Chemical Characteristics, Spatiotemporal Distribution, and Source Apportionment of PM2.5 Surrounding Industrial Complexes in Southern Kaohsiung

This study investigated PM2.5, specifically, its chemical characteristics and spatiotemporal variation, and identified its potential sources in the Linhai industrial complex of southern Kaohsiung. Seasonal 12-h PM2.5 samples were collected simultaneously at three sampling sites. The results showed that high PM2.5 concentrations mainly occurred during winter and spring, with concentrations at the downwind sites always exceeding those at the upwind sites due to the transport of secondary aerosol by the prevailing winds. 31.4–56.8% of the PM2.5 consisted of water-soluble ions (WSIs), which were dominated by secondary inorganic aerosol (SIA) (SO4 + NO3 + NH4). High mass ratios between the SIA and the PM2.5 (SIA/PM2.5) were also observed during winter and spring, suggesting that SIA was the primary contributor to the high levels of PM2.5. A neutralization ratio (NR) < 1 indicated that the PM2.5 was acidic, and a nitrogen oxidation ratio (NOR) > 0.1 and a sulfur oxidation ratio (SOR) > 0.25 showed that SIA frequently formed during winter and spring. The metallic elements, which accounted for 12.0–20.2% of the PM2.5, tended to exhibit higher concentrations during daytime than nighttime, and the enrichment factors (EFs) revealed that the trace metals (Ni, Cr, Cu, and Zn) were mainly anthropogenic in origin. Carbonaceous content formed 9.3–24.3% of the PM2.5, and high mass ratios between the organic and the elemental carbon (OC/EC) were also observed during winter and spring. Moreover, the mass ratios between the malonic and the succinic acid (M/S) were always > 1.0 during winter and spring and < 1.0 during summer and fall, demonstrating that organic acids dominated the SOA during the first two seasons. The major sources of PM2.5 in the Linhai industrial complex were steel plants, followed by secondary sulfate and nitrate, vehicular exhaust, petrochemical plants, and incinerators.


INTRODUCTION
PM 2.5 pollution has been a hot issue nationwide due to the severe threat to human health. Long-term exposure to PM 2.5 could cause severe impacts on human health since it is highly suspected as carcinogenic, mutagenic, and teratogenic, which make PM 2.5 the Type I carcinogenic pollutant issued by World Health Organization (WHO). Additionally, PM 2.5 could highly possibly penetrate alveoli in the lower respiratory tract of human lungs (Sacks et al., 2010;Christopher et al., 2018) and enter the cardiovascular system (Loop et al., 2018) particularly in metropolitan and industrial areas (Kim et al., 1999;Khan et al., 2010;Lee et al., 2018). Thus, the improvement of poor particulate air quality has been one of the hottest environmental tasks nationwide and worldwide.
As the biggest industrial city in southern Taiwan, Kaohsiung suffers from the highest concentrations of criteria air pollutants, particularly PM 2.5 , among the major cities in Taiwan in the past two decades, due to the most densely distributed heavy industries. During 2016, 26.3% of the whole year suffered from severe PM 2.5 pollution that was much higher than daily ambient air quality standard of 35 µg m -3 ruled by the Taiwan Environmental Protection Administration (TWEPA), even higher than 15 µg m -3 of yearly standard. Thus, the exploration of the spatiotemporal distribution of PM 2.5 , as well as its chemical characteristic, has attracted much attention, aiming to depict its chemical fingerprint and track the major source. Yang et al. (2017) indicated that the dominant source of PM 2.5 at the KAQZ was secondary inorganic aerosols, traffic exhausts, and road dust in winter and spring seasons. Shen et al. (2019) also investigated the chemical characteristic of PM 2.5 in the urban, petrochemical industry, rural region of Kaohsiung City and Pingtung County, respectively, during highly polluted event days. The identified dominant sources included anthropogenic emissions such as ironworks, coal-fired power plants, petrochemical industry, etc. Since the contribution of industrial emission to primary PM 2.5 and gaseous precursors (i.e., SO 2 , NO x ) of PM 2.5 plays a key role in the formation of PM 2.5 , it is crucial to investigate the spatiotemporal distribution and chemical characteristics of PM 2.5 surrounding the anthropogenic source in Kaohsiung for understanding the fingerprint of PM 2.5 in highly polluted regions.
The Linhai industrial complex, as the heavy industrial center in southern Taiwan, has approximately 56% of Taiwan's heavy industries with the most prominent infrastructures, including the key industries of ironworks, oil refineries, petrochemical plants, utility power plants, freight transport, and shipbuilding. Additionally, Kaohsiung Linhai Harbor is the busiest seaport with the largest cargo-loading capacity in Taiwan, while the nearby Kaohsiung International Airport is the second busiest airport in Taiwan in terms of passenger numbers. Kaohsiung Linhai Harbor is well connected to other major cities by national highways. However, the systematical investigation on the spatiotemporal distribution and chemical fingerprint of PM 2.5 surrounding the Linhai industrial complex are lacking in the past years, which is imperative to conduct for forming relevant strategies to eliminate the PM 2.5 pollution.
The influence of prevailing winds on the spatiotemporal distribution of PM 2.5 in southern Kaohsiung is significant. Previous studies reported that higher PM 2.5 concentrations are mostly observed in winter (northeasterly winds) because higher concentration of PM 2.5 could be blown from Northeast Asia, North Asia, and even Central Taiwan in winter and spring, promoting and aggregating the local ambient PM 2.5 concentration (Sacks et al., 2010;Loop et al., 2018). By contrast, clean air carried by southwesterly winds in summer influence relatively less on the formation of PM 2.5 in Taiwan . Additionally, the formation of secondary PM 2.5 may exhibit significant diurnal variation due to the sunlight radiation; anthropogenic emission of SO 2 , NO x , volatile organic compounds (VOCs), etc. diurnally varied as well, thereby leading to the diurnal pattern of concentration and chemical characteristic of PM 2.5 (Chen et al., 2019). Therefore, it is critical to investigate the seasonal and diurnal variation of PM 2.5 concentration and their chemical fingerprint.
Accordingly, this study aims to investigate the chemical characteristics, the spatiotemporal variation of PM 2.5 surrounding the largest industrial complexes of southern Kaohsiung. The seasonal and diurnal variations of PM 2.5 's chemical composition, including water-soluble ions (WSIs), metallic elements, anhydrosugars, and organic acids, were thoroughly explored. Finally, the seasonal and diurnal sources of PM 2.5 and their contributions were identified using backward trajectories, principle component analysis (PCA), and the receptor model of chemical mass balance (CMB).

