Zinc concentration prediction in rice grain using back-propagation neural network based on soil properties and safe utilization of paddy soil: A large-scale field study in Guangxi, China.

Zn is an essential nutrient for humans, with crucial biological functions. However, Zn concentration in rice grains is generally low. Therefore, a cereal-based diet may lead to Zn deficiency in people, further leading to a series of health problems, such as immune and brain dysfunction. Previous studies seldom focused on the accumulation of Zn in rice grains based on large-scale field research. In the present study, a large-scale field survey of paddy (n = 40,853) and paired soil-rice samples (n = 1332) was conducted in Guangxi, China. Zn concentration in soil and rice grains was determined, and the associations of its spatial distributions with lithology, soil properties, and Mn nodules were investigated. According to the daily rice intake of different age and sex groups and the values of recommended Zn intake and tolerable Zn upper intake level recommended by National Health Commission of China, the Zn threshold value of the rice grain is 15.47-24.49 mg·kg. Moreover, a back-propagation neural network (BPNN) model was used to predict the Zn bioaccumulation factor (BAF) of rice grains with high accuracy. Soil Zn concentration, Mn concentration, pH, and total organic carbon derived from Pearson's correlation analysis were used as input variables in the BPNN model. Compared with the multiple linear regression model, the developed BPNN model using the training (1198 samples) and testing (134 samples) datasets showed better performance in estimating rice Zn BAF, with R = 0.93, normalized mean error of 0.009, normalized root mean square error of 0.21. When the BPNN model was applied to the 40,853 paddy soil samples, 85.7% of the agriculture lands were within the rice threshold values. These findings further our understanding of the development and utilization of Zn-rich rice and soil.