Shrub cover serves as a pivotal ecological indicator for monitoring shrub encroachment in arid and semi-arid grassland ecosystems. However, the lack of spatially explicit data on shrub proliferation dynamics has constrained spatiotemporal pattern analysis and evidence-based conservation planning. To address this gap, this study proposed an integrated monitoring framework for Inner Mongolian grassland (1985-2023), combining a deep learning-derived (Segment Anything Model) plot-scale Shrub Identification Model (SAM-derived SIM) and a machine learning-driven (Random Forest) regional Shrub cover Estimation Model (RF-driven SEM). Plot-level shrub cover estimates were validated through visual interpretation, demonstrating exceptional agreement (R-2 = 0.96, RMSE = 2.22 %). Shrub cover datasets, curated from 5941 standardized 30-m tiles based on high-resolution drone imagery, were used for SEM training and validation. The evaluation showed robust accuracy in regional-scale shrub cover estimation, with R-2 values of 0.64 (RMSE = 3.62 %) in 2015, 0.58 (RMSE = 3.35 %) in 2020, and 0.74 (RMSE = 4.61 %) in 2023. Multi-decadal analysis revealed a persistent increase in regional mean shrub cover at an average annual rate of 0.06 % from 1985 to 2023. Additionally, spatial analysis exhibited increasing shrub cover across two-thirds of the study area, with half (50.94 %, p < 0.05) being significant, demonstrating the spatially pervasive expansion patterns. The developed novel framework established effective shrub encroachment monitoring pathways by integrating drone-derived plot-scale cover data into spatially regional ecological patterns. By revealing the spatially explicit encroachment patterns, our research aims to provide valuable insights for policymakers to implement targeted grazing management and restoration strategies.