摘要
为提高工业集聚区土壤污染刻画模型的精度,以河北省某在产工业聚集区土壤重金属作为研究对象,对比研究距离权重法(IDW)、普通克里金(OK)、支持向量机(SVM)和梯度提升决策树(GBDT)等不同插值方法在处理浓度非平稳、有偏数据的刻画精度问题。结果显示:该场地土壤中主要污染物为As,GBDT在刻画土壤As污染时表现出最高的精度(
随着我国工业化进程的高速发展,工业聚集区已成为我国支撑制造行业发展的主要载体,但以产业集聚为特征的工业模式给区域环境保护带来巨大挑
精准刻画污染物空间分布特征是准确认识场地土壤污染状况的关键。由于工业聚集区场地具有高污染异质、多污染类型、多过程耦合的特点,重金属在表生环境的迁移累积研究仍然存在一些困
本研究以河北省某在产城市工业聚集区土壤为研究对象,利用有偏钻孔数据,识别场地土壤的污染因子,分别采用反距离权重法(inverse distance weighting,IDW)、普通克里金(OK)、支持向量机(support vector machine,SVM)和梯度提升决策树(gradient boosting decision tree,GBDT)等4种不同的空间插值方法构建非平稳耦合模型,分析各模型刻画结果的精度,并通过蒙特卡罗模拟方法对工业聚集区重金属污染土壤进行健康风险评估,旨在为场地污染土壤的精准刻画提供新思路以及为在产工业聚集区的污染诊断和科学防控提供高效的技术工具。
研究区位于河北省无极县境内(38°13′53.30″N,114°58′16.81″E),占地面积0.91 k

图1 研究区采样点位(A)和土壤岩性(B)
Fig.1 Locations of sample sites(A), soil layering profile(B) in the study area
该工业聚集区于2006年开始建立并运营,目前现存企业21家,主要以原料化工、材料制造、机械设备、纺织印染企业为主。结合实际企业走访调查及生产工艺过程分析,主要潜在污染因子包括As、Pb、Cd、Ni、Cu和Hg。
为了明确工业活动对土壤重金属累积的影响,结合工业聚集区内企业分布情况以及HJ 25.1—2019《建设用地土壤污染状况调查技术导则》进行点位布设。2022年2月共布设172个土壤采样点位。结合土壤岩性和HJ/T 166—2004《土壤环境监测技术规范》要求,采集0~8 m土壤样品:杂填层采样间隔0.5 m,砂土层和粉质粘土层采样间隔2.0 m(
对采集到的土壤样品进行风干、粉碎、筛分(孔径2 mm)和强酸消解,地下水样品采用0.45 μm滤膜过滤后,进行重金属含量分
根据HJ/T 166—2004《土壤环境监测技术规范》,采用土壤污染指数(pollution index,PI;公式中以IP表示)对供试样品中的单一重金属污染现状进行评价;采用Nemerow综合污染指数(comprehensive pollution index,NPI;公式中以INP表示)对土壤中的多种重金属复合污染现状进行评价:
(1) |
(2) |
式(
反距离权重法(IDW)是一种加权平均方法,以插值点与相邻采样点之间的距离为权
(3) |
(4) |
(5) |
式(
普通克里金(OK)模型是基于变异函数理论和结构分析,对有限区域内的区域化变量进行无偏最优估
支持向量机(SVM)是通过在N维空间中找到最佳超平面
(6) |
(7) |
(8) |
(9) |
其中,f(x)为模型输出值,w为决策面的法向量,决定了决策面的方向;b决定了决策面的位置。xi∈
梯度提升决策树(GBDT)是一种集成机器学习算法,通过最小化误差梯度来拟合提升决策树。GBDT模型是将许多弱学习器(单个决策树)组合在一起,得出一个强学习
(10) |
(11) |
(12) |
(13) |
(14) |
式(
机器学习模型预测中使用的环境变量有:采样点位经纬度、高程、土壤剖面深度、土壤岩性、黏粒、粉粒、砂粒、砾石含量、土壤含水量、pH、污染物含量和功能区类型。其中,土壤点位信息、土壤理化性质和重金属含量通过实测数据获得,功能区类型根据现场场地调查结合企业工业类型划分。功能区类型和土壤岩性使用特征编码进行数据预处理。
本研究以均方根误差(RMSE, 公式中以ERMS表示)、平均绝对误差(MAE, 公式中以EMA表示)、平均相对误差(MRE, 公式中以EMR表示)和相对分析误差(RPD,公式中以RPD表示)作为评价指标,评价不同模型的精
(15) |
(16) |
(17) |
(18) |
(19) |
