評估都會地區居民之懸浮微粒及氮氧化物的暴露狀況：比較ISC3、AERMOD及土地利用迴歸模式的預測結果

Abstract

過去流行病學研究大多使用空氣品質監測站污染物的實測值評估空氣污染物對於居民健康的影響，但卻不容易定義出個體詳細的暴露狀況。應用空氣污染暴露模式連結個體的家戶地點以重建其暴露資料為解決辦法之一。 本研究以台北縣市為研究區域，使用兩種空氣擴散模式及土地利用回歸模式，預測17個空氣品質監測站及66個研究對象在2000、2003和2007年的NOX、PM10及PM2.5等污染物的濃度。將相關之排放、氣象與地形等資料輸入空氣擴散模式ISC3 和AERMOD以模擬空氣污染的濃度。另一方面，根據交通、土地利用、氣象及人口資料作為土地利用回歸模式的預測因子以預測污染物濃度。為了評估模式的適用性，本研究比較模式模擬值與實測值兩者之間的差異，且應用監測數值驗證模擬結果，此外也評估了三種方法對於研究對象的預測結果。 以空氣品質監測站的結果而言，顯示ISC3對於NOX模擬值與實測值的R2介於0.56至0.75之間，AERMOD介於0.56至0.72之間，而LUR則介於0.50至0.65之間。以PM10來說，ISC3 至少能夠解釋23%的變異，AERMOD則是28%，而LUR的R2比ISC3 和AERMOD還高。此外，三種方法對於PM2.5之模擬值與實測值的R2皆大於0.69，適合用於模擬PM2.5的濃度。另一方面，研究對象的預測結果顯示ISC3 和AERMOD模擬值的相關性很好，然而空氣擴散模式與土地利用回歸模式的R2卻不佳。 不同的方法能夠運算污染物濃度在不同空間尺度上的變化，可能會產生不同的預測值進而影響流行病學的研究結果。雖然土地利用回歸模式的模擬結果比空氣擴散模式好，但是空氣擴散模式考慮到污染物在環境中的傳輸、變化，且對於時間變異的解釋力較佳，因此未來可結合兩種方法來改善暴露評估的結果，提供可靠的預測資料以評估空氣污染暴露對於人體健康的影響。

Abstract BACKGROUND: Epidemiological studies of assessing the health effects from exposures to air pollution have been hampered by difficulties in characterizing individualized exposure levels for subjects. One possible solution is applying intraurban air pollution exposure models to link subjects’ home addresses to reconstruct individual exposures. METHODS: The study was conducted in Taipei, Taiwan. This study utilized two air dispersion models and a land use regression model to predict nitrogen oxides, particulate matter < 10 (PM10) and < 2.5(PM2.5) μm in aerodynamic diameter concentrations for 17 air quality monitoring (AQM) stations and 66 study subjects in 2000, 2003 and 2007. Two air dispersion models including ISC3 and AERMOD used emission inventory, meteorological data and topography data to simulate air pollution concentrations. Land use regression models were used to predict concentrations rely on the geographic variables as predictors, such as traffic, land use type, meteorological parameters and census data. To evaluate flexibility, the predicted air pollutant concentrations were compared with fixed-sites measurements. Subsequently, the estimations of three approaches for all study subjects were then compared. RESULTS: At the air quality monitoring sites, the R2 between modeled and measured NOX concentrations ranged from 0.56 to 0.75 for ISC3, 0.56 to 0.72 for AERMOD and 0.50 to 0.65 for LUR. For PM10, the ISC3 explained at least 23% variability of the measurement whereas AERMOD had at least 28% explained variance. The R2 value of LUR was higher than ISC3 and AERMOD. For PM2.5, the R2 of three methods were all greater than 0.69 that are applicable to simulate PM2.5 concentrations. For subjects, the correlations between ISC3 and AERMOD predictions are good; however, the R2 between air dispersion models and LUR was not well for NOX, PM10 and PM2.5. CONCLUSIONS: Each approach calculated the variability of concentrations in different spatial-scales, it could produce different estimations to influence the epidemiological results. Although the model performance of LUR was better than air dispersion models, the air dispersion models considered air pollutants transport and reaction in the environment also provided higher temporal variability. Therefore, air dispersion models and LUR could be integrated to improve the exposure estimates and provided reliable predictions to assess the health impacts on cohorts.