Beyond disturbance: habitat quality, regeneration and diversity shape vegetation rewilding potential in urban parks

Amid rapid urbanization and environmental change, rewilding has become a key strategy to restore ecological autonomy in cities. Urban parks, though heavily managed, still possess latent recovery capacity that supports biodiversity and resilience. This study surveys 21 urban parks in Qingdao across a disturbance gradient to map urban wilderness and evaluate Vegetation Rewilding Potential (VRP). Using principal component analysis (PCA), XGBoost modeling, and SHAP interpretation, we quantify VRP and its drivers under high-use, fragmented conditions. Results show that overall VRP is low, with only 14.6% of plots having high potential. Spatial distribution pattern reveals that VRP jointly shaped by habitat quality, disturbance type, and disturbance intensity. Crucially, we reveal that VRP is decoupled from human traffic; over-managed, low-use spaces often exhibit lower VRP than highly frequented parks due to intensive maintenance. The predominant drivers of VRP are regeneration seedling conversion capacity (14.4%), litter thickness (8.9%), and species richness (6.6%), each demonstrating nonlinear threshold effects. By integrating micro-scale wilderness mapping with interpretable machine learning, this study overcomes traditional evaluation limits to provide a refined diagnostic tool. Ultimately, this framework shifts the rewilding paradigm from remote wilderness to everyday urban landscapes, establishing a scientific baseline for transitioning managed green spaces toward self-sustaining ecological autonomy.