| 夏菲洋,和长城,陆晓松,王玉军,杨敏,范婷婷.基于机器学习对铜和锌在土壤中的老化预测和关键因子识别[J].农业环境科学学报,2024,43(11):2534-2544. |
| 基于机器学习对铜和锌在土壤中的老化预测和关键因子识别 |
| Research on machine learning-based prediction of available Cu and Zn and key factor identification during the aging process |
| 投稿时间：2024-10-08 |
| DOI：10.11654/jaes.2024-0843 |
| 中文关键词: 铜 锌 生物有效性预测 极限梯度提升（XGBoost） 动力学过程 老化 |
| 英文关键词: copper zinc bioavailability prediction extreme gradient boosting（XGBoost） kinetic equation aging process |
| 基金项目:国家重点研发计划项目(2021YFC1809102) |
| 作者 | 单位 | E-mail | | 夏菲洋 | 生态环境部南京环境科学研究所, 生态环境部土壤环境管理与污染控制重点实验室, 南京 210042 | | | 和长城 | 生态环境部南京环境科学研究所, 生态环境部土壤环境管理与污染控制重点实验室, 南京 210042 | | | 陆晓松 | 生态环境部南京环境科学研究所, 生态环境部土壤环境管理与污染控制重点实验室, 南京 210042 | | | 王玉军 | 中国科学院南京土壤研究所, 南京 211135 | | | 杨敏 | 生态环境部南京环境科学研究所, 生态环境部土壤环境管理与污染控制重点实验室, 南京 210042 | | | 范婷婷 | 生态环境部南京环境科学研究所, 生态环境部土壤环境管理与污染控制重点实验室, 南京 210042 | bluebird3602@126.com |
|
| 摘要点击次数: 382 |
| 全文下载次数: 402 |
| 中文摘要: |
| 为探究铜（Cu）和锌（Zn）在不同类型土壤中的老化过程及其主要影响因素，本研究开展了为期90 d的培养实验，向12种不同类型的土壤外源添加Cu和Zn。基于传统动力学模型、逐步线性回归和机器学习模型，构建了土壤中Cu和Zn有效态变化的预测模型。此外，基于沙普利可加性模型解释方法（Shapley Additive Explanations，SHAP），分析了影响Cu和Zn有效态含量的关键土壤因子的作用。结果表明，Cu和 Zn有效态含量在培养前 30 d内迅速下降，随后速率减缓，且 pH对老化速率影响显著，在碱性土壤中下降更明显。动力学分析显示Cu的老化过程主要受微孔扩散控制，而Zn的老化机制较为复杂，不完全依赖扩散作用。多变量逐步线性回归分析表明，土壤电导率和粒径组成对金属有效态变化有显著影响。此外，本文比较了随机森林、支持向量回归、极限梯度提升和符号回归 4种机器学习模型对 Cu和 Zn有效态含量的预测能力，发现极限梯度提升模型的预测精度最高。通过SHAP分析发现，铁氧化物和有机质含量分别是影响Cu和Zn有效态的最关键因素。pH对Cu和Zn有效态含量的影响存在显著差异，Cu的有效态含量预测中铁氧化物与pH值之间呈现出显著的交互作用。总体而言，本文通过结合动力学模型、逐步线性回归分析与机器学习方法，揭示了Cu和Zn在土壤中老化的主要驱动因素及其相互作用。 |
| 英文摘要: |
| To explore the aging process of copper（Cu）and zinc（Zn）in various soil types and their key influencing factors, this study conducted a 90-day incubation experiment with exogenous additions of Cu and Zn to 12 different soil types. Predictive models for available Cu and Zn were developed using kinetic models, stepwise linear regression, and machine learning approaches. The SHAP（Shapley Additive Explanations）method was employed to analyze the impact of key soil factors on the bioavailability of Cu and Zn. The results indicated that available Cu and Zn rapidly declined within the first 30 days, followed by a slower decrease, with pH having a significant effect on the aging rate, particularly in alkaline soils. Kinetic models revealed that the aging process of Cu was primarily controlled by micropore diffusion, while the aging process of Zn was more complex and not entirely dependent on diffusion. Stepwise linear regression analysis indicated that soil conductivity and particle size distribution significantly influenced the bioavailability of Cu and Zn. In addition, a comparison of four machine learning models[random forest, support vector regression, eXtreme gradient boosting（XGBoost）, and symbolic regression]demonstrated that the XGBoost model had the highest predictive accuracy. SHAP analysis further identified that iron oxides and organic matter content were the most critical factors affecting available Cu and Zn. The effect of pH on available Cu and Zn varied significantly, with a strong interaction between iron oxides and pH in the prediction of available Cu. Overall, this study combined kinetic models, stepwise linear regression, and machine learning methods to reveal the major driving factors and their interactions in the aging process of Cu and Zn in soils. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
|
|
|
