County-Wide Mortality Assessments Attributable to PM2.5 Emissions from Coal Consumption in Taiwan

Over one-third of energy is generated from coal consumption in Taiwan. In order to estimate the health impact assessment attributable to PM2.5 concentrations emitted from coal consumption in Taiwan. We applied a Gaussian trajectory transfer-coefficient model to obtain county-wide PM2.5 exposures from coal consumption, which includes coal-fired power plants and combined heat and power plants. Next, we calculated the mortality burden attributable to PM2.5 emitted by coal consumption using the comparative risk assessment framework developed by the Global Burden of Disease study. Based on county-level data, the average PM2.5 emissions from coal-fired plants in Taiwan was estimated at 2.03 ± 1.29 (range: 0.32–5.64) μg/m3. With PM2.5 increments greater than 0.1 μg/m3, there were as many as 16 counties and 66 air quality monitoring stations affected by coal-fired plants and 6 counties and 18 monitoring stations affected by combined heat and power plants. The maximum distances affected by coal-fired and combined heat and power plants were 272 km and 157 km, respectively. Our findings show that more counties were affected by coal-fired plants than by combined heat and power plants with significant increments of PM2.5 emissions. We estimated that 359.6 (95% CI: 334.8–384.9) annual adult deaths and 124.4 (95% CI: 116.4–132.3) annual premature deaths were attributable to PM2.5 emitted by coal-fired plants in Taiwan. Even in six counties without power plants, there were 75.8 (95% CI: 60.1–91.5) deaths and 25.8 (95%CI: 20.7–30.9) premature deaths annually attributable to PM2.5 emitted from neighboring coal-fired plants. This study presents a precise and effective integrated approach for assessing air pollution and the health impacts of coal-fired and combined heat and power plants.


Introduction
In Taiwan, coal is the main source of electricity generated by Taipower Company, independent power plants, combined heat and power (CHP) plants, and industrial plants (such as steel, cement, chemical, and paper plants) [1]. In 2000, total coal consumption was 45.5 MT (mega-tons) per year, with about 32.2 MT per year used in coal-fired (CP) and CHP plants. Coal-fired plants alone accounted for 70.7% of total coal consumption. Over a recent and one plant (CP#05) used a subcritical boiler (shown in Supplementary File S1) [32,33]. One coal-fired plant and nine combined heat and power plants were located in Kaohsiung City. Table 1 shows that the most coal-fired plant emissions came from a plant in Taichung City, which accounted for 1235. 44  where all the SO 2 and NOx are transformed to ammonium sulfate ((NH 4 ) 2 SO 4 ) and ammonium nitrate (NH 4 NO 3 ) with enough ambient ammonium ions (NH 4 + ). The fine fractions of PM 10 are 0.85 for (NH 4 ) 2 SO 4 and 0.58 for NH 4 NO 3 compounds [28]. These two ratios were based on the field measured data in a coastal area of Taiwan from Tsai and Cheng [34]. Int  All parameters used in the GTx model are listed in Table 2. Using settings for the year 2013, we also collected observations from 117 meteorological datasets and 71 air quality monitoring networks (AQMNs). We modeled the PM2.5 at 71 AQMNs for all coal-using  All parameters used in the GTx model are listed in Table 2. Using settings for the year 2013, we also collected observations from 117 meteorological datasets and 71 air quality monitoring networks (AQMNs). We modeled the PM 2.5 at 71 AQMNs for all coal-using plants. In the "traj" mode of the GTx model, wind field data were obtained by applying a weighted interpolation method to meteorological data, and the pollutant concentrations were modeled in the "pm" mode of the GTx. Here, we estimated a modeling period of 168 h over one whole year, 2013, using the GTx model. The model's assumptions are shown in Table 1, including emissions data treatments, background concentrations from overseas, and dry deposition velocity of NOy. Adding emission data from boats for line sources; 3.
Excluding diesel-fueled trucks driving on township roads for line sources; 4.
Excluding emission data from boats and suspension dust for area sources; 5.
Emission data estimates for unpaved road dust by using area and wind speed. 2.
The mean fractional error (MFE) for PM 2.5 model validation should be in intervals less than 55%. 3.
The correlation coefficient (R) for PM 2.5 model validation should be higher than 0.5.

