Decomposition and attribution analysis of the coupled evolution characteristics of groundwater and land subsidence in the Beijing-Tianjin-Hebei Plain

This study investigated the relationship between groundwater storage and land subsidence and analyzed the potential causes of these variations in the Beijing-Tianjin-Hebei Plain from 2018 to 2022. Independent Component Analysis (ICA) was applied to separate signals from Interferometric Synthetic Aperture Radar (InSAR) surface deformation data and Gravity Recovery and Climate Experiment (GRACE) and GRACE-Follow On (GRACE-FO) groundwater storage anomalies. A spatiotemporal feature coupling analysis was then conducted on the periodic and trend components. The CNN-LSTM-attention neural network further quantified the drivers of groundwater changes in the trending subsidence area. (1) periodic components showed a strong positive correlation (cross-correlation coefficient = 0.73, with land subsidence lagging behind groundwater storage changes by 1 month) and a 5–6 month delayed response to precipitation, with high-score areas clustered in the precipitation-rich eastern and southern BTHP. (2) The trend components revealed synchronized declines (2018–2021), followed by rebounds, with groundwater recovery outpacing subsidence mitigation. High-score zones were aligned with regions of intense groundwater extraction (e.g., the southern BTHP). (3) In the trending subsidence area, CNN-LSTM-attention model achieved higher accuracy (test set R²: 0.54 vs −0.99, RMSE: 13.11 mm vs 43.12 mm) using reconstructed groundwater signals, confirming precipitation and anthropogenic extraction as significant contributing factors. This study provides a reference for exploring the relationship between changes in groundwater reserves and land subsidence. • InSAR and GRACE data integration reveals groundwater-subsidence correlations. • Independent Component Analysis decomposes subsidence and groundwater dynamics. • Spatiotemporal coupling analysis on periodic and trending independent components. • ICA united CNN-LSTM-attention performs GWSA attribution in trending subsidence areas.