Prediction of groundwater level under the influence of groundwater exploitation using a data-driven method with the combination of time series analysis and long short-term memory: a case study of a coastal aquifer in Rizhao City, Northern China

Abstract

The excessive exploitation of groundwater not only destroys the dynamic balance between coastal aquifer and seawater but also causes a series of geological and environmental problems. Groundwater level prediction provides an efficient way to solve these intractable ecological problems. Although several hydrological numerical models have been employed to conduct prediction, no study has accurately predicted the groundwater level change under the consideration of groundwater exploitation, especially in coastal aquifers. This is due to the characteristics of spatially and temporally complex hydrological processes. This study proposes a novel data-driven method based on the combination of time series analysis and a machine learning method for accurately predicting the variation of groundwater level in a coastal aquifer under the influence of groundwater exploitation. The partial autocorrelation function and continuous wavelet coherence were used to analyze the monitoring data of groundwater level at three wells, which indicated that the historical monitored data and the dataset of precipitation could be considered as the input variables to construct the hydrological model. Then, three models based on the different inputs were constructed, namely, the LSTM, PACF-LSTM, and PACF-WC-LSTM models. The performances of the three models were compared by the calculation of four error metrics. The results showed that the performance of the PACF-LSTM and PACF-WC-LSTM models was better than that of the LSTM model and that the PACF-WC-LSTM model achieved the best prediction performance. Accurately predicting the variation of groundwater level provides the basis for managing groundwater resources and preserving the ecological environment.