Estimation of the Population-Attributable Fraction of Mortality Attributable to PM 2.5
Our study considered the mortality from four major diseases defined according to the International Classification of Diseases, Tenth Revision (ICD-10) codes. These include ischemic heart diseases (denoted as IHD, I20-I25), stroke (I60-I67, I69.0, I69.1, I69.2, and I69.3), lung cancer (denoted as LC, C33, and C34), and chronic obstructive pulmonary disease (denoted as COPD, J40-J44). We estimated the population-attributable fraction (PAF) of cause-specific mortality associated with PM 2.5 for each disease by year and by county level.
According to previous longitudinal studies and our GBD 2013 nonlinear exposureresponse model, we assumed a linear relationship between PM 2.5 exposure (moderate exposure level: <50 µg/m 3 annual average) and the corresponding disease outcomes to derive relative risk (RR) estimates for our analyses [12]. The risk effect sizes were estimated by 10 µg/m 3 increments of PM 2.5 for each disease outcome. These RR estimates for 4 disease outcomes were calculated by 10 µg/m 3 PM 2.5 increment. The standard level of PM 2.5 in our draft is recommended by the World Health Organization (10 µg/m 3 ) [22]. For the uncertainty interval of our RR estimation, we adopted data by meta-analysis of the results of previous longitudinal studies, assuming a constant association between point estimates and standard errors in our RR estimation processes to allow for a monotonic relation between PM 2.5 and health outcomes and to obtain a conservative confidence interval (CI) [36][37][38][39][40][41].
The PAF measures what proportion of the disease burden in a given population would be prevented or postponed if the PM 2.5 exposure level were shifted to an alternative optimal exposure level. We used the following formula to calculate the PAF of cause-specific mortality for a specific disease at the county level: where c(i) is the estimated 10-year average level of PM 2.5 in county i, which reflects the cumulative exposure of PM 2.5 , and RRc(i), j is the RR for disease j at exposure level c(i), as determined using our linear model. We further multiplied the PAF by the cause-specific number of deaths to obtain the mortality burden attributable to PM 2.5 , as seen in the following equation [25]: where ∆M α,k : the change in annual mortality (or pre-mature mortality) [deaths per year] due to coal pollution for each cause of death α in each city/county k; y 0α,k : is the causespecific baseline death rate or PAF [% per year] in the city/county; ∆x is the populationweighted change in PM 2.5 concentration [µg/m 3 ]; β is the cause-specific concentrationresponse function (CRF) relating a one-unit change; P k is the total population of the city/county k. All the analyses included age, gender, and county. Meanwhile, the age group for adult deaths included the population over 25 years old, and the age group for premature deaths included the population between 25 and 70 years old.

Uncertainty and Sensitivity Analyses
Statistical simulation was performed to deal with the uncertainty induced by sampling variability. In the model, we randomly sampled 1000 sets of PM 2.5 exposures and corresponding RRs from prior normal distributions of PM 2.5 concentrations and RRs. Each set of sampled PM 2.5 concentrations and RRs was used to estimate the PAF and the amount of PM 2.5 for adult deaths (or premature deaths) for each county separately, by age group. We ranked the resulting 1000 PAFs and the number of PM 2.5 -attributable deaths and considered the corresponding 2.5th and 97.5th percentiles as the range for 95% CIs.

