Soil organic carbon (SOC) is a critical component of the terrestrial ecosystem carbon cycle. Its spatial distribution and formation mechanisms are essential for understanding carbon sink functions and informing management strategies in arid regions. However, current research has mainly focused on local areas, lacking systematic analyses of large-scale SOC spatial distribution and its driving mechanisms in arid zones-particularly the interactions between hydrothermal stress and other environmental factors. To address this gap, this study focuses on Xinjiang, a typical arid region in China. Based on 2757 measured soil samples, we integrated remote sensing imagery and multi-source environmental covariates to construct a Random Forest (RF) model for spatial prediction of SOC content. We further innovatively combined SHapley Additive exPlanations (SHAP analysis) and piecewise Structural Equation Modeling (pSEM) to analyze the dominant factors and response mechanisms controlling SOC in arid regions. The results of this innovative study are: (1) the Random Forest model achieved reliable prediction with an R-2 of 0.67, revealing that SOC spatial distribution exhibits a geomorphic gradient pattern of high levels in mountains (over 15 g kg(-1)), moderate levels in oases, and low levels in deserts (below 5 g kg(-1)), primarily controlled by hydrothermal conditions, with a total estimated stock of 3.19 Pg C for the top 30 cm; (2) evapotranspiration (ET) and soil moisture (SM) are the most important positive drivers of SOC, while soil temperature (ST) and vapor pressure deficit (VPD) exert significant negative effects; and (3) human activities regulate local hydrothermal stress through irrigation, but SOC stability is highly dependent on management practices. This study reveals the coupled pathways of hydrothermal stress mechanisms and human regulation in SOC formation, providing a theoretical basis and methodological support for carbon stock assessment and precision management in arid regions.