Process Contributions to Secondary Inorganic Aerosols during Typical Pollution Episodes over the Pearl River Delta Region , China

The Integrated Process Rate (IPR) analysis embedded in CAMx model was used to quantify contributions from different atmospheric processes to the formations and accumulations of ambient PM2.5 and the secondary inorganic aerosol (SIA) during two typical particulate pollution episodes in different seasons in the Pearl River Delta (PRD) region. Process analysis results indicated that primary fine particle emissions were the major sources of high ambient PM2.5 in urban areas with intensive anthropogenic activities. Aerosol process and advection transport were another two major processes contributing to the increasing PM2.5 and SIA over the PRD region. Regarding formation of SIA species, elevations of nitrate and ammonium at Guangzhou (urban), Heshan (rural) and Panyu (suburban) sites were largely associated with aerosol process, while those at Huizhou (urban) site were dominated by advection process, but elevated sulfate concentrations at these four sites were all dominated by advection process. The difference can be attributed to spatial variations of SO2, NOx and NH3 emissions, site locations and meteorological conditions. Advection, aerosol chemistry, deposition and vertical diffusion were important pathways to remove SIA at these four sites. Within the hours with most growing PM2.5 concentrations, aerosol process was the most important contributor to the formation of new SIA throughout the entire planetary boundary layer.


INTRODUCTION
With the rapid industrialization and urbanization in the past decades, haze has become a nationwide pollution problem in China due to its significant effects on visibility, human health and climate change (IPCC, 2001;Menon et al., 2002;Nel, 2005).In response to the severe pollution, central government of China issued a new ambient air quality standard (GB 3095-2012) by taking PM 2.5 into account and announced the aim to reduce concentrations of PM 2.5 by up to 25% by 2017 with respect to 2012 levels in Beijing-Tianjin-Hebei (BTH) area, Yangtze River Delta (YRD) and Pearl River Delta (PRD) region.
The PRD region is one of the most densely urbanized and developed regions in China.With only 0.4% of China's land area and 3.2% of the population, it contributes 9.3% of national gross domestic product and 25.2% of total trade in 2013.Accompanied by the increased economic activities and intensive industrialization, PRD has severe air pollution problems in the last decade.The aerosol optical depth (AOD) retrieved from MODIS satellite data showed a hotspot in this area (So et al., 2005;Wu et al., 2005;Lin et al., 2015), with higher records during autumn, winter and spring (GDEMC and HKEPD, 2013).Thanks to the continuous emission control, the number of haze days in PRD declined in recent years, from over 100 in 2007 to 65 in 2013 (http://www.grmc.gov.cn).However, annual average PM 2.5 levels in PRD (42 µg m -3 in 2014) still exceeded the national standard (35 µg m -3 ) (GDEMC and HKEPD, 2014;GB 3095-2012).To further alleviate PM 2.5 pollution in the PRD, more efficient and evidence-based control strategies should be formulated and adopted on the basis of in-depth understanding of interactions among meteorology, primary emissions, physical and chemical processes and PM 2.5 formations in this region.
Such interactions can be simulated by a three-dimensional (3-D) air quality model (AQM) which has been widely used in the PRD region (e.g., Cheng et al., 2007;Feng et al., 2007;Wang et al., 2009;Kwok et al., 2010;Zhang et al., 2011).Process analysis (PA) is an embedded tool that analyzes contributions from different physical and chemical processes within the model, thereby can provide comprehensive insight to the formation of pollution (Jeffries and Tonnesen, 1994).In China, attempts have been made to study the formation and fate of particulate matter in recent years using PA.For example, Liu et al. (2010) [10][11][12]2004.However, previous studies generally focused on PM 10 or a single haze period, while governing processes in the formation of haze events characterized by high concentrations of PM 2.5 and its temporal and spatial variations across the PRD region remained largely unknown.
In this study, the Comprehensive Air Quality Model with Extensions (CAMx) with the embedded PA was applied to two typical haze periods in the PRD.Four locations with distinct backgrounds or emission characteristics in the PRD region were selected for analysis.The objective was to identify governing chemical and physical processes that contribute to elevated PM 2.5 concentrations and their variations over the PRD region.Considering the large proportion of secondary inorganic aerosol (SIA) during haze periods (Lai et al., 2007;Yang et al., 2011;Fu et al., 2014;Huang et al., 2014), this study mainly focused on SIA.

