Machine learning models predict organic fertilization effects on soil organic carbon stocks in global agroecosystems

Organic fertilizers are more environmentally friendly than traditional fertilizers and affect the dynamics of soil organic carbon (SOC) stocks; however, the impact of organic fertilizers on SOC dynamics in global agroecosystems is poorly understood. Here, we use machine learning (ML) models to investigate the effect of organic fertilizer application on SOC stocks under different agronomic practices and environmental conditions on a global scale. Three tree-based ML algorithms—Random Forest (RF), Gradient Boosting Regressor (GBR), and Extreme Gradient Boosting (XGB)—were compared for their efficacy in predicting organic fertilizer effects on SOC stocks in global agroecosystems. Results indicate that GBR outperformed RF and XGB, with the GBR model producing higher training and testing R 2 values (0.96 and 0.73, respectively) and lower training and testing root mean square errors (2.1 and 2.23, respectively). Applied organic fertilizer dosage, cumulative organic fertilizer application amount, and soil bulk density were key factors affecting SOC stocks in global agroecosystems. Climate factors such as mean annual temperature and mean annual precipitation also significantly influenced the prediction of SOC stocks in global agroecosystems. This study underscores the potential of ML models in unravelling complex factors affecting SOC stocks in global agroecosystems. Future research should prioritize long-term monitoring to enhance our understanding of factors affecting SOC stocks in agroecosystems and inform sustainable soil management practices. • Machine learning models are used to study organic fertilization effects on soil organic C (SOC) • Gradient Boosting Regressor model achieved the highest accuracy in predicting SOC stocks • Organic fertilizer dosage, cumulative amount, and soil bulk density are key drivers of SOC • SOC stocks increased with organic fertilizer dosage in global agroecosystems