整合多變量方法評估水體及底泥品質時空特性之研究

Assessment of Spatial-Temporal Characteristics of Water and Sediment Quality using Integrated Multivariate Methods

王嶽斌(Yeuh-Bin Wang)

指導教授 ： 劉振宇教授

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/博士(2015年)

https://doi.org/10.6342/NTU.2015.11013

摘要

多變量統計方法為分析環境監測資料重要而有效率的工具，可將監測數據加值深度分析，以發掘環境資料內涵，獲取更多且有價值的環境資訊。本研究以金門地下水品質、淡水河水質及二仁溪底泥品質為研究區域，探討監測數據資料之特徵，並發掘其品質問題，並研提問題對策或建議。
本研究在金門地下水水質分析案例中，係運用因子分析(Factor analysis, FA)、群集分析(Cluster analysis, CA) 及自組織映射圖(Self organizing map, SOM)等多變量分析工具詮釋水質資料，探討空間及時間變異特性，分析結果顯示硝酸鹽及有機污染因子、氧化因子及鹽化因子為支配當地地下水水質變異之控制因子。此外，金門地下水硝酸鹽污染嚴重問題，本研究建立邏輯式迴歸(Logistic regression, LR)統計模式，顯示地下水硝酸鹽氮超過飲用水水質標準之機率與井位土壤種類、地下水之導電度(EC) 及酸鹼值(pH)等參數顯著相關，利用該特性可快速評估民井地下水是否受到硝酸鹽氮污染。
其次，在淡水河水質監測資料分析之研究案例，係運用多變量分析工具FA、CA及SOM可整合詮釋水質監測資料，探討水質時空變化特性，發掘污染分布特性及水質改善變化趨勢，除顯示淡水河水質過去二十年期間已有大幅改善外，並建議升級污水廠及加強大漢溪上游水土保持對策，可進一步改善水質。另基於空間水質相似性，可據以簡化10處水質監測站，而不會遺失重要的水質資訊；另再結合區別分析(Discriminate analysis, DA)可鑑別出在決定污染程度上有顯著差異的水質項目，研究成果可供簡化水質監測計畫參考。另外，本案例研究亦再就綜合考慮近三年水質之時間及空間相似性，發現短期內淡水河水質變化受降雨量影響顯著，在中、下游河段，因降雨使有機污染濃度因稀釋作業而降低，但在中、上游河段水質，因逕流沖刷使懸浮固體物濃度隨降雨而增加，該現象可供現行水質監測採樣時間調整參考。
最後，二仁溪底泥品質監測資料分析之案例研究，係運用SOM及FA探討在雨、乾季時之空間變化特性，結果顯示乾季時支流三爺宮溪受傳統電鍍業重金屬廢水污染嚴重；在雨季時，受有異常多元芳香烴污染明顯。另進一步分析底泥重金屬累積，以Cr及Cu重金屬污染累積情形最為嚴重，評估污染生態風險，在三爺宮溪已達中度或高度生態危害風險，但對人體健康危害HI (Hazardous Index)值小於1。至於多元芳香烴污染程度，除在S6採樣點受異常事件污染外，全流域尚無顯著的生態風險，但人體致癌健康風險達10-4數量級，屬中度健康危害風險。運用正矩陣因子法(Positive Matrix Factorization, PMF)除分析主要污染源之貢獻比例外，可再針對風險評估結果，進一步推估污染物主要來源對危害風險之貢獻比例，結果二仁溪及支流三爺宮溪經攝食途徑造成重金屬危害健康風險之最主要貢獻，分別為地質及非點源污染(46%)、電鍍及表面金屬處理業點源污染(47%)為最高；另二仁溪底泥多元芳香烴毒性當量之貢獻比例，以石化工業區燃燒來源達56%最高，量化主要污染來源對污染物總量及對應所造成之健康風險的貢獻比例，可供改善底泥品質對策研擬參考。

