The Advanced Research WRF version 3 (WRF-ARW) is used to simulate meteorological processes and advection of atmospheric components in the WRF/CUACE v1.0 model. The WRF-ARW is a state-of-the-science mesoscale meteorological model, making simulations that are based on actual atmospheric conditions or idealized conditions feasible (Langkamp and Böhner, 2011). The equation set for the WRF-ARW is fully compressible and Eulerian non-hydrostatic with a run-time hydrostatic option. It is conservative for scalar variables. The prognostic variables consist of velocity components u and v in Cartesian coordinates, vertical velocity w, perturbation potential temperature, perturbation geopotential, and perturbation surface pressure of dry air, as well as several optional prognostic variables depending on the model physical options (Skamarock et al., 2008; Wong et al., 2012).

The CUACE is a unified chemistry module, which treats most of the physical and chemical processes, except advection and convection processes that are done by its host model. The main processes treated in the CUACE module include emissions, gas chemistry, dry and wet deposition, vertical mixing, aerosol–cloud interaction, and clear air (i.e., aerosols produced by chemical transformation of their precursors together with particle nucleation, condensation, and coagulation) (An et al., 2016; Zhou et al., 2012; Gong et al., 2003).

The CUACE is typically configured with the second generation of the Regional Acid Deposition Model (RADM2) as its gas chemistry module, which represents 63 species through 21 photochemical reactions and 136 gas phase reactions. As the gaseous chemistry (RADM2) in the CUACE module is not computationally economic and it is hard coded, which means that it is not conducive to adapting chemical reactions in the future, the CBM-Z photochemical mechanism (Zaveri and Peters, 1999) with a better computational efficiency is added with the KPP protocol (Damian et al., 2002) to replace the RADM2 mechanism. The CBM-Z mechanism contains 55 species, 114 reactions, and 20 photochemical reactions. It is based on the widely used carbon bond mechanism (CBM-IV) and uses the lumped structure approach for condensing organic species and reactions. CBM-Z extends the CBM-IV to include revised inorganic chemistry; explicit treatment of the lesser reactive paraffins, methane and ethane; revised treatments of reactive paraffin, olefin, and aromatic reactions; inclusion of alkyl and acyl peroxy radical interactions and their reactions with NO3; inclusion of organic nitrates and hydroperoxides; and revised isoprene chemistry. Currently, stratospheric chemistry is not included in the CUACE module. Species (i.e., CH4, CO, O3, NO, NO2, HNO3, N2O5, and N2O) above a specified pressure level are fixed to climatological values. Between the specified pressure level and the tropopause level, the species was relaxed with a 10 d relaxation factor.

The Canadian Aerosol Module (CAM) (Gong et al., 2003) is adopted as its aerosol module. There are in total seven types of aerosols treated in CAM, i.e., black carbon, primary organic carbon, sulfates, nitrates, ammonium, soil dust, and sea salts. The sea salt emissions are calculated online using the parametrization scheme developed by Gong et al. (2003). Soil dust emissions are simulated using the Marticorena–Bergametti–Alfaro scheme (Alfaro and Gomes, 2001; Marticorena and Bergametti, 1995). With the exception of ammonium, the aerosol size spectrum is divided into 12 bins with fixed boundaries of 0.005–0.01, 0.01–0.02, 0.02–0.04, 0.04–0.08, 0.08–0.16, 0.16–0.32, 0.32–0.64, 0.64–1.28, 1.28–2.56, 2.56–5.12, 5.12–10.24, and 10.24–20.48 µm. A detailed description of aerosol physical and chemical processes in the CAM module can be found in Gong et al. (2003).

The CUACE module currently parameterizes particle dry deposition velocity according to the method of Zhang et al. (2001) (Z01), which tends to overestimate the dry deposition, especially for fine particles (Petroff and Zhang, 2010). In this study, we use the scheme developed by Petroff and Zhang (2010) (PZ10) to replace the original scheme in the CUACE module. The most significant difference between the Z01 and PZ10 scheme is the treatment of Rs, which stands for the dry velocity contributed by surface resistance, consisting of Brownian diffusion, turbulent impaction, interception, and rebound. According to the study of Wu et al. (2018), the dry deposition velocity of fine particles is strongly affected by the Brownian diffusion and turbulent impaction. Thereby, it could be inferred that the Z01 scheme is prone to overestimate the effect of Brownian diffusion and turbulent impaction. In a recent study by Emerson et al. (2020), with an observationally constrained approach, the Z01 scheme was revised to have a weaker effect of Brownian diffusion and as a result showed better performance in simulating the dry deposition velocity of fine particles.

