Table 1 summarizes major mechanisms for sulfate and nitrate formation currently treated in the original CMAQ v5.0.1 (R1–R15) in a highly simplified manner. In the gas phase (R1–R6), sulfuric acid (H2SO4) and HNO3 are generated mainly through the oxidation of SO2 and nitrogen oxide (NOx) by OH. Additional HNO3 can be formed through subsequent reactions involving reactive nitrogen species such as nitrogen trioxide (NO3), N2O5, and NTR and OH, hydroperoxyl radical (HO2), and H2O as well as the nighttime oxidation reaction of volatile organic compounds (VOCs) by NO3. H2SO4 and HNO3 can condense on the surface of pre-existing aerosol, forming sulfate (SO42-) and nitrate (NO3-). For in-cloud chemistry (R7–R13), the original CMAQ includes the dissolution equilibria of SO2, H2SO4, ammonia (NH3), NOx, NO3, N2O5, nitrous acid (HNO2), HNO3, peroxynitric acid (HNO4), and several oxidants such as OH, H2O2, and O3, the dissociation equilibria of SO2, bisulfite (HSO3-), HNO2, HNO3, and NH3 ⋅ H2O, and five aqueous-phase kinetic reactions to produce S (VI) through the oxidation of S (IV) (dissolved SO2, HSO3- and sulfite (SO32-)) by H2O2, methylhydroperoxide (MHP), peroxyacetic acid (PAA), O3, and oxygen (O2) catalyzed by ferric iron (Fe3+) and manganese ion (Mn2+). Once clouds dissipate, SO42- formed in the aqueous phase becomes part of aerosol. The original CMAQ only includes two heterogeneous reactions (R14–R15) to produce HNO3, one involving N2O5 and H2O and the other involving nitrogen dioxide (NO2) and H2O. The mechanism of heterogeneous chemistry is much more complex than the homogeneous gas- and aqueous-phase mechanisms. It involves many processes including water condensation onto the particle surfaces, adsorption and accommodation of gases into the liquid–gas interface, diffusion, and surface reactions (Reid and Sayer, 2003). Heterogeneous reaction rates are dependent on relative humidity (RH) (Dentener et al., 1996; Henson et al., 1996; Stutz et al., 2004) because of the significant role of the water film on the aerosol surface in the gas uptake. The formation of ammonium (NH4+) is closely related to that of SO42- and NO3-, as it results from the neutralization of SO42- and NO3- by dissolved NH3 in the particulate phase through aerosol equilibrium treated in ISORROPIA II of Fountoukis and Nenes (2007).

Heterogeneous chemistry might have played a significant role in the January 2013 haze episode for three reasons. First, the total amount of SO42- formed through gas- and aqueous-phase chemistry is too low to explain the observed abrupt increases in the concentrations of SO42- by 70–130 µg m-3 within a few hours during the haze episode. The observed concentrations of SO2 are in the range of 10–216 µg m-3 (Fig. 1). The gas-phase oxidation of SO2 by OH radicals can convert SO2 to H2SO4 at a maximum rate of 2 % h-1 under sunny conditions, leading to 0.2–5.5 µg m-3 h-1 H2SO4 (which is equivalent to 0.2–5.4 µg m-3 h-1 SO42-). Aqueous-phase chemistry as shown in Table 1 can enhance SO42- formation in precipitating clouds, which did not occur frequently during the episode. Only two precipitations are recorded in central China, on 20–21 and 30–31 January, which contribute 92 % of the total precipitation in January (data derived from http://cdc.cma.gov.cn). Meanwhile the weak photochemical activity during dim haze days, characterized by extremely low or even zero O3 concentrations (He et al., 2014; Wang et al., 2014c), does not support that gas- and aqueous-phase chemistry are dominant pathways for sulfate and nitrate production. As shown in Table 1, the original CMAQ only includes two heterogeneous reactions to produce HNO3 and does not include any heterogeneous reactions to produce SO42-. The original model evaluation against ground-based measurements (as shown in Sect. 4.2.1) shows significant underpredictions of SNA (e.g., normalized mean biases (NMBs) of -40 to -60 %). These data analysis and modeling results indicate that the heterogeneous chemistry probably have played a significant role to produce high SNA during the haze pollution. Second, there exist strong correlations between RH and sulfur and nitrogen oxidation ratios (SOR and NOR) during haze in January 2013 (Sun et al., 2014; Wang et al., 2014c; Zheng et al., 2014b), which resemble the RH-dependence of heterogeneous chemistry. Third, transmission electron microscopy studies have shown that the particles sampled during haze days in the NCP are mostly combined with obvious coatings containing significant sulfur and nitrogen elements, probably generated via some reactions on the particle surfaces (Li and Shao, 2009, 2010; Li et al., 2011). This suggests that surface reactions, probably caused by heterogeneous chemistry, play a significant role in haze formation. Based on the above three reasons, heterogeneous chemistry is regarded as the most important missing reaction pathway and nine new heterogeneous reactions (R16–R24) were therefore incorporated into CMAQ to improve its capability in reproducing the high SNA concentrations observed during the haze episode through increasing sulfate and nitrate formation. Simulations from the original and the revised CMAQ are compared to study the role of heterogeneous chemistry in producing sulfate and nitrate during this haze episode, as presented in Sects. 4.2 and 4.3.

