近年來空氣污染日趨嚴重，是一個備受全球重視與討論的議題。空氣污染之嚴重程度除了受都市化情況影響外，也會因氣候型態的差異而有所不同，例如氣流的擴散或化學反應等因素均會導致污染物的累積和跨縣市的傳播。本研究利用台灣2011年74個測站的空氣污染資料，分別針對10微米懸浮微粒(PM10)、二氧化硫(SO2)、一氧化碳(CO)、臭氧(O3)、二氧化氮(NO2)及空氣污染指標(pollutant standards index, PSI)，探討這74個測站之污染情形的分群結構。本研究將空污資料視為函數型資料，並利用Li 與Chiou (2011) 所提出的子空間投影之函數型分群演算法(subspace projected functional clustering, SPFC)，根據資料的平均趨勢及相異時間點的共變異結構對74 個測站進行分群，最後討論其分群結果。此外，本文亦進一步探討各污染物及空氣污染指標的最適分群結果。由最後的分群結果可以發現，利用子空間投影分群演算法除了可以找出各污染物的幾種主要變化型態外，亦發現台灣空污情形的分佈具有地域性。因此，本研究的分群結果可提供未來空污管理或決策時的參考依據。 In recent years, air pollution is getting worse and becomes an important issue in the world. The severity of air pollution is affected not only by urbanization but also climatic differences. For instance, the airflow diffuser or chemical reaction leads to cumulative pollutant and flow across the counties. In this study, we aim to investigate the distribution structures of several air pollutants in Taiwan through a cluster analysis. We apply the subspace projected functional clustering (SPFC) algorithm proposed by Li and Chiou (2011) to the 2011 air pollution data, including five pollutants PM10, SO2, CO, O3, NO2 and the pollution standards index (PSI), of 74 monitoring stations in Taiwan. We consider the daily collected air pollution data as functional data and cluster with the stations according to both the means and the modes of variation differentials among clusters. In addition, we also disscuss the optimal number of clusters for each pollutant.In summary, the clustering results show that the air pollution in Taiwan is mainly affected by climate and topography, and the grouping structures of some pollutants are related. The reasonable clustering results in this study can provide useful information for the environmental protection.
