Abstract <p>Exploring the main controlling factors of soil organic carbon (SOC) in farmland is crucial for evaluating soil quality, optimizing agricultural management, and ensuring food security. Current studies used machine learning methods to determine the relative importance of environmental variables, but fail to visualize nonlinear relationships between SOC and covariates. Moreover, the main controlling factors of SOC in riparian farmland may be particularly complex because of the dual factors of lakes and agricultural activities. [Materials and methods] Peixian County in the North China Plain is a typical agricultural area, adjacent to Weishan Lake. This study aims to explore the main control factors of SOC in Peixian by using gradient boosting decision tree (GBDT) model. Results show that 16 variables can explain 68% of SOC variation. Annual average precipitation, annual minimum temperature, the distance to rural settlements, and the distance to the lake (Dis_lake) emerge as the main influencing factors of SOC. Particularly, the <i>R</i>² of the two-way interaction effect between Dis_lake and of any variable exceeds 0.4. When some of the environmental variable values are extremely high or extremely low, SOC no longer changes accordingly. For instance, when Dis_lake surpasses 9000 m, SOC no longer decreases. These findings highlight the impact of lake-related factors in the spatial variation of riparian farmland. The results from the GBDT analysis provide valuable guidance for agricultural management in the region.</p>