As shown in Table 1, following the work of Wang et al. (2012a), nine heterogeneous reactions involving H2O2, HNO3, HO2, N2O5, NO2, NO3, O3, OH and SO2 (R16–R24) have been incorporated into the original CMAQ. These reactions are assumed to occur on the surface of aerosols. Heterogeneous chemistry is commonly parameterized using a pseudo-first-order rate constant and is assumed to be irreversible (Zhang and Carmichael, 1999; Jacob, 2000). The rate constant k (s-1) for heterogeneous loss of gaseous pollutants is determined by (Jacob, 2000; Wang et al., 2012a) ki=dp2Di+4viγi-1Sp where i represents the reactant for heterogeneous reactions, dp is the effective diameter of the particles (m), Di is the gas-phase molecular diffusion coefficient for reactant i (m2 s-1), vi is the mean molecular speed of reactant i in the gas phase, γi is the uptake coefficient for reactant i (dimensionless), and Sp is the aerosol surface area per unit volume of air (m2 m-3). The parameters dp, Di, vi, and Sp are calculated in CMAQ, and the parameter γi is determined for different reactants based on laboratory measurements reported in the literature, as presented below.

The values of γ for different gaseous pollutants may vary by several orders of magnitude, because of different surface properties, particle compositions, temperature, RH, and laboratory conditions. For a specific combination of particle and gaseous pollutants, the value of γ is highly dependent on RH and increases rapidly as a function of RH (Dentener et al., 1996; Henson et al., 1996; Stutz et al., 2004). For example, Mogili et al. (2006) found that the γ of N2O5 increased by a factor of 4 as RH increased in an environmental aerosol chamber. Liu et al. (2008) reported that the γ of HNO3 on calcium carbonate was enhanced in laboratory experiments by a factor of 15 over a wide range of RHs (from 20–80 %). Enhanced γ of HNO3 with increasing RH have also been reported on many types of particles including oxides, clay and dust. Considering the significant effect of RH on γ, some modeling studies used RH-dependent γ (Song and Carmichael, 2001; Wei, 2010). For example, Song and Carmichael (2001) used a value of γ of 0.005 for SO2 when the RH was lower than 50 % and of 0.05 when RH was higher than 50 %.

The γfor heterogeneous reactions used in this work are determined mainly based on the work of Wang et al. (2012a), which used lower and upper limits to represent a range of γ values reported in the laboratory measurement. On the basis of the lower and upper limits, we then use a piecewise function to represent the RH-dependence of γ. Field measurements during the January 2013 haze episode in Beijing indicate that the SOR and NOR are highly dependent on RH. They are relatively stable when RH is lower than 40–50 % and rapidly increase when RH is higher. The RH value of 50 % is close to the deliquescence point of particles for a mixture of organic compounds and ammonium sulfate (Peckhaus et al., 2012), which constitute about 80 % of PM2.5 in China (Yang et al., 2011). In this work, we assume the value of γ to be the lower limit for RH ≤ 50 % and that it increases linearly to the upper limit as RH increases to RHmax, which approximates the correlation between RH and γ. The γ values of the reactions contributing to sulfate and nitrate (R19–R21, R24) are calculated as the following equation: γi=γlow,RH∈[0,50%]γlow+(γhigh-γlow)/(RHmax⁡-0.5)×(RH-0.5),RH∈(50%,RHmax⁡]γhigh,RH∈(RHmax⁡,100%] where i represents the reactant for heterogeneous reactions, RHmax is the RH value at which the γ reaches the upper limit, and γlow and γhigh are the lower and upper limits of γ values taken from Table 2 of Wang et al. (2012a) with one exception for R24.