關鍵字

水質評估 ； 因子分析 ； 自組織映射圖 ； 底泥品質 ； 風險評估 ； 正矩陣因子法

並列摘要

Multivariate methods are very efficient to interpret environmental monitoring data in depth. We can explore latent intension and get more worthy insight from monitoring data by using multivariate methods. This study used groundwater quality of Kinmen Island, water quality of Tamsui River, and sediment quality of Erjen River as case studies to comprehend the characteristics of water and sediment quality monitoring data by applying multivariate methods. These studies could find out the core issues to the water and sediment quality from the comprehended characteristics, and then thses studies suggested measures to improve thses issues.
The first case study applied multivariate methods, involving cluster analysis (CA), factor analysis (FA), discriminate analysis (DA) and self organizing map (SOM), to interpret the spatial and temporal characteristics of groundwater quality in Kinmen. The nitrate and organic contamination is the major factor dominating groundwater quality in Kinmen. For further assessing nitrate contamination, this study used logistic regression (LR) to find that the soil type, pH, and EC have close relationship of nitrate contamination. The established LR model can be used for preliminary evaluation of nitrate contamination in groundwater. And the application of the model is to predict the probability of exceeding nitrate threshold and to draw the probability map of nitrate contamination. The model can also be applied to develop a handy tool using EC and pH for preliminary evaluation of nitrate contamination in private wells water.
The secondary case study also integrated these aforementioned multivariate methods to evaluate the spatial and temporal variance of water quality in the Tamsui River. This work indicated that the water quality of Tamsui River has been improving to better status and monitoring station can be simplified. This work plotted a spatial pattern using the four latent factor scores and identified 10 redundant monitoring stations near each upstream station with the same score pattern. Finally, for further improving water quality of the Tamsui River, this study also used positive matrix factorization (PMF) to identify the ratio of contribution from the each major pollution. The result of this work can suggest Taiwan EPA adopt some measures to eliminate major pollution.
The third case study explored and compared spatial characteristics of sediment quality of the Erjen River in rainy and dry season by coupling FA and SOM methods. The result of FA and SOM indicated the wastewater that discharged from metal electroplate plants polluted seriously the sediment of the Sanyegong Creek in dry season, but PAHs also polluted unusually the sediment in rainy season. The work also assessed accumulation of heavy metal by using Igeo index and found out two heavy metals, Cr and Cu, accumulated heavily in sediment. The biological risk of heavy metal was evaluated as moderate and high risk in the Erjen River, but hazardous index value of the health risk caused by heavy metal was less than 1. Furthermore, this work used positive matrix factorization method (PMF) to estimate the contribution ratio of the each major heavy metal pollution source to health risk. The geological and nonpoint source of heavy metal contributed 46% health risk in the main stream of the Erjen River and the wastewater from metal electroplate plant also contributed 46% health risk in the tributary stream of the Erjen River, the Sanyegong Creek. As to the assessment of PAHs pollution in sediment, the biological risk caused by PAHs was very little except the unusual polluted event in S6 site. But the carcinogenic risk was the 10-4 level and could be assessed as moderate health risk. The petrochemical industry complex source contributed 56% toxicity caused by PAHs in sediment of the Erjen River.

並列關鍵字

water quality assessment ； factor analysis ； self organizing map ； sediment quality ； risk assessment ； positive matrix factorization

參考文獻

巨廷工程顧問股份有限公司。2011。金門地區地下水資源之管理及運用策略研究計畫成果報告。金門縣政府。

Chen, Y. C., Chen, C. Y., Hwang, H. J., Chang, W. B., Yeh, W. J., Chen, M. H. (2004). Comparison of the metal concentrations in muscle and liver tissues of fishes from the Erren River, Southwestern Taiwan, after the restoration in 2000. Journal of Food and Drug Analysis, 12, 358-366.

Aelion, C. M., Conte, B. C. (2004). Susceptibility of Residential wells to VOC and nitrate contamination. Environmental Science and Technology, 38(6), 1648-1653.

Alvarvez-Guerra, M., Gonzalez-Pinuela C., Andres, A., Galan, B., Viguri, J. R. (2008). Assessment of self-organizing map artificial neural networks for the classification of sediment quality. Environmental International, 34, 782-790.

Astel, A., Tsakovski, S., Barbien, P., Simeonov, V. (2007). Comparison of self-organizing maps classification approach with cluster and principal components analysis for a large environmental data sets. Water Research, 41, 4566-4578.