Ecological-state and modeled ecosystem-service trajectories of persistent cropland-to-forest transitions across policy-context windows in Hunan, China

Assessments of cropland-to-forest transitions often rely on vegetation greenness or composite indices, obscuring whether structural greening reflects functional recovery and how initial conditions shape post-transition trajectories. We examined remotely detected persistent cropland-to-forest transitions across three policy-context windows (2000–2006, 2007–2013, and 2014–2020) in Hunan Province, China. The Grain for Green Program provided the institutional context for these windows rather than verified treatment assignment. Six component indicators—greenness (EVI), wetness (WET), bare-surface exposure (NDBSI), modeled water yield, modeled soil retention, and an NPP-based productivity proxy—were evaluated using balanced event-time alignment, multivariable covariate adjustment, directional synchrony, and spatially validated XGBoost-SHAP. Canopy greenness increased (0.0033–0.0037 yr −1 ) and surface exposure declined (−0.0034 to −0.0039 yr −1 ) consistently across all windows. By contrast, directional agreement between EVI changes and modeled service-indicator changes was generally around or below 50%, and WET showed a near-flat trend in the latest window. After covariate adjustment, between-window differences in EVI and NDBSI were no longer statistically significant, whereas differences in WET and modeled services remained significant. Pre-transition baselines and climate/topography accounted for 50%–96% of model-explained variance. County-level socioeconomic context generally accounted for a small share of model-explained variance, with indicator- and window-specific exceptions. Out-of-sample predictive performance varied across indicators and windows under the two spatial validation strategies, with baseline state and precipitation generally dominating predictions. Because the detected transitions were not administratively verified program parcels, the results describe land-transition dynamics under evolving policy contexts rather than causal program impacts. These findings show that structural greening does not ensure synchronous functional improvement and support component-level, service-specific monitoring.