In situ observations have found significant enhancement of SO2 oxidation rates under wet conditions, indicating possible missing heterogeneous reactions on deliquescent particles (Zheng et al., 2014b). However, the coefficients of SO2 uptake by aerosols (R24) are only established for ice surfaces and mineral dust particles (Kolb et al., 2010). As the parameterization of heterogeneous reaction of SO2 on soot, organics and SNA aerosols are not well established yet, we first arbitrarily selected the uptake coefficients from Wang et al. (2012a) and conducted four sensitivity runs (S1, S2, S3 and S4) by adjusting the uptake coefficients with a successive approximation approach. The parameters and evaluations of the four sensitivity runs are presented in the Supplement (Table S1 and Fig. S1). The γ values of the lower and upper limits of SO2 recommended by Wang et al. (2012a) are 1.0 × 10-4 and 2.6 × 10-4, respectively, whereas other works recommended lower γ values for SO2 – e.g., 4.0 × 10-5 in Crowley et al. (2010), 1.35 × 10-5 in Shang et al. (2010), and 0.6 × 10-5 to 2.45 × 10-4 in Wu et al. (2011). We found that using the γ in Wang et al. (2012a) for R24 (sensitivity run S1) produced unreasonably high sulfate for this haze episode. We finally chose the value from S3 in our work, which can best match observations.

We assume the RHmax of sulfate-related heterogeneous reaction (R24) to be 100 %, and that of nitrate-related heterogeneous reactions (R19–R21) to be 70 %. This assumption is made on the basis of the observational result that the SOR increases when the RH rises from 50 to 100 % and the NOR increases when the RH rises from 50 to 70 % and then stays stable when the RH continues to increase. The similar relationship between sulfur (nitrogen) conversion ratios and RH has also been reported in another pollution episode that occurred in the winter of 2011 in Beijing (Sun et al., 2013). For other heterogeneous reactions, we use the mean of lower and upper limit values in the model and assume that they remain constant under different RHs.

WRF/CMAQ simulations are performed over East Asia at a horizontal resolution of 36 × 36 km (see Fig. 2). The simulation period is from 1 to 31 January 2013 with an additional 7 days used as a spin-up period to minimize the influence of initial conditions.

The physics options selected for the WRF simulation are summarized in Table 2. They are selected based on a number of initial simulations with different option combinations to ensure the best performance for meteorological predictions against observations during this episode. The meteorological initial and boundary conditions (ICs and BCs) are based on the National Centers for Environmental Prediction Final Analysis (NCEP-FNL) reanalysis data. The surface roughness is corrected by increasing the friction velocity by 1.5 times only in the boundary layer scheme to reduce the high biases in wind speed (Mass and Ovens, 2010).

The configurations and options used in the CMAQ model are summarized in Table 3. The gas-phase mechanism module is the CB05 gas-phase mechanism with active chlorine chemistry and updated toluene mechanism of Whitten et al. (2010). The aqueous-phase chemistry is based on the updated mechanism of the Regional Acid Deposition Model (RADM) model (Walcek and Taylor, 1986; Chang et al., 1987). The aerosol mechanism applied in this study is the AERO6 aerosol module. The photolytic rates are calculated in-line using simulated aerosols and ozone concentrations. The ICs and BCs are generated from the GEOS-Chem model (Bey et al., 2001).

Anthropogenic emissions for China in 2013 used in this work are derived from the MEIC model (Multi-resolution Emission Inventory of China, http://www.meicmodel.org). The MEIC model is a dynamic and technology-based emission model developed by Tsinghua University which estimates anthropogenic emissions for about 700 emitting sources over China with unified methodology (Zhang et al., 2007, 2009; Lei et al., 2011a). The MEIC model is an update of the bottom-up emission inventory developed by the same group (Zhang et al., 2007, 2009; Lei et al., 2011a) with several updates such as unit-based emission data for power plants (Wang et al., 2012b) and cement plants (Lei et al., 2011b), high-resolution vehicle emission inventory at county level (Zheng et al., 2014a), and new NMVOC mapping approach for different chemical mechanisms (Li et al., 2014). In the MEIC model, the latest available emission data with real statistics at provincial level is for 2012. In this work, emissions for the year of 2013 are used from the extrapolation of the 2012 estimates and updated based on brief statistics at country level.