Sampling Protocol
Seasonal 24 Quartz fibrous filters of 47 mm were used in the PM 2.5 samplers for collecting 12-h PM 2.5 for consecutive five days in each season. Daytime PM 2.5 sampling started from 07:00 to 19:00, while nighttime PM 2.5 sampling started from 07:00 to 19:00 (next day). After sampling, PM 2.5 filters were then transported back to Air Pollution Laboratory in the Institute of Environmental Engineering at National Sun Yat-sen University as soon as possible and temporarily conditioned in a thermostatic chamber with a constant temperature (T) of 20-25°C and a relative humidity (RH) of 40 ± 5% for at least 48 h for weighing by an analytical microbalance with the precision of 1 µg (MSA6.6S, Sartorius) to determine the mass concentration of PM 2.5 .

Chemical Analytical Methods
After weighing, PM 2.5 filters were divided into four identical portions for further chemical analysis. One quarter of the quartz fibrous filter was used for analyzing the watersoluble ions, which was put into a 50 mL PE bottle poured with 30 mL distilled deionized water (D.I. H 2 O) with conductivity > 18 MΩ to dissolve ionic species into water from PM 2.5 by an ultrasonic vibrator for approximately 120 min. An ion chromatography (IC) ( DX-120, Dionex) was then used to analyze the concentrations of major anions (F -, Br -, Cl -, SO 4 2-, and NO 3 -) with 1.8 mM Na 2 CO 3 /1.7 mM NaHCO 3 solution as an eluent. Another ion chromatography (ICS-1100, Dionex) was further used to analyze the concentrations of major cations (NH 4 + , K + , Na + , Ca 2+ , and Mg 2+ ) with 20 mM methane sulfonic acid as an eluent.
nother quarter of the quartz fibrous filter was used for analyzing metallic content, which was initially digested in a 20 mL mixed acid solution (HNO 3 :HClO 3 = 3:7) at a temperature of 150-200°C for 120 min, and then diluted to 25 mL with D.I. H 2 O. During the digestion process, D.I. H 2 O was added to the residual solution at least two times in order to eliminate the acidic content of the digestion solution. The metallic elements in PM 2.5 , including Cr, Mn, Fe, Ni, Zn, Cd, Pb, Mg, K, Ca, Ti, Al, As, V, Cu, Ba, and Sb, were then analyzed with an inductively coupled plasma/atomic emission spectrometry (ICP/AES) ( Optima 2000 DV, PerkinElmer).
The carbonaceous contents of PM 2.5 were analyzed with an elemental analyzer (EA) ( Vario EL III, Elementar). One quarter of the quartz fibrous filter was cut into two identical parts. One eighth of the quartz fibrous filter was preheated in advance by hot nitrogen gas (340-345°C) for 30 min to expel the organic carbon (OC) fraction, after which the amount of elemental carbon (EC) was then determined. Another eighth of the quartz fibrous filter was analyzed without preheating, and the carbon content was thus characterized as total carbon (TC). OC was further determined by extracting EC from TC.
Another quarter of the quartz fibrous filter was analyzed for anhydrosugars (i.e., levoglucosan, mannosan, and galactosan) and organic acids (i.e., oxalic acid, malonic acid, and succinic acid) with a high-performance ionic chromatography (HPIC) (ICS-5000+, Dionex). The quartz fibrous filters were initially extracted with the distilled deionized water with conductivity > 18.2 MΩ in a PE bottle under ultrasonic vibration for 120 min with a prebaked quartz fibrous filter (0.3 µm pore size). Anhydrosugars were analyzed by using 200, 100, 18, and 5 mM NaOH as the eluents at a flow rate of 1 mL min -1 .
Organic acids were analyzed by using 25 mM H 2 SO 4 and 5 mM NaOH as the eluents at a flow rate of 1 mL min -1 .

