Attribution analysis and threshold identification of land subsidence in the Beijing Plain based on MT-InSAR technology and explainable artificial intelligence

Excessive groundwater extraction is the primary driver of land subsidence (LS) across the Beijing Plain. The South-to-North Water Diversion Project (SWDP) has transformed Beijing's water supply system, presenting new characteristics in groundwater level (GWL) fluctuations and LS evolution. This study used Multi-temporal Interferometric Synthetic Aperture Radar (MT-InSAR) technology, combined with Envisat-ASAR (ASAR), Radarsat-2 (R2), and Sentinel-1 (S1) data, to investigate the evolution of LS in the Beijing Plain before and after the SWDP. The Random Forest (RF) and SHapley Additive exPlanations (SHAP) explainable artificial intelligence methods were employed to reveal the response characteristics of different aquifer GWLs to LS and identify their stage-dependent response thresholds. The research results indicate that before the SWDP, the second confined aquifer was the dominant explanatory factor driving LS. After the project implementation, the third confined aquifer GWL became the dominant explanatory factor in the attribution analysis. The dominant aquifer exhibited identifiable stage-dependent threshold responses to LS during different hydrological periods. This study provides a scientific basis for the control of GWLs and LS regulation based on aquifer management.