Abstract: Given the limitation of single-orbit Interferometric Synthetic Aperture Radar (InSAR) monitoring in measuring line-of-sight (LOS) deformation, accurately identifying deformation zones and characterizing their spatiotemporal evolution in tailings dams remained challenging. This study proposes an integrated method combining ascending and descending Small Baseline Subset (SBAS) InSAR and Variational Mode Decomposition (VMD) to reconstruct the two-dimensional (2D) deformation field and achieve high-precision decomposition of time-series deformation characteristics of the dam. Using a phosphogypsum tailings reservoir in Yunnan as a case study, Sentinel-1 ascending and descending SAR data from 2022 to 2024 were processed jointly using SBAS-InSAR to derive vertical and east-west 2D deformation fields. VMD is used to decompose the vertical time-series deformation signals into trend, periodic, and residual components, and the results were compared and validated against traditional decomposition models. The results show that (1) Ascending and descending deformation results identify three deformation zones in the tailings dam, with subsidence spatially correlated to the dam boundaries. These zones primarily occur along embankment slopes and dried beach areas, with maximum cumulative settlement exceeding 426 mm. (2) The overall deformation of the tailings dam is dominated by vertical settlement, showing an initial acceleration followed by a deceleration trend, with periodic fluctuations. The east-west deformation exhibited a synergistic pattern, westward displacement of the main dam body, eastward movement of the partition sub-dams, and bidirectional displacement in the dried beach area. (3) The long-term settlement exhibited an accelerated-then-decelerated evolution pattern with significant spatial heterogeneity, characterized by minimal deceleration in the central region. The periodic component exhibited distinct seasonality with a marked hysteresis relative to rainfall events. VMD resolved issues of amplitude underestimation (associated with forced annual periods) and linear overfitting observed in conventional models. The high reconstruction accuracy (R² > 0.99) validates its reliability for deformation signal decomposition and temporal characteristic analysis, significantly improving the accuracy and reliability of time-series deformation characteristic analysis. The proposed method demonstrates strong generalizability for deformation monitoring and temporal evolution analysis of tailings reservoirs.