Mapping Susceptibility and Risk of Land Subsidence by Integrating InSAR and Hybrid Machine Learning Models: A Case Study in Xi'an, China

Land subsidence is a widespread geo-hazard, and it can be effectively monitored with the Interferometric Synthetic Aperture Radar (InSAR) technique. Assessing land subsidence plays a significant role in ensuring safety and enhancing disaster prevention. It requires not only focusing on the extent or rate of deformation but also evaluating the susceptibility and risk of subsidence. In this article, we propose a new comprehensive subsidence susceptibility and risk assessment strategy by integrating InSAR observation and hybrid machine learning models, which is first and successfully employed over Xi'an area. In this study, four machine learning models are compared to determine the optimal model, and found that the Random Forest (RF) performs the best in predicting InSAR-derived spatial deformation (Root Mean Square Error = 3.53 mm) and susceptibility (Area Under the Curve = 0.97). Also, the authenticity and reliability of the susceptibility results from RF model are improved by further Isotonic regression calibration processes. Then, subsidence risk map is obtained from the hazard and vulnerability assessment using the Analytic Hierarchy Process by combining multiple conditional factors. Remarkably, the results revealed that regions with the very high and high risk level of subsidence are 12.33 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{km}^{2}$</tex-math></inline-formula> and correspond to 1.34<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> of the total. It further found that the groundwater level and its changes are the domain factors in Xi'an from machine learning method. This work provides an integrated assessment approach for subsidence from a new perspective, and the findings can serve as theoretical support for early warning and disaster prevention.