Both of the Z01 and PZ10 schemes use the “resistance” analogy, but with quite different formulas. The PZ10 scheme improved the surface resistance and collection efficiency of the Z01 scheme to overcome the problem of overestimating the dry deposition velocity of fine particles. The PZ10 scheme is detailed as follows: 1Vd=Vdrift+1Ra+Rs. Here Vd is the dry deposition velocity; Vdrift represents drift velocity, which is equal to the sum of gravitational settling and phoretic velocity and is expressed as 2Vdrift=Vg+Vphor, where Vg is the gravitational settling velocity and Vphor accounts for the phoretic effects that are related to differences in temperature, water vapor, or electricity between the collecting surfaces and the air (Wu et al., 2018).

The aerodynamic resistance (Ra) and surface resistance (Rs) are calculated differently for vegetated and unvegetated surfaces. For vegetated surfaces, Ra is parameterized as 3Ra=1κ⋅u∗ln⁡zR-dh-d-ΨhzR-dLO+Ψhh-dLO, where κ is the von Karman constant (0.4), u∗ is the friction velocity above the canopy, zR is the reference height, h is the canopy height, d is the displacement height of the canopy, LO is the Obhukov length, and Ψh is the integrated form of the stability function for heat.

Surface resistance (Rs) is generally expressed as the reciprocal of the surface deposition velocity (Vds), which is parameterized as 4Vds=u∗Eg1+QQg-α2tanh⁡(η)η1+Qg+α2tanh⁡(η)η, where Eg=Egb+Egt is the total collection efficiency on the ground below the vegetation. Egb and Egt represent Brownian diffusion and turbulent impaction, respectively. Egb is parameterized as 5Egb=Sc-2314.516ln⁡(1+F)21-F+F2+13Arctan2F-13+π63-1, where F is a function of the Schmidt number (Sc) and is parameterized as F=Sc13/2.9. Egt is expressed as 6Egt=2.5×10-3CIT⋅τph+2, where CIT is a constant taken as 0.14, and τph+ is a function of non-dimensional relaxation time of the particle (Petroff et al., 2010).

In Eq. (4), the non-dimensional timescale parameter, Q, represents the ratio of the turbulent transport timescale to vegetation collection timescale, and Qg is the analogy of Q used for the transfer to the ground. Q≪1 characterizes a situation where turbulent mixing is efficient and the transfer of particles is limited by the collection efficiency on leaves. Meanwhile, Q≫1 corresponds to a situation where particles are efficiently collected by leaves and transfer of turbulent mixing is limited. Q and Qg are defined as 7Q=LAI⋅ET⋅hlmp(h),8Qg=Eg⋅hlmp(h), where LAI is the two-sided leaf area index, ET is the total collection efficiency by various physical processes, and lmp is the mixing length for particles. ET is expressed as 9ET=Uhu∗EB+EIN+EIM+EIT, where Uh is the horizontal mean wind speed at canopy height h, and EB, EIN, EIM, and EIT are the collection efficiencies by Brownian diffusion, interception, inertial impaction, and turbulent impaction, respectively. The term η is taken as 10η=α24+Q, where α is the aerodynamic extinction coefficient and is expressed as 11α=kx⋅LAI12k21-dh213ϕm23h-dLO, where kx is the inclination coefficient of the canopy elements and ϕm is the non-dimensional stability function for momentum.

For non-vegetated surfaces, the aerodynamic resistance Ra is calculated as 12Ra=1κ⋅u∗ln⁡zR-dz0-ΨhzR-dLO+Ψhz0LO, and the surface deposition velocity Vds is expressed as 13Vds=u∗(Egb+EIT).

