-AACM showed a larger column concentration over ocean. The overestimation is also reflected in the comparison of the surface concentration. This is probably caused by insufficient oxidation to nitrate and a higher injection height of emissions which leads to a larger transportation distance, as suggested in Badia et al. (2017). Generally, the distribution of tropospheric NO₂ by the model is consistent with satellite observations.

Figure 9Seasonal mean column concentration of NO₂ from IAP-AACM, left column, and GOME-2A, right column (in 10¹⁴ molec cm⁻²).

Download

In order to evaluate the column burden of aerosol in IAP-AACM, we compared the AOD from IAP-AACM with MODIS satellite data as shown in Fig. 10. The calculation of the light-extinction coefficient, b_ext550 (1 Mm⁻¹ at 550 nm), follows Eq. (1) given by Li et al. (2011):

where f_SNA (RH) and f_SS (RH) represent the hygroscopic growth factor for SNA and sea salt respectively, and the variables in brackets are the mass concentration of aerosol species (OC – organic carbon, BC – black carbon, FD – fine dust, CD – coarse dust, and SS – sea salt).

In Fig. 10, the model reproduces the spatial features of AOD exhibited by satellites globally. For example, the high value around 60^∘ S, ranging from 0.1 to 0.3, is due to high concentrations of sea salt. The maximum in SAmerica and SAfrica is due to the large amount of carbonaceous aerosol produced by biomass burning. The desert maximum over 0.5 is caused by mineral dust in NAfrica, Arabian Peninsula, and western China. High AOD in NAmerica, Europe, India, and EA is caused by anthropogenic aerosols. Furthermore, there is a good agreement of the seasonal variations with satellite observations. For example, the AOD in the desert areas of the NH reaches a maximum in March–April–May (MAM), as there are frequent dust storms during this period. SAmerica and SAfrica exhibit the highest AOD during JJA as more biomass burning takes place in this season. However, there are several biases between the model and observations. The model shows a weaker AOD in Southeast Asia than the observations, where the value is mainly controlled by biomass burning. The AOD from IAP-AACM is lower than observations by about ∼0.4 in eastern China, which is mainly due to the negative bias in anthropogenic aerosols simulation. Furthermore, the underestimation of RH₂ in eastern China (shown in Fig. 2) is also responsible for the lower simulation of AOD. The comparison of monthly gridded average AOD (shown in Fig. S4) shows a discrepancy in EA, due to the bias of dust simulation in spring.

Figure 10Seasonal mean AOD from IAP-AACM (left column) and MODIS (right column).

Download

3.3.1 Distribution of pollutants in EA

Figure 11a shows the annual distribution of the four pollutants SO₂, NO₂, PM₁₀, and PM_2.5, against 45 city stations from the nested simulation. In general, the simulation shows better agreement with sites in eastern China than in western China. The distribution of PM_2.5 and its precursors shows high levels in eastern China and low levels in western China, which is related to large emissions in the east. The simulation is in good agreement with observations at most sites. Concentrations of precursor gases and particles in Tibet are greatly underestimated, as the model's coarse resolution cannot represent the emission for the region. As shown in Fig. 11b, model results for NO₂, SO₂, and PM_2.5 are mostly within a factor of 2, with NMB values within ±0.3. The “NO₂” values reported by routine monitoring sites are ${\mathrm{NO}}_{\mathrm{2}}^{\ast }$, which partially includes HNO₃ and ${\mathrm{NO}}_{\mathrm{3}}^{-}$. This implies that the model may overestimate “NO₂”. PM₁₀ concentrations are underestimated at all sites with a NMB of −0.51. The model results for PM₁₀ and PM_2.5 include primary PM_2.5, BC, POA, SOA, and SNA. As natural dust contributes a lot to particles in Northwest China, it is underestimated in this area.

Figure 11(a) Surface annual mean concentration of the nested domain (in µg m⁻³). The circles represent sites observations. The top row is NO₂ and SO₂, and the bottom row is PM₁₀ and PM_2.5. (b) Scatterplots of annual mean concentrations in the nested domain (in µg m⁻³). The order of the panels (b) are in accordance with the order in panel (a). The abscissa shows the observations and the ordinate shows the simulation.

