Short-term groundwater level prediction based on spatio-temporal attention mechanism LSTM model: A case study of the Shougang Park area on the eastern bank of the Yongding River
Driven by ecological water replenishment, precipitation, and reduced groundwater extraction, the groundwater level in the Beijing Plain has been continuously rising. While this trend has improved the aquatic ecological environment and promoted restoration, it also poses risks and challenges to urban construction and the safety of subsurface infrastructure. Therefore, accurate prediction of groundwater levels is crucial for managing groundwater resources and ensuring the safety of surface and subsurface structures in Beijing. Specifically, using the area surrounding Shougang Park on the eastern bank of the Yongding River as a case study, this research enhances the traditional LSTM model by incorporating both spatial and temporal characteristics of groundwater dynamics. A spatio-temporal attention mechanism-based LSTM (STA-LSTM) model was developed for short-term groundwater level forecasting, achieving accurate dynamic predictions of regional groundwater levels. The results show that: (1) The STA-LSTM model demonstrated superior prediction performance across both temporal and spatial scales compared to the LSTM, SA-LSTM, and TA-LSTM models, achieving optimal metrics: MAE of 0.08 m, RMSE of 0.11 m, and NSE of 0.98; (2) The model effectively captured groundwater level dynamics during intense precipitation and ecological water replenishment events, achieving simulation accuracy greater than 99% and 97%, respectively; (3) Prediction accuracy decreased over time due to cumulative errors, with MAE values of 0.10 m, 0.18 m, and 0.40 m for the 7th, 14th, and 28th days, respectively; (4) Predictions of groundwater dynamics under five designed precipitation scenarios and two ecological water replenishment scenarios revealed a strong correlation between groundwater level fluctuations and the intensity and duration of these events. This study demonstrates that the STA-LSTM model effectively addresses the complexity and uncertainty inherent in groundwater dynamics and provides reliable technical support for the short-term forecasting of urban shallow groundwater levels that are significantly influenced by precipitation and ecological replenishment.
ZHOUC, TANGQ, ZHAOY, et al. Reduction of subsidence and large-scale rebound in the Beijing Plain after anthropogenic water transfer and ecological recharge of groundwater: evidence from long time-series satellites InSAR[J]. Remote Sensing, 2024, 16(9): 1528.
GATTINONIP, SCESIL. The groundwater rise in the urban area of Milan (Italy) and its interactions with underground structures and infrastructures[J]. Tunnelling and Underground Space Technology, 2017, 62: 103-114.
[19]
TARIQ AMIN CHAUDHARYM. Implications of rising groundwater level on structural integrity of underground structures-investigations and retrofit of a large building complex[J]. Structural Survey, 2012, 30(2): 111-129.
[20]
LUD, LIX, DUX, et al. Numerical simulation and analysis on the mechanical responses of the urban existing subway tunnel during the rising groundwater[J]. Tunnelling and Underground Space Technology, 2020, 98: 103297.
MEERSCHAERTM M, TADJERANC. Finite difference approximations for fractional advection-dispersion flow equations[J]. Journal of Computational and Applied Mathematics, 2004, 172(1): 65-77.
[24]
DISCACCIATIM, MIGLIOE, QUARTERONIA. Mathematical and numerical models for coupling surface and groundwater flows[J]. Applied Numerical Mathematics, 2002, 43(1): 57-74.
[25]
SUTRADHARA, PAULINOG H, GRAYL J. Transient heat conduction in homogeneous and non-homogeneous materials by the Laplace transform Galerkin boundary element method[J]. Engineering Analysis with Boundary Elements, 2002, 26(2): 119-132.
OMARP J, GAURS, DWIVEDIS B, et al. Groundwater modelling using an analytic element method and finite difference method: an insight into Lower Ganga river basin[J]. Journal of Earth System Science, 2019, 128(7): 195.
PANM, ZHOUH, CAOJ, et al. Water level prediction model based on GRU and CNN[J]. IEEE Access, 2020, 8: 60090-60100.
[33]
TAOH, HAMEEDM M, MARHOONH A, et al. Groundwater level prediction using machine learning models: a comprehensive review[J]. Neurocomputing, 2022, 489: 271-308.
[34]
WUNSCHA, LIESCHT, BRODAS. Groundwater level forecasting with artificial neural networks: a comparison of long short-term memory (LSTM), convolutional neural networks (CNNs), and non-linear autoregressive networks with exogenous input (NARX)[J]. Hydrology and Earth System Sciences, 2021, 25(3): 1671-1687.
[35]
VUM T, JARDANIA, MASSEIN, et al. Reconstruction of missing groundwater level data by using long short-term memory (LSTM) deep neural network[J]. Journal of Hydrology, 2021, 597: 125776.
[36]
ZHANGJ, ZHUY, ZHANGX, et al. Developing a long short-term memory (LSTM) based model for predicting water table depth in agricultural areas[J]. Journal of Hydrology, 2018, 561: 918-929.
[37]
SHINM J, MOONS H, KANGK G, et al. Analysis of groundwater level variations caused by the changes in groundwater withdrawals using long short-term memory network[J]. Hydrology, 2020, 7(3): 64.
[38]
KRATZERTF, KLOTZD, BRENNERC, et al. Rainfall-runoff modelling using long short-term memory (LSTM) networks[J]. Hydrology and Earth System Sciences, 2018, 22(11): 6005-6022.
SUNK, HUL, GUOJ, et al. Enhancing the understanding of hydrological responses induced by ecological water replenishment using improved machine learning models: a case study in Yongding River[J]. Science of the Total Environment, 2021, 768: 145489.
[45]
GRAVESA. Long short-term memory[J]. Supervised Sequence Labelling with Recurrent Neural Networks, 2012: 37-45.