Comparison of PM 2.5 Observation and Estimates by GTx Modeling
The PM 2.5 concentrations estimated by the GTx model were cross-validated with in situ observations in our previous study [26]. In brief, overall, 25,684 daily-site data pairs showed the MFB, MFE, and R indices in Taiwan were 7%, 37%, and 0.53, respectively, and they satisfied the model criteria [35]. For spatial comparison of PM 2.5 , the modeled PM 2.5 findings were higher than observed PM 2.5 findings in mountain areas in northern, central, and eastern Taiwan and the coastal area in Miaoli County. Conversely, the model PM 2.5 figures were lower than the observed PM 2.5 figures in southern Taiwan (Figure 2A

Spatial Distribution of Population and PM2.5 Observations in Taiwan
At the end of 2013, there were a total of 23,373,517 people residing in Taiwan. Figure  3A shows the citywide distribution of the population in Taiwan. New Taipei City is a tier 1 level (of 5 grades), with 3,954,929 people. Kaohsiung City, Taichung City, Taipei City,

Spatial Distribution of Population and PM 2.5 Observations in Taiwan
At the end of 2013, there were a total of 23,373,517 people residing in Taiwan. Figure 3A shows the citywide distribution of the population in Taiwan. New Taipei City is a tier 1 level (of 5 grades), with 3,954,929 people. Kaohsiung City, Taichung City, Taipei City, and Taoyuan City were listed in the tier 2 level, with residents numbering 2,779,877, 2,701,661, 2,686,516, and 2,044,023, respectively [42]. Moreover, the spatial distribution in cities and counties based on 10-year averaged PM 2.5 concentrations is shown in Figure 3B. Meanwhile, the overall average of observed PM 2.5 concentration with standard deviation (SD) was 30.59 ± 4.05 µg/m 3 . For spatial variation, in Chiayi and Kaohsiung cities, both listed in the tier 1 level (of 5 grades), the PM 2.5 averages were higher than 44.0 µg/m 3 . In addition, Tainan Table 3).

Counties Affected by PM 2.5 from CP and CHP Plants and Maximum Distance
The "maximum (affected) distance" was calculated by using the Euclidian distance method between each CP or CHP plant and AQMN under different modeled PM 2.5 levels (shown in Supplementary File S1). In our models, results ranged from 4 to 16 counties and from 12 to 66 AQMNs affected by coal-fired plants, with 0 to 6 counties and 0 to 18 AQMNs affected by combined heat and power plants under increments greater than 0.1 µg/m 3 PM 2.5 . The maximum affected distances ranged from 155 to 272 km for coal-fired plants and from 0 to 157 km for combined heat and power plants ( Table 4). The CP#02 plant, located in Taichung City, had the maximum air quality impact affecting 16 counties and 66 AQMNs. In this case, the distance from CP#02 plant to Linyuan AQMN in the city of Kaohsiung was 193 km. In addition, CP#04 had affected the maximum distance of 272 km, traveling from the CP#04 plant in Kaohsiung City to Xindian AQMN in New Taipei City (Table 4). Under the same conditions, however, the CHP#27 plant in Yuanlin County also had the maximum air quality impact, affecting 6 counties and 18 AQMNs. In that case, the distance from the CHP#27 plant to Longtan AQMN in Taoyuan City was 157 km. In addition, there were 14 combined heat and power plants that did not affect any AQMN due to PM 2.5 increments less than 0.1 µg/m 3 ( Table 4). The PM 2.5 impact from coal-fired and combined heat and power plants differed significantly due to emissions rates, and the stack and flow parameters. Using the same approach, the maximum affected distances were 306 km from CP#01 in New Taipei City and 163 km from CP#02 in Taichung City, at 0.01, and 1.0 µg/m 3 PM 2.5 increments, respectively. Additionally, the maximum affected distances were estimated to be 284 km from CHP#07 in Taoyuan City and 3 km from CHP#23 in Miaoli County at 0.01 and 1.0 µg/m 3 PM 2.5 increments, respectively (Results not shown).
The estimated adult deaths per year due to PM 2.5 from all coal-fired plants ( Figure 4A (Table S4). Estimated premature deaths per year attributable to all coal-fired plants ( Figure 4B) showed New Taipei City ranked in tier 1 (of 7 grades) with a range of 15 to 25 deaths per year. New Taipei, Taichung, and Tainan Cities ranked as tier 2 with a range from 10 to 15 deaths per year, and Taoyuan City, Yunlin County, and Chiayi County ranked in the tier 3 level with a range from 7.5 to 10 deaths per year. More details for disease-specific premature deaths attributable to PM 2.5 from all coal-fired plants are shown in Table S5 tou County, Yunlin County, and Chiayi County ranked in the tier 2 level with a range of 20 to 30 deaths per year, and Taipei City, Ilan County, and Pingtung County ranked in the tier 3 level with a range from 15 to 20 deaths per year. More details for disease-specific adult deaths attributable to PM2.5 from all coal-fired plants are shown in Table S3