Download

Figure 12Annual surface distributions from the nested IAP-AACM compared with regional models from MICS-Asia. Each row from top to bottom represents IAP-AACM, WRF-Chem, CMAQ, and NAQPMS respectively. The left column is SO₂ (ppb), the middle column is NO₂ (ppb), and the right column is PM_2.5 (µg m⁻³).

Download

Model inter-comparison can give some insights into model uncertainties. Here a comparison between IAP-AACM and several regional models of MICS-Asia is presented (Fig. 12). The MICS-Asia models shown here are WRF-Chem(v3.9; Tuccella et al., 2012), CMAQ(v4.7.1; Mebust et al., 2003), and NAQPMS (Wang et al., 2006b). The simulations are for 2010 with the same meteorological fields, emissions, and horizontal resolution (45 km). Overall, the IAP-AACM shows a similar annual distribution to the MICS-Asia models in EA, as the emission inventories used in IAP-AACM are largely the same as for the MICS-Asia models. The simulation of SO₂ in IAP-AACM is similar to WRF-Chem and CMAQ. The simulation of NO₂ in IAP-AACM is lower in source areas (e.g. eastern China and Japan) but higher over downwind areas compared with CMAQ. This is possibly related to the higher injection height in IAP-AACM. Although the models use the same dynamic framework, SO₂ and NO₂ in IAP-AACM are lower than in NAQPMS. This could be related to the different dry deposition schemes used by as Zhang et al. (2003) and Wesely (1989) in IAP-AACM and NAQPMS respectively. Furthermore, the PM_2.5 from NQAPMS is higher than in IAP-AACM in northwest China, as it includes dust aerosol in NQAPMS. Overall, the simulation in the nested domain of the global model is comparable to the regional model.

3.3.2 Trace gas evaluation in cities

To gain deeper insight into the performance of IAP-AACM in cities, the nested simulation was compared with daily averaged observations in 12 cities across China. We first focused on NO₂ and SO₂, as they are precursors of SNA aerosols. The monthly variation of SO₂ and NO₂ against observations is shown in Figs. 13 and 14. Three-quarters of cities show an annual concentration of SO₂ of around 25 µg m⁻³, which is only half of the NO₂ concentration in summer and autumn owing to the strict SO₂ emission reduction policy that has been implemented since 2005. For SO₂, the model shows good agreement with observations except in Wuhan. This probably implies an overestimation of emissions in this city. Furthermore, IAP-AACM reproduces the seasonal variation well, showing good comparison to observations with R values over 0.5 in most cities, as shown in Table 6. In particular, the cities in NC have a high R values of between 0.76 and 0.89.

Figure 13Mean seasonal variation of SO₂ (µg m⁻³) over China. The black and red lines represent the monthly mean concentration of the city-averaged observations and simulation respectively. Gray shaded areas and red vertical bars show ±1 standard deviation over the sites for the observations and for model results respectively. MO and MM represent the annual mean concentration of the observations and simulation respectively.

Download

Figure 14The same as Fig. 13, except for NO₂.

Download

Table 6Summary of the statistics for the global and nested domains. D1 and D2 represent the results for domain 1 and domain 2 respectively. “City Ave” refers to the average of all cities.

Download Print Version | Download XLSX

As illustrated in Fig. 14, the model shows good performance for NO₂ in cities in the PRD and SWC, matching observations well with RMSE values of less than 20 µg m⁻³. The model captures the daily variations well with R values of between 0.49 and 0.7 in NC, the YRD, and the PRD. However, the model overestimates NO₂ in NC, the YRD, and CC in summer. The overestimation of NO₂ in summer is associated with deposition removal process and multiphase chemistry in IAP-AACM. The overestimation of NO₂ and the underestimation of nitrate in daytime in summer and autumn relates to the over-decomposition of nitric acid (HNO₃) under high-temperature conditions in the thermodynamic equilibrium module. Moreover, heterogeneous chemical reactions in the model could be partly responsible for the overestimation in summer. The heterogeneous chemical module coupled in IAP-AACM has been applied in North China in winter (Li et al., 2018). The mechanism significantly improved sulfate simulation under highly polluted conditions (contributing 50 %–80 % of the total concentration of sulfate) and reduced the overestimation of nitrate. However, the simulations that excluded heterogeneous chemical processes showed better performance for NO₂ (shown in Fig. S5). This indicates that a more reasonable mechanism should be considered in model development. Here we also give a comparison of NO₂ VTC over the nested domain between IAP-AACM and GOME-2A in Fig. S6. Generally, the model captures seasonal variations of NO₂ VTC well. In China, the NO₂ VTC is higher during SON and DJF and lower in JJA due to unfavourable diffusion conditions and weaker photochemical reactions in autumn and winter.

