The locations of eighteen power plants that are considered by KMOE for temporary shutdowns or emission reduction measures are shown in Fig. 1. Detailed stack information such as stack height, stack radius, exit temperature and other data used in this study is listed in Table 1. As shown in Fig. 1, the power plants located in the western coastal area are distributed more densely (i.e., P03, P05, P09, P11, P08, and P12); two are located in eastern coastal area (i.e., P06 and P07), four are located in the southern coastal area (i.e., P01, P04, P14, P17, and P18); the others are in more inland areas of Korea (P13, P15, P16, and P02). Also note that Seoul, the capital of South Korea, is near the western coastal power plants, and Daejeon is located in the central inland area.

Fig. 1. 
The location of power plants, meteorological stations, and air quality monitoring station used in this study.

The stack heights are generally higher than 100 m, with the highest stack about 200 m height, P17 (Goseong), and P03 (Incheon), with a stack height of 199 m (Table 1). The average distance between power plant and meteorological observation site is 13.2 km, with the maximum distance of 31.8 km at P11 (Taean) among the eighteen power plants, and the shortest distance, 2.4 km, at P07 (Donghae). The average distance between the power plants and air quality station is 6.9 km, with the maximum distance of 25.6 km at P08 (Boryeong), and the minimum 0.7 km at P02 (Daegu), respectively.

Point source emission data available from the Korean National Institute of Environmental Research were used for this study (NIER, 2016). Table 2 shows the emission rates for each of the 18 power plants. PM₁₀ emissions totaled 3,455 ton/year from 216 power plant stacks for the base year of 2013. As indicated in Table 2, P11 (Taean) showed the highest PM₁₀ emissions (835.13 ton/yr), and the second and third highest emissions were at P17, and P08, respectively. Within the SMA, the highest PM₁₀ emission rate was 560.94 ton/yr at P08 (Boryeong) (Table 2).

TAPM, employed in this study, is a PC-based fast air quality model, driven by a user-friendly graphical interface to configure inputs, run the model, and analyze outputs. TAPM is composed of two prognostic subsections: meteorological and chemical modules. The overview of TAPM is described here. More details on TAPM equations and descriptions can be found in Hurley (2008a) and Hurley et al. (2008).

Meteorological module of TAPM predicts gridded three-dimensional meteorology and air pollution concentrations. The meteorological component of TAPM is an incompressible, non-hydrostatic, primitive equation model with a terrain-following vertical coordinate for three-dimensional simulations. The model solves the primitive meteorological equations: momentum, temperature, specific humidity of water vapor, and cloud/rain/snow components. It includes parameterizations for cloud/rain/snow micro-physical processes, turbulence closure, urban/vegetative canopy and soil, and radiative fluxes (Hurley, 2008a). Turbulence closure in the mean equations uses a gradient diffusion approach with diffusivity K; and includes a countergradient correction for temperature. An E-ε turbulence scheme is used to calculate K using prognostic equations for the turbulence kinetic energy (E) and its dissipation rate (ε). Plume buoyancy, momentum, and building wake effects are also included for point sources (Hurley et al., 2008).

The chemical component of TAPM consists of an Eulerian grid-based set of prognostic equations for both gas and particulate components. The TAPM model, an Eulerian grid-based model, includes gasphase photochemical reactions based on semi-empirical mechanism called the Generic Reaction Set (GRS) of Azzi et al. (1992) with the hydrogen peroxide modification of Venkatram et al. (1997), yielding the following 10 reactions for 13 species.

where hv denotes photo-synthetically active, and thirteen species used here are smog reactivity (R_smog), radical pool (RP), hydrogen peroxide (H₂O₂), nitric oxide (NO), nitrogen dioxide (NO₂), ozone (O₃), Sulphur dioxide (SO₂), stable non-gaseous organic carbon (SNGOC), stable gaseous nitrogen products (SGN), stable non-gaseous nitrogen products (SNGN), stable non-gaseous sulphur products (SNGS). The semiempirical GRS has been employed for photochemical analysis in numbers of previous studies (Kim et al., 2005; Venkatram et al., 1997). The concept of using R_smog rather than volatile organic compounds (VOCs) in the reaction equations follows from the work of Johnson (1984). The concentration of R_smog is defined as a reactivity coefficient multiplied by VOC concentration. For example, Johnson (1984) used [R_smog]= 0.0067[VOC] for typical 1980s. Other detailed descriptions including reaction coefficients and yield coefficients such as α and η are found in Hurley (2008a) and Hurley et al. (2008).