GTx Modeling
The air quality modeling results depend completely on emissions and meteorological data. In our study, the GTx model for PM 2.5 concentrations used the emissions inventories from CP and CHP plants (from TEDS 9.0) and observed meteorological data (from Taiwan's Environmental Protection Agency and the Central Weather Bureau in 2013) ( Table 2). Before scenario modeling through the GTx model, we first verified the modeled results using PM 2.5 observations obtained from 71 Taiwan EPA AQMNs. If we double checked the comparisons with 71 individual AQMS sites, the results showed that 83% (59/71) satisfied the comparison criteria for MFB, 93% (66/71) for MFE, and 89% (63/71) for R (data not shown). In our models, there were still many over-and underestimates related to geography in Taiwan. The overestimated regional statistics were on average 5.3, 1.4, and 7.3 µg/m 3 higher in northern, central, and eastern Taiwan, respectively. Underestimated figures were 3.3 and 2.3 µg/m 3 lower in the Yun-Chia area and southern Taiwan [26]. Several possible reasons might result in the differences between models and observations, including (1) the possibility that emissions data did not capture the correct condensable fraction of PM emitted from stationary sources [43,44]; (2) secondary aerosol growth mechanisms might be more complex than model defaults [28][29][30][31]; (3) the emissions data for all sources might not be revised simultaneously, as shown in Table 2; (4) hourly emissions from sources were unknown; and (5) the background concentrations of PM might be misused.
According to a previous report [45], 37% of PM 2.5 came from outside of Taiwan, and 18.2%, 21.2%, and 23.5% of PM 2.5 were contributed from point, line, and area sources, respectively. The report also indicated that about 4.5% of PM 2.5 contributed from power plants and 13.7% of PM 2.5 contributed to other industrial sources throughout Taiwan. However, there is no detailed county-based PM 2.5 source apportionment study with source/trajectory or receptor models in Taiwan. In our analysis with the GTx model, we estimated the median county-averaged contribution of 4.1% (range: 1.4-12.0%) of PM 2.5 from 5 coal-fired power plants. If taking into account 5 coal-fired power plants (CPs) and 33 coal-fired combined heat and power plants (CHPs), the median county-average contribution was 6.1% (range: 2.4-16.1%) of PM 2.5 (shown in Supplementary File S2). Our PM 2.5 estimates from 5 CPs and/or 5 CPs + 33 CHPs indicated good agreement with the previous report.
In particular, results showed that there were no CP and CHP plants located in Keelung City, Taipei City, Hsinchu City, Nantou County, Chiayi City, and Pingtung County (Table 1), yet CP or CHP plants contributed 1.09, 0.87, 1.40, 5.64, 2.79, and 1.87 µg/m 3 to annual average PM 2.5 ( Table 3). The highest county-level estimate (5.64 µg/m 3 ) was in Nantou County, with emissions from CP and CHP plants accounting for almost 16% of the 10-year averaged PM 2.5 concentration. Results from GTx modeling in Nantou County also showed that nearly 75% (4.22 of 5.64 µg/m 3 ) of PM 2.5 concentrations were emitted from coal-fired plants( Table 3). Nantou County, the only inland district in Taiwan, accounted for 95% of the total area over 100 m high (including 66% of the total area above 1000 m) [46]. High altitude terrain, geographic basin structures, and monsoons might be leading factors causing low dispersion and accumulation effects when the various pollutants travel via air parcels into Nantou County from neighboring cities or counties to the west (Figure 1) [47].

