Abstract: Ground subsidence has a negative impact on urban development, public safety, and property. Accurate characterization and prediction of ground subsidence patterns are critical for effective disaster prevention and mitigation. This study focuses on the southern region of Taihe County, Anhui Province, utilizing 175 Sentinel-1 images and the Mintpy time-series analysis tool, with SBAS-InSAR as the core, to derive time-series subsidence data from 2018 to 2023. To effectively capture the non-linear and non-stationary features of the data, the study uses the ensemble empirical mode decomposition (EEMD) algorithm to perform multi-scale decomposition of the data, effectively isolating subsidence trends across different frequencies. The TimesNet time series model was then applied to capture the characteristics and periodic distributions of the non-stationary time series data, enabling high-precision land subsidence predictions.The results indicate that the study area is basically stable, though localized subsidence phenomena exist, with subsidence rates ranging from 5 to 50 mm/a. In short-term prediction task, the model achieved root mean square error (RMSE) and mean absolute error (MAE) values of 0.54 mm and 0.31 mm, respectively, demonstrating excellent performance in short-term prediction. For long-term predictions, the model achieved an MAE of 0.83 mm on the validation set, effectively addressing the common challenge of underfitting in long-term tasks. This indicates that the model successfully captures time-series characteristics and provide reliable predictions of land subsidence using time-series InSAR data.