Spatiotemporal evolution and driving factors of agricultural carbon emissions at county level in Guangdong Province based on machine learning

The county level serves as a fundamental regional unit for in-depth analysis of the relationships between agricultural carbon emissions and urbanization. Understanding the spatiotemporal evolution and driving mechanisms of agricultural carbon emissions at the county level is crucial for guiding green and low-carbon development in rural areas and advancing rural revitalization efforts. This study aims to elucidate the spatiotemporal evolution patterns and driving factors of county-level agricultural carbon emissions in Guangdong Province from 2000 to 2022. Using the latest emission inventory system, this study systematically calculated the agricultural carbon emissions for 124 counties (districts) in Guangdong Province. By integrating the GIS and machine learning technique including local spatial autocorrelation analysis and the random forest (RF) algorithm, we collaboratively identified the key driving factors and spatiotemporal evolution patterns of agricultural carbon emissions at the county level. The results are shown as follows: 1) the overall agricultural carbon emissions in Guangdong Province exhibited a fluctuating downward trend over the study period, with obvious spatiotemporal heterogeneity observed at the county level. The emissions hierarchy, from highest to lowest, was Pearl River Delta > Western Guangdong > Eastern Guangdong > Northern Guangdong; 2) Geographic Information System (GIS)-based local spatial autocorrelation analysis revealed distinct spatial clustering characteristics of agricultural carbon emissions. High-value clusters were predominantly located in Western Guangdong, whereas low-value clusters were concentrated in the core area of the Pearl River Delta. Notably, the degree of internal clustering within each region showed a continuous upward trend over time; 3) a notable distinction between per capita agricultural carbon emissions and total emissions was observed, with high-value per capita emissions clusters appeared mainly in Northern Guangdong, while low-value clusters were concentrated in the Pearl River Delta, indicating a “spatial spillover” effect where emissions in densely populated urban areas impact surrounding regions; 4) the random forest (RF) model, constructed using agricultural indicators selected through Pearson correlation analysis, identified the primary factors influencing agricultural carbon emissions. These included land ploughing, the usage of fertilizers and pesticides, and total agricultural machinery power. Among these indicators, the contribution ratio of fertilizer and pesticide usage showed a gradual decrease, while the advancement of agricultural mechanization emerged as a positive driving factor in reducing agricultural carbon emissions. The results of this study highlight the complexity and regional variability of agricultural carbon emissions in Guangdong Province, and the integration of local spatial autocorrelation and RF algorithms provides a robust analytical framework for understanding these dynamics. In addition, this research offers valuable policy insights and quantitative tools for regional agricultural carbon emission reduction and promotes green and low-carbon development strategies. Overall, our findings underscore the importance of targeted, region-specific policies to address the unique challenges and opportunities within different areas of Guangdong Province. By leveraging advanced analytical techniques and comprehensive data sets, policymakers can better understand and mitigate the factors driving agricultural carbon emissions, paving the way for sustainable and environmentally friendly agricultural practices.