Volatile Organic Compounds Monitored Online at Three Photochemical Assessment Monitoring Stations in the Pearl River Delta (PRD) Region during Summer 2016: Sources and Emission Areas

Abstract

Volatile organic compounds (VOCs) were monitored online at three photochemical assessment monitoring stations (MDS, WQS and HGS) in the Pearl River Delta region during the summer of 2016. Measured levels of VOCs at the MDS, WQS and HGS sites were 34.78, 8.54 and 8.47 ppbv, respectively, with aromatics and alkenes as major ozone precursors and aromatics as major precursors to secondary organic aerosol (SOA). The positive matrix factorization (PMF) model revealed that VOCs at the sites mainly came from vehicle exhaust, petrochemical industry, and solvent use. Vehicle exhaust and industrial processes losses contributed most to ozone formation potentials (OFP) of VOCs, while industrial processes losses contributed most to SOA formation potentials of VOCs. Potential source contribution function (PSCF) analysis revealed a north-south distribution for source regions of aromatics occurring at MDS with emission sources in Guangzhou mainly centered in the Guangzhou central districts, and source regions of aromatics at WQS showed an east-west distribution across Huizhou, Dongguan and east of Guangzhou, while that at HGS showed a south-north distribution across Guangzhou, Foshan, Zhaoqing and Yangjiang. This study demonstrates that multi-point high time resolution data can help resolve emission sources and locate emission areas of important ozone and SOA precursors.

1. Introduction

Volatile organic compounds (VOCs) are important precursors to secondary pollutants such as ozone (O₃) and secondary organic aerosol (SOA) in the troposphere through complex photochemical cycles [1]. In recent years, the gradual deterioration of O₃ pollution in China’s major cities has attracted great concerns [2]. Particularly in summer, excessive O₃ concentration is the main cause of poor air quality index. Previous studies demonstrated that in these megacities O₃ formation is largely VOC-limited and the increase in anthropogenic emissions of VOCs is an important reason for the deteriorating O₃ pollution. The ozone formation potentials (OFP) of different VOCs vary largely with their OH radical reaction rates [3,4,5]. Many VOCs, like aromatic hydrocarbons and biogenic VOCs, are also precursors of SOA [6,7], which shares 20–70% mass concentrations of PM_2.5 in ambient air [8,9]. Therefore, for the coordinated prevention and control of O₃ and PM_2.5 prevention, it is critical to determine which are the most important precursors to O₃ and SOA among the hundreds of VOC species occurring in the ambient [6,7,10], and, moreover, what are the major sources and emission areas of these precursors.

The Pearl River Delta (PRD) region is the world’s largest megacity and one of the most urbanized and industrialized regions in China. In the past decades, the rapid urbanization and industrialization in the region have given rise to the emission of a large number of air pollutants, leading to air quality problems with PM_2.5 and O₃ pollution. In recent years, while PM_2.5 pollution has been alleviated with intensified pollution control, O₃ pollution however are becoming more and more prominent. As O₃ formation in this area is largely more sensitive to VOCs than to nitrogen oxides (NOx) in the PRD region [11,12,13,14,15], a large number of studies have been carried out in the region to investigate anthropogenic sources of VOCs [10,14,16,17,18,19,20,21,22,23]. However, most of the studies on VOCs sources in the PRD region are focused on urban areas through offline analysis with samples collected in canisters, and studies on VOCs sources in rural areas and on regional scale are quite limited [14,24]. Moreover, online VOC monitoring with much higher time resolution and a larger data pool would facilitate better understanding of the emission sources and diagnosing emission areas [25,26,27] and would therefore yield results that benefit formulating emission control measures.

In this study, approximate one-month online VOC measurement was carried out concurrently at an urban, a suburban and a background site in the PRD region during the photochemically active summertime to determine ambient levels of VOCs and identify important precursors to O₃ and SOA. PMF model was used for the source apportioning, and PSCF was further used to explore the potential source areas of the aromatics, which are important precursors to both O₃ and SOA.

2. Methods

2.1. Desciption of Monitoring Sites

Three sites (MDS, WQS and HGS; Figure 1) among Photochemical Assessment Monitoring Stations (PAMS) in the PRD region are selected in this study. They are located in the cities of Guangzhou and Jiangmen.

