Reactive nitrogen chemistry in aerosol water as a source of sulfate during haze events in China

Multiphase chemistry of NO₂ and alkaline matter in aerosol water explains rapid sulfate formation and severe haze in China.

Abstract

Fine-particle pollution associated with winter haze threatens the health of more than 400 million people in the North China Plain. Sulfate is a major component of fine haze particles. Record sulfate concentrations of up to ~300 μg m⁻³ were observed during the January 2013 winter haze event in Beijing. State-of-the-art air quality models that rely on sulfate production mechanisms requiring photochemical oxidants cannot predict these high levels because of the weak photochemistry activity during haze events. We find that the missing source of sulfate and particulate matter can be explained by reactive nitrogen chemistry in aerosol water. The aerosol water serves as a reactor, where the alkaline aerosol components trap SO₂, which is oxidized by NO₂ to form sulfate, whereby high reaction rates are sustained by the high neutralizing capacity of the atmosphere in northern China. This mechanism is self-amplifying because higher aerosol mass concentration corresponds to higher aerosol water content, leading to faster sulfate production and more severe haze pollution.

INTRODUCTION

Persistent haze shrouding Beijing and the North China Plain (NCP) during cold winter periods threatens the health of ~400 million people living in a region of ~300,000 km². Characteristic features of the winter haze in northern China include stagnant meteorological conditions with low mixing heights, high relative humidity (RH), large emissions of primary air pollutants, and fast production of secondary inorganic aerosols, especially sulfate (see section M1) (1–5). Analyzing surface-based observations at a site in Beijing during January 2013 (see section M2) and using concentration ratios of sulfate to sulfur dioxide ([SO₄²⁻]/[SO₂]) as a proxy for the sulfate production rate (5), we find that sulfate production increases as PM_2.5 (particulate matter with a diameter of less than 2.5 μm) levels increase (Fig. 1A). Ratios are six times higher during the most polluted periods (PM_2.5 > 300 μg m⁻³) than during clean to moderately polluted conditions (ratios of 0.1 when PM_2.5 < 50 μg m⁻³).

Fig. 1. Characteristic features of a major haze event in Beijing, China (January 2013).

(A) Sulfate/sulfur dioxide ratio ([SO₄²⁻]/[SO₂]) and (B) midday ozone (O₃) concentrations (10:00 to 15:00 local time) and nitrogen dioxide (NO₂) concentrations at different fine-particle (PM_2.5) concentration levels. (C) Correlation of the unexplained sulfate concentration (ΔSO₄²⁻ = observation − model) (see sections M2 and M3) with aerosol droplet solution and AWC (color-coded). (D) Anion-to-cation ratio in PM_2.5 as observed in Beijing (solid symbols) and in other studies in northern China (open symbols) (table S1) compared to the characteristic ratios reported for aircraft measurements in the Arctic of outflow from East Asia, Europe, and North America (solid red, blue, and green lines; here, only NH₄⁺, NO₃⁻, and SO₄²⁻ were considered) (87).

Traditional air quality models, however, fail to capture this key feature of NCP winter haze events even after accounting for aerosol-radiation-meteorology feedback (see sections M2 and M3) (6–8). The chemical mechanisms used in these models usually comprise gas-phase oxidation of sulfur dioxide by OH radicals and aqueous-phase reaction pathways in cloud water, involving H₂O₂ and O₃, resulting in sulfate production rates that scale with the intensity of solar ultraviolet (UV) radiation (9, 10). During NCP haze days, UV radiation is significantly reduced because of the aerosol dimming effect, resulting in a decrease of most oxidant concentrations (5). Figure 1B shows that the midday O₃ values drop from ~22 parts per billion (ppb) under clean conditions to ~1 ppb during the haze period (and also lose their typical diurnal variation, fig. S1). The reduced oxidant levels and increased sulfate production suggest the existence of a missing sulfate production pathway. Even after considering the gas phase and cloud/fog chemistry, there is still a large gap between modeled and observed sulfate (Fig. 1C). Adding an apparent heterogeneous process with sulfate production rates that scale with aerosol surface area and RH can greatly improve model predictions (see sections M3 to M5) (7), but the chemical mechanism of the missing sulfate production pathway has not yet been identified.

RESULTS AND DISCUSSION

We find that reactive nitrogen chemistry in aerosol water can explain the missing source of sulfate in NCP winter haze. Aerosol water is a key component of atmospheric aerosols, which serves as a medium that enables aqueous-phase reactions (11–13). The aerosol water content (AWC) in Beijing, calculated using measurements of RH and aerosol composition and the ISORROPIA-II thermodynamic equilibrium model (see section M6) (14–16), is well correlated with the missing sulfate content, the difference between measured and modeled sulfate (Δ[SO₄²⁻]) (Fig. 1C) (see sections M2 to M4), suggesting its involvement in the sulfate production. Note that because of the salt-induced freezing point depression (17), aerosol water will not freeze for a winter temperature of ~271 K in Beijing.

Taking the impact of mass transfer and ion strength into account, we make a conservative estimation of sulfate production rate for different reactions in the aerosol water under relevant atmospheric trace species concentration conditions (see sections M4 and M7 to M9) and find NO₂ to be the most important oxidant in Beijing during haze periods (Fig. 2B). In the presence of aerosol water, gas-phase NO₂ can partition into the condensed phase, react with SO₂ dissolved in the aqueous phase, and produce sulfate as well as nitrite (R1) (18).

$$2{\text{NO}}_2(\text{aq})+{{\text{HSO}}_3}^{-}(\text{aq})+{\mathrm{H}}_2\mathrm{O}(\text{aq})\to 3{\mathrm{H}}^{+}(\text{aq})+2{{\text{NO}}_2}^{-}(\text{aq})+{{\text{SO}}_4}^{2-}(\text{aq})$$ (R1)

Fig. 2. Aqueous-phase sulfate production by sulfur dioxide oxidation under characteristic conditions.

Sulfate production rates for (A) cloud droplets and (B) Beijing haze plotted against pH value.