Figure 15The same as Fig. 13 components of PM_2.5 include primary PM_2.5, BC, POA, SOA, and SNA.

Download

3.3.3 Aerosol composition evaluation in cities

As shown in Fig. 15, the model performs very well regarding the simulation of PM_2.5. The model reproduces PM_2.5 variation over the 12 cities well, particularly in NC, the YRD, and SWC, with an R of 0.70–0.79, 0.71–0.80, and 0.77 respectively. The model results are close to or slightly lower than site observations with a city averaged NMB of −0.12. The concentration in NC on winter days is below the observations. Underestimation of PM_2.5 in severe haze periods is common in CTMs, mainly as a result of the deficiency in the simulation of SNA and SOA (Zheng et al., 2015; Donahue et al., 2006). Additionally, the simulation of meteorology (e.g. RH, wind speed, and planetary boundary layer height) is more uncertain during severe haze periods. There is a clear underestimation in the PRD and Urumqi, where mean values are less than half of the observations, with a NMB around −0.5. In the PRD, it could be related to the underestimation of precursor emissions and the decomposition of HNO₃. In Urumqi, it could be more closely related to the unrepresentative simulation with coarse grids. Compared with the scale of urban areas and local emissions in Urumqi, the model grid area is too large. Thus, the averaged value of grids is significantly lower than the local site records. Furthermore, dust is an important component of PM_2.5 in Urumqi, and this is not included in the result.

Figure 16Daily variation of aerosol components (µg m⁻³) over China. The black and red lines represent the daily mean concentration of the city-averaged observations and simulation respectively. BJ, XZ, NJ, and WH refer to Beijing, Xinzhou, Nanjing, and Wuhan respectively. R, MO, and MM stand for correlation coefficient, mean concentration of observation, and mean concentration of model respectively. (“Organic” on the y axis of the lower panel refers to organic matter.)

Download

To assess the performance of IAP-AACM with respect to representing aerosol components, we compare the model results with four stations in NC, the YRD, and CC in Fig. 16. Generally, the model represents the variation of BC well with R values ranging from 0.5 to 0.8 – the values are close to observations. As a primary species, the simulation of BC depends on the emission inventory and meteorological conditions. Unlike BC, OM is underestimated at these stations, with a negative bias of 8–12 µg m⁻³. For SNA aerosols, sulfate is close to observations in the northern cities (Beijing and Xinzhou), but is underestimated by about 10 µg m⁻³ in southern and central cities (Nanjing and Wuhan). As the concentration of SO₂ in Wuhan is overestimated, this suggests an underestimation of transportation or the insufficient oxidation of sulfate; the insufficient conversion of sulfate has been discussed widely in recent years (Cheng et al., 2016; He et al., 2014). Moreover, SO₂ emitted by coal power plants plays a vital role in the formation of sulfate. The coarse grid resolution is insufficient to reproduce the rapid conversion of H₂SO₄ to particles in the plume. The gas-phase oxidation (${\mathrm{SO}}_{\mathrm{2}}+\mathrm{OH}\to {\mathrm{H}}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}\left(\mathrm{g}\right)$) is very sensitive to meteorological variables (particularly radiation and temperature) and the gas (OH and NO_x) concentration around the stacks (Stevens et al., 2012). The results for ammonium show similar characteristics. The simulation of nitrate is highly underestimated with a difference of 6–16 µg m⁻³. The underestimation is due to a high frequency of “zero” values during the daytime in summer and autumn. As discussed in Sect. 3.3.2, this is caused by the over-decomposition of HNO₃ under high-temperature conditions in the thermodynamic equilibrium module. Schaap et al. (2011) found the same phenomenon in the LOTOS-EUROS model using ISORROPIA and recommended improvements to the equilibrium module, including coarse-mode nitrate.