Machine learning-based assessment of soil organic carbon dynamics in soybean–wheat rotations in eastern China

Abstract

Soil organic carbon (SOC) is a critical component of agroecosystems, influencing carbon cycling, soil fertility, and structure, thereby affecting crop productivity. This study evaluated the spatial distribution and dynamics of SOC stocks in eastern China under soybean–wheat rotations using advanced machine learning models. Data were collected from Anhui, Hebei, Henan, Jiangsu, Shandong, Tianjin, and Beijing, measuring SOC at two soil depths (0–15 cm and 15–30 cm) before sowing and after harvest during 2022–2024. Among the models tested, Random Forest (RF) provided the most accurate SOC predictions, particularly in the 0–15 cm layer (R2 = 0.89, RMSE = 0.95, PRD = 3.41). Results revealed SOC increases following soybean cultivation and decreases after wheat harvest, with regional variations shaped by environmental factors such as standardized height (standh), NDVI, temperature seasonality (tempSeason), and LS factor. The higher biomass and extensive root system of soybean significantly enhanced SOC, whereas wheat’s lower biomass contributed to SOC depletion. These findings underscore the influence of crop type and management practices on soil carbon stocks, highlighting the potential of soybean-inclusive rotations to improve soil health and mitigate climate change impacts.