Answers are grounded in this paper and cite their sources.
Long-span cross-sea bridges operate under complex marine, atmospheric, and geotechnical conditions, making continuous, high-precision deformation-monitoring essential for structural safety. This study proposes an integrated framework that combines enhanced small baseline subset (E-SBAS) time-series InSAR with a hybrid deep-learning model to characterize and predict the deformation of the Hong Kong–Zhuhai–Macao Bridge from 2018 to 2026. Using 172 Sentinel-1A SAR images, E-SBAS results show that the bridge superstructure remained largely stable during 2018–2024, with deformation rates typically within ±5 mm/year, while the artificial islands, especially the Hong Kong Boundary Crossing Facilities (HKBCFs), experienced significant but decelerating consolidation-related subsidence exceeding 40 mm/year. Seasonal deformation is closely associated with thermoelastic expansion and contraction, with amplitudes of 10–18 mm, whereas long-term trends reflect concrete creep and shrinkage, as well as foundation compression. A temporal convolutional network–transformer–cross-attention model was developed to jointly learn multiscale temporal patterns and integrate meteorological variables. The model achieves high predictive accuracy (R2≈0.93, root-mean-square error≈2.7 mm, and mean absolute error≈1.6 mm) and effectively reconstructs long-term and seasonal deformation behavior. Predictions for 2024–2026 indicate continued but markedly decelerating settlement at the HKBCF, consistent with the progressive stabilization of nondredged reclamation zones. This study demonstrates the first comprehensive integration of TS-InSAR and advanced deep learning for cross-sea bridge monitoring, offering a scalable framework for intelligent deformation forecasting and long-term safety assessment of major coastal infrastructure.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: