Data-Driven Techniques for Regional Groundwater Level Forecasts

Abstract:

Data-Driven Techniques for Regional Groundwater Level Forecasts

 ^{Fi-John Changa, Li-Chiu Changb, Fong He Tsaia, Hung-Yu Shenb}

^a Department of Bioenvironmental Systems Engineering, National Taiwan University, Taipei 10617, Taiwan, ROC.

^b Department of Water Resources and Environmental Engineering, Tamkang University, New Taipei City 25137, Taiwan, ROC.^.

Correspondence to: Fi-John Chang (email: changfj@ntu.edu.tw)

The alluvial fan of the Zhuoshui River in Taiwan is a good natural recharge area of groundwater. However, the over extraction of groundwater occurs in the coastland results in serious land subsidence. Groundwater systems are heterogeneous with diverse temporal-spatial patterns, and it is very difficult to quantify their complex processes. Data-driven methods can effectively capture the spatial-temporal characteristics of input-output patterns at different scales for accurately imitating dynamic complex systems with less computational requirements. In this study, we implement various data-driven methods to suitably predict the regional groundwater level variations for making countermeasures in response to the land subsidence issue in the study area. We first establish the relationship between regional rainfall, streamflow as well as groundwater levels and then construct intelligent groundwater level prediction models for the basin based on the long-term (2000-2013) regional monthly data sets collected from the Zhuoshui River basin. We analyze the interaction between hydrological factors and groundwater level variations; apply the self-organizing map (SOM) to obtain the clustering results of the spatial-temporal groundwater level variations; and then apply the recurrent configuration of nonlinear autoregressive with exogenous inputs (R-NARX) to predicting the monthly groundwater levels. As a consequence, a regional intelligent groundwater level prediction model can be constructed based on the adaptive results of the SOM. Results demonstrate that the development of the regional intelligent groundwater level prediction model produces high accuracy and stability, which is beneficial to authorities for sustainable water resources management in a basin scale. Keywords: Artificial neural network (ANN); Groundwater; Regional; Forecasting.