Predictive Meteorological Factors for Elevated PM 2.5 Levels at an Air Monitoring Station Near a Petrochemical Complex in Yunlin County, Taiwan

Since 1991, air pollution has gained special attention in Taiwan after a petrochemical complex was constructed in Mailiao Township, Yunlin County. We explored the association between the magnitude of PM 2.5 and meteorological factors during 2012-2016. Our findings revealed that 1) mean PM 2.5 levels gradually decreased from 30.70 μg/m 3 in 2013 to 25.36 μg/m 3 in 2016; 2) wind speed is the main determinant of air quality—air quality significantly im-proved when it was faster than 4 m/sec; and 3) wind direction is another determinant of air quality— when the wind direction was southerly, air quality improved. Elevated PM 2.5 levels were defined as those hourly levels higher than the third quartile (36 μg/m 3 ). The significantly negative predictive factors for elevated PM 2.5 levels were the summer or autumn seasons, rainfall, increased wind speed, and wind direction from 150˚ to 230˚ from the north. The significantly positive predictive factors for elevated PM 2.5 levels were working hours from 6 a.m. to 2 p.m., a temperature between 11˚C and 25˚C, relative humidity between 40% and 68%, and wind direction (e.g., northerly wind, northeasterly


Introduction
Air pollution is a major health concern and has been extensively studied. Exposure to pollutants such as fine particulate matter (PM 2.5 ) and ozone has been associated with an increase in mortality and hospital admissions due to respiratory and cardiovascular diseases [1] [2] [3] [4]. A study reported that PM 2.5 causes approximately 3% of mortality due to respiratory cancer (cancer of the trachea, bronchus, and lung) and approximately 1% of mortality due to respiratory infection in children aged under 5 years; this amounts to approximately 0.8 million deaths and 6.4 million years of life lost; furthermore, this burden occurs predominantly in Asia (65%) [5]. The global burden of disease attributable to outdoor air pollution uses the annual average concentration of PM 2.5 as the indicator of air pollution [6]. Only 1% of global PM 2.5 exposure occurs in outdoor environments in the developed world, whereas a staggering 14% occurs in outdoor environments in the developing world [7]. Another study reported that ambient PM 2.5 pollution is a major mortality risk factor in Taiwan. The same study also reported that substantial geographic variations in PM 2.5 attributable to the mortality fraction were found, and Yunlin County had the highest percentage (21.8%) of deaths attributable to PM 2.5 [8].
According, the PM 2.5 pollution was an important issue for the citizens of Yunlin County. It is the motivation of this study. According to the statistics of the Environmental Protection Administration, Executive Yuan (Taiwan), of the PM 2.5 emissions from various sources, motor vehicles emissions accounted for 36%, overseas imports accounted for 27%, industry (coal-fired power generation, petrochemical, and steel-making) accounted for 25%, and other sources accounted for 12% [9].
Air quality is severely influenced by weather conditions. A 19-year observational study reported that 14,700 excess deaths from PM 2.5 were attributable to weather-related increases in air quality in the United States [10]. Studies [11]. The orography is played a pivotal role over the variations of PM [12]. Taiwan is located in an area affected by the monsoon climate. During summer, it is affected by the moist and warm air flow brought by the southwest monsoon, and the average temperature reaches 28˚C in July. Because Mailiao Township is located near the coast, the usual wind direction is westerly (from sea to land) during the daytime and easterly (from land to sea) during the nighttime.
In this study, we explored the effects of meteorological variables, such as temperature, wind speed and direction, RH, and daily rainfall on elevated PM 2.5 levels (higher than the third quartile) of all hourly records at the air monitoring station in Mailiao during 2012-2016. The novelty and contributions of this study were meteorological factors such as wind speed and direction, and daily rainfall on the effect of elevated PM 2.5 levels near the huge petrochemical complex.

Meteorological Records
First, the hourly mean data for temperature, wind speed and direction, RH, rainfall, and PM 2.5 from the air quality monitoring station were downloaded from the website of the Environmental Protection Administration; these data were open for public use. Second, we downloaded data from the nine air quality monitoring stations within 50 km for hourly spatial changes [13].

