Source attribution of particulate matter pollution over North China with the adjoint method

We compare the model simulation of PM2.5 concentration and chemical composition to surface measurements by the China Meteorological Administration (CMA) for the winter month of January. A number of studies have analyzed the episodic haze events during January 2013 (He et al 2014, Wang et al 2014b, Zhang et al 2014a). We focus on the monthly averages representing the mean atmospheric condition to target emission control strategies for a long-term air quality improvement. Previous studies have shown that the haze pollution events in January 2013 over North China were characterized by large amounts of sulfate aerosol and models underestimated the observed sulfate enhancements likely due to missing mechanisms of heterogeneous chemistry (He et al 2014, Wang et al 2014b, Zheng et al 2015). The standard GEOS-Chem model simulates sulfate production from aqueous-phase oxidation of SO₂ by hydrogen peroxide (H₂O₂) and ozone (O₃) in clouds, and gas-phase oxidization of SO₂ by OH (Park et al 2004). By adding heterogeneous uptake of SO₂ on deliquesced aerosols under high-relative humidity conditions, Wang et al (2014b) partly improved the model sulfate simulation. In this study we examine the mechanism of aqueous-phase oxidation of S(IV) (the sum of dissolved SO₂, HSO₃⁻, and SO₃²⁻) by dissolved nitrogen dioxide (NO₂) that has been documented in the literature (Lee and Schwartz 1983, Clifton et al 1988, Sarwar et al 2013) but not yet included in GEOS-Chem.

The mechanism proposed by Lee and Schwartz (1983) gives the aqueous-phase reaction: ${{\rm{HSO}}}_{3}^{-}+2{{\rm{NO}}}_{2}+{{\rm{H}}}_{2}{\rm{O}}\to {{\rm{SO}}}_{4}^{2-}+3{{\rm{H}}}^{+}+2{{\rm{NO}}}_{2}^{-}.$ Clifton et al (1988) further studied the reaction and found rate constants 1–2 orders of magnitude higher than Lee and Schwartz (1983). The rate constants were summarized by Sarwar et al (2013) as a function of pH: $k=-7.1557\times {10}^{4}{{\rm{pH}}}^{2}+5.0\times {10}^{6}{\rm{pH}}.$ Although NO₂ is only weakly soluble, the reaction can be important under conditions with high NO₂ concentrations and pH values (Littlejohn et al 1993). We implement this aqueous-phase reaction on both cloud droplets and deliquescent aerosols using the rate constant function as described above. The pH values are assumed to be 4.5 in clouds (Alexander et al 2009) and 5.6 for deliquescent aerosols based on aerosol measurements taken at a suburban site near Beijing in winter 2012 (T-M Fu, unpublished data). Figure 1 shows the simulated mean sulfate aerosol concentrations in surface air over North China for January 2013 from the standard and improved simulations. With this reaction, the simulation shows on average 68% increases in January mean sulfate concentrations at the surface of North China. Using a pH value of 4.5 for deliquescent aerosols would decrease the sulfate concentrations by 14% relative to the results with pH value of 5.6. In addition, we apply the aqueous-phase oxidation of SO₂ by H₂O₂ and O₃ on deliquescent aerosols, further increasing the surface sulfate concentrations over North China by 22% (figure 1(c)).

Zoom In Zoom Out Reset image size

Figure 2 compares monthly mean observed versus simulated total PM2.5 concentrations at 34 CMA monitoring sites for January 2013. Aerosol chemical composition measurements including sulfate, nitrate, ammonium, black carbon (BC), and organic carbon (OC) at 13 CMA monitoring sites for January 2006–2007 (Zhang et al 2012) are also shown in figure 2. The model generally captures the spatial distributions of measured PM2.5 and aerosol composition concentrations with correlation coefficients of 0.65–0.84 and small biases for sulfate, nitrate and ammonium (within 10%; up to −23% for BC).

Zoom In Zoom Out Reset image size

The model underestimates OC concentrations by 43% on average, particularly over the sites in the western China and southern China. This large OC underestimate has also been discussed by Fu et al (2012). Using surface measurements including those presented here, they pointed out a national-scale underestimate of the Chinese OC anthropogenic emissions and in western China a missing OC source associated with biofuels for heating during the cold months. In addition, our study does not include secondary organic carbon, which accounts for 21% of the annual mean surface OC in China as simulated by GEOS-Chem (Fu et al 2012).

The model successfully simulates the high sulfate concentrations over North China in winter owing to the improved aqueous-phase chemistry for sulfate production as described above. The finer model horizontal resolution also improves the comparison by better capturing the spatial variability in PM distributions. We find BC and OC concentrations in a simulation at the coarser 2° × 2.5° resolution are respectively 48% and 65% lower than the measurements.

We use the adjoint model to calculate the sensitivity of surface PM2.5 concentration over Beijing (the 1/4° × 5/16° grid cell covering the center of Beijing: 39.9°N, 116.3°E) to all aerosol sources over North China. Figure 3 shows the adjoint sensitivities for the mean surface PM2.5 concentration at Beijing averaged from January 2013 to 2015 separated into different source sectors and chemical species at the model underlying resolution. The magnitudes represent approximately how much the January mean PM2.5 at Beijing would reduce in the absence of the specific emissions. Summing all values over the model domain gives 122 μg m⁻³, about 86% of the simulated mean PM2.5 concentration (142 μg m⁻³). The discrepancy is mainly attributed to the nonlinear response of PM2.5 to emissions (Henze et al 2007, 2009). We find contributions from natural emissions (e.g. NH₃ from oceans and NO_x from soils) and sources outside the model domain are small (less than 2%).

