以自組特徵映射網路探討地下水資源時空變化特性

Exploring the Spatial-Temporal Variability of Groundwater Resources by Self-Organizing Maps

賴佳薇(Jia-Wei Lai)

指導教授 ： 張斐章

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/碩士(2017年)

https://doi.org/10.6342/NTU201701382

摘要

臺灣地區之地下水分佈廣且蘊含量豐富，促使地下水成為重要之水資源。屏東平原地下水為臺灣重要地下水資源之一，該地區地下水資源豐富，廣厚含水層透水性良好，卻因降雨特性的改變與地形限制間接影響了水資源之補注，再加上當地產業結構型態大量抽取地下水，故需掌握地下水資源變化情況。本研究以屏東平原作為研究區域，蒐集1999年至2015年間地下水位與雨量資料進行分析，並蒐集相關地質資訊討論關聯性，首先以月尺度之雨量與地下水位之關聯，比較長時間雨量與地下水位變化趨勢特徵，再以自組特徵映射網路聚類分析模式進行豐枯水期區域地下水位站特性分群，利用自組特徵映射網路神經元間鄰近關係的特性，合併探討神經元之相似性，找出各聚類的變異特性。分析結果發現：(1)該區域降雨分布不均，且雨量與地下水位變化量有高相關性；(2)空間分布結果顯示，豐水期與枯水期分類結果皆與地質之透水性關係密切，但枯水期之聚類D僅有含水層3-1層德興(2)位於沿海地區，此區域被檢測出有嚴重海水入侵；(3)時間分布結果顯示，豐水期雨量為屏東平原地下水補注主要來源之一，其中透水性較差之聚類各旬平均水位變化量與雨量(MA2)之趨勢變化有較高之相關性，而透水性較佳之聚類水位變化量與當旬平均雨量之變化趨勢相近，且初期地下水位上升快；枯水期因降雨量稀少，使用水多依靠地下水，經枯水期聚類結果顯示，聚類A、B與C之地下水受到各用水需求影響，聚類E位於透水性高之區域，初期水位高易流出，隨水位降低流出速率漸緩。本研究之成果，有助於了解豐枯水期地下水位之變動情況，可作為屏東平原水資源規劃管理之參考依據。

關鍵字

自組特徵映射網路 ； 豐枯水期 ； 地下水資源 ； 時空變化 ； 屏東平原

並列摘要

The abundant groundwater is an import resource in Taiwan. However, characteristics of rainfall plus the high-slope terrain, as well as seasonal over-exploitation greatly affect the groundwater recharge and result in severe problems. To make sustainable groundwater management plans, this study chose the Pingtung Plain as the study area to investigate how rainfall and geomorphology affect the changes in the groundwater level. Located in the southwest of Taiwan, the Pingtung Plain is rich in groundwater resources and has thick and highly permeable aquifers across the basin. This study collected the rainfall and groundwater level data from 1999 to 2015 in 121 stations in the plain and applied statistical methods to explore the relationship among rainfall, groundwater level and geochemical conditions using the monthly scale, so that the mechanisms of hydrology and sources of groundwater recharge can be better understood. Then this study applied two Self-Organizing Maps (SOMs) to explore the patterns of each clustering based on the connective algorithm between the neurons, and further separated and categorized the characteristics of regional groundwater level stations in wet and dry seasons. Analysis results indicated that: (1) There was an excessive and uneven temporal distribution of precipitation, to which the groundwater level variation was highly related; (2) The results of SOM appeared a spatial distribution reflecting the permeability of the geology across the plain that greatly influenced the groundwater recharge in both the wet and the dry seasons. Nonetheless, during the dry seasons, the Dexing 2 station was the only one station located in the coastal area with severe seawater intrusion problems in clustering D; (3) According to the temporal distribution of the SOM, results showed that rainfall in the wet seasons was one of the main sources for groundwater recharge. Moreover, the average groundwater level variations during the 10-day period (MA2) showed greater correlation with the trends of the rainfall in low-permeability clusters. Additionally, the average groundwater level variations during the 10-day period were similar to the rainfall in the high-permeability clusters, and the groundwater level rose rapidly particularly in the initial stage. In the dry seasons, the groundwater level descended much faster in the clusters A, B and C that reflected the reality of large groundwater exploitation by various human activities during the dry seasons. In the cluster E, the high permeability characteristics of this cluster made the outflow rates of the groundwater high when the groundwater level was high. On the contrary, the outflow rates declined when the groundwater level became lower due to the over-exploitation of the groundwater in the dry seasons. In conclusion, this study assessed the variation of the groundwater level in wet and dry seasons. It can be used to provide important information for sustainable groundwater resource management in the Pingtung Plain.

並列關鍵字

Self-Organizing Maps (SOM) ； Wet and dry seasons ； Groundwater resources ； Spatial-temporal variability ； The Pingtung Plain

參考文獻

38. 林承賢，2012，「以類神經網路建構濁水溪流域地下水位推估模式」，國立臺灣大學生物環境系統工程學系碩士論文。

39. 林政華，2012，「以類神經網路探討雲林沿海地區地下水砷濃度與水質特徵」，國立臺灣大學生物環境系統工程學系碩士論文

40. 林時猷、許少華，2004，「以數值模擬評估濁水溪沖積扇北彰化地區之地下水水文細節及人為抽水分佈」，農業工程學報，50(3)：48-58。

41. 洪耀明、萬絢、蘇苗彬、林裕益，2009，「崩塌地地下水位變化之即時預測」，水保技術，4(3)，181-189。

42. 徐年盛、江崇榮、汪中和、劉振宇、劉宏仁、黃建霖，2011，「地下水系統水平衡分析與補注源水量推估之研究」，中國土木水利工程學刊，23(4)，347-357。