基于随机森林模型识别浅层地下水TDS异常的方法研究

doi:10.13745/j.esf.sf.2024.2.20

摘要/Abstract

摘要：

准确识别人类活动引起的地下水水化学异常对于确定地下水水化学组分的背景值,合理开展地下水污染评价至关重要。溶解性总固体(TDS)作为地下水水化学的综合指标,其值的高低直接反映了地下水水质的好坏。目前,水化学图法在地下水TDS的异常值识别中取得了较好的效果,但是,其基本原理是基于主要离子组分构成的水化学类型异常必然导致TDS异常的假设,而进行的反向异常识别,可能存在过度识别的情况。为此,本文以沙颍河流域浅层地下水为研究对象,从TDS成因机制出发,提出了采用随机森林模型结合数理统计的正向识别方法,对研究区内浅层地下水TDS的异常值进行识别,并开展了多种方法异常值识别效果的对比研究。结果表明,机器学习法能够有效地识别出地下水TDS异常值,其识别出的地下水TDS阈值与其他方法较为一致。但相比之下,机器学习法从TDS成因机制角度识别异常,能够有效避免水化学图存在的过度识别问题,而且能够区分高、低异常,为TDS异常识别提供了另外一种有效的思路和方法,丰富了地下水环境背景值的研究思路。

关键词: 地下水环境背景值, TDS, 异常值, 机器学习法, 沙颍河流域

Abstract:

Accurately identifying groundwater hydrochemical outliers caused by human activities is crucial for determining the nature background levels of groundwater chemical components and conducting rational assessment of groundwater pollution. The total dissolved solids (TDS), serving as a comprehensive indicator of groundwater hydrochemistry, its value directly reflect the quality of groundwater. Currently, the hydrochemical diagrams method has achieved favorable results in identifying outliers of TDS in groundwater. However, its fundamental principle is reverse identification based on the assumption that the hydrochemical type anomalies composed of the major ion components inevitably result in TDS anomalies, which will potentially leading to over-recognition during the anomaly identification. Therefore, the random forest model, commencing with a positive identification of the genesis mechanisms of TDS, combined with statistical method was employed to identify anomalies in TDS of shallow groundwater based in the Shaying River Basin. A comparative analysis of anomaly identification effectiveness among various methods was conducted, whose results demonstrated that the machine learning method could effectively identify the outliers of TDS in groundwater, with the identified TDS thresholds aligning with those derived from alternative methods. In contrast, the machine learning method, grounded in TDS genesis mechanisms, effectively identified anomalies by mitigating errors in hydrochemical diagrams. This approach successfully distinguished between high and low outliers, offering an alternative and efficacious method for TDS anomaly identification and expanding research perspectives on groundwater environmental background values.

Key words: groundwater environmental background values, TDS, outliers, machine learning method, the Shaying River Basin

中图分类号:

P641
X523

褚宴佳, 何宝南, 陈珍, 何江涛. 基于随机森林模型识别浅层地下水TDS异常的方法研究[J]. 地学前缘, 2025, 32(2): 456-468.

CHU Yanjia, HE Baonan, CHEN Zhen, HE Jiangtao. Research on identifying the outliers of the TDS in shallow groundwater based on the random forest model[J]. Earth Science Frontiers, 2025, 32(2): 456-468.

图/表 9

图1 机器学习法识别TDS异常值工作流程

Fig.1 Workflow of machine learning method for identifying TDS outlier

图2 沙颍河流域地理位置和采样点分布

Fig.2 Geographical location and sampling sites distribution of the Shaying River Basin

图3 研究区TDS浓度及其潜在影响因素空间分布图 a—研究区TDS浓度空间分布图和散点图;b—研究区含水层渗透系数空间分布图;c—研究区地下水埋深空间分布图;d—研究区水力梯度空间分布图;e—研究区年平均降雨量空间分布图;f—研究区年平均蒸散量空间分布图。

Fig.3 The spatial distribution and descriptive statistical of TDS and the 5 potential influencing factors

