From desert margins to irrigated oases: spatiotemporal dynamics and driving mechanisms of integrated ecological risk in an arid river basin using coupled RSEI-LERI and XGBoost-SHAP frameworks (2001–2025)

Introduction Arid and semi-arid regions face compounding ecological pressures from land degradation, climate variability, and intensifying human activities, yet single-index assessments remain insufficient for capturing coupled ecological dynamics. Methods This study developed a Comprehensive Ecological Risk Index (CERI) by integrating the Remote Sensing Ecological Index (RSEI) and Landscape Ecological Risk Index (LERI) through entropy weighting. The framework was applied to Bayannur City in the upper Yellow River Basin from 2001 to 2025 using multi-source MODIS data on the Google Earth Engine platform. Space-Time Cube modeling, Emerging Hot Spot Analysis, Markov transition matrices, and XGBoost-SHAP were used to characterize spatiotemporal trajectories and quantify driving mechanisms. Results The results revealed a persistent north-south risk polarization. Consecutive Hot Spots covering 18,235.89 km 2 dominated the Yinshan-Ulan Buh Desert zone, whereas Consecutive Cold Spots covering 8,543.10 km 2 were concentrated in the Hetao Irrigation District. The combined proportion of high- and very high-risk areas declined from 57.78% to approximately one-third over the study period, with the most pronounced improvement during 2020-2025. NDBSI consistently acted as the primary risk amplifier, while NDVI and surface water exerted sustained suppressive effects. Discussion These findings provide spatially explicit evidence for identifying ecological compensation zones, optimizing restoration investments, and supporting adaptive ecological risk monitoring across dryland river basins.