多變量統計與時間序列分析於地下水質管理上之應用-以嘉南平原地下水分區為例

Application of multivariate statistical analysis and time series analysis on the management of groundwater quality : an example of Chianan Plain Groundwater Subregion

蘇秋生(Chiu-Sheng Su)

指導教授 ： 吳庭年

崑山科技大學/工程學院/環境工程研究所/碩士(2009年)

https://doi.org/10.6828/KSU.2009.00061

摘要

多變量統計分析之特性可將複雜度高的資料，簡化成具相關性的少數因子，或依資料間之相似度予以分群，及依資料在時間上的變化趨勢做預測。本研究利用多變量分析工具，解析地下水質監測資料，藉以探討地下水質污染特性、來源與趨勢變化，以期達到地下水資源之永續利用。以嘉南平原地下水分區為例，採用環保署「台灣省地下水質監測站網整體規劃」所設立之地下水質監測井之監測數據，藉由SPSS12.0統計軟體進行主成份因子及群集分析，歸納、整理和分類出水質特性與污染成因之關聯性及局部區域特徵，統計解析之水質監測項目包含酸鹼度、導電度、總硬度、總溶解固體、總有機碳、氨氮、硝酸鹽氮、氯鹽、硫酸鹽、鐵、錳、砷、鈉、鉀、鈣、鎂。由主成份因子分析結果獲致嘉南平原地下水分區水質之4個代表主成份因子：鹽化因子、砷環境因子、有機污染因子、溶礦因子。以4個主成份因子可替代原嘉南平原地下水分區水質16個分析項目，解釋整體變異量達82.4％。利用群集分析方法將區域內監測井，依監測數據性質相似度及群組間相異性質分為4群，彙整各群集內監測井水質特性，並配合監測井之相關位置，顯示內陸區域地下水質平均狀況優於沿海鄉鎮，嘉南平原沿海部分區域發現有海水入侵及鹽化情況。而高雄市多個行政區及鳳山、仁德、新營、太保等人口聚集之都會區地下水質呈現遭受有機污染之潛勢，另七股、北門、布袋等地區地下水質含砷量高於飲用水質標準，鑑定出之群集區位與部分文獻記載之烏腳病流行地區吻合。
針對本研究區域中烏腳病之流行地區，利用時間序列分析作為歷史時間資料礦掘之工具，因地下水含砷量為烏腳病之致病原因，藉由時間序列分析地下水砷濃度在時間上之變化趨勢，採用ARMA及ARIMA模式，解析本研究區域沿海鹽化群集與砷環境群集之39口監測井，輸入各監測井砷濃度之所有歷史數據資料，監測資料時間從2000到2007年，試圖建立砷濃度時間變化之預測模式，經篩選比較粗估模式，以模式選擇準則—配適度AIC(Akaike’s Information Criterion)準則評估後，建立之最簡單ARMA(1,1)模式優於其他模式，為本研究區域砷濃度在時間變化趨勢上最適合評估之模式，可提供地下水資源管理之參考。

關鍵字

因子分析 ； 群集分析 ； 時間序列分析 ； 地下水水質 ； 砷污染

並列摘要

Statistical multivariate analysis can reduce data dimension on the basis of variable correlation, classify data clusters according to their similarity, and predict the temporal trend of specific variable variations. For the sustainable use of groundwater, this study employed multivariate analysis as a tool of anatomizing groundwater quality data to realize the characteristics, sources, and temporal trend of groundwater contamination. The established monitoring wells in Chianan Plain groundwater subregion were all subjected to principal component analysis and cluster analysis by the SPSS 12.0 software. As a result, the characteristics of groundwater quality as well as the linkage of contaminant sources and local distribution can be discovered. Groundwater quality data including pH, electrical conductivity (EC), hardness, total dissolved solid (TDS), total organic carbon (TOC), ammonia, nitrate, chloride, sulfate, Fe, Mn, As, Na, K, Ca and Mg were extracted for statistical analysis. By using principal component analysis, the obtained four principal components (PCs) account for 82.4% of the variance or information contained in the original data set. The obtained four PCs represent four identified patterns of groundwater contamination as salinization, arsenic dissolution, organic pollution, and mineralization. The 2-step cluster analysis is utilized to classify the similarity among samples, and eighty four monitoring wells were accordingly classified into four clusters. Associating with the locations of monitoring wells, the result showed that the overall groundwater quality in highland region is superior to the coastal area. Seawater intrusion or salinization is the common case in Chianan Plain coastal area, and the potential organic pollution of groundwater is found around the crowd districts in Kaohsiung, Fengshan, Rende, Shinying, Taiban cities. The measured arsenic level of groundwater exceeds drinking water standard in Chigu, Beimen, and Budai coastal regions where geographically matches up with the reported historical Blackfoot disease region.
Time series analysis is used as data mining tool, and the Chianan Blackfoot disease region was selected as study area. Blackfoot disease is caused by the high uptake of arsenic in groundwater, and thus the temporal trend of arsenic concentration in groundwater is examined by time series analysis. ARMA and ARIMA, the common time series modeling methods, were employed to interpret the information beneath the monitoring data. Thirty-nine monitoring wells around the Chianan Blackfoot disease region were subjected to time series analysis, and the input data was extracted from historical monitoring data of arsenic concentration in groundwater during the time period of 2000 and 2007. The Akaike’s information criterion (AIC) is generally served as a criterion for assessing the quality of model fitting. It is based on residual log-likelihood function for model comparison. As a usual rule, the smaller AIC and simpler model tends better fitness for a give data set. Through further verification, the selected ARMA(1,1) model fits the data set well over the other three models ARMA(2,1), ARMA(1,2) and ARIMA(1,1,1). The result showed that this developed numerical model can effectively interpret and forecast the arsenic level in groundwater from area affected by salinization and high arsenic level in Chianan Plain based on the known information.

並列關鍵字

factor analysis ； cluster analysis ； time series analysis ； groundwater quality ； arsenic pollution

參考文獻

[31] 何秉均、尹彰，“基隆港外海有義波高時間序列不同轉換法之ARMA模式模擬研究”，海洋工程學刊，第2卷，第1期，第73-92頁，2002年。

[33] 陳文福，台灣的地下水，遠足文化事業股份有限公司出版，台北，2005年。

[34] 歐東坤，“嘉南地區地下水砷濃度之研究”，國立台灣大學，地質研究所，碩士論文，2005年。

[35] 劉麗純，“嘉南平原高砷地下水腐植物質的特性”，國立台灣大學，地質科學研究所，碩士論文，2007年。

[47] 黃俊英，多變量分析，華泰文化事業股份有限公司，第五版，台北，1995年。