The MDS site (113.33° E, 23.11° N) is 50 m above the ground on the roof of Guangdong Provincial Environmental Monitoring Center in the Haizhu District of Guangzhou. The station is located in a downtown area surrounded by a number of traffic arteries, and there are petrochemical, shipping, automobile manufacturing and other industrial parks clustered in the east and south of the site.

The WQS site (113.55° E, 22.71° N) is 30 m above the ground on the roof of the Science and Technology Building of Fok Yingdong Research Institute, Hong Kong University of Science and Technology in the Nansha District of Guangzhou. The station is located in the southernmost part of Guangzhou and in the center of the “Golden Triangle” of Guangzhou, Hong Kong and Macao. It is 50, 75 and 65 km away from core urban areas of Guangzhou, Hong Kong and Macao, respectively. There are enterprises in thermal power generation, chemical industry, electronics, shipping and other industries around the point.

The HGS site (112.93° E, 22.73° N) is 20 m above the ground on a hill top of the Huaguo Mountain in the Taoyuan Town of Jiangmen. The station is 80 km away from Guangzhou, 50 km away from Foshan and 30 km away from Jiangmen. It is located within the core areas of the PRD region with relatively intensive pollution emissions in its surrounding cities and pollutions there are obviously affected by atmospheric transport.

According to the dominant southerly wind direction during the observation period (July–August 2016), the upwind WQS is taken as the background site, while the MDS site is an urban point and HGS is a suburban site influenced by regional transport.

2.2. Online Measurement of VOCs

Fifty-seven VOC species, including 29 alkanes, 10 alkenes, 17 aromatic hydrocarbons and acetylene were monitored hourly at the three sites with three online VOCs analyzers, two of which (TH-300B, Tianhong Inc., Wuhan, China) were installed at the HGS and MDS sites, and one of which (Clarus-500 & TurboMatrix, PerkinElmer Inc., Waltham, MA, USA) was installed at the WQS site. For the two TH300-B analyzers, air samples are firstly collected at a flow rate of 60 mL/min for 5 min and then analyzed by the gas chromatography-flame ionization detector/mass selective detector (GC-FID/MSD), with a sampling and analysis cycle of 1 h. The moisture in the air samples was cryogenically removed and the VOCs in air samples were captured by a collecting column. The FID was used to detect C₂–C₅ hydrocarbons through a 20 m × 0.32 mm × 3.0 μm PLOT column, and MSD was used to detect C₅–C₁₂ hydrocarbons through a 60 m × 0.25 mm × 1.4 μm DB-624 column. In order to improve the accuracy to identify sources of VOCs, some typical source tracers, including chloromethane, trichloroethylene, tetrachloroethylene and methyl tert-butyl ether, in each sample were also monitored. The Perkin Elmer analyzer consists of an automatic concentrator, a thermal desorber, and a GC with dual FID detectors with a 50 m × 0.32 mm sodium sulfate deactivated alumina column for the C₂–C₆ components, and a 50 m × 0.22 mm × 1 μm methyl silane column for the C₆–C₁₂ components, respectively.

Stringent quality assurance/quality control criteria were followed during the online monitoring. The analyzers have built-in auto-linearization and auto-calibration programs developed for the quality control. The certified calibration standard mixture (1 ppm, Spectra Gases Inc., Bransburg, NJ, USA) was dynamically diluted to 10 ppb and injected every day to check the performance of the calibration curves, which had concentration-response correlation coefficients above 0.995 in this study. The method detection limits (MDLs) for the 57 VOC species ranged from 0.001 to 0.326 (ppbv).

2.3. Data Collection

In this study, for VOCs monitoring at the three stations the time spans in 2016 were as follows: from 9 July to 1 August at the MDS site; from 4 July to 3 August at the WQS site; and from 1 July to 1 August at the HGS site. Except for missing data caused by instrument failure or maintenance, 387, 707 and 690 groups of valid VOCs data were obtained at the MDS, WQS and HGS sites, respectively. Hourly data for NOx (chemiluminescence method), O₃ (UV spectrophotometry method), SO₂ (UV fluorescence method), CO (infrared spectroscopy method) and the meteorological data (including temperature, relative humidity (RH), wind speed and wind direction measured with Vaisala WXT535 Multi-Parameter Weather Sensor) were collected from the stations administered by the Ecological and Environmental Monitoring Center of Guangdong Province.

