Abstract Coal resources serve as the material foundation for societal development. However, prolonged mining operations induce significant damage to coal mining areas, with accompanying geo-environmental issues becoming increasingly severe, thereby highlighting the critical importance of surface deformation monitoring technologies. Conventional subsidence prediction models exhibit strong dependencies on data quality and model assumptions, while demonstrating limited capability in handling complex spatiotemporal features, ultimately resulting in compromised prediction accuracy. This study employed the Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) technique to monitor surface subsidence in two adjacent working faces of a coal mine in Anhui Province from November 2023 to December 2024. We subsequently developed a hybrid deep learning model integrating Long Short-Term Memory (LSTM) networks and Transformer architecture. The results reveal that: (1) Severe incoherence caused by large-gradient subsidence primarily occurs at the edges of subsidence funnels in goaf areas. The maximum subsidence rate reached −82.4 mm/y, with a cumulative subsidence of −115.6 mm observed in the study area. Comparative experiments demonstrated the superior performance of our model, with the mean values of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) being 1.9615 mm and 1.5936 mm, respectively. This study proposes a novel methodology for mining-induced subsidence monitoring and prediction, which can serve as both (1) a technical reference for post-mining subsidence assessment, and (2) a decision-making basis for targeted remediation in critical deformation zones.
Cui et al. (2026) studied this question.