Backward Trajectory Simulation
In order to trace air masses transported toward the sampling sites, backward trajectories were plotted by using the wind field including wind direction and wind speed obtained from local air quality monitoring stations and weather stations set up by Taiwan Environmental Protection Administration (TEPA) and Taiwan Central Weather Bureau (TCWB), respectively. In this mode, the starting time of backward trajectory simulation was selected as the occurrence time of the highest PM 2.5 concentration during the sampling period, and traced back by the airflow trajectory till 24:00 in the current day. As shown in the output maps, 1 h was used as the time interval of the track data.

Principal Component Analysis (PCA)
Principal component analysis is one of the methods of multivariate statistical analysis, which has been widely applied to resolve the main loading factors in sequence for characterizing the principal components with larger variances. The principle of PCA uses the obtained eigenvalues and eigenvectors to screen for a larger variance of each component (Turias et al., 2006;Han et al., 2015). The level-of-correlation loading factor is shown below: (1) High correlation-loading factor higher than 0.7, (2) Moderate correlation-loading factor between 0.5-0.7, and (3) Low correlation-loading factor less than 0.5.
The principal component analysis applied chemical analytical data to determine the main loading factors. The largest loading factor is the first principal component, and followed by the sequence of the second, the third, and the fourth principal components. However, the main purpose of the principal component analysis was to apply less variance to represent big data, commonly simplify to 3-4 main loading factors.

Chemical Mass Balance (CMB) Receptor Model
The source apportionment of PM 2.5 was resolved by using a chemical mass balance receptor model based on the principle of mass conservation. A CMB Version 8.0 model was applied by verifying the following fit indices, such as χ 2 (≤ 4.0), R 2 (≥ 0.8), the percentage of mass (80-120%), and T statistics (≥ 2.0). The residual and uncertainty ratios were required between -2.0 and 2.0, and their values should be as close as zero, in order to obtain better analytical results. The higher degrees of freedom, the better the resolution of CMB analysis. In this study, the source apportionment of PM 2.5 was carried out by using CMB Version 8.0 model. Comprehensive knowledge of the physical and chemical characteristics of fine particles (PM 2.5 ) measured at both the source and the receptor sites are required to identify the types of PM 2.5 sources and quantify their contributions at the receptor site. The relationship between the concentrations of chemical species measured at the receptor site and those emitted from the sources can be presented by Eq. (1): where C i is the concentration of species i measured at the receptor site (µg m -3 ). α ij is the coefficient of species i generated or sunk during the transportation process. F ij is the mass fraction of species i in the profile of the source j (%). S j is the mass concentration of all species at the receptor site assigned to the source j (µg m -3 ). Fig. 2 illustrates the seasonal and diurnal variation of PM 2.5 concentration measured at three sampling sites in the Linhai industrial complex. The average PM 2.5 concentration at National Open University sampled in winter, spring, summer, and fall were 41.4, 38.5, 12.0, and 16.8 µg m -3 , respectively. While, those at Fongyang Elementary School were 30.1, 33.3, 19.1, and 29.0 µg m -3 , respectively, and those at Fonglin Elementary School were 38.6, 27.8, 13.2, and 25.4 µg m -3 , respectively. In addition, although the concentrations in summer and fall nearly approached the yearly standard of 15 µg m -3 by Taiwan Environmental Protection Agency, those were still far higher than the yearly average PM 2.5 concentration standard of 10 µg m -3 ruled by WHO, as illustrated in Fig. 2.

