Predicting Groundwater Level Based on Machine Learning: A Case Study of the Hebei Plain

Abstract

In recent years, the groundwater level (GWL) and its dynamic changes in the Hebei Plain have gained increasing interest. The GWL serves as a crucial indicator of the health of groundwater resources, and accurately predicting the GWL is vital to prevent its overexploitation and the loss of water quality and land subsidence. Here, we utilized data-driven models, such as the support vector machine, long-short term memory, multi-layer perceptron, and gated recurrent unit models, to predict GWL. Additionally, data from six GWL monitoring stations from 2018 to 2020, covering dynamical fluctuations, increases, and decreases in GWL, were used. Further, the first 70% and remaining 30% of the time-series data were used to train and test the model, respectively. Each model was quantitatively evaluated using the root mean square error (RMSE), coefficient of determination (R2), and Nash–Sutcliffe efficiency (NSE), and they were qualitatively evaluated using time-series line plots, scatter plots, and Taylor diagrams. A comparison of the models revealed that the RMSE, R2, and NSE of the GRU model in the training and testing periods were better than those of the other models at most groundwater monitoring stations. In conclusion, the GRU model performed best and could support dynamic predictions of GWL in the Hebei Plain.