2.4. Data Analysis

2.4.1. PMF Receptor Model

Positive matrix factorization (PMF) receptor model is a method widely used to quantify source contributions to target pollutants observed at a given receptor site. It is a multi-factor analysis tool that can decompose the matrix of field observation data into two matrices: a source contribution matrix and a source profile matrix [28,29]. In this work, the USEPA PMF model version 5.0 was used [30], with input principles as follows: (1) certainty and uncertainty values for VOC species were estimated according to the methods in the PMF 5.0 user guide; (2) VOC sources were explored using a data matrix based on the Equation (1) below; (3) the PMF solution minimizes the objective function Q based on the uncertainties (μ) with a convergence based on Equation (2) below:

$${\mathrm{X}}_{\mathrm{ij}}={\displaystyle \sum }_{\mathrm{k}=1}^{\mathrm{p}}{\mathrm{g}}_{\mathrm{ik}}{\mathrm{f}}_{\mathrm{kj}}+{\mathrm{e}}_{\mathrm{ij}}$$

(1)

$$\mathrm{Q}={\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle \sum }_{\mathrm{j}=1}^{\mathrm{m}}{\left[\frac{{\mathrm{x}}_{\mathrm{ij}}-{{\displaystyle \sum }}_{\mathrm{k}=1}^{\mathrm{p}}{\mathrm{g}}_{\mathrm{ik}}{\mathrm{f}}_{\mathrm{kj}}}{{\mathrm{u}}_{\mathrm{ij}}}\right]}^2$$

(2)

where matrix X is the VOC species concentration data, which can be decomposed into a factorial solution of two matrices including g (source contribution) and f (source profile); e is the residual value; i is the number of VOCs samples, j is the number of VOC species, and k is the number of VOC sources, and p is the number of independent sources.

The PMF model requires two input files: sample species concentration values and sample species uncertainty values. In the model, data values below the MDLs were substituted with MDL/2; missing data values were substituted with median concentrations [31]. If the concentration is less than or equal to the MDL provided, the uncertainty is calculated using the following equation, Unc = 5/6 × MDL; if the concentration is greater than the MDL provided, the calculation is Unc = ((Error Fraction × mixing ratios)² + (MDL/2)²)^1/2 [32,33]. In this study, most species had all concentration levels above the MDLs. In the PMF analysis, 20% extra modelling uncertainty was used to account for temporal changes in source profiles and other sources of variability. The species also need to be further filtered according to the signal-to-noise ratio (S/N). If S/N > 2, species will be classified as “strong”. If 0.2 < S/N < 2, species will be classified as “weak”. Otherwise, species will be classified as “bad” and ruled out from the PMF model. To determine the optimum number of VOC sources, we tested combinations of 3–8 sources, to determine the best fit in the PMF model. The number of sources is optimal when the Q(true)/Q(expected) ratio is close to 1. In this study, 5 sources were selected for the WQS site and 6 sources were selected for the MDS and HGS sites, with the Q(true)/Q(expected) ratios ranging from 0.9 to 1.0.

2.4.2. Potential Source Contribution Function (PSCF)

In order to track possible emission area of pollution sources for ambient VOCs observed at each site, the PSCF analysis was conducted using the software TrajStat developed based mainly on conditional probability function [34], which is defined as the conditional probability that the corresponding value of a certain element exceeds the set threshold when the air mass passing through a certain region reaches the observation site. A certain study area is divided into grids with a specific resolution, and a threshold is set for VOC concentrations. When the concentration of VOCs corresponding to the trajectory is higher than the threshold, the trajectory can be considered as a pollution trajectory.

