Machine learning and structural equation modeling for revealing the influence factors and pathways of different water management regimes acting on brown rice cadmium

Excessive cadmium (Cd) in brown rice has detrimental effects on rice growth and human health. Water management is a cost-effective, eco-friendly measure to suppress Cd accumulation in rice. However, there is no acknowledged water management regime that reduces Cd accumulation in brown rice without compromising the yield. Meanwhile, the major factors affecting brown rice Cd and the pathways of water management affecting rice Cd are not clear. This study explored major factors affecting brown rice Cd using machine learning (ML) and examined the pathways of water management affecting rice Cd using a structural equation model (SEM). Three water management systems were set up, namely flooding, water-saving, and wetting irrigation. Results showed that water-saving irrigation increased dry matter and reduced Cd content and translocation. Root uptake during the grain filling stage and Cd remobilization before the grain filling stage contributed 36 % and 64 % of the Cd accumulation in brown rice, respectively. ML explained 97 % of the variance, suggesting that crop covariates were the most important (e.g., the brown rice bioconcentration factor (12 %), stem Cd (9 %), root-to-stem translocation factor (7 %)), followed by soil covariates (e.g., reducing substances 12 %) and water management (3 %). All SEM explanatory variables collectively explained 94 % of the variation, with a predictive power of 76 %. Water treatments indirectly affected soil available Fe and Mn (indirect effect coefficient = 0.909), iron plaques (indirect effect coefficient = 0.866), soil available Cd (indirect effect coefficient = -0.671), and Cd intensity of xylem sap (BI