The chemical component of TAPM has 3 modes in options: trace, gas and dust modes. The dust mode should be employed for the simulation of particulate matter. In dust mode, particle sizes are categorized in TAPM model for four size ranges: 2.5, 10, 20, and 30 μm, and the calculations of Particulate Matter concentrations are actually done for PM_2.5, PM₁₀, PM_10-20 and PM_20-30. Here Particulate Matter for PM₁₀ and PM_2.5 can be defined as sum of primary particulate matter plus secondary particulate concentrations consisting of (SNGOC), (SNGN), and (SNGS) as indicated in above equations of (1) and (7)-(10). Gas and aqueousphase chemical reactions for sulfur dioxide and particles with the aqueous-phase reactions based on Seinfeld and Pandis (1998) are implemented. The dry deposition formulation follows the approach of Physick (1994) for gaseous pollutants, and Seinfeld and Pandis (1998) for aerosols. Wet deposition processes were considered for highly soluble gases and aerosols such as SO₂, and H₂O₂, PM_2.5, PM₁₀, PM₂₀, and PM₃₀ are considered in this model (Hurley, 2008a).

TAPM has been verified for a number of regions, industrial and urban scales in Australia, Seoul and overseas. TAPM had been used internationally for model verification studies against two US tracer experiments (Kincaid and Indianapolis), several annual US dispersion datasets (Bowline, Lovett and Westvaco), and annual meteorology and/or dispersion in various regions throughout Australia (Hurley et al., 2008; Park 2004).

The domain for TAPM model evaluation study was based on the grids of 80×80×25 at three domain (12 km, 4 km, 2 km, centered at each of the 18 power point location) to evaluate the model performance for each of the 18 power plants. Therefore 18 simulations should be carried out for this case. In this way, the boundary conditions will be effectively provided for finer nested domain to simulate precisely the impact of point sources. However, for the scenario study, the model domain over Korea was centered at the center of Daejeon city (latitude 36°21ʹ14.83ʺ N and longitude 127°23ʹ4.36ʺ E) with the grid of 120×120×25 grid points was employed. For the initial conditions of the meteorological variables, LAPS (Local Administrator Password Solution) data with a resolution of 75 km×100 km were used (ftp://ftp.csiro.au/TAPM). Surface and land use data (1 km×1 km) from the USGS (United States Geological Survey) and initial meteorological data (75 km×100 km) from the Local Administrator Password Solution from the Australia Meteorology Administration were used as inputs to the TAPM model (Hurley, 2008b).

To simulate the impacts of a reduction in emissions from power plants, we ran the TAPM model in two steps. In the first step, we carried out model performance simulations for the model validation study, prior to the impact simulations. The average concentrations of PM₁₀, PM_2.5, SO₂, NO₂, and O₃ in and around power plants during the study period were used for the initial background model conditions. The calculated initial background concentrations are listed in Table 3. Here it is noted that initial background concentrations are surface measurement only, and thus applied 4-day spin-up times to minimize the initial condition influences for both surface and non-surface concentrations. The resultant model outputs were evaluated by comparing the simulated meteorological variables and PM₁₀ concentrations against measurements. The statistical evaluation parameter, the index of agreement (IOA), was used to evaluate model performance.

In the second modeling step, the emission reduction impact study was performed. We set the base case by forcing the initial background conditions to zero in this study and applied emissions of power plants only for scenario studies. Current study is a very simple approach without heavy computational costs base on semi-empirical chemical reactions of TAPM, and thus it is contrasted to the comprehensive sensitivity simulations for calculation of source-receptor relations employed by previous comprehensive studies (Baker et al., 2016; Zhou et al., 2012; Bergin et al., 2008).

