Modelling groundwater level variation at the Zhuoshui River basin by artificial neural network techniques
Date Issued
2012
Date
2012
Author(s)
Lin, Cheng-Hsien
Abstract
In the past decade water demand has increased drastically due to the rapid development in economy and industry in Taiwan. Groundwater has become an important water source during drought periods and/or at the areas short of water storage facilities owing to its advantages such as low-cost and easy-to-access. So far, few studies have discussed about the estimation of the groundwater level variations at the mountainous areas of the Central Taiwan. Therefore, it will be beneficial to develop a reliable model for precisely estimating groundwater level variations at mountainous areas. The mountainous area at the upstream of the Zhuoshui River basin is used as a case study. First, the hydro-system and hydro-environment of the mountainous area are investigated. Second, the correlation among rainfall, streamflow and groundwater level variation is analyzed. The time lag between groundwater level variation and rainfall is one day while the time lag between groundwater level variation and steamflow is within one day. Third, the geological structure of the groundwater monitoring wells is used to explore the relationship of groundwater level variations between shallow groundwater wells and deep groundwater wells. The results show that the groundwater level variations of shallow groundwater wells have low relationship with those of deep ones in the clay layer, while the groundwater level variations of shallow and deep groundwater wells have similar trends.
This research also investigates the effective impacts of rainfall amount on groundwater level variations and uses the Thiessen polygon method to calculate the average rainfall over the basin area based on different thresholds for threshold screening purpose. The correlations among groundwater level variations and rainfall filtered by different thresholds are analyzed and classified into four types: slow-ascending type; ladder-type; slow-descending type; and random type. In addition, the depth and geological structure of groundwater wells are used to find out the causes of those four types.
Because the impacts on the variations of groundwater level are nonlinear, we uses both the backpropagation neural network (BPNN) due to its superior nonlinear mapping ability and high model accuracy and the adaptive network fuzzy inference system (ANFIS) with a fuzzy rule base to construct estimation models for groundwater level variations. We conduct a comparison study among different model input combinations: rainfall only; streamflow only; and rainfall and streamflow, and the results show that all the BPNN and ANFIS models perform well. Besides, the groundwater level variations of groundwater wells near the Zhuoshui River are much influenced by the lateral recharge from the river and the estimation models with streamflow as the only input perform better. While the groundwater level variations of groundwater wells far from the Zhuoshui River are influenced less by the lateral recharge from the river and the estimation models with rainfall as the only input perform better. The estimation models with rainfall and streamflow as inputs perform the best. Understanding the interactive recharge mechanisms between mountainous water resources and groundwater can facilitate future discussion on mountainous water resource conservation strategy for alleviating land subsidence in downstream areas.
Subjects
Rainfall
Streamflow
Groundwater Level
Artificial Neural Networks
Thiessen polygons method
SDGs
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-101-R99622031-1.pdf
Size
23.54 KB
Format
Adobe PDF
Checksum
(MD5):305a8dff7c90b118929f02dba3234283