Analysis of spatiotemporal variations and driving mechanisms of ecosystem health in arid oasis urban agglomerations using machine learning

With escalating climatic variability and anthropogenic pressures, arid ecological systems encounter substantial environmental uncertainty. The Urban Agglomeration on the Northern Slopes of Tianshan Mountains (UANSTM), located in Northwest China’s arid zone, exhibits a characteristic orographic-oasis-arid composite ecological structure. Ecosystem health assessment faces unique challenges including salinization and sand-fixation services. This investigation integrates windbreak sand stabilization services with salinity-incorporated remote sensing ecological indices, establishing a suitable ecosystem health assessment framework for arid regions. This approach enables enhanced characterization of desert-oasis system ecological attributes and facilitates analysis of spatiotemporal evolutionary patterns in ecosystem health during 2000–2020. Through integrating spatial autocorrelation analysis, XGBoost-SHAP machine learning, and PLS-SEM modeling, we explore natural and anthropogenic driving mechanisms. Results indicate: (1) Ecosystem vitality demonstrated sustained enhancement from 0.21 to 0.25, whereas ecosystem resilience exhibited decline from 0.45 to 0.44 following 2010, indicating an inherent trade-off between productivity augmentation and systemic stability.(2) According to machine learning outcomes, temperature, precipitation, and soil sand content emerged as the three most critical determinants, with evapotranspiration and human footprint indices additionally ranking among the top three factors during various temporal periods, demonstrating comparable significance.(3) Temperature emerged as the primary determinant of climatic configurations, accounting for 40% of the contribution, whereas soil sand content constituted the dominant factor in edaphic patterns with a 30% contribution rate. These dual frameworks exhibit alternating control over ecosystem health dynamics, while anthropogenic impacts demonstrate increasing significance. This study enhances ecosystem health assessment frameworks for arid regions and provides guidance for sustainable development strategies.