We carried out five scenarios with varying reductions in power plant emissions in different areas. Table 4 illustrates the scenarios in this study: [Scenario I], complete shutdown for P08 (Boryeong); [Scenario II], complete shutdown for P03 (Incheon), P05 (Ansan), P09 (Dangjin) and P11 (Taean); [Scenario III], partial shutdown for P17 (Goseong) and P08 (Boryeong) and a complete shutdown for P06 (Gangneung), and P10 (Seocheon); [Scenario IV], halving the gas phase emissions with no reduction of PM emission; and finally [Scenario V], 20% emissions reductions of both PM and gas emissions in all power plants during October.

[Scenario I] focuses on Seoul by shutting down P08 (Boryeong), with electricity power generation (EPG) of 30,778,882 MWh (14.6% reduction among total coal-fired EPG) and the highest emission rate (560.94 ton/yr for PM₁₀) but lowest stack height among the power plants located within the SMA (Table 4). [Scenario II] also simulates a shutdown of the four power plants (49.7% reduction of EPG) located within the SMA. Scenarios [I] and [II] both compare Daejeon in contrast to Seoul. [Scenario III] examines the effects of the shutdown of plants more than 30 years old located over the whole south Korea (8.5% reduction of EPG), whose operation had been suspended by KMOE. [Scenario IV] estimates the contributions from secondary production, by reducing 50% of the NO_x and SO_x emissions of all 18 power plants. In Scenario V, the emissions of PM₁₀, NO_x, SO₂ were assigned to 20% reduction amounts of all 18 power plants in the assumption that the generated electric power is proportion to the consumed fuel amount. The Scenario V is chosen to reflect the recent PM mitigation plans of Korean MOE, among which temporary shutdown of power plants during October is included (http://www.seoul.co.kr/news/newsView.php?id=20180629018012&wlog_ tag3=naver). Each of the above five scenarios are compared to quantify the impacts of the changes in each scenario. Additional details on each scenario are given in Table 4.

Two episodes are selected as being representative for high and low aerosol episodes, respectively. The episode cannot be a perfect but suitable one for this numerical study as estimating specific sources’ contribution to local and regional air quality.

First, a long-lasting haze episode such as April 8-10 in 2016 tends to be determined by unfavorable meteorological conditions like a stagnant High pressure system accompanied with the weak boundary layer wind, which could suppress vertical mixing and ventilation, largely driven by the synoptic conditions. This kind of synoptic setting, quite typical for severe haze events, is quite consistent with the previous studies (Seo et al., 2017; Oh et al., 2015; Wang et al., 2014). The moving Low pressure system before the haze event preceded very stagnant synoptic condition and weak pressure gradient, which helped to intensify the accumulation of aerosol loadings. Fig. 2 shows spatial distributions of 850 hPa geopotential heights and wind field for April 8 to 9. Boundary layer mean wind speed (850 hPa) increased specifically around Manchuria on April 8, which facilitated transport of anthropogenic and dust aerosols from the northeastern China to Korea with strong northwesterly up to 20 m/s. Later, wind speed gradually decreased until the end of haze (Fig. 2). The first period (April 8) of the haze event characterized aerosols external transport from China by strong northwesterly. After then, relatively calm and stagnant conditions contributed to build up of domestic haze in Korea.

Fig. 2. 
Spatial distributions of 850 hPa geopotential heights and wind fields for (a) April 7 to 12, and (b) June 1 to 6, 2016.

Meanwhile the June period from 1 to 7 is selected as a low-aerosol event. The synoptic features as shown in Fig. 2 indicate even lighter wind speed with overall lower pressure gradient. It is interesting to note that the most domain of Korea was influenced by the easterly on June 2, which shifted to southeasterly specifically in the southern peninsula later. The most striking difference between April haze and June clean periods is the boundary layer wind direction with the others largely similar. For Scenario V, impact assessment was carried out for October 1 to 11, 2017, when the prevailed wind direction was dominantly observed northeasterlies.