The backward trajectories were calculated for time points during a day (UTC 0:00, 3:00, 6:00, 9:00, 12:00, 15:00, 18:00 and 21:00). The meteorological parameters were collected from the National Oceanic and Atmospheric Administration (NOAA) Global Data Assimilation System (GDAS). The study field was divided into cells of 0.1° × 0.1°. The PSCF is defined by Equation (3) below:

$${\mathrm{PSCF}}_{\mathrm{ij}}=\frac{{\mathrm{m}}_{\mathrm{ij}}}{{\mathrm{n}}_{\mathrm{ij}}}\times {\mathrm{W}}_{\mathrm{ij}}$$

(3)

where i and j represent the latitude and the longitude, respectively; n_ij represents the total endpoint number of trajectories falling in the ijth grid cell. m_ij represents the endpoint number of VOC concentration trajectories in the same grid cell. VOC concentration trajectories represent any VOC concentration above the reference values [25], which were the average VOC concentrations. To avoid uncertainties resulting from the influence of lower n_ij values, an arbitrary weight function W_ij was applied to the PSCF values [27]. W_ij is defined by Equation (4) below:

$${\mathrm{W}}_{\mathrm{ij}}=\{\begin{array}{cc}1.00& 20<{\mathrm{n}}_{\mathrm{ij}}\\ 0.75& 10<{\mathrm{n}}_{\mathrm{ij}}\le 20\\ 0.50& 5<{\mathrm{n}}_{\mathrm{ij}}\le 10\\ 0& 0<{\mathrm{n}}_{\mathrm{ij}}\le 5\end{array}$$

(4)

2.4.3. Ozone Formation Potential (OFP)

OFP was calculated as the Equation (5) below based on the maximum incremental reactivity (MIR) values of each VOC species [35]:

$${\mathrm{OFP}}_{\mathrm{i}}={\left[\mathrm{VOC}\right]}_{\mathrm{i}}\times {\mathrm{MIR}}_{\mathrm{i}}$$

(5)

where OFP_i represents the O₃ formation potential of species i; [VOC]_i is the concentration of species i; and MIR_i (in g O₃/g VOC) is the MIR of species i.

2.4.4. Secondary Organic Aerosol Formation Potential (SOAFP)

SOAFP is a parameter used to estimate the contribution of individual VOCs to SOA formation [36,37]. In the present study, SOAFP was calculated by multiplying the concentration of a species by its fractional aerosol coefficient (FAC), as shown by equations below:

$${\mathrm{SOAFP}}_{\mathrm{i}}={\left[\mathrm{VOCs}\right]}_{\mathrm{i},\mathrm{initial}}\times {\mathrm{FAC}}_{\mathrm{i}}$$

(6)

$${\left[\mathrm{VOCs}\right]}_{\mathrm{i},\mathrm{st}}={\left[\mathrm{VOCs}\right]}_{\mathrm{i},\mathrm{initial}}\times {(1-\mathrm{F}}_{\left[\mathrm{VOC}\right]\mathrm{i}})$$

(7)

where SOAFP_i represents the amount of aerosol formed from species i; FAC_i refers to fractional aerosol coefficient of species i; [VOC]_i,initial refers to the initial concentration of the species i; [VOC]_i,st refers to the concentration of the species i after oxidation; and F_[VOC]i refers to the fraction of reacted for the species i.

3. Results and Discussion

3.1. Variations in Mixing Ratios of VOCs

The concentrations of VOCs were 34.78, 8.54 and 8.47 ppbv on average at the MDS, WQS and HGS sites, respectively, during our field campaign in the summer (Figure 2). Mixing ratios of VOCs at the MDS site in the core urban area were substantially higher than that at the background (WQS) or suburban (HGS) sites. Apart from much stronger vehicle emissions and more decoration-related painting activity in the densely populated urban areas, the prevailing southerly winds during the field campaign might also bring pollutants from the upwind industrial areas to the MDS site. The concentrations of VOCs at WQS and HGS site on average were quite near each other. The mean mixing ratio of total VOCs observed at the HGS site in this study was quite near those observed previously in background areas in the North China Plain, in the Yangtze River Delta region or the PRD region [19,38,39], but was much lower than that measured at the same site during autumn (22 October 22 to 20 November 20) in 2014 [40], possibly due to changes in emission strength from 2014 to 2016 through enhanced emission control or due to lower boundary layer heights in autumn than in summer. However, the mean concentration of VOCs at the WQS site was similar to that observed previously during November–December 2009 [23] at the same area. The inconsistent trends of mixing ratios of VOCs in ambient air at different sites in the rapidly developing PRD region also implied that it is necessary to establish a network with online measurements to track the changing VOCs in ambient air.