由于本研究区域周围有村庄以及住宅区,环境敏感受体需同时考虑成人和儿童。土壤重金属考虑的暴露风险途径包括经口摄入、皮肤接触和吸入土壤颗粒物。本研究地块土壤暴露评估参照HJ 25.3—2019《建设用地土壤污染风险评估技术导则》中第二类用地暴露评估模型计算土壤暴露量,计算公式以及涉及的蒙特卡罗概率风险评估的取值参考文献[
参数 Parameter | 含义 Description | 单位 Unit | 类型 Type | 取值 Values | |
---|---|---|---|---|---|
成人 Adults | 儿童 Children | ||||
BW | 平均体质量 Average body weight | kg | Lognormal | LN(67.48,12.60) | LN(16.68,1.48) |
EF | 暴露频率 Exposure frequency | d/a | Point | 93.75 | 350 |
ED | 暴露期 Exposure duration | a | Normal | N(25,8.25) | N(6,1.25) |
IRoral | 每日经口摄入土壤量 Ingestion rate | mg/d | Triangular | TRI(4,30,52) | TRI(66,103,161) |
InhR | 每日空气呼吸量 Inhalation rate |
| Lognormal | LN(16.30,2.81) | LN(7.5,1.5) |
SAS | 暴露皮肤表面积 Exposed skin area |
c | Lognormal | LN(5 427,579) | LN(1 592,141) |
致癌风险超过1.00×1
(20) |
(21) |
(22) |
(23) |
式(
本研究运用蒙特卡罗模拟法的不确定性分析方法中的概率分析,将污染物浓度、暴露受体参数、人体参数作为不确定性因素,确定其概率分布类型,运用蒙特卡罗方法随机模拟各敏感参数,计算出土壤污染物对健康风险的概率分布。蒙特卡罗计算模拟进行10 000次迭代,敏感性分析用于各输入参数对健康风险结果的贡献情况。
重金属 Heavy metal | 最小值 Min | 最大值 Max | 中位数 Median | 平均值 Mean | 标准差 SD | 筛选值 Screening values | 背景值Background values |
---|---|---|---|---|---|---|---|
Cu | 2.00 | 768.00 | 14.00 | 20.20 | 35.54 | 18 000 | 32.20 |
Ni | 4.00 | 85.00 | 26.50 | 26.96 | 6.54 | 900 | 50.10 |
Pb | 10.00 | 1 350.00 | 16.00 | 19.49 | 40.35 | 800 | 34.50 |
Cd | 0.01 | 35.10 | 0.06 | 0.16 | 0.97 | 65 | 0.68 |
As | 0.90 | 1 040.00 | 2.38 | 20.50 | 82.20 | 60 | 15.20 |
Hg | ND | 0.80 | 0.02 | 0.028 | 0.061 | 38 | 0.10 |
注Note:ND:未达到检出限None detected.
采用单因子污染指数PI和Nemerow污染指数NPI进行土壤中污染物评价。参照土壤污染等级划分标准,土壤平均PI值依次为Hg(0.003 5)<Cd(0.008 1)<Cu(0.010)<Pb(0.049)<V(0.15)<Ni(0.18)<1<As(1.03),仅有As处于轻微污染,其余均为未污染。在所有样本中,有7.73%As处于污染水平(PI>2),其中As最高PI值可达52.00,表明本场地土壤中As污染程度最重。本场地土壤重金属NPI的范围为0.053~37.150,其中,5.23%的土壤为重度污染,1.70%的土壤为中度污染,1.70%的土壤为低度污染。

图2 土壤中污染物的污染指标
Fig.2 The pollution indexes of contaminants in soil
为了克服土壤As浓度非平稳分布对平稳假设的影响,以及钻孔数据有偏造成的刻画结果精度低的难题,本研究在As浓度数据基础上,通过耦合水文地质参数、重金属含量等多源辅助环境数据来构建As污染可视化模型,以期提高模型模拟场地土壤As空间分布的刻画精度。首先使用Pearson相关性分析方法,将土壤As浓度与多源辅助数据进行关联,厘清各参数对As在土壤中的迁移扩散和局部富集的重要性,从而为解耦空间非平稳关系以及有偏数据权重调整建立基础。

图3 重金属与土壤基本理化性质的Pearson相关性分析