Measured background PM₁₀ concentrations during the study period were analyzed to characterize the vicinity of 18 measurement sites routinely monitored by KMOE. PM₁₀ concentrations at eighteen sites for both cases (April and June) are shown in the boxplots (Fig. 3). The median PM₁₀ concentration ranged from 35 μg/m³ to 73 μg/m³ in April and from 23 μg/m³ to 68.5 μg/m³ in June, respectively. The mean PM₁₀ concentration in April was 67% higher than the mean concentration in June. The mean PM₁₀ concentration in Seoul was 58 μg/m³; in Shanghai the mean PM₁₀ concentration was 149 μg/m³ (Wang et al., 2013).

Fig. 3. 
Boxplots of PM₁₀ concentration during 7-12 April and 1-6 June, 2016 at the monitoring sites around 18 power plants.

Figs. 4 and 5 show the horizontal distribution of the PM₁₀ with the wind vector averaged for two simulated periods centered at the selected four power plants: P01 (Busan), P02 (Daegu), P07 (Donghae), and P09 (Dangjin), under the initial concentrations (Table 3) and the emission from the power plants (Table 2). PM₁₀ concentrations during the first simulation period are higher than that for the second simulation period because of differences in the initial concentration and meteorological conditions.

Fig. 4. 
Predicted wind and concentration fields for an averaged values during 7-12 April, 2016 at four power plants (P01, P02, P07, and P09) area.

Fig. 5. 
Predicted wind and concentration fields for an averaged values during 1-6 June, 2016 at four power plants (P01, P02, P07, and P09) area.

Figs. 6 and 7 show the horizontal distribution of PM₁₀ with the wind vector simulated on 10 April and 4 June 2016 with zero background PM₁₀ concentration and the emissions from the power plants, respectively. The patterns are strongly governed by the wind: upwind regions have lower concentrations while downwind regions have higher concentrations; the area where there is a confluence of wind has high concentrations and areas where the wind direction diverges have lower concentrations. Especially the overall concentration in June showed much higher due to the relatively weaker wind speed rather than in April. On 10 April 2016, northerly winds are dominant in Korea, so the high PM₁₀ concentration region moves southward with the maximum concentrations in the south such as P14 (Yeosu), P18 (Hadong), and P17 (Goseong). Meanwhile, on 4 June 2016, southerly and easterly winds are dominant in Korea, so the high PM₁₀ concentration region moves northward with the maximum concentrations in the northwest.

Fig. 6. 
Predicted wind and concentration fields on April 10, 2016 over the country at 00 LST, 04 LST, 08 LST, 12 LST, 16 LST, and 20 LST.

Fig. 7. 
Predicted wind and concentration fields on June 4, 2016 over the country at 00 LST, 04 LST, 08 LST, 12 LST, 16 LST, and 20 LST.

Fig. 8 shows the horizontal distribution PM₁₀ with the wind vector simulated during the two simulation period with zero background PM₁₀ concentration and the emissions from the power plants. The second simulation period had higher concentrations than the first simulation period by approximately 3 times. During the first simulation period, PM₁₀ concentrations are higher in southern Korea and lower in northern Korea. The power plant sites of P14 (Yeosu), P17 (Goseong), and P18 (Hadong) have higher PM₁₀ concentrations, corresponding to the wind convergence. During the second simulation period, PM₁₀ concentrations are higher in the west and lower in the southeast. Lower PM₁₀ concentrations are governed by easterly winds, while higher concentrations are governed by northerly or westerly winds.

Fig. 8. 
Predicted wind and concentration fields for an averaged values during (a) 7-12 April, and (b) 1-6 June, 2016.

In order to validate the simulated meteorological variables and PM₁₀ concentrations for each power plant, data for the nearest meteorological station and the nearest air quality station were used. The index of agreement (IOA) between the simulated value at the power plant and the observed value at the nearest meteorological or air quality monitoring station for each power plant was used to evaluate the simulated result. The IOA is a frequently used measure of how well the predicted variation about the observed mean are represented, with a value greater than about 0.50 considered to indicate a good prediction.