Spatiotemporal Variation of PM2.5 Concentration
The correlation coefficient of manual sampling (three PM 2.5 sampling sites of this study) and automatic sampling (three adjacent TEPA air quality stations) results was as high as 0.98.
In terms of spatial distribution, NOU had the highest average PM 2.5 concentration in winter and spring with the prevailing northeastern winds, and FY has the highest average PM 2.5 concentration in summer and fall with the prevailing southern winds, implying that NOU and FY located at the downwind sites of PM 2.5 sources were highly affected by local sources in the Linhai industrial complex.
The diurnal variation of PM 2.5 concentrations was highly influenced by the synoptic-scale meteorology and anthropogenic activities. This study found that nighttime Auto sampling PM 2.5 concentrations were mostly higher than those in the daytime at NOU and FY, but an opposite trend was observed at FL, in spring and summer, implying that sampling sites located at the downwind of the Linhai industrial complexhad higher PM 2.5 concentrations at nighttime. On the contrary, daytime PM 2.5 concentrations were always higher than those at nighttime in fall and winter. It is speculated that the three sampling sites were greatly influenced by human activities in the daytime in fall and winter. In order to further understand the spatiotemporal variation of PM 2.5 in the Linhai industrial complex, this study also conducted 24-h sampling of PM 2.5 in four seasons simultaneously at ten sampling sites, including National Open University ( Two local maximum of PM 2.5 concentrations were observed at HMG (98.0 µg m -3 ) and CPC (51.2 µg m -3 ) in winter while blowing prevailing northeastern winds. High PM 2.5 concentration at HMG was mainly attributed to ship exhausts at the main entrance of Kaohsiung Harbor, while CPC was located at the downwind site of two industrial complexes. Overall, the spatiotemporal distribution of PM 2.5 concentrations clearly showed a trend of PM 2.5 concentration at the downwind sites were always higher than those at the upwind sites, indicating that prevailing winds could blow both primary and secondary aerosols from the upwind region to the downwind sites.

