Monitoring the spatio-temporal changes of topsoil organic carbon content in Mollisol croplands using Landsat time series (2000–2023) and ensemble learning

Mollisols play a crucial role in global sustainable development due to their high fertility and large carbon stocks. However, the high spatio-temporal resolution dynamic monitoring of soil organic carbon (SOC) content in global Mollisol croplands remains limited, particularly regarding its driving factors and regional variations. In this study, we utilized 35,760 Landsat satellite images from 2000 to 2023 to develop an ensemble learning model (R 2 c = 0.71, RMSE c = 4.25 g/kg, R 2 v = 0.61, RMSE v = 4.98 g/kg) with 14 features, including spatial location, topography, and spectral indices, to estimate the annual topsoil SOC content dynamics in the global Mollisol croplands. The results showed that the topsoil SOC content averaged 21.30 g/kg, with higher levels in Eurasia than in the Americas. From 2000 to 2023, global topsoil SOC content exhibited a significant fluctuating increase, rising by 3.17% overall, with an average annual percentage change of 0.04 g/kg/year. Considerable regional variation was observed, with a sustained 5.66% increase in the Americas but fluctuating declines in Eurasia, including a 2.21% decrease in Northeast China. These regional disparities reflect the coupled effects of vegetation dynamics, soil–water-atmosphere interactions, and human activities. Further analyses reveal the dual sensitivity of SOC dynamics to agro-environmental controls and socio-economic drivers, including cultivation practices, policy shifts, and socio-political stability. The findings of this study represent a new baseline for precision agricultural management and global soil carbon monitoring.