Landform, soil properties, soil cadmium (Cd) pollution and rainfall are the important factors affecting the spatial variation of rice Cd. In this study, we conducted big data mining and model analysis of 150,000 rice-soil sampling sites to examine the effects by the above four factors on the spatial variation of rice Cd in Hunan Province, China. Specifically, the variable coefficient of rice Cd in space was significantly correlated with the partition scale according to the logistic fitting. The improved random forest results suggested that elevation (DEM) and pH were the two most important factors affecting the spatial variation of rice Cd, followed by relief, soil Cd content and rainfall. Typically, variance partitioning analysis (VPA) revealed that both the soil property and the interactive effects between the soil property and Cd pollution were the principal contributors to the rice-Cd variation, with the respective contributing rates of 30.5 % and 29.0 %. Meanwhile, the partial least square-structural equation modelling (PLS-SEM) elucidated 4 main paths of specific indirect effects on rice-Cd variation. They were landform → physicochemical property → soil acidity → rice-Cd variation, landform → soil acidity → rice-Cd variation, physicochemical property → soil acidity → rice-Cd variation, and soil texture → soil acidity → rice-Cd variation. This work can provide a general guidance for scientific zoning, accurate prediction and prevention of Cd pollution in paddy fields.