統計診斷技術應用於地下水質管理－以南部科學園區為例

目前有許多統計軟體可以把巨量資料且複雜性高的資料，快速簡化成相關性資料且容易辨別。本研究運用主成分分析與群集分析來分析台灣南部科學園區地下水質監測資料，探討地下水質污染特性來源與分布，以期達到地下水質管理之預警功能。研究資料來自台灣南部科學園區設立之地下水質監測井網監測數據，分析水質監測項目包含pH、亞硝酸鹽氮、氟化物、氨氮、砷、酚類、硝酸鹽氮、溫度、導電度、錳、總有機碳、總溶解固體、鐵。由主成分分析結果得到南部科學園區台南園區影響地下水質變異的5個主成分因子：鹽化因子、砷因子、氟因子、溶礦因子、農業污染因子可代表影響地下水質之潛在污染成因；南部科學園區路竹園區影響地下水質變異的5個主成分因子：鹽化因子、酸性因子、溶礦因子、砷因子、氟因子。以相同數據項目進行群集分析，採用兩階段分群法作為群集合併原則，依不同污染屬性之地下水質進行分類，將區域內監測井，依性質相似度及群組間相異性質分為5個群集，5個群集區域分別對應影響地下水質之5種主成分因子以了解不同潛在汙染成因影響地下水質之區域分布。研究區域為高科技產業地區，園區內高科技廠商使用之氟化物，與地下水質氟的來源有密切關連性，而地下水質氟濃度一般極低，不易由監測數據判斷是否有洩漏源存在，本研究利用時間序列工具進行歷史時間性之資料礦掘，剖析地下水質氟濃度之時間變化趨勢。時間序列分析採用ARMA及ARIMA模式，解析研究區域內判定為氟因子影響區域範圍之7口監測井，輸入各監測井氟濃度之所有歷史數據監測資料，監測資料從2005到2009經篩選比較粗估模式，由ACF圖及PACF圖判定顯示出並無特別顯著突出的峰值，再以模式選擇準則─配適度AIC及SBC兩種準則評估，判定ARMA(1,1)模式為本研究區域氟濃度在時間變化趨勢上最適合評估模式，且顯現預估之地下水質氟濃度有逐年遞減之變化趨勢。

關鍵字

主成分分析；群集分析；時間序列分析；地下水水質；氟

並列摘要

Nowadays, statistical software is capable of simplifying large and complex data into easy-defined and relevant information quickly. This research applied principal component analysis (PCA) and cluster analysis (CA) to anatomize groundwater quality around Southern Taiwan Science Park, and also the possible contamination sources and their spatial distribution were discovered to restore the alarm function of groundwater quality management system. Data source originated from Southern Taiwan Science Park groundwater quality monitoring well databank. Monitoring items of groundwater quality include pH, nitrite (NO2-), fluoride (F-), ammonia (NH3), arsenic (As), phenols, nitrate (NO3-), temperature, electrical conductivity (EC), manganese (Mn), total organic carbon (TOC), total dissolved solid (TDS), iron (Fe). PCA results illustrated 5 principal components (PCs) contributing variance of groundwater quality in Tainan district of Southern Taiwan Science Park, and 5 PCs are salinization, arsenic, fluoride, mineralization and agricultural pollution that can directly reflect potential sources of groundwater contamination. 5 PCs contributing variance of groundwater quality in Lu-ju district of Southern Taiwan Science Park were identified as salinization, industrial leakage, mineralization, arsenic, and fluoride. Monitoring data of groundwater quality is subsequently subjected to CA, which is following two-step clustering procedures to aggregate and classify different groups based on their relevance and similarity. Monitoring wells are divided into 5 clusters to correspond with the 5 identified PCs in each district of Southern Taiwan Science Park, and thus the affecting domain of each potential sources of groundwater contamination can be spatially allocated. There are many high technology manufacturers located at the studied area, and fluoride is commonly used in their manufacturing processes. In general, fluoride concentration is very low in groundwater that is challenging to examine the presence of leakage source based on monitoring data. In this study, time series analysis was employed to anatomize the temporal trend of fluctuating fluoride concentrations in groundwater. ARMA and the ARIMA models were used for time series analysis. Only 7 monitoring wells located in fluoride affecting domain were subjected to time series analysis, and monitoring data are extracted from 2005 to 2009. The results showed that there is no remarkable peak on ACF and PACF charts. Based on evaluation of AIC and SBC criteria, ARMA(1,1) model is the most suitable one to predict temporal fluoride concentration. The predicting equation illustrated that the fluoride concentration is temporally decreasing in groundwater.

並列關鍵字

principal component analysis ； cluster analysis ； time series analysis ； groundwater quality ； fluoride

參考文獻

[2] 張文亮，水井診斷維護管理與永續經營，地景企業股份有限公司，台北，1999。

[21] 蘇秋生，“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”，碩士論文，崑山科技大學，2009。

[29] 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。

[31] 楊浩二，多變量統計方法，華泰文化事業股份有限公司，台北，1995。

[3] 南部科學工業園區，台南園區地下水監測網，富立業工程顧問股份有限公司制作維護， http://www.stsipa.gov.tw/web/。

國際替代計量

統計診斷技術應用於地下水質管理－以南部科學園區為例

全文下載

主題瀏覽