Chemical Characteristics of PM2.5 Water-soluble Ion Analysis
In this study, PM 2.5 samples collected at National Open University, Fongyang Elementary School, and Fonglin Elementary School in four seasons were further analyzed for five chemical contents, including water-soluble ions, metallic elements, carbonaceous contents, anhydrosugars, and organic acids.
As illustrated in Fig. 4, WSIs accounted for 31.8-58.4%, 31.1-56.1%, 32.1-55.6% of PM 2.5 sampled at NOU, FY, and FL, respectively. For all PM 2.5 samples, the most abundant ionic species in PM 2.5 were SO 4 2-, NO 3 -, and NH 4 + , namely secondary inorganic aerosol (SIA). The seasonal variations of WSIs and PM 2.5 mass concentration were very much similar during the sampling period. Moreover, the concentrations of SO 4 2-, NO 3 -, and NH 4 + had a tendency to rise significantly since fall and winter. Overall speaking, SIA accounted for 66.0-82.9% of WSIs in PM 2.5 at all sampling sites (Table 1). As far as the seasonal variation of SIA, high concentrations of SIA were observed in winter at three sampling sites, while low concentrations of SIA were found in summer. The highest SIA concentration of 15.1 µg m -3 was observed at NOU at nighttime, while the lowest SIA concentration of 3.19 µg m -3 was observed at FL in the daytime. It implied that the particulate air quality of southern Kaohsiung was deteriorated mainly by ammonium sulfate ((NH 4 ) 2 SO 4 ) and ammonium nitrate (NH 4 NO 3 ), which were formed in the atmosphere by SO 2 , NO x , and NH 3 emitted from local sources and/or long-range transport. Similar to PM 2.5 concentration, SIA in PM 2.5 sampled at the downwind sites were commonly higher than those at the upwind sites, implying that air contaminants tended to accumulate in the downwind region and thus increase the concentrations of PM 2.5 .
In order to further ascertain the chemical characteristics of PM 2.5 in the Linhai industrial complex, the equivalent concentrations of SO 4 2-, NO 3 -, and NH 4 + in PM 2.5 were used to determine the neutralization rate (NR), sulfur oxidation rate (SOR), and nitrogen oxidation rate (NOR). The neutralization ratios (NRs) of PM 2.5 which represents the aerosol acidity originating from particulate NO 3and SO 4 2-, were firstly evaluated to analyze the acidity of PM 2.5 using the following equations: The SOR and NOR as the sulfur and nitrogen oxidation ratios are described by the following equations: where nss-SO 4 2is the excess non-sea-salt sulfate, calculated by subtracting the amount of SO 4 2in seawater from that of SO 4 2in fine atmospheric particles; nss-NO 3is the excess non-sea-salt nitrogen, calculated by subtracting the amount of NO 3in seawater from that of NO 3in PM 2.5 . The diurnal variation of NR, SOR, and NOR of PM 2.5 at three sampling sites in four seasons are summarized in Table 2. It clearly showed that the average NRs were always < 1.0 in all seasons, indicating that NH 4 + cannot solely neutralize nss-SO 4 2and NO 3in PM 2.5 . In these cases, PM 2.5 in southern Kaohsiung was thought of as acid particles since water-soluble ions were dominated by acidic species (e.g., SO 4 2and NO 3 -) during the sampling periods in all seasons.
Moreover, the average SORs were mostly > 0.25 in winter and spring, while those in summer and fall were always < 0.25. However, different results were found for NORs, showing that the average NORs were always > 0.1 in all seasons. Therefore, PM 2.5 was dominated by both secondary sulfate and nitrate in winter and spring, resulting in the deterioration of particulate air quality. It suggested that secondary sulfate and nitrate could be converted from SO 2 and NO 2 via  atmospheric chemical reactions and transported to southern Kaohsiung during the northeastern monsoon periods. During the winter and spring days, sulfate and nitrate existing in the atmosphere had pretty much high potential to react with ammonium and form secondary ammonium sulfate and ammonium nitrate, also known as secondary inorganic aerosols which could be formed chemically by homogeneous reaction (mainly gas-gas), heterogeneous reaction (including gas-solid and gas-liquid), and photochemical reaction in the atmosphere and thus increased the amount of SIA in PM 2.5 during the winter and spring days.
As shown in Table 1, the [Cl -]/[Na + ] molar ratios ranged from 0.85 to 1.03, while the chloride deficit ranged from 43.9 to 53.4%, during the sampling periods, suggesting that PM 2.5 was less affected by the oceanic spray. Such a phenomenon was mainly attributed to the fact that the PM 2.   ] is about 1.8 (or the molar ratio of 1.17) of sea salts in the oceanic spray mainly due to low chloride deficit (Cheng et al., 2000;Seinfeld and Pandis, 2006). Chloride could react with acidic substances (e.g., HNO 3 , H 2 SO 4 , and other acids) and form HCl which is then evaporated to the atmosphere. The mass ratio of [Cl -]/[Na + ] is thus reduced gradually while sea salts migrate toward inland in the atmosphere due to chloride deficit. The chloride deficit can be determined by the following Eqs. (5) and (6) (Kerminen et al., 1998): In terms of seasonal variation, the chloride deficits of PM 2.5 at National Open University, Fongyang Elementary School, and Fonglin Elementary School in winter and spring were consistently higher than those in summer and fall ( Table 1). The results indicated that chloride deficit was correlated to East Asian monsoons, mainly due to high atmospheric concentrations of acidic gases in southern Kaohsiung during the northeastern monsoon seasons (Yuan et al., , 2018. On the contrary, in summer and fall, the southwestern monsoons blow relatively clean air masses from the seas to the sampling sites, resulting in lower atmospheric concentrations of acidic gases and thus lower chloride deficit. Fig. 5 illustrates the seasonal and diurnal variation of metallic element concentrations in PM 2.5 sampled at three sampling sites in southern Kaohsiung. It showed that crustal elements (Ca, K, Mg, Fe, and Al) were the dominant metallic elements in PM 2.5 . The seasonal variation of metallic elements had the same trend as PM 2.5 concentration. In terms of spatial distribution, the concentrations of metallic elements at National Open University were generally higher than those at Fongyang Elementary School and Fonglin Elementary School. Additionally, trace elements (Ti, V, Cr, Mn, Ni, Cu, As, Pb, and Cd) emitted mainly from anthropogenic sources were lower than crustal elements in all seasons.

