請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91875
標題: | 應用混合正矩陣因子法改良細懸浮微粒於時間與空間相依性之來源解析 Improving time- and space-dependent source apportionments of PM2.5 using hybrid Positive Matrix Factorization modeling approaches |
作者: | 黃淳聖 Chun-Sheng Huang |
指導教授: | 吳章甫 Chang-Fu Wu |
關鍵字: | 受體模式,暴露評估,在線式微粒成分數據,空氣污染事件,二次氣膠,空氣污染熱點, receptor model,exposure assessment,online particle composition measurements,air pollution episode,secondary aerosol,air pollution hotspot, |
出版年 : | 2023 |
學位: | 博士 |
摘要: | 受體模式之正矩陣因子法(Positive Matrix Factorization,PMF)廣泛應用於解析細懸浮微粒(PM2.5,氣動粒徑等於或小於2.5微米的微粒)的污染來源。然而,受到輸入數據特性(如:採樣解析度)和其既有模式原理(如:使用物種的共線性問題)的影響,傳統PMF對於PM2.5在時間和空間角度的模式解析上具有一定限制,這些限制也影響了模式結果後續在暴露評估和空氣品質管理上的應用。為改善上述限制,在本論文中,將三種數學方法與PMF模式結合以改善PM2.5的來源解析。
第一個研究為從時間的角度上探討空氣污染事件議題。受到低時間解析度數據,以及應用長期輸入數據集的影響,使得PMF無法解析出導致短期PM2.5污染事件的主要貢獻因子。本研究應用了逐時解析度之在線式微粒成分監測數據,以及移動式數據集技術(或稱「移動窗格技術」;Moving window technique),並結合限縮型PMF(Constrained Positive Matrix Factorization),針對臺灣一PM2.5污染事件案例進行解析。結果顯示,於事件發生期間,在2020年12月12日對PM2.5濃度影響最大的是額外分離之「區域傳輸源」(PM2.5日平均占比 = 61%),而在12月13日則是常態污染之「工業混和硫酸鹽源」(43%)。由於12月13日的PM2.5事件主要是由常態污染源所導致,因此,管理境內的工業排放源對於降低高PM2.5濃度十分重要。 第二個研究的主軸,為量化原生性污染源對PM2.5二次氣膠在不同時間尺度上的貢獻。由於PMF經常解析出獨立之二次氣膠特徵因子,因此,此研究提出了兩階段PMF建模方法,並應用穩健迴歸模式(Robust regression)來重新分配PMF因子內的二次氣膠。模式結果顯示,3小時內的二次氣膠主要由「油類燃燒源」貢獻(二次氣膠相關物種總和 = 2.67 μg/m3)。而在24小時內,二次氣膠的最大貢獻來源是「工業源」(1.65 μg/m3)。研究結果凸顯出需依據不同時間尺度,針對二次氣膠相關物種進行管控的重要性,也為控制原生性污染源和抑制二次氣膠的形成提供了有價值的資訊。 第三個研究為從空間的角度上,針對本地排放源的污染熱點區域進行辨識,這是典型PMF模式的限制之一。研究應用地理統計模型-普通克利金法(Ordinary kriging)來估算PM2.5質量濃度的空間分布,並將其放入限縮型PMF中,以解析和分配污染源貢獻PM2.5的空間變異。本研究透過土地利用特徵評估PMF解析之「道路揚塵/土木建設」在空間分布的合理性。結果顯示,該污染源所貢獻的PM2.5空間分布與「主要道路長度」和「建築工地數量」呈現正相關(相關係數 ≥ 0.40)。土地利用迴歸模式之交叉驗證判定係數(cross-validation R2)為0.48。本研究提出的方法能夠辨識本地排放源的污染熱點區域,並為當地政府提供空氣污染源之管理標的。 考量到PMF具有區分空氣污染來源特徵的優點,建議未來可進一步探討結合PMF與各種數學方法,以提升其未來發展和應用。透過結合各類方法有助於改善PMF的既有限制,並提升其在暴露評估和空氣污染研究中的成效。 The receptor model of Positive Matrix Factorization (PMF) is widely employed for source apportionment of fine particulate matter (PM2.5, particulate matter with an aerodynamic diameter of equal to or less than 2.5 μm). However, due to the properties of input data (e.g., sampling time resolution) and inherent modeling principles (e.g., collinearity of utilized species), conventional PMF has limitations in time- and space-dependent source apportionments of PM2.5, affecting the subsequent application in exposure assessment and air quality management. To address these issues, in this dissertation, three mathematical approaches were integrated with PMF modeling for improvement of PM2.5 source apportionments. In the first study, the issue of air pollution episode from the temporal perspective was explored. Owing to the use of low time resolution data and an extended input dataset, the major contributor to the short-term PM2.5 episode cannot be identified by PMF. Here, the online particle composition measurements with hourly resolution were applied, and the moving window technique were incorporated with a constrained PMF to resolve a PM2.5 episode case in Taiwan. The results showed that the most significant contributor to the PM2.5 episode on 12/12/2020 was an additionally differentiated factor of regional transport (daily average PM2.5 contribution = 61%), while on 12/13 was the regular pollution of industry/ammonium sulfate related (43%). Since the PM2.5 episode on 12/13 was mainly caused by the regular pollution, managing local industrial emission sources is crucial to reduce elevated PM2.5 levels. In the second study, the temporally-resolved contributions of PM2.5 secondary aerosols from primary emission sources were quantified. As an individual secondary feature factor was commonly resolved from PMF, a two-stage PMF modeling approach was proposed, and a robust regression model was applied to re-apportion the secondary aerosols in retrieved PMF factors. The results showed that the majority of secondary aerosols (sum of secondary aerosol-related species = 2.67 μg/m3) within three hours were mainly contributed by oil combustion, while the largest contributor of secondary aerosols (1.65 μg/m3) over 24 hours was industry. This highlights the importance of regulating secondary aerosol-related species with various time spans as control targets, providing valuable information for controlling primary emission sources and inhibiting the formation of secondary aerosols. The third study focused on characterizing the hotspot regions of a local emission source from the spatial perspective, which cannot be achieved by typical PMF analysis. A geostatistical model of ordinary kriging was introduced to estimate spatially distributed PM2.5 mass concentrations, which were then applied in the constrained PMF for apportionment of source-specific PM2.5 with spatial variation. The spatial distribution of road dust/civil construction retrieved from PMF was evaluated using land use features. The positive Pearson correlations (coefficients ≥ 0.40) were found between the spatially distributed source-specific PM2.5 and land use characteristics of major road length and the number of construction sites. A leave-one-out cross-validation R2 of 0.48 was achieved using the land use regression model. The proposed approach identifies the hotspot regions of local emission sources, offering targets for the management of air pollution sources by local city governments. Given the advantages of PMF in distinguishing features of air pollution sources, it is recommended to further explore the integration of PMF with mathematical approaches to enhance its future development and applications. The integration with multiple approaches can assist in overcoming PMF's inherent limitations and strengthen its effectiveness in exposure assessments and air pollution studies. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91875 |
DOI: | 10.6342/NTU202304557 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 環境與職業健康科學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-112-1.pdf 此日期後於網路公開 2028-12-24 | 40.52 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。