Light blue– and gray-shaded areas indicate characteristic pH ranges for cloud water under clean to moderately polluted conditions and aerosol water during severe haze episodes in Beijing, respectively. The colored lines represent sulfate production rates calculated for different aqueous-phase reaction pathways with oxidants: hydrogen peroxide (H₂O₂), ozone (O₃), transition metal ions (TMIs), and nitrogen dioxide (NO₂). Characteristic reactant concentrations and model calculations for clouds and haze are taken from the literature and observations, and specified in Materials and Methods (see sections M7 to M9) (12, 21).

Under heavy haze conditions (PM_2.5 ≥ 300 μg m⁻³), the sulfate production rates of the NO₂ reaction pathway (R1) are ~1 to 7 μg m⁻³ h⁻¹, much higher than the reaction rates involving other important aqueous oxidants such as O₃ and H₂O₂. According to Zheng et al. (7), an additional sulfate production of ~3 μg m⁻³ h⁻¹ is needed to explain the observations in the severe winter haze periods of Beijing (see section M9), which falls right into the range of production rates from the chemical reaction mechanism we proposed (R1). As illustrated in Fig. 2, sulfate production in aerosol water under haze conditions differs from that in cloud droplets, where the major oxidation pathways are reactions with H₂O₂ and O₃, and NO₂ plays only a minor role (12, 19). Thus, traditional air quality models usually include only the H₂O₂ and O₃ reaction pathways of sulfate production in the aqueous phase, whereas the NO₂ reaction pathway is neglected (7, 20).

The AWC is typically three to five orders of magnitude lower than the water content of cloud or fog (21). On this basis, how can the NO₂ reaction pathway become important in such tiny amounts of water? The increased importance is due to the relatively high aerosol pH and elevated NO₂ concentrations in Beijing and the NCP during haze periods. As shown in Fig. 2, aqueous oxidation rates of S(IV) by NO₂ and O₃ are strongly pH-dependent. High pH increases the solubility and the effective Henry’s law constant of SO₂, pulling more SO₂ into the aerosol water and thus increasing the reaction rate. When pH increases by one unit, the reaction rates increase by one and two orders of magnitude for NO₂ and O₃, respectively. The H₂O₂ reaction does not show a strong dependence because high pH reduces its reaction rate coefficient, which offsets the effect of increased solubility. Compared with North America and Europe, aerosols in the NCP are more neutralized (22), as shown by a higher cation-to-anion ratio (Fig. 1D). This neutralized feature is also well documented and is the reason that acid rain rarely occurs in northern China (see section M10) (22). Using the ISORROPIA-II thermodynamic equilibrium model (see section M6) (14–16) and in situ aerosol bulk composition measurements, we obtain average pH values of 5.4 to 6.2 for aerosol water under NCP haze conditions (see sections M6 and M9). Similar calculations based on size-segregated aerosol composition measurements even show a higher effective pH and sulfate production rates (fig. S2).

Elevated NO₂ is another key factor that leads to fast sulfate formation. Substantial amounts of NO₂ come from direct emission of NO_x (= NO + NO₂). Although the NO₂-to-NO_x ratios were reduced because of weak photochemistry during the haze event, the stagnant weather trapped more NO₂ near the surface, resulting in elevated NO₂ concentrations that were, on average, three times higher than those under clean conditions (~66 ppb, Fig. 1B). These periods of highest NO₂ levels occurred when the concentrations of other photochemical oxidants that can produce sulfate (H₂O₂, O₃, and OH) were low (Fig. 1B and fig. S1). Changes in pH and precursor concentrations together lead to the transition from an H₂O₂-dominated regime of aqueous sulfate production in cloud droplets to an NO₂-dominated regime of aqueous sulfate production in haze (Fig. 2). Earlier studies had already suggested that the NO₂ reaction pathway may contribute to sulfate formation in fogs under polluted conditions (23, 24), but during the Beijing haze event of January 2013, fog was not observed and sulfate production by NO₂ occurred in aerosol water instead.

Sulfate production rates from the NO₂ reaction pathway (R1) calculated on the basis of measurement data (hourly concentrations of NO₂, SO₂, PM_2.5, and RH; sections M2 and M4) show a positive dependence on the PM_2.5 concentration, varying from 0.01 μg m⁻³ h⁻¹ under relatively clean conditions to nearly 10 μg m⁻³ h⁻¹ in the most polluted periods (pink circles in Fig. 3). These reaction rates can account for the systematic underprediction of models and explain the large missing source of sulfate in the Beijing haze (black diamonds in Fig. 3). Under clean conditions, the OH reaction (green crosses in Fig. 3) dominates the oxidation pathways of SO₂. As particle concentrations increase and more sulfate is produced, the photochemistry slows down, leading to less OH, and sulfate production via this pathway decreases. From this aspect, the OH reaction has a negative feedback, which is self-buffered against heavy pollution. As PM_2.5 concentrations and RH increase simultaneously [a special feature of haze events in NCP (5)], the OH reaction becomes weaker, whereas the aqueous-phase reaction of NO₂ starts to play a more important role. For the January 2013 conditions at PM_2.5 ~100 to 200 μg m⁻³, the aqueous-phase reaction becomes the dominant oxidation pathway (Fig. 3). In contrast to the OH reaction, the NO₂ reaction shows a positive feedback mechanism, where higher particle matter levels lead to more aerosol water, which accelerates sulfate production and further increases the aerosol concentration. This positive feedback intensifies the PM_2.5 levels during haze periods, resulting in a series of record-breaking pollution events.

Fig. 3. Importance of the NO₂ reaction pathway for sulfate production in the Beijing haze (January 2013).

Sulfate production rates calculated for the aqueous-phase NO₂ reaction pathway (pH 5.8) and the gas-phase OH reaction pathway compared to the missing source of sulfate. Crosses and circles represent hourly production rates calculated on the basis of measurement data, and the diamonds represent the average missing source (arithmetic mean ± SD) (see sections M3 to M5) (7). The pink-shaded area shows the maximum and minimum sulfate production rates by the NO₂ reaction pathway bounded by the aerosol water pH ranging from 5.4 to 6.2 during haze periods.

