Mine subsidence monitoring and prediction integrating SBAS-InSAR technology and BO-Prophet model

Coal mining causes significant environmental disruptions due to excessive resource extraction. Real-time monitoring and prediction of mining-induced surface deformation are critical for ensuring mining safety. Conventional monitoring methods struggle to achieve large-scale real-time observation and have limitations in terrain adaptability and weather resistance. Conventional prediction methods are constrained as follows: numerical simulation is limited by the complexity of physical parameters; mathematical statistics requires substantial data and struggles to comprehensively reflect geotechnical properties; hybrid algorithms involve complex modeling and rely on individual models. This study develops an integrated framework combining Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) and Bayesian optimization-Prophet (BO-Prophet) models for mining subsidence monitoring and forecasting. Using the Yinying Coal Mine as the study area, multi-source radar datasets are processed through SBAS-InSAR. This technique identifies four major subsidence areas and generates time-series deformation data. Cross-validation confirms the reliability of SBAS-InSAR monitoring results: the Pearson correlation coefficient between ascending-orbit (Sentinel-1A) and descending-orbit (Radarsat-2) results of 3,000 high-coherence points in the subsidence area reaches 0.946; compared with four Global Navigation Satellite System (GNSS) stations, the maximum absolute errors are 4.684, 3.328, 3.194, and 2.462 mm respectively, with consistent deformation trends, and both methods confirm its reliability. Comparative analysis reveals superior prediction accuracy of the BO-Prophet model over Bayesian optimization-long short-term memory (BO-LSTM) at six characteristic points. The BO-Prophet model achieves a mean absolute error (MAE) of 2.02 mm, representing a 25.2% reduction from BO-LSTM (2.70 mm). Its root mean square error (RMSE) measures 2.36 mm, demonstrating a 35.6% reduction compared to BO-LSTM (3.67 mm). Subsidence predictions for area B using BO-Prophet show high spatial-temporal consistency with SBAS-InSAR monitoring results. Correlation analysis demonstrates a correlation coefficient (R²) exceeding 0.96 between predicted and observed values. The integration of SBAS-InSAR and BO-Prophet shows strong potential for mining subsidence monitoring and forecasting. This combined approach enhances early warning capabilities and supports disaster mitigation strategies in mining areas.