Prediction of maize copper content in copper-poor area based on machine learning

Abstract

As an essential micronutrient, copper (Cu) plays a crucial role in various biological functions in both plant growth and human health. Long-term consumption of a diet based on low Cu-containing grains may lead to Cu deficiency in human body, resulting in a range of health issues. The absorption of Cu by crops largely depends on bioavailable Cu rather than total Cu content in soil. The safe development of Cu-enriched grain resources is an urgent issue to be solved. Therefore, 6,980 topsoil and 109 pairs of maize-rhizosphere soil samples were collected and tested in Linshui County, Sichuan, China. The results indicated that the soil Cu content in the study area ranged from 3.33 to 173.00 mg kg⁻¹, with the average value of 25.40 mg kg⁻¹, which was significantly lower than the Cu background value of 32.00 mg kg^-1 in Sichuan Province. However, the Cu content of maize, with the average value of 1.77 mg kg⁻¹, was significantly higher than the average Cu content of Chinese maize (0.9 mg kg⁻¹). There was no significant positive correlation between Cu content in rhizosphere soil and Cu content in maize grains. Combined with geodetector and correlation analysis, the result showed that the factors in influencing the Cu bioaccumulation factor (BAF) of maize were TFe₂O₃, Mn, OM, Al₂O₃, SiO₂ and pH. The Multiple Linear Regression (MLR) and Random Forest (RF) model were used to predict the maize Cu-BAF, the RF model showed better stability and accuracy. Prediction generated by the RF model indicated that, 99.98% of the county's farmland had maize Cu content exceeding 0.9 mg kg^-1, and 6.39% of the farmland had maize Cu content exceeding 2.5 mg kg^-1. This study provides important references for scientific cultivation and holds profound implications for advancing the application of machine learning algorithms in agriculture.