Modeling and Analysis of Source Contribution of PM10 during Severe Pollution Events in Southern Taiwan

This work simulates the hourly variations of PM10 (suspended particles with diameter < 10 μm) during severe pollution events in southern Taiwan (Kaohsiung City and Pingtung County) in spring, autumn and winter of 2005 by using the Air Pollution Model (TAPM). Comparisons between simulations and measurements at three sites (industrial, urban and rural) were satisfactory. The synoptic weather chart indicated that prevailing winds were northwest (spring), north (autumn), and northeasterly (winter). Meteorological conditions suggest that PM10 typically accumulated and triggered a pollution episode on days with high surface pressure and low winds. Estimations using the TAPM model suggest that point-source emissions were the predominant contributors (about 49.1%) to PM10 concentrations at Hsiung-Kong site industrial site in Kaohsiung City, followed by area sources (approximately 35.0%) and transport from neighboring areas (7.8%). Because Pingtung City (urban) and Chao-Chou town (rural) are located downwind of Kaohsiung City when north or northeasterly winds prevail, the two sites also experience severe pollution events despite the lack of industrial sources; transport from neighboring areas contributed roughly 39.1% to PM10 concentrations at Pingtung site and 48.7% at Chao-Chou site. Since traffic emissions contributed little (around 8%) to PM10 concentrations at the three sites, reducing PM10 emissions from industrial sources in Kaohsiung City should be an effective way of improving air quality for Kaohsiung City and downwind areas such as Pingtung County.


INTRODUCTION
Kaohsiung City, in southern Taiwan, is a heavily industrialized harbor city, with an area of 153.6 km 2 and a population of 1.49 million.Six major industrial parks are located in and around Kaohsiung City (Fig. 1).These parks are home to oil refineries, petrol/plastic industries, power plants, iron/steel/metal plants, recycling factories, and large municipal waste incinerators.Due to intensive industrial and traffic activities, the Kaohsiung metropolitan area has the poorest air quality in Taiwan − either increased ground-level concentrations of particulate matter (PM) or ozone (O 3 ) associated with unfavorable meteorological conditions − particularly between late fall and mid-spring (Chen et al., 2004).Pingtung County, to the south of Kaohsiung City, is primarily agricultural with several small industries located in northern Pingtung County.However, air quality in northern (e.g., Pingtung City) and central (e.g., Chao-Chou town) Pingtung County is as bad as that in the Kaohsiung area, despite the fact that population densities, traffic volumes and industrial emissions in Pingtung County are significantly lower than those of the Kaohsiung area.This poor air quality is mainly because northern and central Pingtung County are downwind of the Kaohsiung area when north or northeasterly winds prevail typically in autumn and winter (Chen et al., 2003(Chen et al., , 2004)).
Primary PM is emitted directly into the atmosphere by anthropogenic sources (e.g., industry, vehicles, combustion sources, bare lands and open burnings), and natural sources (e.g., volcanic eruptions, wildfires and marine aerosols).Atmospheric PM can carry acids and toxic species (e.g., polycyclic aromatic hydrocarbons and heavy metals) and can have adverse effects on human health (Cheng et al., 1996).
Epidemiological studies have demonstrated a strong relationship between elevated concentrations of PM 10 and mortality and morbidity (Lin and Lee, 2004;Arditsoglou and Samara, 2005).Identification of air pollution sources is important in developing clean-air strategies (So and Wang, 2004;Wang and Shooter, 2004;Viana et al., 2006).In non-attainment areas, air-pollution models are frequently employed to predict hourly pollution levels and thereby help establish cost-effective means of reducing atmospheric PM concentrations and controlling emission sources (Park et al., 2004;Zawar-Reza et al., 2005;Luhar et al., 2006;Wilson and Zawar-Reza, 2006;Cheng et al., 2007).
In this work, hourly PM 10 variations were simulated using TAPM-3.0(Hurley, 2005) for severe pollution events during the three

MODEL DOMAIN AND MONITORING SITES
The three monitoring sites were Hsiung-Kong (industrial), Pingtung (urban) and Chao-Chou (rural) sites; all are located in southern Taiwan (Fig. 1).

Governing equations and grid setup
The Air Pollution Model (TAPM) is a three-dimensional, prognostic, Eulerian, incompressible, non-hydrostatic, primitive equation model in terrain-following coordinates for simulating atmospheric motion and pollutant transport using nested grids.Hurley et al. (2003) and Hurley (2005) described in detail the governing equations for mass, momentum, and energy, which are briefly elucidated as follows.