The study of heterogeneous chemical reactions mostly focuses on the surface of dust aerosols, but the parameterization schemes of heterogeneous chemical reactions on different types of aerosol have not been well established (Zheng et al., 2015). The following are the heterogeneous chemical reactions on aerosol surfaces that added to the CUACE module in this study (“Aerosol” in the reactions stands for all the aerosols in the model): R1H2O2 (gas)⟶AerosolProducts,R2HNO3 (gas)⟶Aerosol0.5NO3-+0.5NOx (gas),R3HO2 (gas)+Fe(II)→Fe(III)+H2O2,R4N2O5 (gas)⟶Aerosol2NO3-,R5NO2 (gas)⟶AerosolNO3-,R6NO3 (gas)⟶AerosolNO3-,R7O3 (gas)⟶AerosolProducts,R8OH (gas)⟶AerosolProducts,R9SO2 (gas)⟶AerosolSO42-. Reactions (R2), (R4)–(R6), and (R9) describe the formation of sulfate and nitrate on the surface of sand dust, and the other four reactions describe mineral aerosols as sinks of gaseous substances. In this study, these nine heterogeneous reactions were extended to all types of aerosol surface in the CUACE, referring to the approach of Zheng et al. (2015) for the CMAQ model. The first-order chemical kinetic equation for calculating the adsorption efficiency of a gas on an aerosol surface is 14dCidt=-kiCi, where Ci represents the concentration of gas i, and ki is the pseudo-first-order rate constant and is supposed to be irreversible. The value of ki is defined referring to Jacob (2000) as 15ki=aDi+4viγi-1A, where a is the aerosol diameter, Di is the diffusion coefficient for gas reactant i, vi is the mean molecule speed of gas reactant i, γi is the uptake coefficient of the heterogeneous reaction for the gas reactant i, and A is the surface area of aerosols in unit volume air. The value of γi is obtained from previous laboratory studies (Table 1), and other parameters are calculated in the WRF/CUACE v1.0 model.

At present, there are four major polluted areas in China, namely, the North China Plain (NCP), the Yangtze River Delta (YRD), the Pearl River Delta (PRD), and Sichuan Basin (SCB). To include all these regions, the simulation area is configured as in Fig. 2. There are two domains in total. The boundary field of the inner domain is obtained by the interpolation of its outer domain. The outer region covers the whole of East Asia and its adjacent areas with a horizontal resolution of 54 km and a total of 120×110 grids centered at 30.46∘ N and 105.82∘ E. The inner region covers most of China on the east side of the Qinghai–Tibet Plateau with a horizontal resolution of 18 km and 193×175 grids. There are 32 vertical layers with the top pressure at about 100 hPa. The main physical and chemical options in the model are shown in Table 2. With WRF used in non-hydrostatic mode, we performed two simulations: one for January, April, July, and October in three years, 2013, 2015, and 2017, to evaluate the model on a long timescale, and one for three periods during which SIA observations were conducted (i.e., 5–16 January 2019 in Langfang, 3–29 December 2013 in Nanjing, and 1–10 January 2017 in Chengdu), to investigate improvements in simulating SIA with heterogeneous chemistry.

The model uses the FNL global reanalysis data of the NCEP (National Centers for Environmental Prediction) to provide the meteorological initial and boundary fields with spatial and temporal resolution of 6 h and 1∘ × 1∘, respectively. The initial and boundary chemistry conditions are based on the vertical profiles of O3, SO2, NO2, VOCs (volatile organic compounds), and other air pollutants from the NOAA Aeronomy Lab Regional Oxidant Model (NALROM) (Liu et al., 1996).

Anthropogenic emissions are derived from the MIX emission inventory representative for 2010 (http://www.meicmodel.org/dataset-mix.html, last access: 18 February 2020) (Li et al., 2017), which is an Asian anthropogenic emissions inventory developed for the third phase of the East Asian Model Comparison Plan (MICS-Asia III) and the United Nations Hemispheric Atmospheric Pollution Transport Plan (HTAP). The inventory provides monthly grid emission data with 0.25∘ spatial resolution for five emission sectors (electricity, industry, civil, transportation, and agriculture), including PM2.5, PM10, nitrogen oxides (NOx), sulfur dioxide (SO2), carbon monoxide (CO), NH3, black carbon (BC), organic carbon (OC), and non-methane volatile organic compounds (NMVOCs). During the simulation span from 2013 to 2017, China carried out strict air pollution control measures, which had a considerable impact on anthropogenic emissions. To make the anthropogenic emissions more suitable for the real emissions scenarios in the simulated years, the emissions in mainland China were replaced with the MEIC emissions inventory representative for 2012, 2014, and 2016 to represent the emissions scenarios in 2013, 2015, and 2017, respectively. Figure S2 in the supplement shows the MEIC emissions of PM2.5, NOx, SO2, and CO in the three years, from which it can be seen that anthropogenic emissions of PM2.5, SO2, and CO decreased remarkably from 2012 to 2016.