Fig.3 Correlation hot map between heavy metals and soil parameters
使用4种不同的插值方法(IDW、OK、SVM和GBDT)对场地As污染进行空间刻画分析。根据RMSE、MAE和MRE的统计结果,GBDT模型的RMSE、MAE和MRE均为最低,其次是SVM、OK和IDW模型(

图4 不同模型的散点图拟合
Fig.4 Scatterplots of different models fitting
项目 Item | IDW | OK | VSM | GBDT |
---|---|---|---|---|
| 0.284 7 | 0.412 2 | 0.773 7 | 0.911 5 |
RMSE | 73.53 | 64.25 | 39.10 | 25.32 |
MAE | 22.61 | 15.41 | 13.54 | 10.19 |
MRE | 3.84 | 3.01 | 2.56 | 2.23 |
RPD | 1.18 | 1.30 | 2.10 | 3.36 |
为了更好地反映As污染的分布情况,采用GBRT对土壤As污染进行空间可视化(

图5 使用GBRT模型刻画不同深度土壤中As的空间分布
Fig.5 Spatial distribution of As in soil at different depths using GBRT modeling
本研究分别使用均值参数的确定性评估方法与蒙特卡罗模拟方法对研究区域进行健康风险评估。不确定性分析中通常认为风险概率分布90%~99.9%分位值为合理范围内的最大暴露,采用95%分位值作为场地最终风险值,具体评估结果见
项目 Item | 成人 Adults | 儿童 Children | |||||
---|---|---|---|---|---|---|---|
确定性评估方法(均值参数)Deterministic assessment method(Mean parameter) | 蒙特卡罗模拟方法(95%分位值)Monte Carlo simulation(95th percentile) | 标准偏差 SD | 确定性评估方法(均值参数)Deterministic assessment method(Mean parameter) | 蒙特卡罗模拟方法(95%分位值)Monte Carlo simulation(95th percentile) | 标准偏差 SD | ||
HQ | Cu |
7.53×1 |
2.70×1 |
1.41×1 |
4.07×1 |
1.50×1 |
7.54×1 |
Ni |
7.37×1 |
1.23×1 |
2.54×1 |
1.92×1 |
3.06×1 |
5.83×1 | |
Pb |
1.56×1 |
5.78×1 |
3.32×1 |
3.44×1 |
1.36×1 |
7.90×1 | |
Cd |
9.27×1 |
1.06×1 |
6.20×1 |
1.53×1 |
5.77×1 |
9.20×1 | |
As |
2.85×1 |
1.14×1 |
1.38×1 |
4.37×1 | 1.79 | 3.37 | |
Hg |
2.86×1 |
1.09×1 |
6.55×1 |
9.80×1 |
3.76×1 |
2.27×1 | |
HI |
4.77×1 |
1.37×1 |
1.19×1 |
5.11×1 | 1.89 | 1.72 | |
CR | Ni |
3.34×1 |
5.78×1 |
1.25×1 |
5.57×1 |
9.22×1 |
1.95×1 |
Pb |
8.79×1 |
3.23×1 |
1.87×1 |
7.71×1 |
3.06×1 |
1.78×1 | |
Cd |
2.19×1 |
2.48×1 |
1.46×1 |
5.17×1 |
1.04×1 |
3.38×1 | |
As |
4.15×1 |
1.66×1 |
2.00×1 |
1.53×1 |
6.26×1 |
1.18×1 | |
TCR |
4.38×1 |
1.69×1 |
1.66×1 |
1.59×1 |
6.37×1 |
5.99×1 |
注: Note:HQ:非致癌风险危险商指数Non carcinogenic risk hazard quotient index;HI:总非致癌风险危险商指数Total non carcinogenic risk hazard quotient index;CR:致癌健康风险指数Carcinogenic health risk index;TCR:总致癌风险指数Total carcinogenic risk index.

