應用獨立成分分析法分離地下水位成份之研究－以蘭陽平原為例

EXTRACTING COHERENT COMPONENT OF GROUNDWATER SYSTEM BY INDEPENDENT COMPONENT ANALYSIS: A CASE STUDY OF LANYANG PLAIN

10.6652/JoCICHE.201512_27(4).0007

蕭金財(Chin-Tsai Hsiao)；蔡瑞彬(Jui-Pin Tsai)；陳宇文(Yu-Wen Chen)；張良正(Liang-Cheng Chang)；陳祐誠(You-Cheng Chen)

獨立成份分析 ； 主成份分析 ； 同調性 ； 地下水 ； independent component analysis ； principal component analysis ； coherence ； groundwater

中國土木水利工程學刊

27卷4期（2015 / 12 / 01）

345 - 355

繁體中文

由於水位時序資料為各種物理機制影響之綜合效應，因此若能分離不同物理機制之效應，將有助於地下水管理問題之分析。本研究應用獨立成分分析法 (Independent Component Analysis, ICA) 於地下水位時序資料分析，用以分離不同物理機制對地下水位之影響。本研究首先應用地下水數值模式進行數值試驗，並以降雨補注模組與兩個不同抽水型態之抽水井群建立模式，再由MODFLOW 模擬其水位變化，並應用模擬水位作為觀測資料，再由獨立成分分析，分離出三大獨立成分，最後再與原降雨補注資料以及兩井群之抽水資料進行比較，以驗證獨立成分分析法之適用性。其次，本研究再應用於現地問題上，以蘭陽平原22 口觀測井，2005 年至2010 年之日觀測資料進行獨立成分分析，成功分離兩大獨立成分，其分別代表降雨補注及河川側流對地下水之影響。本研究成果可從混合不同物理機制之地下水訊號中，分離出不同之獨立成分，有助於區域地下水分析與系統管理之參考。

Groundwater fluctuations occur in response to a number of different physical mechanisms. If groundwater fluctuations can be broken down into several independent components, groundwater management problems can be better analyzed and addressed. This study investigates the ability of independent component analysis (ICA) to divide the spatial and temporal fluctuations of groundwater levels into several independent components, which are caused by various physical mechanisms. We applied a numerical model with one recharge source and two groups of pumping wells with different frequencies of the assigned time series for each of them. The model was simulated by MODFLOW and the simulation results at the observation wells were the ICA input. The method divided the mixed signals into three independent components, thus demonstrating the applicability of ICA. After verifying the applicability of ICA, we applied this method to a set of 22 daily observed groundwater levels from 2005 to 2010 within the Lanyang Plain. This study obtained two independent components representing the effects of rainfall and rivers. The results indicate that the proposed method can be used to interpret the signals of groundwater levels and to identify the major physical mechanism. Therefore, this method can serve as a useful reference for the analysis and management of regional groundwater systems.