Quantifying Multifactorial Effects on Land Subsidence Using Interpretable Machine Learning: a Case Study in Cangzhou, China

Understanding how multiple factors affect land subsidence helps to take a scientific approach to preventing and controlling land subsidence. Previous studies have mainly focused on the monitoring and prediction of land subsidence, with less research on the causes of land subsidence. This study proposes an analytical framework to analyze the correlation between land subsidence anomalies and different influencing factors. First, the spatial local outlier measure (SLOM) algorithm is used to calculate the land subsidence anomalies, then the relationship between the land subsidence anomalies and the influencing factors is modeled using the Random Forest algorithm, and finally the contribution of multiple factors to land subsidence anomalies is analysed using the SHapley Additive exPlanation (SHAP) method. The research dataset includes land subsidence monitoring and related socio-economic factors from 2017 to 2019 in Cangzhou City, Hebei Province, and the effects of remote sensing of nighttime lighting, precipitation, DEM, slope, and aspect on land subsidence are determined. The results show that the anomalies of land subsidence in Cangzhou City account for about 10% of all detected points, which are mainly distributed in the west and south of Cangzhou. The analysis identifies human activities and precipitation as the primary drivers, with multi-year average SHAP value contributions of 22.82% and 23.69%, respectively.