图6 土壤总非致癌和总致癌的风险概率分布以及各参数敏感度结果
Fig.6 Probability distribution for hazard index and total carcinogenic risk in soil and sensitivity results for various parameters
由于地下环境的复杂性,尽管部分研究改善了污染物浓度的空间非平稳性和钻井数据的偏差校正,但影响土壤污染深层迁移行为和因素的贡献仍不明确。本研究结果发现,As的深层迁移受土壤理化性质以及水文地质条件影响。Pearson相关性结果表明(
根据模拟结果可知,不同的插值方法得到的刻画精度差异显著。OK和IDW难以建立污染物含量与环境因素之间的空间相关性,忽略了污染物迁移扩散和水文地质条件带来的非平稳问题。因此,计算得到的精度很
关于重金属污染土壤环境风险评估方面,目前国内外普遍使用的导则都是基于输入参数的确定性值,易受数据波动及其他不确定性参数的影响,造成场地健康风险评价结果存在极大的差异。因此,使用均值参数的确定性风险评估方法得到健康风险评价值往往难以反映区域的整体情况。然而,蒙特卡罗模拟可通过使用随机参数值进行迭代计算,降低了由参数选取、数据波动带来的不确定性,并通过参数因子敏感度判断健康风险的主控因素,对人体健康风险进行更准确的评
参考文献 References
SU Y,YU Y Q.Spatial agglomeration of new energy industries on the performance of regional pollution control through spatial econometric analysis[J/OL].Science of the total environment,2020,704:135261[2023-10-25].https://doi.org/10.1016/j.scitotenv.2019.135261. [百度学术]
PODGORSKI J,BERG M.Global threat of arsenic in groundwater[J].Science,2020,368(6493):845-850. [百度学术]
ZHU L,ZHANG X X,ZHANG J C,et al.Saltwater intrusion weakens Fe-(oxyhydr)oxide-mediated (im)mobilization of Ni and Zn in redox-fluctuating soil-groundwater system[J/OL].Water research,2022,221:118799[2023-10-25].https://doi.org/10.1016/j.watres.2022.118799. [百度学术]
HOU S Y,ZHAO X,LIU Y,et al.Spatial analysis connects excess water pollution discharge,industrial production,and consumption at the sectoral level[J/OL].NPJ clean water,2022,5:4[2023-10-25].https://doi.org/10.1038/s41545-022-00152-7. [百度学术]
RAN H Z,GUO Z H,YI L W,et al.Pollution characteristics and source identification of soil metal(loid)s at an abandoned arsenic-containing mine,China[J/OL].Journal of hazardous materials,2021,413:125382[2023-10-25].https://doi.org/10.1016/j.jhazmat.2021.125382. [百度学术]
ZENG J Q,LI C X,WANG J T,et al.Pollution simulation and remediation strategy of a zinc smelting site based on multi-source information[J/OL].Journal of hazardous materials,2022,433:128774[2023-10-25].https://doi.org/10.1016/j.jhazmat.2022.128774. [百度学术]
范俊楠,贺小敏,杜维,等.基于标准化方法和统计分析法的土壤重金属环境基线值分析比较[J].华中农业大学学报,2021,40(1):160-167.FAN J L,HE X M,DU W,et al. Analyzing and comparing environmental baseline values of heavy metals in soil based on standardized method and statistical method[J]. Journal of Huazhong Agricultural University, 2021,40 (1): 160-167 (in Chinese with English abstract). [百度学术]
陶欢,廖晓勇,曹红英,等.场地土壤污染物含量三维刻画的研究进展[J].地理学报,2022,77(3):559-573.TAO H,LIAO X Y,CAO H Y,et al.Research progress of three-dimensional delineation of soil pollutants at contaminated sites[J].Acta geographica sinica,2022,77(3):559-573 (in Chinese with English abstract). [百度学术]
LIU G,ZHOU X,LI Q,et al.Spatial distribution prediction of soil As in a large-scale arsenic slag contaminated site based on an integrated model and multi-source environmental data[J/OL].Environmental pollution,2020,267:115631[2023-10-25].https://doi.org/10.1016/j.envpol.2020.115631. [百度学术]
GUO Z H,ZHANG Y X,XU R,et al.Contamination vertical distribution and key factors identification of metal(loid)s in site soil from an abandoned Pb/Zn smelter using machine learning[J/OL].Science of the total environment,2023,856:159264[2023-10-25].https://doi.org/10.1016/j.scitotenv.2022.159264. [百度学术]
ZHAI Y Z,JIANG Y,CAO X Y,et al.Valuation of ecosystem damage induced by soil-groundwater pollution in an arid climate area:framework,method and case study[J/OL].Environmental research,2022,211:113013[2023-10-25].https://doi.org/10.1016/j.envres.2022.113013. [百度学术]
QIAO P W,LI P Z,CHENG Y J,et al.Comparison of common spatial interpolation methods for analyzing pollutant spatial distributions at contaminated sites[J].Environmental geochemistry and health,2019,41(6):2709-2730. [百度学术]
ZHAO B,ZHU W X,HAO S F,et al.Prediction heavy metals accumulation risk in rice using machine learning and mapping pollution risk[J/OL].Journal of hazardous materials,2023,448:130879[2023-10-25].https://doi.org/10.1016/j.jhazmat.2023.130879. [百度学术]
