圖書典藏及影音數位平台

高屏溪流域，水資源相當豐沛，廣厚地下含水層透水性良好，為台灣主要地下水資源地區之一，惟因降雨時空分布極不平均，造成豐枯水期變化明顯，加上近年來氣候變遷的因素，豐枯水期降雨差距更大，降雨特性的改變，直接或間接影響地下水補注效益，爰地下水資源管理亟需有所因應與規劃。 本署在高屏溪流域具有眾多長期的地面水及地下水監測資料，若能透過加值分析，提供實用性的資訊，分析萃取地面/地下水資源交互作用機制，藉由資訊整合平台展示集水區水文資訊，即時掌握及預測地下水位變化趨勢，將能即早針對當地地下水文情勢，提出因應對策。本計畫藉由蒐集與整理高屏溪流域內眾多流量站、地下水觀測站、水文地質等長期觀測資料，分析全流域的降雨、河川流量、地下水間的關係，利用資料探勘技術建置全流域智慧型長短期預測模式，並開發展示於資訊整合平台，俾作為台灣地下水資源管理方式之參考依據。

The Gaoping River Basin is one of the main areas with ample groundwater resources in Taiwan. The basin possesses abundant water, and the aquifers across the basin are thick and highly permeable. To make sustainable groundwater management plans, it is critical to investigate how rainfall and geomorphology affects the changes in groundwater level. This project aimed to explore the interactive mechanisms between surface water and groundwater, understand groundwater resources status and establish regional groundwater level forecasting models. This project was implemented in two years, and the main tasks included the collection and analysis of hydro-geo-meteorological data, the exploration of the interactive mechanisms of hydrological systems in the Gaoping River Basin, the construction of regional groundwater level forecasting models, and the delivery of technical transfer training. The first-year project has been completed in 2015 with main results including: the collection of hydro-meteorological data, the partition of the study area into eastern and western zones in consideration of the hydrological characteristics, the impact analysis results of the hydrological events, water balance analysis, and the platform construction of demonstration of the project results which was primarily based on Google Map. This is the second-year project, and the main tasks are addressed as follows. The first task was the collection and review of national and international literatures of groundwater forecasting models, and the results indicated that Artificial Neural Networks (ANN) and MODFLOW were two main methods for groundwater forecasting. The second task was to assess the relationship among long-period rainfall, river flow and groundwater level, and the results are addressed as follows. (1) The current groundwater level was highly correlated with one-two-month antecedent surface water (i.e., river flow and rainfall), which showed a time delay phenomenon and revealed there was a direct relationship between surface water and groundwater level variations. (2) The 1st aquifer of the eastern zone and the 2nd aquifer of the western zone could be easily recharged from rainfall as well as discharged into rivers. (3) The SOM clustering results of the continuous 10-day data identified similar trends in groundwater level variation, accumulative rainfall and river flow. 10-day average rainfall of 180 mm could be considered as the threshold of groundwater level variation, which could be a guideline for the management of surface-groundwater resources in the Gaoping River Basin. (4) The SOM clustering results of 30-day moving average data indicated that significant differences in groundwater level could be identified before and after 2005. The analysis on the horizontal spatial distribution of groundwater level variation indicated the groundwater level variation decreased from the east to the west while the analysis on the vertical spatial distribution of groundwater level, rainfall and river flow indicated rainfall influenced river flow highly but influenced water levels of aquifers indirectly. Groundwater levels were much affected by the long-term trends of rainfall in both wet and dry seasons. The basin-wide monthly groundwater level forecasting models were constructed and the demonstration platform was also built, in which the SOM was used to classify the groundwater distribution in the study basin for obtaining a topological map and the Recurrent Configuration of Nonlinear Autoregressive with Exogenous Inputs (R-NARX) was applied to estimating the total variation of basin-wide groundwater level. The estimated amount was used to compare and calibrate the closest cluster in the SOM topological map such that the forecasting model could be developed through training, validating and testing phases. The forecast results suggested that the simulations were quite accurate and clearly demonstrated a map for the basin-wide groundwater level status. Besides, the proposed model performed better than numerical models such as MODFLOW through tackling the problems of assuming numerous parameters, consuming long simulation time and verifying the hypothesis. This approach can be used to make regional groundwater level forecasts for the coming 1-3 months, assess the changes in basin-wide groundwater level, and provide scientific-based information for water resources management and allocation strategies. After constructing groundwater forecasting models and the demonstration platform, a technical transfer training equipped with an introduction of the theories used and models proposed in this project was held, and the “groundwater demonstration and forecasting model user guide” was provided so that WRA members could better understand the proposed groundwater level forecasting models and related methodologies used in this project.