Assessing Health Impacts
We applied the GBD approach to estimate the adult deaths and premature deaths attributable to PM using different age intervals in our study because the disease-specific mortalities were age dependent. The results indicated that there were significant differences in PAFs for adult deaths and premature deaths for IHD (16.9% of adult deaths vs 24.2% of premature deaths) and stroke (26.7% of adult deaths vs 39% of premature deaths) (Table S1). Unlike cardiovascular system diseases (IHD and stroke), the differences between PAFs for adult deaths and premature deaths due to the effect of PM on the respiratory system (LC and COPD) showed no gaps. In 2013, strokes were the leading cause of adult deaths and premature deaths attributable to ambient PM 2.5 exposure in Taiwan, accounting for an estimated 44.9% (2996.8 of 6672.9 per year) of adult deaths and 47.7% (1174.8 of 2461.7 per year) of premature deaths on average. Compared with a previous study [22], our estimate of total deaths fell reasonably into the interval (range: 6282 to 7869 deaths per year on average) between the main analysis and a nine-year time lag. We extracted the estimates in 2013 from the study results of Hwang et al. [23], and the deaths attributable to PM 2.5 ranged from 2026 to 7066 deaths (median: 4665 deaths) per year. Our assessments also fell into estimated intervals but with more accuracy. In terms of geographic analysis, Kaohsiung City was the city with the largest impact on total deaths assessed, with PM emissions estimated to account for 15.7% of adult deaths (1046.3 of 6672.9 per year) and 16.7% of premature deaths (410.3 of 2461.7 per year) on average (Tables S2 and S4).

Limitations
There were several limitations to our study. First, to reasonably link PM 2.5 exposures to health outcomes, we used 10-year PM 2.5 averages for a long-term exposure level capable of promoting or aggravating disease. Second, following previous studies, we only included IHD, stroke, lung cancer, and COPD [12,14,22,25]. Actually, many other potential health impacts were not included in our estimates, such as hospital admissions, emergency department visits, doctor visits, restricted activity/reduced performance, medication use, symptoms, physiological changes in the cardiovascular system, impaired pulmonary function, and subclinical effects proposed by previous reports [48]. Lastly, our GTx modeling might contain several uncertainties in its current version, including incomplete or missing emissions data, background concentrations, and secondary aerosol growth mechanisms, as mentioned above. These uncertainties in each AQMN might cause positive or negative effects on PM 2.5 modeling. Furthermore, we checked an overall 25,684 daily-site data pairs with observed and GTx modeled PM 2.5 from Taiwan, and the results showed that the MFB, MFE, and R indices satisfied the model criteria [35]. Therefore, we believe that our estimates are credible. Indeed, most air quality modeling studies selected the Weather Research and Forecasting Model coupled with Chemistry (WRF-Chem) based model system because of its excellent meteorological parameter forecasting. However, our model system (GTx model) actually used real meteorological data from 117 sites, as shown in Table 2. Our GTx model was mainly developed for Taiwan. Although the GTx model selected more simple photochemical reactions for secondary PM 2.5 productions from gaseous pollutants, the final estimates for PM 2.5 were acceptable for us with several validations.
In fact, the population spatial distribution represented the urbanization levels, and all coal-fired power plants emitted PM 2.5 spatial distributions just represented the air quality levels by those major point sources (Figure 3). These two levels did not match originally. Tsai et al. [49] and Lu et al. [50] used different receptor models (Positive Matrix Factorization and Chemical Mass Balance, respectively) to analyze different coastal areas (Taichung and Tainan, respectively). Therefore, different major sources (road dust in Taichung and traffic emissions in Tainan, respectively) might be possible. Actually, the above two references for PM 2.5 source apportionments were totally different from our study. Our study framework was based on the air quality model (GTx model) and the global burden of disease approach. In fact, any air quality could replace the GTx model. As I know, if the entire emission data, meteorological data, and geological data were fully prepared for GTx modeling in other countries. Our method could be applied outside of Taiwan. Indeed, a previous report showed that the line or mobile source contribute 21.2% of PM 2.5 in Taiwan [45]. Therefore, transportation systems should be discussed, including electronic transport tools [51][52][53].
In our analyses, we did not take into account the ambient concentrations and health impacts of CO, POPs, and others. Yang et al. [54] provided the short-term adverse effects of air pollution and concluded that CO increases hospital admissions for cardiovascular diseases. Raub et al. [5] reviewed the sources of CO indoors/outdoors, such as in workplace/home with poor ventilation and street intersections/internal combustion engines/industrial sources, respectively. The complications of CO poisoning were also reviewed, including immediate death, myocardial impairment, hypotension, arrhythmias, and pulmonary edema. Previous studies showed that several congeners of dioxins/furans could be found in gaseous pollutants emitted from CP or CHP plants [55,56].