For the vertical interpolation, we used the settings of Wang et al. (2010) and Zhou et al. (2017). The industrial emissions were allocated as 50 %, 30 %, and 20 % in layers one to three of the model, respectively, and the power plant emission sources were allocated as 14 %, 46 %, 35 %, and 5 % in model layers two to five, respectively. The emissions from transportation, residential areas, and agriculture were 95 % and 5 %, respectively, in the first and second layers of the model. Then, the inventory was distributed into hourly emissions using the monthly, weekly, and hourly profiles established by Tsinghua University (2006). VOCs released from vegetation were calculated online using the MEGAN model (Guenther, 2006).

Heterogeneous chemical reactions have been shown to have important effects on the formation of SIA, especially during severe haze events with high humidity (Li et al., 2011; Wang et al., 2006; Zhao et al., 2013). The ground observations of SIA from 5 to 16 January 2019 in Langfang (NCP), from 3 to 29 December 2013 in Nanjing (YRD), and from 1 to 10 January 2017 in Chengdu (SCB) were collected for the evaluation of SIA simulations. Following the model configurations in Sect. 4.2, we performed WRF/CUACE v1.0 simulations with (Exp_WH) and without (Exp_WoH) heterogeneous chemistry on the three periods.

Figure 5 illustrates the hourly variations of observed SIA concentration from the Exp_WH and Exp_WoH experiments. For Langfang site, the simulation without heterogeneous chemistry (Exp_WoH) barely captured the sulfate increase (Fig. 5a). This was substantially improved when heterogeneous chemistry was included (Exp_WH), although some observed peak values are not well captured, such as those on 14 January. The overestimation of nitrate was also improved, with the NMBs changing from 124.1 % to 96.0 % (Fig. 5b). It should be noted that the responses of sulfate and nitrate to heterogeneous chemistry are inverse, which might be attributed to the complex thermodynamic processes of SIA formation (Zheng et al., 2015). Sulfate and nitrate will compete for ammonium, which is now the only cation in the CUACE model, resulting in less ammonium nitrate and more ammonium sulfate because of the more thermodynamically stable features of ammonium sulfate. As a result of the dramatic increase in sulfate in Exp_WH, the ammonium concentrations increase slightly relative to those in Exp_WoH to achieve anion–cation balance, which leads to more overestimations in the Exp_WH experiment (Fig. 5c). For Nanjing and Chengdu site, the underestimation of sulfate (Fig. 5d and g) and overestimation of nitrate (Fig. 5e and h) were also improved to varying degrees, with bias of sulfate changing from -95.3 % to -68.4 % in Nanjing and from -88.7 % to -80.1 % in Chengdu and the bias of nitrate changing from 83.0 % to 54.6 % in Nanjing and from 67.6 % to 23.5 % in Chengdu. Nonetheless, deviations in SIA simulations are still too large to neglect in those regions.

It is necessary to compare the MM5/CUACE model with the new WRF/CUACE model for the purpose of assessing the viability of the newly developed model. To this end, a simulation was performed using the MM5/CUACE model for a winter month, i.e., January 2013, during which a long-lasting haze event occurred in central and eastern China. The domain setting, anthropogenic emission inventory and initial and boundary fields of meteorology and chemistry are as the same as those of the WRF/CUACE in Sect. 5.1. It should be known that the gas-phase chemistry mechanism and particle dry deposition scheme in the MM5/CUACE model are RADM2 and Z01, respectively, that updated to CBM-Z and PZ10 in the new WRF/CUACE model. The physical parameterization used in the MM5/CUACE is shown in Table S3 in the Supplement.

Figure 6 presents a comparison of the modeled and observed daily concentrations of PM2.5, O3, NO2, and SO2 in the four regions. It can be seen that the concentrations of PM2.5, NO2, and SO2 simulated in WRF/CUACE are closer to the observations relative to those of MM5/CUACE model (change in bias from -23.0 % to -19.2 % for PM2.5, from 14.7 % to -2.4 % for NO2, and from -46.2 % to -37.5 % for SO2). The daily variations of the three species are also relatively better captured by the WRF/CUACE model (reflected by the R values changing from 0.45 to 0.62 for PM2.5, from 0.41 to 0.49 for NO2, and from 0.19 to 0.32 for SO2). For O3, the differences of statistical metrics between the two models are not obvious. The MM5/CUACE model performed with a slightly smaller bias of -10.7 % but with a lower R value of 0.50, which are 14.3 % and 0.55, respectively, in the WRF/CUACE simulation. In summary, the new WRF/CUACE model performed better than the MM5/CUACE model in simulating air pollutants.