Protected areas play a crucial role in global biodiversity conservation, yet they are increasingly threatened by climate change and human activities. Traditional assessments that focus exclusively on vegetation productivity (vigor) fail to capture essential dimensions of ecosystem health, such as biodiversity structure (organization) and adaptive capacity (resilience). Here, we propose a comprehensive vigor-organization-resilience framework that integrates remote sensing, machine learning, and field data to evaluate ecosystem health across 431 Chinese National Nature Reserves from 1990 to 2020. Our findings reveal that while 60.3% of reserves maintained healthy ecosystems, particularly forests, which outperformed grasslands, we identified a concerning “greening but weakening” paradox. Specifically, vegetation vigor increased by 29.4%, yet ecosystem resilience declined significantly by 26.8%, and organizational integrity decreased moderately by 6.1%. Notably, grasslands exhibited an increase in organizational structure (+7.4%), whereas forests experienced a decline (−9.9%), indicating biome-specific stress responses. Furthermore, the gap in ecosystem health between protected and adjacent non-protected areas narrowed from 3.7% to 1.8%, with measurable spillover effects contributing to an overall improvement in external ecosystem health (+2.8%). Topography, soil properties, and management practices accounted for 60% of the spatial variability in ecosystem health, while climate variability and disturbances were responsible for 51% of the temporal changes observed. By quantifying these decoupled dimensions of ecosystem health, our approach facilitates targeted conservation interventions that are essential for achieving functional, rather than merely area-based, outcomes in line with the Global Biodiversity Framework’s 30×30 initiative.