Metallic Element Analysis
Diurnal variation results showed that daytime metallic elements in PM 2.5 were mostly higher than those at nighttime particularly in winter and spring, while an opposite trend was observed in summer and fall. The metallic contents of PM 2.5 were dominated by crustal materials (Ca, Fe, Al, and Ti), oceanic sprays (Mg and K), anthropogenic sources (Cd, V, Cu, Cr, Pb, and As), and mobile sources (Zn), as shown in Fig. 5. The diurnal variation of Mg, K, Ca, V, Fe, Cu, Zn, Al, As, Cd, and Pb mainly emitted from crustal materials, oceanic sprays, and vehicular exhausts were insignificant. However, a significant diurnal variation of trace elements (Ti, Cr, Mn, and Ni) was observed, suggesting that the anthropogenic emission from industrial process influenced the metallic component of PM 2.5 (Shen et al., 2019).
This study further ascertains the characteristics of metallic elements in PM 2.5 by using enrichment factor (EF), which is defined as the relative abundance of metallic elements to reference element to explore the correlation between PM 2.5 and crustal materials at the sampling sites. In the present study, calcium (Ca) as the most abundant crustal metal in PM 2.5 was used as the reference element to determine EF (Yuan et al., 2006) and identify whether PM 2.5 was dominated by crustal materials or anthropogenic sources (Chester et al., 2009). If EF > 10, metallic elements in PM 2.5 were contributed mainly from anthropogenic sources. If EF < 10, metallic elements in PM 2.5 were derived mainly from crustal materials.
The enrichment factors of metallic elements in PM 2.5 sampled at the sampling sites are depicted in Fig. 6. It showed that the EF S of Ni and Cr elements were mostly > 10 in all seasons except for fall, while the EFs of Cu and Zn were mostly < 10 in all seasons except for spring. The results indicated that trace metals, including Ni, Cr, Cu, and Zn, came mainly from anthropogenic sources. Previous literature reported that Ni and Cr came mainly from ironworks, metal processing, and coal burning, while Cu and Zn came mainly from smelters and metal processing. Unlike Ni, Cr, Cu, and Zn, the EF S of Al, Ti, Mg, Fe, Pb, K, and Ba were always < 10, indicating that these metallic elements were derived mainly from crustal sources. Overall speaking, the most severe metallic pollution contributed to PM 2.5 was observed in spring, while the lowest one occurred in fall.

Carbonaceous Content Analysis
The seasonal and diurnal variation of carbonaceous content of PM 2.5 sampled at three sampling sites is illustrated in Fig. 7. Similarly, the diurnal variation of carbonaceous contents and mass concentrations of PM 2.5 were quite similar. It showed that the concentrations of OC in PM 2.5 were always higher than those of EC at three sampling sites in all seasons. Among them, EC mainly came from inert carbonaceous materials, diesel engines, fuel combustion, industrial combustion, and exhaust gases (Kim et al., 1999). Elemental carbon has a chemical structure similar to impure graphite, originating primarily from direct emissions of combustion (Chen et al., 2016). Furthermore, the mass ratios of organic carbon and elemental carbon (OC/EC) in winter and spring were always higher than those in summer and fall (Fig. 7). Since fall, the OC/EC mass ratios increased gradually at all sampling sites. This study further used the minimum of OC/EC mass ratios ((OC/EC) min ) (see Fig. 8) to estimate the secondary organic carbon (SOC) in PM 2.5 (Cao et al., 2004). The concentration of SOC can be determined by either Eq. (7) or (8) (Kerminen et al., 1998): Herein, the SOC of PM 2.5 can be obtained by subtracting primary organic carbon, (EC) × (OC/EC) pri , from OC. However, in reality, (OC/EC) pri in Eq. (4) was not available, thus (EC) × (OC/EC) min was deemed as primary organic carbon (POC) in this study (Wu et al., 2009). In practice, the (OC/EC) min is readily obtained from the minimum slope of the OC versus EC, as illustrated in Fig. 8.
The seasonal and diurnal variation of estimated SOC concentrations in PM 2.5 at three sampling sites in southern Kaohsiung is illustrated in Fig. 9. It showed that the concentrations of SOC in winter and spring were commonly higher than those in summer and fall, which caused a severe impact on ambient carbonaceous aerosols along the coastal regions during the northeastern monsoon periods. Elemental carbon could also be generated from atmospheric photochemical reactions during long-range transport, which formed SOC in the atmosphere . In addition to long-range transport, volatile organic compounds emitted from industrial complexes were responsible for the formation of SOC during the winter days, since atmospheric photochemical reactions could convert VOCs to SOC.

Anhydrosugar Analysis
Anhydrosugars, including levoglucosan, mannosan, and galactose, are the most efficient organic tracers for organic aerosols emitted from biomass burning in the atmosphere. As illustrated in Fig. 10, levoglucosan was the only anhydrosugar   being detected, while mannosan and galactose were not detectable (ND) in PM 2.5 in the present study. The average concentrations of levoglucosan in 12-h PM 2.5 samples at all sampling sites ranged from 30.3 to 55.3 ng m -3 . High concentrations of levoglucosan in the range of 47.2-54.9 ng m -3 were observed mainly in spring, showing that biomass burning was one of the dominant contributors to atmospheric PM 2.5 , which concurred with the more frequent occurrence of biomass burning in spring than other seasons in the Indochinese Peninsula, as shown in Fig. S1 Bhattarai et al., 2019). Moreover, no apparent diurnal variation trend of levoglucosan concentration was observed at all sampling sites in four seasons. In terms of spatial distribution, the concentrations of levoglucosan were ordered as NOU > FL > FY in spring since the polluted air mass was blown northerly from northern Taiwan and/or long-range transported from southeastern and northeastern Asia. However, an opposite trend was observed in the other three seasons (Fig. 10).

