Wintertime Formation of Large Sulfate Particles in China and Implications for Human Health

Outdoor air pollution causes millions of premature deaths annually worldwide. Sulfate is a major component of particulate pollution. Winter sulfate observations in China show both high concentrations and an accumulation mode with a modal size >1 μm. However, we find that this observed size distribution cannot be simulated using classical gaseous and aqueous phase formation (CSF) or proposed aerosol-processing formation (APF) mechanisms. Specifically, the CSF simulation underestimates sulfate concentrations by 76% over megacities in China and predicts particle size distributions with a modal size of ∼0.35 μm, significantly smaller than observations. Although incorporating the APF mechanism in the atmospheric chemical model notably improves sulfate concentration simulation with reasonable parameters, the simulated sulfate particle size distribution remains similar to that using the CSF mechanism. We further conduct theoretical analyses and show that particles with diameters <0.3 μm grow rapidly (2–3 s) to 1 μm through the condensation of sulfuric acid in fresh high-temperature exhaust plumes, referred to as in-source formation (ISF). An ISF sulfate source equivalent to 15% of sulfur emissions from fossil fuel combustion largely explains both observed size distributions and mass concentrations of sulfate particles. The findings imply that ISF is a major source of wintertime micron-sized sulfate in China and underscore the importance of considering the size distribution of aerosols for accurately assessing the impacts of inorganic aerosols on radiative forcing and human health.


Supporting Text 1. Configuration and test of the MOSAIC aerosol model
The Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) aerosol model can be executed using the modal or sectional approach 1 .In the WRF-Chem model, MOSAIC is implemented using the sectional framework and the model allows users to choose 4 or 8 discrete size bins to represent the aerosol size distributions.In this study, 8 bins were selected in the simulations (Supporting Table S1).MOSAIC includes the treatments of all major aerosol species, such as sulfate, nitrate, ammonium, elemental carbon, organic mass, sodium, chloride, calcium, water and other components.In MOSAIC, aerosol species are mixed internally in each bin, so the particles in a bin have the same chemical compositions.MOSAIC simulates both aerosol mass and number for each bin.
MOSAIC includes the treatments of various chemical and microphysical processes of aerosols, including: nucleation, coagulation, condensation, thermodynamic equilibrium, phase equilibrium, and so on.For thermodynamic module, in MOSAIC, Multicomponent Taylor Expansion Method (MTEM) 2 is used to calculate the mean activity coefficients of electrolytes in the aqueous aerosols and Multicomponent Equilibrium Solver for Aerosols (MESA) 3 is used to solve the solid-liquid phase equilibrium.In addition, Adaptive Step Time-split Euler Method (ASTEM) is employed to deal with the gas-particle partitioning problem.It is coupled with the thermodynamic module MESA-MTEM to integrate the mass transfer equations for all size bins 1 .In MOSAIC, firstly, the particle growth or shrinkage induced by dynamic gas-particle partitioning of trace gases is estimated in a Lagrangian manner, and then, the transfer of particles between bins is determined by employing a two-moment method 1 .
We compared the simulated sulfate mass size distribution with measurements at six background sites or unpolluted remote sites in China to reflect the capability of the model in terms of reproducing the size distributions of sulfate under general conditions (Supporting Fig. S9).Although the model-simulated accumulation-mode peaks of sulfate are slightly smaller than the observation, the model generally captures the size distribution of sulfate in the accumulation mode under general conditions.In this study, CSF pathways lead to smaller sulfate size bins compared to the observations.We tested the simulated results for other components and found that WRF-Chem model also show similar biases of size distribution for nitrate and ammonium under CSF simulation.After introducing the ISF mechanism, the simulated size distribution of ammonium also significantly ameliorated.However, the simulated size distribution of nitrate does not change substantially, suggesting other mechanisms might controlling the size distribution of nitrate during winter hazes in China.

Supporting Text 2. Calculation of β in the APF simulation
The calculation method for β in the APF simulation is as follows.Firstly, we calculated the total sulfate contribution from the ISF in the ISF simulation for China.Then, we performed APF pre-simulations, where all settings are consistent with APF except for the value of β, which is set to 1. Subsequently, in the APF pre-simulation, we determined an appropriate parameter γ through a combination of previous literature and sensitivity analysis.The contribution of sulfate from the APF mechanism can be determined by computing the difference in sulfate mass before and after the APF process at each time step in the model.
By summing up the monthly contributions of APF mechanism from all grid cells across different vertical layers within China in APF pre-simulation, we obtained the total APF contribution to sulfate for each specific month.Finally, the ratio of the total sulfate contributions from the ISF and APF processes in ISF simulation and APF pre-simulation is the value of β for each month.Upon obtaining the values of β, we incorporated them into the simulation of APF for each month, conducted additional simulations, and got the final APF results in the article.