Conclusions
Based on our GTx modeling, PM 2.5 contributed from coal-fired and combined heat and power plants ranged between 0.32 and 5.64 µg/m 3 , with an average of 2.03 ± 1.29 µg/m 3 in city/county-based levels. The PM 2.5 contribution from all coal-fired plants was greatest in Nantou County, estimated at 5.64 ± 1.37 µg/m 3 , equal to 15% of the 10-year PM 2.5 average level. Our results showed coal-fired plants always contributed more PM 2.5 than combined heat and power plants in affected counties. Of the overall average PM 2.5 in Taiwan, 1.33 ± 0.89 came from coal-fired plants and 0.70 ± 0.45 µg/m 3 from combined heat and power plants. While PM 2.5 increments greater than 0.1 µg/m 3 , coal-fired plants affected as many as 16 counties and 66 AQMNs and combined heat and power plants affected 6 counties and 18 AQMNs. The maximum distances affected by coal-fired and combined heat and power plants were 272 km and 157 km, respectively. According to the health statistics, PM 2.5 emissions from coal-fired plants accounted for 359.6 (95%CI: 334.8-384.9) adult deaths and 124.4 (95% CI: 116.4-132.3) premature deaths per year. Our results also demonstrated severe air quality and probable health impacts in six counties without power plants. This study provides an effective and precise integrated approach to assessing air quality and associated health impacts attributable to PM 2.5 emitted from coal-fired and combined heat and power plants.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/ijerph19031599/s1, Table S1: Disease-specific population attributable fraction for adult death and pre-mature death due to PM 2.5 in different cities/counties, Taiwan. Table S2: Disease-specific total deaths attributable to ambient PM 2.5 in different cities/counties, Taiwan. Table S3: Disease-specific deaths attributable to PM 2.5 from all coal-fired plants in different cities/counties, Taiwan. Table S4: Disease-specific pre-mature deaths attributable to ambient PM 2.5 in different cities/counties, Taiwan. File S1: Data_Conc+Dist-CP+CHP-revision.xlsx file: In 73 AQMN stations, the PM 2.5 modeling concentrations and impact distances from 5 coal-fired power plant (CP) and 33 combined heat and power plant (CHP) were shown. File S2: Data_HIA-CP+CHP-revision.xlsx file: In 19 cities/counties of Taiwan, the county-based PM 2.5 exposures and adult and premature deaths from ambient and coal-fired power plant emitted PM 2.5 were shown.