Data Were Recorded by Hour
Concentration of PM 2.5 (μg/m 3 ): Since 2012, hourly mean concentrations have been continually measured at the monitoring station using a β-ray attenuation method for PM 2.5 . The instrument used for PM 2.5 analysis was VEREWA F701 [13]. The standard "satisfactory rate" for PM 2.5 results was 99% in 2012. An accuracy difference between the hand-standard results and automatic monitoring method results of less than 9% is considered "satisfactory" [13].
Wind speed and direction: The hourly mean wind speed was measured using cup anemometers (Model 014A; Met One Instruments, Inc., Grants Pass, USA), and the mean wind direction was measured using a wind vane (Model 024A; Met One Instruments, Inc., Grants Pass, USA) [14]. The means of wind speed and direction for day, month, and seasons were calculated using a vector method.
Seasons: We defined seasons as spring (February to April), summer (May to July), autumn (August to October), and winter (November to January).
The study protocol was reviewed and approved by the Research Ethics Committee of Buddhist Dalin Tzu Chi Hospital, Taiwan (No. B10601004).

Statistical Analysis
All statistical operations were performed using the R 3.0. We assessed the goodness of fit of the final logistic regression model according to the estimated area under the receiver operating characteristic curve (AUC). Statistical tools of regression diagnostics were applied to discover any problems associated with the regression model or data. Two-sided p ≤ 0.05 was considered statistically significant. Hourly mean data of the nine air monitoring stations within 50 km were downloaded and stratified by seasons. The Kriging method was applied to estimate the spatial data to explore changes by hour.

Results
In total, 43,847 hourly records from 2012 to 2016 from the air quality monitoring station in Mailiao were used; 42,499 (96.9%) records were enrolled for analysis after excluding the records with missing data. The third quartile for PM 2.5 was set at 36 μg/m 3 , which was defined as elevated PM 2.5 levels in this study. The mean PM 2.5 levels at this station had a lower trend according to Theil-Sen analysis (p < 0.001) ( Table 1).
The daily means of PM 2.5 were higher at the end of the year and beginning of the following year and lower during the summer ( Figure 1). Moreover, the data stratified by months showed lower levels during summer ( Figure 2).   The data were stratified and compared by season and daytime or nighttime.
The means of PM 2.5 were significantly lower during summer as well as daytime and nighttime (p < 0.001) than during other seasons. Spring and summer are the rainy seasons in Taiwan (from 7.0% to 8.6%) (p < 0.001). The hottest season is summer (28.7˚C ± 2.1˚C), followed by autumn (25.7˚C ± 3.1˚C) and the coldest season is winter (18.0˚C ± 3.0˚C) (p < 0.001). The percentage of higher PM 2.5 levels was lowest in the summer. The mean RH (77% -87%) is the highest during summer. Regarding wind direction, the mean wind directions were southerly or southeasterly during summer. During autumn and winter, the mean wind directions were easterly to northeasterly during both daytime and nighttime. During spring, the mean wind direction was southerly during the daytime and easterly during the nighttime. The highest mean wind speed was noted at nighttime during winter (1.81 m/s), and the lowest mean wind speed was noted in the daytime during spring (0.37 m/s) ( Table 2). The hourly analysis showed elevated     (Table 3). Regarding easterly winds, the hourly spatial data of the surrounding nine monitoring stations within 50 km were stratified by seasons and analyzed using the Kriging method ( Figure 4). During the daytime, air pollution was blown inland by westerly winds or diffusion. During the nighttime, easterly land winds blew the pollution into the area of the studied monitoring station.
The relationships between the air quality monitoring stations are shown in

