Diagnosing and Projecting Farmland Ecosystem Health in Arid Regions: An Interpretable Machine Learning and Scenario Simulation Approach Within a Novel Integrity-Based Framework

Agricultural ecosystem health is critical for regional sustainable development and ecological security, yet existing assessment frameworks often lack explicit quantification of ecosystem integrity. Focusing on the Ili River Valley—a major arid Northwest China grain base—this study developed a novel multi-dimensional VOR-ES-I framework (Vitality–Organization–Resilience–Ecosystem Services–Integrity), elevating ecosystem integrity as a core independent dimension alongside traditional components. Using multi-source spatiotemporal data, the framework assessed farmland ecosystem health dynamics, while an interpretable XGBoost-SHAP model identified key drivers and their non-linear mechanisms. Future 2030 patterns were projected by coupling the assessment with the PLUS model under different scenarios. The results showed the composite Ecosystem Health Index (EHI) generally increased during 2000–2024, with a spatial gradient of higher values in the mountains (ranging from 0.53 to 0.80) and lower in the central plains (ranging from 0.11 to 0.52). Potential evapotranspiration (PET) and slope were the top drivers, with notable synergistic interactions and threshold effects (e.g., PET ~557 mm). Climatic factors (especially temperature) grew more influential over time, while socio-economic drivers had weaker direct effects. Scenario projections indicated the Farmland Protection scenario would best enhance ecosystem health (mean EHI = 0.290), whereas the Urban Development scenario might reduce EHI and intensify ecological degradation risks in expansion zones. This study contributes a refined VOR-ES-I framework and methodology, strengthening integrity diagnosis and complex driver interpretation. It provides a scientific basis for integrated assessment, sustainable management, and spatial planning of arid oasis farmland ecosystems.