Emission inventory and boundary conditions
An emission inventory was obtained using TEDS-6.03(2006), which was issued by Taiwan's EPA.sources in each region.The methods of obtaining emission rates from various sources, except the updated emission rates in TEDS-6.03,were similar to those used by TEDS-4.2 (Chen et al., 2003).However, reactive hydrocarbons, R smog , replace the NMHC rate in the TAPM (Johnson, 1984), in which R smog is defined as the product of a reactive coefficient (or a multiplicative factor) and NMHC emissions.A multiplicative factor of 0.0067 was used for converting the emission rates of NMHC to those of R smog (Johnson, 1984), and a multiplicative factor of 1.0 was applied to emission rates of all other primary pollutants, such as PM 10 , NO x , and SO 2 .Background conditions for pollutants were set to 25 μg/m 3 for PM 10 , 0.7 ppbv for R smog (Hurley, 2003), 1 ppbv for SO 2 and 3 ppbv for NO 2 (Chang and Cardelino, 2000).
Notably, the initial values of eddy terms were set to zero because of a thermally stable condition at midnight.
Four-dimensional data assimilation (FDDA) was adopted to compare simulated surface winds with ground observations and correct horizontal momentum equations using the Newtonian relaxation (or nudging) procedure (Stauffer and Seaman, 1994;Hurley, 2002).
Model performance was assessed relative to actual measurements using the correlation coefficient (R) and index of agreement (IOA) as follows (Willmott et al., 1985).
where P i and O i are predicted and measured values, respectively, with sample size N, and O is average of measured data.= 0.75-0.86.Notably, the agreement between predictions and measurements is regarded as good when IOA exceeds 0.5 (Hurley et al., 2001 and2003).

Fig. 1 .
Fig. 1.Model domain and the three monitoring sites in southern Taiwan.
worst seasons (spring, fall, and winter) in Kaohsiung City and Pingtung County.Each simulation covers three consecutive days or 72 h.Measured data from three monitoring sites were collected and compared with simulation results.Meteorological conditions and source contributions at each monitoring site were analyzed.
work, pollutant emission rates were considered and input to the model via boundary conditions (discussed later).The TAPM is run on a personal computer for and air pollution components.Each horizontal layer had 35 × 35 grids, nested from the outside to the inside with size of 20, 7.5, 2 and 0.5 km.The entire simulation domain covered over 700 km × 700 km (Fig. 1); the fine grids covered the southern Taiwan over the domain of 262.5 km × 262.5 km, and were centered at 120 o 22' E and 22 o 23' N. The vertical domain was 8000 m high and comprised 25 horizontal layers at altitudes of 10 TAPM classifies the surface vegetation (land-use) into 29 classes.Surface data in the model were acquired using geographical charts that were issued by the Ministry of the Interior, Taiwan, to determine area fractions of land-use in each grid.The TAPM model is initialized at each grid point with values for wind velocities, temperatures and humidity ratios that were interpolated from synoptic analyses.The initial pollutant concentrations at inflow boundaries on the outermost grids are set to background values, while zero-gradient conditions are applied at outflow boundaries.
Figs. 2(a)-(c) present a synoptic surface weather chart for March 9 and hourly
Figs. 5(a) to 5(c) present a synoptic surface weather chart for October 13 and hourly variations in pressure and wind speed for October 12-14, 2005.A high-pressure system and north to northwest winds prevailed in southern Taiwan (Fig. 5(a)).Surface pressure was 1008.3-1012.1 hPa (Fig. 5(b)).Winds were frequently weak (< 2 m/s), particularly at midnight and in the early morning (Fig. 5(c)).Relatively strong winds of roughly 4 m/s were occasionally observed at 12:00-14:00.Notably, temperature was 26-32 o C and relative humidity was 57-88% during this period Fig. 6(a) presents simulated surface wind

Fig. 7
Fig. 7 compares the 3-day hourly simulations of surface PM 10 concentrations with measured values at the three sites during October 12-14, 2005.The PM 10 concentrations were relatively high at midnight and in the early morning.The highest (and mean) measured values were about 236 (123) μg/m 3 at Hsiung-Kong, 246 (105) μg/m 3 at Pingtung, and 129 (74) μg/m 3 at Chao-Chou.Similar to the case in spring, high PM 10 events in autumn were related to high-pressure systems and weak winds.The simulated concentrations agreed reasonably well with measured values, with R = 0.47-0.56 and IOA = 0.69-0.75.

Table 1 .
Emission rates of pollutants in TEDS-6.03 in southern Taiwan (kton per year)

Table 2 .
Estimated source contributions at industrial, urban, and rural sites.