應用時頻分析與水筒模式預測地下水監測井群之水位變化－以濁水溪沖積扇為例

Predicting Groundwater Levels at Monitoring Wells Using Time-frequency analysis and Tank Model - A Case Study of the Choushui River Alluvial Fan

10.6937/TWC.202406_72(2).0002

許少華(SHAO-HUA MARKO HSU)；吳宗燁(ZONG-YE WU)；朱政哲(CHENG-JE CHU)；陳星合(XING-HE CHEN)；韋正(ABDOUL RACHID OUÉDRAOGO)

時頻分析 ； 水筒模式 ； 預測模型 ； 地下水水位 ； 濁水溪沖積扇 ； Time-frequency analysis ； Tank Model ； Prediction Model ； Groundwater Levels ； Choushui River Alluvial Fan

台灣水利

72卷2期（2024 / 06 / 01）

21 - 49

繁體中文;英文

本研究以濁水溪沖積扇之地下水監測井井群的水位時間序列資料作為基礎，加上多處雨量站之降雨時間序列資料，並整合頻率分析、降雨對地下水位的交叉相關影響，建立以資料探勘（Data Mining）之機器學習方式延伸水筒模式（Tank Model）之降雨逕流關係之演算模型。經由井群間地下水水位漲落之相似情況建構整個含水層之上下游水流網路系統，應用於地下水自由含水層（第一層，非拘限含水層）之估算。前述所提時間序位資料採用2012年至2020年共八年之濁水溪沖積扇之小時地下水水位與日降雨量，頻率分析以小時地下水水位之時間序列資料透過傅立葉轉換，分析因人為抽水所造成地下水位變化之日頻率振幅大小作為抽水之影響與濁水溪沖積扇範圍內因潮汐所造成之地下水位變化。地下水位消退來自於自然流失與人為抽水之共同影響，而地下水位的抬升來自於降雨入滲補注和上游地下水之自然入流，整合地下水位消退量和補注量為模型中使用的基本參數，結合線性水庫和水筒模式之概念，建立了本研究使用的地下水位模型，透過機器學習便能模擬預測地下水位之變化。本研究發展之地下水位預測模式，具有良好之精確度可證明本模型之可行性與準確性。透由整體之數學關係式可瞭解濁水溪沖積扇地下水位受到降雨量、抽水、潮汐之關係強弱，並可掌握影響濁水溪沖積扇地下水位變化的關鍵因素。

This study utilizes the time series data of groundwater monitoring wells in the alluvial fan of the Choushui River as the basis, supplemented by rainfall time series data from multiple rain-gauge stations. By integrating frequency analysis and the cross-correlation impact of rainfall on groundwater levels, a computational model is established using machine learning techniques in data mining to extend the rainfall-runoff relationship of the tank model. The similarity of groundwater level fluctuations between well clusters is utilized to construct a groundwater flow network system for the entire aquifer, applied to estimate the unconfined aquifer (first layer) groundwater in the alluvial fan. The aforementioned time series data covers hourly groundwater levels and daily rainfall from 2012 to 2020 in the Choushui River alluvial fan. Frequency analysis of hourly groundwater level time series data is conducted using Fourier transform to analyze the daily frequency amplitude of groundwater level changes caused by pumping (AMP) and tidal influences (AMT) within the alluvial fan area. Groundwater level recession results from natural losses and pumping effects, while groundwater level rise is due to rainfall infiltration recharge and natural inflow from upstream groundwater. Integrating groundwater recession and recharge amounts as basic parameters in the model, along with the concepts of linear reservoir and tank models, establishes the groundwater level model used in this study, capable of simulating and predicting groundwater level changes through machine learning. The developed groundwater level prediction model in this study demonstrates good accuracy, proving the feasibility and precision of the model. The overall mathematical relationships provide insights into the strength of relationships between rainfall, pumping, tides, and groundwater levels in the Choushui River alluvial fan, enabling the identification of key factors influencing groundwater level changes.