The NO₂ reaction with SO₂ in aerosol water produces not only sulfate but also nitrite (R1), which may undergo subsequent oxidation or disproportionation reactions forming nitrate. This is consistent with the high nitrate concentrations observed during the haze event, which are of similar magnitude to the sulfate concentrations (up to ~160 μg m⁻³) (5) and have also not yet been explained by air quality models (7). Depending on the pH value of the haze droplets, nitrite can also form nitrous acid (HONO) and undergo reversible partitioning with the gas phase (25, 26). Moreover, the release of HONO removes H⁺, which may help to sustain the droplet acidity and efficient sulfate production (27, 28). For the same haze periods, record high aerosol nitrite concentrations of up to ~12 μg m⁻³ were observed in Shandong, northern China, with nitrite concentrations well correlated with NO_x at high RH (>50%) (25). The nitrite-to-HONO molar ratio of ~3 in Shandong is a hundred times higher than the ratio observed during a pollution event in Nanjing, Yangtze River Delta of China, where only trace amounts of nitrite were detected in the aerosol phase (up to ~0.4 μg m⁻³), and the aerosol water there was more acidic (pH ~4) (24, 25). This further confirms the more neutralized feature of aerosol particle composition and haze in northern China.

Our study unfolds a new and more comprehensive conceptual model of sulfate formation in NCP haze events, including not only the traditional OH, H₂O₂, and O₃ reaction pathways in atmospheric gas phase and cloud chemistry but also the NO₂ reaction pathway in aerosol water (Fig. 4). The NO₂ pathway may not be limited to winter haze because it may also be important at night and during fog events in polluted regions with high boundary layer concentrations of PM_2.5 and NO₂ and elevated RH (23, 24). The importance of multiphase chemistry holds for a wide range of aerosol pH. When aerosols become more acidic, the sulfate production can be maintained at a high rate through TMI reactions (Fig. 2B). The important role of aerosol pH in the multiphase reaction pathway highlights the need to better understand the sources of ammonia and alkaline aerosol components from natural and anthropogenic emissions (soil dust, seawater, agriculture, energy, industrial, and traffic). Furthermore, reactive nitrogen chemistry in aerosol water might also play a role in nitrate and secondary organic aerosol production during haze days when photochemistry is reduced.

Fig. 4. Conceptual model of sulfate formation in haze events in NCP.

The traditional OH, H₂O₂, and O₃ reaction pathways in atmospheric gas phase and cloud chemistry are included here, as well as the NO₂ reaction pathway in aerosol water with elevated pH and NO₂ concentrations proposed in this study.

Our results reveal the complex nature of haze pollution events in China, where NO_x is not only a precursor for nitrate but also an important oxidant for sulfate formation. Thus, reductions of NO_x emissions are expected to reduce nitrate, sulfate, and PM_2.5 much more than anticipated by traditional air quality models. A large decrease in PM_2.5 has already been observed in relation to traffic and energy control measures during the Beijing Olympic Games in 2008 and other events in the NCP. Heavy haze conditions with high pollutant concentration levels and largely neutralized aerosol water are key features of atmospheric chemistry in the NCP. These features will need to be considered in future air quality and pollutant emission control strategies in northern China, and perhaps also in other regions.

MATERIALS AND METHODS

M1. The January 2013 haze in Beijing

The severe haze episode in January 2013 is one of the worst atmospheric pollution events ever recorded in China (2, 9). In Beijing, the daily fine-particle (PM_2.5) concentration reached up to 400 μg m⁻³, exceeding the World Health Organization guideline value by 16 times. The weak East Asian winter monsoon, which resulted in weakened surface winds and the anomalous southerly winds, was responsible for the haze events (4). The southerly winds transported more water vapor from the sea to eastern China. The anomalous high-pressure system at 500 hPa suppressed convection. Thus, the air in January was more stagnant, trapping more air pollutants and water vapor near the surface. The high PM_2.5 concentration reduced the solar radiation and atmospheric photochemistry, resulting in decreases in the concentration of photochemical products such as OH and O₃ (fig. S1).

A major feature of PM_2.5 pollution during this haze event was the large contribution from secondary species, including inorganic (mainly sulfate, nitrate, and ammonium) and organic species (2, 5). However, contribution from secondary inorganic species (such as sulfate and nitrate) showed an increasing trend with increasing pollution levels, whereas contribution from organics decreased (5).

The fast production of sulfate, however, cannot be reproduced by model simulations, which have implemented aerosol-meteorology-radiation feedback and a state-of-the-art chemical mechanism (6, 7), that is, the gas-phase oxidation by OH (29, 30) and aqueous-phase oxidation in clouds by H₂O₂ and O₃ (12, 21). Stabilized Criegee intermediates (sCIs) were also suggested as an oxidant of SO₂ but contribute minor H₂SO₄ production compared with the conventional OH reaction in the midday (31–33). Further analysis has shown that the model simulation can be improved by introducing an apparent heterogeneous process with a reaction rate coefficient scaled with RH (see section M3) (7).