Organic Acid Analysis
The seasonal and diurnal variation of organic acids in PM 2.5 at three sampling sites in southern Kaohsiung is depicted in Fig. 11. It showed that the seasonal variation of organic acid concentrations was similar to PM 2.5 mass concentrations. Overall, the concentrations of organic acids in winter and spring were commonly higher than those in summer and fall. Among the three measured organic acids, oxalic acid was the most abundant organic acid in PM 2.5 . The concentrations of oxalic acid at three sampling sites ranged from 176.2 to 647.7 ng m -3 in winter and spring. Previous literature reported that the mass ratios of malonic and succinic acids (M/S) have been applied to determine whether atmospheric PM 2.5 was mainly attributed to secondary organic aerosol (Kawamura et al., 1993). This study revealed a significant seasonal variation of M/S mass ratios. During the sampling periods of winter and spring, the M/S mass ratios of PM 2.5 ranged from 0.98 to 1.21, which were much higher than those ranging from 0.49 to 0.81 in summer and fall. High M/S mass ratios indicated that PM 2.5 in winter and spring was highly correlated with secondary organic aerosols that were long-range transported from the far north toward southern Kaohsiung (Tsai et al., 2013;Dhananjay et al., 2019).

Simulated Backward Trajectories
In the present study, backward trajectories were applied to trace air masses transported toward southern Kaohsiung. Air trajectories were plotted by using Xiaogang air quality monitoring station (located at Xiaogang Junior High School) as the ending point for tracking the backward trajectories . In winter (on December 4-8), air trajectories came mainly from western Taiwan (i.e., Taichung, Yunlin, and Chiayi), which were transported from the north toward the Xiaogang site on December 4-5. However, air trajectories on December 6 and 8 were blown from the Taiwan Strait, turned from the north to the south, and then moved toward the Xiaogang site.
In spring (on March 12-16), air trajectories on March 12-14 came from the Taiwan Strait, passed through southern Tainan, and arrived at southern Kaohsiung, which transported mainly from the north to the south. Air trajectories on March 15 transported from Pingtung County from the western Taiwan Strait, which mainly transported from the east to the west. While air trajectories on March 16 were transported from the northwestern to the southeastern Taiwan Strait.
In summer (on July 10-13), air trajectories were relatively disordered compared to the other three seasons. Air trajectory on July 10 was mainly transported from the northwestern to the southeastern Tainan and then moved toward the Xiaogang site. Air trajectory on July 11 came from the Taiwan Strait, which was blown to the eastern and western Pingtung and then transported toward the Xiaogang site. Air trajectory on July 13 was blown from the northwestern to the southeastern Taiwan Strait.
In fall (on November 17-21), air trajectory on November 17 came from the northeastern to the southwestern Taitung, crossing the Central Range, and then blown downward the Xiaogang site. Air trajectory on November 18 was transported from the north to the south to Pingtung County, and finally blown from the east to the west and moved toward the Xiaogang site. Air trajectory on November 19 was blown mainly from the western coastline of Taiwan and moved from the north to the south. Air trajectories on November 20-21 were mainly transported from the Kaohsiung offshore toward the Xiaogang site, which moved mainly from the west to the east.

