• 集群首页
  • 关于我们
  • 期刊群

基于机器学习预测水稻籽实砷含量及土地资源安全利用

Safe Utilization of Land Resources Based on Machine Learning to Predict the As Content in Rice Grains

  • 摘要: 砷(As)是一种高致癌风险的类金属,过量摄入会对人体健康造成严重危害。As含量超标的水稻摄入是人体As暴露的主要途径。由于水稻吸收As影响因素复杂,使得仅依据土壤As含量进行的土地资源分类管理会出现问题,没有起到保障水稻安全生产和人体健康的目的。本文选择广东省紫金县中部及北部为研究区,对65套水稻籽实无机砷和根系土As、pH、TFe2O3等地球化学指标进行了系统研究。结果表明,4.6%根系土As含量超过了《土壤环境质量 农用地土壤污染风险管控标准(试行)》(GB 15618—2018)筛选值(30mg/kg),水稻无机砷含量超过《食品安全国家标准食品中污染物限量》(GB 2762—2022)规定的限量值(0.35mg/kg)比例高达23%,且水稻籽实无机砷含量与根系土As并没有正相关性,在优先保护区中,水稻无机砷超标率为19%,而在严格管控区中,水稻无机砷超标率为0。进一步研究发现,水稻籽实无机砷的生物富集系数影响因素有TFe2O3、Mn和SiO2等地球化学指标,对比研究了随机森林(RF)模型、人工神经网络(ANN)模型和多元回归(MLR)模型对水稻籽实As的生物富集系数(BCFAs)的预测效果,结果显示随机森林模型具有更强的稳定性和准确性。利用1∶25万面积性调查数据和随机森林模型,对水稻籽实BCFAs进行了预测,并计算得出水稻籽实无机砷含量。据此,提出了紫金县水稻种植区划分和土地资源安全利用方案,实现水稻安全种植。

     

    Abstract: Arsenic (As) is a metalloid with high carcinogenic risk, and excessive intake can cause severe harm to human health. The consumption of rice with excessive As content is a primary pathway for human As exposure. Due to the complex factors influencing As uptake in rice, classification management of land resources based solely on soil As content is problematic and fails to ensure safe rice production and protect human health. This study selected central and northern Zijin County in Guangdong Province as the research area, conducting a systematic investigation of 65 sets of inorganic As in rice grains and geochemical indicators such as As, pH, and TFe2O3 in rhizosphere soil. The results indicated that 4.6% of the rhizosphere soil samples exceeded the screening value (30mg/kg) specified in the Soil Environmental Quality: Risk Control Standard for Soil Contamination of Agricultural Land (GB 15618−2018). The proportion of rice samples exceeding the limit for inorganic As (0.35mg/kg) as stipulated by GB 2762−2022 was as high as 23%. Furthermore, no positive correlation was observed between inorganic As content in rice grains and As content in rhizosphere soil. In priority protection zones, the exceedance rate for inorganic As in rice was 19%, while in strictly controlled zones, it was 0. Further research revealed that geochemical indicators such as TFe2O3, Mn, and SiO2 influence the bioaccumulation factor (BCF) of inorganic As in rice grains. A comparative study of the predictive performance of the random forest (RF) model, artificial neural network (ANN) model, and multiple linear regression (MLR) model for the BCFAs in rice grains showed that the RF model exhibited stronger stability and accuracy. Utilizing 1∶250000 regional survey data and the RF model, the BCFAs of rice grains was predicted, and the inorganic As content in rice grains was calculated. Based on these findings, a zoning plan for rice cultivation areas and a strategy for the safe utilization of land resources in Zijin County were proposed to achieve safe rice cultivation.

     

/

返回文章
返回