應用多重時間解析度資料推估汙染物之來源並以靴拔重抽法評估模式解析結果之不確定性：以臺北某空氣品質測站為例

Source Apportionment of Multiple Time Resolution Data at an Air Quality Monitoring Station in Taipei and Utilizing Bootstrap Analysis for Uncertainty Estimation

黃淳聖(Chun-Sheng Huang)

指導教授 ： 吳章甫

國立臺灣大學/公共衛生學院/環境衛生研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.00857

摘要

本研究以受體模式（Receptor model）之一－正矩陣因子法（Positive Matrix Factorization, PMF），應用於具多重時間解析度（Multiple time resolution）之揮發性有機化合物（volatile organic compounds, VOCs）與細懸浮微粒PM2.5（fine particulate matter）監測資料，進行汙染物之來源推估（Source apportionment）。除此之外，並以靴拔重抽法（Bootstrap analysis）評估模式解析結果之不確定性（Uncertainty），以獲得更加準確的解析結果。
本研究使用萬華測站於2014年10月5日至2014年11月6日與2014年12月9日至2015年1月19日所監測之每小時揮發性有機化合物資料，共1104筆，同時每24小時採集一筆細懸浮微粒PM2.5樣本，共46筆，應用於正矩陣因子法，以解析出測站周邊之可能汙染來源。但由於正矩陣因子法模式的過度配適（over-fitted），可能會使得模式解析出部分不顯著或不存在的汙染來源，進而影響後續汙染源指紋圖譜（source profile）之闡釋。因此，本研究利用靴拔重抽法以評估模式解析結果之不確定性。相較於其他解析結果，7個汙染源解析結果擁有最佳的再現性（reproducibility）。並且，在與僅使用揮發性有機化合物資料分析結果比較後，發現加入低時間解析度細懸浮微粒PM2.5資料對於汙染源辨識有所助益，卻也會增加其在靴拔重抽法分析中的不確定性。
根據汙染源指紋圖譜，推估萬華測站周邊可能有7種主要汙染來源，分別為交通排放（vehicle）、工業（industry）、老化海鹽�地域傳播（aged sea salt/transported）、芳香族（aromatics）、燃料揮發（fuel evaporation）、天然氣�二次硝酸鹽（natural gas/secondary nitrate）與二次硫酸鹽�長程傳輸（secondary sulfate/long-range transport）。在萬華測站周邊，揮發性有機化合物之主要貢獻來源為交通排放，佔26%。細懸浮微粒PM2.5之主要貢獻來源則為二次硫酸鹽�長程傳輸，佔36%。
然而，模式之不確定性可能會增加物種濃度估計的變異性，進而影響後續所推估之污染源貢獻比例。由於正矩陣因子法模式之限制，會解析出部分混合型之汙染來源。因此，後續研究可透過對模式解析結果進行限縮（Constraint），以獲取更加精確的解析結果。此外，若能夠加入春季與夏季的資料，提供更長期的監測，亦能夠使得暴露評估的資訊更加完整。

關鍵字

受體模式 ； 正矩陣因子法 ； 多重時間解析度資料 ； 來源推估 ； 靴拔重抽法 ； 不確定性

並列摘要

This study applied multiple time resolution data of volatile organic compounds (VOCs) and fine particulate matter (PM2.5) in the receptor model of positive matrix factorization (PMF) for source apportionment. Furthermore, bootstrap analysis was utilized for uncertainty estimation to obtain a more accurate modeling result.
A total of 1104 hourly VOC data and 46 24-hr PM2.5 samples were collected from the Wanhua monitoring station in Taiwan during October 5th to November 6th in 2014 and December 9th in 2014 to January 19th in 2015. These data were applied in the PMF model for source apportionment. However, the over-fitted modeling results from the PMF analysis may create insignificant or phantom factors, and influence the subsequent interpretation of source profile. Bootstrap analysis was further used for assessing the uncertainty from analytic results in this study. Comparing with the other modeling solutions, the 7-factor solution with the best reproducibility was included for the following analysis. In addition, comparing with VOC-only data modeling, the results showed that adding the low-resolution PM2.5 data was beneficial for source identification but increased the uncertainty in bootstrap analysis.
Based on the retrieved source profiles, seven sources were interpreted as: vehicle, industry, aged sea salt/transported, aromatics, fuel evaporation, natural gas/secondary nitrate and secondary sulfate/long-range transport. At Wanhua monitoring site, the largest contributor of VOCs was vehicle source, accounting for 26%. PM2.5 was mainly contributed by secondary sulfate/long-range transport (36%).
Nevertheless, the uncertainty from PMF model may cause the variability of concentration estimates, and affect the posterior source contribution estimates of each source. The limitation from PMF model may also lead to the analytic result with mixed sources. Therefore, through adding constraints in PMF analyses should be considered in future studies. Furthermore, inclusion of spring and summer season data will also provide a long-term monitoring dataset for sufficient exposure assessment.

並列關鍵字

Receptor Model ； Positive Matrix Factorization (PMF) ； Multiple Time Resolution Data ； Source Apportionment ； Bootstrap ； Uncertainty

參考文獻

郭承彬, 2013. 多時間解析度資料於推估懸浮微粒與揮發性有機氣體來源之效用, (碩士論文), 臺灣大學.

Acikara, Ö. B. (2013). Ion-Exchange Chromatography and Its Applications. Edited by Dean F. Martin and Barbara B. Martin, 31.

Almeida, S., Pio, C., Freitas, M., Reis, M., & Trancoso, M. (2005). Source apportionment of fine and coarse particulate matter in a sub-urban area at the Western European Coast. Atmospheric Environment, 39(17), 3127-3138.

Amato, F., Pandolfi, M., Escrig, A., Querol, X., Alastuey, A., Pey, J., Perez, N., & Hopke, P. (2009). Quantifying road dust resuspension in urban environment by multilinear engine: a comparison with PMF2. Atmospheric Environment, 43(17), 2770-2780.

Anderson, T. W. (1984). An introduction to multivariate statistical analysis (2nd edition ed. Vol. 2): Wiley New York.