Green manure management effects on soil labile carbon and rice productivity: A meta-analysis and machine learning synthesis

Context Green manure is increasingly recognized as a sustainable practice for improving soil fertility, enhancing carbon cycling, and sustaining rice productivity. However, its effects on soil labile carbon pools and rice yield vary widely across climate conditions, soil backgrounds, agricultural management measures, and residue quality. The relative importance of green manure characteristics and management practices in regulating these responses remains insufficiently quantified. Objective This study aimed to quantify the effects of green manure incorporation on soil labile carbon pools and rice yield, identify the roles of climate, soil characteristics, management measures, and green manure traits, and determine the key drivers of carbon pool activation and yield formation by integrating meta-analysis with machine learning. Methods A meta-analysis was conducted using 472 paired observations from 94 peer-reviewed studies. We quantified the effects of green manure on soil organic carbon (SOC), microbial biomass carbon (MBC), dissolved organic carbon (DOC), particulate organic carbon (POC), and rice yield. Subgroup analyses were performed according to climate zone, mean annual temperature, mean annual precipitation, initial soil characteristics, experimental duration, green manure incorporation rate, incorporation timing, green manure type, and green manure C, N, P, and K contents. Random forest, LightGBM, and gradient boosting decision tree models were used to identify dominant predictors. Results and conclusions Green manure significantly increased SOC, MBC, DOC, and rice yield by 14.6%, 26.5%, 29.4%, and 21.9%, respectively, demonstrating its dual role in activating soil carbon pools and enhancing rice productivity. These responses were jointly regulated by climate, soil properties, management practices, and green manure quality, with stronger effects generally observed under warm and humid climates, appropriate application rates and incorporation timing, favorable soil backgrounds, and suitable residue nutrient composition. Machine learning identified mean annual temperature, green manure incorporation rate, and experimental duration as key drivers of SOC; soil available phosphorus, green manure carbon content, and soil pH as the most influential variables within the MBC model; initial SOC, soil pH, and experimental duration as major regulators of DOC; and green manure carbon content, soil pH, and soil total nitrogen as key determinants of rice yield. Significance This study provides quantitative evidence that green manure improves soil labile carbon pools and sustains rice productivity. Optimizing green manure type, application rate, incorporation timing, and residue nutrient quality according to regional climate and soil conditions can support sustainable, region-specific paddy management.