图4 基于机器学习法识别TDS异常值的结果 a—随机森林模型对地下水TDS的预测结果;b—TDS预测值与真实值的散点图;c—TDS预测值的常规误差与TDS真实值的散点图;d—TDS预测值的常规误差累计频率曲线图。

Fig.4 Results of the TDS outlier identification based on machine learning method

表1 TDS异常值识别结果统计

Table 1 Statistical results of the TDS outlier identification

TDS异常值识别方法	异常个数/ 个	最小值/ (mg·L^-1)	最大值/ (mg·L^-1)	平均数/ (mg·L^-1)	中位数/ (mg·L^-1)	5%分位数/ (mg·L^-1)	95%分位数/ (mg·L^-1)
剔除前		217.05	2 746.61	727.87	624.58	309.07	1 470.30
拉依达准则法剔除后	17	217.05	1 522.68	681.15	612.65	303.20	1 269.56
Piper三线图+拉依达准则法剔除后	125	217.05	1 522.68	671.35	609.62	285.51	1 263.27
机器学习法剔除后	61	217.05	1 642.08	653.94	598.03	309.07	1 169.69

图5 TDS浓度分布拐点图和箱型图 a—3种方法异常值剔除前后数据拐点图;b—3种方法异常值剔除前后数据箱型图。

Fig.5 Inflection point plot and box-plot of the TDS concentration distribution

图6 基于机器学习法的异常识别结果数据统计和空间分布图 a—高异常点空间分布图;b—低异常点空间分布图;c—正常点数据分布直方图;d—高异常点数据分布直方图; e—低异常点数据分布直方图; f—正常值和高、低异常点数据分布箱型图。

Fig.6 The descriptive statistical and spatial distribution of the TDS outlier identification results of machine learning method

图7 高、低异常点成因分析和人为污染验证结果 a—NO3-N浓度分布箱型图;b—Cl-浓度分布箱型图;c—基于污染指数的人为污染程度量化结果。

Fig.7 Analysis of the genesis of high and low outliers and the verification results of artificial pollution

