| 吕陈,张乃驰,裴晨浩,申捷,喻恺,刘存,胡文友,王小治,吴同亮,王玉军.基于XRF光谱特征筛选与机器学习的土壤砷检测与风险评估[J].农业环境科学学报,2025,44(11):2796-2805. |
| 基于XRF光谱特征筛选与机器学习的土壤砷检测与风险评估 |
| Rapid detection and risk assessment of soil arsenic based on feature wavelength selection and modeling of XRF spectroscopy |
| 投稿时间:2025-04-09 |
| DOI:10.11654/jaes.2025-0340 |
| 中文关键词: 土壤砷 X射线荧光光谱 竞争性自适应重加权算法 偏最小二乘回归模型 土壤污染风险评估 |
| 英文关键词: soil arsenic X-ray fluorescence spectroscopy competitive adaptive reweighted sampling partial least squares regression soil health risk assessment |
| 基金项目:国家重点研发计划项目(2021YFC189100);国家自然科学基金项目(41977027,42177015) |
| 作者 | 单位 | E-mail | | 吕陈 | 扬州大学环境科学与工程学院, 江苏 扬州 225127 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 | | | 张乃驰 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 中国科学院大学, 北京 100049 | | | 裴晨浩 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 中国科学院大学, 北京 100049 | | | 申捷 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 南昌航空大学环境与化学工程学院, 南昌 330063 | | | 喻恺 | 南昌航空大学环境与化学工程学院, 南昌 330063 | | | 刘存 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 | | | 胡文友 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 | | | 王小治 | 扬州大学环境科学与工程学院, 江苏 扬州 225127 | xzwang@yzu.edu.cn | | 吴同亮 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 | | | 王玉军 | 土壤与农业可持续发展全国重点实验室, 中国科学院南京土壤研究所, 南京 211135 中国科学院大学, 北京 100049 | yjwang@issas.ac.cn |
|
| 摘要点击次数: 1139 |
| 全文下载次数: 775 |
| 中文摘要: |
| 便携式X射线荧光光谱(XRF)可应用于快速、经济的土壤重金属分析,但受限于基体效应和谱线干扰,仍存在检测限偏高、准确度不足等瓶颈问题。本研究提出一种基于特征波长建模的XRF光谱技术,以土壤As为研究对象,通过融合XRF光谱解析技术与机器学习算法,构建高精度As浓度反演模型,并评估其在区域土壤快速健康风险评价中的适用性。基于35份国家标准土壤样本,采用Savitzky-Golay(SG)滤波与自适应迭代惩罚最小二乘法(airPLS)相结合,可将XRF对As的最低检测限(LOD)降低至1.7 mg·kg-1。进一步结合相关能谱选择(CSS)与竞争性自适应重加权采样(CARS)进行特征波长筛选,并分别构建偏最小二乘回归(PLSR)、随机森林(RF)和极端梯度提升(XGBoost)3种机器学习反演模型。模型对比结果表明:基于特征筛选的PLSR模型(SG-CSS-CARS-PLSR)表现最优,在测试集上R2=0.914,RMSE=14.743;在浙江绍兴某典型矿区土壤As浓度的外部验证中亦展现出优异性能(R2=0.925,RMSE=5.984),显著优于XRF仪器内置的基本参数法,整体预测精度提升达36.8%。将最优模型预测的结果应用于该矿区土壤健康风险评价分析,发现研究区域内As对成年人和儿童的健康风险指数平均值分别为9.76×10-2和2.28×10-1,致癌风险平均值分别为4.43×10-4和1.03×10-3,与常规的基于三酸消解-AFS法的评价结果一致(P>0.05),且分析周期缩短约65%,检测成本降低约60%,证实该方法可以用于区域土壤重金属的快速检测与健康风险评估。 |
| 英文摘要: |
| X-ray fluorescence spectroscopy(XRF)offers a rapid and cost-effective approach for soil heavy metal analyses. However, its accuracy for low-concentration heavy metal quantification is significantly compromised by soil matrix effects and spectral interference. This study proposes a novel XRF spectral technique for soil arsenic(As)analysis and corresponding health risk assessment, by integrating XRF spectral refinement technology with machine learning algorithms. Based on the analyses of 35 national standard soil samples, the combined Savitzky-Golay filtering and Adaptive Iterative Re-weighted Penalized Least Squares methods(SG-airPLS)reduce XRF’s limit of detection(LOD)for As to 1.7 mg·kg-1. The method were further refined by combining Correlated Spectral Selection and Competitive Adaptive Reweighted Sampling(CSS-CARS)with three different machine learning models, including partial least squares regression (PLSR), random forest(RF), and extreme gradient boosting(XGBoost)algorithms. The comparison results show that SG-CSS-CARSPLSR model exhibited optimal performance for soil As quantification, achieving an R2 of 0.914 and RMSE of 14.743 on the test set. In external validation using soil samples from a typical mining area in Shaoxing, Zhejiang Province, the model showed superior accuracy(R2= 0.925, RMSE=5.984), outperforming the XRF instrument's built-in fundamental parameters(FP)algorithm(R2=0.625, RMSE=9.470), with an overall prediction accuracy improvement of 36.8%.The health risk assessment for the mining area were conducted by the SG-CSSCARS-PLSR model compared to the traditional tri-acid digestion AFS method. The results revealed average health risk indices for adults and children as 9.76×10-2 and 2.28×10-1, respectively, and average carcinogenic risks of 4.43×10-4 and 1.03×10-3, indicating potential Asrelated health risks in the study area, which were not statistically different from tri-acid digestion AFS method(P>0.05), while reducing analysis time by 65% and costs by 60%. This approach demonstrates strong practical applicability and sustainability advantages for regional soil contamination monitoring and risk management. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
|
|
|