M2. Sampling location and experimental methods

We performed aerosol measurements from 1 to 31 January 2013 on the roof of the Environmental Science Building (40°00′17″N, 116°19′34″E, ~10 m above the ground) on the campus of Tsinghua University, an urban background site in Beijing. Table S2 summarizes the aerosol-related parameters and experimental methods. Concentrations of PM_2.5 and PM₁₀ were measured by an online PM-712 monitor (Kimoto Electric Co. Ltd.) equipped with a U.S. Environmental Protection Agency PM₁₀ inlet and a PM_2.5 virtual impactor (34). SO₄²⁻ and other ions in PM_2.5 were measured by an online ACSA-08 monitor (Kimoto Electric Co. Ltd.) and the filter-based analysis. The PM_2.5 filter samples were collected from 12 to 24 January by median-volume samplers (Laoying) on prebaked Quartz filters (2500 QAT-UP; Pall Corporation) with a flow rate of 100 liters min⁻¹ (35). The filter samples were analyzed by the Dionex ion chromatograph (DX-600 for cations and ICS-2000 for anions) (Dionex Corporation) for the concentration of water-soluble inorganic ions, including Na⁺, K⁺, Ca²⁺, Mg²⁺, NH₄⁺, SO₄²⁻, NO₃⁻, and Cl⁻. Organic carbon (OC) and elementary carbon concentrations in PM_2.5 were measured by a Sunset Model 4 semicontinuous carbon analyzer (Beaverton) with a National Institute for Occupational Safety and Health–type temperature protocol (36). A factor of 1.6 was adopted to convert the mass of OC into the mass of organics (37, 38). Gaseous air pollutants SO₂, NO₂, and O₃ were measured by the Atmospheric Environment Monitoring Network (39). The meteorology data were measured by the Milos 520 Weather Station (VAISALA Inc.). More details can be found in the work of Zheng et al. (5).

M3. WRF-CMAQ model simulation

The Weather Research and Forecasting—Community Multiscale Air Quality (WRF-CMAQ) model system was used to determine the missing source of sulfate through comparison with observational data. WRF is a new-generation mesoscale numerical weather prediction system designed to serve a wide range of meteorological applications (www.wrf-model.org/), and CMAQ is a three-dimensional Eulerian atmospheric chemistry and transport modeling system that simulates multipollutants (www.cmascenter.org/cmaq/). The releases of WRF v3.5.1 and CMAQ v5.0.1 were used in this study.

The model simulation was configured the same as in the work of Zheng et al. (7), as detailed in table S3. To better characterize the stagnant meteorological conditions, we applied observational nudging for temperature (T) and RH [above the planetary boundary layer (PBL)], and wind (within and above the PBL). The surface roughness is corrected according to Mass and Ovens (40) by increasing the friction velocity by 1.5 times in the PBL scheme, which significantly reduced the high biases in wind and RH simulations. In general, the simulated T, RH, and wind at the ground surface agree with observations (7). For the gas-phase reactions, we used the CB05 mechanism with active chlorine chemistry and the updated toluene mechanism of Whitten et al. (41). For the aqueous-phase reactions in clouds, we used the updated mechanism of the RADM model (20, 42). Reactions relevant for the sulfate formation were discussed in detail in section M4.

Here, the WRF-CMAQ modeling results were used in the following analysis: (i) the modeled sulfate concentration [SO₄²⁻] was used to calculate Δ[SO₄²⁻], the difference between observed and modeled [SO₄²⁻], and (ii) the modeled OH and H₂O₂ concentrations were used in the estimation of sulfate formation from the gas-phase reaction of OH with SO₂ and the aqueous-phase reaction of H₂O₂ with HSO₃⁻.

M4. Production of sulfate in WRF-CMAQ model

According to current understanding, secondary sulfate is produced through the oxidation of SO₂. In the WRF-CMAQ model, oxidation occurs both in the gas phase and in the cloud/fog droplets. In the gas phase, the major pathway is the OH-initiated reaction (12, 43)

$$\text{OH}+{\text{SO}}_2+\mathrm{M}\to {\text{HO}}_2+ \text{Sulfate}$$ (M1)

in which the second-order kinetic constant can be expressed as

$$\mathit{k}(\mathit{T})=\left\{\frac{{\mathit{k}}_0(\mathit{T})[\mathit{M}]}{1+\frac{{\mathit{k}}_0(\mathit{T})[\mathit{M}]}{{\mathit{k}}_{\infty }(\mathit{T})}}\right\}{0.6}^{\mathit{Z}} \text{and} \mathit{Z}={\{1+{\left[{\text{log}}_{10}\right(\frac{{\mathit{k}}_0(\mathit{T})[\mathit{M}]}{{\mathit{k}}_{\infty }(\mathit{T})}\left)\right]}^2\}}^{-1}$$ (M2)

where [M] is the concentration of N₂ and O₂, and k₀(T) and k_∞(T) represent the low- and high-pressure limiting rate constants, respectively. Their temperature dependence can be expressed as

$${\mathit{k}}_0(\mathit{T})={\mathit{k}}_0^{300}{\left(\frac{\mathit{T}}{300}\right)}^{-\mathit{n}} \text{and} {\mathit{k}}_{\infty }(\mathit{T})={\mathit{k}}_{\infty }^{300}{\left(\frac{\mathit{T}}{300}\right)}^{-\mathit{m}}$$ (M3)

where k₀³⁰⁰ = 3.3 × 10⁻³¹ cm⁶ molecule⁻² s⁻¹ and n = 4.3, and k_∞³⁰⁰ = 1.6 × 10⁻¹² cm³ molecule⁻¹ s⁻¹ and m = 0.

In cloud/fog droplets, the following aqueous reactions have been included in the WRF-CMAQ model (7). The expression of their reaction rate, as well as the rate coefficients, is summarized in table S4.

$${{\text{HSO}}_3}^{-}+{\mathrm{H}}_2{\mathrm{O}}_2\to {{\text{SO}}_4}^{2-}+{\mathrm{H}}^{+}+{\mathrm{H}}_2\mathrm{O}$$ (M4)

$${\text{SO}}_2+{\mathrm{O}}_3+{\mathrm{H}}_2\mathrm{O}\to {{\text{SO}}_4}^{2-}+2{\mathrm{H}}^{+}+{\mathrm{O}}_2$$ (M5)

$${{\text{HSO}}_3}^{-}+{\mathrm{O}}_3\to {{\text{SO}}_4}^{2-}+{\mathrm{H}}^{+}+{\mathrm{O}}_2$$ (M6)

$${{\text{SO}}_3}^{2-}+{\mathrm{O}}_3\to {{\text{SO}}_4}^{2-}+{\mathrm{O}}_2$$ (M7)

$${\text{SO}}_2+{\mathrm{H}}_2\mathrm{O}+0.5{\mathrm{O}}_2+ \text{Fe}(\text{III})/\text{Mn}(\text{II})\to {{\text{SO}}_4}^{2-}+2{\mathrm{H}}^{+}$$ (M8)

$${{\text{HSO}}_3}^{-}+{\text{CH}}_3\text{OOH}\to {{\text{SO}}_4}^{2-}+{\mathrm{H}}^{+}+{\text{CH}}_3\text{OH}$$ (M9)

$${{\text{HSO}}_3}^{-}+{\text{CH}}_3\text{COOOH} \to {{\text{SO}}_4}^{2-}+{\mathrm{H}}^{+}+{\text{CH}}_3\text{COOH}$$ (M10)