图8 地下水TDS高、低异常点的水化学特征

Fig.8 Hydrochemical characteristics of high/low TDS outliers in groundwater

参考文献 44

[1]	何宝南, 何江涛, 孙继朝, 等. 区域地下水污染综合评价研究现状与建议[J]. 地学前缘, 2022, 29(3): 51-63. DOI
[2]	郭高轩, 辛宝东, 刘文臣, 等. 我国地下水环境背景值研究综述[J]. 水文地质工程地质, 2010, 37(2): 95-98.
[3]	孟凡生, 张铃松, 姚志鹏, 等. 黑龙江流域水环境背景值研究进展[J]. 中国环境监测, 2021, 37(6): 14-20.
[4]	GAO Y Y, QIAN H, HUO C C, et al. Assessing natural background levels in shallow groundwater in a large semiarid drainage basin[J]. Journal of Hydrology, 2020, 584: 124638.
[5]	GEMITZI A. Evaluating the anthropogenic impacts ongroundwaters: a methodology based on the determination of natural background levels and threshold values[J]. Environmental Earth Sciences, 2012, 67(8): 2223-2237.
[6]	宇庆华, 曹玉和. 地下水化学背景值研究中的异常值判定与处理[J]. 吉林地质, 1991, 10(2): 75-79.
[7]	寇文杰, 谢振华, 赵立新, 等. 探讨地下水背景值确定方法及其容易忽视的几个问题[J]. 安徽农业科学, 2013, 41(8): 3603-3605.
[8]	MOREAU M, DAUGHNEY C. Defining natural baselines for rates of change in New Zealand’s groundwater quality: dealing with incomplete or disparate datasets, accounting for impacted sites, and merging into state of the-environment reporting[J]. Science of the Total Environment, 2021, 755: 143292.
[9]	MORGENSTERN U, DAUGHNEY C J. Groundwater age for identification of baseline groundwater quality and impacts of land-use intensification: the National Groundwater Monitoring Programme of New Zealand[J]. Journal of Hydrology, 2012, 456/457: 79-93.
[10]	耿婷婷, 李颖智, 张涛, 等. 地下水环境背景值研究进展的分析与建议[J]. 环境科学与管理, 2018, 43(5): 33-35.
[11]	王磊, 何江涛, 张振国, 等. 基于信息筛选和拉依达准则识别地下水主要组分水化学异常的方法研究[J]. 环境科学学报, 2018, 38(3): 919-929.
[12]	张小文, 何江涛, 彭聪, 等. 地下水主要组分水化学异常识别方法对比: 以柳江盆地为例[J]. 环境科学, 2017, 38(8): 3225-3234.
[13]	彭聪, 何江涛, 廖磊, 等. 应用水化学方法识别人类活动对地下水水质影响程度: 以柳江盆地为例[J]. 地学前缘, 2017, 24(1): 321-331. DOI
[14]	曹文庚, 付宇, 南天, 等. 机器学习在地下水环境背景值与污染风险评价的应用和展望[J]. 地质学报, 2023, 97 (7): 2408-2424.
[15]	SAKO A, OUANGARÉ C A C. Hydrogeochemical characterization and natural background level determination of selected inorganic substances in groundwater from a semi-confined aquifer in Midwestern Burkina Faso, West Africa[J]. Environmental Monitoring and Assessment, 2023, 195(4): 519. DOI PMID
[16]	李晓媛, 张翼龙, 王丽娟, 等. 河套盆地地下水环境背景值研究[J]. 干旱区资源与环境, 2020, 34(3): 180-187.
[17]	刘慧文, 姜纪沂, 刘春平, 等. 新疆伊犁河谷平原区地下水环境背景值研究[J]. 南水北调与水利科技, 2016, 14(4): 107-111.
[18]	廖磊, 何江涛, 曾颖, 等. 柳江盆地浅层地下水硝酸盐背景值研究[J]. 中国地质, 2016, 43(2): 671-682.
[19]	刘左, 潘欢迎. 湖北省平原岗区地下水环境背景值初步研究[J]. 安全与环境工程, 2023, 30(3): 208-221.
[20]	廖磊, 何江涛, 彭聪, 等. 地下水次要组分视背景值研究: 以柳江盆地为例[J]. 地学前缘, 2018, 25(1): 267-275. DOI
[21]	何建国, 王莉, 章梅, 等. 徐州沛县地下水应急水源地溶解性总固体(TDS)本底值调查研究[J]. 地下水, 2021, 43(3): 27-29, 65.
[22]	刘文波, 冯翠娥, 高存荣. 河套平原地下水环境背景值[J]. 地学前缘, 2014, 21(4): 147-157.
[23]	曾颖. 秦皇岛柳江盆地浅层地下水常规组分背景值研究[D]. 北京: 中国地质大学(北京), 2015.
[24]	张小文, 何江涛, 刘丹丹, 等. 滹沱河冲洪积扇浅层地下水水质外界胁迫作用分析[J]. 水文地质工程地质, 2018, 45(5): 48-56.
[25]	HE B N, HE J T, ZENG Y, et al. Coupling of multi-hydrochemical and statistical methods for identifying apparent background levels of major components and anthropogenic anomalous activities in shallow groundwater of the Liujiang Basin, China[J]. Science of the Total Environment, 2022, 838: 155905.
[26]	PENG C, HE J T, WANG M L, et al. Identifying and assessing human activity impacts on groundwater quality throughhydrogeochemical anomalies and NO₃^-, NH₄⁺, and COD contamination: a case study of the Liujiang River Basin, Hebei Province, P.R. China[J]. Environmental Science and Pollution Research International, 2018, 25(4): 3539-3556.
