Scale-associated differences in building-oriented land subsidence susceptibility under a nested spatial design using SBAS-InSAR and interpretable machine learning

The exponential growth of urban development has brought about several environmental problems, land subsidence being one of them. Land subsidence within building-containing urban areas is an important concern for urban safety management. Nevertheless, the nonlinear associations between multiple impact factors and building-oriented land subsidence susceptibility (BOLSS) across different spatial scales, together with the corresponding planning and monitoring implications, remain insufficiently understood. In this study, the small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) technique, a random forest (RF) model, and geographic information system (GIS) tools were used to map the susceptibility of building-containing spatial units to land subsidence, hereafter referred to as BOLSS. SHapley Additive exPlanations (SHAP), partial dependence plots, and variable interaction plots were then used to analyze the associations of 15 selected impact factors with BOLSS at the built-up area (BUA) and southwest-BUA scales in Chongqing. The findings show that the factors associated with BOLSS differ between the BUA and southwest-BUA scales. Specifically, built-environment factors make the largest contributions at the BUA scale, whereas natural-environment factors and changes in built-environment conditions make larger contributions at the southwest-BUA scale. In addition, most factors show nonlinear relationships with BOLSS, with scale-dependent differences. This BOLSS mapping provides a preliminary visualization tool for identifying building-containing areas with relatively high subsidence susceptibility. The scale-dependent analysis offers useful information for prioritizing urban monitoring and planning measures.