Integrating multi-source data to assess land subsidence sensitivity and management policies

Uneven land subsidence will cause damage to urban buildings and infrastructure and pose risks to human production and life. This study proposes a new methodology for factor identification and prediction of urban land subsidence that integrates physics-driven models and data-driven models. The driving factor layer was generated using multi-source datasets, and a comprehensive factor identification system was implemented. The machine learning method was used to calculate the comprehensive weight of each driving factor, and the identification and evaluation model of the land subsidence factor in Heze City was established. Then, the SARIMA model and the MODFLOW model were used to predict the development trend of the main influencing factors of land subsidence. Finally, the development trend of land subsidence was evaluated based on the physics-based machine learning model. The results show that the maximum average annual subsidence rate from 2016 to 2020 deduced by InSAR is 130.22 mm/yr, and the subsidence area accounts for 85.09%. It is estimated that the maximum annual average rate of land subsidence will be 65.14 mm/yr by 2025, and the subsidence area will account for 92.56%. The development degree of land subsidence has a significant relationship with groundwater exploitation and coal mining planning. By 2025, the proportion of areas with unchanged sensitivity levels will be 61.39%, and the prevention and control plan can be effectively implemented in most areas. The research provides a new methodology for evaluating the effectiveness of government planning policies and promoting scientific prevention and control.