Understanding spatial drivers of soil organic carbon in arid oasis ecosystems is essential for guiding precision soil management and enhancing land sustainability. This study integrates 644 surface samples and 9 soil profiles with multisource environmental data in the hyperarid eastern Tarim Basin, employing geostatistics and machine learning (random forest, support vector machines, ordinary least squares, back propagation) to quantify driving mechanisms. Key findings: 1) extreme soil organic carbon spatial polarization (0.50-21.70 g kg(-1), mean = 4.47 g kg(-1)) with northern and southern alluvial zones containing 2.1 times higher soil organic carbon than central deserts (p < 0.01); 2) random Forest achieved optimal prediction (coefficient of determination = 0.81, root mean square error = 1.32 g kg(-1)) by resolving nonlinear soil organic carbon-environment interactions; 3) pedogenic properties (texture, cation exchange capacity, salini- ty; 47.5%) dominated soil organic carbon variation, followed by anthropogenic drivers (land use intensity, 14.2%) and soil taxo-nomy (10.9%), while climate and topography showed minimal control (< 8%). Human-modified processes override climatic constraints in shaping soil organic carbon patterns, providing actionable insights for clay-organic stabilization and irrigation optimization. This methodology establishes a transferable framework for deciphering soil organic carbon dynamics in global drylands, directly informing climate-resilient land management.