A Hybrid Coupled Model for Groundwater-Level Simulation and Prediction: A Case Study of Yancheng City in Eastern China

The over-exploitation of groundwater has led to a significant drop in levels, which may lead series geological disasters and ecological environmental problems such as ground subsidence cracks. Therefore, through studying the dynamic change characteristics groundwater, we can grasp changes level over time invert hydrogeological parameters, provides an important basis for management resources. In this study, confined aquifer III between 2005 2014 Yancheng City was selected research object, Back Propagation (BP) neural network, Spatial-temporal Auto Regressive Moving Average (STARMA) model, BP-STARMA model were used predict spatial temporal evolution trends groundwater. order compare prediction effectiveness fitting accuracies three models measured from perspectives space. results Relative Squared Error (RSE), Normal Mean (NMSE), Root-Mean-Squared (RMSE), Absolute (MAE) assess robustness model. showed that RMSE reduced by 39.92%, 38.35%, 30.25%, 31.55%, 13.57% compared with STARMA 22.2%, 8.7%, 15.9%, 28.5%, 4.42% BP network respectively. Collectively, shows better spatiotemporal than models, is more applicable spatially continuous time-discrete sequences, sequences respond natural geographic phenomena.