Discussion
The novel finding of this study was that the annual mean PM 2.5 concentration had significantly negative trend from 30.70 μg/m 3 in 2013 to 25.36 μg/m 3 in 2016. The contributions of this study were wind speed and direction, and daily rainfall on the effect of elevated PM 2.5 levels near the huge petrochemical complex. Daytime (from 6 a.m. to 2 p.m.) was the significant factor associated with hourly PM 2.5 concentrations higher than the third quartile (36 μg/m 3 ). Another key proposed approach was the hourly spatial data of the surrounding nine monitoring stations within 50 km were stratified by seasons and analyzed using the Kriging method. Another key finding was that the significantly positive meteorological factors associated with hourly levels of PM 2.5 higher than the third quartile (36 μg/m 3 ) were cool weather (temperature between 11˚C -25˚C or 51.8˚F -77˚F), RH of 40% -68%, northerly winds (330˚ -360˚ and 0˚ -30˚ from the north), northeasterly winds (30˚ -60˚ from the north), easterly winds (60˚ -90˚ from the north) with a wind speed of 4 -6 m/s, and easterly winds (90˚ -120˚ from the north) with a wind speed of 2 -4 m/s. Furthermore, the significantly negative meteorological factors associated with higher PM 2.5 levels were summer and autumn, daily rainfall, wind speed, and southerly winds (150˚ -230˚ from the north) with a wind speed of m/s. These findings were further discussed in the below sections. After decades of industrialization, air pollution has become a major environmental problem in Taiwan. Poor air quality has both acute and chronic effects on human health. In 2012, the Air Pollution Control Act (APCA) was formulated to control air pollution, maintain public health and living environments, Figure 6. Area under the receiver operating characteristic curve was 0.783 for predicting elevated PM 2.5 levels in our study. Open Journal of Air Pollution and improve the quality of life in Taiwan [15]. The improvement in annual PM 2.5 concentration might contribute to the APCA's formulation. In Taiwan, when PM 2.5 levels are higher than 36 μg/m 3 , the air quality indicator is yellow; people with a history of heart, respiratory, and cardiovascular disease are vulnerable to elevated PM 2.5 levels and should decrease their outdoor activities 90 [13]. In this study, the third quartile for PM 2.5 was set at 36 μg/m 3 , which was defined as elevated PM 2.5 levels. In this study, the daily average level of PM 2.5 was 26.69 ± 15.88 μg/m 3 . Although this data was slightly higher than the World Health Organization (WHO) ambient air quality guidelines (25 μg/m 3 for the 24-hourly mean) [16], it was lower than the standard level (35 μg/m 3 ) in Taiwan.

Diurnal Variation: Daytime Had Elevated PM2.5 Levels
In this study, we observed a diurnal variation in PM 2.5 with elevated levels during the daytime. This result differs from another study that reported that PM 2.5 levels were higher in concentration during the nighttime and lower during the daytime in Beijing [17]. The reasons for this might be work-related, such as working in a petrochemical complex and high human activities, and the usual wind direction is westerly (from sea to land) during the daytime. In this study, daytime was defined from 6 a.m. to 5 p.m. However, working hours defined as 8 a.m. to 5 p.m.
The phenomena of PM 2.5 concentrations start to rise from 6:00 a.m. until 2:00 p.m. that cannot be easily explained by atmospheric condition or emission, mobile-source influence. It is one of limitations in this study.

Wind Direction (Northerly or Northeasterly; 330˚ -360˚, 0˚ -30˚, or 30˚ -120˚)
A relevant study reported that wind direction is a critical parameter affecting PM 2.5 levels [18]. The current study reported that wind speed was independently and negatively correlated with elevated PM 2.5 levels, but that different wind directions had different effects associated with elevated PM 2.5 levels, as shown in another study [19]. In this study, we found that northerly or northeasterly winds (0˚ -120˚ and 330˚ -360˚ from the north) were positively correlated with elevated PM 2.5 levels (OR: 1.57, 95% CI: 1.43 -1.74). This might be because the monitoring station is located to the southeast of the petrochemical complex, and also because of the monsoon climate in Taiwan (northeast in winter and southeast in summer). When the wind blew from the south or southwest (150˚ -230˚ from the north), higher wind speed correlated negatively with PM 2.5 levels. These wind directions might be caused by atmospheric influences such as the monsoon climate. When the wind blew from the east (60˚ -90˚ from the north) with a wind speed between 4 -6 m/s as well as from the east to southeast (90˚ -120˚ from the north) with a wind speed between 2 -4 m/s, this was positively associated with elevated PM 2.5 levels (OR: 1.45, 95% CI: 1.22 -1.73). The spatial hourly data of the surrounding nine stations were analyzed using the Kriging method [20]. A possible explanation might be that PM 2.5 air pollution could be diffused from higher concentrations to lower areas as well as by westerly winds Open Journal of Air Pollution during the daytime (from sea to land). The usual land wind blew the air pollution during the nighttime.

