Application of LSTM network to predict InSAR-derived time-series deformation in the reclamation area of Busan, South Korea

Ground deformation is emerging as a growing concern in the reclamation area of Busan, South Korea. Here, we investigate deformation patterns in the reclamation area of Busan by exploiting interferometric synthetic aperture radar (InSAR) technique and deep-learning-based prediction model. Specifically, we employ the persistent scatterer (PS)-InSAR technique to characterize deformation rates observed in Sentinel-1 scenes spanning from June 2018 to June 2023. In addition, we apply a long short-term memory (LSTM) network to predict further deformation, and its performance is compared with that of a gated recurrent unit (GRU). The obtained results reveal significant subsiding patterns in the southern regions, reaching cumulative deformation values of 90 mm. Notably, the LSTM network exhibits a slight improvement over the GRU network, achieving a root mean squared error (RMSE) of 2.79 mm. This paper offers a comprehensive analysis of the long-term evolution and further prediction of deformation in the reclamation area of Busan, providing valuable insights to support remediation efforts.