[27]	单晓杰, 何江涛, 张小文, 等. 基于人为活动影响识别的区域地下水水质演化预测: 以石家庄地区为例[J]. 现代地质, 2020, 34(1): 189-198.
[28]	HUANG G X, SUN J C, ZHANG Y, et al. Impact of anthropogenic and natural processes on the evolution of groundwater chemistry in a rapidly urbanized coastal area, South China[J]. Science ofthe Total Environment, 2013, 463/464: 209-221.
[29]	王小平, 张飞, 于海洋, 等. 基于多元线性模型、支持向量机(SVM)模型和地统计方法的地表水溶解性总固体(TDS)估算及其精度对比: 以艾比湖流域为例[J]. 环境化学, 2017, 36(3): 666-676.
[30]	ALDREES A, JAVED M F, BAKHEIT TAHA A T, et al. Evolutionary and ensemble machine learning predictive models for evaluation of water quality[J]. Journal of Hydrology: Regional Studies, 2023, 46: 101331.
[31]	JAMEI M, AHMADIANFAR I, CHU X F, et al. Prediction of surface water total dissolved solids using hybridized wavelet-multigene genetic programming: new approach[J]. Journal of Hydrology, 2020, 589: 125335.
[32]	SHVARTSEV S L. Geochemistry of fresh groundwater in the main landscape zones of the Earth[J]. Geochemistry International, 2008, 46(13): 1285-1398.
[33]	XUN Z, HUA Z, LIANG Z, et al. Some factors affecting TDS and pH values in groundwater of the Beihai coastal area in Southern Guangxi, China[J]. Environmental Geology, 2007, 53(2): 317-323.
[34]	张小文, 何江涛, 黄冠星. 石家庄地区浅层地下水铁锰分布特征及影响因素分析[J]. 地学前缘, 2021, 28(4): 206-218. DOI
[35]	SARKAR S, MUKHERJEE A, GUPTA S D, et al. Predicting regional-scale elevated groundwater nitrate contamination risk using machine learning on natural and human-induced factors[J]. ACS ES&T Engineering, 2022, 2(4): 689-702.
[36]	张焕宝, 贺海洋, 杨仕教, 等. 基于机器学习的埃达克质岩构造背景判别研究[J]. 地学前缘, 2024, 31(4): 417-428. DOI
[37]	KNOLL L, BREUER L, BACH M. Nation-wide estimation of groundwater redox conditions and nitrate concentrations through machine learning[J]. Environmental Research Letters, 2020, 15(6): 064004.
[38]	张林泉. 线性回归模型的置信区间与预测区间应用分析[J]. 吉首大学学报(自然科学版), 2013, 34(6): 15-18.
[39]	PACHECO F, VAN DER WEIJDEN C H. Contributions of water-rock interactions to the composition of groundwater in areas with a sizeable anthropogenic input: a case study of the waters of the Fundão area, central Portugal[J]. Water Resources Research, 1996, 32(12): 3553-3570.
[40]	RAHMAN A, MONDAL N C, FAUZIA F. Arsenic enrichment and its natural background in groundwater at the proximity of active floodplains of Ganga River, Northern India[J]. Chemosphere, 2021, 265: 129096.
[41]	陈荦. 沙颍河流域地下水流与硝酸盐运移模拟及其对地表水污染的贡献研究[D]. 南京: 南京大学, 2013.
[42]	HE B N, HE J T, WANG L, et al. Effect of hydrogeological conditions and surface loads on shallow groundwater nitrate pollution in the Shaying River Basin: based on least squares surface fitting model[J]. Water Research, 2019, 163: 114880.
[43]	XIA Q W, HE J T, HE B N, et al. Effect and genesis of soil nitrogen loading and hydrogeological conditions on the distribution of shallow groundwater nitrogen pollution in the North China Plain[J]. Water Research, 2023, 243: 120346.
[44]	CUI R Y, ZHANG D, HU W L, et al. Nitrogen in soil, manure and sewage has become a major challenge in controlling nitrate pollution in groundwater around plateau lakes, Southwest China[J]. Journal of Hydrology, 2023, 620: 129541.