Progressive resource allocation outperforms uniform strategies in enhancing urban vegetation growth offset: Evidence from DSTFormer-SHAP modeling across China's 19 urban agglomerations

Urbanization degrades vegetation through direct land conversion while simultaneously stimulating residual vegetation growth through indirect environmental modifications. The growth offset coefficient (GO), measuring the degree to which indirect enhancement compensates for direct loss, serves as a key indicator of vegetation-growth compensation capacity; however, the nonlinear response, temporal lag structures, and optimal management strategies governing GO remain poorly understood. This study develops DSTFormer, a spatiotemporal Transformer integrated with SHAP analysis, and applies it to panel data from 227 cities across China's 19 major urban agglomerations from 2001 to 2022. Results indicate that GO increased at 0.0043 yr −1 , with indirect enhancement (Ind) strengthening (0.0065 yr −1 ) faster than direct loss (DI) deepening (−0.0021 yr −1 ), although the net urbanization effect on vegetation remained negative throughout the study period. SHAP attribution identified forest landscape cohesion (FLCOHESION) as the leading positive predictive factor and impervious patch number (IMNP) as the main fragmentation-related constraint associated with GO variation. Temporal lag structures differed systematically across agglomeration types: the full sample and Optimization-Upgrading Agglomerations showed stronger recent contributions, whereas Cultivation-Development Agglomerations retained stronger historical contributions. Under identical total adjustment budgets, progressive temporal allocation was associated with larger predicted GO gains than uniform allocation. Spatial clustering further identified three regulatory regimes requiring differentiated interventions: a landscape-composition and infrastructure-environment mixed regime, a connectivity-fragmentation constraint regime, and a built-environment morphology and management-environment coupling regime. These findings suggest that improving GO is associated with harnessing nonlinear leverage effects, pursuing synergistic co-optimization, and adopting temporally progressive, agglomeration-specific greening strategies.