Spatial pattern and driving mechanisms of dryland landscape ecological risk: Insights from an integrated geographic detector and machine learning model

• Drylands are highly sensitive and vulnerable to global climate and land use changes. • Reveals 30-year landscape ecological risk patterns in the arid Tarim River Basin. • Introduces machine learning to identify nonlinear ecological risk drivers in drylands. • Highlights spatial clustering of ecological risks with dominant natural drivers. Drylands are among the most vulnerable and sensitive regions to global climate and land use changes. As one of the largest inland arid river basins, the Tarim River Basin exhibits diverse land use patterns that have exacerbated ecological risks, degraded the environment, and heightened ecosystem vulnerability. This study analyzed the spatiotemporal evolution of landscape ecological risks in the basin from 1990 to 2020 using a landscape ecological risk model. A novel combination of the geographic detector and machine learning was employed to identify the nonlinear driving mechanisms of ecological risk. Key findings include: (1) Unused land dominated the basin, with land use showing “three increases and three decreases” trends: increases in cropland, construction land, and unused land, and decreases in grassland, water bodies, and forested areas. (2) High-risk areas were predominant, increasing by 2.31%, while low- and medium-risk areas declined by 0.60% and 1.49%, respectively. (3) Ecological risks transitioned from dispersed to aggregated patterns, with significant clustering of high- and low-risk zones. (4) Elevation, NDVI, and distance to urban centers were key drivers in single-factor analyses, while dual-factor interactions, particularly involving NDVI, consistently enhanced risk. These findings elucidate the spatiotemporal dynamics of land use and ecological risks, offering insights for ecological management and sustainable development.