Modeling System Set up
The Weather Research and Forecasting (WRF) model version 3.2 (Skamarock et al., 2005), the Sparse Matrix Operator Kernel Emissions processor for the PRD region (SMOKE-PRD) (Wang et al., 2011) and the CAMx model version 5.4 (Environ, 2011) were configured as the regional air quality modeling system in this study.The lambertconformal projection was used as the basic projection scheme.Three levels of nested domains were used in CAMx model (Fig. S1) and WRF model with grid resolutions of 27 km, 9 km, and 3 km, respectively.The outermost domain (D1) coveres most parts of East Asia, Southeast Asia and the northern part of Western Pacific.The middle domain (D2) coveres Guangdong province and its surroundings, and the innermost domain (D3) coveres the whole PRD region.CAMx has 18 vertical layers from the surface to 50 hPa, while WRF has 26 vertical layers with 18 layers under 1000 meters height.Heights of the lower 12 vertical layers that were used in the IPR analysis in the CAMx model are 36,80,131,200,298,437,635,916,1098,1316,1576 and 1887 m, respectively.More information of model configurations in WRF and CAMx are listed in Table S1 of the supporting information (SI).Since Asymmetric Convective Model (ACM) module is not supported in the process analysis, Ktheory was used instead to calculate the vertical diffusion.Clean air profiles available in CAMx were used as boundary conditions for D1.For the nested domain, boundary conditions were generated from parent domains.
The Integrated Processes Rate (IPR) analysis, one component of PA, identifies detailed physical and chemical process rates for selected grid cells and pollutants.Processes considered in this study were: vertical advection (VADV), vertical diffusion (VDIF), horizontal advection (HADV), horizontal diffusion (HDIF), deposition (TDEP), aqueous aerosol chemistry (AQUE), aerosol chemistry (AERO) and emission (EMIS).In IPR analysis, AERO refers to the effect of inorganic aerosol chemistry and organic aerosol chemistry in the aerosol module.TDEP refers to the effect of dry deposition only occurring in the surface layer and wet deposition.It should be noted that, unlike the upper layer, vertical transport in the surface layer only interacts with the second layer via VADV and VDIF.

Emission Inventories and Processing
The 2010 Guangdong emission inventory developed by Pan et al. (2014) was adopted for Guangdong area in D2 and D3.This emission inventory, containing NO x , SO 2 , CO, VOC, NH 3 , PM 10 , PM 2.5 , BC and OC pollutants, covered all major emission sources in Guangdong province, including mobile sources, combustion sources, dust sources, VOCs solvent usage, agricultural sources, industrial processes, biomass burning, and other sources.For Hong Kong area in D2 and D3, annual emissions provided by the Hong Kong Environment Protection Department (HKEPD, http://www.epd.gov.hk/epd/english/top.html) was used.The MEGAN (Guenther et al., 2012) model was used to create hourly biogenic VOC emissions in all domains.For anthropogenic emissions, the SMOKE-PRD modeling system was used to process the bulk annual emission amount into hourly and gridded model-ready emission data for all domains, with updated temporal and spatial profiles (Yin et al., 2015).Specifically, the most up-to-date local chemical profiles of VOC sources in the PRD region were also adopted in this study (Zheng et al., 2013).
The Multi-resolution Emission Inventory for China (MEIC, http://www.meicmodel.org) was applied for D1 and non-Guangdong regions in D2 and D3.For the region in D1 but outside China, the Regional Emission inventory in Asia (REAS, http://web.nies.go.jp/REAS/) emission inventory (Kurokawa et al., 2013) was adopted.MEIC and REAS are in the form of monthly emission inventory with horizontal resolutions of 0.25 degree.A re-gridded program was developed to process the MEIC and REAS data into modelready emission data for the target domain by using temporal allocation profiles in SMOKE-PRD and national spatial allocation profiles.

Observation Data and Evaluation Protocol
To evaluate the performance of CAMx, measurements of PM 2.5 , PM 10 , NO x , NO 2 and SO 2 concentrations at 12 monitoring stations of the Pearl River Delta Regional Air Quality Monitoring Network (RAQM), hourly concentrations of sulfate, nitrate and ammonium in April 2013 from Panyu station (Zhao et al., 2014), and 24-h average concentrations of sulfate, nitrate and ammonium in November 2013 in Nansha, Nanhai, Guangzhou and Dongguan were adopted.Locations of monitoring stations were shown in Fig. 1 and details of their surrounding environments were given in Table S2 of SI.The evaluation protocol proposed by the U.S. Environmental Protection Agency (EPA) was followed for validation, with indicators including the mean bias (MB), mean error (ME), normalized mean bias (NMB), normalized mean error (NME), root mean square error (RMSE), fraction bias (FB), fraction error (FE), and correlation coefficient (R) (Boylan and Russell, 2006).