Zoom In Zoom Out Reset image size

Figure 3 shows large source contributions from residential and industrial emissions of SO₂, OC, and anthropogenic fine dust, consistent with the intensified use of coal over North China during the heating season (Zhang et al 2014a, Zhang et al 2013). Large source contributions are also seen from agricultural emissions of NH₃ reflecting formation of ammonium sulfate and ammonium nitrate aerosols (Wang et al 2011, Kharol et al 2013). We find in the model increasing NO_x emissions would enhance the wintertime ozone titration, and thus suppress sulfate production from SO₂ oxidation by ozone, leading to some negative sensitivity with respect to NO_x emissions.

We sum the adjoint sensitivities across different dimensions and interpret them as percentage contributions to the total sensitivity. By chemical species, the January mean PM2.5 over Beijing is generated by emissions of OC (27.3%), anthropogenic fine dust (27.2%), SO₂ (14.2%), NH₃ (13.2%), BC (9.6%), and NO_x (8.5%). Zhang et al (2013) also showed largest fractions of OC (28.6%) and dust (28.9%) in wintertime PM2.5 measurements in Beijing. Previous studies suggested that secondary inorganic aerosols (sulfate, ammonium, and nitrate) over North China are most sensitive to NH₃ emissions and least to NO_x emissions in winter (Kharol et al 2013, Wang et al 2013). Our results also show weakest sensitivity for NO_x emissions, but higher sensitivity for SO₂ than NH₃ due to the enhanced aqueous-phase sulfate production.

By emission sector, residential (49.8%) and industrial sources (26.5%) show the largest sensitivities, followed by agriculture (11.8%), transportation (5.9%), and power plants (6%). The estimated contributions heavily rely on the bottom-up sectorial emissions with uncertainties in both the emission magnitude and sectorial contribution. The contribution from transportation can be largely underestimated because secondary organic aerosols (SOA) are not included. Recent laboratory studies show emissions from gasoline vehicles cause significant formation of SOA that can be 6–10 times higher than their primary OC emissions (Nordin et al 2013, Gordon et al 2014). In a sensitivity simulation with OC emissions from transportation increased by a factor of 10, we find that the resulting contribution of transportation can reach 12%.

Comparison with previous studies using the PMF method (Zhang et al 2013, Huang et al 2014) is complicated by the different source categories identified. Huang et al (2014) found dominant contributions from coal burning and secondary aerosols for 5–25 January 2013. Zhang et al (2013) estimated major source contributions from coal burning (57%), soil dust (16%), industrial pollution (12%), and secondary inorganic aerosol (6%) for PM2.5 over Beijing in winter 2010. Our adjoint results are qualitatively consistent with the PMF results; both show the largest contribution from coal burning, which dominates the residential and industrial emissions. The PMF results also indicated the importance of secondary aerosols but the adjoint method is able to attribute them to precursor emissions. Future research is needed to reconcile the source category definitions among different source attribution methods.

Lastly, when we consider the source influences summed by region, the geographical distribution of adjoint sensitivity indicates strong pollution transport from the southern Hebei Province and from the west (Shanxi Province and Inner Mongolia; see the top left panel of figure 4 for the regions) to Beijing. The transport pathways are consistent with results from back-trajectory models (Wang et al 2004, Zhang et al 2013). Summing sensitivities over the municipality of Beijing accounts for 52.8% of the total sensitivity. Sources over Hebei Province account for 26.1% of the total sensitivity, 6.7% from Tianjin, and the rest 14% transported from sources further away.

Zoom In Zoom Out Reset image size

We further apply the adjoint to Tianjin (39.13°N, 117.20°E) and Shijiazhuang (38.05°N, 114.48°E; capital city of Heibei Province) over North China. The left panels of figure 4 show the sensitivities of mean surface PM2.5 concentrations at Beijing, Tianjin, and Shijiazhuang to the spatial distribution of aerosol emissions for January 2013–2015. The right panels of figure 4 summarize the percent contributions integrating geographically over municipalities of Beijing, Tianjin, Hebei and other regions. All three cities show strong regional pollution transport influences spreading over North China. Beijing and Tianjin show local PM2.5 source contributions of 45–53%, and large contributions of ∼26% from sources over Hebei. PM2.5 over Shijiazhuang is less impacted by sources in Beijing (less than 2%), but show considerable influences from Shanxi Province (13%) owing to the prevailing southwest winds (Wang et al 2014a).

The Action Plan on Air Pollution Prevention and Control released by the Chinese State Council in September 2013 set a strict target for the Beijing–Tianjin–Hebei region with the PM2.5 concentrations to be reduced by 25% by 2017 relative to 2012 (Chinese State Council 2013). We can see for Beijing in January, it would require at least 46% reduction of all emissions in Beijing if only a local control policy is applied, or on average 29% reduction of emissions in the Beijing–Tianjin–Hebei region with a joint regional pollution control policy.