Prediction of the groundwater level is needed for management of groundwater resources or monitoring in specific site such as pollution area. Neural network of machine learning technique is one of the fancy and powerful tool to estimate or predict the groundwater level in hydrology. There are various kinds of architecture in neural networks such as Artificial Neural Network (ANN), Deep Neural Network (DNN), Long Short Term Memory (LSTM), and Stacked Long Short Term Memory (S-LSTM). These models were applied to predict the groundwater levels in riverside area where much of groundwater is consumed. Model performances from each neural network model were obtained to be compared with each other. While ANN model which is the most basic neural network had a range of RMSE errors from 0.0331 m to 0.0562 m, LSTM showed the best model performances among the four neural network models. Additionally, deeper networks like DNN and S-LSTM did not always show better performances than simple networks, which can imply that the most complex model is not necessarily the best network. Because various neural network models give their efficiency differently according to type of data or system, determination of proper network would be important when machine learning technique is applied for prediction of the groundwater level.
Acknowledgement: This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIP) (No. 2017R1A2B3002119)