As a global epicenter of soil erosion, China's Loess Plateau has undergone a profound ecological transformation over the past four decades, shifting from degradation to large-scale vegetation recovery. The Potential Natural Vegetation (PNV) framework has emerged as a critical tool for assessing ecological restoration potential. However, its effectiveness in defining restoration boundaries and guiding vegetation types remains under discussion. Here, we integrate long-term observations with future climate scenarios (SSP1-1.9, SSP2-4.5, SSP5-8.5) to evaluate PNV spatial dynamics using the Comprehensive and Sequential Classification System (CSCS). Our results reveal a stable PNV pattern over the past 40 years, with only 7.8 % area change, suggesting climate-driven vegetation types remained largely constant. From 2001 to 2016, the grassland line (GL) shifted rapidly toward the semidesert PNV boundary, then largely stalled between 2016 and 2021. Concurrently, vegetation composition has continued to shift: forest-type pixels (GPP >0.7) rose to 69.01 % in forest PNV, while steppe-type pixels (GPP 0.3-0.7) declined, indicating an ongoing convergence toward ecologically appropriate vegetation. Spatial mismatch analysis identified 15.92 % of the region as suitable for optimization. Under three GCMs, forest area is projected to increase by 2 %-13.77 % by 2061, though the high-emission scenario (SSP5-8.5) may reduce forest cover under GFDL-ESM4. Future afforestation zones are expected to expand northwestward but remain south of the Hu Line. This study confirms CSCS-based PNV mapping as an effective, practical framework for guiding longterm restoration planning in semiarid ecosystems.