結合資料科學輔助地下水模式建置-以臺北盆地為例

Application of Data Science to Support Groundwater Modeling – A Case Study in Taipei Basin

曾華廷(Hua-Ting Tseng)

指導教授 ： 余化龍

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(2020年)

https://doi.org/10.6342/NTU202003781

摘要

欲更了解地下水的流動方式，地下水模型的建置一直都是常用的方式，然而過往許多研究在模型的輸入方面，諸如地下水分層的層數，含水層阻水層邊界的位置，水文地質的參數與邊界條件等，都是採取較為主觀的方法，如此導致在同一個研究區，每個研究者設計出來的模式都不盡相同，模擬之結果也都不一樣，甚至有極大的差異。故本研究嘗試以資料的觀點，以相較客觀的方式決定模式的架構與輸入，建立地下水模型，以臺北盆地為研究區。
本研究以數種資料分析的方法建立MODFLOW模式，應用的面向分別為模式架構建立、水利傳導係數推估以及補注區判釋。模式架構方面以類別型貝式最大熵法推估岩性資料至模式各網格，並以kernel density estimation決定阻水層與含水層之分界，結果將台北盆地分為五層，由上至下分別為含水層、阻水層、含水層、阻水層、含水層；水利傳導係數推估方面應用抽水試驗水利傳導係數資料做為確定性資料並考量上述所推估之岩性資料作為不確性資料以連續型貝式最大熵法推估MODFLOW各層各網格之水平與垂直水利傳導係數。補注區則利用經驗證交函數分析地下水位資料與降雨資料，配合表層岩性進行判釋，結果顯示台北盆地補注區位在大漢溪與新店溪上游接近盆地邊緣的扇頂區，分別是柑園、省民一帶與清溪、新店區一帶。
模式率定以Python裡Soptpy函式庫進行，分穩態率定與暫態率定，RMSE分別為0.44與0.97，皆位於一般可接受標準1之內。

關鍵字

MODFLOW ； 貝式最大熵法 ； 經驗正交函數 ； 地下水模式分層 ； 補注區判釋 ； 水文地質參數推估

並列摘要

To understand groundwater system further, groundwater modelling have always been a common way. However, in the past, many studies have used subjective method to decide model input such as the number of groundwater layers, the boundary of aquifer and aquitard, the hydrogeological parameters and the boundary conditions. As a result, the models designed by each researcher are different in the same research area, so the results of the simulation are different as well. Therefore, this research attempts to determine the structure and input of the model in a relatively objective way from data perspective, and build a groundwater model. Taking the Taipei Basin as the study area.
In this study, several data analysis methods are used to support MODFLOW modelling. These methods are used to decide model structure, interpolate hydraulic conductivity in each MODFLOW grid and identify the recharge area. In terms of model structure, the lithology data is estimated by the categorical Bayesian maximum entropy method into each MODFLOW grid, then determined the boundary of the aquifer and aquitard by the kernel density estimation. As a result, the Taipei Basin is divided into five layers, from top to bottom is aquifer, aquitard, aquifer, aquitard, aquifer; In hydraulic conductivity estimation part, use pumping test hydraulic conductivity data as hard data and consider the above estimated lithology data as soft data. Using continuous Bayesian maximum entropy method estimates the horizontal and vertical hydraulic conductivity into each grid in each layer of MODFLOW. As for the identification of recharge area, the Empirical orthogonal function was used to analyze groundwater level data and rainfall data, and the surface lithology was took into consideration as well. The results showed that the recharge area of the Taipei Basin was located in the alluvial fan-top area near the edge of the basin at the upstream of Dahan River and Xindian River where is around GanYuan, XingMin and QingXi, XinDian respectively.
The mode calibration is carried out by the Soptpy package in Python. It is divided into steady-state calibration and transient calibration. The RMSE is 0.44 and 0.97 respectively, which are within the general acceptable standard 1.

並列關鍵字

MODFLOW ； Bayesian maximum entropy ； Empirical orthogonal function ； Groundwater model stratification ； Recharge area identification ； Hydrogeological parameter estimation

參考文獻

Berger, A. L., Pietra, V. J. D., Pietra, S. A. D. (1996). A maximum entropy approach to natural language processing. Computational linguistics, 22(1), 39-71.

Bogaert, P. (2002). Spatial prediction of categorical variables: the Bayesian maximum entropy approach. Stochastic environmental research and risk assessment, 16(6), 425-448.

Bogaert, P., D'Or, D. (2002). Estimating soil properties from thematic soil maps. Soil Science Society of America Journal, 66(5), 1492-1500.

Boughton, W. (1989). A review of the USDA SCS curve number method. Soil Research, 27(3), 511-523.

Chen, S. X. (2000). Probability density function estimation using gamma kernels. Annals of the Institute of Statistical Mathematics, 52(3), 471-480.