Abstract: The source region of the Yellow River is an important water conservation area and alpine ecological barrier in China. Changes in the regional soil carbon pool affect not only carbon sink capacity, but also ecological security and assessment of water ecological functions. To address the limited understanding of the long-term dynamics of soil organic carbon (SOC) and the lack of quantitative studies at the regional scale in this area, this study constructed a regional Denitrification-Decomposition (DNDC) model for the source region of the Yellow River to simulate SOC dynamics from 1990 to 2024, and combined a random forest model with SHapley Additive exPlanations (SHAP) to identify the influencing factors. The results showed that the calibrated DNDC model performed well in the study area. The correlation coefficients between simulated results and reference data ranged from 0.55 to 0.75, and relative root mean square error was controlled within 30%, indicating that the model could reasonably reproduce the spatial distribution of SOC in the study area. From 1990 to 2024, SOC in the source region of the Yellow River exhibited a significant increasing trend. Regional mean SOC density increased from 23.8×10⁴ kgC/ha to 25.6×10⁴ kgC/ha, with an average annual increase rate of 515.05 kgC/ha/yr. Spatially, SOC showed a pattern of being high in the southeast and low in the northwest, and the increase was mainly concentrated in the 0-10 cm topsoil layer. Attribution analysis showed that soil clay content, with a contribution rate of 23.3%, and initial SOC content, with a contribution rate of 16.6%, were the key background factors affecting SOC variation. Dynamic factors such as grazing intensity, air temperature and NDVI also showed strong associations with SOC increase. The results help to improve understanding of soil carbon sink dynamics in the source region of the Yellow River and provide a scientific basis for soil carbon sink assessment and regional ecological management.