Modeling and Monitoring InSAR Time-Series Deformation in Hangzhou Bay via an SE-Attention Temporal Convolutional Residual Transformer

SBAS-InSAR can extract millimeter-level surface deformation from long SAR time series and, through time-series analysis, enable high-precision monitoring. Although Transformer-based deep learning has shown strong performance in time-series forecasting, its use in practical InSAR time-series processing remains limited. To bridge this gap, this paper proposes a Transformer that integrates a temporal convolutional residual network with a squeeze-And-excitation (SE) attention mechanism for SBAS-InSAR deformation series prediction. Using the area along Hangzhou Bay in China's Yangtze River Delta as the study region, experiments show that the model accurately captures the spatiotemporal variability of the InSAR data and achieves high-Accuracy deformation prediction. Comparative studies with other time-series methods indicate that the proposed model outperforms baseline models in terms of root-mean-square error (RMSE) and fitting accuracy, and, in particular, exhibits stronger stability and generalization in forecasting.