Season (Summer and Autumn)
A study reported that elevated PM 2.5 levels occurred during the winter and spring [21]. In the current study, we found that summer and autumn were correlated negatively with PM 2.5 levels. Although higher temperatures, more rainfall, higher wind speeds, and lower RH occurred during the summer and autumn, this finding may be explained by the climate in Yunlin being affected by the northeast monsoon during the winter and spring (November to April); furthermore, the wind direction is mainly north-northeasterly, followed by northeas-

Cool Weather (11˚C -25˚C or 51.8˚F -77˚F)
In the current study, we found that cool weather (11˚C -25˚C) was associated with higher PM 2.5 levels, compared with temperatures lower than 11˚C and higher than 25˚C, an inverted U-shaped effect. One study reported that temperature was positively correlated with PM 2.5 concentration in four seasons in Nagasaki, Japan (Wang and Ogawa, 2015) [18]. Another study reported that temperature has a negative relationship with PM 2.5 in summer and autumn and then turned to positive in spring and winter in Nanjing, China [22]. In Chen's study, the relation between PM 2.5 and temperature was not linear, likely an inverted V-shaped effect [22], which was similar our result. However, the different phenomena cannot be easily explained by atmospheric condition, photochemical activity, or emission influence. Therefore, the mechanism between temperature and PM 2.5 concentrations warrant in the future study.

RH (40% -68%)
In the current study, we found that RH between 40% and 68% was a positive factor for PM 2.5 levels above the third quartile. A similar finding was reported in that RH had an inverted U-shaped relationship with PM 2.5 concentration (peaking at an RH of 45% -70%) [23]. In Lou's study reported that the dry (RH = 45% -60%) and low-humidity (RH = 60% -70%) conditions are positively affected PM 2.5 and exerted an accumulation effect (Lou et al., 2017). Previous study also reported that positive correlations between RH and PM 2.5 were identified [24].
The explanation might be strong evaporation and transpiration in the presence of water or wetlands could form a microclimate with lower temperature and higher RH compared with the surrounding environment, which may decrease the gas-to-particle conversion rate and favor particle deposition [25]. Open Journal of Air Pollution

Rainfall
A study reported that rainfall had washout effects on atmospheric particulate pollution and was recognized as one of the main mechanisms for reducing PM 2.5 pollution [26]. Rainfall possesses a threshold and also the lag effect for reducing PM 2.5 [26]. In the current study, we found that rain was the independently negative factor for PM 2.5 levels above the third quartile.

Limitation
Our study has some limitations. First, the data about the original sources of PM 2.5 included atmospheric condition, emission, or mobile-source influence with multi-resources data such as pollution source information, real-time population grid data, meteorological data, and traffic data were not available in this study, and that is a limitation. Second, we find some phenomena that cannot be easily explained by atmospheric condition, emission, or mobile-source influence.
Therefore, further studies should be carried out with multi-resources data, such as pollution source information, real-time population grid data, meteorological data and traffic data, to provide reasonable interpretations for these unexplainable phenomena. Third, the result cannot apply to other places, however we provided a formula based on the final multiple logistic regression model in the Appendix to calculate the probability of elevated PM 2.5 level.
Furthermore, we found that meteorological factors were associated with elevated PM 2.5 levels, and provided a formula for calculating the probability of elevated PM 2.5 (>36 μg/m 3 ) levels at this key monitoring station. Similar functions based on our method could be established to calculate the probability of elevated PM 2.5 levels in other cities. Further studies should investigate multi-resources data, such as PM 2.5 source information, real-time population data, and traffic data, which may provide deeper interpretations for air pollution prediction.

Conclusion
The annual PM 2.5 concentration gradually decreased from 30.70 μg/m 3 in 2013 to 25.36 μg/m 3 in 2016. The meteorological conditions have important effect on PM 2.5 mass concentration. We found that the factors associated with elevated PM 2.5 levels were daytime, cool weather (11˚C -25˚C), winter and spring, RH between 40% -68%, lower wind speeds, wind direction, and days with no rain.
Furthermore, by the relationship with meteorological conditions, it can be depicted that daytime with cool weather, no rain, relatively dry (RH = 40% -68%), lower wind speeds (breeze), northerly wind, and in winter and spring brought the most pollutants to Yunlin county; and daytime in summer, with rain, relatively humidity (RH > 68%), higher wind speeds, and southerly wind brought the lower probabilities of pollution to Yunlin county. If people want to know the more accurate probability of elevated PM 2.5 level (>36 μg/m 3 ), the predictive formula is attached in the Appendix. Therefore, people should protect them-