The seismic effect of well water level is complex and variable, and even if both wells are located in an area with similar tectonic and hydrogeological conditions, they exhibit slightly varying response characteristics to the same earthquake. Wells BB and RC, located about 100 km apart in the southwest of the Huayingshan fault zone in the Sichuan and Chongqing regions, exhibited obvious similarities and differences in their co-seismically response and sustained recovery characteristics during the Wenchuan Ms8.0 earthquake. Based on the dislocation theory and fluid–solid coupling theory, this study developed the seismic stress–strain model and the response model of pore pressure to seismic stress using Coulomb 3.3 and COMSOL 6.3, respectively. Simulation findings indicate that both BB and RC are located in the expansion zone, where their water levels show a co-seismic step-down. The amplitudes of BB and RC water levels are 83 cm and 81 cm, which are approximately 10 cm smaller than the actual values. The recovery times are 60 d for BB and 3 h for RC, closely resembling the actual values. Furthermore, the numerical results from different scenarios show that the recovery time of pore pressure is reduced by several times when the permeability of the confining layer overlying the observed aquifer increases by one order of magnitude or the thickness decreases, and this change is more sensitive to the permeability. It is clear that the confining condition has an important impact in the response time of sustained changes in well water levels, which may also help to explain the variations in the characteristics of sustained changes in wells BB and RC.

The earthquake-induced water level responses in the fault zone may be distinctly different, even when the underground wells are very close. How to qualitatively and quantitatively analyze the differences and controlling factors of the groundwater response to earthquakes in the fracture zone is a hot topic in seismic hydrogeology. This study utilizes three adjacent groundwater monitoring wells, located across distinct structural domains of the Dachuan-Shuangshi Fault, to systematically investigate the different sensitivities of earthquake-induced water level responses and their main influencing factors. The statistical results reveal that monitoring wells located on opposing fault blocks demonstrate higher co-seismic sensitivity compared to the well situated within the fault fracture zone. The water level co-seismic responses are governed by multiple controlling factors, rather than being dominated by individual parameters. Therefore, we employed random forest to quantitatively assess the importance of influencing factors related to hydraulic parameters, aquifer confinement, fault architecture, tidal characteristics, and barometric efficiency. The results showed that hydraulic properties and aquifer confinement are the primary factors influencing the differential sensitivity of water level co-seismic responses. In contrast, the influence of barometric efficiency on water level co-seismic responses is relatively minor. These findings provide critical insights into the understanding of the mechanism and characteristics of seismic hydrological responses in fault zones and provide support for optimizing the placement of groundwater monitoring in seismotectonic environments.

Deep groundwater dynamics, influenced by multiple factors, exhibit multi-scale periodicity, nonlinearity, and non-stationarity, posing significant challenges for reliable groundwater level prediction and anomaly detection. To address this issue, this study develops a hybrid deep learning model for simulating and predicting normal groundwater dynamics, which integrates multi-source observations—including barometric pressure, solid tide, and historical groundwater levels—with convolutional neural networks (CNNs), Long Short-Term Memory (LSTM), and Attention mechanisms. An Exponentially Weighted Moving Average (EWMA) control chart is then applied to the prediction residuals for statistical anomaly detection. The results show that CNN kernels capture multi-period local features induced by barometric and tidal effects; LSTM gating mechanisms learn long-term temporal dependencies; and the Attention mechanism dynamically assigns weights to both temporal steps and input features. Compared with single-source inputs or standalone LSTM/CNN models, the hybrid model enhances prediction accuracy (MAE < 0.02, RMSE < 0.03, R² > 0.99), while integration with the EWMA control chart helps to identify the possible anomalies. Collectively, these results provide a reliable means for predicting deep groundwater levels and detecting hydrological anomalies under complex multi-scale periodic forcings, holding significant practical value in groundwater resource management and disaster warning.