China's large-scale ecological restoration programs have markedly increased vegetation cover, but their ecological suitability and long-term sustainability remain insufficiently evaluated. To support the development of science-based restoration strategies, this study introduced potential natural vegetation (PNV) as a reference and developed an innovative hybrid model combining random forest algorithm and process-based model (i.e., LPJ-GUESS) to map PNV distributions. Results showed that the hybrid model achieved high predictive accuracy, with an overall classification accuracy of 0.89 and kappa coefficient of 0.87. A comparative analysis of the spatial distribution of the restored (and existing) vegetation and that of the PNV under both the current and the future period was conducted. It revealed that 6.7% of restored forests and 48.8 % of restored grasslands were mismatched with PNV, indicating widespread misallocation of restoration efforts under current ecological restoration programs. Moreover, 2.5 % of existing forests and 34.2 % of existing grasslands were mismatched with PNV in the current period (1993-2022), while 0.5 %-1.4 % of existing forests and 50.8 %70.4 % of existing grasslands are projected to be mismatched with PNV in the future period (2071-2100) under different SSP scenarios. About 32.4 %-41.5 % of China's land area was identified as unstable PNV region under future climate change, within which various vegetation transitions are likely to occur. These findings highlight the necessity to align ecological restoration programs with ecological suitability and sustainability criteria to achieve long-term resilience and cost-effectiveness. The hybrid modeling framework offers a robust tool for guiding adaptive restoration planning across China.