Predicting Cd accumulation in rice and identifying nonlinear effects of soil nutrient elements based on machine learning methods

Abstract

The spatial mismatch of Cd content in soil and rice causes difficulties in environmental management for paddy soil. To investigate the influence of soil environment on the accumulation of Cd in rice grain, we conducted a paired field sampling in the middle of the Xiangjiang River basin, examining the relationships between soil properties, soil nutrient elements, Cd content, plant uptake factor (PUFCd), and translocation factors in different rice organs (root, shoot, and grain). The total soil Cd (CdT) and available Cd (CdA) contents and PUFCd showed large spatial variability with ranges of 0.31–6.19 mg/kg, 0.03–3.07 mg/kg, and 0.02–3.51, respectively. Soil pH, CdT, CdA, and the contents of soil nutrient elements (Mg, Mn, Ca, P, Si, and B) were linearly correlated with grain Cd content (Cdg) and PUFCd. The decision tree analysis identified nonlinear effects of Si, Zn and Fe on rice Cd accumulation, which suggested that low Si and high Zn led to high Cdg, and low Si and Fe caused high PUFCd. Using the soil nutrient elements as predictor variables, random forest models successfully predicted the Cdg and PUFCd and performed better than multiple linear regressions. It suggested the impacts of soil nutrient elements on rice Cd accumulation should receive more attention.