| 牟力言,刘春湘,陈敏,秦莉,林大松,Batsaikhan Bayartungalag.基于机器学习识别中微量元素与我国七大片区产地稻米镉、砷的富集规律[J].农业环境科学学报,2023,42(10):2165-2174. |
| 基于机器学习识别中微量元素与我国七大片区产地稻米镉、砷的富集规律 |
| Enrichment patterns of cadmium and arsenic in rice from seven major regions in China based on machine learning recognition of minor and trace elements |
| 投稿时间：2023-04-05 |
| DOI：10.11654/jaes.2023-0261 |
| 中文关键词: 镉 砷 决策树 随机森林 生物有效性模型 |
| 英文关键词: cadmium arsenic decision tree algorithm random forest bioavailability model |
| 基金项目:“一带一路”创新人才交流外国专家项目（DL2022051004L） |
| 作者 | 单位 | E-mail | | 牟力言 | 农业农村部环境保护科研监测所, 天津 300191
农业农村部环境保护科研监测所湘潭综合实验站, 湖南 湘潭 411100 | | | 刘春湘 | 农业农村部环境保护科研监测所, 天津 300191
农业农村部环境保护科研监测所湘潭综合实验站, 湖南 湘潭 411100 | | | 陈敏 | 农业农村部环境保护科研监测所, 天津 300191
农业农村部环境保护科研监测所湘潭综合实验站, 湖南 湘潭 411100 | | | 秦莉 | 农业农村部环境保护科研监测所, 天津 300191
农业农村部环境保护科研监测所湘潭综合实验站, 湖南 湘潭 411100 | ql-tj@163.com | | 林大松 | 农业农村部环境保护科研监测所, 天津 300191
农业农村部环境保护科研监测所湘潭综合实验站, 湖南 湘潭 411100 | | | Batsaikhan Bayartungalag | 蒙古科学院地理与生态地质研究所, 乌兰巴托 1568683 | |
|
| 摘要点击次数: 702 |
| 全文下载次数: 689 |
| 中文摘要: |
| 本研究在全国大尺度空间范围内，基于机器学习识别稻米镉（Cd）、砷（As）富集的重要影响因素，探究了中微量元素对稻米Cd、As超标的贡献率并构造了生物有效性模型。首先，通过决策树算法构造中微量元素判别Cd、As超标的预测模型，其预测精度分别为95.55%、97.55%，表明中微量元素是识别稻米Cd、As超标的重要指标；其次，利用随机森林算法筛选影响稻米Cd、As富集的主控因子，不同区域的主控因子表现出明显差异，其单一因子主要驱动的Cd富集在不同区域的差异表现为：华东片区pH的贡献占主导、华南片区的交换性钙（Ca）和东北片区的土壤有机质（SOM）分别占主要贡献，而有效铁（Fe）对As富集表现出特异性的区域贡献（如华东、华南和西南片区）；最后，将各区域确定的主控因子引入构建土壤-稻米生物有效性模型，其中，Cd、As的生物有效性九因子模型在不同片区的决定系数最高，分别为0.680、0.664 （P<0.05）。本研究为大尺度地域水平上稻米Cd、As重金属污染防控和环境管理提供了科学依据和决策支撑。 |
| 英文摘要: |
| This study identified important influencing factors of cadmium(Cd)and arsenic(As)enrichment in rice based on machine learning on a large spatial scale nationwide, explored the contribution rate of medium and trace elements to rice Cd and As exceeding standards, and constructed a bioavailability model. First, a prediction model was constructed using a decision tree algorithm to identify trace elements that exceeded Cd and As limits, with prediction accuracies of 95.55% and 97.55%, respectively. This indicates that trace elements were essential for identifying excessive Cd and As in rice. Second, the random forest algorithm was used to screen the main control factors affecting rice Cd and As enrichment, and the main control factors showed significant differences in different regions. Cd enrichment differences, mainly driven by a single factor in different regions, were as follows:the contribution of pH in East China was dominant, exchangeable calcium in South China, and soil organic matter in Northeast China accounted for the main contribution. Effective iron exhibited a specific regional contribution to As enrichment(such as in the East China, South China, and Southwest regions). The main control factors determined in each region were introduced to construct a soil rice bioavailability model; overall, the nine-factor models for Cd and As bioavailability had the highest determination coefficients in different regions, with 0.680 and 0.664(P<0.05), respectively. The models quantified the explanatory power of different factors on Cd and As enrichment patterns in rice from rice-producing areas. This study provides the scientific basis and decision-making support for preventing and controlling Cd and As heavy metal pollution in rice and environmental management at a large-scale regional level. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
|
|
|