Anthropogenic emissions from the other Asian countries and biomass burning emissions are taken from the MIX emission inventory prepared for the Model Inter-comparison Study Asia Phase III (MICS-ASIA III). Biogenic emissions are calculated by the MEGAN v2.1 (Guenther et al., 2012). Sea salt emission and dust emission are calculated online on the basis of the algorithms developed by Gong (2003) and a physical-based dust emission algorithm FENGSHA (http://www.airqualitymodeling.org/cmaqwiki/index.php?title=CMAQv5.0_Windblown_Dust), respectively.

Using the WRF/CMAQ modeling system, the impacts of heterogeneous chemistry and the meteorological anomaly of 2013 on the significant production of sulfate and nitrate aerosols during the January 2013 haze episode are investigated with three simulations, as shown in Table 4. The simulation Original CMAQ uses the officially released version of CMAQ v5.0.1. In the simulation Revised CMAQ, nine important heterogeneous reactions are implemented in the model to explore the effects of heterogeneous chemistry. To further evaluate the impacts of the 2013 meteorological anomaly on sulfate and nitrate production, another simulation with revised CMAQ is designed to use the same 2013 emissions but with the WRF meteorological predictions for 2012 (Revised CMAQ with 2013Emis&2012Met). The uptake coefficients of heterogeneous chemistry used in the latter two simulations are presented in Table S2 of the Supplement.

The model evaluation is performed in terms of domain-wide performance statistics and site-specific temporal variations. The performance statistics are conducted following the evaluation protocol of Zhang et al. (2006, 2011). The statistical parameters include correlation coefficient (R), mean bias (MB), root mean square error (RMSE), NMB and normalized mean error (NME).

Table 5 summarizes the observational data sets used for model evaluation in this study. Three observational data sets are used including the meteorological data from the National Climate Data Center (NCDC), the real-time gaseous and particulate concentrations in 74 cities from the China National Environmental Monitoring Center (CNEMC), and hourly concentrations of chemical species of PM2.5 from the ground-based measurement at the Tsinghua University site (THU) located in northwestern Beijing. A detailed description of these data sets can be found in the Supplement.

Table 6 presents the statistical performances of the meteorological predictions, including temperature at 2 m (T2), RH at 2 m (RH2), wind speed at 10 m (WS10), wind direction at 10 m (WD10), and daily mean precipitation (Precip). The near-surface temperature agrees reasonably well with observations with MBs of -0.8 ∘C. Simulated RH2 agrees well with observations across most of China with an NMB of 9.9 % and an MB of 6.7 %. WS10 is overpredicted slightly with an NMB of 9.5 % and an MB of 0.3 m s-1 for the 36 km domain. The MB of Precip is 1.1 mm and the NMB is 58.8 % with a relatively poor performance compared with other meteorological variables. Precip is usually predicted with large biases by meteorological models (Zhang et al., 2011, 2012; Wang et al., 2014b), indicating the limited capability of a model to accurately reproduce the precipitating processes. The simulated meteorological variables show generally good agreement with observations, and the overall performances are consistent with similar work conducted for China using the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5) or WRF models (Liu et al., 2010; Wang et al., 2010, 2014b; Zhang et al., 2011; Wang et al., 2012a; Fu et al., 2014). The simulated meteorological variables agree well with observations in terms of temporal variations and magnitudes at the THU site (as shown in Fig. 3), confirming the reliability of meteorological prediction at location with SNA observation data.

Figure 4 compares the temporal variations of PM2.5, SO42-, NO3-, NH4+, OC and EC at the THU site simulated by the original and revised CMAQ with observations. The sulfate and nitrate simulations with heterogeneous chemistry are improved significantly in terms of both magnitude and temporal variation. In particular, the significant discrepancies in PM2.5, SO42-, NO3- and NH4+ between the observed and simulated concentrations during severely polluted days are improved, although a couple of observed peak values are still not captured well. The synergic improvement of SNA predictions illustrates the significant role heterogeneous chemistry plays in the haze pollution events. The revised CMAQ shows better performance with NMBs of 0.4, 6.3, 5.7 and -4.1 %, for PM2.5, SO42-, NO3- and NH4+, respectively. The MBs of sulfate and nitrate are reduced, changing from -17.8 to 2.1 µg m-3 and from -12.3 to 1.8 µg m-3, respectively. As expected, the simulated level of PM2.5 is also improved with MBs changing from -40.8 to 0.8 µg m-3.