Figure 3, Figure 4 and Figure 5 display the temporal variations of VOCs along with that of wind speed and direction, temperature and RH, O₃, NO, NO₂, SO₂ and CO at the MDS, WQS and HGS sites.

At the MDS site, higher concentrations of O₃ appeared from 24 July to 26 July and from 29 July to 1 August, but there are differences between the two periods: concentrations of VOCs and NOx showed no obvious rise from 24 July to 26 July, while they had a substantial rise during 29 July−1 August. A sudden increase in SO₂ was observed on 29 July, implying enhanced emission from coal combustion.

At the WQS site, higher concentrations of O₃ appeared from 7 July to 9 July and also from 29 July to 1 August when wind speed was low and the concentrations of VOCs and NOx were both at higher levels.

At the HGS site, higher O₃ concentrations occurred on 8 July, 25 July and 31 July. However, concentrations of VOCs and NOx were also higher on 8 July and 31 July, while concentrations of VOCs and NOx were at a low level on 25 July when northerly winds dominated, higher O₃ levels on that day were possibly due to the regional transport.

Figure 6 shows the diurnal variations of VOCs, NOx, and O₃ concentrations at the three sites. It can be found that O₃ peaked in the afternoon, while NOx and VOCs concentrations seemed to be opposite to that of O₃, with the lowest concentrations in the afternoon. Concentration of VOCs rose at night mainly because there was no photochemical consumption at night and that the boundary layer height was also reduced. On noon time when photochemical reaction was the strongest in a day while temperature and the boundary layer height reached the highest, a large amount of VOCs were consumed during the photochemical formation of O₃ and the higher boundary layer also resulted in decreased VOCs concentrations.

3.2. Compositions and Contributions to OFP and SOAFP

Figure 2 also shows the group compositions (alkanes, alkenes, alkyne and aromatics) of VOCs observed at the three sites. Alkanes were the most abundant with the proportions of 62.8%, 43.8% and 42.9%, followed by aromatics with shares of 20.2%, 36.6% and 35%, alkenes with shares of 12.1%, 19.4% and 17.1%, and alkynes with shares of 4.9%, 0.2% and 5%, respectively, at the MDS, WQS and HGS sites. The higher proportion of alkanes at the MDS site might reflect more contributions from liquefied petroleum gas (LPG)-related sources and vehicle emissions in the urban area [41], while the higher proportion of aromatics in the WQS and HGS sites might be related to stronger emissions in the surrounding areas from industrial point sources, most of which had been moved from urban areas to industrial parks in rural or suburban areas.

Figure 7 shows top 10 VOCs species in average mixing ratios at the three sites. At the MDS site, the LPG-related propane and n-butane were the most abundant VOC species, followed by toluene, isobutane, ethylene, isopentane, m/p-xylenes, ethane, acetylene and n-pentane. At the WQS site, isopentane, which is tracer of oil evaporation [21], was the most abundant VOC species, followed by toluene, n-pentane, 1-hexene, m/p-xylenes, n-dodecane, o-xylene. At the HGS site, toluene, which is a widely used solvent in industry [42], became the most abundant VOC species, followed by propane, isoprene, m/p-xylenes, ethylene, n-butane, isopentane, acetylene, n-pentane, ethane. It worth noting that isoprene at the HGS site ranked the third, indicating reasonably more biogenic emissions at this site on a vegetated mountain.