M5. Apparent heterogeneous uptake of SO₂ on aerosol surfaces

To improve the model simulation for the winter haze events in Beijing and the NCP, Zheng et al. (7) suggested the use of an apparent heterogeneous uptake coefficient (γ) of SO₂ on aerosol surfaces as a function of RH (Eq. M11). γ is defined as the ratio of the number of collisions that result in the reaction SO₂ (g) + Aerosol → SO₄²⁻ to the theoretical total number of collisions. The overall rates of SO₂ uptake and sulfate production are as fast as if the whole aerosol surface is covered by dust (44). Although the exact mechanism supporting this fast production rate is still unknown (7), this apparent heterogeneous source of sulfate can account for the underprediction of modeled sulfate, with significant reduction of normalized model bias from −54.2 to 6.3%.

$$\mathrm{\gamma }=\{\begin{array}{c}{\mathrm{\gamma }}_{\text{low}}, 0\%\le \text{RH} \le 50\%\hfill \\ {\mathrm{\gamma }}_{\text{low}}+({\mathrm{\gamma }}_{\text{high}}-{\mathrm{\gamma }}_{\text{low}})/(1-0.5)\times (\text{RH} -50\%),50\%< \text{RH} \le 100\%\hfill \end{array}$$ (M11)

where γ_low = 2 × 10⁻⁵ and γ_high = 5 × 10⁻⁵.

With γ, a sulfate production rate R_H,g can be determined by Eq. M12:

$${\mathit{R}}_{\mathrm{H},\mathrm{g}}=\frac{\mathit{d}[{{\text{SO}}_4}^{2-}]}{\mathit{d}\mathit{t}}={[\frac{{\mathit{R}}_{\mathrm{p}}}{{\mathit{D}}_{\mathrm{g}}}+\frac{4}{\mathrm{\gamma }\mathrm{\nu }}]}^{-1}{\mathit{S}}_{\text{aerosol}}[{\text{SO}}_2]$$ (M12)

where R_p is the radius of aerosol particles, D_g is the gas-phase molecular diffusion coefficient of SO₂, ν is the mean speed of gaseous SO₂ molecules, and S_aerosol is the surface area concentration of aerosol particles.

M6. ISORROPIA-II model calculation

The ISORROPIA-II model (15) was used to calculate the AWC and pH. The ISORROPIA-II is a thermodynamic equilibrium model that predicts the physical state and composition of atmospheric inorganic aerosols. It can be used in two modes: the reverse mode and the forward mode. The reverse mode calculated the thermodynamic equilibrium based on aerosol-phase concentrations, whereas the forward mode relied on both aerosol-phase and gas-phase concentrations (16). Its ability in predicting AWC and pH has been demonstrated by Guo et al. (45) and Xu et al. (16).

To evaluate the aerosol pH and AWC, we performed both reverse-mode and forward-mode model simulations and used their averages for further analyses. The gaseous NH₃ was not measured in our January campaign, but long-term measurement (46) shows a compact correlation between NH₃ and NO_x concentrations in the winter season of Beijing. Accordingly, we estimated the NH₃ concentration from the observed NO_x concentration with an empirical equation derived from Meng et al. (46), that is, NH₃ (ppb) = 0.34 × NO_x (ppb) + 0.63. The contribution of organic compounds to AWC, W_org (the mass concentration of aerosol water associated with organics), was estimated by the same approach of Guo et al. (45)

$${\mathit{W}}_{\text{org}}=\frac{\text{OM}}{{\mathrm{\rho }}_{\text{org}}}\cdot {\mathrm{\rho }}_{\mathrm{w}}\cdot \frac{{\mathrm{\kappa }}_{\text{org}}}{(100\%/\text{RH} -1)}$$ (M13)

where OM is the mass concentration of organic matter, ρ_w is the density of water (ρ_w = 1.0 × 10³ kg m⁻³), ρ_org is the density of organics (ρ_org = 1.4 × 10³ kg m⁻³) (45), and κ_org is the hygroscopicity parameter (47) of organic aerosol compositions. We adopted a κ_org of 0.06 based on previous cloud condensation nuclei measurements in Beijing (48).

M7. Kinetics of mass transport

For multiphase reactions, the overall reaction rate depends not only on the rate of chemical reactions but also on the mass transport in different medium and across the interface. To account for the effects of mass transport, we adopted the formulation of a standard resistance model (12)

$$\frac{1}{{\mathit{R}}_{\mathrm{H},\text{aq}}}=\frac{1}{{\mathit{R}}_{\text{aq}}}+\frac{1}{{\mathit{J}}_{\text{aq},\text{lim}}}$$ (M14)

where R_H,aq is the sulfate production rate, R_aq is the aqueous-phase reaction rate, and J_aq,lim is the limiting mass transfer rate. For the oxidation of S(IV) by a given oxidant O_xi (12)

$${\mathit{R}}_{\text{aq}}=({\mathit{k}}_0[\mathrm{S}{\mathrm{O}}_2\cdot {\mathrm{H}}_2\mathrm{O}]+{\mathit{k}}_1[{{\text{HSO}}_3}^{-}]+{\mathit{k}}_2[{{\text{SO}}_3}^{2-}])[{\mathrm{O}}_{\text{xi}}]$$ (M15)

where [SO₂·H₂O], [HSO₃⁻], [SO₃²⁻], and [O_xi] are the respective aqueous-phase concentrations, and k₀, k₁, and k₂ are the corresponding second-order reaction rate coefficients as detailed in table S4. The aqueous [O_xi] is assumed to be in equilibrium with its gas-phase concentration and can be determined by Henry’s law (12)

$$[\mathit{X}]={\mathit{p}}_{\infty }(\mathit{X})\cdot \mathit{H}*(\mathit{X})$$ (M16)

where p_∞(X) (atm) is the partial pressure of species X in the bulk gas phase and H^* (M atm⁻¹) is the effective Henry’s constant (table S5).

