Enhanced Data Utilization Approach to Improve the Prediction Performance of Groundwater Level Using Semianalytical and Data Process Models

Abstract

Methods for the estimation of groundwater level (GWL) are often based on knowledge- and data-based approaches, which are largely affected by the representativeness of input parameters and the quantity and quality of previously collected data sets, respectively. In this study, a new hybrid GWL prediction model combining knowledge- and data-based approaches is proposed using a signal decomposing technique to improve the efficiency and accuracy of GWL prediction with less data dependency. The target site condition was urban areas near a river, where the river stage is the dominant influence on GWL. For this purpose, the engineering groundwater-prediction model (EGPM) as a knowledge-based method and multiple linear regression, artificial neural network (ANN), and wavelet ANN (WANN) as data-based methods were employed and adopted to establish the proposed hybrid GWL prediction model. Case studies in the Korean and Japanese contexts were performed to compare and assess results predicted by existing methods and the proposed hybrid method. It was shown that the proposed hybrid method can enhance the GWL prediction performance with improved accuracy of the prediction and efficiency of database utilization. It was also indicated that the required data length can be alleviated while producing a satisfactory prediction throughout the frequency range of GWL fluctuations.