Supporting Text 3. Aerosol number size distribution comparison and current challenges
In power plant plumes, the presence of gaseous H2SO4 can play an important role in the generation of new particles.Previous studies conducted in North America have suggested that H2SO4 generated from power plant precursor emissions can lead to new particle formation, particularly in cleaner background conditions 4 .However, it's important to consider the background aerosol conditions, as pre-existing particles in the ambient environment act as condensation sinks, hindering the formation of new particles 4 .
Consequently, regions with cleaner conditions are more conducive to new particle formation.Our study focused specifically on the heavily polluted winter seasons in China, where the presence of a substantial number of pre-existing particles in the ambient air poses challenges for new particle formation.Recent observational data collected during China's winter season has revealed that particles within power plant plumes tend to exhibit larger sizes compared to background particles in the atmosphere 5 .As a result, we propose that H2SO4 from power plants in China during winter primarily contributes to the condensation and growth of larger particles.
Comparing observed and simulated particle number concentrations is of importance.
Nevertheless, this comparison currently encounters two significant challenges.Firstly, the default particle size bins in the current model are relatively coarse for accurately comparing with observed number concentrations.New particle formation typically occurs below 0.05 μm, necessitating the inclusion of multiple detailed size bins below this threshold in the model to enable precise simulations.However, the WRF-Chem model with the MOSAIC aerosol mechanism, by default, divides aerosol sizes into only 8 bins.The first two size bins are 0.039-0.078μm and 0.078-0.156μm.The current model includes only one particle size bin for all particles with a diameter < 0.05 μm, which is insufficient for accurately simulating new particle formation and comparing it with observations.Secondly, there is currently a scarcity of high-precision and multivariate observational data available for power plant and other combustion plumes and their surrounding areas in China, which limits our ability to compare with observed data.Acquiring such data is crucial for enhancing our understanding of this mechanism.

Supporting Text 4. Evaluation of model-simulated wintertime air pollutants
The CSF simulation provided reasonable results for meteorological variables over China in the winter (Supporting Table S5).At the Wuhan site, the CSF simulations of SO2 (normalized bias +87.1%) and nitrate (+33.0%) were overestimated, and the CSF simulations of ammonium (-24.7%) were underestimated.The APF and ISF simulations showed that the overestimation of SO2 improved from +87.1% (CSF) to +65.1% (APF) and +59.4% (ISF).We noticed that the normalized bias of SO2 in the APF and ISF simulations was still large, which may be primarily caused by the uncertainty of emission heights and emission inventory of SO2.Previous studies have suggested that the nearsurface concentrations of SO2 simulated by atmospheric chemistry models are sensitive to vertical profiles of emissions [6][7][8] .Here, we conducted a set of sensitivity analyses on anthropogenic emission heights and found that simulations using current sector-specific emission vertical profiles could enable well-predicted SO2 concentrations at most stations in eastern China (Supporting Fig. S5).Thus, the current emission heights employed in the model were relatively reasonable on a large scale.Owing to differences in economic development and industrial structures, vertical emission profiles may differ among regions.
The application of unified emission heights in the model may lead to uncertainties in the simulation results.In addition, the SO2 emission inventory also has uncertainties that may affect the simulated SO2 concentration near the ground at the Wuhan site.

Supporting Text 5. Distributions of APF and ISF simulated sulfate
Although the APF or ISF mechanism can greatly improve the simulated inorganic aerosol mass concentrations, and the total APF and ISF sources of sulfate are almost the same (see Methods), the spatial distributions of sulfate are substantially different (Supporting Fig. S10).For example, the surface concentration of ISF sulfate was higher over the North China Plain (ISF: 12.9 μg m -3 vs. APF: 7.8 μg m -3 ), but lower over the Sichuan Basin (ISF: 18.8 μg m -3 vs. APF: 22.4 μg m -3 ).This is because these two mechanisms are affected by different factors.The ISF mechanism is mainly affected by the sulfur emission distribution, whereas the APF mechanism is affected by not only SO2 emissions but also the spatial distribution of the aerosol surface area.The higher aerosol surface concentrations over the Sichuan Basin than over the North China Plain lead to higher sulfate results over the Sichuan Basin in the APF simulation.Overall, Fig. 1 in the main text shows that the ISFsimulated sulfate concentrations were better than the APF results.Supporting Fig. S9 Observed and simulated sulfate mass size distributions at six background sites or unpolluted remote sites over China in winter.The three-month average values were employed to compare with the observation data (Obs.),which was obtained from literature [23][24][25] triangles) used in this study.Panel c. Location of the provinces in eastern China.The background is the spatial distribution of population density in 2015 9 .S9 Supporting Fig. S2 Particulate mass deposition fraction curves in different regions of the human respiratory system based on ICRP model 10 .Lines ranging from light green to dark green represent different regions of the respiratory tract: nasal region /head airways, tracheobronchial region, and alveolar region.The light pink line (TB+AL) represents the sum of the tracheobronchial and alveolar regions, and the dark pink line (total) represents the sum of all three respiratory regions.February 2015 under APF (panel a) and ISF (panel b).Fraction of model results within a factor of two of the observations (FAC2) and the number of stations (N) are displayed on each subplot.Supporting Fig. S6 Observed and simulated sulfate mass size distributions at 12 sites over China in winter.If the observation time of one site is within the simulation period, the simulated mean value of the data during exactly observation time was used to compare with the observation data (obs.)(listed in Supporting Table S4); otherwise, the three-month average values were employed to compare with the observation data.APF, ISF, S1, S2 represent different sulfate formation mechanisms (see the main text for details).

Table S2 .
and sampled between 2012-2014.CSF represents the classical sulfate formation mechanism.Main model configurations of the WRF-Chem model.

Table S3 .
Summary of the sulfate, nitrate, ammonium mass concentration observations. S23Supporting

Table S4 .
Summary of the sulfate, nitrate, ammonium size distribution observations. S25Supporting

Table S5 .
Statistics of meteorological variables in WRF-Chem at all sites overChina.