Selection of Analysis Periods and Locations
Two typical pollution episodes with different weather conditions during 13-15 April and 19-21 November 2013 were selected as study periods.As shown in Fig. 2, the central PRD region, including Guangzhou, Foshan, and Dongguan, were severely polluted on 15 April.The average PM 2.5 concentration over the PRD region was 82.3 µg m -3 during 13-15 April, and the highest PM 2.5 concentration of 226.0 µg m -3 was recorded at Dongguan at 07:00 a.m. on 15 April.Another pollution episode on 19-21 November was selected, during which the average region-wide PM 2.5 concentration was 80.3 µg m -3 with the highest PM 2.5 concentration of 139.0 µg m -3 at Zhongshan.For IPR analysis, four sites in the PRD region, i.e., Heshan, Panyu, Guangzhou and Huizhou site, were selected.Heshan is a rural site located in the downwind area of PRD, Panyu is a suburban site, Guangzhou is an urban site with intensive anthropogenic emissions, while Huizhou is located in the eastern PRD often but shows different PM 2.5 formation (Fan et al., 2014).These four sites with different environmental background can basically capture the formation patterns of PM 2.5 across the PRD region.

Model Performance
Table 1 summarizes overall performance statistics for the hourly concentrations of NO 2 , NO x , SO 2 , PM 2.5 and PM 10 in April and November.In general, the model underestimated NO 2 and NO x in both months and SO 2 in April, but it performed relatively better for NO 2 and NOx than SO 2 with lower NMEs and higher R. Biases of simulated NO 2 , NO x and SO 2 were attributed to the uncertainties inherent in emission inventories and inaccuracies of simulated meteorological fields (Liu et al., 2010).
For PM 2.5 , simulated concentrations showed a high consistency with observations, with FEs of 40.8% and 44.9% and FBs of -2.4% and -4.7% in April and November, respectively, indicating that PM 2.5 was slightly underestimated in these two months.According to the recommended values of FB (≤ ±30%) and FE (≤ 50%) by Boylan and Russell (2006), simulated PM 2.5 showed a satisfactory performance.In comparison, PM 10 was underestimated in April but overestimated in November.Given the fact that dust fugitive was a major contributor to PM 10 emissions in Guangdong province, the negative bias in April and positive bias in November might be partially explained by uncertainties of temporal allocations of dust sources in the SMOKE-PRD system.
Fig. 3 shows that the model well reproduced temporal variations of PM 2.5 and SIA at Panyu site in April, with R of 0.70, 0.65, 0.65 and 0.58 for PM 2.5 , nitrate, sulfate, and ammonium, respectively.Changes of SIA species during the pollution episode on 14-15 April were captured by CAMx model.Nitrate and ammonium were generally overestimated in April, especially during 9-14 April.By contrast, sulfate was underestimated in general, particularly during 16-24 April.Table 2 summarizes comparisons between observed and simulated SIA at Nansha, Nanhai, Guangzhou and Dongguan in November.The simulated and observed SIA were at similar levels, with the mean bias of -22.8% to 44.2% for ammonium, -29.0% to 31.2% for nitrate and -20.1% to 31.0% for sulfate.The relatively small biases of simulated SIA at these four stations indicated that the spatial distribution of SIA was well reproduced in the PRD region, which was reliable for following IPR analysis.Results also indicated that the model overestimated nitrate and ammonium and underestimated sulfate when SIA was relatively high.Positive biases of nitrate and ammonium were also found in other simulation studies (Fountoukis et al., 2011;Walker et al., 2012;Wang et al., 2013;Barbaro et al., 2015) which might be caused by the uncertainties in NO x emissions, •OH concentration, oxidation rate of NO 2 by •OH, removal rate of HNO 3 and under-predicted planet boundary layer (PBL) (Heald et al., 2012;Pirovano et al., 2012;Walker et al., 2012).

Process Contributions to Ground-Level PM 2.5 and SIA in the PRD Region
The IPR results of hourly contributions from atmospheric processes to the formations of PM 2.5 and its major compositions including ammonium (PNH4), nitrate (PNO3) and sulfate (PSO4) at Guangzhou, Heshan, Panyu, and Huizhou site during 13-15 April and 19-21 November are shown in Figs.4-7.Concentrations of PM 2.5 and SIA contributed by atmospheric processes are shown in Table S2 of SI.To investigate the relationships between different processes at these four sites, correlation coefficients between EMIS/HADV and other processes were calculated, as shown in Table S3 of SI.

