Decoding the spatiotemporal dynamics of cropland carbon emission drivers in China: A machine learning-based analysis

The agriculture sector significantly contributes to greenhouse gas emissions, highlighting the necessity of diminishing its carbon footprint for successful climate change mitigation. Nonetheless, insufficient understanding of the spatiotemporal patterns and diverse factors influencing cropland carbon emissions (CCE) hinders the formulation of targeted, evidence-based mitigation solutions. This study assesses CCE across 2852 Chinese counties from 2001 to 2020, using a comprehensive framework that integrates four principal sources: material inputs, paddy fields, straw combustion, and soil emissions. To investigate the fundamental dynamics, alterations in CCE drivers were examined using a sophisticated machine learning model. The findings demonstrate that China's cropland emissions rose consistently, reaching a peak of 1070.54 Tg in 2014, thereafter decreasing to 890.87 Tg in 2020. Emission trajectories exhibited significant variation among counties. Nationally, emissions were mostly influenced by farmland usage, whereas the impact of elevation decreased over time, indicating a reduced significance of topography. Regionally, emission drivers displayed considerable variance; for instance, the contribution of arable land area in South China rose markedly, from 0.44 to 0.90. While cropland scaling generally decreased emissions, urbanization and GDP growth were significant contributors to emissions in the Loess Plateau and Qinghai-Tibet Plateau. This study provides insights for the formulation of region-specific mitigation programs by detecting regional and temporal variations in emission drivers. The results enhance cropland practices to promote carbon neutrality and sustainable agricultural development.