Abstract <p>The construction of large hydropower projects in southwestern China has produced numerous high-steep rock slopes; however, accurately assessing their long-term creep-induced degradation remains challenging. This study addresses this issue by examining the high-steep slopes at the upper reservoir inlet of the Lianghekou pumped storage power station (PSPS). Satellite-based interferometric synthetic aperture radar (InSAR) technology, optimized for steep mountainous terrain, was employed to extract slope deformation data. The InSAR-derived data were integrated with multi-point extensometer measurements. These resulting multi-source datasets were then jointly applied to constrain a numerical model, enabling the quantitative evaluation of the evolution of key mechanical parameters under the combined effects of slope excavation and reservoir impoundment. A generative adversarial network (GAN) was subsequently employed to augment the dataset. Furthermore, a dual-layer ensemble learning regression model featuring adaptive lightweight architecture and threat-aware dynamic topology optimization was developed. The results demonstrate that the strength and stiffness of the rock mass decreased by approximately 4.43% to 29.80% following excavation and impoundment. The degradation exhibited a strong correlation with excavation, a moderate correlation with reservoir impoundment, and a weak correlation with rainfall. These findings confirm the substantial synergistic benefits of integrating multi-source monitoring data for slope deformation analysis. The findings not only advance the understanding of the “excavation–impoundment–degradation” chain mechanism in hydropower engineering slopes but also establishes a theoretical framework and technical foundation for real-time early warning and dynamic risk assessments of landslides.</p>