Abstract
The high cadmium (Cd) accumulation ability of wheat has garnered significant attention in China. It is crucial to identify the key factors affecting Cd accumulation in wheat and to develop predictive models to derive the threshold concentration of Cd in soil for safe wheat production. A total of 311 soil-wheat paired datasets were collected from both literature and field surveys in China, in which the ranges of Cd in soil and wheat grain were 0.068-13.500 mg/kg and 0.006-2.190 mg/kg, respectively. Correlation analyses and Partial Least Squares Path Model indicated that soil Cd, soil pH, and CEC together controlled the transfer of Cd from soil to wheat. Multiple linear regression models were successfully established using soil Cd contents or bioavailable Cd (extracted by CaCl(2) or calculated using a multi-surface speciation model), pH, and CEC as input variables to predict wheat Cd (RMSE = 0.242-0.327, MAE = 0.188-0.249). Furthermore, the Extreme Random Tree model (RMSE = 0.221, MAE = 0.165) outperformed the other seven machine learning algorithms. The thresholds for both soil total Cd and bioavailable Cd for safe wheat production were further back-calculated according to the permissible value of Cd in wheat grain, which demonstrated enhanced protection accuracy compared to the current soil quality standard. Our findings facilitate a quantitative assessment of Cd accumulation risk in wheat, offering a valuable reference for the safe production of wheat.