School of the Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China; State Key Laboratory of Biogeology and Environmental Geology, China University of Geosciences, Beijing 100083, China.
Beijing Institute of Ecological Geology, Beijing 100120, China.
Sci Total Environ. 2024 Oct 15;947:174582. doi: 10.1016/j.scitotenv.2024.174582. Epub 2024 Jul 10.
Trace elements in plants primarily derive from soils, subsequently influencing human health through the food chain. Therefore, it is essential to understand the relationship of trace elements between plants and soils. Since trace elements from soils absorbed by plants is a nonlinear process, traditional multiple linear regression (MLR) models failed to provide accurate predictions. Zinc (Zn) was chosen as the objective element in this case. Using soil geochemical data, artificial neural networks (ANN) were utilized to develop predictive models that accurately estimated Zn content within wheat grains. A total of 4036 topsoil samples and 73 paired rhizosphere soil-wheat samples were collected for the simulation study. Through Pearson correlation analysis, the total content of elements (TCEs) of Fe, Mn, Zn, and P, as well as the available content of elements (ACEs) of B, Mo, N, and Fe, were significantly correlated with the Zn bioaccumulation factor (BAF). Upon comparison, ANN models outperformed MLR models in terms of prediction accuracy. Notably, the predictive performance using ACEs as input factors was better than that using TCEs. To improve the accuracy, a two-step model was established through multiple testing. Firstly, ACEs in the soil were predicted using TCEs and properties of the rhizosphere soil as input factors. Secondly, the Zn BAF in grains was predicted using ACE as input factors. Consequently, the content of Zn in wheat grains corresponding to 4036 topsoil samples was predicted. Results showed that 85.69 % of the land was suitable for cultivating Zn-rich wheat. This finding offers a more accurate method to predict the uptake of trace elements from soils to grains, which helps to warn about abnormal levels in grains and prevent potential health risks.
植物中的微量元素主要来自土壤，随后通过食物链影响人类健康。因此，了解植物和土壤中微量元素的关系至关重要。由于植物从土壤中吸收的微量元素是非线性过程，传统的多元线性回归（MLR）模型无法提供准确的预测。在这种情况下，选择锌（Zn）作为目标元素。利用土壤地球化学数据，采用人工神经网络（ANN）建立预测模型，准确估计小麦籽粒中的 Zn 含量。共采集了 4036 个表层土壤样品和 73 对根际土壤-小麦样品进行模拟研究。通过 Pearson 相关分析，Fe、Mn、Zn 和 P 的总含量（TCEs）以及 B、Mo、N 和 Fe 的有效含量（ACEs）与 Zn 生物积累因子（BAF）显著相关。相比之下，ANN 模型在预测精度方面优于 MLR 模型。值得注意的是，使用 ACEs 作为输入因子的预测性能优于使用 TCEs。为了提高准确性，通过多次测试建立了两步模型。首先，使用 TCEs 和根际土壤特性作为输入因子预测土壤中的 ACEs。其次，使用 ACE 作为输入因子预测籽粒中的 Zn BAF。结果预测了 4036 个表层土壤样品对应的小麦籽粒中 Zn 含量。结果表明，85.69%的土地适合种植富 Zn 小麦。这一发现提供了一种更准确的方法来预测从土壤到谷物中微量元素的吸收，有助于警告谷物中异常水平，并预防潜在的健康风险。
