A hybrid coupling model of groundwater level simulation considering hydrogeological parameter: a case study of Nantong City in Eastern China

Abstract Groundwater level dynamic monitoring data have the characteristics of spatio-temporal non-smoothness and strong spatio-temporal correlation. However, the current groundwater level prediction model is insufficient to consider the spatio-temporal factors of the groundwater level and the autocorrelation of spatio-temporal series, particularly the lack of consideration of hydrogeological conditions in the actual study area. Thus, this study constructed a model based on the hydrogeological conditions and the spatio-temporal characteristics of the dynamic monitoring data of groundwater in the porous confined aquifer III in Nantong, the northern wing of the Yangtze River Delta, China. The spatial autocorrelation coefficient of the hydrogeology important parameter, permeability coefficient K, is used to optimize the distance weighting coefficient of monitoring wells obtained by the K-nearest neighbor (KNN) algorithm and then reconstruct the spatio-temporal dataset and long short-term memory (LSTM) network. A spatio-temporal groundwater level prediction model LSTM-K-KNN that introduces the spatial autocorrelation of hydrogeological parameters was constructed. The reliability and accuracy of LSTM-K-KNN, LSTM, autoregressive integrated moving average (ARIMA) model, and support vector machine (SVM) were evaluated by a cross-validation algorithm. Results showed that the prediction accuracy of LSTM-K-KNN is 19.86, 43.64, and 52.38% higher than that of the other single prediction models (LSTM, ARIMA, and SVM).