A prediction model of soil organic carbon into river and its driving mechanism in red soil region

Soil erosion contributes to the irreversible loss of soil organic carbon (SOC) into rivers (SOCR), posing risks to food security and carbon cycle assessments. Red soil regions, characterized by high carbon sink potential and selenium enrichment, are particularly vulnerable. However, existing studies largely rely on small-scale experiments, with limited understanding of basin-scale SOCR dynamics and their driving factors. This study integrates the Soil and Water Assessment Tool (SWAT) for sediment yield simulation and a Soil Organic Carbon Content (SOCC) model to quantify SOCR at the basin scale. A Random Forest-based prediction model was developed to explore the spatial-temporal variability and driving mechanisms of SOCR in the Dongjiang River Basin (DRB), a representative red soil region in southern China. Results indicate significant spatial-temporal heterogeneity, with higher SOCR observed in downstream, human-disturbed areas during flood seasons. The model demonstrates excellent performance (R<SUP>2</SUP>>0.9). Key drivers of SOCR variability include sediment yield, cultivated land area (CULT), and urban land area (TOWN), with urbanization showing stronger sensitivity than cultivation due to factors such as city size and impervious surfaces. The proposed framework reveals the dynamic change characteristics of SOCR and its driving mechanism, which has the potential to be generalized to other basins with similar studies, and provides a technical support for land resource management and carbon cycling in the erosion-prone red soil region.