Mapping Soil Organic Carbon in High‐standard Farmland Construction Areas Using Machine Learning Algorithms: Threshold and Interaction Effects of Environmental Variables

ABSTRACT Accurate mapping soil organic carbon (SOC) in high‐standard farmland (HSF) construction areas is essential for optimizing agricultural management practices and achieving carbon sequestration. However, how to quantify the construction activities of HSF and predict SOC is still unclear. In this study, we proposed a framework to quantify the impacts of HSF construction activities on SOC from four perspectives: landscape pattern, agricultural infrastructure, soil property, and agricultural management. Using 298 high‐standard and conventional farmland samples collected in Peixian County, China, machine learning algorithms were employed to explore the threshold effect and interaction effect of HSF construction activities on SOC, and then map the SOC content. Results showed that SOC content in HSF was significantly higher than in conventional farmland ( p < 0.05 ). The gradient boosting decision tree model outperformed other models, with an R 2 of 0.49 on the test set. The SOC content was high in the eastern part and low in the western part of Peixian County, which was influenced by farmland landscape configuration and the degree of completeness of agricultural infrastructure. Non‐linear relationship analysis indicated that being close to water bodies and green land was beneficial to the accumulation of farmland SOC content, with average effective influence ranges of 448 m and 430 m, respectively. Friedman's H statistic reflected complex interaction mechanisms among environmental variables, with total phosphorus content demonstrating the highest interaction strength. These findings highlight that HSF construction increased SOC content. In particular, farmland with high connectivity, comprehensive irrigation and drainage measures, and proximity to water bodies and shelterbelts was conducive to carbon sequestration.