The limiting mass transfer rate J_aq (M s⁻¹) is calculated by Eqs. M17 and M18 (12)

$${\mathit{J}}_{\text{aq},\mathit{l}\mathit{i}\mathit{m}}= \text{min}\{{\mathit{J}}_{\text{aq}}({\text{SO}}_2),{\mathit{J}}_{\text{aq}}({\mathrm{O}}_{\text{xi}})\}$$ (M17)

$${\mathit{J}}_{\text{aq}}(\mathit{X})={\mathit{k}}_{\text{MT}}(\mathit{X})\cdot {\mathit{p}}_{\infty }(\mathit{X})\cdot \mathit{H}*(\mathit{X})$$ (M18)

where X refers to SO₂ or the oxidant O_xi such as O₃, H₂O₂, and NO₂.

The mass transfer rate coefficient k_MT (s⁻¹) can be calculated by (12)

$${\mathit{k}}_{\text{MT}}(\mathit{X})={[\frac{{{\mathit{R}}_{\mathrm{p}}}^2}{3{\mathit{D}}_{\mathrm{g}}}+\frac{4{\mathit{R}}_{\mathrm{p}}}{3\mathrm{\alpha }\mathrm{\nu }}]}^{-1}$$ (M19)

where R_p is the aerosol radius, and $\frac{{{\mathit{R}}_{\mathrm{p}}}^2}{3{\mathit{D}}_{\mathrm{g}}}$ and $\frac{4{\mathit{R}}_{\mathrm{p}}}{3\mathrm{\alpha }\mathrm{\nu }}$ are the continuum regime resistance and the free-molecular (or kinetic) regime resistance, respectively. D_g is the gas-phase molecular diffusion coefficient, and ν is the mean molecular speed of X. α is the mass accommodation coefficient of X on the droplet surface, which accounts for imperfect sticking of impinging molecules to the surface, and we adopted literature values of 0.11, 0.23, 2.0 × 10⁻³ (12), and 2.0 × 10⁻⁴ (49) for SO₂, H₂O₂, O₃, and NO₂, respectively. Aqueous-phase mass transfer can be ignored for the size range considered here (D_p ≤ 2.5 μm) (12). An equivalent R_p of 0.15 and 15 μm was assumed for aerosols and cloud droplets, respectively.

M8. Influences of ionic strength on aqueous sulfate-producing reactions

Aerosol liquid water constitutes an aqueous electrolyte that can be extremely concentrated with high ionic strength (I) up to 100 M (50). This highly concentrated chemical environment will affect the inclination of a species to participate in the aqueous-phase chemical reactions, reflected by an apparent reaction rate constant (k) different from that in an ideal solution (21).

The influence of I on k is complicated and is not yet fully understood. According to current understanding, I affects k through its integrated effects on the activity coefficient (a) of reactants and products for most reactions (51). For example, for a reaction where reactants A and B form an activated complex (AB)*, which then quickly decomposes into a final product P (A + B → (AB)* → P), the k-I dependence can be described as (51)

$$\text{log}\frac{\mathit{k}}{{\mathit{k}}^{\mathit{I}=0}}= \text{log} {\mathit{a}}_{\mathrm{A}}+ \text{log} {\mathit{a}}_{\mathrm{B}}- \text{log}{\mathit{a}}_{(\text{AB})}$$ (M20)

where k^I=0 refers to the kinetic constant at I of 0 M. Theories used to predict a_i at given I varied with the ranges of I and the species nature, that is, being an ion or a neutral species, as summarized in table S6.

Major aqueous sulfate-producing reactions considered here include S(IV) oxidation by H₂O₂, O₃, TMI + O₂, and NO₂ (Fig. 2). Among these reactions, the influence of I has been studied experimentally, with I ranging up to ~5 M for H₂O₂ and slightly higher than 1 M for O₃ and TMI + O₂. The observed k-I relationships of these reactions agree with theoretical predictions summarized in table S7 (52–56). As shown in fig. S3, with increasing ionic strength, the rate constant for H₂O₂ decreases first when I is below ~1 M and begins to increase once I goes above ~1 M (52, 57). For O₃, the rate constant is positively related to I. For TMI + O₂ [here, only Fe(III) and Mn(II) are considered as effective catalyzing TMIs (58, 59)], the rate constant decreases significantly with increasing I, even without considering the sulfate inhabitation effect (referring to the effect that the formation of the sulfate-TMI complex would reduce catalytically active TMI concentrations) (55, 60).

Currently, however, no k-I relationship was reported for NO₂, whereas some plausible estimation can be made on the basis of the principle of theories discussed above. Two kinds of mechanisms have been suggested for the NO₂-S(IV) reactions. One is the oxygen-atom transfer reaction (61)

$$2{\text{NO}}_2+{\text{SO}}_3^{2-}\leftrightarrow {({\mathrm{O}}_2\mathrm{N}-{\text{SO}}_3-{\text{NO}}_2)}^{2-}$$ (M21a)

$${({\mathrm{O}}_2\mathrm{N}-{\text{SO}}_3-{\text{NO}}_2)}^{2-}+{\text{OH}}^{-}\leftrightarrow {(\text{HO}-{\text{SO}}_3-{({\text{NO}}_2)}_2)}^{3-}$$ (M21b)

$${(\text{HO}-{\text{SO}}_3-{({\text{NO}}_2)}_2)}^{3-}+{\text{OH}}^{-}\to {{2\text{NO}}_2}^{-}+{\text{SO}}_4^{2-}+{\mathrm{H}}_2\mathrm{O}$$ (M21c)

in which the first reaction (Eq. M21a) is the rate-controlling step.

