應用時頻分析與分群方法於含水層判識之研究

Abstract

台灣地區地下水供應超過全部用水量的三分之一，因此妥善管理運用地下水資源，以降低地下水相關災害及減緩地表水之供水負擔，乃是當務之急。而良好地下水管理的基礎首在水文地質架構的掌握，目前全台地區地下水觀測站網已累積大量地下水位資料，而地下水位反映了含水層系統內部之壓力變化，其乃是外界刺激(如降雨、抽水與地潮)及含水層特性的綜合反應，因此可藉由水位變化分析及物理機制的解釋，輔助進行水文地質架構之判識。因此本研究乃以地下水位資料為基礎，整合時頻分析與分群方法進行含水層判識。 前述時頻分析乃以傅立葉轉換(Fourier Transform, FT)為基礎，可分析構成地下水位變化之各種不同週期與振幅的信號分量，本研究以小時資料為基礎，應用Visual Signal軟體進行時頻分析，另外考量降雨、人為抽水及地潮等主要地下水外界干擾源，各自對受壓及非受壓含水層的不同影響等，而以各水井一天一次、一天1.93次與一年一次不同頻率下之水位振幅為主要反應指標，其中日頻率水位振幅與地潮頻率水位振幅明顯者為受壓含水層，而年頻率水位振幅大者為非受壓含水層，並以此三種頻率之水位振幅為基礎，應用K-mean分群法，以各頻率振幅相對大小進行含水層初步判識；接著針對分群方法不易判識之水井，再以年、日頻率振幅大小，觀察其水位變化，決定出年頻率振幅/日頻率振幅之含水層分類門檻值，並以此門檻值進行含水層種類判識。 為驗證前述發展之判識流程與門檻值訂定，本研究以濁水溪沖積扇、屏東平原、蘭陽平原等地區作實例驗證，各區域驗證結果之辨識率分別為，濁水溪沖積扇：0.9，屏東平原：0.88，蘭陽平原：0.86，此結果證明此門檻值之準確性與可行性。傳統上水文地質架構的分析，多以地質鑽探資料為主，由於成本限制其數量大多有限，本研究發展的判識方法可與傳統的地質剖面分析相輔相成，並可增加地下水位觀測資料的附加價值。

Groundwater supplies more than one third of the water demand in Taiwan. Therefore, proper groundwater management to reduce groundwater-related disasters and mitigate the water supply shortage is an important issue. A proper groundwater management mainly relies on the understanding of hydrogeological structure. In Taiwan, the groundwater monitoring network has collected large amount of groundwater level data. The groundwater level data can reflect the pressure changes in the aquifers responding to the factors such as, precipitation, pumping, earth tide and aquifer characteristics. Hence, analysis of the groundwater level data combined with physic mechanism interpretation can be used to identify the aquifer hydrogeological structure. Accordingly, this study uses time-frequency analysis and clustering methods to develop a procedure to identify the confined and unconfined aquifers. The time-frequency analysis used Fourier Transform (FT) to analyze the groundwater level data and obtained different periods and amplitudes of the signal. The hourly data are used in this study and the commercially available software, Visual Signal, is used for the analysis. Considering the different responses of confined and unconfined aquifers to the precipitation, pumping and earth tide, the amplitudes of the frequency for once-a-day, 1.93 times-a-day and once-a-year are considered as the main response indicators. The results show that confined and unconfined aquifers have clear daily and annual frequency of groundwater level respectively. The K-mean clustering method based on the response factors followed with rational analysis of each clustered groups, is applied to identify the confined or unconfined aquifers. The proposed methodology was applied to identify the aquifer systems of Zhuoshui River alluvial fan, Pingtung Plain, and Yilan Plain. The results showed that the ratio of the daily amplitude to the yearly amplitude of the groundwater level can be a good index to identify the aquifer. The critical value of the day-to-year ratio is 60. The correct identification rates are 0.9, 0.88, and 0.86 for the three studied areas respectively when applying the ratio to identify the aquifer system. This result demonstrated the feasibility the proposed method. Traditionally, geological drilling data are used to identify the hydrogeological structure. However, the drilling data are always limited due to budget restriction. The proposed method is a valuable alternative to assist the hydrogeological structure identification and a great value added to the groundwater level data.