It should be noted that the revised CMAQ model still significantly underestimated the peak PM2.5 concentration on 13 January 2013. Zheng et al. (2014b) argued that the abrupt increase of PM2.5 concentration on 13 January represented rapid recovery from an interruption to the continuous pollution accumulation over the region rather than local chemical production. Our model also failed to predict the high PM2.5 concentration on 13 January over the polluted region (e.g., Langfang and Shijiazhuang, see Supplement), but agreed well with observation in upwind cities (e.g., Chengde). In this case, the model may have underestimated the regional transport in polluted areas given the fact that the wind speed was underestimated at the THU site.

The revised CMAQ can capture the enhancement of relative contribution of SNA from clean days to polluted days, as shown in Fig. 5. Observations show that the fractions of SNA increase rapidly to 42.2 % and 50.3 % on polluted and heavily polluted days, which are well reproduced by the revised CMAQ with fractions of 49.0 % and 52.6 %. For comparison, the original CMAQ gives SNA fractions of 32.1 % and 30.8 %, which are considerably lower. During polluted and heavily polluted days, there exist significant discrepancies in SNA percentage contributions between the original and revised CMAQ, indicating the important role of heterogeneous chemistry in haze pollution. It should be noted that the good agreement between the revised CMAQ and observations is highly dependent on the selections of uptake coefficients, as discussed in Sect. 2.2. However, all sensitivity runs can reproduce the enhancement of relative contribution of sulfate in haze days (Fig. S1), implying the importance of heterogeneous chemistry. Laboratory measurements of uptake coefficients on the surfaces of mixed and deliquescent aerosols will help to confirm our findings in the future.

The evolution patterns of SNA simulated by the revised CMAQ are also generally consistent with other field observations on haze episodes in China, which confirmed the significance of heterogeneous chemistry in the haze formation process over China. The enhanced SNA contribution in haze days compared to clean days were also observed in other field campaigns, where the heterogeneous chemistry was attributed as the most probable pathway of observed abrupt increases in SNA aerosols as the oxidation rates of gas-phase and aqueous-phase chemistry were too slow (Zhao et al., 2013b; Ji et al., 2014; Quan et al., 2014; Wang et al., 2014c). Strong correlations between RH and sulfur and nitrogen oxidation ratios (SOR and NOR) were found during haze episodes (Wang et al., 2012c, 2014c; Sun et al., 2014; Zheng et al., 2014b) with sharp increase of SOR and NOR when RH exceeds 50 %, lending support to our assumptions in the revised CMAQ.

The revised CMAQ gives very similar OC and EC predictions as original CMAQ, with large underpredictions in OC during the haze episodes but overpredictions in EC throughout the simulation period for the reasons discussed previously in Sect. 4.2.2. The percentage contributions for EC and OIN are also slightly decreased especially in the polluted and heavily polluted days. This is because the mode-averaging particle diameter is larger due to the enhanced formation of SNA when the heterogeneous reactions are included. The particle settling velocity is increased and thus dry deposition rates are larger, which helps reduce the overpredictions of these species.

The simulation results with and without heterogeneous chemistry are compared over the whole domain to evaluate the impact of heterogeneous reactions during the January 2013 haze episode. Table 8 summarizes the statistical performance for surface concentrations of CO, NO2, SO2, PM2.5 and PM10 from the simulation with original CMAQ and revised CMAQ for 74 cities in China. The original CMAQ model can generally reproduce the concentrations of aerosol and gaseous pollutants over the whole domain. The model underpredicts the concentrations of CO and PM10 with NMBs of -20.6 and -11.2 %, respectively, and overpredicts those of SO2, NO2 and PM2.5 with NMBs of 51.2, 13.4 and 8.1 %, respectively. As expected, the overpredictions in SO2 and NO2 are improved in the revised CMAQ model because the added heterogeneous reactions enhance their conversions to sulfate and nitrate. The positive biases of SO2 and NO2 are reduced from 51.2 to 38.5 % and 13.4 to 11.2 %. We further found that high NMB in SO2 prediction is mainly contributed by provincial capital cities. As the most developed cities within China, the provincial capital cities tend to prohibit coal use in urban areas or use high-quality coal with low sulfur content, which has not been accurately represented in regional emission inventories which are compiled at the provincial level. As a result, SO2 emissions from those capital cities may have been overestimated.