As for OFP contributed by VOC species at the three sites, consistent with previous studies [43,44,45,46,47,48], aromatics and alkenes were major contributors, and xylenes and toluene were among the most important aromatics, while ethylene, propylene, butene and isoprene were the most important alkenes (Figure 7). Ethylene contributed the most to OFP at the MDS site, followed by m/p-xylenes, toluene, propylene, o-xylene, n-butane, isoprene, cis-2-butene, isobutane and propane. At the HGS site isoprene contributed the most to OFP, followed by m/p-xylenes, ethylene, toluene, o-xylene, propylene, ethylbenzene, isopentane, styrene and n-butane. At the WQS site, m/p-xylenes contributed the most to OFP, followed by 1-hexene, 1,2,4-trimethylbenzene, cis-2-butene, o-xylene, toluene and 1-butene.

As aromatics are most important anthropogenic precursors to SOA [49,50,51], toluene and xylenes were reasonably found to contribute the most to SOAFP at the three sites. At the MDS and HGS sites, toluene, ethylbenzene and xylenes contributed more than 80% of the total SOAFP, while at the WQS site, toluene, m/p-xylenes, o-xylene, n-dodecane, 1,2,4-trimethylbenzene, 1,2,3-trimethylbenzene, ethylbenzene, p-diethylbenzene, m-ethyltoluene and m-diethylbenzene are the top 10 species in their contributions to SOAFP.

3.3. Source Attributions by PMF

Figure 8 shows the source profiles resolved by PMF at the MDS site. Factor 1 shows a higher contribution by C₈- and C₉-aromatics and substantial contributions by trichloroethylene and tetrachloroethylene, therefore it may be related to industrial process losses [31]. Factor 2 is characteristic of higher proportions of aromatics such as toluene, styrene and ethylbenzene that are generally used as solvents [52,53], and it can be termed solvent use. Chloromethane, which is a marker of biomass burning [53,54,55], contributes the most in the Factor 3, and some alkenes and aromatic hydrocarbons also occurs in the Factor 3 [56]. Therefore, Factor 3 is considered as biomass burning. Ethylene, acetylene, ethane and 1-butene, as well as 2-methylhexane, n-heptane and methyl cyclohexane, which might be related to oil cracking process [57,58], are present in the Factor 4. Therefore, Factor 4 is regarded as petrochemical industry emissions. Factor 5 shows highest contributions by propane, isobutane and n-butane, which are markers of LPG-related emissions [59] and also among most abundant species from on-road vehicle fleets with a considerable portion of LPG-fueled buses and taxies [41], meanwhile MTBE, a tracer of gasoline-related emissions [43], and vehicle exhaust related C₅ and C₆ alkanes [43,60], are also present in the Factor 5. So Factor 5 represents vehicular emissions. Isoprene from plant emissions [55] contributes the most in the Factor 6, which is thus biogenic emission. Figure 9 and Figure 10 show source profiles resolved by PMF at the WQS and the HGS sites, respectively. Similarly emission sources are identified based on the factor profiles.

Figure 11 shows contributions by major emission sources at the three sites. At the MDS site, vehicle exhaust contributed the most with a percentage of 43%, followed by petrochemical emissions (30%) and solvent use (13%). At the WQS site, vehicle exhaust also contributed the most but with a percentage of 26%, while petrochemical industry, solvent use, industrial processes losses and biomass burning accounted for 24%, 21%, 18%, and 11%, respectively. At the HGS site, vehicle exhaust was again the largest contributor (34%), followed by petrochemical emissions (25%), industrial processes losses (13%), solvent use (12%) and biogenic emissions (12%). Higher contribution of biogenic emission at the HGS site is reasonable as the site is mostly surrounded by forests.

With contributions to VOCs by different sources, the contribution of different emission sources to OFP can be further calculated. Solvent use instead contributed most to OFP at the MDS site due to a large number of more photochemically reactive VOCs species in this source. At the WQS site, the contribution of petrochemical emissions, solvent use and industrial process losses contributed more to OFP than vehicle exhaust, and accounted for 29%, 27% and 16% of total OFP, respectively, indicating that industry-related emissions played a major role in OFP at the WQS site. At the HGS site, the biogenic emission however ranked first in the contributions to OFP, reaching an average percentage of 26%.