Source Identification and Apportionment of PM2.5
The identification and apportionment of PM 2.5 sources based on PCA at three sampling sites are shown in Table 3. The PCA result showed that the chemical species statistically clustered in each major factor were distinguished in different seasons. In winter, the highest principal component (PC1) contained Mg 2+ , Al, Ti, Pb, and Ba, explaining 37.1% of the total variance, which were closely associated with steel plants or metal processing (Amodio et al., 2013;Zhou et al., 2019). The second principal component (PC2) included NO 3 -, SO 4 2-, NH 4 + , and OC, accounting for 14.1% of the variance, were identified as secondary aerosols. The third and fourth principal components (PC3 and PC4) consisted of Cl -, Na + , and Mg, Ca were linked to oceanic spray and fugitive dust, respectively. In spring, PC1 with the total variance of 33.4% consisted of NH 4 + , Ni, Cu, and EC, which were probably related to coal-burning boilers. PC2 that had V and Pb, explaining 10.9% of the variance, represented metal processing. In addition, V was also probably emitted from coal burning (Pei et al., 2016). PC3 with K and Ba, accounting for 10.2% of the total variance, was probably emitted from steel plants. PC4 with Na + and Clstood for ocean sprays. In summer, the PC1 made up of Al, Ti, V, Mn, Pb, Cu, and Ba, accounting for 45.7% of the total variance, was attributed to steel plants or metal processing. PC2 (12.6%) and PC3 (8.3%) containing NH 4 + , and K, SO 4 2-, Ca 2+ , and Ca, respectively, both of which were closely related to road dust (Zhang et al., 2018). Na + and Clin PC4 (7.1%) mainly originated from the oceanic spray. In fall, PC1 (32.4%) involving OC and EC was directly related to vehicular exhausts (Watson et al., 2001;Corbin et al., 2018;Li et al., 2018). Metallic elements of Pb, Ba, Cr in PC2 (14.8%) were also considered to be emitted from steel plants or metal processes. V, Fe, and Ni in PC3 (13.0%) were considered to be bound in the primary fine particles emitted from fuel oil combustion or vehicular exhausts (Khan et al., 2016;Elhadi et al., 2018). The metallic elements of Ti and Al in PC4 (8.5%) mainly come from road dust. To sum up, the chemical components of PM 2.5 statistically resolved majorly originated from anthropogenic activities in each season, while that derived from secondary aerosols was only found in the winter season, suggesting that photochemical conversion of the precursors were probably more intense in winter (Shen et al., 2019).
The chemical species screened by PCA were selected for further analysis using the CMB receptor model. The results of resolved prominent sources of PM 2.5 at three sampling sites showed that the largest source of PM 2.5 in the Linhai industrial complex was steel plants, accounting for 19.8-26.5% of total PM 2.5 , suggesting that metal processing, blast furnace, fuel combustion, etc. play a vital role in the emission or formation of PM 2.5 . Secondary aerosols, as the second largest source, contributed 13.3-17.8% of total PM 2.5 . It is worth noting that the proportion of secondary sulfates exceeded that of secondary nitrates, revealing that the formation of PM 2.5 was more influenced by stationary sources than vehicular exhausts in the Linhai industrial complex. Additionally, in summer and fall, the proportion originating from secondary aerosol contribution most, about 13.3-15.4% of PM 2.5 , while in winter and fall that contributed rose to 16.8-17.8%, suggesting that the northeastern monsoon periods (in winter and spring), long-range transport or poor dispersion condition resulted in the formation of secondary aerosols and thus rose their contribution to PM 2.5 .
Other anthropogenic sources of petrochemical industries (8.9-13.4%), industrial boilers (9.0-10.7%), vehicular exhausts (7.4-11.8%), and incinerators (6.1-8.3%) also dominated the PM 2.5 source profile (Fig. 14). Besides the industrial activities, the contribution of biomass burning in winter and spring, contributing 4.4-4.5% of PM 2.5 was commonly higher than that of 3.2-3.4% in summer and fall, suggesting more agricultural activities in winter and spring than those in summer and fall. To sum up, the PM 2.5 around the Linhai industrial complex was mainly influenced by primary aerosol emitted from anthropogenic activities, and subsequently by secondary aerosols.

CONCLUSIONS
This study investigated the seasonal and diurnal variation in the PM 2.5 concentration and chemical composition in the Linhai industrial complex in southern Kaohsiung. The average concentrations ranged from 27.8 to 31.4 µg m -3 during the northeastern monsoon periods of winter and spring but were much lower, between 12.0 and 29.0 µg m -3 , during summer and fall. The concentrations at the downwind sites always exceeded those at the upwind sites. The SORs were frequently > 0.25 during winter and spring, whereas the NORs were always > 0.1 regardless of the season. Al, Mg, Fe, Ti, Pb, K, and Ba were the dominant metallic elements in the PM 2.5 , and the trace elements, Ni, Cr, Cu, and Zn, were mainly emitted from anthropogenic sources. The OC/EC ratios were higher during winter and spring than summer and fall, and the SOC concentration displayed a similar trend. High concentrations of levoglucosan, ranging from 47.2 to 54.9 ng m -3 , were observed in winter and spring due to biomass burning during the northeastern monsoon periods, and the organic acids in the PM 2.5 also significantly increased during these two seasons. The high M/S ratios for the PM 2.5 , which ranged   from 0.98 to 1.21, were attributable to long-range-transported primary and secondary organic compounds. The prominent resolved PM 2.5 sources in the Linhai industrial complex were steel plants (19.8-26.5%), followed by secondary aerosols (13.3-17.8%), petrochemical industries (8.9-13.4%), industrial boilers (9.0-10.7%), vehicular exhaust (7.4-11.8%), and incinerators (6.1-8.3%). During the northeastern monsoon periods (in winter and spring), primary pollutants emitted locally and/or long-range transported toward southern Kaohsiung resulted in the formation of secondary inorganic and organic aerosol, thereby increasing their contributions to the PM 2.5 .