• Guangzhou Site
For urban Guangzhou site, EMIS was found to be one of the dominant contributors to PM 2.5 formation during the two episodes.Contributions from EMIS showed obvious diurnal patterns with higher concentrations in the daytime and lower concentrations in the nighttime.Located in the most densely populated areas, Guangzhou site had significant emissions from various anthropogenic sources, including fugitive dust, mobile and combustion sources.HADV was another major process contributing to the PM 2.5 formation, while VADV and VDIF were the two major atmospheric processes responsible for the loss of PM 2.5 .Throughout the study periods in April and November, VADV and HADV  showed highly negative correlation (Table S2).Diurnal profile of VDIF was highly consistent with that of EMIS, with a correlation coefficient of -0.95.It was suggested that intensive EMIS was the main driving force for VIDF that diffused PM 2.5 to the upper layers.Compared with VDIF, the effect of HDIF at Guangzhou site was negligible.In both months, parts of the ground-level PM 2.5 was explained by AERO, with average rates of 27.1 ± 12.1 µg m -3 h -1 and 18.2 ± 7.5 µg m -3 h -1 in April and November, respectively, indicating that chemical production of PM 2.5 in Guangzhou cannot be neglected.PNH4, PNO3, and PSO4 were also identified (Fig. 4).It was found that contribution patterns of HADV and VADV to SIA were similar with those to PM 2.5 .However, differences were noticeable concerning the contribution patterns of AERO to SIA.AERO played an important positive role in the formation of PNH4 and PNO3 but was peripheral to the evolution of PSO4.Instead, PSO4 at Guangzhou site was largely influenced by HADV and VADV.This is because NO x emissions, which were generally emitted from mobile and combustion sources, were intensive in the downtown area of Guangzhou, while SO 2 emissions were relatively lower.Similar results were also found by Xue et al. (2014) using observation-based model.Moreover, the contribution of AERO was more significant in April, due to higher NH 3 emissions (Zheng et al., 2012) and humidity in spring.
• Heshan Site Located in the rural southwestern PRD, Heshan saw significant contributions from advection transport (HADV and VADV).During the pollution episode on 15 April, ground-level PM 2.5 was mainly enhanced by VADV from the upper air under the surface divergence.During the episode in November when Heshan was the downwind of the central PRD under the prevailing northeasterly wind, high ambient PM 2.5 was mainly contributed by HADV.As a typical rural site, anthropogenic emissions at Heshan site were relatively small, indicating that ambient PM 2.5 and its precursors were largely transported from upwind areas.
AERO also made contributions to the change of PM 2.5 concentrations in Heshan site by enhancing the ground-level SIA at night but depleting SIA in the daytime.Examining hourly contributions of SIA, we can find that PNO3 dominated the dynamic aerosol process, while AERO of PSO4 provided a negligible contribution.The depletion of SIA in the day time was resulted from vaporizations of PNO3 and PNH4 due to the higher temperature (Griffith et al., 2015).Additionally, contributions of AERO to PNO3 and PNH4 were more significant during pollution episodes, indicating the important role of chemical production in high PM 2.5 concentrations at Heshan site.

• Panyu Site
Panyu is a typical suburban site and advection process was revealed as the most significant contributor to the change of PM 2.5 concentrations.In April and November, high PM 2.5 concentrations mainly resulted from VADV which brought PM 2.5 from upper layers under the prevailing northeasterly winds (62 degree on average for April and 52 degree on average for November).In addition, VDIF was an important process to remove the ground-level PM 2.5 , particularly during severely polluted periods.EMIS also made part of contributions to the ground-level PM 2.5 .
Due to the relatively high NO x emissions and abundant SIA precursor transported from upwind areas, contributions of AERO to PNO3 and PNH4 at Panyu site were obvious, especially during the time when ambient PM 2.5 was high.Contribution patterns of AERO at Panyu site were similar with those at Heshan site, as shown in Figs. 5 and 6.Owing to relatively large local SO 2 emissions (Fig. S3), AERO of PSO4 was larger than that at Heshan site.The rate even reached about 8.00 µg m -3 h -1 during the peak period.Additionally, advection transport was another major contributor to the change of SIA at Panyu, indicating the importance of transport in the formation of ambient SIA in suburban areas of the PRD region.
• Huizhou Site EMIS, HADV, VADV, and VDIF were all significant contributors to the PM 2.5 formation at Huizhou site.This site is located in the downtown area of Huizhou city where there were moderate PM 2.5 emissions from fugitive dust, mobile and combustion sources.Compared with other sites, loss of surface PM 2.5 by DIFF was noticeable at this site.
AERO, HADV, and VADV either increase or remove SIA at Huizhou site depending on the variation of weather conditions.Contribution patterns of PNO3 showed that AERO dominated the change of nitrate, especially in April.For instance, AERO constituted a major positive contribution to the change of nitrate during the two peaks in April, while it also depleted most nitrate at noontime on 14 April.Contribution patterns of AERO to PNH4 in both periods were consistent with those of PNO3, though the rates were smaller.On the contrary, PSO4 formation was mainly caused by atmospheric advection transport during the two periods, which were consistent with the previous study (Qin et al. 2015).

