A Bayesian network simulates the responses of soil organic carbon to environmental factors at a catchment scale

Understanding the cause-and-effect relationship and the factor interactions between soil organic carbon content (SOC) and environmental properties remain challenging at a catchment scale. Using an extensive dataset (n = 1126), we aimed to develop a Bayesian network (BN) to simulate the response of surficial SOC (0–5 cm) to multiple interacting environmental factors across a semi-humid catchment (1.1 km²) on the Chinese Loess Plateau. We found that landscape position, silt content, and Normalized Difference Vegetation Index (NDVI) controlled SOC, with the functional pathways being that landscape position affected silt content and NDVI, then silt content and NDVI collectively affected SOC. Low landscape position, low silt content and high NDVI were associated with high SOC. The interactive effect of silt content and NDVI on SOC was antagonistic whereas the strongest contribution of silt content could be partly eclipsed by NDVI. Moreover, when SOC was in high state, low state silt content, high state NDVI, and low state altitude were convincingly inferred with probability in 49%, 65%, and 63%, respectively. These findings highlight the role of the landscape position–vegetation–soil texture interactions in accounting for SOC variation at the catchment scale. We also emphasize the potential of the BN as an effective tool for detecting cause-and-effect relations in terrestrial ecosystem.