Accordingly the contribution of different emission sources to SOAFP can also be calculated. At MDS site, SOAFP was largely attributed to petrochemical emission (43%) and solvent use (36%), while vehicle exhaust only contributed 11%. At the WQS site, the contribution percentages of solvent use, industrial process losses and petrochemical emission are 41%, 25% and 20%, respectively. At the HGS site, petrochemical emission and industrial processes losses also contributed the most to SOAFP, with shares of 41% and 23%, respectively. In general, SOAFP was largely contributed by VOCs from industry related emissions.

3.4. Source Areas by PSCF Analysis

As aromatics are found to be among most important ozone and SOA precursors as discussed above and aromatics are more stable in ambient air than alkenes [49], in this study we further conducted PSCF analysis to explore the location of probable emission sources for aromatics including benzene, toluene, ethylbenzene and xylenes observed in July–August when ozone pollution is more serious. This kind of analysis would help clarify the source emission areas and regional transport channels as well.

As showed in Figure 12, the warmer the color in an area, the greater the value of PSCF and thereby the greater the emission intensity in the grid. The potential source areas of the aromatics observed at the MDS site show a north–south distribution, and the emission sources in Guangzhou were mainly centered in the Guangzhou central districts. The potential source areas of the aromatics at the WQS site were distributed from east to west across Huizhou, Dongguan and east of Guangzhou, where a great many industrial enterprises were located. Also there were emission sources in the northeast Guangzhou, in Qingyuan in the northwest to Guangzhou, and in Foshan in the southwest to Guangzhou that contributed aromatics observed at the site. The potential source areas of the aromatics at the HGS site had south-north distribution across Guangzhou, Foshan, Zhaoqing and Yangjiang, and there were the source areas, though less important, in Huizhou and Dongguan. The PSCF identified aromatics emission areas is highly consistent with the actual industrial layout in the cities, indicating that the PSCF results can better reflect the potential source areas. From south to north, there is a belt with high WPSCF values, which is mainly related to furniture processing enterprises, which give emissions of aromatics [42], in Panyu in the south of Guangzhou.

4. Conclusions

In this study, VOCs were monitored online during the summer of 2016 at three representative sites, namely MDS (urban), WQS (background) and HGS (suburban), in the PRD region. The mixing ratios, chemical compositions, O₃ and SOA formation potentials, sources and potential source areas of VOCs in the PRD region were investigated based on the hourly data. The results revealed average VOCs mixing ratios of 34.78 ppbv, 8.54 ppbv and 8.47 ppbv at the MDS, WQS and HGS sites, respectively. In terms of mixing ratios, alkanes were the most abundant VOC groups at the three sites, followed by aromatics and alkenes; propane and n-butane were the most abundant VOC species at the MDS site in mixing ratios, isopentane and toluene were the most abundant at the WQS site, and toluene and propane were the top two at the HGS site.

Aromatics (xylenes and toluene) and alkenes (ethylene, propylene, butene and isoprene) among the VOC species contributed most to OFP at the three sites, and toluene and xylenes contributed the most to SOAFP at the three sites. Source apportionment by the PMF model revealed that vehicle exhaust was the largest contributor at the three sites, accounted for 43%, 26% and 34% of observed VOCs at the MDS, WQS and HGS site, respectively. Meanwhile, petrochemical emissions (30% at MDS, 24% at WQS, and 25% at HGS), solvent use (13% at MDS, 21% at WQS and 12% at HGS) and industrial processes losses (3% at MDS; 18% at WQS and 13% at HGS) also contributed substantially, with reasonably much lower contribution by industrial processes losses at the urban MSD site (3%) and much biogenic emission contribution (12%) at the HGS site surrounded by forests. However, at all the three sites OFP and SOAFP were mostly contributed by VOCs from industry-related emission sources, indicating the importance to control reactive organic gases from industry sources for the coordinated prevention of PM_2.5 and O₃ pollution in the region.

Based on the PSCF analysis with the high time resolution data, potential source areas of the aromatics, which are important precursor to both O₃ and SOA could be targeted. This is a good example that online data can be used to locate potential source areas for pollutants of concern. It worth noting that this study is only based on observations at three sites during a summer month. It is quite certain that online observation data at more sites and in more seasons would give a more clear and reliable picture for potential source areas of important O₃ and SOA precursors in the region, which is highly beneficial to reduce emissions in an accurate and effective way.