Application of Artificial Neural Network to Prediction of Groundwater Level and to Evaluation of Influence Factors in Yangpyeong Riverside Area

Abstract

Sustainable use of groundwater resource is crucial issues in recent years due to increasing use of groundwater. In the Yangpyeong riverside area, where anthropogenic activities such as operation of the groundwater heat pump system (GWHPs) and the water curtain cultivation (WCC) are concentrated, prediction of changes in the groundwater levels is necessary to suggest management plans for protecting the groundwater resource in quantitative aspects. As a means of the groundwater level forecasting, the ANN model was applied in this study. It was revealed that the surface water level, the WCC and the GWHPs operation showed higher correlation with the groundwater level fluctuation in the study area rather than the precipitation. Based on the abovementioned influence factors, network architecture was constructed in train and test period, and prediction of the groundwater level at 8 wells was performed. The forecasted groundwater levels showed good matches with the observed data with low RMSE values in range of 0.03~0.06 m. Additionally, the contribution and the relative importance of each influence factor were computed to decompose the combined effects. Weights method and PaD method which are pre-existing method to quantify the input variables in the ANN model were applied at first to calculate the contribution or the relative importance. An extraction method which can compare spatial and temporal contributions of each influence factor was developed in this study, and verified the suitability on the spatial and temporal comparison. As a results, the surface water level accounted for dominant effect on the groundwater level fluctuation (64.09~83.30 %), the WCC showed low effect (16.04~26.76 %), and the GWHPs had a little or no effect (0.29~6.26 %). Especially, effect of the WCC was changed drastically depending on times, showing greater contribution than that of the surface water level in a period of the WCC operation in the winter season. Evaluation of influence factors on the groundwater level by the ANN will help understanding driving forces to change the groundwater level especially for an aquifer system which have complex and nonlinear features.