Machine learning approaches for predicting land subsidence in Ca Mau: XGBoost, Random Forest, and MAF

Abstract

Land subsidence is a natural hazard that is causing serious damage to the Mekong Delta (VMD) of Vietnam, with subsidence rates of up to 5 cm per year in densely populated cities such as Ca Mau, increasing the risk of salinity intrusion and tidal flooding. These ground movements not only amplify the impacts of sea level rise but also threaten infrastructure, agricultural sustainability, and long - term climate resilience. While traditional monitoring methods such as GNSS and land leveling surveys are highly accurate, they are often spatially inadequate and cost - in effective for regional - scale applications. In this context, remote sensing technologies, specifically Interferometric Synthetic Aperture Radar (InSAR) techniques such as Persistent Scatter Interferometry (PSI), have emerged as powerful tools for understanding surface deformation patterns over large areas. Integrating InSAR - derived observations with machine learning (ML) techniques offers new opportunities for predictive modeling of subsidence phenomena. Ensemble algorithms such as Random Forest (RF) and Extreme Gradient Boosting (XGBoost) have demonstrated robust performance in identifying spatially distributed land deformation susceptibility, especially when applied to high - dimensional geospatial data. In this study, we evaluate the predictive performance of three distinct methods - MAF (Moving Average Filter), RF, and XGBoost - for predicting land subsidence in Ca Mau using PSI - based displacement data. A dataset of 5,000 deformation monitoring points from Sentinel - 1 imagery from 2014 to 2019 is used to train and evaluate the models. Among these models, XGBoost demonstrated the best performance with the lowest RMSE (4.67) and MAE (3.23), and the highest R² (0.9869), significantly outperforming both RF and MAF. These findings highlight the robustness of machine learning approaches, particularly XGBoost, in predicting land subsidence and supporting sustainable land use planning and climate adaptation strategies in vulnerable deltaic environments.