The other mechanism is an electron transfer reaction (62–64)

$${\text{NO}}_2+{\text{SO}}_3^{2-}\to {{\text{NO}}_2}^{-}+{{\text{SO}}_3}^{•-}$$ (M22)

followed by the sulfur auto-oxidation processes (12). The first reaction (Eq. M22) is the rate-controlling step because follow-up reactions with radicals are very fast.

For both mechanisms, the rate-controlling step of the NO₂-S(IV) reaction is a reaction of an ion with a neutral molecule. According to Herrmann (51), the k-I relation for this type of reaction should follow

$$\text{log}\frac{\mathit{k}}{{\mathit{k}}^{\mathit{I}=0}}=\mathit{b}\mathit{I}$$ (M23)

where b is the kinetic salting coefficient. According to Kameoka and Pigford (65), b is expected to be positive. Given a positive b, increasing I will lead to an increase in k, as shown by the red dashed curve in fig. S3 (we take an arbitrary value of 0.5 M⁻¹ for b, which is not determined from experiments and is used only for the purpose of illustration).

During the severe Beijing hazes (when PM_2.5 is higher than 300 μg m⁻³), ionic strength in aerosol liquid water ranged from 13 to 43 M, as predicted by the ISORROPIA-II model (see section M6) (15). Direct extrapolation of the observed/predicted k-I relationship (fig. S3) into such high ranges of ionic strength may not be appropriate. Thus, the rate constants are taken as for diluted solution, although based on current observations and theories (table S7 and fig. S3), this treatment will lead to conservative estimations of the real sulfate production rates of the NO₂, O₃, and H₂O₂ pathways in Fig. 2 but an overestimation of the TMI + O₂ pathway.

M9. Data used in Fig. 2

Detailed descriptions on how to derive the sulfate production rate for different aqueous-phase oxidation pathways of SO₂ (that is, by H₂O₂, O₃, TMI + O₂, and NO₂) in Fig. 2 can be found in the study of Seinfeld and Pandis (12). The aqueous-phase reactions involved are listed as Reactions R1 and M4 to M8. The rate expressions and rate coefficients and the constants that we used for calculating the apparent Henry’s constant are summarized in tables S4 and S5, respectively.

In Fig. 2, the “cloud droplets” scenario is taken from the work of Seinfeld and Pandis (12) and Herrmann et al. (21): [SO₂(g)] = 5 ppb, [NO₂(g)] = 1 ppb, [H₂O₂(g)] = 1 ppb, [O₃(g)] = 50 ppb, [Fe(III)] = 0.3 μM, [Mn(II)] = 0.03 μM, liquid water content = 0.1 g m⁻³, and cloud droplet radius R_p = 15 μm. In this scenario, the temperature T is taken to be the same as that used in the “Beijing haze” scenario described below.

The “Beijing haze” scenario was taken according to the measurement data during the most polluted haze periods (PM_2.5 > 300 μg m⁻³). The average values were used in our calculation: [SO₂(g)] = 40 ppb, [NO₂(g)] = 66 ppb, [H₂O₂(g)] = 0.01 ppb, [O₃(g)] = 1 ppb, AWC = 300 μg m⁻³, aerosol droplet radius R_p = 0.15 μm, and T = 271 K. The concentrations of [Fe(III)] and [Mn(II)] are pH-dependent (fig. S4). The pH dependence is mostly due to the precipitation equilibrium of Fe(OH)₃ and Mn(OH)₂

$$[\text{Fe}(\text{III})]=\frac{{\mathit{K}}_{{\text{sp}, \text{Fe}(\text{OH})}_3}}{{[{\text{OH}}^{-}]}^3} \text{and} {[\text{Mn}(\text{II})]}_{\text{sat}}=\frac{{\mathit{K}}_{{\text{sp}, \text{Mn}(\text{OH})}_2}}{{[{\text{OH}}^{-}]}^2}$$ (M24)

where ${\mathit{K}}_{{\text{sp}, \text{Fe}(\text{OH})}_3}$ and ${\mathit{K}}_{{\text{sp}, \text{Mn}(\text{OH})}_2}$ are the precipitation constants of Fe(OH)₃ and Mn(OH)₂, respectively (66). When all Fe(OH)₃ and Mn(OH)₂ are dissolved, further decrease of pH will not increase [Fe(III)] and [Mn(II)], resulting in a plateau at low pH (fig. S4). The total soluble Fe and Mn are estimated to be 18 and 42 ng m⁻³, respectively, based on data in the literature and observations in Beijing (2, 59, 66–70).

The severe haze events observed in Beijing are a regional phenomenon (5). Because Beijing is located in the northwestern edge of the polluted area, the air pollution in cities south of Beijing is even more severe. NO₂ concentrations in Beijing are comparable to the southern cities, whereas SO₂ concentrations in the latter are typically two to four times higher than those in Beijing (fig. S5). Thus, the sulfate production rate from the NO₂ pathway that we predicted with the Beijing data in Fig. 2 is a conservative estimation for the contribution of this pathway to the sulfate formation in the whole NCP.

Considering that regional transport from cities south of Beijing has contributed to the severe haze episode in January 2013 (5, 71) and judging from the air pollution trend shown in fig. S5, the polluted air parcels could typically have been processed under severe haze conditions in Beijing and in cities south of it for more than 3 days before the peak pollution hour in Beijing. Because cloud amounts were low (7, 72, 73) during the January 2013 winter haze periods, we assumed that aerosols spent 100% of their lifetime under noncloud conditions (RH < 100%) (21). Under these conditions, to produce the observed sulfate concentration (~200 μg m⁻³), the required average heterogeneous production rate can be determined to be ~3 μg m⁻³ h⁻¹. This estimation also falls into the ranges of 1 to 7 μg m⁻³ h⁻¹ that we predicted for the NO₂ pathway in Fig. 2.