PENG Y X,CHEN J,XIE E Z,et al.Three-dimensional spatial prediction of Zn in the soil of a former tire manufacturing plant using machine learning and readily attainable multisource auxiliary data[J/OL].Environmental pollution,2023,318:120931[2023-10-25].https://doi.org/10.1016/j.envpol.2023.120931. [百度学术]
ZHOU Y,JIANG D D,DING D,et al.Ecological-health risks assessment and source apportionment of heavy metals in agricultural soils around a super-sized lead-zinc smelter with a long production history,in China[J/OL].Environmental pollution,2022,307:119487[2023-10-25].https://doi.org/10.1016/j.envpol.2022.119487. [百度学术]
SUN Y,LIANG X J,XIAO C L.Assessing the influence of land use on groundwater pollution based on coefficient of variation weight method:a case study of Shuangliao City[J].Environmental science and pollution research international,2019,26(34):34964-34976. [百度学术]
ZHANG H,YIN A J,YANG X H,et al.Use of machine-learning and receptor models for prediction and source apportionment of heavy metals in coastal reclaimed soils[J/OL].Ecological indicators,2021,122:107233[2023-10-25].https://doi.org/10.1016/j.ecolind.2020.107233. [百度学术]
HUANG B,YUAN Z J,LI D Q,et al.Effects of soil particle size on the adsorption,distribution,and migration behaviors of heavy metal(loid)s in soil:a review[J].Environmental science processes & impacts,2020,22(8):1596-1615. [百度学术]
CHEN R H,TENG Y G,CHEN H Y,et al.A coupled optimization of groundwater remediation alternatives screening under health risk assessment:an application to a petroleum-contaminated site in a typical cold industrial region in Northeastern China[J/OL].Journal of hazardous materials,2021,407:124796[2023-10-25].https://doi.org/10.1016/j.jhazmat.2021.124796. [百度学术]
GUO T,ZHOU Y J,CHEN S C,et al.The influence of periphyton on the migration and transformation of arsenic in the paddy soil:rules and mechanisms[J/OL].Environmental pollution,2020,263:114624[2023-10-25].https://doi.org/10.1016/j.envpol.2020.114624. [百度学术]
ZONG Y T,XIAO Q,LU S G.Distribution,bioavailability,and leachability of heavy metals in soil particle size fractions of urban soils (northeastern China)[J].Environmental science and pollution research international,2016,23(14):14600-14607. [百度学术]
O'CONNOR D,PAN S Z,SHEN Z T,et al.Microplastics undergo accelerated vertical migration in sand soil due to small size and wet-dry cycles[J].Environmental pollution,2019,249:527-534. [百度学术]
WANG J X,ZHANG W J.Evaluating the adsorption of Shanghai silty clay to Cd(Ⅱ),Pb(Ⅱ),As(Ⅴ),and Cr(Ⅵ):kinetic,equilibrium,and thermodynamic studies[J/OL].Environmental monitoring and assessment,2021,193(3):131[2023-10-25].https://doi.org/10.1007/s10661-021-08904-7. [百度学术]
XUE Q,RAN Y,TAN Y Z,et al.Arsenite and arsenate binding to ferrihydrite organo-mineral coprecipitate:implications for arsenic mobility and fate in natural environments[J].Chemosphere,2019,224:103-110. [百度学术]
YANG X,LIU S Y,LIANG T,et al.Ball-milled magnetite for efficient arsenic decontamination:insights into oxidation-adsorption mechanism[J/OL].Journal of hazardous materials,2022,427:128117[2023-10-25].https://doi.org/10.1016/j.jhazmat.2022. 128117. [百度学术]
GAYE B,ZHANG D Z,WULAMU A.Improvement of support vector machine algorithm in big data background[J/OL].Mathematical problems in engineering,2021,2021:1-9[2023-10-25].https://doi.org/10.1155/2021/5594899. [百度学术]
ZHENG J T,WANG P,SHI H Y,et al.Quantitative source apportionment and driver identification of soil heavy metals using advanced machine learning techniques[J/OL].Science of the total environment,2023,873:162371[2023-10-25].https://doi.org/10.1016/j.scitotenv.2023. 162371. [百度学术]
MUKHERJEE I,SINGH U K,SINGH R P,et al.Characterization of heavy metal pollution in an anthropogenically and geologically influenced semi-arid region of East India and assessment of ecological and human health risks[J/OL].Science of the total environment,2020,705:135801[2023-10-25].https://doi.org/10.1016/j.scitotenv.2020. 135801. [百度学术]
CHEN H X,XU F N,CHEN Z Z,et al.Arsenic and cadmium removal from water by a calcium-modified and starch-stabilized ferromanganese binary oxide[J].Journal of environmental sciences,2020,96:186-193. [百度学术]