Table 5 shows the IOAs between modeled and observed PM₁₀ concentrations and the meteorological variables such as wind speed, wind direction, and air temperature for the study period. Temperature in June showed the highest IOA with an average of 0.89 while PM₁₀ concentration showed the lowest IOA with an average of 0.43. The IOA was similar to IOA found in other studies in urban areas (Hurley et al., 2008, Park et al., 2004). The IOA deviation (station IOA - station averaged IOA) for two different periods shows nearly the same patterns. At P01 (Busan) and P02 (Deagu), IOAs for all variables are higher than the station-averaged value. At P12 (Gunsan) and P13 (Iksan), IOAs for the meteorological variable are higher than the stationaveraged value, but those for PM₁₀ concentration are lower than the station-averaged IOAs.

The results for simulated mean and maximum PM₁₀ concentrations to assess the contribution of each power plant in the study region, the contribution to the maximum surface concentration in April and June ranged from 3.79 μg/m³ (P05: Ansan, April) to 17.00 μg/m³ (P11: Taean, June) (Table 6). At P09 (Dangjin), P11 (Taean), P17 (Goseong), and P18 (Hadong), PM₁₀ emissions can be classified as a large power plant with regard to emissions, exceeding 300 ton/year, so that their high emissions are related to elevated maximum PM₁₀ concentrations (ranging from 12.84 to 17.00 μg/m³ in June). At power plants such as P01 (Busan), P02 (Daegu), P04 (Ulsan), P05 (Ansan), P15 (Gimcheon), P05 (Ansan), and P16 (Gumi), the maximum concentrations are lower (ranging from 3.79 to 9.59 μg/m³ in April and June). In particular, the power plant at P08 (Boryeong) had the lowest stack height (42 m) (see Table 1), but the second highest PM₁₀ emission (560.94 ton/yr) (see Table 2). Due to the lower stack height and larger emission amount, the maximum surface concentration at P08 (Boryeong) was 12.12 μg/m³, the highest in April. As considering that the locations showing the maximum concentrations were more than 100 km far from half of all stacks, it should be also mentioned the secondary production during long-range transport is also important in PM formation in addition to their contribution of the primary PM sources.

The maximum and average concentration and the location where those concentration occurred for each scenarios are shown in Table 7. The results of sensitivity experiment, such as increased/decreased concentrations, relative to BAU (Business As Usual) is shown for each scenario in Table 8. In Scenario I the maximum decrease in PM₁₀ concentrations in the SMA due to a shutdown of the P08 (Boryeong) was less than 0.00 to 0.54 μg/m³ (0.0-4.4%). There is a small impact of the P08 (Boryeong) on the SMA because the stack height is low and the SMA is distant from the plant. In scenario II, simulating a shutdown of four power plants the SMA, reduced emissions (1,320 tons/year) comprise more than one-third of the total emissions from 18 power plants (3,455 tons/year). The impact of shutting down four power plants on the decrease in PM₁₀ concentrations in the SMA was as high as 18.9% in June, and a decrease of 23.9% was also simulated in the Daejeon region, proportional to the reduced emissions in the SMA.

For scenarios III, there is a large decrease in PM₁₀ concentrations around the power plants of P17 (Goseong), P08 (Boryong), P06 (Gangneung) from 6.4 to 31.8%, compared to the decrease in the SMA and Daejeon area from 0.9 to 7.7%, relatively far away from the power plants, even though the emission reductions of 60 to 190 tons/year in each power plant due a shutdown of aged plants are not significant compared to the total emission amount from all power plants considered in our study. The maximum PM₁₀ concentrations in in Seoul, Daejeon, P17 (Goseong), P08 (Boryeong), P06 (Gangneung) and P10 (Seocheon) decreased from 0.3 to 19.7% in scenario IV when SO₂, NO_x emissions from all 18 power plants were reduced (Table 8). In Scenario V, the maximum and averaged concentrations were higher than those in April and lower than those in June for BAU. In compared with Scenario IV, the reduction rates were higher than those in April and lower than those in June except Seoul and Gangneung.