Figure 6 illustrates the concentration of SNA and PM2.5 simulated by the original and revised CMAQ and Fig. 7 further explores the difference in heavily polluted regions. Heterogeneous chemistry enhances SNA concentrations significantly in the most polluted regions in China (the Northeast Plain (NP), NCP, Middle-Lower Yangtze Plain (MLYP), and Sichuan Basin (SB)), leading to the increased PM2.5 concentration over those regions. In southern China (e.g., the Pearl River Delta (PRD)), sulfate concentration is still increased but nitrate concentration is decreased by 5–20 µg m-3, resulting in a reduction of PM2.5 concentration by 10–20 µg m-3. The contrasting responses to heterogeneous chemistry in different regions are because of the complex thermodynamic processes of SNA formation, which differ greatly under NH3-rich and NH3-poor conditions. The polluted regions listed above (NP, NCP, MLYP, and SB) are all NH3-rich regions (Wang et al., 2011; Zhao et al., 2013a), as shown in Fig. 7a, which comprise 24.5 % land areas in China but contribute to 47.4 % cultivated lands (National Bureau of Statistics, 2013) and 48.3 % NH3 emissions (derived from the MEIC model). The abundant NH3 emissions provide sufficient amounts of ammonium to neutralize the increased amounts of sulfate and nitrate formed through heterogeneous chemistry; therefore, the total amount of PM2.5 in these regions increases with enhancement of both sulfate and nitrate. This causes the positive bias of simulated PM2.5 to be larger in the NH3-rich regions, mainly contributed by the overpredictions of EC and OIN. In southern China, which is an NH3-poor region in January (Wang et al., 2011), sulfate and nitrate compete for ammonium and the formation of ammonium sulfate occurs first owing to its more thermodynamically stable characteristics – increased levels of sulfate would thus lead to a decrease of nitrate. This phenomenon could even lead to a decrease in the total concentration of PM2.5, because to neutralize with the same amount of ammonium, the mass of sulfate required is smaller than that of nitrate.

The haze episode in January 2013 was the most serious pollution event in recent years. Why it should happen in 2013 but not in other years is an intriguing question. Emissions of SO2, NOx, and PM2.5 were stable during 2011–2013 (derived from the MEIC model), indicating that emissions are not the critical driving force. The anomalous meteorological conditions (low temperature, high RH, and low wind speed) in January 2013 are identified as the key influence factor of haze formation by affecting radiation, horizontal transport, vertical mixing, and the atmospheric reaction rates of air pollutants (Ding and Liu, 2014; Wang et al., 2014d). As described in Sect. 2.2, meteorological conditions (specifically RH) can affect heterogeneous chemistry by increasing the uptake coefficients of gases. In this section, the impact of the 2013 meteorological conditions on the production of sulfate and nitrate is evaluated using the revised CMAQ with heterogeneous chemistry.

The meteorological conditions of 2012 are selected to represent typical weather conditions because they were very close to the 10-year average climatology conditions with regard to temperature, RH, wind speed, and sea level pressure (data derived from http://cdc.cma.gov.cn) in the region of the NCP. Figure 8 illustrates the spatial distributions of the monthly mean temperature, RH, PM2.5, sulfate, and nitrate simulated by the revised CMAQ with the meteorological fields of 2012 and 2013. The simulated temperature of 2013 in North and East China is 2–3 ∘C lower than that in 2012 and the simulated RH is 5–25 % higher. High RH promotes heterogeneous conversions to generate more sulfate and nitrate and therefore, to increase the total concentration of PM2.5. Significant differences in RH occur in the NCP region, where increases by 15–30 % in RH correspond to increases of PM2.5 concentration by 70–150 µg m-3.

Traditional chemistry mechanisms play a relatively small role during haze formation because of the low solar radiation and low-temperature conditions, and few precipitating clouds, whereas heterogeneous chemistry mechanisms are enhanced by the extremely high RH, which leads to the significant production of sulfate and nitrate aerosols. This provides a perspective to understand how adverse meteorological conditions can affect air quality through reaction pathways that are sensitive to specific meteorological variables. The meteorological anomaly of 2013 occurred not only for temperature and RH, but also for other variables – for example, the shallower PBL and lower wind speed than a typical year. The abnormal changes in these variables also have adverse effects on haze pollution. For example, the height of the PBL across China in 2013 was about 200 m lower than in 2012, which could weaken and confine the vertical mixing of pollutants and thus aggravate surface pollution. Researches on the impact of these factors have been reported in other studies (e.g., Wang et al., 2014d; Zhang et al., 2014b).