M10. Aerosol acidity in northern China

The neutralized feature of aerosols in East Asia (including Beijing and the NCP) has been well documented in the literature, with high cation-to-anion ratios (45, 74–86). Aircraft measurements in the Arctic further support this conclusion, showing that the aerosols in the Arctic characterized as coming from the outflow of East Asia were mostly neutralized, whereas the aerosols transported from North America to the Arctic were highly acidic (lines in Fig. 1D) (87).

The low aerosol acidity in East Asia can be attributed to its high NH₃ and mineral dust emissions. For example, in 2008, the emission ratios of NH₃ to 2× SO₂ (molar ratio of NH₃ emission divided by twice the SO₂ emission) are 0.37, 0.86, and 1.04 for North America, Europe, and East Asia, respectively (87). Northern China is expected to have a higher atmospheric neutralizing capacity than other parts of East Asia because of the high NH₃ emission in this region. Both satellite data (88, 89) and emission inventories (90) show that northern China is one of the most NH₃-rich areas in East Asia because of its intense agriculture activities. The NCP (that is, the cities of Beijing and Tianjin and the provinces of Hebei, Henan, and Shandong) accounts for 30 to 40% of the total NH₃ emissions in China (90) while contributing ~20% of the emissions for SO₂ and NO_x (91).

The high mass fraction of mineral dust (10 to 20%) is a distinct characteristic of PM_2.5 in the NCP (2, 70, 92, 93). The influence of mineral aerosols is also higher for northern China than for southern China (94–98). Major sources of mineral aerosols include urban fugitive dust (resuspended road dust, construction dust, etc.) and the long-range transported Asian dust (99–101). The mineral dusts are observed to be internally mixed with sulfate, nitrate, and ammonia, suggesting their participation in atmospheric processing (102–105). Without mineral components (Ca²⁺, Mg²⁺, K⁺, and Na⁺), aerosol pH in northern China may drop below 5.6, showing an acidic nature (37).

In addition, despite the high emission of acidic gases (SO₂ and NO_x), rainwater in northern China has average pH values higher than 5.6 (fig. S6), suggesting an alkaline tendency in contrast to other areas (for example, the United States) (12, 22). A pH of 5.6 is often taken as the “natural” acidity of rainwater (water in equilibrium with CO₂), which has been considered as the demarcation line of acidic precipitation (12). The high pH values in rainwater of northern China are caused by alkaline aerosol particles (12), which have a large buffering capacity to offset the effects of anthropogenic acidity.

M11. Contribution from sCIs and NO₃ radicals

We have also investigated the reactions of S(IV) with NO₃ radicals and sCIs, both of which show minor contributions (0.03 and 0.69%) compared to our proposed mechanism S(IV) + NO₂(aq).

Oxidation by NO₃ radicals

An average NO₃ radical concentration of ~2.5 × 10⁻³ parts per trillion was determined by our modeling results for the severe pollution periods. Taking an effective Henry’s constant of 0.6 M atm⁻¹ for NO₃ and a reaction rate constant of 1.4 × 10⁹ M⁻¹ s⁻¹ (106), we determined a sulfate production rate of 5.1 × 10⁻⁴ μg m⁻³ h⁻¹ for the NO₃ reaction. This rate is only 0.03% compared to the proposed mechanism S(IV) + NO₂(aq) at pH 5.8 (Fig. 2) and is thus negligible.

Oxidation by sCI

According to Mauldin et al. (31), we calculated R_sCI, the sulfate formation rate from the sCI mechanism, by

$${\mathit{R}}_{\text{sCI}}={\mathit{k}}_{{\text{sCI}+\text{SO}}_2}[\text{sCI}][{\text{SO}}_2]$$ (M25)

where [sCI] and [SO₂] are the concentrations of sCI and SO₂, respectively, and the reaction rate coefficient ${\mathit{k}}_{\text{sCI}+{\text{SO}}_2}$ is 6 × 10⁻¹³ cm³ molecule⁻¹ s⁻¹. The concentration of sCI can be determined by the following equation (31)

$$[\text{sCI}]={\mathit{Y}}_{\text{sCI}}{\mathit{k}}_{{\mathrm{O}}_3+\text{alkene}}[{\mathrm{O}}_3][\text{alkene}]{\mathrm{\tau }}_{\text{sCI}}$$ (M26)

where Y_sCI is the yield of sCI ~0.5, ${\mathit{k}}_{{\mathrm{O}}_3+\text{alkene}}$ is the corresponding rate coefficient for the reaction of O₃ with individual alkenes, and τ_sCI is the lifetime of the sCI, which is 0.2 s (31). After taking a typical alkene profile for the haze period (107), we determined an R_sCI of ~0.69% of the proposed S(IV) + NO₂(aq) mechanism.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at hhttp://advances.sciencemag.org/cgi/content/full/2/12/e1601530/DC1

fig. S1. Weakened photochemistry by aerosol dimming effects during January 2013 in Beijing.

fig. S2. Importance of the NO₂ reaction pathway for sulfate production in the Beijing haze (January 2013).

fig. S3. Influence of ionic strength (I) on rate of aqueous sulfate-producing reactions.

fig. S4. Estimation of Fe³⁺ and Mn²⁺ concentrations as a function of aerosol water pH during Beijing hazes.

fig. S5. Regional pollution across the NCP during January 2013.

fig. S6. Annual precipitation pH of China in 2013.

fig. S7. The same as Fig. 2 but with a lower limit of reaction rate constants reported by Lee and Schwartz (18).

table S1. Previously reported concentrations of cations and anions in PM_2.5 during winter for cities in NCP used in Fig. 1D.

table S2. Summary of field observation and methods in this study.

table S3. Domain, configurations, and major dynamic and physical options used in WRF v3.5.1.

table S4. Rate expression and rate coefficients of relevant aqueous-phase reactions.

table S5. Constants for calculating the apparent Henry’s constant (H*).

table S6. Summary of suggested activity coefficient (a)–ionic strength (I) dependence.

table S7. Influence of ionic strength (I) on rate of aqueous sulfate-producing reactions.

References (108–128)