The risk of tailings dam failure is essentially derived from the gradual accumulation of damage within the dam structure under surface deformation fields. Traditional deformation prediction methods based on single-point time series are struggle to capture the spatial variations in damage distribution and spatiotemporal coupling characteristics, resulting in an inadequate ability to describe the structural failure process of the dam. Therefore, a novel method is proposed for extracting spatiotemporal features of structural damage and conducting area-wide prediction for tailings dams by integrating small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) technology with convolutional long short-term memory (ConvLSTM) neural networks. The high-piled tailings dam of Wenzhuang tailings reservoir in Yunnan Province is selected as the research object. A total of 99 Sentinel-1A images from January 2021 to December 2024 were first processed by SBAS-InSAR to obtain the full-domain spatiotemporal grid deformation field of the dam, revealing the spatial differentiation law of deformation rate and cumulative deformation. Then, the time-series deformation grid sequence was constructed into a five-dimensional feature tensor, and an encoder-predictor-decoder spatiotemporal coupling model was established using the ConvLSTM network. The model synchronously learns the spatial neighborhood features and long-term temporal dependence of the deformation field, realizing spatiotemporal areal prediction of the dam structural damage evolution process. The results indicate that deformation in the study area is concentrated on the crest of the embankment dam and the sedimentation plain, with a maximum vertical deformation rate of -33.28 mm/a and a maximum cumulative deformation of -134.7 mm. The deformation exhibits significant spatial heterogeneity and temporal progression, revealing structural differences within the dam body and progressive consolidation damage characteristics. Comparisons between the deformation field predicted by the developed model for December 2024 with the SBAS-InSAR measured values show that the mean absolute error (MAE) was 6.44 mm, the root mean square error (RMSE) was 9.51 mm, and the coefficient of determination (R2) reached 0.87, and the Structural Similarity Index (SSIM) was 0.89, indicating that the model can accurately reconstruct the spatial structure and evolutionary trends of the damage deformation field. Prediction results for December 2025 indicate that deformation values on the southern side of the dam will reach -160.4mm, with the overall dam structure remaining in a progressive deformation phase. The methodology presented in this paper combines deformation prediction with the extraction of structural damage features, offering a new technical approach for the detailed diagnosis of the damage evolution process in tailings dams and for early warning of dam failure risks.