Spatial Differences in Process Contributions to SIA
As discussed above, AERO, HADV, and VADV were the major processes responsible for the change of SIA in the PRD region.To characterize their spatial variations across the PRD region, correlations between process contributions and changes of SIA and spatial distributions of AERO were analyzed in Table 3 and Fig. 8, respectively.Since HADV and VADV represent the divergence and convergence of atmosphere, respectively, they showed highly negative   correlation, which might obscure the relationship between advection transportation and changes of SIA (Table S5).In order to eliminate this impact, total advection process (TADV, the sum of HADV and VADV) was used in the following analysis.
It was revealed that the accumulation of SIA was influenced by TADV in different degrees.TADV showed high contributions to the elevated SIA concentrations at Guangzhou, Heshan, Panyu, and Huizhou site in both months, indicating that advection transport had significant impacts on the accumulations of SIA over the entire PRD region.As Guangzhou site is located in the downtown area with high anthropogenic emissions while Huizhou site is located in the upwind area of the PRD region, TADV at Guangzhou site might be partly originated from the surrounding areas while TADV at Huizhou might be greatly influenced by super-regional transport.TADV was also responsible for the dispersion of SIA in the PRD region.
AERO were also found highly correlated to the elevation of ambient SIA, with correlation coefficients from 0.31 to 0.68, indicating high contributions from the aerosol process over the PRD region.The correlations were even higher than those of TADV at Guangzhou, Heshan, and Panyu, which means that SIA growths at these sites were more dominated by AERO.In terms of SIA species, the increasing PNO3 and PNH4 at Guangzhou, Heshan, and Panyu site were dominated by AERO.In urban areas with high NO x emissions, such as Guangzhou site, abundant NO x emissions were conducive to AERO of PNO3 and PNH4.In suburban and rural areas characterized by low NO x emissions (As shown in Fig. S3), such as Panyu and Heshan site, the dominated AERO was largely due to the transport of precursors of SIA (HNO3) from upwind areas.On the contrary, the increasing PNO3 and PNH4 at Huizhou site were mainly influenced by TADV, which was highly related to the super-regional transport.In term of PSO4, TADV generally dominated at all four sites.Nonetheless, AERO of PSO4 at Panyu site was the largest among these four sites since it is close to an industrial area (Fig. S3).
Examination of spatial distributions of AERO species revealed similar results.As shown in Fig. 8, during the two episodes, contributions of AERO to PSO4 both peaked within narrow areas where SO 2 emissions are intensive, such as ports, power plants and major industries.For other areas with lower SO 2 emissions, PSO4 growth was highly associated with advection transport.AERO of PNO3 and PNH4 were both concentrated in the central PRD region that covered Guangzhou, Heshan, and Panyu sites.Those areas with high AERO of PNO3 and PNH4 were somewhat overlapped with areas with high emissions of NO x and NH 3 .In addition, their spatial distributions were also affected by the transport of ambient NH 3 , NOx, and HNO 3 .For instance, AERO of PNO3 and PNH4 were also significant in downwind areas in November due to the influence of regional transport.

