摘要
土壤是人类生存发展的基础,随着我国工业化、城市化的不断推进,土壤重金属污染的风险也逐年增加.为研究基于非采样多源异构数据进行场地土壤重金属污染模拟预测的科学性,探究不同变量对不同重金属元素污染的重要程度与影响规律,本文选取浙江省104块典型污染场地为研究对象,基于质量平衡原则构建场地土壤重金属污染风险识别指标体系,包括企业生产、大气沉降、植物富集、土壤淋溶4个大类变量和20个小类变量,利用详实的场地土壤污染调查成果,采用随机森林(RF)、特征重要性分析方法,建立预测模型并探究变量的重要性.结果表明:(1)基于场地均值的RF预测模型的决定系数均值(R_(mean)^(2)=0.75)相较于基于场地最大值的模型的决定系数(R_(mean)^(2)=0.62)较高,该模型方法对场地重金属污染均值的预测能力优于场地最大值,基于均值的预测模型精度从高到低依次为镍>汞>镉>铅>铜>砷,基于最大值的预测模型精度从高到低依次为镉>镍>砷>铅>铜>汞,基于内梅罗综合污染指数的预测模型决定系数(R^(2)=0.84)较单一重金属预测模型的平均水平较高,低于镍(R^(2)=0.92)、汞(R^(2)=0.91)的基于均值的预测模型;(2)模型对极端高值的预测能力较差,多源异构数据库对该类样本的的表征能力较弱,是导致基于最大值的模型整体精度低于基于均值的模型的主要原因;(3)以平均特征重要性作为评判依据,工业环境影响等级(15.57%)、土壤黏粒占比(9.55%)、公路密度(8.44%)、企业运行时间(7.31%)、太阳辐射强度(7.06%)、企业创立时间(6.56%)是与土壤重金属污染密切相关的变量指标.本研究旨在探索研究一种新的土壤重金属污染分析思路,即使用多源异构数据确定可能存在重金属污染风险的潜在区域,分析不同重金属元素与特征变量之间的重要性关系,为场地土壤重金属污染调查与修复提供科学和高效的手段.
Soil is considered the fundamental factor involved in the survival and development of human beings.The risk of heavy metal pollution in soil has been also increasing yearly along with the continuous advancement of industrialization and urbanization in China.In order to study the scientific perspective of heavy metal pollution prediction of site soil based on non-sampling multi-source heterogeneous data as well as exploring the importance and influence rules of different variables in different heavy metal pollution,104 typical polluted sites in Zhejiang Province have been selected as the research objects.The site soil heavy metal pollution risk identification index system has been constructed based on the principle of mass balance,including four major variables of enterprise production,atmospheric deposition,plant enrichment,and soil leaching along with twenty Secondary variables.Random Forest and Feature Importance analyzing methods have been used to establish prediction models and explore the importance of variables using detailed site soil pollution survey results.The results have shown that:①The mean value of the determination coefficient(R_(mean)^(2)=0.75)of the prediction model based on the site mean value of Random Forest has been found higher than that of the models based on the site maximum value(R_(mean)^(2)=0.62).The predicting ability of the model method towards the mean values of heavy metal pollution in sites has been found better than that of the maximum values in sites.The preciseness of the prediction models based on the mean values has been ranked as nickel,mercury,cadmium,lead,copper and arsenic from high to low,while the rank based on the maximum values is cadmium,nickel,arsenic,lead,copper and mercury from high to low.The determination coefficient of the prediction model based on the Nemerow comprehensive pollution index(R^(2)=0.84)has been found higher than the average level of the single heavy metal prediction models and lower than the mean-based prediction models of nickel(R^(2)=0.92)and mercury(R^(2)=0.91);②The model has shown poor predicting ability for extremely high values,and the multi-source heterogeneous database has shown weak representing ability for such samples,which could be the main reason why the overall accuracy of the model based on the maximum value has been found lower than that of the model based on the mean value;③Based on the average importance of the characteristics,the industrial-environmental impact grade(15.57%),the proportion of soil clay particles(9.55%),the highway density(8.44%),the enterprise operation time(7.31%),the solar radiation intensity(7.06%),and the enterprise establishment time(6.56%)have been addressed as the variable indicators closely related to soil heavy metal pollution.The purpose of this study has been set to provide a new approach to the analysis of heavy metal pollution in soil,which includes using multi-source heterogeneous data to identify potential areas where heavy metal pollution risks may exist,analyzing the important relationship between different heavy metal elements and feature variables,and providing more accurate information for the investigation of heavy metal pollution in soil and environmental protection.
作者
许洋
陈健松
王志栋
姜芳茗
张清宇
唐阔
蒋洪强
邓劲松
XU Yang;CHEN Jiansong;WANG Zhidong;JIANG Fangming;ZHANG Qingyu;TANG Kuo;JIANG Hongqiang;DENG Jinsong(College of Environmental and Resource Sciences,Zhejiang University,Hangzhou 310058;Zhejiang Ecological Civilization Academy,Anji 313300;Zhejiang Hangzhou Ecological and Environmental Monitoring Center,Hangzhou 310012;Eco-environment Low Carbon Development Center of Zhejiang Province,Hangzhou 310012;Technical Center for Soil,Agriculture and Rural Ecology and Environment,Ministry of Ecology and Environment,Beijing 100012;Chinese Academy of Environmental Planning,Beijing 100012)
出处
《环境科学学报》
CAS
CSCD
北大核心
2023年第9期357-368,共12页
Acta Scientiae Circumstantiae
基金
国家重点研发计划(No.2020YFC1807501)。
关键词
土壤重金属污染
预测模型
多源异构数据
随机森林
特征重要性
soil heavy metal pollution
forecast model
multi-source heterogeneous data
random forest
feature importance