Atmospheric Process Contributions to SIA at Different Layers
As SIA was the dominated components to ambient PM 2.5 in the PRD region, further investigation on process contributions to SIA in different vertical layers is important to understand the PM 2.5 formation in this region.Therefore, vertical-dependent process contributions to SIA at the hour with the largest increasing rate of PM 2.5 concentrations during the pollution episode were examined (9:00 on 15 April and 19:00 on 20 November at Guangzhou, 6:00 on 15 April and 21:00 on 20 November at Heshan, 3:00 on 15 April and 20:00 on 20 November at Panyu, and 3:00 on 15 April and 21:00 on 20 November at Huizhou).
As shown in Fig. 9, positive contributions of AERO to SIA at Guangzhou site were observed throughout the PBL and the rates ranged from 6% to 35% of net SIA changes in April and from 12% to 14% in November.The maximum AERO occurred at the surface layer and declined along the upper layers.Similar declining patterns of AERO throughout the PBL can also be found at other sites.This was attributed to the fact that precursors emissions (NO x and NH 3 ) were mainly emitted in the ground level over the PRD region.Interestingly, for Heshan and Panyu site in April, AERO was significant in the fourth layer at night, leading to peak SIA concentrations at the 200 m height at Heshan site and on the 131 m height at Panyu site.Noteworthy is that this aloft high SIA concentrations occurred during the nighttime.It can be explained by "pollutant cloud" phenomenon which appears when long distance transport of precursors is significant under the backdrop of shrinking nocturnal boundary layer that isolated the pollutant aloft from the surface (Pecorari et al., 2013).Similarly, obvious contributions of AERO were found in layer 10-12 (about 1316 m to 1887 m height) at Heshan, Guangzhou, and Panyu site in November, which might be due to the superregional transport of aerosol precursors under the prevailing northeasterly wind.VADV and HADV played important but opposite roles in different layers throughout the PBL due to the atmospheric convergence and divergence.SIA changes caused by these two processes were somewhat weaker in the upper layers.

CONCLUSIONS
The Integrated Process Rate analysis embedded in CAMx model was used to investigate contributions from different atmospheric processes to the formations of PM 2.5 and SIA at four typical sites in the PRD region: Guangzhou (urban site), Heshan (rural site), Panyu (suburban site) and Huizhou (urban site).Two typical particulate pollution episodes in spring and winter were selected for IPR analysis.
EMIS was an important process that contributed to PM 2.5 formation in urban areas.Guangzhou site had the largest contribution rates of EMIS among the four sites, followed by Huizhou site.The highly positive relationship between EMIS and PM 2.5 formation suggested that local PM 2.5 emissions control in urban areas should be continued or even strengthened.EMIS at other sites had fewer impacts on PM 2.5 formations.
Compared to EMIS, AERO has wider impacts on formations of PM 2.5 and SIA across the PRD region.It was positively associated with the elevation of ambient SIA in urban areas and rural areas.The relationship was intensified at Guangzhou site due to its intensive NO x emissions.AERO at suburban Panyu and rural Heshan sites were also outstanding as a result of rich precursors transported from upwind areas.In terms of SIA species, contributions of AERO to the elevated PNO3 and PNH4 were significant in the PRD region, with the exception of Huizhou where elevated PNO3 and PNH4 were mainly transported by TADV.Spatial distribution also revealed that AERO of PNO3 and PNH4 were both concentrated in the central PRD region and downwind areas, while AERO of PSO4 was merely concentrated in narrow areas with intensive SO 2 emissions.In addition, AERO constituted the major positive contribution to the change of SIA throughout the PBL in the hour with most growing PM 2.5 concentrations.Within the PBL, AERO generally peaked on the ground level and decreased in upper layers.
Advection transport (including HADV and VADV) was another important process that contributed to domain-wide formations of PM 2.5 and SIA, indicating that regional transport was significant over the entire PRD region.Particularly, PSO4 elevations at all four sites were largely influenced by advection transport.For Huizhou site, superregional transport might be the major driving force for the SIA formation.

Fig. 3 .
Fig. 3. Times series of simulated SIA and observed SIA at Panyu site.

Fig. 9 .
Fig. 9. Process contributions to SIA at layer 1 to layer 12 at Guangzhou, Heshan, Panyu and Huizhou site in the hour with most growing PM 2.5 concentrations; Descriptions inside the parentheses are the local standard time and PBL layer of the analysis hour.
used the PA tool embedded in Community Multiscale Air Quality (CMAQ) model to identify the most influential processes and chemical reactions of ozone and PM 10 in China in different seasons.Li et al. (2014) investigated governing chemical and physical processes contributing to the change of PM 2.5 during a haze episode in the YRD region.Fan et al. (2014) applied PA tool to discern the different formation of PM 10 in various cities in the PRD region during October

Table 1 .
Model performance of CAMx in April and November.

Table 2 .
Comparison between simulated SIA and observed SIA at four off-line aerosol sampling stations.(unit: µg m -3 )

Table 3 .
Correlation coefficients between processes and changes of SIA concentrations.